text
stringlengths 4
2.78M
| meta
dict |
|---|---|
---
abstract: 'We describe an apparatus for quickly and simply producing $\Rb87$ Bose-Einstein condensates. It is based on a magnetic quadrupole trap and a red detuned optical dipole trap. We collect atoms in a magneto-optical trap (MOT) and then capture the atom in a magnetic quadrupole trap and force rf evaporation. We then transfer the resulting cold, dense cloud into a spatially mode-matched optical dipole trap by lowering the quadrupole field gradient to below gravity. This technique combines the efficient capture of atoms from a MOT into a magnetic trap with the rapid evaporation of optical dipole traps; the approach is insensitive to the peak quadrupole gradient and the precise trapping beam waist. Our system reliably produces a condensate with $N\approx2\times10^6$ atoms every $16\second$.'
author:
- 'Y.-J. Lin'
- 'A. R. Perry'
- 'R. L. Compton'
- 'I. B. Spielman'
- 'J. V. Porto'
title: 'Rapid production of $^{87}$Rb BECs in a combined magnetic and optical potential'
---
A range of techniques have been developed to produce quantum degenerate gases, of varying degrees of complexity and difficulty. Almost all current methods rely on the same basic approach: laser cooling of atoms [^1], followed by evaporative cooling in a conservative trap [@Anderson1995; @Davis1995a]. When designing there are many often competing considerations: reliability, speed, simplicity, large optical access, and reasonably large number. We describe here a simple approach that quickly produces a relatively large Bose-Einstein condensate (BEC) of $\Rb87$ atoms.
Magnetic quadrupole traps for neutral atoms [@migdall85a] have several strengths. They have large trap volumes, which can be well matched to the size of laser cooled atom clouds. The effectively linear potential provides tight confinement, allowing for efficient evaporative cooling, and the quadrupole field can be generated from a simple arrangement of two electromagnetic coils which allows for good optical access to the sample. The quadrupole trap’s one major drawback is Majorana spin-flip losses and the resulting heating near the zero-field point at the center of the trap. This limits forced evaporative cooling to relatively low phase-space densities [@Davis1995]. Still, the first two dilute-gas BEC’s were produced in traps derived from the simple quadrupole trap, taking advantage of its strengths by avoiding the Majorana losses in two different ways: by providing a time-orbiting bias field (“TOP” trap) to shift the zero away from the atom cloud [@petrich95a; @Anderson1995], or by using a repulsive optical potential (i.e., an optical plug) to push the atoms away from the zero [@Davis1995a]. The TOP trap is still widely used, and the optically plugged trap has successfully been revisited [@naik05blochraizen].
Far detuned optical dipole traps have a different set of strengths. They can be spin-state independent, made in flexible geometries that provide good optical access, have tight confinement and efficient evaporation, and they don’t require magnetic coils. One drawback is that simultaneously deep and large-volume traps require prohibitively large laser power. One must carefully “mode match” the laser-cooled atoms to the optical dipole trap, and all-optical approaches tend to produce relatively small BEC’s [@barrett01a; @weber03a; @cennini03a]. By loading into large-volume, shallow traps and transferring into smaller volume traps [@weber03a], larger BEC’s can be made [@Kinoshita2005], but this requires superlative high-density laser cooling. It has been suggested that BEC’s could be efficiently produced using a quadrupole trap as a reservoir to directly feed a red-detuned dipole trap [@comparat06a]. We demonstrate here such a hybrid technique which combines the advantages of quadrupole and optical traps while avoiding their individual weaknesses. We can then load the BEC from the hybrid trap into an all-optical dipole trap without an additional blue-detuned optical plug.
The basic approach is to load laser cooled atoms into a quadrupole trap, and use forced rf evaporation [@Davis1995] to reduce the phase space density until Majorana spin-flips cause significant loss and heating. We then transfer atoms from the quadrupole trap to a single-beam optical dipole trap. This step can be fairly efficient, transferring a large fraction of the atoms from the quadrupole trap into the optical trap. To limit Majorana losses during and after the transfer, the center of the optical trap is offset by roughly a beam waist from the field zero of the magnetic trap. Further, offset quadrupole field provides harmonic confinement along the beam direction; this approach therefore yields 3D confinement without a crossed dipole trap. Since the atom cloud is adiabatically cooled during the transfer, only a relatively small optical trap depth ($\sim
49$ $\mu$K in our case) is required. The change of the trap shape during the adiabatic transfer, from linear quadrupole to harmonic, modifies the density of states in such a way as to increase the phase space density at constant entropy, leading to fairly high initial phase space density in the dipole trap. Forced evaporative cooling by lowering the optical trap power [@barrett01a] then leads to quantum degeneracy. This scheme is simple, flexible and allows for significant optical access. We reliably produce $^{87}$Rb BEC’s with $N=2\times10^6$ atoms in a time $t\approx16\second$.
Theory
======
In order to discuss this approach, we first provide a brief overview of trapped gas thermodynamics and cooling for our trap geometry. (More detailed discussions can be found in Refs. [@luiten96a; @ketterle96b] and further references therein.) The thermodynamic properties of the gas are determined by the partition function $\zeta = V_0/\Lambda^{3}$, where $\Lambda=(2 \pi
\hbar^2 / m k_B T)^{1/2}$ is the thermal de Broglie wavelength and $$V_0 = \int \mathrm{exp}[-U({\bf r})/k_B T] \ d^3r \label{eq:V0}$$ is an effective trap volume, where $U({\bf r})$ is the trapping potential with the energy minimum $U({\bf r_{\rm min}})=0$ at position $\bf r_{\rm min}$, and $T$ is the temperature. Corrections to $V_0$ due to the finite trap depth $\epsilon_t = \eta k_B T$ during evaporation can be treated using a truncated Boltzmann distribution [@luiten96a]. The large-$\eta$ limit, Eqn. \[eq:V0\], is reasonable for most purposes when $\eta \gtrsim
8$, and below this range provides a qualitative description. From $\zeta$ and $V_0$, one can calculate, for example, the free energy $A = -N k_B T \ln{\zeta} $, the entropy $S = -\partial A/\partial
T$, the peak phase space density $D=N/\zeta$ and the density distribution $$n({\bf r}) =n_0\ \mathrm{exp}[{-U({\bf r})/k_B T}]=\frac{N}{ V_0}
\mathrm{exp}[{-U({\bf r})/k_B T}].$$ Dynamic rates relevant for evaporation, such as $n$-body loss, collision rates, evaporation rates, etc., can be calculated from similar integrals [@luiten96a].
![[*Top:*]{} $V_0$ as a function of temperature for several different field gradients. Thick lines: numerical calculation for field gradients $B^\prime = 1.60$, 0.45, 0.32 and 0.306 T/m. (For $\Rb87$ magnetically trapped in $|F,m_F\rangle=|1,-1\rangle$, $m g=\mu \times 0.30492$ T/m.) Dashed lines: analytical approximations, Eqns. \[eq:V0high\] and \[eq:V0low\]. The dipole trap depth for all plots (indicated by the arrow) is $U_0/k_B=49$ $\mu$K, and the beam waist is 65 $\mu$m along $\hat x$ and 78 $\mu$m along $\hat z$ (which accounts for the difference in measured trap frequencies along $\hat x$ and $\hat z$). [*Bottom:*]{} The entropy per particle, $S(T)/N k_B$, for the same field gradients in the dipole-plus-quadrupole trap (solid lines) and for quadrupole only traps (dashed lines). The short horizontal marks crossing the curves indicate the regions where the atoms are transferred into the dipole trap: the upper set marks where 10% of the atoms are in the dipole trap, and the lower set marks where 90% of the atoms are in the dipole trap. []{data-label="fig:V0andS"}](V0andS){width="3.5in"}
The effective potential for atoms in the combined quadrupole plus dipole trap, including gravity is
$$U({\bf r}) = \mu B^\prime \sqrt{\frac{x^2}{4}+\frac{y^2}{4}+z^2} -
U_0 \mathrm{exp}\{{-2
\left[x^2+(z-z_0)^2\right]/w_0^2}\} + m g z +E_0.$$
where $B^\prime$ is the quadrupole field gradient along $\hat{z}$, $U_0, w_0$ and $z_0$ are the dipole beam trap depth, waist and offset from the zero field point at $x,y,z=0$. $E_0$ is the energy difference between the zero field point absent the dipole trap and the total trap minimum, giving the trap minimum $U({\bf r_{\rm min}})=0$. $\mu$ and $m$ are the magnetic moment and mass of the atom, respectively, and $g$ is the acceleration due to gravity. Here the dipole beam is aligned along $\hat y$, and is displaced vertically (along $\hat{z}$) below the magnetic field zero. (We have ignored the focusing of the beam along $\hat{y}$, since for our waist $w_0=65~\micron$ the Raleigh length $\pi w_0^2 / \lambda
\approx 9\mm$ is large.) The effective volume as a function of temperature can be calculated from Eqn. \[eq:V0\].
At high temperature, by approximating $U({\bf r})$ as $$U({\bf r}) \simeq \mu B^\prime \sqrt{\frac{x^2}{4}+\frac{y^2}{4}+z^2} + E_0 + m g z,$$ we have $$V_0(T) \simeq
\frac{ 32 \pi e^{-E_0/k_B T}}{\left[ 1- (m g/ \mu B^\prime)^2\right]^2} \left(\frac{k_B T}{\mu B^\prime}\right)^3,
\ \ {\rm for}\ \ T \gg U_0/k_B ,\label{eq:V0high}$$ where $E_0 \simeq U_0$ for a typical dipole beam offset $z_0 \simeq w_0$. This approximation is only valid for field gradients that compensate gravity, $\mu B^\prime > m g $. The thermodynamic effect of gravity can be viewed as re-scaling the field gradient to $\mu B^\prime_{eff} = \mu B^\prime [1- (m g/\mu
B^\prime)^2]^{2/3}$, which vanishes at $\mu B^\prime = mg$. In the absence of the dipole trap and ignoring gravity, the effective volume reduces to the simple quadrupole form $V_0(T) = 32 \pi (k_B
T/\mu B^\prime)^3$, but for most parameters we encounter, both corrections are important.
![Cross-sections of trapping potential with the offset $E_0$ subtracted, $U({\bf r})-E_0$, at several points during an adiabatic expansion ($B^\prime = 1.60$, 0.90, 0.45, and 0.306 T/m): (a) Along $\hat z$ (gravity is along $-\hat z$) at $x,y=0$, the potential minima $z = z_{\rm min}$ are indicated by filled circles, and (b) Along $\hat y$ (the dipole beam direction) at $x=0$, each taken at $z = z_{\rm min}$. The dashed line is the potential at 1.60 T/m without the dipole trap. The dipole trap parameters are the same as for Fig. \[fig:V0andS\]. []{data-label="fig:Uexpansion"}](Uvsyz){width="3.5in"}
At low temperature the dipole potential can be approximated as a harmonic trap $$U({\bf r}) \simeq \frac{1}{2}\left[U^{\prime \prime}_x x^2+U^{\prime
\prime}_y y^2+U^{\prime \prime}_z (z-z_{\rm min})^2 \right],$$ where $U_i^{\prime \prime}$ are the curvatures at the bottom of the total trap, and the trap minimum is at position $z_{\rm min}$. This gives $$V_0(T) \simeq \frac{\ (2\pi k_B T)^{3/2}}{\sqrt{U^{\prime \prime}_x U^{\prime \prime}_y U^{\prime \prime}_z}},\ \ {\rm for}\ \ T\ll U_0/k_B. \label{eq:V0low}$$ The curvatures along $\hat x$ and $\hat z$ are dominated by the dipole trap, $\omega_{x,z} \simeq 2 \sqrt{U/m w_0^2}$, and the curvature along $\hat y$ is dominated by the magnetic trap, $\omega_x = (1/2)\sqrt{\mu B^\prime/m z_{\rm min}}$. As shown in Fig. \[fig:V0andS\], both the high and low temperature approximations for the volume $V_0$ and entropy $S$ are quite good over a range of temperatures. The high temperature approximation is accurate well below $U_0/k_B$. The success of our hybrid approach depends critically on managing two processes: limiting Majorana loss in the quadrupole trap and effective adiabatic (approximately constant entropy) transfer from the quadrupole trap to the harmonic dipole trap. Majorana loss is particularly detrimental, as it causes both loss and heating. Such loss can be difficult to describe accurately, but a simple argument [@petrich95a] leads to the estimate that the Majorana loss rate $\Gamma_m$ scales as $\Gamma_m \propto \hbar/ m
l^2$, where $l$ is the radial half width half maximum cloud size in the quadrupole trap. Using the proportionality constant measured in [@petrich95a], we can estimate $$\begin{aligned}
\Gamma_m & =& 3.6 \frac{ \hbar}{m l^2}\nonumber \\
& = & 1.85
\frac{ \hbar}{m}\left(\frac{\mu B^\prime}{k_B T}\right)^2,
\label{eq:Majorana}
\end{aligned}$$ where we have ignored gravity. This estimate is independent of the elastic collision rate in the trap since it assumes thermal equilibrium, an invalid assumption at low densities or large loss rates. Nonetheless, it suggests that for field gradients of $\sim
1.40$ T/m, the lifetime should be on the order of one second at 20 $\mu$K. Near this temperature fast evaporation (high elastic collision rate) is advantageous. This loss rate can be mediated somewhat by adiabatically expanding the trap, since the temperature scales more weakly than the field gradient at constant entropy, $T
\propto (B^\prime)^{2/3}$, and weakening the trap will decrease the loss rate, at the expense of a lower collisional rate.
The adiabatic transfer is more subtle than a mere expansion of the trap volume, since in addition to expanding the trap, the shape of the trap is modified. This leads to changes in phase-space density, even at constant entropy [@stamper-kurn98a], an effect used to obtain a BEC of cesium from a very cold, but dilute gas [@weber03a]. In our case, the effect is quite strong, as the transfer process both decreases the temperature and increases phase space density.
The trap potential during an example expansion sequence is shown in Fig. \[fig:Uexpansion\]. The change of the trap from linear to quadratic dramatically changes the dependence of the entropy on temperature, from $S\propto \ln{T^{9/2}}$ at high $T$ and $B^\prime$, to $S\propto \ln{T^{3}}$ at low $T$ and $B^\prime$ (see Fig. \[fig:V0andS\]). The crossover temperature, $T_{\mathrm x}$, between these two regimes occurs roughly when the high and low temperature approximations for $V_0(T)$ (Eqns. \[eq:V0high\] and \[eq:V0low\]) are equal. The steep nature of $S(T)$ near $T_{\mathrm x}$ for $\mu B^\prime \simeq m g$ indicates that adiabatic expansion from a wide range of initial temperatures leads to a roughly constant final $T\simeq T_{\mathrm x}$, which is about $0.1 U_0/k_B$. The steep part of $S(T)$ near $T_{\mathrm x}$ also corresponds to the region over which the atoms are transferred into the dipole trap. Above the crossover region, a negligible fraction of atoms are in the dipole trap, but below this region, nearly all the atoms are in the dipole trap.
The adiabatic expansion and transfer is similar to forced evaporation, except that here the “evaporated" atoms during the expansion are contained in the low density tails of the weakly confining quadrupole trap. Many considerations are similar to evaporation, such as the final temperature being set by the dipole trap “depth" for temperatures $T_{\mathrm x} \simeq 0.1 U_0/k_B$. The ratio $\simeq 0.1$ is insensitive to $U_0$ and $w_0$, and is approximately a constant for 10 $\mu$K $<U_{0}<$ 45 $\mu$K with $w_0$=65 $\mu$m, and for 65 $\mu$m$<w_{0}<$ 190 $\mu$m with $U_0$=45 $\mu$K. As with evaporation, there is in principle no upper bound on the efficiency of the process [@ketterle96b], so that nearly all the atoms could be cooled in the dipole trap from any initial temperature. This would take nearly infinite time, however, and in practice the actual process depends on the details of the loss mechanisms and the elastic collision rates throughout the trap. Therefore, the loss will be more pronounced for a trap at lower densities with lower collision rates.
Figure \[fig:PSDandGamma\] shows the phase space density per particle, $D/N$, as a function of temperature for a range of $B^\prime$. To illustrate the expansion, several trajectories at constant entropy are shown for a few different initial temperatures. These plots show that if thermal equilibrium could be maintained and absent any loss mechanism, extremely large increases in phase space density are possible. In practice, the collision rate in the low density tails of the atom cloud eventually drops too low to maintain equilibrium. The actual efficiency of the process, and the optimal trajectory, will depend on this decoupling, as well as Majorana and one-body loss.
![[*Top:*]{} The phase space density per particle $D(T)/N$ in the dipole-plus-quadrupole trap for the same field gradients as in Fig. \[fig:V0andS\], assuming no loss and thermal equilibrium (solid lines). Example adiabatic trajectories for $D(T)/N$, calculated at constant entropy starting at $B^\prime = 1.6$ T/m, are shown for different starting temperatures 30, 50, and 70 $\mu$K (dashed lines). [*Bottom:*]{} The average elastic collision rate, $\Gamma_{\mathrm el}/N$ for the same field gradients (solid lines). The collision rate is shown for the same constant entropy trajectories as above (dashed lines). []{data-label="fig:PSDandGamma"}](PSDandGamma){width="3.5in"}
Experiment
==========
The cooling process begins with a thermal beam of rubidium atoms cooled and decelerated by a zero crossing Zeeman slower. The slowed atoms are loaded into a six beam MOT and then transferred to the magnetic quadrupole trap where they are cooled by forced rf evaporation. In the final stage of cooling the atoms are transfered into the hybrid optical-dipole plus magnetic trap where they reach degeneracy.
Vacuum system
-------------
Our apparatus consists of two ultra-high vacuum (UHV) zones: a rubidium oven and the main experimental chamber. Each chamber is pumped with a single $55\liter\second^{-1}$ ion pump. In addition, the experimental chamber is pumped with a Titanium sublimation (TiSub) pump attached to a bellows. When operated, the TiSub filament can be positioned directly in-line with Zeeman slower, allowing it to coat the inside of the main vacuum chamber while avoiding the optical windows of the chamber. After evaporation it is retracted from the slower axis.
The oven is separated from the main experimental chamber by an in-vacuum shutter, a differential pumping tube ($1.1\cm$ inner diameter, $7\cm$ long), a pneumatic gate valve (normally open), and the Zeeman slower (a $0.9\meter$ long stainless tube with a $3.5
\cm$ inner diameter). Together these tubes give a calculated conductance of $1.3\liter\second^{-1}$ [@OHanlon2003]. The differential pumping between the chamber and the oven allow the chamber pressure to be significantly below the oven pressure. The oven has a nominal pressure of $5\times10^{-10}\mbar$ measured using the ion pump current. The ion pump current in the experimental chamber is zero to the accuracy of the pump controller (we observe a $45\second$ magnetic trap lifetime for $T=170\uK$ atoms, consistent with loss dominated by Majorana transitions, see Fig. \[fig:QuadrupoleLifetime\]).
![Cut-away diagram of the main chamber. The main chamber is equipped with recessed windows in which we have mounted 24-turn water cooled coils (A), and single turn rf evaporation loops (B). []{data-label="fig:Chamber"}](ChamberLayout){width="3.375in"}
The main chamber consists of a single stainless steel assembly with two $14\cm$ diameter recessed windows in the top and bottom (Fig. \[fig:Chamber\]). The windows are recessed as much as possible without blocking the line-of-sight from the remaining 16 mini-conflat ($d = 3.8\cm$) and 6 larger conflat ($d = 7\cm$) viewports.
Oven
----
![Schematic diagram of rubidium oven. The left portion of the figure shows the heated rubidium oven, indicating the regions of different temperature. The right part of the figure is a cut-away of the main oven chamber showing the end of the collimation tube and the in-vacuum cold cup. []{data-label="fig:Oven"}](OvenLayout){width="3.375in"}
The rubidium atomic beam originates from a heated rubidium reservoir (Fig. \[fig:Oven\]) and is collimated before entering the Zeeman slower. Efficient collimation is desirable, since it lengthens the time between rubidium reloads, and extends the lifetime of the ion pump.
The rubidium reservoir is contained within a $7.5\cm$ long stainless steel (SS) bellows and was initially loaded with a $5\g$ rubidium glass ampule. The ampule is broken under vacuum after baking. During operation the reservoir is held at $80\C$, setting the rubidium vapor pressure $\approx 6\times10^{-5}\mbar$ and the mean free path (MFP) to be $\lesssim 1\meter$ (molecular flow regime everywhere).
The reservoir is connected to a 4-way mini-ConFlat style cross and then to the collimating tube. The SS collimation tube has a $6\mm$ diameter and is about $22.5\cm$ in length. The 4-way cross and the first $7.5\cm$ of the collimation tube are maintained at $120\C$, providing a thermal gradient between the reservoir and the collimation tube. The remaining $15\cm$ experiences a thermal gradient due to the balance of thermal conduction and black-body radiation; we estimate the temperature of the free end of the tube to be $70\C$ (all points in the tube are above rubidium’s $39\C$ melting point). The oven chamber is thermally isolated from the $120\C$ oven by a glass thermal break. This oven is not a recirculating design [@Hau1994; @Walkiewicz2000; @Pailloux2007], but we expect that the elevated temperature of the collimation tube with respect to the reservoir keeps the tube free of rubidium.
We estimate that the collimation by the tube provides about a $25\times$ improvement in flux directed into our MOT compared to a simple aperture with the same total flux [@Beijerinck1975]. We estimate that 1% of all atoms departing the oven fall within the $\approx 1\cm$ radius of the MOT beams (neglecting the transverse heating and expansion of the beam from the Zeeman slowing process).
The outgoing atomic beam first passes through a $6\mm$ aperture in a copper cold cup; after a further $2\cm$ the beam proceeds through an in-vacuum shutter (not pictured), through a differential pumping tube and then enters the Zeeman slower. The cold cup is chilled to $\approx-30\C$ by a commercial thermoelectric (TEC) based CPU cooler and the thermal link into the vacuum chamber is a high current vacuum feed thru. We find that the cold cup capturing excess rubidium is essential to the long-term operation of our ion pumps, which can fail due to excess leakage current as they pump rubidium (similar alkali poisoning has been observed elsewhere [@RolstonPrivate]). Even for the collimated beam, nearly half of the outgoing rubidium is incident on the cold-cup.
To further inoculate our oven ion pump from rubidium poisoning, we permanently maintain the pump at $70\C$ to drive off rubidium. As of this writing, the oven with its initial $5\g$ rubidium sample has been under vacuum and operating essentially daily since Feb. 2006, with an operating pressure of $5\times10^{-10}\mbar$.
Zeeman slower
-------------
![Comparison of the calculated (continuous curve) and measured field profile (points) for our Zeeman slower. The measurement was performed with $50\amp$ in both the forward and reverse portions of the slower which would result in slowing at 50% of the peak acceleration. (We found the optimum operating condition to be $72\amp$ in the positive field portion and $42\amp$ in the negative field region.) []{data-label="fig:SlowerProfile"}](SlowerProfile){width="3.375in"}
We slow and cool the collimated atomic beam using a Zeeman slower [@Phillips1982]. Operating at maximum efficiency and with an optimal field profile, our $69\cm$ long slower would stop rubidium atoms with a peak velocity $v_{\rm
max}=390\meter\second^{-1}$ losing about $67\times10^{3}$ photon recoil momenta ($\lambda = 780\nm$). Our realized slower operates at nearly $70\%$ of the maximum deceleration.
The slower field profile is a zero crossing design [@witte92a], and is generated from two single-layer, helically wound coils which provide the positive and negative sections of the field profile. We modeled the field profile for the full variable-pitch coils and found an optimized winding pattern with 84 total turns (54 and 30 turns for the positive and negative field region, respectively). Ideally, the optimized designed profile would slow atoms up to $84\%$ of the maximum acceleration. The coils were wound from electrically insulated $1/8"$ ($3\mm$) diameter copper refrigerator tubing onto a $d=7.6$ cm aluminum form enclosing the ConFlat vacuum tube. In operation, cooling water forced through the copper tubes removes the heat dissipated by the $72\amp$ and $42\amp$ flowing through the positive and negative field portions of the slower, respectively.
The calculated and measured field profiles are shown in Fig. \[fig:SlowerProfile\]. (An additional compensation coil, located beyond the end of the slower, zeros the field in the region of the MOT is not shown.) While a multi-layer design could more optimally match the design field profile [@Dedman2004], we opted for the simplicity afforded with a single layer coil.
The Zeeman shifted atoms are slowed by a $30\mW$ laser beam counter-propagating with respect to the atomic beam (the slower laser has $\sigma^+$ polarization referenced to the positive field region of the slower). This laser is detuned $162\MHz$ below the ${|5{\rm S}_{1/2},F=2,m_F=2\rangle}$ to ${| 5{\rm
P}_{3/2},F=3,m_F=3\rangle}$ $\Rb87$ cycling transition. It enters the Zeeman slower with a $1\cm$ $1/e^2$ radius and reaches a focus in the oven collimation tube. A $15\mW$ repump laser beam, copropagating with the slower laser, prepares and maintains the atoms in the cycling transition. This $\sigma^-$ repump beam is detuned $165\MHz$ below the ${|5{\rm S}_{1/2},F=1\rangle}$ to ${|5{\rm P}_{3/2},F=2\rangle}$ transition, and is resonant with the slowed atoms in the low field portions of the slower. We find that the performance of the slower is fairly insensitive to $\approx50\MHz$ changes in the repump detuning.
The rubidium beam entering the slower is at $T\approx400\kelvin$ set by the $120\C$ portions of the oven. Assuming a Maxwell-Boltzman distribution the most probable velocity is $v=340\meter\second^{-1}$. Operating at about 70% of the maximum acceleration for rubidium, the peak capture velocity is $280\meter\second^{-1}$, indicating we can slow 25% of $\Rb87$ atoms in the full distribution.
Magneto-optical trap
--------------------
We capture our Zeeman slowed beam of atoms in a six beam MOT at the center of the main chamber, $15\cm$ from the end of the slower. The required quadrupole magnetic field is generated from the same coils used for magnetic trapping. For the MOT we run $25\amp$ in the quadrupole coils, giving a field gradient of $\simeq0.12\tesla\meter^{-1}$ along $\hat z$.
Each of the six MOT beam has about $20\mW$ of power and are detuned by $-21.5\MHz$ ($3.5$ linewidths) from the cycling transition. The MOT repumper has a total of $16\mW$ of power and is detuned by $-5\MHz$ from the ${|F=1\rangle}$ to ${|F=2\rangle}$ transition. The MOT cooling and repumping light is combined in fiber, and each of the six beams are separately collimated. The six collimated MOT beams are directed into the vacuum chamber by reflecting them from moveable flipper mirrors, which can be moved out of the optical path after the atoms have been transferred to the magnetic trap and the MOT light is no longer needed. This provides optical access for beams needed in the experiment once the BEC has been made.
The MOT loads to $N=9\times10^8$ atoms in $3\second$, and saturates to about $2.7\times10^9$ atoms in $20\second$. We infer the atom number by collecting light scattered from the MOT. (When the cloud is optically thick this can under-estimate the total number of atoms.) After $3\second$ of MOT loading, we perform polarization gradient cooling by turning off the magnetic trap, decreasing the intensity of the repump power to $100\uW$, and linearly increasing the detuning of the MOT beams to 114 MHz ($19$ linewidths) over the course of $19\ms$. At the end of the ramp the cloud has a temperature of $29(6)\uK$ [^2]. We then completely extinguish the repump light, depumping the atoms into the $F=1$ manifold.
Magnetic Trap {#sect:MagTrap}
-------------
Once the atoms are laser cooled, we transfer them to the hybrid of magnetic quadrupole and optical dipole trap. The quadrupole trap is constructed from a pair of $24$-turn coils placed within recessed windows above and below the main vacuum chamber (Fig. \[fig:Chamber\]). The coils are wound from Kapton insulated $3/16"$ ($0.48\cm$) hollow square copper tube. We flow water through the coils in parallel with a pressure differential of about $14~{\rm bar}$ ($200\ {\rm PSI}$) to remove the $\approx4.8\kW$ dissipated in the coils at a full current of $320\amp$. The coils can be effectively modeled as two 24 turn loops separated by $12\cm$ and each with a $5.8\cm$ radius (This is not the ideal anti-Helmholtz configuration that could be realized by further recessing our top and bottom windows. Such recessing would have obscured optical access). We measured a gradient of $4.8
\mT\meter^{-1}\amp^{-1}$ in the stiff direction (vertical).
We capture our laser cooled atoms by abruptly turning on the magnetic trap. For our MOT parameters, the optimal field gradient for best BEC production was $0.48\tesla\meter^{-1}$ along $\hat z$. Compressing to $1.6\tesla\meter^{-1}$ in $0.1\second$ increases the collisional rate and heats the atoms to $190(35)\uK$. Simultaneously with the adiabatic compression, we turn on a dipole beam along $\hat y$ to increase the density and therefore the evaporation speed. The calculated peak collision rate and phase space density are $15\second^{-1}$ and $5\times10^{-7}$. We then turn on the rf, and apply a linear sweep of the frequency from $20\MHz$ down to $3.75\MHz$ in $2.9\second$. During this rf-forced evaporation, the temperature and number decrease to $30(5)\uK$ and $N=8\times10^7$, and the calculated density increases to $4.5\times10^{12}\cm^{-3}$.
To characterize the loss from Majorana spin flips, we measure the quadrupole trap lifetime $\tau_m$ versus the temperature $T$ at $B^{'}=1.55\tesla\meter^{-1}$ absent the dipole trap (Fig. \[fig:QuadrupoleLifetime\]). During the lifetime measurement at each $T$, we apply a rf knife to maintain a constant $T$ given the heating from Majorana spin flips. At our densities the calculated collision time is approximately $200$ times shorter than $\tau_m$, which ensures thermal equilibrium. $\tau_m$ is found to scale as $\tau_m=AT^{p}$ with the best fit of $p$=1.6(1). A fit with fixed $p=2.0$ yields $A=5.8\times10^{-3} \second\uK^{-2}$, approximately a factor of 2 of that given by Eqn. \[eq:Majorana\], which was calibrated at a $\approx 50\%$ larger field gradient, $B^{'}=2.4$ $\tesla\meter^{-1}$.
![Lifetime $\tau_m$ and collision times $\Gamma_{\rm{el}}^{-1}$ in the compressed quadrupole trap at the field gradient $B^{'}=1.55\tesla\meter^{-1}$. Solid circles: measured lifetime in the quadrupole trap showing Majorana losses. Empty circles: calculated collision time from measured number and trap parameters.[]{data-label="fig:QuadrupoleLifetime"}](mjlifetime){width="3.375in"}
![Temperature $T$ versus quadrupole field gradient $B^{'}$ during the transfer into the dipole trap. Circles indicate the measurement, where the data point at the lowest $T$ corresponds to the end of the transfer at $\mu B^{'}<mg$, and thus free evaporation. For comparison, the shaded region denotes calculation for a constant entropy per atom, with the uncertainty from dipole trap parameters.[]{data-label="fig:tempgrad"}](tempgrad){width="3.2in"}
{width="6.27in"}
Transfer to the dipole trap
---------------------------
Next, we adiabatically ramp down the quadrupole field gradient $B^{'}$ from $1.6\tesla\meter^{-1}$ to $0.30\tesla\meter^{-1}$ in $3\second$, leaving the rf power on and ramping down the frequency to 2 MHz withing the first $1\second$, after which it remains constant. We turn off the rf at the end of the magnetic decompression when $B^{'}$ is smaller than the gravity, $m g /\mu = 0.305\tesla\meter^{-1}$, and therefore loading the atoms into the dipole trap. The $3.5\watt$, $\lambda=1550\nm$ dipole beam has a waist $w_0 = 65 \micron$. We offset the center of dipole beam by $z_0 \sim w_0$ from the zero of magnetic field in order to decrease the Majorana loss, where the reduction factor scales as the density ratio $n(0)/n({\bf r}_{\rm
min})=\exp{(-E^{'}/k_B T)}$. $n(0)$ and $n({\bf r}_{\rm min})$ are the densities at zero field point and at the trap center ${\bf r}_{\rm min}$, respectively, and $E^{'}$ is the energy difference between ${\bf r}=0$ and ${\bf r}_{\rm min}$. We transfer the atoms into the dipole trap with a depth $U= 49(6)\uK$ and $E^{'}=36(5)\uK$. The trap frequencies are $270\Hz$ and $320(30)\Hz$ along $\hat x$ and $\hat z$, respectively, and $30\Hz$ along $\hat{y}$ which is provided by the magnetic confinement at a field of $20\uT$ along $\hat z$ at the trap bottom.
During the $3\second$ loading to the dipole trap, the truncation parameter $\eta$ increases from 6 to 10, arising from the rf knife and $\approx 0.2\second$ of free evaporation at the end. The atom number decreases by a factor of $\approx 4$. Figure \[fig:tempgrad\] shows the temperature $T$ versus the quadrupole field gradient $B^{'}$ during the loading. The entropy per atom $S/Nk_{B}$ is initially 31.5, then slightly decreases by $\lesssim 1.5$, and the cloud cools from $T=30\uK$ to $5.2\uK$.
After loading the atoms into the dipole trap, we then perform forced evaporation by lowering the dipole trap depth from $49(6)\uK$ to a variable $U$ in $4\second$, and simultaneously decrease $B^{'}$ from $0.30\tesla\meter^{-1}$ to $0.27\tesla\meter^{-1}$. The trap depth versus time is approximately quadratic, slower toward the end the evaporation. At $U = 4.7(7)\uK$, a bimodal distribution in the time-of-flight (TOF) image shows a $6\ \%$ condensation fraction of the total number $N=4\times 10^6$ with a temperature $T=0.32(5)\uK$ (calculated $T_c = 0.5\uK$). At $U=1.4(2)\uK$, the condensate is nearly pure with $N= 1.8 \times 10^6$. Figure \[fig:evap\] shows the atom number $N$ and the temperature $T$ versus the trapping time, and the evolution of phase space density $D$ versus $N$ in the following steps: (i) rf evaporation in the magnetic trap (ii) transfer to the dipole trap (iii) evaporation in the dipole trap.
We have also successfully produced condensates with similar atom numbers adopting the same hybrid technique using different of dipole beams parameters. They include: (1) a different waist $w_0\sim 90\micron$ still at $\lambda= 1550$ nm, and (2) a different wavelength $\lambda= 1064$ nm and waist $w_0\sim 70\micron$. The robustness of our approach to these changes illustrates the generality and versatility of such hybrid magnetic and optical trapping approach.
Conclusion
==========
We have described a simple and effective technique of producing $\Rb87$ Bose-Einstein condensates which merges the best aspects of magnetic and optical trapping. A condensate of $\sim 2\times 10^6$ atoms is produced in a cycle time of $16\second$. This approach applies for a wide range of trap parameters, such as the quadrupole field gradient, the dipole beam trap depth and the waist. We expect that this technique can be generally applicable to other magnetically trappable atomic species.
This work was partially supported by ONR, ODNI, ARO with funds from the DARPA OLE program, and the NSF through the JQI Physics Frontier Center. R.L.C. acknowledges supports from NIST/NRC.
[25]{} natexlab\#1[\#1]{}bibnamefont \#1[\#1]{}bibfnamefont \#1[\#1]{}citenamefont \#1[\#1]{}url \#1[`#1`]{}urlprefix\[2\][\#2]{} \[2\]\[\][[\#2](#2)]{}
, , , , , **** ().
, , , , , , , ****, ().
, , , , , ****, (), ISSN .
, , , , , ****, ().
, , , , ****, (), ISSN .
, ****, (), ISSN ; , ; , .
, , , ****, (), ISSN .
, , , , , ****, (), ISSN .
, , , , ****, (), ISSN .
, , , ****, ().
, , , , , , ****, (), ISSN .
, , , ****, ().
, ****, ().
, , , , , , ****, ().
, ** (, ), ed.
, , , ****, ().
, , , ****, ().
, , , ****, ().
, ****, ().
, .
, ****, ().
, , , , ****, ().
, , , , , , ****, ().
, , , , , , , ****, ().
[^1]: Bose-condensation of atomic hydrogen is an exception where the initial cold atomic sample was prepared not by laser cooling, but instead by thermalization with a buffer gas [@Fried1998].
[^2]: Uncertainties reflect the uncorrelated combination of 1-$\sigma$ statistical and systematic uncertainties.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'This paper studies a particular class of higher order conformally invariant differential operators and related integral operators acting on functions taking values in particular finite dimensional irreducible representations of the Spin group. The differential operators can be seen as a generalization to higher spin spaces of $k$th-powers of the Euclidean Dirac operator. To construct these operators, we use the framework of higher spin theory in Clifford analysis, in which irreducible representations of the Spin group are realized as polynomial spaces satisfying a particular system of differential equations. As a consequence, these operators act on functions taking values in the space of homogeneous harmonic or monogenic polynomials depending on the order. Moreover, we classify these operators in analogy with the quantization of angular momentum in quantum mechanics to unify the terminology used in studying higher order higher spin conformally invariant operators: for integer and half-integer spin, these are respectively bosonic and fermionic operators. Fundamental solutions and their conformal invariance are presented here.'
author:
- |
Chao Ding$^1$[^1][^2], Raymond Walter$^{1,2 *}$[^3], and John Ryan$^1$[^4]\
*$^1$Department of Mathematics, University of Arkansas, Fayetteville, AR 72701, USA*\
*$^2$Department of Physics, University of Arkansas, Fayetteville, AR 72701, USA*
title: Higher Order Fermionic and Bosonic Operators
---
[**Keywords:**]{}Higher order fermionic and bosonic operators, Conformal invariance, Fundamental solutions, Intertwining operators, Ellipticity.
Introduction
============
Classical Clifford analysis started as a generalization of aspects of one variable complex analysis to $m$-dimensional Euclidean spaces. At the heart of this theory is the study of the Dirac operator $D_x$ on ${\mathbb{R}^m}$, a conformally invariant first order differential operator which generalizes the role of the Cauchy-Riemann operator. Moreover, this operator is related to the Laplace operator with $D_x^2=-\Delta_x$. The classical theory is centered around the study of functions on ${\mathbb{R}^m}$ and taking values in a spinor space [@At; @Br], and abundant results have been found. See for instance [@Br; @De; @P; @G; @R; @R1].\
P.A.M. Dirac constructed a first order relativistically covariant equation describing the dynamics of an electron by using Clifford modules; hence differential operators constructed using Clifford modules are called Dirac operators. In the presence of an electromagnetic field, the Dirac Hamiltonian for an electron acquires an additional contribution formally analogous to internal angular momentum called spin, from which the Spin group and related notions take their name; for the electron, spin has the value $\frac{1}{2}$ [@Lawson]. Indeed, in dimension four with appropriate signature, null-solutions of the Dirac operator $D_x$ from classical Clifford analysis correspond to solutions for the relativistically covariant dynamical equation of a massless particle of spin $\frac{1}{2}$, also called the Weyl equation. The Dirac equation for the electron, which has mass, may be considered an inhomogeneous equation satisfied by the Dirac operator $D_x$. The Dirac equation is not only relativistically covariant, but also conformally invariant. The construction of conformally invariant massless wave equations, in terms of invariant operators with conformal weights over spin fields, is well described by [@E]. The general importance of conformal invariance in physics has long been recognized [@RevModPhys].\
Rarita and Schwinger [@Ra] introduced a simplified formulation of the theory of particles of arbitrary half-integer spin $k+\frac{1}{2}$ and in particular considered its implications for particles of spin $\frac{3}{2}$. In the context of Clifford analysis, the so-called *higher spin theory* was first introduced through the Rarita-Schwinger operator [@B], which is named analogously to the Dirac operator and reproduces the wave equations for a massless particle of arbitrary half-integer spin in four dimensions with appropriate signature [@Ro]. (The solutions to these wave equations may not be physical [@VZ1; @VZ2].) The higher spin theory studies generalizations of classical Clifford analysis techniques to higher spin spaces [@B1; @Br1; @B; @D; @E; @Li]. This theory concerns the study of the operators acting on functions on ${\mathbb{R}^m}$, taking values in arbitrary irreducible representations of $Spin(m)$. These arbitrary representations are defined in terms of polynomial spaces that satisfy certain differential equations, such as $j$-homogeneous monogenic polynomials (half-integer spin) or $j$-homogeneous harmonic polynomials (integer spin). More generally, one can consider the highest weight vector of the spin representation as a parameter [@HighestWeightExample], but this is beyond our present scope. The present paper contributes to the study of conformally invariant operators in the higher spin theory.\
In principle, all conformally invariant differential operators on locally conformally flat manifolds in higher spin theory are classified by Slovák [@J]; see also [@VS-ref]. This classification is non-constructive, showing only between which vector bundles these operators exist and what is their order; explicit expressions of these operators are still being found. Eelbode and Roels [@E] point out that the Laplace operator $\Delta_x$ is not conformally invariant anymore when it acts on $C^{\infty}({\mathbb{R}^m},{\mathcal{H}_1})$, where ${\mathcal{H}_1}$ is the degree one homogeneous harmonic polynomial space (correspondingly ${\mathcal{M}_1}$ for monogenic polynomials). They construct a second order conformally invariant operator on $C^{\infty}({\mathbb{R}^m},{\mathcal{H}_1})$, the (generalized) Maxwell operator. In dimension four with appropriate signature it reproduces the Maxwell equation, or the wave equation for a massless spin-1 particle (the massless Proca equation) [@E]. De Bie and his co-authors [@B1] generalize this Maxwell operator from $C^{\infty}({\mathbb{R}^m},{\mathcal{H}_1})$ to $C^{\infty}({\mathbb{R}^m},{\mathcal{H}_j})$ to provide the higher spin Laplace operators, the second order conformally invariant operators generalizing the Laplace operator to arbitrary integer spins. Their arguments also suggest that $D_x^k$ is not conformally invariant in the higher spin theory. This raises the following question: what operators generalize $k$th-powers of the Dirac operator in the higher spin theory? We know these operators exist, with even order operators taking values in homogeneous harmonic polynomial spaces and odd order operators in homogeneous monogenic polynomial spaces [@J]. This paper explicitly answers the question with the condition that the target space is a degree one homogeneous polynomial space, encompassing the spin-1 and spin-$\frac{3}{2}$ cases. More generally, one can consider bosonic and fermionic operators corresponding to either integer or half-integer spins, taking values in polynomial spaces of appropriate degree of homogeneity that are either harmonic or monogenic; however, their function theory is not fully examined here.\
The paper is organized as follows: We briefly introduce Clifford algebras, Clifford analysis, and representation theory of the Spin group in Section 2. In Section 3, we introduce the $k$-order higher spin operators ${\mathcal{D}_{1,k}}$ as the generalization of $D_x^k$ when acting on $C^{\infty}({\mathbb{R}^m}, U)$, where $U={\mathcal{H}_1}$ (spin-1) or $U={\mathcal{M}_1}$ (spin-$\frac{3}{2}$) depending on whether $k$ is even or odd. We overview classification, existence, and uniqueness results for higher spin operators. Nomenclature is given for the higher order higher spin operators that we consider: bosonic and fermionic operators. The construction and conformal invariance of the operators ${\mathcal{D}_{1,k}}$ are given with the help of the concept of *generalized symmetry*, as in [@B1; @E]. Then we provide the intertwining operators for ${\mathcal{D}_{1,k}}$, which also reveal that ${\mathcal{D}_{1,k}}$ is conformally invariant. These intertwining operators are special cases of Knapp-Stein intertwining operators ([@CO; @KS]) in higher spin theory. Section 4 presents the fundamental solutions (up to a multiplicative constant) of ${\mathcal{D}_{1,k}}$ with the help of Schur’s Lemma from representation theory. We also present an argument that the fundamental solution to ${\mathcal{D}_{1,k}}$ seen as a type of convolution operator is also conformally invariant. These convolution type operators can also be recovered as Knapp-Stein operators ([@CO; @KS]) in higher spin theory. The expressions of the fundamental solutions also suggest that ${\mathcal{D}_{1,k}}$ is a generalization of $D_x^k$. With the observation that the bases of the target spaces $U$ have simple expressions, we prove that ${\mathcal{D}_{1,k}}$ is an elliptic operator in Section 5.
Acknowledgement {#acknowledgement .unnumbered}
===============
The authors wish to thank an anonymous referee for helpful suggestions that significantly improved the manuscript. The authors are also grateful to Bent Ørsted for communications pointing out that the intertwining operators of our conformally invariant differential operators and our convolution type operators can be recovered as Knapp-Stein intertwining operators and Knapp-Stein operators in higher spin theory.
Preliminaries
=============
Clifford algebra
----------------
A real Clifford algebra, ${\mathcal{C}l_m},$ can be generated from $\mathbb{R}^m$ by considering the relationship $$\underline{x}^{2}=-\|\underline{x}\|^{2}$$ for each $\underline{x}\in \mathbb{R}^m$. We have $\mathbb{R}^m\subseteq {\mathcal{C}l_m}$. If $\{e_1,\ldots, e_m\}$ is an orthonormal basis for $\mathbb{R}^m$, then $\underline{x}^{2}=-\|\underline{x}\|^{2}$ tells us that $$e_i e_j + e_j e_i= -2\delta_{ij},$$ where $\delta_{ij}$ is the Kronecker delta function. An arbitrary element of the basis of the Clifford algebra can be written as $e_A=e_{j_1}\cdots e_{j_r},$ where $A=\{j_1, \cdots, j_r\}\subset \{1, 2, \cdots, m\}$ and $1\leq j_1< j_2 < \cdots < j_r \leq m.$ Hence for any element $a\in {\mathcal{C}l_m}$, we have $a=\sum_Aa_Ae_A,$ where $a_A\in \mathbb{R}$. Similarly, the complex Clifford algebra ${\mathcal{C}l_m}({\mathbb{C}})$ is defined as the complexification of the real Clifford algebra $${\mathcal{C}l_m}({\mathbb{C}})={\mathcal{C}l_m}\otimes{\mathbb{C}}.$$ We consider real Clifford algebra ${\mathcal{C}l_m}$ throughout this subsection, but in the rest of the paper we consider the complex Clifford algebra ${\mathcal{C}l_m}({\mathbb{C}})$ unless otherwise specified.\
The Pin and Spin groups play an important role in Clifford analysis. The Pin group can be defined as $$Pin(m)=\{a\in \mathcal{C}l_m: a=y_1y_2\dots y_{p},\ y_1,\dots,y_{p}\in\mathbb{S}^{m-1},p\in\mathbb{N}\},$$ where $\mathbb{S} ^{m-1}$ is the unit sphere in ${\mathbb{R}^m}$. $Pin(m)$ is clearly a group under multiplication in $\mathcal{C}l_m$.\
Now suppose that $a\in \mathbb{S}^{m-1}\subseteq \mathbb{R}^m$, if we consider $axa$, we may decompose $$x=x_{a\parallel}+x_{a\perp},$$ where $x_{a\parallel}$ is the projection of $x$ onto $a$ and $x_{a\perp}$ is the rest, perpendicular to $a$. Hence $x_{a\parallel}$ is a scalar multiple of $a$ and we have $$axa=ax_{a\parallel}a+ax_{a\perp}a=-x_{a\parallel}+x_{a\perp}.$$ So the action $axa$ describes a reflection of $x$ in the direction of $a$. By the Cartan-Dieudonn$\acute{e}$ Theorem each $O\in O(m)$ is the composition of a finite number of reflections. If $a=y_1\cdots y_p\in Pin(m),$ we define $\tilde{a}:=y_p\cdots y_1$ and observe that $ax\tilde{a}=O_a(x)$ for some $O_a\in O(m)$. Choosing $y_1,\ \dots,\ y_p$ arbitrarily in $\mathbb{S}^{m-1}$, we see that the group homomorphism $$\begin{aligned}
\theta:\ Pin(m)\longrightarrow O(m)\ :\ a\mapsto O_a,\end{aligned}$$ with $a=y_1\cdots y_p$ and $O_ax=ax\tilde{a}$ is surjective. Further $-ax(-\tilde{a})=ax\tilde{a}$, so $1,\ -1\in Ker(\theta)$. In fact $Ker(\theta)=\{1,\ -1\}$. See [@P1]. The Spin group is defined as $$Spin(m)=\{a\in \mathcal{C}l_m: a=y_1y_2\dots y_{2p};\ y_1, \dots,y_{2p}\in\mathbb{S}^{m-1},p\in\mathbb{N}\}$$ and it is a subgroup of $Pin(m)$. There is a group homomorphism $$\begin{aligned}
\theta:\ Spin(m)\longrightarrow SO(m)\ ,\end{aligned}$$ which is surjective with kernel $\{1,\ -1\}$. It is defined by $(1)$. Thus $Spin(m)$ is the double cover of $SO(m)$. See [@P1] for more details.\
For a domain $U$ in $\mathbb{R}^m$, a diffeomorphism $\phi: U\longrightarrow \mathbb{R}^m$ is said to be conformal if, for each $x\in U$ and each $\mathbf{v,w}\in TU_x$, the angle between $\mathbf{v}$ and $\mathbf{w}$ is preserved under the corresponding differential at $x$, $d\phi_x$. For $m\geq 3$, a theorem of Liouville tells us the only conformal transformations are Möbius transformations. Ahlfors and Vahlen show that given a Möbius transformation on $\mathbb{R}^m \cup \{\infty\}$ it can be expressed as $y=(ax+b)(cx+d)^{-1}$ where $a,\ b,\ c,\ d\in {\mathcal{C}l_m}$ and satisfy the following conditions [@Ah]: $$\begin{aligned}
&&1.\ a,\ b,\ c,\ d\ are\ all\ products\ of\ vectors\ in\ \mathbb{R}^m;\\
&&2.\ a\tilde{b},\ c\tilde{d},\ \tilde{b}c,\ \tilde{d}a\in\mathbb{R}^m;\\
&&3.\ a\tilde{d}-b\tilde{c}=\pm 1.\end{aligned}$$ Since $y=(ax+b)(cx+d)^{-1}=ac^{-1}+(b-ac^{-1}d)(cx+d)^{-1}$, a conformal transformation can be decomposed as compositions of translation, dilation, reflection and inversion. This gives an *Iwasawa decomposition* for Möbius transformations. See [@Li] for more details.
The Dirac operator in $\mathbb{R}^m$ is defined to be $$D_x:=\sum_{i=1}^{m}e_i\partial_{x_i}.$$ Note $D_x^2=-\Delta_x$, where $\Delta_x$ is the Laplacian in $\mathbb{R}^m$. A ${\mathcal{C}l_m}$-valued function $f(x)$ defined on a domain $U$ in ${\mathbb{R}^m}$ is left monogenic if $D_xf(x)=0.$ Since multiplication of Clifford numbers is not commutative in general, there is a similar definition for right monogenic functions. Sometimes we will consider the Dirac operator $D_u$ in vector $u$ rather than $x$.\
Let $\mathcal{M}_j$ denote the space of $\mathcal{C}l_m$-valued monogenic polynomials, homogeneous of degree $j$. Note that if $h_j(u)\in{\mathcal{H}_j}$, the space of $\mathcal{C}l_m$-valued harmonic polynomials homogeneous of degree $j$, then $D_u h_j(u)\in\mathcal{M}_{j-1}$, but $D_u up_{j-1}(u)=(-m-2j+2)p_{j-1}(u),$ so $$\mathcal{H}_j=\mathcal{M}_j\oplus u\mathcal{M}_{j-1},\ h_j=p_j+up_{j-1}.$$ This is an *Almansi-Fischer decomposition* of ${\mathcal{H}_j}$. See [@D] for more details. In this Almansi-Fischer decomposition, we define $P_j$ as the projection map $$\begin{aligned}
P_j: \mathcal{H}_j\longrightarrow \mathcal{M}_j.\end{aligned}$$ Suppose again $U$ is a domain in $\mathbb{R}^m$. Consider a differentiable function $f: U\times \mathbb{R}^m\longrightarrow \mathcal{C}l_m,$ such that for each $x\in U$, $f(x,u)$ is a left monogenic polynomial homogeneous of degree $j$ in $u$, then the Rarita-Schwinger operator [@B; @D] is defined by $$R_jf(x,u):=P_jD_xf(x,u)=(\frac{uD_u}{m+2j-2}+1)D_xf(x,u).$$ Though we have presented the Almansi-Fischer decomposition, the Dirac operator, and the Rarita-Schwinger operator here in terms of functions taking values in the real Clifford algebra $\mathcal{C}l_m$, they can all be realized in the same way for spinor-valued functions in the complex Clifford algebra ${\mathcal{C}l_m}({\mathbb{C}})$; we discuss spinors in the next section.
Irreducible representations of the Spin group
---------------------------------------------
The following three representation spaces of the Spin group are frequently used as the target spaces in Clifford analysis. The spinor representation is the most commonly used spin representation in classical Clifford analysis and the other two polynomial representations are often used in higher spin theory.
### Spinor representation of $Spin(m)$
Consider the complex Clifford algebra $\mathcal{C}l_m(\mathbb{C})$ with even dimension $m=2n$. Then ${\mathbb{C}}^m$ or the space of vectors is embedded in ${\mathcal{C}l_m}({\mathbb{C}})$ as $$\begin{aligned}
(x_1,x_2,\cdots,x_m)\mapsto \sum^{m}_{j=1}x_je_j:\ {\mathbb{C}}^m\hookrightarrow \mathcal{C}l_m({\mathbb{C}}).\end{aligned}$$ Define the *Witt basis* elements of ${\mathbb{C}}^{2n}$ as $$f_j:=\displaystyle\frac{e_j-ie_{j+n}}{2},\ \ f_j^{\dagger}:=-\displaystyle\frac{e_j+ie_{j+n}}{2}.$$ Let $I:=f_1f_1^{\dagger}\dots f_nf_n^{\dagger}$. The space of *Dirac spinors* is defined as $$\mathcal{S}:={\mathcal{C}l_m}({\mathbb{C}})I.$$ This is a representation of $Spin(m)$ under the following action $$\rho(s)I:=sI,\ for\ s\in Spin(m).$$ Note that ${\mathcal{S}}$ is a left ideal of ${\mathcal{C}l_m}({\mathbb{C}})$. For more details, we refer the reader to [@De]. An alternative construction of spinor spaces is given in the classical paper of Atiyah, Bott and Shapiro [@At].
### Homogeneous harmonic polynomials on $\mathcal{H}_j({\mathbb{R}^m},\mathbb{C})$
It is a well-known fact that the space of complex-valued harmonic polynomials defined on several vector variables is invariant under the action of $Spin(m)$, since the Laplacian $\Delta_m$ is an $SO(m)$ invariant operator. But it is not irreducible for $Spin(m)$. It can be decomposed into the infinite sum of $j$-homogeneous harmonic polynomials, $0\leq j<\infty$. Each of these spaces is irreducible for $Spin(m)$. This brings us the most familiar representations of $Spin(m)$: spaces of $j$-homogeneous complex-valued harmonic polynomials defined on $\mathbb{R}^m$, henceforth denoted by ${\mathcal{H}_j}:=\mathcal{H}_j({\mathbb{R}^m},\mathbb{C})$. The following action has been shown to be an irreducible representation of $Spin(m)$ [@L]: $$\begin{aligned}
\rho\ :\ Spin(m)\longrightarrow Aut({\mathcal{H}_j}),\ s\longmapsto (f(x)\mapsto f(sx\tilde{s})).\end{aligned}$$ This can also be realized as follows $$\begin{aligned}
Spin(m)\xlongrightarrow{\theta}SO(m)\xlongrightarrow{\rho} Aut({\mathcal{H}_j});\\
a\longmapsto O_a\longmapsto \big(f(x)\mapsto f(O_ax)\big),\end{aligned}$$ where $\theta$ is the double covering map and $\rho$ is the standard action of $SO(m)$ on a function $f(x)\in{\mathcal{H}_j}$ with $x\in\mathbb{R}^m$. The function $\phi(z)=(z_1+iz_m)^j$ is the highest weight vector for ${\mathcal{H}_j}({\mathbb{R}^m},{\mathbb{C}})$ having highest weight $(j,0,\cdots,0)$ (for more details, see [@G]). Accordingly, the spin representations given by $\mathcal{H}_j({\mathbb{R}^m},\mathbb{C})$ are said to have integer spin $j$; we can either specify an integer spin $j$ or the degree of homogeneity $j$ of harmonic polynomials.
### Homogeneous monogenic polynomials on $\mathcal{C}l_m$
In $\mathcal{C}l_m$-valued function theory, the previously mentioned Almansi-Fischer decomposition shows that we can also decompose the space of $j$-homogeneous harmonic polynomials as follows $${\mathcal{H}_j}={\mathcal{M}_j}\oplus u{\mathcal{M}_{j-1}}.$$ If we restrict ${\mathcal{M}_j}$ to the spinor valued subspace, we have another important representation of $Spin(m)$: the space of $j$-homogeneous spinor-valued monogenic polynomials on ${\mathbb{R}^m}$, henceforth denoted by ${\mathcal{M}_j}:={\mathcal{M}_j}({\mathbb{R}^m},\mathcal{S})$. More specifically, the following action has been shown to be an irreducible representation of $Spin(m)$: $$\begin{aligned}
\pi\ :\ Spin(m)\longrightarrow Aut({\mathcal{M}_j}),\ s\longmapsto (f(x)\mapsto sf(sx\tilde{s})).\end{aligned}$$ When $m$ is odd, in terms of complex variables $z_s=x_{2s-1}+ix_{2s}$ for all $1\leq s\leq \frac{m-1}{2}$, the highest weight vector is $\omega_j(x)=(\bar{z_1})^jI$ for ${\mathcal{M}_j}({\mathbb{R}^m},\mathcal{S})$ having highest weight $(j+\frac{1}{2},\frac{1}{2},\cdots,\frac{1}{2})$, where $\bar{z_1}$ is the conjugate of $z_1$, $\mathcal{S}$ is the Dirac spinor space, and $I$ is defined as in Section $2.2.1$; for details, see [@L]. Accordingly, the spin representations given by $\mathcal{M}_j({\mathbb{R}^m},\mathcal{S})$ are said to have half-integer spin $j+\frac{1}{2}$; we can either specify a half-integer spin $j+\frac{1}{2}$ or the degree of homogeneity $j$ of monogenic spinor-valued polynomials.
**The higher order higher spin operator $\mathcal{D}_{1,k}$**
=============================================================
Motivation
----------
We have mentioned that the Laplace operator (acting on a ${\mathbb{C}}$-valued field) is related to the Dirac operator (acting on a spinor-valued field) and they are both conformally invariant operators [@R1]. Moreover, the $k$th-power of the Dirac operator $D_x^k$ for $k$ a positive integer, is shown also to be conformally invariant in the spinor-valued function theory [@R1]. However, the Dirac operator $D_x$ and the Laplace operator are no longer conformally invariant when acting on functions taking values in the higher spin spaces, in the sense explained in the next paragraph; see [@B1; @E], and [@Ding] for the Dirac operator case. The first generalization of the Dirac operator to higher spin spaces is instead the so-called Rarita-Schwinger operator [@B; @D], and the generalization of the Laplace operator to higher spin spaces is the so-called higher spin Laplace or Maxwell operator given in [@B1; @E].
Let us look deeper into this lack of conformal invariance of the Dirac operator $D_x$ when acting on functions taking values in the higher spin spaces. Given a function $f(x,u)\in C^{\infty}({\mathbb{R}^m},{\mathcal{M}_j})$ such that $D_x f(x,u) = 0$, we apply inversion $x\longmapsto \displaystyle\frac{x}{||x||^2}$ to it. There is also a reflection of $u$ in the direction $x$ given by $\displaystyle\frac{xux}{||x||^2}$; this reflection involves $x$, which changes the conformal invariance of $D_x$ such that $D_x f(x,u) = 0$ does not hold in general. This explanation also applies for the Laplace operator $\Delta_x$ in the higher spin theory. The explanation we just mentioned further implies that the $k$th-power of the Dirac operator $D_x^k$ is not conformally invariant in the higher spin theory. In this section, we will provide the generalization of $D_x^k$ when it acts on $C^{\infty}({\mathbb{R}^m}, U)$, where $U={\mathcal{H}_1}$ or $U={\mathcal{M}_1}$ depending on the order. We provide nomenclature for these higher order operators in higher spin theory. We begin by examining existence and uniqueness of conformally invariant differential operators in higher spin spaces.
Existence of conformally invariant operators
--------------------------------------------
There is a well developed literature on the existence of conformally invariant operators [@Fegan; @Branson; @J; @Slovak-Soucek; @VS-ref]. In [@J], Slovák demonstrated the existence of conformally invariant differential operators in higher spin spaces. Then Souček considered Slovák’s results in a form more suitable for Clifford analysis. In this section, we review Souček’s results. For more details, we refer the reader to [@VS-ref].\
Let $M={\mathbb{R}^m}\cup\{\infty\}$ be the conformal compactification of ${\mathbb{R}^m}$, $\Gamma_m$ be the Clifford group, and $V(m)$ be the group of Ahlfors-Vahlen matrices. We know that all conformal transformations in ${\mathbb{R}^m},\ m>2$ can be expressed in the form $\varphi (x)=(ax+b)(cx+d)^{-1}$ with $\begin{pmatrix} a & b \\ c & d
\end{pmatrix}\in V(m)$. Let $G$ denote the identity component of the group $V(m)$. The group $G$ acts transitively on $M$ and the isotropic group of the point $0\in{\mathbb{R}^m}$ is clearly the subgroup $H$ of all matrices in G with the form $\begin{pmatrix} a & 0 \\ c & d
\end{pmatrix}$. Hence $M\cong G/H$.
For a matrix $A\in H$, the element $a\in\Gamma_m$ has a nonzero norm and can be written as the product of $\displaystyle\frac{a}{||a||}\in Spin(m)$ and $||a||\in\mathbb{R}^+$. If $\lambda$ is a dominant integral weight for $Spin(m)$ with the corresponding irreducible representation $V_{\lambda}$ and $\omega\in\mathbb{C}$ is a conformal weight, we denote $\rho_{\lambda}(\omega)$ the irreducible representation of $H$ on $V_{\lambda}$ given by $$\begin{aligned}
\rho_{\lambda}(\omega)(h)[v]=||a||^{-2\omega}\rho_{\lambda}(\frac{a}{||a||})[v];\ v\in V_{\lambda},\ h\in H;\ h=\begin{pmatrix} a & 0 \\ c & d
\end{pmatrix}.\end{aligned}$$ Below we discuss differential operators acting on sections of homogeneous vector bundles over $M=G/H$. We shall consider only bundles associated to irreducible representations of the isotropic group $H$. Hence they are specified by a highest weight $\lambda$ giving an irreducible representation of $Spin(m)$ and by a conformal weight $\omega\in\mathbb{C}$. Such a bundle will be denoted by $V_{\lambda}(\omega)$. The following lemma gives the action of $G$ on $C^{\infty}({\mathbb{R}^m},V_{\lambda}(\omega))$.
[@VS-ref]\[action\] The action of $G$ on $C^{\infty}({\mathbb{R}^m},V_{\lambda}(\omega))$ is given by $$\begin{aligned}
[g\cdot f](x)=||cx+d||^{-2\omega}\rho_{\lambda}(\frac{\widetilde{cx+d}}{||cx+d||})f((ax+b)(cx+d)^{-1}),\end{aligned}$$ where $g^{-1}=\begin{pmatrix} a & b \\ c & d
\end{pmatrix}\in G$. $``\cdot"$ stands for the action of $g$ on function $f$.
We now consider conformally invariant differential operators between sections of $\Gamma(M,V_{\lambda}(\omega))$ and $\Gamma(M,V_{\lambda'}(\omega'))$ of order $\omega'-\omega$, separately for the even and odd dimension cases.
### Even dimension $m=2n$
The highest weights of fundamental representations $\Lambda^i({\mathbb{C}}^m)$ are $$\{\lambda_i=(1,1,\cdots,1,0,\cdots,0):\ i=1,2,\cdots,n-2, \text{first}\ i\ \text{entries\ of\ $n$-tuple\ are\ $1$}\}$$ and highest weights of the basic spinor representations $\mathcal{S}^{\pm}$ of $Spin(m)$ [@G] are $n-$tuples $$\sigma^{\pm}=(\displaystyle\frac{1}{2},\displaystyle\frac{1}{2},\cdots,\pm\displaystyle\frac{1}{2}).$$ Then the $(n+1)$-tuple $(B,D_i,A,C)$ specifies the irreducible representation $\rho_{\lambda}(\omega)$ for $Spin(m)$, where $$\begin{aligned}
\label{evenlambda}
\lambda=\displaystyle\sum_{i=1}^{n-2}(D_i-1)\lambda_i+(A-1)\sigma^{+}+(C-1)\sigma^-\end{aligned}$$ and the conformal weight is given by $$\begin{aligned}
\label{evenomega}
\omega=n-\big[B+\sum_{i=1}^{n-2}D_i+\frac{A+C}{2}\big].\end{aligned}$$ Let us now state Souček’s theorem on classification of nonstandard operators, in the even dimension case.
[@VS-ref]\[VS1\] Let $(\lambda,\omega)$ and $(\lambda',\omega')$ be computed using Equations (\[evenlambda\]) and (\[evenomega\]), where the positive integers $D_i, A, C, D_i',A',C', i=1,2,\cdots,n-2,$ may adopt any values in the columns to their right in the following table, and where $(\lambda',\omega')$ are determined by primed coefficients. In the table, $a,b,c,d_i,i=1,2,\cdots,n-2$ are nonnegative integers, $d=\sum_id_i$, and the integer $e$ is defined by $e=a+b+c+d$.
-------------------- ------------- ------------------- ---------- ------------ ----------- -----------
B $-b-d$ $-b-d+d_{n-2}$ $\cdots$ $-b-d_1$ $-b$ b
$B'$ $-e$ $-e-d_{n-2}$ $\cdots$ $-e-d+d_1$ $-e-d$ $-e-d$
$D_1=D_1' $ b b $\cdots$ b $b+d_1$ $d_1$
$D_2=D_2'$ $d_1$ $d_1$ $\cdots$ $d_1+d_2$ $d_2$ $d_2$
$\vdots$ $\vdots$ $\vdots$ $\vdots$ $\vdots$ $\vdots$ $\vdots$
$D_{n-3}=D_{n-3}'$ $d_{n-4}$ $d_{n-4}$ $\cdots$ $d_{n-3}$ $d_{n-3}$ $d_{n-3}$
$D_{n-2}=D_{n-2}'$ $d_{n-3}$ $d_{n-3}+d_{n-2}$ $\cdots$ $d_{n-2}$ $d_{n-2}$ $d_{n-2}$
$A=C'$ $a+d_{n-2}$ a $\cdots$ a a a
$C=A'$ $c+d_{n-2}$ c $\cdots$ c c c
-------------------- ------------- ------------------- ---------- ------------ ----------- -----------
Then there exists (up to a multiple) unique nontrivial conformally invariant differential operators between sections of $\Gamma(M,V_{\lambda}(\omega))$ and $\Gamma(M,V_{\lambda'}(\omega'))$; its order is equal to $\omega'-\omega$. This is a complete list of the so-called nonstandard conformally invariant differential operators on spaces of even dimension.
### Odd dimension $m=2n+1$
The highest weights of fundamental representations $\Lambda^i({\mathbb{C}}^m)$ are $$\{\lambda_i=(1,1,\cdots,1,0,\cdots,0):\ i=1,2,\cdots,n-1, \text{where\ the\ first}\ i\ \text{entries\ of\ $n$-tuple\ are\ $1$}\}$$ and highest weights of the basic spinor representation $\mathcal{S}$ of $Spin(m)$ [@G] are $n-$tuples $$\sigma=(\displaystyle\frac{1}{2},\displaystyle\frac{1}{2},\cdots,\displaystyle\frac{1}{2}).$$ Then the $(n+1)$-tuple $(B,D_i,A,C)$ specifies the irreducible representation $\rho_{\lambda}(\omega)$ for $Spin(m)$, where $$\begin{aligned}
\label{oddlambda}
\lambda=\displaystyle\sum_{i=1}^{n-1}(D_i-1)\lambda_i+(A-1)\sigma\end{aligned}$$ and the conformal weight is given by $$\begin{aligned}
\label{oddomega}
\omega=\frac{2n+1}{2}-\big[B+\sum_{i=1}^{n-1}D_i+\frac{A}{2}\big].\end{aligned}$$ Let us now state Souček’s theorem on classification of nonstandard operators, now in the odd dimension case.
[@VS-ref]\[VS2\] Let $(\lambda,\omega)$ and $(\lambda',\omega')$ be computed using Equations (\[oddlambda\]) and (\[oddomega\]), where the positive integers $D_i, A, C, D_i',A',C', i=1,2,\cdots,n-1,$ may adopt any values in the columns to their right in the following table, and where $(\lambda',\omega')$ are determined by primed coefficients. In the table, $a,b,c,d_i,i=1,2,\cdots,n-1$ are nonnegative half-integers or integers (at least one of them being half integral), $d=\sum_id_i$, and the integer $e$ is defined by $e=a+b+d$.
-------------------- ------------- ------------------- ---------- ------------ ----------- -----------
B $-b-d$ $-b-d+d_{n-1}$ $\cdots$ $-b-d_1$ $-b$ b
$B'$ $-e$ $-e-d_{n-1}$ $\cdots$ $-e-d+d_1$ $-e-d$ $-e-d$
$D_1=D_1' $ b b $\cdots$ b $b+d_1$ $d_1$
$D_2=D_2'$ $d_1$ $d_1$ $\cdots$ $d_1+d_2$ $d_2$ $d_2$
$\vdots$ $\vdots$ $\vdots$ $\vdots$ $\vdots$ $\vdots$ $\vdots$
$D_{n-2}=D_{n-2}'$ $d_{n-3}$ $d_{n-3}$ $\cdots$ $d_{n-2}$ $d_{n-2}$ $d_{n-2}$
$D_{n-1}=D_{n-1}'$ $d_{n-2}$ $d_{n-2}+d_{n-1}$ $\cdots$ $d_{n-1}$ $d_{n-1}$ $d_{n-1}$
$A=A'$ $a+d_{n-2}$ a $\cdots$ a a a
-------------------- ------------- ------------------- ---------- ------------ ----------- -----------
Then there exists (up to a multiple) unique nontrivial conformally invariant differential operators between sections of $\Gamma(M,V_{\lambda}(\omega))$ and $\Gamma(M,V_{\lambda'}(\omega'))$; its order is equal to $\omega'-\omega$. This is a complete list of the so-called nonstandard conformally invariant differential operators on spaces of odd dimension.
### Applications to our cases
Theorem \[VS1\] and \[VS2\] show existence of conformally invariant differential operators as follows.
[@VS-ref]\[VS\] Let $(\lambda,\omega)$ and $(\lambda',\omega')$ be one of couples for which there is a (nontrivial) invariant differential operator $$D:\ \Gamma(M,V_{\lambda}(\omega))\longrightarrow \Gamma(M,V_{\lambda'}(\omega'))$$ (the nonstandard operators are listed in Theorem \[VS1\] and \[VS2\]; the complete list is in [@J]).
Let $T_{\lambda,\omega}(g),\ g^{-1}=\begin{pmatrix} a & b \\ c & d
\end{pmatrix}\in G$ (similarly for $T_{\lambda',\omega'}(g)$) be the operator acting on smooth maps from ${\mathbb{R}^m}$ to $V_{\lambda}$ by $$\begin{aligned}
\big[T_{\lambda,\omega}(g)f\big](x)=||cx+d||^{-2\omega}\rho_{\lambda}(\frac{\widetilde{cx+d}}{||cx+d||})\big[f((ax+b)(cx+d)^{-1})\big].\end{aligned}$$ Then $$\begin{aligned}
D(T_{\lambda,\omega}(g)f)=T_{\lambda',\omega'}(g)(Df),\ g\in G,\ f\in C^{\infty}({\mathbb{R}^m},V_{\lambda}).\end{aligned}$$
This paper considers differential operators acting on functions $f(x,u)\in C^{\infty}({\mathbb{R}^m},{\mathcal{H}_j})$ or $f(x,u)\in C^{\infty}({\mathbb{R}^m},{\mathcal{M}_j})$. Here we only show existence of conformally invariant differential operators on spaces of *even dimension $m$*; the odd dimensional case is similar. We work out the allowable highest weights, conformal weights, and orders on operators acting on these function spaces.
From Theorem \[VS1\], we notice that highest weight $\lambda$ is determined by $D_i,\ A$ and $C$. From the table, we also have $\lambda=\lambda'$. In other words, conformally invariant operators only exist between $C^{\infty}({\mathbb{R}^m},{\mathcal{H}_j})$ and itself or $C^{\infty}({\mathbb{R}^m},{\mathcal{M}_j})$ and itself. We consider each in turn.\
**Integer Spin Case: $C^{\infty}({\mathbb{R}^m},{\mathcal{H}_j})\longrightarrow C^{\infty}({\mathbb{R}^m},{\mathcal{H}_j})$**\
As an irreducible representation of $Spin(m)$, ${\mathcal{H}_j}$ has highest weight of $n$-tuple $\lambda=\lambda'=(j,0,\cdots,0).$ The group action $\rho_{\lambda}$ is defined in Section $2.2.2$. From the table in Theorem \[VS1\] and $$\begin{aligned}
&&\lambda=\displaystyle\sum_{i=1}^{n-2}(D_i-1)\lambda_i+(A-1)\sigma^{+}+(C-1)\sigma^-,\\
&&\lambda_i=(1,1,\cdots,1,0,\cdots,0),\ i=1,2,\cdots,n-2,\\
&&\sigma^{\pm}=(\displaystyle\frac{1}{2},\displaystyle\frac{1}{2},\cdots,\pm\displaystyle\frac{1}{2}),\end{aligned}$$ we know that $D_1=j+1,\ D_i=1,\ i=2,\cdots,n-2$ and $A=C=1$. There is exactly one possibility for all entries but the last column in the table, for which there is a sequence of possibilities indexed by a nonnegative integer $b$. The last column corresponds to the $(2b+2n+2j-2)$-th order conformally invariant differential operator, with $$d=j+n-2,\ a=c=1,\ e=b+j+n,\ B=b,\ B'=-b-2j-2n+2,$$ and conformal weights $\omega=1-b-j$ and $\omega'=b+j+2n-1$. Hence, we have $$\begin{aligned}
&&\mathcal{D}_{1,2b+2n+2j-2} T_{\lambda,1-b-j}=T_{\lambda,b+j+2n-1}\mathcal{D}_{1,2b+2n+2j-2},\\
&&i.e.,\ \mathcal{D}_{1,2b+2n+2j-2}||cx+d||^{2j+2b-2}=||cx+d||^{-2b-2j-4n+2}\mathcal{D}_{1,2b+2n+2j-2}.\end{aligned}$$ To make the above intertwining operators coincide with the forms of the intertwining operators we have at the end of Section 3, we let $2b+2n+2j-2=2s$ and since $m=2n$, we have $$\mathcal{D}_{1,2s}||cx+d||^{2s-m}=||cx+d||^{-m-2s}\mathcal{D}_{1,2s}.$$\
**Half-integer Spin Case: $C^{\infty}({\mathbb{R}^m},{\mathcal{M}_j})\longrightarrow C^{\infty}({\mathbb{R}^m},{\mathcal{M}_j})$**\
As an irreducible representation of $Spin(m)$, ${\mathcal{M}_j}$ has highest weight as $n$-tuple $\lambda=\lambda'=(j+\displaystyle\frac{1}{2},\displaystyle\frac{1}{2},\cdots,\pm\displaystyle\frac{1}{2}).$ The group action $\rho_{\lambda}$ is the action defined as in Section $2.2.3$. From the table and $$\begin{aligned}
&&\lambda=\displaystyle\sum_{i=1}^{n-2}(D_i-1)\lambda_i+(A-1)\sigma^{+}+(C-1)\sigma^-,\\
&&\lambda_i=(1,1,\cdots,1,0,\cdots,0),\ i=1,2,\cdots,n-2,\\
&&\sigma^{\pm}=(\displaystyle\frac{1}{2},\displaystyle\frac{1}{2},\cdots,\pm\displaystyle\frac{1}{2}),\end{aligned}$$ we can find that $D_1=j+1, D_i=1, i=2,\cdots,n-2$ and $A=1,C=0$( or $A=0, C=1$ depending on the last entry of $\lambda$ is $\displaystyle\frac{1}{2}$ or $-\displaystyle\frac{1}{2}$). Similar as the previous case, there is just one possibility for all but the last column in the table and there is a sequence of possibilities indexed by a nonnegative integer $b$ for the last column. The last column corresponds to the $(2j+2n+2b-3)$-th order conformally invariant differential operator, with $$d=j+n-2,\ a=1,c=0\ (or\ a=0,\ c=1),\ e=b+j+n-1,\ B=b,\ B'=-2j-2n-b+3,$$ and conformal weights $\omega=-b-j+\displaystyle\frac{3}{2}$ and $\omega'=j+2n+b-\displaystyle\frac{3}{2}$. Hence, we have $$\begin{aligned}
\mathcal{D}_{1,2j+2n+2b-3} T_{\lambda,-b-j+\frac{3}{2}}=T_{\lambda,j+2n+b-\frac{3}{2}}\mathcal{D}_{1,2j+2n+2b-3},\end{aligned}$$ in other words, $$\begin{aligned}
\mathcal{D}_{1,2j+2n+2b-3}||cx+d||^{2b+2j-3}\frac{\widetilde{cx+d}}{||cx+d||}=||cx+d||^{-2j-4n-2b+3}\frac{\widetilde{cx+d}}{||cx+d||}\mathcal{D}_{1,2j+2n+2b-3}.\end{aligned}$$ To make the above intertwining operators coincide with the intertwining operators we have at the end of Section 3, we let $2j+2n+2b-3=2s+1$ and since $m=2n$, we have $$\mathcal{D}_{1,2s+1}\displaystyle\frac{\widetilde{cx+d}}{||cx+d||^{m-2s}}=\displaystyle\frac{\widetilde{cx+d}}{||cx+d||^{m+2s+2}}\mathcal{D}_{1,2s+1}.$$
Similar arguments apply for the odd dimensional cases. This establishes existence of the conformally invariant differential operators we wish to consider. Further, even order conformally invariant differential operators only exist between $C^{\infty}({\mathbb{R}^m},{\mathcal{H}_j})$ and odd order ones only exist between $C^{\infty}({\mathbb{R}^m},{\mathcal{M}_j})$. Intertwining operators of conformally invariant differential operators in Theorem \[interwining op of D1k\] can also be recovered. Once we establish conformal invariance of the operators that we construct between the desired higher spin spaces, uniqueness up to multiplicative constant of these higher order higher spin operators is established by the preceding theorems.
Construction and conformal invariance
-------------------------------------
We have established by arguments of Slovák [@J] and Souček [@VS-ref], for integers $j\ge0$ and $k>0$ there exist conformally invariant differential operators in the higher spin setting $$\mathcal{D}_{j,k}:\ C^{\infty}({\mathbb{R}^m}, U)\longrightarrow C^{\infty}({\mathbb{R}^m}, U),$$ where $U=\mathcal{H}_j$ if $k$ is even and $U=\mathcal{M}_j$ if $k$ is odd. We introduce some nomenclature suggestive of massless spin fields in mathematical physics, which we hope is adopted by others studying higher spin theory in Clifford analysis. As a Spin representation $\mathcal{H}_j$ is associated with integer spin $j$ and particles of integer spin are called bosons, so the operators $\mathcal{D}_{j,k}:\ C^{\infty}({\mathbb{R}^m}, \mathcal{H}_j)\longrightarrow C^{\infty}({\mathbb{R}^m}, \mathcal{H}_j)$ are named *bosonic operators*. Thus in the spin 0 case we have the Laplace operator and its $k$-powers, the spin 1 case the Maxwell operator and its generalization to order $k=2n$, and general higher spin Laplace operators and their generalization to order $k=2n$. Correspondingly, as a Spin representation $\mathcal{M}_j$ is associated with half-integer spin $j+\frac{1}{2}$ and particles of half-integer spin are called fermions, so the operators $\mathcal{D}_{j,k}:\ C^{\infty}({\mathbb{R}^m}, \mathcal{M}_j)\longrightarrow C^{\infty}({\mathbb{R}^m}, \mathcal{M}_j)$ are named *fermionic operators*. Thus in the spin $\frac{1}{2}$ case we have the Dirac operator and its $k=2n+1$ powers, the spin $\frac{3}{2}$ case the simplest Rarita-Schwinger operator and its generalization to order $k=2n+1$, and general Rarita-Schwinger operators and their generalization to order $k=2n+1$. Note that our notation indexes according to degree of homogeneity of the target space $j$ and differential order $k$, so fractions are not used in the notation; if we indexed according to spin, fractional spins would need to be used for odd order operators.
We will consider the higher order spin 1 and spin $\frac{3}{2}$ operators ${\mathcal{D}_{1,k}}: C^{\infty}({\mathbb{R}^m}, U)\longrightarrow C^{\infty}({\mathbb{R}^m}, U),$ where $U={\mathcal{H}_1}$ for $k$ even and $U=\mathcal{M}_1$ for $k$ odd. Note that the target space $U$ here is a function space. That means any element in $C^{\infty}({\mathbb{R}^m},U)$ is of the form $f(x,u)$ with $f(x,u)\in U$ for each fixed $x\in{\mathbb{R}^m}$ and $x$ is the variable which ${\mathcal{D}_{1,k}}$ acts on. The construction and conformal invariance of these two operators are considered as follows.
### $k$ even, $k=2n$, $n>1$ (The bosonic case) {#k-even-k2n-n1-the-bosonic-case .unnumbered}
For positive integer $n$, the unique $2n$-th order conformally invariant differential operator of spin-$1$ $\mathcal{D}_{1,2n}:C^{\infty}(\mathbb{R}^m,\mathcal{H}_1)\longrightarrow C^{\infty}(\mathbb{R}^m,\mathcal{H}_1)$ has the following form, up to a multiplicative constant: $$\begin{aligned}
\mathcal{D}_{1,2n}=\Delta_x^n-\frac{4n}{m+2n-2}\langle u,D_x\rangle \langle D_u,D_x\rangle\Delta_x^{n-1}.\end{aligned}$$
For the case $n=1$, we retrieve the Maxwell operator from [@E].\
Our proof of conformal invariance of this operator follows closely the method of [@E]. In order to explain what conformal invariance means, we begin with the concept of a generalized symmetry (see for instance [@Eastwood]):
An operator $\eta_1$ is a generalized symmetry for a differential operator $\mathcal{D}$ if and only if there exists another operator $\eta_2$ such that $\mathcal{D}\eta_1=\eta_2\mathcal{D}$. Note that for $\eta_1=\eta_2$, this reduces to a definition of a (proper) symmetry: $\mathcal{D}\eta_1=\eta_1\mathcal{D}$.
One determines the first order generalized symmetries of an operator, which span a Lie algebra [@E; @Miller]. In this case, the first order symmetries will span a Lie algebra isomorphic to the conformal Lie algebra $\mathfrak{so}(1,m+1)$; in this sense, the operators we consider are conformally invariant. The operator $\mathcal{D}_{1,2n}$ is $\mathfrak{so}(m)$-invariant (rotation-invariant) because it is the composition of $\mathfrak{so}(m)$-invariant (rotation-invariant) operators, which means the angular momentum operators $L_{ij}^x+L_{i,j}^u$ that generate these rotations are proper symmetries of $\mathcal{D}_{1,2n}$. The infinitesimal translations $\partial_{x_j}, j=1,\cdots , n,$ corresponding to linear momentum operators are proper symmetries of $\mathcal{D}_{1,2n}$; this is an alternative way to say that $\mathcal{D}_{1,2n}$ is invariant under translations that are generated by these infinitesimal translations. Readers familiar with quantum mechanics will recognize the connection to isotropy and homogeneity of space, the rotational and translational invariance of Hamiltonian, and the conservation of angular and linear momentum [@Sakurai]; see also [@Br] concerning Rarita-Schwinger operators.
The remaining two of the first order generalized symmetries of $\mathcal{D}_{1,2n}$ are the Euler operator and special conformal transformations. The Euler operator $\mathbb{E}_x$ that measures degree of homogeneity in $x$ is a generalized symmetry because $\mathcal{D}_{1,2n}\mathbb{E}_x=(\mathbb{E}_x+2n)\mathcal{D}_{1,2n}$; this is an alternative way to say that $\mathcal{D}_{1,2n}$ is invariant under dilations, which are generated by the Euler operator. The special conformal transformations are defined in Lemma \[SCT\] in terms of harmonic inversion for ${\mathcal{H}_1}$-valued functions; harmonic inversion is defined in Definition \[HI\] and is an involution mapping solutions of $\mathcal{D}_{1,2n}$ to $\mathcal{D}_{1,2n}$. Readers familiar with conformal field theory will recognize that invariance under dilation corresponds to scale-invariance and that special conformal transformations are another class of conformal transformations arising on spacetime [@CFT]. An alternative method of proving conformal invariance of $\mathcal{D}_{1,2n}$ is to prove the invariance of $\mathcal{D}_{1,2n}$ under those finite transformations generated by these first order generalized symmetries (rotations, dilations, translations, and special conformal transformations) to show invariance of $\mathcal{D}_{1,2n}$ under actions of the conformal group; this may be phrased in terms of Möbius transformations and the Iwasawa decomposition. However, the first-order generalized symmetry method emphasizes the connection to mathematical physics and is more amenable to our proof of a certain property of harmonic inversion. It is also that used by earlier authors [@B1; @E].
\[HI\] The harmonic inversion is a conformal transformation defined as $$\begin{aligned}
\mathcal{J}_{2n}:C^{\infty}(\mathbb{R}^m,\mathcal{H}_1)\longrightarrow C^{\infty}(\mathbb{R}^m,\mathcal{H}_1):f(x,u)\mapsto \mathcal{J}_{2n}[f](x,u):=||x||^{2n-m}f(\frac{x}{||x||^2},\frac{xux}{||x||^2}).\end{aligned}$$
Note that this inversion consists of Kelvin inversion $\mathcal{J}$ on $\mathbb{R}^m$ in the variable $x$ composed with a reflection $u\mapsto \omega u\omega$ acting on the dummy variable $u$ (where $x=||x||\omega$) and a multiplication by a conformal weight term $||x||^{2n-m}$; it satisfies $\mathcal{J}_{2n}^2=1.$\
Then we have the special conformal transformation defined in the following lemma. The definition is an infinitesimal version of the fact that finite special conformal transformations consist of a translation preceded and followed by an inversion [@CFT]: an infinitesimal translation preceded and followed by harmonic inversion. The second equality in the lemma shares some terms in common with the generators of special conformal transformations in conformal field theory [@CFT], and is a particular case of a result in [@ER].
\[SCT\] The special conformal transformation defined as $\mathcal{C}_{2n}:=\mathcal{J}_{2n}\partial_{x_j}\mathcal{J}_{2n}$ satisfies $$\begin{aligned}
\mathcal{C}_{2n}=2\langle u,x\rangle\partial_{u_j}-2u_j\langle x,D_u\rangle +||x||^2\partial_{x_j}-x_j(2\mathbb{E}_x+m-2n).\end{aligned}$$
A similar calculation as in *Proposition A.1* in [@B1] will show the conclusion.
Then, we have the main proposition as follows.
\[propeven\] The special conformal transformations $\mathcal{C}_{2n}$, with $j\in\{1,2,\dots,m\}$ are generalized symmetries of $\mathcal{D}_{1,2n}$. More specifically, $$\begin{aligned}
[\mathcal{D}_{1,2n},\mathcal{C}_{2n}]=-4nx_j\mathcal{D}_{1,2n}.\end{aligned}$$ In particular, this shows that $$\begin{aligned}
\label{crucialeven}
\mathcal{J}_{2n}\mathcal{D}_{1,2n}\mathcal{J}_{2n}=||x||^{4n}\mathcal{D}_{1,2n},\end{aligned}$$ which is the generalization of the case of the classical higher order Laplace operator $\Delta_x^n$ [@Ax]. This also implies $\mathcal{D}_{1,2n}$ is invariant under inversion.
If the main proposition holds, then the conformal invariance can be summarized in the following theorem:
\[theoremeven\] The first order generalized symmetries of $\mathcal{D}_{1,2n}$ are given by:
1. The infinitesimal rotation $L_{i,j}^x+L_{i,j}^u$, with $1\leq i<j\leq m$.
2. The shifted Euler operator $(\mathbb{E}_x+\displaystyle\frac{m-2n}{2})$.
3. The infinitesimal translations $\partial_{x_j}$, with $1\leq j\leq m$.
4. The special conformal transformations $\mathcal{J}_{2n}\partial_{x_j}\mathcal{J}_{2n}$, with $1\leq j\leq m$.
These operators span a Lie algebra which is isomorphic to the conformal Lie algebra $\mathfrak{so}(1,m+1)$, whereby the Lie bracket is the ordinary commutator.
The proof is similar as in [@ER] via transvector algebras. Notice that the shift in the shifted Euler operator $\mathbb{E}_x+\omega$ defines the conformal weight (defined in Section $3.2$) $\omega=\displaystyle\frac{m-2n}{2}$.
### Detailed proof of Proposition \[propeven\]: {#detailed-proof-of-proposition-propeven .unnumbered}
First, let us prove a few technical lemmas. It is worth pointing out that since we are dealing with degree-$1$ homogeneous polynomials in $u$, terms involving second derivatives with respect to $u$ disappear.
\[lemma1\] For all $1\leq j\leq m$, we have $$\begin{aligned}
[\Delta_x^n,\mathcal{C}_{2n}]=-4nx_j\Delta_x^n+4n\langle u,D_x\rangle\partial_{u_j}\Delta_x^{n-1}-4nu_j\langle D_u,D_x\rangle\Delta_x^{n-1}.\end{aligned}$$
We prove this by induction. First, we have ([@B1]) $$\begin{aligned}
[\Delta_x,\mathcal{C}_{2}]=-4x_j\Delta_x+4\langle u,D_x\rangle\partial_{u_j}-4u_j\langle D_u,D_x\rangle.\end{aligned}$$ Assuming the lemma is true for $\Delta^{n-1}$, applying the fact that for general operators $A,\ B$ and $C$ $$\begin{aligned}
[AB,C]&=&A[B,C]+[A,C]B\end{aligned}$$ and $$\begin{aligned}
\mathcal{C}_{2n}&=&\mathcal{C}_{2}+(2n-2)x_j,\end{aligned}$$ we have $$\begin{aligned}
&&[\Delta_x^n,\mathcal{C}_{2n}]
=\Delta_x^{n-1}[\Delta_x,\mathcal{C}_{2n}]+[\Delta_x^{n-1},\mathcal{C}_{2n}]\Delta_x.\end{aligned}$$ Since $$\mathcal{C}_{2n}=\mathcal{C}_{2n-2}+2x_j,$$ a straightforward calculation leads to the conclusion.
\[lemma2\] For all $1\leq j\leq m,$ we have $$\begin{aligned}
&&[\langle u,D_x\rangle\langle D_u,D_x\rangle \Delta_x^{n-1},\mathcal{C}_{2n}] \\
&=&-4nx_j\langle u,D_x\rangle\langle D_u,D_x\rangle \Delta_x^{n-1}+(m+2n-2)\big(\langle u,D_x\rangle\partial_{u_j}-u_j\langle D_u,D_x\rangle\big)\Delta_x^{n-1}.\end{aligned}$$
First, we have [@B1]: $$\begin{aligned}
&&[\langle u,D_x\rangle\langle D_u,D_x\rangle, \mathcal{C}_{2}]\\
&=&2||u||^2\partial_{u_j}\langle D_u,D_x\rangle-4x_j\langle u,D_x\rangle\langle D_u,D_x\rangle+\big(\langle u,D_x\rangle\partial_{u_j}-u_j\langle D_u,D_x\rangle\big)(2\mathbb{E}_u+m-2)\\
&=&-4x_j\langle u,D_x\rangle\langle D_u,D_x\rangle+m\big(\langle u,D_x\rangle\partial_{u_j}-u_j\langle D_u,D_x\rangle\big),\end{aligned}$$ Then $$\begin{aligned}
&&[\langle u,D_x\rangle\langle D_u,D_x\rangle \Delta_x^{n-1},\mathcal{C}_{2n}] \\
&=&\langle u,D_x\rangle\langle D_u,D_x\rangle[\Delta_x^{n-1},\mathcal{C}_{2n}]+[\langle u,D_x\rangle\langle D_u,D_x\rangle,\mathcal{C}_{2n}]\Delta_x^{n-1}\end{aligned}$$ together with the previous lemma proves the conclusion.
With the help of *Lemma* \[lemma1\] and \[lemma2\], a straightforward calculation shows that $$[\mathcal{D}_{1,2n},\mathcal{C}_{2n}]=-4nx_j\mathcal{D}_{1,2n}.$$ Since $\mathcal{D}_{1,2n}$ is conformally invariant by Theorem \[theoremeven\] and Slovak’s results provide the uniqueness and intertwining operators of conformally invariant differential operators, we have $$\mathcal{J}_{2n}\mathcal{D}_{1,2n}\mathcal{J}_{2n}=||x||^{4n}\mathcal{D}_{1,2n}$$ from the intertwining operators under (harmonic) inversion. This is a generalization of the Laplacian case [@Ax].
### $k$ odd, $k=2n-1$, $n>1$ (The fermionic case) {#k-odd-k2n-1-n1-the-fermionic-case .unnumbered}
For positive integer $n$, the unique $(2n-1)$-th order conformally invariant differential operator of spin-$\frac{3}{2}$ $\mathcal{D}_{1,2n-1}:C^{\infty}(\mathbb{R}^m,\mathcal{M}_1)\longrightarrow C^{\infty}(\mathbb{R}^m,\mathcal{M}_1)$ has the following form, up to a multiplicative constant: $$\begin{aligned}
\mathcal{D}_{1,2n-1}=D_x\Delta_x^{n-1}-\frac{2}{m+2n-2}u\langle D_u,D_x\rangle \Delta_x^{n-1}-\frac{4n-4}{m+2n-2}\langle u,D_x\rangle \langle D_u,D_x\rangle \Delta_x^{n-2}D_x.\end{aligned}$$
When $n=1$, we have the Rarita-Schwinger operator appearing in [@B; @D] and elsewhere.\
The same strategy in the even case applies: we only must show the special conformal transformation defined below is a generalized symmetry of $\mathcal{D}_{1,2n-1}$. We have the definition for monogenic inversion as follows.
Monogenic inversion is a conformal transformation defined as $$\begin{aligned}
&&\mathcal{J}_{2n+1}\ :\ C^{\infty}(\mathbb{R}^m,\mathcal{M}_1)\longrightarrow C^{\infty}(\mathbb{R}^m,\mathcal{M}_1);\\
&&f(x,u)\mapsto \mathcal{J}_{2n+1}[f](x,u):=\frac{x}{||x||^{m-2n}}f(\frac{x}{||x||^2},\frac{xux}{||x||^2}).\end{aligned}$$
Note that this inversion also consists of Kelvin inversion $\mathcal{J}$ on $\mathbb{R}^m$ in the variable $x$ composed with a reflection $u\mapsto \omega u\omega$ acting on the dummy variable $u$ (where $x=||x||\omega$) and a multiplication of a conformal weight term $\displaystyle\frac{x}{||x||^{m-2n}}$; it satisfies $\mathcal{J}_{2n+1}^2=-1$ instead. Similarly, monogenic inversion is an involution mapping solutions for ${\mathcal{D}_{1,2n-1}}$ to solutions for ${\mathcal{D}_{1,2n-1}}$ ([@Ro]). Then we have the following lemma.
The special conformal transformation is defined as $$\begin{aligned}
\mathcal{C}_{2n-1}:=\mathcal{J}_{2n-1}\partial_{x_j}\mathcal{J}_{2n-1}=-e_jx-2\langle u,x\rangle\partial_{u_j}+2u_j\langle x,D_u\rangle -||x||^2\partial_{x_j}+x_j(2\mathbb{E}_x+m-2n),\end{aligned}$$ $\mathcal{C}_{2n-1}=\mathcal{C}_{2n-3}-2x_j=-\mathcal{C}_{2n-2}-e_jx-2x_j$.
As similar calculation as in *Proposition A.1* in [@B1] will show the conclusion.
Then we arrive at the main proposition, stating that the special conformal transformations are generalized symmetries of operator $\mathcal{D}_{1,2n-1}$.
\[propodd\] The special conformal transformations $\mathcal{C}_{2n-1}$, with $j\in\{1,2,\dots,m\}$ are generalized symmetries of $\mathcal{D}_{1,2n-1}$. More specifically, $$\begin{aligned}
[\mathcal{D}_{1,2n-1},\mathcal{C}_{2n-1}]=(4n-2)x_j\mathcal{D}_{1,2n-1}.\end{aligned}$$ In particular, this shows that $\mathcal{J}_{2n-1}\mathcal{D}_{1,2n-1}\mathcal{J}_{2n-1}=||x||^{4n-2}\mathcal{D}_{1,2n-1}$, which is the generalization of the case of the classical higher order Dirac operator $D_x^{2n-1}$ [@Ax]. This also implies $\mathcal{D}_{1,2n-1}$ is invariant under inversion.
\[theoremodd\] The first order generalized symmetries of $\mathcal{D}_{1,2n-1}$ are given by:
1. The infinitesimal rotation $L_{i,j}^x+L_{i,j}^u$, with $1\leq i<j\leq m$.
2. The shifted Euler operator $(\mathbb{E}_x+\displaystyle\frac{m-2n+1}{2})$.
3. The infinitesimal translations $\partial_{x_j}$, with $1\leq j\leq m$.
4. The special conformal transformations $\mathcal{J}_{2n-1}\partial_{x_j}\mathcal{J}_{2n-1}$, with $1\leq j\leq m$.
These operators span a Lie algebra which is isomorphic to the conformal Lie algebra $\mathfrak{so}(1,m+1)$, whereby the Lie bracket is the ordinary commutator.
### Detailed proof of Proposition \[propodd\]: {#detailed-proof-of-proposition-propodd .unnumbered}
To prove Proposition \[propodd\], as in the even case, we need a few technical lemmas.
\[lemma 5\] For all $1 \leq j \leq m$, we have $$\begin{aligned}
&&[D_x\Delta_x^{n-1},\mathcal{C}_{2n-1}] \\
&=&(4n-2)x_jD_x\Delta_x^{n-1}+(4n-4)\big(u_j\langle D_u,D_x\rangle-\langle u, D_x\rangle \partial_{u_j}\big)D_x\Delta_x^{n-2}-2u\partial_{u_j}\Delta_x^{n-1}.\end{aligned}$$
\[lemma 6\] For all $1 \leq j \leq m$, we have $$\begin{aligned}
&&[u\langle D_u,D_x\rangle \Delta_x^{n-1}, \mathcal{C}_{2n-1}] \\
&=&(4n-2)x_ju\langle D_u,D_x\rangle \Delta_x^{n-1}-(m+2n-2)u\partial_{u_j}\Delta_x^{n-1}-(2n-2)ue_j\langle D_u,D_x \rangle \Delta_x^{n-2}.\end{aligned}$$
\[lemma 7\] For all $1 \leq j \leq m$, we have $$\begin{aligned}
&&[\langle u, D_x\rangle \langle D_u,D_x\rangle \Delta_x^{n-2}D_x,\mathcal{C}_{2n-1}]
=(4n-2)x_j\langle u, D_x\rangle \langle D_u,D_x\rangle \Delta_x^{n-2}D_x\\
&&-(m+2n-2)\big( \langle u,D_x\rangle \partial_{u_j}-u_j\langle D_u,D_x\rangle \big)\Delta_x^{n-2}D_x+ue_j\langle D_u,D_x\rangle \Delta_x^{n-2}D_x.\end{aligned}$$
We combine *Lemma* \[lemma 5\], \[lemma 6\] and \[lemma 7\] to get $$\begin{aligned}
[\mathcal{D}_{1,2n-1},\mathcal{C}_{2n-1}]=(4n-2)x_j\mathcal{D}_{1,2n-1}.\end{aligned}$$ This implies $\mathcal{J}_{2n-1}\mathcal{D}_{1,2n-1}\mathcal{J}_{2n-1}=||x||^{4n-2}\mathcal{D}_{1,2n-1}$.
### Conformal Invariance and Intertwining Operators, Both Cases {#conformal-invariance-and-intertwining-operators-both-cases .unnumbered}
Strictly speaking, Theorem \[VS\] together with the constructions in this subsection can provide the intertwining operators for the bosonic and fermionic operators in this paper. However, for the sake of concreteness and to highlight the alternative approach centering upon Möbius transformations, here we rely on the Iwasawa decomposition for Möbius transformations to determine these intertwining operators. Let $\mathcal{D}_{1,k,x,u}$ and $\mathcal{D}_{1,k,y,w}$ be the higher order higher spin operators with respect to $x,\ u$ and $y,\ w$, respectively and $y=\phi(x)=(ax+b)(cx+d)^{-1}$ is a Möbius transformation. Let $$\begin{aligned}
J_k=\frac{\widetilde{cx+d}}{||cx+d||^{m-2n}},\ \ for\ k=2n+1;\\
J_k=\frac{1}{||cx+d||^{m-2n}},\ \ for\ k=2n;\\
J_{-k}=\frac{\widetilde{cx+d}}{||cx+d||^{m+2n+2}},\ \ for\ k=2n+1;\\
J_{-k}=\frac{1}{||cx+d||^{m+2n}},\ \ for\ k=2n,\\\end{aligned}$$ with $n=1,2,3,\cdots.$ See [@P]. Then we make the following claim.
\[interwining op of D1k\] Let $y=\phi (x)=(ax+b)(cx+d)^{-1}$ be a Möbius transformation. Then $$\begin{aligned}
J_{-k}\mathcal{D}_{1,k,y,w} f(y,w)=\mathcal{D}_{1,k,x,u}J_k f(\phi(x),\frac{(cx+d)u(\widetilde{cx+d})}{||cx+d||^2}),\end{aligned}$$ where $w=\displaystyle\frac{(cx+d)u(\widetilde{cx+d})}{||cx+d||^2}$.
We only prove the bosonic (order $k=2n$) case, as the fermionic (order $k=2n+1$) case is similar. According to the Iwasawa decomposition, we need only prove this with respect to orthogonal transformation and inversion, since translation and dilation are trivial. Note that our argument here requires the invariance under harmonic inversion established earlier.
### **Orthogonal transformations** $a\in Pin(m)$ {#orthogonal-transformations-ain-pinm .unnumbered}
If $x=ay\tilde{a},$ $u=aw\tilde{a}$, then $\mathcal{D}_{1,2n,x,u}f(x,u)=a\mathcal{D}_{1,2n,y,w}\tilde{a}f(y,w).$
$$\begin{aligned}
&&\mathcal{D}_{1,2n,x,u}f(x,u)=\bigg(\triangle_x-\frac{4n}{m+2n-2}\langle u,D_x\rangle \langle D_u,D_x\rangle \bigg)\Delta_x^{n-1}f(x,u) \\
&=&\bigg(a\triangle_y\tilde{a}-\frac{4n}{m+2n-2}a\langle w,D_y\rangle\tilde{a}a \langle D_w,D_y\rangle\tilde{a} \bigg)a\Delta_y^{n-1}\tilde{a}f(y,w) \\
&=&a\bigg(\triangle_y-\frac{4n}{m+2n-2}\langle w,D_y\rangle \langle D_w,D_y\rangle \bigg)\Delta_y^{n-1}\tilde{a}f(y,w) \\
&=&a\mathcal{D}_{1,2n,y,w}\tilde{a}f(y,w).\end{aligned}$$
### Inversions {#inversions .unnumbered}
\[inversion\] Let $x=y^{-1}$ and $u=\displaystyle\frac{ywy}{||y||^2}$, then $$\begin{aligned}
\mathcal{D}_{1,2n,y,w}||x||^{m-2n}f(y,w)=||x||^{m+2n}\mathcal{D}_{1,2n,x,u}f(x,u).\end{aligned}$$
Recall that after we showed $[\mathcal{D}_{1,2n},\mathcal{J}_{2n}\partial_{x_j}\mathcal{J}_{2n}]=-4nx_j\mathcal{D}_{1,2n}$ for $\mathcal{J}_{2n}$ the harmonic inversion, we claimed and later showed that $\mathcal{J}_{2n}\mathcal{D}_{1,2n}\mathcal{J}_{2n}=||x||^{4n}\mathcal{D}_{1,2n}$. This can also be written as $$\begin{aligned}
\mathcal{D}_{1,2n,y,w}||x||^{m-2n}f(y,w)=||x||^{m+2n}\mathcal{D}_{1,2n,x,u}f(x,u).\end{aligned}$$
Theorem \[interwining op of D1k\] now follows using the Iwasawa decomposition. See [@Ding] for the first order case.
**Fundamental solutions of $\mathcal{D}_{1,k}$**
================================================
To get the fundamental solutions of $\mathcal{D}_{1,k}$, we use techniques from [@B]. It is worth pointing out that the reproducing kernels of ${\mathcal{M}_1}$ and ${\mathcal{H}_1}$ below have simple expression, but we insist on using techniques used in [@B], since it also works for more general cases when we have $\mathcal{M}_j$ or $\mathcal{H}_j$ instead. This will be found in an upcoming paper. This method only provides us the fundamental solutions up to a multiplicative constant. We prefer this constant to be determined in the more general case in an upcoming paper.
### $k$ even, $k=2n$ (The bosonic case) {#k-evenk2n-the-bosonic-case .unnumbered}
Recall that the reproducing kernel for $j$-homogeneous harmonic spherical polynomials $Z_{j}(u,v)$ is called the *zonal spherical harmonic* of degree $j$, and is invariant under reflections (and consequently rotations) in the variables $u$ and $v$ [@Ax]. In our circumstance, $$Z_1(u,v)=\displaystyle\frac{(m-2)^2\omega_{m-1}}{m}\langle u,v\rangle$$ is the zonal spherical harmonic of degree $1$, where $\omega_{m-1}$ is the surface area of the $(m-1)$-dimensional unit sphere and $\langle u,v\rangle$ is the standard inner product in Euclidean space. It can be considered as the identity of $End(\mathcal{H}_1)$ and satisfies $$\begin{aligned}
P_1(v)=(Z_1(u,v),P_1(u))_u:=\int_{S^{m-1}} \overline{Z_1(u,v)}P_1(u)dS(u),\end{aligned}$$ where $(\ ,\ )_u$ denotes the Fischer inner product with respect to $u$; we define the Fischer inner product of two functions by the integral of their product over the sphere, consistent with other work in higher spin theory [@B; @D]. A homogeneous $End(\mathcal{H}_1)$-valued $C^{\infty}$-function $x\rightarrow E(x)$ on $\mathbb{R}^m\backslash \{0\}$ satisfying $\mathcal{D}_{1,2n}E(x)=\delta(x)Z_1(u,v)$ is referred to as a fundamental solution for the operator $\mathcal{D}_{1,2n}$. We will show that such a fundamental solution has the form $E_{1,2n}(x,u,v)=c_1||x||^{2n-m}Z_1(\displaystyle\frac{xux}{||x||^2},v)$. Since $Z_1(u,v)$ is a trivial solution of ${\mathcal{D}_{1,2n}}$, according to the invariance of $\mathcal{D}_{1,2n}$ under inversion, we obtain a non-trivial solution $\mathcal{D}_{1,2n}E_{1,2n}(x,u,v)=0$ in $\mathbb{R}^m\backslash \{0\}$. Clearly the function $E_{1,2n}(x,u,v)$ is homogeneous of degree $2n-m$ in $x$, so $\mathcal{D}_{1,2n}E_{1,2n}(x,u,v)$ is homogeneous of degree $-m$ in $x$ and it belongs to $L_1^{loc}(\mathbb{R}^m)$. Because $\delta(x)$ is the only (up to a multiple) distribution homogeneous of degree $-m$ with support at the origin, we have in the sense of distributions: $$\begin{aligned}
\mathcal{D}_{1,2n}E_{1,2n}(x,u,v)=\delta(x)P_1(u,v)\end{aligned}$$ for some $P_1(u,v)\in \mathcal{H}_1\otimes \mathcal{H}_1^*$. Then we have $$\begin{aligned}
&&\int_{\mathbb{S}^{m-1}}\mathcal{D}_{1,2n}\overline{E_{1,2n}(x,u,v)}Q_1(v)dS(v)\\
&=&\delta(x)\int_{\mathbb{S}^{m-1}}\overline{P_1(u,v)}Q_1(v)dS(v).\end{aligned}$$ Now, for all $Q_1\in\mathcal{H}_1$, we have $$\begin{aligned}
&&\int_{\mathbb{S}^{m-1}}\mathcal{D}_{1,2n}\overline{E_{1,2n}(x,u,v)}Q_1(v)dS(v)\\
&=&\mathcal{D}_{1,2n}\int_{\mathbb{S}^{m-1}}c_1||x||^{2n-m}\overline{Z_1(\frac{xux}{||x||^2},v)}Q_1(v)dS(v)\\
&=&\mathcal{D}_{1,2n}\int_{\mathbb{S}^{m-1}}c_1||x||^{2n-m}\overline{Z_1(\frac{xux}{||x||^2},\frac{xv'x}{||x||^2})}Q_1(\frac{xv'x}{||x||^2})dS(v'),\end{aligned}$$ where in the last line we made a change of variables in the second argument of $Z_1$. Since $Z_1(u,v)$ is invariant under reflection and $\displaystyle\frac{xux}{||x||^2}$ is a reflection of variable $u$ in the direction of $x$, the last line in the last equation becomes $$\begin{aligned}
&&\mathcal{D}_{1,2n}\int_{\mathbb{S}^{m-1}}c_1\overline{Z_1(u,v')}||x||^{2n-m}Q_1(\frac{xv'x}{||x||^2})dS(v')\\
&=&c_1\mathcal{D}_{1,2n}||x||^{2n-m}Q_1(\frac{xux}{||x||^2}).\end{aligned}$$ Hence, we obtain $$\begin{aligned}
\delta(x)\int_{\mathbb{S}^{m-1}}\overline{P_1(u,v)}Q_1(v)dS(v)
=c_1\mathcal{D}_{1,2n}||x||^{2n-m}Q_1(\frac{xux}{||x||^2}).\end{aligned}$$
As the reproducing kernel $Z_1(u,v)$ is invariant under the $Spin(m)$-representation $H:\ f(u,v)\mapsto sf(su\tilde{s},sv\tilde{s})\tilde{s}$, the kernel $E_{1,2n}(x,u,v)$ is also $Spin(m)$-invariant: $$\begin{aligned}
sE_{1,2n}(sx\tilde{s},su\tilde{s},sv\tilde{s})\tilde{s}=E_{1,2n}(x,u,v).\end{aligned}$$
From this it follows that $P_1(u,v)$ must be also invariant under $H$. Let now $\phi$ be a test function with $\phi(0)=1$. Let $L$ be the action of $Spin(m)$ given by $L: f(u)\mapsto sf(\tilde{s}us)\tilde{s}.$ Then $$\begin{aligned}
&&\langle \mathcal{D}_{1,2n}\big(c_1||x||^{2n-m}L(\frac{x}{||x||})L(s)Q_1(u)\big),\phi(x)\rangle\\
&=&\int_{\mathbb{S}^{m-1}}\overline{P_1(u,v)}L(s)Q_1(v)dS(v)\\
&=&L(s)\int_{\mathbb{S}^{m-1}}\overline{P_1(u,v)}Q_1(v)dS(v)\\
&=&\langle L(s)\big(\mathcal{D}_{1,2n}c_1||x||^{2n-m}L(\frac{x}{||x||})Q_1(u)\big),\phi(x)\rangle.\end{aligned}$$ In this way we have constructed an element of $End(\mathcal{H}_1)$ commuting with the $L$-representation of $Spin(m)$ that is irreducible; see Section 2.2.2. By Schur’s Lemma ([@F]), it follows that $P_1(u,v)$ must be the reproducing kernel $Z_1(u,v)$ if we choose $c_1$ properly. Hence $$\begin{aligned}
\mathcal{D}_{1,2n}E_{1,2n}(x,u,v)=\delta(x)Z_1(u,v).\end{aligned}$$
### $k$ odd, $k=2n-1$ (The fermionic case) {#k-oddk2n-1-the-fermionic-case .unnumbered}
The reproducing kernel $Z_{k}(u,v)$ for degree $k$ homogeneous monogenic spherical polynomials, those in $\mathcal{M}_k$, is called the *zonal spherical monogenic* [@B2]. (There should be no confusion using the same notation for zonal spherical harmonics and monogenics.) In our circumstance, for $u,v\in\mathbb{S}^{m-1}$, $$Z_1(u,v)=\displaystyle\frac{1}{\omega_{m-1}}\bigg(\displaystyle\frac{2\mu+1}{2\mu}C^{\mu}_1(t)+(u\wedge v)C_0^{\mu+1}(t)\bigg),$$ where $\mu=\displaystyle\frac{m}{2}-1$, $t=\langle u,v\rangle$, $u\wedge v=uv+\langle u,v\rangle$, and $C^{\mu}_k(t)$ are the Gegenbauer polynomials [@B2]. With similar arguments and the fact that $Z_1(u,v)$ is also $Spin(m)$-invariant under the same $Spin(m)$-action as in the even case, one can show that $$E_{1,2n-1}(x,u,v)=c'_1\frac{x}{||x||^{m-2n+2}}Z_1(\frac{xux}{||x||^2},v)$$ is the fundamental solution of $\mathcal{D}_{1,2n-1}$, where $c'_1$ is a non-zero real constant.\
Since $E_{1,k}(x,u,v)$ is the fundamental solution of ${\mathcal{D}_{1,k}}$, we have $$\begin{aligned}
\int_{{\mathbb{R}^m}}\int_{\mathbb{S}^{m-1}}E_{1,k}(x-y,u,v){\mathcal{D}_{1,k}}\psi(x,u)dS(u)dx^m=\psi(y,v),\end{aligned}$$ where $\psi(x,u) \in C^{\infty}({\mathbb{R}^m}, U)$ with compact support in $x$ for each $u\in \mathbb{R}^{m}$, $U={\mathcal{M}_1}$ when $k$ is odd and $U={\mathcal{H}_1}$ when $k$ is even. Hence, we have ${\mathcal{D}_{1,k}}E_{1,k}=Id$ and $E_{1,k}={\mathcal{D}_{1,k}}^{-1}$ in the distribution sense. Now, $$\begin{aligned}
J_{-k}\mathcal{D}_{1,k,y,w} \psi(y,w)=\mathcal{D}_{1,k,x,u}J_k \psi(\phi(x),\frac{(cx+d)u(\widetilde{cx+d})}{||cx+d||^2}),\end{aligned}$$ where $y=\phi (x)=(ax+b)(cx+d)^{-1}$ is a Möbius transformation and $w=\displaystyle\frac{(cx+d)u(\widetilde{cx+d})}{||cx+d||^2}$ as in Theorem 3, we get $$J_k^{-1}\mathcal{D}_{1,k,x,u}^{-1}J_{-k}=\mathcal{D}_{1,k,y,w}^{-1}.$$ Alternatively, $$J_k^{-1}E_{1,k,x,u}J_{-k}=E_{1,k,y,w}.$$ This gives us the intertwiners of the fundamental solution $E_{1,k}$ under Möbious transformations, which also reveals that the fundamental solutions are conformally invariant under Möbius transformations.
**Ellipticity of the operator $\mathcal{D}_{1,k}$**
===================================================
Notice that the bases of the target space ${\mathcal{H}_1}$ and $\mathcal{M}_1$ have simple expressions. We can use techniques similar to those in [@B1; @E] to show that the operators $\mathcal{D}_{1,k}$ are elliptic. First, we introduce the definition for an elliptic operator.
A linear homogeneous differential operator of $k$-th order $\mathcal{D}_{1,k}: C^{\infty}(\mathbb{R}^m, V_1)\longrightarrow C^{\infty}(\mathbb{R}^m, V_2)$ is elliptic if for every non-zero vector $x\in \mathbb{R}^m$ its principal symbol, the linear map $\sigma_x(\mathcal{D}_{1,k}):V_1\longrightarrow V_2$ obtained by replacing its partial derivatives $\partial_{x_j}$ with the corresponding variables $x_j$, is a linear isomorphism.
Then we prove ellipticity of ${\mathcal{D}_{1,k}}$ in the even and odd cases individually.
### $k$ even, $k=2n$ (The bosonic case) {#k-even-k2n-the-bosonic-case .unnumbered}
\[even elliptic\] The operator $\mathcal{D}_{1,2n}:=\bigg(\triangle_x-\frac{4n}{m+2n-2}\langle u,D_x\rangle \langle D_u,D_x\rangle \bigg)\Delta_x^{n-1}$ is an elliptic operator.
In [@E] it was shown that the operator $\triangle_x-\frac{4}{m}\langle u,D_x\rangle \langle D_u,D_x\rangle $ is elliptic. In our case, the term in the parentheses is the same as the previous one up to a constant coefficient, so a similar argument shows $\triangle_x-\frac{4n}{m+2n-2}\langle u,D_x\rangle \langle D_u,D_x\rangle $ is elliptic. Since the symbol of $\Delta_x^{n-1}$ is non-negative, $\bigg(\triangle_x-\frac{4n}{m+2n-2}\langle u,D_x\rangle \langle D_u,D_x\rangle \bigg)\Delta_x^{n-1}$ is elliptic.
### $k$ odd, $k=2n-1$ (The fermionic case) {#k-odd-k2n-1-the-fermionic-case .unnumbered}
\[odd elliptic\] The operator $$\begin{aligned}
\mathcal{D}_{1,2n-1}:=D_x\Delta_x^{n-1}-\frac{2}{m+2n-2}u\langle D_u,D_x\rangle \Delta_x^{n-1}-\frac{4n-4}{m+2n-2}\langle u,D_x\rangle \langle D_u,D_x\rangle \Delta_x^{n-2}D_x\end{aligned}$$ is an elliptic operator.
To prove the theorem, we show that, for fixed $x\in \mathbb{R}^m,$ the symbol of the operator $\mathcal{D}_{1,2n-1}$, which is given by $$\begin{aligned}
x||x||^{2n-2}-\frac{2}{m+2n-2}u\langle D_u,x\rangle ||x||^{2n-2}-\frac{4n-4}{m+2n-2}\langle u,x\rangle \langle D_u,x\rangle ||x||^{2n-4}x,\end{aligned}$$ is a linear isomorphism from ${\mathcal{M}_1}$ to ${\mathcal{M}_1}$. As the symbol is clearly a linear map, it remains to be proven that the map is injective. Recall that ${\mathcal{M}_1}$ is actually ${\mathcal{M}_1}({\mathbb{R}^m},\mathcal{S})$, however, if we can prove the symbol is injective for ${\mathcal{M}_1}(\mathcal{C}l_m)$, then this also implies that it is injective for ${\mathcal{M}_1}({\mathcal{S}})\subset{\mathcal{M}_1}(\mathcal{C}l_m({\mathbb{C}}))$. From the Almansi-Fischer decomposition $\mathcal{H}_1={\mathcal{M}_1}\oplus u\mathcal{M}_0$, it is easy to obtain that $dim {\mathcal{M}_1}=m-1$. Since $\{e_ju_m+e_mu_j\}_{j=1}^{m-1}$ are in ${\mathcal{M}_1}$ and it is also a linearly independent set in ${\mathcal{M}_1}$. Therefore, it is actually a basis of ${\mathcal{M}_1}$. Hence, an arbitrary element of $\mathcal{M}_1$ can be written as $\sum_{j=1}^m\alpha_j(e_ju_m+e_mu_j)$ with $\alpha_j\in \mathbb{C}$ for all $1 \leq j\leq m$. We next show that the following system of equations has a unique solution: $$\begin{aligned}
\big(x||x||^{2}-\frac{2u\langle D_u,x\rangle ||x||^{2}}{m+2n-2}-\frac{4n-4}{m+2n-2}x\langle u,x\rangle \langle D_u,x\rangle\big)\big(\sum_{j=1}^m\alpha_j(e_ju_m+e_mu_j)\big)=0.\end{aligned}$$ With $c_1=\frac{2}{m+2n-2}$, $c_2=\frac{4n-4}{m+2n-2}$, $a_i=(c_1e_i||x||^2+c_2xx_i)$, $b_j=x_me_j+x_je_m$, and $1 \leq i,j\leq m-1$, this equation system can be written in matrix notation as follows: $$\begin{bmatrix}
-x||x||^2e_m-a_1b_1 & -a_1b_2 & \ldots & -a_1b_{m-1} \\
-a_2b_1 & -x||x||^2e_m-a_2b_2 & \ldots & -a_2b_{m-1} \\
\vdots & \vdots & \ddots & \vdots \\
-a_{m-1}b_1 & -a_{m-1}b_2 & \ldots & -x||x||^2e_m-a_{m-1}b_{m-1}
\end{bmatrix}
\begin{bmatrix}
\alpha_1 \\
\alpha_2 \\
\vdots \\
\alpha_{m-1}
\end{bmatrix}
=0.$$ In order to show that this system has an unique solution, it suffices to prove that $$\begin{vmatrix}
-x||x||^2e_m-a_1b_1 & -a_1b_2 & \ldots & -a_1b_{m-1} \\
-a_2b_1 & -x||x||^2e_m-a_2b_2 & \ldots & -a_2b_{m-1} \\
\vdots & \vdots & \ddots & \vdots \\
-a_{m-1}b_1 & -a_{m-1}b_2 & \ldots & -x||x||^2e_m-a_{m-1}b_{m-1}
\end{vmatrix}
\neq 0.$$ Using the notation $\vec{a}=(a_1,a_2,\dots,a_{m-1})^{T}$ and $\vec{b}=(b_1,b_2,\dots,b_{m-1})^{T}$, the determinant can be written more compactly as $$\begin{aligned}
P(x)=det\big(-x||x||^2e_m\mathbb{I}_{m-1}-\vec{a}\cdot\vec{b}^T\big)\neq 0.\end{aligned}$$ As a function, $$\begin{aligned}
P(x)&=&det\big(-x||x||^2e_m\mathbb{I}_{m-1}-\vec{a}\cdot\vec{b}^T\big)=det\big(-x||x||^2e_m\mathbb{I}_{m-1}-\vec{a}\cdot\vec{b}^T\big)^T \\
&=&-x||x||^2e_m-\vec{a}^T\cdot\vec{b}
=-x||x||^2e_m-\sum_{j=1}^{m-1}(c_1e_j||x||^2+c_2xx_j)(x_me_j+x_je_m) \\
&=&-x||x||^2e_m+(c_1+c_2)||x||^2x_m-(c_1+c_2)x||x||^2e_m\end{aligned}$$ Checking each $e_j$-th component with $1\leq j\leq m$, it is easy to see $P(x)$ is non-zero if $x$ is non-zero. This completes the proof.
[99]{} L. V. Ahlfors, *Möbius transformations in $\mathbb{R}^n$ expressed through 2$\times$2 matrices of Clifford numbers*, Complex Variables, 5, 1986, 215-224.
M.F. Atiyah, R. Bott, A. Shapiro, *Clifford modules*, Topology, Vol. 3, Suppl. 1, 1964, pp. 3-38.
S. Axler, P. Bourdon and W. Ramey, *Harmonic function theory*, second edition, Graduate Texts in Mathematics, Springer, New York, 2001.
F. Brackx, R. Delanghe, F. Sommen, *Clifford Analysis*, Pitman, London, 1982.
F. Brackx, D. Eelbode, L. Van de Voorde, *Higher spin Dirac operators between spaces of simplicial monogenics in two vector variables*, Mathematical Physics, Analysis and Geometry, 2011, Vol. 14, Issue 1, pp. 1-20.
T. Branson, *Second order conformal covariants*, Proc. Am. Math. Soc., 1998, 126, pp. 1031Ð1042.
J. Bureš, F. Sommen, V. Souček, P. Van Lancker, *Rarita-Schwinger Type Operators in Clifford Analysis*, J. Funct. Anal. 185 (2001), No. 2, pp. 425-455.
J.L. Clerc, B. Ørsted, *Conformal covariance for the powers of the Dirac operator*, https://arxiv.org/abs/1409.4983
H. De Bie, D. Eelbode, M. Roels, *The higher spin Laplace operator*, arXiv:1501.03974 \[math-ph\]
H. De Bie, F. Sommen, M. Wutzig, *Reproducing kernels for polynomial null-solutions of Dirac operators*, http://arxiv.org/abs/1503.03969
R. Delanghe, F. Sommen, V. Souček, *Clifford Algebra and Spinor-Valued Functions: A Function Theory for the Dirac Operator*, Kluwer, Dordrecht, 1992.
H. De Schepper, D. Eelbode, T. Raeymaekers, *On a special type of solutions of arbitrary higher spin Dirac operators*, J. Phys. A: Math. Theor., 43, 2010, 325208-325221.
P. Di Francesco, P. Mathieu, D. Sénéchal, *Conformal Field Theory*, Graduate Texts in Contemporary Physics, New York: Springer-Verlag, 1997.
C. Ding, J. Ryan, *On Some Conformally Invariant Operators in Euclidean Space*, arXiv:1509.00098 \[math.CV\], submitted.
C. F. Dunkl, J. Li, J. Ryan, P. Van Lancker, *Some Rarita-Schwinger type operators*, Computational Methods and Function Theory, 2013, Vol. 13, Issue 3, pp. 397-424.
M. Eastwood, *Higher symmetries of the Laplacian*, Ann. Math., 161 No. 3, 2005, pp. 1645-1665.
D. Eelbode, M. Roels, *Generalised Maxwell equations in higher dimensions*, Complex Analysis and Operator Theory, Dec. 2014, pp. 1-27. dx.doi.org/10.1007/s11785-014-0436-5
D. Eelbode, T. Raeymaekers, *Construction of conformally invariant higher spin operators using transvector algebras*, Journal of Mathematical Physics, Vol. 55, Issue 10, 2014, DOI:http://dx.doi.org/10.1063/1.4898772
H.D. Fegan, *Conformally invariant first order differential operators*, Q. J. Math., 1976, Vol. 27, 513Ð538.
T. Fulton, F. Rohrlich, L. Witten, *Conformal invariance in physics*, Rev. Mod. Phys., 1962, Vol. 34, Issue 3, pp. 442Ð456.
W. Fulton, J. Harris, *Representation theory. A first course*, Graduate Texts in Mathematics, Readings in Mathematics 129, New York: Springer-Verlag, 1991.
J. Gilbert, M. Murray, *Clifford Algebras and Dirac Operators in Harmonic Analysis*, Cambridge University Press, Cambridge, 1991.
A.W. Knapp, E.M. Stein, *Intertwining operators for semisimple groups*, Annals of Mathematics, Vol. 93, No.3, 1971, pp. 489-578
B. Lawson, M.L. Michelson, *Spin Geometry*, Princeton University Press, Princeton, New Jersey, 1989.
J. Li, J. Ryan, *Some operators associated to Rarita-Schwinger type operators*, Complex Variables and Elliptic Equations: An International Journal, Volume 57, Issue 7-8, 2012, pp. 885-902.
W. Miller, *Symmetry and Separation of Variables*, Addison-Wesley Publishing Company, Providence, Rhode Island, 1977.
J. Peetre, T. Qian, *Möbius covariance of iterated Dirac operators*, J. Austral. Math. Soc. Series A 56 (1994), pp. 403-414.
I. Porteous, *Clifford algebra and the classical groups*, Cambridge University Press, Cambridge, 1995.
W. Rarita, J. Schwinger, *On a Theory of Particles with Half-integral Spin*, Phys. Rev., Vol. 60, Issue 1, 1941, pp. 60-61.
M. Roels, *A Clifford analysis approach to higher spin fields*, Master’s Thesis, University of Antwerp, 2013.
J. Ryan, *Conformally covariant operators in Clifford analysis*, Z. Anal. Anwendungen, 14, 1995, pp. 677-704.
J. Ryan, *Iterated Dirac operators and conformal transformations in ${\mathbb{R}^m}$*, Proceedings of the XV International Conference on Differential Geometric Methods in Theoretical Physics, World Scientific, 1987, pp. 390-399.
J.J. Sakurai, J. Napolitano, *Modern Quantum Mechanics*, Second Edition, San Francisco: Addison-Wesley, 2011.
J. Slovák, *Natural Operators on Conformal Manifolds*, Habilitation thesis, Masaryk University, Brno, Czech Republic, 1993.
J. Slovák, V. Souček, *Invariant Operators of the First Order On Manifolds with a Given Parabolic Structure*, Séminaires & Congrès, 2000, Vol. 4, p. 251-276.
V. Souček, *Higher spins and conformal invariance in Clifford analysis*, Lecture in Seiffen, Proc. Conf. Seiffen 1996.
P. Van Lancker, F. Sommen, D. Constales, *Models for irreducible representations of Spin(m)*, Advances in Applied Clifford Algebras, Vol. 11, Issue 1 supplement, 2001, pp. 271-289.
G. Velo, D. Zwanzinger, *Propagation and Quantization of Rarita-Schwinger Waves in an External Electromagnetic Potential*, Phys. Rev., Vol. 186, Issue 5, 1969, pp. 1337-1341.
G. Velo, D. Zwanzinger, *Noncausality and Other Defects of Interaction Lagrangians for Particles with Spin One and Higher*, Phys. Rev., Vol. 188, Issue 5, 1969, pp. 2218-2222.
[^1]: These authors contributed equally to this manuscript.
[^2]: Electronic address: [[email protected]]{}.
[^3]: Electronic address: [[email protected]]{}; R.W. acknowledges this material is based upon work supported by the National Science Foundation Graduate Research Fellowship Program under Grant No. DGE-0957325 and the University of Arkansas Graduate School Distinguished Doctoral Fellowship in Mathematics and Physics.
[^4]: Electronic address: [[email protected].]{}
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Let $N$ be an irreducible, compact 3-manifold with empty or toroidal boundary which is not a closed graph manifold. We show that it follows from the work of Agol, Kahn-Markovic and Przytycki-Wise that $\pi_1(N)$ admits a cofinal filtration with ‘fast’ growth of Betti numbers as well as a cofinal filtration of $\pi_1(N)$ with ‘slow’ growth of ranks.'
address: |
Mathematisches Institut\
Universität zu Köln\
Germany
author:
- Stefan Friedl
title: 'A note on the growth of Betti numbers and ranks of $3$-manifold groups'
---
Introduction
============
A *filtration of a group $\pi$* is a sequence $\{\pi_i\}_{i\in \N}$ of finite index subgroups of $\pi$ such that $\pi_{i+1}\subset \pi_i$ for every $i$. We say that a filtration is *cofinal* if $\cap_{i\in \N}\pi_i$ is trivial, we call it *normal* if $\pi_i \lhd \pi$ for every $i$, and we say it is *almost normal* if there exists a $k$ such that $\pi_i\lhd \pi_k$ for every $i\geq k$. A group which admits a cofinal normal filtration is called *residually finite*. Given a filtration $\{\pi_i\}_{i\in \N}$ of a group $\pi$ it is of interest to study how the following measures of ‘complexity’ grow:
the first Betti number $b_1(\pi_i)=\dim H_1(\pi_i;\Q)$,
the $\F_p$-Betti numbers $b_1(\pi_i,\F_p)=\dim H_1(\pi_i;\F_p)$,
the rank $d(\pi_i)$, i.e. the minimal size of a generating set,
the order of $\tor H_1(\pi_i;\Z)$. Such growth functions have been studied for 3-manifold groups by many authors over the years. We refer to [@CE10; @CW03; @De10; @EL12; @Gi10; @GS91; @KMT03; @La09; @La11; @Le10; @Lu94; @LL95; @KS12; @Ra10; @Ri90; @ShW92; @SiW02a; @SiW02b; @Wa09] for a sample of results in this direction. It is clear that given any group $\pi$ we have $d(\pi)\geq b_1(\pi)$, i.e. given a filtration the ranks grow at least as fast as the Betti numbers.
Now let $N$ be a 3-manifold. Throughout this paper we will use the following convention: a 3-manifold will always be assumed to be connected, compact, orientable and irreducible with empty or toroidal boundary. By [@He87] the group $\pi_1(N)$ is residually finite. In this paper we are interested in how fast Betti numbers can grow in a cofinal filtration of $\pi_1(N)$ and how slowly the ranks can grow in a cofinal filtration of $\pi_1(N)$.
First note that given any cofinal normal filtration $\{\pi_i\}_{i\in \N}$ of $\pi=\pi_1(N)$ it follows from the work of Lück [@Lu94 Theorem 0.1] and Lott and Lück [@LL95 Theorem 0.1] that \[equ:ll\] \_[i]{} b\_1(\_i)=0,i.e. the first Betti number grows sublinearly. The same equality also holds for almost normal cofinal filtrations of $\pi_1(N)$ if we apply the aforementioned results to an appropriate finite cover of $N$.
Note that (\[equ:ll\]) does not necessarily hold for cofinal filtrations of $\pi_1(N)$ which are not almost normal. In fact Girão [@Gi10] (see proof of [@Gi10 Theorem 3.1]) gives an example of a cusped hyperbolic 3-manifold together with a cofinal filtration of $\{\pi_i\}_{i\in \N}$ of $\pi=\pi_1(N)$ such that $$\lim_{i\to \infty} \frac{1}{[\pi:\pi_i]} b_1(\pi_i)>0.$$
It is an interesting question how quickly $\frac{1}{[\pi:\pi_i]} b_1(\pi_i)$ converges to zero, and to what degree the convergence depends on the choice of normal cofinal filtration of $\pi=\pi_1(N)$. This question for example was recently studied by Kionke and Schwermer [@KS12].
We will use recent work of Agol [@Ag12] (which in turn builds on work of Kahn-Markovic [@KM12] and Wise [@Wi12]) to prove the following theorem which says that ‘most’ 3-manifolds admit cofinal filtrations with ‘fast’ sublinear growth of first Betti numbers.
\[mainthm1\] Let $N\ne S^1\times D^2$ and $N\ne T^2\times I$ be a 3-manifold which is neither spherical nor covered by a torus bundle. Then the following hold:
Given any function $f\colon \N\to \R_{\geq 0}$ such that $$\lim_{n\to \infty} \frac{f(n)}{n}=0$$ there exists an almost normal cofinal filtration $\{\pi_i\}_{i\in \N}$ of $\pi$ such that $$b_1(\pi_i) \geq f([\pi:\pi_i]) \mbox{ for every }i\in \N.$$
There exists a normal cofinal filtration $\{\pi_i\}_{i\in \N}$ of $\pi=\pi_1(N)$ and an $\eps\in (0,1)$ such that $$b_1(\pi_i)\geq [\pi:\pi_i]^\eps \mbox{ for every }i\in \N.$$
We now turn to the construction of cofinal filtrations with ‘slow’ growth of ranks. First note that if $H$ is a finite index subgroup of a finitely generated group $G$, then it follows from the Reidemeister-Schreier method (see e.g. [@MKS76 Corollary 2.7.1]) that $$d(H)\leq [G:H]\cdot (d(G)-1)+1\leq [G:H]\cdot d(G).$$ In particular if $\{\pi_i\}_{i\in \N}$ is a cofinal filtration of a group $\pi$, then $$\frac{1}{[\pi:\pi_i]} d(\pi_i)\leq d(\pi) \mbox{ for every }i.$$ Put differently, the rank grows at most linearly with the degree.
We will again use the recent work of Agol, Kahn-Markovic and Wise together with work of Przytycki-Wise [@PW12] to prove the following theorem which says that ‘most’ 3-manifolds admit cofinal filtrations with ‘slow’ growth of ranks.
\[mainthm2\] Let $N$ be a 3-manifold which is not a closed graph manifold.
Given any function $f\colon \N\to \R_{\geq 0}$ with $$\lim_{n\to \infty} f(n)=\infty$$ there exists an almost normal cofinal filtration $\{\pi_i\}_{i\in \N}$ of $\pi$ such that $$d(\pi_i)\leq f([\pi:\pi_i]) \mbox{ for every }i\in \N.$$
There exists a normal cofinal filtration $\{\pi_i\}_{i\in \N}$ of $\pi_1(N)$ and an $\eps\in (0,1)$ such that $$d(\pi_i)\leq [\pi:\pi_i]^{\eps}\mbox{ for every }i\in \N.$$
Acknowledgment {#acknowledgment .unnumbered}
--------------
We wish to thank Jack Button and Wolfgang Lück for helpful conversations. We are also grateful to the anonymous referees for carefully reading an earlier version of the paper and for giving helpful feedback.
Proofs
======
3-manifold groups
-----------------
The world of 3-manifold topology was shaken up considerably by the recent breakthroughs due to Agol, Kahn-Markovic, Przytycki-Wise and Wise. In particular the following is a consequence of these recent results:
\[thm:akmw\] Let $N$ be a 3-manifold.
Suppose that $N\ne S^1\times D^2$ and $N\ne T^2\times I$ and suppose that $N$ is neither spherical nor covered by a torus bundle. Then $\pi_1(N)$ is large, i.e. $\pi_1(N)$ contains a finite index subgroup which admits an epimorphism onto a non-cyclic free group.
Suppose that $N$ is not a closed graph manifold. Then $N$ is virtually fibered, i.e. $N$ admits a finite index cover which fibers over $S^1$.
The first statement is a consequence of the ‘Virtually Compact Special Theorem’ of Agol [@Ag12] (building on work of Kahn-Markovic [@KM12] and Wise [@Wi12]) and older work of Kojima [@Ko87] and Luecke [@Lue88]. The second statement is also a consequence of the ‘Virtually Compact Special Theorem’ together with further work of Agol [@Ag08] and Przytycki-Wise [@PW12]. The fact that graph manifolds with boundary are fibered follows from earlier work of Wang–Yu [@WY97] (see also [@Li11; @PW11]). We refer to the survey paper [@AFW12] for details and how this theorem follows precisely from the aforementioned papers.
Growth of the first Betti number of large groups {#section:largeb1growth}
------------------------------------------------
In this section we will several times make use of the basic fact that if $\varphi\colon G\to H$ is a group homomorphism with finite cokernel, then a transfer argument shows that $H_1(G;\Q)\to H_1(H;\Q)$ is surjective, and therefore $b_1(G)\geq b_1(H)$. We start out with the following lemma.
\[lem:fastb1growth\] Let $\G$ be a residually finite group which admits an epimorphism $\a\colon \G \to F$ onto a non-cyclic free group. Let $g\colon \N\to \R_{\geq 0}$ be a function such that $$\lim_{n\to \infty} \frac{g(n)}{n}=0.$$ Then there exists a normal cofinal filtration $\{\G_i\}_{i\in \N}$ of $\G$ such that $$b_1(\G_i)\geq g([\G:\G_i]) \mbox{ for every }i\in \N.$$
Let $\G$ be a residually finite group which admits an epimorphism $\a\colon \G \to F$ onto a non-cyclic free group. Let $g\colon \N\to \R_{\geq 0}$ be a function such that $\lim_{n\to \infty} \frac{g(n)}{n}=0$. After possibly replacing $g$ by $$n\mapsto \max\{g(1),\dots,g(n)\}$$ we can and will assume that $g$ is monotonically increasing. [^1]
Let $\{G_i\}_{i\in \N}$ be any normal cofinal filtration of $\G$. We denote the projection maps $\G\to \G/G_i$, $i\in \N$, by $\rho_i$. We write $d_i:=[\G:G_i]$, $i\in \N$. We pick an epimorphism $\phi\colon F\to \Z$ and given $n\in \N$ we denote by $\phi_n\colon F\xrightarrow{\phi}\Z\to \Z/n$ the canonical projection. We also write $\psi_n=\phi_n\circ \a$.
Since $ \lim_{n\to \infty} \frac{g(n)}{n}=0$ we can iteratively pick $n_i\in \N$ with $$\frac{g(n_id_i)}{n_id_i}< \frac{1}{d_i}, \mbox{ i.e. such that } g(n_id_i)<n_i$$ and such that $n_{i+1}|n_i$ if $i>1$. We now define $$\G_i:=\ker\{ \rho_i\times \psi_{n_i}\colon \G\to \G/G_i\times \Z/n_i\}.$$ Note that $n_id_i\geq [\G:\G_i]$ and note that $\{\G_i\}_{i\in \N}$ is a cofinal normal filtration of $\G$. Given any $i\in \N$ we then have $$\ba{rcl} \frac{1}{g([\G:\G_i])} b_1(\G_i)&\geq& \frac{1}{g(n_id_i)}b_1(\G_i)\\
&\geq & \frac{1}{n_i}b_1(\ker\{ \rho_i\times \psi_{n_i}\colon \G\to \G/G_i\times \Z/n_i\})\\[2mm]
&\geq & \frac{1}{n_i}b_1(\ker\{ \phi_{n_i}:F\to \Z/n_i\})\\
&=&\frac{1}{n_i}(n_ib_1(F)-1)\geq 1.\ea$$
We are now in a position to prove Theorem \[mainthm1\].
Let $N\ne S^1\times D^2$ and $N\ne T^2\times I$ be a 3-manifold which is neither spherical nor covered by a torus bundle. By Theorem \[thm:akmw\] (1) the group $\pi=\pi_1(N)$ is large, i.e. it admits a finite index subgroup $\G$ which surjects onto a non-cyclic free group. Since this property is preserved by going to finite index subgroups we can assume that $\G$ is a normal subgroup of $\pi$. We write $k=[\pi:\G]$.
Let $f\colon \N\to \R_{\geq 0}$ be a function with $\lim_{n\to \infty} \frac{f(n)}{n}=0$. After possibly replacing $f$ by $$n\mapsto n\sup \left\{ \frac{f(n)}{n},\frac{f(n+1)}{n+1},\dots \right\}$$ we can and will assume that $\frac{f(n)}{n}$ is monotonically decreasing.
We apply Lemma \[lem:fastb1growth\] to $\G$ and the function $g(n)=kf(n)$ and we denote by $\{\G_i\}_{i\in \N}$ the resulting cofinal normal filtration of $\G$. Note that $\{\G_i\}_{i\in \N}$ is a cofinal almost normal filtration of $\pi$, and that $$\ba{rcl} b_1(\G_i)&\geq & f([\G:\G_i])[\pi:\G]\\
&=&\frac{f([\G:\G_i])}{[\G:\G_i]}[\G:\G_i][\pi:\G]\\
&\geq& \frac{f([\pi:\G_i])}{[\pi:\G_i]}[\G:\G_i][\pi:\G]=f([\pi:\G_i]).\ea$$
By Lemma \[lem:fastb1growth\] there exists a cofinal normal filtration $\{\G_i\}_{i\in \N}$ of $\G$ such that $$b_1(\G_i) \geq k^{\frac{1}{2k}}\sqrt{[\G:\G_i]} \mbox{ for every }i\in \N.$$ We pick a complete set of representatives $a_1,\dots,a_k$ for $\pi/\G$. Given $i\in \N$ we define $$\pi_i:=\bigcap\limits_{j=1}^k a_j\G_ia_j^{-1}.$$ Note that $\{\pi_i\}_{i\in \N}$ is a normal cofinal filtration of $\pi$. Also note that $$\pi_i=\ker\{ \G\to \G/a_1\G_ia_1^{-1} \times \dots \times \G/a_k\G_ia_k^{-1}\}.$$ It thus follows that $$[\pi:\pi_i]=[\pi:\G]\cdot [\G:\pi_i]\leq [\pi:\G]\cdot [\G:\G_i]^k=k\cdot [\G:\G_i]^k.$$ Finally note that $b_1(\pi_i)\geq b_1(\G_i)$, we thus see that for every $i$ we have $$b_1(\pi_i)\geq b_1(\G_i)\geq k^{\frac{1}{2k}}\sqrt{[\G:\G_i]} \geq [\pi:\pi_i]^{\frac{1}{2k}}.$$
It seems unlikely that one can turn the almost normal sequence of Theorem \[mainthm1\] (1) into a normal sequence without paying a price. For example consider the group $$\pi= \Z/2 \ltimes (F\times F)$$ where $F$ is a free non-cyclic group and $1\in \Z/2$ acts by commuting the two copies of $F$. If we apply the principle of the proof of Theorem \[mainthm1\] (1) to $\G=F\times F$ and $\a\colon F\times F\to F$ the projection on the first factor and $\G_n:=\ker\{F\times F\to F\to \Z/n\}$, then if we normalize these groups we really take the kernel $\ker\{F\times F\to F\to \Z/n\times \Z/n\}$ but now the growth of the Betti numbers is sublinear (in fact it grows with the square root of the index).
Growth of the rank of virtually fibered 3-manifolds {#section:slowgrowth}
---------------------------------------------------
In the following we mean by a surface group $G$ the fundamental group of a compact orientable surface. We will make use of the following two facts:
For any surface group $G$ we have $b_1(G)=d(G)$.
If $H$ is a finite index subgroup of a surface group $G$, then an Euler characteristic argument shows that $ b_1(H)\leq l\cdot b_1(G)$. We can now formulate and prove the following lemma.
\[lem:dsemidirect\] Let $\G=\Z\ltimes G$ be the semidirect product of $\Z$ with a surface group $G$. Let $f\colon \N\to \R_{\geq 0}$ be a function with $ \lim_{n\to \infty} f(n)=\infty$. Then there exists a normal cofinal filtration $\{\G_i\}_{i\in \N}$ of $\G$ such that $$d(\G_i)\leq f([\G:\G_i]) \mbox{ for every }i\in \N.$$
Let $G$ be a surface group. We write $r=b_1(G)$. Note that surface groups are residually finite, in particular there exists a cofinal filtration $\{G_i\}_{i\in \N}$ of $G$ by characteristic finite index subgroups of $G$. (Recall that a subgroup of $G$ is called characteristic if it is preserved by every automorphism of $G$.) We write $d_i:=[G:G_i]$, $i\in \N$.
We denote by $\phi\colon \G=\Z\ltimes G\to \Z$ the projection onto the first factor and given $n\in \N$ we denote by $\phi_n\colon \G=\Z\ltimes G\to \Z/n$ the composition of $\phi$ with the surjection onto $\Z/n$. Since $\lim_{n\to \infty} f(n)=\infty$ we can iteratively pick $n_i\in \N$ such that $$f(n_id_i)\geq 1+d_ir$$ and such that $n_i|n_{i+1}$ for $i>1$. We then define $ \G_i:=n_i\Z\ltimes G_i$. Note that $\G_i, i\in \N$ is normal in $\G=\Z\ltimes G$ since $G_i\subset G$ is characteristic. In particular the $\{\G_i\}_{i\in \N}$ form a normal cofinal filtration of $\G$. It now follows that $$\ba{rcl} d(\G_i)=d(n_i\Z\ltimes G_i)\leq 1+d(G_i)&=&1+b_1(G_i) \\
&\leq& 1+d_ir\\
&\leq& f(n_id_i)= f([\G:\G_i]).\ea$$
We are now in a position to prove Theorem \[mainthm2\].
Let $N$ be a 3-manifold which is not a closed graph manifold. We write $\pi=\pi_1(N)$. By Theorem \[thm:akmw\] (2) there exists a finite cover $\ti{N}$ which fibers over $S^1$, i.e. $\G:=\pi_1(\ti{N})\cong \Z\ltimes G$, where $G$ is a surface group. Since finite covers of fibered 3-manifolds are again fibered, we can assume that $\G:=\pi_1(\ti{N})$ is a normal subgroup of $\pi$.
Let $g\colon \N\to \R_{\geq 0}$ be a function with $ \lim_{n\to \infty} g(n)=\infty$. We then apply Lemma \[lem:dsemidirect\] to $\G=\Z\ltimes G$ and $f(n):=\frac{1}{[\pi:\G]}g(n)$. The resulting filtration is an almost normal cofinal filtration of $\pi$ with the desired property.
By Lemma \[lem:dsemidirect\] there exists a normal cofinal filtration $\{\G_i\}_{i\in \N}$ of $\G$ such that $$d(\G_i)\leq [\G:\G_i]^{\frac{1}{2}}\mbox{ for all $i$.}$$ Given $i\in \N$ we write $n_i:=[\G:\G_i]$. We now denote by $a_1,\dots,a_k$ a complete set of representatives of $\pi/\G$. Given any $i\in \N$ we define $$\pi_i:=\bigcap\limits_{j=1}^k a_j\G_ia_j^{-1}\subset \G_i.$$ Note that $\{\pi_i\}_{i\in \N}$ is now a normal cofinal filtration of $\pi$. Given $i\in \N$ we write $s_i:=[\G_i:\pi_i]$. Note that $n_i\cdot s_i=[\G:\G_i]\cdot [\G_i:\pi_i]\leq n_i^k$. We thus see that $s_i\leq n_i^{k-1}$. Using this observation we obtain that $$\ba{rcl} d(\pi_i) \leq [\G_i:\pi_i]\cdot d(\G_i)&=&s_i \cdot n_i^{\frac{1}{2}}\\[1mm]
&=& s_i^{\frac{2k-1}{2k}}s_i^{\frac{1}{2k}}\cdot n_i^{\frac{1}{2}}
\leq s_i^{\frac{2k-1}{2k}}\cdot n_i^{\frac{k-1}{2k}}n_i^{\frac{1}{2}}\\[1mm]
&=& s_i^{\frac{2k-1}{2k}}n_i^{\frac{2k-1}{2k}}= k^{-\frac{2k-1}{2k}} (s_in_ik)^{\frac{2k-1}{2k}}\\
&=& k^{-\frac{2k-1}{2k}}\cdot [\pi:\pi_i]^{\frac{2k-1}{2k}}.\ea$$ It follows that the sequence $\{\pi_i\}_{i\in \N}$ together with $\eps=\frac{2k-1}{2k}$ has the desired properties.
[asd]{}
I. Agol, [*Criteria for virtual fibering*]{}, J. Topol. 1 (2008), no. 2, 269–284
I. Agol, [*The virtual Haken conjecture*]{}, with an appendix by I. Agol, D. Groves and J. Manning. Preprint (2012).
M. Aschenbrenner, S. Friedl and H. Wilton, [*$3$-manifold groups*]{}, Preprint (2012)
F. Calegari and M. Emerton, [*Mod-$p$ cohomology growh in $p$-adic analytic towers of $3$-manifolds*]{}, Groups, Geometry and Dynamics, to appear.
B. Clair and K. Whyte, [*Growth of Betti numbers*]{}, Topology 42 (2003), no. 5, 1125–1142.
J. DeBlois, [*Rank gradient of cyclic covers*]{}, Preprint (2010)
M. Ershov and W. Lück, [*The first $L^2$-Betti number and approximation in arbitrary characteristic*]{}, Preprint (2012)
D. Girão, [*Rank gradient in cofinal towers of certain Kleinian groups*]{}, Preprint (2010), to appear in Groups, Geometry and Dyanmics.
F. González-Acuña and H. Short, [*Cyclic branched coverings of knots and homology spheres*]{}, Rev. Mat. Univ. Complut. Madrid 4 (1991), no. 1, 97–120.
J. Hempel, [*Residual finiteness for $3$-manifolds*]{}, Combinatorial group theory and topology (Alta, Utah, 1984), 379–396, Ann. of Math. Stud., 111, Princeton Univ. Press, Princeton, NJ, 1987
J. Kahn and V. Markovic, [*Immersing almost geodesic surfaces in a closed hyperbolic three manifold*]{}, Ann. of Math. 175 (2012), 1127–1190.
S. Kionke and J. Schwermer, [*On the growth of the first Betti number of arithmetic hyperbolic 3-manifolds*]{}, Preprint (2012)
T. Kitano, T. Morifuji and M. Takasawa, [*$L^2$-torsion invariants and homology growth of a torus bundle over $S^1$*]{}, Proc. Japan Acad. Ser. A Math. Sci. 79 (2003), no. 4, 76–79.
S. Kojima, [*Finite covers of $3$-manifolds containing essential surfaces of Euler characteristic $=0$*]{}, Proc. Amer. Math. Soc. 101 (1987), no. 4, 743–747.
M. Lackenby, [*New lower bounds on subgroup growth and homology growth*]{}, Proc. Lond. Math. Soc. (3) 98 (2009), no. 2, 271–297.
M. Lackenby, [*Finite covering spaces of $3$-manifolds*]{}, Proceedings of the International Congress of Mathematicians 2010, edited by R. Bhatia, A. Pal, G. Rangarajan and V. Srinivas (2011).
T. Le, [*Homology torsion growth and Mahler measure*]{}, Preprint (2010), to appear in Comment. Math. Helv.
Y. Liu, [*Virtual cubulation of nonpositively curved graph manifolds*]{}, Preprint (2011), to appear in J. of Topology J. Lott and W. Lück, [*$L^2$-topological invariants of 3-manifolds*]{}, Invent. Math., 120(1):15-60, 1995
W. Lück, [*Approximating $L^2$-invariants by their finite-dimensional analogues*]{}, Geom. Funct. Anal., 4(4):455-481, 1994.
J. Luecke, [*Finite covers of $3$–manifolds containing essential tori*]{}, Trans. Amer. Math. Soc. 310 (1988), 381–391.
W. Magnus, A. Karrass and D. Solitar, [*Combinatorial Group Theory: Presentations of Groups in Terms of Generators and Relations*]{}, Second revised edition. Dover Publications, Inc., New York (1976)
P. Przytycki and D. Wise, [*Graph manifolds with boundary are virtually special*]{}, Preprint (2011), to appear in J. of Topology.
P. Przytycki and D. Wise, [*Mixed 3-manifolds are virtually special*]{}, Preprint (2012).
J. Raimbault, [*Exponential growth of torsion in abelian coverings*]{}, Algebraic & Geometric Topology, to appear.
R. Riley, [*Growth of order of homology of cyclic branched covers of knots*]{}, Bull. London Math. Soc. 22 (1990), no. 3, 287–297.
P. Shalen and P. Wagreich, [*Growth rates, $\Z_p$-homology, and volumes of hyperbolic $3$-manifolds*]{}, Trans. Amer. Math. Soc. 331 (1992), no. 2, 895–917.
D. Silver and S. Williams, [*Mahler measure, links and homology growth*]{}, Topology 41 (2002), no. 5, 979–991.
D. Silver and S. Williams, [*Torsion numbers of augmented groups with applications to knots and links*]{}, Enseign. Math. (2) 48 (2002), no. 3-4, 317–343.
L. Wall, [*Homology growth of congruence subgroups*]{}, PhD Thesis, Oxford University (2009).
S. Wang and F. Yu, [*Graph manifolds with non-empty boundary are covered by surface bundles*]{}, Math. Proc. Cambridge Philos. Soc. 122 (1997), no. 3, 447–455.
D. Wise, [*The structure of groups with a quasi-convex hierarchy*]{}, 189 pages, preprint (2012), downloaded on October 29, 2012 from\
`http://www.math.mcgill.ca/wise/papers.html`
[^1]: Note that if $\lim_{n\to \infty} \frac{g(n)}{n}=0$ and if we set $f(n):=\max\{g(1),\dots,g(n)\}$, then $\lim_{n\to \infty} \frac{f(n)}{n}=0$ as well. Indeed, let $\eps>0$. By assumption there exists an ${N}$ such that $\frac{g(n)}{n}<\eps$ for all $n\geq {N}$. We now let $M$ be any integer greater than ${N},\frac{2}{\eps}g(1),\dots,\frac{2}{\eps}g({N}-1)$. For every $n\geq M $ we then have $$\ba{rcl} \frac{1}{n} f(n)&=& \max\{\frac{1}{n}g(1),\dots,\frac{1}{n}g({N}-1),\frac{1}{n}g({N}),\dots,\frac{1}{n}g(M)\}\\[2mm]
&\leq& \max\{\frac{1}{M}g(1),\dots,\frac{1}{M}g({N}-1),\frac{1}{{N}}g({N}),\dots,\frac{1}{M}g(M)\}<\eps.\ea$$
| {
"pile_set_name": "ArXiv"
} |
---
author:
- '山崎 雅人（東京大学カブリ数物連携宇宙研究機構）'
title: '団代数と超対称ゲージ理論[^1]'
---
団代数と物理学者の邂逅
======================
筆者は理論物理学者であるが， 「数物連携」と名のつく研究所に所属しており 数学者とも交流が深い． 理論物理学者と純粋数学者とでは 研究の手法も動機も大きく異なるが， 昨今の素粒子理論の研究においては， 純粋数学において育てられた概念に，思わぬところで 出くわすことも少なくない．
本特集で取り上げられている**団代数**（クラスター代数） もその一つのめざましい例である．本稿では， 超対称場を研究する物理学者たちが， いかにして団代数に出会ったかを 説明したい．
電場と磁場のハーモニー
======================
我々が考えたいのは我々の住む4次元時空， すなわち空間3次元，時間１次元において定義された **ゲージ場の理論**である．
ゲージ理論はゲージ群$G$を指定することによって得られるのであった． 簡単な例として，ゲージ群が$U(1)$の 場合を考えよう．これはおなじみの電磁気学であり， $U(1)$ゲージ場は光子を表す． 電磁気学では，光子だけではなく， 光子と相互作用する物質場，例えば電子を 考える．これはゲージ群のもとで，とある 電荷$e$をもった場のことである．
電磁場は磁場と電場からなっていたが，電子は電場のみに対して チャージを持っていた．逆に，磁場のみに対してチャージ（これを磁荷$g$と書こう）を持つ のが**磁気モノポール**（磁気単極子）である． ディラックが気づいたのは，モノポールと電子の両者を考えることで， マックスウェル方程式は電場と磁場の入れ替えについて対称な形になることであった．さらに彼は， 電荷と磁荷の満たすべき量子化条件 $$\begin{aligned}
e g \in 2\pi \hbar\, \mathbb{Z}\end{aligned}$$ を導いたのであった[^2]．
ここまでは電荷ないし磁荷のみを持つ理論を考えたが， 両方を同時に持つ粒子（**ダイオン**）を考えてもよい．その場合の量子化条件は， 二つの粒子の電荷・磁荷の組（以下単にチャージと呼ぶ）を $\gamma_{i=1,2}=(e_i, g_i)$としたとき，その ペアリング$\langle \gamma_1 , \gamma_2 \rangle$の量子化条件として表される： $$\begin{aligned}
\langle \gamma_1, \gamma_2 \rangle := e_1 g_2- e_2 g_1\in 2\pi \hbar\, \mathbb{Z}\ .\label{pairing}\end{aligned}$$ すぐにわかるように，このペアリングは 完全半対称$\langle \gamma_1 , \gamma_2 \rangle = -\langle \gamma_2 , \gamma_1 \rangle$である． ゲージ群をより一般の可換ゲージ群$U(1)^r$にした時も，チャージ$e, g$がそれぞれ$r$成分を持つ ベクトル$\vec{e}, \vec{g}$になることを除けば量子化条件は同様である． 式で導入したチャージのペアリングは後で見るように 箙の定義に用いるので重要である．
4次元$\mathcal{N}=2$理論のクーロンブランチ
==========================================
ここまでの議論では 例えばゲージ場のラグランジアンの 具体的な形を直接必要としたわけではない． 従って，もともとゲージ群が非可換の設定から出発しても，最終的にゲージ群がその可換部分に破れている限り，議論は同様であると期待される．
ゲージ群が破れると聞いてまず思い出すのはヒッグズ効果である：例えば$SU(3)$ゲージ群の基本表現に属する クオークが期待値を持てばゲージ群は破れる．しかし，この場合ゲージ群は一般には完全に破れてしまい，可換ゲージ群すら残らない．
そこで，通常のクオークの代わりにゲージ群に対して随伴表現で変換する場$\sigma$（つまりゲージ群の元$g$のもと$\sigma\to g^{-1} \sigma g$と変換する場）が存在し，その場が（一般の）期待値を持ったとしよう．このとき，ゲージ群の非可換部分は破れるが，ゲージ群の可換部分群$U(1)^{r}$が全て残ることになる（随伴表現はゲージ群の可換部分群に対して変換を受けない）．ここで$r$はゲージ群のランクとよばれる量である．
このような状況のうち典型的なものとして，**4次元$\mathcal{N}=2$理論**を考える． 超対称性はボソンとフェルミンを入れ替える対称性であるが，$\mathcal{N}=2$対称性では 独立な超対称変換が二つ存在するので，その二つを組み合わせるとボソンが別のボソンに移されることになる． 特に，ボソンであるゲージ場$A_{\mu}$に別のボソンのペアが存在し，それが先に述べたスカラー 場$\sigma$である．ゲージ場はゲージ群の随伴表現で変換するので，それのペアである $\sigma$も期待通り随伴表現で変換する． $\mathcal{N}=2$対称性を持つ理論では，$\sigma$が期待値を持つ真空が存在し，そこでは可換なゲージ場が残るので**クーロンブランチ**（クーロン枝）[^3]と呼ばれる．我々はこの真空における4次元$\mathcal{N}=2$理論に対する低エネルギーでの振る舞いを調べることにしよう．そのような低エネルギー有効理論を解くのが，90年代半ばに現れたザイバーグ・ウィッテン理論 であり，これまで本誌上でもたびたび取り上げられてきた．団代数は，この古典的な設定とその一般化を議論する中で現れてきたのである．
BPS粒子から箙へ
===============
クーロンブランチではゲージ群は可換であるが， 我々の出発点は非可換ゲージ群をもつゲージ理論である． 物質場の数が多すぎない時，理論は漸近自由性を持ち 我々の興味のある低エネルギーでは強結合領域にありその 直接の解析は容易ではない．
ここでは，理論そのものを直接調べる代わりに，理論のスペクトラムを調べることにしよう． つまり，どういうチャージを持った安定な粒子が，幾つ存在するかを考えるのである． $\mathcal{N}=2$超対称性からの制限を最大限に活用するために，特に$\mathcal{N}=2$超対称性のうちの 最大限（今の場合は半分の）超対称性を保つ粒子のみを考えることにしよう． このような粒子のことを **BPS粒子**（今の場合は正確には$\frac{1}{2}$-BPS粒子）と呼ぶ．
BPS粒子の特徴として，チャージ$\gamma$を持つBPS粒子の質量は，そのチャージから 定まる複素数$Z_{\gamma}$（**セントラルチャージ**）の絶対値$\big| Z_{\gamma} \big|$で決まる． 一方，$Z_{\gamma}$の位相部分は，$\mathcal{N}=2$対称性のうちどの$\mathcal{N}=1$超対称性 を保つかを指定している．さらに，$Z_{\gamma}$は，$\gamma$に対して線形である： $Z_{\gamma_{1}+\gamma_2}=Z_{\gamma_1}+Z_{\gamma_2}, Z_{n \gamma}=n Z_{\gamma}$.
クーロンブランチでは破れずに残った可換ゲージ群$U(1)^r$が存在するので， 粒子はそのゲージ群についてのチャージ$\gamma\in \Gamma$を持つ． ここで$\Gamma$は許されるチャージの全体であり，完全反対称ペアリングを持ちそのもとで 量子化条件式を満たす．
我々は相対論的な局所場の理論から 出発しているので，CPT定理が成立し，特に チャージ$\gamma$を持つ粒子が存在すれば， 逆のチャージ$-\gamma$を持つ反粒子も存在しなけれればならない． したがって，粒子を数え上げる時には二つのうちどちらか一方だけを考えてやればよく， $\Gamma$は二つの交わりを持たない和に分解する： $$\begin{aligned}
\Gamma=\Gamma_{\rm 正} \cup \Gamma_{\rm 負} \ .\end{aligned}$$
もっとも，この分解は一意ではない． 我々の目的のためには， とある偏角$\zeta$を定めて先に導入した複素数$Z_{\gamma}$が複素平面上で 偏角が$[\zeta, \zeta+\pi]$にあるものを集めて$\Gamma_{\rm 正}$とすればよい（図\[half\_plane\]）： $$\begin{aligned}
\gamma\in \Gamma_{\rm 正} \longleftrightarrow \textrm{Im} (e^{- i\zeta} Z(\gamma))>0 \ .
\label{positive}\end{aligned}$$
![チャージのなす集合$\Gamma$を$Z_{\gamma}$の偏角に応じて 正負に分割する．[]{data-label="half_plane"}](gamma_1.eps)
さて，ここから箙を定義しよう． $\Gamma_{\rm 正}$を生成する基底を$\{ \gamma_i \}$としよう：[^4] $$\begin{aligned}
\Gamma_{\rm 正} = \displaystyle\bigoplus_i \mathbb{Z}_{\ge 0} \gamma_i \ . \label{positive_span}\end{aligned}$$ この時， $$\begin{aligned}
b_{ij}:=\langle \gamma_i, \gamma_j \rangle
\label{bdef}\end{aligned}$$ により半対称行列$B=\left(b_{i,j} \right)$を，従って箙を定義するのである（本特集中西氏の 記事参照 ）．この箙を**BPS箙**（えびら）と呼ぶことにしよう． 既に述べたようにペアリングは反対称であったから，$b_{ij}$も 反対称行列を与えることに注意されたい．
式の定義は天下りだが，その背後にはちゃんと物理的意味がある． BPS粒子は4次元$\mathcal{N}=2$理論の超対称ゲージ理論の中に存在する粒子であるが，それ自体 半分の超対称性を保っている．したがって，その粒子の上にいる人の立場に立てば， 4次元$\mathcal{N}=1$を持つ超対称量子力学が現れたように見える（粒子に対する場の理論は量子力学である）． 先に定義した箙はこの超対称量子力学（**箙超対称力学**）の定義データを与えるのである．
具体的には次のようにすれば良い：箙の頂点にはチャージ上で定義した基底の一つの元$\gamma_i$が対応する． $\gamma=\sum_i n_i \gamma_i$なるチャージを考えた時には $i$番目の頂点にはゲージ群$U(n_i)$を考え，二つの頂点$i$からでて$j$までを結ぶ辺には， $U(n_i) \times U(n_j)$のもとで$(n_i, \bar{n}_j)$として変換する場を対応させる[^5][^6]．
箙の変異
========
さてここまでで箙とその物理的意味を説明してきたが， ここまでの説明には不満足な点がある．それは， 正と負のチャージへの分割が一意ではないということだ． このことは式が偏角$\zeta$に依存することからも明らかである． 特に，変化が起こるのは$\zeta$が基底のある元$\gamma_k$の偏角を越えて変化すると， $\Gamma_{\rm 正}$が，従って箙が変化する（図\[half\_plane\_change\]）．
それまで正のチャージを持っていた$\gamma_k$が 負のチャージを持つようになったとしよう． このとき，明らかに$-\gamma_k$を新たな基底に加えなければならない: $$\begin{aligned}
\gamma'_k=-\gamma_k \ .
\label{gchange_1}\end{aligned}$$ しかし，を保つ必要があるので 他の基底の元も取り替える必要がある．その結果は $$\begin{aligned}
\gamma_i \to \gamma_i + \left[b_{ik} \right]_{+} \gamma_k
\label{gchange_2}\end{aligned}$$ であることが知られている． ただし，ここで$[x]_{+}:=\textrm{max}(x,0)$． このとき，箙はの定義から $$\begin{aligned}
\begin{split}
&b_{ik} \to -b_{ik} \ , \\
&b_{ij} \to b_{ij} + \left[b_{ik} \right]_{+} b_{kj}+ \left[b_{jk}\right]_{+} b_{ik}
\end{split}
\label{eq.mutate_rule}\end{aligned}$$ と変化する．これは中西氏の解説中(5)式で導入された箙の変異 $B\to B'=\mu_k(B)$に他ならない． こうして我々は箙の変異に辿りついた．
![$\zeta$の値を変化させることでチャージの正負への分割を変更する． この時，$\Gamma_{\rm 正}$の基底も取り替える必要がある．[]{data-label="half_plane_change"}](gamma_2.eps)
の証明はここでは行わないが，例えば箙の表現論の立場から[@Alim:2011kw]のsection 3.1.2を参照されたい．また，より直接の説明として，先に説明した箙量子力学の双対性を用いたものがある．同じ箙からは2次元超対称場の理論も定義することができ，箙の変異はその双対性を表している[@Benini:2014mia]．それを1次元量子力学にに次元還元したのがここでの箙の変異を表すのだ．
クラスター$y$変数とループ演算子
===============================
クラスター代数はクラスター$x$変数や**$y$変数**（係数）と呼ばれる変数が重要な役割を果たした． 実は，我々の設定ではクラスター$y$変数はクーロンブランチの座標として現れるのである．
ここでは一般の4次元$\mathcal{N}=2$理論を考える代わりに， $A_{N-1}$型の6次元$(2,0)$理論を点付き（つまり，穴のある）リーマン面$C$の上にコンパクト化した理論を考えることにしよう． $$\begin{aligned}
\textrm{6次元理論} : \mathbb{R}^4\times C \longrightarrow
&
\,
\textrm{4次元理論} : \mathbb{R}^4 \ .
\label{6d4d}\end{aligned}$$ ここで，6次元$A_{N-1}$型$(2,0)$理論はその正体が明らかでない謎の理論であり直接の役には立たないが， その$S^1$コンパクト化が5次元$\mathcal{N}=2$ $SU(N)$ゲージ理論を与えることはよくわかっている． そこで，$\mathbb{R}^4$のうち一方向を$S^1$にコンパクト化することを考えよう： $$\begin{aligned}
\begin{split}
& \textrm{6次元$A_N$型理論} : \mathbb{R}^3\times S^1 \times C \\
& \longrightarrow \,
\textrm{5次元$SU(N)$ゲージ理論} : \mathbb{R}^3 \times C \ .
\end{split}
\label{5d_compact}\end{aligned}$$ となり，それは実行できる．では先に$S^1$コンパクト化したが，順序を変えて 先に$C$にコンパクト化することすると，の 4次元理論が3次元理論に $S^1$コンパクト化されることになる： $$\begin{aligned}
\textrm{4次元理論} : \mathbb{R}^3\times S^1 \longrightarrow
\textrm{3次元理論} : \mathbb{R}^3 \ .
\label{4d3d}\end{aligned}$$
さて，5次元理論のラグランジアンを用いて$C$上のBPS方程式を解析すると， リーマン面$C$上に**ヒッチン・モジュライ**と呼ばれるモジュライ空間が現れる． この空間は複素構造の取り方によって幾つかの記述があるが[^7]， そのうちの一つは，$PSL(N, \mathbb{C})$平坦接続の空間であるというものである． つまり，$C$上の複素接続$\mathcal{A}$で， $$\begin{aligned}
\mathcal{F}=d \mathcal{A} + \mathcal{A} \wedge \mathcal{A}=0\end{aligned}$$ を満たすものの全体を，ゲージ変換で割ったものである[^8][^9]． こうして現れた$PSL(N, \mathbb{C})$平坦接続の空間には， 自然な座標（**フォック・（ゴンチャロフ）座標**）[@FG]が 存在することが知られている．
ここでは簡単のため$N=2$の場合を考えることにしよう． 頂点を$C$の穴にもつ 三角形分割（そのような三角形分割を**理想三角形分割**と呼ぶ）を考え， 各三角形ごとに 図\[trig\]の箙を書くことで$C$上に書かれた箙が$B=\left(b_{ij} \right)$が得られる（但し，$C$は十分な数の穴を持ち，そのような三角形分割が存在すると仮定する）． このとき，箙の辺$i$（理論三角形分割の辺）複素変数 $y_{\gamma_i}$を対応させるとそれがヒッチン・モジュライの座標となり， そのもとでのシンプレクティック形式は簡単な形 $$\begin{aligned}
\big\{ y_{\gamma_i}, y_{\gamma_j} \big\}= b_{i,j}
\label{omega}\end{aligned}$$ で与えられるというのが数学的な結果である． なお，以下$y_{\gamma_1+ \gamma_2}:=y_{\gamma_1}+ y_{\gamma_2}$ $y_{n\gamma}=n y_{\gamma})$と定義することで， 一般の$\gamma\in \Gamma$に対し$y_{\gamma}$を定義しておくと便利である．
式を古典力学でおなじみのポアソン括弧とみなすことにすれば， $y_{\gamma_i}$は有限次元の古典力学の相空間の座標に他ならない． と比較すれば，辺$i$は正チャージの基底$\gamma_i$に対応さて， 図\[trig\]から定まる箙は4次元$\mathcal{N}=2$理論のBPS箙と同一視 するのが自然である．
さらに，座標$y_{\gamma}$はチャージ$\gamma$によって指定される 赤外理論での**ループ演算子**の期待値と同定される[@GMN]． ここでループ演算子とは，一次元的に広がった演算子のことであり，その代表例は ウィルソンラインであり，ゲージ場を閉路に沿って積分したものである： 4次元理論でのループ演算子は，4次元理論を$S^1$上で次元還元した時， その$S^1$方向に巻きついているとすると3次元理論の粒子になる． 4次元でのクーロンブランチは随伴表現に値を持つ場$\sigma$の期待値によってパラメーター付けされていたのであった．3次元に次元還元すると，ゲージ場の$S^1$方向のゲージ場の積分は 3次元に新たな複素スカラー場$\bar{\sigma}$を与え，$\sigma, \bar{\sigma}$がペアをなして 3次元のクーロンブランチをパラメーター付けする． したがって，ループ演算子の期待値の複素化 がクーロンブランチを指定する座標を与えるというのは自然であると納得できる．
![理想三角形分割と，それから定めるBPS箙．[]{data-label="trig"}](quiverrule_2014_11.eps)
厳密に3次元を考えるのではなく，$S^1$の半径$R$を 有限に保った時には 何が起こるだろうか？ にたちもどると， これは，5次元理論から6次元理論への持ち上げである． 超弦理論の言葉では，これはIIA型超弦理論の5次元のブレーン（D4ブレーン）が M理論の6次元のブレーン（M5ブレーン）へ持ち上がる過程である．
このとき，3次元での粒子は4次元のループ演算子に持ち上がる． 粒子とは異なり，一般の電荷・磁荷を持つ ループ演算子はお互いに交換しない演算子になることが知られている[@Polyakov; @GMN]から， 座標$y_{\gamma_i}$は交換しない演算子$\hat{y}_{\gamma_i}$に置き換わるはずである． 自然な量子化は有限次元相空間のポアソン括弧を 演算子の交換関係に置き換えて得られる： $$\begin{aligned}
\left[ \hat{y}_{\gamma_i}, \hat{y}_{\gamma_j} \right]= i \hbar \, b_{i,j} \ .
\label{omega_q}\end{aligned}$$ あるいは，$\hat{Y}_{\gamma}=e^{\hat{y}_{\gamma}}$で定義される変数を用いることにすると， いわゆる**量子トーラス**が得られる： $$\begin{aligned}
\hat{Y}_{\gamma_1} \hat{Y}_{\gamma_2} = q^{\langle \gamma_1, \gamma_2 \rangle }\hat{Y}_{\gamma_2}\hat{Y}_{\gamma_1} \ .
\label{qTorus}\end{aligned}$$ ただし，$q:=e^{i \hbar}$であり，古典極限は$q\to 1$となる． このように，4次元理論（従ってM理論）への持ち上げは，ヒッチン・モジュライを量子化するのである[^10]．
箙の変異とフリップ
==================
ここまで，M理論を考えることでヒッチン・モジュライが量子化されることを みてきたが，それでは箙の変異はどう量子化されるのだろうか．
![理想三角形分割に対するフリップはBPS箙の変異を引き起こす[]{data-label="trig_mutation"}](flip_2014_11.eps)
ここでは，この変異を組み合わせ論的に説明しておこう． 箙は理想三角形分割から定まっていたのであるから， 箙の曖昧さは理想三角形分割の曖昧さに起因していることになる[^11]．
理想三角形分割の任意性は，理想四角形の対角線を取り替える操作（**フリップ**と呼ばれる）を繰り返すことによって 尽くされることが知られている．そこで，フリップのもとで変数 $\hat{Y}_{\gamma}$が変化するかを調べればよい．
まず，辺$k$においてフリップをした時，対応する箙$B$は 頂点における変異$\mu_k(B)$に移り変わる（図\[trig\_mutation\]）． このときチャージは及びに従って 変化するので，対応する変数$\hat{Y}_{\gamma_k}$も $$\begin{aligned}
\mu_k:
\begin{array}{l}
\hat{Y}_{\gamma_k} \to \hat{Y}_{-\gamma_k}=\hat{Y}_{\gamma_k}^{-1} \ ,\\
\hat{Y}_{\gamma_i} \to \hat{Y}_{\gamma_i+ [b_{ik}]_{+} \gamma_k} \quad (i\ne k)
\end{array}
\label{Xchange_1}\end{aligned}$$ と変化するのが自然である． 実際，$\hat{Y}_{\gamma_i}$がを満たす時，こうして変換した後の$\hat{Y}_{\gamma}$は新しい箙$B'=\mu_k(B)$に対するを満たすことが確認できる（なお，このために $$\begin{aligned}
\hat{Y}_{\gamma_i+ [b_{ik}]_{+} \gamma_k}
=q^{-\frac{1}{2}[b_{ik}]_{+} b_{ki}} \hat{Y}_{\gamma_k}^{[b_{ik}]_{+} } \hat{Y}_{\gamma_i} \end{aligned}$$を用いる． ）．しかしこれで話は終わりではない． 実際には $$\begin{aligned}
K_{\gamma_k}: \hat{Y}_{\gamma_i} \to \Psi_q(\hat{Y}_{\gamma_k})^{-1} \hat{Y}_{\gamma_i} \Psi_q(\hat{Y}_{\gamma_k})
\label{Xchange_2}\end{aligned}$$ なる変換をさらに行った合成 $$\begin{aligned}
\bar{\mu}_k := K_{\gamma_k} \mu_k\end{aligned}$$ がフリップによって引き起こされる変換なのである． ここで，$\Psi_q(x)$は**量子ダイログ関数**[@FK]と呼ばれる特殊関数であり， $$\begin{aligned}
\begin{split}
&\Psi_q(qx;q)=(1+q^{1/2} x)^{-1} \Psi_q(x;q) , \\
&\Psi_q(0;q)=1
\end{split}\end{aligned}$$ という関数関係式によって定義される． より具体的に書くと，$\bar{\mu}_k$の作用は，$\hat{Y}_i:=\hat{Y}_{\gamma_i}$と書くことにして $$\begin{aligned}
\hat{Y}_k \to \hat{Y}_k^{-1} \end{aligned}$$ 及び$i\ne k$の時，$s_{jk}:= \textrm{sgn}(b_{jk} )$と書くと $$\begin{aligned}
\hat{Y}_j \!\to \!
\prod_{n=0}^{|b_{jk}|-1} \left(1+q^{-(n+\frac{1}{2}) s_{jk}} Y_k^{-s_{jk} }\right)^{-s_{jk}} Y_ j\end{aligned}$$ で与えられる． これは**量子クラスター代数**における**量子$y$変数**の変換則に他ならない．
特に， $q=1$とすると． 次の変換則が得られる： $$\begin{aligned}
\begin{split}
&Y_k \to Y_k^{-1} \ , \\
&Y_j \to (1+ Y_k^{-\textrm{sgn}(b_{jk})} )^{-b_{jk}} Y_j\ .
\end{split}\end{aligned}$$ これはすでにいくつかの記事にも現れた （古典）クラスター代数の係数（**$y$変数**）の変換則に他ならない．
ここまでではただ1回の箙の変異を考えたが， 変異を繰り返すこともできる． その結果箙$B$は別の箙$B'$に変化し， $$\begin{aligned}
B'=\mu_k \cdots \mu_1(B) \ ,\end{aligned}$$ 演算子$\bar{\mu}_k$の積は$B$の量子トーラスから$B'$の量子トーラスへの写像 $$\begin{aligned}
\bar{\mu}_k\cdots \bar{\mu}_1 \end{aligned}$$ を与える．この形の演算子は，コンツェビッチとソイベルマンによる **壁越え現象**における公式の記述に用いられ， 関連して量子ダイログ関数の恒等式やYシステムの構成に用いられる． また，4次元$\mathcal{N}=2$理論の境界に現れる 3次元$\mathcal{N}=2$理論の分配関数としても 解釈することができる[@Terashima:2013fg]． 後者は3次元多様体の幾何にも関係しておりそれ自体 興味深い理論である（本特集寺嶋氏の記事並びに[@Yamazaki_book; @Yamazaki_Science]を参照）．
団代数の彼方へ
==============
以上，団代数の構造が超対称ゲージ理論の一つの文脈でどのように現れるかを解説してきた． 団代数とは，4次元$\mathcal{N}=2$理論の**ループ演算子のなす代数**，及びそれが チャージ・ラティス$\Gamma_{\rm 正}$の取り換えでどう変化するかを記述するものに他ならないのであった．より一般の超対称ゲージ理論に対しても，ループ演算子のなす演算子を 考えることにより同様の構造が得られると期待される（例えば[@KW]を参照）．
興味深いことに，純粋に数学上の興味から生まれた団代数は，超対称場の理論の物理的考察においても重要な構造であるのだ．筆者自身，当初は団代数のことを食わず嫌いでなんとなく敬遠していたが，折に触れて団代数のことは小耳に挟んでいた．その後，自分の研究の中でその強力さに気づき，宗旨替えをして現在に至っている[^12]．この記事をきっかけに，読者の皆さんが 団代数に少しでも親しみを感じて頂ければと願う．
本特集からも明らかなように 団代数は様々な文脈で姿をみせる．超対称場の理論での現れは団代数の 数多くの姿のうちの一つでしかないということもできるだろう[^13]．しかし，ここまで議論してきた超対称ゲージ理論の物理は 骨組みとしての代数的な団代数の構造そのものよりもはるかに豊穣な内容を 含んでいる．それは，団代数の多様な側面を一つにまとめてみせると同時に， 新たな数学の発展を刺激してきた．団代数からさらに何を汲み取ることができるのか，その先には何が待ち構えているのか，それを思うとき筆者の胸は高鳴る．
[99]{}
M. Alim, S. Cecotti, C. Cordova, S. Espahbodi, A. Rastogi and C. Vafa, Adv. Theor. Math. Phys. [**18**]{}, 27 (2014) \[arXiv:1112.3984 \[hep-th\]\].
F. Benini, D. S. Park and P. Zhao, Commun. Math. Phys. [**340**]{}, 47 (2015) doi:10.1007/s00220-015-2452-3 \[arXiv:1406.2699 \[hep-th\]\]. D. Gaiotto, G. W. Moore and A. Neitzke, Adv. Theor. Math. Phys. [**17**]{}, 241 (2013) \[arXiv:1006.0146 \[hep-th\]\]; arXiv:0907.3987 \[hep-th\].
V. Fock and A. Goncharov, Publ. Math. Inst. Hautes Études Sci., [**103**]{}, 1 (2006). D. Xie and K. Yonekura, JHEP [**1410**]{}, 134 (2014) \[arXiv:1404.7521 \[hep-th\]\].
Y. Ito, T. Okuda and M. Taki, JHEP [**1204**]{}, 010 (2012) \[arXiv:1111.4221 \[hep-th\]\].
A. M. Polyakov, Mod. Phys. Lett. A [**3**]{}, 325 (1988).
L. D. Faddeev and R. M. Kashaev, Mod. Phys. Lett. A [**9**]{}, 427 (1994) \[hep-th/9310070\].
M. Kontsevich and Y. Soibelman, arXiv:0811.2435 \[math.AG\].
Y. Terashima and M. Yamazaki, PTEP [**023**]{}, B01 (2014) \[arXiv:1301.5902 \[hep-th\]\].
場の理論の分解学，数理科学2012年10月号，山崎雅人, arXiv:1705.06377 \[hep-th\]. 場の理論の構造と幾何，山崎雅人，サイエンス社SGCライブラリ，2015年
A. Kapustin and B. Willett, arXiv:1302.2164 \[hep-th\]. “Special issue on cluster algebras in mathematical physics”, P. D. Francesco, M. Gekhtman, A. Kuniba and M. Yamazaki (ed.), <http://iopscience.iop.org/1751-8121/47/47>
[^1]: 雑誌数理科学2015年3月号特集「団代数をめぐって：新たな共通構造の認識」
[^2]: これらのことについては，例えば数理科学誌2014年7月号の特集 「モノポールの謎」を参照されたい．
[^3]: 枝と呼ばれるのは特定の点（例えば共形不変性を持つ点）から複数の「枝」が伸びてくるからなのではないかと想像する．
[^4]: このような基底が存在するかどうかは明らかではない．例えば$\mathcal{N}=4$理論はそのような有限な基底を持たない．
[^5]: 正確にはスーパーポテンシャルにより相互作用も指定する必要がある．
[^6]: 超対称量子力学の真空のモジュライ空間は数学的にはポテンシャル付き箙の 安定な表現のなすモジュライ空間であり，木村氏の解説に現れる箙多様体と近い関係にある．
[^7]: ヒッチン系はハイパーケーラ多様体であり$\mathbb{P}^1$でパラメーター付けされる 複素構造を持つ．
[^8]: 正確にはこのほかに$\mathcal{A}$と交換するスカラー場が存在し，それらが クーロンブランチ以外の真空のブランチを記述している[@XieYonekura]．
[^9]: 但し，$C$の穴では$\mathcal{A}$のホロノミーを指定する境界条件が課される．
[^10]: もっとも， この主張は最終的には場の理論的による直接計算によって確かめられるべきものである．この方向については例えば[@Okuda]を参照．
[^11]: ここでは三角形分割を天下りかつ組み合わせ論的に 与えたが，実際にはより物理的な説明がある：BPS状態はリーマン面$C$上の測地線で与えられ，その測地線の族を考えることで， 理想三角形分割を構成することができる． また，その三角形分割は$\zeta$の値に依存し，これを変えることで 三角形分割の変化を引き起こすことができる[@GMN]． その構成は（より厳密な設定で）本特集中の岩木氏の記事で取り扱われている．
[^12]: 筆者の場合，一般的にいって，そうした紆余曲折の後に学んだ知識の方が役立つことが多い．
[^13]: クラスター代数の数理物理における広がりをのぞくには例えば特集[@special]の論文たちを参考されたい．
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We present the top property measurements in the CDF. Most of measurements utilize close to the integrated luminosity of 3 fb$^{-1}$.'
author:
- 'Hyun Su Lee (On Behalf of the CDF collaboration)'
title: Top quark properties at CDF
---
Introduction
============
During the last decade after discovery of top quark [@r_run1topdiscCDF; @r_run1topdiscD0], top quark has been inclusively studied. By now, the mass of the top quark has been measured to be 173.1$\pm$1.3 [$\mathrm{GeV}/c^{2}$ ]{}[@masscombo] which is the most precisely measured quark mass and [$t\bar{t}$ ]{}pair production cross section has been measured as less than 10 % of uncertainties [@cross]. However, many of another top quark property have not yet been well explored due to the limited statistics. In the ongoing data taking at Fermilab’s Tevatron proton-antiproton collider with Collider Detector at Fermilab (CDF), an increasing of integrated luminosity can make us to measure the property and also discover the unexpected phenomena from top sector. We describes a few of the CDF’s progress of top quark property measurements in the following.
Top property measurements
=========================
Top quark production
--------------------
The predominant production of top quark in the Tevatron is the [$t\bar{t}$ ]{}pair production. The standard model (SM) predicts the [$t\bar{t}$ ]{}production processes to be [$q\bar{q}$ ]{}annihilation (${\ensuremath{q\bar{q}} }\rightarrow {\ensuremath{t\bar{t}} }$) and $gg$ fusion ($gg
\rightarrow {\ensuremath{t\bar{t}} }$), occurring at the Tevatron with relative fractions of $\sim$85% and $\sim$15%, respectively [@ggfrac]. A measurement of this fraction tests the SM predictions and our understanding of gluon parton distribution functions in the proton. We measure this quantity both lepton jets and dilepton final state. In the lepton jet channel, we have two different measurement based on different discriminant. One method builds distriminant using number of low-momentum track to take advantage of the higher probability for a gluon than for a quark to radiate a low-momentum gluon [@gg_lj1]. The other method builds discriminant with artificial neural net using eight variables those are sensitive to the production mechanism [@gg_lj2]. Both two measurement use [$955~\mathrm{pb}^{-1}$]{} of CDF data, and is combined by using the Feldman-Cousin prescription [@fc]. Figure \[ggfrac\_fc\] shows Feldman-Cousin bands for the combination of two analysis with 68% and 95% confidence level (C.L.) for the fraction of $gg$ to produce [$t\bar{t}$ ]{}. We measure the fraction to be $G_{f}=0.07^{+0.15}_{-0.07}$, and we find the 95% C.L. limit to be $G_{f} < 0.38$ [@gg_lj2]. This is consistent with SM.
![\[ggfrac\_fc\] Feldman-Cousins bands for the combination of the analysis for $gg$ fraction in the lepton jet channel with statistical and systematic uncertainties for 68% C.L. and 95% C.L..](ggfrac.eps){width="80mm"}
We use angle between two lepton to build discriminant for extraction of $gg$ fraction in the dilpeton channel [@gg_dil]. We build Feldman-Cousin bands and fit data using [$2.0~\mathrm{fb}^{-1}$]{} of [$p\bar{p}$ ]{}collision as you can see in Fig. \[ggfrac\_fc\_dil\]. We measure the $gg$ fraction in the dilepton channel to be $G_{f} = 0.53^{+0.36}_{-0.38}$ which is consistent with SM.
![\[ggfrac\_fc\_dil\] Feldman-Cousins bands for $gg$ fraction in the dilepton channel with statistical and systematic uncertainties for 68% C.L. and 95% C.L..](ggfrac_dil1.eps){width="80mm"}
Since new production mechanism for top quark pairs can make the shape of [$t\bar{t}$ ]{}invariant mass as resonances or general shape distortions, the generic method to search the such contribution is to compare the shape of the observed differential [$t\bar{t}$ ]{}cross section $d\sigma/dM_{{\ensuremath{t\bar{t}} }}$ with SM expectation. The mass of the top-antitop system is reconstructed for each event by combining the four vectors of the four leading jets, lepton, and missing transverse energy. The unfolding technique implemented to correct the reconstructed distribution as for direct comparison with theoretical differential cross section. In the update with [$2.7~\mathrm{fb}^{-1}$]{} data, we have in-situ jet energy scale (JES) measurement using di-jet mass of $W$ boson decay, which have been used in the top quark mass measurement [@mass], that we can significantly reduce the JES systematics. As one can see in Fig. \[dsigma\], we do not find any significant difference with SM expectation. We check the consistency using the Anderson-Darling (AD) statistics [@AD]. We calculate a p-value of 0.28 using AD statistics which have a good agreement with the SM [@cdfdsigma].
![\[dsigma\] Unfolded differential cross section of [$t\bar{t}$ ]{}invariant mass using CDF data is compared with SM prediction.](dsigmmadm.eps){width="80mm"}
We search for resonant top-antitop pair production and subsequent decay in the all-hardronic channel with [$2.8~\mathrm{fb}^{-1}$]{} of data [@mttbar]. We use the probability from per-event matrix element calculation as discriminant to reduce and control the large background from QCD-multijet as an input of neural net with other kinematic variables. We reconstruct invariant mass of [$t\bar{t}$ ]{}with matrix element technique and have consistent result with SM prediction as one can see in Fig. \[mtt\]. We then set the 95% C.L. limits on Z’ production as 805 GeV in case of a leptophobic topcolor resonance candidate.
![\[mtt\] Invariant mass of [$t\bar{t}$ ]{}from data and SM expectation.](mtt.eps){width="80mm"}
Top like new physics particle search
------------------------------------
Due to the large mass of the top quark, the super-symmetric partner of top quark (stop) can be lighter than the top quark even to be a lightest squark. For a light stop and R-parity conservation of super-symmetry particles, stop quark dominantly decay to b-quark and chargino, and chargino decay to W boson and neutralino. While neutralino can not be detected, the decay of pair produced stop quark have same final state with [$t\bar{t}$ ]{}decay. We search the pair production of stop in the dilepton final state with [$2.7~\mathrm{fb}^{-1}$]{} of data [@stops]. We reconstruct the stop mass in the underconstraint system to extract the stop components in the [$t\bar{t}$ ]{}dilepton decay events. Figure \[stop\] shows our data is compatible with SM prediction without stop, and then we set the 95% C.L. upper limit of stop in certain condition of SUSY parameter space [@stops].
![\[stop\] Reconstructed stop mass comparing data to monte carlo.](stop.eps){width="80mm"}
Top properties
--------------
One of basic quantities of top quark is the electric charge, which is expected to have a value of 2/3$e$ in the SM. However, one of exotic model have decay of top to a $W^{-}$ instead of $W^{+}$ having a charge of 4/3$e$ [@tcex]. We have a measurement using [$1.5~\mathrm{fb}^{-1}$]{} of data in the lepton jet channel. The measurement identifies the charge of the two W bosons and two b-quarks in each data event, and then determine which W bosons and b-quarks decayed from the same parent top quark. The charge of the top is then obtained by multiplying the charge of the W with the charge of the jet associated with a b-quark. Figure \[charge\] shows the measured distribution of pairs of charge product which is compatible with SM like top charge. We then exclude exotic model-like top at 87% C.L. [@tcharge].
![\[charge\] Product of the W charge and the associated jet charge for data and MC (SM signal MC distribution).](charge.eps){width="80mm"}
Top quark width have been measured with [$1~\mathrm{fb}^{-1}$]{} of data in the lepton jet channel. Main idea is to use top mass reconstruction and templates for different top width and to fit it to data. We use the top quark mass considered known as $M_{top} = {\ensuremath{175~\mathrm{GeV}/c^{2}}}$ and templates are produced for range of different top width. We extract top width from reconstructed top mass distribution compared to signal with different top width and background using unbinned likelihood fit. We then have measurement consistent with SM as one can see in Fig. \[width\], so we set a limit on top width using Feldman-Counsins [@fc] prescription to be top width($\Gamma$)$<$13.1 GeV of 95% C.L. upper limit [@twidth].
![\[width\] Likelihood fit to the data with 1-tag (up) and 2-tag (down) categories.](Data1tagfit.eps "fig:"){width="60mm"}![\[width\] Likelihood fit to the data with 1-tag (up) and 2-tag (down) categories.](Data2tagfit.eps "fig:"){width="60mm"}
Top decay
---------
The SM predicts that the top quark decays almost entirely to a W-boson and a bottom quark, and that the Wtb vertex is a V-A charged weak current interaction. A consequence of this is that the top quark is expected to decay 70.4% of the time of longitudinal and the rest to left handed polarized W-bosons [@wpred]. Any new particles involved in the same decay topologies and non-standard coupling could create a different mixture of polarized W-bosons. Therefore, a measurement of this fraction is a test of the V-A nature of the Wtb vertex. In the CDF, there are several measurements using different technique in the lepton jet channel with [$1.9~\mathrm{fb}^{-1}$]{} of data. One method builds template using $cos(\theta)$ [@wtemp], where the $\theta$ is the angle between lepton and b-quark in the W rest frame, this is sensitive to W helicity. The other method use matrix element technique [@wmat] which we calculate a likelihood for each event then product per event likelihood to build total likelihood. Figure \[whel\] shows W-helicity measurements at CDF. All of measurements are consistent with SM prediction.
![\[whel\] The summary of W-helicity measurement at CDF.](whel.eps){width="60mm"}
Several exotic physics models, such as SUSY and two Higgs doublet, predict flavor changing neutral currents (FCNC) in top decay. In the standard model, this decay mode is highly suppressed. Therefore, any signal from FCNC decay chain indicate an evidence of new physics. A search for FCNC decays has been performed at CDF with [$1.9~\mathrm{fb}^{-1}$]{}. This analysis utilizes a template fit to a mass $\chi^2$ variable constructed from kinematic constraints present in FCNC top quark decays. A simultaneous fit is performed to the data using two signal and one control region as one can see in Fig. \[fcnc\]. The control region constrains uncertainties in the shape and normalization of the templates. As one can see in this plot, our data is well explained without FCNC components and then, we set 95% C.L. upper limit on the branching fraction of ($t\rightarrow Zq$) $<$ 3.7% [@tfcnc].
![\[fcnc\] Mass $\chi^2$ distribution for signal and control regions.](money3_rebin_maxsig.eps){width="80mm"}
Charged Higgs $H^{\pm}$ bosons are predicted in supersymmetric and GUT extensions of the SM. If a charged Higgs boson in sufficiently light, it can be produced in top quark decays. In the presence of a charged Higgs boson, the $t \rightarrow H^{+}b$ decay would compete with the SM top quark decay, thereby altering the expected number of events in different final states of [$t\bar{t}$ ]{}. In the certain final state, which is low $\tan \beta$, the dominant decay of charged higgs is $H^{+}\rightarrow c\bar{s}$. We has searched for the decays in the lepton + jets events with [$2.2~\mathrm{fb}^{-1}$]{} by fully reconstructing [$t\bar{t}$ ]{}decay and exploiting the difference between the dijet mass spectra in $W
\rightarrow q\bar{q}$ and $H^{+} \rightarrow c\bar{s}$ decays [@chigg]. The invariant dijet mass spectrum in data is shown in Fig. \[chiggs\] (up) which no significant deviation from the SM is observed. Therefore we set the limits on branching fraction of $t\rightarrow
H^{+}b\rightarrow c\bar{s}b$ as one can see in Fig. \[chiggs\] (down). In this plot, we extend our search to generic charged boson search which possibly have smaller mass than W boson.
![\[chiggs\] Distribution of di-jet mass in [$t\bar{t}$ ]{}decay (up) and the upper limit on the branching fraction of $t\rightarrow H^{+}b$ at 95% C.L. as a function of charged higgs mass (down).](chigg_dijet.eps "fig:"){width="60mm"} ![\[chiggs\] Distribution of di-jet mass in [$t\bar{t}$ ]{}decay (up) and the upper limit on the branching fraction of $t\rightarrow H^{+}b$ at 95% C.L. as a function of charged higgs mass (down).](chigg_up.eps "fig:"){width="60mm"}
Conclusions
===========
Number of top quark properties not only standard model top signature but also exotic model signature have been searched and measured. However many measurements are still limited by the statistical uncertainty. Although we do not find evidence conflicting with SM top quark, we expect to have interesting measurement with more data in near future.
I would like to thank the CDF colleagues for their effort to carry out these challenging physics analysis. I also thank the conference organizers for a very rich week of physics.
[99]{} F. Abe [*et al.*]{} (CDF Collaboration), Phys. Rev. Lett. [**74**]{}, 2626 (1995).
S. Abachi [*et al*]{} (D0 Collaboration), Phys. Rev. Lett. [**74**]{}, 2632 (1995).
Tevatron EW Working Group (CDF and D0 Collaboration), CDF NOTE, [**9717**]{} (2009), (*Preprint* ).
CDF Collaboration, CDF NOTE [**9448**]{} (2009).
M. Cacciari [*et al.*]{}, J. High Energy Phys. [**0404**]{} 068 (2004).
T. Aaltonen [*et al*]{} (CDF Collaboration), Phys. Rev. D [**78**]{}, 111101 (2008).
T. Aaltonen [*et al*]{} (CDF Collaboration), Phys. Rev. D [**79**]{}, 031101 (2008).
G. J. Feldman and R. D. Cousins, Phys. Rev. D [**57**]{}, 3873 (1998).
CDF Collaboration, CDF NOTE [**9432**]{} (2008).
T. Aaltonen (CDF Collaboration), Phys. Rev. D [**79**]{}, 092005 (2009).
T. W. Anderson and D. A. Darling, The Annals of Math. Stat. [**23**]{}, 193 (1952).
T. Aaltonen (CDF Collaboration), Phys. Rev. Lett. [**102**]{}, 222003 (2009).
CDF Collaboration, CDF NOTE [**9844**]{} (2009).
CDF Collaboration, CDF NOTE [**9439**]{} (2009).
D. Chang, W. Chang, and E. Ma, Phys. Rev. D [**59**]{}, 091503 (1999).
CDF Collaboration, CDF NOTE [**8967**]{} (2008).
T. Aaltonen [*et al.*]{} (CDF Collaboration), Phys. Rev. Lett. [**102**]{}, 042001 (2009).
J. Aguilar-Saavedra [*et al.*]{}, Eur. Rev. J. C [**50**]{}, 519 (2007).
CDF Collaboration, CDF NOTE [**9215**]{} (2007).
CDF Collaboration, CDF NOTE [**9144**]{} (2007).
T. Aaltonen [*et al*]{} (CDF Collaboration), Phys. Rev. Lett. [**101**]{}, 192002 (2008).
CDF Collaboration, CDF NOTE [**9322**]{} (2008).
| {
"pile_set_name": "ArXiv"
} |
---
author:
- 'D.E. Charrier'
- 'T. Weinzierl'
bibliography:
- 'efficient-aderdg.bib'
title: 'Stop talking to me—a communication-avoiding ADER-DG realisation[^1]'
---
=1
[^1]: The underlying project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 671698 (ExaHyPE). All software is freely available from [www.exahype.eu](www.exahype.eu).
| {
"pile_set_name": "ArXiv"
} |
---
author:
- |
[^1]\
Racah Institute of Physics, The Hebrew University, 91904 Jerusalem, Israel\
E-mail:
- |
Denis Allard\
Laboratoire Astroparticule et Cosmologie, Université Paris Diderot/CNRS, 10 rue A. Domon et L. Duquet, F-75205 Paris Cedex 13, France\
E-mail:
- |
Etienne Parizot\
Laboratoire Astroparticule et Cosmologie, Université Paris Diderot/CNRS, 10 rue A. Domon et L. Duquet, F-75205 Paris Cedex 13, France\
E-mail:
- |
Tsvi Piran\
Racah Institute of Physics, The Hebrew University, 91904 Jerusalem, Israel\
E-mail:
title: 'PROBING THE EXTRAGALACTIC COSMIC RAYS ORIGIN WITH GAMMA-RAY AND NEUTRINO BACKGROUNDS'
---
Introduction {#sec:intro}
============
The interaction of ultra-high-energy cosmic-rays (UHECRs) with the photon backgrounds during their propagation in intergalactic space produces cosmogenic $\gamma$-ray photons and neutrinos ($\nu$s). The flux of these secondary messengers is highly sensitive to the spectral shape, maximal energy, composition and cosmological evolution of the UHECR sources, and therefore, one can derive important constraints on the UHECR origin from a multi-messenger approach that takes these into account ; .
The recent Fermi-LAT data [@2015ApJ...799...86A], together with statistics of the photon counts in the skymap pixels [e.g. @Malyshev11 and references therein] have enabled different authors [@Ack2016; @Zechlin16 hereafter A16 and Z16] to estimate the flux contributed by point sources (PS) well below the Fermi-LAT detection limits. These studies show that resolved and unresolved PS account for the majority of the EGB. Since a $\gamma$-ray background due to extragalactic cosmic rays (EGCRs) is unavoidable, it is crucial to verify that the proposed UHECR source models do not violate the existing constraints.
Recent measurements by the Pierre Auger Observatory (Auger) indicate that the composition of UHECRs is mixed (predominantly light) at the ankle of the cosmic-ray spectrum, and it gets progressively heavier as the energy increases . This composition trend can be interpreted as the signature of a low maximal energy-per-unit-charge ($E_{\rm max}/Z\lesssim 10^{19}$ eV) of the nuclei accelerated at the dominant sources of UHECRs. Below $10^{18}$ eV, the KASCADE-Grande experiment reported an ankle-like feature in the energy spectrum of light (proton-helium) elements with a break at $\sim 10^{17}$ eV [@Apel13; @Bertaina15]. This “light ankle” can be naturally understood as the emergence of a light EGCR component, taking over the steeper Galactic cosmic-ray (GCR) component.
We discuss here the viability of a class of mixed-composition models in which the KASCADE-Grande and Auger data are understood in terms of a transition between a GCR component and a single EGCR component with a soft proton spectrum and low $E_{\rm max}$. This soft proton component, responsible for the KASCADE-Grande light ankle, would be the dominant contributor to the cosmogenic $\gamma$-ray flux. This model was shown to be compatible with the spectrum and composition data at all energies [@Globus2015a; @Globus2015b hereafter G15a, G15b], and it is consistent with the anisotropy constraints on galactic protons [@Tinyakov16].
\[sec:model\]Source model and propagated cosmic-ray spectrum
============================================================
Any phenomenological EGCR model that account for the data needs a very hard spectrum at the sources, to reproduce the evolution of the composition above the ankle observed by Auger, and a softer proton component, to account for the light ankle seen by KASCADE-Grande.
![Left panel: UHECR injection spectrum for the various nuclei, as obtained in G15a, with the fit of the proton component with spectral index $\beta = 2.0$ (dashed blue), and with its modified shape in the case of $\beta=2.5$. Right panel: Propagated UHECR spectra for $2.0\leq\beta\leq2.5$, compared to KASCADE-Grande and Auger data, for a GRB-like cosmological evolution (blue lines) or non-evolving sources (violet shaded area). The total (GCR+EGCR) light component is compared to that deduced from KASCADE-Grande data (using the EPOS-LHC [@EPOS1] hadronic model), for GRB-like evolution with $\beta=2.4$ and 2.5 (dashed lines).[]{data-label="fig:modelc"}](cr.pdf "fig:"){width="0.46\linewidth"} ![Left panel: UHECR injection spectrum for the various nuclei, as obtained in G15a, with the fit of the proton component with spectral index $\beta = 2.0$ (dashed blue), and with its modified shape in the case of $\beta=2.5$. Right panel: Propagated UHECR spectra for $2.0\leq\beta\leq2.5$, compared to KASCADE-Grande and Auger data, for a GRB-like cosmological evolution (blue lines) or non-evolving sources (violet shaded area). The total (GCR+EGCR) light component is compared to that deduced from KASCADE-Grande data (using the EPOS-LHC [@EPOS1] hadronic model), for GRB-like evolution with $\beta=2.4$ and 2.5 (dashed lines).[]{data-label="fig:modelc"}](cr-lightankle.pdf "fig:"){width="0.46\linewidth"}
The effective spectrum from the mixed composition model (G15b) is displayed in the left panel of Fig. \[fig:modelc\]. The much softer source spectrum for the nucleons is due to the free escape of neutrons produced by the photo-disintegration of nuclei at the source, and its exact shape depends on various physical parameters (see G15a for details). Since the extragalactic protons around $10^{17}$ eV contribute significantly to the expected cosmogenic $\gamma$-ray flux in the Fermi energy range, we explore, for the sake of generality, (i) different slopes for the proton component $\beta$ (as could result from different physical parameters describing the sources) while keeping the same maximal rigidity and spectral shape for heavier nuclei; (ii) different cosmological evolutions, assuming an average source power proportional to $(1+z)^{\alpha}$ up to a redshift $z_{\max}$. We consider a range of spectral indices $2.0\leq\beta\leq2.5$. The two proton spectra with the extreme values of $\beta$ are represented by thick dashed and dotted blue lines, respectively. The implied range of UHECR emissivities above $10^{17}$ eV is $L_{\rm CR}^{17}\sim[5.7-14]\cdot10^{44}\,\mathrm{erg}\,\,\mathrm{Mpc}^{-3}\,\, \mathrm{yr}^{-1}$.
The right panel of Fig. \[fig:modelc\] depicts the *propagated* UHECR spectra for $2.0\leq\beta\leq2.5$, for EGCR sources evolving as GRBs [@2010MNRAS.406.1944W blue lines] and for non evolving sources (violet shaded area). In the case of a GRB-like or SFR-like cosmological evolutions, proton spectral indices $\beta \simeq 2.4-2.5$ provide a good fit to the KASCADE-Grande data when summing the light EGCR component with the GCR light component obtained in G15b (dashed line in the right panel of Fig. \[fig:modelc\]). Softer proton indices are required in the case of a non-evolving scenario.
\[t\] ![ Left panel: $\gamma$-ray fluxes from EGCRs (dashed-dotted lines), for GRB-like evolution (blue) and non-evolving (violet) sources, as computed with our mixed-composition model and spectral indices of the soft proton component $2.0\leq\beta\leq2.5$. Also represented the $\gamma$-ray fluxes from SFG, MisAGN and blazar sources (see labels) as modelled by [@Ack2012; @Inoue11; @Ajello15] respectively. The corresponding sum of UHECR, SFG, misAGN and blazar components is represented by thick solid lines [or with a dotted line when 1-$\sigma$ lower bound are adopted for the SFG+misAGN+blazar model, see @Ajello15], and compared to the EGB estimated from Fermi-LAT data, for both foreground models A and B. Right upper panel: cosmogenic $\nu$ fluxes associated with our mixed-composition scenarios in the case of GRB-like evolution (blue) and non-evolving (violet) sources, compared with the current IceCube sensitivity [@Aartsen16] and the expected sensitivities of ARIANNA (5 years, 50 MHz option, [@Hallgren16]) and GRAND (3 years, [@Martineau15]). Right lower panel: same, i) for 100% proton scenarios compatible with the Fermi constraints (plain lines, same colour code), and ii) for a sub-dominant proton component (contributing 5% of the UHECRs at 10 EeV) evolving as FR-II galaxies (dashed-dotted line) [@Wall05]. []{data-label="fig:EGB"}](fig_egrb.pdf "fig:"){width="0.42\linewidth"} ![ Left panel: $\gamma$-ray fluxes from EGCRs (dashed-dotted lines), for GRB-like evolution (blue) and non-evolving (violet) sources, as computed with our mixed-composition model and spectral indices of the soft proton component $2.0\leq\beta\leq2.5$. Also represented the $\gamma$-ray fluxes from SFG, MisAGN and blazar sources (see labels) as modelled by [@Ack2012; @Inoue11; @Ajello15] respectively. The corresponding sum of UHECR, SFG, misAGN and blazar components is represented by thick solid lines [or with a dotted line when 1-$\sigma$ lower bound are adopted for the SFG+misAGN+blazar model, see @Ajello15], and compared to the EGB estimated from Fermi-LAT data, for both foreground models A and B. Right upper panel: cosmogenic $\nu$ fluxes associated with our mixed-composition scenarios in the case of GRB-like evolution (blue) and non-evolving (violet) sources, compared with the current IceCube sensitivity [@Aartsen16] and the expected sensitivities of ARIANNA (5 years, 50 MHz option, [@Hallgren16]) and GRAND (3 years, [@Martineau15]). Right lower panel: same, i) for 100% proton scenarios compatible with the Fermi constraints (plain lines, same colour code), and ii) for a sub-dominant proton component (contributing 5% of the UHECRs at 10 EeV) evolving as FR-II galaxies (dashed-dotted line) [@Wall05]. []{data-label="fig:EGB"}](fig_nu.pdf "fig:"){width="0.42\linewidth"}
Gamma-ray and neutrinos counterparts {#sec:gamma}
====================================
The interactions of the propagating EGCRs leads to the production of cosmogenic $\gamma$-rays in the GeV-TeV range, and $\nu$s in the PeV-EeV range, through the development of electromagnetic cascades. The Monte-Carlo procedure used to calculate the cosmic-ray, $\nu$ and $\gamma$-ray spectra is presented in .
The cosmogenic $\gamma$-rays spectra corresponding to the UHECR spectra of Fig. \[fig:modelc\], are shown in Fig. \[fig:EGB\] for a mixed-composition model with proton spectral indices $2.0\leq\beta\leq2.5$, for sources with no cosmological evolution (violet lines) and with a GRB-like evolution (in blue). These $\gamma$-ray fluxes represent only a small contribution to the total EGB, which is reproduced from @2015ApJ...799...86A for two different models of the Galactic $\gamma$-ray foreground, referred to as model A and model B by the authors. These two models roughly differ by $\sim 20-30\%$, which can be seen as a rough estimate of their systematics in the subtraction process.
To determine whether a given EGCR source model is compatible with the $\gamma$-ray data, we need to take into account other known contributions to the EGB. The contribution of star-forming galaxies (SFG) and misaligned active galactic nuclei (misAGN), based on the models by @Inoue11 and @Ack2012, are shown in the left panel of Fig. \[fig:EGB\] (omitting the uncertainty bands for clarity). Also shown is the $\gamma$-ray spectrum arising from blazars, adapted from @Ajello15, which appears in good agreement with the PS contribution estimated by A16 and Z16 over the whole energy range. We find that, for the SFR and weaker evolving scenarios, the sum of all components (UHECR, misAGN, SFG and blazars) never exceeds the total EGB, in the case of model B. In the case of model A, the sum is above the EGB. However, it falls below it if one adopts the $1\sigma$ lower bound on the misAGN+SFG+blazars contribution (see @Globus2017 for the estimates of the fluxes of the different contributions in the same energy bands as in A16 and Z16).
The right panel of Fig. \[fig:EGB\] shows the resulting $\nu$ spectra for different EGCR models, together with the sensitivity of current and planned experiments. The mixed-composition models predict $\nu$ fluxes too low to be detected by IceCube [@Aartsen16] or ARIANNA [@Hallgren16], even in the case of a GRB-like cosmological evolution. They would require a sensitivity such as that expected for the GRAND observatory [@Martineau15] or CHANT satellite concept [@Neronov16]. Pure proton scenarios can be seen on Fig. \[fig:EGB\] to yield detectable fluxes, while still being allowed by the current IceCube limits and Fermi-LAT data. (For these calculations, we assumed a pure proton $E^{-2}$ spectrum with an exponential cutoff at $E_{\rm max} = 60$ EeV, which is known to reproduce reasonably well the Auger spectrum above the ankle). It is interesting to note that the $\nu$s that could be produced by a hypothetical subdominant EGCR proton sources, with large enough $E_{\rm max}$ and cosmological evolution, would contribute a detectable $\nu$ flux around $10^{18}$ eV, thus making EeV $\nu$s a powerful probe for revealing the existence of trans-GZK proton accelerators, even if they do not dominate the observed UHECR flux.
![Fermi-LAT constraints on EGCR source evolution in the case of our mixed-composition scenario and proton index $\beta=2.5$. The different colors show the percentage of the sum of all components (UHECR+PS+misAGN+SFG) to the EGB (Models A and B) in the 10.4–50 GeV energy band, in the ($\alpha$, $z_{\max}$) parameter space, where $z_{\max}$ is the maximum redshift up to which sources experience a cosmological evolution in $(1+z)^{\alpha}$. Some possible EGCR sources [see e.g. @2016ApJ...822...56G for the references to the cosmological evolutions] are shown. GRB: gamma-ray bursts. SFR: star-formation rate. MLLAGN: Medium-low-luminosity AGNs. MHLAGN: Medium-High-Luminosity AGNs. HLAGN: High Luminosity AGNs. []{data-label="fig:summary"}](summary.pdf){width="0.65\linewidth"}
Conclusions
===========
The UHECR model considered in G15b gives a coherent picture of the GCR-to-EGCR transition, and appears to be compatible with the Fermi-LAT measurements and the estimates of the PS contributions by A16 and Z16. The mixed-composition model appear to be less constrained by the Fermi-LAT than the electron-positron dip (pure-proton) scenario [@Bere16; @Supanitsky16; @2016ApJ...822...56G] that rules out SFR-like and stronger cosmological evolutions (see also [@Heinze16] for more radical conclusions on the dip model). Our results are summarized in Fig. \[fig:summary\], that shows the allowed parameter space of different evolutionary scenarios for the mixed composition model. This estimate is based on the summed contribution of all components in the 10.4-50 GeV band, where the contribution from UHECRs is the largest. Only very strong evolutions, e.g. similar to the very luminous AGNs, are excluded by the current observations. For the evolutionary models allowed by Fermi, the $\nu$s fluxes above $10^{17}$ eV associated with the mixed-composition scenario are well below the current IceCube limits. These fluxes are within the reach only of the most sensitive planed $\nu$ observatories.
Finally, we note that while the PS contributions are now understood to dominate the extragalactic $\gamma$-ray fluxes in the GeV-TeV range, the uncertainties on the different contributions [notably for sources other than blazars, see e.g @DiMauro13; @Lacki14; @Tamborra14] as well as on the Galactic foreground are still too large to efficiently constrain the cosmological evolution of UHECR sources. Since the $\gamma$-ray fluxes associated with mixed-composition UHECRs never exceed $\sim 20$% of the EGB (at least for source evolutions not significantly larger than SFR, see Table 2 in @Globus2017), the EGB and its other contributions should be [ determined ]{} to this level of precision in order to estimate whether a UHECR mixed-composition model is excluded. Moreover, the Fermi-LAT estimates of the Galactic foreground are based on the GALPROP framework [@Strong00]. These calculations rely on several simplifying assumptions in particular in the description of the Galactic cosmic-ray source distribution or the magnetic halo, as well as on several [*ad-hoc*]{} parameters that are tuned to reproduce cosmic-ray data. Alternative models [e.g. @Nava17 and references therein] have been shown to fairly account for the primary-to-secondary ratios as well as some puzzling features in the observed $\gamma$-ray Galactic signal. These models have a smaller halo extension and would probably result in a lower Galactic foreground, leaving more room for EGCR contributions.
NG and TP acknowledges the I-CORE Program of the Planning and Budgeting Committee and The Israel Science Foundation (grant 1829/12), the advanced ERC grant TReX, and the Lady Davis foundation.
[99]{} Aab A. et al. (Pierre Auger Collaboration), 2014, Phys. Rev. D 90, 122006 Aartsen, M. G. et al. \[IceCube collaboration\], 2016, Phys. Rev. Lett. 117 (24), 241101, arXiv:1607.05886 \[astro-ph.HE\] Ackermann M. e. a., 2015, ApJ, 799, 86 Ackermann M. e. a., 2016, Phys Rev Letters,116, 151105 Ackermann, M., Ajello, M., Allafort, A., et al. 2012b, ApJ, 755, 164 Ahlers M., Anchordoqui L. A. and Sarkar S., 2009, Phys. Rev. D 79(8), 083009. Ahlers M., Salvado J., 2011, Phys.Rev. D84, 085019 Ajello, M., Gasparrini, D., Sánchez-Conde, M. , Zaharijas, G., Gustafsson M., et al., 2015, ApJ Letters, 800:L27 Allard D., Ave M., Busca N., Malkan M. A., Olinto A. V., Parizot E., Stecker F. W. and Yamamoto T., 2006, Journal of Cosmology and Astro-Particle Physics 9, 5. Anchordoqui L. A. , Goldberg H., Hooper D., Sarkar S., Taylor A., 2007, Phys. Rev. D, vol. 76 pp. 123008 W. Apel et al. (KASCADE-Grande Collaboration), Nucl. Instrum. Methods Phys. Res., Sect. A 620, 202 (2010). W. Apel et al. (KASCADE-Grande Collaboration),Phys. Rev. D 87 081101 (2013). Berezinsky V. S., Gazizov A. & Kalashev O., 2016, Astropart. Phys. 84, 52 M. Bertaina et al., 2015, proc. of the 34th Int. Cosmic Ray Conf., The Hague, The Netherlands Coppi, P. S., Aharonian, F. A., 1997, ApJ, 487, 9 Decerprit G., Allard D., 2011, A&A, 535, A66 Di Mauro, M., Calore, F., Donato, F., Ajello, M., & Latronico, L. 2013, ApJ, 780, 161 Engel R., Seckel D. and Stanev T., 2001, Phys. Rev. D 64(9), 093010. E., [Eichler]{} D., 2016, ApJ, [822, 56]{} N., [Allard]{} D., [Mochkovitch]{} R., [Parizot]{} E., 2015, MNRAS, [451, 751]{} N., [Allard]{} D., [Parizot]{} E., 2015a, Phys Rev D, 92, 021302 N., [Allard]{} D., [Parizot]{} E., Piran, T., 2017, ApJ, 839, 22 Hallgren A. for the ARIANNA collaboration, 2016, EPJ Web of Conferences, 116 03003 Heinze J., Boncioli D., Bustamante M., Winter W., 2016, ApJ, 825, 122 Inoue, Y. 2011, ApJ, 733, 66 Kotera K., Allard D. and Olinto A. V., 2010, Journal of Cosmology and Astroparticle Physics, vol. 10 pp. 013 Lacki, B. C., Horiuchi, S., & Beacom, J. F. 2014, ApJ, 786, 40 Malyshev D. & Hogg D. W., 2011, ApJ 738, 181 Martineau-Huynh O. et al., 2015, Proceedings of The 34th International Cosmic Ray Conference, The Hague, The Netherlands, arXiv:1508.01919 Nava, L., Benyamin, D., Piran, T. & Shaviv, N. J., 2017, MNRAS, 466, 3674 Neronov A. et al., 2016, arXiv:1606.03629 Protheroe R. J and Johnson P. A., 1996, Astroparticle Physics, vol. 4 pp. 253 Seckel D. and Stanev T., 2005, Physical Review Letters 95(14), 141101 Strong, A. W., Moskalenko, I. V., & Reimer, O., 2000, ApJ, 537, 763 Stecker F. W., 1979 ApJ 228, 919 Supanitsky A. D., 2016, Phys. Rev. D 94, 063002 Tamborra, I., Ando, S., & Murase, K. 2014, J. Cosmology Astropart. Phys., 9, 43 Tinyakov, P. G., Urban, F. R., Ivanov, D., Thomson, G. B., Tirone, A. H., 2016, MNRAS, 460, 3479 Wall J. V., Jackson C. A., Shaver P. A., Hook I. M. and Kellermann K. I., 2005, A&A 434, 133 Wanderman D., Piran T., 2010, MNRAS, 406, 1944 K. Werner, M. F. Liu &T. Pierog, 2006, Phys. Rev. C74 044902 Zechlin, H-S et al., 2016, ApJ Letters, [826, L31]{}
[^1]: **Oral presentation CRI184 scheduled on Tue., July 18.**
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We describe a cryogenic cavity-optomechanical system that combines Si$_3$N$_4$ membranes with a mechanically-rigid Fabry-Perot cavity. The extremely high quality-factor frequency products of the membranes allow us to cool a MHz mechanical mode to a phonon occupation of $\bar{n} < 10$, starting at a bath temperature of 5 kelvin. We show that even at cold temperatures thermally-occupied mechanical modes of the cavity elements can be a limitation, and we discuss methods to reduce these effects sufficiently to achieve ground state cooling. This promising new platform should have versatile uses for hybrid devices and searches for radiation pressure shot noise.'
address: 'JILA, University of Colorado and National Institute of Standards and Technology, and Department of Physics, University of Colorado, Boulder, Colorado 80309, USA'
author:
- 'T. P. Purdy, R. W. Peterson, P.-L. Yu, and C. A. Regal'
title: 'Cavity optomechanics with Si$_3$N$_4$ membranes at cryogenic temperatures'
---
Introduction
============
For decades the mechanical effects of light have been used to coax gas-phase atoms towards their quantum mechanical ground state of motion. Recently, experimenters in the field of cavity optomechanics have learned how to extend the mechanical effects of light to more massive mesoscale objects. A variety of microfabricated devices, with integrated optical and mechanical or electrical and mechanical resonators, have now been employed to backaction cool mechanical objects to near their ground state: Successful devices use lithographic electrical circuits [@Teufel2011; @Rocheleau2010; @Massel2012], nanoscale optical-mechanical silicon resonators [@Chan2011], or whispering gallery mode resonators [@Verhagen2012]. These experiments combine cryogenic cooling of the mechanical object with backaction cooling of a specific mechanical mode by harnessing radiation pressure within a cavity.
Parallel attempts have been made to cool mirrors or other dielectric objects to their ground state, and detect motion at quantum limits, within a canonical high-finesse Fabry-Perot cavity [@Tittonen1999a; @Metzger2004; @Arcizet2006b; @Corbitt2007; @Thompson2008; @Groblacher2009b; @Yang2011]. These devices are generally characterized by lower frequency, higher quality factor mechanics, larger-mass, and power handling capabilites closer to macroscopic cavities. Low-frequency mechanics combined with high quality factors enable longer absolute coherence times, and hence prospects for long-lived quantum memory elements. A larger motional mass is important for searches for gravitationally induced quantum collapse [@Marshall2003]. Another frontier in optomechanical systems is the realization of a quantum-limited continuous position measurement of a macroscopic object [@Caves1980a; @Braginsky1992; @Jacobs1994; @Heidmann1997; @Murch2008; @Yamamoto2010; @Zwickl2011]. This interest is in part motivated by large interferometers used for gravitational wave searches that will soon be limited by so-called radiation pressure shot noise [@Caves1980a; @Braginsky1992]. Thus far, operating optomechanical systems at high power has been one limitation to observation of radiation pressure shot noise, and Fabry-Perot cavities comprised of super mirrors and low-optical-loss mechanics are an excellent candidate for achieving the required intensities. However, despite the promising combination of optical and mechanical properties Fabry-Perot systems afford, solid-state mechanical resonators in Fabry-Perot cavities have not yet entered the quantum regime. One reason is that non-integrated cavities are difficult to make cryogenically compatible while maintaining optical alignment and vibrational stability.
Fabry-Perot cavities are also particularly amenable to cooling high-tension, planar membrane mechanical objects [@Thompson2008], which offer a promising route for creating hybrid systems for quantum information. In particular, there is considerable interest in combining optomechanical and electromechanical systems to realize mechanically mediated quantum state transfer between microwave and optical photons [@Regal2011; @SafaviNaeini2011; @Taylor2011; @Yu2012; @Teufel2011]. One of the successful platforms for electromechanics is a 10 MHz membrane coupled to a LC resonator, and hence one possible electro-optical transducer is a system where electrical and optical cavities are parametrically coupled to the same mechanical membrane resonator. However, no membrane mechanical system, and in fact no mechanical object with a frequency below $\sim70$ MHz, has been brought to the quantum regime with an optical platform [@Chan2011; @Verhagen2012].
Here we describe the details of a robust three-component Fabry-Perot cavity at cryogenic temperatures that incorporates silicon nitride membrane microresonators [@Thompson2008]. Our simple and near-monolithic design rigidly attaches the cavity mirrors and the membrane to a common base while maintaining the stringent alignment requirements of a high-finesse cavity [@Purdy2012]. With cryogenic pre-cooling, we can harness the extremely high frequency quality-factor products seen in higher-order modes of Si$_3$N$_4$ membranes [@Zwickl2007; @Wilson2009; @WilsonRae2011; @Yu2012], allowing, in-principle, ground state optomechanical cooling. By laser cooling this device in a $^4$He flow cryostat at 5 kelvin, we achieve cooling of a few MHz mechanical mode to the lowest occupations yet achieved for an oscillator with this low a frequency, $\bar{n} < 10$. The limitation to cooling is other background mechanical modes of the cavity structure; we discuss how to understand the effects of these modes based upon the transmitted intensity spectrum and future routes to eliminating these modes.
![(a) Membrane-at-the-end cavity and simultaneous image of the intracavity optical mode and the Si$_3$N$_4$ membrane. The nodes of the (4,4) mechanical mode are indicated in blue. (b) Device design. Along the axial direction of the cavity from left to right are: Piezoelectric actuator (yellow), flat cavity mirror (transparent), membrane chip, silicon holder (green) with piezoelectric actuator (yellow), and curved cavity mirror (not visible).[]{data-label="fig1a"}](fig1a.eps)
Device and experiments
======================
Our cavity consists of two superpolished fused silica substrates with a high-reflectivity coating. The 100 ppm throughput mirrors can have down to a few ppm scattering-absorption, creating a cavity with a finesse up to 31,000. One mirror has a radius of curvature of 5 cm, and the other mirror is flat. For the work presented here, the mirrors are placed $L=5.1$ mm apart and the membrane is placed 0.9 mm from the flat mirror. This creates a “membrane-at-the-end” cavity as shown in Fig. \[fig1a\](a). An invar spacer connects the mirrors, the membrane, and the piezoelectric elements that translate the membrane and one mirror along the cavity axis (Fig. \[fig1a\](b)). In contrast to early devices [@Purdy2012], the cavity design presented here allows direct access to the membrane; by pulling the entire central metal section vertically out of the cavity, the membrane can be switched out without disturbing the high-finesse cavity mirrors.
The cavity is assembled by aligning the elements at room temperature using an optical signal and then epoxying each piece in place in turn. We align both cavity mirrors to retroreflect a single fixed laser beam. The membrane is then inserted into the cavity, held temporarily on a 5-axis micrometer stage. The transverse alignment of the membrane is assessed by in-situ imaging of both the optical mode spot, formed by a 1064 nm laser source, and the membrane, illuminated with an LED source at 940 nm where the mirror reflectivity is low (Fig.\[fig1a\](a)). Any tilt in the membrane plane relative to the optical axis leads to a displacement and distortion of the optical mode spot, which also varies with the location of the membrane along the optical standing wave. The membrane tilt is adjusted so the location and shape of the optical mode spot are not perturbed as the membrane travels over an optical wavelength.
The cavity must be constructed in such a way to maintain alignment upon cooling the device down to cryogenic temperatures. To this end, we employ mainly low thermal expansion materials such as invar, fused silica, and the popular cryogenic epoxy Stycast 2850FT. Further, a key element of the construction is symmetric, uniform thickness epoxy joints that do not result in relative tilt of the elements due to differential coefficients of thermal expansion, nor in excess localized mechanical stress build-up.
The membrane inserted into the cavity is a square $d=500$ $\mu$m Si$_3$N$_4$ membrane from Norcada Inc. We use $t=40$ nm thick membranes, as measured directly via ellipsometry. The membrane is suspended on a 5 mm square silicon chip with a thickness of 500 $\mu$m. We focus on cooling either the ($m$,$n$)=(2,2) or (4,4) drum modes (where $m$, $n$ are the number of antinodes in $x$, $y$) of the membrane. These modes have resonant frequencies of 1.6 MHz and 3.2 MHz and square mode sizes of 250 $\mu$m and 125 $\mu$m respectively. These sizes can be compared to the cavity mode, which is typically measured to have an intensity profile with a $1/e^2$ diameter of 180 $\mu$m at the membrane position. The mechanical mode frequencies are sufficiently large to theoretically cool to $\bar{n}=0.02$ given the sideband resolution, assuming the full finesse of 31,000, even for the 1.6 MHz mode. The silicon chip is mounted to holder also made from silicon. A compromise between high-$Q_m$ and mechanical stability is achieved by attaching the membrane chip only at three corners using Stycast 2850FT. The mechanical damping rate is measured in-situ by observing the mechanical ringdown lifetime with an optical probe at a wavelength where the cavity has low finesse. The realized values of $Q_m$ vary between $10^6$ and $10^7$ at 5 kelvin depending on chip mounting and membrane cleanliness. The matched thermal expansion of the chip and holder lowers the stress on the small epoxy joints. However, after thermal cycling from room temperature to cryogenic a few times, the epoxy joints are weakened and eventually the angular alignment of the membrane is compromised.
The entire device is cooled using a $^4$He flow cryostat from Advanced Research Systems Inc. Additional custom radiation shielding, including cold windows, is necessary to thermalize the membrane to the temperature of the copper cold finger (Fig. \[fig1b\]). Alignment is monitored upon cooling the device by assessing the finesse of the cavity and the consistency of the optical mode position as a function of the membrane translation within the cavity standing wave. This analysis shows that we typically maintain sub-milliradian tilts between the components of the cavity.
![Experiment layout. At the top is the laser and associated filter cavity. The three tunable acousto-optical modulators (AOMs) are for: The high-frequency branch of the cavity-laser lock, detuning the cooling laser frequency with respect to the locking laser frequency, and setting and maintaining the intensity of each beam. The locking laser is sent through an in-fiber electro-optical modulator (EOM) that applies frequency sidebands used for implementing a Pound-Drever-Hall lock. Multiple beams are combined and sent directly via free-space into the cryogenic cavity.[]{data-label="fig1b"}](fig1b.eps)
The cooling and probing of the membrane motion uses a 1064 nm Nd:YAG laser from Innolight (Fig. \[fig1b\](a)). While already a low-noise source, we additionally pass the light through a filter cavity (40 kHz linewidth) to remove intensity and frequency noise at our MHz frequencies of interest. Significant laser-cooling of our mechanical resonators requires careful attention to laser noise [@Yang2011]. For a sense of scale, laser frequency noise of $1~\mathrm{Hz/\sqrt{Hz}}$ will result in a contribution in the output spectrum equivalent to a typical mechanical signal near the ground state. A single tone red-detuned from the cavity resonance is used for both cooling and detection via monitoring the transmitted intensity directly on a photodetector. We also make use of the orthogonally polarized mode for locking the relative frequency of the laser and the cavity. At low frequencies we use the piezoelectric actuator attached to the mirror to stabilize the overall optical path length inside the cavity. High frequency noise is eliminated by servoing the laser frequency with an AOM. The overall bandwidth is 100 kHz, and aggressive filtering is used to eliminate response at the membrane mechanical resonance frequencies. Light from the two polarization modes is combined and split before and after the cavity on polarizing beam-splitters with cross-coupling of less than $10^{-3}$.
With the membrane near the end of the cavity (Fig. \[fig1a\](a)), the coupling between the membrane motion and the cavity resonance is more complicated than a membrane in the center of the cavity. The cavity resonance frequency, linewidth, and input/output port asymmetry all depend upon the placement of the membrane within the cavity standing wave, even without absorptive loss. These phenomena have been extensively numerically modeled in the work of Refs. [@Wilson2009; @Wilson2011]. A membrane placed near an end mirror can in fact have somewhat larger coupling than a membrane placed near the middle. However, operating at positions of enhanced coupling results in degraded finesse; hence, in our experiments we actually operate at the local coupling maxima where $\kappa$ is minimized.
Results
=======
In this manuscript we present results from two different cavity devices. The first data shown in Fig. \[fig2\](a) uses a device optimized for cooling the (2,2) mode. As we increase the red-detuned input optical power, the mode is cooled and damped; the data show just the last amount of cooling. The mechanical quality factor is the relatively large value of $Q_m=13.6\times10^6$, and hence the mode is damped from an initial linewidth $\Gamma_m=116$ mHz to an optically damped $\Gamma=5.5$ kHz (a factor of 47,000).
![(a) Displacement spectrum inferred from cavity transmission near the (2,2) mechanical mode. Each curve is taken with a different cooling laser power, with cavity photon occupation ranging from $3 \times 10^5$ to $6\times10^6$ from top to bottom. For this device we operate with a cavity linewidth of $\kappa/2\pi=1.2$ MHz and a detuning of $\Delta/2\pi=-1.6$ MHz. (b) Relative transmitted intensity spectrum corresponding to the same data as in (a). The horizontal range has been widened to show the large peak on the left stemming from a mechanical mode of the cavity support structure. Note, detector noise that is a dominant contributor for the low power used in this dataset is not subtracted. Conclusions about phonon occupation for this device are displayed in Fig. \[fig4\].[]{data-label="fig2"}](fig2.eps)
If we expand the frequency range of the plot we see, however, that the overall relative transmitted intensity spectrum is not white (Fig. \[fig2\](b)). In these data there is a large peak to the left of the thermal peak of the (2,2) membrane mode. We attribute this peak to a thermally occupied mode of a part of the cavity mirrors/coatings or mounting structure (which we will refer to as “cavity mechanical modes"). Figure \[fig3\](a) shows a similar set of cooling curves for a second device focusing on (4,4) mode cooling. Here we see the cavity mechanical modes appear mainly as a weakly modulated spectrum at the highest optical powers (black curve). Similar features are evident near the (2,2) mode if a higher laser power is used to attain a lower shot noise level. We suspect the lower-$Q$ modes in Fig. \[fig3\](a) arise from the fused silica mirror substrates based upon previous measurements in empty high-finesse cavities with similar mirrors [@Wilson2011; @Zhao2012], where at room temperature the modes were found to have a quality factor of 700. Particularly high-$Q$ modes, such as the mode in Fig. \[fig2\] ($Q_m>20,000$) we suspect may be from the single-crystal silicon substrate holding the membrane; our simulations of the free modes of the silicon substrate indicate there are indeed coupled modes around 1.6 MHz. In general, we have seen a variety of different cavity mechanical spectra depending on, for instance, substrate or mirror mounting techniques, coupling of the membrane motion compared to the cavity end mirrors, or the membrane mechanical quality factor.
![ (a) Cavity cooling data for the (4,4) mode device. For this device we operate with a cavity linewidth of $\kappa/2\pi=1.4$ MHz and a detuning of $\Delta/2\pi=-2.8$ MHz. As discussed in the text, based upon a calculation that includes the cavity noise, for these data we reach a minimum phonon occupation of $\bar{n}=6$; note a simple conversion between the intensity spectrum and motion (Eqn. \[eqn1\]) yields $\bar{n}=6$ for the light blue curve (8th from the top). Detector noise is not a dominant source here, but the detection efficiency was low due to loss in the detection path. The small spike at 3.145 MHz in the lowest (black) dataset is electronic in origin. (b) Integrated mechanical response as a function of cooling power and cryostat temperature (points), along with fits to the data (lines). The three data sets correspond to cryostat thermometer readings of 4.9 (blue), 10 (green), and 15 (red) kelvin. The final point on the 4.9 kelvin data represents the light blue curve in (a), i.e. corresponds to $\bar{n}=6$ for a naive conversion. (c) Bath temperature extracted from the effective temperature in (b) as a function of cryostat thermometer temperature. The vertical error bars are the uncertainty extracted from the uncertainty in $G$ and mass, and the horizontal error bars represent the systematic uncertainty of the thermometer.[]{data-label="fig3"}](fig3.eps)
The first step in the analysis of our data is a calibration of the membrane motion and an understanding of the membrane’s bath temperature. We can extract the membrane coupling in [*three*]{} different ways: (1) Using thermally driven motion as a known displacement standard, (2) careful measurement of the cavity and membrane geometry including knowledge of membrane’s position within the optical standing wave mode and a corresponding model of the expected coupling, and (3) the optical damping observed for a given intracavity photon number. In the Appendix we compare these models for the (2,2) mode data and find good agreement. The coupling for the (2,2) mode is found to be $G/2\pi=1.9\times10^{16}$ Hz/m compared to an end-mirror coupling of $\omega_c/2\pi L=5.5\times10^{16}$ Hz/m. The (4,4) mode data presented here is not a particularly well-coupled device due to the final optical and mechanical mode overlap; for this device $G/2\pi=3.9\times10^{15}$ Hz/m.
To verify that the membrane is thermalizing to the expected cryostat temperatures, we look at Fig. \[fig3\](b) that compiles data from the (4,4) device. We plot the integrated mechanical response from a Lorentzian fit to the data as a function of the mechanical damping (proportional to intracavity photon number) for three different cryostat temperatures. Here we plot only the larger response data where effects of cavity mechanical mode noise can be neglected. In Fig. \[fig3\](c) we show the extracted bath temperature from these three datasets as a function of the measured cryostat temperature (taking into account slight variations in mechanical and cavity parameters with the changing bath temperature). We find a linear trend that extrapolates to zero within the uncertainty, indicating that the membrane mechanical mode does indeed equilibrate to the cryostat cold finger temperature.
To start we can use our extracted coupling $G$ to apply a naive conversion between the transmitted intensity spectrum and the mechanical motion of the mode of interest. When cavity mechanical motion and radiation pressure shot noise can be neglected, we can convert the relative transmitted intensity spectrum $S_{I}(\omega)$ to a (one-sided) displacement spectrum $S_{z}(\omega)$ using the following relation: $$S_{z}(\omega)=\frac{S_{I}(\omega)}{\bar{I}^2}\frac{1}{|\Pi(\omega)|^2 G^2}.
\label{eqn1}$$ Here $\Pi(\omega)=\chi_c(\omega)-\chi_c^*(-\omega)$ where $\chi_c(\omega)=\frac{1}{\kappa/2-\imath(\omega + \Delta)}$ is the response function for a cavity with linewidth $\kappa$ (energy decay rate) and detuning of the laser from the effective cavity resonance $\Delta$. The extracted $S_z(\omega)$ yields a thermally driven, optomechanically damped Lorentzian on top of a floor coming from optical shot and photodetector dark noise (Fig. \[fig2\](a) for example). We extract the effective temperature of the mode by fitting the data to a Lorentzian. We use the area under the fit to determine the mean square displacement, which is proportional to the effective temperature, and the effective phonon occupation $\bar{n}$. Applying this procedure to the (2,2) mode data in Fig. \[fig2\](a) we find the apparent effective phonon occupation displayed as the red squares in Fig. \[fig4\](d), which would indicate the thermal component of the mechanical motion is brought to 1.4 quanta.
As discussed above, an important limit to laser cooling is classical noise in the relative frequency of the cooling laser and cavity resonance frequency. Such effects include laser phase noise [@Rabl2009b; @SchleierSmith2011; @Kippenberg2011] and mechanical and thermal noise in the cavity mechanical modes [@Pinard2000; @Verhagen2012; @Zhao2012]. This relative frequency noise is converted to intensity noise in the cavity for off resonant laser drives, such as cooling tones. The intensity noise applies a radiation pressure force to the mechanical oscillator leading to an extra displacement and greater effective temperature. Further, extracting the temperature from the transmitted intensity spectrum in the presence of classical noise becomes complex [@Rabl2009b; @SchleierSmith2011; @Yang2011].
As a first study of the effect of the classical noise on measurements of $\bar{n}$, we present a more thorough analysis of the data of Fig. \[fig2\](b). Here, we model the cavity mechanical mode noise as dominated by a single thermally-occupied mode. We then determine the corresponding effect on the (2,2) mode. Importantly, we assume that the observed cavity mechanical motion does not directly drive the membrane mode, but rather the only coupling is through the intracavity field. If this assumption were not true, it is possible the entire output spectrum could be attributed to membrane motion. Further, we assume the effective mass of the cavity mode is much larger than that of the membrane mode. The physical mass of expected cavity modes is greater than $10^6$ times larger than that of membrane modes. Hence, even without complete knowledge of the coupling to the cavity mode, this is an excellent assumption. Thus, the cavity mechanical mode motion, to a good approximation, is unaffected by the optomechanical interaction. In the Appendix we present a calculation for this situation. Our model allows us to estimate the displacement spectrum of the membrane mode from the measured cavity transmission spectrum, when the simple relation between transmission and displacement spectra (Eqn. \[eqn1\]) no longer holds.
We estimate the functional form of the noise spectrum as a Lorentzian centered at the position of the large cavity mechanical mode peak. Because the noise is concentrated away from the membrane mode resonance, we may assume (at least for the weakly damped data) that the area in the transmission spectrum due to the cavity mechanical mode is largely independent of the membrane motion, and is equal to $S_{i}(\omega)$, the transmission spectrum without the membrane mode of interest present. In the Appendix it is shown that this extra intensity noise causes a radiation pressure force that drives the membrane mode to an additional displacement given by: $$S_{z,\delta f}(\omega)=\frac{4\omega_m^2 g_0^2 \bar{N}^2}{|\mathcal{N}(\omega)|^2} \frac{S_{i}(\omega)}{\bar{I}^2}Z_\mathrm{zp}^2.$$ In this expression, $g_0=G Z_\mathrm{zp}$ where $Z_\mathrm{zp}=\sqrt{\frac{\hbar}{2 m \omega_m}}$ is the zero point motion for an oscillator with frequency $\omega_m$ and effective mass $m$, and $\bar{N}$ is the intracavity photon number. The function $\mathcal{N}(\omega)$ represents the optomechanically modified mechanical response and is displayed in the Appendix.
 at an intracavity cavity photon number of $3.3\times10^6$ (teal curve 7th from the top in Fig. \[fig2\](b)). (b) Expected contributions to the relative transmission spectrum due to the cavity mechanical modes, $S_{i}(\omega)$ for the same parameters as in (a). (c) Modeled displacement spectrum of the (2,2) mechanical mode (solid green). The dashed red line shows the displacement spectrum expected if we were to ignore $S_{i}(\omega)$. (d) Extracted integrated motion (converted to a phonon occupation) for the (2,2) membrane mechanical mode as a function of cooling laser power. Red squares show only the contribution of the thermally driven Lorentzian peak, which correctly predicts a small on-resonance motional spectral density, but gives an erroneously low phonon occupation. The green diamonds include contribution from red squares and the motion induced by the cavity mechanical mode motion. The curve in (c) corresponds to the point with a 3 kHz mechanical damping rate.[]{data-label="fig4"}](fig4.eps)
Curves displaying an example of the modeled spectrum $S_{I}(\omega)$ and the cavity mechanical noise contribution $S_{i}(\omega)$ are shown in Fig. \[fig4\](a)&(b). The inferred value of $S_{z}(\omega)$ for these parameters is plotted in Fig. \[fig4\](c) (green line). Since the majority of the noise peak in $S_{I}(\omega)$ comes directly from $S_{i}(\omega)$, the height of the cavity mechanical mode peak relative to the thermally driven motion in the displacement spectrum is much smaller than the corresponding ratio in the transmission spectrum. The total estimated displacement of the membrane mode is shown in Fig. \[fig4\](d) as the green diamonds. Specifically, we plot the integrated motion $\langle z^2 \rangle$ for curves like that shown in Fig. \[fig4\](c), converted to an equivalent phonon occupation. As the laser power increases, the contribution of the cavity mechanical mode noise drive increases because the intensity fluctuations grow in proportion to the optical intensity. Thus, a minimum phonon occupation of $\bar{n}\approx 5$ occurs at an intermediate intracavity power. And as noted earlier, without this cavity mechanical mode (or if this mode were much farther away in frequency space) we would achieve $\bar{n}$ approaching unity, as displayed by the red squares.
Next we consider the temperature achieved for the (4,4) mode data of Fig. \[fig3\](a) taking into account the cavity noise. Here our best estimate is obtained by modeling the cavity mechanical noise spectrum as a white floor. For these data, shot noise for the highest-power set (black curve) corresponds to a relative intensity noise of $6\times10^{-15}$ and hence we would place the classical cavity noise floor at $\sim 4\times10^{-15}$. First, it is useful to consider the magnitude of motion this value corresponds to; as discussed above, we suspect in this case the cavity mechanical motion is dominated by the mirror fused silica substrates. Using the $G$ values discussed above, $S_I/\bar{I}^2=4\times10^{-15}$ corresponds to membrane motion of $0.9\times10^{-17}~\mathrm{m/\sqrt{Hz}}$ or a real end-mirror motion of $0.7\times10^{-18}~\mathrm{m/\sqrt{Hz}}$. Assuming a scaling of the mirror substrate motion with $\sqrt{T_\mathrm{bath}}$, this cavity mirror motion is within a range expected for thermally occupied modes of fused silica substrates based upon previous measurements of room temperature cavities [@Wilson2011]. Finally, we apply the same model discussed above and in the Appendix to these data. Again we assume that there is no physical coupling between the cavity mechanical modes and the membrane; since we assign the cavity mechanical modes in this situation to the mirror substrates, we believe this is a very good assumption. We use the estimated flat spectrum at $4\times10^{-15}$ as $S_{i}(\omega)/\bar{I}^2$ and the data of Fig. \[fig3\](a), and we find the phonon occupation reaches a minimum value of $\bar{n}=6$ as a function of cooling laser intensity.
In conclusion, we have presented two experimental data sets in which we have cooled Si$_3$N$_4$ membrane modes, as a conservative estimate, within a factor of 10 of the quantum mechanical ground state. The dominant uncertainty in our temperature measurement comes from complex classical noise spectra added at the few phonon level by other mechanical modes within the cavity. We expect with minimal additional optimization of our cavities we can remove deleterious effects of these additional cavity modes. Already we have demonstrated the kind of system parameters required for ground state cooling; for example a fruitful set of parameters would combine the coupling and quality factors achieved for the device in Fig. \[fig2\] with the absence of the isolated high-$Q$ peak in Fig. \[fig3\]. We believe engineering the silicon substrate to advantageously define the relative frequencies of the membrane and substrate modes will allow us to consistently achieve desired parameters in future designs [@Jockel2011]. It has also been shown that relative frequency noise of the cooling laser and cavity resonance frequency can be removed via active feedback schemes involving higher-order cavity modes [@Zhao2012]. Further, final studies of these devices will likely be conducted at even colder cryogenic temperatures compatible with superconducting circuits; here thermal occupation of cavity mirrors and the substrate should be reduced.
Importantly, thus far we have not observed physical absorption heating of our devices that leads to a significant increase in $T_\mathrm{bath}$. This fact, combined with the efficient detection that these devices can afford, makes this system promising for the observation of the radiation pressure shot noise. Even with the presence of cavity frequency noise, shot-noise from an sufficiently-strong optical tone placed on the cavity resonance should be efficiently transduced to the membrane motion.
This work was supported by the DARPA QuASAR program, ONR YIP, and JILA NSF-PFC. We thank Konrad Lehnert’s group for helpful input. CR thanks the Clare Boothe Luce Foundation for support. TP thanks the NRC for support.
Calculation of thermo-mechanical noise in an optomechanical system
==================================================================
To analyze our optomechanical system we consider a standard optomechanical Hamiltonian with the addition of cavity mechanical modes of the cavity structure indexed by $i$: $$\begin{aligned}
\label{eq:Hamiltonian}
\fl \qquad H=\hbar \omega_m c^{\dag} c + \hbar \omega_c a^{\dag} a +\hbar G Z_\mathrm{zp} (c+c^{\dag}) a^{\dag} a \\
+ \hbar \sum_{i}{\left(\omega_i b_i^{\dag} b_i+G_i Z_{zp,i}(b_i+b_i^{\dag}) a^{\dag} a\right)} + H_{\kappa}+H_{\Gamma} \nonumber\end{aligned}$$ Here $c$ is the annihilation operator for the mechanical mode of interest (the membrane mode), with oscillation frequency $\omega_m$ and harmonic oscillator length $Z_\mathrm{zp}$. $a$ is the annihilation operator for the cavity mode, at frequency $\omega_c$. $G$ is the optomechanical coupling constant, and we define a single photon coupling rate $g_0=G Z_\mathrm{zp}$. $b_i$ is the annihilation operator for the $i$th cavity mechanical mode with frequency $\omega_i$, harmonic oscillator length $Z_{zp,i}$, and optomechanical coupling $G_i$. Additionally, the term $H_{\kappa}$ represents the input and output optical coupling of the cavity with total cavity decay rate $\kappa=\kappa_L+\kappa_R+\kappa_\mathrm{int}$. The contributing decay rates stem from the input port ($\kappa_L$), the output port $(\kappa_R$), and the internal loss ($\kappa_\mathrm{int}$). The term $H_{\Gamma}$ represents the thermal drive on all of the mechanical modes.
The optomechanical interaction can be linearized and fast oscillations at the optical frequency accounted for by defining $a(t)=(\bar{a}+d(t))e^{-\imath \omega_L t}$. $\bar{a}=\frac{\sqrt{\kappa_L} \bar{a}_\mathrm{in}}{\kappa/2-\imath \Delta}$ is the large classical amplitude of the intracavity field and $d(t)$ represents small fluctuations about this value. $\bar{a}_\mathrm{in}$ is the coherent state amplitude of the input laser field driven at frequency $\omega_L$.
From the Hamiltonian in Eqn. (\[eq:Hamiltonian\]), we derive a set of Heisenberg-Langevin equations of motion and transform them into the frequency domain using the Fourier transformation $f(\omega)=\int^{\infty}_{-\infty}e^{\imath \omega t} f(t) dt$, $f^{\dag}(\omega)=\int_{-\infty}^{\infty}e^{\imath \omega t} f^{\dag}(t) dt$. $$\fl d(\omega)=\chi_c(\omega) \left(-\imath \bar{a} G z(\omega) -\imath \bar{a} \sum{G_i z_i(\omega)}+ \sqrt{\kappa_L}\xi_L(\omega) +\sqrt{\kappa_\mathrm{int}}\xi_\mathrm{int}(\omega)+\sqrt{\kappa_R} \xi_R(\omega)\right)$$ $$\fl \frac{z(\omega)}{Z_\mathrm{zp}}=\imath g_0 \Big(\chi_m^*(-\omega)-\chi_m(\omega) \Big)\Big(\bar{a} d^{\dag}(\omega) +\bar{a}^*d(\omega)\Big)+\sqrt{\Gamma_m}\Big(\chi_m(\omega)\eta(\omega)+\chi_m^*(-\omega)\eta^{\dag}(\omega)\Big)$$ $$\fl \frac{z_i(\omega)}{Z_{zp,i}}=\sqrt{\Gamma_i}\left(\chi_{i}(\omega)\eta_i(\omega)+\chi_{i}^*(-\omega)\eta_i^{\dag}(\omega) \right)$$ $z$ and $z_i$ represent the small displacements of the mechanical modes about their optomechanically shifted equilibrium positions $\bar{z}$ and $\bar{z_i}$, such that $Z_\mathrm{zp}\left(c+c^{\dag}\right)=\bar{z}+z$ and $Z_{zp,i}\left(b_i+b_i^{\dag}\right)=\bar{z}_i + z_i$. We use here the mechanical susceptibilities $\chi_{m}(\omega)=\frac{1}{\Gamma_m/2-\imath(\omega-\omega_{m})}$ and $\chi_{i}(\omega)=\frac{1}{\Gamma_i/2-\imath(\omega-\omega_{i})}$. $\chi_c(\omega)=\frac{1}{\kappa/2-\imath(\Delta+\omega)}$ is the cavity susceptibility, where $\Delta=\omega_c-\omega_L-G \bar{z}-\sum{G_i \bar{z}_i}$ is the detuning of the laser input frequency from optomechanically shifted cavity resonance frequency. The operators $\xi_L e^{-\imath \omega_L t}$, $\xi_R e^{-\imath \omega_L t}$, and $\xi_\mathrm{int} e^{-\imath \omega_L t}$ are Langevin noise operators representing vacuum fluctuations entering the cavity from the input, loss, and output ports (see for example [@WallsandMilburn; @Clerk2010]). $\eta$ and $\eta_i$ are the Langevin noise operators representing the thermal and vacuum noise driving the mechanical modes. To simplify the equations of motion we drop small terms of order $d^2$, $d z$, and $d z_i$. Additionally, we work in the limit where $G_i^2 Z_{zp,i}^2 \bar{a}^*\bar{a}/\kappa \ll \bar{n}_{th,i} \Gamma_i$ where $\bar{n}_{th,i}=\frac{k_b T_\mathrm{bath}}{\hbar \omega_i}$ and $T_\mathrm{bath}$ is the bath temperature of the modes. In this limit the optical drive does not perturb the mechanical state of the cavity mechanical modes. However, signatures of the cavity mechanical modes are still imprinted on the optical mode.
We can then solve the equations of motion and calculate the mechanical displacement spectrum $S_z^{(2)}(\omega)=\left<z(-\omega) z(\omega)\right>$. $$\begin{aligned}
\label{eq:zz}
\fl \qquad \frac{\left<z(-\omega) z(\omega)\right>}{Z_\mathrm{zp}^2}= \frac{1}{\mathcal{N}(-\omega) \mathcal{N}(\omega)}\bigg\{&
\Gamma_m \left(\frac{\bar{n}_\mathrm{th}+1}{|\chi_m(\omega)|^2}+\frac{\bar{n}_\mathrm{th}}{|\chi_m(-\omega)|^2}\right) \\
&+ 4\omega_m^2 g_0^2 \kappa \bar{a}^*\bar{a}|\chi_c(-\omega)|^2 \nonumber \\
& +4\omega_m^2 g_0^2(\bar{a}^*\bar{a})^2 |\Pi(\omega)|^2 \left< \delta f (-\omega) \delta f(\omega)\right> \bigg\} \nonumber\end{aligned}$$ where $\mathcal{N}(\omega)=\frac{1}{\chi_m(\omega)\chi_m^*(-\omega)} -\imath 2 \omega_m g_0^2 \bar{a}^*\bar{a} \Pi(\omega) $ and $\Pi(\omega)=\chi_c(\omega)-\chi_c^*(-\omega)$. We make use of the operator expectation values $\left<\xi_L(-\omega)\xi_L^{\dag}(\omega)\right>=\left<\xi_\mathrm{int}(-\omega)\xi_\mathrm{int}^{\dag}(\omega)\right>=\left<\xi_R(-\omega)\xi_R^{\dag}(\omega)\right>=1$, and $\left<\eta(-\omega) \eta^{\dag}(\omega)\right>=\bar{n}_\mathrm{th}+1$, $\left<\eta^{\dag}(-\omega) \eta(\omega)\right>=\bar{n}_\mathrm{th}$. The first term of Eqn. (\[eq:zz\]) contains the thermal motion of the membrane. The second term includes the membrane motion induced by radiation pressure from optical shot noise. The effect of the cavity mechanical modes is seen in the third term, through $\left< \delta f (-\omega) \delta f(\omega)\right>= \sum{G_i^2 \left< z_i(-\omega) z_i(\omega)\right>}$, which represents the noise spectrum of the cavity frequency shifts induced by the cavity mechanical modes.
We next compute the spectrum of intensity fluctuations for light directly detected on a photodetector at the output port of the cavity. Let $S_{II}^{(2)}=\left<\left(I(-\omega)-\bar{I}\right)\left(I(\omega)-\bar{I}\right) \right>$ be the two-sided power spectrum of the detected photocurrent where $I(t)=\epsilon\hbar \omega_L \mathcal{R} a^{\dag}_\mathrm{out}(t) a_\mathrm{out}(t)+(1-\epsilon) a_n^{\dag}(t)a_n(t)+I_d(t)$, with mean value $\bar{I}=\left<I\right>$. $\mathcal{R}=\frac{q_e}{\hbar \omega_L}$ is the photodetector sensitivity where $q_e$ is the electron charge. $\epsilon$ is the detection efficiency, $a_n(t)=\xi_n(t)e^{-\imath \omega_L t}$ is a Langevin noise operator representing vacuum fluctuations entering the detector through the loss port associated with the detector inefficiency, and $I_d$ is the photodetector dark current. The output optical field $a_\mathrm{out}(t)=(\bar{a}_\mathrm{out}+d_\mathrm{out}(t))e^{-\imath \omega_L t}$ is evaluated via the input-output relations: $\bar{a}_\mathrm{out}=\sqrt{\kappa_R}\bar{a}$ and $d_\mathrm{out}=\xi_R+\sqrt{\kappa_R} d$. $$\begin{aligned}
\fl \qquad \frac{S_{II}^{(2)}(\omega)}{\bar{I}^2}=\frac{1}{\kappa_R (\bar{a}^*\bar{a})^2}\Big< &\left[ \bar{a}^*(\sqrt{\kappa_R} d(-\omega)-\xi_R(-\omega))+\bar{a} (\sqrt{\kappa_R}d^{\dag}(-\omega)-\xi_R^{\dag}(-\omega))\right]\nonumber \\
& \times \left[ \bar{a}^*(\sqrt{\kappa_R} d(\omega)-\xi_R(\omega))+\bar{a}(\sqrt{\kappa_R}d^{\dag}(\omega)-\xi_R^{\dag}(\omega))\right] \Big>\nonumber\\
& + \frac{1-\epsilon}{\epsilon}\frac{1}{\kappa_R \bar{a}^*\bar{a}}+\frac{\left<I_d(-\omega) I_d(\omega)\right>}{\bar{I}^2}\nonumber\\
& = \left<\Psi(-\omega) \Psi(\omega)\right> + \frac{1-\epsilon}{\epsilon}\frac{1}{\kappa_R \bar{a}^*\bar{a}}+\frac{\left<I_d(-\omega) I_d(\omega)\right>}{\bar{I}^2}\nonumber\end{aligned}$$ where $\Psi(\omega)=\Psi_q(\omega)+\Psi_m(\omega)+\Psi_i(\omega)$ contains the following components $$\begin{aligned}
\fl \qquad \Psi_{q}(\omega)=\frac{1}{\bar{a}^*\bar{a}}&\bigg(\bar{a}^*\sqrt{\kappa_L}\chi_c(\omega) \xi_L(\omega)+\bar{a}\sqrt{\kappa_L}\chi_c^*(-\omega) \xi_L^{\dag}(\omega)+\frac{\bar{a}^*}{\sqrt{\kappa_R}}(\kappa_R \chi_c(\omega)-1)\xi_R(\omega)\nonumber\\
+ & \frac{\bar{a}}{\sqrt{\kappa_R}}(\kappa_R \chi_c^*(-\omega)-1)\xi_R^{\dag}(\omega)+\bar{a}^*\sqrt{\kappa_\mathrm{int}}\chi_c(\omega) \xi_\mathrm{int}(\omega)+\bar{a}\sqrt{\kappa_\mathrm{int}}\chi_c^*(-\omega) \xi_\mathrm{int}^{\dag}(\omega)\bigg)\nonumber\end{aligned}$$ $$\begin{aligned}
\fl \qquad \Psi_{m}=-\imath G \Pi(\omega) z(\omega)\nonumber\end{aligned}$$ $$\begin{aligned}
\fl \qquad \Psi_{i}=-\imath \Pi(\omega) \delta f(\omega).\nonumber\end{aligned}$$ We can thus compute the functions $$\begin{aligned}
\fl \qquad \left<\Psi_{q}(-\omega) \Psi_{q}(\omega) \right>=\frac{1}{\kappa_R \bar{a}^*\bar{a}}\nonumber\end{aligned}$$ $$\begin{aligned}
\fl \qquad \left<\Psi_{m}(-\omega) \Psi_{m}(\omega) \right>= G^2|\Pi(\omega)|^2 \left<z(-\omega) z(\omega)\right>\nonumber\end{aligned}$$ $$\begin{aligned}
\fl \qquad \left<\Psi_{i}(-\omega) \Psi_{i}(\omega) \right>=|\Pi(\omega)|^2 \left< \delta f (-\omega) \delta f(\omega)\right>\nonumber\end{aligned}$$ and the nonzero cross terms are $$\fl \qquad \left<\Psi_q(-\omega) \Psi_m(\omega) \right>+\left<\Psi_m(-\omega) \Psi_q(\omega) \right>=-4 \omega_m g_0^2 \mathrm{Im} \left[\frac{\Pi(\omega)}{\mathcal{N}(\omega)} \chi_c(-\omega) \right]$$ $$\fl \qquad \left<\Psi_i(-\omega) \Psi_m(\omega) \right>+\left<\Psi_m(-\omega) \Psi_i(\omega) \right>=-4 \omega_m \bar{a}^*\bar{a} g_0^2 \mathrm{Im}\left[\frac{\Pi(\omega)}{\mathcal{N}(\omega)} \right] |\Pi(\omega)|^2 \left< \delta f (-\omega) \delta f(\omega)\right>.$$ The photodetector signals we record in the experiment are one-sided power spectra $S_{I}(\omega)=S_{II}^{(2)}(-\omega)+S_{II}^{(2)}(\omega)$. Similarly we define $S_{i}(\omega)/\bar{I}^2=2 \left<\Psi_{i}(-\omega) \Psi_{i}(\omega) \right>$, the one-sided spectrum of transmitted intensity noise due to the fluctuations of the cavity mechanical modes, and from this we infer $S_{z}(\omega)=\left<z(-\omega) z(\omega)\right>+ \left<z(\omega) z(-\omega)\right>$ the one-sided displacement spectrum.
If the contribution of the cavity mechanical modes is small ($S_i(\omega) \rightarrow 0 $) and the radiation pressure shot noise is small, then the displacement spectrum can be easily inferred from the transmission spectrum. $$S_{z,\delta f \rightarrow 0}(\omega)+ S_{\mathrm{noise floor}}(\omega) = \frac{1 }{G^2 |\Pi(\omega)|^2} \frac{S_{I}(\omega)}{\bar{I}^2}$$ Where $S_{\mathrm{noise floor}}(\omega)=\frac{2}{G^2 |\Pi(\omega)|^2}\left( \frac{1}{\epsilon}\frac{1}{\kappa_R \bar{a}^* \bar{a}}+\left<I_d(-\omega) I_d(\omega)\right> /\bar{I}^2 \right)$ is the detection noise floor due to optical shot noise and detector noise. If frequency noise is not negligible and $S_{i}(\omega)$ can be measured empirically then the contribution to $S_{z}(\omega)$ from the optomechanically transduced fluctuations of the cavity mechanical modes can be computed from Eqn. (\[eq:zz\]). $$\begin{aligned}
\frac{S_{z,\delta f}(\omega)}{Z_\mathrm{zp}^2}=\frac{4\omega_m^2 g_0^2 (\bar{a}^*\bar{a})^2}{|\mathcal{N}(\omega)|^2} \frac{S_{i}(\omega)}{\bar{I}^2} \label{Eqn:final}\end{aligned}$$ Note $\bar{a}^*\bar{a}$ is the intracavity photon number $\bar{N}$. From $S_{z}(\omega)$ the root mean square displacement of the membrane mode can be computed. $$\frac{\left<z^2\right>}{Z_\mathrm{zp}^2}=\int_0^{\infty} \frac{S_{z}(\omega)}{Z_\mathrm{zp}^2} \frac{d\omega}{2 \pi}=2 \left( \bar{n}+\frac{1}{2}\right)$$
We can compare our final expression for the impact of mechanical modes within the cavity (Eqn. \[Eqn:final\]) to the result expected for equivalent frequency noise on a laser at the input port of the cavity. We find our expression can be translated into a formula equivalent to the laser frequency-noise result derived in, for example, Ref. [@SchleierSmith2011].
Calibration of membrane motion
==============================
Here we present the three different methods we use to calibrate the mechanical motion and corresponding temperature. We compare the methods by assessing the membrane coupling extracted in each case for the (2,2) data presented in Fig. \[fig2\]; based on these analyses we deduce an uncertainty in $G$ of 5$\%$.
Bath temperature and optical damping
------------------------------------
The thermal motion of the mechanical resonator is equated to the effective temperature $T$ via $\langle z^2 \rangle=\frac{k_b T}{m \omega_m^2}$. Optomechanical cooling theory in the large damping limit tells us $T=T_\mathrm{bath}\frac{\Gamma_m}{\Gamma}$, where $\Gamma$ can simply be determined by the measured linewidth of the optically damped mechanical resonator. We compare $\langle z^2 \rangle$ as determined via integrating the spectrum $\int_0^\infty S_z(\omega) \frac{d\omega}{2\pi}$ to that from $\frac{k_b T_\mathrm{bath}}{m \omega_m^2}\frac{\Gamma_m}{\Gamma}$. $T_\mathrm{bath}=4.9$ K is input based upon the cryostat thermometer, where the accuracy of this measurement was motivated via the trend in Fig. \[fig3\](c). $S_z(\omega)$ can be extracted from the measured intensity spectrum via Eqn. \[eqn1\] with one free parameter $G$.
The inputs to this calculation are: $\kappa/2\pi = 1.2$ MHz, which is determined by a measurement of the cavity linewidth in ringdown for the position of the membrane during the measurement. $\Delta/2\pi = 1.6$ MHz, which is determined from the detuning set with respect to the on-resonant locking light, accounting for a birefringent splitting of 0.4 MHz. The mechanical frequency $\omega_m/2\pi = 1.575$ MHz. The mechanical quality factor $Q_m=13.6\times10^6$. The effective mass $m=\rho d^2 t/4$, which is determined using $\rho=2700$ $\mathrm{kg/m^3}$ [@Verbridge2006]. (However, given the range of values in the literature for LPCVD Si$_3$N$_4$ [@Wilson2011] we would place a $10\%$ systematic uncertainty on the mass. Note, the mass uncertainty is relevant for $G$ but not $g_0$ [@Gorodetsky2010].)
This comparison tells us $G/2\pi =1.8 \times 10^{16}$ Hz/m.
Membrane-at-the-end model
-------------------------
With knowledge of the position of the membrane within the cavity standing wave we can model the expected coupling. In each iteration of the experiment, we scan the membrane within the cavity to sit at the position for which the cavity linewidth $\kappa$ is minimized [@Wilson2011]. With knowledge of the cavity and membrane parameters we can calculate the expected coupling at this operating position we refer to as $Z_\mathrm{min}$. $Z_\mathrm{min}$ is registered within the standing wave $\sim0.9$ mm from the flat mirror of the cavity. The cavity has an overall length of 5.1 mm, which is measured via the ratio of the transverse mode spacing to the free spectral range of 29.4 GHz. Based on direct ellipsometry measurements of our film we base our calculation on a $t=40$ nm thick membrane with index $n=2.0$. Hence, at $Z_\mathrm{min}$ we predict $\mathrm{d}\omega_c/\mathrm{d}z=2\pi \times 2.9\times10^{16}$ Hz/m.
We then apply a correction based upon the measured transverse mode overlap between the mechanical mode of interest and the $\mathrm{TEM}_{00}$ cavity mode (an example of such a measurement is shown in Fig. \[fig1a\]). To account for the mode overlap we calculate $G=\eta \frac{d\omega_c}{dz}$ using [@Gillespie] $$\eta_{mn} = \left | \iint \mathrm{d}x\mathrm{d}y I(x,y)a_{mn}(x,y)/a_0 \right |$$ where $a_{mn}=a_0 \sin(m\pi x / d) \sin(n\pi y /d)$ and $I(x,y)$ is a normalized intensity function $I(x,y)=\frac{2}{\pi \mathrm{w}_x \mathrm{w}_y}\mathrm{exp}(-2(x-x_0)^2/\mathrm{w}_x^2)\mathrm{exp}(-2(y-y_0)^2/\mathrm{w}_y^2)$. We measure the location of the (2,2) mode while cold to be at the coordinate position $(x_0,y_0)=(108, 99)$ $\mu$m. The size of the mode in $x$,$y$ is measured to be $\mathrm{w}_x=92$ $\mu$m and $\mathrm{w}_y=88$ $\mu$m. Hence $\eta=0.67$, and $G/2\pi = 2.0 \times 10^{16}$ Hz/m.
Optical damping and intracavity photon number
---------------------------------------------
The optical damping $\Gamma$ we observe for a given intracavity photon number is also a measure of $G$. To infer the intracavity photon number $\bar{N}$ from the measured output flux of the cavity we understand the asymmetry of the cavity and the internal loss. Usually we orient the cavity according to Fig. \[fig1a\] where the membrane is at the output (right) side of the cavity, providing the most signal at the output. However, for these particular measurements we happened to measure out the less-transmissive port of the cavity, i.e. the membrane and flat mirror were actually at the input (left) side of the cavity. For a measurement of the photocurrent $\bar{I}$ at the output port the photon number is given by $$\bar{N}=\frac{P_\mathrm{out}}{\hbar \omega_c \kappa}\frac{\kappa}{\kappa_R}=\frac{1}{\kappa_R}\frac{\bar{I}}{q_e}\frac{1}{\epsilon}$$ for a low-reflectivity membrane [@Wilson2011], where $\epsilon=\epsilon_d \epsilon_p$ is a combination of two efficiency factors. In our setup for these measurements, the detector efficiency is $\epsilon_d=0.87$ and the propagation losses from the cavity output to the detector are given by $\epsilon_p=0.88$.
To find $\kappa_R$ we must understand all the contributions to $\kappa=\kappa_R+\kappa_L+\kappa_\mathrm{int}$. $\kappa_\mathrm{int}$ is dominated by clipping of the transverse mode on the silicon frame that results from imperfect alignment when cooling to cryogenic temperatures, and hence for modeling $G$ and $\kappa$ we consider this a loss that is independent of membrane position. At the position that corresponds to the minimum value of $\kappa$ ($Z_\mathrm{min}$), the theoretical analysis of the three-element cavity described above indicates $\kappa_\mathrm{min}=0.79$ MHz. With a well-aligned cavity we often achieve this value; in this particular cavity at room temperature we achieve $\kappa_\mathrm{min}=0.85$ MHz, and when cooled down we find $\kappa_\mathrm{min}=1.17$ MHz. This indicates an internal loss contribution of $\kappa_\mathrm{int}=0.33 \kappa$. The ratio between $\kappa_R$ and $\kappa_L$ can be calculated from the asymmetry of the cavity, which can be determined from a calculation of the resonant reflection $R$ and transmission $T$ at our operating position of $Z_\mathrm{min}$. We use the expression $\kappa_L/\kappa_R=\frac{(1+\sqrt{R})^2}{T}$ to find $\kappa_L/\kappa_R=1.9$ [@Wilson2011]. Thus $\kappa_R=0.23 \kappa$ taking into account $\kappa_\mathrm{int}$.
We then compare the measured optically damped linewidth $\Gamma$ to the expected calculated $\Gamma$ given by [@Marquardt2007a; @WilsonRae2007] $$\Gamma = G^2Z_\mathrm{zp}^2\kappa \bar{N} \left(|\chi_c(\omega_m)|^2-|\chi_c(-\omega_m)|^2 \right).$$ This gives us $G/2\pi =1.9 \times 10^{16}$ Hz/m.
References {#references .unnumbered}
==========
[10]{}
J. D. Teufel [*et al.*]{}, Nature [**475**]{}, 359 (2011).
T. Rocheleau [*et al.*]{}, Nature [**463**]{}, 72 (2010).
F. Massel [*et al.*]{}, arXiv:1205.3106v1 (2012).
J. Chan [*et al.*]{}, Nature [**478**]{}, 89 (2011).
E. Verhagen [*et al.*]{}, Nature [**482**]{}, 63 (2012).
I. Tittonen [*et al.*]{}, Phys. Rev. A [**59**]{}, 1038 (1999).
C. H. Metzger and K. Karrai, Nature [**432**]{}, 1002 (2004).
O. Arcizet [*et al.*]{}, Nature [**444**]{}, 71 (2006).
T. Corbitt [*et al.*]{}, Phys. Rev. Lett. [**98**]{}, 150802 (2007).
J. D. Thompson [*et al.*]{}, Nature [**452**]{}, 72 (2008).
S. Gröblacher [*et al.*]{}, Nature Phys. [**5**]{}, 285 (2009).
C. Yang, Ph.D. thesis, Yale University, 2011.
W. Marshall, C. Simon, R. Penrose, and D. Bouwmeester, Phys. Rev. Lett. [ **91**]{}, 130401 (2003).
C. M. Caves, Phys. Rev. Lett. [**45**]{}, 75 (1980).
V. B. Braginsky and F. Y. Khalili, in [*Quantum Measurement*]{}, edited by K. S. Thorne (Cambridge University, Cambridge, UK, 1992).
K. Jacobs, P. Tombesi, M. J. Collett, and D. F. Walls, Phys. Rev. A [**49**]{}, 1961 (1994).
A. Heidmann, Y. Hadjar, and M. Pinard, Appl. Phys. B [**64**]{}, 173 (1997).
K. W. Murch, K. L. Moore, S. Gupta, and D. M. Stamper-Kurn, Nature Phys. [ **4**]{}, 561 (2008).
K. Yamamoto [*et al.*]{}, Phys. Rev. A [**81**]{}, 033849 (2010).
B. M. Zwickl, Ph.D. thesis, Yale University, 2011.
C. A. Regal and K. W. Lehnert, J. Phys.: Conf. Ser. [**264**]{}, 012025 (2011).
A. H. Safavi-Naeini and O. Painter, New J. Phys. [**13**]{}, 013017 (2011).
J. M. Taylor, A. S. Sorensen, C. M. Marcus, and E. S. Polzik, Phys. Rev. Lett. [**107**]{}, 273601 (2011).
P.-L. Yu, T. P. Purdy, and C. A. Regal, Phys. Rev. Lett. [**108**]{}, 083603 (2012).
T. P. Purdy, R. W. Peterson, P.-L. Yu, and C. A. Regal, in [*CLEO: Science and Innovations*]{} (OSA Technical Digest (online), Optical Society of America, 2012), No. CTh3M.2.
B. M. Zwickl [*et al.*]{}, Appl. Phys. Lett. [**92**]{}, 103125 (2007).
D. J. Wilson, C. A. Regal, S. B. Papp, and H. J. Kimble, Phys. Rev. Lett. [ **103**]{}, 207204 (2009).
I. Wilson-Rae [*et al.*]{}, Phys. Rev. Lett. [**106**]{}, 047205 (2011).
D. J. Wilson, Ph.D. thesis, California Institute of Technology, 2011.
Y. Zhao, D. J. Wilson, K.-K. Ni, and H. J. Kimble, Optics Express [**20**]{}, 3586 (2012).
P. Rabl, C. Genes, K. Hammerer, and M. Aspelmeyer, Phys. Rev. A [**80**]{}, 063819 (2009).
M. H. Schleier-Smith [*et al.*]{}, Phys. Rev. Lett. [**107**]{}, 143005 (2011).
T. J. Kippenberg, A. Schliesser, and M. Gorodetsky, arXiv:1112.6277v1 (2011).
T. B. M. Pinard, P. F. Cohadon and A. Heidmann, Phys. Rev. A [**63**]{}, 013808 (2000).
A. Jöckel [*et al.*]{}, Appl. Phys. Lett. [**99**]{}, 143109 (2011).
D. F. Walls and G. J. Milburn, [*Quantum Optics*]{} (Springer, New York, 1994).
A. A. Clerk [*et al.*]{}, Rev. Mod. Phys. [**82**]{}, 1155 (2010).
S. S. Verbridge [*et al.*]{}, J. Appl. Phys. [**99**]{}, 124304 (2006).
M. L. Gorodetsky [*et al.*]{}, Opt. Express [**18**]{}, 23236 (2010).
A. Gillespie and F. Raab, Phys. Rev. D [**52**]{}, 577 (1995).
F. Marquardt, J. P. Chen, A. A. Clerk, and S. M. Girvin, Phys. Rev. Lett. [ **99**]{}, 093902 (2007).
I. Wilson-Rae, N. Nooshi, W. Zwerger, and T. J. Kippenberg, Phys. Rev. Lett. [**99**]{}, 093901 (2007).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We investigate the joint asymptotic behavior of so-called blocks estimator of the extremal index, that determines the mean length of clusters of extremes, based on the exceedances over different thresholds. Due to the large bias of these estimators, the resulting estimates are usually very sensitive to the choice of the threshold and thus difficult to interpret. We propose and examine a bias correction that asymptotically removes the leading bias term while the rate of convergence of the random error is preserved.'
author:
- 'Holger Drees[^1]'
title: Bias correction for estimators of the extremal index
---
[^2]
Introduction
============
When one analyzes a risk related to extreme values of a stationary time series, then the clustering behavior of extremes can be as least as important as the tail behavior of the marginal distribution. For example, while a flood control basin may cope with a single day of extreme rainfall, an extended period of heavy rain will more likely lead to a flooding of the surrounding area. Similarly, large negative returns on a stock index over several days may sum up to an overall loss which is much worse than the most extreme crash ever experienced on a single day.
Obviously, there is no single parameter which captures all facets of serial dependence between extreme values, and in different applications different features may be of interest. Recently, Drees and Rootzén (2010) introduced a very flexible class of empirical processes that are capable of describing quite general aspects of extremal dependence. In the present paper, it is demonstrated how the asymptotic theory of these empirical processes can be used to immensely improve the performance of well-known estimators of the so-called extremal index, that is the reciprocal value of the asymptotic mean cluster size.
More specifically, let a stationary time series $X_i, 1\le i\le n$, with marginal distribution function (d.f.) $F$ be observed. We assume that $F$ belongs to the maximum domain of attraction of some extreme value d.f. $G_\gamma$, i.e., for an accompanying sequence of independent and identically distributed (i.i.d.) random variables (r.v.s) $\tilde X_i, 1\le i\le n$, with d.f. $F$ there exist normalizing constants $a_n>0$ and $b_n\in{\mathbb{R}}$ such that $$\label{eq:DOA}
P\Big\{ \frac{\max_{1\le i\le n} \tilde X_i -b_n}{a_n}\le x\Big\}
\;\longrightarrow\; G_\gamma(x),\quad x\in{\mathbb{R}},$$ as $n\to\infty$. It is well known that (up to a scale and location parameter) $G_\gamma$ must be of the form $G_\gamma(x)=\exp\big(-(1+\gamma x)^{-1/\gamma}\big)$ for all $x$ such that $1+\gamma x>0$. Let $$u_n(x) := a_n x+b_n.$$ Moreover, assume the following mild mixing condition (a weakened version of Leadbetter’s condition $D$):
Then there exists a constant $\theta\in[0,1]$, the so-called [*extremal index*]{}, such that $$\label{eq:maxdep}
P\Big\{ \frac{\max_{1\le i\le n} X_i -b_n}{a_n}\le x\Big\}
\;\longrightarrow\; G_\gamma^\theta(x),\quad x\in{\mathbb{R}},$$ provided that the left hand side converges (to an arbitrary limit) for some $x\in{\mathbb{R}}$. In what follows, we will always rule out the degenerate case $\theta=0$ which, in the limit, corresponds to clusters of extremes with infinite mean length.
If the extremal index $\theta$ is strictly positive, then usually it may be interpreted as the reciprocal value of a limiting cluster size. To see this, note that from and one may conclude $$\frac{P\{\max_{1\le i\le n} X_i>u_n(x)\}}{1-F^{n\theta}(u_n(x))}
\;\longrightarrow\; 1\quad \forall\, x\in{\mathbb{R}},$$ with the convention $0/0:=1$. Indeed, Hsing (1993, Theorem 3.1) proved that under a stronger mixing condition this convergence holds uniformly in $x$. If the following condition holds:
then $$\label{eq:unifconv}
\sup_{u\in{\mathbb{R}}} \Big| \frac{P\{\max_{1\le i\le n}
X_i>u\}}{1-F^{n\theta}(u)} -1\Big|
\;\longrightarrow\; 0.$$ Now a Taylor expansion yields $1-F^{n\theta}(u)\sim n\theta \bar
F(u)=\theta E(C_n(u))$ uniformly for all $u\in[u_n,F^\leftarrow
(1))$ where $C_n(u):=\sum_{i=1}^n {{\boldsymbol{1}_{\textstyle\{X_i>u\}}}}$ denotes the total number of exceedances over $u$, provided $u_n\to F^\leftarrow(1):=
\sup\{x\in{\mathbb{R}}\mid F(x)<1\}$ such that $\bar F(u_n) :=
1-F(u_n)=o(1/n)$. Hence, in view of , it follows $$\label{eq:recimean}
\frac 1{E(C_n(u)\mid C_n(u)>0)} = \frac{P\{\max_{1\le i\le n}
X_i>u\}}{n\bar F(u)} \;\longrightarrow\; \theta$$ uniformly for all $u\in[u_n,F^\leftarrow(1))$.
Convergence suggests to estimate $\theta$ by replacing the unknown probability and expectation on the left hand side by empirical counterparts. Since we cannot estimate $P\{\max_{1\le i\le n} X_i>u\}$ consistently if we observe merely $n$ consecutive r.v.s $X_i$, $1\le i\le n$, we must first replace $n$ with $r_n=o(n)$ in and adjust $u$ accordingly. Thus we split the sample into $m_n=\lfloor
n/r_n\rfloor$ blocks of length $r_n$ and estimate $\theta$ by $$\label{eq:blockestdef}
\hat\theta_n := \frac{\sum_{j=1}^{m_n}
{{\boldsymbol{1}_{\textstyle\{\textstyle\max_{(j-1)r_n<i\le jr_n}
X_i>u_n\}}}}}{\sum_{j=1}^{m_n} \sum_{i=(j-1)r_n+1}^{jr_n}
{{\boldsymbol{1}_{\textstyle\{X_i>u_n\}}}}},$$ for a sequence of thresholds $u_n$ satisfying $r_n\bar F(u_n)\to 0$, but $n\bar F(u_n)\to\infty$.
This so-called [*blocks estimator*]{} of the extremal index has been intensively studied in the literature. Hsing (1993) and Weissman and Novak (1998) proved its consistency and asymptotic normality under suitable mixing conditions. Variants of the blocks estimator were also examined by Smith and Weissman (1994) and Robert et al.(2009). As alternatives to blocks estimators, so-called [*runs estimators*]{} of $\theta$ have been proposed. While, in the numerator of the right hand side of , the number of clusters of extremes is defined as the number of blocks of length $r_n$ which contain at least one exceedance, in the runs approach two exceedances are considered to belong to different clusters if they are separated by at least $\tilde r_n$ consecutive observations that do not exceed $u_n$: $$\tilde\theta_n := \frac{\sum_{i=1}^{n-\tilde r_n}
{{\boldsymbol{1}_{\textstyle\{X_i>u_n, X_j\le u_n \text{ for all } i+1\le j\le i+\tilde r_n
\}}}}}{\sum_{i=1}^{n-\tilde r_n} {{\boldsymbol{1}_{\textstyle\{X_i>u_n\}}}}}.$$ The asymptotic behavior of this estimator was examined by Hsing (1993), Smith and Weissman (1994) and Weissman and Novak (1998), among others. Yet another approach was suggested by Ferro and Segers (2003), who used interarrival times between exceedances to estimate the extremal index.
In all these papers, the behavior of the estimators was analyzed for a [*fixed*]{} sequence of thresholds. Below we will argue that the analysis of the [*joint*]{} behavior of blocks estimators for different thresholds does not only provide deeper insight, but that it is the key to a remarkable reduction of the bias.
Indeed, all the estimators mentioned above are plagued by serious bias problems, which often renders inconclusive the analysis of the strength of extremal dependence. As a typical example, consider the following autoregressive time series of order 1 with Cauchy innovations $\varepsilon_t$: $X_t=\varphi X_{t-1}+\eps_t$ with $\varphi=0.6$. Figures \[sampleplot\] (a) and (b) display blocks and runs estimates of $\theta$ based on the exceedances over $F_n^\leftarrow(u)=X_{n-\lceil nu\rceil+1:n}$ as a function of $u$ for several block lengths $r_n$, resp. run lengths $\tilde r_n$. (Here $F_n$ denotes the empirical d.f. and $X_{i:n}$ the $i$th smallest order statistic.) The true value $\theta=1-\varphi$ is indicated by the horizontal lines. The estimates are almost monotone functions in $u$ and monotonically increasing in the block lengths $r_n$, resp. run lengths $\tilde
r_n$. (The latter monotonicity holds by construction if $n$ is divisible by $r_n$ resp. if the last $\tilde r_n$ observations do not exceed the threshold.) Since there is no region where the estimates remain stable, it is not obvious how to choose the threshold appropriately. Without an objective procedure for choosing the threshold, it will thus be difficult to justify any particular estimate for the extremal index.
\[sampleplot\]
{height="50mm"}{height="50mm"}{height="50mm"}
In Section \[sect:biascorr\] we suggest a method to combine blocks estimators that are based on the exceedances over different thresholds in a suitable way such that the leading bias term cancels out for many well-known time series models. In Figure \[sampleplot\] (c) the resulting estimates based on exceedances over $F_n^\leftarrow(u)$ are shown (again as a function of $u$) for the same block lengths. Obviously, the estimates are not only almost constant for a wide range of thresholds, but they also vary much less with the block length than the original blocks estimator.
The remainder of the paper is organized as follows. In Section 2, we establish a limit result for processes of blocks estimators indexed by the threshold. To this end, we represent the blocks estimators as functionals of a suitably defined empirical cluster process. Then the joint asymptotic behavior of the blocks estimators easily follows from a general limit theorem of such processes proved in Drees and Rootzén (2010). In the main Section \[sect:biascorr\] we first show that often the leading bias term of the blocks estimators is a power function of the threshold. We then introduce a method to remove the leading bias term of the blocks estimators in that case without deteriorating the rate of convergence of the random error part. All proofs are collected in Section \[sect:proofs\].
Joint asymptotics of blocks estimators
======================================
In this section we want to analyze the joint asymptotic behavior of blocks estimators over a whole [*continuum*]{} of thresholds. Since here we are interested in the extremal [*dependence*]{} (and not in the marginal tails), the results should be invariant under strictly increasing transformations of the observations. Hence it is natural to parameterize the thresholds in terms of the marginal quantile function $F^\leftarrow$, that is to consider $$\hat\theta_{n,t}^* := \frac{ \sum_{j=1}^{m_n} {{\boldsymbol{1}_{\textstyle\{
\max_{(j-1)r_n<i\le jr_n}
X_i>F^\leftarrow(1-v_nt)\}}}}}{\sum_{j=1}^{m_n}
\sum_{i=(j-1)r_n+1}^{jr_n}{{\boldsymbol{1}_{\textstyle\{X_i>F^\leftarrow(1-v_nt)\}}}}}, \quad
0<t\le 1.$$
For later applications, though, it is more convenient to examine a version where the unknown quantile function is replaced with an empirical analog: $$\hat\theta_{n,t} := \frac{ \sum_{j=1}^{m_n} {{\boldsymbol{1}_{\textstyle\{
\max_{(j-1)r_n<i\le jr_n} X_i>X_{n-\lceil
nv_nt\rceil:n}\}}}}}{\sum_{j=1}^{m_n}
\sum_{i=(j-1)r_n+1}^{jr_n}{{\boldsymbol{1}_{\textstyle\{X_i>X_{n-\lceil nv_nt\rceil:n}\}}}}},
\quad 0<t\le 1.$$ If there are no ties among the largest $\lceil nv_n\rceil$ observations and none of them are among the last $n-m_nr_n$ observations, then $\hat\theta_{n,t}$ can be rewritten as $$\hat\theta_{n,t} = \frac 1{\lceil nv_nt\rceil} \sum_{j=1}^{m_n} {{\boldsymbol{1}_{\textstyle\{
\max_{(j-1)r_n<i\le jr_n} X_i>X_{n-\lceil nv_nt\rceil:n}\}}}}.$$ In particular, this representation holds with probability tending to 1 if we assume that $F$ is continuous on some neighborhood of $F^\leftarrow(1)$ and $r_nv_n\to 0$, which we will do throughout the remainder of the paper.
For sufficiently large $n$, we then have $$\hat\theta_{n,t}^* = \frac{ \sum_{j=1}^{m_n} {{\boldsymbol{1}_{\textstyle\{
\max_{(j-1)r_n<i\le jr_n} U_i>1-v_nt\}}}}}{\sum_{j=1}^{m_n}
\sum_{i=(j-1)r_n+1}^{jr_n}{{\boldsymbol{1}_{\textstyle\{U_i>1-v_nt\}}}}}$$ where the random variables $U_i=F(X_i)$, $1\le i\le n$, have a distribution which equals the uniform distribution in a neighborhood of 1. Thus this blocks estimator can be expressed in terms of certain empirical processes of cluster functionals that have been introduced and analyzed by Drees and Rootzén (2010). To this end, define standardized excesses $$U_{n,i} := \frac {(U_i-(1-v_n))^+}{v_n} = \frac
{(U_i-(1-v_n))\vee 0}{v_n}, \quad 1\le i\le n,$$ blocks thereof $$Y_{n,j} := (U_{n,i})_{(j-1)r_n<i\le j r_n}, \quad 1\le j\le m_n,$$ and functionals on ${\mathbb{R}}_\cup:=\bigcup_{l\in{\mathbb{N}}} {\mathbb{R}}^l$ by $$\begin{aligned}
f_t(x_1,\ldots,x_l) & := & {{\boldsymbol{1}_{\textstyle\{\max_{1\le i\le l}
x_i>1-t\}}}}\\
g_t(x_1,\ldots,x_l) & := & \sum_{i=1}^l {{\boldsymbol{1}_{\textstyle\{
x_i>1-t\}}}}.\end{aligned}$$ Then $$\hat\theta_{n,t}^* = \frac{m_n^{-1} \sum_{j=1}^{m_n}
f_t(Y_{n,j})}{m_n^{-1} \sum_{j=1}^{m_n}
g_t(Y_{n,j})} = \frac{E(f_t(Y_{n,1}))+(nv_n)^{1/2} m_n^{-1} Z_n(f_t)}{E(g_t(Y_{n,1}))+(nv_n)^{1/2} m_n^{-1}
Z_n(g_t)},
\label{eq:empprocrep}$$ where for a generic functional $h:{\mathbb{R}}_\cup\to{\mathbb{R}}$ we define $$Z_n(h) := \frac 1{\sqrt{nv_n}} \sum_{j=1}^{m_n}\big( h(Y_{n,j})-E
h(Y_{n,j})\big).$$ Under suitable conditions on the time series and the family $\mathcal{H}$ of functionals $h$, Drees and Rootzén (2010) proved convergence of the empirical processes $(Z_n(h))_{h\in\mathcal{H}}$ to a centered Gaussian process with continuous sample paths.
Here we recall conditions that ensure the convergence of the processes $\big(Z_n(f_t),Z_n(g_t)\big)_{0\le t\le 1}$. Note that $(Z_n(g_t))_{0\le t\le 1}$ is the usual tail empirical process, whose asymptotic behavior has been investigated by Rootzén (1995, 2009) and Drees (2000).
\[theo:empproc\]
1. Under the conditions (C1) and (C2), $(Z_n(f_t))_{0\le
t\le 1}$ converge weakly to $Z_f:=(\sqrt{\theta} B_t)_{0\le
t\le 1}$ with $B$ denoting a standard Brownian motion.
2. If the conditions (C1), (C2), (C3.1) and (C4) are met and $r_n=o(\sqrt{nv_n})$, then $(Z_n(g_t))_{0\le t\le 1}$ converge to a centered Gaussian process $(Z(g_t))_{0\le t\le 1}$ with covariance function $c_g$.
3. If the conditions (C1)–(C4) are satisfied and $r_n=o(\sqrt{nv_n})$, then $(Z_n(f_t),Z_n(g_t))_{0\le t\le
1}$ converge weakly to $(Z_f(t),Z_g(t))_{0\le t\le
1}$ with $$\begin{aligned}
Cov(Z_f(s),Z_f(t)) & = & \theta(s\wedge t),\\
Cov(Z_g(s),Z_g(t)) & = & c_g(s,t),\\
Cov(Z_f(s),Z_g(t)) & = & c_{fg}(s,t),\quad 0\le s,t\le 1.
\end{aligned}$$
The covariance conditions (C3.1) and (C3.2) are fulfilled if all finite dimensional marginal distributions $(X_1,\ldots,X_k)$ belong to the domain of attraction of some multivariate extreme value distribution, $\lim_{n\to\infty} \limsup_{m\to\infty} \beta_{n,m}=0$, and the following condition holds:
In this case, Segers (2003) has shown that the conditional distributions $P^{(U_{n,i})_{1\le i\le k}|U_{n,1}\ne 0}$ of $(U_{n,i})_{1\le i\le
k}$ given that the first observation exceeds the threshold converge weakly to the distribution of $(W_i)_{1\le i\le
k}=(V_i\vee 0)_{1\le i\le k}$, where $(V_i)_{1\le i\le k}$ is the so-called tail sequence pertaining to the time series $U_i$, $i\in{\mathbb{N}}$. The limiting covariance functions $c_g$ and $c_{fg}$ are then given by $$\begin{aligned}
c_g(s,t) & = & s\wedge t + \sum_{k=2}^\infty\big( P\{W_1>1-s,
W_k>1-t\}+P\{W_1>1-t, W_k>1-s\} \big), \label{eq:cg}\\
c_{fg}(s,t) & = & \left\{ \begin{array}{l@{\quad}l}
P\{W_1>1-t,\max_{j\ge 1}W_j>1-s\}\\
\hspace*{1em}+\sum_{k=2}^\infty P\{W_1>1-s,W_k>1-t,\max_{j\ge 2}W_j\le
1-s\}, & s<t,\\
t & s\ge t.
\end{array} \right. \label{eq:cfg}
\end{aligned}$$ [ ]{}
Using the joint convergence of $Z_n(f_t)$ and $Z_n(g_t)$ and the representation , one can easily derive a limit theorem for the processes $(\hat\theta_{n,t}^*)_{0< t\le 1}$ of blocks estimators.
\[cor:blockasymp\] Under the conditions of Theorem \[theo:empproc\] (iii) $$\big( \sqrt{nv_n}
t(\hat\theta_{n,t}^*-\theta_{n,t})\big)_{0<t\le 1} \;\to\;
Z:=Z_f-\theta Z_g \quad\text{weakly as } n\to\infty$$ with $$\theta_{n,t} := \frac{E(f_t(Y_{n,1}))}{E(g_t(Y_{n,1}))} =
\frac{P\{\max_{1\le i\le r_n} X_i>F^\leftarrow (1-v_nt)\}}{r_nv_nt}.$$ The limit process $Z$ is Gaussian with $E(Z(t))=0$ and $$\label{eq:covdef}
Cov(Z(s),Z(t)) = \theta\big(s\wedge
t-c_{fg}(s,t)-c_{fg}(t,s)\big)+\theta^2 c_g(s,t) =: c(s,t).$$
Note that the centering constant $\theta_{n,t}$, which is the leading term in the representation , converges to $\theta$ uniformly for all $t\in(0,1]$ by Hsing’s (1993) result . However, the convergence can be rather slow leading to a large bias of the blocks estimator as observed in Figure \[sampleplot\].
In the next section we will see how to combine all blocks estimators $\hat\theta_{n,t}^*$ non-linearly such that the resulting estimator has a much smaller bias. As the threshold $F^\leftarrow(1-v_nt)$ is unknown, for any given threshold $u_n$ in the definition it is not known for which index $t$ one has $\hat\theta_n =\hat\theta_{n,t}^*$. Hence, we first need an analog to Corollary \[cor:blockasymp\] for the estimator $\hat\theta_{n,t}$ with random threshold $X_{n-\lceil
nv_nt\rceil:n}$.
To this end, we analyze the difference between the deterministic threshold $1-v_nt$ (after standardization of the marginals) and its random counterpart $1-U_{n-\lceil nv_nt\rceil:n}$. It has been shown in Drees (2000), proof of Corollary 3.1, that $\sqrt{nv_n}\big((1-U_{n-\lceil nv_nt\rceil:n})/v_n-t\big)_{0\le
t\le 1}$ converges to a Gaussian process if $(Z_n(g_t))_{-\eps\le
t\le 1}$ converges to a Gaussian process. Note that the latter convergence follows from an analog to Theorem \[theo:empproc\] (ii), because the conditions (C3.1) and (C4) have been formulated for $s,t\in[-\eps,1]$ (while for Theorem \[theo:empproc\] (ii) to hold it suffices to require the conditions for $s,t\in[0,1]$). This suffices to establish a limit theorem for $\hat\theta_{n,t}$. It turns out that under a suitable continuity condition on $\theta_{n,t}$, the blocks estimator with estimated threshold has the same asymptotic behavior as $\hat\theta_{n,t}^*$.
\[cor:blockasympest\] Suppose the conditions of Theorem \[theo:empproc\] (iii) are met. Then $$\label{eq:blockpreasympest}
\big( \sqrt{nv_n}
t(\hat\theta_{n,t}-\theta_{n,(1-U_{n-\lceil
nv_nt\rceil:n})/v_n})\big)_{0\le t\le 1} \;\to\;
Z \quad\text{weakly as } n\to\infty.$$
If, in addition, to each $t_0\in (0,1)$ and each $M_1>0$ there exists $M_2>0$ such that $$\label{eq:thetantcont}
\sup_{s,t\ge t_0, |s-t|\le M_1(nv_n)^{-1/2}} \Big|
\frac{\theta_{n,s}-\theta}{\theta_{n,t}-\theta}-1\Big| \le
M_2(nv_n)^{-1/2},$$ then $$\label{eq:blockasympest}
\big( \sqrt{nv_n}
t(\hat\theta_{n,t}-\theta_{n,t})\big)_{0\le t\le 1} \;\to\;
Z \quad\text{weakly as } n\to\infty.$$
Bias correction {#sect:biascorr}
===============
As in Figure \[sampleplot\], the blocks estimator $\hat\theta_{n,t}$ often exhibits a clear trend, that is caused by its bias, when it is plotted versus $t$. In this section we show how to combine blocks estimators for different thresholds such that the leading bias term vanishes while the order of magnitude of the random error is preserved. To this end, we make structural assumptions on the form of the bias $\theta_{n,t}-\theta$ as a function of $t$. The following examples demonstrate that in time series models discussed in the literature the leading bias term often equals a power of $t$ with positive exponent.
\[ex:Weissman\] Let $Z_i$, $i\in{\mathbb{N}}$, be iid r.v.s with d.f. $F$, and let $\xi_i$, $i\in{\mathbb{N}}$, denote a series of iid Bernoulli rvs, independent of $(Z_i)_{i\in{\mathbb{N}}}$, with $P\{\xi_i=0\}=\psi=1-P\{\xi_i=1\}$. Weissman and Novak (1998, p. 285) proved that then the time series $X_0:=Z_0$, $X_t:=\xi_t
Z_t+(1-\xi_t)X_{t-1}$, $t\in{\mathbb{N}}$, is stationary with marginal d.f. $F$ and extremal index $\theta=1-\psi$. Moreover, if $F$ is eventually continuous, then for all $t_0>0$ $$\theta_{n,t} = \frac{1-(1-v_nt)(1-\theta v_nt)^{r_n-1}}{r_nv_n
t} = \theta -\frac{\theta^2}2 r_nv_n t+\frac{1-\theta}{r_n} +
O(v_n+r_n^2v_n^2)$$ uniformly for $t\in[t_0,1]$. If $r_n^2v_n\to\infty$, then the linear function $-\theta^2 r_nv_n
t/2$ is the leading bias term.
\[ex:movingmaxima\] Consider a finite order moving maxima time series $$X_t= \max_{0\le j\le q} (\psi_jZ_{t-j})$$ with non-negative coefficients $\psi_j\ge 0$. W.l.o.g. we may and will assume that $ \max_{0\le j\le q} \psi_j=1$. Further assume that the innovations $Z_t$ are iid with heavy tailed d.f. $F_Z$ satisfying $$\bar F_Z(z) := 1-F_Z(z)= c_1z^{-\beta_1}\big(1+c_2
z^{-\beta_2} + o(z^{-\beta_2})\big)$$ for some $\beta_1,\beta_2,c_1>0$ and $c_2\ne 0$.
If $\beta_2<\beta_1$, then $$\begin{aligned}
F(x) & := & P\{X_t\le x\}\\
& = & P\{Z_{t-j}\le x/\psi_j\; \forall\, 0\le j\le q\} \\
& = &
\prod_{j=0}^q\Big(1-c_1(x/\psi_j)^{-\beta_1}\big(1+c_2(x/\psi_j)^{-\beta_2}+o(x^{-\beta_2})\big)\Big)\\
& = & 1- c_1 \sum_{j=0}^q \psi_j^{\beta_1}x^{-\beta_1} - c_1c_2
\sum_{j=0}^q \psi_j^{\beta_1+\beta_2} x^{-(\beta_1+\beta_2)} +
o\big(x^{-(\beta_1+\beta_2)}\big)
\end{aligned}$$ as $x\to\infty$, and thus for all fixed $A>0$ $$\begin{aligned}
\bar F_Z(x/A) & = & \frac{A^{\beta_1}}{\sum_{j=0}^q
\psi_j^{\beta_1}} \bar F(x) + \frac{c_2}{c_1^{\beta_2/\beta_1}}
\frac{A^{\beta_1+\beta_2}}{\big(\sum_{j=0}^q
\psi_j^{\beta_1}\big)^{1+\beta_2/\beta_1}} \Big( 1- \frac{\sum_{j=0}^q
\psi_j^{\beta_1+\beta_2}}{\sum_{j=0}^q
\psi_j^{\beta_1}A^{\beta_2}}\Big) (\bar F(x))^{1+\beta_2/\beta_1}\\
& & +
o(x^{(-\beta_1+\beta_2)}).
\end{aligned}$$
To determine $\theta_{n,t}$, check that with $$d := \frac{c_2}{c_1^{\beta_2/\beta_1}}
\frac1{\big(\sum_{j=0}^q
\psi_j^{\beta_1}\big)^{1+\beta_2/\beta_1}} \Big( 1- \frac{\sum_{j=0}^q
\psi_j^{\beta_1+\beta_2}}{\sum_{j=0}^q
\psi_j^{\beta_1}}\Big)$$ it follows that $$\begin{aligned}
\lefteqn{P\big\{\max_{1\le t\le r_n} X_t\le F^\leftarrow(1-v_n
t)\big\}}\\
& = & P\Big\{ Z_{t-j} \le \frac{F^\leftarrow(1-v_nt)}{\psi_j} \;
\forall\, 1\le t\le r_n, 0\le j\le q\Big\} \\
& = & P\Big\{ Z_m \le \frac{F^\leftarrow(1-v_nt)}{\max_{0\vee (1-m)\le
j\le q \wedge (r_n-m)} \psi_j}\; \forall\, 1-q\le m\le r_n\Big\}
\\
& = & \prod_{m=1-q}^0
F_Z\Big(\frac{F^\leftarrow(1-v_nt)}{\max_{1-m\le j\le q}
\psi_j}\Big) \cdot \prod_{m=1}^{r_n-q}
F_Z\Big(\frac{F^\leftarrow(1-v_nt)}{\max_{0\le j\le q}
\psi_j}\Big) \cdot \prod_{m=r_n-q+1}^{r_n}
F_Z\Big(\frac{F^\leftarrow(1-v_nt)}{\max_{0\le j\le r_n-m}
\psi_j}\Big)\\
& = & \prod_{m=1-q}^0 (1+O(v_n))\cdot \bigg( 1- \frac
1{\sum_{j=0}^q
\psi_j^{\beta_1}} v_n t - d (v_nt)^{1+\beta_2/\beta_1}+
o(v_n^{1+\beta_2/\beta_1})\Big)^{r_n-q}\\
& & \cdot \prod_{m=r_n-q+1}^{r_n}(1+O(v_n))\\
& = & 1- \frac 1{\sum_{j=0}^q
\psi_j^{\beta_1}} r_n v_n t - dr_n(v_nt)^{1+\beta_2/\beta_1}+
O(v_n+r_n^2v_n^2)+
o(r_nv_n^{1+\beta_2/\beta_1}).
\end{aligned}$$ Hence, if $r_nv_n^{\beta_2/\beta_1}\to\infty$ but $r_nv_n^{1-\beta_2/\beta_1}\to 0$ (which implies $\beta_2<\beta_1/2$), then for all $t_0>0$ $$\theta_{n,t} = \frac{1-P\big\{\max_{1\le t\le r_n} X_t\le F^\leftarrow(1-v_n
t)\big\}}{r_nv_n t} = \frac 1{\sum_{j=0}^q \psi_j^{\beta_1}} + d(v_nt)^{\beta_2/\beta_1} +
o(v_n^{\beta_2/\beta_1})$$ uniformly for $t\in [t_0,1]$. Here the the constant $d$ is strictly negative if $\psi_j\in (0,1)$ for some $j\in\{0,\ldots,q\}$. Hence, in this case, $\theta:= 1/\sum_{j=0}^q \psi_j^{\beta_1}$ is the extremal index and the leading term of the bias $\theta_{n,t}-\theta$ is a multiple of $t^{\beta_2/\beta_1}$.
Now we investigate the general case, i.e. we do not assume that $\beta_2<\beta_1$. By similar calculations as above, we obtain that $$\begin{aligned}
\lefteqn{\bar F(x) = 1+c_1 \sum_{j=0}^q \psi_j^{\beta_1}
x^{-\beta_1} + O(x^{-(\beta_1+\beta_2)}+x^{-2\beta_1})}\\
& \Longrightarrow & \bar F_Z(x/A) = \frac{A^{\beta_1}}{\sum_{j=0}^q
\psi_j^{\beta_1}} \bar F(x) +
O(x^{-(\beta_1+\beta_2)}+x^{-2\beta_1}).
\end{aligned}$$ Therefore $$\begin{aligned}
\lefteqn{P\big\{\max_{1\le t\le r_n} X_t\le F^\leftarrow(1-v_n
t)\big\}}\\
& = & (1+O(v_n))^{2q}\cdot\Big( 1-\frac{v_nt}{\sum_{j=0}^q
\psi_j^{\beta_1}} +
O(v_n^{1+\beta_2/\beta_1}+v_n^2)\Big)^{r_n-q}\\
& = & 1- \frac{r_nv_nt}{\sum_{j=0}^q \psi_j^{\beta_1}} + \frac
12\bigg(\frac{r_nv_nt}{\sum_{j=0}^q \psi_j^{\beta_1}}\bigg)^2 +
O\big(v_n+(r_nv_n)^3+r_n(v_n^{1+\beta_2/\beta_1}+v_n^2)\big),
\end{aligned}$$ which in turn implies $$\theta_{n,t} = \theta-\frac{\theta^2}2r_nv_nt+o(r_nv_n)$$ if $r_nv_n^{\max(1/2,1-\beta_2/\beta_1)}\to \infty$. Hence, in this case the leading bias term is a linear function of $t$.
Theorem 4.1 of Hsing (1993) suggests that indeed for $m$-dependent time series with $m$-dimensional regularly varying marginal distributions the leading bias term usually is a linear function of $t$ if $r_n\to\infty$ sufficiently fast.
We propose the following estimator of the extremal index with reduced bias: $$\label{eq:redbiasest}
\hat \theta_{n,\mu} := \frac{\displaystyle \int_{(0,1]^2} \hat\theta_{n,s}
\hat\theta_{n,t}\, \mu(ds,dt)}{\displaystyle \int_{(0,1]^2} \hat\theta_{n,s}+
\hat\theta_{n,t}\, \mu(ds,dt)},$$ where $\mu$ is some finite signed measure on $(0,1]^2$ satisfying the following conditions:
\[ex:mu\]
1. Let $F,G$ be d.f.s of probability measures $Q_F$ and $Q_G$ on $(0,1]$ such that $\int_{(0,1]} t^{-1} \, Q_F(dt)<\infty$, $\int_{(0,1]} t^{-1} \, Q_G(dt)<\infty$ and $\int_{(0,1]} t^\delta \, Q_F(dt)\ne \int_{(0,1]} t^\delta \,
Q_G(dt)$ for all $\delta>0$. (The latter condition is, for instance, fulfilled if $Q_F$ equals the distribution $Q_G^{T_b}$ of the map $T_b(x):=x/b$ under $Q_G$ for some $b>1$.) Then the signed measure $\mu=Q_F^{T_a}\otimes Q_G-Q_F\otimes Q_G^{T_a}$ for some $a>1$ (i.e., $\mu\big((0,x]\times(0,y]\big)=F(ax)G(y)-F(x)G(ay)$ for all $x,y\in(0,1]$) satisfies the conditions (M1)–(M3): $$\begin{aligned}
\mu\{(s,t)\mid st\le u\} & = & \int_{(0,1]} G(u/s)\, Q_F^{T_a}(ds)-\int_{(0,1]} G(au/s)\, Q_F(ds)=0,\\
\int_{(0,1]^2} s^\delta+t^\delta\, \mu(ds,dt)
& = & \int_{(0,1]} s^\delta\, Q_F^{T_a}(ds) + \int_{(0,1]} t^\delta\, Q_G(dt)
-\int_{(0,1]} s^\delta\, Q_F(ds) - \int_{(0,1]} t^\delta\, Q_G^{T_a}(dt)\\
& = & (a^{-\delta}-1) \Big(\int_{(0,1]} s^\delta\, Q_F(ds)-\int_{(0,1]} t^\delta\, Q_G(dt)\Big)\\
& \ne & 0,\\
\int_{(0,1]^2} (st)^{-1}\, |\mu|(ds,dt) & = & \int_{(0,1]} s^{-1}\,
Q_F^{T_a}(ds) \int_{(0,1]} t^{-1}\, Q_G(dt)+ \int_{(0,1]} s^{-1}\,
Q_F(ds) \int_{(0,1]} t^{-1}\, Q_G^{T_a}(dt)\\
& = & (1+a)\int_{(0,1]} s^{-1}\,
Q_F(ds)\cdot \int_{(0,1]} t^{-1}\, Q_G(dt)<\infty.
\end{aligned}$$
2. The above example is a special case of the following more general construction. Let $T:(0,1]^2\to D:=\{(u,v)\mid 0<u\le
v\le 1\}$, $T(x,y):=(xy,y)$, and let $T^{-1}:D\to (0,1]^2$, $T^{-1}(u,v)=(u/v,v)$ denote its inverse. Choose some measure $\nu$ on $(0,1]$ satisfying $\int_{(0,1]} s^{-1}\,\nu(ds)<\infty$, and Markov kernels $K_1$ and $K_2$ from $(0,1]$ to $(0,1]$ such that $K_i(u,[u,1])=1$. Then the signed measure $\mu:=(\nu\otimes K_1)^{T^{-1}}-(\nu\otimes
K_2)^{T^{-1}}$ meets the conditions (M1) and (M3), because $\pi=T\circ pr_1$ with $pr_1$ denoting the projection on the first coordinate, and thus $\mu^\pi=\Big(\big((\nu\otimes
K_1)^{T^{-1}}\big)^T\Big)^{pr_1}-\Big(\big((\nu\otimes
K_2)^{T^{-1}}\big)^T\Big)^{pr_1} = \nu-\nu=0$ and $\int_{(0,1]^2}
(st)^{-1} \, (\nu\otimes K_i)^{T^{-1}}(ds,dt) = \int_{(0,1]^2} u^{-1}\,
(\nu\otimes K_i) (du,dv)<\infty$.
Our main result shows that the bias of $\hat\theta_{n,\mu}$ (and hence its estimation error) is of smaller order than the bias of $\theta_{n,t}$ if the bias dominates the random error and its leading term is a power function.
\[theo:main\] Suppose that conclusion of Corollary \[cor:blockasympest\] holds and that $$\label{eq:biascond}
\theta_{n,t}=\theta_n+c_nt^\delta + R_n(t) \quad \forall\,
t\in (0,1]$$ for some $\delta>0$ with $d_n:=\sup_{0< t\le 1}t|R_n(t)|=o(c_n)$ and $(nv_n)^{-1/2}=o(c_n)$. If the conditions (M1)–(M3) are fulfilled, then $$\begin{aligned}
\hat\theta_{n,\mu} & =^d & \theta_n+(nv_n)^{-1/2}
\frac{\displaystyle \int_{(0,1]^2} s^\delta t^{-1} Z(t)+t^\delta s^{-1} Z(s)\,
\mu(ds,dt)}{\displaystyle \int_{(0,1]^2} s^\delta +t^\delta\,
\mu(ds,dt)}\\
& &
+ \frac{\displaystyle \int_{(0,1]^2} s^\delta R_n(t)+t^\delta R_n(s)\,
\mu(ds,dt)}{\displaystyle \int_{(0,1]^2} s^\delta +t^\delta\,
\mu(ds,dt)} + o_P((nv_n)^{-1/2}+d_n).
\end{aligned}$$ In particular, if $d_n=o((nv_n)^{-1/2})$, then $$\label{eq:main}
\sqrt{nv_n}(\hat\theta_{n,\mu} -\theta_n) \;\longrightarrow\;
\frac{\displaystyle \int_{(0,1]^2} s^\delta t^{-1} Z(t)+t^\delta s^{-1} Z(s)\,
\mu(ds,dt)}{\displaystyle \int_{(0,1]^2} s^\delta +t^\delta\,
\mu(ds,dt)}.$$
\[rem:mainrem\] If $\sup_{0< t\le 1}|R_n(t)|=o\big((nv_n)^{-1/2}\big)$, then assertion holds if merely convergence is required instead of , that is, the smoothness assumption on $\theta_{n,t}$ is not needed.
In the leading bias term which depends on the threshold is removed, while the random error is still of the order $(nv_n)^{-1/2}$. To analyze the latter, w.l.o.g. we may assume that the signed measure $\mu$ is symmetric, because $\hat\theta_{n,\mu}=\hat\theta_{n,\tilde\mu}$ for $\tilde \mu(ds,dt):=\mu(ds,dt)+\mu(dt,ds)$ and $\tilde\mu$ satisfies (M1)–(M3) iff $\mu$ meets these conditions. Then the right-hand side of equals $$\frac{\displaystyle \int_{(0,1]^2} s^\delta t^{-1} Z(t)\,
\mu(ds,dt)}{\displaystyle \int_{(0,1]^2} s^\delta\,
\mu(ds,dt)}$$ which is a centered Gaussian rv with variance $$\sigma^2_\mu :=
\frac{\displaystyle \int_{(0,1]^2} \int_{(0,1]^2} (s\tilde s)^\delta (t\tilde t)^{-1} c(t,\tilde t)\,
\mu(ds,dt)\,\mu(d\tilde s,d\tilde t)}{\Big(\displaystyle \int_{(0,1]^2} s^\delta\,
\mu(ds,dt)\Big)^2}.$$ If $\mu$ is the symmetrized version of the signed measure discussed in Example \[ex:mu\] (i) with $f$ and $g$ denoting Lebesgue densities of $Q_F$ and $Q_G:=Q_F^{T_b}$, respectively, then $$\begin{aligned}
\sigma^2_\mu & = &
\bigg(\frac{ab}{(1-a^{-\delta})(1-b^{-\delta})}\bigg)^2\times\\
& & \times
\int_0^1\int_0^1 \Big(
a^{-(\delta+1)}f(bt)+b^{-(\delta+1)}f(at)-f(abt)-(ab)^{-(\delta+1)}f(t)\Big)\times\\
& & \hspace*{1cm} \times
\Big(
a^{-(\delta+1)}f(b\tilde t)+b^{-(\delta+1)}f(a\tilde t)-f(ab\tilde t)-(ab)^{-(\delta+1)}f(\tilde
t)\Big)(t\tilde t)^{-1}c(t,\tilde t)\, dt\,d\tilde t.
\end{aligned}$$ To estimate this asymptotic variance is essentially as difficult as to determine the asymptotic variance of the original blocks estimators. To this end, one may employ ideas developed in Drees (2003), but a bootstrap approach, that will be worked out in a forthcoming paper, seems more promising.
Finally, we would like to mention that our approach is obviously not capable of removing the part $\theta_n-\theta$ of the bias which does not depend on the threshold but on the block length $r_n$.
Proofs {#sect:proofs}
======
Theorem \[theo:empproc\]. We apply Theorem 2.10 of Drees and Rootzén (2010) to prove asymptotic equicontinuity of the processes and Theorem 2.3 to establish convergence of the finite dimensional marginal distributions. To this end, we must verify the conditions required in these theorems.
\(i) The assumptions (B1) and (B2) of Drees and Rootzén (2010) follow from our conditions (C1) and (C2). For the functionals $f_t$, condition (C2) of Drees and Rootzén (2010) is trivial. Condition (C3) of Drees and Rootzén (2010) reads as $$\frac{P\{\max_{1\le i\le r_n} U_{n,i}>(1-s)\vee
(1-t)\}}{r_nv_n} \;\to\; \theta(s\wedge t)$$ (cf. Drees and Rootzén (2010), (4.1)). This is immediate from , which implies $$\label{eq:recimeannew}
\frac{P\{\max_{1\le i\le r_n}U_{n,i}>1-t\}}{r_n v_n t} \;\to\; \theta$$ uniformly for $t\in (0,1]$.
Likewise, condition (D3) of Drees and Rootzén (2010) is equivalent to $$\lim_{\delta\downarrow 0} \limsup_{n\to\infty} \sup_{0\le s\le t\le 1, t-s\le\delta}
\frac{P\{1-t<\max_{1\le i\le r_n} U_{n,i}\le 1-s\}}{r_nv_n}=0,$$ which again is a direct consequence of the uniform convergence .
The remaining conditions can be verified by the arguments given in Drees and Rootzén (2010), Section 4 and the proof of Corollary 4.3. (Note that $Z_n(f_t)$ equals the random variable $\tilde Z_n(1-t)$ defined in Example 4.2 (with $k=1$) of that paper.)
\(ii) This assertion is a reformulation of the results on the univariate tail empirical process given in Example 3.8 of Drees and Rootzén (2010).
\(iii) The equicontinuity of the joint process immediately follows from the equicontinuity of $(Z_n(f_t))_{0\le t\le 1}$ and $(Z_n(g_t))_{0\le t\le
1}$ and a similar remark applies to the conditions (C1) and (C2) of Drees and Rootzén (2010). The remaining condition (C3) follows from (C3.1) and (C3.2) of the present paper and the calculations in part (i) above.
Remark 2.2. The conditions (C3.1) and (C3.2) follow by similar arguments as in Remark 3.7 (ii) of Drees and Rootzén (2010) (cf. also Corollary 2.4 of that paper). Here we have $$\begin{aligned}
c_g(s,t) & = & E\Big(
{{\boldsymbol{1}_{\textstyle (1-s,1]}}}(W_1){{\boldsymbol{1}_{\textstyle (1-t,1]}}}(W_1)\\
& & \hspace*{1cm} { } +\sum_{k=2}^\infty
{{\boldsymbol{1}_{\textstyle (1-s,1]}}}(W_1){{\boldsymbol{1}_{\textstyle (1-t,1]}}}(W_k)+{{\boldsymbol{1}_{\textstyle (1-t,1]}}}(W_1){{\boldsymbol{1}_{\textstyle (1-s,1]}}}(W_k)\Big),
\end{aligned}$$ which equals the right hand side of , and $$c_{fg}(s,t) = E\Big(
{{\boldsymbol{1}_{\textstyle (1-s,1]}}}(\max_{i\ge 1}W_i)\sum_{k=1}^\infty
{{\boldsymbol{1}_{\textstyle (1-t,1]}}}(W_k)-{{\boldsymbol{1}_{\textstyle (1-s,1]}}}(\max_{i\ge 2}W_i)\sum_{k=2}^\infty
{{\boldsymbol{1}_{\textstyle (1-t,1]}}}(W_k)\Big).$$ If $s\ge t$ and the first sum does not vanish, then the first indicator equals 1. Together with a similar reasoning for the second sum, one obtains $$c_{fg}(s,t)=E({{\boldsymbol{1}_{\textstyle (1-t,1]}}}(W_1))=t.$$ In the case $s<t$, direct calculations show that $$\begin{aligned}
c_{fg}(s,t) & = & E\Big(
{{\boldsymbol{1}_{\textstyle (1-s,1]}}}(\max_{i\ge 1}W_i) {{\boldsymbol{1}_{\textstyle (1-t,1]}}}(W_1)\\
& & \hspace*{1cm} { } +\big({{\boldsymbol{1}_{\textstyle (1-s,1]}}}(\max_{i\ge 1}W_i)-{{\boldsymbol{1}_{\textstyle (1-s,1]}}}(\max_{i\ge 2}W_i)\big)
\sum_{k=2}^\infty {{\boldsymbol{1}_{\textstyle (1-t,1]}}}(W_k)\Big)
\end{aligned}$$ is equal to the right hand side of .
Corollary \[cor:blockasymp\]. Using $E(g_t(Y_{n,1}))=r_nv_nt$ and representation , we obtain by simple calculations $$\label{eq:errorexpan}
\sqrt{nv_n}t(\hat\theta_{n,t}^*-\theta_{n,t}) =
\frac{n}{m_nr_n} \cdot \frac{Z_n(f_t)-\theta_{n,t}Z_n(g_t)}{1+
\sqrt{nv_n}/(m_nr_nv_nt) Z_n(g_t)}.$$ By the equicontinuity of $(Z_n(g_t))_{0\le t\le 1}$ and $Z_n(g_0)=0$, there exists a sequence $\eta_n\to 0$ such that $\sup_{0\le t\le (nv_n)^{-1/4}}
|Z_n(g_t)|=O_P(\eta_n)$. Hence, for $t_n:=(\eta_n/(nv_n))^{1/2}$, $$\sup_{t_n\le t\le 1} \frac{|Z_n(g_t)|}{\sqrt{nv_n}t}
=O_P\Big(\frac{\eta_n}{\sqrt{nv_n}t_n}\Big)+\sup_{(nv_n)^{-1/4}\le
t\le 1} \frac{|Z_n(g_t)|}{(nv_n)^{1/4}}=o_P(1),$$ so that the denominator of the second fraction tends to 1 uniformly for $t\in[t_n,1]$. Moreover, since both $\hat\theta_{n,t}^*$ and $\theta_{n,t}$ are bounded, $$\label{eq:thetabd1}
\sup_{0\le t\le t_n}
\sqrt{nv_n}t|\hat\theta_{n,t}^*-\theta_{n,t}|=o_P(1).$$ Finally, the continuity of $Z$ implies $$\label{eq:Zbd1}
\sup_{0\le t\le t_n} |Z(t)|=o_P(1).$$ Therefore, in view of –, Theorem \[theo:empproc\] and the uniform convergence of $\theta_{n,t}$ to $\theta$ prove the assertion.
Corollary \[cor:blockasympest\]. Check that under the conditions of Theorem \[theo:empproc\] (iii) the following equivalences hold on a set with probability tending to 1: $X_i>X_{n-\lceil nv_nt\rceil:n} \iff U_i>U_{n-\lceil nv_nt\rceil:n}
\iff U_{n,i} > 1-(1-U_{n-\lceil nv_nt\rceil:n})/v_n$, and thus $\hat\theta_{n,t}=\hat\theta^*_{n,s_n(t)}$ with $s_n(t) := (1-U_{n-\lceil
nv_nt\rceil:n})/v_n$. An application of Vervaat’s (1972) Theorem 1 to the assertion of Theorem \[theo:empproc\] (ii) yields $$\label{eq:tailempproc}
\sqrt{nv_n} (s_n(t)-t)_{0\le t\le 1} \;\longrightarrow\; Z_g$$ (cf. the proof of Corollary 3.1 of Drees (2000)). In particular, $s_n(t)/t\to 1$ uniformly for all $t\in[(nv_n)^{-1/3},1]$.
Moreover, by continuity $\sup_{0\le t\le (nv_n)^{-1/3}}
|Z(t)|\to 0$ and thus by Corollary \[cor:blockasymp\] $$\begin{aligned}
\sqrt{nv_n}t\big(\hat\theta_{n,t}-\theta_{n,s_n(t)}\big)1_{[(nv_n)^{-1/3},1]}(t) & = &
\sqrt{nv_n}s_n(t)\big(\hat\theta_{n,s_n(t)}^*-\theta_{n,s_n(t)}\big)\cdot \frac{t}{s_n(t)}1_{[(nv_n)^{-1/3},1]}(t)
\nonumber\\
&\to& Z(t) \label{eq:thetahatconv1}
\end{aligned}$$ uniformly for $t\in[0,1]$.
Next note that $$\label{eq:thetahatconv2}
\sup_{0\le t\le (nv_n)^{-3/4}} \sqrt{nv_n} t
|\hat\theta_{n,t}-\theta_{n,s_n(t)}| \le (nv_n)^{-1/4} \;\longrightarrow\;
0,$$ while for $(nv_n)^{-3/4}\le t\le (nv_n)^{-1/3}$ $$\begin{aligned}
\lefteqn{\sqrt{nv_n} |\hat\theta_{n,t}-\theta_{n,s_n(t)}|} \nonumber \\
& = & \Big|\frac{nv_nt}{\lceil nv_nt\rceil}\Big(
Z_n(f_{s_n(t)}) + \frac{m_n}{\sqrt{nv_n}} P\big\{\max_{1\le
i\le r_n} X_i>F^\leftarrow(1-v_ns)\big\}|_{s=s_n(t)}\Big) -
\sqrt{nv_n} t\theta_{n,s_n(t)}\Big| \nonumber \\
& \le & | Z_n(f_{s_n(t)})| + \Big|\frac{nv_nt}{\lceil
nv_nt\rceil} \cdot \frac{m_nr_n}n -1\Big| \sqrt{nv_n} s_n(t) \theta_{n,s_n(t)}
+ \sqrt{nv_n} |s_n(t)-t| \theta_{n,s_n(t)}.
\label{eq:thetahatconv3}
\end{aligned}$$ The first term on the right-hand side tends to 0 uniformly by Theorem \[theo:empproc\] (i) and the continuity of $Z_f$, the last term converges to 0 by and the continuity of $Z_g$. Furthermore, by $$\sup_{(nv_n)^{-3/4} \le t\le (nv_n)^{-1/3}} \Big|\frac{nv_nt}{\lceil
nv_nt\rceil} \cdot \frac{m_nr_n}n -1\Big| \sqrt{nv_n} s_n(t) =
O_P\big( (nv_n)^{-1/4}+ r_n/n\big)\cdot
O_P\big((nv_n)^{1/6}\big) \to 0.$$ Combining this with –, we arrive at the first assertion.
It remains to prove that under the additional continuity condition on $\theta_{n,t}$ $$\sqrt{nv_n}\sup_{0\le t\le
1}t|\theta_{n,s_n(t)}-\theta_{n,t}|\;\longrightarrow\; 0$$ in probability. To this end, first check that $$\begin{aligned}
|\theta_{n,s}-\theta_{n,t} | & \le & \Big|\frac 1{r_nv_n s}-\frac 1{r_nv_n
t}\Big| P\big\{\max_{1\le i\le r_n} X_i>F^\leftarrow(1-v_ns)\big\}\\
& & \hspace{2cm} +
\frac 1{r_nv_nt} P\big\{ F^\leftarrow(1-v_n(s\vee t))< \max_{1\le i\le r_n} X_i\le F^\leftarrow(1-v_n(s\wedge
t))\big\}\\
& \le & \frac{|t-s|}{r_nv_n st}\cdot r_nv_n s + \frac
1{r_nv_nt}\cdot r_nv_n|t-s|\\
& \le & 2\frac{|t-s|}t.
\end{aligned}$$ Hence, again by and the continuity of $Z_g$ for each $\delta>0$ there exists $\eta>0$ such that $$P\big\{ \sqrt{nv_n}\sup_{0\le t\le
\eta}t|\theta_{n,s_n(t)}-\theta_{n,t}|>\delta\big\}<\delta.$$ On the other hand, by , assumption and Hsing’s result $$\sqrt{nv_n}t|\theta_{n,s_n(t)}-\theta_{n,t}| =
\sqrt{nv_n} t \Big|\frac{\theta_{n,s_n(t)}-\theta}{\theta_{n,t}-\theta}-1\Big|\cdot
|\theta_{n,t}-\theta| =
O_P\big(|\theta_{n,t}-\theta|\big)\to 0$$ uniformly for $t\in[\eta,1]$, which completes the proof.
Theorem \[theo:main\]. By condition (M1) $$\label{eq:intvanish}
\int_{(0,1]^2} (st)^\delta\, \mu(ds,dt)=0, \qquad
\mu((0,1]^2)=0.$$ Thus $$\begin{aligned}
\hat\theta_{n,\mu}-\theta_n & = & \frac{\displaystyle \int
(\hat\theta_{n,s}-\theta_n)(\hat\theta_{n,t}-\theta_n)\,
\mu(ds,dt)}{\displaystyle \int
(\hat\theta_{n,s}-\theta_n)+(\hat\theta_{n,t}-\theta_n)\,
\mu(ds,dt)}\\
& = & \frac{\displaystyle \int
\big(\hat\theta_{n,s}-\theta_{n,s}+c_n s^\delta+R_n(s)\big)\big(\hat\theta_{n,t}-\theta_{n,t}
+c_nt^\delta+R_n(t)\big)\,
\mu(ds,dt)}{\displaystyle \int
\hat\theta_{n,s}-\theta_{n,s}+\hat\theta_{n,t}-\theta_{n,t}+c_n(s^\delta+t^\delta)+R_n(s)+R_n(t)\,
\mu(ds,dt)}.
\end{aligned}$$ In view of and the integrability condition (M3), the right-hand side has the same distribution as $$\frac{\displaystyle \int \big((nv_n)^{-1/2} s^{-1}(Z(s)+o_P(1))+c_ns^\delta+R_n(s)\big)
\big((nv_n)^{-1/2} t^{-1}(Z(t)+o_P(1))+c_nt^\delta+R_n(t)\big)\,
\mu(ds,dt)}{\displaystyle c_n\Big(\int
s^\delta+t^\delta\,\mu(ds,dt)+o_P(1)\Big)+(nv_n)^{-1/2}\Big(\int
s^{-1}Z(s)+t^{-1} Z(t)\, \mu(ds,dt)+o_P(1)\Big)}.$$ Because of , the conditions (M2) (M3), $(nv_n)^{-1/2}=o(c_n)$ and $d_n=o(c_n)$ this fraction equals $$\frac{\displaystyle \int (nv_n)^{-1/2} \big(t^\delta s^{-1}
Z(s)+s^\delta t^{-1}Z(t)\big)+s^\delta R_n(t)+t^\delta R_n(s)
\,
\mu(ds,dt)+o_P\big((nv_n)^{-1/2}\big)}{\displaystyle \int
s^\delta+t^\delta\,\mu(ds,dt)+o_P(1)}.$$ Now the first assertion is obvious and convergence is an immediate consequence of the additional assumption $d_n=o((nv_n)^{-1/2})$ and the integrability condition (M3).
Remark \[rem:mainrem\]. Recall the definition of $s_n(t)$ from the proof of Corollary \[cor:blockasympest\]. For $0<\delta\le 1$ and $0<u\le v$ one has $v^\delta-u^\delta\le v^{\delta-1}(v-u)\le u^{\delta-1}(v-u)$ and hence $|(s_n(t))^\delta-t^\delta)|\le t^{\delta-1}|s_n(t)-t|$. For $\delta>1$, the mean value theorem implies $
|(s_n(t))^\delta-t^\delta)|\le\delta |s_n(t)-t|$. Combining both inequalities with convergence , we conclude $|(s_n(t))^\delta-t^\delta)|=O_P\big( (nv_n)^{-1/2} t^{-1}\big)$. Moreover, under the given conditions, $R_n(s_n(t))=o_P\big((nv_n)^{-1/2} \big)$. Hence, $\hat\theta_{n,t}-\theta_n = (nv_n)^{-1/2}
t^{-1}(Z_n(t)+o_P(1))+c_n t^\delta$ and we proceed as in the proof of Theorem \[theo:main\] to establish .
[**References**]{}
Drees, H. (2000). Weighted approximations of tail processes for $\beta$–mixing random variables. [*Ann. Appl. Probab.*]{} [**10**]{}, 1274–1301.
Drees, H., and Rootzén, H. (2010). Limit Theorems for Empirical Processes of Cluster Functionals. [*Ann. Statist.*]{} [**38**]{}, 2145–2186.
Ferro, C.A.T., and Segers, J. (2003). Inference for clusters of extreme values. [*J. Roy. Statist.Soc. B*]{}, [**65**]{}, 545–556.
Hsing, T. (1993). Extremal index estimation for a weakly dependent stationary sequence. [ *Ann. Statist.*]{} [**21**]{}, 2043–2071.
Robert, C.Y., Segers, J, and Ferro, C.A.T. (2009). A sliding blocks estimator for the extremal index. [*Electron. J. Stat.*]{} [**3**]{}, 993-–1020.
Rootzén, H. (1995). The tail empirical process for stationary sequences. Preprint, Chalmers University Gothenburg.
Rootzén, H. (2009): Weak convergence of the tail empirical function for dependent sequences. [*Stoch. Proc. Appl.*]{} [**119**]{}, 468-490.
Segers, J. (2003). Functionals of clusters of extremes. [*Adv. Appl. Probab.*]{} [ **35**]{}, 1028–1045.
Smith, R.L., and Weissman, I. (1994). Estimating the extremal index. [*J. Roy. Statist. Soc. B*]{} [**56**]{}, 515–528.
Weissman, I., and Novak, S.Yu. (1998). On blocks and runs estimators of the extremal index. [*J. Statist. Plann. Inference*]{} [**66**]{}, 281–288.
[^1]: University of Hamburg, Department of Mathematics, SPST, Bundesstr. 55, 20146 Hamburg, Germany; email: [email protected]
[^2]: [*Keywords and phrases:*]{} absolute regularity, clustering of extremes, extremal index, empirical cluster process, bias reduction.\
[*AMS 2000 Classification:*]{} Primary 60G70; Secondary 60F17, 62G32.\
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We study the homology of Riemannian manifolds of finite volume that are covered by a product $({\mathbb{H}}^2)^r = {\mathbb{H}}^2 \times \ldots \times {\mathbb{H}}^2$ of the real hyperbolic plane. Using a variation of a method developed by Avramidi and Nyguen-Phan in [@FlatCycles], we show that any such manifold $M$ possesses, up to finite coverings, an arbitrarily large number of compact oriented flat totally geodesic $r$-dimensional submanifolds whose fundamental classes are linearly independent in the real homology group $H_r(M;{\mathbb{R}})$.'
address: 'Karlsruhe Institute of Technology, 76131 Karlsruhe, Germany'
author:
- Pascal Zschumme
bibliography:
- 'literature.bib'
title: Geometric construction of homology classes in Riemannian manifolds covered by products of the hyperbolic plane
---
=1
Introduction {#sec:introduction}
============
Let $M$ be a Riemannian manifold of finite volume that is covered by a product $({\mathbb{H}}^2)^r = {\mathbb{H}}^2 \times \ldots \times {\mathbb{H}}^2$. If $r=1$, then $M$ is a hyperbolic surface and its homology is well understood. Otherwise, $M$ can be a complicated object. For example, let $d > 0$ be a square-free integer and consider the real quadratic number field $F = {\mathbb{Q}}(\sqrt{d})$ with its two distinct real embeddings $\sigma_1,\sigma_2 \colon F \hookrightarrow {\mathbb{R}}$. Let ${\mathcal{O}}_F$ be the ring of integers of $F$. Then the group $\operatorname{SL}_2({\mathcal{O}}_F)$ acts properly discontinuously on the product ${\mathbb{H}}^2 \times {\mathbb{H}}^2$ by $$\gamma \cdot (z_1,z_2) := (\sigma_1(\gamma) \cdot z_1, \sigma_2(\gamma) \cdot z_2),$$ where $\sigma_i(\gamma) \cdot z_i$ is the action of $\operatorname{SL}_2({\mathbb{R}})$ on ${\mathbb{H}}^2$ by fractional linear transformations. For any torsion-free subgroup of finite index $\Gamma \subset \operatorname{SL}_2({\mathcal{O}}_F)$, the quotient $\Gamma \backslash ({\mathbb{H}}^2 \times {\mathbb{H}}^2)$ is a Riemannian manifold of finite volume that is covered by ${\mathbb{H}}^2 \times {\mathbb{H}}^2$. It is called a [*Hilbert modular surface*]{} and is an irreducible locally symmetric space of higher rank.
The homology of such a locally symmetric space is in general hard to compute, and even if one can do so, the geometric meaning of the homology classes is often lost during the computation. We choose a more geometric approach going back to Millson [@Millson], in which one studies homology classes that are the fundamental classes of totally geodesic submanifolds.
Promising candidates for such submanifolds are the compact flat totally geodesic submanifolds of dimension equal to the rank of the locally symmetric space. It is known that these submanifolds exist in any nonpositively curved locally symmetric space of finite volume, and Pettet and Souto proved in [@Pettet Theorem 1.2] that they are [*non-peripheral*]{}, which means they cannot be homotoped outside of every compact subset of the locally symmetric space.
This suggests that these submanifolds might contribute to the homology of the locally symmetric space. Avramidi and Nyguen-Phan [@FlatCycles] have investigated this question for the locally symmetric space $M = \operatorname{SL}_n({\mathbb{Z}}) \backslash \operatorname{SL}_n({\mathbb{R}}) / \operatorname{SO}(n)$ and proved that, in fact, up to finite coverings, the compact flat totally geodesic submanifolds of dimension equal to the rank of $M$ contribute to the homology group of $M$ with coefficients in a field.
We prove the following theorem, which shows that this is also true for all Riemannian manifolds of finite volume that are covered by products of ${\mathbb{H}}^2$:
\[thm:main\] Let $M$ be a Riemannian manifold of finite volume that is covered by $({\mathbb{H}}^2)^r$. Then for any $n \in {\mathbb{N}}$, there exists a connected finite covering $M' \to M$ and compact oriented flat totally geodesic $r$-dimensional submanifolds $\mathcal{F}_1,\ldots,\mathcal{F}_n \subset M'$ such that the images of the fundamental classes $[\mathcal{F}_1],\ldots,[\mathcal{F}_n]$ in $H_r(M';{\mathbb{R}})$ are linearly independent.
In particular, it follows that the $r$-th Betti number of a Riemannian manifold of finite volume that is covered by $({\mathbb{H}}^2)^r$ can be made arbitrarily large by going to a finite covering space of the manifold.
Our proof proceeds as follows: Using induction on $\dim(M)$, one finds irreducible Riemannian manifolds $M_1,\ldots,M_k$ for some $k \in {\mathbb{N}}$ and a finite covering $M_1 \times \ldots \times M_k \to M$. An application of the Künneth theorem for homology now shows that it suffices to prove \[thm:main\] for irreducible manifolds. So assume that $M$ is irreducible. If $r = 1$, then $M$ is a hyperbolic surface and there exists a finite covering surface $M'$ of $M$ whose genus is at least $n$. The surface $M'$ has $n$ distinct simple closed geodesics whose real homology classes are linearly independent in $H_1(M';{\mathbb{R}})$. This proves the claim for $r=1$. If $r > 1$, then Margulis’ arithmeticity theorem implies that $M$ is [*arithmetic*]{}, by which we mean that it is finitely covered by a quotient of $({\mathbb{H}}^2)^r$ by an arithmetically defined lattice in $\operatorname{SL}_2({\mathbb{R}})^r$.
The goal of this article is to explain our proof of \[thm:main\] for arithmetic manifolds. It is structured as follows: In \[sec:algebraic-groups\], we fix our notation for algebraic groups, discuss arithmetically defined lattices, and state Margulis’ arithmeticity theorem. In \[sec:quaternion-algebras\], we describe the arithmetically defined lattices in $\operatorname{SL}_2({\mathbb{R}})^r$ using quaternion algebras. In \[sec:flats\], we study flats in symmetric spaces, and in \[sec:geometric-cycles\] we discuss geometric cycles. Finally, in \[sec:geometric-construction\], we describe our construction of the covering $M' \to M$ and the submanifolds $\mathcal{F}_1, \ldots, \mathcal{F}_n \subset M'$ for an arithmetic manifold $M$, which is based on the techniques developed by Avramidi and Nyguen-Phan in [@FlatCycles].
The material covered in this article evolved from the author’s doctoral thesis [@ZschummePhD], in which the reader can find more details on our construction and the proof of \[thm:main\].
Algebraic groups and Margulis’ arithmeticity theorem {#sec:algebraic-groups}
====================================================
We consider algebraic groups as special cases of group schemes and identify them with their functors of points (see [@Waterhouse; @Milne-Book]). By this, we mean the following: Let $R$ be a commutative ring. A [*group scheme over $R$*]{} is a functor ${\mathbf{G}} \colon {\textnormal{Alg}}_R \to {\textnormal{Grp}}$ from the category of commutative $R$-algebras to the category of groups that is representable by a finitely generated $R$-algebra. We denote this algebra by ${\mathcal{O}}({\mathbf{G}})$. One can think of ${\mathbf{G}}$ as a group functor defined by polynomial equations with coefficients in $R$. In fact, by choosing a surjection $\pi \colon R[X_1,\ldots,X_n] \to {\mathcal{O}}({\mathbf{G}})$, we obtain for each commutative $R$-algebra $A$ a natural inclusion $${\mathbf{G}}(A) {\xrightarrow{\raisebox{-1pt}[0ex][0ex]{$\scriptscriptstyle\cong$}}}\hom({\mathcal{O}}({\mathbf{G}}), A) \xhookrightarrow{\pi^*} \hom(R[X_1,\ldots,X_n], A) {\xrightarrow{\raisebox{-1pt}[0ex][0ex]{$\scriptscriptstyle\cong$}}}A^n,$$ which identifies the group ${\mathbf{G}}(A)$ with the vanishing set of the ideal $\ker(\pi) \subset R[X_1,\ldots,X_n]$ in $A^n$. For a topological $R$-algebra $A$, we put on ${\mathbf{G}}(A)$ the unique weakest topology for which the above inclusion (and thus any such inclusion) is continuous with respect to the product topology on $A^n$.
The [*extension of scalars*]{} of a group scheme ${\mathbf{G}}$ over $R$ to some ring extension $S/R$ is the group scheme ${\mathbf{G}}_S \colon {\textnormal{Alg}}_S \to {\textnormal{Grp}}$, $A \mapsto {\mathbf{G}}(\operatorname{res}_S(A))$. Here, $\operatorname{res}_S(A)$ denotes $A$ as an $S$-algebra. If the extension map $\sigma \colon R \hookrightarrow S$ is not clear from the context, then we write $\operatorname{res}_\sigma(A)$ instead of $\operatorname{res}_S(A)$ and ${\mathbf{G}}_\sigma$ instead of ${\mathbf{G}}_S$.
An [*algebraic group*]{} over a field $K$, or in short a [*$K$-group*]{}, is simply a group scheme ${\mathbf{G}}$ over $K$. Its group ${\mathbf{G}}(\overline{K})$ of points with values in an algebraic closure $\overline{K}$ of $K$ is then an affine variety in the space $\overline{K}^n$ equipped with the Zariski topology and the group operations are polynomial maps. The algebraic group ${\mathbf{G}}$ is said to be [*connected*]{} or [*finite*]{} if this affine variety is connected or finite, respectively.
Let $L/K$ be a finite separable field extension. The [*restriction of scalars*]{} of an algebraic group ${\mathbf{G}}$ over $L$ to $K$ is the functor $\operatorname{Res}_{L/K}{\mathbf{H}} \colon {\textnormal{Alg}}_K \to {\textnormal{Grp}}$, $A \mapsto {\mathbf{H}}(A \otimes_K L)$. This is an algebraic group over $K$ (see [@Milne-Book p. 57]) and the natural isomorphism $K \otimes_K L {\xrightarrow{\raisebox{-1pt}[0ex][0ex]{$\scriptscriptstyle\cong$}}}L$ induces an isomorphism of topological groups $(\operatorname{Res}_{L/K}{\mathbf{H}})(K) {\xrightarrow{\raisebox{-1pt}[0ex][0ex]{$\scriptscriptstyle\cong$}}}{\mathbf{H}}(L)$.
Every algebraic group ${\mathbf{G}}$ over a number field $F$ has an [*integral form*]{}. This is a group scheme ${\mathbf{G}}_0$ over the ring of integers ${\mathcal{O}}_F$ of $F$ together with an $F$-isomorphism $({\mathbf{G}}_0)_F {\xrightarrow{\raisebox{-1pt}[0ex][0ex]{$\scriptscriptstyle\cong$}}}{\mathbf{G}}$. A subgroup $\Gamma \subset {\mathbf{G}}(F)$ is called an [*arithmetic subgroup*]{} if it is commensurable with the image of ${\mathbf{G}}_0({\mathcal{O}}_F)$ in ${\mathbf{G}}(F)$ for some integral form ${\mathbf{G}}_0$ of ${\mathbf{G}}$. This notion is independent of the choice of the integral form, because the groups of ${\mathcal{O}}_F$-points of any two integral forms of ${\mathbf{G}}$ are commensurable with each other (see [@Platonov Proposition 4.1]).
By an important theorem of Borel and Harish-Chandra [@Borel-Harish-Chandra], any arithmetic subgroup $\Gamma \subset {\mathbf{H}}({\mathbb{Q}})$ of a semisimple algebraic group ${\mathbf{H}}$ over ${\mathbb{Q}}$ is a lattice in ${\mathbf{H}}({\mathbb{R}})$. The following definition describes all lattices in the group of real points of an algebraic group over ${\mathbb{R}}$ that are constructed in this way:
\[def:arithmetically-defined-subgroup\] Let ${\mathbf{G}}$ be a connected semisimple ${\mathbb{R}}$-group without ${\mathbb{R}}$-anisotropic almost ${\mathbb{R}}$-simple factors. An irreducible lattice $\Delta \subset {\mathbf{G}}({\mathbb{R}})^0$ is [*arithmetically defined*]{} if there exists a connected almost ${\mathbb{Q}}$-simple ${\mathbb{Q}}$-group ${\mathbf{H}}$, an ${\mathbb{R}}$-epimorphism $$\Phi \colon {\mathbf{H}}_{\mathbb{R}}\twoheadrightarrow {\mathbf{G}}$$ so that $(\ker{\Phi})({\mathbb{R}})$ is compact, and an arithmetic subgroup $\Gamma \subset {\mathbf{H}}({\mathbb{Q}})$ such that $\Delta$ is commensurable with $\Phi({\mathbf{H}}(\Gamma))$.
Non-arithmetically defined lattices are known to exist in some real algebraic groups of rank one. For example, in ${\mathbf{SL}}_2({\mathbb{R}})$ this follows from the existence of uncountably many non-commensurable hyperbolic surfaces (see [@ArithmeticGroupsWhatWhyHow p. 63]).
Margulis’ celebrated arithmeticity theorem states that in a real algebraic group of higher rank, or similarly in a real Lie group of higher rank, all irreducible lattices are arithmetically defined (see [@Margulis-DiscreteSubgroups Theorem IX.1.11]):
\[thm:margulis-arithmeticity\] Let ${\mathbf{G}}$ be a connected semisimple ${\mathbb{R}}$-group without ${\mathbb{R}}$-anisotropic almost ${\mathbb{R}}$-simple factors and with $\operatorname{rank}_{\mathbb{R}}({\mathbf{G}}) > 1$. Then every irreducible lattice in ${\mathbf{G}}({\mathbb{R}})^0$ is arithmetically defined.
Unit groups in quaternion algebras {#sec:quaternion-algebras}
==================================
The arithmetically defined lattices in $\operatorname{SL}_2({\mathbb{R}})^r$ can be described using quaternion algebras. Let $K$ be a field of characteristic zero. A [*quaternion algebra over $K$*]{} is an algebra $D$ over $K$ for which there exists a vector space basis $\{1,i,j,k\}$ and $a,b \in K^\times$ such that $$\label{eq:quaternion-algebra-relations}
i^2 = a,
\quad
j^2 = b,
\quad
k = ij = -ji.$$ We then call $\{1,i,j,k\}$ a [*quaternionic basis*]{} for $D$. The [*reduced norm*]{} of an element $x = x_0 + x_1 i + x_2 j + x_3 k \in D$ is $$\label{eq:reduced-norm}
N(x) = x_0^2 - ax_1^2 - bx_2^2 + abx_3^2 \in K.$$ The reduced norm of $x \in D$ is preserved by every automorphism of $D$ and is therefore independent of the choice of the quaternionic basis. For example, in the case of the matrix algebra $D = M_2(K)$, we have $N(x) = \det(x)$. An element $x \in D$ is invertible if and only if $N(x) \neq 0$. We write $D^1$ for the group of units of reduced norm one in $D$.
The [*general linear group*]{} associated to a quaternion algebra $D$ over $K$ is the algebraic group ${\mathbf{GL}}_D \colon {\textnormal{Alg}}_K \to {\textnormal{Grp}}$, $A \mapsto (A \otimes_K D)^\times$. We extend the reduced norm to tensor products $A \otimes_K D$ for $K$-algebras $A$ using and define the [*special linear group*]{} ${\mathbf{SL}}_D \colon {\textnormal{Alg}}_K \to {\textnormal{Grp}}$, $A \mapsto (A \otimes_K D)^1$.
We now study quaternion algebras over number fields. Let $F$ be a number field. The analog for a quaternion algebra $D$ over $F$ of the ring of integers of $F$ is an [*order*]{}. This is a subring $\Lambda \subset D$ that is a finitely generated ${\mathcal{O}}_F$-submodule of $D$ and spans $D$ over $F$. For example, the subring $M_2({\mathcal{O}}_F) \subset M_2(F)$ is an order. A quaternion algebra has many orders, but the groups of units of any two of its orders are commensurable with each other (see [@GroupRingGroupsI Lemma 4.6.9]).
For an order $\Lambda \subset D$, we define its [*general linear group*]{} ${\mathbf{GL}}_\Lambda \colon {\textnormal{Alg}}_{{\mathcal{O}}_F} \to {\textnormal{Grp}}$, $A \mapsto (A \otimes_{{\mathcal{O}}_F} \Lambda)^\times$. This is a group scheme over ${\mathcal{O}}_F$ and an integral form of the algebraic group ${\mathbf{GL}}_D$. If $\Lambda$ is the ${\mathcal{O}}_F$-span of a quaternionic basis for $D$, then we again extend the reduced norm using and define the [*special linear group*]{} ${\mathbf{SL}}_\Lambda \colon {\textnormal{Alg}}_{{\mathcal{O}}_F} \to {\textnormal{Grp}}$, $A \mapsto (A \otimes_{{\mathcal{O}}_F} \Lambda)^1$, which is an integral form of ${\mathbf{SL}}_D$. It follows that for any order $\Lambda \subset D$, the group of units of reduced norm one $\Lambda^1 := \Lambda \cap D$ is an arithmetic subgroup of ${\mathbf{SL}}_D(F)$.
For any real embedding $\sigma \colon F \hookrightarrow {\mathbb{R}}$, the algebra $D \otimes_F \operatorname{res}_\sigma({\mathbb{R}})$ is either isomorphic to $M_2({\mathbb{R}})$ or is a division algebra. In the first case, we say $D$ is [*split at $\sigma$*]{} and otherwise [*ramified at $\sigma$*]{}. This now leads us to the following:
\[def:subgroup-derived-from-quaternion-algebra\] A subgroup $\Delta \subset \operatorname{SL}_2({\mathbb{R}})^r$ is said to be [*derived from a quaternion algebra*]{} if there exists a totally real number field $F$, a quaternion algebra $D$ over $F$ that is split at exactly $r$ distinct real embeddings $\sigma_1,\ldots,\sigma_r \colon F \hookrightarrow {\mathbb{R}}$, an isomorphism $\tau_i \colon D \otimes_F \operatorname{res}_{\sigma_i}({\mathbb{R}}) {\xrightarrow{\raisebox{-1pt}[0ex][0ex]{$\scriptscriptstyle\cong$}}}M_2({\mathbb{R}})$ for each $i \in \{1,\ldots,r\}$, and an order $\Lambda \subset D$ such that $$\Delta = \bigl\{ \bigl(\tau_1(x), \ldots,\tau_r(x)\bigr) : x \in \Lambda \bigr\}.$$
\[rem:derived-commensurable\] The maps $\tau_i \colon D \otimes_F \operatorname{res}_\sigma({\mathbb{R}}) {\xrightarrow{\raisebox{-1pt}[0ex][0ex]{$\scriptscriptstyle\cong$}}}M_2({\mathbb{R}})$ in \[def:subgroup-derived-from-quaternion-algebra\] are not unique, but any two choices for $\tau_i$ differ only by conjugation with a matrix in $\operatorname{GL}_2({\mathbb{R}})$ by the Skolem–Noether theorem. So up to commensurability and conjugation in $\operatorname{GL}_2({\mathbb{R}})^r$, the subgroup of $\operatorname{SL}_2({\mathbb{R}})^r$ derived from a quaternion algebra depends only on the isomorphism class of the quaternion algebra.
A subgroup derived from a quaternion algebra is readily seen to be an arithmetically defined lattice in $\operatorname{SL}_2({\mathbb{R}})^r$. In fact, the converse direction also holds (see [@ZschummePhD Theorem 5.44]):
\[thm:classification-arithmetically-defined-lattices\] A subgroup of $\operatorname{SL}_2({\mathbb{R}})^r$ is an arithmetically defined lattice if and only if it is commensurable with a subgroup derived from a quaternion algebra.
We write $D = (a,b)_F$ for the quaternion algebra determined by . The field $E := F(\sqrt{a})$ is a [*splitting field for $D$*]{}. By this, we mean that it is a field extension $E/F$ so that $D \otimes_F E \cong M_2(E)$. A quadratic extension $E/F$ is a splitting field for $D$ if and only if for all of the places $v$ of $F$ where $D$ is ramified, the local completion $E_v/F_v$ is a quadratic extension (see [@Arithmetics-Reid Theorem 7.3.3 and its proof]). Here, $E_v$ is the completion of $E$ at some place $w$ of $E$ lying above $v$. Note that for any two places $w$ and $w'$ of $E$ lying above $v$, the corresponding completions of $E$ are $F_v$-isomorphic to each other by [@Neukirch Proposition II.9.1], which justifies the notation $E_v$. Moreover, for any $a \in F^\times$ for which $F(\sqrt{a})$ is a splitting field for $D$, there exist some $b \in F^\times$ with $D \cong (a,b)_F$ (see [@CSAGaloisCohomology Proposition 1.2.3]).
Next, we show that the constants $a,b \in F^\times$ defining the isomorphism class of the quaternion algebra $(a,b)_F$ can always be chosen in a certain way. We will need this in our computations later in \[sec:geometric-construction\].
\[thm:quaternion-algebra-trick\] Let $F$ be a number field. Then for every quaternion algebra $D$ over $F$, there exist $a,b \in {\mathcal{O}}_F$ in the ring of integers of $F$ such that $D$ is isomorphic to $(a,b)_F$ and such that for any real embedding $\sigma \colon F \hookrightarrow {\mathbb{R}}$ at which $D$ is split, we have $\sigma(a) > 0$.
We use the Grunwald–Wang theorem [@Roquette p. 29] to construct a splitting field $E/F$ for $D$ as follows: By this theorem, there exists a quadratic extension $E/F$ whose local completions $E_v/F_v$ at the places $v$ of $F$ where $D$ is ramified are quadratic extensions, and such that at all real places $v$ of $F$ at which $D$ is split, they are trivial extensions. Then $E/F$ is a splitting field for $D$ by the criterion stated above.
Since $E/F$ is a quadratic extension, we have $E = F(\sqrt{a})$ for some $a \in F^\times$. Let $\sigma \colon F \hookrightarrow {\mathbb{R}}$ be a real embedding at which $D$ is split. By the construction of $E$, we have $E_v \cong {\mathbb{R}}$ for the place $v$ of $F$ corresponding to $\sigma$. Hence the image of the extension $\widetilde{\sigma} \colon E \hookrightarrow {\mathbb{C}}$ of $\sigma$ given by $\widetilde{\sigma}(x + y\sqrt{a}) = \sigma(x) + \sigma(y)\sqrt{\sigma(a)}$ for $x,y \in F$ stays in ${\mathbb{R}}$, and so we must have $\sigma(a) > 0$.
Since $E/F$ is a splitting field for $D$, we find $b \in F^\times$ with $D \cong (a,b)_F$. Finally, we can also achieve that $a,b \in {\mathcal{O}}_F$, because of $(a,b)_F \cong (c^2a,c^2b)_F$ for all $c \in F^\times$ by [@Arithmetics-Reid p. 78].
Adeles and congruence subgroups {#sec:adeles-and-congruence-subgroups}
===============================
We will use adeles and congruence subgroups to construct subgroups of finite index in arithmetic groups. Let $F$ be a number field with ring of integers ${\mathcal{O}}_F$. We denote by ${\mathbb{A}_{f,F}}$ the ring of finite adeles of $F$ and by ${\mathbb{O}_{f,F}}$ the ring of integral finite adeles of $F$ (see [@Platonov pp. 10–13]). We consider $F$ as a subring of ${\mathbb{A}_{f,F}}$ by the diagonal embedding $F \hookrightarrow {\mathbb{A}_{f,F}}$, and similarly ${\mathcal{O}}_F$ as a subring of ${\mathbb{O}_{f,F}}$ by the diagonal embedding ${\mathcal{O}}_F \hookrightarrow {\mathbb{O}_{f,F}}$. This turns ${\mathbb{A}_{f,F}}$ and ${\mathbb{O}_{f,F}}$ into topological algebras over $F$ and ${\mathcal{O}}_F$, respectively.
Let now ${\mathbf{G}}$ be a group scheme over ${\mathcal{O}}_F$. For a subgroup $U \subset {\mathbf{G}}({\mathbb{O}_{f,F}})$ and a nonzero ideal ${\mathfrak{a}}\subset {\mathcal{O}}_F$, we write $$U({\mathfrak{a}}) := \ker\bigl(U \to {\mathbf{G}}({\mathbb{O}_{f,F}}/{\mathfrak{a}}{\mathbb{O}_{f,F}})\bigr),$$ where ${\mathfrak{a}}{\mathbb{O}_{f,F}}$ is the ideal in ${\mathbb{O}_{f,F}}$ generated by ${\mathfrak{a}}$. We have (see [@ZschummePhD Proposition 4.39]):
\[thm:adelic-points-basis\] Let ${\mathbf{G}}$ be a group scheme over the ring of integers ${\mathcal{O}}_F$ of a number field $F$. Then the family of groups ${\mathbf{G}}({\mathbb{O}_{f,F}})({\mathfrak{a}})$ for all nonzero ideals ${\mathfrak{a}}\subset {\mathcal{O}}_F$ is a basis of open neighborhoods of the identity in both of the group ${\mathbf{G}}({\mathbb{O}_{f,F}})$ and ${\mathbf{G}}({\mathbb{A}_{f,F}})$.
For a subgroup of the integral points $\Gamma \subset {\mathbf{G}}({\mathcal{O}}_F)$, the group $\Gamma({\mathfrak{a}})$ coincides with $\ker(\Gamma \to {\mathbf{G}}({\mathcal{O}}_F/{\mathfrak{a}}{\mathcal{O}}_F))$ and has finite index in $\Gamma$. We call $\Gamma({\mathfrak{a}})$ the [*principal congruence subgroup of $\Gamma$ of level ${\mathfrak{a}}$*]{}. More generally, any subgroup of $\Gamma$ that contains a principal congruence subgroup has finite index in $\Gamma$ and is called a [*congruence subgroup of $\Gamma$*]{}.
It is possible that there exist subgroups of finite index in ${\mathbf{G}}({\mathcal{O}}_F)$ that are not congruence subgroups. Examples of such subgroups in ${\mathbf{SL}}_2({\mathbb{Z}})$ were already known to Fricke and Klein in the $19$th century (see [@CongruenceSubgroupProblem p. 299]). The group scheme ${\mathbf{G}}$ is said to have the [*congruence subgroup property*]{} if every subgroup of finite index in ${\mathbf{G}}({\mathcal{O}}_F)$ is a congruence subgroup. Chevalley [@Chevalley] proved in 1951 that ${\mathbf{GL}}_1$ has this property:
\[thm:chevalley\] Let $F$ be a number field. Then for every subgroup of finite index $\Gamma \subset {\mathcal{O}}_F^\times$, there exists a nonzero ideal ${\mathfrak{a}}\subset {\mathcal{O}}_F$ such that ${\mathcal{O}}_F^\times({\mathfrak{a}}) \subset \Gamma$.
Polar regular elements and flats {#sec:flats}
================================
We will use polar regular elements, as introduced by Mostow in [@StrongRigidity p. 12], to algebraically describe the maximal flat subspaces of a symmetric space.
Let $G$ be a connected linear semisimple Lie group. By [@Helgason p. 431], each $g \in G$ has a unique decomposition $$\label{eq:real-jordan-decomposition}
g = g_u g_h g_e$$ with $g_u,g_h,g_e \in G$ such that $g_u$,$g_h$, and $g_e$ correspond in one (and thus any) embedding $G \hookrightarrow \operatorname{GL}_n({\mathbb{R}})$ to a unipotent, a hyperbolic, and an elliptic matrix, respectively, and such that they all commute with each other. We call a matrix in $\operatorname{GL}_n({\mathbb{R}})$ [*semisimple*]{} if it is diagonalizable over ${\mathbb{C}}$. A semisimple matrix is called [*hyperbolic*]{} if all its eigenvalues are real and positive, and it is called [*elliptic*]{} if all its eigenvalues have absolute norm one. We call the [*real Jordan decomposition of $g$*]{}.
\[def:polar-regular\] An element $g \in G$ is [*polar regular*]{} if for all $g' \in G$, we have $$\dim(C_G(g_h)) \leq \dim(C_G(g'_h)),$$ where $C_G(g_h)$ denotes the centralizer of $g_h$ in $G$.
Let $X_G$ be the symmetric space associated to $G$. A [*flat*]{} in $X_G$ is a connected totally geodesic submanifold of $X_G$ whose curvature tensor vanishes. A flat is called [*maximal*]{} if it is of maximal dimension among all flats in $X_G$. We write $G_A := \{ g \in G : g \cdot A = A\}$ for the stabilizer group of flat $A \subset X_G$.
By [@StrongRigidity Lemma 5.2], we have the following relationship between polar regular elements in $G$ and maximal flats in $X_G$:
\[thm:polar-regular-mostow\] Let $G$ be a connected linear semisimple Lie group and let $g \in G$ be polar regular. Then there exists a unique maximal flat $A \subset X_G$ in the symmetric space associated to $G$ such that $g \cdot A = A$. Moreover, the centralizer $C_G(g)$ is a subgroup of $G_A$ and acts transitively on $A$.
\[ex:flats-product-hyperbolic-space\] Let $G = \operatorname{SL}_2({\mathbb{R}})^r$. Then $X_G = ({\mathbb{H}}^2)^r$. The maximal flats in $({\mathbb{H}}^2)^r$ are the products of geodesic lines in ${\mathbb{H}}^2$. An element $g = (g_1,\ldots,g_r) \in \operatorname{SL}_2({\mathbb{R}})^r$ is polar regular if and only if each $g_i$ has two distinct real eigenvalues.
Let $\Gamma \subset G$ be a lattice. We say that a flat $A \subset X_G$ is [*$\Gamma$-compact*]{} if the quotient $\Gamma_A \backslash A$ is compact. Then the image of $A$ in $\Gamma \backslash X_G$ is also compact. By [@StrongRigidity Lemma 8.3’], the set of $\Gamma$-compact maximal flats is dense in the space of all maximal flats:
\[thm:density\] Let $G$ be a connected linear semisimple Lie group. Let $\Gamma \subset G$ be a lattice and let $A \subset X_G$ be a maximal flat. Then for every open neighborhood of the identity $U \subset G$, there exists some $u \in U$ such that $u \cdot A$ is a $\Gamma$-compact maximal flat in $X_G$ that is stabilized by a polar regular element of $\Gamma$.
Geometric cycles {#sec:geometric-cycles}
================
Let $G$ be a linear semisimple Lie group and let $X_G$ be the symmetric space associated to $G$. For a closed subgroup $H \subset G$, we can always find a maximal compact subgroup $K \subset G$ such that $K_H := K \cap H$ is a maximal compact subgroup of $H$. We write $X_H := H/K_H$. Then $X_H$ is diffeomorphic to a Euclidean space and the inclusion $H \hookrightarrow G$ induces a closed embedding $$j_H \colon X_H \hookrightarrow X_G$$ whose image is a totally geodesic submanifold of $X_G$ (see [@Schwermer p. 213]).
Consider now a torsion-free lattice $\Gamma \subset G$ and the corresponding locally symmetric space $\Gamma \backslash X_G$. The map $j_H$ descends to an immersion into the locally symmetric space $\Gamma \backslash X_G$, but this map will in general not be an embedding. In the arithmetic setting, we can always obtain an embedding by passing to a subgroup of finite index (see [@Schwermer Theorem D]):
\[thm:geometric-cycles\] Let ${\mathbf{G}}$ be a connected semisimple ${\mathbb{Q}}$-group and let ${\mathbf{H}}$ be a connected reductive ${\mathbb{Q}}$-subgroup of ${\mathbf{G}}$. Then any arithmetic subgroup $\Gamma \subset {\mathbf{G}}({\mathbb{Q}})$ has a torsion-free subgroup of finite index $\Gamma_0 \subset \Gamma$ such that for any subgroup of finite index $\Gamma' \subset \Gamma_0$, the map $$j_{H \mid \Gamma'} \colon (\Gamma' \cap H) \backslash X_H \to \Gamma' \backslash X_G$$ induced by the above map $j_H \colon X_H \hookrightarrow X_G$ for $G = {\mathbf{G}}({\mathbb{R}})$ and $H = {\mathbf{H}}({\mathbb{R}})$ is a closed embedding and its image is an orientable totally geodesic submanifold.
In the situation of the above theorem, we call the image of $j_{H|\Gamma'}$ in $\Gamma' \backslash X_G$ a [*geometric cycle*]{}. An effective strategy to show that the fundamental class of a geometric cycle is nontrivial in the homology of $\Gamma' \backslash X_G$ is to find another geometric cycle in $\Gamma' \backslash X_G$ such that their intersection product is nontrivial.
Construction of flat submanifolds {#sec:geometric-construction}
=================================
We now explain our construction of compact flat totally geodesic submanifolds with linearly independent homology classes in an arithmetic Riemannian manifold covered by $({\mathbb{H}}^2)^r$. This will finish the proof of \[thm:main\]. As we have seen, it suffices to consider for this task a quotient $M = \Delta \backslash ({\mathbb{H}}^2)^r$, where $\Delta \subset \operatorname{SL}_2({\mathbb{R}})^r$ is a subgroup derived from a quaternion algebra.
Throughout this section, we fix the following assumptions: $F$ is a totally real number field and we denote by $\{\sigma_1,\ldots,\sigma_d\}$ the set of all distinct real embeddings $F \hookrightarrow {\mathbb{R}}$. Further, $D = (a,b)_F$ is a quaternion algebra over $F$ that is split at the first $r$ embeddings and ramified at the remaining ones. $\Lambda \subset D$ is an order and $\Gamma \subset \Lambda^1$ is a torsion-free subgroup of finite index. For each $i \in \{1,\ldots,r\}$, we fix an isomorphism $\tau_i \colon D \otimes_F \operatorname{res}_{\sigma_i}({\mathbb{R}}) {\xrightarrow{\raisebox{-1pt}[0ex][0ex]{$\scriptscriptstyle\cong$}}}M_2({\mathbb{R}})$ so that $$\Delta = \{ (\tau_1(x), \ldots, \tau_r(x)) : x \in \Gamma \} \subset \operatorname{SL}_2({\mathbb{R}})^r.$$ We assume that $F \subset {\mathbb{R}}$ and $\sigma_1$ is the identity embedding. Further, we assume that $a,b \in {\mathcal{O}}_F$ and $\sigma_1(a), \ldots, \sigma_r(a) > 0$. This is justified by \[thm:quaternion-algebra-trick\]. We also assume that $\Lambda$ is the ${\mathcal{O}}_F$-span of a quaternionic basis $\{1,i,j,k\}$ for $D$ and that with respect to this basis, $\tau_1$ is given by $$\tau_1(x+yi+zj+wk)
= \begin{pmatrix}x + y\sqrt{a} & z + w\sqrt{a} \\ b(z - w\sqrt{a}) & x-y\sqrt{a}\end{pmatrix}.$$ This is justified by \[rem:derived-commensurable\]. In particular, we have $\tau_1(D) \subset M_2(F(\sqrt{a}))$. In this setting, we have an action of $D^1$ on $({\mathbb{H}}^2)^r$ by $$g \cdot (z_1,\ldots,z_r) := (\tau_1(g)\cdot z_1, \ldots, \tau_r(g) \cdot z_r),$$ where $\tau_i(g) \cdot z_i$ is the action of $\operatorname{SL}_2({\mathbb{R}})$ on ${\mathbb{H}}^2$ by fractional linear transformations. We extend this action to an action of $D^\times$ on $({\mathbb{H}}^2)^r$ through the maps $\tau_i$ as above by defining the action of $\operatorname{GL}_2({\mathbb{R}})$ on ${\mathbb{H}}^2$ as follows: $$\begin{pmatrix}
a_{11} & a_{12} \\
a_{21} & a_{22}
\end{pmatrix}
\cdot
z :=
\begin{cases}
(a_{11}z+a_{12})(a_{21}z+a_{22})^{-1}, & \text{if } a_{11}a_{22}-a_{12}a_{21} > 0, \\[5pt]
\overline{(a_{11}z+a_{12})(a_{21}z+a_{22})^{-1}}, & \text{otherwise. }
\end{cases}$$ We write $(D^\times)_A := \{ x \in D^\times : x \cdot A = A \}$ for the stabilizer group of a flat $A \subset ({\mathbb{H}}^2)^r$. One can check that for our action of $D^\times$ on $({\mathbb{H}}^2)^r$, the following analog of \[thm:polar-regular-mostow\] holds (see [@ZschummePhD Proposition 3.31] for a proof):
\[thm:polar-regular-quaternion-algebra\] Let $x \in D^\times$ be such that $\tau_1(x),\ldots,\tau_r(x) \in \operatorname{GL}_2({\mathbb{R}})$ each have two distinct real eigenvalues. Then there exists a unique maximal flat $A \subset ({\mathbb{H}}^2)^r$ with $x \cdot A = A$. Moreover, the centralizer $C_{D^\times}(x)$ is a subgroup of finite index in $(D^\times)_A$ and acts by orientation-preserving isometries on $A$.
Building a configuration of flats {#subsec:configuration}
---------------------------------
The following lemma shows that two generic geodesic lines in ${\mathbb{H}}^2$ can be slightly perturbed without changing the way they intersect. We denote by $\partial_\infty {\mathbb{H}}^2$ the boundary at infinity of the hyperbolic plane (see [@Eberlein p. 27]).
\[thm:perturbation-lemma\] Let $L_1$ and $L_2$ be two geodesic lines in ${\mathbb{H}}^2$ whose endpoints in $\partial_\infty {\mathbb{H}}^2$ are pairwise distinct. Then $L_1$ and $L_2$ are either disjoint or intersect transversally in a single point. Moreover, there exists an open neighborhood of the identity $U \subset \operatorname{SL}_2({\mathbb{R}})$ such that for all $u,v \in U$, the geodesic lines $u \cdot L_1$ and $v \cdot L_2$ also have pairwise distinct endpoints in $\partial_\infty {\mathbb{H}}^2$ and intersect in the same way.
(0,0) circle (1); (0,0) circle (1cm); [ pt = 1pt (+180:1) – (:1) (:) circle (); (:) – (:--.08) ]{}; [ pt = 1pt (+180:1) – (:1) (:) circle (); (:) – (:--.08) ]{};
at (200:1) ; at (70:1) ; at (100:1) ; at (0:1cm) ; at (-0.4,-0.275) [$L_1$]{}; at (0.45,-0.025) [$L_2$]{};
(0,0) circle (1); (0,0) circle (1cm); [ pt = 1pt (+180:1) – (:1) (:) circle (); (:) – (:--.08) ]{}; [ pt = 1pt (+180:1) – (:1) (:) circle (); (:) – (:--.08) ]{};
at (230:1cm) ; at (100:1cm) ; at (80:1cm) ; at (-30:1cm) ;
at (0.28,-0.2) [$M_2$]{}; at (-0.23,-0.5) [$M_1$]{};
Note that $L_1$ and $L_2$ intersect if and only if their endpoints in $\partial_\infty {\mathbb{H}}^2$ are linked, in which case their intersection consists of a single point and is transverse (see \[fig:geodesics-endpoints\]). We denote the four endpoints of $L_1$ and $L_2$ by $v_1,v_2,v_3,v_4 \in \partial_\infty{\mathbb{H}}^2$. The geodesic compactification $\overline{{\mathbb{H}}^2} := {\mathbb{H}}^2 \sqcup \partial_\infty{\mathbb{H}}^2$ as defined in [@Eberlein pp. 28–30] is Hausdorff, and so we can find pairwise disjoint open neighborhoods $V_i \subset \overline{{\mathbb{H}}^2}$ of the points $v_i$. The group $\operatorname{SL}_2({\mathbb{R}})$ acts continuously on $\overline{{\mathbb{H}}^2}$, hence the maps $\phi_i \colon \operatorname{SL}_2({\mathbb{R}}) \to \overline{{\mathbb{H}}^2}$ given by $\phi_i(g) := g \cdot v_i$ are continuous. So $U = \bigcap_{i=1}^4 \phi_i^{-1}(V_i)$ is an open neighborhood of the identity in $\operatorname{SL}_2({\mathbb{R}})$. By the construction of $U$, the endpoints of two geodesic lines $u \cdot L_1$ and $v \cdot L_2$ with $u,v \in U$ in $\partial_\infty{\mathbb{H}}^2$ are linked if and only if whose of $L_1$ and $L_2$ are linked. So the statement of the lemma follows.
\[thm:configuration\] For any $n \in {\mathbb{N}}$, there exist maximal flats $A_1,\ldots,A_n$ and $B_1,\ldots,B_n$ in $({\mathbb{H}}^2)^r$ so that the following conditions are satisfied:
1. The flats $A_i$ and $B_j$ intersect transversally in a single point if $i \leq j$, and they are disjoint if $i > j$.
2. Each $A_i$ is $\Gamma$-compact.
3. Each $A_i$ is stabilized by an element $\alpha_i \in D^1$ such that $\tau_1(\alpha_i),\ldots,\tau_r(\alpha_i)$ each have two distinct real eigenvalues.
4. Each $B_j$ is stabilized by an element $\beta_j \in D^\times$ such that $\tau_1(\beta_j),\ldots,\tau_r(\beta_j)$ each have two distinct real eigenvalues and $\tau_1(\beta_j)$ is diagonalizable over $F(\sqrt{a})$.
We start with the first condition. For this, we choose geodesic lines $L_1,\ldots,L_n$ and $M_1,\ldots,M_n$ in ${\mathbb{H}}^2$ with pairwise distinct endpoints in $\partial_\infty{\mathbb{H}}^2$ such that for each $i,j \in \{1,\ldots,n\}$, we have that $L_i$ and $M_j$ intersect if and only if $i \leq j$ (See \[fig:geodesics-pattern\] for an example with $n=3$). Set $A_i := L_i \times \cdots \times L_i$ and $B_j := M_j \times \cdots \times M_j$. The first condition is now satisfied.
By \[thm:perturbation-lemma\], there exists an open neighborhood of the identity $U \subset \operatorname{SL}_2({\mathbb{R}})$ so that we may perturb each $L_i$ and $M_j$ by isometries in $U$ without invalidating the first condition. The product $U^r$ is an open neighborhood of the identity in $\operatorname{SL}_2({\mathbb{R}})^r$. So for each $i \in \{1,\ldots,n\}$, there exists by \[thm:density\] an element $u_i \in U^r$ such that $u_i \cdot A_i$ is a $\Gamma$-compact flat that is stabilized by an element $\alpha_i \in D^1$ for which $\tau_1(\alpha_i),\ldots,\tau_r(\alpha_i)$ each have two distinct real eigenvalues. We now replace each $A_i$ by $u_i \cdot A_i$ and the first three conditions are satisfied.
(0,0) circle (1); (0,0) circle (1); [ pt = 1pt (+180:1) – (:1) (:) circle (); (:) – (:--.08) ]{} node\[blue,xshift=-8,yshift=-10\] ; [ pt = 1pt (+180:1) – (:1) (:) circle (); (:) – (:--.08) ]{} node\[blue,xshift=-10,yshift=-13\] ; [ pt = 1pt (+180:1) – (:1) (:) circle (); (:) – (:--.08) ]{} node\[blue,xshift=-12,yshift=-18\] ; [ pt = 1pt (+180:1) – (:1) (:) circle (); (:) – (:--.08) ]{} node\[red,xshift=5,yshift=-5\] ; [ pt = 1pt (+180:1) – (:1) (:) circle (); (:) – (:--.08) ]{} node\[red,xshift=5,yshift=-5\] ; [ pt = 1pt (+180:1) – (:1) (:) circle (); (:) – (:--.08) ]{} node\[red,xshift=5,yshift=-5\] ; at (-0.737,0.065) [$L_3$]{}; at (-0.6605,-0.2137) [$L_2$]{}; at (-0.5566,-0.5306) [$L_1$]{}; at (0.7709,0.1904) [$M_1$]{}; at (0.7273,-0.0991) [$M_2$]{}; at (0.6017,-0.405) [$M_3$]{};
For the last condition, we choose an element $x_0 \in D^\times$ with $(x_0)^2 = a$ and $x_0 \notin F$. Then for each $k \in \{1,\ldots,r\}$, we have $(\tau_k(x_0))^2 = \sigma_k(a)I_2$ and $\tau_k(x_0) \notin {\mathbb{R}}\cdot I_2$, where $I_2 \in M_2({\mathbb{R}})$ is the identity matrix. Recall that $\sigma_k(a) > 0$ by assumption. So the minimal polynomial of $\tau_k(x_0)$ over ${\mathbb{R}}$ is $\bigl(X + \sqrt{\sigma_k\smash[b]{(a)}}\bigr)\bigl(X - \sqrt{\sigma_k\smash[b]{(a)}}\bigr) \in {\mathbb{R}}[X]$. Hence, each $\tau_k(x_0)$ has two distinct real eigenvalues and $\tau_1(x_0)$ is diagonalizable over $F(\sqrt{a})$. So by \[thm:polar-regular-quaternion-algebra\], there exists a unique maximal flat $B_0 \subset ({\mathbb{H}}^2)^r$ that is stabilized by $x_0$. The group $\operatorname{SL}_2({\mathbb{R}})^r$ acts transitively on the set of all maximal flats in $({\mathbb{H}}^2)^r$, and so for each $j \in \{1,\ldots,n\}$, we can find some $T_j \in \operatorname{SL}_2({\mathbb{R}})^r$ with $B_j = T_j \cdot B_0$.
By the real approximation theorem [@Milne-Book Theorem 25.70], the image of $D^1$ in $\operatorname{SL}_2({\mathbb{R}})^r$ under the maps $\tau_1,\ldots,\tau_r$ is dense. Since the subsets $UT_j \subset \operatorname{SL}_2({\mathbb{R}})^r$ are open, there exist $x_j \in D^1$ and $v_j \in U$ with $(\tau_1(x_j),\ldots,\tau_r(x_j)) = v_j T_j$. Next, from $v_j \cdot B_j = v_jT_j \cdot (T_j^{-1} \cdot B_j) = x_j \cdot B_0$, we conclude that $$x_jx_0x_j^{-1} \cdot (v_j \cdot B_j) = x_j x_0 \cdot B_0 = x_j \cdot B_0 = v_j \cdot B_j.$$ This shows that $v_j \cdot B_j \subset ({\mathbb{H}}^2)^r$ is stabilized by $\beta_j := x_j x_0 x_j^{-1} \in D^\times$. We now replace each $B_j$ by $v_j \cdot B_j$ and then all four conditions are satisfied.
Controlling the intersections in the quotient {#subsec:controlling}
---------------------------------------------
Our next goal is to find a finite covering space of $\Gamma \backslash ({\mathbb{H}}^2)^r$ in which the image of the flats $A_i$ and $B_j$ from \[thm:configuration\] are embedded submanifolds and to control the intersections of these submanifolds.
In order to ease notation, we fix throughout this subsection two maximal flats $A, B \subset ({\mathbb{H}}^2)^r$ that are either disjoint or intersect transversally in a single point and we fix $\alpha \in D^1$ and $\beta \in D^\times$ that stabilizes $A$ and $B$, respectively. We assume that $\tau_1(\alpha),\ldots,\tau_r(\alpha)$ and $\tau_1(\beta),\ldots,\tau_r(\beta)$ each have two distinct real eigenvalues and that $\tau_1(\beta)$ is diagonalizable over $F(\sqrt{a})$. We also assume that $A$ is $\Gamma$-compact.
\[thm:centralizer\] There exists a subgroup of finite index $\Gamma_{{\textnormal{cent}}} \subset \Gamma$ such that every element of $\Gamma_{\textnormal{cent}}$ which stabilizes $A$ commutes with $\alpha$, and every element of $\Gamma_{\textnormal{cent}}$ which stabilizes $B$ commutes with $\beta$.
By \[thm:polar-regular-quaternion-algebra\], the centralizer $C_{D^\times}(\alpha)$ is a subgroup of finite index in $(D^\times)_A$. So there exist $y_1,\ldots,y_m \in (D^\times)_A$ with $$(D^\times)_A = C_{D^\times}(\alpha) \sqcup C_{D^\times}(\alpha)y_1 \sqcup \ldots \sqcup C_{D^\times}(\alpha)y_m.$$ We now use the rings ${\mathbb{A}_{f,F}}$ and ${\mathbb{O}_{f,F}}$ from \[sec:adeles-and-congruence-subgroups\]. For each $i \in \{1,\ldots,m\}$, the set $C_{{\mathbf{GL}}_D({\mathbb{A}_{f,F}})}(\alpha)y_i \subset {\mathbf{GL}}_D({\mathbb{A}_{f,F}})$ is a closed subset that does not contain the identity. Hence, by \[thm:adelic-points-basis\], we find a nonzero ideal ${\mathfrak{a}}\subset {\mathcal{O}}_F$ such that for all $i \in \{1,\ldots,m\}$, we have ${\mathbf{GL}}_{\Lambda}({\mathbb{O}_{f,F}})({\mathfrak{a}}) \cap C_{{\mathbf{GL}}_D({\mathbb{A}_{f,F}})}(\alpha)y_i = \emptyset$. Consequently, we have $$\Gamma({\mathfrak{a}})_A \subset (D^\times)_A \cap {\mathbf{GL}}_\Lambda({\mathbb{O}_{f,F}})({\mathfrak{a}}) \subset C_{D^\times}(\alpha).$$ Similarly, we find a nonzero ideal ${\mathfrak{b}}\subset {\mathcal{O}}_F$ with $\Gamma({\mathfrak{b}})_B \subset C_{D^\times}(\beta)$. Then $\Gamma_{\textnormal{cent}}:= \Gamma({\mathfrak{a}}) \cap \Gamma({\mathfrak{b}})$ is of finite index in $\Gamma$, and the proof is complete.
We can now show that the images of $A$ and $B$ in some finite covering space of $\Gamma \backslash ({\mathbb{H}}^2)^r$ are embedded submanifolds:
\[thm:embedding\] There exists a subgroup of finite index $\Gamma_{{\textnormal{emb}}} \subset \Gamma_{{\textnormal{cent}}}$ so that for every subgroup of finite index $\Gamma' \subset \Gamma_{{\textnormal{emb}}}$, the natural maps $$\Gamma'_A \backslash A \to \Gamma' \backslash ({\mathbb{H}}^2)^r
\quad \text{and} \quad
\Gamma'_B \backslash B \to \Gamma' \backslash ({\mathbb{H}}^2)^r$$ are closed embeddings whose images are orientable flat totally geodesic $r$-dimensional submanifolds of $\Gamma' \backslash ({\mathbb{H}}^2)^r$.
The algebraic group ${\mathbf{SL}}_D$ is connected and semisimple. Because of $\alpha \in {\mathbf{SL}}_D(F)$, there exists by [@Milne-Book pp. 33–35] a unique smallest $F$-subgroup $C_{{\mathbf{SL}}_D}(\alpha)$ of ${\mathbf{SL}}_D$ such that for all fields $K$ with $F \subset K$, we have $$\bigl(C_{{\mathbf{SL}}_D}(\alpha)\bigr)(K) = C_{{\mathbf{SL}}_D(K)}(\alpha).$$ Since $\tau_1(\alpha)$ is a diagonalizable matrix, we see that $C_{{\mathbf{SL}}_D}(\alpha)$ becomes isomorphic to ${\mathbf{GL}}_1$ over an algebraic closure of $F$, so $C_{{\mathbf{SL}}_D}(\alpha)$ is connected and reductive.
Let ${\mathbf{G}} := \operatorname{Res}_{F/{\mathbb{Q}}}({\mathbf{SL}}_D)$ and ${\mathbf{H}} := \operatorname{Res}_{F/{\mathbb{Q}}}\bigl(C_{{\mathbf{SL}}_D}(\alpha)\bigr)$. Then ${\mathbf{G}}$ is a connected semisimple ${\mathbb{Q}}$-group and ${\mathbf{H}}$ is a connected reductive ${\mathbb{Q}}$-subgroup of ${\mathbf{G}}$. For each $i \in \{r+1,\ldots,d\}$, we choose an isomorphism $\rho_i \colon D \otimes_F \operatorname{res}_{\sigma_i}({\mathbb{R}}) {\xrightarrow{\raisebox{-1pt}[0ex][0ex]{$\scriptscriptstyle\cong$}}}{\mathcal{H}}$, where ${\mathcal{H}}$ is the real Hamilton quaternion algebra. The isomorphisms $\tau_1,\ldots,\tau_r$ and $\rho_{r+1},\ldots,\rho_d$ induce an isomorphism $$\begin{aligned}
G := {\mathbf{G}}({\mathbb{R}}) {\xrightarrow{\raisebox{-1pt}[0ex][0ex]{$\scriptscriptstyle\cong$}}}\operatorname{SL}_2({\mathbb{R}})^r \times ({\mathcal{H}}^1)^{d-r},\end{aligned}$$ which maps the group $H := {\mathbf{H}}({\mathbb{R}})$ to $\prod_{i=1}^r C_{\operatorname{SL}_2({\mathbb{R}})}(\tau_i(\alpha)) \times \prod_{i=r+1}^{d} C_{{\mathcal{H}}^1}(\rho_i(\alpha))$. Let $K \subset G$ be the preimage of $\operatorname{SO}(2)^r \times ({\mathcal{H}}^1)^{d-r}$ under this isomorphism. Then $K$ is a maximal compact subgroup of $G$ and $K_H := K \cap H$ is a maximal compact subgroup of $H$. Consider now the quotient spaces $X_G := G/K$ and $X_H := H/K_H$ and the embedding $j_H \colon X_H \hookrightarrow X_G$ induced by the inclusion $H \hookrightarrow G$. Fix a point $x_0 \in A$. Then the diffeomorphism $$X_G {\xrightarrow{\raisebox{-1pt}[0ex][0ex]{$\scriptscriptstyle\cong$}}}({\mathbb{H}}^2)^r,
\quad
gK \mapsto g \cdot x_0$$ maps $j_H(X_H)$ onto the flat $A$. Note that ${\mathbf{H}}({\mathbb{Q}}) = C_{D^1}(\alpha)$. So by \[thm:centralizer,thm:polar-regular-quaternion-algebra\], we have $\Gamma'_A = \Gamma' \cap {\mathbf{H}}({\mathbb{Q}})$ for every subgroup of finite index $\Gamma' \subset \Gamma_{\textnormal{cent}}$. Hence, by \[thm:geometric-cycles\], there exists a subgroup of finite index $\Gamma_0 \subset \Gamma_{\textnormal{cent}}$ such that for all subgroups of finite index $\Gamma' \subset \Gamma_0$, the map $\Gamma'_A \backslash A \to \Gamma' \backslash ({\mathbb{H}}^2)^r$ is a closed embedding whose image is an orientable flat totally geodesic submanifold.
Similarly, we obtain a subgroup of finite index $\Gamma_1 \subset \Gamma_{\textnormal{cent}}$ such that for every subgroup of finite index $\Gamma' \subset \Gamma_1$, the map $\Gamma'_B \backslash B \to \Gamma' \backslash ({\mathbb{H}}^2)^r$ is a closed embedding whose image is an orientable flat totally geodesic submanifold. We set $\Gamma_{\textnormal{emb}}:= \Gamma_0 \cap \Gamma_1$ and the proof is complete.
\[thm:technical-disjoint-union\] Let $\Gamma' \subset \Gamma_{{\textnormal{emb}}}$ be a subgroup of finite index. Then $\Gamma'A$ is a disjoint union of copies of $A$, that is, for any $\gamma \in \Gamma'$, we have $\gamma A = A$ or $\gamma A \cap A = \emptyset$. Similarly, $\Gamma'B$ is a disjoint union of copies of $B$.
Let $\gamma \in \Gamma'$ and assume that $\gamma A \cap A \neq \emptyset$. Then there exist $x_1,x_2 \in A$ with $x_2 = \gamma x_1$, and so we have $\Gamma' x_1 = \Gamma' x_2$. Since the map $\Gamma'_A \backslash A \to \Gamma' \backslash ({\mathbb{H}}^2)^r$ is injective by \[thm:embedding\], it follows that $\Gamma'_{A} x_1 = \Gamma'_{A} x_2$. So there exists some $\delta \in \Gamma'_A$ with $x_1 = \delta x_2$. Hence, we have $\delta\gamma x_1 = x_1$, and because $\Gamma'$ is torsion-free, this implies that $\gamma = \delta^{-1}$. So we have $\gamma \in \Gamma'_A$, or, in other words, $\gamma A = A$. The statement for $\Gamma'B$ can be proven analogously.
\[rem:induced-orientations\] For any subgroup of finite index $\Gamma' \subset \Gamma_{{\textnormal{emb}}}$, the images of the flats $A$ and $B$ in $\Gamma' \backslash ({\mathbb{H}}^2)^r$ can be oriented as follows: We choose orientations $A^+$ on $A$ and $B^+$ on $B$. Then we define $\Gamma'$-invariant orientations on $\Gamma'A$ and $\Gamma'B$ by $(\gamma A)^+ := \gamma A^+$ and $(\gamma B)^+ := \gamma B^+$ for any $\gamma \in \Gamma'$. Note that by \[thm:embedding\], the maps $\Gamma'A \to \Gamma'_A \backslash A$ and $\Gamma'B \to \Gamma'_B \backslash B$ are covering maps and their images are diffeomorphic to the images of $A$ and $B$ in $\Gamma' \backslash ({\mathbb{H}}^2)^r$, respectively. Since $\Gamma_{{\textnormal{emb}}} \subset \Gamma_{\textnormal{cent}}$, we have by \[thm:centralizer,thm:polar-regular-quaternion-algebra\] that $\Gamma'_A$ and $\Gamma'_B$ act by orientation-preserving isometries on $A$ and $B$, respectively. So we get induced orientations on the images of $A$ and $B$ in $\Gamma' \backslash ({\mathbb{H}}^2)^r$.
Our next task is to find a finite covering space of the locally symmetric space $\Gamma \backslash ({\mathbb{H}}^2)^r$ in which we can control the intersection of the images of $A$ and $B$. We start with some technical results:
\[thm:technical-double-cosets\] Let $\Gamma' \subset \Gamma_{{\textnormal{emb}}}$ be a subgroup of finite index. Then for any $\gamma_1,\gamma_2 \in \Gamma'$, we have $\Gamma'_{B}\gamma_1 A = \Gamma'_{B}\gamma_2 A$ if and only if there exists some $\delta \in \Gamma'_{B}$ with $\gamma_1 A = \delta \gamma_2 A$.
If there exists some $\delta \in \Gamma'_{B}$ with $\gamma_1 A = \delta \gamma_2 A$, then we also have $\Gamma'_{B}\gamma_1 A = \Gamma'_{B}\gamma_2 A$. Conversely, if $\Gamma'_B\gamma_1 A = \Gamma'_{B}\gamma_2 A$, then $\gamma_1 A \cap \delta \gamma_2 A \neq \emptyset$ for some $\delta \in \Gamma'_{B}$, and so $\gamma_1 A = \delta \gamma_2 A$ by \[thm:technical-disjoint-union\].
\[thm:finiteness\] For every subgroup of finite index $\Gamma' \subset \Gamma_{\textnormal{emb}}$, we have $$\# \bigl\{ \Gamma'_B\gamma A : \gamma \in \Gamma' \text{ with } \gamma A \cap B \neq \emptyset \bigr\} < \infty.$$
Let $\pi \colon ({\mathbb{H}}^2)^r \to \Gamma' \backslash ({\mathbb{H}}^2)^r$ be the projection map. We write $\mathcal{F} := \pi(A)$ and $\mathcal{G} := \pi(B)$. Since $\mathcal{F}$ and $\mathcal{G}$ are closed totally geodesic submanifolds of $\Gamma' \backslash ({\mathbb{H}}^2)^r$ and $\mathcal{F}$ is compact, it follows that $\mathcal{F} \cap \mathcal{G}$ is a compact manifold. In particular, $\mathcal{F} \cap \mathcal{G}$ has only finitely many path-connected components. Thus, it suffices to show that for all $\gamma_0,\gamma_1 \in \Gamma'$ for which there is a continuous path in $\mathcal{F} \cap \mathcal{G}$ connecting a point in $\pi(\gamma_0 A \cap B)$ to a point in $\pi(\gamma_1 A \cap B)$, we have $$\Gamma'_{B}\gamma_0{A} = \Gamma'_{B}\gamma_1{A}.$$ Let $c \colon [0,1] \to A' \cap B'$ be such a path and choose preimages $x_i \in \gamma_i A \cap B$ with $\pi(x_i) = c(i)$ for $i \in \{0,1\}$. Since $j_B \colon \Gamma'_{B} \backslash B \to B'$ is a diffeomorphism, $c$ induces a path $$c_B := j_B^{-1} \circ c \colon [0,1] \to \Gamma'_{B} \backslash B.$$ The map $p_B \colon \Gamma' B \to \Gamma'_B \backslash B$, $\gamma \cdot b \mapsto \Gamma_B' b$ is well-defined by \[thm:embedding\] and is a covering map. By the lifting property of $p_B$ and the fact that $p_B(x_0) = c_B(0)$, there exists a path $\widetilde{c} \colon [0,1] \to \Gamma'B$ with $\widetilde{c}(0) = x_0$ such that the diagram $$\begin{tikzcd}[column sep=large]
\Gamma'B \ar{r}{p_B} & \Gamma'_B\backslash B \ar{r}{j_B} & B' \\
{}& {}[0,1] \ar[dashed]{ul}{\widetilde{c}} \ar{u}{c_B} \ar[swap]{ur}{c}
\end{tikzcd}$$ commutes. From $c([0,1]) \subset A'$, we deduce that $\widetilde{c}([0,1]) \subset \Gamma' A$. But $\Gamma' A$ is a disjoint union of copies of $A$ by \[thm:technical-disjoint-union\], and so $\widetilde{c}(0) = x_0 \in \gamma_0A$ implies that the image of $\widetilde{c}$ must be fully contained in $\gamma_0 A$. In particular, $\widetilde{c}(1) \in \gamma_0 A$.
On the other hand, using $\pi(\widetilde{c}(1)) = c(1) = \pi(x_1)$ and the injectivity of $j_B$, we see that $$\Gamma'_{B} \widetilde{c}(1)
= p_B(\widetilde{c}(1))
= p_B(x_1)
= \Gamma'_{B} x_1.$$ Because of $x_1 \in \gamma_1 A$, this shows that $\widetilde{c}(1) \in \delta \gamma_1 A$ for some $\delta \in \Gamma'_{B}$. In conclusion, we have $\widetilde{c}(1) \in \gamma_0 A \cap \delta \gamma_1 A$, and so \[thm:technical-disjoint-union\] implies that $\delta \gamma_1 A = \gamma_0 A$. Hence, by \[thm:technical-double-cosets\], we have $$\Gamma'_{B}\gamma_0{A} = \Gamma'_{B}\gamma_1{A}. \qedhere$$
\[thm:surjective-map\] Let $\Gamma' \subset \Gamma_{{\textnormal{emb}}}$ be a subgroup of finite index. Then there exist $\gamma_1,\ldots,\gamma_m \in \Gamma'$ such that the intersection of the images of $A$ and $B$ in $\Gamma' \backslash ({\mathbb{H}}^2)^r$ is the image of the projection map $$\bigcup_{i=1}^m \gamma_i A \cap B \to \Gamma' \backslash ({\mathbb{H}}^2)^r.$$
By \[thm:finiteness\], there exist $\gamma_1,\ldots,\gamma_m \in \Gamma'$ with $$\label{eq:projection-surjective-map}
\bigl\{ \Gamma_B'\gamma A : \gamma \in \Gamma' \text{ with } \gamma A \cap B \neq \emptyset \bigr\}
=
\bigl\{
\Gamma'_B\gamma_1 A,
\ldots,
\Gamma'_B\gamma_m A
\bigr\}.$$ Let $\pi \colon ({\mathbb{H}}^2)^r \to \Gamma' \backslash ({\mathbb{H}}^2)^r$ be the projection map. For each $i \in \{1,\ldots,m\}$, we have $\pi(\gamma_i A \cap B) \subset \pi(A) \cap \pi(B)$. Conversely, let $z \in \pi(A) \cap \pi(B)$. Then there is some $x \in \Gamma' A \cap B$ with $z = \Gamma'x$. Let $\gamma \in \Gamma'$ with $x \in \gamma A$. By , we have $\Gamma'_B \gamma A = \Gamma'_B \gamma_i A$ for some $i \in \{1,\ldots,m\}$. So by \[thm:technical-double-cosets\], there exists some $\delta \in \Gamma'_B$ with $\gamma A = \delta \gamma_i A$. Hence, there is some $y \in A$ with $x = \delta \gamma_i y$. From $\gamma_i y = \delta^{-1}x \in B$ and $\Gamma' x = \Gamma' \delta \gamma_i y = \Gamma' \gamma_i y$, we deduce that $z = \Gamma'x \in \pi(\gamma_iA \cap B)$.
The next two lemmas will be used in the proof of \[thm:product-formula-closure\].
\[thm:centralizer-intersection\] For the centralizers of $\alpha$ and $\beta$ in $D \otimes_F {\mathbb{A}_{f,F}}$, we have $$C_{D \otimes_F {\mathbb{A}_{f,F}}}(\beta) \cap C_{D \otimes_F {\mathbb{A}_{f,F}}}(\alpha)
= {\mathbb{A}_{f,F}}.$$
We only show $C_D(\alpha) \cap C_D(\beta) = F$. This suffices, because taking the tensor product with ${\mathbb{A}_{f,F}}$ commutes with taking the centralizer. So suppose to the contrary that there exists $x \in D$ with $x \notin F$, $\alpha x=x \alpha$, and $\beta x=x \beta$. Note that $\alpha$ and $\beta$ do not commute with each other, because otherwise we would have $A = B$ by \[thm:polar-regular-quaternion-algebra\], in contradiction to our assumptions on $A$ and $B$. This implies $\alpha,\beta \notin F$, and so we see that $\{1,x,\alpha,\beta\}$ is a linearly independent subset with four distinct elements of the linear subspace $C_{D}(x) \subset D$. Since $\dim_F(D) = 4$, it follows that $C_D(x) = D$, and so we have $x \in Z(D) = F$. But this contradicts the assumption $x \notin F$.
To simplify the notation, we will from now write $C_{G}(g) := \{ h \in G : gh = hg\}$ for a group $G$ whenever multiplication with $g$ is defined, even if $g \not\in G$.
\[thm:roots-of-unity-transitively\] The group $\mu_2({\mathbb{O}_{f,F}}) := \{ \nu \in {\mathbb{O}_{f,F}} : \nu^2 = 1\}$ of second roots of unity in ${\mathbb{O}_{f,F}}$ acts transitively by $\nu \cdot (u,v) := (\nu^{-1} u, \nu v)$ on the fibers of the map $$C_{{\mathbf{SL}}_\Lambda({\mathbb{O}_{f,F}})}(\beta) \times C_{{\mathbf{SL}}_\Lambda({\mathbb{O}_{f,F}})}(\alpha) \to {\mathbf{SL}}_\Lambda({\mathbb{O}_{f,F}}),
\quad
(u,v) \mapsto uv.$$
Let $(u_1,v_1),(u_2,v_2) \in C_{{\mathbf{SL}}_\Lambda({\mathbb{O}_{f,F}})}(\beta) \times C_{{\mathbf{SL}}_\Lambda({\mathbb{O}_{f,F}})}(\alpha)$ with $u_1v_1 = u_2v_2$. Then $\nu := u_2^{-1}u_1 = v_2v_1^{-1}$ commutes with both $\alpha$ and $\beta$, and so $\nu$ is a scalar in ${\mathbb{O}_{f,F}}$ by \[thm:centralizer-intersection\]. Since $N(\nu) = 1$ and $N(\nu) = \nu^2$, we have $\nu \in \mu_2({\mathbb{O}_{f,F}})$. It follows that $\nu^{-1} u_1 = u_1 \nu^{-1} = u_2$ and $\nu v_1 = v_2$. Hence we have $\nu \cdot (u_1,v_1) = (u_2,v_2)$.
Recall from \[thm:polar-regular-quaternion-algebra\] that $C_{D^\times}(\alpha)$ and $C_{D^\times}(\beta)$ act by orientation-preserving isometries on the flats $A$ and $B$, respectively. The next two propositions will, combined with \[thm:surjective-map\], allow us to control the intersection of the images of $A$ and $B$ in a finite covering space of $\Gamma \backslash ({\mathbb{H}}^2)^r$.
\[thm:product-formula-closure\] For every $\gamma \in \Lambda^1$ that is in the closure of $C_{\Lambda^1}(\beta)C_{\Lambda^1}(\alpha)$ in ${\mathbf{SL}}_\Lambda({\mathbb{O}_{f,F}})$, there exist $x \in C_{D^\times}(\beta)$ and $y \in C_{D^\times}(\alpha)$ such that $\gamma = xy$ and such that $x$ acts by orientation-preserving isometries on $({\mathbb{H}}^2)^r$.
**Step 1:** We first find $x' \in C_{{\mathbf{SL}}_\Lambda({\mathbb{O}_{f,F}})}(\beta)$ and $y' \in C_{{\mathbf{SL}}_\Lambda({\mathbb{O}_{f,F}})}(\alpha)$ such that $\gamma = x'y'$. This is possible because $C_{{\mathbf{SL}}_\Lambda({\mathbb{O}_{f,F}})}(\beta)$ and $C_{{\mathbf{SL}}_\Lambda({\mathbb{O}_{f,F}})}(\alpha)$ are both closed in ${\mathbf{SL}}_\Lambda({\mathbb{O}_{f,F}})$, so their product is also closed and contains the set $C_{\Lambda^1}(\beta)C_{\Lambda^1}(\alpha)$, hence also the closure point $\gamma$ of this set.
**Step 2:** Next, we find some $c \in {\mathbb{A}_{f,F}}$ with $cx' \in D^\times$. To achieve this, we observe that $x'$ is a solution in $D \otimes_F {\mathbb{A}_{f,F}}$ of the homogeneous system of linear equations $$\begin{aligned}
x' \alpha =&\; (\gamma \alpha \gamma^{-1}) x', \\
x' \beta =&\; \beta x'.
\end{aligned}$$ The coefficients of this system are in $F$. Let $\mathcal{B}$ be an $F$-basis for the space of solutions of this system in $D$. Then the solution space in $D \otimes_F {\mathbb{A}_{f,F}}$ is the ${\mathbb{A}_{f,F}}$-span of $\mathcal{B}$. In particular, $\mathcal{B} \neq \emptyset$. Moreover, the function $x \mapsto N(x)$ on the solution space in $D$ can be expressed in coordinates with respect to $\mathcal{B}$ by some multivariate polynomial $P \in F[X_1,\ldots,X_m]$. Because of $N(x') \neq 0$, we have $P \neq 0$. So since $F$ is an infinite field, there exists a solution with nonzero reduced norm in $D$, that is, there exists an element $x \in D^\times$ satisfying $$\begin{aligned}
x \alpha =&\; (\gamma \alpha \gamma^{-1}) x, \\
x \beta =&\; \beta x.\end{aligned}$$ The element $x^{-1}x' \in D \otimes_F {\mathbb{A}_{f,F}}$ commutes with $\beta$. It also commutes with $\alpha$ because from the above two linear systems of equations, we deduce that $$x^{-1}x'\alpha = x^{-1} (\gamma \alpha \gamma^{-1}) x' = x^{-1} (x \alpha x^{-1}) x' = \alpha x^{-1} x'.$$ So by \[thm:centralizer-intersection\], we have $x = cx' \in D^\times$ for some $c \in {\mathbb{A}_{f,F}}$ as required.
Let $y :=c^{-1}y'$. Then $y = c^{-1}(x')^{-1}\gamma = x^{-1}\gamma \in D^\times$, and so we have $\gamma = xy$ with $x \in C_{D^\times}(\beta)$ and $y \in C_{D^\times}(\alpha)$. It remains to show that $x$ acts by orientation-preserving isometries on $({\mathbb{H}}^2)^r$. We do this in the next two steps.
**Step 3:** We now show that $C_{\Lambda^1}(\beta)x' \cap \mu_2({\mathbb{O}_{f,F}}) U \neq \emptyset$ for every open neighborhood of the identity $U \subset {\mathbf{SL}}_\Lambda({\mathbb{O}_{f,F}})$. Assume to the contrary that $U \subset {\mathbf{SL}}_\Lambda({\mathbb{O}_{f,F}})$ is an open neighborhood of the identity with $C_{\Lambda^1}(\beta)x' \cap \mu_2({\mathbb{O}_{f,F}}) U = \emptyset$. Then we have $$\bigl(C_{\Lambda^1}(\beta) x' \times y'C_{\Lambda^1}(\alpha)\bigr)
\cap
\bigl(\mu_2({\mathbb{O}_{f,F}}) U \times C_{{\mathbf{SL}}_\Lambda({\mathbb{O}_{f,F}})}(\alpha)\bigr) = \emptyset.$$ The set $\mu_2({\mathbb{O}_{f,F}}) U \times C_{{\mathbf{SL}}_\Lambda({\mathbb{O}_{f,F}})}(\alpha)$ is invariant under the action of $\mu_2({\mathbb{O}_{f,F}})$ defined in \[thm:roots-of-unity-transitively\] and $\mu_2({\mathbb{O}_{f,F}})$ acts transitively on the fibers of the multiplication map by this lemma. Hence, for the images under this map, we obtain $$\label{eq:doublecoset-intersection-pre}
C_{\Lambda^1}(\beta) \gamma C_{\Lambda^1}(\alpha) \cap \mu_2({\mathbb{O}_{f,F}}) U C_{{\mathbf{SL}}_\Lambda({\mathbb{O}_{f,F}})}(\alpha) = \emptyset.$$ Since $U$ is an open neighborhood of the identity in ${\mathbf{SL}}_\Lambda({\mathbb{O}_{f,F}})$, there exists by \[thm:adelic-points-basis\] a nonzero ideal ${\mathfrak{a}}\subset {\mathcal{O}}_F$ with ${\mathbf{SL}}_\Lambda({\mathbb{O}_{f,F}})({\mathfrak{a}}) \subset U$. Consequently, we have $\Lambda^1({\mathfrak{a}}) \subset U$ and so implies that $$\label{eq:doublecoset-intersection}
C_{\Lambda^1}(\beta) \gamma C_{\Lambda^1}(\alpha) \cap \Lambda^1({\mathfrak{a}}) = \emptyset.$$ On the other hand, $\gamma$ is in the closure of $C_{\Lambda^1}(\beta)C_{\Lambda^1}(\alpha)$ and so we have $C_{\Lambda^1}(\beta)C_{\Lambda^1}(\alpha) \cap \Lambda^1({\mathfrak{a}})\gamma \neq \emptyset$. Hence, there exist $\gamma_\beta \in C_{\Lambda^1}(\beta)$, $\gamma_\alpha \in C_{\Lambda^1}(\alpha)$ and $\gamma_0 \in \Lambda^1({\mathfrak{a}})$ with $\gamma_\beta\gamma_\alpha = \gamma_0 \gamma$. Since $\Lambda^1({\mathfrak{a}})$ is normal in $\Lambda^1$, we obtain $$\gamma_\beta^{-1}\gamma_0^{-1}\gamma_\beta = \gamma_\beta^{-1} \gamma \gamma_\alpha^{-1}
\in C_{\Lambda^1}(\beta) \gamma C_{\Lambda^1}(\alpha) \cap \Lambda^1({\mathfrak{a}})$$ in contradiction to . This proves the claim and thus finishes this step.
**Step 4:** Finally, we show that $x = cx'$ acts by orientation-preserving isometries on $({\mathbb{H}}^2)^r$. Assume to the contrary that this is not the case. Then there must exist some $i \in \{1,\ldots,r\}$ with $\det(\tau_i(x)) = \sigma_i(N(x)) = \sigma_i(c^2) < 0$. Let $E := F(\sqrt{a})$. We can extend $\sigma_i$ to a real embedding $\widetilde{\sigma_i} \colon E \hookrightarrow {\mathbb{R}}$ as in the proof of \[thm:quaternion-algebra-trick\], because by assumption we have $\sigma_i(a) > 0$.
Recall that $\tau_1(D^\times) \subset M_2(E)$. So $\tau_1$ induces an $F$-algebra homomorphism $D \to M_2(E)$ and hence also an $F$-homomorphism ${\mathbf{GL}}_D \to \operatorname{Res}_{E/F}{\mathbf{GL}}_2$. By [@Platonov p. 15], we have $(\operatorname{Res}_{E/F}{\mathbf{GL}}_2)({\mathbb{A}_{f,F}}) \cong {\mathbf{GL}}_2({\mathbb{A}_{f,E}})$, and so we get a continuous group homomorphism $$\Phi \colon {\mathbf{GL}}_D({\mathbb{A}_{f,F}}) \to {\mathbf{GL}}_2({\mathbb{A}_{f,E}})$$ that extends $\tau_1$ on $D^\times$. The matrix $\Phi(\beta) = \tau_1(\beta) \in \operatorname{GL}_2(E)$ is by assumption diagonalizable over $E$ with two distinct eigenvalues. So there exists a one-dimensional subspace $L \subset E^2$ which is invariant under $\tau_1(\beta)$. The corresponding eigenspace in $({\mathbb{A}_{f,E}})^2$ of $\tau_1(\beta)$ is ${\mathbb{A}_{f,E}}\cdot L$, and so every matrix in $M_2({\mathbb{A}_{f,E}})$ that commutes with $\tau_1(\beta)$ stabilizes ${\mathbb{A}_{f,E}}\cdot L$. Let now $v \in L$ be a nonzero vector and let $\ell \in \{1,2\}$ be such that the $\ell$-th coordinate of $v$ is $v_\ell \neq 0$. Consider the map $$\label{eq:definition-phi}
s \colon C_{{\mathbf{GL}}_D({\mathbb{A}_{f,F}})}(\beta) \to {\mathbf{GL}}_1({\mathbb{A}_{f,E}}),
\quad
g \mapsto \biggl(\frac{(\Phi(g)v)_\ell}{v_\ell}\biggr)^2.$$ Note that $s$ is multiplicative and so its image is contained in ${\mathbf{GL}}_1({\mathbb{A}_{f,E}}) = {\mathbb{A}_{f,E}}^\times$. Moreover, $s$ is continuous because $\Phi$ is continuous and ${\mathbb{A}_{f,E}}$ is a topological $E$-algebra. We have $s(C_{D^\times}(\beta)) \subset (E^\times)^2$, and so by writing $x' = c^{-1}cx'$, we see that $$s(\mu_2({\mathbb{O}_{f,F}}) C_{\Lambda^1}(\beta)x') \subset (E^\times)^2 c^{-2} s(cx')$$ is contained in $E^\times$ and has only negative images under $\widetilde{\sigma_i}$ because of $\widetilde{\sigma_i}(c^2) < 0$. Note that $V_+ := \{ v \in {\mathcal{O}}_E^\times: \widetilde{\sigma_i}(v) > 0 \}$ is a subgroup of finite index in ${\mathcal{O}}_E^\times$. So by \[thm:chevalley\], there exists a nonzero ideal ${\mathfrak{a}}\subset {\mathcal{O}}_E$ with ${\mathcal{O}}_E^\times({\mathfrak{a}}) \subset V_+$. Hence ${\mathbb{O}_{f,E}}^\times({\mathfrak{a}}) \cap E^\times \subset {\mathcal{O}}_E^\times({\mathfrak{a}}) \subset V_+$, and so we obtain $$\mu_2({\mathbb{O}_{f,F}}) C_{\Lambda^1}(\beta)x' \cap s^{-1}({\mathbb{O}_{f,E}}^\times({\mathfrak{a}})) = \emptyset.$$ Since $s$ is continuous, the preimage $s^{-1}({\mathbb{O}_{f,E}}^\times({\mathfrak{a}}))$ is an open neighborhood of the identity in $C_{{\mathbf{GL}}_D({\mathbb{A}_{f,F}})}(\beta)$, and so there exists a nonzero ideal ${\mathfrak{b}}\subset {\mathcal{O}}_F$ with $C_{{\mathbf{GL}}_\Lambda({\mathbb{O}_{f,F}})}(\beta)({\mathfrak{b}}) \subset s^{-1}({\mathbb{O}_{f,E}}^\times({\mathfrak{a}}))$. Let $U := {\mathbf{SL}}_\Lambda({\mathbb{O}_{f,F}})({\mathfrak{b}})$. Then we have $\mu_2({\mathbb{O}_{f,F}}) C_{\Lambda^1}(\beta)x' \cap U = \emptyset$, or equivalently, $$C_{\Lambda^1}(\beta)x' \cap \mu_2({\mathbb{O}_{f,F}}) U = \emptyset.$$ This contradicts the result from the previous step. So $x$ must act by orientation-preserving isometries on $({\mathbb{H}}^2)^r$ and the proof is complete.
\[thm:product-formula-corollary\] There exists a subgroup of finite index $\Gamma_{{\textnormal{prod}}} \subset \Gamma_{{\textnormal{emb}}}$ such that every $\gamma \in \Gamma_{{\textnormal{prod}}}$ with $\gamma A \cap B \neq \emptyset$ can be written as $\gamma =xy$ with $x \in C_{D^\times}(\beta)$ and $y \in C_{D^\times}(\alpha)$ so that $x$ acts by orientation-preserving isometries on $({\mathbb{H}}^2)^r$.
By \[thm:finiteness\], there exist $\gamma_1,\ldots,\gamma_m \in \Gamma_{{\textnormal{emb}}}$ with $$\bigl\{ (\Gamma_{{\textnormal{emb}}})_B\gamma A : \gamma \in \Gamma_{{\textnormal{emb}}},\, \gamma A \cap B \neq \emptyset \bigr\}
=
\bigl\{
(\Gamma_{{\textnormal{emb}}})_B\gamma_1 A,
\ldots,
(\Gamma_{{\textnormal{emb}}})_B\gamma_m A
\bigr\}.$$ We now choose for every $i \in \{1,\ldots,m\}$ a subgroup $\Gamma_{(i)} \subset \Gamma_{{\textnormal{emb}}}$ as follows: If $\gamma_i$ is in the closure of $C_{\Lambda^1}(\beta)C_{\Lambda^1}(\alpha)$ in ${\mathbf{SL}}_\Lambda({\mathbb{O}_{f,F}})$, then we set $\Gamma_{(i)} := \Gamma_{\textnormal{emb}}$. Otherwise, there exists by \[thm:adelic-points-basis\] a nonzero ideal ${\mathfrak{a}}_i \subset {\mathcal{O}}_F$ with $C_{\Lambda^1}(\beta)C_{\Lambda^1}(\alpha) \cap {\mathbf{SL}}_\Lambda({\mathbb{O}_{f,F}})({\mathfrak{a}}_i)\gamma_i = \emptyset$ and we set $\Gamma_{(i)} := \Gamma_{\textnormal{emb}}({\mathfrak{a}}_i)$. Consider $$\Gamma_{{\textnormal{prod}}} := \Gamma_{(1)} \cap \ldots \cap \Gamma_{(m)}.$$ Let $\gamma \in \Gamma_{{\textnormal{prod}}}$ with $\gamma A \cap B \neq \emptyset$. Then we have $(\Gamma_{{\textnormal{emb}}})_B \gamma A = (\Gamma_{{\textnormal{emb}}})_B \gamma_i A$ for some $i \in \{1,\ldots,m\}$. So by \[thm:technical-double-cosets\], there exists some $\delta \in (\Gamma_{{\textnormal{emb}}})_B$ with $\gamma A = \delta \gamma_i A$. Hence we have $\gamma^{-1} \delta \gamma_i A = A$. Let $\tau := \gamma^{-1} \delta \gamma_i$. Then $\tau \in (\Gamma_{{\textnormal{emb}}})_A$ and $$\label{eq:conjugate-equation}
\delta^{-1}\tau = (\delta^{-1}\gamma^{-1}\delta)\gamma_i.$$ Since $\Gamma_{{\textnormal{prod}}}$ is normal in $\Gamma_{{\textnormal{emb}}}$, we have $\delta^{-1}\gamma^{-1}\delta \in \Gamma_{{\textnormal{prod}}}$. Moreover, we have $\delta^{-1} \in (\Gamma_{{\textnormal{emb}}})_B \subset C_{\Lambda^1}(\beta)$ and $\tau \in (\Gamma_{{\textnormal{emb}}})_A \subset C_{\Lambda^1}(\alpha)$. Hence, shows that $$C_{\Lambda^1}(\beta)C_{\Lambda^1}(\alpha) \cap \Gamma_{{\textnormal{prod}}}\gamma_i \neq \emptyset.$$ Now because of $\Gamma_{{\textnormal{prod}}} \subset \Gamma_{(i)}$, we have that $\gamma_i$ must be in the closure of $C_{\Lambda^1}(\beta)C_{\Lambda^1}(\alpha)$ in ${\mathbf{SL}}_\Lambda({\mathbb{O}_{f,F}})$. So by \[thm:product-formula-closure\], we can write $\gamma_i = x_iy_i$ with $x_i \in C_{D^\times}(\beta)$ and $y_i \in C_{D^\times}(\alpha)$ such that $x_i$ acts by orientation-preserving isometries on $({\mathbb{H}}^2)^r$. Observe that $$\gamma = \delta\gamma_i\tau^{-1} = \delta x_i y_i \tau^{-1}.$$ So we have $\gamma = xy$ for $x := \delta x_i \in C_{D^\times}(\beta)$ and $y := y_i \tau^{-1} \in C_{D^\times}(\alpha)$, and $x$ acts by orientation-preserving isometries on $({\mathbb{H}}^2)^r$ because of $N(\delta) = 1$.
We can now show the following result about the intersection of the images of $A$ and $B$ in a finite covering space of the locally symmetric space $\Gamma \backslash ({\mathbb{H}}^2)^r$:
\[thm:intersection\] Let $\Gamma' \subset \Gamma_{{\textnormal{prod}}}$ be a subgroup of finite index. Then the images $\mathcal{F}$ of $A$ and $\mathcal{G}$ of $B$ in $\Gamma' \backslash ({\mathbb{H}}^2)^r$ intersect if and only if $A$ and $B$ intersect in $({\mathbb{H}}^2)^r$. Further, the intersection $\mathcal{F} \cap \mathcal{G}$ is transverse and consists of finitely many points all of which have the same intersection number.
We denote as above by $\mathcal{F}$ and $\mathcal{G}$ the images of $A$ and $B$ in $\Gamma' \backslash ({\mathbb{H}}^2)^r$, respectively. By \[thm:surjective-map\], there exist $\gamma_1,\ldots,\gamma_m \in \Gamma'$ such that the projection map induces a surjection $$\bigcup_{i=1}^m \gamma_i A \cap B \twoheadrightarrow \mathcal{F} \cap \mathcal{G}.$$ Because of $\Gamma' \subset \Gamma_{\textnormal{prod}}$, we can apply \[thm:product-formula-corollary\] and write each $\gamma_i$ as $\gamma_i = x_iy_i$ for some $x_i \in C_{D^\times}(\beta)$ and $y_i \in C_{D^\times}(\alpha)$ such that $x_i$ acts by orientation-preserving isometries on $({\mathbb{H}}^2)^r$. Now choose orientations $A^+$ on $A$ and $B^+$ on $B$, and let $\mathcal{F}$ and $\mathcal{G}$ carry the induced orientations as described in \[rem:induced-orientations\]. By \[thm:polar-regular-quaternion-algebra\], we have $x_i \cdot B^+ = B^+$ and $y_i \cdot A^+ = A^+$, and so we obtain $$\begin{aligned}
\gamma_i \cdot A^+ \cap B^+ = x_i \cdot A^+ \cap B^+ = x_i \cdot (A^+ \cap x_i^{-1} \cdot B^+) = x_i \cdot (A^+ \cap B^+).\end{aligned}$$ This shows that $\mathcal{F}$ and $\mathcal{G}$ intersect if and only if $A$ and $B$ intersect. Furthermore, we see that in this case the intersection of $\mathcal{F}$ and $\mathcal{G}$ is transverse and the intersection number is the same in each point of intersection, because for each $i \in \{1,\ldots,m\}$, the action of $x_i$ maps the intersection $A^+ \cap B^+$ to the intersection $\gamma_iA^+ \cap B^+$ while also preserving the orientation of the ambient space $({\mathbb{H}}^2)^r$.
Finishing the proof of the main theorem {#subsec:finishing}
---------------------------------------
We now use \[thm:intersection\] from the previous subsection to show that the flats that we have constructed in \[thm:configuration\] give us a family of linearly independent real homology classes. For this, we will use de Rham cohomology (see [@Bott-Tu]).
Let $M$ be a smooth oriented $n$-manifold. We write $H_{{\operatorname{dR}}}^*(M)$ for the de Rham cohomology groups of a smooth oriented manifold $M$ and $H_{{\operatorname{dR},\operatorname{c}}}^*(M)$ for the de Rham cohomology groups of $M$ with compact support.
Let $S \subset M$ be a closed oriented $k$-submanifold. We denote by $i \colon S \hookrightarrow M$ the inclusion map. The [*closed Poincaré dual of $S$*]{} is the unique cohomology class $\eta_S \in H_{{\operatorname{dR}}}^{n-k}(M)$ so that for all $\omega \in H_{{\operatorname{dR},\operatorname{c}}}^k(M)$, we have $$\label{eq:poincare-dual}
\int_M \omega \wedge \eta_S = \int_S i^*w.$$ If additionally $S$ is compact and $M$ is diffeomorphic to the interior of a compact manifold with boundary, then we can also define the [*compact Poincaré dual of $S$*]{}. This is the unique cohomology class $\eta'_S \in H_{{\operatorname{dR},\operatorname{c}}}^{n-k}(M)$ for which holds with $\eta'_S$ instead of $\eta_S$ and for all $\omega \in H_{{\operatorname{dR}}}^k(M)$. Note that the natural map $H_{{\operatorname{dR},\operatorname{c}}}^{n-k}(M) \to H_{{\operatorname{dR}}}^{n-k}(M)$ sends $\eta'_S$ to $\eta_S$ (see [@Bott-Tu p. 51]), and that $\eta'_S$ coincides with the image of the fundamental class $[S] \in H_k(S;{\mathbb{R}})$ under the composition $$H_k(S;{\mathbb{R}}) \to H_k(M;{\mathbb{R}}) {\xrightarrow{\raisebox{-1pt}[0ex][0ex]{$\scriptscriptstyle\cong$}}}H_c^{n-k}(M;{\mathbb{R}}) {\xrightarrow{\raisebox{-1pt}[0ex][0ex]{$\scriptscriptstyle\cong$}}}H_{{\operatorname{dR},\operatorname{c}}}^{n-k}(M),$$ where the second map is the Poincaré duality isomorphism and the third map is the de Rham isomorphism for cohomology with compact support.
The Poincaré dual of a transverse intersection of submanifolds is related to the wedge product as follows (see [@Bott-Tu p. 69]):
\[thm:wedge-product-intersection\] Let $M$ be a smooth oriented manifold. Then for any two closed oriented submanifolds $S_1$ and $S_2$ of $M$ that intersect transversally, we have $$\label{eq:wedge-product-intersection}
\eta_{S_1} \wedge \eta_{S_1} = \eta_{S_1 \cap S_2}.$$
In the above proposition, we equip $S_1 \cap S_2$ with the natural orientation induced by the orientations on $M$, $S_1$, and $S_2$. If $S_1$ and $S_2$ are of complementary dimensions in $M$, then this orientation is simply given by the intersection numbers of $S_1$ and $S_2$.
We now use this to finish the proof of our main result. Recall that we call a Riemannian manifold $M$ covered by $({\mathbb{H}}^2)^r$ [*arithmetic*]{} if it is finitely covered by a quotient of $({\mathbb{H}}^2)^r$ by an arithmetically defined lattice in $\operatorname{SL}_2({\mathbb{R}})^r$. The following theorem completes our proof of \[thm:main\]:
\[thm:main-arithmetic\] Let $M$ be an arithmetic Riemannian manifold covered by $({\mathbb{H}}^2)^r$. Then for any $n \in {\mathbb{N}}$, there exists a connected finite covering $M' \to M$ and compact oriented flat totally geodesic $r$-dimensional submanifolds $\mathcal{F}_1,\ldots,\mathcal{F}_n \subset M'$ such that the images of the fundamental classes $[\mathcal{F}_1],\ldots,[\mathcal{F}_n]$ in $H_r(M';{\mathbb{R}})$ are linearly independent.
As explained before, we can and will assume that $M = \Gamma \backslash ({\mathbb{H}}^2)^r$ with $\Gamma$ as in the beginning of this section. By \[thm:configuration\], there exist maximal flats $A_1,\ldots,A_n$ and $B_1,\ldots,B_n$ in $({\mathbb{H}}^2)^r$ such that for all $i,j \in \{1,\ldots,n\}$, \[thm:intersection\] can be applied to the flats $A = A_i$ and $B = B_j$. So there exists a subgroup of finite index $\Gamma' \subset \Gamma$ such that the images of $A_1,\ldots,A_n$ and $B_1,\ldots,B_n$ in $M' := \Gamma' \backslash ({\mathbb{H}}^2)^r$ are closed orientable flat totally geodesic $r$-dimensional submanifolds. We denote them by $\mathcal{F}_1,\ldots,\mathcal{F}_n$ and $\mathcal{G}_1,\ldots,\mathcal{G}_n$ and choose orientations on them as in \[rem:induced-orientations\]. By \[thm:wedge-product-intersection\], we have $$\int_{M'} \eta'_{\mathcal{F}_i} \wedge \eta_{\mathcal{G}_j}
= \int_{M'} \eta_{\mathcal{F}_i} \wedge \eta_{\mathcal{G}_j}
= \int_{M'} \eta_{\mathcal{F}_i \cap \mathcal{G}_j}
= \sum_{p \in \mathcal{F}_i \cap \mathcal{G}_j} I_p(\mathcal{F}_i, \mathcal{G}_j).$$ Furthermore, by what we know about the intersections of $\mathcal{F}_i$ and $\mathcal{G}_j$ from \[thm:intersection\], this sum is nonzero if and only if $\mathcal{F}_i \cap \mathcal{G}_j \neq \emptyset$, which is the case if and only if $i \leq j$. It follows that the matrix $$\biggl( \int_{M'} \eta'_{\mathcal{F}_i} \wedge \eta_{\mathcal{G}_j} \biggr)_{ij} \!\in M_n({\mathbb{R}})$$ is upper triangular with nonzero entries on the diagonal, hence is in $\operatorname{GL}_n({\mathbb{R}})$. Since the map $H_{{\operatorname{dR},\operatorname{c}}}^r(M') \times H_{{\operatorname{dR}}}^{r}(M') \to {\mathbb{R}}$, $(\omega,\tau) \mapsto \int_{M'} \omega \wedge \tau$ is bilinear, it follows that the cohomology classes $
\eta'_{\mathcal{F}_1}, \ldots, \eta'_{\mathcal{F}_n} \in H_{{\operatorname{dR}},c}^{r}(M')$ are linearly independent. They are mapped by the Poincaré duality isomorphism to the images of the fundamental classes $[\mathcal{F}_1],\ldots,[\mathcal{F}_n]$ in $H_r(M';{\mathbb{R}})$, and so these homology classes are also linearly independent.
Acknowledgments {#sec:acknowledgments .unnumbered}
===============
I wish to thank my doctoral advisor Enrico Leuzinger for many helpful comments and discussions. Further, I want to thank Stefan Kühnlein for reading an early manuscript of this work and giving important feedback, Roman Sauer and Steffen Kionke for answering my questions on algebraic groups, and Aurel Page for bringing my attention to the Grunwald–Wang theorem.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The total entropy production of stochastic systems can be divided into three quantities. The first corresponds to the excess heat, whilst the second two comprise the house-keeping heat. We denote these two components the transient and generalised house-keeping heat and we obtain an integral fluctuation theorem for the latter, valid for all Markovian stochastic dynamics. A previously reported formalism is obtained when the stationary probability distribution is symmetric for all variables that are odd under time reversal which restricts consideration of directional variables such as velocity.'
author:
- 'Richard E. Spinney and Ian J. Ford'
date: 'December 20, 2011'
title: 'Non-equilibrium thermodynamics of stochastic systems with odd and even variables'
---
For over $100$ years the statement of the second law of thermodynamics stood simply as the Clausius inequality. However in recent years advances in technology have encouraged the thermodynamic consideration of small systems which has led to the generalisation of the concept of entropy production: it may be associated with individual dynamical realisations revealing a wealth of relations valid out of equilibrium. Such extensions had their origins in the dissipation function of Evans et al. for thermostatted systems that led to the Fluctuation Theorem [@Evans93; @Evans95; @Evans02; @Carberry04] with similar, but asymptotic relations for chaotic systems [@Gallavotti95] which were extended to Langevin dynamics [@kurchan] followed by general Markovian stochastic systems [@GCforstochastic]. Crooks and Jarzynski [@Jarzynski97; @crooksoriginal; @Crooks98] then derived work relations for a variety of dynamics which held for finite times. These were followed by similar generalised relations for the entropy production associated with transitions between stationary states [@hatanosasa], the total entropy production [@seifertoriginal] and the heat dissipation required to maintain a stationary state [@IFThousekeeping]. More recently the relationship between the latter quantities has been explored [@Jarpathintegral; @Esposito07; @Ge09; @Ge10] resulting in a formalism involving a division of the total entropy change into two distinct terms, the adiabatic and non-adiabatic entropy productions [@adiabaticnonadiabatic0; @adiabaticnonadiabatic1; @adiabaticnonadiabatic2], each of which obeys appropriate fluctuation relations and which map onto the house-keeping and excess heats, respectively, of Oono and Paniconi [@oono]. We seek to take such a formalism and generalise its scope by the explicit inclusion of both even (e.g. spatial) and odd (e.g. momentum) variables that transform differently under time reversal. In doing so we define a new quantity which obeys an integral fluctuation theorem for all time.
Specifically, we consider the dynamics of a general set of variables $\textbf{\em x}=(x^1,x^2,\ldots x^n)$ that behave differently under time reversal such that $\boldsymbol{\varepsilon}\textbf{\em x}=(\varepsilon^1x^1,\varepsilon^2x^2,
\ldots \varepsilon^nx^n)$ where $\varepsilon^i=\pm1$ for even and odd variables $x^i$ respectively. Odd variables arise in the discussion of directional quantities and consequently such a consideration is essential when discussing velocities, from the most simple lattice Boltzmann model to considerations of full phase space. The entropy production of a path of duration $\tau$ depends on two probabilities. The first is the path probability, $P^{\rm F}[\vec{\textbf{\em x}}]$, defined as the probability of the forward trajectory, $\vec{\textbf{\em x}}=\textbf{\em x}(t)$ for $0\leq t\leq \tau$, with a distribution of starting configurations, $P^{\rm F}(\textbf{\em x}(0),0)$, that acts as an initial condition for the general master equation (relevant examples arise, for example, in the context of full phase space [@Brenig; @kampenfluct] and in lattice Boltzmann models): $$\frac{\partial P^{\rm F}(\textbf{\em x},t)}{\partial t}=
\sum_{\textbf{\em x}'}T(\textbf{\em x}|\textbf{\em x}',\lambda^{\rm F}(t))
P^{\rm F}(\textbf{\em x}',t)
\label{master}$$ where $T(\textbf{\em x}|\textbf{\em x}',\lambda^{\rm F}(t))$ is a matrix of transition rates between configurations $\textbf{\em x}'$ and $\textbf{\em x}$, defining the normal dynamics, parameterised by the forward protocol $\lambda^{\rm F}$ at time $t$. We use notation $T(\textbf{\em x}|\textbf{\em x})=-\sum_{\textbf{\em x}'\neq
\textbf{\em x}}T(\textbf{\em x}'|\textbf{\em x})$ which describes the mean escape rate. The path probability of some sequence of $N$ transitions to configurations $\textbf{\em x}_i$ from $\textbf{\em x}_{i-1}$ at times $t_i$, such that $t_0=0$ and $t_{N+1}=\tau$, can then be computed as a function of transition rates and exponential waiting times $$\begin{aligned}
P^{\rm F}[\vec{\textbf{\em x}}]
&=P^{\rm F}(\textbf{\em x}_0,0)e^{\int_{t_{0}}^{t_{1}}dt'T(\textbf{\em x}_{0}|
\textbf{\em x}_{0},\lambda^{\rm F}(t'))}\nonumber\\
&\!\!\!\!\!\!\times\prod_{i=1}^{N} T(\textbf{\em x}_{i}|\textbf{\em x}_{i-1},
\lambda^{\rm F}(t_{i}))dt_i
e^{\int_{t_{i}}^{t_{i+1}}dt'T(\textbf{\em x}_{i}|\textbf{\em x}_{i},
\lambda^{\rm F}(t'))}.
\label{pathprob}\end{aligned}$$ We compare this probability to that of another trajectory $\vec{\textbf{\em x}}^*$, protocol $\lambda^*$, initial condition $P^{*}(\textbf{\em x}^*(0),0)$ and chosen dynamics, denoted $P^{*}$, and write $$A[\vec{\textbf{\em x}}]=\ln\left[{P^{\rm F}[\vec{\textbf{\em x}}]}
/{P^{*}[\vec{\textbf{\em x}}^{*}]}\right].
\label{entform}$$ Such a quantity may obey an integral fluctuation theorem (IFT) which may be derived by explicit summation over all possible paths, $\vec{\textbf{\em x}}$, for which $P^{\rm F}[\vec{\textbf{\em x}}]\neq 0$ as follows $$\begin{aligned}
\langle \exp{\left[-A[\vec{\textbf{\em x}}]\right]}\rangle^{\rm F}
&= \sum_{\vec{\textbf{\em x}}} P^{\rm F}[\vec{\textbf{\em x}}]
\exp{\left[-A[\vec{\textbf{\em x}}]\right]}=\sum_{\vec{\textbf{\em x}}}
P^{\rm F}[\vec{\textbf{\em x}}]\frac{P^{*}[\vec{\textbf{\em x}}^{*}]}
{P^{\rm F}[\vec{\textbf{\em x}}]}\nonumber\\
&=\sum_{\vec{\textbf{\em x}}^{*}} P^{*}[\vec{\textbf{\em x}}^{*}]=1.
\label{IFT}\end{aligned}$$ We assume a one to one mapping between $\vec{\textbf{\em x}}$ and $\vec{\textbf{\em x}}^{*}$ (a condition equivalent to a Jacobian of unity in the transformation) so that we may consider the summation over $\vec{\textbf{\em x}}^{*}$ to be equivalent to that over $\vec{\textbf{\em x}}$. We also require that $P^{*}[\vec{\textbf{\em x}}^{*}]= 0$ for all $P^{\rm F}[\vec{\textbf{\em x}}]= 0$ such that the final summation contains all possible paths $\vec{\textbf{\em x}}^{*}$, meaning the required normalisation of $P^{*}[\vec{\textbf{\em x}}^{*}]$ then yields the result of unity. A key result is the implication $\langle A[\vec{\textbf{\em x}}]\rangle^{\rm F}\geq 0$ by Jensen’s inequality.
A common choice for $P^{*}$, and that used to construct the total entropy production, is that of the normal dynamics under the reversed protocol, denoted $P^{*}=P^{\rm R}$. Given the specification of the normal dynamics we point out that all further specifications, including the choice of protocol, can be systematically derived from the appropriate path transformation $\vec{\textbf{\em x}}^{*}$ which we must choose carefully in conjunction with the dynamics so as to obey the above conditions. At this point we must be clear that given a transition $\textbf{\em x}\to \textbf{\em x}'$ under the normal dynamics, the transition $\textbf{\em x}'\to\textbf{\em x}$ is not, in general, possible under those same dynamics. Explicitly, we can construct models such that $T(\textbf{\em x}'|\textbf{\em x})\neq 0$ whilst $T(\textbf{\em x}|\textbf{\em x}')= 0$ (as an intuitive example: Hamiltonian dynamics cannot produce a negative positional step whilst the velocity is positive). The correct path, $\vec{\textbf{\em x}}^{*}$, to consider is the time reversed trajectory proper which includes a reversal of sign for all odd variables. This is the choice $\textbf{\em x}^*(t)=\textbf{\em x}^{\dagger}(t)=\boldsymbol{\varepsilon}
\textbf{\em x}(\tau\!-\!t)$ and it satisfies the condition $P^{*}[\vec{\textbf{\em x}}^{*}]=
P^{\rm R}[\vec{\textbf{\em x}}^{\dagger}]= 0$ for all $P^{\rm F}[\vec{\textbf{\em x}}]= 0$ required for an IFT. The reversed protocol $\lambda^{*}=\lambda^{\rm R}$ may be similarly obtained from the forward protocol, which may be treated as an even dynamical variable, meaning it transforms to yield $\lambda^{*}(t)=\varepsilon\lambda^{\rm F}(\tau\!-\!t)
=\lambda^{\rm F}(\tau\!-\!t)=\lambda^{\rm R}(t)$. And finally we require the choice of initial condition for the reverse path. This may be informed physically: we seek to characterise the irreversibility of the forward path and so initiate the reverse behaviour by time reversing the coordinates, $\textbf{\em x}(\tau)$, and distribution, $P^{\rm F}(\textbf{\em x}
(\tau),\tau)$, at the end of the forward process and evolve forward in time from there. The distribution can also be found by applying the transformation rules used to obtain the trajectory $\vec{\textbf{\em x}}^{\dagger}$ from $\vec{\textbf{\em x}}$ such that $P^{*}(\textbf{\em x}^{*}(0),0)=P^{R}(\textbf{\em x}^{\dagger}(0),0)
=\boldsymbol{\hat{\varepsilon}}P^{\rm F}(\boldsymbol{\varepsilon}
\textbf{\em x}(\tau),\tau)=P^{\rm F}(\boldsymbol{\varepsilon}
\boldsymbol{\varepsilon}
\textbf{\em x}(\tau),\tau)=P^{\rm F}(\textbf{\em x}(\tau),\tau)$ where $\boldsymbol{\hat{\varepsilon}}$ denotes the time reversal operation on the distribution. In this instance the path probability is therefore $$\begin{aligned}
P^{\rm R}[\vec{\textbf{\em x}}^{\dagger}]
&=P^{\rm R}(\textbf{\em x}^{\dagger}_0,0)e^{\int_{t_{0}}^{t_{1}}dt'
T(\textbf{\em x}^{\dagger}_{0}|\textbf{\em x}^{\dagger}_{0},
\lambda^{\rm R}(t'))}\nonumber\\
&\!\!\!\!\!\!\!\!\!\!\!\!\times\prod_{i=1}^{N} T(\textbf{\em x}^{\dagger}_{i}|
\textbf{\em x}^{\dagger}_{i-1},\lambda^{\rm R}(t_{i}))dt_i
e^{\int_{t_{i}}^{t_{i+1}}dt'
T(\textbf{\em x}^{\dagger}_{i}|\textbf{\em x}^{\dagger}_{i},
\lambda^{\rm R}(t'))}.\end{aligned}$$ We have $\textbf{\em x}_i^{\dagger}=\boldsymbol{\varepsilon}\textbf{\em x}_{N-i}$ so we may rearrange to give $$\begin{aligned}
&P^{\rm R}[\vec{\textbf{\em x}}^{\dagger}]=P^{\rm F}(\textbf{\em x}_N,\tau)
e^{\int_{t_{N}}^{t_{N+1}}dt'T(\boldsymbol{\varepsilon}\textbf{\em x}_{0}|
\boldsymbol{\varepsilon}\textbf{\em x}_{0},\lambda^{\rm R}(t'))}\\
&\!\!\!\!\!\!\!\!\!\times\prod_{i=1}^{N} e^{\int_{t_{N\!-\!i}}^{t_{N\!-\!i\!+\!1}}dt'
T(\boldsymbol{\varepsilon}
\textbf{\em x}_{i}|\boldsymbol{\varepsilon}
\textbf{\em x}_{i},\lambda^{\rm R}(t'))}
T(\boldsymbol{\varepsilon}\textbf{\em x}_{i\!-\!1}|
\boldsymbol{\varepsilon}\textbf{\em x}_{i},
\lambda^{\rm R}(t_{N\!-\!i\!+\!1}))dt_i\nonumber.\end{aligned}$$ We then perform a change of variable $t'\to \tau-t'$ and use $\lambda^{\rm R}(t_i)=\lambda^{\rm F}(t_{N-i+1})$ such that $$\begin{aligned}
&P^{\rm R}[\vec{\textbf{\em x}}^{\dagger}]=P^{\rm F}(\textbf{\em x}_N,\tau)
e^{-\int_{t_{1}}^{t_{0}}dt'T(\boldsymbol{\varepsilon}\textbf{\em x}_{0}|
\boldsymbol{\varepsilon}\textbf{\em x}_{0},\lambda^{\rm F}(t'))}\\
&\times\prod_{i=1}^{N} e^{-\int_{t_{i+1}}^{t_{i}}dt'T(\boldsymbol{\varepsilon}
\textbf{\em x}_{i}|\boldsymbol{\varepsilon}\textbf{\em x}_{i},
\lambda^{\rm F}(t'))}
T(\boldsymbol{\varepsilon}\textbf{\em x}_{i-1}|
\boldsymbol{\varepsilon}\textbf{\em x}_{i},
\lambda^{\rm F}(t_{i}))dt_i\nonumber.\end{aligned}$$ A comparison of $P^{\rm F}[\textbf{\em x}]$ and $P^{\rm R}[\textbf{\em x}^{\dagger}]$ characterises the irreversibility of the forward path and defines the total entropy production (using units $k_B=1$) $$\begin{aligned}
\Delta S_{\rm tot}&=\ln{{P^{\rm F}[\vec{\textbf{\em x}}]}}-\ln{P^{\rm R}[
\vec{\textbf{\em x}}^{\dagger}]}\nonumber\\
&=\ln{\!\frac{P^{\rm F}(\textbf{\em x}_0,0)}
{P^{\rm F}(\textbf{\em x}_N,\tau)}}+
\sum_{i=0}^{N}\ln{\frac{e^{\int_{t_{i}}^{t_{i+1}}dt'T(\textbf{\em x}_{i}|
\textbf{\em x}_{i},
\lambda^{\rm F}(t'))}}{e^{\int_{t_{i}}^{t_{i+1}}dt'T(\boldsymbol{\varepsilon}
\textbf{\em x}_{i}|\boldsymbol{\varepsilon}\textbf{\em x}_{i},
\lambda^{\rm F}(t'))}}}\nonumber\\
&+\sum_{i=1}^{N}\ln{\!\frac{T(\textbf{\em x}_{i}|\textbf{\em x}_{i-1},
\lambda^{\rm F}(t_i))}{T(\boldsymbol{\varepsilon}\textbf{\em x}_{i-1}|
\boldsymbol{\varepsilon}\textbf{\em x}_{i},\lambda^{\rm F}(t_i))}}
\label{Stot}\end{aligned}$$ which by its definition and Eq. (\[IFT\]) obeys [@seifertoriginal] $$\langle \exp{[-\Delta S_{\rm tot}]}\rangle^{\rm F}=1.$$ We find that this form of $\Delta S_{\rm tot}$ is more complicated than previous descriptions [@Harris07; @adiabaticnonadiabatic0] unless $\boldsymbol{\varepsilon}\textbf{\em x}=\textbf{\em x}$. Note that if detailed balance holds, such that $P^{\rm eq}(\textbf{\em x})
T(\textbf{\em x}'|\textbf{\em x})=P^{\rm eq}(\boldsymbol{\varepsilon}
\textbf{\em x}')T(\boldsymbol{\varepsilon}\textbf{\em x}|\boldsymbol{
\varepsilon}\textbf{\em x}')$, we expect $P^{\rm eq}$, the equilibrium state for a given $\lambda^{\rm F}(t)$, to satisfy $P^{\rm eq}(
\textbf{\em x})=P^{\rm eq}(\boldsymbol{\varepsilon}\textbf{\em x})$ due to time-reversal invariance, along with $T(\textbf{\em x}|\textbf{\em x})
=T(\boldsymbol{\varepsilon}\textbf{\em x}|\boldsymbol{\varepsilon}
\textbf{\em x})$. For a system in equilibrium, we therefore conclude that $\Delta S_{\rm tot}=0$ for all paths.
Next we consider alternative specifications of $P^{*}$. We consider the adjoint dynamics which lead to the same stationary state, $P^{\rm st}
(\textbf{\em x},\lambda^{\rm F}(t))$, as the normal dynamics, but generate flux of the opposite sign in that stationary state. It can be shown [@adiabaticnonadiabatic0; @Harris07; @Jarpathintegral] that this requires an adjoint transition rate matrix $T^{\rm ad}$ described by $$T^{\rm ad}(\textbf{\em x}|\textbf{\em x}',\lambda^{\rm F}(t))
=T(\textbf{\em x}'|\textbf{\em x},
\lambda^{\rm F}(t))\frac{P^{\rm st}(\textbf{\em x},\lambda^{\rm F}(t))}{
P^{\rm st}(\textbf{\em x}',\lambda^{\rm F}(t))}.
\label{adj1}$$ However, in the same way that the normal dynamics may not, in general, permit transitions $\textbf{\em x}'\to\textbf{\em x}$ or $\boldsymbol{\varepsilon}\textbf{\em x}\to
\boldsymbol{\varepsilon}\textbf{\em x}'$, similarly the adjoint dynamics may not, in general, permit transitions $\textbf{\em x}\to\textbf{\em x}'$ or $\boldsymbol{\varepsilon}\textbf{\em x}'\to
\boldsymbol{\varepsilon}\textbf{\em x}$. Thus we must consider the representation of the adjoint dynamics as either Eq. (\[adj1\]) or $$\!\!\!\!\!T^{\rm ad}(\boldsymbol{\varepsilon}\textbf{\em x}'|\boldsymbol{\varepsilon}
\textbf{\em x},\lambda^{\rm F}(t))=T(\boldsymbol{\varepsilon}\textbf{\em x}|
\boldsymbol{\varepsilon}\textbf{\em x}',\lambda^{\rm F}(t))\!\frac{P^{\rm st}(
\boldsymbol{\varepsilon}\textbf{\em x}',\lambda^{\rm F}(t))}{P^{\rm st}(
\boldsymbol{\varepsilon}\textbf{\em x},\lambda^{\rm F}(t))}
\label{adj2}$$ depending on the specific transition being considered. Explicitly, when choosing $P^{*}[\vec{\textbf{\em x}}^*]$, we should not consider $P^{\rm ad}[\vec{\textbf{\em x}}]$ or $P^{\rm ad}[\vec{\textbf{\em x}}^{\dagger}]$ since these might violate the required condition $P^{*}[\vec{\textbf{\em x}}^{*}]= 0$ for all $P^{\rm F}[\vec{\textbf{\em x}}]= 0$, required for an IFT.
Under the adjoint dynamics, however, an appropriate transformation of $\vec{\textbf{\em x}}$ is $\textbf{\em x}^{*}(t)=
\textbf{\em x}^{\rm R}(t)=\textbf{\em x}(\tau\!-\!t)$. Applying the transformation rules used to obtain $\vec{\textbf{\em x}}^{\rm R}$ yields the reverse protocol as before $\lambda^{*}(t)=\lambda^{\rm F}(\tau\!-\!t)=\lambda^{\rm R}(t)$ and the initial distribution $P^{*}(\textbf{\em x}^{*}(0),0)=
P^{\rm ad,R}(\textbf{\em x}^{\rm R}(0),0)=
P^{\rm F}(\textbf{\em x}(\tau),\tau)$. The path probability is then $$\begin{aligned}
&P^{\rm ad,R}[\vec{\textbf{\em x}}^{\rm R}]
=P^{\rm ad,R}(\textbf{\em x}^{\rm R}_0,0)
e^{\int_{t_{0}}^{t_{1}}dt'T^{\rm ad}(\textbf{\em x}^{\rm R}_{0}|
\textbf{\em x}^{\rm R}_{0},\lambda^{\rm R}(t'))}\nonumber\\
&\!\!\!\times\prod_{i=1}^{N} T^{\rm ad}(\textbf{\em x}^{\rm R}_{i}|
\textbf{\em x}^{\rm R}_{i-1},
\lambda^{\rm R}(t_{i}))dt_ie^{\int_{t_{i}}^{t_{i+1}}dt'T^{\rm ad}(
\textbf{\em x}^{\rm R}_{i}|
\textbf{\em x}^{\rm R}_{i},\lambda^{\rm R}(t'))}\nonumber\\
&=P^{\rm F}(\textbf{\em x}_N,\tau)e^{-\int_{t_{1}}^{t_{0}}dt'
T^{\rm ad}(\textbf{\em x}_{0}|
\textbf{\em x}_{0},\lambda^{\rm F}(t'))}\\
&\!\!\!\times\prod_{i=1}^{N} e^{-\int_{t_{i+1}}^{t_{i}}dt'T^{\rm ad}(
\textbf{\em x}_{i}|\textbf{\em x}_{i},\lambda^{\rm F}(t'))}
T^{\rm ad}(\textbf{\em x}_{i-1}|\textbf{\em x}_{i},
\lambda^{\rm F}(t_{i}))dt_i\nonumber.\end{aligned}$$ We then construct a quantity of the form given in Eq. (\[entform\]), utilise Eq. (\[adj1\]) and the property $T^{\rm ad}(\textbf{\em x}|\textbf{\em x})=
T(\textbf{\em x}|\textbf{\em x})$, valid by means of balance, to obtain $$\begin{aligned}
\Delta S_{\rm 1}&=\ln{{P^{\rm F}[\vec{\textbf{\em x}}]}}-\ln{P^{\rm ad,R}[
\vec{\textbf{\em x}}^{\rm R}]}\nonumber\\
%&=\ln{\!\frac{P^{\rm F}(\textbf{\em x}_0,0)}
%{P^{\rm F}(\textbf{\em x}_N,\tau)}}+
%\sum_{i=1}^{N}\ln{\!\frac{T(\textbf{\em x}_{i}|\textbf{\em x}_{i-1},
%\lambda^{\rm F}(t_i))}{T^{\rm ad}(\textbf{\em x}_{i-1}|\textbf{\em x}_{i},
%\lambda^{\rm F}(t_i))}}\nonumber\\
&=\ln{\!\frac{P^{\rm F}(\textbf{\em x}_0,0)}
{P^{\rm F}(\textbf{\em x}_N,\tau)}}
+\sum_{i=1}^{N}\ln{\!\frac{P^{\rm st}(\textbf{\em x}_{i},
\lambda^{\rm F}(t_i))}
{P^{\rm st}(\textbf{\em x}_{i-1},\lambda^{\rm F}(t_i))}}\end{aligned}$$ which through its definition and Eq. (\[IFT\]) obeys $$\langle \exp{[-\Delta S_{1}]}\rangle^{\rm F}=1$$ which exists in the literature as the Hatano-Sasa relation [@hatanosasa] or IFT for the non-adiabatic entropy production [@adiabaticnonadiabatic0; @adiabaticnonadiabatic1; @adiabaticnonadiabatic2]. Let us now consider, once again under the adjoint dynamics, the path transformation choice $\textbf{\em x}^{*}(t)=\textbf{\em x}^{\rm T}(t)
=\boldsymbol{\varepsilon}\textbf{\em x}(t)$. Applying the transformation rules we obtain the protocol $\lambda^{*}(t)=
\varepsilon\lambda^{\rm F}(t)=\lambda^{\rm F}(t)$ and initial distribution $P^{*}(\textbf{\em x}^{*}(0),0)=P^{\rm ad,F}(\textbf{\em x}^{\rm T}(0),0)=
\boldsymbol{\hat{\varepsilon}}
P^{\rm F}(\boldsymbol{\varepsilon}\textbf{\em x}(0),0)
=P^{\rm F}(\textbf{\em x}(0),0)$. The path probability for this case is therefore $$\begin{aligned}
&P^{\rm ad,F}[\vec{\textbf{\em x}}^{\rm T}]=
P^{\rm ad,F}(\textbf{\em x}^{\rm T}_0,0)
e^{\int_{t_{0}}^{t_{1}}dt'
T^{\rm ad}(\textbf{\em x}^{\rm T}_{0}|\textbf{\em x}^{\rm T}_{0},
\lambda^{\rm F}(t'))}\nonumber\\
&\quad\times\prod_{i=1}^{N}
T^{\rm ad}(\textbf{\em x}^{\rm T}_{i}|\textbf{\em x}^{\rm T}_{i-1},
\lambda^{\rm F}(t_{i}))dt_ie^{\int_{t_{i}}^{t_{i+1}}dt'
T^{\rm ad}(\textbf{\em x}^{\rm T}_{i}|
\textbf{\em x}^{\rm T}_{i},\lambda^{\rm F}(t'))}\nonumber\\
&=P^{\rm F}(\textbf{\em x}_0,0)e^{\int_{t_{0}}^{t_{1}}dt'T^{\rm ad}(
\boldsymbol{\varepsilon}\textbf{\em x}_{0}|
\boldsymbol{\varepsilon}\textbf{\em x}_{0},
\lambda^{\rm F}(t'))}\\
&\quad\times\prod_{i=1}^{N}
T^{\rm ad}(\boldsymbol{\varepsilon}\textbf{\em x}_{i}|
\boldsymbol{\varepsilon}\textbf{\em x}_{i-1},\lambda^{\rm F}(t_{i}))dt_i
e^{\int_{t_{i}}^{t_{i+1}}dt'
T^{\rm ad}(\boldsymbol{\varepsilon}\textbf{\em x}_{i}|
\boldsymbol{\varepsilon}\textbf{\em x}_{i},\lambda^{\rm F}(t'))}\nonumber.\end{aligned}$$ By Eq. (\[entform\]) this then allows us to define $$\begin{aligned}
&\Delta S_{\rm 2}=\ln{{P^{\rm F}[\vec{\textbf{\em x}}]}}-\ln{P^{\rm ad,F}
[\vec{\textbf{\em x}}^{\rm T}]}\nonumber\\
&=\sum_{i=0}^{N}\ln{\frac{e^{\int_{t_{i}}^{t_{i+1}}dt'T(\textbf{\em x}_{i}|
\textbf{\em x}_{i},\lambda^{\rm F}(t'))}}{e^{\int_{t_{i}}^{t_{i+1}}dt'
T(\boldsymbol{\varepsilon}\textbf{\em x}_{i}|
\boldsymbol{\varepsilon}\textbf{\em x}_{i},
\lambda^{\rm F}(t'))}}}\nonumber\\
&+\sum_{i=1}^{N}\ln{\!\frac{P^{\rm st}(\boldsymbol{\varepsilon}
\textbf{\em x}_{i-1},
\lambda^{\rm F}(t_i))}{P^{\rm st}(\boldsymbol{\varepsilon}\textbf{\em x}_i,
\lambda^{\rm F}(t_i))}\frac{T(\textbf{\em x}_{i}|
\textbf{\em x}_{i-1},\lambda^{\rm F}(t_i))}{
T(\boldsymbol{\varepsilon}\textbf{\em x}_{i-1}|
\boldsymbol{\varepsilon}\textbf{\em x}_{i},
\lambda^{\rm F}(t_i))}}
\label{S2}\end{aligned}$$ which similarly must obey $$\langle \exp{[-\Delta S_2]}\rangle^{\rm F}=1.
\label{S2IFT}$$ Unlike $\Delta S_1$, the quantity $\Delta S_2$ is new in the literature. We must immediately recognise that $\Delta S_{\rm tot}\neq \Delta S_1+\Delta S_2$ differing by a quantity $$\Delta S_3=\sum_{i=1}^{N}\ln{\!{\frac{P^{\rm st}(\textbf{\em x}_{i-1},
\lambda^{\rm F}(t_i))P^{\rm st}(\boldsymbol{\varepsilon}\textbf{\em x}_{i},
\lambda^{\rm F}(t_i))}{P^{\rm st}(\textbf{\em x}_i,\lambda^{\rm F}(t_i))
P^{\rm st}(\boldsymbol{\varepsilon}\textbf{\em x}_{i-1},
\lambda^{\rm F}(t_i))}}}
\label{S3}$$ such that $\Delta S_{\rm tot}=\Delta S_1+\Delta S_2+\Delta S_3$. If $\boldsymbol{\varepsilon}\textbf{\em x}=\textbf{\em x}$ then $\Delta S_3=0$ and $\Delta S_2$ reduces to the adiabatic entropy production appearing in [@adiabaticnonadiabatic0; @adiabaticnonadiabatic1; @adiabaticnonadiabatic2]. More importantly we must recognise that $\Delta S_{\rm tot}\!-\!\Delta S_1\!=
\!\Delta S_2\!+\!\Delta S_3=\ln{P^{\rm ad,R}[\vec{\textbf{\em x}}^{\rm R}]}-
\ln{P^{\rm R}[\vec{\textbf{\em x}}^{\rm \dagger}]}$ or $\Delta S_{\rm tot}\!-
\!\Delta S_2\!=\!\Delta S_1\!+\!\Delta S_3=
\ln{P^{\rm ad,F}[\vec{\textbf{\em x}}^{\rm T}]}-
\ln{P^{\rm R}[\vec{\textbf{\em x}}^{\rm \dagger}]}$ cannot be written in the form required for Eq. (\[IFT\]) and so do not obey an IFT and do not necessarily have any bounds on the sign of their mean. We proceed by following the formalism of Seifert [@seifertoriginal; @seifertprinciples] and write $$\Delta S_{\rm tot}=\ln{\frac{P^{\rm F}(\textbf{\em x}(0),0)}
{P^{\rm F}(\textbf{\em x}(\tau),\tau)}}+\frac{\Delta Q}{T_{\rm env}}=
\Delta S_{\rm sys}+\frac{\Delta Q}{T_{\rm env}},$$ where $T_{\rm env}$ is the temperature of the environment, and that of Oono and Paniconi, such that total heat transfer to the environment, $\Delta Q$, is the sum of the excess heat and house-keeping heat $\Delta Q=
\Delta Q_{\rm ex}+\Delta Q_{\rm hk}$ [@oono]. The house-keeping heat is associated with the entropy production in stationary states and arises from a non-equilibrium constraint that breaks detailed balance. The sum $\Delta S_2
+\Delta S_3$ is manifestly the entropy production in the stationary state and since we are considering Markov systems, both $\Delta S_2$ and $\Delta S_3$ are only non-zero when detailed balance is broken. Hence it is sensible to associate $\Delta S_2 +\Delta S_3$ with the house-keeping heat such that $$\Delta Q_{\rm hk}=(\Delta S_2\!+\!\Delta S_3)T_{\rm env}.$$ $\Delta S_1$ is zero for all trajectories in the stationary state consolidating the definition of the excess heat as the heat transfer associated with an entropy flow that exactly cancels the change in system entropy in the stationary state such that $$\Delta Q_{\rm ex}=(\Delta S_1\!-\!\Delta S_{\rm sys})T_{\rm env}.$$ However, the prevailing definition of the house-keeping heat does not make clear its properties when the system is not in a stationary state. A reported formalism suggests that it is associated with the adiabatic entropy production which serves as a general measure of the breakage of detailed balance [@adiabaticnonadiabatic0; @adiabaticnonadiabatic1; @adiabaticnonadiabatic2]. When considering cases where $\boldsymbol{\varepsilon}\textbf{\em x}=\textbf{\em x}$, this is a consistent approach and the mean house-keeping heat obeys strict positivity requirements suggesting the entropy additively increases due to non-equilibrium constraints and a lack of detailed balance on top of that arising from relaxation. However, with the inclusion of odd variables this simple picture no longer holds, with an ambiguity illustrated by the fact that any of $\Delta S_2$, $\Delta S_3$ or $\Delta S_2\!+\!\Delta S_3$ could be argued to be a measure of the departure from detailed balance. In the light of Eq. (\[S2IFT\]) we propose that it is sensible to divide the house-keeping heat into two quantities which map onto $\Delta S_2$ and $\Delta S_3$. It is important to observe that, on average, the rate of change of $\Delta S_3$ vanishes in the stationary state by means of balance: the path integral over an increment in $\Delta S_3$ explicitly vanishes. Consequently we define the ‘transient house-keeping heat’ and the ‘generalised house-keeping heat’ $$\Delta Q_{\rm hk,T}=\Delta S_3T_{\rm env}\quad
\quad\Delta Q_{\rm hk,G}=\Delta S_2T_{\rm env}$$ such that $\Delta Q_{\rm hk}=\Delta Q_{\rm hk,T}\!+\!\Delta Q_{\rm hk,G}$. Since $\langle d\Delta S_3/d\tau\rangle ^{\rm F,st}\!=\!0$, the generalised house-keeping heat, when averaged, has the mean properties previously attributed to the house-keeping heat: it describes the heat flow required to maintain a non-equilibrium stationary state and is rigorously non-negative. Our central result therefore is $$\langle \exp{[-\Delta Q_{\rm hk,G}/T_{\rm env}]}\rangle^{\rm F}=1$$ so $\langle\Delta Q_{\rm hk,G}\rangle^{\rm F}\geq 0$ for all times, protocols and initial conditions. As a corollary we also state that in general $$\langle \exp{[-\Delta Q_{\rm hk}/T_{\rm env}]}\rangle^{\rm F}\neq 1
\label{noIFT}$$ providing no bounds on $\langle\Delta Q_{\rm hk}\rangle^{\rm F}$ except in the stationary state when $\Delta S_1=0$ and $\Delta Q_{\rm hk}/T_{\rm env}=\Delta S_{\rm tot}$ or generally when $P^{\rm st}(\boldsymbol{\varepsilon}\textbf{\em x},
\lambda^{\rm F}(t))=P^{\rm st}(\textbf{\em x},\lambda^{\rm F}(t))$. As such the view that the mean rate of entropy production is the sum of two specific non-negative contributions as in [@adiabaticnonadiabatic0; @adiabaticnonadiabatic1; @adiabaticnonadiabatic2], is incomplete. The contribution associated with a non-equilibrium constraint requires further unravelling, particularly when out of stationarity.
![\[lattice\]Allowed moves between positions $X_i$ and $\pm$ velocity states are shown by arrows, with associated rates $T$. Periodic boundaries allow jumps from $X_L+$ to $X_1+$ and $X_{1}-$ to $X_{L}-$. A given path contributes to the transient and generalised house-keeping heats, $T_{\rm env}\Delta S_3$ and $T_{\rm env}
\Delta S_2$, respectively, due to transitions between, and residence times $\Delta t$ at, each phase space point, as indicated. These correspond to individual terms in the summations in Eqs. (\[S2\]) and (\[S3\]). ](lattice){width="\columnwidth"}
To explore the nature of the house-keeping heat we consider its behaviour in the approach to the stationary state of a simple model of particle dynamics on a ring. The phase space consists of $L$ identical spatial positions $X_{1},X_{2}\ldots X_{L}$ and two velocities labelled $+$ and $-$ as shown in Fig. \[lattice\] with the time reversal properties $\boldsymbol{\varepsilon}X_i\pm=X_i\mp$ necessitated by the one-way nature of many of the transitions. The stationary state probabilities that arise from these dynamics are $P^{\rm st}(X_i+)=A/(L(A+B))$ and $P^{\rm st}(X_i-)=B/(L(A+B))$. Any difference between the velocity reversal rates $A$ and $B$ gives rise to a non-equilibrium stationary state by providing a stationary particle current, which for $A>B$ runs from left to right. Such dynamics amount to a very simple lattice Boltzmann model. Contributions $\Delta S_2$ and $\Delta S_3$ associated with particle behaviour consisting of instantaneous transitions and waiting periods are indicated. We consider particle behaviour over a small time interval $dt$, and compute the mean entropy production rates to leading order in $dt$. Examining the path probability in Eq. (\[pathprob\]) we need only consider $N=0$ or $N=1$ transitions. Identifying leading order terms in the products of $P$, $T$, exponentiated waiting times and $\Delta S_3$ that make up the average of the form given in Eq. (\[IFT\]) yields $$\frac{d\langle\Delta S_3\rangle^{\rm F}}{dt}=\sum_{i=1}^{L}
2P(X_i+)B\ln{\frac{A}{B}}+2P(X_i-)A\ln{\frac{B}{A}}.
\label{S3dt}$$ For non-stationary $P$ its sign is unbounded: for example if all the probability were uniformly distributed initially amongst the $+$ velocity states it would equal $2B\ln(A/B)$, whilst if it were distributed over the $-$ states it would be $-2A\ln(A/B)$ instead. Such non-zero contributions to $\Delta S_3$ require an asymmetric stationary state in odd variables which thus explains their absence when the stationary velocity distribution is assumed to be symmetric, such as in overdamped Langevin descriptions (see [@IFThousekeeping] and examples in [@adiabaticnonadiabatic2]). However, in the stationary state with $P=P^{\rm st}$, $d\langle\Delta
S_3\rangle^{\rm F}/dt$ is demonstrably equal to zero as claimed. By similar means $$\begin{aligned}
\frac{d\langle\Delta S_2\rangle^{\rm F}}{dt}=\sum_{i=1}^{L}&P(X_i+)
\left[A-B-B\ln{\frac{A}{B}}\right]\nonumber\\
&+P(X_i-)\left[B-A-A\ln{\frac{B}{A}}\right]
\label{S2dt}\end{aligned}$$ which is positive for all positive $A$ and $B$ and reduces to $d\langle\Delta S_2\rangle^{\rm F,st}/dt=(A-B)^2/(A+B)$ in the stationary state. We note that the sum of Eqs. (\[S3dt\]) and (\[S2dt\]) has no bound on its sign and relates to the inequality in Eq. (\[noIFT\]). Further, $d\langle\exp{[-\Delta S_{2}]}
\rangle^{\rm F}/dt=0$ and $\langle \exp{[-\Delta S_2(t=0)]}
\rangle^{\rm F}=1$ which explicitly demonstrates the expected IFT for any normalised $P(X_i\pm)$. Finally, we note that for $A=B$, all contributions vanish in detail as this corresponds to equilibrium where there is no entropy production.
We have extended the formalism found in [@hatanosasa; @IFThousekeeping; @adiabaticnonadiabatic0; @adiabaticnonadiabatic1; @adiabaticnonadiabatic2] and split the total entropy production into two rigorously positive contributions and a third contribution which has no bounds on its sign. We have argued that this final quantity is, in the mean, a transient contribution to the house-keeping heat and it is the mean generalised house-keeping heat that is rigorously positive for all times. It is not straightforward to consolidate this with the two causes of time reversal asymmetry namely relaxation to the stationary state and imposed non-equilibrium constraints: $\Delta S_3$ exists only in the presence the latter, but is, in the mean, its own measure of relaxation to the stationary state. It could be argued that the non-adiabatic entropy production and Hatano-Sasa relation do not fully capture the entropy production due to transitions between stationary states, but associating $\Delta S_3$ with one or other form of entropy production is not entirely satisfactory as it occurs when the line between them is blurred. Nevertheless, either interpretation elucidates a new layer of complexity in the theory of entropy production in stochastic systems. Further exploration in the context of continuous stochastic processes is to be reported elsewhere [@SpinneyFord]. The authors acknowledge financial support from EPSRC. \[1\]\[1\][\#1]{}
[26]{}ifxundefined \[1\][ ifx[\#1]{} ]{}ifnum \[1\][ \#1firstoftwo secondoftwo ]{}ifx \[1\][ \#1firstoftwo secondoftwo ]{}““\#1””@noop \[0\][secondoftwo]{}sanitize@url \[0\][‘\
12‘\$12 ‘&12‘\#12‘12‘\_12‘%12]{}@startlink\[1\]@endlink\[0\]@bib@innerbibempty @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [ ()]{} [****, ()](\doibase 10.1103/PhysRevE.76.031132) [****, ()](\doibase 10.1103/PhysRevE.80.021137) [****, ()](\doibase 10.1103/PhysRevE.81.051133) @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [ ()]{} @noop [****, ()]{} @noop [ ]{}
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'In this paper, we continue our study of quasihyperbolic metric in Banach spaces. The main results of the paper present a criterion for smoothness of geodesics of quasihyperbolic type metrics in Banach spaces, under a Dini type condition on the weight function, which improves an earlier result of the two first authors. We also answer to a question posed by the two first authors in an earlier paper with R. Klén, and present results related to the question on smoothness of quasihyperbolic balls.'
address:
- 'Department of Mathematics and Systems Analysis, Aalto University, P.O. Box 11100, FI-00076 Aalto, Finland'
- 'University of Eastern Finland, Department of Physics and Mathematics, Box 111, FI-80101 Joensuu, Finland'
- 'School of Science, Zhejiang Sci-Tech University, 310018 Hangzhou, China'
author:
- 'Antti Rasila, Jarno Talponen, Xiaohui Zhang'
title: Observations on quasihyperbolic geometry modeled on Banach spaces
---
Introduction
============
The quasihyperbolic metric in the $n$-dimensional Euclidean space ${\mathbb{R}}^n$ is a natural generalization of the hyperbolic metric, which was first introduced by F.W. Gehring and his students Palka [@GehringPalka76] and Osgood [@GehringOsgood79] in 1970’s. The significance of this metric arises from its several useful properties. In particular, quasihyperbolic metric is generally well-behaved in quasiconformal mappings and related classes of transformations, and it naturally arises in the generalization of the Schwarz-Pick lemma for quasiconformal mappings in ${\mathbb{R}}^n$. In addition, bounds for this metric can be obtained by using the [*distance ratio metric*]{}, a quantity that has a simple and natural definition and is easy to compute.
Notably, unlike the conformal modulus and related approaches relying on $n$-dimensional volume integration, the quasihyperbolic metric can be defined in a wide range of metric spaces, which includes infinite dimensional Banach spaces. This observation has led to the concept of [*(dimension) free quasiconformality*]{}, which was developed by Väisälä (see [@Vai99]). The definition is based on the quasiconformal Schwarz-Pick lemma, where the Grötzsch modulus function is replaced with an arbitrary strictly increasing function of the non-negative real numbers onto itself. This definition coincides with other definitions of quasiconfromality in ${\mathbb{R}}^n$. It has led to study of this class of mappings in Banach spaces, but it is also workable in a more general metric settings (see [@hrwz]). Besides its role in the theory of quasiconformal mappings, the quasihyperbolic metric is also related to domain classification problems, which have independent interest and applications related to certain function spaces and partial differential equations.
Because of these important applications, there has been significant interest in the quasihyperbolic metric itself, and related concepts such as quasihyperbolic geodesics and balls. However, analytic properties of such objects are usually not easy to see directly from the definition of the quasihyperbolic distance. In this context, it is generally assumed that the quasihyperbolic metric would exhibit behavior similar to the hyperbolic metric, at least in a sufficiently small scale. For example, the quasihyperbolic metric is conformal in the sense that small quasihyperbolic balls are geometrically close to the balls of the norm metric.
The first two authors have investigated these questions in a series of articles [@KRT2; @RasilaTalponen12; @RT], the latests of which is joint work with R. Klén. The purpose of this paper is to further refine and improve this line of research by presenting a more precise condition for smoothness of geodesics of quasihyperbolic type weighted metrics, giving additional results on tangential properties of quasihyperbolic balls and, finally, presenting some open questions related to the connection of the quasihyperbolic metric with the geometry of the underlying Banach space.
Preliminaries
=============
In this section, we recall certain basic definitions required to formulate our main results. We refer to the monographs in the references for suitable background information.
Quasihyperbolic metric
----------------------
Let ${\mathrm{X}}$ be a Banach space with ${\mathrm{dim}}{\mathrm{X}}\ge 2$, and suppose that $\Omega \subsetneq {\mathrm{X}}$ is a domain. For $x\in \Omega$, we denote by $d(x)$ distance $d(x,\partial \Omega)$. Then the *quasihyperbolic length* of a rectifiable arc $\gamma$ in $\Omega$ is defined by $$\ell_k(\gamma):=\int_{I} \frac{\|d \gamma\|}{d(\gamma(t))}.$$ The *quasihyperbolic distance* of two points $x,y\in \Omega$ is the number $$k(x,y):=k_\Omega(x,y):=\inf_\gamma \ell_k(\gamma)$$ where the infimum is taken over all rectifiable arcs $\gamma$ joining the points $x,y\in\Omega$.
Next we recall the following definitions that are central in the geometry of Banach spaces.
Uniform convexity
-----------------
The *modulus of convexity* $\delta_{{\mathrm{X}}}(\epsilon),\ 0<\epsilon \leq 2,$ is defined by $$\delta_{{\mathrm{X}}}(\epsilon):=\inf\{1-\|x+y\|/2:\ x,y\in {\mathrm{X}},\ \|x\|=\|y\|=1,\ \|x-y\|=\epsilon\},$$ A Banach space ${\mathrm{X}}$ is called *uniformly convex* if $\delta_{{\mathrm{X}}}(\epsilon)>0$ for all $\epsilon>0$, and *uniformly smooth* if $$\lim_{\tau\to 0^{+}}\frac{\rho_{{\mathrm{X}}}(\tau)}{\tau}=0.$$ Furthermore, we call a a space ${\mathrm{X}}$ uniformly convex of power type $p\in [2,\infty)$ if $\delta_{{\mathrm{X}}}(\epsilon)\geq K\epsilon^{p}$, for some $K>0$. Note that the modulus $\delta_X$ measures the convexity of the unit ball. A set $C$ is [*strictly convex*]{} if it is convex and $d(sx+(1-s)y,\partial C)>0$ for all $x,y\in \partial C,\ x\neq y,$ and $0<s<1$.
Uniform smoothness
------------------
The *modulus of smoothness* $\rho_{{\mathrm{X}}}(\tau),\ \tau>0$ is defined by $$\rho_{{\mathrm{X}}}(\tau):=\sup\{(\|x+y\|+\|x-y\|)/2 -1 :\ x,y\in {\mathrm{X}},\ \|x\|=1,\ \|y\|=\tau\}.$$ Again, we call a a space ${\mathrm{X}}$ [*uniformly smooth*]{} of power type $p\in [1,2]$ if $\rho_{{\mathrm{X}}}(\tau)\leq K\tau^p$, for some $K>0$.
LUR spaces
----------
Recall that ${\mathrm{X}}$ is [*locally uniformly rotund*]{} (LUR) if for all $x,x_n \in {\mathbf{S}}_{\mathrm{X}}$, $n\in{\mathbb{N}}$, with $\lim_{n\to\infty} \|\frac{x+x_n }{2}\| = 1$ it follows that $$\lim_{n\to\infty} \| x_n - x\|=0.$$ Note that a uniformly convex space is LUR and this in turn implies that the space is strictly convex.
By following the argument of the main lemma of [@MartioVaisala], we see the following useful observations. If $\lambda$ and $\gamma$ are rectifiable paths with unit quasihyperbolic speed in a convex domain of a strictly convex space and $$\label{eq: unit}
\left\|\frac{\lambda' + \gamma'}{2d(\frac{\lambda+\gamma}{2})}\right\|=1$$ then there is a representation $$\label{eq: presentation}
\lambda' (t)= F(t) d(\lambda(t) ),\quad \gamma' (t) = F(t) d(\gamma(t))$$ at points $t$ of differentiability of these paths. Here $F(t)$ is a norm-$1$ vector. Moreover, $$\begin{gathered}
\left\|\frac{\lambda' + \gamma'}{2d(\frac{\lambda+\gamma}{2})}\right\|
\leq \left\|\frac{\lambda' + \gamma'}{2\frac{d(\lambda)+d(\gamma)}{2}}\right\|
\leq \frac{\|\lambda' \|+ \|\gamma' \|}{d(\lambda)+d(\gamma)}\\
=\left\|\frac{F (d(\lambda) + d(\gamma))}{d(\lambda)+d(\gamma)}\right\|
=\|F\|
=\left\|\frac{F+F}{2}\right\| = \left\| \frac{\frac{\lambda'}{d(\lambda)}+\frac{\gamma'}{d(\gamma)}}{2}\right\|.\end{gathered}$$ The first inequality follows from the concavity of the distance function and the first equality from . Suppose that $\lambda$ and $\gamma$ have unit quasihyperbolic speed. Then we have $$\left\|\frac{\lambda'}{d(\lambda)}+\frac{\gamma'}{d(\gamma)}\right\|
\leq \frac{\|\lambda'\|}{d(\lambda)}+\frac{\|\gamma'\|}{d(\gamma)}
=\frac{d(\lambda)}{d(\lambda)}+\frac{d(\gamma)}{d(\gamma)}=2.$$ The geometric interpretation of these facts is that the quasihyperbolic length of point-wise average of paths is dominated by the average of the quasihyperbolic lengths of the mentioned paths. If ${\mathrm{X}}$ is a strictly convex space then the above readily yields that quasihyperbolic geodesics are unique and that quasihyperbolic balls are strictly convex.
\[lm: LUR\] Let ${\mathrm{X}}$ be a LUR Banach space. Suppose that $x,y_n \in {\mathrm{X}}$, $x\neq 0$, $n\in{\mathbb{N}}$, are vectors satisfying $$\lim_{n\to\infty} \|x\|+ \|y_n \| - \|x+y_n \| =0$$ where $(y_n )$ is a norm bounded sequence. Then $d(y_n ,[x])\to 0$ as $n\to\infty$.
Let $x$ and $y_n$ be vectors as above. Without loss of generality $\|x\|=1$. According to the Hahn-Banach theorem we may fix $f_n \in {\mathrm{X}}^*$, $\|f_n \|=1$, such that $f_n (x+ y_n )=\|x+y_n \|$ for $n\in{\mathbb{N}}$. Then $f_n (x) \to 1$ and $f_n (y_n ) - \|y_n \| \to 0$ as $n\to\infty$.
Write $(f_n \otimes x)(z)=f_n (z)x$, $z\in {\mathrm{X}}$. Note that $$\|(f_n \otimes x)(x+y_n )\|=|f_n (x+y_n )| \|x\| =\|x+y_n\|.$$ Observe that $$x +y_n = f_n (x + y_n )x -(f_n (x+y_n )x -(x+ y_n ))$$ where $$f_n ((f_n (x+y_n )x -(x+ y_n )))=f_n (x+y_n )f_n (x) - f_n (x+y_n ) \to 0 .$$
From any subsequence $(n_k )$ we may pick a further subsequence $(n_{k_j} )$ such that $f_{n_{k_j}}(x + y_{n_{k_j} })\to c$ as $n\to\infty$. Then, using the LUR condition with $cx$ in place of $x$, and noting that $$\left\|\frac{cx + (x+ y_{n_{k_j}})}{2}\right\|\to \|cx\|,\quad j\to\infty$$ yields that $x + y_{n_{k_j} } \to cx$. This suffices for the statement of the lemma, since $(n_k )$ was arbitrary.
The following lemma is from [@AFT].
[@AFT]\[lem:inequal\_series\] let $\lambda$ be a positive real number and let $\sum_{k=0}^\infty x_k$ be a convergent series with non-negative terms. Suppose that $$\lambda x_n \geq \sum_{k=n+1}^\infty x_k, \quad (n=0, 1, 2, \cdots)$$ Then, for $0<\alpha\leq1$ we have $$\sum_{k=0}^\infty x_k^\alpha \leq \frac{1}{(\lambda+1)^\alpha-\lambda^\alpha}\left(\sum_{k=0}^\infty x_k\right)^\alpha$$ with equality in the case $x_k=(\lambda/(\lambda+1))^k, k\geq0$.
Radon-Nikodym Property
----------------------
A Banach space is said to have the [*Radon-Nikodým property (RNP)*]{}, if any rectifiable and absolutely continuous path starting from the origin can be recovered by Bochner integrating its derivative.
For basic information about these concepts we refer to [@Diestel] and [@DiestelUhl77], see also [@FA_book].
Main Results
============
The first of our main results is the following improvement of [@RT Theorem 3.1].
Let $X$ be a uniformly convex Banach space whose modulus of convexity has power type. Let $\nu$ be the modulus of continuity of the weight function $w$. We assume that the function $\nu$ satisfies the condition $$\label{eq: dini_condition}
\limsup_{s\to0+}\frac{\int_0^s \frac{\nu(t)}{t}dt}{\nu(s)}<\infty.$$ Then every $d_w$-geodesic $\gamma$ is $C^1$ excluding the endpoints.
Above in we have a strengthening of a Dini type condition.
It is easy to see from the condition that there exists a constant $C$ and a positive integer $k_0>0$ such that for all $k\geq k_0$, $k\in\mathbb{N}$, $$\int_0^{2^{-k}} \frac{\nu(t)}{t}dt<C\nu(2^{-k}).$$ Since $$\sum_{j=k+1}^\infty\nu(2^{-j})\leq 2 \int_0^{2^{-j}}\frac{\nu(t)}{t}dt,$$ we have that for all $k\geq k_0$ $$\sum_{j=k+1}^\infty\nu(2^{-j})\leq C\nu(2^{-k}).$$ Then by Lemma \[lem:inequal\_series\], we get $$\sum_{j=k_0}^\infty \nu(2^{-j})^\alpha\leq \frac{1}{(C+1)^\alpha-C^\alpha}\left(\sum_{j=k_0}^\infty \nu(2^{-j})\right)^\alpha<\infty,$$ which implies that $$\label{eq: nu_convergence}
\sum_{j=1}^\infty \nu(2^{-j})^\alpha<\infty.$$
Let $$\beta(h)=\frac{1}{c}\left(\frac{2}{\omega_0}\right)^{\frac{1}{p}}\nu(3h)^{\frac{1}{p}},\quad p\geq2$$ be the same function as in Rasila-Talponen [@RT]. Using with $\alpha=\frac{1}{p}$, we see that $$\sum_{j=1}^\infty \beta\left(\frac{h}{2^j}\right)<\infty.$$ Now the theorem follows from the argument of Rasila-Talponen [@RT].
It is known that some convexity properties (e.g. uniform convexity and the RNP with strict convexity) of the underlying Banach space are transferred to the quasihyperbolic geometry in the case of a convex domain. It was asked in [@KRT] whether for a locally uniformly rotund (LUR) Banach space the quasihyperbolic-metric in a symmetric convex domain in fact induces via a Minkowski functional a norm which is LUR. This question is next settled affirmatively in the reflexive case.
Let ${\mathrm{X}}$ be a LUR reflexive Banach space and let $\Omega \subset {\mathrm{X}}$ be a convex domain. Let $x_0 , y, y_n \in \Omega$, $n\in{\mathbb{N}}$, such that $$k(x_0 , y )= k(x_0 , y_n ),\quad n\in{\mathbb{N}},$$ $$k(x_0 , \frac{y+y_n }{2})\to k(x_0 , y),\ n\to\infty .$$ Then $y_n \to y$ in norm as $n\to\infty$. Moreover, if $\Omega$ is symmetric then the Minkowski functional of the ball ${\mathbf{B}}_k (0 , r)$ is an equivalent LUR norm on ${\mathrm{X}}$ for any $r>0$.
The latter part of the statement follows from the first one by first observing that ${\mathbf{B}}_k (0 , r)\subset \Omega$ is a symmetric convex bounded subset including the origin as an interior point. Therefore the norm $|||\cdot |||$ induced by the Minkowski functional is equivalent to the given norm of the Banach space ${\mathrm{X}}$. Thus, in ${\mathbf{B}}_k (0 , r)$ the norm topology and the topology induced by the quasihyperbolic metric coincide, and in particular there is no need to distinguish between different modes of sequential convergence (norm vs. quasihyperbolic). By using the fact that the function $x\mapsto d(x, \partial \Omega)$ is bounded on ${\mathbf{B}}_k (0 , r)$ from above and from below by a positive constant, we obtain that $|||y_n ||| \to 1$ if and only if $k(0, y_n )\to 1$.
Let $x_0 ,y\in \Omega$ and $\ell = k(x_0 , y)$. Fix $y_n \in \Omega$, $n\in{\mathbb{N}}$, such that $k(x_0 , y_n)=\ell$ for all $n\in{\mathbb{N}}$ and $k(x_0 , \frac{y+y_n }{2})\to \ell$, $n\to\infty$.
Note that ${\mathrm{X}}$ is strictly convex, being LUR. Since ${\mathrm{X}}$ is a strictly convex reflexive Banach space and $\Omega$ is its convex domain, there is, up to a reparametrization, a unique quasihyperbolic geodesic $\lambda$ between $x_0$ and $y$, see [@RasilaTalponen12; @RT]. Let us investigate the unique quasihyperbolic unit speed quasihyperbolic geodesics $\lambda, \lambda_n \colon [0, \ell ] \to \Omega$ between $x_0 , y$ and $x_0 , y_n$, respectively. We will follow closely the arguments in [@MartioVaisala; @RasilaTalponen12; @RT]; see also the Preliminaries section for the Radon-Nikodym Property (RNP) and [@DiestelUhl77].
Recall that according to the RNP of ${\mathrm{X}}$ the paths $\lambda, \lambda_n$ are differentiable a.e. and can be recovered by integrating the derivatives in the Bochner sense: $$\lambda_n (t)=\lambda_n (0)+\int_{0}^t \lambda_{n}' (s)\ ds,\quad t\in [0,\ell].$$ According to the parametrization of the paths we have that $$\left\|\frac{\lambda' (t)}{d(\lambda(t))} \right\| = \left\|\frac{\lambda_{n}' (t)}{d(\lambda_{n}(t))} \right\| = 1,\quad n\in{\mathbb{N}},\ \text{for\ a.e.}\ t\in [0,\ell].$$
We wish to show that $$\|y - y_{n}\|\ \to 0,\quad n\to \infty .$$ It suffices to show that for *any* subsequence $(n_k)$ there exists a further subsequence $(n_{k_j})$ such that the above convergence holds when passing to this subsequence and letting $j\to\infty$ (because then clearly $\limsup_{n\to\infty}\|y-y_n \|=0$).
Since $B_k (x_0 , \ell)$ is convex we have that $k(x_0 , \frac{y+y_{n_k}}{2})\leq \ell$. We observe that $$1= \frac{1}{2}\left(\frac{\|\lambda' \|}{d(\lambda)}+\frac{\|\lambda_{n_k}' \|}{d(\lambda_{n_k})}\right)
\geq \frac{\frac{1}{2}(\|\lambda' \| + \|\lambda_{n_k}' \|)}{\frac{1}{2}(d(\lambda) +d(\lambda_{n_k}))}
\geq\frac{\frac{1}{2}\left\|\lambda' + \lambda_{n_k}' \right\|}{d\left(\frac{\lambda +\lambda_{n_k}}{2}\right)}$$ and $$\ell=\int_{0}^\ell \frac{1}{2}\left(\frac{\|\lambda' \|}{d(\lambda)}+\frac{\|\lambda_{n_k}' \|}{d(\lambda_{n_k})}\right)\geq
\int_{0}^\ell \frac{\frac{1}{2}\left\|\lambda' + \lambda_{n_k}' \right\|}{d\left(\frac{\lambda +\lambda_{n_k}}{2}\right)}
\geq k\left(x_0 , \frac{y+y_{n_k}}{2}\right) \to \ell$$ as $k\to\infty$.
It follows that $$\frac{\frac{1}{2}\left\|\lambda' + \lambda_{n_k}' \right\|}{d\left(\frac{\lambda +\lambda_{n_k}}{2}\right)} \to 1$$ in $L^1$ and in measure as $k\to\infty$. Thus there is a further subsequence $(n_{k_j})$ such that $$\frac{\frac{1}{2}\left\|\lambda' (t) + \lambda_{n_{k_j}}' (t) \right\|}{d\left(\frac{\lambda(t) +\lambda_{n_{k_j}}(t)}{2}\right)} \to 1$$ for a.e. $t\in [0,\ell]$ as $j\to\infty$.
Using an adaptation of considerations after we get $$\left\|\frac{\lambda'}{d(\lambda)} + \frac{\lambda_{n_{k_j}}'}{d(\lambda_{n_{k_j}})}\right\|\to 2$$ for a.e. $t$ as $j\to\infty$. It follows from the LUR assumption that $$\left\|\frac{\lambda'}{d(\lambda)} - \frac{\lambda_{n_{k_j}}'}{d(\lambda_{n_{k_j}})}\right\|\to 0$$ for a.e. $t$ as $j\to\infty$. Lebesgue’s dominated convergence theorem then yields $$\label{eq: LebgDCT}
\int_{0}^\ell \left\|\frac{\lambda'}{d(\lambda)} - \frac{\lambda_{n_{k_j}}'}{d(\lambda_{n_{k_j}})}\right\|\ dm\to 0,\quad j\to\infty .$$ By [@MartioVaisala Lemma 3.3], the path $\lambda_{n_{k_{j}}}$ converges uniformly to the geodesic path $\lambda$ and hence, $$\label{eq: pathUC}
\left\|\frac{\lambda_{n_{k_{j}}}'(t)}{d(\lambda_{n_{k_{j}}}'(t))}-\frac{\lambda_{n_{k_{j}}}'(t)}{d(\lambda'(t))}\right\|\to 0, \quad j\to\infty,$$ uniformly. Since $$y_{n_{k_j}}=x_0 + \int_{0}^\ell \lambda_{n_{k_j}}' (t)\ dt$$ and $$y=x_0 + \int_{0}^\ell \lambda' (t)\ dt$$ by the Radon-Nikodym Property, we have $$\begin{aligned}
\|y_{n_{k_j}}-y\|&=&\left\|\int_0^\ell(\lambda_{n_{k_{j}}}'(t)-\lambda'(t))dt\right\|\\
&\leq&\int_0^\ell\left\|\lambda_{n_{k_{j}}}'(t)-\lambda'(t)\right\|dt\\
&\leq&\max\limits_{0\leq t\leq \ell}\{d(\lambda(t))\}\int_0^\ell\left\|\frac{\lambda_{n_{k_{j}}}'(t)}{d(\lambda(t))}-\frac{\lambda'(t)}{d(\lambda(t))}\right\|dt\\
&\leq&\max\limits_{0\leq t\leq \ell}\{d(\lambda(t))\}\left(\int_0^\ell\left\|\frac{\lambda_{n_{k_{j}}}'(t)}{d(\lambda_{n_{k_{j}}}(t))}-\frac{\lambda'(t)}{d(\lambda(t))}\right\|dt\right.\\
& &\left. +\int_0^\ell\left\|\frac{\lambda_{n_{k_{j}}}'(t)}{d(\lambda_{n_{k_{j}}}(t))}-\frac{\lambda_{n_{k_{j}}}'(t)}{d(\lambda(t))}\right \|dt \right)\\
&\to&0\end{aligned}$$ which follows from and .
Thus $y_{n} \to y$ in norm as $k\to \infty$.
The following result has some bearing on some previous results on the existence and smoothness of geodesics and the smoothness of the quasihyperbolic metric in convex domains, see [@RasilaTalponen12; @RT].
\[thm: unique\_smooth\] Let ${\mathrm{X}}$ be a Banach space. Let $\Omega \subset {\mathrm{X}}$ be a domain and $\gamma \colon [0,\ell]\to \Omega$ be a quasihyperbolic unit speed quasihyperbolic geodesic between points $x_0 , x \in \Omega$ (and of quasihyperbolic length $\ell$). Suppose that the Gateaux derivative $(k (x_0 , \cdot))' (x)$ exists and the vector derivative $\gamma' (\ell )=\lim_{t\to \ell^- } \frac{\gamma(\ell)-\gamma(t)}{\ell- t}$ exists. Then $$\sup_{z \in {\mathbf{S}}_{\mathrm{X}}} (k (x_0 , \cdot))' (x) \left[\frac{z}{d(x)}\right] = (k (x_0 , \cdot))' (x) [\gamma' (\ell )]$$ where $\|\gamma'(\ell)\|=d(x)$.
The latter part of the claim is clear from the continuity of the distance function. Therefore the claim reduces to checking that $$\|(k (x_0 , \cdot))' (x)\|_{{\mathrm{X}}^*} = (k (x_0 , \cdot))' (x) \left[\frac{\gamma' (\ell )}{d(x)}\right]$$ where $\left\|\frac{\gamma' (\ell )}{d(x)} \right\|=1$.
Clearly $$\label{eq: dx}
\frac{1}{d(x)}\geq \|(k (x_0 , \cdot))' (x) \|_{{\mathrm{X}}^*} \geq (k (x_0 , \cdot))' (x) \left[\frac{\gamma' (\ell )}{d(x)}\right]> 0.$$ Note that according to the definition of the geodesic $k (x_0 , \gamma(t))=t$ for each $t\in [0,\ell]$ and therefore the chain rule gives us that $$\frac{d}{dt} k (x_0 , \gamma(t))\Big|_{t=\ell} = (k(x_0 , \cdot ))' (x) [\gamma' (\ell)]=1.$$ This means that the first two inequalities in hold as equalities. This proves the claim.
Let ${\mathrm{X}}$ be a Gateaux smooth Banach space. Let $\Omega \subset {\mathrm{X}}$ be a domain and $\gamma \colon [0,\ell]\to \Omega$ be a quasihyperbolic unit speed quasihyperbolic geodesic between points $x_0 , x \in \Omega$ (and of quasihyperbolic length $\ell$). Assume that $y=\gamma (t)$ for some $0<t<1$. Suppose $k (x_0 , \cdot)$ and $k(y,\cdot)$ are Gateaux differentiable at $x$ and $\gamma$ has left derivative at $\ell$. Then $k (x_0 , \cdot)' (x)$ and $k (y , \cdot)' (x)$ coincide.
In particular, the geometric interpretation of this fact is that the unique tangent spaces of the spheres ${\mathbf{S}}_k (x_0, \ell)$ and ${\mathbf{S}}_k (y, \ell-t)$ at $x$ coincide.
The functionals $d(x)(k (x_0 , \cdot)' (x)),\ d(x)(k (y , \cdot)' (x)) \in {\mathbf{S}}_{{\mathrm{X}}^*}$ attain their norm at $\frac{\gamma' (\ell )}{d(x)}$ according to Theorem \[thm: unique\_smooth\] and its proof. Thus, the Gateaux smoothness of ${\mathrm{X}}$ together with the Smulyan lemma yields that these functionals coincide.
The following result is an immediate consequence of the smoothness of quasihyperbolic balls.
Suppose that ${\mathrm{X}}$ is a uniformly smooth Banach space and $\Omega \subset {\mathrm{X}}$ is a convex domain. Then the geometric conclusion of the previous result holds: If $B_k (y,s) \subset B_k (x,r)$ and $z \in S_k (x,r) \cap S_k (y,s)$ then the quasihyperbolic spheres $S_k (x,r)$ and $S_k (y,s)$ have the same tangent space at $z$.
In particular, the assumption on ${\mathrm{X}}$ is valid above if ${\mathrm{X}}$ is a Hilbert space, e.g. ${\mathbb{R}}^n$ with the usual norm.
The statement follows from the fact that under the assumptions for any $x_0 \in \Omega$ the quasihyperbolic metric $k(x_0 ,\cdot)$ is continuously Frechet differentible in $\Omega$ away from $x_0$, see Theorem 2.7 in [@KRT2]. Indeed, recall that the quasihyperbolic balls in convex domains are convex. This means that the tangent space $T$ of $B_k (x,r)$ at $z$ exists. Moreover, the tangent space is unique according to the Frechet differentiablility of $k(x_0 ,\cdot)$. Similar fact holds also for $B_k (y,s)$. Clearly $T$ is also the unique tangent space of $B_k (y,s) \subset B_k (x,r)$ at $z$.
Final remarks
=============
Professor Beata Randrianantoanina has asked in a personal communication the following general question relating the geometry of Banach spaces to the properties of quasihyperbolic manifolds modeled on these spaces.
Which Banach space properties can be characterized by the corresponding quasihyperbolic metric induced norm, following [@KRT2]?
The authors have also previously risen the question about the characterization of reflexive Banach spaces in terms of quasihyperbolic metrics. Namely, every reflexive Banach space has the property that for every convex domain the corresponding quasihyperbolic metric space is geodesic, see [@RasilaTalponen12].
Does the converse implication hold?
It is useful to observe that if holds suitably asymptotically, for a pair of sequences of paths, then the conlusion holds asymptotically as well. This is convenient in particular in the setting of uniformly convex spaces.
Corrigendum. The following result appears in [@RasilaTalponen12].
Let ${\mathrm{X}}$ be a Banach space, $\Omega\subsetneq{\mathrm{X}}$ a domain. Then each $j$-ball ${\mathbf{B}}_{j}(x_{0},r),\ x_{0}\in \Omega,$ is starlike for radii $r\leq \log 2$.
Although the statement of the theorem is correct, the proof contains a blunder with the grouping of terms. The corrected proof goes as follows.
Let $x_{0},y\in \Omega$, $x_{0}\neq y$, such that $j(x_{0},y)\leq \log 2$. This is to say that $$\frac{\|x_{0}-y\|}{d(x_{0})\wedge d(y)}\leq 1.$$ By using simple calculations involving the triangle inequality we get $$\begin{aligned}
j(x_{0},ty+(1-t)x_{0})&\leq &\log\left(1+\frac{t\|x_{0}-y\|}{d(x_{0})\wedge (d(y)-(1-t)\|x_{0}-y\|)}\right)\\
&\leq &\log\left(1+\frac{t\|x_{0}-y\|}{t\|x_{0}-y\|}\right)=\log 2,\quad t>0 .\end{aligned}$$
Acknowledgments
---------------
This research was supported by the Academy of Finland Project \#268009. The second named author was also financially supported by the Finnish Cultural Foundation and Väisälä Foundation. The third named author was also supported by NNSF of China Project No. 11601485.
[DGZ]{}
, and [G. Tenenbaum]{}: Entropy: An inequality. *Tokyo J. Math.* **11** (1988), 323–328.
and [V. Zizler]{}: [*Functional Analysis and Infinite-Dimensional Geometry*]{}, Volume 8 of CMS Books in Mathematics, Canadian Mathematical Society, Springer, 2001.
, [*The geometry of discrete groups*]{}, Graduate Texts in Mathematics [**91**]{}, Springer-Verlag, New York, 1983.
and [V. Zizler]{}: [*Smoothness and renormings in Banach spaces*]{}. Longman Scientific & Technical, 1993.
: [*Geometry of Banach spaces – selected topics.*]{} Lecture Notes in Mathematics, Vol. 485. Springer-Verlag, Berlin-New York, 1975.
and [J.J. Uhl]{}: [*Vector measures*]{}. Math. Surveys 15, AMS, Providence, R.I. 1977.
: Unpublished note to Matti Vuorinen in 1979.
and [B.G. Osgood]{}: Uniform domains and the quasi-hyperbolic metric. [*J. Anal. Math.*]{} [**36**]{} (1979), 50–74.
and [B.P. Palka]{}: Quasiconformally homogeneous domains. [*J. Anal. Math.*]{} [**30**]{} (1976), 172–199.
and [J. Talponen]{}: Smooth approximations of norms in separable Banach spaces. [*Q. J. Math.*]{} [**65**]{} (2014), 957–969.
: [*Lectures on analysis on metric spaces.*]{} Universitext. Springer-Verlag, New York, 2001.
: [ Semisolidity and locally weak quasisymmetry of homeomorphisms in metric spaces]{}. To appear in [*Studia Math.*]{} arXiv:1409.5631
: Local convexity properties of quasihyperbolic balls in punctured space. [*J. Math. Anal. Appl.*]{} [**342**]{} (2008), 192–201.
: On hyperbolic type metrics. [*Ann. Acad. Sci. Fenn. Math. Diss.*]{} [**152**]{}, 2009.
and [J. Talponen]{}: Quasihyperbolic geometry in Euclidean and Banach spaces. [*J. Anal.*]{} [**18**]{} (2010), 261–278.
and [J. Talponen]{}: On the smoothness of quasihyperbolic balls. [*Ann. Acad. Sci. Fenn. Math.*]{} 42 (2017), 439–452.
Quasiconformal and bi-Lipschitz homeomorphisms, uniform domains and the quasihyperbolic metric. *Trans. Amer. Math. Soc.* **292** (1985), 169–191.
and [J. Väisälä]{}: [Quasihyperbolic geodesics in convex domains II]{}. [*Pure Appl. Math. Q.*]{} [**7**]{} (2011), 395–409.
and [J. Talponen]{}: [Convexity properties of quasihyperbolic balls on Banach spaces]{}. [*Ann. Acad. Sci. Fenn. Math.*]{} [**37**]{} (2012), 215–228. and [J. Talponen]{}: [On Quasihyperbolic Geodesics in Banach Spaces]{}. [*Ann. Acad. Sci. Fenn. Math.*]{} [**39**]{} (2014), 163–173.
. [Lecture Notes in Math., Vol. 1319, Springer-Verlag]{}, [Berlin]{}, 1988.
: Metrics and quasiregular mappings. [*Proceedings of International Workshop on Quasiconformal Mappings and Their Applications, December 27, 2005- Jan 1, 2006,*]{}, 291–325, 354, Narosa Publ Co, 2007, ISBN 81-7319-807-1.
: Free quasiconformality in Banach spaces. IV. [*Analysis and topology*]{}, 697–717, World Sci. Publ., River Edge, NJ, 1998.
: The free quasiworld. Freely quasiconformal and related maps in Banach spaces. Quasiconformal geometry and dynamics (Lublin, 1996), 55–118, [*Banach Center Publ.*]{}, 48, Polish Acad. Sci., Warsaw, 1999.
: Tangential properties of quasihyperbolic geodesics in Banach spaces. [*Ann. Acad. Sci. Fenn. Math.*]{} [**38**]{} (2013), 825–837.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The hyperbolic (and more generally, Lorentzian) Kac-Moody (KM) Lie algebras $\cA$ of rank $r+2 > 2$ are shown to have a rich structure of indefinite KM subalgebras which can be described by specifying a subset of positive real roots of $\cA$ such that the difference of any two is not a root of $\cA$. Taking these as the simple roots of the subalgebra gives a Cartan matrix, generators and relations for the subalgebra. Applying this to the canonical example of a rank 3 hyperbolic KM algebra, $\cF$, we find that $\cF$ contains all of the simply laced rank 2 hyperbolics, as well as an infinite series of indefinite KM subalgebras of rank 3. It is shown that $\cA$ also contains Borcherds algebras, obtained by taking all of the root spaces of $\cA$ whose roots are in a hyperplane (or any proper subspace). This applies as well to the case of rank 2 hyperbolics, where the Borcherds algebras have all their roots on a line, giving the simplest possible examples.'
address:
- 'Dept. of Math. Sci., The State University of New York, Binghamton, New York 13902-6000 '
- 'Max-Planck-Institut für Gravitationsphysik, Albert-Einstein-Institut, Mühlenberg 1, D-14476 Golm, Germany'
author:
- 'Alex J. Feingold'
- Hermann Nicolai
title: 'Subalgebras of Hyperbolic Kac-Moody Algebras'
---
[^1]
[*We dedicate this work to the memory of our friend Peter Slodowy.*]{}
Introduction
============
More than thirty years after their discovery, indefinite and, more specifically, hyperbolic Kac-Moody (KM) algebras remain an unsolved challenge of modern mathematics. For instance, there is not a single such algebra for which the root multiplicities are known in closed form. Even less is known about the detailed structure of the imaginary root spaces, and a manageable explicit realization analogous to the current algebra realization of affine algebras appears to be beyond reach presently. (However, there are intriguing hints that a “physical” realization might be found in an extension of Einstein’s theory of general relativity, see [@DHN] and references therein). Given this lack of knowledge, any new information about these algebras may be of potential value and importance.
In this note we point out that hyperbolic KM algebras possess a very rich structure of nonisomorphic infinite dimensional subalgebras, all of which can be generated by a set of root vectors $E_\alpha$ and $F_\alpha$, where certain roots $\alpha$ from the root system of the original algebra are taken as new simple roots. If these roots are real, then the subalgebra can be an indefinite KM algebra; however, we will also exhibit examples where the new simple roots are imaginary, and the resulting subalgebra is a Borcherds algebra [@B1; @Jur1; @Nie]. Two subalgebras are said to be $W$-equivalent if their respective simple roots are related by a Weyl transformation from the Weyl group of the original algebra. Geometrically, some of the non-$W$-equivalent subalgebras of indefinite KM algebras can be understood in terms of conical sections. Generally there are three ways of slicing indefinite (and hyperbolic) KM algebras corresponding to the intersections of the light-cone with a hyperplane in the dual of the Cartan subalgebra. That dual space is equipped with an indefinite bilinear form, $(\alpha,\beta)$, and depending on whether the hyperplane is space-like, light-like or time-like (meaning that its normal vector, $\nu$, satisfies $\nu^2 = (\nu,\nu) < 0$, $\nu^2 = 0$, or $\nu^2 > 0$) one obtains elliptic, parabolic or hyperbolic sections corresponding to a decomposition of the original KM algebra with respect to a finite, an affine or an indefinite subalgebra. (The latter need not be hyperbolic, even if the original algebra is hyperbolic.)
Previous attempts to understand indefinite KM algebras were almost exclusively based on a decomposition into irreducible representations of their affine subalgebras. (Decompositions with respect to finite subalgebras seem to have received scant attention so far.) Although the theory of representations of affine algebras is reasonably well understood, this approach has not yielded much information about the behavior of the algebra “deep within the light-cone”. A complete analysis was carried out up to affine level two in [@FF] for the rank 3 hyperbolic algebra, $\cF$ (also referred to as $HA_1^{(1)}$), containing the affine algebra $A_1^{(1)}$. This work was generalized in [@KMW] to the rank 10 hyperbolic KM algebra $E_{10}$, where the first two levels with respect to the affine subalgebra $E_9 = E_8^{(1)}$ were analyzed. This analysis has been extended as far as level 5 for $\cF$ and also carried out for some levels for some other algebras [@BKM1; @BKM2; @BKM3; @Ka1; @Ka2; @Ka3; @Ka4; @Ka5; @KaM1; @KaM2], but appears rather hopeless as a method for giving all levels. Finally, DDF operators have been used in [@GN] to probe some root spaces in detail.
We conclude this introduction with a brief description of the main results in each section. In section 2 we give the necessary background and details about the rank 3 hyperbolic KM algebra, $\cF$, which the most decisive example for understanding the subject. In section 3 we prove the key theorem which identifies a class of subalgebras of KM algebras by the choice of a set of positive real roots whose differences are not roots. We apply that theorem to the algebra $\cF$ to find inside it all simply laced rank 2 hyperbolic KM algebras, as well as an infinite series of inequivalent rank 3 indefinite KM algebras. We have generalizations of these results where the role of the algebra $\cF$ is played by an arbitrary rank $r+2$ Lorentzian KM algebra, $\cA$, and we find two infinite series of inequivalent indefinite KM subalgebras, one series of rank $r+1$ and one series of rank $r+2$. In section 4 we go back to the special case of $\cF$, and exploit the beautiful geometry of its Weyl group, $W$, which is the hyperbolic triangle group $T(2,3,\infty) = PGL_2(\bZ)$, through its action on the Poincaré disk model of the hyperbolic plane. This allows us to find, in principle, a large class of KM subalgebras of $\cF$ which are inequivalent to those already found in section 3. A complete classification of such subalgebras is beyond the scope of this paper, but we give a number of representative examples to illustrate the idea. We also discuss how the geometry of the Poincaré disk gives information about the generators and relations defining the Weyl groups of these subalgebras, and how they are related to the Weyl group $W$. We then use this geometrical picture to relate the Weyl groups of the series of rank 3 indefinite subalgebras of $\cF$ found in section 3 to $W$. (Thanks to Tadeusz Januszkiewicz for suggesting and explaining this geometrical way of finding subgroups of $W$.) In section 5 we give a method of finding Borcherds generalized KM algebras inside $\cF$, inside and Lorentzian algebra $\cA$, and even inside rank 2 hyperbolics.
The rank 3 algebra $\cF$
========================
The canonical example of a hyperbolic KM algebra is the rank 3 algebra $\cF$ studied in [@FF], whose Cartan matrix is $$\bmatrix 2 & -1 & 0 \\ -1 & 2 & -2 \\ 0 & -2 & 2 \endbmatrix\ ,$$ whose simple roots are $\alpha_{-1}$, $\alpha_0$, $\alpha_1$, and whose Dynkin diagram is $$\bullet\underset{\hskip -10pt\alpha_{-1}} \ \hbox{-----} \
\bullet\underset{\hskip -10pt\alpha_{0}}\
\vbox{\hbox{-----}\vskip -10pt \hbox{-----}
\vskip -10pt\hbox{-----} \vskip -10pt\hbox{-----}\vskip -3pt}\
\bullet\underset{\hskip -10pt\alpha_{1}}\ .$$ This algebra is the minimal rank hyperbolic KM algebra with both finite and affine KM subalgebras. Rank 2 hyperbolic KM algebras are also quite interesting ([@F; @KaM2]) because of their connections with real quadratic fields and Hilbert modular forms [@LM]. They have infinitely many $A_1$ subalgebras, infinitely many rank 2 hyperbolic KM subalgebras, and our new work shows that they contain Borcherds subalgebras as well. But our main inspiration has been the hyperbolic algebra $\cF$, which seems to be the simplest such algebra that incorporates all the essential features of higher rank examples. Further evidence that $\cF$ deserves most serious study is that it contains all the symmetric rank 2 hyperbolics, as well as many other rank 3 hyperbolics. From a physicist’s point of view, it is attractive because it may appear as a hidden symmetry in (an extension of) Einstein’s theory of gravity in four space-time dimensions (see [@DHN] for a review of recent developments and further references).
Many interesting results about the structure of $\cF$ were obtained in [@FF], as well as a relationship with Siegel modular forms of genus 2, but a complete understanding of the root multiplicities has remained elusive despite considerable further work ([@BKM1; @Ka3; @Ka4; @Ka5; @KaM1]). It is apparent from the Cartan matrix and Dynkin diagram that $\cF$ contains an affine subalgebra $\cF_0$ of type $A_1^{(1)}$ with simple roots $\alpha_0$, $\alpha_1$. The approach in [@FF] is based on the decomposition $$\cF = \bigoplus_{n\in\bZ} \cF_n$$ with respect to $\cF_0$, that is, with respect to the “level” which is the value of the central element of $\cF_0$. The feature of $\cF$ which first attracted the attention of AJF was its Weyl group, $W$, which is a reflection group $$\la r_{-1},r_0,r_1\ |\ r_{-1}^2 = r_0^2 = r_1^2 = 1, (r_{-1} r_0)^3 = 1,
(r_{-1} r_1)^2 = 1\ra$$ isomorphic to the hyperbolic triangle group, $T(2,3,\infty)$ and to $PGL_2(\bZ)$. The action of $W$ shows that there are infinitely many subalgebras of type $A_1^{(1)}$ inside $\cF$, corresponding to cusps of the modular group, or to lines of null roots on the light-cone. However, these lines of null roots are all related by the Weyl group, and therefore the corresponding affine subalgebras of $\cF$ are all $W$-equivalent.
To give more details, let $H$ be the Cartan subalgebra of $\cF$ and let $H^*$ be its dual space, which has a basis consisting of the simple roots $\{\alpha_{-1}, \alpha_0,\alpha_1\}$, and a symmetric bilinear form whose matrix with respect to this basis is the Cartan matrix above. That form is indefinite, and can be very conveniently described as follows. On the space $S_2(\bR)$ of $2 \x 2$ symmetric real matrices define a bilinear form by $$\bpm a & b\\b & c\epm \cdot
\bpm a' & b'\\b' & c'\epm
= 2bb' - ac' - a'c$$ so the associated quadratic form is $$2b^2 - 2ac = -2 \det \bpm a & b\\b & c\epm.$$ Then the root lattice of $\cF$ is isometric to $S_2(\bZ)$ where the entries $a$, $b$ and $c$ are integers, and the weight lattice of $\cF$ is isometric to $S'_2(\bZ)$ where $a$ and $c$ are integers but $b\in \frac{1}{2}\bZ$. We make the correspondence explicit by choosing $$\alpha_{-1} = \bpm 1 & 0\\0 & -1\epm, \quad
\alpha_0 = \bpm -1 & -1\\-1 & 0\epm,\quad
\alpha_1 = \bpm 0 & 1\\1 & 0\epm.$$ The Weyl group action of $A \in PGL_2(\bZ)$ on $N = \bpm a & b\\b & c\epm$ is given by $A(N) = ANA^t$ and the explicit matrices which represent the three simple reflections are $$r_{-1} = \bpm 0 & 1\\1 & 0\epm, \quad
r_0 = \bpm -1 & 1\\0 & 1\epm,\quad
r_1 = \bpm 1 & 0\\0 & -1\epm.$$ The root system of $\cF$ is $$\Phi = \{N\in S_2(\bZ)\ |\ \det(N) \geq -1\}$$ which is just the elements $N$ of the root lattice for which the norm squared is less than or equal to 2, $-2\det(N) \leq 2$. The real roots (Weyl conjugates of the simple roots) are $$\Phi_{real} = \{N\in S_2(\bZ)\ |\ \det(N) = -1\}$$ which lie on a single sheeted hyperboloid. The light-cone is the set of points $N\in S_2(\bR)$ where $\det(N) = 0$. All real roots have multiplicity one, and in the case of $\cF$, the same is true of all roots on the light-cone. This comes from $\cF_0$ (with simple roots $\alpha_0$ and $\alpha_1$) whose underlying finite dimensional Lie algebra is the rank one Lie algebra $sl_2$ of type $A_1$. Any light-cone root, $N\in\Phi$, satisfies $\det(N) = 0$ and is $W$ equivalent to a unique one on level 0 of the form $\bpm a &0\\0 &0\epm = -a(\alpha_0+\alpha_1)$ with $0\neq a\in\bZ$. The roots $\Phi$ decompose into the disjoint union of positive and negative roots, and we can distinguish these level $0$ light-cone roots easily according to whether $a$ is negative or positive. The roots on level 1 are of the form $\bpm a &b\\b &1\epm = -(a+1)(\alpha_0+\alpha_1) + b\alpha_1 - \alpha_{-1}$ with $a-b^2\geq -1$ so $a\geq -1$. The affine Weyl group of $\cF_0$ is an infinite dihedral group generated by $r_0$ and $r_1$, and it preserves each level. Each root on level 1 is equivalent by the affine Weyl group to one of the form $\bpm a &0\\0 &1\epm = -(a+1)(\alpha_0+\alpha_1) - \alpha_{-1}$, and because level 1, $\cF_1$, is a single irreducible $\cF_0$-module (the basic module), the multiplicity of that root is a value of the classical partition function, $p(a+1)$ [@FL]. Each root on level 2 is equivalent by the affine Weyl group to one of the form $\bpm a &0\\0 &2\epm$ or $\bpm a &1\\1 &2\epm$. The $\cF_0$-module structure of level 2 is much more subtle, being the antisymmetric tensors in $\cF_1\otimes\cF_1$ with one irreducible module removed. If we form a generating function of the level 2 root multiplicities in those two columns, $$\sum_{a=0}^\infty {\rm Mult}\bpm a &1\\1 &2\epm\ q^{2a} +
{\rm Mult}\bpm a &0\\0 &2\epm\ q^{2a+1}$$ it is remarkable that the first 20 coefficients are again given by the partition function $p(\det(N)+1)$. The exact formula for that generating function $$\left[\sum_{n\geq 0} p(n) q^n\right] \left[\prod_{j\geq 1} (1 - q^{4j-2})\right]
\frac{q^{-3}}{2}\left[ \prod_{j\geq 1} (1 + q^{2j-1}) - \prod_{j\geq 1} (1 - q^{2j-1})
- 2q \right]$$ was one of the main results of [@FF], but further work has shown the increasing complexity of higher levels as $\cF_0$-modules.
There are other ways in which one might naturally decompose $\cF$, for example, with respect to the finite dimensional subalgebra of type $A_2$ with simple roots $\alpha_{-1}$ and $\alpha_0$. In that decomposition the analogue of level of a root $\alpha = \sum a_i \alpha_i$ would be the coefficient $a_1$ of $\alpha_1$. Each graded piece of the decomposition would be a finite dimensional $A_2$ module, and it is not hard to use a computer to find the irreducible modules which occur and their multiplicities for quite a few levels[^2]. The question is whether there is a useful (recursive) description of those module multiplicities which sheds more light on the hyperbolic root multiplicities.
Indefinite subalgebras from subroot systems
===========================================
The main point of the present contribution is that indefinite KM algebras possess infinitely many non-isomorphic subalgebras, which are of indefinite type (in fact, KM subalgebras of equal rank with inequivalent Cartan matrices). Unlike finite or affine subalgebras, these subalgebras are themselves not yet well understood, but the corresponding decompositions may nevertheless provide valuable new viewpoints.
We will use the following theorem, which allows us to find subalgebras by locating a set of simple roots for the subalgebra within the root system of the larger algebra. One cannot just choose an arbitrary subset of positive roots and declare them to be the simple roots of a subalgebra. For example, if root vectors $E_i$ correspond to simple roots $\beta_i$ and root vectors $F_j$ correspond to their negatives $-\beta_j$, then one of the required Serre relations, $[E_i,F_j] = \delta_{ij} H_i$ could be violated for $i\neq j$ if $\beta_i - \beta_j$ were a root of the larger algebra.
\[subrootsys\] Let $\Phi$ be the set of roots of a Kac-Moody Lie algebra, ${{\mathfrak{g}}}$, with Cartan subalgebra, ${{\mathfrak{h}}}$, and let $\Phi_{real}^+$ be the positive real roots of ${{\mathfrak{g}}}$. Let $\beta_1,\cdots,\beta_n\in\Phi_{real}^+$ be chosen such that for all $1\leq i \neq j \leq n$, we have $\beta_i - \beta_j\notin \Phi$. For $1\leq i\leq n$ let $0\neq E_i\in{{\mathfrak{g}}}_{\beta_i}$ and $0\neq F_i\in{{\mathfrak{g}}}_{-\beta_i}$ be root vectors in the one-dimensional root spaces corresponding to the positive real roots $\beta_i$, and the negative real roots $-\beta_i$, respectively, and let $H_i = [E_i,F_i]\in{{\mathfrak{h}}}$. Then the Lie subalgebra of ${{\mathfrak{g}}}$ generated by $\{E_i,F_i,H_i\ |\ 1\leq i\leq n\}$ is a Kac-Moody algebra with Cartan matrix $C = [C_{ij}] = [2(\beta_i,\beta_j)/(\beta_j,\beta_j)]$. Denote this subalgebra by ${{\mathfrak{g}}}(\beta_1,\cdots,\beta_n)$.
By construction, the elements $H_i$ are in the Cartan subalgebra, ${{\mathfrak{h}}}$, so the following relations are clear: $$[H_i,E_j] = \beta_j(H_i) E_j, \quad
[H_i,F_j] = -\beta_j(H_i) F_j, \quad
[H_i,H_j] = 0.$$ Because all of the $\beta_i$ are positive real roots, the matrix $C$ satisfies the conditions to be a Cartan matrix. For $i\neq j$, the bracket $[E_i,F_j]$ would be in the $\beta_i - \beta_j$ root space, but we have chosen the $\beta_i$ such that this difference is not a root of ${{\mathfrak{g}}}$, so that bracket must be zero, giving the relations $$[E_i,F_j] = \delta_{ij} H_i.$$ To check that $(ad\ E_i)^{1-C_{ji}}\ E_j = 0$ for all $i\neq j$, it would suffice to show that $(1 - C_{ji})\beta_i + \beta_j$ is not in $\Phi$. Since $\beta_j$ is a real root of ${{\mathfrak{g}}}$, and since $\beta_j - \beta_i\notin\Phi$, the $\beta_i$ root string through $\beta_j$ is of the form $$\beta_j, \beta_j + \beta_i, \cdots, \beta_j + q\beta_i$$ where $-q = 2(\beta_j,\beta_i)/(\beta_i,\beta_i) = C_{ji}$. Therefore, $$\beta_j + (q+1)\beta_i = \beta_j + (1 - C_{ji})\beta_i$$ is not in $\Phi$. The relations $(ad\ F_i)^{1-C_{ji}}\ F_j = 0$ for all $i\neq j$, follow immediately since $-(1 - C_{ji})\beta_i - \beta_j$ is not in $\Phi$. This shows that all of the Serre relations determined by the Cartan matrix $C$ are satisfied by the generators given above.
Our first application of this theorem is to show that the rank 3 algebra $\cF$ contains all the simply laced rank 2 hyperbolic KM algebras as subalgebras. The decomposition of $\cF$ with respect to its affine algebra $\cF_0$ corresponded to slicing with respect to planes parallel to an edge of the light-cone, and the affine Weyl group acted on those planes by moving roots along parabolas. Decomposition with respect to the finite dimensional algebra $A_2$ corresponds to slices which intersect the light-cone in circles or ellipses. So it should come as no surprise that there should be subalgebras whose decompositions correspond to slices which intersect the light-cone in hyperbolas. Consequently we will “locate” the rank 2 hyperbolic KM algebras by identifying their simple root systems inside the root system of $\cF$. There are two equally good choices for each algebra, distinguished in the theorem below by a choice of plus or minus sign.
\[rank2thm\] Fix any integer $m \geq 1$. In the root system $\Phi$ of $\cF$ we have the positive root vectors $$\beta_0 = \alpha_{-1} = \bpm 1 & 0\\0 & -1\epm,$$ $$\beta_1 = \beta_1^{(m)} = m(\alpha_0 + \alpha_1) \pm \alpha_1
= \bpm -m &\pm 1\\\pm 1 &0\epm.$$ Then the KM Lie subalgebra ${{\mathfrak{g}}}(\beta_0,\beta_1)$ constructed in Theorem \[subrootsys\] has Cartan matrix $$\bbm 2 & -m \\ -m & 2 \ebm$$ which is rank 2 hyperbolic for $m \geq 3$. To indicate the dependence on $m$ we will denote this subalgebra by $\cH(m)$.
The result follows from Theorem \[subrootsys\] because $\beta_0, \beta_1^{(m)} \in\Phi_{real}^+$ and $$\beta_1^{(m)} - \beta_0 = m(\alpha_0 + \alpha_1) \pm \alpha_1 - \alpha_{-1}
= \bpm -m-1 &\pm 1\\\pm 1 &1\epm$$ has determinant $-m-2 < -2$ for $m\geq 1$, so is not a root of $\cF$.
While the real root vector $E_0$ of $\cH(m)$ may be taken to be the simple root generator $e_{-1}$ of $\cF$, the real root vector $E_1$ may be written as the $2m$-fold commutator $$E_1 = [e_1,[e_0,[e_1,\cdots [e_1,[e_0 , e_1] \cdots ]]]$$ if $\beta_1^{(m)} = m(\alpha_0 + \alpha_1) + \alpha_1$, and as the commmutator $$E_1 = [e_0,[e_1,[e_0,\cdots [e_0,[e_1 , e_0] \cdots ]]]$$ if $\beta_1^{(m)} = m(\alpha_0 + \alpha_1) - \alpha_1$. (Because $\beta_1^{(m)}$ is real, all reorderings of these expressions are equivalent.)
The above theorem generalizes to any Lorentzian KM algebra, $\cA$, that is, one which is obtained from an affine algebra by the procedure of “over-extension”, attaching an additional node by one line only to the affine node in an affine Dynkin diagram. For any root $\alpha$ such that $(\alpha,\alpha)\neq 0$, define $\alpha^\vee = 2\alpha/(\alpha,\alpha)$. Then the Cartan integers which are the entries of the Cartan matrix of the rank $r+2$ algebra $\cA$ are given by $(\alpha_i,\alpha_j^\vee)$.
\[rankr1thm\] Let $\{\alpha_i\ | -1\leq i\leq r\}$ be the simple roots of a Lorentzian KM algebra, $\cA$, so $\alpha_1,\dots,\alpha_r$ are the simple roots of a finite dimensional simple Lie algebra, $\alpha_0$ is the affine simple root which generates an affine root system when included in the list, and $\alpha_{-1}$ satisfies $(\alpha_{-1},\alpha_{-1}) = 2$,$(\alpha_{-1},\alpha_0) = -1$, $(\alpha_{-1},\alpha_i) = 0$ for $1\leq i\leq r$. Write the affine null root $\delta = \sum_{j=0}^r n_j \alpha_j$ where $n_0 =1$. Fix any integer $m \geq 0$ and define $$\beta_0 = \alpha_{-1}, \quad
\beta_1 = \beta_1^{(m)} = m\delta + \alpha_1, \quad
\beta_j = \alpha_j \;\;\; {\rm for} \; 2\leq j\leq r.$$ Then the KM Lie subalgebra $\cA(\beta_0,\cdots,\beta_r)$ constructed in Theorem \[subrootsys\] has Cartan matrix $$C = [(\beta_i,\beta_j^\vee)] =
\bbm 2 & -m & 0 & \cdots & 0 \\ -m & 2 & * & \cdots & * \\
0 & * & 2 & \cdots & * \\ . & & & & \\ . & & & & \\ 0 & * & * & \cdots & 2
\ebm$$ which is rank $r+1$ of indefinite type for $m\geq 2$, where the submatrix obtained by removing the first row and column of $C$ is the finite type Cartan matrix $C_r^{fin}$ determined by $\alpha_1,\dots,\alpha_r$. We also denote this subalgebra by $\cA_{r+1}^{indef}(m)$.
As in Theorem \[rank2thm\], for $m \geq 0$, the roots $\beta_i$, $0\leq i\leq r$, satisfy the conditions of Theorem \[subrootsys\] and therefore determine a KM subalgebra of $\cA$ whose Cartan matrix is as shown because $(\delta,\delta) = 0$, $(\delta,\alpha_i) = 0$ for $0\leq i\leq r$, and $(\delta,\alpha_{-1}) = -1$. Note that the submatrix obtained by removing the first row and column of $C$ is the finite type Cartan matrix $C_r^{fin}$ determined by $\alpha_1,\dots,\alpha_r$. Let $C_{r-1}^{fin}$ be the finite type Cartan matrix obtained from $C_r^{fin}$ by deleting its first row and column. Then we see that $$\det(C) = 2\ \det(C_r^{fin}) - m^2\ \det(C_{r-1}^{fin}).$$ Using the table of values of these determinants on page 53 of [@Kac], we find that $\det(C) < 0$ in all cases for $m\geq 2$, guaranteeing that the subalgebra will be indefinite.
For example, if the finite dimensional algebra with simple roots $\alpha_1,\dots ,\alpha_r$ is of type $A_r$, then it is easy to see that $\det(C) = 2(r+1) - m^2 r$, which will be negative when $m^2 > 2(r+1)/r$. For $r\geq 2$, this is true for $m \geq 2$, and for $r = 1$ this is true for $m\geq 3$. The $r = 1$ case is rather extreme since we only have $\beta_0 = \alpha_{-1}$ and $\beta_1 = m\delta + \alpha_1$ giving the rank 2 Cartan matrix already studied in Theorem \[rank2thm\].
The above subalgebras $\cA_{r+1}^{indef}(m)$ of $\cA$ for different values of $m\geq 2$ have inequivalent Cartan matrices. Therefore, $\cA$ possesses infinitely many nonisomorphic indefinite KM subalgebras corresponding to the infinitely many ways of slicing the forward light-cone in root space by time-like hyperplanes such that the hyperboloidal intersections become more and more steeply inclined with increasing $m$. For rank 3 all subalgebras were again hyperbolic, but the indefinite KM algebras obtained in this way for higher rank in general are no longer hyperbolic even if the original subalgebra is hyperbolic.
The above construction can be further modified as follows to obtain subalgebras of rank $r+2$.
\[rankr2thm\] Let the notation be as in Theorem \[rankr1thm\]. Fix any integer $m \geq 1$ and define $$\gamma_{-1} = \alpha_{-1}, \quad
\gamma_0 = \gamma_0^{(m)} = (m-1)\delta + \alpha_0, \quad
\gamma_j = \alpha_j \;\;\; {\rm for} \; 1\leq j\leq r.$$ Then the KM Lie subalgebra $\cA(\gamma_{-1},\cdots,\gamma_r)$ constructed in Theorem \[subrootsys\] has Cartan matrix $$C = [(\beta_i,\beta_j^\vee)] =
\bbm 2 & -m & 0 & \cdots & 0 \\ -m & 2 & * & \cdots & * \\
0 & * & 2 & \cdots & * \\ . & & & & \\ . & & & & \\ 0 & * & * & \cdots & 2
\ebm$$ which is rank $r+2$ of indefinite type for $m\geq 1$, where the submatrix obtained by removing the first row and column of $C$ is the affine type Cartan matrix $C_{r+1}^{aff}$ determined by $\alpha_0,\dots,\alpha_r$. We also denote this subalgebra by $\cA_{r+2}^{indef}(m)$. The affine subalgebra with simple roots $\gamma_0,\cdots,\gamma_r$ has minimal null root $\sum_{j=0}^r n_j \gamma_j = m\delta$.
The proof is as in the previous theorem, but the submatrix obtained by removing the first row and column of $C$ is the affine type Cartan matrix $C_{r+1}^{aff}$ determined by $\alpha_0,\dots,\alpha_r$, and $C_r^{fin}$ is the finite type Cartan matrix obtained from $C_{r+1}^{aff}$ by deleting its first row and column, so $$\det(C) = 2\ \det(C_{r+1}^{aff}) - m^2\ \det(C_r^{fin}).$$ Since $\det(C_{r+1}^{aff}) = 0$, and $\det(C_r^{fin}) > 0$, we find that $\det(C) < 0$ in all cases for $m\geq 1$, guaranteeing that the subalgebra will be indefinite. Of course, the case when $m = 1$ just gives the original algebra $\cA$, so the only new content is for $m\geq 2$.
For the algebra $\cF$, and for $m \geq 2$, this procedure yields infinitely many inequivalent rank 3 subalgebras with the Cartan matrices $$\bbm 2 & -m & 0\\ -m & 2 & -2\\ 0 & -2 & 2 \ebm$$ not previously known to exist inside of $\cF$. In the next section we will explore the geometrical meaning of these subalgebras using the Weyl group of $\cF$.
Indefinite subalgebras of $\cF$ from the Weyl group
===================================================
A beautiful geometrical way to find subalgebras of $\cF$ is to examine the action of the Weyl group, $W$, which preserves each surface of fixed determinant. In fact, it preserves each sheet of the two sheeted hyperboloids with fixed positive integral determinant. Any one of these sheets can be taken as a model of the hyperbolic plane, isometric to the Poincaré disk, whose boundary corresponds to the light-cone. In Figure 1 we have shown the fixed points of the simple reflections which generate $W$, the triangular fundamental domain $\cD$ for $W$, and several other reflection lines corresponding to some other positive real roots of $\cF$ which are obtained by applying Weyl group elements to the simple roots. A complete tesselation of the disk would be obtained by including all reflection lines, which are in one-to-one correspondence with all of the positive real roots of $\cF$, $\Phi^+_{real}$. In that tesselation there would be a bijection between the elements of the Weyl group, $W$, and the images of the fundamental domain $\cD$. In Figure 1 we see $\cD$ is the triangle with angles $\pi/2$, $\pi/3$ and $0$, and sides labelled by the simple roots. Six images of $\cD$ form a larger triangle with all vertices on the boundary and all angles zero. The three sides of that larger triangle are labelled $\alpha_1$, $r_0 \alpha_1 = \alpha_1 + 2\alpha_0$ and $r_{-1} r_0 \alpha_1 = \alpha_1 + 2\alpha_0 + 2\alpha_{-1}$. Since that triangle has all angles zero, and contains 6 copies of $\cD$, it means that reflections with respect to these three roots generate a subgroup of $W$ of index 6 isomorphic to the hyperbolic triangle group $T(\infty,\infty,\infty)$. Using Theorem \[subrootsys\], those roots determine a rank 3 hyperbolic subalgebra of $\cF$, inequivalent to any of the algebras in the series found at the end of the last section.
More generally, for any $n\geq 1$, let $S = \{\beta_1,\cdots,\beta_n\}$ be a subset of $n$ roots from $\Phi^+_{real}$, corresponding to reflection lines in the Poincaré disk, such that $\beta_i - \beta_j\notin\Phi$. Then the KM subalgebra $\cF_S = \cF(\beta_1,\cdots,\beta_n)$ determined by Theorem \[subrootsys\] has Weyl group, $W_S$, generated by the reflections $r_{\beta_1},\cdots,r_{\beta_n}$. The Cartan matrix of $\cF_S$ is $C_S = [(\beta_i,\beta_j)]$ since all real roots of $\cF$ are of squared length $2$. The fundamental domain $\cD_S$ of $W_S$ is a union of images of $\cD$, and the number of such images will be equal to the index of $W_S$ in $W$. Using Figure 1 we can find choices of $S$ which give either finite or infinite index sub-Weyl groups. Our first example was mentioned above. We leave it to the reader to check the condition that $\beta_i - \beta_j\notin\Phi$ for each choice of $S$.
\[example1\] If $S$ consists of the roots $$\beta_1 = \alpha_1, \quad \beta_2 = r_0 \alpha_1, \quad
\beta_3 = r_{-1} r_0 \alpha_1,$$ then $\cF_S$ is a rank 3 subalgebra with Cartan matrix $$\bbm 2 & -2 & -2\\-2 & 2 & -2\\-2 & -2 & 2\ebm$$ and Weyl group $W_S = T(\infty,\infty,\infty)$ of index 6 in $W$.
There are other hyperbolic triangles with all angles zero in Figure 1, for example, the one with sides labelled $\alpha_1$, $\alpha_0$ and $r_{-1} r_0 r_1 r_0 \alpha_1 = 3\alpha_1 + 4\alpha_0 + 4\alpha_{-1}$. There are also 6 images of the fundamental triangle in this one, so these three reflections generate a subgroup of index 6 in $W$. The Cartan matrix generated from these three roots, taken as simple, is the same as the one above. Is there an automorphism of $\cF$ which interchanges these two isomorphic subalgebras? Are these two index 6 subgroups of $W$ conjugate?
\[example2\] If $S$ consists of the roots $$\beta_1 = r_1 r_0 \alpha_1 = 3\alpha_1 + 2\alpha_0,$$ $$\beta_2 = r_0 r_{-1} r_1 r_0 \alpha_1 = 3\alpha_1 + 6\alpha_0 + 2\alpha_{-1},$$ $$\beta_3 = r_{-1} r_0 r_1 r_0 \alpha_1 = 3\alpha_1 + 4\alpha_0 + 4\alpha_{-1},$$ then $\cF_S$ is a rank 3 subalgebra with Cartan matrix $$\bbm 2 & -10 & -10\\-10 & 2 & -10\\-10 & -10 & 2\ebm$$ and Weyl group $W_S = T(\infty,\infty,\infty)$ of infinite index in $W$ because there are no relations between the generating reflections and the area enclosed by the reflection lines contains an infinite number of copies of $\cD$.
\[example3\] If $S$ consists of the roots $$\beta_1 = \alpha_1,\qquad
\beta_2 = r_0 r_1 r_{-1} r_0 \alpha_1 = 3\alpha_1 + 6\alpha_0 + 2\alpha_{-1},$$ $$\beta_3 = r_0 r_{-1} r_0 r_1 r_0 \alpha_1 = 3\alpha_1 + 6\alpha_0 + 4\alpha_{-1},$$ then $\cF_S$ is a rank 3 subalgebra with Cartan matrix $$\bbm 2 & -6 & -6\\-6 & 2 & -2\\-6 & -2 & 2\ebm$$ and Weyl group $W_S = T(\infty,\infty,\infty)$ of infinite index in $W$ because there are no relations between the generating reflections and the area enclosed by the reflection lines contains an infinite number of copies of $\cD$.
\[example4\] If $S$ consists of the roots $$\beta_1 = r_0 \alpha_{-1} = \alpha_0 + \alpha_{-1},\qquad
\beta_2 = r_1 \alpha_0 = 2\alpha_1 + \alpha_0,$$ $$\beta_3 = r_0 r_1 r_0 \alpha_1 = 3\alpha_1 + 4\alpha_0,$$ then $\cF_S$ is a rank 3 subalgebra with Cartan matrix $$\bbm 2 & -3 & -2\\-3 & 2 & -2\\-2 & -2 & 2\ebm$$ and Weyl group $W_S = T(\infty,\infty,\infty)$ of infinite index in $W$ because there are no relations between the generating reflections and the area enclosed by the reflection lines contains an infinite number of copies of $\cD$.
\[example5\] If $S$ consists of the four roots $$\begin{array}{lll}
\beta_1 &= r_0 \alpha_{-1} &= \alpha_0 + \alpha_{-1}, \\
\beta_2 &= r_{-1} r_1 r_0 \alpha_1 &= 3\alpha_1 + 2\alpha_0 + 2\alpha_{-1},\\
\beta_3 &= r_1 r_0 \alpha_1 &= 3\alpha_1 + 2\alpha_0, \\
\beta_4 &= r_0 r_1 r_0 \alpha_1 &= 3\alpha_1 + 4\alpha_0
\end{array}$$ then $\cF_S$ is a rank 3 subalgebra with Cartan matrix $$A = [A_{ij}] = \bbm 2 & -2 & -4 & -2 \\ -2 & 2 & -2 & -10 \\
-4 & -2 & 2 & -2 \\ -2 & -10 & -2 & 2 \ebm$$ and Weyl group $$W_S = \la r_{\beta_1},r_{\beta_2},r_{\beta_3},r_{\beta_4}\ |\ r_{\beta_i}^2 = 1\ra.$$ >From Figure 1 we see that there are 12 fundamental triangles enclosed by these reflecting lines, so the index of $W_S$ in $W$ is 12.
In $\cF_S$, the four $H_i$ are linearly dependent in the three dimensional space $H$, as are the new simple roots $\beta_1,\cdots,\beta_4$. In fact, we have $2\beta_1 - \beta_2 + 2\beta_3 - \beta_4 = 0$. It is certainly possible for a Cartan matrix to be degenerate and still define a Kac-Moody algebra, but Theorem \[subrootsys\] gives the Serre relations because for $1\leq i\neq j\leq 4$, $\beta_i - \beta_j$ is not a root of $\cF$. There are several ways to solve the dependency problem ([@Kac; @Jur1; @Nie]), for example, by adjoining some derivations to the Cartan subalgebra so as to make the simple roots linearly independent, but then the resulting algebra will have Cartan subalgebra larger than the Cartan subalgebra of $\cF$. The same considerations occur when generalizing these ideas to the higher rank algebras $\cA$.
It would be interesting if one could use the geometry of the tesselated disk to classify the subgroups of the Weyl group of $\cF$ and use that to classify the subalgebras coming from Theorem \[subrootsys\].
We would like to finish this section by showing how the series of subalgebras of $\cF$ found at the end of the last section fit into the geometrical point of view given in this section. When Theorem \[rankr2thm\] is applied to $\cF$, for $m\geq 1$, the set $S$ is $$\gamma_{-1} = \alpha_{-1}, \quad
\gamma_0 = \gamma_0^{(m)} = (m-1)\delta + \alpha_0, \quad
\gamma_1 = \alpha_1$$ so the Weyl group $W_S$ is generated by two of the simple generators of $W$, $r_{-1}$ and $r_1$, along with one other reflection, $r_{\gamma_0^{(m)}}$. It is not hard to check that $$\gamma_0^{(1)} = \alpha_0,\ \gamma_0^{(2)} = r_0 \alpha_1,\
\gamma_0^{(3)} = r_0 r_1 \alpha_0,\
\gamma_0^{(4)} = r_0 r_1 r_0 \alpha_1,\ \cdots,$$ which can be seen in Figure 1. The case of $m = 1$ just gives $W$, but when $m = 2$ we see that the three reflecting lines enclose a fundamental domain $\cD_S$ containing three images of $\cD$, and the angles of the triangle are $\pi/2$, $0$ and $0$, so in that case $W_S = T(2,\infty,\infty)$ is of index 3 in $W$. But for $m\geq 3$, the three sides do not form a triangle, and there are infinitely many copies of $\cD$ in their fundamental domain, and the index of $W_S$ in $W$ is infinite. We would also like to mention that $$\bbm -1 & m \\ 0 & 1 \ebm$$ is the matrix in $PGL_2(\bZ)$ which represents the reflection $r_{\gamma_0^{(m)}}$ as discussed in section 2.
Borcherds algebras as subalgebras of hyperbolic KM algebras
===========================================================
Although one might have been surprised to find so many inequivalent indefinite KM subalgebras in any Lorentzian algebra $\cA$, it is perhaps even more surprising to find Borcherds algebras as proper subalgebras in $\cF$, in any Lorentzian algebra $\cA$, and even in rank 2 hyperbolics. As is well known, Borcherds (or generalized KM) algebras can be defined in terms of a generalized Cartan matrix and a set of generators and relations just like standard KM algebras, but are distinguished by the existence of [*imaginary simple roots*]{}. These correspond to zero or negative diagonal entries in the Cartan matrix, with corresponding modifications of the Serre relations [@B1]. (See also [@Jur1; @Nie].) Moreover, the multiplicity of a simple imaginary root may be greater than one.
To explain the basic idea let us return to Theorem \[rank2thm\]. As we have seen, for each $m$, the hyperbolic algebra $\cH(m)$ can be embedded into $\cF$ by identifying its two simple roots $\beta_0$ and $\beta_1$ in the root system of $\cF$, and the corresponding simple root generators as multiple commutators of the generators of $\cF$. Let us also choose the $\pm$ sign to be $-$ in the definition of $\beta_1$. The root system of $\cH(m)$ is contained in the linear subspace spanned by $\beta_0$ and $\beta_1$ inside the root lattice of $\cF$. As we already explained, this subspace gives a hyperbolic section of the light-cone in the root lattice of $\cF$. The root space $\cH(m)_\beta$ associated with any root $\beta$ in this hyperplane is contained in, but in general will not equal the root space $\cF_\beta$. Rather we will have strict inequality for “sufficiently imaginary” roots $\beta$, viz. $${\rm Mult}_{\cH(m)} (\beta) < {\rm Mult}_{\cF} (\beta).$$ (Actually the difference in dimension of these two root spaces will grow exponentially as $\beta$ is moved deeper into the light-cone.) As an example, let us take the rank 2 “Fibonacci” algebra $\cH(3)$ (so called because of its connection with the Fibonacci numbers [@F]), and in it the root $\beta = 2 \beta_0 + 2 \beta_1$. Then, from the table on page 214 of [@Kac], where this root is denoted by $(2,2)$, we have $${\rm Mult}_{\cH(3)} (\beta) = 1.$$ On the other hand, with the identification of Theorem \[rank2thm\], we have $\beta = 4 \alpha_1 + 6 \alpha_0 + 2 \alpha_{-1}$ as a root of $\cF$. The Weyl group reflection $r_0$ sends $\beta$ to the root $4 \alpha_1 + 4 \alpha_0 + 2 \alpha_{-1}$ which is denoted by $(4,4,2)$ in [@Kac], page 215, where the multiplicity of the root is given as $7$, so $${\rm Mult}_{\cF} (\beta) = 7,$$ showing that the root space $\cF_\beta$ contains six independent vectors not contained in $H(3)_\beta$.
We are thus led to define a new algebra inside $\cF$, as follows. Let $\Phi(\cH(m))$ be the set of roots of $\cF$ which are in the plane spanned by $\beta_0$ and $\beta_1$, so they are the same as the roots of $\cH(m)$. Let ${{\mathfrak{h}}}_m$ be the span of $H_0$ and $H_1$ from Theorems \[subrootsys\] and \[rank2thm\], that is, the Cartan subalgebra of $\cH(m)$. Then define the subspace $$\cG(m) = {{\mathfrak{h}}}_m \oplus \bigoplus_{\beta\in\Phi(\cH(m))} \cF_\beta$$ which is a proper Lie subalgebra of $\cF$ which contains $\cH(m)$ properly. The only subtle point involved in checking the closure of $\cG(m)$ under brackets is that $[\cF_\beta,\cF_{-\beta}]\subseteq {{\mathfrak{h}}}_m$, but this follows immediately from the invariance of the bilinear form on $\cF$.
We can think of $\cG(m)$ as an extension of $\cH(m)$ in the following way. For $\beta = a_0\beta_0 + a_1\beta_1 \in\Phi(\cH(m))^+$, $0\leq a_0,a_1\in\bZ$, define the $\cH(m)$-height $\cH t(\beta) = a_0 + a_1$. Note that this is not the same as the height of $\beta$ as a root of $\cF$. Define a sequence of extensions, $\cG^{(i)}(m)$, recursively, beginning with $\cG^{(1)}(m) = \cH(m)$. For $i>1$ let $\cG^{(i)}(m)$ be the Lie algebra containing $\cG^{(i-1)}(m)$ and the additional generators taken from the complement of $\cG^{(i-1)}(m)_\beta$ in $\cF_\beta$ for all $\beta\in\Phi(\cH(m))^+$ with $\cH t(\beta) = i$. This amounts to adding a new imaginary simple root for each such $\beta$ with simple multiplicity [@BGGN; @BGN] (not to be confused with the root multiplicity of $\beta$) $$\mu(\beta) = \dim(\cF_\beta) - \dim(\cG^{(i-1)}(m)_\beta).$$ Then, we have $$\cG(m) = \bigcup_{i=1}^\infty \cG^{(i)}(m) .$$ However, to work out the complete system of imaginary simple roots, their multiplicities, and the associated generalized Cartan matrix even for these simple examples would be an extremely difficult task. This would in particular require full knowledge of the root multiplicities of $\cF$. Fortunately, this is not necessary because we can invoke the following theorem of Borcherds [@B2] (also see [@Jur1]).
A Lie algebra ${{\mathfrak{g}}}$ is a Borcherds algebra if it has an almost positive definite contravariant form ${{\langle \ |\ \rangle}}$, which means that ${{\mathfrak{g}}}$ has the following properties:
1. (Grading) ${{\mathfrak{g}}}=\bigoplus_{n\in{{\mathbb Z}}}{{\mathfrak{g}}}_n$ with $\dim{{\mathfrak{g}}}_n<\infty$ for $n\ne0$;
2. (Involution) ${{\mathfrak{g}}}$ has an involution $\theta$ which acts as $-1$ on ${{\mathfrak{g}}}_0$ and maps ${{\mathfrak{g}}}_n$ to ${{\mathfrak{g}}}_{-n}$;
3. (Invariance) ${{\mathfrak{g}}}$ carries a symmetric invariant bilinear form ${{( \ |\ )}}$ preserved by $\theta$ and such that ${{( {{\mathfrak{g}}}_m | {{\mathfrak{g}}}_n )}}=0$ unless $m+n=0$;
4. (Positivity) The contravariant form ${{\langle x|y \rangle}}:=-{{( \theta(x)|y )}}$ is positive definite on ${{\mathfrak{g}}}_n$ if $n\neq0$.
Thus, we have
For all $m\geq 3$, $\cG(m)$ is a Borcherds algebra such that $$\cH(m) \subset \cG(m) \subset \cF.$$
All properties listed in Theorem \[thm-B1\] are satisfied for $\cG(m)$ because they are manifestly true for $\cF$.
In Theorem \[rankr2thm\] we have seen how any Lorentzian KM algebra $\cA$ of rank $r+2$ contains an infinite series of inequivalent indefinite KM subalgebras, $\cA_{r+1}^{indef}(m)$, for $m\geq 2$. Let $\Phi(\cA_{r+1}^{indef}(m))$ be the set of roots of $\cA$ which are in the hyperplane spanned by the simple roots $\beta_i$, $0\leq i\leq r$, of $\cA_{r+1}^{indef}(m)$, and let ${{\mathfrak{h}}}_m$ be its Cartan subalgebra. Then defining $$\cG(\cA_{r+1}^{indef}(m)) =
{{\mathfrak{h}}}_m \oplus \bigoplus_{\beta\in\Phi(\cA_{r+1}^{indef}(m))} \cA_\beta$$ gives a proper Lie subalgebra of $\cA$ generalizing the previous construction.
For all $m\geq 2$, $\cG(\cA_{r+1}^{indef}(m))$ is a Borcherds algebra such that $$\cA_{r+1}^{indef}(m) \subset \cG(\cA_{r+1}^{indef}(m)) \subset \cA.$$
Let’s denote more briefly by $\cG$ any of the “hyperplane Borcherds subalgebras" just constructed inside $\cA$, and let $\cH$ denote the indefinite KM subalgebra properly contained in $\cG$. Then we have the following decomposition $$\cG = \cM_- \oplus \cH \oplus \cM_+$$ where $\cM_+$ and $\cM_-$ are supported on positive and negative roots, respectively. This decomposition corresponds to a similar decomposition found in [@Jur2] for all Borcherds algebras, $$\tilde\cG = \cM_- \oplus (\cH \oplus \tilde{{\mathfrak{h}}}) \oplus \cM_+ ,$$ where $\tilde{{\mathfrak{h}}}$ is an infinite dimensional extension of the Cartan subalgebra of the KM algebra $\cH$ which makes all the imaginary simple roots linearly independent. This extension is analogous to the extension of the Cartan subalgebra mentioned at the end of the previous section. But this extension would not be contained in $\cA$. As shown in [@Jur2], $\cM_+$ and $\cM_-$ are free Lie algebras. It would be interesting to determine their structure as $\cH$-modules. Similar structures were studied in [@BGGN].
Finally, we would like to note that there are Borcherds algebras inside the rank 2 hyperbolics, and in particular, in $\cH(3)$, whose positive roots are shown in Figure 2. Note that the positive real roots are shown by open circles and the positive imaginary roots by solid dots. The figure also shows the simple reflection lines and root multiplicities. We draw the reader’s attention to the central vertical line in the figure, and define the subalgebra $\cG$ which is the direct sum of all the root spaces along that line, including the negative root spaces not shown and the one-dimensional subspace of the Cartan subalgebra spanned by $h_1 + h_2$. This is the simplest example of a Borcherds algebra embedded inside a hyperbolic KM algebra. In this case we have the decomposition $$\cG = \cM_- \oplus sl_2 \oplus \cM_+$$ and it would not be hard to determine the number of free generators in the root spaces of $\cM_+$ by using the formulas in [@Jur2] for the dimensions of graded subspaces in free Lie algebras with graded generators.
Appendices
==========
Figure 1: Poincaré Disk Model of Hyperbolic Plane Tesselated
By the Hyperbolic Triangle Group $T(2,3,\infty)$
Figure 2: Hyperbolic Root System For The Fibonacci Algebra $\cH(3)$
[BKM1]{}
R. E. Borcherds [*Generalized Kac-Moody algebras*]{}, J. Algebra 115 (1988), 501–512
R. E. Borcherds [*A characterization of generalized Kac-Moody algebras*]{}, J. Algebra 174 (1995), 1073–1079.
O. Bärwald, R.W. Gebert, M. Günaydin, H. Nicolai [*Missing modules, the Gnome Lie Algebra, and $E_{10}$*]{}, Commun. Math. Phys. 195 (1998), 29–65
O. Bärwald, R.W. Gebert, H. Nicolai [*On the imaginary simple roots of the Borcherds algebra ${{\mathfrak{g}}}_{II_{9,1}}$*]{}, Nucl. Phys. B [**510**]{} \[PM\] (1998), 721–738.
G. M. Benkart, S.-J. Kang, K. C. Misra, [*Graded Lie algebras of Kac-Moody type*]{}, Adv. Math. 97 (1993), 154–190.
G. M. Benkart, S.-J. Kang, K. C. Misra, [*Indefinite Kac-Moody algebras of special linear type*]{}, Pacific J. Math.
G. M. Benkart, S.-J. Kang, K. C. Misra, [*Indefinite Kac-Moody algebras of classical type*]{}, Adv. Math. 105 (1994), 76–110.
T. Damour, M. Henneaux and H. Nicolai, [*Cosmological Billiards*]{}, hep-th/0212256
A. J. Feingold, [*A hyperbolic GCM Lie algebra and the Fibonacci numbers*]{}, Proc. Amer. Math. Soc. [**80**]{}, (1980), 379–385.
A. J. Feingold, I. B. Frenkel, [*A hyperbolic Kac-Moody Algebra and the theory of Siegel modular forms of genus 2*]{}, Math. Ann. [**263**]{} (1983), 87–144.
A. J. Feingold, I. B. Frenkel, J. F. X. Ries, [*Spinor Construction of Vertex Operator Algebras, Triality and $E_8^{(1)}$*]{}, Contemp. Math. [**121**]{}, Amer. Math. Soc., Providence, RI, 1991.
A. J. Feingold, J. Lepowsky, [*The Weyl-Kac character formula and power series identities*]{}, Advances in Mathematics [**29**]{} (1978), 271-309.
I. B. Frenkel, J. Lepowsky, A. Meurman, [*Vertex Operator Algebras and the Monster*]{}, Pure and Applied Math., 134, Academic Press, Boston, 1988.
I. B. Frenkel, Y. Zhu, [*Vertex operator algebras associated to representations of affine and Virasoro algebras*]{}, Duke Math. J. [**66** ]{} (1992), 123–168.
R. W. Gebert, H. Nicolai, [*On $E_{10}$ and the DDF construction*]{}, Commun. Math. Phys. [**172**]{} (1995), 571–622.
P. Goddard, D. I. Olive, [*Kac-Moody and Virasoro algebras in relation to quantum physics*]{}, Internat. J. Mod. Phys. A, Vol. 1, No. 2 (1986), 303–414.
E. Jurisich, [*An exposition of Generalized Kac-Moody algebras*]{}, Contemporary Math [**194**]{} (1996).
E. Jurisich, [*Generalized Kac-Moody Lie algebras, free Lie algebras and the structure of the Monster Lie algebra*]{}, J. Pure Appl. Algebra [**126**]{} (1998), 233–266.
V. G. Kac, [*Infinite Dimensional Lie Algebras*]{}, Third Edition, Cambridge University Press, 1990.
V. G. Kac, R. V. Moody, M. Wakimoto, [*On $E_{10}$*]{}, Differential Geometrical Methods in Theoretical Physics, Proceedings, NATO Advanced Research Workshop, 16th International Conference, Como, Amsterdam, Kluwer, 1988, Editors: K. Bleuler, M. Werner, pp. 109–128.
V. G. Kac, A. K. Raina, [*Bombay Lectures on Highest Weight Representations*]{}, World Scientific, Singapore, 1987.
Seok-Jin Kang, [*Gradations and structure of Kac-Moody Lie algebras*]{}, dissertation, Yale Univeristy, 1990.
Seok-Jin Kang, [*Kac-Moody Lie algebras, spectral sequences, and the Witt formula*]{}, Trans. Amer. Math. Soc. 339 (1993), 463–495.
Seok-Jin Kang, [*On the hyperbolic Kac-Moody Lie algebra $HA_1^{(1)}$*]{}, Trans. Amer. Math. Soc. 341 (1994), 623–638.
Seok-Jin Kang, [*Root multiplicities of the hyperbolic Kac-Moody Lie algebra $HA_1^{(1)}$*]{}, J. Algebra 160 (1993), 492–523.
Seok-Jin Kang, [*Root multiplicities of Kac-Moody algebras*]{}, Duke Math. J. 74 (1994), 645–666.
Seok-Jin Kang, D. J. Melville, [*Root multiplicities of the Kac-Moody algebras $HA_n^{(1)}$*]{}, J. Algebra 170 (1994), 277–299.
Seok-Jin Kang, D. J. Melville, [*Rank 2 symmetric hyperbolic Kac-Moody algebras*]{}, Nagoya Math. J. 140 (1995), 41–75.
J. Lepowsky, R. V. Moody, [*Hyperbolic Lie Algebras and quasi-regular cusps on Hilbert modular surfaces*]{}, Math. Ann. [**245**]{} (1979), 63–88.
P. Niemann, [*Some Generalized Kac-Moody Algebras with Known Root Multiplicities*]{}, Memoirs of the AMS, Vol. 157, No. 746, Amer. Math. Soc., Providence, RI, 2002.
A. Tsuchiya, Y. Kanie, [*Vertex operators in conformal field theory on ${\mathbb P}^1$ and monodromy representations of braid group*]{}, Conformal Field Theory and Solvable Lattice Models, Adv. Studies in Pure Math. [**16**]{}, Academic Press, New York, 1988, pp. 297–372.
Weiqiang Wang, [*Rationality of Virasoro vertex operator algebras*]{}, Duke Math. J. IMRN, Vol. 71, No. 1 (1993), 197–211.
[^1]: AJF wishes to thank the Max-Planck-Institut für Gravitationsphysik, Albert-Einstein-Institut, Potsdam, Germany, for support during two stimulating visits.
[^2]: The level decomposition with respect to the coefficient $a_1$ has been used in [@DHN], and is known up to level $a_1 = 56$ (T. Fischbacher, private communication).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'A variety of environmental media were analyzed for fallout radionuclides resulting from the Fukushima nuclear accident by the Low Background Facility (LBF) at the Lawrence Berkeley National Laboratory (LBNL) in Berkeley, CA. Monitoring activities in air and rainwater began soon after the onset of the March 11, 2011 tsunami and are reported here through the end of 2012. Observed fallout isotopes include $^{131}$I, $^{132}$I,$^{132}$Te,$^{134}$Cs, $^{136}$Cs, and $^{137}$Cs. Isotopes were measured on environmental air filters, automobile filters, and in rainwater. An additional analysis of rainwater in search of $^{90}$Sr is also presented. Last, a series of food measurements conducted in September of 2013 are included due to extended media concerns of $^{134, 137}$Cs in fish. Similar measurements of fallout from the Chernobyl disaster at LBNL, previously unpublished publicly, are also presented here as a comparison with the Fukushima incident. All measurements presented also include natural radionuclides found in the environment to provide a basis for comparison.'
address:
- 'Nuclear Science Division, Lawrence Berkeley National Laboratory, Berkeley, CA 94720'
- 'Department of Nuclear Engineering, University of California, Berkeley, CA 94720'
- 'Department of Physics, University of São Paulo, São Paulo- SP 05314970 , Brazil'
author:
- 'A.R. Smith'
- 'K.J. Thomas'
- 'E.B. Norman'
- 'D.L. Hurley'
- 'B.T. Lo'
- 'Y.D. Chan'
- 'P.V. Guillaumon'
- 'B.G. Harvey'
bibliography:
- 'bibtex.bib'
title: 'Measurements of Fission Products from the Fukushima Daiichi Incident in San Francisco Bay Area Air Filters, Automobile Filters, Rainwater, and Food'
---
Fukushima Daiichi Nuclear Power Plant, Fukushima, fallout, air monitoring, rainwater, automobile filters, Chernobyl
Introduction
============
Since the early 1980’s, the Low Background Facility (LBF) at Lawrence Berkeley National Laboratory (LBNL) has been analyzing laboratory environmental air sampler filters for the presence of any gamma-emitters, such as naturally occurring, $^{7}$Be and $^{210}$Pb. After the announcement of the accident at the Fukushima Daiichi Nuclear Power Plant, detailed monitoring began to search for fallout isotopes in environmental air filters at the local LBF station at LBNL and in rainwater in Oroville, CA. Several other groups have reported similar Fukushima-related measurements in air filters and rainwater in the U.S. [@Norman2011; @Bandstra2011; @Leon2011; @MacMullin2012; @Biegalski2012]. The fallout also provided a demonstration for a pilot program in monitoring fallout isotopes in automobile filters, which are also continuously monitored by the LBF. An additional analysis was later performed upon rainwater collected in the spring of 2011 in an effort to search for the presence of $^{90}$Sr in the San Francisco Bay Area. Authors involved in this study, A.R. Smith and E.B Norman, also performed similar measurements in California in the aftermath of the Chernobyl accident in 1986 which were never published formally, and are presented here in comparison to those on Fukushima. Last, due to extended reports and local concerns of contaminated water being leaked in 2013 into the ocean at Fukushima, a series of food measurements were performed in September of 2013 to search for fission products in fish and other food items from the Pacific region purchased at local Bay Area retail locations.
The Low Background Facility at Lawrence Berkeley National Laboratory
--------------------------------------------------------------------
The LBF at LBNL typically performs a variety of low background gamma spectroscopy services to a variety of experiments and end users. The LBF operates in two unique, low-background laboratory spaces: a local surface laboratory constructed of low-activity concrete; and an underground site with over 500 m.w.e (meters water-equivalent) overburden for shielding against cosmic ray muons. A primary component of its activities include low background counting of candidate construction materials for ultralow background experiments such as those searching for dark matter, neutrinos, and neutrinoless double beta decay – which often have strict requirements for radiopurity. Of primary concern in these experiments are the natural primordial radioisotopes and their decay chains ($^{238,235}$U,$^{232}$Th, $^{40}$K), common man-added radioactivity ($^{60}$Co, $^{137}$Cs, etc.), and various cosmogenic radioisotopes. Other services include neutron activation analysis for trace analysis of many stable isotopes, neutron flux/transmutation doping, radon emanation, R&D hosting, environmental measurements, and other various applications such as accelerator characterization and radiological waste assay. More information on the Low Background Facility at LBNL can be found in [@Thomas2013] and [@ThomasLRT].
Air Sampling
============
An air sampler was put into operation outside LBNL LBF on 3/14/2011, to track the arrival and continued presence at LBNL of gamma-emitting fission products from the Fukushima Daiichi nuclear power plant disaster that was caused by the earthquake/tsunami on 3/11/2011. This sampling station has continued operation to the present and results through the end of 2012 are summarized here.
Air sampling began on the morning of 3/14/11, using a nominal 24-hr collection period, followed by immediate counting on an ORTEC low-background packaged HPGe detector (2.26 kg n-type HPGe crystal, relative efficiency of 115%). Air flow through the sampler is actively regulated at a constant flow rate, producing 244.7 m$^{3}$ throughput in a 24 hour collection period. Aerosols are collected upon a 10.16 cm diameter HEPA filter rated for 0.3 $\mu$m aerosol sizes and counted directly atop the top face of the detector. For the sake of these measurements, it was assumed that essentially all fission products in the air are attached to aerosols greater than this size. All filters analyzed with the n-type HPGe detector showed statistically significant peaks from the naturally produced airborne gamma-emitters, such as $^{7}$Be and $^{210}$Pb. Initial counting (within an hour after end of sampling) of filters also showed prominent peaks from the short-lived daughters of U-series $^{222}$Rn and Th-series $^{220}$Rn. Analysis of gamma ray peaks from short-lived Rn daughters could provide useful information with respect to collection of fission product nuclides, especially for short collection periods. The analysis procedure included several short counting intervals to quantify Rn daughter activities, followed by an overnight count to increase the possibility of detecting very small peaks, as may be expected from the first signs of fission product radionuclides from the damaged Japanese reactors. Filter exchanges were initially performed in 24 hour cycles. Sampling intervals were later extended to 2 or more days, one week, and eventually to one month in order to continue to produce actual values for the decreasing activities of gamma-emitters rather than only upper limits. Through this strategy, activities over a range of 1000-fold per 24 hour collection period were documented for gamma-emitters from which the maximum observed count rate in diagnostic peaks were as low as 1 count per minute. Throughout the monitoring period, observed fallout isotopes include $^{131}$I (t$_{1/2}$ = 8.0 days), $^{132}$I (t$_{1/2}$ = 2.3 hours) ,$^{132}$Te (t$_{1/2}$ = 3.2 days),$^{134}$Cs (t$_{1/2}$ = 2.06 years), $^{136}$Cs (t$_{1/2}$ = 13.16 days) and $^{137}$Cs (t$_{1/2}$ = 30.07 years). A sample gamma spectrum from such an air filter is displayed in [**Figure \[fig:airfilter\]**]{}.
The overnight count of the filter collected on March 15-16, 2011 showed the initial presence of a very small peak at 364 keV, the most intense gamma emitted in the decay of $^{131}$I, but without any companion fission products could not provide unequivocal evidence to support official arrival. When alone, the $^{131}$I could have originated from a nearby source, such as a local hospital. The following day provided the full confirmation of fission products from Fukushima from a filter exposed March 16-17, in which $^{132}$Te and $^{131}$I were both observed. The filter collected during March 17-18, 2011 showed the highest activities observed for $^{131}$I and $^{132}$Te, at 14.3$\pm$0.1 mBq m$^{-3}$ and 20.9$\pm$0.1 mBq m$^{-3}$ respectively, as well as the first peak in their temporal distributions. The filter collected from March 23-24 exhibited the second highest $^{131}$I and $^{132}$Te activities, at 12.5$\pm$0.1 mBq m$^{-3}$ and 1.44$\pm$0.03 mBq m$^{-3}$ respectively, and may represent the Òsecond lapÓ of the most radioactive plume as it passed by California again. Note all uncertainties listed here are statistical only, and a $\sim$10% systematic uncertainty should be assumed as well. The relative temporal and absolute airborne activities based upon the air filter measurements are shown for the full duration of the LBF’s Fukushima fallout monitoring activities through the end of 2012 in [**Figure \[fig:allfalloutfull\]**]{}. Not plotted are $^{136}$Cs, since it only was present on a small amount of filters, and $^{132}$I, since it is the short-lived daughter of $^{132}$Te. The activities of $^{210}$Pb and $^{7}$Be, both naturally occurring radioisotopes found in the air, are shown in [**Figure \[fig:pbbe\]**]{}. The natural isotopes can be used as a comparison to the Fukushima fallout in [**Figure \[fig:allfalloutfull\]**]{} to demonstrate that the observed fission radioisotope activities at the LBF station were, briefly at their initial arrival, comparable to natural radioisotopes that are always present, but fell much below natural background activities quickly. In addition, it is also worth noting that only $^{210}$Pb and $^{7}$Be were plotted, and there are many other $^{222}$Rn daughters such as $^{214}$Bi not shown in [**Figure \[fig:pbbe\]**]{} since most $^{222}$Rn daughters are quite short lived. The measurements of airborne fallout activities presented here are also in reasonable agreement with the results of another local group at UC Berkeley who made measurements in the area [@Bandstra2011].
![A HPGe gamma ray spectrum from an air filter exposed March 23-24, 2011 at the LBF facility at LBNL containing both natural and fission nuclides. This spectrum represents a full 24 hour exposure period and was counted for a livetime of 1250 minutes immediately upon being removed from the air sampler. More information on the peaks displayed here are listed in [**Table \[tab:isotopes\]**]{}.[]{data-label="fig:airfilter"}](26099-AirFilter){width="\linewidth"}
![Long term airborne concentrations fission products from March 11, 2011 to the end of 2012 as measured on HEPA filters at the LBF in Berkeley, CA. Horizontal error bars, when visible, represent filter exposure periods. Missing error bars are smaller than the visible data marker. Plotted uncertainties seen here are statistical only, and a conservative systematic uncertainty of $\sim$10% should also be assumed.[]{data-label="fig:allfalloutfull"}](AllFalloutFullloglog){width="\linewidth"}
Automobile Air Filters
----------------------
Since 2002, the Low Background Facility has been analyzing automobile filters obtained from the local Berkeley Police Department to perform regular analysis as a pilot program for detecting radioisotopes with potential homeland security applications. Over 1200 automobile filters have been counted over the course of the program. The release of fission products from Fukushima have allowed for a ‘proof-of-principle’ for this radioisotope monitoring technique. Although they do not have the collection efficiency of HEPA filters (automobile filters collect at approximately one-third the efficiency as the 0.3 $\mu$m filters used in the previous section), they can still produce reliable airborne measurements since their airflow can even be estimated based upon the odometer reading and fuel consumption of a vehicle. However, the use of automobile filters perhaps provides more important qualitative advantages. First, with the use of public sector vehicles that run somewhat regular patrol routes, there is a low-cost network of filters *already deployed* and screening essentially every city. With regular screening of filters removed from vehicles at normal maintenance exchanges, any unusual radioactivity found in an air filter could trigger ‘first-alert’ procedures that would call for further investigation. After determining the true presence of an isotope of concern, routes from logbooks can be determined and teams could be scrambled somewhat quickly to perform more thorough sweeps of areas looking for potential sources of radioactivity simply by driving vehicles and quickly analyzing filters upon return, or perhaps with more sophisticated portable filtration and detection systems. Second, in the event of a large-scale disaster, networks of automobiles could be deployed to determine the level and extent of contamination across large regions and could be analyzed via a large network of existing detectors at laboratories and universities nationally.
The Low Background Facility at LBNL obtains automobile filters from the Berkeley Police Department every few weeks. At regular maintenance intervals, air filters are removed and (ideally) the odometer readings are recorded. In the LBF pilot program, the filters are initially screened at the vehicle maintenance shop using a First Response Detector (FRD) composed of a 2x2 NaI detector within a 2 inch thick steel shield. Here, only gross counts of filters are noted. If any filters were to exhibit gross counting rates above background, such filters would take priority in screening. In this study no filters have ever registered above the normal variation of background readings on the FRD– even those with radioisotopes from Fukushima. Once returned to the LBF, the filters are counted on an HPGe system in which isotopes are identified and quantified. In a full-scale operation, one could imagine placing many FRD systems at maintenance shops around the country, in which mechanics could test the filter and could flag unusual filters for additional laboratory study.
Along with a similar program operating in Butte County at the Oroville location of the LBNL LBF, more than 1500 auto filters have been evaluated. Prior to the Fukushima disaster, the only man-added radionuclide ever observed on any filters was the rare occurrence of very small amounts of $^{137}$Cs, a relic of atmospheric nuclear weapons testing in the mid 20th century, via re-suspension of surface soil particles. No filter has ever triggered above the background threshold upon the FRD detector, even with the added Fukushima fallout. Nevertheless, the Fukushima Daiichi release of fission products provided a satisfactory test and proof-of-principle for using automobile filters as an early detection system. Although the fallout in the San Francisco Bay Area was comparable to other natural background radioactivity found in the air (such as the $^{214}$Bi, $^{210}$Pb, $^{7}$Be, etc.), several gamma ray peaks were very easily identifiable on auto filters as seen in [**Figure \[fig:autofilter\]**]{}. A plot showing the counting rates of $^{134}$Cs, $^{137}$Cs, $^{131}$I, and $^{132}$Te from the onset of the radiation release in March 2011 through December 2012 is shown in [**Figure \[fig:auto\]**]{}, with corresponding counting rates from natural radioisotopes provided in [**Figure \[fig:autonatural\]**]{}. The counting rates for automobile filters can be converted to true airborne activities using geometrical efficiency calibrations, airflow data, and collection efficiency. However, the data presented here is intended as merely a qualitative demonstration, since mileage information for the full set of filters was unavailable. By comparison to the LBF HEPA filter data in [**Figure \[fig:allfalloutfull\]**]{}, a rough idea of the sensitivity can be established. After approximately the 100 day mark from 3/11/2011, the $^{134}$Cs and $^{137}$Cs activities in [**Figure \[fig:allfalloutfull\]**]{} from HEPA filters dropped below 10$\mu$Bq m$^{-3}$ and as seen in [**Figure \[fig:auto\]**]{}, the same isotopes still had quantifiable counting rates above background in the automobile filters. A second demonstration of detection appeared in a set of filters removed in early September of 2012 in [**Figure \[fig:auto\]**]{} where $^{131}$I reappeared, likely released as a result of a local medical procedure. This demonstrates the ability of this pilot program to detect even small amounts of radioactivity. A stationary monitor, say in the center of a city on a tall building, may not see such a release as a result of dilution and distance from the source. However, in this scenario, a police cruiser was likely near the hospital in which this isotope was released as part of a medical treatment. However, it should be noted, short lived isotopes including $^{131}$I (t$_{1/2}=$ 8 days) could be difficult to detect using automobile filters, since they can decay away before regular maintenance may occur to replace filters in vehicles.
![A gamma ray spectrum from April 7, 2011 obtained from counting an automobile filter from a Berkeley Police Department patrol vehicle. Fission products were easily identifiable in the aftermath of the radiation release. More information on the peaks displayed here are listed in [**Table \[tab:isotopes\]**]{}.[]{data-label="fig:autofilter"}](26165S-AutoFilter){width="\linewidth"}
![Counting rates for fission products from Fukushima as seen on automobile filters in Berkeley, CA. The day plotted for filters represents the date after 3/11/11 when the filter was removed from a vehicle. Counting rates were not converted to airborne activity, as the full mileage (and hence the airflow) information was not available for the full data set, so the information is presented as a qualitative demonstration of the use of automobile filters for radionuclide monitoring.[]{data-label="fig:auto"}](autoallfallout-loglin){width="\linewidth"}
Rainwater Measurements
======================
Also collected in the aftermath of the Fukushima announcement was rainwater in Oroville, CA. Collection periods were recorded and samples were transferred from an outdoor collector to a Marinelli-style counting beaker for counting. No other preparation was performed on the samples prior to being counted on an ORTEC HPGe spectrometer (2.1 kg p-type HPGe crystal, 85% relative efficiency) at the underground location of the Low Background Facility with over 500 m.w.e. overburden. When applicable, results were decay corrected for the shorter lived isotopes that for decay prior to counting in order to report the activity at the time of collection. The rainfall measurements collected in the immediate days following the Fukushima nuclear accident are shown in [**Figure \[fig:raincompare\]**]{} and the full period of monitoring in [**Figure \[fig:ororainloglog\]**]{}. The activities found in the rainwater in Oroville, CA were remarkably close in activity to the measurements made by Norman, et. al. [@Norman2011] on rainwater collected in the San Francisco Bay Area, which is around 120 miles away– which is quite interesting, and are included in [**Figure \[fig:raincompare\]**]{} for comparison. This suggests that the radioactivity was spread uniformly in the precipitation as the storm front passed through northern California over the course of several hours.
![Comparison of rainwater samples collected during the same time period at two locations approximately 120 miles apart. The hollow points represent rainwater collected in the east San Francisco Bay Area by Norman et. al. and replotted from [@Norman2011]. The solid points represent rainwater collected and analyzed by the LBF in Oroville, CA as seen in [**Figures \[fig:raincompare\]**]{} and [**\[fig:ororainloglog\]**]{}. The data shows remarkably similar activities in both locations, despite their distance form one another and time between rainstorms passing through each.[]{data-label="fig:raincompare"}](ORO-EastBay-Compare){width="\linewidth"}
The Search for Strontium-90
===========================
The rainwater samples in Norman, et al. [@Norman2011] contained the man-made isotopes $^{131}$I, $^{132}$I,$^{132}$Te,$^{134}$Cs,$^{136}$Cs, and $^{137}$Cs. They were collected between March 16-26, 2011 in Oakland, Albany, and Berkeley, California on the eastern side of the San Francisco Bay. Samples were placed directly into Marinelli beakers for gamma ray counting, with no additional preparation prior to gamma ray counting on a 60% relative efficient HPGe gamma spectrometer. An efficiency calibration utilizing natural La, Lu, and K isotopes was used for analysis [@Perillo1996]. The samples were left sealed and archived in the laboratory after analysis, where they remained until the summer of 2012. The highest activity found in these samples was that of $^{131}$I which peaked at 16 Bq/L on 3/24/2011. This period of rainfall also exhibited the highest activities of the other detected isotopes.
These rainwater samples presented in Norman, et al. [@Norman2011] were reexamined during the summer of 2012 for the presence of $^{90}$Sr. Isotopes of strontium, in particular, are a health hazard as they are readily absorbed into bones due to its chemical similarity to calcium. Most of the fallout from Fukushima is easily observed due to the presence of high energy gamma rays which make detection via HPGe gamma spectroscopy trivial for most systems, especially those optimized for low background counting. The fission product $^{90}$Sr (t$_{1/2}$ = 28.8 years) however is much more difficult to detect. It decays via $\beta$- with Q$_{\beta}$ = 546 keV – and *no subsequent emission of gammas*. The low Q$_{\beta}$ and no gamma ray emission makes for a much trickier isotope to detect. However, the daughter, $^{90}$Y (t$_{1/2}$ = 2.67 days), also decays via $\beta$- with very weak emission of gamma rays, but has a much higher end point energy of Q$_{\beta}$ = 2.280 MeV which makes detection easier.
The method used to isolate $^{90}$Sr involved introducing stable strontium to the rainwater samples and creating a precipitate, which would act as a carrier to also extract any $^{90}$Sr that may be present. Four rainwater samples, each approximately 1L in volume, collected between March 18-23 in Albany, Oakland, and Berkeley CA were selected for $^{90}$Sr testing. A control sample of tap water was also analyzed. Approximately 100-140 mg of SrCl$_{2}$ was dissolved in each sample followed by 200-250 mg of K$_{2}$CO$_{3}$ , which immediately turns the water sample a cloudy white color due to the formation of a SrCO$_{3}$ precipitate. The precipitate was then filtered out of solution onto a paper filter and dried, which was weighed before and after to obtain a precipitate mass. For every 1 mg of SrCl$_{2}$, it is expected to produce 0.931 mg of SrCO$_{3}$. The experimental conversion we achieved averaged 91.61% of the expected precipitate mass for the five samples. The filter and precipitate was then laminated in a strip of packing tape. X-ray fluorescence was performed on the samples to confirm that the precipitate samples did indeed contain strontium.
If $^{90}$Sr were present in the rainwater samples then the chemistry performed to produce a Sr precipitate would break the secular equilibrium between the decays of $^{90}$Sr (t$_{1/2}$ = 28.8 years) and $^{90}$Y (t$_{1/2}$ = 2.67 days).The samples were immediately counted after preparation and again one week later on a planar HPGe detector equipped with a thin beryllium entrance window (0.254 mm), which allows for transmission of low energy gamma and x-rays. The efficiency of a thin planar detector drops off rapidly for higher energy gammas ($>$100 keV). Betas however, maintain a constant efficiency at these higher energies since they are absorbed over a short distance within the HPGe crystal, so a $\beta$ particle above an MeV would have a far greater efficiency compared to gammas in that energy region. The dimensions of the planar HPGe used in these measurements was 32 mm in diameter and 13 mm thick. We then compared a ratio of the counts in the 0-0.546 MeV energy interval (the $^{90}$Sr endpoint) to that in the 0.546-2 MeV interval (approximate $^{90}$Y endpoint) immediately after filtering and again after one week to search relative changes between the two energy regions, which would be expected if $^{90}$Sr were present and feeding into $^{90}$Y decays. The immediate and one-week-old counting ratios are shown in [**Figure \[fig:ratiocomparison\]**]{}, which did not yield any indication for the presence $^{90}$Sr via $^{90}$Y growth. The four samples were then stacked on top of each other for a week long counting to set limits, based upon $^{90}$Y $\beta$’s, calculated using Equation \[eq\], where N is the number of counts in the background, $\epsilon$ is the beta detection efficiency determined using a calibrated $^{90}$Sr source, V is the volume of rainwater (3.35 L) from which the precipitate was extracted, and LT is the livetime of counting. This produced a one sigma upper limit of 8.98 mBq/L for the presence of $^{90}$Sr in the rainwater samples analyzed.
$${ A }_{ ul }\quad =\quad \frac { \sqrt { 2N } }{ \epsilon \cdot V\cdot LT }
\label{eq}$$
Isotope Activity (Bq/L) Fission Yield (%) [@englandrider] Volatility [@Sill198897; @langowski]
------------ ----------------------------- ----------------------------------- --------------------------------------
$^{90}$Sr $<$ 8.99 $\times$ 10$^{-3}$ 5.78 Low
$^{137}$Cs 0.136 $\pm$ 0.041 6.19 High
: Volatility comparisons of $^{90}$Sr and $^{137}$Cs, alongside their approximate activities or limits set for their presences in rainwater from this work and [@Norman2011]. These details are in support of $^{90}$Sr not being a major concern as a transported isotope to the Bay Area.
\[tab:volatility\]
Soil and Sediment Samples
=========================
Soil samples have been collected by LBNL personnel and analyzed at the LBF since the late 1950s. The chosen analytic technique has been gamma ray spectroscopy, first using NaI(Tl) scintillation crystal detectors, but switching to high resolution HPGe semiconductor detectors in the 1980s when they became available with relative efficiencies of at least 30%. Initially, the motivation was to document the areal distribution in surface and near-surface soils of fission product radionuclides from atmospheric nuclear weapons testing. Later applications included determination of the natural terrestrial radionuclides (U,Th,K), and determination of the radon emanation (principally the $^{238}$U series $^{222}$Rn) from these materials. Analyses continue in all of the above applications.
The current study is designed to provide information for identifying and characterizing the radionuclides transported to the Berkeley Lab from the nuclear power plant disaster at Fukushima, Japan in the aftermath of the 3/11/2011 earthquake/tsunami. Two pairs of on-site soil samples provide some relevant information, as listed in [**Table \[tab:soil\]**]{}. The first pair represents surface soils from a site near LBNL building 90 and a site near LBNL building 72, at which the surfaces had not been disturbed since before the onset of atmospheric nuclear weapons testing that began in the mid-1940s. The second pair were sampled in late April 2011; one from the undisturbed site near LBNL building 90, and the other from the face of a shallow cut in deeply weathered Orinda Formation bedrock along one side of the parking lot at building 72, wherein the present surface has only been exposed since the late 1970s during a major expansion of the building.
The absence of $^{134}$Cs in the 1998 samples identifies the activity of $^{137}$Cs remaining from mid 20$^{th}$ century atmospheric nuclear weapons tests. The results presented in the air monitoring results of this report indicate approximately equal activities of $^{134}$Cs and $^{137}$Cs were deposited at our sampling sites; hence the $^{134}$Cs activities observed in these April 2011 samples allows estimates to be made for the $^{137}$Cs activities delivered from the Fukushima accident.
A number of sediment samples obtained from laboratory roads were also analyzed. The asphalt road connecting building 72 with the main part of LBNL extends almost horizontally across a steep hillside. An asphalt berm (curb) along the downslope edge of the road protects the slopes below from rainfall runoff from the road surface. Partway along this stretch of roadway (near the turnoff to LBNL building 31) is a low spot at which particulates, mainly abraded from the road surface, collect against the curb. Normal LBF procedure is to collect at least two samples per year from these deposits - - one just before the onset of the rainy season (September or October), and one at the end of the rainy season (May or June). One to two kilograms of material is collected from the upstream end of one of these narrow deposits. After air-drying, the material is sieved, and that fraction which passes through a 1/16Ó mesh screen becomes the sample to be analyzed. Typically, the sieved material represents at least 80% of the total collected material. For analysis, these sediments are packed into one of the same counting containers as are used for soil samples. These procedures were followed for the samples listed in [**Table \[tab:sediment\]**]{}.
Food Measurements in October 2013
=================================
The recent reports of radioactivity leaking from the Fukushima Nuclear Power Plant into 2013 prompted the measurement of various food samples to search for fission products, in particular fish since much of the media reports were related to continued releases of contaminated water into the ocean. There have already been studies of fission products in fish, such as those used to track migration in tuna using $^{134,137}$Cs by Madigan et. al. [@Madigan2012; @Madigan2013].
To explore the presence of such products in the current food supply, fish from the Pacific Ocean and a few various other foods were obtained at local San Francisco Bay Area retail locations in September 2013. Samples containing a significant water weight, such as fish and yogurt, were first baked to reduce the overall sample mass and enable easier handling; but the wet weight was recorded for reporting and use in analysis. Samples were then placed in plastic bags and counted in a cylindrical geometry at the face of a HPGe detector (2.1 kg p-type HPGe crystal, 85% relative efficiency) counted for a few days for each sample. For efficiency determinations, a calibration sample was created using a NIST Standard Reference Material– ‘River Sediment 4350B,’ which contains certified activities of $^{137}$Cs, $^{60}$Co, $^{152, 154}$Eu, $^{226}$Ra, $^{241}$Am, and others [@nist4350B]. The uncertainties of the isotopes in the river sediment vary from 4 to 21%, where the primary isotope of interest for this set of measurements, $^{137}$Cs, had an uncertainty of 6.4%. To achieve a volume similar to the food samples analyzed, 53.5 grams of the river sediment was thoroughly mixed with 141.7 grams of flour and counted in the same geometry.
The results of the food sampling are shown in [**Table \[tab:fukushimafood\]**]{}. In most of the Pacific Fish samples, $^{137}$Cs was present, with the highest activity found in tuna from the Philippines, which had a $^{137}$Cs activity of 0.24(4) Bq kg$^{-1}$. No samples had a detectable presence of $^{134}$Cs, which would have indicated that Fukushima products were present. Therefore, without $^{134}$Cs, the detected $^{137}$Cs is attributed to legacy activities such as surface nuclear weapons testing. It is also worth noting, that all samples had much higher levels $^{40}$K present, which is a naturally-occurring isotope. Comparing $^{137}$Cs to $^{40}$K is useful, since they both belong to the same column on the periodic table, and hence have similar affinities in various tissues and minerals. The tuna from the Phillipines for instance, had a $^{40}$K activity of 105(3) Bq kg$^{-1}$ – more than four hundred times the activity of the $^{137}$Cs. Comparisons such as this are useful in evaluating the relative danger the $^{137}$Cs presents to the public without referencing limits set by regulations, as it allows one to make direct comparisons to natural background radiation. The last sample listed in [**Table \[tab:fukushimafood\]**]{} are weeds collected after the rainfall described by Norman in 2011 [@Norman2011] that had absorbed $^{134, 137}$Cs from Fukushima fallout. These weeds were cooked and washed very thoroughly in an attempt to determine whether the activity was inside the plants or merely surface-contaminated, and recounting proved that the activity was indeed absorbed inside the plant matter. The weeds were recounted in October 2013 along with the food samples in [**Table \[tab:fukushimafood\]**]{} as a comparison. Then the weeds were recounted again in November of 2013 along with the Philippines tuna sample as seen in [**Figure \[fig:weedstuna\]**]{}, this time on a lower background system; the same 115% HPGe system at the LBNL LBF used in the air monitoring efforts. The results of the low background counting of these samples are seen in [**Table \[tab:merweedstuna\]**]{}. These weeds were used as a proxy for Fukushima fallout– the ratio of $^{134}$Cs/$^{137}$Cs was compared to the $^{137}$Cs we detected in our food samples, which confirmed that if the $^{137}$Cs was indeed from Fukushima, we should have easily seen $^{134}$Cs over our detection limits. Therefore, the only detected $^{137}$Cs was from pre-Fukushima, legacy sources. The U.S. Food and Drug Administration (FDA) Derived Intervention Level (DIL) for the combined $^{134,137}$Cs activity in food [@fdadil] is currently 1200 Bq kg $^{-1}$. All food samples measured were more than 1000-times smaller than the 1200 Bq kg $^{-1}$ FDA DIL and pose no concern to the public and were far less than natural gamma-emitting radioisotopes present, namely $^{40}$K. Other studies that have found Fukushima isotopes in fish, such as Madigan, et. al.[@Madigan2012; @Madigan2013], have gone a step further and to show that Fukushima-sourced dose rates due to ingestion of even their highest-activities of $^{134,137}$Cs found in tuna are absolutely minuscule in comparison to the natural dose from $^{210}$Po as discussed in great detail in Fisher et. al. [@Fisher03062013].
--------------------- -------------- -------------- --------------
Sample $^{134}$Cs $^{137}$Cs $^{40}$K
Bq kg$^{-1}$ Bq kg$^{-1}$ Bq kg$^{-1}$
Local Weeds 04/2011 7.0(5) 20.8(7) 294(8)
Tuna Phillipines $<$0.07 0.20(1) 95.3(6)
--------------------- -------------- -------------- --------------
: The results of recounting the local weeds and the tuna sample from the Phillipines on a lower background system at the LBF, the spectra of which are seen in [**Figure \[fig:weedstuna\]**]{}. Since the weeds were collected in April of 2011 after absorbing Fukushima fallout, the ratio of $^{134}$Cs to $^{137}$Cs was used as a proxy to compare to the tuna sample, which had a consistent limit for $^{134}$Cs detection as compared to the $^{137}$Cs present. This suggests that if the $^{137}$Cs in the tuna sample was indeed from Fukushima, then our system would have seen $^{134}$Cs as it would be present above the listed detection limits. Therefore the $^{137}$Cs is from legacy, pre-Fukushima sources.[]{data-label="tab:merweedstuna"}
![A comparison of gamma spectra of local weeds (pictured in dotted blue) collected in April 2011 in Oakland, CA (recounted here in October 2013, at the LBNL LBF) along with a sample of Tuna from the Philippines (pictured in solid red) purchased locally at a San Francisco Bay Area retail location. $^{134}$Cs and $^{137}$Cs are readily visible and the remaining peaks are all natural background radioisotopes. The activities derived from these two samples are listed in [**Table \[tab:merweedstuna\]**]{} after recounting on this lower background system.[]{data-label="fig:weedstuna"}](weedstuna){width="\linewidth"}
Chernobyl Comparisons
=====================
Chernobyl Air Monitoring
------------------------
Similar air monitoring efforts were also made by the Low Background Facility in the aftermath of the Chernobyl accident in 1986. Following the accident an air filtration system was installed and operated. This system had a variable motor speed that was manually set to draw 2 CFM of air through filters 5 inches by 9 inches in size. The filters were then folded in half and counting in the annulus of a Marinelli style beaker, such that they were wrapped flat along the side of a HPGe spectrometer (p-type, 30% relative efficiency). Filters were typically exposed in 24 hour intervals and counted immediately after removal. Results of the airborne monitoring performed in the aftermath of the Chernobyl incident and are displayed in [**Figure \[fig:chernobylfallout\]**]{}. In addition to the isotopes seen from Fukushima, there was also an extended release of $^{103}$Ru (t$_{1/2}$ = 39.26 days, primary gamma ray of 497.9 keV at 90.9% intensity). The results of this monitoring showed the highest activities for fallout in the Bay Area on May 5, 1986 with airborne activities of: $^{132}$Te - 16.6$\pm$0.1 mBq m$^{-3}$, $^{131}$I - 95.6$\pm$0.5 mBq m$^{-3}$, $^{134}$Cs - 23.4$\pm$0.3 mBq m$^{-3}$, $^{137}$Cs - 41.8$\pm$0.4 mBq m$^{-3}$, and $^{103}$Ru - 25.3$\pm$0.5 mBq m$^{-3}$ (statistical uncertainties only, a $\sim$10% systematic uncertain should also be assumed).
![Airborne concentrations fission products starting April 26, 1986 following the Chernobyl disaster as measured on a HEPA filter at the LBF in Berkeley, CA. Horizontal error bars, when visible, represent filter exposure periods. Missing error bars are smaller than the visible data marker. Plotted uncertainties seen here are statistical only, and a conservative systematic uncertainty of $\sim$10% should also be assumed.[]{data-label="fig:chernobylfallout"}](chernobylfallout){width="\linewidth"}
When comparing the highest levels of fallout radionuclides from the Chernobyl and Fukushima accidents we can easily see the Chernobyl activities measured at the San Francisco Bay Area of California were about an order of magnitude higher than those from Fukushima. Roughly the same seven-day temporal period appears to separate successive maxima for these fallout nuclides, seen most clearly in [**Figure \[fig:chernobylfallout\]**]{}, a graphical presentation of the Chernobyl data, and [**Figure \[fig:allfalloutfull\]**]{} for those from Fukushima.
Chernobyl Food Monitoring
-------------------------
Also measured in the late 1980’s were a series of imported and domestic food products. The study from Norman and Harvey [@Norman88; @Norman89] was previously unpublished and is summarized here. After hearing extended reports of fallout isotopes originating from the Chernobyl disaster appearing in foods, a small survey of approximately 50 imported foods available in the San Francisco Bay Area was conducted. The samples originated from 21 European countries as well as 4 from the U.S.A. The study was conducted with a 109 cm$^{3}$ (approximately 0.6 kg) HPGe detector. All samples were counted directly on the face of the detector and all had masses between 10-500 grams. Efficiencies and attenuation affects were measured using calibrated point sources in various positions with varying thicknesses of absorbers to determine a counting efficiency to $\pm$20%. A table of results is presented in [**Table \[tab:chernobylfood\]**]{}.
Many of the products showed no activity above background and of those that did show radioisotopes, $^{134}$Cs and $^{137}$Cs were the only residual fission products still present from Chernobyl as others had decayed away. In order to confirm the origin of the two isotopes, a ratio of the $^{137}$Cs/$^{134}$Cs activities were used to estimate the age they were produced. All food product samples, except for one, produced a $^{137}$Cs/$^{134}$Cs ratio of 2.95$\pm$0.30 (as of October 1987). Due to the difference in half-lives of the isotopes, this ratio can be used as an identifier of their source, since it will be unique for each reactor in its fuel cycle. Decay correcting this ratio to the date of the Chernobyl incident, April 25, 1986, the ratio is 1.88$\pm$0.19 which was in agreement with ratios present in prompt fallout such as those in [**Figure \[fig:chernobylfallout\]**]{}, which yield a ratio of 1.90$\pm$0.08. All food measurements were below the maximum permissible limits allowed in food at the time, which was 10 pCi g$^{-1}$ (370 Bq kg$^{-1}$) after being reduced from 75 pCi g$^{-1}$ (2.8 $\times 10^{3}$ Bq kg$^{-1}$) [@usda]. The authors extrapolated these measurements to exposure, if one were to eat 2 kg/day of foods at this level of activity. Factoring in that both of the isotopes have a biological half-life of 70 days [@lanl], they estimated that at equilibrium one would have a total activity of 0.5 $\mu$Ci (2 $\times 10^{7}$ Bq kg$^{-1}$) of $^{134}$Cs and 1.5 $\mu$Ci (5.6 $\times 10^{7}$ Bq kg$^{-1}$) of $^{137}$Cs, which were well below the limits of 2.0 $\mu$Ci (7.4 $\times 10^{7}$ Bq kg$^{-1}$) of $^{134}$Cs and 3.0 $\mu$Ci (1.1 $\times 10^{8}$ Bq kg$^{-1}$) of $^{137}$Cs set for non-radiation workers at the time [@lanl].
Conclusion
==========
Monitoring of fallout from the Fukushima nuclear accident was performed on a variety of media at the LBNL Low Background Facility starting soon after the tsunami on March 14, 2011 and reported here through the end of 2012 on air filters, automobile air filters, and rainwater. At the local LBNL LBF location, HEPA air filters were used in sampling durations ranging from 24 hours to one month while monitoring radioisotope concentrations. Rainwater was also collected and analyzed through the end of 2012 in Oroville, CA. Additionally, the fission products monitored provided a useful demonstration of the use of automobile filters as a low cost means of monitoring for radioisotopes. More extensive analysis was performed upon rainwater samples from March 2011 that contained the highest measured activities in the Berkeley area in an effort to search for the presence of $^{90}$Sr in the rainwater that arrived in the initial weeks following the incident, in which it was not detected and one sigma limits were placed at $<$ 8.98 mBq L$^{-1}$. A series of food measurements were also performed in October 2013 upon imported food products from the Pacific region, and although background $^{137}$Cs was present from sources prior to Fukushima, it was far below any level of concerns due to radioactivity– both from FDA DIL’s and comparison to natural $^{40}$K. Measurements made by the same authors of air filters and food products in the aftermath of the Chernobyl disaster are also presented here in comparison to the Fukushima fallout. The main conclusion drawn from these sets of data is that the peak fallout activities from Chernobyl in 1986 upon the San Francisco Bay Area were approximately an order of magnitude more than the levels seen from Fukushima in 2011.
The fission products measured in this work in the San Francisco Bay Area of California were not found at any time to be of concern to public or environmental health. These conclusions are justified not only through governmental limits of exposure and activities, but also by direct comparison of natural radioactivities also present in all samples measured.
Acknowledgements
================
We wish to thank M. B. Norman for useful discussions that prompted our searches for Fukushima fallout in food samples.
This material is based upon work supported by the Department of Energy National Nuclear Security Administration under Award Number(s) DE-NA0000979 and by the Director, Office of Energy Research, Office of High Energy and Nuclear Physics, Division of Nuclear Physics, of the US Department of Energy under Contract No. DE-AC02-05CH11231.
References
==========
Supplementary Figures and Tables
================================
Isotope Peak keV (intensity) Source t$_{1/2}$
------------ ----------------------- -------------- -------------
$^{132}$Te 228 (88%) fission 3.2 days
$^{212}$Pb 238 (43%) Th series 10.64 hours
$^{131}$I 284 (6%), 364 (82%), fission 8.02 days
637 (7%), 723 (2%)
$^{214}$Pb 352 (38%) U series 26.8 min
$^{136}$Cs 340 (42%), 819 (100%) fission 13.16 days
$^{7}$Be 477 (11%) cosmogenic 53.12 days
n/a 511 annihilation n/a
$^{132}$I 522 (16%), 667 (97%), fission 2.295 hours
773 (76%)
$^{134}$Cs 563 (8%), 569 (15%), fission 2.06 years
604 (98%), 795 (86%)
$^{208}$Tl 583 (84%) Th series 3.05 min
$^{214}$Bi 609 (46%) U series 19.9 min
$^{137}$Cs 661 (85%) fission 30.07 years
: Radioisotopes shown in Figures \[fig:airfilter\] and \[fig:autofilter\], along with the gamma ray energies, intensities, and half-life. Isotope data adapted from [@toi].[]{data-label="tab:isotopes"}
![Airborne concentrations of two naturally-produced radioisotopes, $^{210}$Pb and $^{7}$Be, from March 11, 2011 to the end of 2012 as measured on HEPA filters at the LBF in Berkeley, CA. These isotopes provide some context to the scale of the activities seen from the fission products in Figure \[fig:allfalloutfull\]. Horizontal error bars, when visible, represent filter exposure periods. Missing error bars are smaller than the visible data marker.Plotted uncertainties seen here are statistical only, and a conservative systematic uncertainty of $\sim$10% should also be assumed.[]{data-label="fig:pbbe"}](Pb210Be7loglog){width="\linewidth"}
![Counting rates for natural radioisotopes as seen on automobile filters in Berkeley, CA. These gross counting rates are of use for comparison to the counting rates in Figure \[fig:auto\], which reveals the fallout counting rates were similar in scale to natural radioisotopes. The day plotted for filters represents the date after 3/11/11 when the filter was removed from a vehicle. Counting rates were not converted to airborne activity, as the full mileage (and hence the airflow) information was not available for the full data set, so the information is presented solely as a qualitative demonstration of the use of automobile filters for radionuclide monitoring.[]{data-label="fig:autonatural"}](autonatural-loglin){width="\linewidth"}
![Observation of Fukushima fallout in rainwater collected in Oroville, CA over the entire monitoring duration to the end of 2012. $^{7}$Be is shown as a comparison to a naturally produced airborne isotope. Horizontal error bars, when visible, represent filter exposure periods. Missing error bars are smaller than the visible data marker. Plotted uncertainties seen here are statistical only, and a conservative systematic uncertainty of $\sim$10% should also be assumed.[]{data-label="fig:ororainloglog"}](OROrainAll-loglog){width="\linewidth"}
![Low energy to high energy ratios (0 to 546 keV region to 546 to 2000 keV region) for Sr precipitate samples immediately after preparation (black) and at one week old (blue). This showed no evidence or the presence of $^{90}$Sr since all measurements agree to within approximately 1$\sigma$.[]{data-label="fig:ratiocomparison"}](ratiocomparison){width="0.7\linewidth"}
------------------ --------- -------------- --------------
Location Date $^{134}$Cs $^{137}$Cs
m/d/yr Bq kg$^{-1}$ Bq kg$^{-1}$
bldg 72 vicinity 7/25/98 $<$0.01 1.06(5)
bldg 90 vicinity 8/5/98 $<$0.01 8.5(1)
bldg 72 vicinity 4/27/11 1.70(4) 2.69(5)
bldg 90 vicinity 4/30/11 0.28(2) 3.47(5)
------------------ --------- -------------- --------------
: Sample results of surface soil tests at LBNL in 1998 and post-Fukushima in spring 2011.[]{data-label="tab:soil"}
---------- -------------- -------------- -------------- -------------- --------------
Date $^{134}$Cs $^{137}$Cs $^{131}$I $^{7}$Be $^{210}$Pb
m/d/yr Bq kg$^{-1}$ Bq kg$^{-1}$ Bq kg$^{-1}$ Bq kg$^{-1}$ Bq kg$^{-1}$
11/18/10 $<$0.02 0.09(1) $<$0.01 84.8(4) 53.5(6)
3/1/11 $<$0.02 0.11(1) $<$0.02 220.0(8) 46.6(7)
4/11/11 20.0(1) 24.6(1) 21.29(8) 318.9(8) 59.0(9)
6/9/11 58.6(2) 73.8(2) 0.15(4) 346.7(8) 92(2)
9/27/11 34.5(1) 48.2(1) $<$0.03 88.8(4) 71.9(9)
10/9/11 22.6(1) 31.8(1) $<$0.03 90.8(4) 52.5(9)
1/4/12 17.43(5) 26.5(1) $<$0.02 67.8(4) 48.8(6)
4/9/12 15.67(5) 26.1(1) $<$0.02 163.1(4) 61.6(7)
9/20/12 14.44(5) 27.3(1) $<$0.02 51.4(3) 82.8(6)
11/25/12 5.26(5) 10.6(1) $<$0.02 104.5(4) 43.3(6)
---------- -------------- -------------- -------------- -------------- --------------
: Sample results of surface sediment tests along the roadside at LBNL building 72 post-Fukushima in spring 2011. The activities in this sediment often become elevated since rainwater drains off the roadway and a natural filtration mechanism occurs.[]{data-label="tab:sediment"}
----------------------- ------- -------------- -------------- --------------
Sample mass $^{40}$K $^{134}$Cs $^{137}$Cs
g Bq kg$^{-1}$ Bq kg$^{-1}$ Bq kg$^{-1}$
Tuna Philippines 444.5 105(4) n.d. 0.27(6)
Fiji Tuna 506 62(2) n.d. 0.12(2)
Local Red Snapper 440 67(3) n.d. 0.10(3)
Local Organic Yogurt 907.2 66(2) n.d. n.d.
Hawaii Swordfish 467.2 99(4) n.d. n.d.
Local Squid 902.6 21(1) n.d. n.d.
Alaskan Salmon 430.9 82(3) n.d. n.d.
Alaskan Cod 444.5 68(3) n.d. 0.13(4)
Atlantic Mackerel 396 78(3) n.d. n.d.
Japanese Dried Bonito 100 195(10) n.d. n.d.
Local Grape Leaves 64.9 249(14) n.d. n.d.
Local Weeds (04/2011) 30.6 294(27) 7.0(6) 18(1)
----------------------- ------- -------------- -------------- --------------
: Food measurements made in search of fallout from the Fukushima nuclear accident in food samples purchased in San Francisco Bay Area retail locations in October 2013, with the exception of the local weeds that were extracted during April 2011, after the initial rainfall containing fallout isotopes, but re-counted here in October 2013. Activities of $^{137}$Cs are listed alongside natural $^{40}$K activity that were also measured. Reported masses for the food samples are the wet weight before baking to expel excess water mass. For entries with no data, it can be assumed the value is less than the approximate MDA’s for this set of measurements, which varied over the following ranges: 0.06-0.15 Bq kg$^{-1}$ for $^{137}$Cs and 0.05-0.13 Bq kg$^{-1}$ for $^{134}$Cs. The absence of $^{134}$Cs suggests that the $^{137}$Cs measured in some of the samples is from legacy sources prior to Fukushima. []{data-label="tab:fukushimafood"}
------------------- ---------------------- -------------- --------------
Country of Origin Product
pCi g$^{-1}$ Bq kg$^{-1}$
England beer n.d. n.d.
crackers n.d. n.d.
Belguim beer n.d. n.d.
shallots n.d. n.d.
Spain white wine (1986) n.d. n.d.
France red wine n.d. n.d.
white whine n.d. n.d.
black olives n.d. n.d.
raspberry jam n.d. n.d.
apricot jam n.d. n.d.
honey (1986) 0.4 14.8
goat cheese 0.2 7.4
rouquefort cheese 0.2 7.4
Italy pasta (brand 1) n.d. n.d.
pasta (brand 2) 1.3 48.1
pasta (brand 3) 0.8 29.6
pasta (brand 4) 1.6 59.2
olive oil n.d. n.d.
red wine (1986) n.d. n.d.
white wine (1986) n.d. n.d.
dried mushrooms 4 148
Switzerland green cheese w/herbs 0.2 7.4
emmenthaler cheese 0.13 4.81
Austria blackberry jam 0.4 14.8
beer n.d. n.d.
West Germany cheese w/herbs n.d.
soft cheese n.d. n.d.
beer n.d. n.d.
gummi bear candy n.d. n.d.
The Netherlands beer n.d.
Norway beer n.d. n.d.
goat cheese 5 185
Sweden lingon berry sauce 0.13 4.81
crispbread n.d. n.d.
Finland rye crackers 0.16 5.92
U.S.S.R. vodka n.d. n.d.
Czechoslovakia beer n.d. n.d.
Poland ham n.d. n.d.
Hungary paprika n.d. n.d.
Roumania feta cheese n.d. n.d.
Yugoslavia beer 0.1 3.7
Greece black olives n.d. n.d.
beer 1.4 51.8
U.S.A. beer n.d. n.d.
pasta n.d. n.d.
cream cheese n.d. n.d.
cheddar cheese n.d. n.d.
------------------- ---------------------- -------------- --------------
: Chernobyl food measurements from 1988 previously unpublished (formally) from Norman and Harvey [@Norman88; @Norman89]. All samples were purchased at retail locations in the San Francisco Bay Area. Overall uncertainties are estimated to be $\pm$25%. n.d. indicates no activity was detected, in which the detection limits for this study varied from 0.02 - 0.1 pCi/g (0.74 - 3.7 Bq/kg) amongst the samples.[]{data-label="tab:chernobylfood"}
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Using the Yale stellar evolution code, models of based on asteroseismic measurements are constructed. A $\chi^{2}$ minimization is performed to approach the best modeling parameters which reproduce the observations within their errors. By combining all non-asteroseismic constraints with asteroseismic measurements, we find that the observational constraints favour a model with **a mass of 1.00$^{+ 0.01}_{- 0.02}$ $M_{\odot}$, an age t = 6.433 $\pm$ 0.04 Gyr, a mixing-length parameter $\alpha$ = 1.75 $\pm$ 0.25 , an initial hydrogen abundance $X_{i}$ = 0.605$^{+ 0.01}_{- 0.005}$ and metal abundance $Z_{i}$ = 0.0275$^{+ 0.002}_{- 0.001}$**. is in post-main sequence phase of evolution. The modes of $l$ = 1 show up the characteristics of avoided crossings, which may be applied to test the internal structure of this type stars. Asteroseismic measurements can be used as a complementary constraint on the modeling parameters. **The models with mass 1.00 - 1.10 $M_{\odot}$ can reproduce the observational constraints. Existing observed data of do not rule out these models.**'
address: 'School of Physics and Chemistry, Henan Polytechnic University, Jiaozuo 454000, Henan, China. '
author:
- Wuming Yang
- Xiangcun Meng
title: 'Models of $\mu$ Her with asteroseismic constraints'
---
Stars: individual: , stars: evolution, stars: oscillations
Introduction
============
Solar-like oscillations have been confirmed for several main-sequence and subgiant stars, such as [@bedd04], [@kjel05], [@egge04b], [@carr05], etc. The large and small frequency separations of p-modes can provide a good estimate of the mean density and age of the stars [@ulr86; @ulr88]. It has been proven that asteroseismology is a powerful tool for determining the fundamental parameters of the stars [@egge05; @egge06].
is a G5 IV subgiant star, in the neighbourhood of the Sun. It is considered to be a solar-type star with mass 1.14 $M_{\odot}$, effective temperature $T_{e}$ = 5596 $\pm$ 80 K, and radius R = 1.77 $\pm$ 0.07 $R_{\odot}$ [@fuhr98]. Recently, @bona08 detected solar-like oscillations on $\mu$ Her and identified individual frequencies in the range of 900 - 1600 $\mu$Hz. These seismological data will provide a constraint on the fundamental parameters of . In this work, we try to determine modeling parameters of based on asteroseismic constraints using the Yale Rotation Evolution Code (YREC7) in its non-rotating configuration.
The observational constraints available for are summarized in Sect. 2, while the details of the evolutionary models and the computational method are given in Sect. 3. The results are presented in Sect. 4 and the conclusion is given in Sect. 5.
Observational constraints
=========================
Effective temperature, luminosity and chemical composition
----------------------------------------------------------
The effective temperature of given by @fuhr98 is 5596 K, however that given by @ivan04 is only 5390 K. Combining other data [@vald04; @take05; @soub08], we adopt an average effective temperature $T_{eff}$ = 5500 $\pm$ 90 K.
The luminosity of a star can be obtained through combining the knowledge of the magnitude and distance. By combining the visual magnitude V = 3.417 $\pm$ 0.014, the bolometric correction BC = - 0.15 $\pm$ 0.05 mag [@fuhr98], the solar absolute magnitude $M_{bol,\odot}$ = 4.746 [@leje98] and the newest Hipparcos parallax $\Pi$ = 120.33 $\pm$ 0.16 mas [@leeu07], we obtained a luminosity for $\mu$ Her of L = 2.70 $\pm$ 0.16 $L_{\odot}$.
In the version of the Catalogue of \[Fe/H\] determinations given by @cayr01, there are five metallicity values for $\mu$ Her. Recent determinations give the values: 0.26 [@soub05] and 0.29 [@take05]. We adopt the average of these determinations, \[Fe/H\] = 0.21 $\pm$ 0.07. This value is close to the value of 0.23 given by [@fuhr98]. For Population I stars, the ratio of surface heavy elements to hydrogen abundance is related to the Fe/H by $[Fe/H]
\simeq log(Z/X)_{s} - log(Z/X)_{\odot}$, where $(Z/X)_{\odot}$ is the ratio of the solar mixture. The most recent ratio of the heavy-element abundance to hydrogen abundance of the Sun, $(Z/X)_{\odot}$, is 0.0171 [@aspl04]. There are, however, some discrepancies between this new value and seismical results [@yang07]. Thus in this work, we adopt the old value, 0.0245 [@grev93]. Consequently, the value of $(Z/X)_{s}$ for $\mu$ Her is about 0.040 $\pm$ 0.006.
Asteroseismic data
------------------
solar-like oscillations for this star have been detected by @bona08 with the SARG echelle spectrograph. Twenty oscillation frequencies have been identified by using modified and standard extraction methods between 900 and 1600 $\mu$Hz. By means of a least square best fit with the asymptotic relation of frequencies for all the identified modes, @bona08 gave that the most likely value of the mean large and small frequency separation is $\Delta\nu$ = 56.50 $\pm$ 0.07 $\mu$Hz and $\delta\nu$ = 5.03 $\pm$ 0.94 $\mu$Hz, respectively. Some of the observational constraints for used in this work are given in Table \[tab1\].
------------------------- -------------------
$\Pi$ \[mas\] 120.33 $\pm$ 0.16
V \[mag\] 3.417 $\pm$ 0.014
$L/L_{\odot}$ 2.70 $\pm$ 0.16
$T_{eff}$ \[K\] 5500 $\pm$ 90
$[Fe/H]_{s}$ 0.21 $\pm$ 0.07
$\Delta\nu$ \[$\mu$Hz\] 56.50 $\pm$ 0.07
$\delta\nu$ \[$\mu$Hz\] 5.03 $\pm$ 0.94
------------------------- -------------------
: Observational data for used in this work. \[tab1\]
Stellar models
==============
Input physics
-------------
A grid of stellar evolutionary models was computed with the YREC7 in its non-rotating configuration [@guen92]. All models are evolved from fully convective pre-main sequence (PMS) to a stage of subgiant. The OPAL EOS2001 [@roge02], OPAL opacity [@igle96], and the @alex94 opacity for low temperature were used. These opacity tables have the solar mixtures given by @grev93. The models took into account diffusion of helium and metals, using the prescription of @thou94. Energy transfer by convection is treated according to the standard mixing-length theory, and the boundaries of the convection zones are determined by the Schwarzschild criterion. See @dema07 in details for the YREC.
Computational method
--------------------
The position of a stellar model in the Hertzsprung-Russell (H-R) diagram depends on five modeling parameters: the stellar mass $M$, the mixing-length parameter $\alpha$, the age of the star $t$, the initial hydrogen abundance $X_{i}$ and metallicity $Z_{i}$. For $\mu$ Her we take the ratio of the heavy-element abundance to hydrogen abundance on the stellar surface as an observable. Thus we have three observables ($L, T_{eff}, (Z/X)_{s}$) and five unknowns ($M, \alpha, t, X_{i}, Z_{i}$).
In order to reproduce the observational constraints, we construct a grid of models with various masses, initial element abundances and mixing-length parameters. For each stellar model, low-degree p-mode frequencies are computed using Guenther adiabatic pulsation code [@guen94]. To find the set of modeling parameters ($M, \alpha,
Z_{i}, X_{i}, t$) that leads to the best agreement with the observational constraints, following [@egge04a], we perform a $\chi^{2}$ minimization. The function $\chi^{2}$ is defined as follows $$\chi^{2} = \chi^{2}_{clas} + \chi^{2}_{osci},
\label{equa1}$$ where $$\chi^{2}_{clas} = (\frac{T^{eff}_{mod}-T^{eff}_{obs}}{\sigma(T^{eff}_{obs})})^{2}+
(\frac{\log(\frac{L}{L_{\odot}})_{mod}-\log(\frac{L}{L_{\odot}})_{obs}}
{\sigma(\log(\frac{L}{L_{\odot}})_{obs})})^{2}+
(\frac{[\frac{Fe}{H}]_{mod}-[\frac{Fe}{H}]_{obs}}{\sigma([\frac{Fe}{H}]_{obs})})^{2},
\label{equa2}$$ and $$\chi^{2}_{osci} = \frac{1}{N}\sum^{N}_{i=1}
(\frac{\nu^{theo}_{i} - \nu^{obs}_{i} - \langle
D_{\nu}\rangle}{\sigma})^{2}.
\label{equa3}$$ Here the $\sigma( )$’s are the errors on the corresponding observations, $N$ is the number of observed frequencies, $\sigma$ = 1.8 $\mu$Hz is the resolution on the observed frequencies, and $\langle D_{\nu}\rangle = \sum^{N}_{i=1}(\nu^{theo}_{i} -
\nu^{obs}_{i})/N$. The model which can minimize $\chi^{2}_{clas}$ and $\chi^{2}_{osci}$ at the same time will be considered to be the best one.
Results
=======
**Firstly**, we computed a grid of evolutionary tracks for models with masses, mixing length parameters, initial metal and hydrogen mass fractions respectively in the ranges of 0.86 $M_{\odot}\leq M \leq$ 1.16 $M_{\odot}$, 1.5 $\leq \alpha \leq$ 2.10, 0.025 $\leq Z_{i} \leq$ 0.033, 0.57 $\leq X_{i} \leq$ 0.70 **with a primary resolution $\delta M=0.02, \delta\alpha=0.2,
\delta Z=0.002$, and $\delta X=0.02$**. Computational results show that the evolutionary tracks for models with $0.9 M_{\odot} < M <
1.15 M_{\odot}$ and $1.5 \leq \alpha \leq 2.10$ approximately span the error box in \[$T_{eff}, L, (Z/X)_{s}$\]. This is because the fact that a decrease of the mass can be compensated by a decrease in hydrogen and metal abundances to get the same position in the H-R diagram [@meng08]. Moreover, the evolutionary tracks imply that $\mu$ Her is in the post-main sequence phase of evolution. In this phase, the tracks are almost parallel to the $T_{eff}$-axis (see Figs. \[fig1\], \[fig2\]). Additionally, a variation of the mixing length parameter mainly changes the radius, but has almost no influence on the luminosity [@kipp90]. Therefore, with increasing the mixing length parameter $\alpha$, the evolutionary track moves almost horizontally to the left of H-R diagram and vice versa. Thus the models with various mixing length parameters can span the same position in the H-R diagram (see Fig. \[fig2\]) at different ages. Table \[tab3\] gives the characteristics of three models, M2a, M2 and M2b, with different mixing length parameters but a same surface metallicity and position in the H-R diagram. These models have different $\chi^{2}_{osci}$, i.e. different oscillation frequencies, reflecting the differences in the internal structure of the models; for example, the differences in the central helium-core mass and density. This indicates that non-asteroseismic observational constraints do not enable us to determine the mixing-length parameter $\alpha$ for , an evolved solar-type star, but asteroseismic observations could provide a constraint on the mixing length parameter.
----------------- -------------------------------------- ------------------ ------------------------------ -------------------
$\log L/L_{\odot}$
Observational $T_{eff}$ \[K\]
constraints $(Z/X)_{s}$
$\Delta\nu$
M1 M2 M3
$M/M_{\odot}$ 1.10$_{- 0.01}$ 1.00$^{+ 0.01}_{- 0.02}$ 0.93 $\pm$ 0.015
Modeling $\alpha$ 1.75 $\pm$ 0.25 1.75 $\pm$ 0.25 1.75 $\pm$ 0.25
parameters $Z_{i}$ 0.029$^{+0.001}$ 0.0275$^{+ 0.002}_{- 0.001}$ 0.026 $\pm$ 0.002
$X_{i}$ 0.650$^{+0.005}$ 0.605$^{+ 0.01}_{- 0.005}$ 0.572 $\pm$ 0.01
$\log L/L_{\odot}$ 0.4550 0.4386 0.4307
$T_{eff}$ \[K\] 5454 5494 5492
Model $Z_{s}$ 0.0272 0.0257 0.0245
characteristics $(Z/X)_{s}$ 0.040 0.040 0.041
$R/R_{\odot}$ 1.894 1.831 1.816
age \[Gyr\] 6.556 $\pm$ 0.03 6.433 $\pm$ 0.04 6.303 $\pm$ 0.03
$\chi^{2}_{osci}$ 1.05 1.07 4.55
$\chi^{2}_{clas}$ 1.18 0.10 0.03
$\chi^{2}$ 2.23 1.17 4.58
$\Delta\nu_{0}$ 56.44 56.58 55.34
$\delta\nu_{02}$ 5.00 4.94 5.09
$\langle D_{\nu}\rangle$ \[$\mu$Hz\] 28.10 31.20 -0.53
He-core mass \[$M_{\odot}$\] 0.100 0.101 0.105
----------------- -------------------------------------- ------------------ ------------------------------ -------------------
: Models for . The superscripts or subscripts of some modeling parameters show confidence limits. \[tab2\]
-------------------------------- ------------------- ------------------- -------------------
M2a M2 M2b
$M/M_{\odot}$ 1.00 1.00 1.00
$Z_{i}$ $\mathbf{0.0275}$ $\mathbf{0.0275}$ $\mathbf{0.0275}$
$X_{i}$ 0.605 0.605 0.605
$\alpha$ 1.60 1.75 1.90
age \[Gyr\] 6.374 $\pm$ 0.04 6.433 $\pm$ 0.04 6.471 $\pm$ 0.03
$\log L/L_{\odot}$ 0.4387 0.4386 0.4387
$T_{eff}$ \[K\] 5494 5494 5493
$(Z/X)_{s}$ 0.040 0.040 0.041
$R/R_{\odot}$ 1.832 1.831 1.832
$\chi^{2}_{clas}$ 0.099 0.097 0.101
$\chi^{2}_{osci}$ $\mathbf{3.50}$ 1.07 3.37
$\Delta\nu_{0}$ 56.65 56.58 56.46
$\log \rho_{c}$ \[g/$cm^{3}$\] 3.636 3.800 3.942
He-core mass \[$M_{\odot}$\] 0.091 0.101 0.109
-------------------------------- ------------------- ------------------- -------------------
: Characteristics of the models illustrated by open circles in Fig.\[fig2\]. \[tab3\]
**For the sets of modeling parameters that lead to agreements with the observational constraints, we calculated the models with a fine resolution $\delta M=0.01, \delta\alpha=0.05, \delta Z=0.0005$, and $\delta X=0.005$ in the vicinity of the parameters. We obtained many models that can almost minimize the $\chi^{2}_{osci}$ and fall within the observational error box. Fig. \[fig3\] shows the $\chi^{2}_{osci}$ as a function of mass and age. From this figure we can see that the models with mass 1 - 1.1 $M_{\odot}$ and age 6.2 - 6.7 Gyr can better reproduce the observed frequencies. We also constructed the models with higher resolutions of modeling parameters than the fine resolution. But results are not sensitive to the resolutions.** Performing the $\chi^{2}$ minimization described above, we found a solution: M = 1.00 $M_{\odot}$, $\alpha$ = 1.75, $Z_{i}$ = 0.0275, $X_{i}$ = 0.605, and t = 6.433 Gyr, marked M2. Table \[tab2\] lists characteristics of this model. **The confidence limits of each modeling parameter correspond to the maximum/minimum values it can reach when other parameters are fixed, in order that the generated models fall within the observational error box.** Corresponding evolutionary tracks are shown in Figure \[fig1\]. Although the value of $\chi^{2}_{clas}$ and $\chi^{2}_{osci}$ of the M2 model is not the lowest one respectively, this model is almost able to minimize $\chi^{2}_{clas}$ and $\chi^{2}_{osci}$ at the same time and has a lowest $\chi^{2}$. **Fig. \[fig3\] shows the model with a mass of 1.04 $M_{\odot}$ has a lowest $\chi^{2}_{osci}$. But the value of $\chi^{2}$ of this model is larger than that of M2.** The mean large and small separation of the M2 model is 56.58 and 4.94 $\mu$Hz, respectively, which are in good agreement with those observed. However, the mass of M2 model, 1.00 $M_{\odot}$, is less than the values given by [@fuhr98] and [@take05]. They showed that the mass of $\mu$ Her is about 1.14 $M_{\odot}$. Table \[tab2\] also gives a model M1 which has a mass as high as 1.10 $M_{\odot}$. This model is able to minimize $\chi^{2}_{osci}$ in the age of 6.556 Gyr but has a higher value of $\chi^{2}_{clas}$. This implies that increasing the hydrogen abundance and mass at the same time for a fixed value of the mixing length parameter leads to an increase in the $\chi_{clas}^{2}$ in order to attain the same value of $\chi_{osci}^{2}$. This scenario was also found by [@egge06]. The mean large and small separation of the M1 model is 56.44 and 5.00 $\mu$Hz, respectively, which are almost same as those of the M2 model.
In Fig. \[fig4\], we plotted the differences between the observed and theoretical frequencies for the M1 and M2 model. Moreover, the theoretical and observed frequencies are compared by plotting echelle diagrams of the M1 and M2 models in Fig. \[fig5\]. The systematic difference $\langle D_{\nu}\rangle$ between computed and observed frequencies has been considered in these figures. The systematic difference $\langle D_{\nu}\rangle$ is 28.1 and 31.2 $\mu$Hz for the M1 and M2 models, respectively. This systematic difference also exists in other stellar models [@chri95; @egge04a; @egge04b; @egge06]. Fig. \[fig4\] shows that the differences between the observed and theoretical frequencies are very similar for both models at high frequency, however they are different at lower frequency. Compared to the M2 model, the M1 model badly reproduces the observed frequencies of 1034.0 and 1061.2 $\mu$Hz, whereas the M2 model does not reproduce the observed frequencies of 947.6 and 1124.8 $\mu$Hz well. These comparisons between individual observed and theoretical asteroseismic frequencies for the M1 and M2 models do not allow us to differentiate which model is better. Additionally, both models have the almost same values of the mean large and small separations reflecting a similarity of the internal structure between the M1 and M2 models. For example, both models have an almost same helium-core mass.
However, Fig. \[fig5\] shows that both the M1 and M2 models badly reproduce the observed frequencies of $l$ = 1. We searched the models which can better reproduce the observed frequencies of $l$ = 1. A model labeled M2b, which was shown in Table \[tab3\], was found. However, the value of $\chi^{2}_{osci}$ for the M2b model is larger than **those** of the M1 and M2 models, which is mainly due to differences in the modes of $l$ = 1 of these models. If the contribution to $\chi^{2}_{osci}$ from the $l$ = 1 mode near 1238 $\mu$Hz is neglected, the value of $\chi^{2}_{osci}$ for the M2b model will decrease from 3.37 to 0.72. Echelle diagram for the M2b model is shown in Fig. \[fig6\]. Except for the $l$ = 1 mode near 1238 $\mu$Hz, the M2b model reproduces the observed frequencies well. Moreover Fig. \[fig6\] shows that the observed and theoretical frequencies with $l$ = 1 have a similar behavior: **a zigzag echelle diagram**. The modes of $l$ = 1 show up a deviation from their expected asymptotic values. Fig. \[fig7\] also shows that the large separations for modes of $l$ = 1 deviate from their expected asymptotic behaviour. This may be because that the formation of the central helium core leads to an increase in the frequencies of the g-modes. ‘When the frequency of a g-mode approaches that of a p-mode, the two modes undergo an avoided crossing, where they exchange physical nature’ [@aize77; @chri95]. Figs. \[fig6\] and \[fig7\] show that the $l$ = 1 modes of the M2b model between 1200 and 1300 $\mu$Hz may undergo the avoided crossing. Noting the helium-core mass of the M1 and M2 models are almost same, but that of the M2 and M2b models are different, the behaviour of the $l$ = 1 modes might be very sensitive to the internal structure of the stars. It may be applied to extract the information of the helium core.
We also considered a $\chi^{2}_{osci}$ which does not contain $\langle D_{\nu}\rangle$. For models falling within the observed constraints on effective temperature, luminosity and surface metallicity, we computed this new $\chi^{2}_{osci}$. A solution with M = 0.93 $M_{\odot}$, $\alpha$ = 1.75, $Z_{i}$ = 0.026, $X_{i}$ = 0.572, and t = 6.303 Gyr was found. The evolutionary track and the position of this model (marked M3) in the H-R diagram are also shown in Fig. \[fig1\]. The characteristics of the M3 model are given in Table \[tab2\]. The systematic difference $\langle D_{\nu}\rangle$ between computed and observed frequencies is -0.53 $\mu$Hz for this model. However the value of the new $\chi^{2}_{osci}$ is 4.55, which is much larger than that of the M1 and M2 models. Moreover the echelle diagram of the M3 model is plotted in Fig. \[fig8\]. This echelle diagram shows that the M3 model can reproduce the frequencies in the range of 1034 $< \nu <$ 1437 $\mu$Hz but badly reproduces the modes at the lower and higher frequencies. Moreover its mean larger separation is only 55.49 $\mu$Hz. Thus this model is in disagreement with the asteroseismic constraints.
Discussion and Conclusion
=========================
The mass of 1.00 $M_{\odot}$ for the M2 and M2b models is less than the value in the previous literatures [@fuhr98; @take05], which may result from that the results of [@fuhr98] and [@take05] were obtained without the asteroseismic constraints. The initial hydrogen mass fraction is 0.605 and 0.650 for the M2 and M1 model, respectively. Although there is a difference of 0.0015 in Z between the M1 and M2 model, the increase of the hydrogen or decrease of helium is mainly compensated by an increase of the mass in order to reach the same location in the H-R diagram: a helium-mass degeneracy [@lebr93]. The M1 and M2 models have the same value of $\chi_{osci}^{2}$ and the almost same asteroseismic features. This implies that the helium-mass degeneracy is difficult to be removed, even if the asteroseismic constraints are taken into account.
Comparing with the observed frequencies, the theoretical frequencies have a systematic shift. **The mode of $l$ = 1 near 1238 $\mu$Hz deviate from its expected asymptotic value in Fig. \[fig6\]. It may undergo an avoided crossing and be a mixed-mode. The mixed-mode could have higher inertia than other p-modes. The shift of this mode may be different from that of other p-modes. Thus we divide the $\chi_{osci}^{2}$ into two parts: one $\chi_{oscm}^{2}$ for mixed-modes and one $\chi_{oscp}^{2}$ for all modes except for mixed-modes. Assuming only the $l=1$ mode near 1238 $\mu$Hz is a mixed-mode, we obtained the value of this new $\chi_{osci}^{2}$ is 1.07, 1.10, 3.61 and 0.72 for M1, M2, M2a and M2b, respectively. Under this assumption, M2b has a smallest $\chi_{osci}^{2}$.**
We confirmed the results that an analysis of the H-R diagram does not allow us to determine the mixing length parameter for an evolved solar-type star [@fern03] but the observed oscillation frequencies could provide a constraint on this parameter. The internal structure of the evolved solar-type stars is sensitive to the mixing length parameter at given ($T_{eff}, L$).
In this work we constructed the models for the using the Yale stellar evolution code. By combining the non-asteroseismic constraints with the asteroseismic observations, we find that a model for $\mu$ Her can reproduce the all non-asteroseismic and asteroseismic constraints well: **the model with a mass of 1.00$^{+ 0.01}_{- 0.02}$ $M_{\odot}$, an age t = 6.433 $\pm$ 0.04 Gyr, a mixing-length parameter $\alpha$ = 1.75 $\pm$ 0.25 , an initial hydrogen abundance $X_{i}$ = 0.605$^{+ 0.01}_{- 0.005}$ and metal abundance $Z_{i}$ = 0.0275$^{+ 0.002}_{- 0.001}$. However, the models with mass 1 - 1.1 $M_{\odot}$ and age $6.2 - 6.7$ Gyr also can reproduce the non-asteroseismic and asteroseismic constraints. Existing observational constraints do not rule out those models.**
The modes of $l$ = 1 show up the characteristic avoided crossings, which may be applied to test the internal structure of an evolved solar-type star. The asteroseismic observations put important constraints on the models for , but they are not enough to really test the differences in the models. More accurate oscillation frequencies, especially the modes of $l$ = 1, are need to investigate the internal structure of this type star.
Alexander, D. R., & Ferguson, J. W. 1994, ApJ, 437, 846 Asplund, M., Grevesesse N., Sauval, A. J., Allende Prieto, C., & Kiselman, D. 2004, A&A, 417, 751 Aizenman, M., Smeyers, P., & Weigert. A. 1977, A&A, 58, 41 Bedding, T. R., Kjeldsen, H., Butler, P. R., et al. 2004, ApJ, 614, 380 Bonanno, A., Benatti, S., & Claudi, R. et al. 2008, ApJ, 676, 1248
Carrier, F., Eggenberger, P., & Bouchy, F. 2005, A&A, 434, 1085 Cayrel, D. S. G., Soubiran, C., & Ralite, N. 2001, A&A, 373, 159 Christensen-Dalsgaard, J., Bedding, T. R., & Kjeldsen, H. 1995, ApJ, 443, L29
Demarque, P., Guenther, D. B., Li, L. H., Mazumdar, A., & Straka, C. W. 2007, AP&SS, 316, 31
Eggenberger, P., Charbonnel, C., & Talon, S. et al. 2004a, A&A, 235, 246 Eggenberger, P., Carrier, F., Bouchy, F. & Blecha, A. 2004b, A&A, 422, 247 Eggenberger, P., Carrier, F., & Bouchy, F. 2005, NewA, 10, 195 Eggenberger, P., & Carrier, F. 2006, A&A, 449, 293 Fernandes, J., & Monteiro, M. J. P. F. G. 2003, A&A, 399, 243 Fuhrmann, K. 1998, A&A, 338, 161
Grevesse, N., & Noels, A. 1993, in Origin and Evolution of the Elements, ed. N. Prantzos, E. Vangioni-Flam, & M. Cassé (Cambridge University Press), p15 Guenther, D. B., Demarque, P., Kim, Y.-C., & Pinsonneault, M. H. 1992, ApJ, 387, 372 Guenther, D. B. 1994, ApJ, 422, 400 Iglesias, C., & Rogers, F. J. 1996, ApJ, 464, 943 Ivanov, V. D., Rieke, M. J., & Engelbracht, C. W. et al. 2004, ApJS, 151, 387 Kippenhahn, R., & Weigert, A. 1990, in Stellar Structure and Evolution (Berlin, Springer-Verlag), p271 Kjeldsen, H., Bedding, T. R., & Butler, R. P. et al. 2005, ApJ, 635,1281 Lebreton, Y., Auvergne, M., & Morel, P. 1993, in Inside the Stars, IAU Coll. 137, ed. A. Baglin, & W. W. Weiss, ASP Conf. Ser., 40, 474 Lejeune, T., Cuisinier, F., & Buser, R. 1998, A&AS, 130, 65 McWilliam, A. 1990, ApJS, 74, 1075 Meng, X., Chen, X., & Han, Z. 2008, A&A, 487, 625
Rogers, F. J., & Nayfonov, A, ApJ 2002, 576,1064
Soubiran, C., & Girard, P. 2005, A&A, 438, 139 Soubiran, C., Bienayme, O., Mishenina, T. V., & Kovtyukh, V. V. 2008, A&A, 480, 91
Takeda, Y., Ohkubo, M., & Sato, B. et al. 2005, PASJ, 57, 27 Thoul, A. A., Bahcall, J. N., Loeb, A. 1994, ApJ, 421, 828
Ulrich, R. K. 1986, ApJL, 306, L37 Ulrich, R. K. 1988, IAUS, 123, 2990
Valdes F., Gupta R., Rose J.A., Singh H.P., Bell D.J. 2004, ApJS,152, 251 Van Leeuwen, F. 2007, A&A, 474, 653
Yang, W. M., & Bi, S. L. 2007, ApJ, 658, L67
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Killing-Yano (KY) two and three forms of a class of spherically symmetric space-times that includes the well-known Minkowski, Schwarzschild, Reissner-Nordstr[ø]{}m, Robertson-Walker and six different forms of de Sitter space-times as special cases are derived in a unified and exhaustive manner. It is directly proved that while the Schwarzschild and Reissner-Nordstr[ø]{}m space-times do not accept any KY 3-form and they accept only one 2-form, the Robertson-Walker space-time admits four KY 2-forms and only one KY 3-form. Maximal number of KY-forms are obtained for Minkowski and all known forms of de Sitter space-times. Complete lists comprising explicit expressions of KY-forms are given.'
address: |
$^{1}$ Department of Physics, Ankara University, Faculty of Sciences, 06100, Tandoğan-Ankara, Turkey\
$^{2}$ Department of Physics Engineering, Hacettepe University, 06800, Beytepe-Ankara, Turkey.
author:
- 'Ö. Aç[i]{}k $^{1}$'
- 'Ü. Ertem $^{1}$'
- 'M. Önder $^{2}$'
- 'A. Verçin $^{1}$'
title: 'Killing-Yano Forms of a Class of Spherically Symmetric Space-Times II: A Unified Generation of Higher Forms'
---
Introduction
============
In the previous study [@ozumav1] (henceforth referred to as I), we developed a constructive method which provided a unified generation of all Killing vector fields for a class of four dimensional ($4D$) spherically symmetric space-times. As the sequel of I, in the present paper we shall investigate the explicit forms of KY two and three forms by solving the KY-equations $$\begin{aligned}
\nabla_{X_{a}}\omega_{(p)} =\frac{1}{p+1} i_{X_{a}}d\omega_{(p)}\;,\end{aligned}$$ for this class of space-time metrics in the $p=2$ and $p=3$ cases.
The underlying base manifold is supposed to be a $4D$ pseudo-Riemannian manifold endowed with the metric $g$ having the Lorentzian signature $(-+++)$ such that $$\begin{aligned}
g=-e^{0}\otimes e^{0}+e^{1}\otimes e^{1}+e^{2}\otimes
e^{2}+e^{3}\otimes e^{3} \;.\nonumber\end{aligned}$$ in a local orthonormal co-frame basis $\{e^{a}\}$. This metric can be parameterized by the (metric) characterizing functions $T=\exp(\lambda(t))$ and $H_{j}=H_{j}(r)$ by choosing $$\begin{aligned}
e^0 &=& H_{0}dt\;,\qquad
e^1=T H_{1}dr\;,\nonumber\\
e^2 &=& T H_{2}d\theta\;,\quad e^3=T H_{2}\sin\theta d\varphi
\;.\nonumber\end{aligned}$$ The corresponding orthonormal vector bases will be denoted by $X_{a}$. The torsion-free connection $1$-forms for this class of metrics and covariant derivatives of basis elements required for explicit calculations can be found in I. We shall mainly use the notation of I, which is in accordance with [@Benn-Tucker].
As they span the kernel of the linear operator $\nabla_{X}-(p+1)^{-1}i_{X}d$, the space $Y_{p}$ of all KY p-forms constitute a linear space [@Semmelman1; @Stepanov]. Any function is a KY 0-form, $\omega_{(1)}$ is the dual of a Killing vector field and $\omega_{(n)}$ is a constant (parallel), that is, it is a constant multiple $\omega_{(n)}=az$ of the volume form, say $z=e^{0123}$ for $n=4$. Therefore while $Y_{0}$ is infinite dimensional, $Y_{n}$ is one dimensional. By a straightforward extension of the argument for determining the maximum number of Killing vectors, the upper bound for the dimension of $Y_p$’s can now be established for each $p$ [@Kastor; @Kastor1]. For this purpose, we should first note that equation (1) suggests an equivalent definition: a p-form is a KY p-form if and only if its symmetrized covariant derivatives vanish, that is, if and only if $$\begin{aligned}
i_{X}\nabla_{Y}\omega_{(p)}+
i_{Y}\nabla_{X}\omega_{(p)}=0\;,\nonumber\end{aligned}$$ is satisfied for all pair of the vector fields $X$ and $Y$. This is also equivalent to $i_{X}\nabla_{X}\omega_{(p)}=0$ and in particular every KY-form is divergent free, or equivalently co-closed, that is, $\delta \omega_{(p)}=-i_{X^{a}}\nabla_{X_{a}}\omega_{(p)}=0$. These statements imply that the covariant derivatives of KY forms are also totally anti-symmetric, with respect to the additional tensorial index. Hence the maximum numbers of linearly independent components and their first covariant derivatives are $C(n,p)$ and $C(n,p+1)$ respectively, where $C(n,p)$ stands for the binomial numbers.
On the other hand “the second covariant derivatives”, that is, the Hessian of KY forms can be written, in terms of the curvature 2-forms $R^{c}_{\;a}$ as $$\begin{aligned}
(\nabla_{X_{a}}\nabla_{X_{b}}-\nabla_{\nabla_{X_{a}}X_{b}})\omega_{(p)}
=\frac{1}{p} i_{X_{b}}(R^{c}_{\;a}\wedge
i_{X_{c}}\omega_{(p)})\;.\nonumber\end{aligned}$$ These imply that the value of any component of a KY-form at any point is entirely determined by the value of its first covariant derivative and by the value of the component itself at the same point. As a result the upper bounds for the numbers of linearly independent KY-forms is determined by the sum $C(n,p)+C(n,p+1)=C(n+1,p+1)$. In particular for $n=4$ the upper bounds are $10,10$ and $5$, respectively, for $p=1, p=2$ and $p=3$. These bounds are attained for the space-times of constant curvature.
The rest of the paper is structured in two main parts. In the next section, the defining equations for KY 2-forms and their integrability conditions are obtained. By the integrability conditions, the set of solutions naturally breaks up into cases and subcases that are considered in the rest of first part consisting five sections. The second part is entirely devoted to determination of KY 3-forms. The final Section VIII presents a summary of the results, and we briefly point out how to calculate the associated Killing tensors and related linear and quadratic first integrals for the considered class of space-times.
KY 2-Forms: Defining Equations, Integrability Conditions
========================================================
For a 2-form $$\begin{aligned}
\omega_{(2)}=\alpha e^{01}+\beta e^{02}+\gamma e^{03}+\delta
e^{12}+\epsilon e^{13}+\mu e^{23} \;,\nonumber\end{aligned}$$ there are two types of equations that result from KY-equation (1) in the case of $p=2$. The first consists of twelve relatively simple equations resulting from the left hand side of the KY equations. Two of them are $\alpha_t=0=\alpha_r$, which imply that $\alpha=\alpha(\theta,\varphi)$, and the other ten equations are as follows: $$\begin{aligned}
\beta_{t}&=&\frac{H'_0}{T H_1}\delta \;,\;\qquad \qquad
\beta_{\theta}=-\frac{H'_2}{H_1}\alpha\;,\nonumber\\
\delta_{r} &=& \dot{T}\frac{H_1}{H_0} \beta\;,\;\qquad \qquad
\delta_{\theta} =-\dot{T}\frac{H_2}{H_0}\alpha\;,\nonumber\\
\gamma_{t} &=&\frac{H'_0}{T H_1}\epsilon \;,\;\qquad \qquad
\gamma_{\varphi}=
-\frac{H'_2}{H_1}\sin\theta \alpha -\cos\theta\beta\;, \\
\epsilon_{r} &=& \dot{T}\frac{H_1}{H_0}\gamma\;,\;\qquad \qquad
\epsilon_{\varphi} =
-\dot{T}\frac{H_2}{H_0}\sin\theta \alpha -\cos\theta\delta\;,\nonumber\\
\mu_{\theta} &=&\dot{T} \frac{H_2}{H_0}\gamma
-\frac{H'_2}{H_1}\epsilon \;, \quad \mu_{\varphi} =-\dot{T}
\frac{H_2}{H_0} \sin\theta\beta +
\frac{H'_2}{H_1}\sin\theta\delta\;.\nonumber\end{aligned}$$ The second type also consists of twelve equations, but four of them are just a sum of two other equations of the same type. Hence, there are eight independent additional equations, six of which can be put in the following forms: $$\begin{aligned}
\alpha_{\theta}
&=&-\frac{H^{2}_{2}}{H_1}\partial_{r}\frac{\beta}{H_{2}}\;,\quad
\alpha_{\varphi}=- \frac{H_{2}^{2}}{H_1}\sin
\theta\partial_{r}\frac{\gamma}{H_{2}}\;,\nonumber\\
\partial_{r}\frac{\beta}{H_{0}}&=&-T^{2}\frac{H_{1}}{H^{2}_{0}}
\partial_{t}\frac{\delta}{T}\;,\quad
\gamma_{\theta} =-T^{2}\frac{H_{2}}{H_{0}}\partial_{t}\frac{\mu}
{T}\;,\\
\partial_{r}\frac{\gamma}{H_{0}}&=&-
T^{2}\frac{H_{1}}{H^{2}_{0}}\partial_{t}\frac{\epsilon}{T}\;,\quad
\epsilon_{\theta} =-\frac{H^{2}_{2}}{H_{1}}\partial_{r}\frac{\mu}
{H_{2}}\;.\nonumber\end{aligned}$$ The last two equations (presented below) do not depend on the metric coefficient functions, and therefore much of the essence of spherical symmetry is encoded in them: $$\begin{aligned}
\beta_{\varphi}
&=&-\sin^{2}\theta\partial_{\theta}\frac{\gamma}{\sin\theta}\;,
\quad\quad \delta_{\varphi}=-
\sin^{2}\theta\partial_{\theta}\frac{\epsilon}{\sin\theta}\;.\end{aligned}$$
Before solving these eighteen equations, there are some integrability conditions that must be met. Two of them, which enable us to find the solutions in a systematic way, are $$\begin{aligned}
\dot{T}H^{\prime}_{0}\alpha &=& 0\;,\\
\;[\rho(r)+\varrho(t)\;]\alpha &=& 0\;,\end{aligned}$$ where $$\begin{aligned}
\rho=\frac{H_{0}^{3}}{H_{1}H_{2}}(\frac{H_{2}^{\prime}}{H_{0}H_{1}})^{\prime}\;,\quad
\varrho=T\ddot{T}-\dot{T}^{2}=T^{2}\partial_t(\frac{\dot{T}}{T})\;.\end{aligned}$$ Condition (5) separately follows from each of $$\begin{aligned}
\beta_{t\theta}=\beta_{\theta t},\;\gamma_{t\varphi}=\gamma_{\varphi
t},\;\delta_{r\theta}=\delta_{\theta
r}\;,\;\epsilon_{r\varphi}=\epsilon_{\varphi r} \;,\nonumber\end{aligned}$$ and condition (6) can be checked from $\delta_{t\theta}=\delta_{\theta t}$ or $\epsilon_{\varphi
t}=\epsilon_{t\varphi}$. The integrability conditions imply that solutions must be investigated in two classes: (A) $\alpha\neq 0$ and (B) $\alpha=0$, such that the first consists of three important subclasses characterized by $$\begin{aligned}
(i)\;H_0^{\prime}=0\;,\;\;\dot{T}\neq 0\;,\quad (ii)\;
H_0^{\prime}\neq0\;,\;\;\dot{T}=0\;,\quad (iii)\;
H_0^{\prime}=0=\dot{T}\;.\nonumber\end{aligned}$$ In fact there are $6\times 6$ integrability conditions, some of which are trivially satisfied and the remaining ones will be considered in the following classes. We should also note that whenever $T$ or $H_{k}$’s are constant, they must be considered to be nonzero to keep the metric non-degenerate.
In the next three sections, we shall consider the first (A) case in which $\alpha$ is different from zero. At the outset of these sections, we should note the following two equations for $\alpha$: $$\begin{aligned}
\alpha_{\theta\theta}+\ell_{1}\alpha &=&0\;,\\
\alpha_{\varphi\varphi}+\ell_{1}\sin^{2}\theta
\alpha+\sin\theta\cos\theta\alpha_{\theta}&=&0\;,\end{aligned}$$ which do not change in the following subcases. These are obtained by differentiating the first two equations of (3) and by making use of the equation for $\beta_\theta$ and $\gamma_\varphi$ from (2). Here the constant $\ell_{1}$ is defined, in terms of $P=H_{2}^{\prime}/H_{1}H_2$, as follows: $$\begin{aligned}
\ell_{1}=-\frac{H_{2}^{2}}{H_{1}}P^{\prime}\;,\end{aligned}$$
We should finally note that, as there are no equations involving the first power of $\mu$, $$\begin{aligned}
\omega_{0}=TH_2e^{23} \;\nonumber\end{aligned}$$ is a solution of the KY-equation without any constraint. This observation means that all the space-times within the considered class of metrics admit of at least one KY 2-form. This fact was also observed for the static space-times in [@Howarth; @Collinson; @1]. In fact, it turns out that $\omega_{0}$ is the only KY 2-form admitted by two important cases, the Schwarzschild and the Reissner-Nordstr[ø]{}m space-times, which do not accept any KY 3-forms.
KY 2-Forms for $H_0^{\prime}=0,\;\dot{T}\neq0\;,\alpha\neq 0$
=============================================================
In the case of nonzero $\alpha$ and $H_0^{\prime}=0$, two equations of (2) show that $\beta$ and $\gamma$ are time independent and the integrability conditions (5-7) require that $\ell$, defined in terms of $P$ by $$\begin{aligned}
\frac{H_{0}^{2}}{H_{1}H_{2}}(H_{2}P)^{\prime}=-\ell=T^{2}\partial_t(\frac{\dot{T}}{T})\;,\end{aligned}$$ must be a constant. By multiplying both sides of the first equality by $H_2H_{2}^{\prime}$ it can easily be integrated to write $$\begin{aligned}
P^{2}=\frac{\ell_1}{H^{2}_2}-\frac{\ell}{H^{2}_0}\;,\end{aligned}$$ where $\ell_1$ is taken as an integration constant, since the defining relation (10) of $\ell_1$ is implied by the condition (12).
Then, in terms of $$\begin{aligned}
{\cal B}&=&\dot{T}\frac{H_2}{H_0}\beta\;,\quad
{\cal D}=H_2P\delta\;,\nonumber\\
{\cal G}&=&\dot{T}\frac{H_2}{H_0}\gamma\;,\quad {\cal
E}=H_2P\epsilon\;,\nonumber\end{aligned}$$ we define $$\begin{aligned}
x={\cal G}-{\cal E}\;,\quad y={\cal B}-{\cal D}\;.\end{aligned}$$ The first four equations appearing in the second column of (2) and two equations of (4) give the following equations for $x$ and $y$: $$\begin{aligned}
y_\theta=0\;,\quad x_\varphi=-\cos\theta y\;,\quad
y_\varphi=-\sin^{2}\theta\partial_\theta\frac{x}{\sin\theta}\;.\end{aligned}$$ These immediately imply $y_{\varphi\varphi}+y=0$ and hence $$\begin{aligned}
y=y_1\cos\varphi+y_2\sin\varphi\;,\quad x=\cos\theta
y_\varphi+y_3\sin\theta\;,\end{aligned}$$ where three $y_i$ functions depend on both $t$ and $r$, and the expression of $x$ is obtained by integration from (14).
### The general forms of $\gamma,\epsilon$ and $\mu$
In terms of $x$ and $y$, the last two equations of (2) are simply $\mu_\theta=x,\;\mu_\varphi=-y\sin\theta$, and therefore they specify the general form of $\mu$ as $$\begin{aligned}
\mu=y_\varphi\sin\theta-y_3\cos\theta+c_0TH_2\;,\end{aligned}$$ where $c_0$ is an integration constant. In specifying the last term of (16), we make use of the $\gamma_\theta$ and $\epsilon_\theta$-equations of (3), which transform to $$\begin{aligned}
\gamma_\theta=T^{2}\frac{H_2}{H_0}\partial_t\frac{x_\theta}{T}\;,\quad
\epsilon_\theta=\frac{H_2^{2}}{H_1}\partial_r\frac{x_\theta}{H_2}\;.\end{aligned}$$ Integration of these equations with respect to $\theta$ yield $$\begin{aligned}
\gamma=T^{2}\frac{H_2}{H_0}\partial_t\frac{x}{T}+g(r,\varphi)\;,\quad
\epsilon=\frac{H_2^{2}}{H_1}\partial_r\frac{x}{H_2}+\varepsilon(t,r,\varphi)\;.\end{aligned}$$ By substituting these expression into $x={\cal G}-{\cal E}$, we obtain $$\begin{aligned}
\varepsilon=\frac{\dot{T}}{H_0}\frac{g}{P}\;,\end{aligned}$$ and a set of three equations for $y_i$, which in terms of $Y_i=y_i/TH_2$ can be written as $$\begin{aligned}
Y_i=T\dot{T}(\frac{H_2}{H_0})^{2}Y_{it}-\frac{H_2^{2}P}{H_1}
Y_{ir}\;, \quad i=1,2,3\;.\end{aligned}$$ In writing equations (20), we have equated the coefficient functions of different trigonometric functions forming a basis, and relation (19) results from the fact that $x$ does not contain a term independent from $\theta$.
We now concentrate on specifying $g$ and $y_i$ functions. For this purpose we consider the coupled $\epsilon_r$-equation of (2) and $\gamma_r$-equation of (3) which, in terms of $P$ and $\ell$, give $$\begin{aligned}
\partial_r(\frac{g}{P})=H_1g\;,\quad
g_r=-\frac{\ell}{H_0^{2}}\frac{H_1}{P}g\;,\end{aligned}$$ in addition to two sets of equations for $y_i$ : $$\begin{aligned}
\partial_r(\frac{H^{2}_2}{H_1}Y_{ir})&=&T\dot{T}\frac{H_1H^{2}_2}{H^{2}_0}Y_{it}\;,\\
\partial_r(H^{2}_2Y_{it})&=&-H^{2}_2Y_{itr}\;.\end{aligned}$$ By dividing the first equation of (21) with $P$, it can be easily integrated to yield $$\begin{aligned}
g=H_2Pu_{\varphi}\;,\end{aligned}$$ where for convenience the last factor of $g$ has been written as the derivative of $u=u(\varphi)$. From the second equation of (21) we then obtain just one of the conditions of (11), when $u_{\varphi}\neq 0$.
After a slight rearrangement, the equation (23) can be integrated to find $$\begin{aligned}
Y_{i}=\frac{q_i(t)}{H_2}+\frac{z_i(r)}{H_2}\;\end{aligned}$$ The substitution of this relation into (20) and (22) yield $$\begin{aligned}
T\dot{T}\dot{q}_{i}-[\ell+(1-\ell_1)\frac{H^{2}_0}{H_2^{2}}]
q_i&=&[\ell+(1-\ell_1)\frac{H_{0}^{2}}{H_1}]z_i+H_{0}^{2}\frac{P}{H_1}z_{i}^{\prime}\;,\\
T\dot{T}\dot{q}_{i}-\ell q_i&=&\ell
z_i-\frac{H_{0}^{2}H_1^{\prime}}{H_1^{3}}z_i^{\prime}+\frac{H_0^{2}}{H^{2}_1}z_i^{\prime\prime}
\;,\end{aligned}$$ for $i=1,2,3$. As we are about to see at the beginning of next section, the constant $\ell_1$ must be $1$. Anticipating this result here, we see that the above equations are separable, and therefore each side of them must be a constant such that $$\begin{aligned}
T\dot{T}\dot{q}_{i}-\ell q_i&=&m_i\;,\nonumber\\
\ell
z_i-\frac{H_{0}^{2}H_1^{\prime}}{H_1^{3}}z_i^{\prime}+\frac{H_0^{2}}{H^{2}_1}z_i^{\prime\prime}
&=&m_i\;,\\
\ell z_i+\frac{H_{0}^{2}}{H_1}Pz_{i}^{\prime}&=&m_i\;.\nonumber\end{aligned}$$ It is not hard to see that the last two sets of (28) can be integrated to obtain $$\begin{aligned}
z_i=c_{i}H_{2}P-\tilde{c}_{i}\;,\quad i=1,2,3\end{aligned}$$ with $m_i=-\ell \tilde{c}_{i}$. The equations in the first line of (28) need not be integrated at this point, since they will be directly solved during the investigation of next subsection.
### The general forms of $\alpha,\beta$ and $\delta$
Relations found by (25) specify $y_i$ for $i=1,2,3$ as follows : $$\begin{aligned}
y_{i}=Tq_i(t)+Tz_i(r)\;.\end{aligned}$$ By noting that $(y_{i}/T)_{tr}=0$, the substitution of the expression of $\gamma$ given by (18) into the second equation of (3) yield $\alpha_\varphi=\ell_1\sin\theta u_\varphi$ in view of (24). This can easily be integrated to write $\alpha=\ell_1\sin\theta
u+f(\theta)$, and then by using this in equations (8) and (9) we obtain $$\begin{aligned}
\ell_1(\ell_1-1)\sin\theta u+f_{\theta\theta}+\ell_{1}f &=&0\;,\nonumber\\
\ell_1[u_{\varphi\varphi}+(\cos^{2}\theta+\ell_{1}\sin^{2}\theta)
u]+\ell_1\sin\theta f+\cos\theta f_{\theta}&=&0\;.\nonumber\end{aligned}$$ Since $u$ is a functions of $\varphi$, these relations imply that $\ell_1$ must be either $0$ or $1$. In fact, the former value is not compatible with the very definition of spherical symmetry, since it would lead to metric coefficient functions depending on some finite powers of angular coordinates. Indeed, equations (8) and (9) show that when $\ell_1=0$, $\alpha$ can be assumed to be a nonzero constant. But the equations in the second column of (2) then show that there is no way to get rid of the above mentioned angular dependence. Therefore from now on, we take $\ell_1=1$ by which the above two equations are reduced to the following forms : $$\begin{aligned}
f_{\theta\theta}+f=0\;,\quad u_{\varphi\varphi}+u+\sin\theta
f+\cos\theta f_{\theta}=0\;.\end{aligned}$$ Hence, in terms of integration constants $c_4,c_5,c_6, \tilde{c}_4$ and $$\begin{aligned}
v=v(\varphi)=c_5\cos\varphi+c_6\sin\varphi\;,\end{aligned}$$ we have $f=c_4\cos\theta+\tilde{c}_4\sin\theta$ and $u=v-\tilde{c}_4$ which implies $u_\varphi=v_\varphi$. These completely specify $\alpha$ as $$\begin{aligned}
\alpha=c_4\cos\theta+\sin\theta v\;.\end{aligned}$$
Using (33) in the second equation of (2) and the first equation of (3) leads us to $$\begin{aligned}
\beta=H_2P\alpha_\theta+H_2\bar{f}(\varphi)\;,\quad \delta=\dot{T}
\frac{H_2}{H_{0}}(\alpha_\theta+\frac{\bar{f}}{P})-\frac{y}{H_2P}
\;,\end{aligned}$$ where $\delta$ is obtained from the definition $y={\cal B}-{\cal
D}$. Substitutions of the solutions (33) and (34) into the $\gamma_\varphi$-equation of (2) yield, in terms of constants $c_7$ and $c_8$ $$\begin{aligned}
\bar{f}=\frac{v_1}{H_{0}}\;,\quad v_1=v_1(\varphi)=c_7\cos\varphi
+c_8\sin\varphi\;,\end{aligned}$$ and $\dot{q}_{1}=c_7/T^{2},\;\dot{q}_{2}=c_8/T^{2}$ since $\gamma$ is independent of time.
### KY 2-forms for time-dependent de Sitter space-time: $\ell\ne 0$ solutions
From the first set of equations of (28), $q_1$ and $q_2$ can be completely specified as $$\begin{aligned}
q_{1}=c_7\frac{\dot{T}}{\ell T}+\tilde{c}_1\;,\quad
q_{2}=c_8\frac{\dot{T}}{\ell T}+\tilde{c}_2\;,\end{aligned}$$ by recalling the relations $m_1=-\ell \tilde{c}_1$ and $m_2=-\ell
\tilde{c}_2$. In view of (29), (30) and (36), $y$ is also completely specified as $$\begin{aligned}
y=\frac{\dot{T}}{\ell}v_1+TH_2Pv_2\;,\end{aligned}$$ where we have defined $$\begin{aligned}
v_2=v_2(\varphi)=c_1\cos\varphi +c_2\sin\varphi\;.\end{aligned}$$
In view of (35) and (37), while $\beta$ and $\delta$ have been completely determined, there remains to determine $q_3(t)$ of the $y_3$-function $$\begin{aligned}
y_{3}=Tq_3(t)+T(c_{3}H_{2}P-\tilde{c}_{3})\;\end{aligned}$$ for complete specification of $\gamma,\epsilon$ and $\mu$. The resulting $\beta,\;\delta$ solutions identically satisfy the $\beta_r$-equation of (3) and $\delta_r$-equation of (2), and therefore they give nothing new. On the other hand, if (37) and (38) are used in the $\gamma$-expression of (18), we see that like $q_1$ and $q_2$, $q_3$ must be equal to $c_9(\dot{T}/\ell T)+\tilde{c}_3$ since $\gamma$ does not depend on $t$. The resulting $\gamma$ and $\epsilon$ also satisfy the $\epsilon_r$-equation of (2), as well as the definition $x={\cal G}-{\cal E}$. But, as they do not take part in any metric coefficient functions, the constants $\tilde{c}_1,\tilde{c}_2$ and $\tilde{c}_3$ become redundant. We are now ready to present all the coefficient functions together : $$\begin{aligned}
\alpha&=&c_4\cos\theta+\sin\theta v\;,\quad
\beta=H_2P\alpha_\theta+\frac{H_2}{H_0}v_1\;,\nonumber\\
\gamma&=& \frac{H_2}{H_{0}}(c_9\sin\theta+\cos\theta
v_{1\varphi})+H_2Pv_\varphi\;,\nonumber\\
\delta &=& \dot{T}
\frac{H_2}{H_{0}}\alpha_\theta-\frac{1}{\ell}\dot{T}H_2Pv_1-Tv_2 \;,
\\
\epsilon&=&-\sin\theta(c_9\frac{1}{\ell}\dot{T}H_2 P+
c_3T)-\cos\theta(\frac{1}{\ell}\dot{T}H_2Pv_{1\varphi}+Tv_{2\varphi})+\dot{T}\frac{H_2}{H_0}
v_\varphi\;,\nonumber\\
\mu&=&\sin\theta(\frac{1}{\ell}\dot{T}v_{1\varphi}+
TH_2Pv_{2\varphi})-\cos\theta(\frac{c_9}{\ell}\dot{T}+c_3TH_2P)+c_0TH_2\;,\nonumber\end{aligned}$$ which determine ten linearly independent KY $2$-forms, one for each $c_i,\;i=0,1,\ldots,9$: $$\begin{aligned}
\omega_{0}&=&TH_2e^{23}\;,\nonumber\\
\omega_{1}&=&-T\cos\varphi e^{12}+T\sin\varphi(\cos\theta e^{13}+H_2P\sin\theta e^{23})\;,\nonumber\\
\omega_{2}&=& -T\sin\varphi e^{12}-T\cos\varphi(\cos\theta e^{13}-H_2P\sin\theta e^{23})\;,\nonumber\\
\omega_{3}&=& -T(\sin\theta e^{13}+H_{2}P\cos\theta e^{23})
\;,\nonumber\\
\omega_{4}&=& \cos\theta e^{01}-H_2\sin\theta\Omega_1\wedge e^{2}\;,\\
\omega_{5}&=& \cos\varphi(\sin\theta
e^{01}+H_2\cos\theta\Omega_1\wedge e^{2})
-H_2\sin\varphi\Omega_1\wedge e^{3}\;,\nonumber\\
\omega_{6}&=& \sin\varphi(\sin\theta
e^{01}+H_2\cos\theta\Omega_1\wedge e^{2})
+H_2\cos\varphi\Omega_1\wedge e^{3}\;,\nonumber\\
\omega_{7}&=&H_2\cos\varphi \Omega_2\wedge
e^{2}-H_2\sin\varphi(\cos\theta\Omega_2\wedge
e^{3}+\frac{\dot{T}}{\ell}\sin\theta e^{23})\;,\nonumber\\
\omega_{8}&=&H_2\sin\varphi \Omega_2\wedge e^{2}+H_2\cos\varphi
(\cos\theta\Omega_2\wedge
e^{3}+\frac{\dot{T}}{\ell}\sin\theta e^{23})\;,\nonumber\\
\omega_{9}&=&H_2\sin\theta \Omega_2\wedge
e^{3}-\frac{\dot{T}}{\ell}\cos\theta e^{23} \;.\nonumber\end{aligned}$$ where the 1-forms $\Omega_1$ and $\Omega_2$ are defined, for the sake of simplicity, as $$\begin{aligned}
\Omega_1=Pe^{0}+\frac{\dot{T}}{H_0}e^{1}\;,\quad
\Omega_2=\frac{1}{H_0}e^{0}-\frac{\dot{T}}{\ell}Pe^{1}\;.\end{aligned}$$
### KY 2-forms for the usual form of de Sitter space-time: $\ell=0$ solutions
When $\ell$ is zero, the integrability conditions (11) and condition (10) for $\ell_1=1$ imply that $$\begin{aligned}
H_{2}P=\varepsilon\;,\quad P^{\prime}=-\frac{H_1}{H^{2}_2}\;,\quad
\dot{T}=\lambda T\;,\end{aligned}$$ where $\varepsilon$ and $\lambda$ are some nonzero constants. One can easily verify that the first two equations of (43) yield $\varepsilon^{2}=1$, therefore $\varepsilon=\pm 1$, and that the first relation implies the second one. The nine equations of (28) are then as follows for $i=1,2,3$ : $$\begin{aligned}
\dot{q}_{i}=\frac{m_i}{\lambda T^{2}}\;,\quad
z^{\prime}_i=\frac{m_i}{H^{2}_0}\frac{H_{1}}{P}\;,\quad
z_i^{\prime\prime}-\frac{H_1^{\prime}}{H_1}z_i^{\prime}=\frac{m_i}{H^{2}_0}H^{2}_1\;.\end{aligned}$$ The first two sets can be easily solved, such that $m_i=2c_iH^{2}_0$ and $$\begin{aligned}
q_{i}=-c_i\frac{H^{2}_0}{\lambda^{2}T^{2}}+a_i\;,\quad
z_i=c_iH^{2}_2+b_{i} \;,\end{aligned}$$ where $a_i$ and $b_i$ are new integration constants. These $z_i$ solutions identically satisfy the last relations of (44) without any extra condition.
By defining $d_i=a_i+b_i$ and $$\begin{aligned}
v_3=v_3(\varphi)=d_1\cos\varphi +d_2\sin\varphi\;,\end{aligned}$$ $y_i$ and hence $x,\;y$ functions can be explicitly written from (15) and (30) as $$\begin{aligned}
y_i&=&T(c_i S_{-}+ d_i)\;,\quad y=T(S_{-}v_2+v_3)\;,\\
x&=&T(S_{-}v_{2\varphi}+v_{3\varphi})\cos\theta +
T(c_3S_{-}+d_{3})\sin\theta\;,\end{aligned}$$ where $v_2$ is given by the relation (38) and $$\begin{aligned}
S_{\pm}=H^{2}_2\pm \frac{H^{2}_0}{\lambda^{2}T^{2}}\;.\end{aligned}$$ The substitution of $\alpha$ and $\beta$ given by (33) and (34) into the $\gamma_\varphi$-equation of (2) yields $$\begin{aligned}
\bar{f}(\varphi)=\frac{T^{2}}{H_0}\partial_t\frac{y}{T}=
2\frac{H_0}{\lambda}v_2 \;.\end{aligned}$$ Using (38),(47) and (48) in (18) and (34) give the following : $$\begin{aligned}
\alpha&=&c_4\cos\theta+\sin\theta v\;,\quad
\beta=H_2P\alpha_\theta+2\frac{H_0}{\lambda}H_2v_2\;,\nonumber\\
\gamma&=&2\frac{H_0}{\lambda}H_2(c_3\sin\theta+\cos\theta
v_{2\varphi}
)+H_2Pv_\varphi\;,\nonumber\\
\delta &=& \frac{\lambda}{H_{0}}TH_2\alpha_\theta+T\frac{1}{H_2P}
(S_{+}v_2-v_3)
\;,\\
\epsilon&=&\varepsilon TS_{+}(c_3\sin\theta +\cos\theta
v_{2\varphi})-\varepsilon T(d_3\sin\theta+\cos\theta
v_{3\varphi})+\frac{\lambda}{H_0}TH_2v_{\varphi}\;,\nonumber\\
\mu&=&T(S_{-}v_{2\varphi}+
v_{3\varphi})\sin\theta-T(d_3+c_3S_{-})\cos\theta+c_0TH_2\;,\nonumber\end{aligned}$$ which determine ten linearly independent KY $2$-forms, one for each $c_i,\;i=0,1,\ldots,6$ and $d_{j},\;j=1,2,3$. The corresponding KY 2-forms can be written as $$\begin{aligned}
\omega_{0}&=&TH_2e^{23}\;,\nonumber\\
\omega_{1}&=& 2\frac{H_0}{\lambda}H_2e^{0}\wedge
\Phi_{\varphi}+\varepsilon TS_{+}e^{1}\wedge\Phi-TS_{-}\sin\theta \sin\varphi e^{23}\;,\nonumber\\
\omega_{2}&=& 2\frac{H_0}{\lambda}H_2e^{0}\wedge \Phi-\varepsilon
TS_{+}e^{1}\wedge\Phi_{\varphi}+TS_{-}\sin\theta \cos\varphi e^{23}\;,\nonumber\\
\omega_{3}&=& 2\frac{H_0}{\lambda}H_2\sin\theta e^{03}+\varepsilon
TS_{+}\sin\theta e^{13}
-TS_{-}\cos\theta e^{23}\;,\nonumber\\
\omega_{4}&=& \cos\theta e^{01}-\sin\theta\Psi\wedge e^{2}\;,\\
\omega_{5}&=& \cos\varphi(\sin\theta e^{01}+\cos\theta\Psi\wedge
e^{2})
-\sin\varphi\Psi\wedge e^{3}\;,\nonumber\\
\omega_{6}&=& \sin\varphi(\sin\theta e^{01}+\cos\theta\Psi\wedge
e^{2})
+\cos\varphi\Psi\wedge e^{3}\;,\nonumber\\
\omega_{7}&=&-\varepsilon T\cos\varphi e^{12}+T\sin\varphi(\varepsilon \cos\theta e^{13}-\sin\theta e^{23})\;,\nonumber\\
\omega_{8}&=&-\varepsilon T\sin\varphi
e^{12}-T\cos\varphi(\varepsilon \cos\theta e^{13}-
\sin\theta e^{23})\;,\nonumber\\
\omega_{9}&=&-\varepsilon T\sin\theta e^{13}-T\cos\theta e^{23}
\;,\nonumber\end{aligned}$$ where the 1-forms $\Phi$ and $\Psi$ are defined, for the sake of simplicity, as $$\begin{aligned}
\Phi=\cos\varphi e^{2}-\cos\theta\sin\varphi e^{3}\;,\quad
\Psi=\varepsilon e^{0} +\frac{\lambda}{H_0}T H_2e^{1}\;,\end{aligned}$$ and $\Phi_\varphi=-(\sin\varphi e^{2}+\cos\theta\cos\varphi e^{3})$.
KY 2-Forms for $\dot{T}=0\;,\;H^{\prime}_{0}\neq 0\;, \alpha\neq 0$
====================================================================
In this case, condition (7) requires $$\begin{aligned}
H^{\prime}_{2}=kH_{0}H_{1}\;,\end{aligned}$$ such that $k$ is a nonzero constant and, in addition to $\alpha_\tau=0=\alpha_r$, we immediately have $\delta_r=0=\delta_\theta$ and $\epsilon_r=0$ from equations (2). Moreover, the $\epsilon_\varphi$-equation of (2) becomes independent of the metric characterizing functions, and leads to $\delta_{\varphi\varphi}+\delta=0$ when substituted into the $\varphi$-derivative of the second equation of (4). Therefore, $\delta$ is a harmonic function of $\varphi$ and when this fact is used in the integration of $\epsilon_\varphi=-\cos\theta \delta$, we obtain $$\begin{aligned}
\delta&=&D_1(\tau)\cos\varphi+D_2(\tau)\sin\varphi\;,\nonumber\\
\epsilon&=&\cos\theta\delta_{\varphi}+E(\tau)\sin\theta\;,\\
\mu
&=&-H_{2}P\sin\theta\delta_{\varphi}+E(\tau)H_2P\cos\theta+U(\tau,r)\;,\nonumber\end{aligned}$$ where we have also made use of the $\mu_\theta$ and $\mu_\varphi$-equations of (2). The $\epsilon_\theta$-equation then yields $P^{\prime}=-H_1/H^{2}_2$, which means that $\ell_1=1$ and $U=H_{2}u(\tau)$. Therefore, $\alpha$ is again given by (35) and $\beta$ can be written, from the second equation of (2) and the first equation of (3), as $\beta=H_2P\alpha_\theta+H_2B(\tau,\varphi)$. It remains to determine only the $\tau$ dependence of $\delta,\epsilon$ and $\mu$.
The $\beta_\tau$-equation of (2) and the third equation of (3) give $$\begin{aligned}
B_\tau=m\delta\;,\quad m_1B=-\delta_\tau\;,\end{aligned}$$ provided that $m$ and $m_1$ defined by $$\begin{aligned}
m=\frac{H_{0}^{\prime}}{H_{1}H_{2}}\;,\quad
m_{1}=(\frac{H_{2}}{H_{0}})^{\prime}\frac{H^{2}_{0}}{H_{1}}=kH_{0}^{2}-mH_{2}^{2}\;,\end{aligned}$$ are new constants, such that $m$ is supposed to be nonzero in this section. Thus, $\delta$ must satisfy $\delta_{\tau\tau}+mm_1\delta=0$, and therefore its coefficient functions can be obtained, depending on the value of $mm_1$, from $$\begin{aligned}
D_{i\tau\tau}+mm_{1}D_{i}=0\;,\quad i=1,2\;.\end{aligned}$$ From here on, the discussion proceeds in to ways: (A) $m_1\neq 0$ and (B) $m_1=0$.
KY 2-Forms of the static de Sitter space-time: $m_1\neq 0$ Solutions
---------------------------------------------------------------------
In this case, the $\gamma_\varphi$-equation of (2), the second equation of (3) and the first equation of (4) specify $\gamma$ as $$\begin{aligned}
\gamma=H_{2}Pv_\varphi-\frac{H_{2}}{m_1}\cos\theta\delta_{\tau\varphi}+H_2\sin\theta
g(\tau)\;.\end{aligned}$$ Only three equations remain unused so far; the $\gamma_\tau,\gamma_\theta$-equations and the equation $\partial_r(\gamma/H_0)=-H_1\epsilon_\tau/H_0^{2}$ of (3). One can easily check that the first two of these equations yield $$\begin{aligned}
g_\tau=mE\;,\quad g=-kE_\tau\;,\quad km_1=1 \;,\quad u_\tau=0\;,\end{aligned}$$ and the last equation gives nothing new. In accordance with the previous solutions, we take the constant $u$ as $u=c_0 T$. The first two equations of (60) also give $E_{\tau\tau}+mm_1E=0$. We can now write the the complete solutions of the case: $$\begin{aligned}
\alpha&=&c_4\cos\theta+\sin\theta v\;,\quad
\beta=H_2P\alpha_\theta-\frac{H_2}{m_1}\delta_\tau\;,\nonumber\\
\gamma&=&H_2Pv_\varphi-\frac{H_2}{m_1}(\cos\theta\delta_{\tau\varphi}+\sin\theta
E_{\tau})
\;,\nonumber\\
\delta &=& D_{1}\cos\varphi+D_2\sin\varphi\;,\\
\epsilon&=&\cos\theta \delta_{\varphi}+E\sin\theta\;,\nonumber\\
\mu&=&-H_{2}P\sin\theta\delta_\varphi+
EH_2P\cos\theta+c_0TH_2\;,\nonumber\end{aligned}$$ where for, say $mm_1=-w_0^{2}<0$, we have $$\begin{aligned}
D_1&=& a_{1}\cosh w_0\tau+a_{2}\sinh w_0\tau\;,\nonumber\\
D_2&=& a_{3}\cosh w_0\tau+a_{7}\sinh w_0\tau\;,\\
E&=& a_{8}\cosh w_0\tau+a_{9}\sinh w_0\tau\;.\nonumber\end{aligned}$$ When $mm_1=w^{2}_0>0$, it is enough to replace the hypergeometric functions of (62) by the corresponding trigonometric functions and the minus sign by a plus in the expression of $H_0^{2}$. The above solutions define ten linearly independent KY 2-forms $$\begin{aligned}
\omega_{0}&=&TH_2e^{23}\;,\nonumber\\
\omega_{1}&=& -\frac{w_0}{m_1}H_2\sinh w_0\tau e^{0}\wedge
\Phi+\cosh w_0\tau (e^1\wedge\Phi+H_{2}P\sin\theta\sin\varphi e^{23}) \;,\nonumber\\
\omega_{2}&=& -\frac{w_0}{m_1}H_2\cosh w_0\tau e^{0}\wedge
\Phi+\sinh w_0\tau (e^1\wedge\Phi+H_{2}P\sin\theta\sin\varphi e^{23})\;,\nonumber\\
\omega_{3}&=& \frac{w_0}{m_1}H_2\sinh w_0\tau e^{0}\wedge
\Phi_\varphi-\cosh w_0\tau (e^1\wedge\Phi_{\varphi}+H_{2}P\sin\theta\cos\varphi e^{23})\;,\nonumber\\
\omega_{4}&=& \cos\theta e^{01}-H_2 P\sin\theta e^{02}\;,\\
\omega_{5}&=& \sin\theta\cos\varphi
e^{01}+H_2P(\cos\theta\cos\varphi e^{02}
-\sin\varphi e^{03})\;,\nonumber\\
\omega_{6}&=& \sin\theta\sin\varphi
e^{01}+H_2P(\cos\theta\sin\varphi e^{02}
+\cos\varphi e^{03})\;,\nonumber\\
\omega_{7}&=&\frac{w_0}{m_1}H_2\cosh w_0\tau
e^{0}\wedge\Phi_\varphi- \sinh w_0\tau (e^{1}\wedge
\Phi_\varphi+H_2 P\sin\theta\cos\varphi e^{23})\;,\nonumber\\
\omega_{8}&=&-\frac{w_0}{m_1}H_2\sinh w_0\tau\sin\theta e^{03}+
\cosh w_0\tau (\sin\theta e^{1}+H_2 P\cos\theta e^{2})\wedge e^{3}\;,\nonumber\\
\omega_{9}&=&-\frac{w_0}{m_1}H_2\cosh w_0\tau\sin\theta e^{03}+
\sinh w_0\tau (\sin\theta e^{1}+H_2 P\cos\theta e^{2})\wedge e^{3}
\;,\nonumber\end{aligned}$$ corresponding, respectively, to $c_i,\;i=0,4,5,6$ and $a_j,\;j=1,2,3,7,8,9$. Here, $\Phi$ is given by (53) and the metric coefficient functions are as follows: $$\begin{aligned}
H_{0}^{2}=m_1^{2}-w^{2}_0H_{2}^{2}\;, \quad
H_1=m_1\frac{H^{\prime}_2}{H_0}\;.\end{aligned}$$
$m_1=0$ Solutions
-----------------
When $m_1$ is zero, we have $H_2=m_0H_0$ where $m_0$ is a nonzero constant and, by virtue of (54) and (57), $k=mm^{2}_0$. Therefore $k$ and $m$ have the same sign, and (54) and (57) amount to the same relation. In that case, the equations of (56) are of the forms $\delta_\tau=0$ and $B_\tau=m\delta$, which imply that $B=m\tau\delta+C_\varphi(\varphi)$ and $$\begin{aligned}
\beta=H_2(P\alpha_\theta+m\tau\delta+C_{\varphi})\;,\quad
\delta=b_1\cos\varphi+b_2\sin\varphi\;.\end{aligned}$$ where $b_i$’s are constants. The $\gamma_\varphi$-equation of (2) and the second equation of (3) provide us with $$\begin{aligned}
\gamma=H_2[Pv_\varphi+\cos\theta(m\tau\delta_\varphi-C(\varphi))+G(\tau,\theta)]\;,\end{aligned}$$ such that $\ell_1=1$. Thus, $\alpha$ is still given by (33) and $\epsilon,\mu$ are as in equations (55). For the above solutions, we have $\partial_r(\gamma/H_0)=0$ and the fifth equation of (3) implies that $\epsilon_\tau=0$, that is, $E$ is independent of time. On the other hand, the $\gamma_\tau$-equation of (2) and the first equation of (4) yield $$\begin{aligned}
G_\tau=mE\sin\theta\;,\quad
C_{\varphi\varphi}=-\sin^{2}\theta\partial_\theta\frac{G}{\sin\theta}\;.\nonumber\end{aligned}$$ We are finally left with the $\gamma_\theta$-equation, which gives $$\begin{aligned}
G_\theta=\sin\theta(m\tau\delta_\varphi-C)-\frac{H_2}{H_0}u_\tau\;.\nonumber\end{aligned}$$ As $G$ is independent of $\varphi$, the only possible solutions of the last three equations are $G_\theta=-c\sin\theta$ and $\delta_\varphi=0=u_\tau$ such that $C=c=$constant. Since $G_{\tau\theta}=G_{\theta\tau}$ implies $E=0$, we can write $\epsilon=0=\delta$ and $u=c_0T$. Although $G=c\cos\theta$ is a solution, $c$ does not take part in any component functions. The results can therefore be written as follows: $$\begin{aligned}
\alpha&=&c_4\cos\theta+\sin\theta v\;,\quad
\beta=H_2P\alpha_\theta\;,\nonumber\\
\gamma&=&H_2v_\varphi\;,\quad \delta=0=\epsilon\;,\quad \mu=
c_0TH_2\;,\nonumber\end{aligned}$$ which define four linearly independent KY 2-forms; $\omega_{0}$ and $$\begin{aligned}
\omega_{1}&=&\cos\theta e^{01}-mm_0H_{2}\sin\theta
e^{02} \;,\nonumber\\
\omega_{2}&=& -\cos\varphi\omega_{1\theta}-H_{2}\sin\varphi e^{03}\;,\nonumber\\
\omega_{3}&=& -\sin\varphi \omega_{1\theta}+H_{2}\cos\varphi
e^{03}\;.\nonumber\end{aligned}$$ (The corresponding 1-forms can be found from Section IV-A or V with $k_1=0$ of the first paper).
KY 2-Forms of The Flat Space-Time
=================================
In this section we take both $\dot{T}$ and $H_0^{\prime}$ to be zero, which imply $\beta_\tau=0=\gamma_\tau,\;\delta_r=0=\delta_\theta$ and $\epsilon_r=0$. Thus, as in the previous section, we have $$\begin{aligned}
\beta&=&H_2P\alpha_\theta+H_2B(\varphi)\;,\nonumber\\
\delta&=&D_1(\tau)\cos\varphi+D_2(\tau)\sin\varphi\;,\\
\epsilon&=&\cos\theta\delta_{\varphi}+E(\tau)\sin\theta\;,\quad\mu
=H_{2}(P\epsilon_{\theta}+u(\tau))\;,\nonumber\end{aligned}$$ provided that $P^{\prime}=-H_1/H^{2}_2$ and hence $\alpha$ is given by (33). On the other hand, the $\beta_r$-equation of (3) gives $$\begin{aligned}
(H_2P)^{\prime}=0\;,\quad
B=-\frac{H_1}{H_0H^{\prime}_2}\delta_\tau\;.\end{aligned}$$ The first relation is equivalent to the first integrability condition of (7), and when it is combined with $P^{\prime}=-H_1/H^{2}_2$, we obtain $P=\varepsilon/H_2$, that is $\varepsilon H_2^{\prime}=H_1$ with $\varepsilon=\pm 1$. Five equations remain untouched so far; the $\gamma_\varphi$-equation of (2), three equations of (3) involving $\gamma$, and the first equation of (4).
We first consider the $\gamma_\theta$-equation of (3) by inserting the $\mu$-solution of (67) into it. The resulting equation implies that $u$ must be a constant, which we take to be $u=c_0T$, and we are left with $$\begin{aligned}
\gamma_\theta=\frac{H^{2}_2P}{H_0}(\sin\theta\delta_{\tau\varphi}-E_\tau\cos\theta)\;.\end{aligned}$$ As $\gamma$ is independent of $\tau$, $E_\tau$ and $D_{j\tau}$ must be constants, and therefore $E=a_0\tau+b_0,\; D_1=b_1\tau+b_3$ and $D_2=b_2\tau+b_4$, where $a_i$ and $b_j$ are constants. Thus, in terms of $$\begin{aligned}
z_1(\varphi)=b_1\cos\varphi+b_2\sin\varphi\;,\quad
z_2(\varphi)=b_3\cos\varphi+b_4\sin\varphi\;,\end{aligned}$$ we can write $\delta=\tau z_1+z_2$, which implies that $B=-z_1/H_0H_2P$ by virtue of the second relation of (68). Therefore, (69) can be integrated to obtain the explicit expression of $\gamma$, presented together with the other solutions below: $$\begin{aligned}
\alpha&=&c_4\cos\theta+\sin\theta v\;,\quad
\beta=H_2P\alpha_\theta-\frac{1}{H_0P}z_1\;,\nonumber\\
\gamma&=& -\frac{H^{2}_2P}{H_0}(\cos\theta
z_{1\varphi}+a_0\sin\theta)+H_2Pv_{\varphi}\;,\nonumber\\
\delta&=&\tau z_1+z_2\;,\\
\epsilon&=&\tau(\cos\theta z_{1\varphi}+a_0\sin\theta)+\cos\theta
z_{2\varphi}+b_0\sin\theta\;,\nonumber\\
\mu&=&H_{2}(P\epsilon_{\theta}+c_0T)\;.\nonumber\end{aligned}$$ The last term of $\gamma$ which arises from the $\theta$-integration mentioned above is specified by the $\beta_\varphi$-equation of (4). It is straightforward to verify that the remaining three equations of $\gamma$ are identically satisfied. The corresponding KY 2-forms are, in addition to $\omega_{0}$, as follows: $$\begin{aligned}
\omega_{1}&=& -\varepsilon\frac{H_2}{H_0}(\cos\varphi
e^{02}-\cos\theta\sin\varphi e^{03})+\tau(\cos\varphi e^{12}
-\sin\varphi A\wedge e^{3})\;,\nonumber\\
\omega_{2}&=& -\varepsilon\frac{H_2}{H_0}(\sin\varphi
e^{02}+\cos\theta\cos\varphi e^{03})+\tau(\sin\varphi e^{12}
+\cos\varphi A\wedge e^{3})\;,\nonumber\\
\omega_{3}&=& \cos\varphi e^{12}-\sin\varphi A\wedge
e^{3}\;,\;\;\;\qquad
\omega_{4}= e^{0}\wedge A\;,\\
\omega_{5}&=& -\cos\varphi e^{0}\wedge A_\theta-\varepsilon\sin\varphi e^{03}\;,\nonumber\\
\omega_{6}&=& -\sin\varphi e^{0}\wedge
A_\theta+\varepsilon\cos\varphi e^{03}\;, \quad
\omega_{7}=\sin\varphi e^{12}
+\cos\varphi A\wedge e^{3}\;,\nonumber\\
\omega_{8}&=&-\varepsilon\frac{H_2}{H_0}\sin\theta e^{03}-\tau
A_\theta\wedge e^{3}\;,\;\;\quad \omega_{9}=-A_\theta\wedge e^{3}
\;,\nonumber\end{aligned}$$ where $A=\cos\theta e^{1}-\varepsilon \sin\theta e^{2}$.
Solutions For $\alpha=0$
========================
In this case the second and fourth equations of (2) give $\beta_{\theta}=0=\delta_{\theta}$ and $$\begin{aligned}
\beta=H_{2}B(t,\varphi)\;,\quad\gamma=H_{2}G(t,\theta,\varphi)\;,\end{aligned}$$ are implied by the first two equations of (3). Fourteen equations remain to be solved. But when $\alpha$ is set to zero, the $\gamma_\varphi$ and $\epsilon_\varphi$-equations of (2) are freed from metric coefficient functions, such that $\gamma_\varphi=-\cos\theta \beta$ and $\epsilon_\varphi=-\cos\theta
\delta$. When these are combined with two equations of (4), they considerably ease the investigation by providing general forms of the solutions. Indeed, differentiating both equations of (4) with respect to $\varphi$ gives $\beta_{\varphi\varphi}+\beta=0=\delta_{\varphi\varphi}+\delta$. Therefore, the general forms of $\beta,\gamma,\delta,\epsilon$ and $\mu$ can be written, in view of (73), as follows: $$\begin{aligned}
\beta &=&H_2(B_1(t)\cos\varphi+B_2(t)\sin\varphi)\;,\nonumber\\
\gamma &=&\cos\theta\beta_\varphi+G(t)H_2\sin\theta\;,\nonumber\\
\delta &=&D_1(t,r)\cos\varphi+D_2(t,r)\sin\varphi\;,\\
\epsilon &=&\cos\theta \delta_\varphi+E(t,r)\sin\theta\;,\nonumber\\
\mu&=&(\dot{T}\frac{H_2}{H_0}\beta_\varphi-
\frac{H^{\prime}_2}{H_1}\delta_\varphi)\sin\theta+U(t,r,\theta)\;,\nonumber\end{aligned}$$ where functions $B_i,D_i, G,E$ and $U$ are to be determined from the remaining nine equations: the five equations in the first column of (2) and the last four equations of (3). The second and fourth relations are obtained by integrations and then by using the results in equations (4). The last relation of (74) is obtained by first using $\beta$ and $\delta$ obtained in the last relation of (2) and then by integrating the resulting equation with respect to $\varphi$.
When the solutions (74) are substituted into the five equations appearing in the first column of equations (2), we obtain $$\begin{aligned}
\dot{B_1}&=&\frac{M}{T}D_1\;,\;\;\quad
\dot{B_2}=\frac{M}{T}D_2\;,\nonumber\\
D_{1r}&=&\dot{T}LB_1\;,\quad
D_{2r}=\dot{T}LB_2\;,\\
\dot{G}&=&\frac{M}{T}E\;,\qquad E_{r}=\dot{T}L G\;,\nonumber\\
U_{\theta}&=&(\dot{T}\frac{H_2^{2}}{H_0}G-\frac{H_2^{\prime}}{H_1}E)\sin\theta\;,\nonumber\end{aligned}$$ where we have defined the functions $$\begin{aligned}
M=\frac{H_{0}^{\prime}}{H_{1}H_{2}}\;,\quad
L=\frac{H_{1}H_{2}}{H_{0}}\;.\end{aligned}$$
A General Case
--------------
To be as general as possible, we shall first seek solutions for which both $\dot{T}$ and $H_0^{\prime}$ can be different from zero. In such a case, provided that $m$ defined by $$\begin{aligned}
m =\frac{M^{\prime}}{M^{2}L}\;\end{aligned}$$ is a constant, from the first four equations of (75) one can easily obtain $$\begin{aligned}
B_i=b_iK\;,\quad D_i=b_i\frac{\dot{K}}{K^{m}M}\;;\quad i=1,2\;,\end{aligned}$$ where $b_1$ and $b_2$ are integration constants, $K=T^{-1/m}$ and $m$ is supposed to be different from zero. In a similar way, the fifth and sixth equations of (75) yield $$\begin{aligned}
G=b_3K\;,\quad E=b_3\frac{\dot{K}}{K^{m}M}\;,\quad
U_\theta=-b_3\frac{\dot{K}R}{K^{m}M}\sin\theta\;,\end{aligned}$$ where $b_3$ is a constant, and we have defined $$\begin{aligned}
R=\frac{H_2}{H_1}(m\frac{H_0^{\prime}}{H_0}+\frac{H_2^{\prime}}{H_2})\;.\end{aligned}$$
In terms of $g=b_1\cos\varphi+b_2\sin\varphi$, the solutions can now be rewritten as $$\begin{aligned}
\beta &=&KH_2g\;,\quad\;\; \gamma=KH_2(g_\varphi\cos\theta+b_3\sin\theta)\;,\nonumber\\
\delta &=&\frac{\dot{K}}{K^{m}M}g\;,\quad
\epsilon=\frac{\dot{K}}{K^{m}M}
(g_\varphi\cos\theta+b_3\sin\theta)\;,\\
\mu
&=&-\frac{\dot{K}R}{K^{m}M}(g_\varphi\sin\theta-b_3\cos\theta)+u(t,r)\;,\nonumber\end{aligned}$$ where the integration of $U_\theta$ done with respect to $\theta$. There remain the last four equations of (3) that have not been used so far. Substitution of these solutions into the last four equations of (3) yield $$\begin{aligned}
\ddot{K}&=&-m_1K^{2m+1}=-m_2K^{2m+1}\;,\nonumber\\
\partial_t(\frac{u}{T})&=&0=\partial_r(\frac{u}{H_2})
\;,\\
\frac{H_1}{H_2^{2}}&=&-M(\frac{R}{H_2M})^{\prime}\;,\nonumber\end{aligned}$$ provided that $m_1$ and $m_2$ defined by $$\begin{aligned}
m_1=M\frac{H_0^{2}}{H_1}(\frac{H_2}{H_0})^{\prime}\;,\quad
m_2=M\frac{H_0}{R}\;,\end{aligned}$$ are constant. In fact, the first equality of (82) implies that $m_1=m_2$, and $$\begin{aligned}
\ddot{K}&=&-m_1K^{2m+1}\;,\quad
R\frac{H_0}{H_1}(\frac{H_2}{H_0})^{\prime}=1\;.\end{aligned}$$ The equations appearing in the second line of (82) give $u=TH_2$, and hence $\mu$ has been completely specified as $$\begin{aligned}
\mu=-\frac{\dot{K}R}{K^{m}M}(g_\varphi\sin\theta-b_3\cos\theta)+c_0TH_2\;.\end{aligned}$$ We have obtained five linearly independent KY 2-forms for a family of space-times characterized by two constants $m$ and $m_1$.
Having determined the coefficient functions of $\omega_{2}$, we now turn to the conditions which restrict the functions determining the metric tensor. In addition to two conditions given by (84), we have two more conditions defining the constants $m$ and $m_1$. The first, given by (77), can be integrated to yield $$\begin{aligned}
H_{0}^{\prime}=kH_{0}^{-1/m}H_{1}H_{2}\;\end{aligned}$$ and hence, $M=kH_0^{m}$. From the first equation of (83) and the second of (84), we find $$\begin{aligned}
R=\frac{k}{m_1}H_0^{m+1}\;,\quad
H_2^{\prime}=kH_0^{m-1}H_2^{2}+\frac{m_1}{k}H_1H_0^{-m-1}\;.\nonumber\end{aligned}$$ Substitutions of these into (73) finally yields $$\begin{aligned}
m_1H_2(m+\frac{1}{H_1})=H_0^{2}-(\frac{m_1}{k})^{2}H_0^{-2m}\;.\end{aligned}$$
Although several subclasses of space-times can be identified for particular values of $m$ and $m_1$, this general consideration will not be pursued any further. It will suffice to exhibit the physically important final case.
KY $2$-forms of the Robertson-Walker space-time
------------------------------------------------
The Robertson-Walker space-time is characterized by $$\begin{aligned}
H_0=1\;,\quad H_2=r\;,\quad H_1^{2}=\frac{1}{1+k_3r^{2}}\;,\end{aligned}$$ such that $T$ is specified by special cosmological models. Two such specifications are $T=t^{1/2}$ and $T=t^{2/3}$ which correspond, respectively, to radiation-dominated and matter-dominated universes. In this final subsection, we shall present KY 2-forms of this space-time such that there is no constraint on $T$. It turns out that such a case is possible only if we take $\alpha=0=\beta=\gamma$ and $H^{\prime}_0=0$. The solutions then are $$\begin{aligned}
\delta &=&T(c_1\cos\varphi+c_2\sin\varphi)\;,\nonumber\\
\epsilon &=&\cos\theta \delta_\varphi+c_3 T\sin\theta \;,\\
\mu
&=&-H_2P\sin\theta\delta_\varphi+T(c_3H_2P\cos\theta+c_0H_2)\;,\nonumber\end{aligned}$$ provided that $P^{\prime}=-H_1/H_2^{2}$, which leads to $H_1$ of (94). The above solutions provide us, in addition to $\omega_0$, with three linearly independent KY 2-forms: $$\begin{aligned}
\omega_1 &=&T(\cos\varphi e^{12}-\cos\theta\sin\varphi e^{13})+H_2P\sin\theta\sin\varphi e^{23}\;,\nonumber\\
\omega_2 &=&T(\sin\varphi e^{12}+\cos\theta\cos\varphi e^{13})-H_2P\sin\theta\cos\varphi e^{23}\;,\\
\omega_3 &=&T(\sin\theta e^{13}+H_2P\cos\theta e^{23})\;.\nonumber\end{aligned}$$
KY 3-Forms
==========
For a 3-form $$\begin{aligned}
\omega_{(3)}=\alpha e^{012}+\beta e^{013}+\gamma e^{023}+\delta
e^{123}\;,\nonumber\end{aligned}$$ the KY-equation gives sixteen equations, five of which have the following simple forms: $$\begin{aligned}
\alpha_{t}=0=\beta_{t}\;,\quad \alpha_{r}=0=\beta_{r}\;,\quad
\alpha_{\theta}=0\;.\nonumber\end{aligned}$$ These imply that $\alpha$ depends only on $\varphi$, and $\beta$ is a function of $\theta$ and $\varphi$. Seven of the remaining equations have the following two-term forms: $$\begin{aligned}
\beta_{\varphi}&=&- \cos\theta \alpha\;,\nonumber\\
\gamma_{t}&=&\frac{H'_0}{T H_1}\delta\;,\quad \gamma_{\theta}
=-\frac{H'_2}{H_1}\beta \;,\; \quad \gamma_{\varphi} =
\frac{H'_2}{H_1}\sin\theta\alpha\;,\\
\delta_{r}&=& \dot{T}\frac{H_1}{H_0}\gamma\;,\quad \delta_{\theta}=-
\dot{T}\frac{H_2}{H_0}\beta\;,\quad \delta_{\varphi} =
\dot{T}\frac{H_2}{H_0}\sin\theta\alpha\;.\nonumber\end{aligned}$$ The last four equations give three independent equations, which can be written as: $$\begin{aligned}
\alpha_{\varphi}=-\sin^{2}\theta\partial_{\theta}\frac{\beta}{\sin\theta}\;,
\quad
\beta_{\theta}=-\frac{H_{2}^{2}}{H_{1}}\partial_{r}\frac{\gamma}{H_{2}}\;,\quad\
\partial_{r}\frac{\gamma}{H_{0}}
=-T^{2}\frac{H_{1}}{H^{2}_{0}}\partial_{t}\frac{\delta}{T}\;.\end{aligned}$$
Two of the most important integrability conditions for these equations are as follows: $$\begin{aligned}
\dot{T}H^{\prime}_{0}\beta = 0\;,\quad
[\rho(r)-\varrho(t)]\beta=0\;,\end{aligned}$$ where $\rho$ and $\varrho$ are given by (7). The first follows from $\gamma_{t\theta}=\gamma_{\theta t}$ and $\delta_{r\theta}=\delta_{\theta r}$, and the second from $\delta_{t\theta}=\delta_{\theta t}$. There are also identical conditions with $\beta$ replaced by $\alpha$ that can be checked from $\gamma_{t\varphi}=\gamma_{\varphi t}\;,
\delta_{r\varphi}=\delta_{\varphi r}$ and $\delta_{t\varphi}=\delta_{\varphi t}$. But noting that $\beta=0$ implies $\alpha=0$, we see that the above conditions include the second (see also relations (94)). There are also some other conditions which should be considered in investigating the cases implied by the above conditions. Therefore, we shall present the general solutions in two classes: (A) $\beta\neq 0$ and (B) $\beta=0$.
The essence of spherical symmetry seems to be encoded in the first equations of (91) and (92), for they do not depend on the metric coefficient functions. We are thus able to start with their general solutions $$\begin{aligned}
\alpha= a_1\cos\varphi+a_2 \sin\varphi\;,\quad \beta =
\cos\theta\alpha_\varphi+a_3\sin\theta\;,\end{aligned}$$ which can easily be verified. Here, $a_i$ are integration constants.
Solutions for $\beta\neq0$
--------------------------
For the fulfilment of conditions (93) in the case of nonzero $\beta$, two sets of conditions must be distinguished: $$\begin{aligned}
{\bf (i)}: \;\;\quad \dot{T} &=& 0\;,\quad
H^{\prime}_{2}=kH_{0}H_{1}\;,\nonumber\\
{\bf (ii)}: \quad H_{0}^{\prime} &=& 0\;,\quad
\dot{T}^{2}-T\ddot{T}=-\ell=\frac{H_{0}^{2}}{H_{1}H_{2}}(H_2P)^{\prime}\;,\nonumber\end{aligned}$$ Here $k$ and $\ell$ are constants such that $k\neq 0$. The well-known maximal symmetric Minkowski and the static form of de Sitter space-times, each having five independent KY $3$-forms are obtained among the ${\bf (i)}$ solutions as special cases. On the other hand, four time dependent forms plus the most well-known form of de Sitter and Robertson-Walker space-times, emerge in the second case. We should emphasize the fact that the former space-time is obtained without any restriction on $T$, which is obtained by taking $H_0^{\prime}$ zero and by starting in such a way that the last two conditions of [**(ii)**]{} are avoided.
### KY 3-Forms for the Minkowski and Static Form of de Sitter Space-time
In the [**(i)**]{} case we have, in addition to $\delta=f(\tau)$, the following five equations: $$\begin{aligned}
\gamma_{\tau} &=&\frac{H'_0}{H_1}f\;,\quad\qquad \gamma_{\theta}
=-kH_0\beta\;,\;\quad \gamma_{\varphi} =
kH_{0}\sin\theta\alpha\;,\nonumber\\
&&\\
f_{\tau}&=&-\frac{H^{2}_{0}}{H_{1}}\partial_{r}\frac{\gamma}{H_{0}}\;,\;\;
\beta_{\theta}=-\frac{H_{2}^{2}}{H_{1}}\partial_{r}\frac{\gamma}{H_{2}}\;.\nonumber\end{aligned}$$ where $\tau=t/T$. Provided that $m$ is a nonzero constant such that $$\begin{aligned}
H_{2}H_{0}^{\prime}-H_{0}H_{2}^{\prime}=mH_{1}\;,\end{aligned}$$ the last two equations of (95) imply that $m\gamma$ must be equal to $H_{2}f_{\tau}-H_{0}\beta_{\theta}$. The first and second equations of (95) in this case require $km=-1$ and $f_{\tau\tau}+m_{1}f = 0$, where the constant $m_1$ is defined by $$\begin{aligned}
H_{0}^{\prime}&=&km_{1}H_{1}H_{2}\;.\end{aligned}$$ The third equation of (95) is then identically satisfied.
As an alternative approach one can first integrate the $\gamma_{\theta}$-equation with respect to $\theta$, and then use it in the $\gamma_{\varphi}$-equation to find $\gamma
=kH_{0}\beta_{\theta}+G(\tau,r)$. The remaining three equations then give the same solution. As a result, under the ${\bf (i)}$ conditions the general forms of the coefficient functions for KY $3$-form are, in addition to that given by (94), as follows: $$\begin{aligned}
\gamma =-k(f_{\tau}H_{2}-H_{0}\beta_{\theta})\;,\quad \delta =
f(\tau)\;.\end{aligned}$$
Depending on the value of $m_{1}$, one can easily write the explicit form of $f$. For $m_1=0$ we have, in terms of integration constants $a,b$, $$\begin{aligned}
f=a+b\tau\;,\quad H_{2}^{\prime}=kH_0 H_1\;.\end{aligned}$$ and $H^{2}_{0}=k^{-2}$ from equations (96) and (97). For nonzero $m_1$, we have $$\begin{aligned}
H^{2}_{0}=k_0+m_1H_{2}^{2}\;,\quad H_0 H_1=k_0
kH_{2}^{\prime}\;,\quad k_0 k^{2}=1\;\end{aligned}$$ and $f$ is as follows ($k_0,b_1$ and $b_2$ are integration constants): $$\begin{aligned}
f=\left\{
\begin{array}{cc}
b_1\cos\omega_0\tau+b_2\sin\omega_0\tau\;; & \quad m_1=\omega_0^{2}>0\;, \\
b_1\cosh\omega_0\tau+b_2\sinh\omega_0\tau\;, &\quad
m_1=-\omega_0^{2}<0\;,
\end{array}
\right.\end{aligned}$$ In any case, we have five linearly independent $3$-forms.
For $T=1,\; H_0=1=H_1$ and $H_2=r$ we get, from the equations (94), (97) and (99), 3-forms of Minkowski space-time: $$\begin{aligned}
\omega_1 &=& w_1 +\sin\theta \sin\varphi e^{023}\;,\quad \omega_2 =
w_2
-\sin\theta \cos\varphi e^{023}\;,\nonumber\\
\omega_3 &=& \sin\theta e^{013}+\cos\theta e^{023}
\;,\\
\omega_4 &=& e^{123}\;,\quad\omega_5 =-r e^{023}+\tau e^{123}\;,
\nonumber\end{aligned}$$ where the $3$-forms $w_1$ and $w_2$ are defined, for brevity, as $$\begin{aligned}
w_1 = \cos\varphi e^{012}-\cos\theta \sin\varphi e^{013}\;,\quad w_2
= \sin\varphi e^{012}+\cos\theta \cos\varphi e^{013} \;.\end{aligned}$$ Note that $k=1$ (hence $m=-1$) and $m_1=0$ for this case. On the other hand, for the values $$\begin{aligned}
T=1\;,\quad H_2=r\;,\quad k_0=1=k \;\end{aligned}$$ we obtain, by virtue of equations (94), (97) and (100), the KY $3$-forms $$\begin{aligned}
\omega_1 &=& w_1 +H_0\sin\theta \sin\varphi e^{023}\;,\quad \omega_2
= w_2
-H_0\sin\theta \cos\varphi e^{023}\;,\nonumber\\
\omega_3 &=& \sin\theta e^{013}+H_0\cos\theta e^{023} \;,\quad
\omega_4 =r\sinh\tau e^{023}+\cosh\tau e^{123}\;,\\
\omega_5 &=&-r\cosh\tau e^{023}+\sinh\tau e^{123}\; \nonumber\end{aligned}$$ for the static form of de Sitter space-time, specified also by $H_1^{2}=1+m_1r^{2}$ and $H_0H_1=1$.
### Solutions for Four Time-Dependent Forms of de Sitter Space-time
From here on, we consider the ${\bf (ii)}$ conditions. The condition $H_{0}^{\prime}=0$ gives $\gamma_{t}=0$ and leaves us with seven equations to be solved. It is easy to integrate the $\gamma_\theta$-equation of (91) and then use it in the $\gamma_\varphi$-equation to find $\gamma=H_2P\beta_\theta+G(r)$. By substituting this solution into the $\beta_\theta$-equation, we find $G=cH_2$ and hence, $\gamma=H_2P\beta_\theta+cH_2$, provided that $P^{\prime}=-H_{1}/H_{2}^{2}$. Here, $c$ is an integration constant. The $\delta_\theta$-equation can also be integrated with respect to $\theta$, and then by substituting the solution into $\delta_\varphi$-equation, we find $$\begin{aligned}
\delta=\dot{T}\frac{H_2}{H_0}\beta_\theta+D(t,r)\;.\end{aligned}$$
The following two equations of (91) and (92) remain to be solved: $$\begin{aligned}
\delta_{r} =\dot{T}\frac{H_{1}}{H_{0}}\gamma\;,\;\quad
\frac{H_{0}}{H_{1}}\gamma_{r} =
-T^{2}\partial_{t}\frac{\delta}{T}\;.\nonumber\end{aligned}$$ By substituting the above $\gamma$ solution and (106) into these equations, we obtain $$\begin{aligned}
D_{r}=c\dot{T}\frac{H_{1}H_2}{H_{0}}\;,\quad
T^{2}\partial_{t}\frac{D}{T}=-cH_{0}H_2P\;,\end{aligned}$$ provided that $$\begin{aligned}
(H_2P)^{\prime}=-\ell \frac{H_{1}H_2}{H_{0}^{2}} \;,\end{aligned}$$ which is just one of the integrability conditions of [**(ii)**]{}. Now it is not difficult to see that for nonzero values of $\ell$, the general solution for $D$ is $D=-(c/\ell)\dot{T}H_{0}H_2P+c_0T$, where $c_0$ is another integration constant. We can now collate the solutions of the case as follows: $$\begin{aligned}
\alpha &=& a_1\cos\varphi+a_2 \sin\varphi\;,\quad\beta=
\cos\theta\alpha_\varphi+a_3\sin\theta
\;,\nonumber\\
\gamma &=&H_2P\beta_\theta+cH_2\;,\\
\delta
&=&\dot{T}\frac{H_2}{H_0}\beta_\theta-\frac{c}{\ell}\dot{T}H_{0}H_2P+c_0T\;.\nonumber\end{aligned}$$ Depending on the values of $\ell$ and other integration constants arising when integrating the equation $T^{2}\partial_t(\dot{T}/T)=\ell$, four different time regimes were presented in Part I (see relations (72) in I). The above solutions provide five independent KY $3$-forms: $$\begin{aligned}
\omega_1 &=& \cos\varphi e^{012}-\cos\theta\sin\varphi
e^{013}+H_2\sin\theta\sin\varphi B\;,\nonumber\\
\omega_2 &=& \sin\varphi e^{012}+\cos\theta\cos\varphi
e^{013}-H_2\sin\theta\cos\varphi B\;,\nonumber\\
\omega_3 &=& \sin\theta e^{013}+H_2\cos\theta B \;,\\
\omega_4 &=&H_2(e^{023}-\frac{H_0}{\ell}\dot{T}Pe^{123})\;,\nonumber\\
\omega_5 &=&T e^{123}\;, \nonumber\end{aligned}$$ where we have defined the 3-form $B=Pe^{023}+H^{-1}_0\dot{T}e^{123}$.
### KY 3-Forms of de Sitter Space-Time with Exponential Time Dependence
When $\ell=0$, we have $\dot{T}=\lambda T$ and $H_{2}^{\prime}=m_0
H_1$ from (108). In such a case, the most general solution of (107) turns out to be $$\begin{aligned}
D=c\frac{\lambda T}{2m_0H_0}[H_{2}^{2}+(\frac{m_0H_0}{\lambda
T})^{2}]+c_1\frac{1}{T}\;,\nonumber\end{aligned}$$ where $c_1$ is an integration constant. The solutions are then given by (109), with $\delta$ replaced by $\delta=\dot{T}(H_2/H_0)\beta_\theta+D$. From $P^{\prime}=-H_1/H_2$, it follows that $m_0^{2}=1$, that is, $m_0=\varepsilon=\pm1$ and we again obtain five linearly independent $3$-forms for de Sitter space-time. The first three forms are the same as those given in (110), and the last two are as follows: $$\begin{aligned}
\omega_4 =H_2e^{023}+\frac{\lambda
T}{2m_0H_0}[H_{2}^{2}+(\frac{m_0H_0}{\lambda T})^{2}]e^{123}\;,\quad
\omega_5 =\frac{1}{T} e^{123}\;.\end{aligned}$$
### KY 3-form of the Robertson-Walker Space-Time
For the solutions obtained so far under ${\bf (ii)}$ conditions, $T$ is restricted as a special function of time. It turns out (see also the next section) that the only possible solution in which there are no constraints on $T$ is $$\begin{aligned}
\omega=Te^{123}\;,\end{aligned}$$ provided that $H_0^{\prime}=0$. In particular, there is only one KY 3-form for the Robertson-Walker space-time.
Solutions for $\beta=0$
-----------------------
For $\beta=0$, the equations (91) and (92) imply that $\alpha=0,\;\gamma=H_{2}f(t)$ and $\delta=D(t,r)$, and the following three equations $$\begin{aligned}
T\dot{f}&=&\frac{H_{0}^{\prime}}{H_{1}H_2}D\;,\nonumber\\
\dot{T}f&=&\frac{H_{0}}{H_{1}H_2}D_r\;,\\
\frac{T^{2}}{f}\partial_t\frac{D}{T}&=&
-\frac{H_{0}^{2}}{H_{1}}(\frac{H_2}{H_0})^{\prime}\;,\nonumber\end{aligned}$$ determine $f$ and $D$. If $H_{0}^{\prime}$ is zero, then $f$ is a constant, say $c$, and we get $\gamma=cH_{2}$ and $\delta=-(c/\ell)\dot{T}H_0H_2P+c_0T$, which are special cases of solutions (109). Moreover, the special case $c=0$ produces the solution (112) for the Robertson-Walker space-time.
For nonzero $H_0^{\prime}$, taking $D$ from the first equation of (113) and substituting it into the other two equations lead us to two separate equations. Each side of these equations must be constant such that $$\begin{aligned}
\frac{H_{0}}{H_{1}H_{2}}(\frac{H_{1}H_2}{H_{0}^{\prime}})^{\prime}=m_1\;,\quad
\frac{H_{0}^{2}H_{0}^{\prime}}{H^{2}_{1}H_2}(\frac{H_2}{H_0})^{\prime}=m_2\;.\end{aligned}$$ This leaves us with two simple equations for $f$: $$\begin{aligned}
\frac{\dot{T}f}{T\dot{f}} =m_1\;,\quad T^{2}\frac{\ddot{f}}{f}=
-m_2\;.\end{aligned}$$ For $m_1=0$, we have $\dot{T}=0$ and $$\begin{aligned}
f_{\tau\tau}+m_2f=0\;,\quad H_{0}^{\prime}=m_{3}H_{1}H_{2}\;,\quad
D=m_{3}^{-1}f_{\tau}.\end{aligned}$$ where $m_3$ is a nonzero constant. These provide two independent $3$-forms: $$\begin{aligned}
\omega_1 &=& \tau H_2e^{023}+\frac{1}{m_3}e^{123}\;,\quad \omega_2 =
H_2e^{023}\;,\nonumber\end{aligned}$$ for $m_2=0$ and $$\begin{aligned}
\omega_1 &=& H_2\cosh w_0\tau
e^{023}+\frac{w_0}{m_3}\sinh w_0\tau e^{123}\;,\nonumber\\
\omega_2 &=& H_2\sinh w_0\tau e^{023}+\frac{w_0}{m_3}\cosh w_0\tau
e^{123}\;,\nonumber\end{aligned}$$ for $m_2=w_0^{2}>0$. For nonzero values of $m_1$, the first equation of (115) can be easily integrated to yield $f=c_1 T^{1/m_1}$, and by substituting this into the second equation, we get $\ddot{K}=-m_2K^{1-2m_1}$, where $K=T^{1/m_1}$. As long as this last condition and that given by (114) are satisfied, the general solutions which define only one KY 3-form are as follows: $$\begin{aligned}
\gamma=c_{1}H_{2}T^{1/m_1}\;,\quad
\delta=\frac{c_{1}}{m_1}\dot{T}T^{1/m_1}\frac{H_1H_2}{H_{0}^{\prime}}\;.\nonumber\end{aligned}$$
Conclusion
==========
By directly starting from the KY-equation, we have developed a constructive method which makes it possible to generate all KY forms for a large class of spherically symmetric space-times in a unified and exhaustive way. Our results for the well-known spherically symmetric space-times are quantitatively summarized in Table I of the first paper and their KY two and three forms are computed in this second paper. In particular, we have found an exactly solvable nonlinear time equation for de Sitter type space-times which enables us to generate all of their KY forms in a unified manner. We have also reported solutions in some detail for sufficiently symmetric new cases which fall within the considered class of metrics.
Our results can be used to reach decisive, or at least conclusive statements in analyzing the algebraic structures of KY-forms [@Gibbons; @Kastor1; @Cariglia], in specifying of the symmetry algebra and related conserved quantities of the Dirac as well as other equations in spherically symmetric curved backgrounds [@Benn-Charlton; @Benn-Kress1; @Benn-Kress2; @Benn-Kress3; @Cebeci]. Finally, as an application, we indicate an approach for calculating Killing tensors and associated first integrals for the considered class of space-times [@Benn4]. As has been mentioned before, to each KY (p+1)-form $\omega$, there corresponds an associated Killing tensor $K$ that can be defined by $K(X,Y)=g_{p}(i_X\omega,i_Y\omega)$, where $g_{p}$ is the compatible metric in the space of $p$-forms induced by the space-time metric $g$. Then $i_{\dot{\gamma}}\omega$ is parallel-transported along the affine-parameterized geodesic $\gamma$ with tangent field $\dot{\gamma}$, and $K(\dot{\gamma},\dot{\gamma})$ is the associated quadratic first integral. The first statement follows from the fact that the covariant and interior derivatives with respect to the same geodesic tangent field commute and the second statement follows from the fact that the cyclicly permuted sum of $\nabla_{X}K(Y,Z)$ vanishes. In particular, Killing tensor fields and associated first integrals for the space-times given in the Table I of I can be computed and used in investigating some integrability problems. Our study on the symmetries of the Dirac equation and related matter, is in progress, and soon will be reported elsewhere [@ozumav2].
This work was supported in part by the Scientific and Technical Research Council of Turkey (TÜBİTAK).
Ö. Aç[i]{}k, Ü. Ertem, M. Önder and A. Verçin, Killing-Yano Forms of a Class of Spherically Symmetric Space-Times I : A Unified Generation of Killing Vector Fields.
I. M. Benn and R. W. Tucker, *An Introduction to Spinors and Geometry with Applications in Physics*, IOP Publishing Ltd, Bristol, 1987.
U. Semmelmann, *Math. Z.* **243** (2003), 503.
S. E. Stepanov, *Theor. Math. Phys.* **134** (2003), 333.
D. Kastor and J. Traschen, *JHEP* [**0408**]{} (2004), 045 \[arXive:hep-th/0406052\].
D. Kastor, S. Ray and J. Traschen, *Class. Quantum Grav.* [**24**]{} (2007), 3759.
L. Howarth and C. D. Collinson, *Gen. Rel. Grav.* **32** (2000), 1845.
C. D. Collinson and L. Howarth, *Gen. Rel. Grav.* **32** (2000), 1767.
G. W. Gibbons, R. H. Rietdijk and J. W. von Holten , *Nucl. Phy. B* [**404**]{} (1993), 42.
M. Cariglia, *Class. Quantum. Grav.* [**21**]{} (2004), 1051.
I. M. Benn and P. Charlton, *Class. Quantum Grav.* **14** (1997), 1037.
I. M. Benn and J. Kress, *Class. Quantum Grav.* **21** (2004), 427.
I. M. Benn and J. M. Kress, Symmetry operators for the Dirac and Hodge-deRham equations, *Proceedings of the 9th DGA Conference*, Prague, August 30 - September 3, 2004, pp. 421-430.
I. M. Benn, P. R. Charlton and J. Kress, *J. Math. Phys.* **38** (1997), 4504.
H. Cebeci, O. Sarioglu and B. Tekin, *Phys.Rev. D.* [**74**]{} (2006), 124021 \[arXive:hep-th/0611011\].
I. M. Benn, *J. Math. Phys.* **47** (2006), 022903.
Ö. Aç[i]{}k, Ü. Ertem, M. Önder and A. Verçin, Symmetries of Dirac equation in curved space-times (in preparation).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
The independence polynomial of a graph $G$ is $$I(G,x)=\sum\limits_{k\ge 0}i_k(G)x^k,$$ where $i_k(G)$ denotes the number of independent sets of $G$ of size $k$ (note that $i_0(G)=1$). In this paper we show a new method to prove real-rootedness of the independence polynomials of certain families of trees.
In particular we will give a new proof of the real-rootedness of the independence polynomials of centipedes (Zhu’s theorem), caterpillars (Wang and Zhu’s theorem), and we will prove a conjecture of Galvin and Hilyard about the real-rootedness of the independence polynomial of the so-called Fibonacci trees.
address: |
Central European University, Department of Mathematics\
H-1051 Budapest\
Zrinyi u. 14, Third Floor\
Hungary & Alfréd Rényi Institute of Mathematics\
H-1053 Budapest\
Reáltanoda u. 13-15.
author:
- Ferenc Bencs
bibliography:
- 'hivatkozat.bib'
title: On trees with real rooted independence polynomial
---
[^1]
Introduction
============
The independence polynomial of a graph $G$ is $$I(G,x)=\sum\limits_{k\ge 0}i_k(G)x^k,$$ where $i_k(G)$ denotes the number of independent sets of $G$ of size $k$ (note that $i_0(G)=1$). In this paper we study the independence polynomials of trees. For trees, it is a well known conjecture that the sequence $(i_k(T))_{k\ge 0}$ is unimodal.
Recall that a sequence $(b_k)_{k= 0}^{n}$ is unimodal ([@Stanley1989]), if there exists an index $k$, such that $$b_0\le b_1\le\dots\le b_{k-1}\le b_k\ge b_{k+1}\ge \dots \ge b_{n}.$$ A stronger property for positive sequences is the so called log-concavity: for any $0< i< n$ we have $b_i^2\ge b_{i-1}b_{i+1}$. An even stronger property is the real-rootedness of the polynomial $p(x)=\sum_{i=0}^n b_ix^i$ (any complex zero of the polynomial is real). This prompted many mathematicians to study trees with real-rooted independence polynomials. In this paper we show a general method to construct such trees or prove real-rootedness.
In particular we will give a new proof for real-rootedness of the independence polynomials of certain families of trees, which includes centipedes (Zhu’s theorem, see [@Zhu2007]), caterpillars (Wang and Zhu’s theorem, see [@Wang2011]), and we will prove a conjecture of Galvin and Hilyard about the real-rootedness of the independence polynomial of the Fibonacci trees (Conj. 6.1. of [@Galvin2017]).
[0.5]{}
(0.71,0.73) rectangle (6.68,2.11); (1,1)– (1,2); (1,1)– (2,1); (2,1)– (3,1); (2,1)– (2,2); (3,1)– (3,2); (5,1)– (6,1); (6,1)– (6,2); (5,1)– (5,2); (4,1) node [$$\dots$$]{};
(1,1) circle (1.5pt); (2,1) circle (1.5pt); (3,1) circle (1.5pt); (5,1) circle (1.5pt); (6,1) circle (1.5pt); (1,2) circle (1.5pt); (2,2) circle (1.5pt); (3,2) circle (1.5pt); (5,2) circle (1.5pt); (6,2) circle (1.5pt);
[0.5]{}
(0.44,0.69) rectangle (6.8,2.23); (1,1)– (0.75,2); (1,1)– (2,1); (2,1)– (3,1); (2,1)– (1.75,2); (3,1)– (2.75,2); (5,1)– (6,1); (6,1)– (5.75,2); (5,1)– (4.75,2); (4,1) node [$$\dots$$]{}; (1,1)– (1.25,2); (2,1)– (2.25,2); (3,1)– (3.25,2); (5,1)– (5.25,2); (6,1)– (6.25,2);
(1,1) circle (1.5pt); (2,1) circle (1.5pt); (3,1) circle (1.5pt); (5,1) circle (1.5pt); (6,1) circle (1.5pt); (0.75,2) circle (1.5pt); (1.75,2) circle (1.5pt); (2.75,2) circle (1.5pt); (4.75,2) circle (1.5pt); (5.75,2) circle (1.5pt); (1.25,2) circle (1.5pt); (2.25,2) circle (1.5pt); (3.25,2) circle (1.5pt); (5.25,2) circle (1.5pt); (6.25,2) circle (1.5pt);
[0.5]{}
(-1.54,-1.78) rectangle (7.95,3.69); (5,3)– (6,3); (6,3)– (7,3); (5,3)– (6,2); (5,1)– (6,1); (4,3)– (5,3); (4,3)– (5,1); (4,0)– (5,0); (5,0)– (6,0); (4,0)– (5,-1); (3,3)– (4,3); (3,3)– (4,0); (0,1)– (1,1); (1,1)– (2,1); (0,1)– (1,0); (0,-1)– (1,-1); (-1,1)– (0,1); (-1,1)– (0,-1); (0,3)– (1,3); (1,3)– (2,3); (0,3)– (1,2); (-1,2)– (0,2);
(6,3) circle (1.5pt); (7,3) circle (1.5pt); (6,2) circle (1.5pt); (5,3) circle (1.5pt); (5,1) circle (1.5pt); (6,1) circle (1.5pt); (4,3) circle (1.5pt); (5,0) circle (1.5pt); (6,0) circle (1.5pt); (5,-1) circle (1.5pt); (4,0) circle (1.5pt); (3,3) circle (1.5pt); (3,3.18) node [$r_4$]{}; (1,1) circle (1.5pt); (2,1) circle (1.5pt); (1,0) circle (1.5pt); (0,1) circle (1.5pt); (0,-1) circle (1.5pt); (1,-1) circle (1.5pt); (-1,1) circle (1.5pt); (-1,1.18) node [$r_3$]{}; (1,3) circle (1.5pt); (2,3) circle (1.5pt); (1,2) circle (1.5pt); (0,3) circle (1.5pt); (0,3.18) node [$r_2$]{}; (-1,2) circle (1.5pt); (-1,2.18) node [$r_1$]{}; (0,2) circle (1.5pt); (-1,3) circle (1.5pt); (-1,3.18) node [$r_0$]{};
Recall that the *$n$-centipede $W_n$* is a graph (Fig. \[fig:centipede\]), such that we take a path on $n$ vertices and we hang 1 pendant edge from each vertex of it. Similarly the *$n$-caterpillar* $H_n$ is the graph (Fig. \[fig:caterpillar\]) obtained by taking a path on $n$ vertices and by hanging 2 pendant edges from each vertex of it. The *Fibonacci trees* were defined by Wagner [@Wagner2007] as follows (Fig. \[fig:fibonacci\]): let $F_0=K_1$ and $F_1=K_2$ with roots $r_0\in V(F_0)$ and $r_1\in V(F_1)$. Then for $n\ge 2$ the $n$th Fibonacci tree $F_n$ is obtained from the disjoint union of $F_{n-1}$, $F_{n-2}$ and a new vertex, labeled by $r_n$ and connecting $r_n$ to the roots of $F_{n-1}$ and $F_{n-2}$. Define $r_n$ as the root of $F_n$.
Methods and motivations
-----------------------
To motivate our method we will use certain results from the theory of matching polynomials. Recall that the matching polynomial of a graph $G$ is defined as: $$\mu(G,x)=\sum\limits_{k=0}(-1)^km_k(G)x^{n-2k},$$ where $m_k(G)$ is the number of matchings with $k$ edges (note that $m_0(G)=1$). One of the best known theorems about matching polynomials is that for any finite graph $G$ and $u\in V(G)$ there exists a rooted tree $(T,r)$, such that $$\begin{aligned}
\frac{\mu(G-u,x)}{\mu(G,x)}=\frac{\mu(T-r,x)}{\mu(T,x)} \end{aligned}$$ A well-known construction for $T$ is the path-tree [@Godsil1993] (a.k.a. Godsil tree), which is the tree on paths of $G$ starting from $u$, and the edges are the strict inclusions. (For an example see Figure \[fig:pt\].)
[0.5]{}
(0.72,0.7) rectangle (3.48,3.64); (1,2)– (2,2); (1,2)– (2,1); (2,2)– (2,1); (1,2)– (2,3); (2,2)– (3,2); (2,3)– (3,2);
(1,2) circle (1.5pt); (1,2.22) node [$1$]{}; (2,2) circle (1.5pt); (2,2.26) node [$2$]{}; (2,3) circle (1.5pt); (2,3.22) node [$3$]{}; (2,1) circle (1.5pt); (2.22,1) node [$5$]{}; (3,2) circle (1.5pt); (3,2.22) node [$4$]{};
[0.5]{}
(0.39,-0.28) rectangle (6.31,3.61); (1,2)– (2,3); (1,2)– (2,2); (1,2)– (2,0); (2,3)– (3,3); (3,3)– (4,3); (4,3)– (5,3); (2,2)– (3,2); (3,2)– (4,2); (2,2)– (3,1); (2,0)– (3,0); (3,0)– (4,0); (4,0)– (5,0);
(1,2) circle (1.5pt); (1,2.23) node [$1$]{}; (2,3) circle (1.5pt); (2,3.23) node [$3$]{}; (2,2) circle (1.5pt); (2,2.23) node [$2$]{}; (2,0) circle (1.5pt); (2,0.23) node [$5$]{}; (3,1) circle (1.5pt); (3,1.23) node [$5$]{}; (3,2) circle (1.5pt); (3,2.23) node [$4$]{}; (4,2) circle (1.5pt); (4,2.23) node [$3$]{}; (3,3) circle (1.5pt); (3,3.23) node [$4$]{}; (4,3) circle (1.5pt); (4,3.23) node [$2$]{}; (5,3) circle (1.5pt); (5,3.23) node [$5$]{}; (3,0) circle (1.5pt); (3,0.23) node [$2$]{}; (4,0) circle (1.5pt); (4,0.23) node [$4$]{}; (5,0) circle (1.5pt); (5,0.23) node [$3$]{};
In this paper we will prove an ”independence version” of this theorem through a quite similar construction. More precisely, we will show that there exists a rooted tree $(T',r)$, such that $$\frac{{I}(G-u,x)}{{I}(G,x)}=\frac{{I}(T'-r,x)}{{I}(T',x)}.$$ We will call the constructed tree a stable-path tree. This construction already appeared in in the work of Scott and Sokal (see [@Scott2005]) and variant of this construction in the work of Weitz (see [@Weitz2006]). We will see that the key property of a stable-path tree is that its independence polynomial is a product of independence polynomials of some induced subgraphs of $G$.
Chudnovsky and Seymour showed that the independence polynomial of any claw-free graph is real-rooted (see [@sey]). Since any induced subgraph of a claw-free graph is also claw-free, this enables us to conclude that any stable-path tree of a claw-free graph has real-rooted independence polynomial. In section \[sec:ap\] we will construct claw-free graphs such that their stable-path trees will be $n$-centipedes, $n$-caterpillars and Fibonacci trees. In the same section we will give further applications of this method.
Notation
--------
We denote the vertex set and edge set of a graph $G$ by $V(G)$ and $E(G)$, respectively. Let $N_G(u)$ denote the set of neighbours of the vertex $u$ and $d(u)$ the degree of the vertex $u$. Let $N_G[u]=N_G(u)\cup\{u\}$ denote the closed neighbourhood of the vertex $u$. If it is clear from the context, then we will write $N(u)$ and $N[u]$ instead of $N_G(u)$ and $N_G[u]$. Let $G-v$ denote the graph obtained from $G$ by deleting the vertex $v$. If $S\subseteq V(G)$, then $G[S]$ denotes the induced subgraph of $G$ on the vertex set $S$, and $G-S$ denotes $G[V(G)-S]$.
This paper is organized as follows:
-----------------------------------
in the next section we will define stable-path trees of graphs, and we will prove some properties of it. In the last section we will prove real-rootedness of independence polynomials of certain graphs.
Tree of stable paths {#sec:main}
====================
In this section we will give two variants of the definition of the stable-path tree, where the first one is a special case of the latter one. For the applications it is enough to get familiar with the first definition. But first let us recall the following properties of the independence polynomial, which we will use intensively in the proofs. For proof see [@Levit2005].
Let $G$ be a graph with connected components $G_1,\dots,G_k$, and let $u\in V(G)$ be a fixed vertex. Then $$\begin{gathered}
I(G,x)=I(G-u,x)+xI(G-N_G[u],x)\\
I(G,x)=\prod_{i=1}^kI(G_i,x)
\end{gathered}$$
Let $G$ be a graph, where we have a total ordering $\prec$ on $V(G)$ and let $u\in V(G)$ fixed. Then we define a tree $(T^<_{G,u},\bar{u})$ as follows. Let us denote by $N(u)=\{u_1\prec \dots \prec u_d\}$, and let $$\begin{gathered}
G^i=G[V(G)\setminus\{u,u_1,v_2,\dots,u_{i-1}\}]\\
(T^i,r^i)=(T^<_{G^i,u_i},\bar{u_i}),
\end{gathered}$$ where we take the induced ordering of the vertices on $V(G^i)$ for $1\le i\le d$. Consider the disjoint unions of $T^i$ with roots $r^i$ and a new vertex with label $\bar{u}$, and add edges $(\bar{u},r^i)$ for $1\le i\le d$. In this way we gain a tree $T^<_{G,u}$ and let $\bar{u}$ be the root of this tree. See an example in Fig \[fig:spt\].
[0.5]{}
(0.72,0.7) rectangle (3.48,3.64); (1,2)– (2,2); (1,2)– (2,1); (2,2)– (2,1); (1,2)– (2,3); (2,2)– (3,2); (2,3)– (3,2);
(1,2) circle (1.5pt); (1,2.22) node [$1$]{}; (2,2) circle (1.5pt); (2,2.26) node [$2$]{}; (2,3) circle (1.5pt); (2,3.22) node [$3$]{}; (2,1) circle (1.5pt); (2.22,1) node [$5$]{}; (3,2) circle (1.5pt); (3,2.22) node [$4$]{};
[0.5]{}
(0.75,0.73) rectangle (4.48,4.6); (1,3)– (2,3); (1,3)– (2,2); (1,3)– (2,1); (2,2)– (3,2); (2,3)– (3,4); (3,4)– (4,4); (2,3)– (3,3);
(1,3) circle (1.5pt); (1,3.22) node [$1$]{}; (2,3) circle (1.5pt); (2,3.22) node [$2$]{}; (2,2) circle (1.5pt); (2,2.22) node [$3$]{}; (2,1) circle (1.5pt); (2,1.22) node [$5$]{}; (3,4) circle (1.5pt); (3,4.23) node [$4$]{}; (4,4) circle (1.5pt); (4,4.22) node [$3$]{}; (3,3) circle (1.5pt); (3,3.22) node [$5$]{}; (3,2) circle (1.5pt); (3,2.22) node [$4$]{};
\[th:fa\] Let $G$ be a graph, $u\in V(G)$. Then for $T=T^<_{G,u}$ we have that $$\frac{{I}(G-u,x)}{{I}(G,x)}=\frac{{I}(T-\overline{u},x)}{{I}(T,x)},$$
We will prove the statement by induction on the number of vertices of $G$. If $G$ has exactly one vertex, then $T^<_{G,u}$ is constructed to be a graph with one vertex.
Let $N(u)=\{u_1\prec \dots \prec u_d\}$, and then let $G^i=G[V(G)\setminus\{u,u_1,v_2,\dots,u_{i-1}\}]$ and $(T^i,r^i)=(T_{G^i,u_i},\bar{u_i})$ for $1\le i \le d$ as in the definition. Then $$\begin{gathered}
\frac{{I}(G,x)}{{I}(G-u,x)}=\frac{{I}(G-u,x)+x{I}(G-N[u],x)}{{I}(G-u,x)}=1+\frac{x{I}(G-N[u],x)}{{I}(G-u,x)}=\\
=1+x\frac{{I}(G-u-u_1,x){I}(G-u-\{u_1,u_2\},x)\dots {I}(G-u-\{u_1,\dots,u_k\},x)}{{I}(G-u,x){I}(G-u-u_1)\dots {I}(G-u-\{u_1,\dots,u_{k-1}\})}=\\
=1+x\frac{{I}(G^1-
u_1,x)}{{I}(G^1,x)}\frac{{I}(G^2-u_2,x)}{{I}(G^2,x)}\dots\frac{{I}(G^d-u_d,x)}{{I}(G^d,x)}=\\
=1+x\frac{{I}(T^1-r^1,x)}{{I}(T^1,x)}\frac{{I}(T^2-r^2,x)}{{I}(T^2,x)}\dots\frac{{I}(T^d-r^d,x)}{{I}(T^d,x)}=\\
=\frac{{I}(T-r,x)+x{I}(T-N[r],x)}{{I}(T-r,x)}=\frac{{I}(T,x)}{{I}(T-r,x)}.
\end{gathered}$$
We would like to remark that in all applications it will be enough to use this definition, however, for the completeness we will give a a more general form.
The following construction already appeared in the work of Scott and Sokal (see [@Scott2005]), where they called the this tree as pruned SAW-tree.
Let $\mathcal{P}_u$ be the set of paths from $u$ in $G$, and let $$A_{G,u}=\{(P,e)\in \mathcal{P}_u\times E(G)~|~P=(v_0,\dots,v_k), v_k\in e\}.$$ A function $\sigma:A_{G,u}\to \mathbb{R}$ is called deep decision if it satisfies that whenever $(P,e),(P,f)\in A_{G,u}$ and $\sigma(P,e)=\sigma(P,f)$, then $e=f$. Then a path $P=(v_0,v_1, \dots, v_k)$ from $u$ is $\sigma$-stable, if whenever $(v_i,v_j)\in E(G)$ and $i+1<j$, then $\sigma(P',(v_i,v_{i+1}))<\sigma(P',(v_i,v_j))$, where $P'=(v_0,\dots, v_i)$ is a subpath. If the path $P=(u,v_1,\dots,v_k)$ is stable with respect to $\sigma$, then $P'=(u,v_1,\dots,v_{k-1})$ is also stable with respect to $\sigma$.
Let $T^\sigma_{G,u}$ be a tree, whose vertices are the $\sigma$-stable paths from $u$, and the edges correspond to the strict inclusion. In that tree the path $(u)$ (with length 0) appears, which we will denote by $\overline{u}$.
To see the relation between the two definitions, let us assume, that $G$ has a total ordering on its vertices, so we may assume, that $(V(G),\prec)=(\{1,\dots,n\},<)$. Then for a $(P,e)\in A_{G,u}$, such that $P=(u,v_1,\dots,v_k)$ and $e=
(v_k,v_{k+1})$ let $\sigma(P,e)=v_{k+1}$. Then it is easy to check that $T^\sigma_{G,u}=T^<_{G,u}$. Indeed the second definition is a generalization of the first one.
For the completeness we will prove Theorem \[th:fa\] also for the generalized $\sigma$-stable-path tree.
\[th:fa\_gen\] Let $G$ be a graph, $u\in V(G)$ and let $\sigma:A_{G,u}\to \mathbb{R}$ be a deep decision. Then for $T=T^{\sigma}_{G,u}$ we have that $$\frac{{I}(G-u,x)}{{I}(G,x)}=\frac{{I}(T-\overline{u},x)}{{I}(T,x)},$$
We will prove the statement by induction on the number of vertices. If $G$ has exactly one vertex, then $T^\sigma(G,u)$ is constructed to be a graph with one vertex.
Furthermore we may assume that $G$ is connected, since if $G_1, \dots, G_k$ are the connected components of $G$, where $u\in V(G_1)$, then by using the multiplicity of the independence polynomial, we have: $$\frac{{I}(G-u,x)}{{I}(G,x)}=\frac{{I}(G_1-u,x){I}(G_2,x)\dots {I}(G_k,x)}{{I}(G_1,x){I}(G_2,x)\dots {I}(G_k,x)}=\frac{{I}(G_1-u,x)}{{I}(G_1,x)}.$$ and by $A_{G,u}=A_{G_1,u}$ we have that $T^\sigma(G_1,u)=T^\sigma(G,u)$, which is the appropriate tree.
For the rest assume that $G$ is connected. Then let $N(u)=\{u_1,\dots,u_d\}$ in such a way, such that $\sigma(\overline{u},(u,u_i))<\sigma(\overline{u},(u,u_j))$, whenever $1\le i<j\le d$. Then for any $1\le i\le d$ and for any $(P,e)\in A_{G-\{u,u_1,\dots,u_{i-1}\},u_i}$ let $\sigma_i$ be defined as follows (where $P=(u_i,v_1,\dots,v_k)$): $$\sigma_i(P,e)=\sigma((u,u_i,v_1,\dots v_k), e).$$ Then $$\begin{gathered}
\frac{{I}(G,x)}{{I}(G-u,x)}=\frac{{I}(G-u,x)+x{I}(G-N[u],x)}{{I}(G-u,x)}=1+\frac{x{I}(G-N[u],x)}{{I}(G-u,x)}=\\
=1+x\frac{{I}(G-u-u_1,x){I}(G-u-\{u_1,u_2\},x)\dots {I}(G-u-\{u_1,\dots,u_d\},x)}{{I}(G-u,x){I}(G-u-u_1)\dots {I}(G-u-\{u_1,\dots,u_{d-1}\})}=\\
=1+x\frac{{I}(G-u-
u_1,x)}{{I}(G-u,x)}\frac{{I}(G-u-\{u_1,u_2\},x)}{{I}(G-u-u_1)}\dots\frac{{I}(G-u-\{u_1,\dots,u_d\},x)}{{I}(G-u-\{u_1,\dots,u_{d-1}\})}=\\
=1+x\frac{{I}(T^{\sigma_1}_{G-u,u_1}-\overline{u_1},x)}{{I}(T^{\sigma_1}_{G-u,u_1},x)}\frac{{I}(T^{\sigma_2}_{G-u-u_1,u_2}-\overline{u_2},x)}{{I}(T^{\sigma_2}_{G-u-u_1,u_2},x)}\dots\frac{{I}(T^{\sigma_d}_{G-u-\{u_1\dots u_{d-1}\},u_d}-\overline{u_d},x)}{{I}(T^{\sigma_d}_{G-u-\{u_1\dots u_{d-1}\},d_k},x)}=\\
=\frac{{I}(T,x)}{{I}(T-r,x)},
\end{gathered}$$ where $T$ is a tree that is obtained from a star with $k$ leaves, whose root is $r$, and the $i$th leaf is glued to the root of $T^{\sigma_i}_{G-u-\{u_1\dots u_{i-1}\},u_i}$. On the other hand this $T$ is isomorphic to $T^\sigma_{G,u}$, since any $\sigma$-stable path $P=(u,u_i,v_1,\dots,v_k)$ (specially if $1\le j<i$, then $u_j\notin\{v_1,\dots,v_k\}$) the path $P'=(u_i,v_1,\dots,v_k)$ is $\sigma_i$-stable. And for any $\sigma_i$-stable path $P'=(u_i,v_1,\dots,v_k)$ is a $P=(u,u_i,v_1,\dots,v_k)$ $\sigma$-stable path. So $$\frac{{I}(T-r,x)}{{I}(T,x)}=\frac{{I}(T^{\sigma}_{G,u}-\overline{u},x)}{{I}(T^{\sigma}_{G,u},x)}$$
We would like to remark that Weitz’s construction of the self-avoiding path tree is a special case of the previously defined stable-path tree of a deep decision. Let $\phi:E(G)\to \{1,\dots,m\}$ bijection, where $m=|E(G)|$. Then for a $(P,e)\in A_{G,u}$ let $\sigma(P,e)=\phi(e)$. Then $T^\sigma_{G,u}$ is the Weitz-tree.
Observe that if we have a deep decision for a connected graph, then we can perform the DFS-algorithm with respect to $\sigma$, in the following way. Whenever we arrive into the vertex $v$ along the path $P$ and there is an unvisited neighbor of $v$, then we will move to that unvisited vertex $w$ for which $\sigma(P,(v,w))$ is the smallest.
Formally, let us assume, that there is a given connected graph $G$, $u\in V(G)$ and a $\sigma$ deep decision from $u$. Then one can construct a spanning tree $F_{G,u,\sigma}$ (call as $\sigma$-DFS tree of $G$) as follows. Let $G_1,\dots,G_k$ be a the connected components of $G-u$, $u_i=\textrm{argmin}_{v\in V(G_i)\cap N_G(u)}(\sigma(u,(u,v)))$ for $1\le i\le k$ and the functions $\sigma_i:A_{G_i,u_i}\to \mathbb{R}$ are $$\begin{gathered}
\sigma_i((u_i,v_1,\dots,v_k),e)=\sigma((u,u_i,v_1,\dots,v_k),e).
\end{gathered}$$ Then we gain $F_{G,u,\sigma}$ as we take the disjoint union of $F_{G_i,u_i,\sigma_i}$ for $1\le i\le k$ and we connect a new vertex called $u$ with $u_i$ for $1\le i\le k$.
By induction we can prove the following properties of a stable-path tree.
\[prop:faszerk\] Let $G$ be a connected graph, $u\in V(G)$, $\sigma$ a deep decision, and let $F$ be a $\sigma$-DFS tree. Denote by $\overline{F}$ the set of paths from $u$ in $F$ (they are $\sigma$-stable paths). Then
1. there exists a sequence $G_1, \dots, G_k$ of induced subgraphs of $G$, such that $${I}(T^\sigma_{G,u},x)={I}(G,x){I}(G_1,x)\dots {I}(G_k,x),$$
2. and $${I}(G,x)=\frac{{I}(T^\sigma_{G,u})}{{I}(T^\sigma_{G,u}-\overline{F},x)}.$$
We will prove the first part by induction on the number of vertices of $G$. The proof of the second part goes similarly. From the proof of the previous theorem (and with its notations) we know that $$\begin{gathered}
{I}(T^\sigma_{G,u},x)=\frac{{I}(G,x)}{{I}(G-u,x)}{I}(T^\sigma_{G,u}-\overline{u},x)=\\
=\frac{{I}(G,x)}{{I}(G-u,x)}{I}(T^{\sigma_1}_{G-u,u_1},x){I}(T^{\sigma_2}_{G-\{u,u_1\},u_2},x)\dots {I}(T^{\sigma_d}_{G-\{u,u_1,\dots u_{d-1}\},u_d},x)=\\
=\frac{{I}(G,x)}{{I}(G-u,x)}\prod_{i=1}^d\prod_{j=0}^{l_i}{I}(G^i_j,x),
\end{gathered}$$ where $G^i_0$ is the connected component of $G-\{u,u_1,\dots,u_{i-1}\}$, which contains $u_i$; and each $G^i_j$ is an induced subgraph of $G^i_0$. So each $G^i_j$ is an induced subgraph of $G$. Let $\{H_1,\dots,H_t\}$ the set of connected components of $G-u$, and $$I=\{\min_{u_i\in V(H_j)}(i)~|~1\le j \le t\}.$$ By definition of $I$ we have that the set $\{G^i_0~|~i\in I\}$ is the set of connected components of $G-u$. This implies that the product $\prod_{i\in I}I(G^i_0,x)=I(G-u,x)$, therefore $$\begin{aligned}
\label{prec}
{I}(T^\sigma_{G,u},x)={I}(G,x)\prod_{i\in I'}{I}(G^i_0,x)\prod_{i=1}^d\prod_{j=1}^{l_i}{I}(G^i_j,x),
\end{aligned}$$ where $I'=\{1,\dots,d\}\setminus I$.
Sometimes, it is useful to follow the induction to determine explicitly the multiplicites of the subgraphs occuring in the formula (\[prec\]).
Applications of stable-path tree {#sec:ap}
================================
In this section we will present various applications of the following corollary of Proposition \[prop:faszerk\]:
\[cor:poly\] Let $G$ be a graph, $v\in V(G)$, and let $\sigma$ be a deep decision. If $G$ is a claw-free graph, then $I(T^\sigma_{G,u},x)$ is real-rooted. Moreover $I(G,x)$ divides $I(T^{\sigma}_{G,u},x)$.
Assume that $G$ is a claw-free graph. Then by Proposition \[prop:faszerk\] we have a sequence of induced subgraphs $G_1,\dots,G_k$ of $G$, such that $$I(T^\sigma_{G,u})=I(G,x)\prod_{i=1}^kI(G_i,x).$$ Since each $G_i$ is an induced subgraph of a claw-free graph, therefore it is also claw-free. Then by the result of Chudnovsky and Seymour, Thm. 1.1. of [@sey], we have that each polynomial $I(G_i,x)$ and the polynomial $I(G,x)$ are real-rooted, so their product is also real rooted.
In this section will show some applications of this corollary. In all applications, of Corollary \[cor:poly\] the vertices of $G$ will be labelled by integers. This labeling will induce a total order on the vertices in the most natural way, the order of two vertices will be the order of their labels.
Trees with real-rooted independence polynomial
----------------------------------------------
In this subsection we will show that some families of trees have real-rooted independence polynomials.
Let us recall that, the *$n$-centipede $W_n$* is a graph such that we take a path on $n$ vertices and we hang 1 pendant edge from each vertex of it.
The *$n$-caterpillar* $H_n$ is a graph such that we take a path on $n$ vertices and we hang 2 pendant edges from each vertex of it.
The *Fibonacci tree* $F_0=K_1$ and $F_1=K_2$ with roots $r_0\in V(F_0)$ and $r_1\in V(F_1)$. Then for $n\ge 2$ the $n$th Fibonacci tree $F_n$ is obtained from the disjoint union of $F_{n-1}$, $F_{n-2}$ and a new vertex, labeled as $r_n$, and connecting $r_n$ to the roots of $F_{n-1}$ and $F_{n-2}$. Define $r_n$ as the root of $F_n$.
The proof of the real-rootedness of the independence polynomial of $W_n$ was in [@Zhu2007], then a unified proof for $W_n$ and $H_n$ appeared in [@Wang2011]. The statement for $F_n$ was verified in [@Galvin2017] for $n\le 22$, and conjectured for arbitrary $n$. Our proofs will follow the following strategy: for each mentioned $T$ tree we will define a claw-free graph $\widetilde{G}$ with integer labels, such that the stable-path tree of $\widetilde{G}$ from one of its vertex will be isomorphic to $T$.
\[pro:centipede\] For any $n$, the independence polynomial of $W_n$ is real-rooted, hence log-concave and unimodal.
Let $\widetilde{W}_n$ be a graph (Fig. \[fig:tilde\_Wn\]), such that we take a path on $\{1,\dots,n\}$ and we attach a triangle to every $(2k+1)$th edge of the path. If $n$ is odd, then we attach a pendant edge to $n$. Also label all the new vertices by numbers bigger than $n$.
[0.5]{}
(0.71,0.44) rectangle (7.63,2.4); (1,1)– (2,1); (2,1)– (3,1); (5,1)– (6,1); (4.5,1) node [$$\dots$$]{}; (3,1)– (4,1); (1,1)– (1.5,2); (2,1)– (1.5,2); (3.5,2)– (3,1); (4,1)– (3.5,2); (6,1)– (7,1); (6.5,2)– (7,1); (6.5,2)– (6,1);
(1,1) circle (1.5pt); (1,0.75) node [$1$]{}; (2,1) circle (1.5pt); (2,0.75) node [$2$]{}; (3,1) circle (1.5pt); (3,0.75) node [$3$]{}; (5,1) circle (1.5pt); (5,0.75) node [$n-2$]{}; (6,1) circle (1.5pt); (6,0.75) node [$n-1$]{}; (4,1) circle (1.5pt); (4,0.75) node [$4$]{}; (1.5,2) circle (1.5pt); (1.5,2.22) node [$n+1$]{}; (3.5,2) circle (1.5pt); (3.5,2.22) node [$n+3$]{}; (7,1) circle (1.5pt); (7,0.75) node [$n$]{}; (6.5,2) circle (1.5pt); (6.5,2.22) node [$2n-1$]{};
[0.5]{}
(0.71,0.44) rectangle (7.63,2.4); (1,1)– (2,1); (2,1)– (3,1); (5,1)– (6,1); (4.5,1) node [$$\dots$$]{}; (3,1)– (4,1); (1,1)– (1.5,2); (2,1)– (1.5,2); (3.5,2)– (3,1); (4,1)– (3.5,2); (6,1)– (7,1); (5.5,2)– (6,1); (5,1)– (5.5,2); (7,1)– (7,2);
(1,1) circle (1.5pt); (1,0.75) node [$1$]{}; (2,1) circle (1.5pt); (2,0.75) node [$2$]{}; (3,1) circle (1.5pt); (3,0.75) node [$3$]{}; (5,1) circle (1.5pt); (5,0.75) node [$n-2$]{}; (6,1) circle (1.5pt); (6,0.75) node [$n-1$]{}; (4,1) circle (1.5pt); (4,0.75) node [$4$]{}; (1.5,2) circle (1.5pt); (1.5,2.22) node [$n+1$]{}; (3.5,2) circle (1.5pt); (3.5,2.22) node [$n+3$]{}; (7,1) circle (1.5pt); (7,0.75) node [$n$]{}; (5.5,2) circle (1.5pt); (5.5,2.22) node [$2n-2$]{}; (7,2) circle (1.5pt); (7,2.22) node [$2n$]{};
These graphs are claw-free, and $$\begin{gathered}
T^<_{\widetilde{W}_n,1}\cong W_n.
\end{gathered}$$ Therefore by Corollary \[cor:poly\] we have the desired statement.
\[pro:caterpillar\] For any $n$, the independence polynomial of $H_n$ are real-rooted, hence log-concave and unimodal.
Let $\widetilde{H}_n$ be a graph (Fig. \[fig:tilde\_Hn\]), such that we take a path on $\{0,\dots,n+1\}$ and we attach a triangle to each edge, which is not the first or the last. Also label all the new vertices with numbers bigger than $n$.
(0.71,0.44) rectangle (7.63,2.4); (1,1)– (2,1); (2,1)– (3,1); (5,1)– (6,1); (4.5,1) node [$$\dots$$]{}; (3,1)– (4,1); (6,1)– (7,1); (2.5,2)– (2,1); (3,1)– (2.5,2); (4,1)– (3.5,2); (5,1)– (5.5,2); (3.5,2)– (3,1); (6,1)– (5.5,2);
(1,1) circle (1.5pt); (1,0.75) node [$0$]{}; (2,1) circle (1.5pt); (2,0.75) node [$1$]{}; (3,1) circle (1.5pt); (3,0.75) node [$2$]{}; (5,1) circle (1.5pt); (5,0.75) node [$n-1$]{}; (6,1) circle (1.5pt); (6,0.75) node [$n$]{}; (4,1) circle (1.5pt); (4,0.75) node [$3$]{}; (7,1) circle (1.5pt); (7,0.75) node [$n+1$]{}; (2.5,2) circle (1.5pt); (2.5,2.22) node [$n+2$]{}; (3.5,2) circle (1.5pt); (3.5,2.22) node [$n+3$]{}; (5.5,2) circle (1.5pt); (5.5,2.22) node [$2n$]{};
These graphs are claw-free, and $$\begin{gathered}
T^<_{\widetilde{H}_n,0}\cong H_n.
\end{gathered}$$ Therefore by Corollary \[cor:poly\] we have the desired statement.
For any $n$, the independence polynomial of $F_n$ are real-rooted, hence log-concave and unimodal.
Let $\widetilde{F}_n$ be a graph (Fig. \[fig:tilde\_Fn\]), such that we take the set $\{0,\dots,n-1\}$ and we connect $i$ and $j$ if $0<|i-j|\le 2$.
(0.71,0.44) rectangle (7.63,2.4); (1,1)– (2,1); (2,1)– (3,1); (5,1)– (6,1); (4.5,1) node [$$\dots$$]{}; (3,1)– (4,1); (6,1)– (7,1); plot\[domain=0:pi,variable=\]([1\*1\*cos(r)+0\*1\*sin(r)]{},[0\*1\*cos(r)+1\*1\*sin(r)]{}); plot\[domain=0:pi,variable=\]([1\*1\*cos(r)+0\*1\*sin(r)]{},[0\*1\*cos(r)+1\*1\*sin(r)]{}); plot\[domain=0:pi,variable=\]([1\*1\*cos(r)+0\*1\*sin(r)]{},[0\*1\*cos(r)+1\*1\*sin(r)]{}); plot\[domain=1.57:pi,variable=\]([1\*1\*cos(r)+0\*1\*sin(r)]{},[0\*1\*cos(r)+1\*1\*sin(r)]{}); plot\[domain=0:1.57,variable=\]([1\*1\*cos(r)+0\*1\*sin(r)]{},[0\*1\*cos(r)+1\*1\*sin(r)]{});
(1,1) circle (1.5pt); (1,0.75) node [$0$]{}; (2,1) circle (1.5pt); (2,0.75) node [$1$]{}; (3,1) circle (1.5pt); (3,0.75) node [$2$]{}; (5,1) circle (1.5pt); (5,0.75) node [$n-3$]{}; (6,1) circle (1.5pt); (6,0.75) node [$n-2$]{}; (4,1) circle (1.5pt); (4,0.75) node [$3$]{}; (7,1) circle (1.5pt); (7,0.75) node [$n-1$]{};
These graphs are claw-free, and $$\begin{gathered}
T^<_{\widetilde{F}_n,0}\cong F_n.
\end{gathered}$$ Therefore by Corollary \[cor:poly\] we have the desired statement.
If someone carefully examine the formula (\[prec\]), then one might get the following identities: $$\begin{gathered}
I(W_{n},x)=I(\widetilde{W}_{n})(1+x)^{\lfloor n/2\rfloor},\\
I(H_n,x)=I(\widetilde{H}_n)(1+x)^{n-2},\\
I(F_n,x)=\prod_{k=0}^n I(\widetilde{F}_{k},x)^{f_{n-k}},
\end{gathered}$$ where $f_{0}=1$, $f_{1}=0$ and $f_{n}=f_{n-1}+f_{n-2}$ for $n\ge 1$.
Some real-rooted graph families
-------------------------------
In this subsection we show another approach to verify real-rootedness of independence polynomials of some graphs. The idea is that for a graph $G$ we construct a stable-path tree $T$, which is real-rooted. Then by Corollary \[cor:poly\] we know that $I(G,x)$ divides $I(T,x)$, so it means that $I(G,x)$ is also real-rooted.
Let us define the following graph families.
The *$n$th apple graph* $A_n$ is a graph (Fig. \[fig:An\]), such that we take a path on $\{1,\dots,n\}$, and we add the edge $(2,n)$.
The *$n$-sunlet* graph $N_n$ is a graph (Fig. \[fig:Nn\]), such that we take a cycle on $\{1,\dots, n\}$, and we attach a new vertex to each vertex of the cycle. Also label all the new vertices with numbers bigger than $n$.
Let $M_n$ be a graph (Fig. \[fig:Mn\]), such that we take a path on $\{1,\dots,n\}$, and we attach 2 triangles to any $2k+1$th edge of the path. If $n$ is odd, then we attach 2 pendant edges to $n$. For the new vertices choose different numbers greater than $n$ as labels.
[0.5]{}
(0.71,0.44) rectangle (7.63,2.4); (1,1)– (2,1); (2,1)– (3,1); (5,1)– (6,1); (4.5,1) node [$$\dots$$]{}; (3,1)– (4,1); (6,1)– (7,1); plot\[domain=1.01:2.13,variable=\]([1\*4.72\*cos(r)+0\*4.72\*sin(r)]{},[0\*4.72\*cos(r)+1\*4.72\*sin(r)]{});
(1,1) circle (1.5pt); (1,0.75) node [$1$]{}; (2,1) circle (1.5pt); (2,0.75) node [$2$]{}; (3,1) circle (1.5pt); (3,0.75) node [$3$]{}; (5,1) circle (1.5pt); (5,0.75) node [$n-2$]{}; (6,1) circle (1.5pt); (6,0.75) node [$n-1$]{}; (4,1) circle (1.5pt); (4,0.75) node [$4$]{}; (7,1) circle (1.5pt); (7,0.75) node [$n$]{};
[0.5]{}
(0.64,-0.17) rectangle (6.63,2.41); (1,1)– (1,2); (1,1)– (2,1); (2,1)– (3,1); (2,1)– (2,2); (3,1)– (3,2); (5,1)– (6,1); (6,1)– (6,2); (5,1)– (5,2); (4,1) node [$$\dots$$]{}; plot\[domain=4.15:5.27,variable=\]([1\*4.72\*cos(r)+0\*4.72\*sin(r)]{},[0\*4.72\*cos(r)+1\*4.72\*sin(r)]{});
(1,1) circle (1.5pt); (2,1) circle (1.5pt); (3,1) circle (1.5pt); (5,1) circle (1.5pt); (6,1) circle (1.5pt); (1,2) circle (1.5pt); (2,2) circle (1.5pt); (3,2) circle (1.5pt); (5,2) circle (1.5pt); (6,2) circle (1.5pt);
[0.5]{}
(0.51,-0.43) rectangle (7.5,2.26); (1,1)– (2,1); (2,1)– (3,1); (5,1)– (6,1); (4.5,1) node [$$\dots$$]{}; (3,1)– (4,1); (1,1)– (1.5,2); (2,1)– (1.5,2); (3.5,2)– (3,1); (4,1)– (3.5,2); (6,1)– (7,1); (6.5,2)– (7,1); (6.5,2)– (6,1); (1.5,0)– (1,1); (1.5,0)– (2,1); (3,1)– (3.5,0); (3.5,0)– (4,1); (6,1)– (6.5,0); (6.5,0)– (7,1);
(1,1) circle (1.5pt); (2,1) circle (1.5pt); (3,1) circle (1.5pt); (5,1) circle (1.5pt); (6,1) circle (1.5pt); (4,1) circle (1.5pt); (1.5,2) circle (1.5pt); (3.5,2) circle (1.5pt); (7,1) circle (1.5pt); (6.5,2) circle (1.5pt); (1.5,0) circle (1.5pt); (3.5,0) circle (1.5pt); (6.5,0) circle (1.5pt);
[0.5]{}
(0.49,-0.35) rectangle (7.64,2.3); (1,1)– (2,1); (2,1)– (3,1); (5,1)– (6,1); (4.5,1) node [$$\dots$$]{}; (3,1)– (4,1); (1,1)– (1.5,2); (2,1)– (1.5,2); (3.5,2)– (3,1); (4,1)– (3.5,2); (6,1)– (7,1); (5.5,2)– (6,1); (5,1)– (5.5,2); (7,1)– (7,2); (1.5,0)– (1,1); (2,1)– (1.5,0); (3.5,0)– (3,1); (4,1)– (3.5,0); (5.58,0)– (5,1); (5.58,0)– (6,1); (7,0)– (7,1);
(1,1) circle (1.5pt); (2,1) circle (1.5pt); (3,1) circle (1.5pt); (5,1) circle (1.5pt); (6,1) circle (1.5pt); (4,1) circle (1.5pt); (1.5,2) circle (1.5pt); (3.5,2) circle (1.5pt); (7,1) circle (1.5pt); (5.5,2) circle (1.5pt); (7,2) circle (1.5pt); (1.5,0) circle (1.5pt); (3.5,0) circle (1.5pt); (5.58,0) circle (1.5pt); (7,0) circle (1.5pt);
A proof for real-rootedness of the independence polynomial of $M_n$ and $N_n$ was given in [@Wang2011].
For any $n$, the independence polynomial of $M_n$ is real-rooted, hence log-concave and unimodal.
By Proposition \[pro:caterpillar\] we have that $H_n$ has real-rooted independence polynomial. However we can see that $$T^<_{M_n,1}\cong H_{n}.$$ By Corollary \[cor:poly\] we know that $I(M_n,x)$ divides $I(H_{n},x)$, which implies, that $I(M_n,x)$ is real-rooted polynomial.
For any $n$, the independence polynomial of $N_n$ is real-rooted, hence log-concave and unimodal.
By Proposition \[pro:centipede\] we have that $W_n$ has real-rooted independence polynomial. However we can see that $$T^<_{N_n,1}\cong W_{2n-1}.$$ By Corollary \[cor:poly\] we know that $I(N_n,x)$ divides $I(W_{2n-1},x)$, which implies, that $I(N_n,x)$ is real-rooted polynomial.
For any $n\ge 4$, the independence polynomial of $A_n$ is real-rooted, hence log-concave and unimodal.
Let $\widetilde{A}_n$ be a graph (Fig. \[fig:tilde\_An\]), such that we take a path on $\{1,\dots,n\}$, and add the edge $(2,4)$.
(0.54,0.25) rectangle (8.6,2.7); (1,1.5)– (2,1.5); (2,1.5)– (3,2); (3,2)– (3,1); (3,1)– (2,1.5); (3,1)– (4,1); (4,1)– (5,1); (7,1)– (8,1); (6,1) node [$$\dots$$]{};
(1,1.5) circle (1.5pt); (1,1.25) node [$1$]{}; (2,1.5) circle (1.5pt); (2,1.25) node [$2$]{}; (3,1) circle (1.5pt); (3,0.75) node [$4$]{}; (3,2) circle (1.5pt); (3,2.22) node [$3$]{}; (4,1) circle (1.5pt); (4,0.75) node [$5$]{}; (5,1) circle (1.5pt); (5,0.75) node [$6$]{}; (7,1) circle (1.5pt); (7,0.75) node [$n-1$]{}; (8,1) circle (1.5pt); (8,0.75) node [$n$]{};
Since $\widetilde{A}_n$ is a claw-free graph, so for any $n\ge 4$ we have that $T_{\widetilde{A}_n,1}$ has a real-rooted independence polynomial. However we can see that $$T^<_{\widetilde{A}_n,1}\cong T^<_{A_n,1},$$ which means that $I(T^<_{A_n,1},x)$ is real-rooted. By Corollary \[cor:poly\] we know that $I(A_n,x)$ divides $I(T^<_{A_n,1},x)$, which implies, that $I(A_n,x)$ is real-rooted polynomial.
Final remarks
=============
We would like to remark, that this method can be also capable of proving the real-rootedness of the independence polynomial of the ladder graph (Thm. 5.1. of [@Zhu2015a]), the polyphenyl ortho-chain ($\bar{O}_n$ of [@alikhani2011]), $k$-ary analogue of the Fibonacci tree (Remark of [@Wagner2007]).
(0.73,0.72) rectangle (4.4,3.3); (1,2)– (2,2); (2,2)– (3,2); (3,2)– (4,2); (1,2)– (2,3); (2,3)– (3,3); (3,3)– (4,3); (2,2)– (3,1); (3,1)– (4,1);
(1,2) circle (1.5pt); (2,2) circle (1.5pt); (3,2) circle (1.5pt); (4,2) circle (1.5pt); (2,3) circle (1.5pt); (3,3) circle (1.5pt); (4,3) circle (1.5pt); (3,1) circle (1.5pt); (4,1) circle (1.5pt);
One might ask that it is true that any tree with real-rooted independence polynomial is a stable path tree of a non-tree graph $G$. The answer is no, as the following example shows:
Let $T$ be a tree on 9 vertices as on the Figure \[ellentree\] and assume that there exists a graph $G$, a deep decision $\sigma$ and a vertex $u\in V(G)$, such that $T=T^\sigma_{G,u}$. Then the independence polynomial of $T$ is $$\begin{gathered}
I(T,x)=(1+3x+x^2)(1+5x+6x^2+x^3)+x(1+2x)^3=\\
(1+x)(1+8x+20x^2+16x^3+x^4),\end{gathered}$$ where the factors are real-rooted and irreducible polynomials in $\mathbb{Q}[x]$. By Proposition \[prop:faszerk\] we have that $I(G,x)$ divides $I(T,x)$, and clearly $G$ cannot be $K_1$ or the empty graph, therefore $I(G,x)$ should be $1+8x+20x^2+16x^3+x^4$. However it can be proved, that there is no such a graph $G$.
[^1]: The author was partially supported by the MTA Rényi Institute Lendület Limits of Structures Research Group.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We use a coarse-grained molecular model to study the self-assembly process of complexes of cationic and neutral lipids with DNA molecules (“lipoplexes”) - a promising nonviral carrier of DNA for gene therapy. We identify the resulting structures through direct visualization of the molecular arrangements and through calculations of the corresponding scattering plots. The latter approach provides a means for comparison with published data from X-ray scattering experiments. Consistent with experimental results, we find that upon increasing the stiffness of the lipid material, the system tends to form lamellar structures. Two characteristic distances can be extracted from the scattering plots of lamellar complexes - the lamellar (interlayer) spacing and the DNA-spacing within each layer. We find a remarkable agreement between the computed values of these two quantities and the experimental data \[J. O. Rädler, I. Koltover, T. Salditt and C. R. Safinya, [*Science*]{}, 1997, [**275**]{} 810-814\] over the entire range of mole fractions of charged lipids (CLs) studied experimentally. A visual inspection of the simulated systems reveals that, for very high fractions of CLs, disordered structures consisting of DNA molecules bound to small membrane fragments are spontaneously formed. The diffraction plots of these non-lamellar disordered complexes appear very similar to that of the lamellar structure, which makes the interpretation of the X-ray data ambiguous. The loss of lamellar order may be the origin of the observed increase in the efficiency of lipoplexes as gene delivery vectors at high charge densities.'
author:
- |
Oded Farago$^{1,2}$ and Niels Gr[ø]{}nbech-Jensen$^{3,4}$\
$^1$Department of Biomedical Engineering and $^2$Ilse Katz Institute for Nanoscale\
Science and Technology, Ben Gurion University, Be’er Sheva 84105, Israel.\
$^3$Department of Applied Science, University of California, Davis,\
California 95616, USA.\
$^4$The Niels Bohr International Academy, The Niels Bohr Institute,\
Blegdamsvej 17, 2100 Copenhagen, Denmark.
title: Molecular simulation analysis of structural variations in lipoplexes
---
When DNA molecules are mixed with neutral and cationic lipids (CLs) in an aqueous environment, they spontaneously aggregate to form macromolecular complexes called “lipoplexes”. These complexes have attracted much attention over the past two decades because of their potential use as nonviral transfection vectors in gene therapy [@general1; @general2; @general3; @general4; @general5]. Transfection is a two-stage process involving adsorption and entry (via endocytosis) of the lipoplex into the cell, followed by the release of the DNA to the cytoplasm and delivery to the nucleus, which makes the DNA available for expression [@transfection1; @transfection2; @transfection3]. CL-DNA complexes exhibit low toxicity and nonimmunogenicity, but their transfection efficiency (TE) remains low compared to that of viral vectors [@transfection1; @transfection4]. This has spurred an intense research activity aimed at enhancing TE. Recognizing that the structure of CL-DNA complexes may strongly influence their function and TE, much of the effort in theoretical and experimental studies has been devoted to understanding the mechanisms governing complex formation, structure, and phase behavior. X-ray diffraction experiments have revealed that CL-DNA complexes exist in a variety of mesoscopic structures. These structures include: (i) a multilamellar phase where DNA monolayers are intercalated between lipid bilayers ($L_{\alpha}^C$) [@safinya_science1], and (ii) the inverted hexagonal phase with DNA encapsulated within monolayers tubes and arranged on a two-dimensional hexagonal lattice ($H_{II}^C$) [@safinya_science2].
One of the major advantages of lipoplexes over viral capsids is their ease of preparation and their almost unlimited DNA-carrying capacity, which stem from the fact that the vector is formed by spontaneous self-assembly in aqueous solutions. The electrostatic attraction between the anionic DNA and the CLs along with the entropic gain associated with the release of tightly bound counterions from the CLs and DNA are the driving forces for the formation of a complex. In a recent publication, we reported on coarse-grained (CG) simulations of self-assembly of CL-DNA complexes [@farago_jensen_jacs]. We demonstrated, in agreement with previous theoretical studies and X-ray scattering experiments [@safinya_science1; @safinya_science2; @harries_may_benshaul; @harries_may_benshaul1], that rigid membranes tend to form lamellar complexes. For soft membranes, the preferred geometry is that of the inverted hexagonal phase. Our simulations also revealed that the phase diagram of the CL-DNA complexes is quite rich and includes, in addition to the lamellar and inverted hexagonal complexes, several other disordered structures with distinct configurational characteristics. We also found a new ordered phase, which has thus far not been observed experimentally, where DNA rods and cylindrical micelles form a 2D square lattice analogous to the 3D cubic NaCl-type structure. Our analysis of the computed self-assembled structures was based on simulation images and on the calculation of the Fourier transform of the DNA positions [@farago_jensen_jacs]. The Fourier transform provides a quantitative measure for how the simulated structures would appear in x-ray scattering experiments. To achieve a better comparison with the experimental data, one should also consider the contribution of the lipids to the scattered intensity. By plotting the separate contributions of each component (something which cannot be done experimentally), one can dissect the information displayed in the scattering plots.
In this paper we analyze the scattering intensity plots (the square of the Fourier transform averaged of all angles [@farago_jensen_jacs]) of the lamellar complexes, which are formed in our simulations when the DNAs are mixed with stiff lipid material. The technical details of the model and the simulations have been presented in ref. [@farago_jensen_jacs]. In short, the model is based on the Noguchi-Takasu implicit solvent CG membrane model [@noguchi_takasu] in which each lipid of length $l_{\rm LIP}$ is represented by a linear rigid molecule [@niels_erratum] consisting of three beads of diameter $\sigma=l_{\rm LIP}/3=6.25$Å, one of which is hydrophilic and the other two are hydrophobic. The CLs are modeled by associating the hydrophilic bead with a positive unit point charge, while the DNA molecules are modeled as infinitely long parallel rigid rods of diameter $D_{\rm DNA}=4\sigma=25$Å with uniform charge density corresponding to -1.7e/Å. The molecules interact via three types of interactions: (i) Unscreened electrostatic interactions which are calculated using the Lekner summation method [@lekner1; @lekner2]. (ii) Short-range repulsive (“hard core”) potential. The bead-bead pair potential $U_{\rm rep,bb}$ is given by Eq. (4) in ref. [@noguchi_takasu], and the bead-DNA potential $U_{\rm rep, bD}(r)= U_{\rm rep,bb}(r-1.5\sigma)$. Since the DNA rods are strongly repelled from each other by electrostatic forces, there was no need to introduce an additional $U_{\rm rep, DD}$. (iii) The Noguchi-Takasu hydrophobic interaction potential, given by Eqs. (5)-(6) in ref. [@noguchi_takasu]. The lipids and the DNA rods are initially randomly distributed within a given volume in the simulation box and, through Molecular Dynamics (MD) simulations at constant temperature, we follow the evolution of the complexes under different conditions. As the objective of the simulations is to attain self-assembled structures representative of equilibrium, we simulate tens of millions of time steps such that at least half of the total simulated time does not change the characteristics of the visible structures. We have also verified that, while the details of the shown structures do depend on initial conditions, the significant characteristics, such as the peaks in the resulting scattering intensities, are well defined.
We study isoelectric complexes where the total charges on DNA and the CLs neutralize each other, with no added counterions. The structure of the complex is determined as a function of two parameters: (i) the fraction of CLs, $\phi_c$, which can be varied by adding different amounts of neutral lipids (NLs), and (ii) the bending modulus, $\kappa_s$, which is the prefactor of the bending energy term introduced in the later version of the Noguchi-Takasu model [@noguchi_takasu_2003] to control the stiffness of the simulated membranes (see Eq. (7) in ref. [@farago_jensen_jacs]).
Fig. \[fig:giant\] shows the diffraction patterns of stiff complexes ($\kappa_s=10$) with different values of $\phi_c$ ranging from $\phi_c=1$ (a) to $\phi_c=4/15$ (l). Following the approach outlined in ref. [@farago_jensen_jacs], we calculate diffraction patterns from the two-dimensional (2D) Fourier transformation $$\begin{aligned}
{\cal F}(\bar{q}) & = & \sum_{j=1}^Nw_j\exp(i\bar{r}_j\bar{q}) \; ,
\label{eq:Fourier}\end{aligned}$$ where $\bar{r}_j$ represents the 2D coordinate of the DNA rod or the 2D coordinate of a bead’s center of mass in the plane perpendicular to the DNA axis, $\bar{q}$ is the reciprocal vector, and $w_j$ represents the electron density of the $j$th particle relative to bulk water. For each value of $\phi_c$, we show a triplet of figures consisting of (left) the simulated scattered intensity from the DNA rods (with $N=N_{DNA}$ in Eq. (\[eq:Fourier\])), (middle) the simulated scattered intensity from the lipids ($N=3N_{\rm LIP}$), and (right) the self-assembled structure of the complex. The displayed scattering intensities are $I(q)\sim\langle|{\cal F}(\bar{q})|^2\rangle_\theta$, where $\bar{q}=q\exp(i\theta)$ and $\langle\dots\rangle_\theta$ denote the average over all angels. All the DNA scattering plots are drawn on the same scale. The lipid scattering plots are also drawn on the same scale, except for (a)-(e) which are multiplied by the factor indicated on the corresponding plot. The relative scattering intensity of the lipids and the DNA depends on their electron densities, as well as on $\phi_c$. Using reasonable values for the electron densities [@raviv1; @raviv2], we find that the scale of the DNA intensities is two order of magnitude larger than that of the lipids. Nevertheless, the scattering plots of the lipids exhibit two well identified peaks located at $q_{\rm LAM}$ and $2q_{\rm LAM}$. These peaks are commonly associated with the lamellar structure. From the position of the first lamellar peak, one can extract the inter-layer lamellar spacing $d$ through $q_{\rm
LAM}=2\pi/d$. The DNA plots generally exhibit three peaks, two of which coincide with the lamellar peaks from the lipid scattering plots, and one which is commonly referred to as the “DNA peak”. The position of the latter at $q_{\rm DNA}$ provides information about the DNA spacing within each layer of the lamellar structure, $d_{\rm
DNA}$. It is assumed that $q_{\rm DNA}=2\pi/d_{\rm DNA}$. Notice that the scattering from the DNAs includes the information about the lamellar spacing observed also in the scattering intensity from the lipids. This can be easily understood by considering the idealized lamellar structure sketched in fig. \[fig:ideal\]. In this structure, the DNA rods form an oblique lattice with lattice vectors equal to $a_1=d_{\rm DNA}$ and $a_2=d/\sin\theta$. The reciprocal lattice is also oblique with the same angle $\theta$ between the lattice vectors $b_1=2\pi/d$ and $b_2=2\pi/d_{\rm DNA}$. The peaks which can be seen in the scattering plots of the DNAs correspond to the reciprocal lattice vectors: $b_1$, $2b_1$, and $b_2$. These are also the peaks which are usually observed in actual scattering experiments [@safinya_science1]. For idealized lamellar structures, the scattering intensity has peaks at other wavevectors $q$ which correspond to linear combinations of integer-multiples of $b_1$ and $b_2$. The positions of these peaks are $\theta$-dependent [@raviv_paper]. In non-idealized complexes, like the ones in our simulations, these peaks are usually very small and are hard to be detected in the scattering plots.
Two open arrows are drawn in each of the lipid scattering plots included in fig. \[fig:giant\]. The first one indicates the position of the larger lamellar peak which is located at $q_{\rm
LAM}=2\pi/d$. The second one denotes the wavevector $2q_{\rm LAM}$, where we indeed find the second lamellar peak. These two arrows are copied into the corresponding DNA scattering plot, verifying that these peaks are also reproduced by the DNA ordering as explained above. Notice that $q_{\rm LAM}$ is only weakly dependent on $\phi_c$. In contrast, the position of the third peak, which is indicated by the solid arrow, varies noticeably with $\phi_c$. This peak is located at $q_{\rm DNA}=2\pi/d_{\rm DNA}$, and the variations in its position reflect the decrease in the DNA spacing with increasing $\phi_c$. From the computed scattering plots shown in fig. \[fig:giant\], we can extract $d$ and $d_{\rm DNA}$ as a function of $\phi_c$. Our results are summarized in fig. \[fig:companion\] (a). The visual impression of this plot is that the results are in overall agreement with the idealized geometry of a lamellar structure shown in fig. \[fig:ideal\]. Specifically, the lamellar spacing is well approximated by the sum of twice the length of the lipids and the diameter of the DNA rods, $d\approx(2l_{\rm LIP}+D_{\rm DNA})\approx
10\sigma=62.5$[Å]{}. Similarly, the characteristic distance $d_{\rm
DNA}$ between DNA rods in each layer approximates a linear relationship with $1/\phi_c$ [@safinya_science1; @oded_prl; @oded_bpj]. This relationship can be derived from the simple geometric consideration that equally spaced DNA rods fill the surface area made available by the lipid bilayer material for any given charge density. There are, however, noticeable deviations from this idealized picture, especially at large $\phi_c$. The lamellar spacing decreases below the ideal value, while the DNA spacing attains values higher than predicted by the linear relationship. The origin of these discrepancies becomes clear by inspection of the corresponding structures shown in fig. \[fig:giant\], where we observe that the long range lamellar order is lost at high $\phi_c$ in favor of a disordered arrangement of smaller DNA-bilayer fragments. The transition is consistent with previously described membrane rupture at high charge densities resulting from the electrostatic stresses that develop in the complex [@oded_prl; @oded_bpj]. The highly charged cationic membrane fragments strongly associate with the DNA rods to form the disordered structures seen in fig. \[fig:giant\] (a)-(d). We should not, therefore, be surprised that the values of $d$ and $d_{\rm DNA}$, which we inferred for lamellar structures, deviate from the expected behavior. The nearly constant behavior of $d_{\rm DNA}$ vs. $\phi_c$ in the disordered phase can be well observed in the structures seen in fig.\[fig:giant\] (a)-(d). The decrease in $d$ in this regime is also consistent with the transition into the disordered phase, where the highly charged cationic membranes become squeezed between the negatively charged DNA rods.
Fig. \[fig:companion\] (b) shows the synchrotron X-ray scattering data reported in ref. [@safinya_science1]. The agreement with the simulation results in (a) is obvious, lending credibility to our CG model as well as to the Fourier space analysis of the resulting structures. The horizontal axes of the figures express the inverse of the membrane charge density using different scales: $1/\phi_c$ in (a) and $L/D$ (the mass ratio between the lipid and DNA material) in (b). The scales are linearly related by: $2.2(1/\phi_c)=L/D$. In both figures, the lamellar spacing approaches the asymptotic value $d\simeq
2l_{\rm LIP}+D_{\rm DNA}$ at small membrane change densities. These asymptotic values are different in the two figures, but this is merely a consequence of the chosen model parameters for $l_{\rm LIP}$ and $D_{\rm DNA}$, which slightly differ from the experimental values. Both figures exhibit a weak, and very similar, monotonic decrease of $d$ with increasing membrane charge density. For fully charge membranes, the value of $d$ is depressed by about 15-20% compared to the low density asymptote. In ref. [@safinya_currop] this decrease has been attributed to the difference in length between DOPC (neutral) and DOTAP (cationic), the latter being about 6[Å]{} shorter than the former. We, however, observe the same behavior with CLs and NLs being geometrically identical. Our simulations point to two more explanations for this observation: For moderately charged membranes the effective bilayer thickness slightly shrinks with $\phi_c$ due to the tensile stress induced by the (negative) electrostatic energy density of a confined charge neutral systems [@lekner1]. For high membrane charge density, the decrease in $d$ is likely related to the loss of lamellar order. In this regime, the derived value of $d=2\pi/q_{\rm LAM}$ does not necessarily coincide with the actual interlayer spacing, since the derivation is based on the presumption that the complex is ideally lamellar.
The agreement between the simulation and experimental results for $d_{\rm DNA}$ is also clear. In both fig. \[fig:companion\] (a) and (b), we observe that the DNA spacing drops from $d_{\rm DNA}\approx
60$[Å]{} at low charge densities to $d_{\rm DNA}\approx 30$[Å]{} at high charge densities, in a manner which is well approximated by a linear relationship with the inverse charge density. At very high charge densities, both figures exhibit the same deviation from a linear relationship between $d_{\rm DNA}$ and the inverse charge density. This feature has been attributed in ref. [@safinya_science1] to the limiting contact distance, $D_{\rm
DNA}$, between DNA rods. This interpretation of the results is correct provided that the hydration shell is included in the contact distance. Our results provide yet another possibility. Visual inspection of the self-assembled structures show that the DNA rods do not experience any hard core interactions. The plateau-like behavior of $d_{\rm DNA}$ at high charge densities is related to the formation of disordered structures which enable a more loose packing of the DNA rods.
What we have described in this paper is based on the very close agreement between the computational and experimental results shown in fig. \[fig:companion\]. These results can be explained by a structural shift from lamellar to fragmented geometry occurring at high membrane charge densities. The fragmentation of the membranes is consistent with the membrane rupture observed in our earlier work described in refs. [@oded_prl; @oded_bpj], where we used a different CG membrane model. This consistency gives us confidence that the loss of structural integrity of the membrane at high charge densities is not an artifact of a particular model. However, we recognize that although this work features the largest complexes ever simulated, there may still be some finite size effects that obscure the comparison with experiments. One such finite size effect is related to the periodic boundary conditions along the DNA axis, which enforce the infinite DNA rods to lie parallel to each other. This constraint, which simplifies the computational scheme, may lead to the formation of structures with artificial spatial correlations between the DNA rods. Another finite size effect is related to the relatively small sizes of the simulated complexes which, therefore, have scattering plots with peaks that are broader than in the corresponding experimental scattering plots. This low resolution makes it difficult to infer the degree of order from the width of the peaks. Yet another consequence of finite sizes is the enhancement of surface effects. While it seems plausible that the increase in the electrostatic tensile stress at high charge densities does proliferate structural defects, it can be that these defects form more easily on the boundaries of the complex and, therefore, they become over-expressed in our smaller complexes. Nevertheless, the close agreement between fig. \[fig:companion\] (a) and (b), over the entire range of charge densities, supports the possibility that structural variations observed in our simulations may take place in nature. Such a structural shift from lamellar to fragmented geometry should have implications for gene therapy. The shift may explain the improvement in transfection efficiency (TE) exhibited by these complexes at high membrane charge densities [@transfection1]. One of the main limiting stages in the transfection process is the release of the genetic material from the complex into the cytoplasm of the host cell. It is indeed reasonable to expect that the DNA rods will be more readily released from the fragmented disordered complexes than from lamellar structures with long range order. Given the remarkable success of our model, which is based on a highly CG representation of the constituting molecular species and their interactions, it is fair to anticipate the application of this model for obtaining direct observations of the mechanisms governing transfection and gene delivery.
Acknowledgments
===============
We thank Uri Raviv for very useful discussions on X-ray scattering and Cyrus Safinya for his critical reading of the paper. This work was supported by the Israel Science Foundation (Grant Number 946/08). NGJ also acknowledges support from The President’s Fund for Visiting Scientists at Ben Gurion University (Israel) and from Danmarks Nationalbank (Denmark).
[99]{}
P. L. Felgner and G. Rhodes, [*Nature*]{}, 1991, [**349**]{}, 351-352.
N. Smyth-Templeton and D. D. Lasic, Eds. [*Gene Therapy. Therapeutic Mechanisms and Strategies*]{}; Marcel Dekker Inc.: New York, 2000.
P. L. Felgner, M. J. Heller, P. Lehn, J.-P. Behr and F. C. Szoka, Eds. [*Artificial Self-Assembling Systems for Gene Delivery*]{}; American Chemical Society: Washington DC, 1996.
A. D. Miller, [*Angew. Chem. Int. Ed. Engl.*]{}, 1998, [**37**]{}, 1768-1785.
L. Huang, M. C. Hung and E. Wagner,Eds., [ *Non-Viral Vectors for Gene Therapy*]{}; Academic Press: San Diego, 1999.
K. Ewert, N. L. Slack, A. Ahmad, H. M. Evans, A. J. Lin, C. E. Samuel and C. R. Safinya, [ *Curr. Med. Chem.*]{}, 2004, [**11**]{}, 133-149.
S. Huebner, B. J. Battersby, R. Grimm and G. Cevc, [*Biophys. J.*]{}, 1999, [**76**]{}, 3158-3166.
M. T. Kennedy, E. V. Pozharski, V. A. Rakhmanova and R. C. MacDonald, [*Biophys. J.*]{}, 2000, [**78**]{}, 1620-1633.
H. F. Willard, [*Science*]{}, 2000, [ **290**]{}, 1308-1309.
J. O. Rädler, I. Koltover, T. Salditt and C. R. Safinya, [*Science*]{}, 1997, [**275**]{} 810-814.
I. Koltover, T. Salditt, J. O. Rädler and C. R. Safinya, [*Science*]{}, 1998, [**281**]{}, 78-81.
O. Farago and N. Gr[ø]{}nbech-Jensen, [ *J. Am. Chem. Soc.*]{}, 2009, [**131**]{}, 2875-2881.
D. Harries, S. May, W. M. Gelbart and A. Ben-Shaul, [*Biophys. J.*]{}, 1998, [**75**]{}, 159-173.
S. May, D. Harries and A. Ben-Shaul, [*Biophys. J.*]{}, 2000, [*78*]{}, 1681-1697.
H. Noguchi and M. Takasu, [ *Phys. Rev. E*]{}, 2001, [**64**]{}, 041913.
H. Tapia-McClung and N. Gr[ø]{}nbech-Jensen, [*J. Polym. Sci. Part B: Polym. Phys.*]{}, 2005, [**43**]{}, 911-916; [*ibid*]{}, 2010, [**48**]{}, 2604-2605.
N. Gr[ø]{}nbech-Jensen, [*Int. J. Mod. Phys. C*]{}, 1997, [**8**]{} 1287-1297
N. Gr[ø]{}nbech-Jensen, [*Comput. Phys. Commun.*]{}, 1999, [**119**]{} 115-121.
H. Noguchi, [*Phys. Rev. E*]{}, 2003, [**67**]{}, 041901.
U. Raviv, D. J. Needleman, Y. Li, H. P. Miller, L. Wilson and C. R. Safinya, [*Proc. Natl. Acad. Sci. USA*]{}, 2005, [**102**]{}, 11167-11172.
U. Raviv, Private Communication, 2010.
T. Ben-Nun, A. Ginsburg, P. Sz[é]{}kely and U. Raviv, [*J. Appl. Cryst.*]{}, 2010, [**43**]{}, 1522-1531.
O. Farago, N. Gr[ø]{}nbech-Jensen and P. Pincus, [*Phys. Rev. Lett.*]{}, 2006, [**96**]{}, 018102.
O. Farago and N. Gr[ø]{}nbech-Jensen, [*Biophys. J.*]{}, 2007, [**92**]{}, 3228-3240.
C. R. Safinya, [ *Curr. Opin. Struct. Biol.*]{}, 2001, [**11**]{}, 440-448.
![Self-assembled complex structures consisting of 32 DNA rods mixed with 800 CLs. The amount of NLs varies from 0 ($\phi_c=1$) at (a) to 2200 ($\phi_c=4/15$) at (l). Each structure is initiated in a random molecular configuration, and has evolved for $10-50\times 10^6$ MD time steps. The structures are viewed along the DNA axes. Color coding: grey - hydrophobic lipid beads, red - charged hydrophilic heads, green - neutral hydrophilic heads, and blue - DNA rods. For each configurations, the scattering intensities of the DNA rods and the lipids are also plotted. The open arrows indicate the position of the lipid peaks at $q_{\rm LAM}$ and $2q_{\rm LAM}$. The solid arrow indicate the position of the DNA in-plane correlation peak at $q_{\rm
DNA}$. []{data-label="fig:giant"}](fig1.eps){width="16cm"}
![Sketch of an idealized lamellar complex.[]{data-label="fig:ideal"}](fig2.eps){width="16cm"}
![(a) The inter-layer lamellar spacing $d$ and the DNA-spacing $d_{\rm DNA}$ vs. $1/\phi_c$, as computed from the peaks indicated by arrows in the scattering plots shown in fig. [\[fig:giant\]]{}. (b) The same quantities derived from synchrotron X-ray scattering data reported in ref. [[@safinya_science1]]{}. $1/\phi_c=1$ in (a) corresponds to $L/D=2.2$ in (b). (b) has been adopted from [[@safinya_science1; @safinya_currop]]{} with permission.[]{data-label="fig:companion"}](fig3.eps){width="16cm"}
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The photoproduction of $\omega$ mesons on nuclei has been investigated using the Crystal Barrel/TAPS experiment at the ELSA tagged photon facility in Bonn. The aim is to study possible in-medium modifications of the $\omega$ meson via the reaction $\gamma + A \rightarrow \omega + X \rightarrow \pi^0 \gamma + X^\prime$. Results obtained for Nb are compared to a reference measurement on a $\rm{LH_2}$ target. While for recoiling, long-lived mesons ($\pi^0$, $\eta$ and $\eta^\prime$), which decay outside of the nucleus, a difference in the lineshape for the two data samples is not observed, we find a significant enhancement towards lower masses for $\omega$ mesons produced on the Nb target. For momenta less than $\rm{500~MeV/c}$ an in-medium $\omega$ meson mass of $\rm{M_{medium}=[722_{-2}^{+2}(stat)_{-5}^{+35}(syst)]}~\rm{MeV/c^2}$ has been deduced at an estimated average nuclear density of $0.6~\rho_0$.'
author:
- |
D. Trnka $^1$, G. Anton $^2$, J. C. S. Bacelar $^3$, O. Bartholomy $^4$, D. Bayadilov $^{4,~8}$, Y.A. Beloglazov $^8$, R. Bogendörfer $^2$, R. Castelijns $^3$, V. Crede $^{4,^\ast}$, H. Dutz $^5$, A. Ehmanns $^4$, D. Elsner $^5$, R. Ewald $^5$, I. Fabry $^4$, M. Fuchs $^4$, K. Essig $^4$, Ch. Funke $^4$, R. Gothe $^{5,^\diamond}$, R. Gregor $^1$, A. B. Gridnev $^8$, E. Gutz $^4$, S. Höffgen $^5$, P. Hoffmeister $^4$, I. Horn $^4$, J. Hössl $^2$, I. Jaegle $^7$, J. Junkersfeld $^4$, H. Kalinowsky $^4$, Frank Klein $^5$, Fritz Klein $^5$, E. Klempt $^2$, M. Konrad $^5$, B. Kopf $^{6,9}$, M. Kotulla $^7$, B. Krusche $^7$, J. Langheinrich $^{5,^\diamond}$, H. Löhner $^3$, I.V. Lopatin $^8$, J. Lotz $^4$, S. Lugert $^1$, D. Menze $^5$, J. G. Messchendorp $^3$, T. Mertens $^7$, V. Metag $^1$, C. Morales $^5$, M. Nanova $^1$, R. Novotny $^1$, M. Ostrick $^5$, L. M. Pant $^{1,^\dagger}$, H. van Pee $^1$, M. Pfeiffer $^1$, A. Roy $^{1,^\ddagger}$, A. Radkov $^8$, S. Schadmand $^{1,^\star}$, Ch. Schmidt $^4$, H. Schmieden $^5$, B. Schoch $^5$, S. Shende $^3$, G. Suft $^2$, V. V. Sumachev $^8$, T. Szczepanek $^4$, A. Süle $^5$, U. Thoma $^{1,4}$, R. Varma $^{1,^\ddagger}$, D. Walther $^5$, Ch. Weinheimer $^{4,^+}$, Ch. Wendel $^4$\
(The CBELSA/TAPS Collaboration)
title: 'First observation of in-medium modifications of the $\omega$ meson'
---
The modification of experimentally observable properties of vector mesons such as mass and width, when embedded in a dense medium, is one of the most fundamental research issues in hadron physics. While for composite systems like molecules, atoms or nuclei, the mass of the system is almost completely (apart from small binding energy effects) governed by the sum of the masses of the constituents, this is no longer true in the hadronic sector. Here, the hadron masses are much larger than the summed masses of the constituents, the u-, d- and s-quarks. One possible interpretation is that the masses of hadrons are generated dynamically [@wilczek]. Furthermore, hadron masses can be associated with the spontaneous breaking of chiral symmetry. In nuclear matter (and at high temperatures) this symmetry is predicted to be at least partially restored. As a consequence, the properties of hadrons are expected to be modified (see e.g. [@meis; @klimt; @brown; @hats; @shu1]).
A variety of theoretical models predict a lowering of the in-medium mass of vector mesons even at normal nuclear matter density $\rho_0$. For the $\omega$ meson a drop of the mass by 20 to 150 $\rm{MeV/c^2}$ and a broadening of the width up to 60 $\rm{MeV/c^2}$ has been predicted (e.g. [@renk; @klingl; @riek; @lyk; @weid; @eff]). However, the discussion in the literature is very controversial. Even upward mass shifts [@zsch] or the appearance of additional peaks [@lutz] have been suggested by some authors. This situation underlines the importance of experimental results.
{width="0.68\columnwidth"} {width="0.68\columnwidth"} {width="0.68\columnwidth"}
Several previous experiments have studied the properties of vector mesons in hot and dense matter. Dilepton spectroscopy allows to measure the in-medium properties without distortion due to final state interactions (FSI). The CERES collaboration at CERN, for example, measured the low-mass $e^+e^-$ pair production in heavy-ion collisions and observed an enhancement in the mass range of $0.3~\rm{GeV/c^2} \le m_{e^+e^-} \le 0.7~\rm{GeV/c^2}$ over the yield expected from the known sources in pp collisions [@ag1; @ag2]. More recently, analyzing $\pi^+\pi^-$ pairs the STAR experiment at RHIC observed a decrease of the $\rho$ meson in-medium mass in peripheral Au+Au collisions [@star]. The KEK-PS E325 collaboration investigated $p+A$ reactions at 12 GeV [@kek] and reported an enhancement in the $e^+e^-$ invariant mass spectra in the region of $0.6 ~\rm{GeV/c^2} \le m_{e^+e^-} \le 0.77~\rm{GeV/c^2}$. Presently, an experiment performed at JLAB using a photon beam is being analyzed [@jlab]. At GSI, it has been proposed [@gsi] to perform pion induced experiments with the HADES [@gsi1] detector system.
All $e^+e^-$ experiments suffer from the small branching ratios (BR) of vector mesons into dileptons, which are in the order of $10^{-5} - 10^{-4}$. In addition, the comparable $e^+e^-$ decay rates for $\omega$- and $\rho$ mesons make it difficult to isolate an $\omega$ signal from the $e^+e^-$ invariant mass spectrum [@riek]. An alternative and promising approach to investigate in-medium modifications of the $\omega$ meson is to study the $\omega \rightarrow \pi^0 \gamma$ decay mode, as pointed out in [@muehl; @messch; @sib]. An essential advantage of this decay channel is the large BR of almost $9 \%$, three orders of magnitude larger than the decay into dileptons. Furthermore, this mode is a clean and exclusive probe to study the $\omega$ in-medium properties since the $\rho \rightarrow \pi^0 \gamma$ BR is only $6.8 \cdot 10^{-4}$ and therefore suppressed by two orders of magnitude relative to the $\omega$ BR into this channel. However, the disadvantage is a possible rescattering of the $\pi^0$ within the nuclear medium, which would distort the deduced $\omega$ invariant mass distribution. Pion rescattering within the nucleus proceeds predominately via the formation of an intermediate $\Delta$ resonance. Due to the kinematics of the $\Delta$ resonance decay the distorted events are predicted to accumulate at approx. 500 $\rm{MeV/c^2}$, far below the nominal $\omega$ invariant mass. This leads to a small contribution of only about $3 \%$ in the mass range of interest, $0.6~\rm{GeV/c^2} < M_{\pi^0 \gamma} < 0.9~\rm{GeV/c^2}$. Moreover, the authors of [@messch] and [@muehl] have demonstrated that the constraint on the kinetic energy $T_{\pi^0} > 150$ MeV suppresses the FSI down to the $1 \%$ level.
Only $\omega$ mesons decaying inside the nucleus carry information on the in-medium properties. To enhance the in-medium decay probability, the vector meson decay length $L_\omega = p_\omega / m_\omega \Gamma_\omega$ should be less than the nuclear radius. This can be achieved by applying a kinematic cut on the 3-momentum of the $\omega$ meson. But still, only a fraction of the $\omega$ mesons will decay inside the nucleus. Thus, one expects the $\pi^0 \gamma$ invariant mass spectra to show a superposition of decays outside of the target at the vacuum mass peak position (782 $\rm{MeV/c^2}$) with modified decays inside the nucleus [@messch].
{width="0.71\columnwidth"} {width="0.71\columnwidth"} {width="0.71\columnwidth"}
The experiment was performed at the [**EL**]{}ectron [**S**]{}tretcher [**A**]{}ccelerator (ELSA) in Bonn, using a 2.8 GeV electron beam. The photon beam was produced via bremsstrahlung. A magnetic spectrometer (tagger) was used to determine the photon beam energies within the tagged photon range of 0.64 to 2.53 GeV. The Nb and $\rm{LH_2}$ targets had thicknesses of 1 mm and 53 mm, respectively, and 30 mm in diameter. The targets were mounted in the center of the Crystal Barrel detector (CB), a photon calorimeter consisting of 1290 CsI(Tl) crystals ($\sim$ 16 radiation lengths $X_0$) with an angular coverage of $30^\circ$ up to $168^\circ$ in the polar angle and a complete azimuthal angle coverage. Inside the CB, covering its full acceptance, a three-layer scintillating fiber detector (513 fibers of 2 mm diameter) was installed for charged particle identification. Reaction products emitted in forward direction were detected in the TAPS detector. TAPS consisted of 528 hexagonally shaped $\rm{BaF_2}$ detectors ($\sim$ 12 $X_0$) covering polar angles between $4^\circ$ and $30^\circ$ and the complete $2\pi$ azimuthal angle. In front of each $\rm{BaF_2}$ module a 5 mm thick plastic scintillator was mounted for the identification of charged particles. The resulting geometrical solid angle coverage of the combined system was $ 99 \%$ of $4\pi$. The $\rm{BaF_2}$ crystals were read out by photomultipliers providing a fast trigger, the CsI(Tl) crystals via photodiodes. For further details see [@cb; @taps; @taps1].
The invariant masses of the mesons were calculated from the measured 4-momenta of the decay photons. The calibration of the Nb and $\rm{LH_2}$ data samples was carefully cross checked by comparing the lineshapes for long-lived mesons, the $\pi^0$, $\eta$, and $\eta^\prime$. The decay lengths ($c\tau$) of 25.1 nm ($\pi^0$), 0.153 nm ($\eta$), and 0.001 nm ($\eta^\prime$) guarantee that these pseudoscalar mesons will not decay inside the nucleus, hence the lineshapes should not exhibit any difference for the two data samples. Fig. \[cal\_check\] shows the comparison of the background subtracted invariant mass distributions for $\pi^0\rightarrow\gamma\gamma$, $\eta\rightarrow\pi^0\pi^0\pi^0\rightarrow6\gamma$ and $\eta^\prime\rightarrow\pi^0\pi^0\eta\rightarrow6\gamma$. Indeed, a difference in the lineshapes is not observed. However, when comparing the $\omega\rightarrow\pi^0\gamma$ invariant mass distributions, we find a significant change in the lineshapes. The left panel of Fig. \[om\_mom\] shows the $\pi^0\gamma$ invariant mass distribution without further cuts except for a three momentum cutoff of $|\vec{p}_{\omega}| < \rm{500~MeV/c}$. The dominant background source is two pion production where one of the four photons escapes the detection. This probability was determined by Monte Carlo simulations to be $14 \%$. The resulting three photon final state is not distinguishable from the $\omega \rightarrow \pi^0 \gamma$ invariant mass. The central panel of Fig. \[om\_mom\] shows the invariant mass distribution obtained after background subtraction. We observe the expected superposition of decays outside of the nucleus at the nominal vacuum mass with decays occuring inside the nucleus, responsible for the shoulder towards lower invariant masses. The high mass part of the $\omega$ mass signal appears identical for the Nb and $\rm{LH_2}$ targets, indicating that this part is dominated by $\omega$ meson decays in vacuum. These decays are eliminated by matching the right hand part of the Nb invariant mass spectrum to the $\rm{LH_2}$ data (see central panel of Fig. \[om\_mom\]) and by subtracting the two spectra from each other. For this normalization the integral of the undistorted spectrum corresponds to $75 \%$ of the counts in the Nb spectrum. This is in good agreement with a theoretical prediction obtained from a transport code calculation [@muehl; @muehl1]. There, about $16 \%$ of the total decays are predicted to occur inside the nuclear medium ($\rho > 0.1\cdot \rho_0$) without any FSI and $3 \%$ of the events are distorted due to FSI in the mass range of $0.6 ~\rm{GeV/c^2} < M_{\pi^0 \gamma} < 0.9~\rm{GeV/c^2}$. In addition, $9 \%$ of the events are moved towards lower masses due to the $\Delta$ decay kinematics. The right panel of Fig. \[om\_mom\] shows the resulting in-medium signal along with a Voigt fit (Breit-Wigner folded with Gaussian) to the data. We obtain an $\omega$ in-medium mass of $\rm{M_{medium}=[722_{-2}^{+2}(stat)_{-5}^{+35}(syst)]}~\rm{MeV/c^2}$. This corresponds to a lowering of the $\omega$-mass by $8~\%$ with respect to the vacuum value at an estimated average nuclear density of $0.6~\rho_0$ in line with the assumptions in [@messch]. Consistency with a scaling of the $\omega$-mass by $m = m_0(1-0.14 \rho / \rho_0)$ is found [@brown]. Within this scenario the width is governed by the experimental resolution of $\Gamma = 55~\rm{MeV/c^2}$ (FWHM). The systematic uncertainty mainly reflects different assumptions for the subtraction of decays of the $\omega$ mesons in vacuum. The fraction of these decays was varied within a broad range from $80 \%$ to $45 \%$ (the central and right panel of Fig. \[om\_mom\] correspond to $75 \%$). The case with $45 \%$ corresponds to the upper bound of the systematic uncertainty (+35 MeV). This extreme scenario would, however, require an increase of the in-medium width of the $\omega$ by almost an order of magnitude.
![(color online). $\pi^0 \gamma$ mass spectrum after background subtraction and FSI suppression ($T_{\pi^0} > 150$ MeV) for different $\omega$ momentum bins. Solid histogram: Nb data, dashed histogram: $\rm{LH_2}$ data. \[om\_dep\]](fig3.eps){width=".83\columnwidth"}
Furthermore, the dependence of the signal on the $\omega$ momentum has been studied. It is expected that only low-momentum $\omega$ mesons (with a corresponding low velocity) decay inside the nucleus and carry information on the in-medium properties of the $\omega$ meson. Fig. \[om\_dep\] shows the $\pi^0 \gamma$ invariant mass distribution after background subtraction and FSI suppression ($T_{\pi^0} > 150$ MeV) for different $\omega$-momentum bins. A pronounced modification of the lineshape is only observed for $\omega$ momenta in the range of $\rm{200~MeV/c}< |$ $\vec{p_\omega}$ $|<\rm{400~MeV/c}$. In Fig. \[om\_dep1\] the mean value of the mass distribution is plotted against the three momentum for the $\rm{LH_2}$ and the Nb data. This result might allow to extract the momentum dependence of the $\omega$-nucleus potential [@muehl; @muehl1].
In summary, we have investigated the in-medium modifications of $\omega$ mesons in photoproduction experiments using the Crystal Barrel/TAPS detector at the ELSA accelerator facility in Bonn. When comparing data from a $\rm{LH_2}$ target with data taken with a Nb target, we find a pronounced modification of the $\omega$ meson mass in the nuclear medium for $\omega$ mesons with momenta less than 500 MeV/c. The in-medium mass has been determined to $\rm{M_{medium}=[722_{-2}^{+2}(stat)_{-5}^{+35}(syst)]}~\rm{MeV/c^2}$ at an estimated average nuclear density of $0.6~\rho_0$. The width is found to be $\Gamma = 55~\rm{MeV/c^2}$ and is dominated by the experimental resolution. The momentum dependence of the signal shows that only low-momentum $\omega$ mesons contribute to the downward mass shift. In contrast, $\omega$ mesons with high momenta decay outside the nucleus, exhibiting an invariant mass distribution corresponding to $\omega$ decays in vacuum. First evidence for a lowering of the $\omega$ mass in the nuclear medium has been observed.
![(color online). Mean value of the $\pi^0 \gamma$ invariant mass as a function of the $\omega$ momentum at an estimated average density of $0.6~\rho_0$ for the Nb data (circles) and the $\rm{LH_2}$ (crosses) along with a fit. The systematic errors are determined by varying the integration range of $\rm{0.65~\rm{GeV/c^2}<M_{\pi^0 \gamma}<0.9~GeV/c^2}$ by $\rm{50~MeV/c^2}$ and are shown as a bar chart for the Nb data. In contrast to the peak positions the mean values do not reach the nominal vacuum mass of $\rm{782~MeV/c^2}$ due to the detector response function. \[om\_dep1\]](fig4.eps){width=".73\columnwidth"}
We gratefully acknowledge stimulating discussions with W. Cassing, S. Leupold, U. Mosel, and in particular with P. Mühlich. We thank the accelerator group of ELSA as well as the technicians and scientists of the HISKP in Bonn, the PI in Bonn and the II. Physikalisches Institut in Giessen. This work was supported by the Deutsche Forschungsgemeinschaft, SFB/Transregio 16, and the Schweizerischer Nationalfond. U. Thoma thanks for an Emmy-Noether grant from the DFG.
[30]{} F. Wilczek [*et al.*]{}, Phys. Today August 10 (2002). V. Bernard, U. Meißner, Nucl. Phys. A [**489**]{} 647 (1988). S. Klimt [*et al.*]{}, Phys. Lett. B [**249**]{} 386 (1990). G. E. Brown, M. Rho, Phys. Rev. Lett. [**66**]{} 2720 (1991). T. Hatsuda, S. H. Lee, Phys. Rev. C [**46**]{} R34 (1992). E. V. Shuryak, G. E. Rho, [*et al.*]{}, Nucl. Phys. A [**717**]{} 322 (2003). T. Renk [*et al.*]{}, Phys. Rev. C [**66**]{} 014902 (2002). F. Klingl [*et al.*]{}, Z. Phys. A [**356**]{} 193 (1996). F. Riek, J. Knoll, Nucl. Phys. A [**740**]{} 287 (2004). G. I. Lykasov [*et al.*]{}, Eur. Phys. J. A [**6**]{} 71 (1999). T. Weidmann [*et al.*]{}, Phys. Rev. C [**59**]{} 919 (1999) . M. Effenberger [*et al.*]{}, Phys. Phys. C [**60**]{} 027601 (1999). S. Zschocke [*et al.*]{}, Phys. Lett. B [**562**]{} 562 (2003). M. Lutz [*et al.*]{}, Nucl. Phys. A [**706**]{} 431 (2002). G. Agakichiev [*et al.*]{}, Phys. Rev. Lett. [**75**]{} 1272 (1995). G. Agakichiev [*et al.*]{}, Phys. Lett. B [**422**]{} 405 (1998). J. Adams [*et al.*]{}, Phys. Rev. Lett. [**92**]{} 092301 (2004). R. Muto [*et al.*]{}, J. Phys. G: Nucl. Part. Phys. [**30**]{} 1023 (2004). Jefferson Lab proposal, E-01-112, unpublished (2001) and C. Tur, priv. comm. (2005) W. Schön [*et al.*]{}, Act. Phys. Pol. B [**27**]{} 2959 (1996). J. Friese, Prog. Part. Nucl. Phys. [**42**]{} 235 (1999). J. G. Messchendorp [*et al.*]{}, Eur. Phys. J. A [**11**]{} 95 (2001). P. Mühlich [*et al.*]{}, Eur. Phys. J. A [**20**]{} 499 (2004). A. Sibirtsev [*et al.*]{}, Phys. Lett. B [**483**]{} 405 (2000). E. Aker [*et al.*]{}, Nucl. Instr. Meth. A [**321**]{} 69 (1992) and H. Kalinowsky [*et al.*]{}, in preperation. R. Novotny [*et al.*]{}, IEEE Trans. Nucl. Sci. [**38**]{} 392 (1991). A. R. Gabler [*et al.*]{}, Nucl. Inst. Meth. A [**364**]{} 164 (1994). P. Mühlich, priv. comm. (2004).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'A quantum symmetric pair is a quantization of the symmetric pair of universal enveloping algebras. Recent development suggests that most of the theory for quantum groups can be generalised to the setting of quantum symmetric pairs. In this paper, we study the $\imath$quantum group at roots of $1$. We generalize Lusztig’s quantum Frobenius morphism in this new setting. We define the small $\imath$quantum group and compute its dimension.'
address:
- 'Department of Mathematics, National University of Singapore, Singapore 119076, Singapore.'
- 'Department of Mathematics, University of Virginia, Charlottesville, Virginia 22904, United states.'
author:
- Huanchen Bao
- Thomas sale
bibliography:
- 'Ref.bib'
title: Quantum symmetric pairs at roots of $1$
---
[A]{}
[\^[inv]{}]{}
[\_]{}
[\_’]{}
[\_]{}
[\_L]{}
[\_M]{}
[B\_i\^[(n)]{}]{}
[\^L]{}
[[Z]{}]{}
[[C]{}]{}
[[N]{}]{}
[[Z]{}]{}
[[F]{}]{}
[h]{}
[n]{}
[[g]{}]{}
[[i]{}]{}
[V]{}
[E]{}
[[x]{}]{}
[u]{}
[(v)]{}
[\^]{}
[\_]{}
[[\^]{}]{}
[\^]{}
[\^]{}
[e]{}
[f]{}
[k]{}
[t]{}
[[\_]{}]{}
[Y\^[w\_0]{}]{}
[\^]{}
[\_]{}
[<\_]{}
[\^]{}
Introduction
============
Let $\U$ be the quantum group of a semisimple complex Lie algebra $\mathfrak{g}$ with a parameter $v$. When $v$ is a root of $1$, the algebra $\U$ is closely related with the modular representation theory of algebraic groups and the representation theory of affine Lie algebras; cf. [@Lu90a §0.6] and [@AJS94].
Let $\mA = \Z[v,v^{-1}]$ with a generic parameter $v$. We denote by $_{\mA}\U$ the $\mA$-form of $\U$, which is an $\mA$-subalgebra of $\U$ generated by the divided powers $E_{i}^{(n)}$ and $F_{i}^{(n)}$ for various simple root $i$ and $n \in \Z_{\ge 0}$. Let $l$ be an odd positive integer, and prime to $3$ if $\U$ has a component of type $G_2$. Let $\mA'$ be the quotient of $\mA$ by the two-sided ideal generated by the $l$-th cyclotomic polynomial, that is, $v$ is an $l$-th root of $1$ in $\mA'$. Let ${}_{\mA'} \U = \mA' \otimes_{\mA} {}_{\mA}\U $.
Let ${}_{\mA'}\U^*$ be the universal enveloping algebra (over $\mA'$) of the Lie algebra $\mathfrak{g}$ (we refer to §\[subsub:Ustar\] for a more precise definition of the algebra ${}_{\mA'}\U^*$ in terms of root data). Lusztig [@Lu90b Theorem 8.10] defined the quantum Frobenius morphism $$\begin{aligned}
\label{eq:LuFr}
{\bold{Fr}}: {}_{\mA'}\U &\longrightarrow {}_{\mA'}\U ^{*}, \\
\notag E_{i}^{(n)} &\mapsto \begin{cases}
E_{i}^{(n/l)}, &\mbox{for $n \in l\mathbb{Z}$}; \\
0, &\mbox{otherwise};
\end{cases}
\\
\notag F_{i}^{(n)} &\mapsto
\begin{cases}
F_{i}^{(n/l)} , &\mbox{for $n \in l\mathbb{Z}$}; \\
0, &\mbox{otherwise}.
\end{cases}\end{aligned}$$
Lusztig also defined the small quantum group $ {\mathfrak{u}}$ as an $\mA'$-subalgebra of ${}_{\mA'}\U$ generated by $F_i^{(a)} $, $E_i^{(a)} $ with $0 \le a < l$. The algebra $ {\mathfrak{u}}$ should be viewed as the “Frobenius kernel” of the morphism (cf. [@GK93 Appendix]).
A quantum symmetric pair (QSP) $(\U, \Ui)$ is a quantization of the symmetric pair of enveloping algebras $(\U(\mathfrak{g}), \U(\mathfrak{g}^\theta))$, where $\theta : \mathfrak{g} \rightarrow \mathfrak{g}$ denotes an involution of Lie algebras. They were originally developed by Letzter [@Le99] to study quantum homogeneous spaces and reflection equations in finite type, some of which was generalized to Kac-Moody type by Kolb [@Ko14]. The algebra $\Ui$ is a coideal subalgebra of $\U$, which we often call the $\imath$quantum group.
Recent development in the theory of quantum symmetric pairs suggests that many of the fundamental constructions and results for quantum groups can be generalised to the setting of quantum symmetric pairs. In [@BW18a; @BW18b; @BW18c], the first author and Wang developed the theory of canonical bases arising from quantum symmetric pairs (called the $\imath$canonical bases). The integral form of the $\imath$quantum group is spanned by the $\imath$canonical basis over $\mA$, and generated by the $\imath$divided powers introduced in [@BW18c] as an $\mA$-algebra. The $\imath$divided powers are the generalization of the usual divided powers, but are far more complicated.
Since $\Ui$ lacks suitable triangular decomposition, we generally consider the modified form $\Uidot$, together with the modified quantum group $\Udot$. The modified quantum group $\Udot$ (resp. $\Uidot$) is obtained by replacing the identity element in $\U$ (resp. $\Ui$) with a collection of idempotents.
We have the natural generalization of the quantum Frobenius morphism in the setting of $\Udot$, as well as the modified small quantum group $\dot{\mathfrak{u}}$ as established in [@Lu §35.1.9].
Let ${}_{\mA}\Uidot$ be the integral form of the modified $\imath$quantum group. In this paper, we study the modified $\imath$quantum group ${}_{\mA'}\Uidot = \mA' \otimes_{\mA} {}_\mA\Uidot $ at the $l$-th root of $1$, where $l$ is a positive odd integer that is relatively prime to $3$ if $\U$ has a component of type $G_{2}$. We remark that while one can consider more general $l$ in the setting of quantum groups [@Lu Chap. 35], it is much more restricted in the setting of $\imath$quantum groups (cf. Remark \[rem:even\]).
We prove that Lusztig’s quantum Frobenius morphism restricts to an algebra homomorphism on modified $\imath$quantum groups in Theorem \[thm:main\]. We define the (modified) small $\imath$quantum group ${}_{\mA'}\dot{\mathfrak{u}}^\imath$ and compute its dimension in Theorem \[thm:main2\]. The $\imath$divided powers of the $\mA$-algebra ${}_\mA\Uidot$ studied in [@BW18c; @BeW18] play important roles in this paper.
This paper is organised as follows. We recall quantum groups and quantum symmetric pairs in Section \[sec:2\]. We define and study the $\imath$quantum group $\U^{*,\imath}$ in Section \[sec:3\]. We explicitly compute the restriction of the quantum Frobenius morphism on ${}_{\mA'}\Uidot$ in Section \[sec:4\]. We define the modified small $\imath$quantum group and compute its dimension in Section \[sec:5\].
[**Acknowledgement:** ]{}The authors would like to thank Weiqiang Wang for his helpful discussions and for encouraging the collaboration. HB is supported by a NUS start-up grant. TS was partially supported by Wang’s NSF Grant DMS-1702254.
Preliminaries {#sec:2}
=============
In this section, we follow the conventions of [@Lu] and [@BW18b].
Cartan data and root data
-------------------------
###
Let $\langle Y, X, \dots \rangle$ be a root datum of finite type $(\I, \cdot)$; cf. [@Lu §1.1.1]. Let $\Phi \subset X$ (resp. $\Phi^+ \subset X$) be the set of roots (resp. positive roots). Let $\Phi^\vee \subset Y$ (resp. $\Phi^{\vee,+} \subset Y$) be the set of coroots (resp. positive coroots).
Let $v$ be an indeterminate and write $\mathcal{A} = \mathbb{Z}[v,v^{-1}].$ We define, for $n \in \mathbb{Z}$ and $a \in \mathbb{Z}_{\geq0}$, $$[n] = \frac{v^{n} - v^{-n}}{v - v^{-1}}, \quad [n]^{!} = [n][n-1]. . .[2][1], \quad \left[ \begin{array}{cc|r} n \\ a \end{array} \right] = \frac{[n]^{!}}{[n-a]^{!}[a]^{!}}.$$ For any $ i \in \I$, we define $v_{i} := v^{(i \cdot i)/2}$ and also $$[n]_i = \frac{v_i^{n} - v_i^{-n}}{v_i - v_i^{-1}}, \quad [n]_i^{!} = [n]_i[n-1]_i. . .[2]_i[1]_i, \quad \left[ \begin{array}{cc|r} n \\ a \end{array} \right]_i = \frac{[n]_i^{!}}{[n-a]_i^{!}[a]_i^{!}}.$$
We also define $$[n]_{v^2_i} = \frac{v_i^{2n} - v_i^{-2n}}{v^2_i - v_i^{-2}}, \text{ and similarly define } \quad [n]_{v^2_i}^{!}, \quad \left[ \begin{array}{cc|r} n \\ a \end{array} \right]_{v^2_i}.$$
### {#subsec:stardatum}
Throughout this paper, we assume that $l > 0$ is an odd integer, and $l$ is prime to $3$ if the Cartan datum has a factor of $G_2$. We define a new Cartan datum $(\I, \circ)$ with the same $\I$ and the pairing $$i \circ j = (i \cdot j)l_{i}l_{j},$$ where $l_i$ is the smallest positive integer such that $l_{i}(i \cdot i)/2 \in l\mathbb{Z}$; cf. [@Lu §2.2.4].
We further define a new root datum $(Y^{*},X^{*}, \dots)$ of type $(\mathbb{I}, \circ)$ following [@Lu §2.2.5]. Define $X^{*} = \{ \la \in X | \mbox{$\langle i, \la \rangle \in l_i\mathbb{Z}$ for all $i \in I$} \}$ and $Y^{*} = \mathrm{Hom}(X^{*}, \mathbb{Z})$ with the obvious bilinear pairing, which we denote by $\langle \cdot,\cdot \rangle^*$. The map $\I \rightarrow X^*$ is given by $i \mapsto i\rq{}^* = l_i i\rq{} \in X$. The map $\I \rightarrow Y^*$ associates to $i \in \I$ the element $i^* \in Y^*$ such that $\langle i^*, \zeta \rangle^* = \langle i, \zeta \rangle / l_i$ for any $\zeta \in X^*$.
We define $v_{i}^{*} :=v_{i}^{(i \circ i)/2}$. We also define $[n]^*_i$, $([n]_i^*)^{!}$, $\left[ \begin{array}{cc|r} n \\ a \end{array} \right]_i^*$ in the obvious way, for $i \in \I$.
\[lem:sametype\] We have $ \langle i^*, j\rq{}^{*} \rangle ^* = \langle i, j\rq{} \rangle$ for any $i, j \in \I$.
Recall $\langle i^*, j\rq{}^{*} \rangle ^* = \langle i, l_j j\rq{}\rangle/ l_i $. If $ \langle i, j\rq{}\rangle = 0$ or $j = i$, we trivially have $\langle i^*, j\rq{}^{*} \rangle ^* = \langle i, j\rq{}\rangle$. Otherwise, we could have $\langle i, j\rq{}\rangle = 1, 2, 3$ thanks for our finite type assumption. Then since $l$ is odd and prime to $3$ if there is a $G_2$ factor, we must have $l_i = l_j$. The lemma follows.
As a consequence of the proof, we must have $l=l_i$ for all $i \in \I$ (cf. [@Lu90b §8.4]).
Quantum groups
--------------
### {#section-7}
Given a root datum $(Y, X,\dots)$ of type $(\mathbb{I}, \cdot)$, let $\U$ be the associated quatnum group over $\Qq$ generated by $E_{i}$, $F_{i}$, and $K_{\mu}$, for all $i \in I$ and $\mu \in Y$. We write $E_{i}^{(n)} = E_{i}^{n}/[n]_{i}^{!}$ and $F_{i}^{(n)} = F_{i}^{n}/[n]_{i}^{!}$ for $n \in \mathbb{Z}_ {\ge 0}$.
Recall [@Lu §23.1] the modified version of $\U$, denoted by $\dot{\U}$. The algebra $\dot{\U}$ has the direct sum decomposition $$\dot{\U} = \bigoplus_{\la, \la' \in X} {}_{\la}\U_{\la'}.$$ The $\mathcal{A}$-form $_{\mathcal{A}}\dot{\U}$ is the $\mA$-subalgebra of $\dot{\U}$ generated by: $$E_{i}^{(n)}1_{\la} = 1_{\la + n i'}E_{i}^{(n)} = 1_{\la + n i'}E_{i}^{(n)}1_{\la} \in {}_{\la + ni'}\U_{\la}
,$$ and $$F_{i}^{(n)}1_{\la} = 1_{\la - ni'}F_{i}^{(n)} = 1_{\la - ni'}F_{i}^{(n)}1_{\la} \in {}_{\la - ni'}\U_{\la}
,$$ for various $n \in \mathbb{Z}_{\geq 0}$, $i \in \I$, $\lambda \in X$.
We define the modified quantum group with scalars in a commutative $\mathcal{A}$-algebra $R$ as follows: $${}_{R}\dot{\U} := R \otimes_{\mathcal{A}} {}_\mathcal{A}\dot{\U}.$$
### {#subsec:com}
We consider the completion ${\Udot}^\wedge$ of $\Udot$ as follows; cf. [@Lu §36.2.3]. Recall any element in $\Udot$ can be written uniquely as a (finite) sum $\sum_{\lambda, \lambda'} x_{\lambda,\lambda'}$ for $x_{\lambda,\lambda'} \in {}_\lambda\U_{\lambda'}$. We consider infinite summation of the form $$\sum_{\lambda, \lambda'} x_{\lambda,\lambda'},$$ as long as there is a finite set $F \in X$ such that $x_{\lambda,\lambda'} =0$ unless $\lambda - \lambda' \in F$. The algebra structure of $\Udot$ extends naturally to an algebra structure of $\Udot^\wedge$. We similarly define ${}_\mA \Udot^\wedge$ and ${}_R \Udot^\wedge$ for any $\mA$-algebra $R$.
### {#subsub:Ustar}
Let $\U^*$ be the quantum group over the field $\Qq$ associated with the root datum $(Y^{*},X^{*}, \dots)$ of type $(\mathbb{I}, \circ)$, generated by (by abuse of notations) $E_{i}$, $F_{i}$, and $K_{\mu}$, for all $i \in I$ and $\mu \in Y^*$.
We abuse the notations and write $E_{i}^{(n)} := E_{i}^{n}/([n]_{i}^*)^{!} \in \U^*$ and $F_{i}^{(n)} := F_{i}^{n}/([n]_{i}^*)^{!} \in \U^*$. We similarly define $\dot{\U}^*$, ${}_{\mathcal{A}}\dot{\U}^*$, $_{R}\U^*$, ${}_{R}\dot{\U}^*$.
We also define similarly the completions ${\Udot}^{*,\wedge}$, ${}_\mA{\Udot}^{*,\wedge}$, as well as ${}_R{\Udot}^{*,\wedge}$ for any $\mA$-algebra $R$.
### {#section-8}
Let $\mA\rq{}$ be the quotient of $\mA$ by the two-sided ideal generated by the $l$-th cyclotomic polynomial $f_{l} \in \mA$. Recall that $(f_1, f_2, f_3, \dots) = (v-1, v+1, v^2+v+1, \dots)$. We denote by $\phi : \mA \rightarrow \mA\rq{}$ the quotient map.
[@Lu §35.1.9]\[thm:LuFr\] There is a homomorphism of $\mA\rq{}$-algebras $\bold{Fr}: {}_{\mA\rq{}}\dot{\U} \rightarrow {}_{\mA\rq{}}\dot{\U}^{*}$ such that $$\bold{Fr}: E_{i}^{(n)}1_{\la} \mapsto
\begin{cases}
E_{i}^{(n/l)}1_{\la},&\text{for $n \in l\mathbb{Z}$, and $\la \in X^{*}$};\\
0, &\text{otherwise}.
\end{cases}$$ and $$\bold{Fr}: F_{i}^{(n)}1_{\la} \mapsto
\begin{cases}
F_{i}^{(n/l)}1_{\la}, &\mbox{for $n \in l\mathbb{Z}$, and $\la \in X^{*}$}; \\
0, &\mbox{otherwise}.
\end{cases}$$
Note that $\bold{Fr}$ extends natually to an algebra homomorphism $$\bold{Fr} : {}_{\mA\rq{}}\dot{\U}^\wedge \rightarrow {}_{\mA\rq{}}\dot{\U}^{*, \wedge}.$$
Quantum symmetric pairs
-----------------------
### {#section-9}
Let $\tau$ be an involution of the Cartan datum $(\I, \cdot)$; we allow $\tau =1$. We further assume that $\tau$ extends to an involution on $X$ and an involution on $Y$, respectively, such that the perfect bilinear pairing is invariant under the involution $\tau$. Let $\I_{\bullet} \subset \I$. We have a subroot datum $(X, Y, \dots)$ of type $(\I_\bullet, \cdot)$. Let $W_{\I_\bullet}$ be the parabolic subgroup of $W$ with $w_{\bullet}$ as its longest element. Let $\Phi_\bullet \subset X$ (resp. $\Phi^+_\bullet \subset X$) be the set of roots (resp. positive roots). Let $\Phi^\vee_\bullet \subset Y$ (resp. $\Phi^{\vee,+}_\bullet \subset Y$) be the set of coroots (resp. positive coroots).
Let $\rho^\vee_{\bullet}$ be the half sum of all positive coroots in the set $\Phi^{\vee}_{\bullet}$, and let $\rho_{\bullet}$ be the half sum of all positive coroots in the set $\Phi _{\bullet}$. We shall write $$\label{eq:white}
\I_{\circ} = \I \backslash \I_{\bullet}.$$ A pair $(\I_{\bullet}, \tau)$ is called [*admissible*]{} (cf. [@Ko14 Definition 2.3]) if the following conditions are satisfied (with respect to the root datum $(X,Y, \dots)$ of type $(\I, \cdot)$):
- $\tau (\I_{\bullet}) = \I_{\bullet}$;
- The action of $\tau$ on $\I_{\bullet}$ coincides with the action of $-w_{\bullet}$;
- If $j \in \I_{\circ}$ and $\tau(j) = j$, then $\langle \rho^\vee_{\bullet}, j' \rangle \in \Z$.
### {#section-10}
Let $ \inv = -w_{\bullet} \circ \tau$ be an involution of $X$ and $Y$. Following [@BW18b], we introduce the $\imath$-weight lattice and $\imath$-root lattice $$\begin{aligned}
\label{XY}
\begin{split}
X_{{\imath}} = X / \breve{X}, & \quad \text{ where } \; \breve{X} = \{ \la - \inv(\la) \vert \la \in X\},
\\
Y^{\imath} &= \{\mu \in Y \big \vert \inv(\mu) =\mu \}.
\end{split}\end{aligned}$$ For any $\la \in X$, we denote its image in $X_{\imath}$ by $\overline{\la}$.
The involution $\tau$ of $\I$ induces an automorphism of the $\Q(v)$-algebra $\U$, denoted also by $\tau$, under which $E_i \mapsto E_{\tau i}, F_i \mapsto F_{\tau i}$, and $K_\mu \mapsto K_{\tau \mu}$.
### {#section-11}
We recall here the definition of quantum symmetric pair $(\U, \Ui)$ following [@BW18b §3.3].
\[def:Ui\] The algebra $\Ui$, with parameters $$\label{parameters}
{\varsigma}_{i} \in \pm v^\Z, \quad \kappa_i \in \Z[v,v^{-1}], \qquad \text{ for } i \in \I_{\circ},$$ is the $\Qq$-subalgebra of $\U$ generated by the following elements: $$\begin{aligned}
F_{i} &+ {\varsigma}_i {\texttt T}_{w_{\bullet}} (E_{\tau i}) \widetilde{K}^{-1}_i
+ \kappa_i \widetilde{K}^{-1}_{i} \, (i \in \I_{\circ}),
\\
& \quad K_{\mu} \,(\mu \in Y^{\imath}), \quad F_i \,(i \in \I_{\bullet}), \quad E_{i} \,(i \in \I_{\bullet}).\end{aligned}$$ The parameters are required to satisfy Conditions -: $$\begin{aligned}
\label{kappa}
\begin{split}
\kappa_i &=0 \; \text{ unless } \tau(i) =i, \langle i, j' \rangle = 0 \; \forall j \in {\I_{\bullet}},
\\
&
\qquad
\text{ and } \langle k,i' \rangle \in 2\Z \; \forall k = \tau(k) \in {\I_{\circ}}\text{ such that } \langle k, j' \rangle = 0 \; \text{ for all } j \in {\I_{\bullet}};
\end{split}
\\
\overline{\kappa_i} &= \kappa_i; \label{kappa2}
\\
{\varsigma}_{i} & ={\varsigma}_{{\tau i}} \text{ if } i \cdot \theta (i) =0;
\label{vs=}
\\
{\varsigma}_{{\tau i}} &= (-1)^{ \langle 2\rho^\vee_{\bullet}, i' \rangle } v_i^{-\langle i, 2\rho_{\bullet}+{w_\bullet}\tau i ' \rangle} {\ov{{\varsigma}_{i} }}. \label{vs2}
$$
Note that we require ${\varsigma}_{i} \in \pm v^\Z$, instead of ${\varsigma}_{i} \in \Z[v,v^{-1}]$ as in [@BW18c]. Note that such ${\varsigma}_i$ always exists by [@BK15 Remark 3.14], thanks to the establishment of the [@BK15 Conjecture 2.7] in [@BW18c Theorem 4.1].
We denote by $\Uidot$ the modified coideal subalgebra as defined in [@BW18b §3.7]. It follows from [@BW18b §3.7] that $\Udot$ is naturally a $\Uidot$-bimodule.
### {#section-12}
We recall here the integral from ${}_\mA \Uidot$ of the modified coideal subalgebra $\Uidot$.
[@BW18c Definition 3.10] We define ${}_\mA \Uidot$ to be the set of elements $u \in \Uidot$ such that $u \cdot m \in{}_\mA\Udot$ for all $m \in {}_\mA\Udot$.
[@BW18c Theorem 5.1, Theorem 7.2, Corollary 7.5]\[prop:idiv\]
1. ${}_\mA \Uidot$ is a free $\mA$-module.
2. ${}_\mA \Uidot$ is generated as an $\mA$-subalgebra of $\Uidot$ by the $\imath$-divided powers $B^{(a)}_{i, \zeta}$ ($ i \in \I$) and $E^{(a)}_{j} 1_{\zeta}$ ($j \in {\I_{\bullet}}$) for $\zeta \in X_\imath$ and $a \ge 0$.
The $\imath$-divided powers were introduced in [@BW18c Theorem 5.1] in order to construct the canonical basis for quantum symmetric kac-Moody pairs. They play a crucial role in this paper.
### {#section-13}
We consider the completion ${\Udot}^{\imath,\wedge}$ of $\Uidot$ analogous to that of $\Udot$ in §\[subsec:com\] by allowing infinite summation $$\sum_{\lambda, \lambda'} x_{\lambda,\lambda'}$$ as long as there is a finite set $G \in X_\imath$ such that $x_{\lambda,\lambda'} =0$ unless $\lambda - \lambda' \in G$. The following lemma is straightforward.
\[lem:emd\] We have the algebra embedding $$\begin{aligned}
\imath: \Udot^{\imath,\wedge} &\longrightarrow {\Udot}^\wedge,\\
x &\mapsto \sum_{\lambda \in X} x \one_{\lambda}.\end{aligned}$$ In particular, $\imath$ restricts to embeddings $\imath: \Uidot \rightarrow {\Udot}^\wedge$, and $\imath: {}_\mA\Udot^{\imath} \rightarrow {}_\mA{\Udot}^\wedge$.
### {#section-14}
For a commutative $\mA$-algebra $R$, we define the modified coideal subalgebra with scalars in $R$ as $${}_R \Uidot = R \otimes_{\mA} {}_\mA \Uidot.$$ We similarly define $ {}_R \Udot^{\imath,\wedge}$.
\[prop:emd\] Let $R$ be a commutative $\mA$-algebra. The following induced embedding after base change is injective $$\label{eq:emb}
\imath: {}_R \Udot^{\imath} \longrightarrow {}_R{\Udot}^\wedge.$$
Recall the canonical basis $\dot{\B}^\imath$ of ${}_\mA \Uidot$ from [@BW18c Theorem 7.2]. For any $b_1\diamondsuit^\imath_\zeta b_2 \in \dot{\B}^\imath$ and $\lambda \in X$ such that $\overline{\lambda} = \zeta \in X_\imath$, we have $$b_1\diamondsuit^\imath_\zeta b_2 1_\lambda = b_1\diamondsuit_\lambda b_2 1_\lambda + \text{lower terms}.$$ Here $b_1\diamondsuit_\lambda b_2$ denotes Lusztigs canonical basis element ([@Lu Theorem 25.2.1]) on ${}_\mA \Udot$. The proposition follows, since the coefficient of the leading term is $1$.
The $\imath$quantum group $\U^{*,\imath}$ {#sec:3}
=========================================
In this section, we define the $\imath$quantum group associated with the root datum $(X^*, Y^*,\dots)$ and the pair $({\I_{\bullet}}, \tau)$.
Root data
---------
Recall the root datum $(X^*, Y^*, \dots)$ in §\[subsec:stardatum\]. Note that since $\tau$ is an involution of the Cartan datum $(\I, \cdot)$, it is naturally an involution the Cartan datum $(\I, \circ)$. The involution $\tau$ restricts to an involution of $X^* \subset X$. The involution $\tau$ extends naturally on $Y^* = \text{Hom}(X^*, \mathbb{Z})$, such that the perfect pairing $\langle \cdot, \cdot \rangle^*$ is $\tau$-invariant. Thanks to Lemma \[lem:sametype\], the pair $({\I_{\bullet}}, \tau)$ is admissible with respect to the root datum $(X^*, Y^*, \dots)$.
Recall the definition of $X_\imath$ and $Y^\imath$ in . We similarly define $X^*_\imath$ and $Y^{*,\imath}$. In particular, we have $$X^*_{{\imath}} = X^* / \breve{X}^*, \quad \text{ where } \; \breve{X}^* = \{ \la - \inv(\la) \vert \la \in X^*\},$$
If $\lambda \in X^*$, then we have $\theta (\lambda) \in X^*$.
Note that $X^*$ is $\tau$-invariant by our assumption on $\tau$. The sublattice $X^*$ is invariant under the Weyl group action as well, thanks to Lemma \[lem:sametype\]. Therefore $X^*$ is invariant under $\theta$.
So we have $\breve{X}^* \subset X^* \cap \breve{X}$ and the following commutative diagram: $$\xymatrix{ X^* \ar[r] \ar[d] & X \ar[d] \\
X^*_\imath \ar[r]& X_\imath}$$
If the root datum $(Y,X, \dots )$ is simply connected, then we have $\breve{X}^* = X^* \cap \breve{X}$.
We need to show that if $\langle i, \lambda -\theta(\lambda) \rangle \in l\mathbb{Z}$, then $\lambda -\theta(\lambda) = \mu -\theta(\mu)$ for some $\mu \in X^*$.
We write $X = \oplus_{i \in \I}\mathbb{Z} \omega_{i}$, where $\omega_i$ denotes the $i$-th fundamental weight. Let $$\lambda = \sum_{i \in \I } a_i \omega_i.$$ Note that since $$\theta(\omega_i) =
\begin{cases}
-\omega_{\tau i}, &\text{if } i \in {\I_{\circ}};\\
\omega_{i}, &\text{if } i \in {\I_{\bullet}},
\end{cases}$$ we have $$\lambda - \theta(\lambda) = \sum_{i \in {\I_{\circ}}} (a_i + a_{\tau i}) \omega_i.$$
Since $\langle i, \lambda -\theta(\lambda) \rangle \in l\mathbb{Z}$, we have $$a_i \equiv - a_{\tau i}, \mod l .$$ Since $l$ is odd, then we must have $$a_i \equiv a_{\tau i} \equiv 0, \mod l, \text{ if } i = \tau i.$$ Therefore by considering $\tau$-orbits in ${\I_{\circ}}$, we can find $1-l \le b_i \le l-1$ for each $i \in {\I_{\circ}}$ such that $$b_i + b_{\tau i} =0, \quad \text{ and } \quad b_i \equiv a_i \mod l.$$
Now we can simply take $$\mu = \sum_{i \in {\I_{\circ}}} a_i \omega_i - \sum_{i \in {\I_{\circ}}} b_i \omega_i.$$
Then since $\langle i, \mu \rangle \in l\mathbb{Z}$ for all $i \in \I$, we have $\mu \in X^*$. Since $b_i + b_{\tau i} = 0$ for all $i \in {\I_{\circ}}$ by definition, we have $$\mu - \theta(\mu) = \sum_{i \in {\I_{\circ}}} (a_i + a_{\tau i } + b_i + b_{\tau i}) \omega_i = \lambda - \theta(\lambda).$$ The lemma follows.
[**For the rest of paper, we assume $\breve{X}^* = X^* \cap \breve{X}$.**]{} Thanks to the previous lemma, the equality holds when the root datum $(Y,X, \dots )$ is simply connected.
Let $\lambda \in X$ such that $\overline{\lambda} \in X^*_\imath$, then $\lambda \in X^*$.
We have $$\lambda = \mu+ \nu -\theta(\nu), \text{ with } \mu, \nu \in X^*.$$ Then obviously we have $\lambda \in X^*$.
The $\imath$quantum group $\U^{*,\imath}$ {#the-imathquantum-group-uimath}
-----------------------------------------
We now define the $\imath$quantum group associated with the root datum $(X^*, Y^*,\dots)$ and the pair $({\I_{\bullet}}, \tau)$.
The algebra $\U^{*,\imath}$, with parameters $$\label{parameters}
{\varsigma}_{i}^* = {\varsigma}_i^{l^2} , \quad \kappa_i^* = \kappa_i^{l^2}, \qquad \text{ for } i \in \I_{\circ},$$ is the $\Qq$-subalgebra of $\U^*$ generated by the following elements: $$\begin{aligned}
F_{i} &+ {\varsigma}_i^* {\texttt T}_{w_{\bullet}} (E_{\tau i}) \widetilde{K}^{-1}_i
+ \kappa^* _i \widetilde{K}^{-1}_{i} \, (i \in \I_{\circ}),
\\
& \quad K_{\mu} \,(\mu \in Y^{\imath,*}), \quad F_i \,(i \in \I_{\bullet}), \quad E_{i} \,(i \in \I_{\bullet}).\end{aligned}$$
The parameters ${\varsigma}^*_i$ and $\kappa_i^*$ satisfy the following conditions: $$\begin{aligned}
\begin{split}
\kappa^*_i &=0 \; \text{ unless } \tau(i) =i, \langle i^*, j'{^*} \rangle^* = 0 \; \forall j \in {\I_{\bullet}},
\\
&
\qquad
\text{ and } \langle k^*,i'^* \rangle^* \in 2\Z \; \forall k = \tau(k) \in {\I_{\circ}}\text{ such that } \langle k^*, j'^* \rangle^* = 0 \; \text{ for all } j \in {\I_{\bullet}};
\end{split}
\\
{\varsigma}^*_{i} & ={\varsigma}^*_{{\tau i}} \text{ if } i \circ \theta (i) =0;
\\
{\varsigma}^*_{{\tau i}} &= (-1)^{ \langle 2(\rho_{\bullet}^*)^\vee, i'^* \rangle^* } (v^*_i)^{-\langle i^*, 2\rho^*_{\bullet}+{w_\bullet}\tau i'^* \rangle^*} \overline{{\varsigma}^*_{i} }.
$$
The lemma follows directly from Lemma \[lem:sametype\], and the fact that $v_i^* = v_i^{l^2}$.
So $(\U^*, \U^{\imath, *})$ is also a quantum symmetric pair as defined in Definition \[def:Ui\]. Hence results from [@BW18b; @BW18c] applies. Therefore, we can similarly define the following as before ($R$ is any commutative $\mA$-algebra) $${}_\mA\Udot^{*,\imath}, {}_R\Udot^{*,\imath}, \Udot^{*,\imath,\wedge}, {}_\mA\Udot^{*,\imath,\wedge}, {}_R\Udot^{*,\imath,\wedge}.$$ In particular, we have the following counterpart of Proposition \[prop:emd\]: $$\label{eq:emb*}
\imath: {}_R \Udot^{*,\imath,\wedge} \lhook\joinrel\longrightarrow {}_R \Udot^{*,\wedge}.$$
\[rem:even\] If we take $l$ to be even, then the pair $({\I_{\bullet}}, \tau)$ may not be admissible with respect to the root datum $(X^*, Y^*,\dots)$. Let us illustrate this phenomenon in the following example.
Let $(\I = \{\alpha_1,\alpha_2\}, \cdot)$ be the Cartan datum of type $B_2$, where $\alpha_2$ denotes the short root. Let $(X, Y, \dots)$ be any root datum of type $(\I, \cdot)$. We take ${\I_{\bullet}}= \{\alpha_2\}$ and $\tau = \id$. It follows that $({\I_{\bullet}},\tau)$ is admissible, which one can see from the Satake diagram. Let $l = 4$. Then $(\I = \{\alpha_1,\alpha_2\}, \circ)$ is actually of type $C_2$ with short root $\alpha_1$. It is easy to see that $({\I_{\bullet}},\tau)$ is not admissible anymore, which one can again observe from the Satake diagram.
Quantum Frobenius homomorphism {#sec:4}
==============================
Let $\mA\rq{}$ be the quotient of $\mA$ by the two-sided ideal generated by the $l$-th cyclotomic polynomial $f_{l} \in \mA$ through out this section. (Recall that $(f_1, f_2, f_3, \dots) = (v-1, v+1, v^2+v+1, \dots)$). We denote by $\phi : \mA \rightarrow \mA\rq{}$ the quotient map.
Recall the Frobenius morphism in Theorem \[thm:LuFr\]. Thanks to the embeddings and , we have $$\xymatrix{ {}_{\mA\rq{}}\Uidot \ar[r]^-\imath & {}_{\mA\rq{}} {\Udot}^\wedge \ar[d]^-{\bold{Fr}}\\
{}_{\mA\rq{}}\Udot^{*,\imath} \ar[r]^-\imath & {}_{\mA\rq{}} {\Udot}^{*,\wedge} }
.$$
Recall the parameters for the algebra $\Ui$ in . [**For the rest of the paper, we shall only consider the case when the parameter $\kappa_i =0$.**]{} We remark that this restriction is only relevant to the computation in §\[subsec:AI1\]. The goal of this section is to establish the following theorem:
\[thm:main\] The quantum Frobenius morphism restricts to an $\mA\rq{}$-algebra homomorphism $${\bold{Fr}}: {}_{\mA\rq{}}\Uidot \longrightarrow {}_{\mA\rq{}}\Udot^{*,\imath} .$$
Recall Proposition \[prop:idiv\] that ${}_{\mA\rq{}}\Uidot $ is generated by the $\imath$-divided powers $B^{(a)}_{i,\zeta}$ ($i \in \I$) and $E^{(a)}_j 1_\zeta$ ($j \in {\I_{\bullet}}$) for $\zeta \in X_i$ and $a \ge 0$. It suffices to prove that
1. \[FrobEj\] $ {\bold{Fr}}(\imath (E^{(a)}_j 1_{\zeta} )) = {\bold{Fr}}( \sum_{\lambda \in X }E^{(a)}_j 1_\lambda) \in \text{Image of } \imath : {}_{\mA\rq{}}\Udot^{*,\imath} \rightarrow {}_{\mA\rq{}} {\Udot}^{*,\wedge}, $
2. \[FrobBi\] $ {\bold{Fr}}(\imath (B^{(a)}_{i,\zeta} )) = {\bold{Fr}}( \sum_{\lambda \in X }B^{(a)}_{i,\zeta} 1_\lambda) \in \text{Image of } \imath : {}_{\mA\rq{}}\Udot^{*,\imath} \rightarrow {}_{\mA\rq{}} {\Udot}^{*,\wedge}.$
The proof reduces to real rank one quantum symmetric pairs. The statement is trivial. We shall prove the statement case by case for real rank one quantum symmetric pairs. We include the Satake diagram for real rank one quantum symmetric pairs for the readers’ convenience.
[| c | c || c | c |]{}
at (0, -0.15) [AI$_1$]{};
& $ \begin{tikzpicture}[baseline=0]
\node at (0,0) {$\circ$};
\node at (0,-.3) {\small 1};
\end{tikzpicture}
$ &
at (0, -0.15) [AII$_3$]{};
& $ \begin{tikzpicture}[baseline=0]
\node at (0,0) {$\bullet$};
\draw (0.1, 0) to (0.4,0);
\node at (0.5,0) {$\circ$};
\draw (0.6, 0) to (0.9,0);
\node at (1,0) {$\bullet$};
\node at (0,-.3) {\small 1};
\node at (0.5,-.3) {\small 2};
\node at (1,-.3) {\small 3};
\end{tikzpicture}
$
\
at (0, -0.2) [AIII$_{11}$]{};
&
at (-0.5,0) [$\circ$]{}; at (0.5,0) [$\circ$]{}; (-0.5, 0.2) to\[out=45, in=180\] (-0.15, 0.35) to (0.15, 0.35) to\[out=0, in=135\] (0.5, 0.2); at (-0.5,-0.3) [1]{}; at (0.5,-0.3) [2]{};
&
at (0, -0.2) [AIV, n$\ge$2]{};
&
\[baseline=6\] at (-0.5,0) [$\circ$]{}; (-0.4,0) to (-0.1, 0); at (0,0) [$\bullet$]{}; at (2,0) [$\bullet$]{}; at (2.5,0) [$\circ$]{}; (0.1, 0) to (0.5,0); (0.5,0) to (1.4,0); (1.6,0) to (1.9,0); (2.1,0) to (2.4,0); (-0.5, 0.2) to\[out=45, in=180\] (0, 0.35) to (2, 0.35) to\[out=0, in=135\] (2.5, 0.2); at (-0.5,-.3) [1]{}; at (0,-.3) [2]{}; at (2.5,-.3) [n]{};
\
at (0, -0.2) [BII, n$\ge$ 2]{};
&
at (1.05,0) [$\circ$]{}; at (1.5,0) [$\bullet$]{}; (1.1,0) to (1.4,0); (1.4,0) to (1.9, 0); (1.9,0) to (2.7,0); (2.7,0) to (2.9, 0); at (3,0) [$\bullet$]{}; (3.1, 0) to (3.7, 0); at (3.8,0) [$\bullet$]{}; at (1,-.2) [1]{}; at (1.5,-.2) [2]{}; at (3.8,-.2) [n]{};
&
at (0, -0.15) [CII, n$\ge$3]{};
&
(0.6, 0.15) to (0.9, 0.15); at (0.5,0.15) [$\bullet$]{}; at (1,0.15) [$\circ$]{}; at (1.5,0.15) [$\bullet$]{}; (1.1,0.15) to (1.4,0.15); (1.4,0.15) to (1.9, 0.15); (1.9, 0.15) to (2.1, 0.15); (2.1,0.15) to (2.7,0.15); (2.7,0.15) to (2.9, 0.15); at (3,0.15) [$\bullet$]{}; (3.1, 0.15) to (3.7, 0.15); at (3.8,0.15) [$\bullet$]{}; at (0.5,-0.15) [1]{}; at (1,-0.15) [2]{}; at (3.8,-0.15) [n]{};
\
at (0, -0.05) [DII, n$\ge$4]{};
&
at (1,0) [$\circ$]{}; at (1.5,0) [$\bullet$]{}; (1.1,0) to (1.4,0); (1.4,0) to (1.9, 0); (1.9,0) to (2.7,0); (2.7,0) to (2.9, 0); at (3,0) [$\bullet$]{}; at (3.5, 0.4) [$\bullet$]{}; at (3.5, -0.4) [$\bullet$]{}; (3.1, 0.1) to (3.4, 0.35); (3.1, -0.1) to (3.4, -0.35); at (1,-.3) [1]{}; at (1.5,-.3) [2]{}; at (3.5, 0.7) [n-1]{}; at (3.5, -0.6) [n]{};
&
at (0, -0.2) [FII]{};
&
at (0,0) [$\bullet$]{}; (0.1, 0) to (0.4,0); at (0.5,0) [$\bullet$]{}; (0.6, 0) to (1.2,0); at (1.3,0) [$\bullet$]{}; (1.4, 0) to (1.7,0); at (1.8,0) [$\circ$]{}; at (0,-.3) [1]{}; at (0.5,-.3) [2]{}; at (1.3,-.3) [3]{}; at (1.8,-.3) [4]{};
\
{#section-15}
Let $i\in {\I_{\circ}}$ be such that $\tau (i)\neq i$. This includes type AIII$_{11}$ and type AIV. We have $\kappa_i=0$, and $B_i = F_i +{\varsigma}_i {\texttt T}_{w_{\bullet}} (E_{\tau i}) {\widetilde{K}}^{-1}_i $. We write $$Y_i : = {\varsigma}_i {\texttt T}_{w_{\bullet}} (E_{\tau i}) {\widetilde{K}}^{-1}_i, \qquad Y_i^{(n)} = Y_i^n / [n]^!_i.$$
We have $F_i Y_i -v_i^{-2} Y_i F_i =[F_i, Y_i] {\widetilde{K}}^{-1}_i =0$. Following [@BW18c §5.5.1], we define $$B_{i,\zeta}^{(n)} = \frac{B_i^n}{[n]_i!} \one_{\zeta}
= \sum_{a=0}^n v_i^{-a(n-a)} Y_i^{(a)}F_i^{(n-a)} \one_\zeta \in {}_\mA \Uidot.$$
\[prop:ineqtaui\] We have that $$\begin{aligned}
{\bold{Fr}}: {}_{\mA\rq{}} {\Udot}^\wedge &\longrightarrow {}_{\mA\rq{}} {\Udot}^{*,\wedge},\\
\sum_{\lambda \in X }B^{(n)}_{i,\zeta} 1_\lambda &\mapsto
\begin{cases}
\sum_{\lambda \in X^* } B^{(n/l)}_{i, \zeta} 1_\lambda, &\text{if } n \in l\Z, \zeta \in X^*_\imath;\\
0, &\text{otherwise.}
\end{cases} \end{aligned}$$ In other words, we have $$\begin{aligned}
{\bold{Fr}}: {}_{\mA\rq{}}\Uidot &\longrightarrow {}_{\mA\rq{}}\Udot^{*,\imath},\\
B^{(n)}_{i,\zeta} &\mapsto \begin{cases}
B^{(n/l)}_{i, \zeta}, &\text{if } n \in l\Z, \zeta \in X^*_\imath;\\
0, &\text{otherwise.}
\end{cases} \end{aligned}$$
It suffices to check that $$\bold{Fr} (B^{(n)}_{i,\zeta} 1_\lambda ) = B^{(n/l)}_{i, \zeta} 1_\lambda, \text{ for } \zeta \in X_\imath^*, \overline{\lambda} = \zeta.$$ We check only the cases where $n = kl \in l\mathbb{Z}$. The other cases are entirely similar.
We can use the quantum binomial formula ([@Lu §1.3.5]) to write, for $\lambda \in X^{*}$ with $\overline{\lambda} =\zeta$, $$\begin{aligned}
\bold{Fr}(B^{(n)}_{i, \zeta} \one_{\lambda}) &= \bold{Fr}((\sum_{a=0}^{n} v_i^{-a(n-a)} Y_i^{(a)}F_i^{(n-a)})\one_\lambda)\\
&\stackrel{\heartsuit}{=} (\sum_{a=0}^{k} (v_i^*)^{-a(n-a)} Y_i^{(a)}F_i^{(n-a)})\one_\lambda\\
&= B_{i,\zeta}^{(k)}\one_{\la},\end{aligned}$$ where $\heartsuit$ follows from Theorem \[thm:LuFr\], and the fact that ${\varsigma}_i^{l^2_i} ={\varsigma}_i^*$.
{#section-16}
Let $i \in {\I_{\circ}}$ such that $\tau (i) = i \neq {w_\bullet}(i)$. This includes the types: BII, DII, AII$_3$, CII, and FII. We have $\kappa_i=0$, and $B_i = F_i +{\varsigma}_i {\texttt T}_{w_{\bullet}} (E_{\tau i}) {\widetilde{K}}^{-1}_i $.
We write $$Y_i = {\varsigma}_i {\texttt T}_{w_{\bullet}}(E_i) \widetilde{K}_i^{-1}, \quad b^{(n)}_i = \sum_{a=0}^n v_i^{-a(n-a)} Y_i^{(a)} F_i^{(n-a)}.$$ Following ([@BW18c (5.12)]), we define $$\label{eq:Bi}
B^{(n)}_{i, \zeta} \one_{\lambda} = b_i^{(n)}\one_{\lambda} + \frac{v}{v-v^{-1}} \sum_{k \ge 1} v_i^{\frac{k (k+1)}{2}} \mathfrak{Z}_{i}^{(k)} b_i^{(n-2k)}\one_{\lambda},$$ where $$\mathfrak{Z}_{i}^{(n)} = -v_{i}^{\frac{1}{2}n(n-1)}(\sum_{a=0}^{n-1}v_{i}^{-2n^{2} + 2na -\frac{1}{2}a(a-1)}Y_{i}^{(n-a)}F_{i}^{(n-a)}\mathfrak{Z}_{i}^{(a)} - F_{i}^{(n)}Y_{i}^{(n)}).$$
\[lem:bi\] We have that $$\begin{aligned}
{\bold{Fr}}: {}_{\mA\rq{}} {\Udot}^\wedge &\longrightarrow {}_{\mA\rq{}} {\Udot}^{*,\wedge},\\
\sum_{\lambda \in X }b^{(n)}_{i} 1_\lambda &\mapsto
\begin{cases}
\sum_{\lambda \in X^* } b^{(n/l)}_{i} 1_\lambda, &\text{if } n \in l\Z;\\
0, &\text{otherwise.}
\end{cases} \end{aligned}$$ In other words, we have $$\begin{aligned}
{\bold{Fr}}: {}_{\mA\rq{}}\Uidot &\longrightarrow {}_{\mA\rq{}}\Udot^{*,\imath},\\
b^{(n)}_{i} 1_\zeta &\mapsto
\begin{cases}
b^{(n/l)}_{i} 1_\zeta, &\text{if } n \in l\Z, \zeta \in X^*_\imath;\\
0, &\text{otherwise.}
\end{cases} \end{aligned}$$
This is the same computation as Proposition \[prop:ineqtaui\].
We have that $$\begin{aligned}
{\bold{Fr}}: {}_{\mA\rq{}} {\Udot}^\wedge &\longrightarrow {}_{\mA\rq{}} {\Udot}^{*,\wedge},\\
\sum_{\lambda \in X } B^{(n)}_{i, \zeta} 1_\lambda &\mapsto
\begin{cases}
\sum_{\lambda \in X^*} b^{(n/l)}_{i} 1_\lambda, &\text{if } n \in l\Z, \zeta \in X^*_\imath;\\
0, &\text{otherwise.}
\end{cases} \end{aligned}$$ In other words, we have $$\begin{aligned}
{\bold{Fr}}: {}_{\mA\rq{}}\Uidot &\longrightarrow {}_{\mA\rq{}}\Udot^{*,\imath},\\
B^{(n)}_{i} 1_\zeta &\mapsto
\begin{cases}
b^{(n/l)}_{i} 1_\zeta, &\text{if } n \in l\Z, \zeta \in X^*_\imath;\\
0, &\text{otherwise.}
\end{cases} \end{aligned}$$
We first have $$\bold{Fr}(\mathfrak{Z}_{i}^{(n)} \one_{\lambda}) = 0, \qquad \text{ for } 1 \leq n \leq l-1, \lambda \in X.$$ Moreover, $$\bold{Fr}(\mathfrak{Z}_{i}^{(l)} \one_{\lambda}) = - \bold{Fr}(Y_i^{(l)} F_i^{(l)} - F_i^{(l)} Y_i^{(l)}) \one_{\lambda}= \varsigma_{i}^{l}[F_{i}, {\texttt T}_{w_{\bullet}}(E_i)] \one_{\lambda} = 0, \text{ for any }\lambda \in X.$$
The last equality follows from case by case computation. It suffices to notice that in none of these cases is $w_{\bullet}(i\rq{}) - i\rq{}$ a root.
For $1 \leq n \leq l - 1,$ $$0 = \bold{Fr}(\mathfrak{Z}_{i}^{(kl)} \one_{\mu\rq{}})\bold{Fr}(\mathfrak{Z}_{i}^{(n)} \one_{\mu} ) = \left[ \begin{array}{cc|r} kl + n \\ n \end{array} \right]_{i}\bold{Fr}(\mathfrak{Z}_{i}^{(kl+n)} \one_{\mu})$$ and $$0 = \bold{Fr}(\mathfrak{Z}_{i}^{(l)} )^{k} \one_{\mu}= k!\bold{Fr}(\mathfrak{Z}_{i})^{(kl)} \one_{\mu}.$$ Therefore we have $$\bold{Fr}( \mathfrak{Z}_{i}^{(n)} \one_{\mu}) = 0 \quad \mbox{for $n \geq 1$}.$$ The proposition follows from and Lemma \[lem:bi\].
{#subsec:AI1}
In this final section, we consider the case $\tau i = i = {w_\bullet}i$. This is of type AI$_1$. In this case, we have $B_i = F_i + {\varsigma}_i E_i \widetilde{K}_i^{-1} $. Following the definition of $\imath$divided powers in this case given in [@BW18a], the precise formula for $B_{i,\zeta}^{(n)}$ has been obtained in [@BeW18] (recall the parameter $\kappa_i = 0$).
Let $\lambda \in X$ such that $\overline{\lambda} = \zeta$. The parity of $\langle i, \lambda \rangle \in \mathbb{Z}$ depends only on $\zeta$, but not on $\lambda$. We shall simply call it the parity of $\langle i, \zeta \rangle$. The computation divides into two cases depends on the parity of $\langle i, \zeta \rangle$. We shall focus on the case when $\langle i, \zeta \rangle$ is even. The odd case is entirely similar.
[@BeW18 Proposition 2.8] Let $\langle i, \zeta \rangle$ be even. Let $m \ge 1$, and $\lambda \in X$ such that $\overline{\lambda} = \zeta$. We have $$\label{eq:2m}
\begin{split}
B_{i, \zeta}^{(2m)} 1_\lambda
= & \sum_{c=0}^{m} \sum_{a=0}^{2m-2c} v_i^{2(a+c)(m-a-\langle i, \lambda \rangle/2) - 2ac - \binom{2c+1}{2}} \\
& \cdot\left[ \begin{array}{cc|r} m-c-a -\langle i, \lambda \rangle/2\\ c \end{array} \right]_{v_i^{2}}E^{(a)}_i F_i^{(2m-2c-a)} 1_\lambda,
\end{split}$$
$$\label{eq:2m-1}
\begin{split}
B_{i, \zeta}^{(2m-1)} 1_\lambda = &\sum_{c=0}^{m-1} \sum_{a=0}^{2m-1-2c} v_i^{2(a+c)(m-a-\langle i, \lambda \rangle/2) - 2ac -a- \binom{2c+1}{2}} \, \cdot \\
\quad &\left[ \begin{array}{cc|r} m-c-a -\langle i, \lambda \rangle/2 -1\\ c \end{array} \right]_{v_i^{2}}E^{(a)}_i F_i^{(2m-1-2c-a)} 1_\lambda.
\end{split}$$
\[prop:teven\] Let $\langle i, \zeta \rangle$ be even. We have, via restriction, $$\begin{aligned}
{\bold{Fr}}: {}_{\mA\rq{}} {\Udot}^\wedge &\longrightarrow {}_{\mA\rq{}} {\Udot}^{*,\wedge},\\
\sum_{\lambda \in X}B_{i, \zeta}^{(n)} 1_\lambda&\mapsto
\left\{ \begin{array}{rcl}
&\sum_{\lambda \in X^*}B_{i, \zeta}^{(n/l)} 1_\lambda, &\mbox{for $n \in l\mathbb{Z}$ and $\zeta \in X_\imath^{*}$}; \vspace{.2cm}\\
&\sum_{\lambda \in X^*} \left[ \begin{array}{cc|r} (l-1)/2 \\ b/2 \end{array} \right]_{v_i^{2}}B_{i, \zeta}^{(k)} 1_\lambda, &\begin{aligned}&\text{for $n = kl + b$, $k$ odd,}\\ &\text{$b$ even, $0 < b < l$, $\zeta \in X_\imath^{*}$}; \end{aligned}\\
&0, &\mbox{otherwise} .
\end{array}\right.
\end{aligned}$$
In other words, we have $$\begin{aligned}
{\bold{Fr}}: {}_{\mA\rq{}}\Uidot &\longrightarrow {}_{\mA\rq{}}\Udot^{*,\imath},\\
B_{i, \zeta}^{(n)} &\mapsto
\left\{ \begin{array}{rcl}
&B_{i, \zeta}^{(n/l)}, &\mbox{for $n \in l\mathbb{Z}$ and $\zeta \in X_\imath^{*}$}; \vspace{.2cm}\\
&\left[ \begin{array}{cc|r} (l-1)/2 \\ b/2 \end{array} \right]_{v_i^{2}}B_{i, \zeta}^{(k)}, &\begin{aligned}&\text{for $n = kl + b$, $k$ odd,}\\ &\text{$b$ even, $0 < b < l$, $\zeta \in X^{*}_\imath$}; \end{aligned}\\
&0, &\mbox{otherwise} .
\end{array}\right.
\end{aligned}$$
Let $\lambda \in X^*$ such that $\overline{\lambda} = \zeta$. We divide the computation into several cases.
\(1) We first consider the case when $n =2m$ is even, where either $ 0<n=2m<l$ or $n=2m=kl$. Recall for the expression of $B_{i, \zeta}^{(n)} 1_\lambda$.
(1.1) We assume $ 0<n=2m<l$. Then it follows from direct computation and [@Lu Lemma 34.1.2] that $${\bold{Fr}}(B_{i, \zeta}^{(n)} 1_\lambda) = 0.$$
(1.2) We assume $n=2m=kl$. Note that $${\bold{Fr}}(E_i^{(a)} F_i^{(2m-2c-a)}) =
\begin{cases}
E_i^{(a/l)} F_i^{((2m-2c-a)/l)}, &\text{if } l \vert a \text{ and } l \vert c;\\
0, &\text{otherwise}.
\end{cases}$$ However, if $l \vert a$ and $l \vert c$, we must have $v_i^{2(a+c)(m-a-\langle i, \lambda \rangle/2) - 2ac - \binom{2c+1}{2}} =1$.
Then, we have $$\begin{aligned}
&\bold{Fr}(B_{i, \zeta}^{(2kl)} 1_\lambda ) \\
= &\sum_{c=0}^{k} \sum_{a=0}^{2k-2c} \left[ \begin{array}{cc|r} kl-cl-al - \langle i, \lambda \rangle/2\\ cl \end{array} \right]_{v_i^{2}}E^{(a)}_i F_i^{(2k-2c-a)} 1_\lambda\\
= &\sum_{c=0}^{k} \sum_{a=0}^{2k-2c} \left( \begin{array}{cc|r} k-c-a - \langle i, \lambda \rangle^*/2\\ c \end{array} \right) E^{(a)}_i F_i^{(2k-2c-a)} 1_\lambda\\
= & B_{i, \zeta}^{(2k)} 1_\lambda.\end{aligned}$$
\(2) We then consider the case when $n =2m-1$ is odd, where either $0<n=2m-1<l$ or $n=2m-1 = (2k-1)l$. Recall for the expression of $B_{i, \zeta}^{(n)} 1_\lambda$ in this case.
(2.1) We assume $n = 2m-1 < l$. Then by direct computation, we must have $${\bold{Fr}}(B_{i, \zeta}^{(n)} 1_\lambda ) = 0 .$$ (2.2) We assume $n =2m-1= (2k-1)l$ for $2k-1 >0$. Then we have $$m-1 = (k-1)l + (l-1)/2, \quad\text{with } 0< (l-1)/2 < l.$$ Then $$\begin{aligned}
B_{i, \zeta}^{(n)} 1_\lambda = &\sum_{c=0}^{(k-1)l + ({l-1})/{2}} \sum_{a=0}^{(2k-1) l-2c} v_i^{2(a+c)(m-a-\langle i, \lambda \rangle/2) - 2ac -a- \binom{2c+1}{2}} \, \cdot \\
\quad &\left[ \begin{array}{cc|r} m-c-a -\langle i, \lambda \rangle/2 -1\\ c \end{array} \right]_{v_i^{2}}E^{(a)}_i F_i^{(2m-1-2c-a)} 1_\lambda.
\end{aligned}$$ Note that $${\bold{Fr}}(E_i^{(a)} F_i^{(2m-1-2c-a)}) =
\begin{cases}
E_i^{(a/l)} F_i^{((2m-1-2c-a)/l)}, &\text{if } l \vert a \text{ and } l \vert c;\\
0, &\text{otherwise}.
\end{cases}$$ However, if $l \vert a$ and $l \vert c$, we must have $v_i^{2(a+c)(m-a-\langle i, \lambda \rangle /2 ) - 2ac -a- \binom{2c+1}{2}} =1$.
Therefore $$\begin{aligned}
&\bold{Fr}(B_{i, \zeta}^{(n)} 1_\lambda ) \\
= &\sum_{c=0}^{(k-1)} \sum_{a=0}^{2k-1-2c} \left[ \begin{array}{cc|r}(k-1)l + ({l-1})/{2}-cl-al - \langle i, \lambda \rangle/2\\ cl \end{array} \right]_{v_i^{2}}E^{(a)}_i F_i^{(2k-1-2c-a)} 1_\lambda\\
\stackrel{\spadesuit}{=} &\sum_{c=0}^{(k-1)} \sum_{a=0}^{2k {-1}-2c} \left( \begin{array}{cc|r} k-1-c-a - \langle i, \lambda \rangle^*/2\\ c \end{array} \right) E^{(a)}_i F_i^{(2k-1-2c-a)} 1_\lambda\\
{=} & {B_{i, \zeta}^{(2k-1)} 1_\lambda},\end{aligned}$$ where $\spadesuit$ follows from [@Lu Lemma 34.1.2].
\(3) At last we consider the case where $n = kl +b$ with $0<b<l$. Note that since $\mA\rq{}$ is an integral domain, it suffices to perform the computation in the field of fractions.
Recall the following induction formula from [@BeW18 (2.5)]: $$\begin{aligned}
B_i B_{i, \zeta}^{(kl+c)} &= [kl+c+1]_{v_i} B_{i, \zeta}^{(kl+c+1)}, \quad &\text{if } kl+c \text{ is odd; }\\
B_i B_{i, \zeta}^{(kl+c)} &= [kl+c+1]_{v_i} B_{i, \zeta}^{(kl+c+1)} + [kl+c]_{v_i} B_{i, \zeta}^{(kl+c-1)}, \quad &\text{if } kl+c \text{ is even. }\end{aligned}$$ Recall $v_i^{l} =1$. Note that $[kl]_{v_i}=0$ and $[kl+b]_{v_i} = [b]_{v_i}\neq 0$ for all $0<b<l$.
(3.1) When $k$ (hence also $kl$) is odd, we have $${\bold{Fr}}(B_{i, \zeta}^{(kl+b)} 1_\lambda) = \begin{cases} (-1) \frac{[b-1]_{v_i}}{[b]_{v_i}} {\bold{Fr}}(B_{i, \zeta}^{(kl+b-2)} 1_\lambda), &\text{if } b \text{ is even;}\\
0, &\text{if } b \text{ is odd.}
\end{cases}$$ Therefore we have
$$\begin{aligned}
{\bold{Fr}}(B_{i, \zeta}^{(kl+b)} 1_\lambda) &= (-1)^{b/2 } \frac{[b-1]_{v_i}}{[b]_{v_i}} \frac{[b-3]_{v_i}}{[b-2]_{v_i}} \cdots \frac{[1]_{v_i}}{[2]_{v_i}} {\bold{Fr}}(B_{i, \zeta}^{(kl)} 1_\lambda) \\
& = \frac{[l-b+1]_{v_i}}{[b]_{v_i}} \frac{[l-b+3]_{v_i}}{[b-2]_{v_i}} \cdots \frac{[l-1]_{v_i}}{[2]_{v_i}} {\bold{Fr}}(B_{i, \zeta}^{(kl)} 1_\lambda) \\
& = \left[ \begin{array}{cc|r} (l-1)/2 \\ b/2 \end{array} \right]_{v_i^{2}} {\bold{Fr}}(B_{i, \zeta}^{(kl)} 1_\lambda).\end{aligned}$$
(3.2) When $k$ (hence also $kl$) is even, note that $$B_i B_{i, \zeta}^{(kl)} =[kl+1]_{v_i} B_{i, \zeta}^{(kl+1)}, \qquad \text{ since } [kl]_{v_i}=0.$$
Therefore by a similar computation as above, we have $${\bold{Fr}}(B_{i, \zeta}^{(kl+b)} 1_\lambda) = 0, \quad \text{ for all } 0<b<l.$$
Let $\langle i,\zeta \rangle$ be odd. We have $$\begin{aligned}
{\bold{Fr}}: {}_{\mA\rq{}} {\Udot}^\wedge &\longrightarrow {}_{\mA\rq{}} {\Udot}^{*,\wedge},\\
\sum_{\lambda \in X}B_{i, \zeta}^{(n)} 1_\lambda&\mapsto
\left\{ \begin{array}{rcl}
&\sum_{\lambda \in X^*}B_{i, \zeta}^{(n/l)} 1_\lambda, &\mbox{for $n \in l\mathbb{Z}$ and $\zeta \in X_\imath^{*}$}; \vspace{.2cm}\\
&\sum_{\lambda \in X^*} \left[ \begin{array}{cc|r} (l-1)/2 \\ b/2 \end{array} \right]_{v_i^{2}}B_{i, \zeta}^{(k)} 1_\lambda, &\begin{aligned}&\text{for $n = kl + b$, $k$ even,}\\ &\text{$b$ even, $0 < b < l$, $\zeta \in X^{*}$}; \end{aligned}\\
&0, &\mbox{otherwise} .
\end{array}\right.
\end{aligned}$$
In other words, we have $$\begin{aligned}
{\bold{Fr}}: {}_{\mA\rq{}}\Uidot &\longrightarrow {}_{\mA\rq{}}\Udot^{*,\imath},\\
B_{i, \zeta}^{(n)} &\mapsto \left\{ \begin{array}{rcl}
&B_{i, \zeta}^{(n/l)}, &\mbox{for $n \in l\mathbb{Z}$ and $\la \in X_\imath^{*}$}; \vspace{.2cm} \\
&\left[ \begin{array}{cc|r} (l-1)/2 \\ b/2 \end{array} \right]_{v_i^{2}}B_{i, \zeta}^{(k)}, &\!\begin{aligned}&\text{for $n = kl + b$, $k$ even,}\\ &\text{$b$ even, $0 < b < l$, $\zeta \in X_\imath^{*}$;}\end{aligned} \\
&0, &\mbox{otherwise}.
\end{array}\right.
\end{aligned}$$
The computation is entirely simialr to that of Proposition \[prop:teven\]. We shall omit the details here.
Small quantum symmetric pairs {#sec:5}
=============================
{#section-17}
Let ${}_{\mA'}\udot$ be the $\mA'$-subalgebra of $_{\mA'} \Udot$ generated by $E^{(n)}_i 1_\lambda$, $F^{(n)}_i 1_\lambda$ for various $i \in \I$, various $n$ such that $0 \le n <l$ and various $\lambda \in X$. Let ${}_{\mA'}\dot{\mathfrak{p}} = {}_{\mA'}\dot{\mathfrak{p}}_{\I_\bullet}$ be the $\mA'$-subalgebra of $_{\mA'} \Udot$ generated by $E^{(n)}_i 1_\lambda$, $F^{(n)}_j 1_\lambda$ for various $i \in \I_\bullet, j \in \I$, various $n$ such that $0 \le n <l$ and various $\lambda \in X$.
Let ${}_{\mA'}\dot{\mathfrak{u}}^\imath$ be the $\mA'$-subalgebra of ${}_{\mA'}\Uidot$ generated by $
B^{(n)}_{i,\zeta}$, $E_{j}^{(n)} 1_\zeta$ for various $i \in \I$, $j \in {\I_{\bullet}}$, $\zeta \in X_\imath$ and $n$ such that $0 \le n <l$. For any $\mA'$-commutative ring $R$, we define ${}_{R}\dot{\mathfrak{u}}^\imath = R \otimes_{\mA'} {}_{\mA'}\dot{\mathfrak{u}}^\imath$.
We call $({}_{\mA'}\dot{\mathfrak{u}}^\imath, {}_{\mA'}\udot)$ the small quantum symmetric pair. It is clear that ${}_{\mA'}\dot{\mathfrak{u}}^\imath$ is a “coideal subalgebra" of ${}_{\mA'}\udot$, that is, we have (via restriction) $$\Delta: {}_{\mA'}\dot{\mathfrak{u}}^\imath \longrightarrow \prod_{\zeta \in X_\imath^*, \lambda \in X^*} {}_{\mA'}\dot{\mathfrak{u}}^\imath 1_\zeta \otimes_{\mA'} {}_{\mA'}\udot 1_\lambda.$$
Here we abuse the terminology “coideal subalgebra", even though ${}_{\mA'}\dot{\mathfrak{u}}^\imath$ is not a subalgebra of ${}_{\mA'}\udot$.
\[thm:main2\] For any $\zeta \in X_\imath$, ${}_{\mA'}\dot{\mathfrak{u}}^\imath 1_\zeta$ is a free $\mA'$-module with rank $l^{\vert \Phi^{+}_\bullet \vert+ \vert \Phi^+ \vert}$.
We have the following linear isomorphism via a base change from [@BW18b Corollary 6.20] $$p_{\imath,\lambda}: \xymatrix{{}_{\mA'} \Uidot 1_\zeta \ar[r]^{\cong}& {}_{\mA'} \dot{\bold{P}}1_\lambda}, \qquad \overline{\lambda} =\zeta, \lambda \in X.$$ Via restriction, we have the map $$p_{\imath,\lambda}: {}_{\mA'} \dot{\mathfrak{u}}^\imath 1_\zeta \longrightarrow {}_{\mA'}\dot{\mathfrak{p}}1_\lambda,$$ whose surjectivity can be obtained similar to [@BW18b Corollary 6.20]. We know the image lies in ${}_{\mA'}\dot{\mathfrak{p}}1_\lambda$ thanks to the precise formulas of $
B^{(n)}_{i,\zeta}$ obtained in [@BW18b; @BeW18].
The theorem follows immediately from the fact that ${}_{\mA'}\dot{\mathfrak{p}}1_\lambda$ is a free $\mA'$-module with rank $l^{\vert \Phi^{+}_\bullet \vert+ \vert \Phi^+ \vert}$ thanks to [@Lu90b Theorem 8.3].
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'In this paper, we discuss a non-orthogonal multiple access (NOMA) scheme to exploit a high diversity gain using repetition, namely repetition-based NOMA. Unlike conventional power-domain NOMA, all the users can have the same transmit power, but different number of repetitions. Thanks to a high diversity gain, a low outage probability can be achieved without instantaneous channel state information (CSI) feedback for power allocation. A closed-form expression for an upper-bound on the outage probability is derived so that the values of key parameters can be decided to maintain the outage probability below a target value. We also consider the average error probability for finite-length codes. Simulation results are compared with the derived bounds and it is shown that the bounds are reasonably tight and can be used to decide key parameters (e.g., code rates) to guarantee target error probabilities.'
author:
- |
Jinho Choi\
[^1]
bibliography:
- 'noma.bib'
title: 'Repetition-based NOMA Transmission and Its Outage Probability Analysis'
---
non-orthogonal multiple access (NOMA); fading; outage probability
Introduction
============
Since non-orthogonal multiple access (NOMA) has a higher spectral efficiency than orthogonal multiple access (OMA), it has been extensively studied [@Ding_CM] [@Choi17_ISWCS], although there are a number of challenges (e.g., optimal user clustering [@Liu16] and beamforming [@Choi17_JCN] [@Seo18]). In [@Liang17] [@ChoiJSAC], the notion of NOMA can be employed in uncoordinated transmissions such as random access for uplink transmissions in order to improve the throughput, which is important for massive machine-type communication (MTC) that provides the connectivity for various Internet-of-Things (IoT) applications [@Bockelmann16].
In this paper, we consider a NOMA scheme that can have a low outage probability without channel state information (CSI) feedback so that it can be used for low-latency communication in MTC (thanks to no CSI feedback as well as a low outage probability). In conventional power-domain NOMA, the power allocation based on instantaneous CSI is essential for successful successive interference cancellation (SIC). If instantaneous CSI is not available, the power allocation can be carried out with statistical CSI to maximize the throughput as in [@Zhang16] (in [@Choi17_CSI] for downlink). In this case, there might be a delay due CSI feedback, and outage events (due to fading) and error propagation in SIC are inevitable. As a result, for reliable transmissions, re-transmissions are required and the resulting access delay becomes random and can be long. Furthermore, as in [@Xu16], when instantaneous CSI can be available using limited CSI feedback, due to quantization error and delay [@Choi02], CSI becomes imperfect, which leads to outage events and error propagation in SIC. Thus, if power-domain NOMA is applied to low-latency communication under time-varying fading (with statistical CSI or limited CSI feedback), there might be outage events. To keep the outage probability low without instantaneous CSI-based power allocation, we consider repetition-based NOMA that exploits a high diversity gain. The main difference from other power-domain NOMA with statistical CSI (e.g., [@Zhang16] [@Choi17_CSI]) is that the gain from NOMA is used to lower the outage probability rather than to increase the spectral efficiency.
In order to guarantee a certain low error probability, good performance prediction techniques are required so that the key parameters of repetition-based NOMA can be decided in advance. To this end, we focus on deriving a tight bound on the outage probability as a closed-form expression in this paper. In uplink NOMA, other users’ signals become interfering signals and their strength depends on fading, which makes the interference power a random variable. Since its probability density function (pdf) is unknown, we consider an approximation with the chi-squared distribution under Rayleigh fading by using a moment matching approach. Based on this approximation, we obtain an upper-bound on the outage probability with finite-length codes [@Polyanskiy10IT]. We also study decoding error probability for finite-length codes using the derived outage probability. Simulation results show that this bound is tight at a low outage probability. Thus, key parameters (e.g., the code rates) can be determined to keep a low outage probability or a high probability of successful SIC.
In summary, the main contributions are two-fold: [*i)*]{} repetition-based NOMA is proposed to exploit a high diversity gain for a low error probability without instantaneous CSI-based power allocation for low-latency communication; [*ii)*]{} a closed-form expression for the outage probability is derived, which allows to decide key parameters of repetition-based NOMA for a desirable performance (i.e., a certain low error rate).
The rest of the paper is organized as follows. In Section \[S:SM\], we present the system model for repetition-based NOMA to be used for uplink transmissions. To see the performance of repetition-based NOMA in terms of key parameters, we analyze the performance and find a tight upper-bound on the outage probability in Section \[S:PA\]. The error probability with finite-length codes is studied in Section \[S:EP\]. Simulation results are presented in Section \[S:Sim\] together with the bounds, which can help determine design criteria for repetition-based NOMA. The paper is concluded with some remarks in Section \[S:Conc\].
### Notation {#notation .unnumbered}
Matrices and vectors are denoted by upper- and lower-case boldface letters, respectively. The superscripts $\rT$ and $\rH$ denote the transpose and complex conjugate, respectively. The Kronecker delta is denoted by $\delta_{l,l^\prime}$, which is 1 if $l = l^\prime$ and 0 otherwise. $\uE[\cdot]$ and ${\rm Var}(\cdot)$ denote the statistical expectation and variance, respectively. $\cC\cN(\ba, \bR)$ represents the distribution of circularly symmetric complex Gaussian (CSCG) random vectors with mean vector $\ba$ and covariance matrix $\bR$. The Q-function is given by $\cQ(x) = \int_x^\infty \frac{1}{\sqrt{2 \pi} } e^{- \frac{t^2}{2} } dt$.
System Model {#S:SM}
============
In this section, we consider a NOMA scheme for uplink based on repetition with multiple radio resource blocks. As in [@Malak19], each block can be seen as a time-frequency resource block and it is assumed that a signal transmitted through a block experiences an independent block fading [@TseBook05]. Unlike conventional power-domain NOMA schemes, it is assumed that the average receive powers of users’ signals in the proposed scheme are the same (to this end, open-loop power control can be used, where each user can set its transmit power based on the average power of the received signal from a base station (BS) based on statistical channel reciprocity [@ChiangBook]). Thus, the proposed scheme may be suitable for uplink with users who cannot arbitrarily increase the transmit power for power-domain NOMA and a BS that does not perform instantaneous CSI-based power allocation.
Suppose that we have a set of $L$ parallel radio resource blocks (in the frequency domain), which is called a frame, for uplink transmissions. A frame is to be shared by multiple users in uplink transmissions for a high spectral efficiency based on NOMA. For NOMA, there are multiple layers and a user can transmit a different number of copies of a packet depending on his/her layer, where each layer[^2] is characterized by the number of copies per user and users in the same layer need to transmit their copies through different (orthogonal) blocks. In each block, there are $B$ signals transmitted (one signal from each layer), where $B$ represents the number of layers that are generated by power-domain NOMA. For example, as shown in Fig. \[Fig:frame\], with a frame consisting of 4 blocks, we can have 3 layers. A user in layer 1 is to transmit 4 copies of a packet through all 4 blocks. There are two users in layer 2 and each user is to transmit two copies of a packet through two blocks. In layer 3, there are 4 users and each user is to transmit a packet through a block. In this example, in each frame consisting of 4 block, there are 7 users in a frame and 3 co-existing signals per block, which clearly shows that the resulting scheme is a NOMA scheme. For convenience, the resulting scheme is referred to as repetition-based NOMA.
![A layered structure of frame with 3 layers and 4 blocks.[]{data-label="Fig:frame"}](frame.pdf){width="\figwidth"}
At the BS, the signals can be decoded with SIC as other NOMA schemes. In this paper, it is assumed that a signal in a lower layer transmits more copies (repetitions) than that in a higher layer. Thus, the BS is to decode the signals from layer 1 to layer $B$ with SIC, where $B$ represents the number of layers. From this, for successful SIC with a high probability, it is expected that the probability of decoding error is the lowest in layer 1 and might increase with $b \in \{1,\ldots, B\}$, where $b$ represents the layer index, thanks to the diversity gain[^3]. Although there can be as many as layers, the number of layers, $B$, has to be limited, while $B$ needs to be proportional to $L$ for a high spectral efficiency.
Suppose that user $k$ is in layer $1$. Let $\cU_l$ be the index set of the users transmitting signals through the $l$th block. In addition, denote by $\cL_k$ the index set of the blocks that are used for multiple transmissions by user $k$. Then, the received signal at the BS through the $l$th block is given by \_l = h\_[l,k]{} \_[l,k]{} + \_[q \_l k]{} h\_[l,q]{} \_[l,q]{} + \_l, l \_k, where $h_{l,k}$ represents the channel coefficient from user $k$ to the BS through the $l$th block, $\bs_{l,k}$ is the $l$th copy of the signal packet transmitted (through the $l$th block) by the $k$th user, and $\bn_l \sim \cC \cN(\b0, N_0 \bI)$ represents the background noise vector. For example, in Fig. \[Fig:frame\], with $L = 4$ and $B = 3$, suppose that user 1 lies in layer 1 and transmits signals through blocks $\cL_1 = \{1,2,3,4\}$. In addition, let users 2 and 3 be in layer 2 with $\cL_2 = \{1,2\}$ and $\cL_3 = \{3,4\}$, respectively, and let users 4, 5, 6, and 7 be in layer 3 with $\cL_4 = \{1\}$, $\cL_5 = \{2\}$, $\cL_6 = \{3\}$, and $\cL_7 = \{4\}$, respectively. In this case, with $k = 1$, we have $\cU_1 \setminus k = \{2, 4\}$, $\cU_2 \setminus k = \{2, 5\}$, $\cU_3 \setminus k = \{3, 6\}$, and $\cU_4 \setminus k = \{3, 7\}$.
It is assumed that the $l$th copy of user $k$’s signal is an interleaved block of the original signal block, denoted by $\bs_k$, i.e., \_[l,k]{} = \_[l,k]{}(\_k), where $\Pi_{l,k} (\cdot)$ denotes the interleaving operation for the $l$th copy at user $k$. In particular, a random permutation can be considered for the interleaving operation. In this case, $\Pi_{l,k}$ is seen as a random permutation matrix. For convenience, let $\Pi^{-1}_{l,k}$ be the deinterleaving operation. Throughout the paper, we also assume that $\uE[\bs_k] = \b0$ and $\uE[\bs_k \bs_q^\rH] = P \bI \delta_{k,q}$, where $P$ represents the signal power of all users.
At the BS, the decoding order with SIC corresponds to the number of layer, $b$, as mentioned earlier. That is, the signals in layer 1 are to be decoded first. To decode the signal from user $k$, the maximal ratio combining (MRC) [@BiglieriBook] [@ChoiJBook2] is used with deinterleaved signals as follows: $$\begin{aligned}
\by_k & = \sum_{l \in \cL_k} h_{l,k}^*
\Pi^{-1}_{l,k}(\br_l)\cr
& = \sum_{l \in \cL_k} |h_{l,k}|^2 \bs_k
+ \sum_{l \in \cL_k} h_{l,k}^* \bw_{l,k},
\label{EQ:byk}\end{aligned}$$ where \_[l,k]{} = \_[q \_l k]{} h\_[l,q]{} \_[l,k]{}\^[-1]{} (\_[l,q]{}) + \_[l,k]{}\^[-1]{}(\_l). Once all the signals in layer 1 are decoded, they can be removed from $\br_l$ using SIC. Then, the BS is to decode the signals in layer 2, and so on. In this case, $\cU_l$ is to be updated by removing the indices of the users in layer 1 because their signals in layer 1 are removed.
Decoding can be unsuccessful, which incurs error propagation and outage events due to incorrect SIC operation [@Zhang16] (which is also true for downlink NOMA as in [@Choi17_CSI]). Thus, it might be important to guarantee successful decoding with a high probability. To this end, it is necessary to have a sufficient number of repetitions or copies for a high diversity gain.
In repetition-based NOMA, since the powers are fixed and no power allocation is carried out, we consider rate allocation for successful decoding with a sufficiently high probability. To this end, it is necessary to know the error probability in terms of the code rate and other parameters. In the next sections, we will focus on the derivation of the error rate.
Performance Analysis {#S:PA}
====================
To guarantee a specific target error probability and decide key parameters (e.g., the code rate) accordingly at each layer, we need to predict the performance under given conditions. To this end, in this section, we focus on the performance analysis with the outage probability to allow such a prediction.
Outage Probability
------------------
In this subsection, we assume that the number of copies for user $k$ is $D = |\cL_k|$. In addition, it is assumed that all the interfering signals in the lower layers are removed by successful SIC. To find a closed-form expression for the outage probability, the following assumptions are mainly considered:
- The interleaving operation makes the copies of the signal $\bs_k$ uncorrelated, i.e., $$\begin{aligned}
\uE[\Pi_{l,k} (\bs_k) ( \Pi_{l^\prime,q} (\bs_q))^\rH ]& = P \bI
\delta_{l,l^\prime}
\delta_{k,q}
\cr
\uE[\Pi_{l,k}^{-1} (\bs_k) ( \Pi_{l^\prime,q}^{-1} (\bs_q))^\rH ]& = P \bI
\delta_{l,l^\prime}
\delta_{k,q}.\end{aligned}$$
- The channels are independent Rayleigh fading channels with = \_h\^2 \_[l,l\^]{} \_[k,q]{}. Thus, $X_{l,k} = |h_{l,k}|^2$ has the following exponential distribution: X\_[l,k]{} \~[Exp]{}(\_h\^2) = ( - ), X\_[l,k]{} 0. \[EQ:hpdf\]
Under the assumption of [**A1**]{}, we have = P \_[q, q\^]{}. From this, it can be shown that $$\begin{aligned}
\uE[\bw_{l,k} \bw_{l^\prime,k}^\rH ]
=
\left(\sum_{q \in \cU_l \setminus k} X_{l,q} P \bI
+ N_0 \right) \delta_{l, l^\prime}.\end{aligned}$$ The instantaneous signal-to-interference-plus-noise ratio (SINR) for user $k$ in becomes $$\begin{aligned}
\gamma_k
& = \frac{(\sum_{l=1}^D X_{l,k})^2P}{\sum_{l=1}^D
X_{l,k}(N_0 + P \sum_{q \in \cU_l \setminus k} X_{l,q} )} \cr
& = \frac{\sum_{l=1}^D X_{l,k} P}{
\frac{\sum_{l=1}^D (\sum_{q \in \cU_l \setminus k} X_{l,q}) X_{l,k}
}{\sum_{l=1}^D X_{l,k}}P
+N_0 }.
\label{EQ:iSINR}\end{aligned}$$
Let $T_k$ denote the SINR threshold for successful decoding. Then, the outage probability becomes \_k = (\_k < T\_k). \[EQ:pout\] Thus, we need to find the distribution of the instantaneous SINR, $\gamma_k$.
SINR Analysis
-------------
For convenience, let $M = |\cU_l \setminus k|$, i.e., the number of the interfering signals is denoted by $M$. If $M = 0$, under the assumption of [**A2**]{} or from , we can show that \_[l=1]{}\^D X\_[l,k]{} = , where $\chi_n^2$ represents a chi-squared random variable with $n$ degrees of freedom. For convenience, let $Z_n = \frac{\chi_{2n}^2}{2n}$. Then, it follows $$\begin{aligned}
\Pr(\gamma_k < T_k)
= \Pr \left( Z_{D} < \frac{T_k}{ D \sSNR} \right),
\label{EQ:OP_M0}\end{aligned}$$ where $\sSNR = \frac{P \sigma_h^2 }{N_0}$ is the signal-to-noise ratio (SNR). Thus, using the cumulative distribution function (cdf) of the chi-squared random variable, the outage probability can be found. Furthermore, as shown in Appendix \[A:2\], the following tight upper-bound on the outage probability can be obtained: $$\begin{aligned}
\Pr(\gamma_k < T_k)
\le \frac{1}{D!} \left( \frac{T_k}{\sSNR} \right)^D
e^{- \frac{c_D T_k}{\sSNR}}
\le\frac{1}{D!} \left( \frac{T_k}{\sSNR} \right)^D,
\label{EQ:M0}\end{aligned}$$ where c\_D = D e\^[-1]{} ( D! )\^[- ]{} 1. \[EQ:c\_D\]
Unfortunately, if $M > 0$, it is difficult to obtain a bound on the outage probability. Thus, we have to resort to an approximation. For convenience, let the interference term in be $$\begin{aligned}
P \frac{\sum_{l=1}^D (\sum_{q \in \cU_l \setminus k} X_{l,q}) X_{l,k}
}{\sum_{l=1}^D X_{l,k}}
= \frac{P \sigma_h^2}{2} \sum_{l=1}^D Y_l \alpha_l,
\label{EQ:A1}\end{aligned}$$ where $\alpha_l
= \frac{X_{l,k}}{\sum_{l=1}^D X_{l,k}} \ge 0$ with $\sum_{l=1}^D \alpha_l = 1$, and $Y_l = \frac{2}{\sigma_h^2}
\sum_{q \in \cU_l \setminus k} X_{l,q}$. Clearly, $Y_l$ is a chi-squared random variables with $2M$ degrees of freedom. Let $W = \sum_{l=1}^D \alpha_l Y_l$. If $\alpha_l = \frac{1}{D}$ for all $l$, we can see that $W$ becomes a scaled chi-squared random variable with $2DM$ degrees of freedom. However, since $\alpha_l$ is a random variable, the resulting approximation may have a lower variance than the actual one. Since $W$ can be seen as a weighted sum of chi-squared random variables, we may consider another chi-squared random variable to approximate $W$. To this end, let = , where $N$ is a parameter to be decided using a moment matching approach. It can be shown that $\uE[W] =
\uE\left[ \sum_{l=1}^D \alpha_l Y_l \right] = 2 M$, because $\uE[\chi_{2n}^2] = 2n$. In addition, $\uE[\Omega] = 2 M$. Thus, regardless of the value of $N$, we can see that $W$ and $\Omega$ have the same mean. We can consider the 2nd moment and find $N$ such that = . \[EQ:2nd\]
\[L:1\] The value of $N$ satisfying is given by N = . \[EQ:D1\]
See Appendix \[A:1\].
In Fig. \[Fig:ex\_cdf\], the empirical cdf of $W$ is shown with the cdf of $\Omega$ for two different pairs of $M$ and $D$. Clearly, thanks to the moment matching (up to the 2nd moment), the two cdfs become quite similar to each other.
![Empirical cdf of $W$ and an approximate cdf with $\Omega$: (a) $(D,M) = (8,2)$; (b) $(D,M) = (32,4)$.[]{data-label="Fig:ex_cdf"}](ex_cdf.pdf){width="\figwidth"}
By replacing $W$ with $\Omega$, the instantaneous SINR in can be approximated as $$\begin{aligned}
\gamma_k
\approx \frac{\sum_{l=1}^D X_{l,k} P}{
\frac{P \sigma_h^2}{2} \Omega + N_0}.
\label{EQ:sinrB}\end{aligned}$$ For a tractable analysis, the instantaneous SINR in can be used.
A Closed-form Expression for Outage Probability
-----------------------------------------------
In this subsection, we find a closed-form expression for the outage probability with the SINR in . Since $\Omega$ is a (scaled) chi-squared random variable with $2NM$ degrees of freedom, the outage probability in can have the following approximation: $$\begin{aligned}
\uP_k
& = \Pr(\gamma_k < T_k ) \cr
& \approx \tilde \uP_k
=
\Pr\left( Z_D < \frac{T_k}{D}
\left( \frac{\chi_{2 N M}^2}{2 N} + \frac{1}{\sSNR}
\right) \right).
\label{EQ:app}\end{aligned}$$ We can have a closed-form expression for $\tilde \uP_k$ as follows.
\[L:2\] For $M \ge 1$, suppose that d = - > 0, \[EQ:cond2\] where $T = T_k$ (for convenience, we omit the index $k$). Then, we have $$\begin{aligned}
\tilde \uP_k \le \psi (D, M, \sSNR, T) +
\left( \frac{d e}{M} \right)^{MN} e^{- N d},
\label{EQ:app2}\end{aligned}$$ where $$\begin{aligned}
& \psi(D,M, \sSNR, T) =
\frac{1}{D!} \left(\frac{T}{\sSNR} \right)^D
\frac{e^{- \frac{c_D T}{\sSNR}}}{(1+ \frac{c_D T}{N})^{M N}}\cr
& \quad \times
\sum_{n=0}^D \binom{D}{n}
\left(\frac{\sSNR}{N + c_D T} \right)^n
\prod_{t=0}^{n-1}(M N+t) .
\label{EQ:psi}\end{aligned}$$ Since the 2nd term on the right-hand side (RHS) in is negligible if $d$ is sufficiently large, the first term becomes a good approximation of $\tilde \uP_k$.
See Appendix \[A:2\].
For the outage probability, we will usually consider the first term on the RHS in (for a large $d$). Note that in , we can see that the diversity gain is $D$ as the outage probability is proportional to $\sSNR^{-D}$ when $M = 0$. For the case of $M \ge 1$, in order to see the diversity gain, we have the following result.
\[L:x\] Suppose that holds. Then, it can be shown that \^D, \[EQ:p\_nu\] where $C$ is a constant that is independent of $D$ and $\nu$ becomes smaller than 1 if ( )\^ 1 + . \[EQ:condx\]
See Appendix \[A:x\].
In , taking $\frac{1}{\nu}$ as a scaled SINR, we can see that the diversity gain[^4] becomes $D$. In general, we can derive design criteria for repetition-based NOMA to keep the outage probability low from (or to hold ). However, since the expression in is a bit complicated, it is not easy to obtain design criteria. Thus, for a more tractable analysis, we can consider the asymptotic $\psi$ when $\sSNR \to \infty$ as follows.
\[L:3\] If $M \ge 1$, we have $$\begin{aligned}
\bar \psi & = \lim_{\sSNR \to \infty} \psi (D, M, \sSNR, T)\cr
& = \binom{MN+D-1}{MN-1}
\left( \frac{N}{N + c_D T}\right)^{MN}
\left( \frac{T}{N + c_D T} \right)^D.
\label{EQ:L3}\end{aligned}$$
See Appendix \[A:3\].
From , when $D$ increases with a fixed $\rho = \frac{T}{N}$, since $\binom{n}{m} \le 2^{n \sH(m/n)}$, where $\sH(p) = -p \log_2p - (1-p) \log_2 (1-p)$, we can further show that $$\begin{aligned}
\bar \psi
& \le 2^{(MN+D-1) \sH \left(
\frac{MN-1}{MN +D -1}
\right)}
\left( \frac{\rho}{1+ c_D \rho} \right)^D
\left( \frac{1}{1+ c_D \rho} \right)^{MN} \cr
& \approx
2^{ D \left(1+ \frac{M}{2} \right) \sH \left(\frac{M}{M+2} \right)}
\left(
\frac{\rho}{ \left(1 + \rho\right)^{1 + \frac{M}{2}} }
\right)^D,\end{aligned}$$ where the approximation is tight if $D$ is sufficiently large with $N = \frac{D+1}{2}$ and $c_D \to 1$. From this, it can be further shown that $$\begin{aligned}
\frac{\log_2 \bar \psi }{D}
& =
\left(1 + \frac{M}{2} \right)
\left(\sH \left(\frac{M}{M+2} \right)
- \log_2 (1 + \rho)
\right) \cr
& \quad + \log_2 \rho.\end{aligned}$$ This implies that if ( ) < \_2 (1 + ) - \[EQ:cc\] holds, $\bar \psi$ decreases exponentially with $D$ (i.e., the diversity order is $D$). Since $\sH(p) \le 1$, a sufficient condition for can be found as follows: 2\^[1 + ]{} < = = . \[EQ:cc2\]
For convenience, let $D_{(b)}$ and $M_{(b)}$ denote the number of copies and the number of interfering signals for a user in layer $b$, respectively, and assume that all the users in a layer have the same number of copies and the same number of interfering signals. Then, according to , with a sufficiently high SNR, at layer $b$, a low outage probability is expected in repetition-based NOMA if M\_[(b)]{} 2 \_2 , \[EQ:MDT\] where $T_{(b)}$ is the threshold for the users in layer $b$. In addition, since $M_{(b)}$ decreases with $b$, the number of copies in layer $b$, $L_{(b)}$, needs to be larger than that in layer $b^\prime$ if $b < b^\prime$, as illustrated in Fig. \[Fig:frame\].
In order to determine key parameters, from , with a fixed $T_{(b)} = T$ for all layers, we can show that D\_[(b)]{} = 4 T 2\^[ ]{} 2\^[ ]{}, \[EQ:Db\] since $M_{(b)} = B - b$. From , we can see that the number of copies can decrease exponentially with $b$. In addition, the number of blocks, $L$, has to be proportional to $2^B$, which means that the number of layers, $B$, cannot be arbitrarily large with respect to a finite $L$. A small $B$ (e.g., $B \le 3$) is also important to keep SIC error propagation limited. Since $L = D_{(b)} K_{(b)}$, where $K_{(b)}$ represents the number of users in layer $b$, we also have $K_{(b)} \propto 2^{\frac{b}{2}}$.
Other Issues
------------
If $B = 1$ (i.e., orthogonal multiple access (OMA) is used), each user has the outage probability as in . However, if we consider $B > 1$ (i.e., NOMA) for a higher spectral efficiency, the performance of layer $b$ is affected by the performance of layers $1,\ldots, b-1$ through error propagation. To see the impact of imperfect SIC through error propagation on performance, consider the error probability with SIC propagation. Let $\epsilon_b$ denote the outage probability of the signal in layer $b$. Then, the error probability with SIC propagation at each layer, denoted by $\rho_b$, becomes $$\begin{aligned}
\rho_1 & = \epsilon_1 \cr
\rho_2 & = (1-\epsilon_1) \epsilon_2 + \epsilon_1 \le \epsilon_1
+ \epsilon_2 \cr
\rho_3 & = \epsilon_1 +
(1-\epsilon_1) \epsilon_2 +
(1-\epsilon_1) (1-\epsilon_2) \epsilon_3 \le
\epsilon_1 + \epsilon_2 + \epsilon_3 \cr
& \vdots\end{aligned}$$ Thus, the error probability with SIC propagation is bounded by $\sum_{b=1}^B \epsilon_b$ or $O(\max_b
\epsilon_b)$ if $B$ is sufficiently small (e.g., 3 or 4), which implies that the impact of error propagation on the performance may not be significant. To see further, suppose that $T_k$ is decided for a low outage probability, which is denoted by $\epsilon_{\rm out}$ (and $\epsilon_{\rm out} = \epsilon_b$ for all $b$), using the upper-bound on the outage probability in for given $D$, $M$, and $\sSNR$. For example, suppose that $L = 4$ and $B = 3$ as in Fig. \[Fig:frame\]. For user 1 in layer 1, with $D = 4$ and $M = 2$, $T_1$ can be obtained for a target outage probability, $\epsilon_{\rm out}$. For user 2 in layer 2, if $T_2$ is decided to keep a target outage probability of $\epsilon_{\rm out}$, the actual outage probability with taking into account error propagation becomes \_[out]{} (2) 2 \_[out]{}. Thus, as long as $\epsilon_{\rm out} \ll 1$ and $B$ is not too large, the actual outage probabilities of all the layers can be an order of $\epsilon_{\rm out}$ as mentioned earlier. As a result, it can be seen that the proposed repetition-based NOMA transmission can not only guarantee a low error probability (without instantaneous CSI-based resource allocation), but also provide a high spectral efficiency.
If capacity-achieving codes [@MacKayBook] are employed, the information outage probability [@TseBook05] can be given by (\_2 (1+ \_k) < R\_k) = (\_k < 2\^[R\_k]{} - 1), where $R_k$ is the code rate of user $k$’s packet. Thus, $T_k = 2^{R_k} - 1$. That is, if $T_k$ is obtained to keep a specific target outage probability from the closed-form expression in , the corresponding code rate, $R_k$, can be easily decided. Therefore, repetition-based NOMA can guarantee a specific target error probability (which is usually low enough to avoid frequent re-transmissions) without using instantaneous CSI-based power allocation.
Error Probability with Finite-Length Codes {#S:EP}
==========================================
In this section, we consider the case that finite-length codes are used in repetition-based NOMA.
Suppose that the BS is to decode the signal from user $k$ using $\by_k$ in . For a given $\gamma = \gamma_k$, according to [@Polyanskiy10IT] [@Tan15], the achievable rate (for complex Gaussian channel [@Durisi16]) for a finite-length code is given by R\^\* (n, ) \_2 (1 + ) - \^[-1]{} () + O( ), \[EQ:R\_PPV\] where $V(\gamma)$ is the channel dispersion that is given by $V(\gamma)
= \frac{\gamma(2 + \gamma)}{(1+ \gamma)^2} (\log_2 e)^2$, $n$ is the length of codeword when a codeword is transmitted within a block, and $\epsilon$ is the error probability. It can be shown that $\bar V > V(\gamma)$, where $\bar V = \frac{1}{(\ln 2)^2} \approx 2.0814$. Thus, ignoring the term of $O\left(\frac{\log_2 n}{n}\right)$ and letting $R = R^*(n,\epsilon)$, a lower-bound on the achievable rate can be obtained as follows: R \_2 (1 + ) - \^[-1]{} (), \[EQ:DB\] which might be tight for a sufficiently high SNR, $\gamma$, because $V(\gamma) \to \bar V$ as $\gamma \to \infty$. Then, an upper-bound on the average error probability is given by $$\begin{aligned}
\bar \epsilon
& \approx \uE\left[ \cQ \left(
\sqrt{\frac{n}{V (\gamma)}} \left(
\log_2 (1+ \gamma)- R \right) \right) \right] \cr
& \le \uE\left[ \cQ \left(
\sqrt{\frac{n}{\bar V}} \left(
\log_2 (1+ \gamma)- R \right) \right) \right] \cr
& = \frac{1}{\sqrt{2 \pi}} \int_{-\infty}^\infty
\Pr \left(\gamma < 2^{ \sqrt\frac{\bar V}{n} x + R} - 1
\right) e^{- \frac{x^2}{2}} dx,
\label{EQ:beps}\end{aligned}$$ where the expectation is carried out over $\gamma$ and the first equality is due to [@VerduBook Eq. (3.57)]. Note that if $n \to \infty$, $ \sqrt\frac{\bar V}{n} \to 0$. Thus, we can have $$\begin{aligned}
\lim_{n \to \infty}
\int_{-\infty}^\infty
\Pr (\gamma < 2^{ \sqrt\frac{\bar V}{n} x + R} - 1
)
\frac{e^{- \frac{x^2}{2}} }{\sqrt{2 \pi}}
dx
=
\Pr (\gamma < 2^{R} - 1),\end{aligned}$$ which means that the average error probability becomes the outage probability.
From , using a closed-form expression for the outage probability in and a numerical integration technique, we can find an upper-bound on $\bar \epsilon$.
Simulation Results {#S:Sim}
==================
In this section, we present simulation results that can provide design criteria. For simulations, we mainly consider the instantaneous SINR in with randomly generated channel coefficients according to Rayleigh fading channels in .
Outage Probability
------------------
In this subsection, we present simulation results of the outage probability.
Fig. \[Fig:plt1\] shows the outage probability as a function of SNR with $D = 16$, $T = 2$, and $M \in \{1,2\}$. For the upper-bound, we use the first term on the RHS in (in general, the 2nd term is negligible). Due to the presence of the interfering signals (as $M \ge 1$), we can see that there is an error floor in the outage probability. That is, although the SNR goes to $\infty$, the outage probability does not approach 0, but a non-zero constant as shown in . We can confirm that is a tight upper-bound from Fig. \[Fig:plt1\] (a) and (b).
![Outage probabilities as functions of SNRs with $D = 16$ and $T = 2$: (a) $M = 2$; (b) $M = 1$.[]{data-label="Fig:plt1"}](plt1.pdf "fig:"){width="\figwidth"}\
0.5cm (a) 3.5cm (b)
The impact of $M$ and $T$ on the outage probability is shown in Fig. \[Fig:plt24\] when $\sSNR = 6$ dB. Since $M$ is the number of interfering signals, the outage probability increases with $M$ as shown in Fig. \[Fig:plt24\] (a). In Fig. \[Fig:plt24\] (b), as expected, the outage probability increases with $T$. Furthermore, since the bound is tight, we can choose $T$ for a sufficiently low target outage probability. Note that the first term on the RHS in in Fig. \[Fig:plt24\] (b) is not an upper-bound when $T$ is large (e.g., $T \ge 8$). When $T = 10$, the 2nd term on the RHS in becomes 0.6727, which is not negligible. As shown in Fig. \[Fig:plt24\] (b), the 2nd term needs to be taken into account for the upper-bound when $T$ is not small. However, since we are mainly interested in a low outage probability (e.g., $\le 10^{-3}$), the impact of the 2nd term on the upper-bound is negligible.
![Outage probabilities as functions of $M$ and $T$ when $\sSNR = 6$ dB: (a) outage probability versus $M$ with $D = 32$ and $T = 4$; (b) outage probability versus $T$ with $D = 16$ and $M = 2$.[]{data-label="Fig:plt24"}](plt24.pdf "fig:"){width="\figwidth"}\
0.5cm (a) 3.5cm (b)
Fig. \[Fig:plt3\] shows the outage probability as a function of $D$ when $M = 3$, $\sSNR = 6$ dB, and $T \in \{2, 4\}$. Clearly, a better performance is achieved with $D$ due to a higher diversity gain. It is shown that as long as $d$ is sufficiently large (due to a large $D$ or small $T$), the bound with the first term in is reasonably tight and can be used to predict the performance in terms of the outage probability.
![Outage probabilities as a function of $D$ with $M = 3$ and $\sSNR = 6$ dB: (a) $T = 4$; (b) $T = 2$.[]{data-label="Fig:plt3"}](plt3.pdf "fig:"){width="\figwidth"}\
0.5cm (a) 3.5cm (b)
Average Error Probability for Finite-Length Codes
-------------------------------------------------
As mentioned earlier, a careful determination of $T_k$ or $R_k$ is necessary to keep the outage probability low (for successful SIC). If finite-length codes are used, we need to consider the average error probability instead of the outage probability. In this subsection, we consider the average error probability in .
In Fig. \[Fig:plt\_RvsER\], we consider a repetition-based NOMA system with $B = 3$ and $n = 512$ (bits). It is assumed that one user in layer 1 (transmitting $L$ copies), two users in layer 2 (each user transmitting $L/2$ copies), and four users in layer 3 (each user transmitting $L/4$ copies). The average error probability in each layer for different values of the code rate, $R$, is shown in Fig. \[Fig:plt\_RvsER\] with the bound from . As shown in Fig. \[Fig:plt\_RvsER\] (a), the signal in layer 1 needs to have $R \le 0.75$ for an average error probability of $10^{-3}$. It is also possible to decide the code rates for the signals in layers 2 and 3 for an average error probability of $10^{-3}$ using the bound from , because the bound is sufficiently tight. In Fig. \[Fig:plt\_RvsER\] (b), we can see that the rates increase more than twice when $L = 16$ compared with the rates when $L = 8$ (which are shown in Fig. \[Fig:plt\_RvsER\] (a)) with a target error probability of $10^{-3}$. If the target error probability further decreases, the rate gap increases. This indicates that the repetition-based NOMA scheme can be more efficient with a large $L$ and a low target error probability thanks to a high diversity gain.
![Average error probabilities as a function of $R$ with $B = 3$ and $\sSNR = 6$ dB: (a) $L = 8$; (b) $L = 16$.[]{data-label="Fig:plt_RvsER"}](plt_RvsER.pdf "fig:"){width="\figwidth"}\
0.5cm (a) 3.5cm (b)
In Figs. \[Fig:plt1\] - \[Fig:plt\_RvsER\], we consider the instantaneous SINR in for simulations. Since is obtained under the assumption of [**A1**]{}, it would be necessary to consider simulations with interleaved finite-length blocks. To this end, quadrature phase shift keying (QPSK) is considered with a block length of $n/2$ (since one QPSK symbol can transmit 2 bits). Random interleaving at symbol-level is considered. In Fig. \[Fig:Eplt2\], we show the average error probability for different values of the code rate when there are $M \in \{2,3\}$ interfering signals, $D = 16$, and $\sSNR = 6$ dB. It is shown that the average error probability with QPSK is slightly lower than that with the instantaneous SINR in . This may result from the fact that the correlation cannot be zero by symbol-level random interleaving and the correlation reduces the interference level.
![Average error probabilities as a function of $R$ with $D = 16$, $n = 512$, $\sSNR = 6$ dB, and $M \in \{2,3\}$.[]{data-label="Fig:Eplt2"}](Eplt2.pdf){width="\figwidth"}
To see the impact of the length of finite-length codes, $n$, on the average error probability, we perform simulations and show the results in Fig. \[Fig:Eplt3\] when with $D = 16$, $M = 2$, $R = 2$, and $\sSNR = 6$ dB. As expected, the average error probability decreases with $n$, while it becomes saturated for a sufficiently large $n$. We can also confirm that the bound in can be used to predict the performance with finite-length codes from Figs. \[Fig:Eplt2\] and \[Fig:Eplt3\].
![Outage probabilities as a function of $n$ with $D = 16$, $M = 2$, $R = 2$, and $\sSNR = 6$ dB.[]{data-label="Fig:Eplt3"}](Eplt3.pdf){width="\figwidth"}
Concluding Remarks {#S:Conc}
==================
In this paper, we discussed a NOMA scheme based on repetition to exploit high diversity gains. The resulting scheme, called repetition-based NOMA, was able to provide a low error probability without instantaneous CSI-based power allocation thanks to high diversity gains. In order to guarantee a target performance, a closed-form expression for an upper-bound on the outage probability was derived so that key parameters (e.g., the code rate) can be decided accordingly. The case of finite-length codes was also considered with the average error probability. Simulation results demonstrated that the derived upper-bound is reasonably tight and can be used to decide key parameters that meet a certain target performance.
Since we mainly focused on the performance analysis to derive a closed-form expression for the outage probability in terms of key parameters, we did not work on other issues, e.g., scheduler design. The design of scheduler would be an interesting topic to be studied in the future, which might be based on the derived closed-form expression for the outage probability in this paper.
Proof of Lemma \[L:1\] {#A:1}
======================
Since $\uE[\chi_{2n}^2] = 2n$ and ${\rm Var}(\chi_{2n}^2) = 4n$, it can be shown that $$\begin{aligned}
\uE[\Omega^2] = \frac{1}{N^2} \left( 4 M^2 N^2 + 4N M \right)
= 4 M \left( M+ \frac{1}{N}\right).
\label{EQ:B2}\end{aligned}$$
The 2nd moment of $W$ is given by $$\begin{aligned}
\uE[W^2] & = \uE \left[ \sum_{l=1}^D \sum_{l^\prime = 1}^D
\alpha_l \alpha_{l^\prime} Y_l Y_{l^\prime} \right]\cr
& = \sum_{l=1}^D \uE[\alpha_l^2]
\uE[Y_l^2] + \sum_{l \ne l^\prime} \uE[\alpha_l\alpha_{l^\prime}]
\uE[Y_l] \uE[Y_{l^\prime}] \cr
& = D \alpha_{(2)} (4 M + 4M^2)
+ 4 M^2 D (D-1) \sigma_{1,2},
\label{EQ:W2o}\end{aligned}$$ where $\alpha_{(2)} = \uE[\alpha_l^2]$ and $\sigma_{1,2} = \uE[\alpha_1 \alpha_2]$. Since $X_{l,k}$ is an exponential random variable under the assumption of [**A2**]{}, $\alpha_l$ is expressed as $\alpha_l = \frac{D_l}{\sum_{l^\prime=1}^D D_{l^\prime}}$, where $D_l \sim {\rm Exp}(1)$ is an independent exponential random variable with parameter 1. The distribution of $\alpha_l$ is the same as that of the minimum of $D-1$ independent standard uniform random variables [@Ahsanullah13 Example 4.6], i.e., $f(\alpha_l) = (D-1) (1-\alpha_l)^{D-2}$, $\alpha_l \in [0,1]$. Thus, we have \_[(2)]{} = . \[EQ:a2D\] In addition, the distribution of $\alpha_1 + \alpha_2 = \frac{D_1 + D_2}{\sum_{l=1}^D D_l}$ is the same as the distribution of the 2nd smallest order statistic among $D-1$ independent standard uniform random variables. From this, since $$\uE[(\alpha_1 + \alpha_2)^2] = \uE[\alpha_1^2]
+ \uE[\alpha_2^2] + 2 \uE[\alpha_1 \alpha_2],$$ we have $$\begin{aligned}
\uE[\alpha_1 \alpha_2]
& = \frac{\uE[(\alpha_1 + \alpha_2)^2]}{2} - \uE[\alpha_1^2] \cr
& = \frac{6}{2 D (D+1)} - \frac{2}{D (D+1)}
= \frac{1}{D (D+1)},\end{aligned}$$ because the 2nd moment of the 2nd smallest order statistic is $\frac{6}{D(D+1)}$ [@Ahsanullah13 Eq. (8.4)]. Then, it can be shown that = (4 M + 4M\^2) + 4 M\^2. \[EQ:W2\] Consequently, from and we can find that the 2nd moments of $W$ and $\Omega$ are the same if $N$ is given as in , which completes the proof.
Proof of Lemma \[L:2\] {#A:2}
======================
Using the Chernoff bound [@Mitz05], it can be shown that $$\begin{aligned}
\Pr(Z_D < z)
& \le \uE[ e^{-t (Z_D - z)}] \cr
& = e^{2 D tz} \left( \frac{1}{1+ 2 t} \right)^D
= \left( \frac{e^{2t z}}{1+ 2 t} \right)^D.\end{aligned}$$ Here, $t > 0$. Letting $z = \frac{1}{1 + 2t}$, we have (Z\_D < z) (z e\^[1-z]{} )\^D, z \[0, 1), \[EQ:CB\_l\] which is reasonably tight. For $z > 1$, it can also be shown that (Z\_D > z) (z e\^[1-z]{} )\^D. \[EQ:CB\_r\]
As in [@Choi20_WCNC], the upper-bound in can be tighter using a correction term as follows: $$\begin{aligned}
\Pr(Z_D \le z) & \le F_D (z) \cr
& = (z c_D e^{1-z c_D } )^D, \ z \in [0, 1/c_D],
\label{EQ:UB}\end{aligned}$$ where $c_D$ is the correction term[^5] that is given in . Thus, if $M = 0$, can be applied to , which results in .
For the case that $M > 0$, we need to take into account the interference. In , let $Y = \frac{\chi_{2 N M}^2}{2}$. Then, using , it can be shown that $$\begin{aligned}
\tilde \uP_k
& =
\int_0^\infty \Pr\left(Z_D < \frac{T}{D}
\left(\frac{y}{N} + \frac{1}{\sSNR} \right)
\right) f_Y (y) d y \cr
& \le \int_0^\kappa
( c_D \phi (y) e^{1- c_D \phi (y)} )^D f_Y (y) d y
+ \int_\kappa^\infty
f_Y (y) d y \cr
& \le \int_0^\infty
( c_D \phi (y) e^{1- c_D \phi (y)} )^D f_Y (y) d y
+ \int_\kappa^\infty
f_Y (y) d y , \quad
\label{EQ:UBk}\end{aligned}$$ where $f_Y (y)$ represents the pdf of $Y$ and $$\begin{aligned}
\phi (y) = \frac{T}{D}
\left( \frac{y}{N} + \frac{1}{\sSNR} \right) \ \mbox{and} \
\kappa = N \left(\frac{D}{c_D T} - \frac{1}{\sSNR} \right).\end{aligned}$$ The first term on the RHS in can be expressed as $$\begin{aligned}
\psi & =
\left( \frac{c_D e T}{D} \right)^D
e^{- \frac{c_D T}{\sSNR}} \cr
& \ \times \int_0^\infty
\left(\frac{y}{N} + \frac{1}{\sSNR}
\right)^D e^{- \frac{c_D T}{N} y}
f_Y( y) dy \cr
& =
\frac{1}{D!} \left(\frac{T}{\sSNR} \right)^D
e^{- \frac{c_D T}{\sSNR}} \cr
&\ \times
\sum_{n = 0}^D \binom{D}{n}
\left(\frac{\sSNR}{N}\right)^n
\int_0^\infty
e^{-\frac{c_D T}{N} y}
\frac{ y^{M N-1+n} e^{-y}}{(M N-1)!} dy \cr
& =
\frac{1}{D!} \left(\frac{T}{\sSNR} \right)^D
e^{- \frac{c_D T}{\sSNR}} \cr
&\ \times
\sum_{n = 0}^D \binom{D}{n} \left( \frac{\sSNR}{N} \right)^{n}
\left(\frac{1}{1+ \frac{c_D T}{N}
} \right)^{M N+n}
\frac{(M N +n-1)!}{(M N-1)!} \cr
& =
\frac{1}{D!} \left(\frac{T}{\sSNR} \right)^D
\frac{e^{- \frac{c_D T}{\sSNR}}}
{(1 + \frac{c_D T}{N})^{M N}}
\cr
&\ \times
\sum_{n = 0}^D \binom{D}{n} \left( \frac{\sSNR}{N
+ c_D T} \right)^{n}
\frac{(M N +n-1)!}{(M N-1)!},
\label{EQ:psi0}\end{aligned}$$ which becomes .
From , the 2nd term on the RHS in is bounded as follows: $$\begin{aligned}
\Pr(Y \ge \kappa)
& = \Pr\left(
\frac{\chi_{2 N M}^2}{2 N M}
\ge \frac{1}{M} \left( \frac{D}{c_D T} - \frac{1}{\sSNR} \right)
\right) \cr
& \le
\left( \frac{d}{M} e^{1 - \frac{d}{M} }
\right)^{NM}
= \left(\frac{d e}{M} \right)^{NM} e^{-d N},\end{aligned}$$ which is the 2nd term on the RHS in . This completes the proof.
Proof of Lemma \[L:x\] {#A:x}
======================
Using the inequality of arithmetic and geometric means, it can be shown that $$\begin{aligned}
\prod_{t=0}^{n-1}(M N+t)
\le
\left(MN + \frac{\sum_{t=0}^{n-1}t }{n} \right)^n
=
\left(MN + \frac{n-1}{2} \right)^n.
\label{EQ:AGineq}\end{aligned}$$ In , since $n \le D$, using , it follows $$\begin{aligned}
& \sum_{n=0}^D \binom{D}{n}
\left(\frac{\sSNR}{N + c_D T} \right)^n
\prod_{t=0}^{n-1}(M N+t) \cr
& \le
\left( 1 +
\frac{\sSNR}{N + c_D T} \left(MN + \frac{D-1}{2} \right) \right)^D.
\label{EQ:bb1}\end{aligned}$$ Substituting into , with $N = \frac{D+1}{2}$, we have $$\begin{aligned}
\psi &
\le
\frac{e^{- \frac{c_D T}{\sSNR}}}{D!}
\left(\frac{T}{\sSNR} \right)^D
\left(1+ \frac{c_D T}{N}\right)^{-M N}\cr
& \ \times
\left( 1 +
\frac{\sSNR}{N + c_D T} \left(MN + \frac{D-1}{2} \right) \right)^D \cr
& \le \frac{(1+ \frac{c_D T}{N})^{-\frac{M}{2}}}{D!} \nu^D,
\label{EQ:uu1}\end{aligned}$$ where $
\nu =
\frac{T}{\sSNR} \left(1 + \frac{2 c_D T}{D+1} \right)^{-\frac{M}{2}}
\left( 1 +
\frac{\sSNR(D(M+1) + M-1) }{D+1 + 2 c_D T}
\right)$. In , the last inequality is due to the fact that $e^{-\frac{c_D T}{\sSNR}} \le 1$. Then, we can see that $\nu < 1$ if holds.
From , we have c\_D T = , 1. \[EQ:cDT\] To determine $C$ in , from , it can be shown that $$\begin{aligned}
\frac{1}{1 + \frac{c_D T}{N}}
& = \frac{1}{1 + \frac{2 c_D T}{D+1}}
= \frac{1}{1 + \frac{2 D \sSNR}{(D+1) (d \sSNR +1)}} \cr
& \le \frac{1}{1 + \frac{\sSNR}{d \sSNR +1}}
= \frac{1 + d \sSNR}{1+(1+d) \sSNR},\end{aligned}$$ we have $C = \left(\frac{1}{1 + \frac{\sSNR}{d \sSNR +1}} \right)^{\frac{M}{2}}
\ge \left( \frac{1}{1+ \frac{c_D T}{N}}\right)^{\frac{M}{2}}$, where $C$ is independent of $D$.
Proof of Lemma \[L:3\] {#A:3}
======================
As $\sSNR \to \infty$, from , after some manipulations, we can show that $$\begin{aligned}
\bar \psi & =
\left( \frac{c_D e T}{D} \right)^D
\int_0^\infty
\left(\frac{y}{N} \right)^D e^{- \frac{c_D T}{N} y}
f_Y( y) dy \cr
& =
\left( \frac{c_D e T}{D N} \right)^D
\frac{\Gamma (MN + D)}{\Gamma (MN)}
\frac{1}{\left( 1 + \frac{c_D T}{N} \right)^{MN + D}}\cr
& =
\frac{1}{D!}
\frac{\Gamma (MN + D)}{\Gamma (MN)}
\left(\frac{T}{N+c_D T} \right)^D
\frac{1}{\left( 1 + \frac{c_D T}{N} \right)^{MN}}\cr
& = \binom{MN+D-1}{MN-1}
\left(\frac{T}{N+c_D T} \right)^D
\left(\frac{N}{ N + c_D T}\right)^{MN},\end{aligned}$$ which becomes .
[^1]: The author is with the School of Information Technology, Deakin University, Geelong, VIC 3220, Australia (e-mail: [email protected]). This research was supported by the Australian Government through the Australian Research Council’s Discovery Projects funding scheme (DP200100391).
[^2]: In conventional power-domain NOMA, each layer is characterized by the transmit power and seen as a logical division of a radio resource block for superposition coding. In the proposed NOMA scheme, each layer is also a logical division that is characterized by the number of copies.
[^3]: It will be shown later that the diversity gain can be equal to the number of copies in the presence of interference under certain conditions.
[^4]: The diversity order is the negative SNR exponent of the outage probability in a high SNR regime [@TseBook05]. In this case, the SNR is replaced with SINR.
[^5]: with the correction term in is an inequality conjecture.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We propose a statistical mechanical derivation of Kähler-Einstein metrics, i.e. solutions to Einstein’s vacuum field equations in Euclidean signature (with a cosmological constant) on a compact Kähler manifold $X.$ The microscopic theory is given by a canonical free fermion gas on $X$ whose one-particle states are pluricanonical holomorphic sections on $X$ (coinciding with higher spin states in the case of a Riemann surface) defined in background free manner. A heuristic, but hopefully physically illuminating, argument for the convergence in the thermodynamical (large $N)$ limit is given, based on a recent mathematically rigorous result about exponentially small fluctuations of Slater determinants. Relations to higher-dimensional effective bosonization, the Yau-Tian-Donaldson program in Kähler geometry and quantum gravity are explored. The precise mathematical details will be investigated elsewhere.'
address: Chalmers University of Technology and University of Gothenburg
author:
- 'Robert J. Berman'
title: 'Kähler-Einstein metrics emerging from free fermions and statistical mechanics '
---
Introduction
============
The basic laws of gravity have an intriguing similarity with the laws of thermodynamics and hydrodynamics - this has been pointed out at several occasions in the physics literature, in particular in connection to the study of black holes (see for example [@bch; @ja; @v]). As a consequence one is lead to ask whether gravity can be seen as an emergent effect of an underlying microscopic theory in a thermodynamical limit [@ja]? The aim of this note is to propose a situation where this question can be answered in affirmatively. We will consider Einstein’s vacuum field equations in Euclidean signature on a compact manifold $X$, whose solutions are usually called *Einstein metrics* in the mathematics literature [@an]. More precisely, these equations will be considered in the presence of a fixed background integrable complex structure $J$ on $X.$ It turns out that the underlying microscopic theory may then be realized as a certain free fermion gas on $X$ (whose definition only involves $J$ but no metric data) and it will be shown how to recover an Einstein metric (with a non-zero cosmological constant) in the thermodynamical limit. The metric is singled out by the fact that it is Hermitian and Kähler with respect to $J$. In other words these are the *Kähler-Einstein metrics* which have been extensively studied during the last decade in the mathematics literature (for a recent survey see [@p-s]).
Physically, metrics as above appear, for example, as gravitational instantons in Hawking’s functional integral approach to quantum gravity [@ha; @ts]. Although we will not restrict $X$ to be a real four-manifold - the physically most relevant case - it is worth pointing out that in this latter case “most” Einstein metrics are Kähler-Einstein metrics. In particular, for a negative cosmological constant $\Lambda$ it may actually be that *all* Einstein metrics are Kähler with respect to *some* complex structure $J,$ as long as $X$ complex structures (the question was raised in [@le2]). This is for example the case for compact quotients of the unit ball, as shown in [@le] using Seiberg-Witten gauge theory. By making the complex structure $J$ dynamical, inspired by the works of Fujiki [@fu] and Donaldson [@do3] in Kähler geometry, we will also explain how the microscopic theory referred to above can be used to give a finite $N$ well-defined approximation to a variant of quantum gravity (related to Liouville gravity when $n=1$) where the role of the diffeomorphism group in ordinary gravity is played by a symplectomorphism group which arises as the gauge group for a bundle over the moduli space of all complex structures on the smooth manifold $X.$
To be a bit more precise, given a complex structure $J$ on $X$ we will obtain a canonical (metric free) probability measure $\mu^{(N)},$ expressed in terms of a Slater determinant, on the $N-$ product $X^{N}$ such that in the large $N-$limit the measure $\mu^{(N)}$ becomes exponentially concentrated on configurations of points approximating the normalized volume form $\mu_{KE}$ of the Kähler-Einstein metric of $(X,J),$ i.e. such that $$\frac{1}{N}\sum_{i=1}^{N}\delta_{x_{i}}\approx\mu_{KE}$$ Since the corresponding Kähler-Einstein metric on $(X,J)$ is uniquely determined by its volume form $\mu_{KE}$ this means that the Kähler-Einstein metric indeed emerges macroscopically. Finally, by working over the moduli space of all complex structures on $X$ the dependence on $J$ is also taken into account.
The main ingredients in the investigation of the thermodynamical limit is the asymptotics of exponentially small fluctuations of Slater determinants for $N-$particle correlations of fermions on complex manifolds in [@berm3] (building on [@b-b; @b-b-w; @bbgz]). On one hand, from a purely mathematical point of view these large $N$ asymptotics concern large deviations for certain critical determinantal random point processes, which generalize Random Matrix ensembles previously extensively studied. On the other hand, from a physical point of view the result can be seen as an effective bosonization of a free fermion gas (see section \[sub:Derivation-of-\]), which in the case of a Riemann surface alternatively can be deduced from the exact bosonization results in [@vv; @b-v-]. The large deviation result for the Slater determinant is then combined with a basic large deviation result for a non-interacting *classical* gas going back to Boltzmann’s fundamental work on entropy (called Sanov’s theorem in the mathematics literature).
It should however be pointed out that the argument in the present note which combines the two mathematically rigorous results referred to above is not completely rigorous. Basically, it involves an interchange of two limits which needs to be mathematically justified. The mathematical details, as well as various extensions, will be investigated elsewhere, but hopefully the heuristic derivation given here is illuminating from a physical point of view as it involves manipulations that are standard in the functional integral approach to quantum field theory.
Incidentally, in the case of a Riemann surface (i.e. the case when the *real* dimension $D$ of $X$ is two) the situation studied in the present note is closely related to the previous mathematical study of various $2D$ ensembles (point vortex systems, plasmas, self-gravitating systems, ...) from the point of view of mean field theory; see [@clmp; @k] and references therein. In particular, the corresponding thermodynamical limit was studied in [@clmp; @k] as a model of 2D turbulence. However, the higher dimensional situation studied in the present work is analytically considerably more complicated as the resulting limiting mean field equations are fully non-linear (see section \[sub:The-minimizer-\]). The reason is that the role of the Laplace operator on a Riemann surface is played by the non-linear *Monge-Ampère operator* for higher dimensional complex manifolds. A different “linear” higher-dimensional generalization of point vortex systems has previously been consider by Kiessling [@ki2], where the role of the Laplace operator is played by the linear Paneitz operator. It involves *conformal geometry* of spheres rather than the *complex (holomorphic) geometry* considered here and the thermodynamical limit is a mean field limit of an explicit gas with logarithmic pair interactions.
It would be interesting to understand the relation between the present note and the ADS/CFT correspondence [@agsmoo], which relates gravity in the bulk of a manifold to a conformal field theory on its boundary. This is a realization of t’Hooft’s holographic principle. Such a principle has recently been put forward by Verlinde in [@v] as the basis of an entropic explanation of gravity. As explained in the concluding section \[sub:Conclusion-and-discussion\] the emergence of the Kähler-Einstein metric from the present microscopic model can be interpreted as coming from a *fermionic maximum entropy principle.*
From the mathematical point of view an important motivation for the present work comes from the Yau-Tian-Donaldson program which relates the analytic problem of existence of extremal metrics in a given Kähler class (i.e. Kähler-Einstein metrics in the case of the canonical class) to algebro-geometric stability conditions (notably $K-$stability; see [@do1; @don1; @t; @p-s] and references therein). For example, the free energy functional derived below turns out to coincide, in the canonical case, with Mabuchi’s $K-$energy, which is usually used to define various notions of $K-$stability. Moreover, the thermodynamical convergence towards a Kähler-Einstein volume form in section \[sub:Convergence-in-the thermo\] is somewhat “dual” to the convergence of canonically balanced metrics conjectured by Donaldson in [@don2] and proved in [@bbgz] (see section \[sub:Duality-and-relation\]).
As a conclusion one of the mathematical aims of the present paper is to introduce a “thermodynamical formalism” for Kähler-Einstein metrics and more generally for Monge-Ampère equations of mean field type. A rigorous mathematical account of the corresponding variational calculus is given in [@berm6].
Acknowledgment {#acknowledgment .unnumbered}
--------------
Thanks to Kurt Johansson for stimulating my interest in general $\beta-$ensembles and to Bengt Nilsson for comments on a draft of the present paper.
Geometric setup
---------------
Let $X$ be a compact Kähler manifold with $\dim_{\C}X=n.$ In other words, we are given a real manifold $(X,J)$ of dimension $D=2n$ equipped with an integrable complex structure $J$ and admitting an Hermitian metric $$\omega=\frac{i}{2}h_{ij}dz^{i}\wedge d\bar{z}^{j}$$ on the complex tangent bundle $TX,$ which is closed: $d\omega=0$ (so that $\omega$ is a symplectic form) where $\omega$ is called the Kähler form (metric). Identifying $\omega$ with a Riemannian metric $g$ compatible with $J,$ i.e. $g=\omega(\cdot,J\cdot)$ (or locally $g=\mbox{Re \ensuremath{h)}}$ the vacuum Einstein equations, in Euclidean signature, with a cosmological constant read: $$\mbox{Ric \ensuremath{\omega=\Lambda\omega}}\label{eq:einsteins eq}$$ when $n>1$ and for general $n$ this is the equation for a Kähler-Einstein metric. After a scaling, we may assume that the cosmological constant $\Lambda$ is $0,1$ or $-1.$ In the following we will be mainly concerned with the latter case, i.e. when the solution $\omega$ is a Kähler metric with constant *negative* Ricci curvature. As shown in the seminal works of Aubin [@au] and Yau [@y] such a metric $\omega$ exists precisely when the first Chern class $c_{1}(K_{X})$ of the canonical line bundle $K_{X}:=\Lambda^{n}(T^{*}X)$ is *positive,* which will henceforth be assumed. The Kähler-Einstein metric $\omega$ is then uniquely determined by the complex structure $J$ and we will denote it by $\omega_{KE}.$ When $n=1,$ i.e. $X$ is a Riemann surface, this amounts to the classical fact that $X$ admits a metric of constant negative curvature precisely when $X$ has genus at least two. This hyperbolic metric is unique in its conformal class (determined by the complex structure $J)$
The starting point of the existence proof of Aubin and Yau is the basic complex geometric fact that the metric $\omega_{KE}$ is uniquely determined by its volume form $\omega_{KE}^{n}/n!,$ that we will normalize to become a probability measure: $$\mu_{KE}:=\frac{\omega_{KE}^{n}/n!}{Vn!}$$ In other words the tensor equation \[eq:einsteins eq\] reduces to a *scalar* equation (for the density of $\mu_{KE})$ and the Kähler-Einstein metric $\omega_{KE}$ may then be recovered by $$\omega_{KE}=\frac{i}{2\pi}\partial\bar{\partial}\log\mu_{KE}$$ i.e. as $\frac{i}{2\pi}$ times the curvature two form of the metric on the canonical line bundle $K_{X}$ defined by $\mu_{KE}.$
The question raised in the introduction may now be reformulated as *“Can the probability measure $\mu_{KE}$ be realized as the (macroscopic) expected distribution of particles in a thermodynamical limit of a (microscopic) statistical mechanical system canonically associated to $X?$* Moreover, the point is to be able to define the microscopic system without specifying any background metric structure so that the Einstein metric and hence (Euclidean) gravity would emerge macroscopically. It turns out that such a statistical mechanical system can indeed be realized by a certain free fermion gas on $X,$ as explained below.
General statistical mechanics formalism
---------------------------------------
We start by recalling some basic statistical mechanical formalism. Mathematically a (classical) gas of $N$ identical particles (i.e. a *random point process with $N$ particles*) is described by a *symmetric* probability measure $\mu^{(N)}$ on the $N-$fold product $X^{N}$ *(the $N-$particle configuration space).* In local holomorphic coordinates $Z=(z_{1},....,z_{n})$ on the complex manifold $X$ this means that $$\mu^{(N)}=\rho^{(N)}(Z_{1},...,Z_{N})dV(Z_{1})\wedge\cdots\wedge dV(Z_{N})$$ where $dV(Z_{1}):=(\frac{i}{2})^{n}dz_{1}\wedge d\bar{z}_{1}\wedge\cdots\wedge dz_{n}\wedge d\bar{z}_{n}$ and where the local *$N-$point correlation function* $\rho^{(N)}$ is invariant under permutations of the $Z_{i}:$s (note that the fact that we have not written out any dependence on $\bar{Z}_{i}$ does not indicate that the objects are holomorphic!). Pushing forward $\mu^{(N)}$ to $X^{j}$ one then obtains the corresponding $j-$point correlation measures $\mu_{j}^{(N)}$ on $X^{j}$ and their local densities $\rho_{j}^{(N)}.$ We will be mainly concerned with the *one-point correlation measure* $\mu_{1}^{(N)}$ on $X,$ i.e. $$\mu_{1}^{(N)}:=\int_{X^{N-1}}\mu^{(N)}$$ In other words, its local density $\rho_{1}^{(N)}(Z)$ represents the probability of finding a particle in the infinitesimal box $dV(Z_{1}).$ Yet another (trivially) equivalent formulation representation of $\mu_{1}^{(N)}$ can be given: $$\mu_{1}^{(N)}=\left\langle \frac{1}{N}\sum_{i}\delta_{x_{i}}\right\rangle ,$$ where the brackets denote the ensemble mean (expectation) of the random variable $$(x_{1},...,x_{N})\mapsto\frac{1}{N}\sum_{i}\delta_{x_{i}}\label{eq:empirical measure}$$ with values in the space $\mathcal{M}_{1}(X)$ of probability measures on $X.$ In other words, if $\phi$ denotes a fixed smooth function then $$\int_{X}\phi\mu_{1}^{(N)}=\frac{1}{N}\sum_{i}\left\langle \phi(x_{i})\right\rangle =\left\langle \phi(x_{1})\right\rangle$$ We will next explain how to define $\mu^{(N)}$ so that the one-point correlation measures converge to the normalized volume form of the Kähler-Einstein metric: $$\mu_{1}^{(N)}\rightarrow\mu_{KE}$$ in the large $N-$limit. More precisely, the convergence will hold in the weak topology on $\mathcal{M}_{1}(X),$ i.e. $$\int_{X}\phi\mu_{1}^{(N)}\rightarrow\int\phi\mu_{KE}$$ for any fixed smooth function $\phi$ on $X.$ This convergence can be interpreted as an answer to the question raised above. In fact, the argument will give a much stronger “exponential” convergence which in particular implies the asymptotic factorization of all $j-$point correlation functions (i.e. *propagation of chaos* holds).
Line bundles and Slater determinants
------------------------------------
To define $\mu^{(N)}$ first recall that we have assumed that the canonical line bundle $K_{X}\rightarrow X$ is *positive* (i.e. ample in the sense of algebraic geometry). We next recall some basic facts about line bundles (see for example [@b-v-; @d-k] for introductions aimed at physicists). To any holomorphic line bundle $L\rightarrow X$ there is a naturally associated $N-$dimensional complex vector space $H^{0}(X,L)$ consisting of global holomorphic section of $L\rightarrow X$ and the limit we will be interested is when $L$ is replaced by a large tensor power $L^{\otimes k}.$ Since $L$ is assumed ample it follows that the dimension $N=N_{k}$ (which will be the number of particles of our gas) grows with $k$ in the following way: $$N_{k}:=\dim_{\C}H^{0}(X,L^{\otimes k})=Vk^{n}+o(k^{n})$$ where the *volume* $V>0.$ In particular,
$$N(=N_{k})\rightarrow\infty\Leftrightarrow k\rightarrow\infty$$ We will often omit the subscript $k$ in $N_{k}$.
In physics, $H^{0}(X,L)$ usually arises as the quantum ground state space of a single *chiral fermion* on $X$ coupled to $L$ [@vv; @b-v-], since it may be realized as the zero of a gauged Dirac operator (once metrics/gauge fields are introduce, as explained below). The corresponding $N-$particle space of fermions is then, according to Pauli’s exclusion principle, represented by the top exterior power $\Lambda^{N}H^{0}(X,L).$ In other words this is the maximally filled many particle fermion state. As a consequence it is one-dimensional and may, up to scaling, be represented by the $N-$body state $$\Psi(x_{1},...x_{N}):=\Psi_{1}(x_{1})\wedge\cdots\wedge\Psi_{N}(x_{N})$$ expressed in terms of a given base $(\Psi_{I})$ in $H^{0}(X,L),$ where $I=1,...,N.$ Locally this means that $\Psi$ may be written as a Slater determinant: $$\Psi(Z_{1},...,Z_{N})=\det(\Psi_{I}(Z_{J}))\label{eq:slater det}$$ which hence transforms as a holomorphic section of the line bundle $L^{\boxtimes N}$ over $X^{N}.$
### Introducing metrics
Usually, one equips $L$ with an *Hermitian metric* $h_{0}.$ Taking the squared point-wise norm $\left\Vert \Psi(Z)\right\Vert ^{2}$ with respect to $h_{0}$ of a section $\Psi$ of $L$ hence gives a scalar function on $X.$ Let us briefly recall the notion of curvature in this context. The (Chern) *curvature form* $\Theta$ of $h_{0}$ is the globally well-defined two-form on $X$ locally defined as follows: if $s$ is a local trivializing holomorphic section of $L,$ then $$\Theta:=-\partial\bar{\partial}\log(\left\Vert s\right\Vert ^{2})\label{eq:def of curvature}$$ Physically, the curvature form $\Theta$ represents a background *magnetic* two-form of bidegree $(1,1)$ to which the fermions are minimally coupled. More precisely, the holomorphic structure on $L$ together with the Hermitian metric $h_{0}$ determines a reduction of $L$ to a $U(1)-$bundle and a unique unitary connection $A$ on $L,$ i.e. a $U(1)-$gauge field such that its field strength $F_{A}=\Theta$ is of type $(1,1)$ [@b-v-; @d-k]. Let us briefly recall the local meaning of this correspondence. In terms of a local holomorphic trivialization $s$ of $L$ we may write $$\Psi(z)=f(z)s$$ where $f(z)$ is a local holomorphic function, i.e. $\bar{\partial}f=0.$ Writing $h_{0}(z):=\left\Vert s\right\Vert ^{2}=e^{-\Phi}$ for a local function $\Phi$ we then have $$\left\Vert \Psi(Z)\right\Vert ^{2}=|f(z)|^{2}e^{-\Phi}=|f_{\Phi}(z)|^{2}$$ where $f_{\Phi}:=f(z)e^{-\Phi/2}$ represents $\Psi$ wrt the unitary (but non-holomorphic) local trivialization $e^{\Phi/2}s.$ Note in particular that $$\bar{\partial}_{\Phi}f_{\Phi}=0,\,\,\,\,\,\,\bar{\partial}_{\Phi}=\bar{\partial}+\frac{1}{2}\bar{\partial}\Phi.\label{eq:local dirac equation}$$ Hence, setting $$A=\frac{1}{2}(\bar{\partial}\Phi-\partial\Phi),\,\,\, F_{A}:=dA=\partial\bar{\partial}\Phi$$ gives a one-form with values in $i\R$ representing the $U(1)-$gauge field (connection) and the equation \[eq:local dirac equation\] implies that $f_{\Phi}$ is a zero-mode for the Dirac operator $\D_{A}$ expressed in the unitary trivialization (where $\D_{A}$ also depends on a choice of metric on $X;$ compare section \[sub:Derivation-of-\] below).
The metric $h_{0}$ is *positively curved* precisely when the *real* two-form $$\omega:=\frac{i}{2\pi}\Theta(=\frac{i}{2\pi}\partial\bar{\partial}\Phi)\label{eq:def of norm curv}$$ is positive definite, i.e. when it defines a *Kähler metric* on $X$ (and $\Phi$ is sometimes called the local Kähler potential of $\omega$). The line bundle $L$ is ample precisely when it admits some positively curved metric. The normalization above ensures that the cohomology class $[\omega],$ which represents the normalized *first Chern class* $c_{1}(L)$ is an integer class, i.e. it lies in the integer lattice $H^{2}(X,\Z)$ of $H^{2}(X,\R).$
The Hermitian metric $h_{0}$ naturally induces metrics on all tensor powers of $L$ etc. Coming back to the Slater determinant above, the point-wise squared norm with respect to the metric $h_{0}$ $$\left\Vert \Psi(Z_{1},...,Z_{N})\right\Vert ^{2}$$ is, from a physical point of view, proportional to the probability of finding (or creating) particles at the point $Z_{1},...$$Z_{N}$ on $X$ in the presence of the corresponding background magnetic field. To normalize it we also need to pick an integration measure $\mu_{0}$ (a volume form) on $X$ so that $$\left\Vert \Psi(Z_{1},...,Z_{N})\right\Vert ^{2}/\mathcal{Z}_{N},\,\,\,\mathcal{Z}_{N}:=\int_{X^{N}}\left\Vert \Psi\right\Vert ^{2}\mu_{0}^{\otimes N}$$ is a probability density on $X^{N}.$ Since, $\Lambda^{N}H^{0}(X,L)$ is one-dimensional the probability density above is, by homogeneity, independent of the choice of base $(\Psi_{I})$ in $H^{0}(X,L),$ but, of course, it does depend on the metric $h_{0}$ on $L$ (or more precisely on the background magnetic field $F_{A}$) and also on the integration measure $\mu_{0}$ on $X.$
The canonical background free ensemble \[sub:The-canonical-background\]
-----------------------------------------------------------------------
The main point of the present paper is the simple observation that in the particular case when $L$ is the canonical line bundle $K_{X}$ there is no need to specify any metric on $K_{X}$ if one defines a probability measure on $X^{N}$ by $$\mu^{(N)}=(\Psi_{1}\wedge\bar{\Psi}_{1}\wedge\cdots\wedge\Psi_{N}\wedge\bar{\Psi}_{N})^{1/k}/\mathcal{Z_{N}}.$$ To see this first note that it follows from the very definition of $K_{X}$ that $(\Psi_{1}\wedge\bar{\Psi}_{1}\wedge\cdots\wedge\Psi_{N}\wedge\bar{\Psi}_{N})^{1/k}$ transforms as a (degenerate) volume form on $X^{N}.$ The points is that if $\Psi$ is a section of $K_{X}(:=\Lambda^{n}(TX))\rightarrow X$ then $\Psi\otimes\bar{\Psi}$ gives a well-defined measure on $X$ (i.e. a degenerate volume form), concretely this is a consequence of the local representation $\Psi=f(z_{1},...,z_{N})dz_{1}\wedge\cdots\wedge dz_{N}.$ Hence after dividing by $$\mathcal{Z_{N}}=\int_{X^{N}}(\Psi_{1}\wedge\bar{\Psi}_{1}\wedge\cdots\wedge\Psi_{N}\wedge\bar{\Psi}_{N})^{1/k}=\int_{X^{N}}|\det(f_{I}(Z_{J}))|^{2/k}dV(Z_{1})\wedge\cdots\wedge dV(Z_{N})$$ one obtains a probability measure $\mu^{(N)}$ on $X^{N_{k}}$ which is canonically associated to $(X,K_{X}^{\otimes k}),$ since by homogeneity it is independent of the base $(\Psi_{I})$ in $H^{0}(X,K_{X}^{\otimes k}).$ Note that when $n=1,$ i.e. $X$ is a Riemann surface of genus at least two the space $H^{0}(X,K_{X}^{\otimes k})$ arises as the space of spin $2k$ particles [@vv; @b-v-] .
\[sub:General-ensembles.\]General $\beta-$ensembles.
----------------------------------------------------
Before turning to the investigation of the thermodynamical convergence towards the Kähler-Einstein volume form $\mu_{KE}$ it should be pointed out that *integer* powers of Slater determinants have been used before to model the *fractional Quantum Hall effect* [@l]. More generally we note that the previous construction may be generalized by introducing general $k-$dependent powers $\beta_{k}$ in the Slater determinant. To see this we come back to the general setting of an ample line bundle $L\rightarrow X$ and now fix a background metric $h_{0}$ on $L$ and a volume form $\mu_{0}$ on $X.$ To this geometric data we associate the probability measure $$\mu^{(N_{k})}=\left\Vert \Psi\right\Vert ^{\beta_{k}}\mu_{0}^{\otimes N}/\mathcal{Z_{N}}.$$ on $X^{N}$ for a fixed choice of parameters $\beta_{k}.$ The case of $L=K_{X}$ considered above is obtained by setting $\beta_{k}=2/k,$ fixing any metric $h_{0}$ on $K_{X}$ and then letting $\mu_{0}=1/h_{0},$ which defines a volume form on $X.$ Then it is easy to see that all factors of $h_{0}$ cancel out leading to the previous canonical construction above. Finally, note that if one defines the *Hamiltonian* $$H^{(N)}:=-\log\left\Vert \Psi\right\Vert$$ then $\mu^{(N_{k})}$ may be represented as a Boltzmann-Gibbs ensemble $$\mu^{(N_{k})}=e^{-\beta_{k}H^{(N)}}\mu_{0}^{\otimes N}/\mathcal{Z_{N}},\label{eq:bolt-gibbs}$$ of a *classical* system in thermal equilibrium with an external heat bath of temperature $T_{k}=1/\beta_{k}.$ From this point of view $\mathcal{Z_{N}}$ is the *partition function* of the system. It depends of the choice of bases $(\Psi_{I})$ in $H^{0}(X,L^{\otimes k})$ (but $\mu^{(N_{k})}$ does not, as explained above). For example, the case when $\beta_{k}=1,$ $2$ or $4$ appears in the study of the Random Matrix ensembles associated to the classical groups (see [@j] and references therein). In this latter cases $X$ is taken as the Riemann sphere, i.e. the compactification of $\C$ and then the Hilbert space $H^{0}(X,L^{\otimes k})$ appears as the corresponding orthogonal polynomials on $\C.$ However, as will be explained below we will be different in the case when $\beta_{k}$ depends on $k$ and is of the order $1/k$ which changes the thermodynamical limit, as compared to ordinary Random Matrix Theory.
It is worth emphasizing that the Hamiltonian $H^{(N)}$ above is not a sum of pair interactions (even to the leading order) when $n>1.$ This is closely related to the fact that the mean field equations obtained in section \[sub:The-minimizer-\] are fully non-linear (i.e. non-linear in the derivative terms, so that the corresponding actions are of higher order) and it makes the analysis of the thermodynamical limit rather challenging.
\[sub:Convergence-in-the thermo\]Convergence in the thermodynamical limit
=========================================================================
It will be illuminating to consider the general setting of the previous setting with $$\beta_{k}=\beta/k$$ for a fixed parameter $\beta$ (where $\beta=2$ appears in the canonical background free case \[sub:The-canonical-background\]). As will be clear this is, in a certain sense, a mean field limit. As explain above we hence fix the geometric data $(h_{0},\mu_{0})$ consisting of Hermitian metric $h_{0}$ on $L\rightarrow X$ and a volume form $\mu_{0}$ on $X$ (the canonical case is then a special case if one takes $h_{0}=1/\mu_{0}$ for any given volume form $\mu_{0}).$ Given this data we furthermore fix a base $(\Psi_{I})$ in $H^{0}(X,L^{\otimes k}),$ for any $k,$ which is orthonormal with respect to Hilbert space structure on $H^{0}(X,L^{\otimes k})$ induced by $(h_{0},\mu_{0}):$ $$\left\langle f,g\right\rangle _{X}:=\int_{X}\left\langle f,g\right\rangle \mu_{0},$$ where the point-wise Hermitian product in the integrand is taken with respect $h_{0}.$ In particular, the corresponding partition function is then (a power of) the induced $L^{\beta/k}$ norm of the corresponding Slater determinant $\Psi$ (formula \[eq:slater det\]): $$\mathcal{Z}_{N}:=\int_{X^{N}}\left\Vert \Psi\right\Vert ^{\beta/k}\mu_{0}^{\otimes N}$$ To prove the convergence we will use the techniques of the *theory of large deviations.* In a nutshell this is a formalism which allows one to give a meaning to the statement that a given sequence of probability measure $\mu^{(N)}$on $X^{N}$ is “exponentially concentrated on a deterministic macroscopic measure $\mu_{*}$ with a rate functional $I(\mu)"$ (see [@to] for an introduction to the theory of large deviations, due to Cramér, Varadhan and others, emphasizing the links to statistical mechanics - relations to functional integrals are emphasized in [@gaw]) . Heuristically, the idea is to think of the large $N-$limit of the $N-$particle space $X^{N}$ of “microstates” as being approximated by a space of “macrostates”, which is the space $\mathcal{M}_{1}(X)$ of all probability measures on $X:$ $$X^{N}\sim\mathcal{M}_{1}(X),$$ as $N\rightarrow\infty.$ The exponential concentration referred to above may then be heuristically written as $$\mu^{(N)}:=\rho^{(N)}(Z_{1},...,Z_{N})dV(Z_{1})\wedge\cdots dV(Z_{N})\sim e^{-NF(\mu)}\mathcal{D}\mu,\label{eq:large dev heur}$$ where $\mathcal{D}\mu$ denotes a (formal) probability measure on the infinite dimensional space $\mathcal{M}_{1}(X)$ (more generally, the exponent $N$ could be replaced by a *rate* $a_{N}$ which is usually a power of $N),$ where $F(\mu)$ is called the *rate functional.* Exponential concentration around $\mu_{*}$ appears when $F(\mu)\geq0$ with $\mu_{*}$ the unique minimizer of $F.$ Mathematically, the “change of variables” from $X^{N}$ to $\mathcal{M}_{1}(X)$ is made precise by using the embedding $$j_{N}:\, X^{N}\rightarrow\mathcal{M}_{1}(X),\,\,\, j_{N}(x,...,x_{N}):=\frac{1}{N}\sum_{i}\delta_{x_{i}}$$ and then pushing forward $\mu^{(N)}$ to $\mathcal{M}_{1}(X)$ with the map $j_{N},$ giving a probability measure $(j_{N})_{*}\mu^{(N)}$ on $\mathcal{M}_{1}(X)$ (i.e. the law of the random variable \[eq:empirical measure\]). The precise mathematical meaning of \[eq:large dev heur\], in the sense of large deviations, is then that
$$\lim_{\delta\rightarrow0}\lim_{N\rightarrow\infty}\frac{1}{N}\log\int_{\mathcal{B}_{\delta}(\mu)}(j_{N})_{*}\mu^{(N)}=-F(\mu),$$ integrating over a small ball $\mathcal{B}_{\delta}(\mu)$ of radius $\delta$ centered at $\mu\in\mathcal{M}_{1}(X)$ (using any metric on $\mathcal{M}_{1}(X)$ which is compatible with the weak topology) and where we have assumed for simplicity that $F$ is continuous (however in our case the definition of the LDP is slightly more complicated as $F$ will only be semi-continuous).
The idea is now to establish the asymptotics \[eq:large dev heur\] for a certain *free energy functional* $F(\mu)$ which is minimized precisely on a measure $\mu_{*}$ which equals the Kähler-Einstein measure $\mu_{KE}$ in the canonical case introduced in section \[sub:The-canonical-background\]. In fact, in this latter case the functional $F(\mu)$ will turn out to be naturally identified with Mabuchi’s K-energy, which plays an important role in Kähler geometry (as explained in section \[sub:Duality-and-relation\])
To this end we will combine two already established asymptotics, concerning the the case when $\beta_{k}=0$ and $\beta_{k}=2$ respectively. In the first case it is a classical result going back to the work of Boltzmann (called Sanov’s theorem in the mathematics literature) that the asymptotics \[eq:large dev heur\] hold with $-F(\mu)$ equal to the relative entropy functional $S(\mu)$: $$\mu_{0}^{\otimes N}\sim e^{NS(\mu)}\mathcal{D}\mu,\label{eq:large dev as beta zero}$$ where $$S(\mu):=-\int_{X}\log(\frac{\mu}{\mu_{0}})\mu(\leq0)\label{eq:def of entropy}$$ if $\mu$ has a density with respect to $\mu_{0}$ and otherwise $S(\mu)=-\infty.$ This result gives a precise meaning to Boltzmann’s notion of entropy as proportional to the logarithmic number (or volume) of microstates corresponding to a given macrostate.
Next, in the case when $\beta_{k}=2$ it was shown very recently in [@berm3] that $$e^{-2H^{(N)}(x_{1},...,x_{N})}\mu_{0}^{\otimes N}\sim e^{-kNE(\mu)}\mathcal{D}\mu\label{eq:large deve as beta 2}$$
In the present work we are interested in the intermediate asymptotic regime where $\beta_{k}=\beta/k.$ Decomposing the corresponding probability measure $\mu^{(N)}$ as
$$\mu^{(N)}:=(e^{-2H^{(N)}(x_{1},...,x_{N})})^{\beta/2k}\mu_{0}^{\otimes N}$$ or more precisely as $$\mu^{(N)}:=\left[(e^{-2H^{(N)}(x_{1},...,x_{N})}\mu_{0}^{\otimes N})^{\beta/2k}\right]\left[\cdot(\mu_{0}^{\otimes N})^{1-\beta/2k}\right]$$ we can, at least heuristically, combine the asymptotics \[eq:large dev as beta zero\] and \[eq:large deve as beta 2\] to get $$\mu^{(N)}\sim e^{-N\beta(E(\mu)-\frac{1}{\beta}S(\mu))}\mathcal{D}\mu\label{eq:rasympt with rate fun}$$ (in order to be mathematically rigorous this heuristic argument needs to be complemented with precise estimates justifying the “interchange” of the large $N$ and small $\delta-$limits)
The convergence of of the one-point correlation measures $\mu_{1}^{(N)}$ towards the minimizer $\mu_{*}$ of the functional $$F(\mu):=E(\mu)-\frac{1}{\beta}S(\mu))$$ can now be shown by standard arguments (given the existence and uniqueness of $\mu_{*}$ which we will deal with in section \[sub:The-minimizer-\]). First note that the partition function may be asymptotically calculated as
$$\mathcal{Z_{N}}\sim\int_{\mathcal{M}_{1}(X)}\mathcal{D}\mu e^{-\beta N(E(\mu)-\frac{1}{\beta}S(\mu))},$$ giving $$-\frac{1}{\beta N}\mathcal{\log Z_{N}}\rightarrow\inf_{\mu\in\mathcal{M}_{1}(X)}(E(\mu)-\frac{1}{\beta}S(\mu)$$ Next, note that upon performing an overall scaling of the original base $(\Psi_{I})$ we may assume that the infimum above vanishes. Now fix a smooth function $\phi$ on $X$ and consider the functional $$\mathcal{F}_{N}(u):=-\log\left\langle e^{-(\phi(x_{1})+...+\phi(x_{N}))}\right\rangle :=-\log\int_{X^{N}}e^{-(\phi(x_{1})+...+\phi(x_{N}))}\mu^{(N_{k})}$$ The following basic general exact variational identity holds $$\frac{1}{\beta N}\frac{d\mathcal{F}_{N}(t\phi)}{dt}_{t=0}=\int_{X}\mu_{1}^{(N_{k})}\phi\label{eq:finite n var ident}$$ Arguing precisely as above and using the trivial asymptotics $$e^{-(\phi(x_{1})+...+\phi(x_{N}))}\sim e^{-N\int_{X}\phi\mu},\label{eq:exp term as mea}$$ hence gives $$\frac{1}{\beta N}\mathcal{F}_{N}(t\phi)\rightarrow\inf_{\mu\in\mathcal{M}_{1}(X)}(E(\mu)-\frac{1}{\beta}S(\mu)+t\int_{X}\phi\mu)$$ Finally, differentiating with respect to $t$ gives $$\frac{1}{\beta N}\frac{d\mathcal{F}_{N}(t\phi)}{dt}_{t=0}\rightarrow0+\int_{X}\mu_{*}\phi$$ and hence, using \[eq:finite n var ident\], we finally get $$\mu_{1}^{(N)}\rightarrow\mu_{*}$$ Next, we will show that the minimizer $\mu_{*}$ can be obtained by solving a mean field type equation which will reduce to the Kähler-Einstein equation in the canonical case. We will start by explaining the notion of pluricomplex energy $E(\mu)$ appearing in the asymptotics \[eq:large deve as beta 2\].
The pluricomplex energy $E(\mu)$
--------------------------------
Assume now that the fixed metric $h_{0}$ on $L$ has positive curvature, i.e. its normalized curvature form is a Kähler form that we denote by $\omega(:=\omega_{0}).$ Since, $h_{0}$ is uniquely determined up to scaling by its curvature form $\omega$ and since the probability measure $\mu^{(N)}$ is insensitive to scaling of $h_{0}$ we may as well say that the geometric data defining the $\beta_{k}$- ensemble is $(\omega_{0},\mu_{0}).$
Now any Kähler metric which is cohomologous to $\omega$ (i.e. in the class $[\omega]=c_{1}(L))$ may by the $\partial\bar{\partial}-$lemma be written as $$\omega_{\phi}:=\omega+\frac{i}{\pi}\partial\bar{\partial}\phi$$ for a smooth function $u.$ In this way the space of all Kähler metrics in $c_{1}(L)$ be identified with the space of (global) *Kähler potentials* $$\mathcal{H}_{\omega}(X):=\{\phi\in\mathcal{C}^{\infty}(X):\,\omega_{\phi}>0\}$$ modulo constants (we will usually mod out by $\R$ sometimes without mentioning it explicitly). Note that these potentials are globally well-defined and depend on the choice of $\omega$ Geometrically, the space $\mathcal{H}_{\omega}(X)$ may be identified with the space $\mathcal{H}(L)$ of *all positively curved Hermitian metrics on $L$* and $\omega_{\phi}$ with the (normalized) curvature form of the metric $$h_{\phi}:=e^{-2\phi}h_{0}$$ on $L$ corresponding to $\phi$ (as follows immediately from formula \[eq:def of curvature\]), i.e. $\phi:=\Phi-\Phi_{0}$ in terms of *local* Kähler potentials of $\omega_{\phi}$ and $\omega$ respectively.
Thanks to Yau’s solution of the Calabi conjecture one can also associate potentials to *volume forms* on $X.$ Indeed, to any volume form $\mu$ on $X$ (which we will always assume normalized so that $\int_{X}\mu=1)$ there is a unique *potential* $\phi(:=\phi_{\mu})$ in $\mathcal{H}_{\omega}(X)/\R$ such that $$\frac{\omega_{\phi}^{n}}{Vn!}=\mu,\label{eq:inhomogenous ma-eq}$$ where $V$ is the *volume* of any Kähler metric in the class $c_{1}(L).$ The equation involves the $n:$th exterior power of $\omega_{\phi}$ and is hence a non-linear generalization of the inhomogeneous Laplace equation, called the inhomogeneous (complex) *Monge-Ampère* *equation* (and the left hand side above is called the *Monge-Ampère measure* of $\phi).$
The previous equation can also be given a variational formulation by noting that there is a functional $\mathcal{E}_{\omega}$ (we will often omit the subscript $\omega)$ on the space $\mathcal{H}_{\omega}(X)$ such that its first variation is given by $$\delta\mathcal{E}(\phi):=d\mathcal{E}_{\phi}=\frac{\omega_{\phi}^{n}}{Vn!},\label{eq:var def of ma action}$$ where $d\mathcal{E}$ is the differential of $\mathcal{E}(\phi)$ seen as a one-form on $\mathcal{H}_{\omega}(X).$ The functional $\mathcal{E}$ is uniquely determined by the normalization $\mathcal{E}(0)=0$ (singled out by the fixed reference Kähler metric $\omega).$ This is a well-known functional in Kähler geometry which seems to first have been introduced by Mabuchi ( (it is denoted by $-F_{\omega}$ in the book [@t]; similar functionals also appeared in the works of Aubin and Yau). We will call it the *Monge-Ampère action*, since it physically appears as an action generalizing the *Liouville action*, as explained in section \[sub:Derivation-of-\]. It is straight-forward to obtain an explicit formula for $\mathcal{E}(\phi)$ by integrating along the line segment $t\phi$ for $0\leq t\leq1$ and get $$\mathcal{E}_{\omega}(\phi)=\frac{1}{(n+1)!V}\int_{X}\phi\sum_{i=1}^{n}(\omega^{n-j}\wedge\omega_{\phi}^{j}),\label{eq:explcit ex for e of phi}$$ but we will only make use of the defining property \[eq:var def of ma action\] in the following.
The functional $\mathcal{E}$ is (strictly) concave on $\mathcal{H}_{\omega}(X)/\R$ (with respect to the flat metric) [@t] and hence the potential $\phi_{\mu}$ may be characterized as the unique (mod $\R)$ maximizer of the the functional $$\phi\mapsto\mathcal{E}(\phi)-\left\langle \phi,\mu\right\rangle ,$$ expressed in terms of the usual pairing $$\left\langle \phi,\mu\right\rangle :=\int_{X}\phi\mu\label{eq:pairing}$$ Finally, we can now, following [@bbgz], define the *(pluricomplex) energy* $E_{\omega}(\mu)$ of the measure $\mu$ (but we will often omit the subscript $\omega)$ as $$E(\mu):=\sup_{\phi\in\mathcal{H}_{\omega}(X)}\mathcal{E}(\phi)-\left\langle \phi,\mu\right\rangle =\mathcal{E}(\phi_{\mu})-\left\langle \phi_{\mu},\mu\right\rangle$$ The first equality in fact makes sense for *any* (possibly singular) measure $\mu$ in $\mathcal{M}_{1}(X)$ and one says that $\mu$ has finite energy if $E(\mu)<\infty.$
When $n=1$ one may actually take the sup defining $E$ over *all* $\phi\in\mathcal{C}^{\infty}(X)$ (i.e. without imposing the constraint $\omega_{\phi}>0).$ Then the convex functional $E(\mu)$ is, by definition, the Legendre transform of the concave functional $\mathcal{E}(\phi)$ on $\mathcal{C}^{\infty}(X)$ (with a non-standard sign convention). It turns out that in the case $n>1$ the functional $E(\mu)$ can also be realized as a Legendre transform by extending $\mathcal{E}(\phi)$ to another (concave and one time differentiable) functional $\mathcal{F}_{\infty}$ on $\mathcal{C}^{\infty}(X)$ [@bbgz; @berm3; @b-b] (which appears in the general asymptotics \[eq:conv of free energy det proc\] below). This fact is an important ingredient in the variational approach to complex Monge-Ampère equations introduced in [@bbgz].
Note that the energy functional $E$ certainly depends on the choice of fixed Kähler metric $\omega.$ In fact, it is not hard to see that $E(\mu)\geq0$ with equality precisely if $\mu=\omega^{n}/Vn!.$ Indeed, it follows from general principles (concerning Legendre transforms) that $$\inf_{\mu\in\mathcal{M}_{1}(X)}E(\mu)=E((\delta\mathcal{E})(0))=E(\frac{\omega^{n}}{Vn!})=\mathcal{E}(0)-0=0$$ It would hence be more appropriate to call $E(\mu)$ the *relative energy of $\mu.$*
### The Riemann surface case
It may be illuminating to consider the case when $n=1,$ i.e. when $X$ is a Riemann surface. Then $\mathcal{E}(\phi)$ coincides with the functional sometimes referred to as the *Liouville action* in the physics literature [@b-v-; @o-v]: $$\mathcal{E}(\phi)=\frac{1}{2}\int_{X}\phi(\omega_{\phi}+\omega)$$ and hence, taking the potential $\phi_{\mu}$ to be normalized so that $\int\phi_{\mu}\omega=0$ we get $$E(\mu)=-\frac{i}{2\pi}\int_{X}\phi_{\mu}\partial\bar{\partial}\phi_{\mu}=\frac{i}{2\pi}\int\partial\phi_{\mu}\wedge\bar{\partial}\phi_{\mu}\label{eq:energy as gradient}$$ which is essentially the usual electrostatic energy of the continuous charge distribution $\mu$ in the neutralizing background charge $-\omega.$ Equivalently, if we define the Green function $g(x,y)$ for the scalar Laplacian $\Delta=\omega^{-1}\frac{i}{\pi}\partial\bar{\partial}$ on $X$ by the properties $g(x,y)=g(y,x)$ and $$\frac{i}{\pi}\partial_{x}\bar{\partial}_{x}g(x,y)=\delta_{x}(y)-\omega(y),\,\,\,\int_{X}g(x,y)\omega(y)=0$$ then we have $\phi_{\mu}(x)=\int_{X}g(x,y)d\mu(y)$ and hence $$E(\mu)=-\frac{1}{2}\int_{X\times X}g(x,y)d\mu(x)\otimes d\mu(y).$$
\[sub:The-minimizer-\]The minimizer $\mu_{*}$ of the free energy functional $F(\mu)$ and mean field equations
-------------------------------------------------------------------------------------------------------------
In this section we will give some formal variational arguments to determine the minimizer of $F(\mu).$ See [@berm6] for a rigorous account and further developments. Recall that the free energy functional $F(\mu)$ (for a fixed parameter $\beta>0)$ on the space $\mathcal{M}_{1}(X)$ of probability measure son $X$ is defined by $$F(\mu):=E(\mu)-\frac{1}{\beta}S(\mu)$$ where $E$ is the energy functional define in the previous section and $S(\mu)$ is the relative entropy \[eq:def of entropy\]. It follows from basic duality arguments that $F$ is strictly convex on $\mathcal{M}_{1}(X)$ (or rather on the subset where $F$ is finite) and hence admits at most one minimizer. Next we note that $\mu$ is a critical point for $F(\mu)$ on $\mathcal{M}_{1}(X)$ if and only if $$-\phi_{\mu}+\frac{1}{\beta}\log(\mu/\mu_{0})-Z_{\mu}=0,$$ where $\phi_{\mu}$ is the potential of $\mu$ and $Z_{\mu}$ is a normalizing constant. Indeed, using the defining properties of $\mathcal{E}$ and $E$ respectively one obtains (by basic Legendre transform considerations) that $$\delta E(\mu)=-\phi_{\mu}$$ as a one-form on the infinite dimensional submanifold $\mathcal{M}_{1}(X)$ of the vector space $\mathcal{M}(X)$ of all signed measures. Moreover, a simple calculation gives $$\delta S(\mu)=-\log(\mu/\mu_{0})+Z_{\mu},$$ where $Z_{\mu}$ is a normalizing constant (coming from the constraint $\int_{X}\mu=1).$ Combining these two variational formulas gives $$\delta F(\mu)=-\log(\mu/\mu_{0})-\phi_{\mu}\label{eq:variational der of f}$$ up to a normalizing constant. In other words, $\mu$ is a critical point for for $F(\mu)$ on $\mathcal{M}_{1}(X)$ if and only if its potential $\phi$ solves the following non-linear partial differential equation: $$\frac{\omega_{\phi}^{n}}{Vn!}=\frac{e^{\beta\phi}\mu_{0}}{Z_{\phi}}\label{eq:mean field equation}$$ As follows from a simple modification of the proof of the Aubin-Yau theorem there is a unique $\phi\in\mathcal{H}_{\omega}(X)/\R$ solving this equation (crucially using that $\beta>0)$ which by strict convexity is hence the unique maximizer of the free energy functional $F.$ It is sometimes convenient to fix the normalization of the solution $\phi$ above by imposing that $$\int_{X}e^{\beta\phi}\mu_{0}=1,$$ i.e. $\phi\in\mathcal{H}_{\omega}(X)$ is the *unique* solution to $$\frac{\omega_{\phi}^{n}}{Vn!}=e^{\beta\phi}\mu_{0}\label{eq:m-a mfe ej z}$$ It should be pointed out that when $n=1$ the previous equation is often called the *mean field equation* [@clmp; @k] and accordingly we will call it the *mean field Monge-Ampère equation* for a general dimension $n.$
Finally, coming back to the canonical case when $\beta=2$ and $L=K_{X}$ we take, as explained in section \[sub:The-canonical-background\] the geometric data $(\omega,\mu_{0})$ such that $\omega$ is the curvature form of the metric on $K_{X}$ defined by the inverse $1/\mu_{0}.$ This means that $\mu_{0}=e^{2f_{\omega}}\omega^{n}/Vn!,$ where $f_{\omega}$ is the *Ricci potential, i.e.* $$\frac{i}{\pi}\partial\bar{\partial}f_{\omega}=\omega+\mbox{Ric\ensuremath{\omega},\,\,\,\ensuremath{\int_{X}e^{2f_{\omega}}\omega^{n}/Vn!=1}},$$ where $\mbox{Ric}\omega=-\frac{i}{\pi}\partial\bar{\partial}(\log\omega^{n})$ is the $(1,1)-$form representing the Ricci curvature of $\omega.$ Then the corresponding Monge-Ampère mean field equation reads $$\omega_{\phi}^{n}=e^{2\phi}e^{2f_{\omega}}\omega^{n}$$ Hence, the solution $\phi$ is such that the Kähler metric $\omega_{\phi}$ satisfies $$\mbox{Ric\ensuremath{\omega_{\phi}=-\omega_{\phi}}},$$ i.e. $\omega_{\phi}$ is a Kähler-Einstein metric with negative Ricci curvature. Coming back to the convergence of the one-correlation measures in the thermodynamical limit, considered in section \[sub:Convergence-in-the thermo\], this means that the limiting measure $\mu_{*}$ indeed equals $\mu_{KE}:=\omega_{KE}^{n}/Vn!,$ as expected.
### \[sub:Derivation-of-\]The asymptotics \[eq:large deve as beta 2\] for $\beta_{k}=2$ and effective bosonization
Before continuing let us briefly explain the idea behind the large deviation asymptotics \[eq:large deve as beta 2\] proved in [@berm3] and its relation to effective bosonization. The starting point is the basic observation that when $\beta_{k}=2$ the one point correlation function $\rho_{1}^{(N)}$ can be represented as a *density of states function:* $$\rho_{1}^{(N)}(Z)=\sum_{I=1}^{N}\left\Vert \Psi_{I}(Z)\right\Vert ^{2}$$ (called the *Bergman kernel at the diagonal* in the mathematics literature). By a fundamental result of Bouche and Tian the leading asymptotics of the corresponding one point correlation measure are given by the Monge-Ampère measure: $$\mu_{1}^{(N)}\rightarrow\omega^{n}/Vn!$$ (see [@z] for a survey of Bergman kernel asymptotics and [@d-k] for a physical point of view) Now using these asymptotics and perturbing by potentials $\phi$ in $\mathcal{H}_{\omega}(X)$ one can reverse the arguments used in the end of section \[sub:Convergence-in-the thermo\] and get $$\frac{1}{N}\mathcal{F}_{N}(\phi)\rightarrow\mathcal{E}(\phi),\,\,\,\phi\in\mathcal{H}_{\omega}(X)$$ using the variational property of $\mathcal{E}.$ Then an argument involving Legendre transforms gives the large deviation asymptotics \[eq:large deve as beta 2\], using that $E(\mu)$ can be realized as an (infinite dimensional) Legendre transform of $\mathcal{E}(\phi).$ More precisely, the argument uses the convergence of (perturbed) free energies $$\frac{1}{N}\mathcal{F}_{N}(\phi)\rightarrow\mathcal{F}_{\infty}(\phi),\,\,\,\phi\in\mathcal{C}^{\infty}(X)\label{eq:conv of free energy det proc}$$ for *any* smooth function $\phi$ (not necessarily with $\omega_{\phi}\geq0$) for a certain functional $\mathcal{F}_{\infty}$ on $\mathcal{C}^{\infty}(X),$ whose Legendre transform is $E(\mu).$ The key point, as shown in [@b-b], is that $\mathcal{F}_{\infty}$ is one time differentiable on $\mathcal{C}^{\infty}(X),$ which hence establishes the absence of a phase transition with respect to perturbations of $\phi$ for the $\beta_{k}=2-$ensemble.
Incidentally, as explained in [@berm3] the large deviation asymptotics \[eq:large deve as beta 2\] can, from a physical point of view, be interpreted as an *effective bosonization* of a fermionic quantum field theory on $X$ (but is should be pointed out that this is only an interpretation: no bosonization is actually used in the derivation \[eq:large deve as beta 2\] as explained above). In other words, the collective theory of $N$ fermions is effectively described by a bosonic field theory, as $N\rightarrow\infty.$ The point is that the asymptotics \[eq:large deve as beta 2\] is equivalent, at least formally, to the asymptotics
$$\left\langle \left\Vert \Psi(x_{1})\right\Vert ^{2}\cdots\left\Vert \Psi(x_{N})\right\Vert ^{2}\right\rangle \sim\left\langle e^{i\phi(x_{1})}\cdots e^{i\phi(x_{N})}\right\rangle ,\label{eq:boson ans}$$
in the large $N-$limit. Here the lhs above is the usual $N-$point function for a fermionic quantum field theory with the usual gauged Dirac action, i.e. it is the following functional integral over Grassman fields:
$$\left\langle \left\Vert \Psi(x_{1})\right\Vert ^{2}\cdots\left\Vert \Psi(x_{N})\right\Vert ^{2}\right\rangle :=\int\mathcal{D}\Psi\mathcal{D}\bar{\Psi}e^{-S_{ferm}(\Psi,\bar{\Psi})}\left\Vert \Psi(x_{1})\right\Vert ^{2}\cdots\left\Vert \Psi(x_{1})\right\Vert ^{2},\label{eq:fermion path int}$$
integrating of over all complex spinors, i.e. smooth sections of the exterior algebra $\Lambda^{0,*}(T^{^{*}}X)\otimes L^{\otimes k}$ and where $S_{ferm}(\Psi,\bar{\Psi})$ is the fermionic action $$S_{ferm}(\Psi,\bar{\Psi})=\int_{X}\left\langle \D_{kA}\Psi,\Psi\right\rangle \mu_{0},$$ expressed in terms of the gauged Dirac operator $\D_{kA}$ on $\Lambda^{0,*}(T^{^{*}}X)\otimes L^{\otimes k}$ induced by the complex structure $J,$ the Hermitian metric on $L,$ i.e. a gauge field $A$ and a choice of Hermitian metric $g$ on $X$ with volume form $\mu_{0},$ i.e. $\D_{kA}=\overline{\partial}+\overline{\partial}^{*}$ a (see [@b-v-] for the Riemann surface case). The integer $N$ is the dimension of the space of zero-modes of $\D_{A}$ on $\Lambda^{0,*}(T^{^{*}}X)\otimes L^{\otimes k}$ which coincides with $H^{0}(X,L^{\otimes k})$ due to Kodaira vanishing in positive degrees (when $k>>1)$ and is hence, in fact, independent of the metric $g.$ As for the rhs in \[eq:boson ans\] the bracket denote integration wrt the (formal) functional measure $\mathcal{D}\phi e^{-S_{bose}(\phi)}$ over all scalar field $\phi$ on $X$ and where $$S_{bose}(\phi)=-\frac{1}{(-i)^{n-1}}\mathcal{E}_{-i\omega}(\phi)\label{eq:ansats for bos act}$$ defining bosonic action which is a “higher-derivative action” when $n>1$ and when $n=1$ it coincides with non-solotonic part of the bosonic action obtained in [@b-v-; @vv].
Relation to the Yau-Tian-Donaldson program and quantum gravity
==============================================================
\[sub:Duality-and-relation\]Duality and relation to the Yau-Tian-Donaldson program and balanced metrics
--------------------------------------------------------------------------------------------------------
In this section we will briefly point out some relations to the influential Yau-Tian-Donaldson program in Kähler geometry [@do1; @don1; @t; @p-s]. In a nutshell the idea of this program is to approximate Kähler-Einstein metrics (and more general extremal metrics), by a limit of finite dimensional objects of an algebro-geometrical nature. There are various versions of this program, but the one which is most relevant for the present paper is Donaldson’s notion of *canonically balanced metrics* introduced in [@don2], which is particularly adapted to Kähler-Einstein metrics (as opposed to general extremal metrics).
To highlight the similarities let us first formulate a more general “$\beta-$analogue” of Donaldson’s setting, starting with an ample line bundle $L\rightarrow X.$ The main point is to replace the infinite dimensional space $\mathcal{H_{\omega}}(X)$ of Kähler potentials for $\omega\in c_{1}(L)$ (we recall that $\mathcal{H_{\omega}}(X)$ may be identified with the space of all positively curved metrics on $L)$ with its *quantization at level $k.$* This latter space, denoted by $\mathcal{H}_{k},$ is the space of all Hermitian metrics on the finite dimensional vector space $H^{0}(X,L^{\otimes k}).$ Upon fixing a reference metric ** $\mathcal{H}_{k}$ is hence isomorphic to the symmetric space $$GL(N_{k})/U(N_{k})\label{eq:symmetric space finite}$$ of all Hermitian $N_{k}\times N_{k}$- matrices. There is a natural injection defined by the *Fubini-Study map* $FS_{k}(H)$ at level $k:$ $$FS_{k}:\,\mathcal{H}_{k}\rightarrow\mathcal{H_{\omega}}(X),\,\,\, FS_{k}(H)(x):=\frac{1}{k}\log\sum_{I=1}^{N}\left\Vert \Psi_{I}(x)\right\Vert ^{2},\label{eq:fs}$$ expressed in terms of the point-wise norms with respect to the fixed metric $h_{0}^{\otimes k}$ on $L^{\otimes k}$ of a base $(\Psi_{I})$ in $H^{0}(X;L^{\otimes k})$ which is orthonormal with respect to $H.$ Moreover, for any given $\beta$ we may define a map in the reversed direction that we will call $Hilb_{k,\beta}:$ $$Hilb_{k,\beta}:\,\mathcal{H_{\omega}}(X)\rightarrow\mathcal{H}_{k}$$ defined as follows: $Hilb_{k,\beta}(\phi)$ is the Hermitian product (or equivalently, Hilbert norm) on $H^{0}(X,L^{\otimes k})$ defined by $$\left\langle f,g\right\rangle _{Hilb_{k,\beta}(\phi)}:=\int_{X}\left\langle f,g\right\rangle e^{-k\phi}e^{\beta\phi}\mu_{0}$$ (note that $\left\langle \cdot,\cdot\right\rangle e^{-k\phi}$ is the Hermitian metric on $L^{\otimes k}$ *naturally* associated to $\phi\in\mathcal{H}_{\omega}(X)$ and the remaining factor $e^{\beta\phi}\mu_{0}$ should be thought of as a specific choice of integration element depending on $\phi).$ An element $H_{k}$ in $\mathcal{H}_{k}$ will be said to be *$\beta-$balanced at level $k$* with respect to $(\omega,\mu_{0})$ if is is a fixed point under the composed map $$T_{k,\beta}:=Hilb_{k,\beta}(\phi)\circ FS_{k}:\,\,\,\mathcal{H}_{k}\rightarrow\mathcal{H}_{k}.\label{eq:donaldsons t map}$$ on $\mathcal{H}_{k}.$ Equivalently, this means that $H_{k}$ is a critical point of the following functional $\mathcal{G}_{k}$ on $\mathcal{H}_{k}:$ $$\mathcal{G}_{k}(H):=-\frac{1}{kN}\log\det(H)-\frac{1}{\beta}\log\int_{X}e^{\beta FS_{k}(H)}\mu_{0},$$ (after normalization). Repeating the arguments in the proof of Theorem 7.1 in [@bbgz] concerning the canonical case when $L=K_{X}$ ((the case referred to as $S_{+}$ in [@bbgz]) essentially word for word, one obtains the existence and uniqueness of a $H_{k}\in\mathcal{H}_{k}$ which is *$\beta-$balanced at level $k$* with respect to $(\omega,\mu_{0})$ and such that $$FS_{k}(H)\rightarrow u_{\beta},$$ in $\mathcal{H_{\omega}}(X)$ when $k\rightarrow\infty$ (or equivalently, $N\rightarrow\infty)$ where $u_{\beta}$ is the unique solution of the Monge-Ampère mean field equation \[eq:m-a mfe ej z\], assuming $\beta>0.$ As explained in [@bbgz] the main point of the proof is to show that any limit point in $\mathcal{H_{\omega}}(X)/\R$ of the sequence $FS_{k}(H)$ is a maximizer of the following functional on $\mathcal{H_{\omega}}(X):$ $$\mathcal{G}(\phi):=\mathcal{E}(\phi)-\frac{1}{\beta}\log\int_{X}e^{\beta\phi}\mu_{0},$$ whose critical points are precisely the solutions of the Monge-Ampère mean field equation \[eq:mean field equation\]. Note that the functional $\mathcal{G}$ is invariant under the natural action by $\R,$ $\phi\rightarrow\phi+c$ and hence maximizing the functional $$\mathcal{E}(\phi)-\frac{1}{\beta}\int_{X}e^{\beta\phi}\mu_{0}\label{eq:liov exp}$$ picks out the maximizers of $\mathcal{G}$ which satisfies the normalization $$\int_{X}e^{\beta\phi}\mu_{0}=1$$ In the Riemann surface case the functional \[eq:liov exp\] with the exponential term is also sometimes referred to as the *Liouville action* (it appears for example in Polyakov’s functional integral quantization of the bosonic string, further developed in [@o-v])
To see the relation to the $\beta-$ensembles introduced in section \[sub:General-ensembles.\] and their thermodynamical limit one should keep in mind the basic linear duality between *functions* $\phi$ and *measures* $\mu$ defined by the basic pairing \[eq:pairing\] In turn, this pairing induces, using the Legendre transform a non-linear duality between *convex functionals* of $\phi$ on one hand and convex functionals of $\mu,$ on the other.
The roles of the spaces $\mathcal{H_{\omega}}(X)$ and $\mathcal{H}_{k}(X)$ are now played by the space $\mathcal{M}_{1}(X)$ and $\mathcal{M}(X^{N_{k}}),$ respectively, where $\mathcal{M}_{N_{k}}(X)$ denotes the space of all symmetric probability measures on the product $X^{N}$ (i.e. all $N_{k}-$particle random point processes on $X).$ The analogue of the Fubini-Study map \[eq:fs\] is the map $$\mathcal{M}(X^{N_{k}})\rightarrow\mathcal{M}_{1}(X),\,\,\,\mu_{N}\mapsto(\mu_{N})_{1}:=\left\langle \frac{1}{N}\sum_{i}\delta_{x_{i}}\right\rangle ,$$ sending a random point process to its one-point correlation measure. Finally, the role of a $\beta-$balanced metric is now played by the measure $\mu^{(N_{k})}\in\mathcal{M}(X^{N_{k}})$ defining the $\beta-$ensemble with $N_{k}$ particles, which was expressed as a Boltzmann-Gibbs ensemble with Hamiltonian $H^{(N_{k})}$ in formula \[eq:bolt-gibbs\]. The point is that $\mu^{(N_{k})}$ can also be defined by a variational principle. Indeed, by the $N-$particle Gibbs principle for canonical ensembles $\mu_{(N_{k})}$ is the unique minimizer of the $N-$particle mean free energy functional on $\mathcal{M}(X^{N_{k}}):$ $$F^{(N)}(\mu_{N})=\frac{1}{N}\int_{X^{N}}\mu_{N}H^{(N)}-\frac{1}{N}S(\mu_{N},\mu_{0}^{\otimes N})\label{eq:N particle free en}$$ i.e. the difference between *mean energy* and *mean entropy*. There is also an analogue of the definition of a balanced metric as a fixed point of the map $T_{k,\beta}$ above. Indeed, it is well-known that any Gibbs-Boltzmann measure can be uniquely determined as a stationary state for a stochastic process $\mu_{t}$ on $X^{N}$ defined by suitable Glauber (or Langevin) dynamics, but we will not develop this point of view here.
Interestingly, performing a Legendre transform of each of the two convex functionals on $\mathcal{M}_{1}(X)$ summing up to the free energy functional $F(\mu)$ (i.e. the pluricomplex energy $E(\mu)$ and minus entropy $-\frac{1}{\beta}S(\mu))$ yields a functional on $\mathcal{H_{\omega}}(X)$ which is nothing but the functional $\mathcal{G}$ above: $$F=E+(-\frac{1}{\beta}S),\,\,\,\,\mathcal{G}=E^{*}+(-\frac{1}{\beta}S)^{*}$$ It should be pointed out that in the canonical case (where the critical points of the functionals are Kähler-Einstein metrics) the two functionals $F$ and $\mathcal{G}$ have already appeared in Kähler geometry from a different point of view. For example, the limiting free energy functional $F(\mu)$ on $\mathcal{M}_{1}(X)$ may be identified with Mabuchi’s $K-$energy $\nu$ of a Kähler metric in $c_{1}(K_{X}):$ $$F(\omega_{\phi}^{n}/Vn!)=\nu(\omega_{\phi})\label{eq:f as mab}$$ The functional $\nu$ was first introduced by Mabuchi as the functional on $\mathcal{H_{\omega}}(X)$ whose gradient with respect to the Mabuchi-Semmes-Donaldson Riemannian metric on $\mathcal{H_{\omega}}(X)$ is the scalar curvature minus its average [@t; @p-s]. But this is easily seen to be equivalent to the variational property \[eq:variational der of f\] of $F$ and hence $F$ and $\nu$ coincide under the identification above (up to an additive constant). The explicit formula for $\nu$ obtained from the identification \[eq:f as mab\] is in fact equivalent to an explicit formula for $\nu$ of Tian and Chen [@t]. Moreover, the functional $-G$ coincides with the Ding called functional (see the book [@t] and references therein). Interestingly, using Legendre transforms as above one arrives at new proofs and generalizations of various useful results in Kähler geometry [@berm6].
Finally, it seems worth pointing out that in the case when $\beta=0$ the notion of balanced metrics still makes sense and was studied by Donaldson in [@don2] with a particular emphasize on the case when $X$ is a Calabi-Yau form. Then $\mu$ may be canonically chosen as $i^{n^{2}}\Omega\wedge\bar{\Omega}/\int i^{n^{2}}\Omega\wedge\bar{\Omega}$ where $\Omega$ is non-vanishing holomorphic $n-$form on $X$ and the curvature forms of the balanced metrics at level $k$ then converge to the unique Ricci flat metric in $[\omega],$ whose existence was established in Yau’s proof of the Calabi conjecture. Relation between these balanced metrics on Calabi-Yau manifolds and black holes were considered in [@d-k2]. However, in the case when $\beta=0$ the $\beta-$ensembles introduced in the present work appear to be less interesting: they are pure Poisson processes without any connections to fermions. The considerably more complicated case when $\beta$ is *negative* is briefly discussed below in connection to Kähler-Einstein metrics with positive Ricci curvature.
Relations to quantum gravity and the space of all complex structures
--------------------------------------------------------------------
In Hawking’s approach to Euclidean quantum gravity [@ha] one considers the (formal) functional measure $$\mathcal{D}ge^{-S(g)}$$ on the space of all Riemannian metrics $g$ (modulo diffeomorphisms), where $S(g)$ is the Einstein-Hilbert action with a cosmological constant $\Lambda$ and where we have set the fundamental constants to be equal to one (but eventually we will discuss semi-classical limits). One also has to integrate over all diffeomorphism types of four-manifolds, but the relation to the present setting appears more closely when one considers the restriction of the integration to metrics on a fixed compact smooth manifold $X.$ As discussed by Hawking, by choosing a conformal gauge, the integration may be decomposed as an integration over a conformal factor and the space of conformal equivalence classes of metrics (see section $5$ in [@ha] and also the very recent paper [@hoo] of t’ Hooft ).
To see the relation to the present setting we let $X$ be a smooth oriented manifold admitting some complex structure $J_{0}$ such that the corresponding canonical line bundle $K_{X_{0}}$ is ample (when $n=1$ this just means that $X$ has genus at least two). For simplicity we will also assume that $H^{1}(X,\R)=0.$ We let $\mathcal{J}(X)$ be the space of *all* complex structures $J$ on $X,$ which are compatible with the orientation. Given $J$ we denote by $\mathcal{K}(X,J)$ be the space of all Kähler metrics on $(X,J)$ in the cohomology class $c_{1}(K_{(X,J)})$ and we let $\mathcal{K}(X)$ be their union over all $J$ in $\mathcal{J}(X).$ The group $DIFF(X)$ of diffeomorphisms on $X$ acts naturally on $\mathcal{J}(X)$ and $K(X)$ and we let $\mathcal{T}(X)$ be the corresponding moduli space $\mathcal{T}(X)=\mathcal{J}(X)/DIFF(X)$ [^1]and consider the associated bundle $$\mathcal{K}(X)/DIFF(X)\rightarrow\mathcal{T}(X)\label{eq:bundle of equiv classes of kahler me}$$ of equivalence classes of Kähler metrics. This bundle may be given a useful complex geometric realization as follows. First we recall that there is a natural holomorphic fibration (the *universal family*) $$\mathcal{X}\rightarrow\mathcal{T}(X)$$ such the fiber $\mathcal{X}_{[J]}$ is a complex manifold biholomorphic to $(X,J)$ (when $n=1$ $\mathcal{X}$ is called the *universal curve* over $\mathcal{T}(X)).$ The bundle \[eq:bundle of equiv classes of kahler me\] may now be identified with an infinite dimensional bundle of metrics over $\mathcal{T}(X)$ such that the fiber over $[J]$ gets identified with the corresponding space of Kähler metrics on $\mathcal{X}_{[J]}.$ Equivalently, $$\mathcal{K}(X)/DIFF(X)\simeq\mathcal{H}(K_{\mathcal{X}/\mathcal{T}(X)})/\R,$$ where the rhs is the space of all fiber-wise positively curved metrics on the relative canonical line bundle $K_{\mathcal{X}/\mathcal{T}(X)}$ over $\mathcal{X}:$ $$K_{\mathcal{X}/\mathcal{T}(X)}\rightarrow\mathcal{X}\rightarrow\mathcal{T}(X),$$ (by definition the restriction of $K_{\mathcal{X}/\mathcal{T}(X)}$ to each fiber is the usual canonical line bundle). We can now consider a (formal) functional measure on the space $\mathcal{K}(X)/DIFF(X)$ of the form $$\mathcal{D}ge^{-\frac{1}{N}F(g)}\label{eq:functional measure on space of kahler}$$ where the positive integer $N$ plays the role of a semi-classical parameter and we simply let $$F(g):=F(\mu_{g}):=F(\frac{dVol_{g}}{V})$$ where $F(\mu)$ is the free energy functional defined in section \[sub:Convergence-in-the thermo\] and where we have used that $g$ may be identified with a Kähler metric $\omega_{g}$ in $c_{1}(K_{\mathcal{X}_{[J]}})$ for some $J,$ so that $dVol_{g}=\omega_{g}^{n}/n!.$ In other words, fiber-wise over $\mathcal{T}(X)$ the functional $F(g)$ may be identified with the free energy functional for a fixed complex manifold dimorphic to $X$ (which in turn as explained in the previous section may be identified with Mabuchi’s $K-$energy). Using the previous arguments we can now give a well-defined and canonical “finite $N$ approximation” to the functional measure \[eq:functional measure on space of kahler\] as follows. First note that for each fiber $\mathcal{X}_{[J]}$ we have the canonical probability measure $\mu^{(N)}$ on $\mathcal{X}_{[J]}^{N}$ (defined in terms of fermions in section \[sub:The-canonical-background\]). Moreover the base $\mathcal{T}(X)$ admits a canonical metric, the Weil-Peterson metric, with finite volume (as follows form the results in [@sc]). Altogether this induces a “universal” probability measure $\tilde{\mu}^{(N)}$ on the finite-dimensional complex manifold $$\mathcal{X}^{(N)}:=\bigcup_{[J]\in\mathcal{T}(X)}\mathcal{X}_{[J]}^{N}$$ and the arguments in section \[sub:Convergence-in-the thermo\] give the asymptotics $$\mathcal{D}ge^{-\frac{1}{N}F(g)}\sim\tilde{\mu}^{(N)}$$ as $N\rightarrow\infty$ in a suitable sense, so that the measure $\tilde{\mu}^{(N)}$ is exponentially concentrated on the moduli space of all Einstein metrics on $X$ which are Kähler for some complex structure. To see the relation to Hawking’s setting one may think of $\mathcal{T}(X)$ as playing the role of the space of all conformal equivalence classes and the Kähler potentials as the logarithm of the conformal factors. This is more then an analogy when $n=1$ since the space $\mathcal{K}(X)$ then coincides with the space of all Riemannian metrics on $X,$ normalized so that their total area is $2\mbox{genus}(X)-2.$ However, in higher dimensions non-trivial integrability conditions appear. Of course, another difference is that the action $F(g)$ is not coming form the Einstein-Hilbert action (in any dimension). As explained in the previous section the action is, in a sense, dual to an action generalizing the Liouville action to higher dimension.
Finally, following Fujiki [@fu] and Donaldson [@do3] we will explain how the action $F$ appears naturally from a symplecto-geometric point of view when the complex structures $J$ are used as dynamical variable instead of Riemannian metrics (we refer to [@fu; @do3] for more details and further references). Fix a symplectic form $\omega$ in the real cohomology class $-c_{1}(X)=$: $c_{1}(K_{(X,J_{0})}))$ in $H^{2}(X,\Z)$ for any $J\in\mathcal{J}(X)$ and consider the subspace $\mathcal{J}(X,\omega)\subset\mathcal{J}(X)$ of all complex structures $J$ on $X$ which are compatible with $\omega,$ i.e. such that $$\mathcal{J}(X,\omega)\rightarrow\mathcal{G}(X,\omega):\,\,\, J\mapsto g_{(\omega,J)}:=\omega(\cdot,J\cdot)\label{eq:map j to g}$$ maps $J$ to a Riemannian metric which is Kähler wrt $J$ and with $\omega$ as its Kähler form. The map \[eq:map j to g\] naturally defines a principal $SDIFF(X,\omega)-$bundle over $\mathcal{G}(X,\omega),$ where $SDIFF(X,\omega)$ is the symplectomorphism group, i.e. $SDIFF(X,\omega)$ acts freely by pull-back on the fibers so that $$\mathcal{J}(X,\omega)/SDIFF(X,\omega)\simeq\mathcal{G}(X,\omega)$$ The infinite dimensional space $\mathcal{J}(X,\omega)$ is itself naturally a Kähler manifold with a Weil-Petersson type Kähler form $\Omega$ such that $SDIFF(X,\omega)$ acts by holomorphic symplectomorphism on $(\mathcal{J}(X,\omega),\Omega).$ More precisely, the action lifts to an Hermitian line bundle over $\mathcal{J}(X,\omega)$ with curvature $\Omega$ can be realized as a certain determinant line bundle (or alternatively as a Deligne pairing [@p-s]). This allows one to apply the powerful formalism of “moment maps” for group actions on Kähler manifolds and geometric invariant theory (in this case the corresponding moment map turns out to be represented by the scalar curvature). Even though the complexified group $SDIFF(X,\omega)_{\C}$ does not exist, in the strict sense, it is still possible to give a meaning to its orbits in $\mathcal{J}(X,\omega)$ which are complex submanifolds of $\mathcal{J}(X,\omega).$ On each orbit the Kähler metric $\Omega$ has an $SDIFF(X,\omega)$- invariant Kähler potential $F(J)$ which is defined up to an additive constant (corresponding to the “norm-functional”/ “Kempf-Ness functional” in the abstract setting of moment maps). The functional $F(J)$ may be uniquely determined by requiring its infimum to be equal to $0$ (in general the “norm functional” is bounded from below precisely when the group action is semi-stable which is indeed the case in our setting). Finally, for any fixed complex structure $J_{0}$ the quotient of the orbit of $SDIFF(X,\omega)_{\C}$ passing through $J_{0},$ i.e. $$\left(SDIFF(X,\omega)_{\C}J_{0}\right)/SDIFF(X,\omega)\label{eq:infinite dim symm space}$$ may, under the map \[eq:map j to g\], be identified with the space of all Kähler metrics in $c_{1}(K_{(X,J_{0})})$ (this is the $N=\infty$ analogue of \[eq:symmetric space finite\]). Under this identification $F(J)$ coincides with the corresponding free energy functional (since, as shown by Fujiki and Donaldson it coincides with Mabuchi’s $K-$energy functional on the space of Kähler metrics). Concretely, the identification used above is obtained by noting that $J$ is in the orbit $SDIFF(X,\omega)_{\C}J_{0}$ iff there is $f\in DIFF(X)$ such that the pull-back $f^{*}\omega$ defines a Kähler form on $(X,J_{0})$ which is cohomologous to $\omega.$
In conclusion this means that from one point of view the space $K(X)/DIFF(X)$ where the functional measure \[eq:functional measure on space of kahler\] lives may be identified with the space of positively curved metrics on the relative canonical line bundle $K_{\mathcal{X}/\mathcal{T}(X)}$ (mod $\R)$ and the action $F(g)$ with the free energy functional, fiber-wise over the moduli space $\mathcal{T}(X).$ From another point of view $K(X)/DIFF(X)$ may be identified with $\mathcal{J}(X,\omega)/SDIFF(X,\omega)$ and the corresponding action $F(g)$ with a (normalized) $SDIFF(X,\omega)-$invariant Kähler potential $F(J)$ on the infinite dimensional Kähler manifold $\mathcal{J}(X,\omega).$ In other words, when formulating the theory directly on the space $\mathcal{J}(X,\omega)$ it becomes a gauge theory with structure group $SDIFF(X,\omega)$ for a bundle over the moduli space of complex structures. The finite $N-$approximation is obtained by replacing $K(X)/DIFF(X)$ with the finite dimensional complex manifold $\mathcal{X}^{(N)}$ fibered over $\mathcal{T}(X)$ and replacing the formal functional measure with a probability measure on $\mathcal{X}^{(N)}.$ Finally we note that another somewhat dual approximation, in the spirit of the Yau-Tian- Donaldson program discussed in the previous section, could also be obtained by instead replacing $\mathcal{K}(X)/DIFF(X)$ with the space of Hermitian metrics on the vector bundle $E_{N}\rightarrow\mathcal{T}(X)$ defined as the the direct image of $K_{\mathcal{X}/\mathcal{T}(X)}^{\otimes k},$ i.e. $$E_{N_{k}}:=\bigcup_{[J]\in\mathcal{T}(X)}H^{0}(\mathcal{X}_{[J]},K_{\mathcal{X}_{[J]}}^{\otimes k})$$ (compare the setup in [@berm3-1]). Note that in this latter case the structure group is $U(N)$ and, in a sense, the Yau-Tian- Donaldson program concerns the problem of making the heuristic statement $U(N)\rightarrow SDIFF(X,\omega),$ as $N\rightarrow\infty,$ mathematically precise.
\[sub:Conclusion-and-discussion\]Conclusion and outlook
-------------------------------------------------------
To a given a compact manifold with a fixed integrable complex structure $J$ we have associated a canonical $N-$particle free fermion gas whose one-particle correlation measures converge in the thermodynamical (large $N)$ limit to the volume form of the Kähler-Einstein metric $\omega_{KE}$ associated to $(X,J).$ More precisely, it was assumed that the canonical line bundle $K_{X}$ be positive (i.e. ample), which corresponds to $\omega_{KE}$ having *negative* Ricci curvature and the one-particle quantum state space of the fermion gas was taken as the $N_{k}-$dimensional space $H^{0}(X,K_{X}^{\otimes k})$ of global holomorphic sections on $X$ with values in $K_{X}^{\otimes k}.$ The argument in fact gave precise exponentially small fluctuations around the Kähler-Einstein volume forms with a rate function $F$ naturally identified with Mabuchi’s $K-$energy (which plays an important rule in Kähler geometry). The convergence in the thermodynamical limit was obtained by introducing an *auxiliary* background Kähler form $\omega$ in the first Chern class $c_{1}(K_{X})$ (or equivalently a metric on $K_{X}$) also determining a volume form on $X.$ This lead to a decomposition of the rate functional $F$ as $$F(\mu)=E(\mu)-S(\mu),$$ (with both terms depending on the choice of $\omega).$ In terms of the statistical mechanics of a classical canonical Boltzmann-Gibbs ensemble $E$ and $S$ appeared as the limiting *mean energy* and *mean entropy,* respectively. Minimizing $F$ then gave an equation of mean field type whose unique solution is given by the Kähler-Einstein volume form. An interpretation of the energy $E$ in terms of *effective bosonization* was also given.
Heuristically, it seems that one could interpret the thermodynamical limit above as saying that the Kähler-Einstein metric emerges from a *fermionic maximum entropy principle*: the particles try to maximize their entropy (i.e. volume in configuration space) under the constraint that they behave as fermions and hence, according to Pauli’s exclusion principle, cannot occupy the same space leading to an effective repulsion.
By making the complex structure $J$ dynamical relations to a variant of quantum gravity were explained using a canonical finite $N$ approximation of the functional measure. It would be interesting to compare with other discretization schemes such as spin foam loop quantum gravity or the approach of CDT (causal dynamical triangulations); see [@n-p; @lo; @a---] and references therein. However, in the present case one only consider fluctuating Riemannian metrics which are Kähler with respect to some complex structure on $X$ (imposing a constraint unless $n=1).$
The precise mathematical details of the thermodynamical convergence will be investigated elsewhere, as well as the case when the anti-canonica line bundle $K_{X}^{-1}$ is positive (so that any Kähler-Einstein metric must have positive Ricci curvature). In the latter case there are well-known obstructions to the existence of Kähler-Einstein metrics and the Yau-Tian-Donaldson program aims at showing that all obstruction may be formulated in terms of algebro-geometric stability. More precisely, according to the the Yau-Tian-Donaldson conjecture the existence of Kähler-Einstein is equivalent $X$ beeing $K-$stable (or to some refined notion of $K-$stability) [@p-s]. From the point of view of the present paper this is related to the fact that the natural candidate for the $N-$particle ensemble in the case when $K_{X}^{-1}$ is positive may be formally written as $$\mu^{(N)}=(\Psi_{1}(x_{1})\wedge\bar{\Psi}_{1}(x_{1})\wedge\cdots\wedge\Psi_{N}(x_{N})\wedge\bar{\Psi}_{N}(x_{N}))^{-1/k}/\mathcal{Z}_{N}\label{eq:prob measure for kx neg}$$ where now $(\Psi_{I})$ is a base in $H^{0}(X,(K_{X}^{-1})^{\otimes k}).$ However, because of the negative exponent above this singular volume form on $X^{N}$ may be non-integrable, i.e. $\mathcal{Z}_{N}=\infty,$ due to singularities appearing when to points merge. This means that the correponding Boltszmann-Gibbs measure $\mu^{(N_{k})}$ at level $k$ may not even exist. In fact, its existence should be closely relatated to the existence of balanced metrics, which in turn is closely related to the notion of $K-$stability (compare the disucssion in section \[sub:Duality-and-relation\]). Anyway, fixing an auxiliary smooth Hermitian metric $h=e^{-\Phi_{0}}$ on $K_{X}^{-1}$ one can look at the $\beta-$ensemble $$\mu_{\beta}^{(N)}=\left|\Psi_{1}\wedge\bar{\Psi}_{1}\wedge\cdots\wedge\Psi_{N}\right|^{\beta/k}((e^{\Phi})^{\otimes N}){}^{(-2-\beta)}/\mathcal{Z}_{\beta,N}\label{eq:renormalized prob meas}$$ (with $\beta=-\gamma$ *negative*) which is integrable for $\gamma$ sufficiently small, coinciding with \[eq:prob measure for kx neg\] for $\beta=2.$ In fact, it can be shown that the measure is integrable as long as $\gamma<\alpha_{X},$ where $\alpha_{X}$ is Tian’s *$\alpha-$invariant* [@t] (also called the log canonical threshold in algebraic geometry ). Repeating the same argument as before the rate functional in \[eq:rasympt with rate fun\] may be written as $$\beta F_{\beta}(\mu)=\beta E(\mu)-S(\mu)\label{eq:free energy times beta in neg}$$ In the case of main interest, i.e. **$\beta=-2,$** this functional may be identified with Mabuchi’s $K-$energy functional in Kähler geometry. Since now $\beta<0$ this means that the energy contribution $\beta E(\mu)$ corresponds to an *attractive* force. Hence, minimizimgs $\beta F_{\beta}(\mu)$ amounts to finding a balance between *minimizing energy* (which brings the particles close together) and *maximizing entropy* (which rather has a repelling effect). Accordingly, if $\beta$ is too small then the attractive effect wins and the rate funcional \[eq:free energy times beta in neg\] becomes unbounded from below. Interestingly, it can be shown that when $X$ admits a Kähler-Einstein metrics with volume form $\mu_{KE}$ and holomorphic vector fields (so that the metric is non-unique) then the rate functional \[eq:free energy times beta in neg\] is bounded from below for $\beta\geq-2$ (with minimizer $\mu_{KE}$ for $\beta=-2)$ and it is unbounded from below when $\beta<-2.$ In fact, the notion of *analytic* $K-$stability which by Tian’s theorem is equivalent to the existence of a unique Kähler-Einstein metric [@t] may be equivalently formulated as the condition that $\beta F_{\beta}$ be bounded from below for all $\beta$ sufficently close to $-2$ (see [@berm6] and references therein). As suggested above it seems natural to expect that the ordinary notion of (non-analytic) $K-$stability is closely related to integrability of the measure \[eq:prob measure for kx neg\] for $N$ sufficently large. In turn, by Gibbs principle this latter property is equivalent to the boundedness from below of the functional $\beta F^{(N)},$ where $F^{(N)}$ is the corresponding $N-$particle Gibbs free energy (formula \[eq:N particle free en\]).
If $(X,J)$ admits a Kähler-Einstein metric (or if appropriate stability conditions hold) it seems natural to expect the one-point correlation measure of $\mu_{\beta}^{(N)}$ to converge to a Kähler-Einstein volume form when first $N\rightarrow\infty$ and then $\beta\rightarrow-2.$ The simplest case appears when $X$ is the Riemann sphere. Then the corresponding ensemble is explicitly given by a one component plasma (or equivalently a point vortex system) studied by Kiessling in [@ki2] where a first order phase transition appears at $\beta=-2.$ Kiessling also considered higherdimensional generalizations in other directions than the one explored here, namely to the conformal geometry of higher dimensional spheres where the correponding Hamiltonian $H^{(N)}$ is a sum of logarithmic pair interaction. As a consequence the corresponding mean field equations are quasi-linear (with the non-linearity coming from the exponential term), as opposed to the present setting where the fully non-linear Monge-Ampère operator appears.
There is also an interesting variant of the probability measure $\mu_{\beta}^{(N)}$ above where one, in the definition \[eq:renormalized prob meas\], instead uses a *singular* metric on $K_{X}^{-1}$ whose curvature is the current $\delta_{D}$ defined by a given Calabi-Yau submanifold $D$ of dimenion $n-1$ in $X$ (such $D$ usually exist); more precisely, $D$ is cut out by a holomorphic section $s$ of $K_{X}^{-1}$ and one takes $\Phi=\log|s|^{2}$ in \[eq:renormalized prob meas\]. Then the minimizers $\mu_{\gamma}(=\omega_{\gamma}^{n}/n)$ of the corresponding scaled free energy functional $\beta F_{\beta,D}(\mu)$ (now writing $\beta=-2\gamma$ for $0<\gamma<1)$ turn out to satisfy an equation very recently introduced by Donaldson [@do-2] (see also [@do-3]): $$\mbox{Ric}\omega_{\gamma}=\gamma\omega_{\gamma}+(1-\gamma)\delta_{D}\label{eq:donaldsons equ}$$ saying that $\omega_{\gamma}$ is a Kähler-Einstein metric on $X-D$ whose Ricci curvature is singular along the submanifold (divisor) $D.$ The equationa above appeared in a program proposed by Donaldson to attack the Yau-Tian-Donaldson conjecture where the first step consists in proving the existence of solutions to the previous equation for** **$\gamma<<1$ and the last step amounts to, assumping $K-$stabilility (or rather a variant like $\bar{K}-$stability), use certain balanced metrics to take the limit $\gamma\rightarrow1$ (corresponding to $\beta\rightarrow-2$ in the present notation) to obtain a bona fida Kähler-Einstein metric on $X$.[^2]
In the light of Donaldson’s program it seems natural to formulate the following conjecture: if $X$ is $K-$stable (or say $\bar{K}-$stable in Donaldson’s sense) and $X$ contains a submanifold $D$ as above, then the corresponding probability measures $\mu_{\beta,D}^{(N)}$ are well-defined for all $\beta<-2$ and the corresponding one-point correlation measures converge to a Kähler-Einstein volume form when first $N\rightarrow\infty$ and then $\beta\rightarrow-2.$
Finally, it would also be interesting to detail the Glauber (Langevin) stochastic dynamics alluded to section \[sub:Duality-and-relation\] and investigate a suitable “hydrodynamical” scaling limit (see for example the recent paper [@d-o-r] for relations between Langevin dynamics and field theories with holomorphic factorization and supersymmetry). This should lead to a deterministic heat-equation type flow on the space of all (smooth) probability measures $\mathcal{M}_{1}(X),$ converging towards the Kähler-Einstein volume form. In fact, a “dual” (in the sense of the previous section) scaling limit of Donaldson’s iteration of the map $T_{k,\beta}$ \[eq:donaldsons t map\] was shown to converge to the Kähler-Ricci flow in [@berm3-1].
[67]{} Aharony, O; Gubser, SS; Maldacena, J; Ooguri, H; Oz, Y: Large N field theories, string theory and gravity. Physics Reports, 2000 - Elsevier
Aliev, A N; Nutku, N: Gravitational instantons admit hyper-Kähler structure. Class. Quantum Grav. 16 (1999) 189210.
Alvarez-Gaumé, L; Bost, J-B; Moore, G; Nelson, P; Vafa, C: Bosonization on higher genus Riemann surfaces. Comm. Math. Phys. 112 (1987), no. 3,
Ambjorn, J; Goerlich, A: Jurkiewicz, J; R. Loll: CDT—an Entropic Theory of Quantum Gravity. arXiv:1007.2560
Andersson, M T: A survey of Einstein metrics on 4-manifolds, in: Handbook of Geometric Analysis, in Honor of S.-T. Yau, Ed. Lizhen Ji, R. Schoen et al, International Press, Boston, (2008).
Aubin, T.: Equations du type Monge-Amp‘ere sur les varietes kahleriennes compactes, Bull. Sci. Math. 102 :1 (1978) 63-95
Bando, S.; Mabuchi, T: Uniqueness of Einstein Kahler metrics modulo connected group actions, in Algebraic geometry, Sendai, 1985 (T. Oda, Ed.), Adv. Stud. Pure Math. 10, Kinokuniya, 1987, 11-40.
Bardeen, J.M; Carter, B; Hawking, S. W: The four laws of black hole mechanics. Comm. Math. Phys. Volume 31, Number 2 (1973), 161-170.
Ben Arous, G.; Zeitouni, O. Large deviations from the circular law. ESAIM Probab. Statist. 2 (1998), 123-134
Berline, N; Getzler, E; Vergne, M: Heat kernels and Dirac operators. Corrected reprint of the 1992 original. Grundlehren Text Editions. Springer-Verlag, Berlin, 2004.
Berman, R:J: determinantal point processes and fermions on complex manifolds: large deviations and bosonization. Arxiv :0812.4224
Berman, R:J: Relative Kahler-Ricci flows and their quantization. arXiv:1002.3717
Berman, R:J: Analytic torsion, vortices and positive Ricci curvature. arXiv:1006.2988
Berman, R:J: A thermodynamical formalism for Monge-Ampere equations, Moser-Trudinger inequalities and Kahler-Einstein metrics. arXiv:1011.3976
Berman, R.J.; Boucksom, S: Growth of balls of holomorphic sections and energy at equilibrium. Inventiones Mathematicae Vol. 181, Issue 2 (2010), p. 337
Berman, R.J.; Boucksom, S; Witt Nyström, D: Fekete points and convergence towards equilibrium measures on complex manifolds. Acta Mathematica. To appear (arXiv:0907.2820).
Berman, R.J; Boucksom, S; Guedj,V; Zeriahi: A variational approach to complex Monge-Ampere equations. http://arxiv.org/abs/0907.4490
Boucksom, S; Essidieux,P: Guedj,V; Zeriahi: Monge-Ampere equations in big cohomology classes. Preprint.
Caglioti.E; Lions, P-L; Marchioro.C; Pulvirenti.M: A special class of stationary flows for two-dimensional Euler equations: a statistical mechanics description. Communications in Mathematical Physics (1992) Volume 143, Number 3, 501-525
Cheltsov, I; Shramov, C: Log canonical thresholds of smooth Fano threefolds. With an appendix by Jean-Pierre Demailly
Deift, P. A. Orthogonal polynomials and random matrices: a Riemann-Hilbert approach. Courant Lecture Notes in Mathematics, 3. New York University, Courant Institute of Mathematical Sciences, New York; American Mathematical Society, Providence, RI, 1999.
Demailly, J-P: Complex analytic and algebraic geometry. Available at www-fourier.ujf-grenoble.fr/\~demailly/books.html
Dijkgraaf, R; Orlando, D; Reffert, S: Relating Field Theories via Stochastic Quantization. Nucl.Phys.B824:365-386,2010
Douglas, M.R; Klevtsov, S: Bergman Kernel from Path Integral. Communications in Mathematical Physics, Vol: 293, Issue: 1, January 2010, Pages: 205-230
Douglas, M.R; Klevtsov, S: Black holes and balanced metrics. arXiv:0811.0367
Donaldson, S. K. Scalar curvature and projective embeddings. I. J. Differential Geom. 59 (2001), no. 3, 479–522.
Donaldson, S. K. Scalar curvature and projective embeddings. II. Q. J. Math. 56 (2005), no. 3, 345–356.
Donaldson, , S. K. : Some numerical results in complex differential geometry. arXiv:math/0512625
Donaldson, S.K.: Discussion of the Kahler-Einstein problem. Notes available at http://www2.imperial.ac.uk/\~skdona/KENOTES.PDF
Donaldson, S.K.: Kahler metrics with cone singularities along a divisor. arXiv:1102.1196
Donaldson, , S. K.: Remarks on gauge theory, complex geometry and \$4\$4-manifold topology. Fields Medallists’ lectures, 384403, World Sci. Ser. 20th Century Math., 5, World Sci. Publ., River Edge, NJ, 1997.
Ellis, R: Entropy, large deviations, and statistical mechanics. Springer, 2005
Fujiki, A: The moduli spaces and Kähler metrics of polarized algebraic varieties. (Japanese) Sūgaku 42 (1990), no. 3, 231243. (Reviewer: Akira Asada),
Fujiki, A; Schumacher, G: The moduli space of extremal compact Kähler manifolds and generalized Weil-Petersson metrics. Publ. Res. Inst. Math. Sci. 26 (1990), no. 1, 101–183.
Gawedzki: K. Lectures on conformal field theory, available at http://www.math.ias.edu/QFT/faIV, or see pp. (727-805) in P. Deligne, D. Freed, (eds.), Classical Field Theory. Quantum Fields and Strings: a Course for Mathematicians, Vol. I, 2 (Princeton, NJ, 1996/1997) Providence, American Mathematical Society 1999.
Griffiths, P; Harris, J: Principles of algebraic geometry. Wiley Classics Library. John Wiley & Sons, Inc., New York, 1994.
Hawking, S.W.: Spacetime foam. Nuclear Physics B Volume 144, Issues 2-3, 13 November-20 November 1978, Pages 349-362
’t Hooft, G: Probing the small distance structure of canonical quantum gravity using the conformal group. arXiv:1009.0669
Hough, J. B.; Krishnapur, M.; Peres, Y.l; Virág, B: Determinantal processes and independence. Probab. Surv. 3 (2006), 206–229
Jacobson, T: Thermodynamics of Spacetime: The Einstein Equation of State. Phys. Rev. Lett. 75, 12601263 (1995)
Jeffres, T; Mazzeo, R; Rubinstein, Y.A:; Kähler-Einstein metrics with edge singularities. Arxiv 1105.5216
Johansson, K: On fluctuations of eigenvalues of random Hermitian matrices. Duke Math. J. 91 (1998), no. 1, 151–204.
Kiessling M.K.H.: Statistical mechanics of classical particles with logarithmic interactions, Comm. Pure Appl. Math. 46 (1993), 27-56.
Kiessling M.K.H.: Statistical mechanics approach to some problems in conformal geometry. Physica A: Statistical Mechanics and its Applications Volume 279, Issues 1-4, 1 May 2000, Pages 353-368
Laughlin, R.B.: Anomalous Quantum Hall Effect: An Incompressible Quantum Fluid with Fractionally Charged Excitations. Phys. Rev. Lett. 50, 13951398 (1983)
LeBrun, C: Einstein Metrics and Mostow Rigidity, Math. Res. Lett. 2 (1995) 1–8
LeBrun, C: Einstein Metrics on Complex Surfaces, in Geometry and Physics, Anderson et al. eds. pp. 167–176, Marcel Dekker, 1997.
Loll, R: Discrete approaches to quantum gravity in four dimensions. arXiv:gr-qc/9805049
Macchi, O: The coincidence approach to stochastic point processes. Advances in Appl. Probability 7 (1975), 83–122.
Nicolai, H; Peeters, K: Loop and spin foam quantum gravity: a brief guide for beginners. Lect.NotesPhys.721:151-184,2007 (arXiv:hep-th/0601129)
Onofri1, E; Virasoro, M.A: On a formulation of Polyakov’s string theory with regular classical solutions. Nuclear Physics B Vol. 201, Issue 1, 14 June 1982, Pages 159-175
Phong, D.H; Sturm, J: Lectures on Stability and Constant Scalar Curvature. arXiv:0801.4179
Scardicchio, A; Torquato, S; Zachary, C.E: Point processes in arbitrary dimension from fermionic gases, random matrix theory, and number theory . arXiv:0809.0449
Scardicchio, A; Torquato, S; Zachary, C.E: Statistical properties of determinantal point processes in high-dimensional Euclidean spaces. arXiv:0810.4977
Schumacher, G: Positivity of relative canonical bundles for families of canonically polarized manifolds, arXiv:0808.3259. Curvature of higher direct images and applications. arXiv:1002.4858
Tian, G; Yau S-T: Kähler-Einstein metrics on complex surfaces with $C_{1}>0:$ Comm. Math. Phys. Volume 112, Number 1 (1987), 175-203.
Tian, G; On Calabi’s conjecture for complex surfaces with positive first Chern class. Invent. Math. Vol. 101, Nr. 1 (1990)
Tian, G: Canonical metrics in Kähler geometry. Notes taken by Meike Akveld. Lectures in Mathematics ETH Zürich. Birkhäuser Verlag, Basel, 2000.
Tseytlin, A.A: Two-dimensional Kähler Einstein spaces and gravitational instantons. Physics Letters B Vol. 97, Issues 3-4, 15 December 1980, 391–395
Touchette, H: The large deviation approach to statistical mechanics. Physics Reports 478 (2009) 169
Verlinde, E: Verlinde, H: Chiral bosonization, determinants and the string partition function. Nuclear Physics B Volume 288, 1987, Pages 357-396
Verlinde, E: On the Origin of Gravity and the Laws of Newton. . arXiv:1001.0785
Wentworth, R.A.: Precise constants in bosonization formulas on Riemann surfaces II. arXiv:1008.2914
Yau, S.T: On the Ricci curvature of a compact Kähler manifold and the complex Monge-Ampère equation. I. Comm. Pure Appl. Math. 31 (1978), no. 3, 339–41
Zabrodin, A; Matrix models and growth processes: from viscous flows to the quantum Hall effect. Preprint in 2004 at arXiv.org/abs/hep-th/0411437
Zelditch, S: Book review of “Holomorphic Morse inequalities and Bergman kernels Journal” (by Xiaonan Ma and George Marinescu) in Bull. Amer. Math. Soc. 46 (2009), 349-361.
Zeitouni O; Zelditch, S: “Large deviations of empirical zero points measures on Riemann surfaces, I: $g=0.$ [\[]{}Preprint[\]]{}.
[^1]: to avoid technical difficulties one should really replace $DIFF(X)$ with its subgroup $DIFF_{0}(X)$ in the connected component of the identity and accordingly the corresponding moduli space with the *Teichmuller space* associated to $X$ but to simplify the discussion we will not make this distinction. For mathematical results about the relevant moduli spaces we refer the reader to [@fs; @sc].
[^2]: The existence of the metric in \[eq:donaldsons equ\] $\omega_{\gamma}$ for $\gamma<<1$ has now been proved in [@berm6] using a variational calculus inspired by (but independent of) the statistical mechanical model proposed in the present paper. Moreover, this result combined with the very recent paper [@jmr] shows that $\omega_{\gamma}$ in fact has *conical* singularies with angle $2\pi\gamma$ transversally to $D$ and is smooth in the directions of $D$ as also conjectured by Donaldson.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Internet of Things (IoT) networks are becoming a part of our daily lives and the number of IoT devices around us are surging. The authentication of millions of connected *things* and the distribution and management of shared keys between these devices pose challenging research problems. Current one-to-one authentication schemes do not take the resource limitations of IoT devices into consideration. Nor do they address the scalability problem of massive machine type communication (mMTC) networks. Group authentication schemes (GAS), on the other hand, have emerged as novel approaches for many-to-many authentication problems. They can be used to simultaneously authenticate many resource-constrained devices. However, existing group authentication schemes are not energy efficient, and they do not provide enough security for widespread use. In this paper, we propose a lightweight GAS that significantly reduces energy consumption on devices, providing almost 80% energy savings compared to the state-of-the-art solutions. Our approach is resistant to the common replay and man-in-the-middle attacks.. The proposed approach also includes a solution for key agreement and distribution in mMTC environments. Moreover, this approach can be used in both centralized and decentralized group authentication scenarios.'
author:
- |
Yucel Aydin, Gunes Karabulut Kurt, ,\
[Enver Ozdemir, ,]{} and [Halim Yanikomeroglu, ]{}
title: A Flexible and Lightweight Group Authentication Scheme
---
Group authentication, Internet of Things, massive machine type communication, secret sharing scheme.
Introduction
============
A *thing* in an Internet of Things (IoT) network can be defined as a physical or virtual node which connects to the Internet and has the ability to communicate with other nodes [@FGMAA; @KCVA]. Security and privacy are crucial points in the advancement of IoT networks. The IoT paradigm extends the capabilities of the Internet to mobile and sensor networks. Each node is connected to the network and is also capable of communicating with each other. The confidentiality and integrity of data and the authentication of nodes are the main security issues for IoT networks [@suo]. Authenticating each node remains a key challenge. Authentication is a process for ascertaining that an entity really is who it claims to be [@smith]. It is one of the most important processes in the access control chain, since all other security and data transmission operations follow after the authentication process.
There will be numerous connected nodes around us in the near future which are connected to us and/or to each other. Most of them have limited computing power and battery capacity. Therefore, the use of traditional cryptographic methods is not possible for the authentication process of resource-constrained IoT nodes. IoT networks typically have a three-layer design, comprised of a sensing layer, a network layer, and an application layer. The sensing layer consists of IoT nodes with various sensing capabilities. The network layer aids transmission of the sensed data to the servers. Typically, gateways are used as the devices that provide the connection between the application layer and sensing layer, along with routers and other packet forwarding devices. Traditional authentication methods can be used in gateways since these devices have relatively high computational power. On the other hand, lightweight authentication solutions are needed for the sensing layer nodes.
![Three-layer Internet of Things (IoT) architecture. The data is collected through the sensing layer and transferred over the network layer to servers and databases in the application layer. Users connect to the system over the cloud.[]{data-label="fig:iot"}](IoT-Architecure.png){width="\linewidth"}
In traditional authentication methods, the client and authentication server usually have the shared key before communication. A random value, which is selected and sent by the server to the client, is encrypted by the client with the shared key, and the encrypted value is sent back to the server. Finally, the server validates the client by decrypting the response. In this process, there is one claimer and one prover. The prover can only authenticate one user at a time. So this approach is not scalable for densely deployed IoT networks where millions of nodes are expected to be operational. Although the problem of providing connectivity to all these nodes is currently being addressed by 3GPP in Release 16 under the name massive machine-type communications (mMTC), the scalability of the authentication process remains to be addressed.
According to IMT-2020’s mMTC requirements, over 1 million nodes can operate in a single km$^2$ [@IMT]. Although the security issues of mMTC networks are visible and currently being studied by the research community [@mtc], to the best of the authors’ knowledge there are no standardization efforts targeting the scalability of device authentication in these networks [@3GPPSec]. Each mMTC node must perform individual authentication with an authentication server according to the current evolved packet system authentication protocol (EPS-AKA) in mobile networks [@mtc2]. This can cause high signaling overhead on the server. One of the bottlenecks in this scenario is that all IoT nodes can request authentication from the server at the same time, and the server cannot respond to all the requests. A fast authentication proposal is required to authenticate all users at the same time.
Group authentication is one of the best solutions to minimize the load on the authentication server. Millions of nodes in IoT can create groups according to their coverage area or functions in the system. Instead of sending all authentication requests to the server, authentication within the group will reduce the load of the server. The many-to-many group authentication idea was proposed by Harn in [@harn] and developed further by Chien in [@chien].
Existing group authentication approaches do not take the resource constraints of the network into account. However, sensing nodes in an IoT environment frequently have limited memory, tight energy constraints, and very limited processing capability. So during the authentication process, the communication overhead on the nodes should be as little as possible. Also, the energy consumption of group authentication algorithm should be as low as possible. For this reason, traditional cryptographic systems, along with existing group authentication methods, are not proper for IoT and lightweight systems must be proposed.
Group authentication in wireless communication environments is more vulnerable to attacks by unauthorized entities. Man-in-the-middle attacks can be performed by anyone who can capture group credentials. Hence, group authentication algorithms must provide security for attacks on wireless channel. Existing group authentication approaches remain vulnerable to such attacks.
Another challenge for IoT networks is the need for secure communication between nodes without any human intervention. For secure communication between millions of mMTC nodes, each node must have a shared key. In such a crowded environment, key distribution and management consume too much time and energy.
In light of these challenges, our main contributions are as listed:
- We propose a flexible lightweight authentication scheme to overcome the possible problems in group authentication and that can be used in centralized or decentralized group authentication scenarios. Group members share their private keys publicly in other group authentication schemes. If an intruder can eavesdrop on group traffic, the intruder can recover the group secret.
- Our proposed approach offers a group key for mMTC. At the end of the group authentication, each group member can recover the same group key for further communications. The mMTC nodes communicate with each other by symmetric key encryption once they have a group key. In terms of privacy and security, the private keys are not used for group authentication, which secures the privacy of group members and prevents intruders from capturing private keys and performing man-in-the-middle, impersonation or replay attack.
- Lightweight schemes are vital for IoT scenarios due to the resource constrained nodes. When we compare our proposal with other group authentication schemes, the energy consumption of one node can be reduced by 80%. Additionally, the energy consumption remains constant even if the number of group members increases.
This paper is organized as follows. The next section provides an overview of security requirements for IoT and an overview of existing group authentication methods. In Section III, our proposed approach for group authentication is presented. The security and performance evaluation is provided in Section IV and Section V, respectively. Conclusions are presented in Section VI. The study is completed by an overview of future work suggestions in Section VII.
Related Background and Literature Overview
==========================================
There are three layers in IoT architecture as shown in Figure \[fig:iot\]. As the first layer, the sensing layer provides information from the field to the upper layers. The components in this layer include sensors of diverse applications or radio frequency identification (RFID) tags and readers [@suo]. The network layer is the mediator between the sensing and application layers and is responsible for the secure transmission of information from the former to the latter. This network layer mostly relies on traditional networks such as the Internet, mobile communication networks, satellite networks, or wireless networks [@suo]. The application layer, for its part, provides services to users. Each layer must have different security requirements according to the capabilities of the layer components. Our proposed approach focuses mostly on the sensing layer due to the computational restrictions on IoT nodes. We propose a lightweight scheme to provide authentication and key agreement for nodes in the sensing layer. Our proposed approach is based on the group authentication technique used to combat scalability problems that arise as the number of components in IoT networks increases.
Studies on group authentication have mostly focused on secret sharing schemes. In general, a group key is divided into a number of shares via a secret sharing algorithm that distributes private keys among users. Users share their private keys with each other, and each user can recover the group key after by using their private keys up to a threshold value. Finally, group users can communicate with each other securely by a symmetric key encryption.
Harn proposed a group authentication method based on Shamir’s secret sharing scheme (SSS) [@harn]. According to Shamir [@shamir], a secret value $D$ is divided into $n$ pieces and the secret can be recovered by having $k$ pieces. However, even if one owns $k-1$ shares, the secret can not be recovered. The scheme is called ($k$,$n$) threshold scheme. Drawing on SSS, Harn proposed three group authentication schemes: one basic scheme and two asynchronous schemes. In the basic scheme, the group manager (GM), the most powerful member of the group with the most resource capabilities, selects a polynomial $f(x)$ of degree $k-1$, $f(x)=a_0+a_1x+a_2x^2+\ldots+a_{k-1}x^{k-1}$. The first coefficient $a_0$ is the group key. The GM computes one private key $f(x_i)$ and one public key $x_i$ for each user $U_i$. Then, the GM shares private keys with related users. In the group authentication phase, each user shares their private key with other users. After having $t\ge k$ private keys, each user can compute the group key by the Lagrange interpolating formula: $$a_0=\sum^{t}_{i=1}f(x_i)\prod^{t}_{r=1, r\neq i}\dfrac{-x_r}{x_i-x_r}.$$ The basic algorithm is secure if the users share their private keys simultaneously. Otherwise any attacker can capture $k$ private keys and recover the polynomial. Then, the attacker also can share a valid private key. Due to the vulnerabilities of the basic scheme, Harn proposed two asynchronous group authentication schemes.
The first of the asynchronous authentication schemes, given in [@harn], is for one-time authentication. In the preparation phase, the GM selects $t$ values such that $kt>n-1$ and selects random polynomials, $f_{l}(x), l=1,2,\ldots,t,$ having degree $k-1$. The GM generates tokens, $f_l(x_i), l=1,2,\ldots,t,$ for each group member $U_i$. The GM then finds the integers $w_j,d_j,j=1,2,\ldots,t,$ such that the secret key, $s$, is $$s=\sum^{t}_{j=1}d_jf(w_j)$$ for every pair of $i$ and $j$. The GM makes these integers $w_j,d_j,j=1,2,\ldots,k,$ and $H(s)$ publicly known where $H(\cdot)$ is a cryptographic hash function. In the group authentication phase, each user $U_i$ uses its tokens and computes $$c_i=\sum^{k}_{j=1}d_jf_j(x_i)\prod^{m}_{r=1, r\neq i}\dfrac{w_j-x_r}{x_i-x_r},$$ and shares the computed value with other group members. After having, $c_r,r=1,2,\ldots,m,$ each group member computes $$s'=\sum^{m}_{r=1}c_r.$$ If $H(s')$ is equal to $H(s)$, all users will have been authenticated. Outside attackers cannot derive any private token from the released $c_i$ value. Therefore, the approach provides security for asynchronous group authentication. However, it is a one-time authentication since the secret is no longer a secret once it has been recovered [@harn].
The second asynchronous authentication scheme was designed for multiple authentications. In this case, the GM selects two large prime numbers, $p$ and $q$, such that $q$ divides $p-1$, and selects $g_i$ which is a generator of $GF(q)$. Next, the GM selects two polynomials, $f_l(x), l=1,2,$ having $t-1$ degree, and generates tokens $f_l(x_i), l=1,2,$ for each group member $U_i$. There are multiple secrets selected by the GM. For each secret, the GM then selects random integers, $w_{i,j},d_{i,j},j=1,2,$ and computes secrets $$s_i=g_i^{\sum^{2}_{j=1}d_{i,j}f_j(w_{i,j})}.$$ The GM makes these integers $w_{i,j}, d_{i,j}$ and $g_i,H(s_i)$ publicly known. In the group authentication phase, each user $U_i$ uses their tokens and computes $$c_i=\sum^{2}_{j=1}d_{i,j}f_j(x_i)\prod^{m}_{r=1, r\neq i}\dfrac{w_{i,j}-x_r}{x_i-x_r},$$ and then computes $e_i=g_i^{c_i}$ shares the computed $e_i$ value with other group members. After having, $e_i,i=1,2,\ldots,m,$ each group member computes $$s_i'=\prod^{m}_{i=1}e_i.$$ If $H(s_i')$ is equal to $H(s_i)$, all users will have been authenticated. According to the security analysis by Harn, it is infeasible to recover $c_i$ value from $e_i$ due to the discrete logarithm problem. In other words, private user tokens are protected from released values. Also, even if an attacker obtains the secret $s_i$, it is infeasible to recover $\sum^{2}_{j=1}d_{i,j}f_j(w_{i,j})$ values because of the discrete logarithm problem (DLP). To put it differently, private user tokens are protected [@harn].
Chien [@chien] has shown that the Harn asynchronous schemes have some critical weaknesses. If an attacker can get $k$ distinct values in $k$ different trials, the secret function chosen by the GM can be solved and all users’ secrets can be obtained. It is proved in [@chien] with simple examples that an attacker can recover group secret values and user private tokens. Chien introduced a new method based on SSS, elliptic curve cryptography (ECC) and pairing-based cryptography in order to ensure a secure group authentication process with multiple trials and multiple authentications. In this new method, according to Chien, the GM selects two additive groups $G_1$ and $G_2$, and one multiplicative group $G_3$ with order $q$. The GM makes a generator $P$ for $G_2$ public. A polynomial with degree $t-1$ is chosen. The constant term of the polynomial will be the master secret $s$. The value of $$Q=s\times P \text{ in } G_2$$ is then computed and shared publicly. For each user, one public key $x_i$ and one private key $f(x_i)$ are chosen and shared with the related users secretly. Users participating in the authentication phase agree on a random point $R_v$ in $G_1$. Then, each user computes $$c_i=f(x_i)\prod^{m}_{r=1, r\neq i}\dfrac{-x_r}{x_i-x_r},$$ and releases $c_i$ $R_v$. After all users release the $c_i\times$ $R_v$, each user computes $$\sum^{m}_{i=1}c_i\times R_v,$$ and verifies whether $$e\left(\sum^{m}_{i=1} c_i\times R_v,P\right){\stackrel{?}{=}} e(R_v,Q)$$ holds. In Chien’s approach, group members can authenticate each other, but they can not recover the group key for further communications. Extra steps are required in order to obtain the group secret value by each group member. According to the security analysis by Chien, the private tokens are protected due to the elliptic curve discrete logarithm problem (ECDLP). Accordingly, it is infeasible to obtain $c_i$ values from $c_iR_v$ as long as ECDLP is hard. In addition, for a valid $c_iR_v$ and another point $R_2$, one cannot compute $c_iR_2$ as long as the elliptic curve Diffie-Hellman problem (ECDHP) is hard. The last point in the security analysis for Chien’s proposed approach is a bilinear pairing inversion problem(BPIP). Given the values $e(R_v,Q)$ and $Q$ one cannot derive the value $\sum^{m}_{i=1} c_i\times R_v$ as long as BPIP is hard. If ECDLP, BPIP or ECDHP is not hard, any attacker can compute a valid value and participate in the group authentication.
In addition to the group authentication schemes described above, several studies have been proposed in order to reduce the computation and communication costs [@BSV; @AB; @MPP]. However, these approaches do not provide many-to-many authentication, i.e., the GM can not authenticate group members at the same time.
An authentication method is proposed in [@BSV] for MTC to authenticate user equipments (UE) through the gateway as quick as possible. The gateway communicates with UEs one by one to have the tokens of UEs. The gateway authenticates UEs as a group via a home subscriber server. One-to-one communication between UEs and gateway causes additional communicational cost.
Asmuth and Bloom proposed a key safeguarding scheme, which is based on the Chinese remainder theorem (CRT) in [@AB]. If anyone has shadows up to $r$, the secret value $y$ can be computed easily using CRT. But anyone who has $r-1$ shadows can not know the secret [@AB].
The authors proposed an algorithm using Paillier threshold cryptography in [@MPP]. They compared their results with Harn’s group authentication method and presented the results from their experiments. The results from [@MPP] show that their algorithm has a better computational time than Harn’s group authentication algorithm. But they did not take into account the computational cost of public and private key encryptions or scalability issues.
The Proposed Method
===================
Preliminaries
-------------
In the proposed algorithm, a group authentication process is performed first, after which group members recover group keys for further communication. Group authentication is crucial for the scalability of the system components and the resource usage of the group members. The group key agreement part of the algorithm is to provide machine-to-machine communication without the need of central authority.
We begin by introducing computationally hard problems, which are used by Harn and Chien, and compare them with our proposed approach.
*Definition 1.* The elliptic curve discrete logarithm problem is defined as [@chien]: given an elliptic curve over a finite field $F_p$ and two points $P,Q$ over an elliptic curve, to find a integer $k$ such that $Q=kP$ is ECDLP.
*Definition 2.* The computational elliptic curve Diffie-Hellman problem is [@chien]: given an elliptic curve over a finite field $F_p$, a point $P$ over an elliptic curve and points $A=aP$, $B=bP$, to find a point $C=abP$ is the computational ECDHP.
*Definition 3.* The bilinear pairing inversion problem is defined as [@chien]: Given $e(P,Q)$ and a point $Q$ to find the point $P$ is the BPIP.
A Use-Case Example: Authentication Scenarios of Cloud-Based IoT
---------------------------------------------------------------
Before giving the details of our proposed approach, we provide an example of a scenario where our scheme could be used. Cloud-based IoT technology consists of three components: the cloud network, IoT gateways and IoT sensors. Each IoT sensor is the sensing node of the system. It provides real-time information from the field. The gateway is the contact point of the sensing layer with traditional networks. It transmits the information to the cloud through the Internet or a mobile network.
![Group authentication use-case example. A cloud-based IoT network with different groups is shown. Each gateway acts as the group manager.[]{data-label="fig:usecase"}](GroupAuthentication.png){width="\linewidth"}
Authentication is one of the most crucial steps in cloud-based IoT technology. Each component must authenticate others before transmitting any information. The cloud must authenticate the gateways, and the gateways must authenticate the sensors as well. If the number of sensors is too high in an IoT network, a group authentication scheme is vital. As shown in Figure \[fig:usecase\], each gateway with its sensors creates a group. The critical point here is that the gateway has a higher computational capability than sensors. The gateway should be the group manager and perform group authentication before data transmission.
In a smart home or smart medical system, the sensors or machines can share data with each other. Therefore, a key agreement scheme is required to ensure the confidentiality of the data. Below, we propose a key agreement scheme after group authentication is completed. As illustrated by the scenario, a central authority is needed for group authentication. But if we assume that sensors will create a group without a gateway, there may not be a central authority. The appropriate selection of the group manager in this scenario becomes essential from the authentication perspective. For such an ad-hoc scenario, below we also propose group authentication without a group manager. Two options are noted at each step of our proposal. Hence, we provide a flexible and lightweight group authentication scheme that also scales with the number of increasing sensor nodes.
The Proposed Method
-------------------
The GM is assumed to be infrastructure-based and has relatively more computational power. In addition to the GM, each group has several other members with resource or computational constraints. Note that in IoT environments, the GM is basically the gateway with specific capabilities. Sensor nodes or RFID tags can be considered to be the other members in a group. The capabilities of these nodes are at a certain restricted rate.
The proposed method has two stages. The first stage involves the authentication process, which is based on ECC and SSS. This first stage consists of two phases, which are called the initialization and confirmation phases. The second stage, which is the key agreement stage, provides a solution to construct a master key for further communications. The details of each phase are presented at below.
[**The Initialization Phase**]{}
1. GM selects a cyclic group $G$ and a generator $P$ for $G$.
2. GM selects $E=Encryption(\cdot)$ and $D= Decryption(\cdot)$ algorithms and a hashing function $H(\cdot)$.
3. A polynomial with degree $t-1$ is chosen by GM and the constant term is determined as master key $s$.
4. GM selects one public key $x_i$ and one private key $f(x_i)$ for each user in the group $U$ where each user is denoted by $U_i$ for $i=1,\ldots,n$.
5. GM computes $Q=s\times P$.
6. GM makes $P, Q, E, D, H(s), H(\cdot), x_i$ public and shares $f(x_i)$ with only user $U_i$ for $i=1,\ldots,n$.
The confirmation phase is executed after the GM shares the values with the related users. There are two different options in the confirmation phase. One is that the GM is responsible for confirming the group members (*the centralized approach*). The other is that any member in the group is responsible for confirming the other members (*the decentralized approach*). The member selection can be done on the basis of the instantaneous resource availability of each node, such as their battery levels.
[Each member computes $f(x_i)\times P$, and shares $f(x_i)\times P \| ID_i$ with GM and other members]{}\
[**The Confirmation Phase**]{}
1. Each user computes $f(x_i)\times P$ and sends $f(x_i)\times P\ |ID_i$ to the GM and other users ($ID_i$ is the identification number of the user and $\|$ symbol shows the concatenation of two values).
2. If the GM verifies the authentication, the GM computes $f(x_i)\times P$ for each user and verifies whether the values are valid or not.
3. If the GM is not included in the verification process, any user in the group computes $$C_i=\left(\prod^{m}_{r=1, r\neq i}\dfrac{-x_r}{x_i-x_r})\right)f(x_i)\times P$$ for each user (m denotes the number of the users in the group and $m$ must be equal or larger than t).
4. User verifies whether $$\sum_{i=1}^{m}C_i {\stackrel{?}{=}} Q$$ holds.
5. If it holds, authentication is done. Otherwise, the process will be repeated from the initialization phase.
Both authentication by the GM and by any group member is given in the Algorithm 1. It is clear that group members should only compute one elliptic curve multiplication operation. The users should send their identification numbers by concatenating with public shares in order to avoid confusion for further communications. This is because these public shares will be used by other users in further communications and in the group key agreement stage, and all members should know which public share belongs to which user.
After the authentication has been performed, users will communicate with each other by using symmetric key encryption. Shared key for symmetric key encryption will be calculated by senders and receivers.
[$U_i$ computes $E_{(f(x_i)f(x_j)P)}[f(x_i)]$ for each $U_j$.]{}\
[Each user computes $D_{(f(x_j),f(x_i)P)}[f(x_i)]$.]{}\
[Each user computes $$s'=\sum_{i=1}^{m}f(x_i)\prod^{m}_{r=1, r\neq i}\dfrac{-x_r}{x_i-x_r}.$$ ]{}\
[Each user computes $H(s')$.]{}\
ECC is used in order to compute shared key between the group members. Let us set the key, $K$ as $$K= (y_iy_jP)$$ where $y_t=f(x_t)$, i.e., $y_t$ is the secret of the user $U_t$. The sender will use their own private key $(y_i)$ and the value sent by receiver $(y_jP)$. The receiver will obtain the same key with a similar operation, i.e., combining its own key $y_j$ with the received data $y_iP$.
After this stage, group members can communicate with each other by a symmetric key encryption method. However, instead of using different keys for each user, the master key that was selected by the GM can be used as the group key. The problem is how the users will recover the master key. We basically exploit SSS and a symmetric key encryption method to share the master key in the group key agreement stage.
[**The Group Key Agreement Stage**]{}
1. Each user shares their own secret key $f(x_i)$ with other users using symmetric key encryption.
2. Each user decrypts the values and obtains $m$ different $f(x_i)$.
3. Each user computes $$s'=\sum_{i=1}^{m}f(x_i)\prod^{m}_{r=1, r\neq i}\dfrac{-x_r}{x_i-x_r}.$$
4. Each user verifies whether $$H(s') {\stackrel{?}{=}} H(s)\text{ holds.}$$
At the end of the group key agreement stage, each member within the group will recover the master key as given in the Algorithm 2. After the group key agreement process, the members of the group will be able to communicate with each other using the master key. In addition, the GM can update $x_i$ and $f(x_i)$ values remotely using the master key in order to avoid replay attacks.
To sum up the foregoing, we propose a comprehensive solution for the authentication of users users belonging to the same group. A group authentication is accomplished with very low computational power on users in the first stage. A master key is recovered by all group users for a distributed environment in the second stage. The details of the security and performance analyses are given in the following parts.
Security Analysis
=================
In this section, we analyze possible attacks to the presented algorithms above. Our proposal provides security for most man-in-the-middle and replay attacks as shown below.
**Theorem 1:** *Group authentication cannot be performed without $t$ valid public and private values.*
*Proof.* Since the stated polynomial $f(x)$ is of degree $t-1$, it is necessary to know $t$ distinct pairs of ($x$,$f(x)$) for the formation of the polynomial again. $f(x)$ cannot be constructed by holding less than $t$ pairs. [$ \hspace{4.1cm} \qed$]{} **Theorem 2:** *The attacker who captures the value of $Q$ and $P$ sent by the group manager publicly cannot have knowledge of secret $s$.*
*Proof.* Given two points $P$ and $Q$ on an elliptic curve group, it is hard to find the $s$ value that provides the relationship $Q = s\times P$. This open problem is called ECDLP. Therefore, it is hard to find $s$ by having $Q$ and $P$. [$ \hspace{2.5cm} \qed$]{}
**Theorem 3:** *The attacker who captures the value of $f(x_i)P$ sent by the group members to the GM cannot have knowledge of $f(x_i)$.*
*Proof.* Due to the hardness assumption of ECDLP, it is hard to find $f(x_i)$ by having $f(x_i)P$. [$ \hspace{3.4cm} \qed$]{}
**Theorem 4:** *The attacker can capture $f(x_i)P$ and $f(x_j)P$ but can not obtain a valid symmetric key in order to establish a communication with user $U_i$.*
*Proof.* The attacker will need $f(x_j$) to compute $f(x_j)f(x_i)P$ but $f(x_j)$ is a secret known only by the user $U_j$. In other words, the attacker should be able to solve the computational ECDHP. [$ \hspace{4.8cm} \qed$]{}
**Theorem 5:** *The attacker can not perform replay and man-in-the-middle attack.*
*Proof.* An attacker can eavesdrop or cut the traffic between the GM and any user and capture $f(x_i)P$. After having $f(x_i)P$, the attacker can be part of the authentication process. Because the attacker does not have any valid $f(x_i)$, the attacker can not communicate with any other group member by using symmetric key encryption after the authentication process. [$ \hspace{7cm} \qed$]{}
Below we also list the vulnerabilities associated with the proposed approach.
**Vulnerability 1:** *The attacker can perform denial of service (DoS) attacks for the authentication process.*
*Proof.* An attacker can share a non-valid value when the members send their shares to the GM who will confirm the authentication. The GM can not compute the group secret value and repeat the process. An attacker can share the non-valid value again and perform denial of authentication. [$ \hspace{0.5cm} \qed$]{}
**Vulnerability 2:** *The node compromise attack can be performed.*
*Proof.* If the attacker could physically capture a group member then the attacker might obtain the secret key of the member. As a result of capturing the secret key, the attacker can generate a valid public key and share it with the GM in order to authenticate itself. If the attacker has a secret key, the attacker can also communicate with the other members of the group by producing symmetric keys. [$ \hspace{2.8cm} \qed$]{}
**Vulnerability 3:** *The group members can perform DoS attacks for confirmation point.*
*Proof.* If the group members send their shares to the confirmation point at the same time, this certain point can be locked. The solution for this kind of DoS attack is still a challenge in group authentication research. [$ \hspace{3.8cm} \qed$]{}
The joint design of group authentication with an authentication handoff between different groups is also crucial for an environment with millions of nodes. The groups exchange several nodes between each other. Repeating the group authentication process for each new node is resource and time consuming. One of our future works will cover the design of a handoff scheme in mMTC or cellular environments, as noted in Section VII.
Performance Analysis
====================
Group authentication is a novel method to increase the performance of the authentication system and to decrease the computational load on the group members. Additionally, the number of communications between a GM and group members is kept to a minimum in in group authentication.
The first parameter we used to compare the performance was the communication overhead. In battery constrained wireless environments, the communication between users should be as minimal as possible. In both Harn’s and Chien’s proposals, the users send a computed value to $m-1$ users ($m$ denote the number of users in the group). If we assume the cost of a communication between two group users is $\sigma$, $(m-1)\sigma$ communications are performed by one group member in both of the proposals in order to complete the group authentication process. In our framework, one user sends the computed value to $m-1$ users as shown in Table \[table:1\].
Harn [@harn] Chien [@chien] Proposal
--------------- ---------------- --------------- --
$(m-1)\sigma$ $(m-1)\sigma$ $(m-1)\sigma$
: Communication Overhead of One User[]{data-label="table:1"}
The second parameter we used to compare the performance was the energy used by one user for group authentication. We used the time complexity approach in [@chien] to compare the computational complexity of our algorithm with the schemes of Harn and Chien’s. In our algorithm, as well as in those of Harn and Chien, the GM is responsible for initializing the authentication. We only took into consideration the computations that were made by group members which were resource-restricted. While each user in Chien’s algorithm should compute $(7m+6785)T_{mul,q}$ [@chien], each user in Harn asynchronous multiple authentication scheme should compute $(45m+1418) T_{mul,q}$ [@chien]. $T_{mul,q}$ is the time for one multiplication in field $\mathbb F_q$ where $q$ is 160 bits, $m$ is the number of user in the group.
In our proposed approach, the group members should only compute one elliptic curve point multiplication ($T_{EM}$). According to Chien [@chien], $T_{EM}$ is roughly equal to $29 T_{mul,p}$ ($T_{mul,p}$ denotes the time for one multiplication in field $p$ where $p$ is 1024 bits). The security of ECC with 160-bit key is roughly equivalent to that of RSA with 1024-bit key or D-H algorithm with 1024-bit key. Therefore; $ T_{mul,p}$ is roughly equal to $ 41 T_{mul,q}$ [@chien]. In our authentication algorithm, group members compute $ 29 T_{mul,p}$, which is $1189 (29\times 41) T_{mul,q}$.
Confirmation of the authentication process was done by group members in Chien and Harn schemes. However in our scheme, the GM or only one user is responsible for the confirmation part of the authentication. As shown in Figure \[fig:M1\], our proposed approach is scalable with the number of group members.
+\[mark=\*,mark repeat=2\] [x\*7+6785]{}; +\[mark=\*,mark repeat=2\] [x\*45+1418]{}; +\[mark=\*,mark repeat=2\] [x\*0+1189]{}; +\[mark=nomark\] [x\*0+0]{};
If we assume $T_{mul,q}$ is equal to 1 second and $P$, which is the maximum power that an IoT node can consume, is equal to 1 Watt, the energy consumption of one user in a group with 100 members is shown in Table \[table:2\].
Harn [@harn] Chien [@chien] Proposal
-------------- ---------------- ------------- --
7485 joules 5918 joules 1189 joules
: Energy Consumption of One User[]{data-label="table:2"}
Conclusion
==========
This study has proposed a novel method for authentication of group communication in mMTC networks. Many-to-many authentication is used for group authentication by several studies but resource-constrained nodes were forced to compute more than their capacity. In the proposed method, group members should only compute one elliptic curve point multiplication. Secret sharing scheme and ECC are used on the basis of the proposed algorithms. Most of the resource-consuming work is done by the group manager or one of the group members, not by all the group members.
Both centralized and decentralized scenarios are supported by the proposed approach. In the centralized group authentication method, most of the resource consuming work is assigned to the group manager. Thus, energy-constrained group members do not need to consume a high amount of energy. In the decentralized group authentication method, only one of the group members, possibly with more computational capacity or more battery capacity than others, consumes more energy. The rest of the group members have a limited computational load, enabling them to function in an energy-efficient manner.
In existing group authentication approaches, attack possibilities still exist. To the best of the authors’ knowledge, there are no proposals against replay, node compromise or DoS attacks under the framework of secret sharing schemes. Our proposal provides provides security against replay attacks if the GMs update the credentials for each authentication.
Open Research Problems
======================
Each group in the group authentication scheme is coordinated by the group manager. This architecture resembles the cellular model, where the communication within each cell is monitored and enabled through the base station. As nodes can be mobile in some IoT networks, defining a handoff process for mobile devices to be authenticated enables fast and reliable communication, possibly with a low latency. Hence, the joint design of group authentication with the authentication handoff is of critical importance to accommodate a large number of users that may also be mobile. Instead of repeating the group authentication process from the start, the new node must be authenticated by using a fast handoff scheme. For future works, such handoff schemes must be designed to support mobility in mMTC environments.
Another research problem lies in the configuration of the base stations. Current networks make use of fixed base stations to enable authenticated communication between mobile users. However, in next-generation (6G) wireless networks, an ultra-agile radio access architecture with mobile base stations (including aerial base stations) is envisioned as a solution for coverage and/or unexpected congestion problems [@bor]. One of the deployment challenges for such dynamic cell structures will be from the authentication perspective. Although current networks do not need the authentication of the base station, in the presence of a mobile aerial base station, these intermediate devices also need to be authenticated to avoid man-in-the-middle attacks. A fast authentication scheme will be required to provide authentication between mobile ground users and aerial base stations and/or between the aerial base station and the terrestrial base stations. Moreover, mutual authentication of all active aerial base stations will also be required. Furthermore, the handoff processes all of the aforementioned connections will need to be carefully designed. We believe that the proposed group authentication technique can serve as a foundation to these open research problems.
[1]{} Al-Fuqaha, A., Guizani, M., Mohammadi, M., Aledhari, M., Ayyash, M, “Internet of Things: A survey on enabling technologies, protocols, and applications”, *IEEE Comm. Surveys and Tutorials*, vol. 17 no. 4, pp. 2347-2376, June 2015. Karagiannis, V., Chatzimisios, P., Vazquez-Gallego, F., Alonso-Zarate, J., “A survey on application layer protocols for the Internet of Things”, *Trans. on IoT and Cloud Computing*, vol. 3 no. 1, pp. 11-17, January 2015. Suo, H., Wan, J., Zou, C., Liu, J., “Security in the Internet of Things: a review”, *Int. Conf. on Computer Science and Electronics Eng.*, pp. 648-651, April 2012. Smith, R. E., *Authentication: From Passwords to Public Keys*, Addison-Wesley Longman, December 2001. ITU-R M.2410, “Minimum requirements related to technical performance for IMT-2020 radio interface(s)”, Nov. 2017. Li, J., Zhang, Y., Chen, J., Li, H., Zhang, W., “Group key agreement in multimedia service for machine type communication”, *Asia-Pacific Services Computing Conf.*, pp. 141-146, December 2014. 3GPP TS 33.501, “Security architecture and procedures for 5G system”. Lai, C., Li, H., Lu, R., Jiang, R., Shen, X., “LGTH: A lightweight group authentication protocol for machine-type communication in LTE networks”, *IEEE GLOBECOM*, pp. 832-837, December 2013. Harn, L., “Group authentication”, *IEEE Trans. on Computers*, vol. 62, no. 9, pp. 1893-1898, September 2013. Chien, H. Y., “Group authentication with multiple trials and multiple authentications”, *Security and Comm. Networks*, vol. 2017, pp. 1-7, May 2017. Shamir, A., “How to share a secret”, *Communications of the ACM*, vol. 22, no. 11, pp. 612-613, November 1979. Broustis, I., Sundaram, G. S., Viswanathan, H., “Group authentication: A new paradigm for emerging applications", *Bell Labs Tech. Journal*, vol. 17, no. 3, pp. 157-173, December 2012. Asmuth, C., Bloom, J., “A modular approach to key safeguarding", *IEEE Trans. on Info. Theory*, vol. 29, no. 2, pp. 208-210, March 1983. Mahalle, P. N., Prasad, N. R., Prasad, R., “Threshold cryptography-based group authentication (TCGA) scheme for the Internet of Things (IoT)", *IEEE Int. Conf. on Wireless Comm., Vehicular Tech., Info. Theory and Aerospace and Electronic Systems (VITAE)*, pp. 1-5, May 2014. Bor-Yaliniz, I., Yanikomeroglu, H., “The new frontier in RAN heterogeneity: Multi-tier drone-cells", *IEEE Comm. Magazine*, vol. 54, no. 11, pp. 48-55, Nov. 2016.
| {
"pile_set_name": "ArXiv"
} |
---
author:
- Dong Gao
- 'Yan-Chun Liang'
- 'Shun-Fang Liu'
- 'Guo-Hu Zhong'
- 'Xiao-Yan Chen'
- 'Yan-Bin Yang'
- Francois Hammer
- 'Guo-Chao Yang'
- 'Li-Cai Deng'
- 'Jing-Yao Hu'
title: 'Multiwavelength study of nearly face-on low surface brightness disk galaxies '
---
Introduction {#sect:intro}
============
Low Surface Brightness Galaxies (LSBGs) were often hard to find owing to their faintness compared with the night sky, thus their properties were seldom studied and their contribution to the galaxy population were underestimated for a long time. However, it has been suggested that LSBGs may comprise up to 1/2 of the local galaxy population (McGaugh, Bothun & Schombert 1995).
An initial quantitative study about LSBGs was done by Freeman (1970), who noticed that the central surface brightness of their 28 out of 36 disc galaxies fell within a rather narrow range, $\mu_{0}$(B)=21.65 $\pm$ 0.3 mag arcsec$^{-2}$. This could be caused by selection effects (Disney 1976; Zwicky 1957). Since then, many efforts have been made to search for more LSBGs from surveys (see the reviews by Bothun, Impey, McGaugh 1997; Impey & Bothun 1997; and the introduction part in Zhong et al. 2008). These include giant LSBGs (Sprayberry et al. 1995), red LSBGs (O’Neil et al. 1997) and the most common cases of “blue LSBGs" as late-type, disk-dominated spirals with $\mu_{0}$(B)$>$ 22 mag arcsec$^{-2}$ (Zhong et al. 2008 and the references therein). The LSBGs were generally thought to be unevolved systems with low metallicity (McGaugh 1994), low star formation rate (van der Hulst et al. 1993), a relatively high gas fraction (de Blok et al. 1996) and large amounts of dark matter (de Blok & McGaugh 1997). The age of LSBG is also an important topic to study.
The ages of the LSBGs have long been a matter of controversy. Broadband photometric studies, complemented by H$\alpha$ emission line data and synthetic stellar population modeling, predict quite a wide range for the ages of blue LSBGs: from 1-2 Gyr (Zackrisson et al. 2005) to 7-9 Gyr (Padoan et al., 1997; Jimenez et al. 1998). Ronnback & Bergvall (1994) carried out the multicolor studies of a sample of extremely blue galaxies with only small radial color gradients. In their work, BVI photometry was interpreted using spectral evolutionary models, and ages typically higher than 3 Gyr were found. Using optical/near-infrared (NIR) broadband photometry together with H$\alpha$ emission line data for a subsample of this sample, Zackrisson et al. (2005) found that the current observations cannot rule out the possibility that these blue LSBGs formed as recently as 1-2 Gyr ago. Schombert et al. (2001) argued on the basis of their V-I colors and relative HI content that the most gas-rich LSBGs should typically have mean stellar ages below 5 Gyr. Padoan et al. (1997) and Jimenez et al. (1998) used UBVRI photometry to conclude that most of the galaxies in their sample appeared to be older than about 7 Gyr. Vorobyov et al. (2009) used numerical hydrodynamic modelling to study the long-term ($\sim$13 Gyr) dynamical and chemical evolution of blue LSBGs adopting a sporadic scenario for star forming. Their modelling strongly suggested the existence of a minimum age for blue LSBGs: 1.5-3.0 Gyr (from model B-V colors and mean oxygen abundances), or 5-6 Gyr (from model H$\alpha$ equivalent widths). The later value may decrease slightly if LSBGs host truncated initial mass functions (IMFs) with a smaller upper mass limit. Haberzettl, Bomans, Dettmar (2005) studied the star formation history of a sample of 7 LSBGs in the HDF-S. The comparison of measured spectral energy distributions (SEDs) with the synthetic spectra extracted from the synthesis evolution model PEGASE (Fioc & Rocca-Volmerange 1997, assuming an exponentially declining star formation rate) suggested the ages of the dominant stellar population between 2 to 5 Gyrs. This implies that the major star formation event of LSBG galaxies took place at much later stages (at z$\sim$0.2 to 0.4).
These studies show that the ages of LSBGs are in debate, and could spread in a wide range, although the working sample is small in these investigations. Surely much more efforts and much larger sample are needed for detailed studies on the ages of LSBGs. Moreover, most of these investigations are based on optical photometry which has obvious shortcoming in estimating ages of galaxies. As we know, the youngest stars may dominate the ultraviolet (UV) and optical part of the spectrum, and an old stellar component will thereby be extremely difficult to detect. With the inclusion of NIR broadband photometry one would expect to be able to constrain the ages more strictly (Bell et al. 2000; Zackrisson et al. 2005). The UV data are also needed to build SEDs of galaxies in a wider range, which will help to derive ages of galaxies more reliably. At present, the modern digital sky survey in large sky coverage provide us much larger sample of galaxies for doing detailed and statistic studies on many fields. Moreover, the multiwavelength data of a large set of astronomical objects from far-ultraviolet (FUV) to NIR have been carried by high efficient surveys and have been released for public use, such as Galaxy Evolution Explorer (GALEX) in FUV and near-ultraviolet (NUV), Sloan Digital Sky Survey (SDSS) in optical and Two Micron All Sky Survey (2MASS) in NIR. These multiwavelength data will help us to well study the properties of a large sample of LSBGs. We focus on the ages of LSBGs in this work.
We will study the ages of a large sample of nearly face-on disc LSBGs selected from SDSS Data Release 7 (DR7, Abazajian et al. 2009) main galaxy sample (MGS, Strauss et al. 2002), and then matched them with the 2MASS /NIR and GALEX/UV data, so that their SEDs could cover from FUV to Ks (1350Å to 2.17$\mu$m) bands. The selected sample of LSBGs are 1,802 with their $\mu_{0}$(B) $\geq$ 22 mag arcsec$^{-2}$. Indeed, this is a follow-up work of Zhong et al. (2008) and Liang et al. (2010). In Zhong et al. (2008), we selected a la SDSS Data Release 4 (DR4, Adelman-McCarthy et al. 2004) main galaxy sample (Strauss et al. 2002), and presented their basic photometric properties including correlations of disk scalelength versus B-band absolute magnitude and distance, stellar populations from colors. In Liang et al. (2010), we studied the spectroscopic properties of this large sample of LSBGs, including dust extinction, strong emission-line ratios, metallicities and stellar mass-metallicity relations. In this work, we firstly extend the sample from DR4 to DR7, then try to obtain their multiwavelength data from FUV to Ks based on the public survey data from GALEX GR4/GR5 and 2MASS Extended Source Catalogue (XSC). Then, we use the evolutionary population synthesis (EPS) model PEGASE to fit their multiwavelength SEDs from FUV to NIR, and derive the ages of these galaxies. At the same time, a large sample of nearly face-on disk high surface brightness galaxies (HSBGs, 5,886) with $\mu_{0}$(B) $<$ 22 mag arcsec$^{-2}$ are selected and applied to same analyses for comparisons.
This paper is organized as follows. In Sect.2, we describe the multiwavelength observations and the selected sample. In Sect.3 we describe the EPS model PEGASE, which is what we used to fit the SEDs of the galaxies, and the adopted parameters. In Sect.4, we present the fitting method and the derived ages of the sample galaxies. In Sect.5, the incompleteness effects are discussed. The discussions are given in Sect.6. We summarize and conclude the work in Sect.7. Throughout this paper, a cosmological model with $H_0$=70 km s$^{-1}$ Mpc$^{-1}$, $\Omega _M$=0.3 and $\Omega _\Lambda =0.7$ has been adopted. All the magnitudes are in AB system.
The sample {#sect:Obs}
==========
Our sample galaxies are selected by matching three databases from multiwavelength surveys, i.e., the SDSS for ugriz optical, 2MASS for JHKs NIR, and GALEX for FUV and NUV bands.
The SDSS sample
---------------
The SDSS[^1] is the most ambitious astronomical survey ever undertaken in imaging and spectroscopy (York et al. 2000; Stoughton et al. 2002). The photometric and spectroscopic observations were conducted using the 2.5-m SDSS telescope at the Apache Point Observatory in New Mexico, USA. The imaging data were done in drift-scan mode in ugriz five bands, with effective central wavelengths of 3551, 4686, 6166, 7480 and 8932 Å, respectively (Gunn et al. 1998), and the 95% completeness limits for point sources are 22.0, 22.2, 22.2, 21.3 and 20.5 mag, respectively. The spectra are flux- and wavelength-calibrated with 4096 pixels from 3800 to 9200 Å at $R\sim$1800 (Stoughton et al. 2002).
We select 21,666 nearly face-on disk LSBGs from the SDSS-DR7 MGS by following the criteria used in Zhong et al. (2008). These are $fracDev_r$ $<$ 0.25 (indicating the fraction of luminosity contributed by the de Vaucouleurs profile relative to exponential profile in the $r$-band is much small), $b/a$ $>$ 0.75 (for nearly face-on ones, $a$ and $b$ are the semi-major and semi-minor axes of the fitted exponential disk, respectively), $M_B$ $<$ -18 (excluding few dwarf galaxies with this B-band absolute magnitude cut) and $\mu_0$(B)$\geq$ 22 mag arcsec$^{-2}$. These selection criteria have also been described in Zhong et al. (2010). At the same time, 30,896 nearly face-on disk HSBGs with $\mu_0$(B)$<$ 22 mag arcsec$^{-2}$ are selected for comparisons. The central surface brightnesses of our sample galaxies are calculated by using the corresponding parameters provided in the SDSS catalog following the method given in Sect.2.2 and Eq.(6) of Zhong et al. (2008). We use Petrosian magnitudes (and their errors) to represent the optical brightness of the SDSS galaxies. The Petrosian magnitudes could recover essentially all of the flux of an exponential galaxy profile (Stoughton et al. 2002), which are just the case of our sample here. All the magnitudes are converted to be in AB system following Hewett et al. (2006).
The SDSS-2MASS matched sample
-----------------------------
We use the 2MASS Extended Source Catalogue (XSC)[^2] to match with the SDSS selected sample. Using the Mt. Hopkins northern 1.3-m telescope and the CTIO southern 1.3-m telescope in Chile, the 2MASS covered almost entire sky in the J(1.25$\mu$m), H(1.65$\mu$m) and Ks(2.17$\mu$m) bands, with spatial resolution of 3$^{\prime \prime }$. The Point-source sensitivity limits (10$\sigma$) are 15.8 (0.8 mJy), 15.1 (1.0 mJy), and 14.3 (1.4 mJy) mag at J, H, and Ks, respectively. The extended source sensitivity (10$\sigma$) is $\sim$1 mag brighter than the point-source limits, or 14.7 (2.1 mJy), 13.9 (3.0 mJy), and 13.1 (4.1 mJy) mag at J, H, and Ks, respectively, with the precise threshold depending on the brightness profile of the objects (Jarrett et al. 2000). The magnitude limit of typical extended sources in the Ks band is about 15.3 AB mag (Lee et al. 2010).
By matching our SDSS sample galaxies with the 2MASS XSC catalog within 3$^{\prime \prime }$ following Blanton et al. (2005) and Lee et al. (2010), we obtain 3,523 LSBGs and 9,483 HSBGs. For the JHKs magnitudes (and their errors) of the galaxies, we adopt those from fit extrapolation (j,h,k-m-ext) provided by the 2MASS database[^3], which are assumed to better trace the galactic luminosity (Jarrett et al. 2000; Blanton et al. 2005). All the magnitudes are converted to be in AB system following Blanton et al. (2005).
The SDSS-2MASS-GALEX matched sample
-----------------------------------
We use the GALEX (Martin et al. 2005) survey GR4/GR5[^4] data to match with our selected sample from SDSS and 2MASS. The GALEX was launched in 2003 April, and will cover the whole sky in the FUV (1350 - 1750 Å) and NUV (1750 - 2800 Å) bands, with spatial resolutions of 6.0$^{\prime \prime }$ in the FUV and 4.5$^{\prime \prime }$ in the NUV. The central wavelengths of FUV and NUV are 1528Å and 2271Å, respectively. In this paper, only the GALEX objects detected both in the NUV and FUV bands were used. The GALEX data products are made available to the general public via the MultiMission Archive at Space Telescope Science Institute (MAST). The GALEX survey design was done for five modes, in which two of them are more related to our work here, i.e., the All-sky Imaging Survey (AIS) with the goal to survey the entire sky subject to a sensitivity of $m_{AB}$$\sim$ 20.5, and the Medium Imaging Survey (MIS), which covers 1000 deg$^2$ with extensive overlap of the SDSS (Martin et al. 2005; Morrissey et al. 2007).
We use the CASJobs[^5] to match our SDSS-2MASS sample with the GALEX GR4/GR5 database by importing coordinate list and then obtain their magnitudes in FUV and NUV (see the help in website[^6]). Firstly, we import a prepared coordinate list into the database, which is for the RA and DEC of our SDSS-2MASS matched sample. Then a new table of this is created in the database. Next we search for neighbors for the objects within a radius of 5$^{\prime \prime }$, and then obtain a new table of GALEX-matched objects which provide the unique GALEX object identifier (objid) as matched\_id, and also the search\_id to mark which original object it corresponds to. In the case of multiple matches, we choose the nearest one with minimum matching distance. Finally, we obtain the magnitudes, fluxes and their errors for these objects. We use the GALEX products given in the “-mcat.fits" file, which is the Merged (FUV+NUV) source catalog. We adopt “nuv\_mag" and “fuv\_mag" (and their errors) as the magnitudes of our matched objects, which are the NUV calibrated magnitude and FUV calibrated magnitude, respectively, and “calibrated" means that values have been converted to AB magnitudes (see help in the pipeline data guide[^7]).
Finally, we select 1,802 LSBGs and 5,886 HSBGs with FUV-to-NIR multiwavelength observations. The histogram distributions of their $\mu_{0}$(B) and redshift are given in Fig. \[his.mu.z\]. The median values of $\mu_{0}$(B) are 22.24 mag arcsec$^{-2}$ for LSBGs and 21.35 mag arcsec$^{-2}$ for HSBGs (mean as 22.28 and 21.25), respectively. The corresponding median values of redshifts are 0.0732 and 0.0801 (mean as 0.0810 and 0.0886), respectively.
Corrections on magnitudes
-------------------------
The effects of foreground Galactic extinction on the observed magnitudes were calculated by using the reddening maps of Schlegel, Finkbeiner & Davis (1998). The corresponding values of the SDSS and 2MASS magnitudes have been provided by Blanton et al. (2003) at the NYU-VAGC[^8] catalog (thanks to them). For the GALEX FUV and NUV magnitudes, the corrections for interstellar extinctions were calculated using $A(FUV)=8.16E(B-V)$ and $A(NUV)=8.90E(B-V)$ (Rey et al. 2007; Kinman et al. 2007), and the $E(B-V)$ were taken from Schlegel, Finkbeiner & Davis (1998).
The magnitudes in our sample are then corrected for the redshifts. We calculated the K-corrections using their K\_CORRECT program, version 4.1.4, originally developed by Blanton et al. (2003) and now extended to handle GALEX data (Blanton & Roweis 2007). We corrected the observed magnitudes (FUV-to-NIR magnitudes) of the sample galaxies into the magnitudes at redshift z = 0. We did not consider the evolutionary correction on the magnitudes of galaxies since it won’t be large in the redshift range of our sample. We did not further consider aperture corrections for the magnitudes among three surveys since the Petrosian magnitudes in SDSS, fit extrapolation magnitudes in 2MASS and calibrated magnitudes in GALEX have been good tracers of the total luminosities of these galaxies.
For the internal reddening of the galaxy, we input dust extinction in the PEGASE model, then get the model SEDs after extinction when do SED fittings. As for the values of internal extinction $A_V$, we adopt two methods, one is constant dust extinction, i.e., $A_V$=0.6 mag for LSBGs and $A_V$=0.8 mag for HSBGs, and another one is assuming $A_V$ as a free parameter varying from 0.1 to 1.6 in the fitting procedures, which come from Liang et al. (2010) about the dust extinction estimates from Balmer decrement for related samples. More information will be given in Sect.4.
![Histogram distributions of $\mu_{0}$(B) (with bin of 0.1) and redshift (with bin of 0.01) for our sample galaxies with FUV-to-NIR multiwavelength observations. The shaded regions refer to the LSBGs with $\mu_{0}$(B) $\geq$ 22 mag arcsec$^{-2}$, and the dashed-lines refer to the HSBGs. []{data-label="his.mu.z"}](ms471fig1a.eps "fig:"){width="5.5cm" height="5.5cm"} ![Histogram distributions of $\mu_{0}$(B) (with bin of 0.1) and redshift (with bin of 0.01) for our sample galaxies with FUV-to-NIR multiwavelength observations. The shaded regions refer to the LSBGs with $\mu_{0}$(B) $\geq$ 22 mag arcsec$^{-2}$, and the dashed-lines refer to the HSBGs. []{data-label="his.mu.z"}](ms471fig1b.eps "fig:"){width="5.5cm" height="5.5cm"}
Evolutionary Synthesis Model PEGASE {#sect:data}
===================================
Evolutionary population synthesis is a powerful tool of interpretation of the integrated spectrophotometric observations of galaxies. The most common method of model-observation comparison for stellar population analysis in galaxies or star clusters is SED fitting, with either least-squares or chi-squared minimization technique (Kong et al. 2000; Gavazzi et al. 2002; Jiang et al. 2003; Li et al. 2004a,b; Fan et al. 2006; Ma et al. 2002a,b, 2007, 2009a,b; Wang et al. 2010). There are several EPS models popularly used in astrophysical studies (Chen et al. 2009, 2010 for references therein). We use PEGASE in this work.
PEGASE is an evolutionary spectral synthesis model for starbursts and evolved galaxies of the Hubble sequence. It is continuous over an exceptionally large wavelength range from 220 Å up to 5$\mu$m. It was extended to the NIR of the atlas of synthetic spectra of Rocca-Volmerange & Guiderdoni (1988) with a revised stellar library including cold star parameters and stellar tracks extended to the thermally-pulsing regime of the asymptotic giant branch (TP-AGB) and the post-AGB phase. The synthetic stellar spectral library is taken from Kurucz (1992), modified by Lejeune et al. (1997) to fit the observed colors. A set of reference synthetic spectra at z = 0, to which the cosmological k- and evolution e- corrections for high-redshift galaxies are applied, is built from fits of observational templates (Fioc & Rocca-Volmerange 1997).
With the PEGASE code, we can compute the stellar SEDs of starbursts and evolved galaxies of the Hubble sequence at any stage of evolution, within the metallicity range Z = 10$^{-4}$ to 10$^{-1}$. Typical parameters of PEGASE are the star formation rate (SFR) and IMF. Assuming a standard IMF, SFR and other initial conditions, such as dust extinction, the PEGASE code will give an evolutionary history and some other important properties of a galaxy (Li et al. 2004a,b).
In our fitting analysis, the internal dust extinctions are considered in two methods, the constant A$_V$ (0.6 for LSBGs and 0.8 for HSBGs) and varying $A_V$ within A$_V$ = 0.1 - 1.6. A zero initial metallicity of interstellar medium (ISM) is taken. We adopt the exponential decreasing SFR ([*SFR(t)*]{} $\propto$ [*e*]{}$^{-t/\tau}$) with $\tau$ = 0.1 - 15 Gyr varying in the fittings. The IMF is assumed to follow the Salpeter (1955) form, $\Phi$(M) = A$\times$ $M^{-\alpha}$ with ${\alpha}$ = 2.35 and a lower cutoff of $M_{l}$ = 0.1 $M_{\odot}$ and an upper cutoff $M_{u}$ = 125 M$_{\odot}$(Sawicki & Yee 1998). As a result, a rest-frame modeled spectra with various star formation histories are generated by running the PEGASE code.
Model Fits And Results
======================
We present the fitting method in Sect. \[sect:analysis0\], and the fitting results in Sects. \[sect:analysis1\], \[sect:analysis2\]. Extinction affects intrinsic colors of the objects and hence accurate ages, so the photometric measurements must be dereddened before running the fitting procedure. We use two methods for considering the internal dust extinction of the galaxies in fitting procedures, the constant and varying $A_V$.
The fitting method {#sect:analysis0}
------------------
To estimate the ages of galaxies accurately, we add UV and NIR photometric data points to the optical data. As Ma et al. (2009b) discussed, Kaviraj et al. (2007) showed that the combination of FUV and NUV photometry with optical observations in the standard broad bands enables one to efficiently break the age-metallicity degeneracy. The optical broadband colors have the more obvious problem of age-metallicity degeneracy (Worthey 1994; MacArthur et al. 2004). Again, de Jong (1996) showed that such degeneracy can be partially broken by adding NIR photometry to optical colors, which has also been stated by Bell et al. (2000), Wu et al. (2005) and Li et al. (2007).
Our observational data consist of integrated luminosities through a given set of filters (FUV, NUV, ugriz, JHKs), thus we convolved the theoretical SEDs with these filter response curves to obtain synthetic ultraviolet, optical and NIR photometry for comparison. The synthetic magnitude in the AB magnitude system for the $i$th filter can be computed as $$m_i=-2.5{\rm log}\frac{\int_{\lambda} F_\lambda\varphi_i(\lambda)d\lambda}{\int_{\lambda}\varphi_i(\lambda)d\lambda}-48.60,$$ where $F_{\lambda}$ is the theoretical SED and $\varphi_i$ is the response curve of the $i$th filter of the sets of filter we used.
We use a $\chi^2$ minimization test to examine which model SEDs are most comparable with the observed SEDs following $$\chi^2=\frac{1}{d}\sum^{10}_{i=1}\frac{[m^{obs}_{\lambda_i}-m^{mod}_{\lambda_i}(t)]^2}{\sigma^2_{i}},$$ where $m^{mod}_{\lambda_i}(t)$ is the integrated magnitude in the $i$th filter of a theoretical SED at age $t$, $m^{obs}_{\lambda_i}(t)$ represents the observed integrated magnitude in the same filter, and $\sigma_i^2$ is the observational uncertainty for the $i$th filter magnitude, $d$ is the number of degrees of freedom. We did not consider the uncertainty associated with the model itself here, which should be insignificant and not affect much our results. More details about the fitting method can be referred to Li et al. (2004a,b) and Ma et al. (2009a,b). The derived ages of our sample galaxies from such SED fitting analyses are given in next two subsections.
Constant dust extinction {#sect:analysis1}
------------------------
We adopt constant $A_V$=0.6 mag for LSBGs and $A_V$=0.8 mag for HSBGs in fittings by following Liang et al. (2010). In Liang et al. (2010), we derive the $A_V$ value for each object from their spectroscopic Balmer decrement H$\alpha$/H$\beta$ (see their Fig.2 for histogram distributions). The median values of $A_V$ of the four subsamples (with $\mu_0(B)$ in units of mag arcsec$^{-2}$: vLSBGs with 22.75-24.5, iLSBGs with 22.0-22.75, iHSBGs with 21.25-22.0, vHSBGs with $<$21.25) are 0.46, 0.63, 0.76, 0.83, respectively. In our present work, most LSBGs have $\mu_0(B)$ within 22-23 mag arcsec$^{-2}$, thus $A_V$=0.6 is acceptable to be their dust extinction. Also $A_V$=0.8 is reasonable to be the dust extinction of HSBGs.
Figure \[Fig2\] shows the resulted SED fittings (left panel for one example of LSBGs; middle panel for one example of HSBGs) and the histogram distributions of the derived ages of the galaxies (the right panel, where the shaded region is for LSBGs and the dashed-line is for HSBGs). The right panel shows that the ages of most of the LSBGs are 1-5 Gyr with median (mean) value of 1.63 (1.75) Gyr. This means that the majority of their stars were formed quite recently. However, about 3% of the LSBGs have larger ages, 5-8 Gyr. This small part of galaxies could probably form their stars at earlier time. Similarly, the ages of most of HSBGs are 1-5 Gyr as well with the median value of 1.47 (mean as 1.59) Gyr. About 3% of them have larger ages as 5-8 Gyr. This similarity may mean that the LSBGs and HSBGs could have no much different star formation history. This result is consistent with the metallicity estimates of similar samples presented in Liang et al. (2010). But we should notice the fact that the HSBGs are slightly younger than the LSBGs, $\sim$0.2 Gyr shown by the median (mean) ages of the samples. [The median (mean) ages of the sample galaxies are presented in Table \[tab1\].]{}
![The fitting results with PEGASE model and constant dust extinction ($A_V$=0.6 for LSBGs and $A_V$=0.8 for HSBGs) for our sample galaxies. The left panel shows the SED fitting for one example of LSBGs (RA=8.06543, DEC=0.90616 in 2000 epoch, the derived age is 1.57 Gyr), the middle panel shows the SED fitting for one example of HSBGs (RA=6.75941, DEC=-9.63505 in 2000 epoch, the derived age is 1.44 Gyr)), and the right panel shows the histogram distributions of the derived ages of LSBGs (shade region) and HSBGs (dashed-line) with bin of 0.2 Gyr. []{data-label="Fig2"}](ms471fig2a.eps "fig:"){width="4.8cm" height="4.8cm"} ![The fitting results with PEGASE model and constant dust extinction ($A_V$=0.6 for LSBGs and $A_V$=0.8 for HSBGs) for our sample galaxies. The left panel shows the SED fitting for one example of LSBGs (RA=8.06543, DEC=0.90616 in 2000 epoch, the derived age is 1.57 Gyr), the middle panel shows the SED fitting for one example of HSBGs (RA=6.75941, DEC=-9.63505 in 2000 epoch, the derived age is 1.44 Gyr)), and the right panel shows the histogram distributions of the derived ages of LSBGs (shade region) and HSBGs (dashed-line) with bin of 0.2 Gyr. []{data-label="Fig2"}](ms471fig2b.eps "fig:"){width="4.8cm" height="4.8cm"} ![The fitting results with PEGASE model and constant dust extinction ($A_V$=0.6 for LSBGs and $A_V$=0.8 for HSBGs) for our sample galaxies. The left panel shows the SED fitting for one example of LSBGs (RA=8.06543, DEC=0.90616 in 2000 epoch, the derived age is 1.57 Gyr), the middle panel shows the SED fitting for one example of HSBGs (RA=6.75941, DEC=-9.63505 in 2000 epoch, the derived age is 1.44 Gyr)), and the right panel shows the histogram distributions of the derived ages of LSBGs (shade region) and HSBGs (dashed-line) with bin of 0.2 Gyr. []{data-label="Fig2"}](ms471fig2c.eps "fig:"){width="4.6cm" height="4.2cm"}
Varying dust extinction {#sect:analysis2}
-----------------------
Now we adopt $A_V$ as a varying parameter from 0.1 to 1.6 in the fittings. This range of $A_V$ values is reasonable for LSBGs and HSBGs by following Liang et al. (2010), and 1.6 is almost the upper limit of the sample (also see Liang et al. 2007 for the SDSS star forming galaxies).
Figure \[Fig3\] shows the resulted SED fittings of two example galaxies (the left panel for the same example of LSBGs, and the middle panel for the same example of HSBGs as in Fig. \[Fig2\]) and the histogram distributions of the derived ages of the galaxies (the right panel, where the shaded region is for LSBGs and the dashed-line is for HSBGs). The right panel also shows that the ages of most of the LSBGs are 1-5 Gyr with median (mean) value of 2.06 (2.18) Gyr, which means that the majority of their stars were formed quite recently. However, about 2% of the LSBGs have larger ages, 5-8 Gyr, which mean that they could form their stars at earlier time. Similarly, the ages of most of HSBGs are also 1-5 Gyr with the median value of 1.86 (mean as 1.92). About 3% of them have larger ages with 5-8 Gyr. This similarity may mean that the LSBGs and HSBGs have no much different star formation history, but the HSBGs are 0.2 Gyr younger than the LSBGs generally. All these results are consistent with those obtained with constant dust extinction, although the varying $A_V$ results in a bit older age ($\sim$0.4 Gyr) than the constant one. [The corresponding ages are also presented in Table \[tab1\].]{}
![The fitting results with PEGASE model and varying dust extinction ($A_V$=0.1-1.6) for our sample galaxies. The left panel shows the SED fitting for one example of LSBGs (same as in Fig. \[Fig2\], but the derived age is 2.10 Gyr), the middle panel shows the SED fitting for one example of HSBGs (same as in Fig. \[Fig2\] but the derived age is 1.96 Gyr), and the right panel shows the histogram distributions of the derived ages of LSBGs (shade region) and HSBGs (dashed-line) with bin of 0.2 Gyr. []{data-label="Fig3"}](ms471fig3a.eps "fig:"){width="4.8cm" height="4.8cm"} ![The fitting results with PEGASE model and varying dust extinction ($A_V$=0.1-1.6) for our sample galaxies. The left panel shows the SED fitting for one example of LSBGs (same as in Fig. \[Fig2\], but the derived age is 2.10 Gyr), the middle panel shows the SED fitting for one example of HSBGs (same as in Fig. \[Fig2\] but the derived age is 1.96 Gyr), and the right panel shows the histogram distributions of the derived ages of LSBGs (shade region) and HSBGs (dashed-line) with bin of 0.2 Gyr. []{data-label="Fig3"}](ms471fig3b.eps "fig:"){width="4.8cm" height="4.8cm"} ![The fitting results with PEGASE model and varying dust extinction ($A_V$=0.1-1.6) for our sample galaxies. The left panel shows the SED fitting for one example of LSBGs (same as in Fig. \[Fig2\], but the derived age is 2.10 Gyr), the middle panel shows the SED fitting for one example of HSBGs (same as in Fig. \[Fig2\] but the derived age is 1.96 Gyr), and the right panel shows the histogram distributions of the derived ages of LSBGs (shade region) and HSBGs (dashed-line) with bin of 0.2 Gyr. []{data-label="Fig3"}](ms471fig3c.eps "fig:"){width="4.6cm" height="4.2cm"}
Incompleteness effects
======================
It is important to test the completeness of the sample, and discuss its effect on the derived ages of the galaxies.
The volume-limited sub-sample
-----------------------------
The incompleteness of the sample could affect the derived results of our analysis, thus it is important to test. Zhong et al. (2008) has carefully discussed the incompleteness of the sample of LSBGs. In their Sect. 2, the good completeness of the SDSS MGS has been discussed. The MGS is a spectroscopic sample selected from the SDSS photometric data. The completeness for objects with spectroscopy observations is high, exceeding 99%, and the fraction of galaxies eliminated by surface brightness cut is very small ($\sim0.1$%). Relative to all the SDSS targets, the SDSS spectroscopic survey is 90% complete (Blanton et al. 2003; Hogg et al. 2004; Strauss et al. 2002; McIntosh et al. 2006). Blanton et al. (2005) had nicely discussed the incompleteness at low surface brightness, and presented the contributions to completeness as a function of surface brightness, the $r$-band Petrosian half-light surface brightness $\mu_{50,r}$. It shows that, for those brighter ones with $\mu_{50,r}<$23 mag arcsec$^{-2}$ (corresponding to $\mu_{0}$(B)=24.5 mag arcsec$^{-2}$), the completeness of the spectroscopy is very close to 100%. But for the photometric and tilling catalogs, they show obvious incompleteness within $\mu_{50,r}$=22-23 mag arcsec$^{-2}$, then the total completeness there could decrease to be about 70%.
Similar to Zhong et al. (2008) in their Sect.5, here we also extract a volume-limited sub-sample to minimize the effects of sample incompleteness on the derived ages of galaxies. We firstly extract a volume-limited sub-sample from the $M_r-z$ plane by considering $z<0.1$ and those brighter ones than the corresponding $M_r$ from the SDSS-DR7 MGS as given in Sect. \[sect:Obs\].1. We obtain 5,608 LSBGs and 8,046 HSBGs. Then, we match them with our sample galaxies having FUV to NIR SEDs (i.e., the 1,802 LSBGs and 5,886 HSBGs). This allows us to obtain a sub-sample including 1,046 LSBGs and 2,853 HSBGs. Their ages are given in Fig. \[fig.Mr.z\] as histogram distributions. The left panel is for the constant dust extinction, same as in Fig. \[Fig2\], and the right panel is for the varying dust extinction, same as in Fig. \[Fig3\]. The left one shows that the median (mean) ages are 1.60 (1.69) Gyr for LSBGs and 1.44 (1.53) Gyr for HSBGs sub-samples. The right panel shows that the corresponding median (mean) ages are 2.10 (2.20) Gyr for LSBGs and 1.90 (1.94) Gyr for HSBGs sub-samples. These derived ages of the volume-limited sub-sample are very similar to those of the whole sample (Figs. \[Fig2\],\[Fig3\]) with only 0.02-0.06 Gyr discrepancy in ages. This verifies the robustness of our results. [The derived ages are also presented in Table \[tab1\].]{}
![The histogram distributions of ages (bin as 0.2) of the volume-limited sub-samples of LSBGs (shade region) and HSBGs (dashed line). The left panel is for the constant dust extinction case (see Sect. 4.2) and the right panel is for the varying dust extinction case (see Sect. 4.3). []{data-label="fig.Mr.z"}](ms471fig4a.eps "fig:"){width="4.6cm" height="4.2cm"} ![The histogram distributions of ages (bin as 0.2) of the volume-limited sub-samples of LSBGs (shade region) and HSBGs (dashed line). The left panel is for the constant dust extinction case (see Sect. 4.2) and the right panel is for the varying dust extinction case (see Sect. 4.3). []{data-label="fig.Mr.z"}](ms471fig4b.eps "fig:"){width="4.6cm" height="4.2cm"}
The pure disc sub-sample {#sec.5.2}
------------------------
It is also necessary to check the effects of the calculated central surface brightness on our results. In the sample criteria (Sect. 2.1), we adopted $fracDev_r$ $<$ 0.25 to guarantee the disk light (in r-band) to be well explained by exponential profile. To entirely remove the effect of bulge light, we select a sub-sample as pure discs with $fracDev_r$ = 0. This sub-sample includes 8,453 LSBGs and 12,096 HSBGs from the SDSS-DR7 MGS. Then we match these sub-sample galaxies with our sample galaxies having FUV to NIR SEDs, then we obtain 505 LSBGs and 1,818 HSBGs. The derived ages of these sub-sample galaxies are given in Fig. \[fig.fraDev\]. The left panel shows the results with constant dust extinction, and the median (mean) ages are 1.65 (1.77) Gyr for LSBGs and 1.42 (1.58) Gyr for HSBGs sub-samples. The right panel shows the results with varying dust extinction, and the corresponding median (mean) ages are 2.10 (2.20) Gyr for LSBGs and 1.83 (1.87) Gyr for HSBGs sub-samples. These results are also very similar to the derived ages of the whole sample of LSBGs and HSBGs (given in Fig. \[Fig2\], Fig. \[Fig3\]), respectively, as well similar to those of the volume-limited sub-sample (given in Fig. \[fig.Mr.z\]). The discrepancy is 0.02-0.08 Gyr. [These results are presented in Table \[tab1\] as well.]{}
We use criterion $fracDev_r<$0.25 to select the disc galaxies in our work as shown in Sect. 2.1. To carefully check whether the central surface brightness of galaxies depends on $fracDev_r$, we obtain the relations of $\mu_{0}$(B) vs. $fracDev_r$ for our LSBGs. It shows very small slope in their least-square fitting relation, $-$0.08 (i.e., $\mu_{0}(B)=-0.08fracDev_r+22.51$). This negative value does show that, for the galaxies having higher $fracDev_r$, their central surface brightness may be overestimated a bit by exponential disc profile, however, the discrepancy is quite small. We also check this relation for HSBGs, the corresponding slope is small as well, $-$0.30. These verify that $fracDev_r<$0.25 could be a reasonable criterion to select disc galaxies, and won’t affect much on our calculations on central surface brightnesses of galaxies. This criterion has already been more strict than what used in literatures, for example, Chang et al. (2006) and Shao et al. (2007) used $fracDev_r<$0.5 to select their spiral galaxies.
![The histogram distributions of ages (bin as 0.2) of the pure disc sub-samples of LSBGs (shade region) and HSBGs (dashed line) with $fracDev_r$ = 0. The left panel is for the constant dust extinction case (see Sect. 4.2) and the right panel is for the varying dust extinction case (see Sect. 4.3). []{data-label="fig.fraDev"}](ms471fig5a.eps "fig:"){width="4.6cm" height="4.2cm"} ![The histogram distributions of ages (bin as 0.2) of the pure disc sub-samples of LSBGs (shade region) and HSBGs (dashed line) with $fracDev_r$ = 0. The left panel is for the constant dust extinction case (see Sect. 4.2) and the right panel is for the varying dust extinction case (see Sect. 4.3). []{data-label="fig.fraDev"}](ms471fig5b.eps "fig:"){width="4.6cm" height="4.2cm"}
The $r \leq 16$ sub-sample {#sec.5.3}
--------------------------
As we know 2MASS is shallow, thus we may worry about the results obtained from 2MASS biased to the massive galaxies. McIntosh et al. (2006) helpfully made cross-correlation on the well-defined and highly complete spectroscopic selection of $r\leq 17.5$ mag galaxies in the SDSS MGS with the 2MASS sources to explore the nature and completeness of the 2MASS (K-band) selection of nearby galaxies. They quantified the completeness of 2MASS galaxies in terms of optical properties from SDSS, and found, for $r\leq 16$ mag, 94.5 per cent of the MGS is found in the 2MASS XSC. An XSC completeness of 97.6 per cent is achievable at bright magnitudes, with blue low-surface-brightness galaxies being the only major source of incompleteness. They concluded that the rapid drop in XSC completeness at $r > 16$ mag reflects the sharp surface-brightness limit of the extended source detection algorithm in 2MASS. A combined $K \leq 13.57$ and $r\leq 16$ mag-limited selection provides the most representative inventory of galaxies in the local cosmos with NIR and optical measurements, and 92.2 per cent completeness.
Therefore, we select a sub-sample with $r\leq 16$ mag from our LSBGs and HSBGs (they have $K \leq 13.57$), and then compare their derived ages with those of the whole sample. Firstly, the galaxies with $r\leq 16$ are selected from the SDSS-DR7 MGS, this results in 1,855 LSBGs and 2,485 HSBGs. These galaxies are further matched with our sample galaxies having FUV to NIR SEDs, and then results in a sub-sample including 585 LSBGs and 1,553 HSBGs, respectively. Their ages are given in Fig. \[fig.r16\] as histogram distributions. The left panel shows the results with constant dust extinction, and the median (mean) ages are 1.54 (1.65) Gyr for LSBGs and 1.42 (1.53) Gyr for HSBGs sub-samples. The right panel shows the results with varying dust extinction, and the corresponding median (mean) ages are 2.07 (2.17) Gyr for LSBGs and 1.93 (1.98) Gyr for HSBGs sub-samples. These derived ages of the sub-sample are similar to those of the whole sample as given in Fig. \[Fig2\] and Fig. \[Fig3\], and to the other two sub-samples as given in Fig. \[fig.Mr.z\] and Fig. \[fig.fraDev\], respectively. The discrepancy is only up to 0.12 Gyr. [These results are given in Table \[tab1\].]{}
![The histogram distributions of ages (bin as 0.2) of the $r \leq 16$ sub-samples of LSBGs (shade region) and HSBGs (dashed line). The left panel is for the constant dust extinction case (see Sect. 4.2) and the right panel is for the varying dust extinction case (see Sect. 4.3).[]{data-label="fig.r16"}](ms471fig6a.eps "fig:"){width="4.6cm" height="4.2cm"} ![The histogram distributions of ages (bin as 0.2) of the $r \leq 16$ sub-samples of LSBGs (shade region) and HSBGs (dashed line). The left panel is for the constant dust extinction case (see Sect. 4.2) and the right panel is for the varying dust extinction case (see Sect. 4.3).[]{data-label="fig.r16"}](ms471fig6b.eps "fig:"){width="4.6cm" height="4.2cm"}
The pure disc with $r \leq 16$ sub-sample {#sec.5.4}
-----------------------------------------
In this subsection, we perform one more test to check the robustness of our estimates by further extracting a subsample with $fracDev_r$=0 and $r\leq 16$ mag, i.e. the pure discs with $r \leq 16$ mag. That is, we select the sub-sample by matching the galaxies in Sect. \[sec.5.2\] and Sect. \[sec.5.3\]. This results in 161 LSBGs and 341 HSBGs. The distributions of their derived ages are presented in Fig. \[fig.frac0.r16\].
The left panel shows the results with constant dust extinction, and the median (mean) ages are 1.60 (1.74) Gyr for LSBGs and 1.39 (1.63) Gyr for HSBGs sub-samples. The right panel shows the results with varying dust extinction, and the corresponding median (mean) ages are 2.17 (2.30) Gyr for LSBG and 1.90 (1.97) Gyr for HSBG sub-samples. These derived ages of the sub-samples are similar to those of the whole sample as given in Fig. \[Fig2\] and Fig. \[Fig3\], and to the other three sub-samples as given in Figs. \[fig.Mr.z\],\[fig.fraDev\],\[fig.r16\]. The discrepancy is only up to 0.13 Gyr. The results are also given in Table \[tab1\].
![The histogram distributions of ages (bin as 0.2) of the pure disc with $r \leq 16$ sub-samples of LSBGs (shade region) and HSBGs (dashed line). The left panel is for the constant dust extinction case and the right panel is for the varying dust extinction case.[]{data-label="fig.frac0.r16"}](ms471fig7a.eps "fig:"){width="4.6cm" height="4.2cm"} ![The histogram distributions of ages (bin as 0.2) of the pure disc with $r \leq 16$ sub-samples of LSBGs (shade region) and HSBGs (dashed line). The left panel is for the constant dust extinction case and the right panel is for the varying dust extinction case.[]{data-label="fig.frac0.r16"}](ms471fig7b.eps "fig:"){width="4.6cm" height="4.2cm"}
All the consistencies among the whole sample and the four sub-samples about the derived ages of the sample galaxies confirm well the robustness of our analyses. To be clear, all the results are summarized in Table \[tab1\].
[llllll]{} & whole & volume-limited & pure disc & $r \leq 16$ & pure disc + $r \leq 16$\
& & & LSBGs & &\
Number & 1,802 & 1,046 & 505 & 585 & 157\
Constant $A_V$ & 1.63 (1.75) & 1.60 (1.69) & 1.65 (1.77) & 1.54 (1.65) & 1.60 (1.74)\
Varying $A_V$ & 2.06 (2.18) & 2.10 (2.20) & 2.10 (2.20) & 2.07 (2.17) & 2.17 (2.30)\
& & & HSBGs & &\
Number & 5,886 & 2,853 & 1,818 & 1,553 & 392\
Constant $A_V$ & 1.47 (1.59) & 1.44 (1.53) & 1.42 (1.58) & 1.42 (1.53) & 1.39 (1.63)\
Varying $A_V$ & 1.86 (1.92) & 1.90 (1.94) & 1.83 (1.87) & 1.93 (1.98) & 1.90 (1.97)\
Discussions {#sect:discussion}
===========
Comparisons with previous studies
---------------------------------
We study the ages of LSBGs using the EPS model PEGASE with exponential decreasing SFR to fit their multiwavelength SEDs from FUV to NIR. The HSBGs are also applied to similar studies as comparisons. Firstly, we find that the derived ages are 1-5 Gyr for most of our large sample of 1,802 LSBGs, which is consistent with most of the previous studies based on smaller sample.
Schombert et al. (2001) studied the V-I colors and relative H[i]{} content of the most gas-rich LSB dwarf galaxies. They suggested that the low stellar densities of their gas-rich LSB dwarfs are due to inefficient conversion of gas mass into stellar mass. The further comparison with star formation models (Boissier & Prantzos 2000) indicated that the blue optical colors of LSB dwarfs can only be explained by a dominant stellar populations less than 5 Gyr in mean age.
Zackrisson et al. (2005) used optical/NIR broadband photometry together with H$\alpha$ emission line data to study the ages of a sample of 9 blue LSBGs. They found that the current observations cannot rule out the possibility that these blue LSBGs formed as recently as 1-2 Gyr ago. Indeed, their Fig.3 shows that the blue colors of the group of blue LSBGs (B-V$\sim$ 0.4 mag, V-J$\sim$ 1.2 mag) can be well represented by an absolute age of 3.0 Gyr, an average age of 2.2 Gyr and a star formation history with $\tau$= 1.0 Gyr. However, they also present possible much older ages, such as 13 Gyr and 7.4 Gyr. Indeed, our results here from more than thousands of LSBGs confirm their young age results.
In Vorobyov et al. (2009), they used numerical hydrodynamic modelling and found the existence of a minimum age for blue LSBGs: 1.5-3.0 Gyr or 5-6 Gyr. They complement hydrodynamic modelling with population synthesis modeling to produce the integrated B-V colors and H$\alpha$ equivalent widths. They adopted a sporadic model for star formation which yields no radial abundance gradients in the model disk (as observed by de Blok & van der Hulst 1998). They also mentioned this sporadic star formation was agreement with the studies that the current star formation of LSBGs is localized to a handful of compact regions (Auld et al. 2006 from H$\alpha$ imaging), and then there is little or no diffuse H$\alpha$ emission coming from the rest of the galactic disk. We may need more observations on LSBGs to further support this.
The most similar and comparable work with us is Haberzettl, Bomans, Dettmar (2005), who studied the star formation history of seven LSBGs in the HDF-S by comparing the measured SEDs with the synthetic spectra extracted from PEGASE. Comparing a library of SEDs to the measured spectra they were able to derive ages between 2 to 5 Gyrs for the dominant stellar population. All these investigations above, as well as ours with a much larger sample, favor a scenario that the major star formation events of LSBG galaxies took place at much later stages (at z$\sim$0.2 to 0.4).
However, Podoan et al. (1997) concluded that most of the LSBGs in their sample appeared to be older than about 7 Gyr. In the work, Podoan et al. (1997) used a new IMF derived from numerical fluid dynamics simulations and a new synthetic stellar population code obtained in their earlier works. Jimenez et al. (1998) further improved their model and obtained similar results about ages of LSBGs (older than 7 Gyr). In fact, a part of our LSBGs ($\sim$2-3%) also have larger ages as 5-8 Gyr, which could be consistent with Podoan et al. (1997) and Jimenez et al. (1998). These could mean that some of the LSBGs may do form their majority of stars in earlier time.
Comparing the LSBGs and HSBGs
-----------------------------
Another aspect of results in this work is that the HSBGs do not show much different ages from the LSBGs. The ages of most of the HSBGs are also 1-5 Gyr with part of them (3%) have larger ages as 5-8 Gyr. This is not unlikely and consistent with some other investigations. Mattson, Caldwell, Bergvall (2007) study the chemical evolution of LSBGs. They conclude that LSBGs probably have the same ages as their high surface brightness counterparts, although the global rate of star formation must be considerably lower in these galaxies. Boissier et al. (2003) compared the observed properties with the predictions of models of the chemical and spectrophotometric evolution of LSBGs, the basic idea behind the models is that LSBGs are similar to “classical" HSB spirals except for a larger angular momentum.
As van den Hoek et al. (2000) discussed, the presence of an old stellar population in many late-type LSBGs as indicated by the optical colors and confirmed by their galactic chemical and photometric evolution model suggested that LSBGs roughly follow the same evolutionary history as HSBGs, but at a much lower rate. The mean age of the stellar population in most LSBGs and HSBGs was similar even though the disks of LSBGs were in a relatively early evolutionary stage.
Although Haberzettl, Bomans, Dettmar (2005) found the HSBGs from Kennicutt (1992) have much larger ages than their LSBGs, in fact, their Fig.2 shows that some of the objects of Kennicutt (1992) do have quite young ages $<$4 Gyr, and many of the objects of Kennicutt (1992) have SEDs as early-type galaxies (E or S0), which should not be very similar objects to our galaxies here as face-on disks. Similarly, the relatively high ages of the HSBGs in Terlevich & Forbes (2002) and Caldwell et al. (2003) could also be due to most of their samples as early-type galaxies.
The similar ages between LSBGs and HSBGs obtained in this work could be consistent with Liang et al. (2010) about the metallicity analysis of the LSBGs and HSBGs, which also come from the parent sample of Zhong et al. (2008). They found that the LSBGs and HSBGs located closely on the stellar mass vs. metallicity and N/O vs. O/H relations of the normal galaxies, but the LSBGs have slightly higher N/O than the HSBGs at give O/H in low metallicity region. This may mean that the LSBGs could have relatively lower SFR than the HSBGs and then show dominant primary nitrogen component there as Molla et al. (2006, their Fig.5) suggested. However, we should notice that the derived ages of HSBGs in this work are about 0.2 Gyr younger than those of the LSBGs generally (see Table \[tab1\]). This may mean that HSBGs may have occurred star formation process more recently than the LSBGs, or the recent star formation process is stronger in HSBGs than in LSBGs. This is also consistent with the results of Chen et al. (2010, in preparation) for the stellar population analysis on spectral absorption lines and continua of Liang et al (2010)’s sample, in which they found that the HSBGs have slightly larger fraction (5%) of young population than the LSBGs. In a word, as Liang et al. (2010) concluded, the large sample shows that LSBGs span a wide range in metallicity and stellar mass, and they lie nearly on the stellar mass vs. metallicity and N/O vs. O/H relations of normal galaxies. The HSBGs show similar trends. These suggest that LSBGs and HSBGs have not had dramatically different star formation and chemical enrichment histories.
Effects of other parameters
---------------------------
We adopt the exponential decreasing SFR in PEGASE for fitting the SEDs of the sample galaxies. The derived star formation rate decay time $\tau$ of LSBGs spread in a wide range from 0.1 to 15 Gyr in both cases of constant (Sect.4.1) and varying (Sect.4.2) dust extinction. In the constant $A_V$ case the median of $\tau$ is about 0.5 Gyr and mean as 0.57 Gyr. In the varying $A_V$ case, the median of $\tau$ is about 0.5 Gyr and mean as 0.62 Gyr. For the HSBGs, the derived $\tau$ values are also within 0.1-15 Gyr with median (mean) of 0.5 Gyr (0.78 Gyr) in constant $A_V$ case, and median (mean) of 0.6 Gyr (0.73 Gyr) in varying $A_V$ case, respectively. It is similar between LSBGs and HSBGs. These values are not unacceptable. Haberzettl, Bomans, Dettmar (2005) estimated $\tau$ to be 500 Myr (2 cases), 1400 Myr (3 cases), 5000 Myr (2 cases) for their 7 LSBGs. Li et al. (2004a) adopted $\tau$ = 3 Gyr for M81, and Li et al. (2004b) adopted $\tau$ = 12 Gyr for M33 in their SED fittings by using PEGASE. Zackrisson et al. (2005) also adopted wide range of $\tau$ in their model (0.5-15 Gyr generally, their Fig.3).
The colors and H$\alpha$ emission properties of disk and irregular galaxies have shown the general picture of their star formation history. As Kennicutt (1983) and Kennicutt et al. (1994) commented, early-type galaxies (types S0-Sb) represent systems which formed most of their gas into stars on timescales much less then the Hubble time, while the disks of late-type systems (Sc-Im) have formed stars at roughly a constant rate since they formed. They parameterized the star formation history as an exponentially declining star formation rate: [*SFR(t)*]{} = $R_0$ [*e*]{}$^{-t/\tau}$, where $\tau$ was adopted as 0-15 Gyr, and the 1/$\tau$=0 case is corresponding to the constant star formation case (Kennicutt 1983). As Zackrisson et al. (2005) analyzed, the short burst scenario (with star formation ending very abruptly) can be ruled out for LSBGs since it will predict too high EW(H$\alpha$). And in the scenarios including constant or increasing star formation rates over cosmological time scales, it still predict too high EW(H$\alpha$) to reach the observations. This may nonetheless be remedied if the slope of the IMF is significantly more bottom-heavy or the upper mass limit substantially lower than typically assumed.
For LSBGs, Vorobyov et al. (2009) used a simple model of sporadic star formation in which the individual star formation sites (SFSs) are distributed randomly throughout the galactic disk, which is reasonable for LSBGs. They further point out, there is evidence that star formation in blue LSBGs does not proceed at a near-constant rate. Their own numerical simulations and modelling by Zackrisson et al. (2005) indicated that the SFR should be declining with time to reproduce the observed EW(H$\alpha$) in LSBGs. Therefore, they assumed an exponentially decreasing SFR in their modelling. Moreover, van den Hoek et al. (2000) found that most of the LSBGs in their sample belonged to the group of late-type galaxies for which exponentially decreasing SFR models were in good agreement with the observations. Haberzettl, Bomans, Dettmar (2005) also pointed out that the spectra of their sample galaxies were all represented by an exponential decreasing star formation rate. Therefore, it is reasonable for us to adopt the exponential decreasing SFR for our LSB sample galaxies by following the discussions above. We also adopt the same form of SFR for the HSBGs to be consistent. Even though, we have tried to use the constant SFR instead in PEGASE to fit the SEDs of all our sample galaxies. The resulted ages of galaxies become a bit larger than those with exponential decreasing SFR. It is $\sim$5-6 Gyr generally. This could be understood from the model prediction, i.e., for the constant SFR case, it may need longer time to assemble the equivalent amount of stellar populations.
The degeneracy between age and metallicity is often a problem in stellar population analysis of galaxies. It is not easy to degenerate them. In our fittings, we obtain the metallicity Z of galaxies from 0.0001 to 0.04 in a wide range for most of the sample galaxies. As we know, the metallicity presented by PEGASE is the value averaged on the Simple Stellar Populations (SSPs) with various metallicities, and the average is sensitive to the dominant populations among all the SSPs. Moreover, by adding UV and NIR photometric data to the optical data, the degeneracy between age and metallicity could be efficiently broken as discussed in Sect. 4.1. Thus the dominate stellar population ages derived from PEGASE on SED fittings here should be robust.
The derived ages of galaxies discussed here represent the time-scale since their dominant stellar populations formed. This could be different from the stellar mean ages of the galaxies. PEGASE model also provides the mean ages of the stars averaged on the bolometric luminosity. We check their values by following the items in Table1, and found that these mean ages of stars are about 0.9 Gyr younger generally. This could be understood from the model prediction since the stars are assumed to form later than that time corresponding to the ages of galaxies.
Conclusions {#sect:conclusion}
===========
We summarize and conclude our work in four items.
1\) A much large sample (1,802) of nearly face-on disk LSBGs are studied their ages by fitting their multiwavelength SEDs from FUV to NIR using the EPS model PEGASE. The exponential decreasing SFR is adopted. The ages of LSBGs spread in a wide range. Most of them have ages of 1-5 Gyr generally no matter the constant or varying dust extinction $A_V$ are considered (the varying dust extinction results in $\sim$0.4 Gyr older ages than the constant one). This age range is consistent with most of previous studies on smaller sample of LSBGs. In addition, a part of the LSBGs ($\sim$2-3%) have larger ages as 5-8 Gyrs. These derived ages are also consistent with some earlier works. We should notice that if a constant SFR is adopted instead, the derived ages of sample galaxies will become a bit larger, $\sim$5-6 Gyr generally, which is for a further check although it has been commented that the exponential decreasing SFR is favored for LSBGs.
2\) A large sample (5,886) of nearly face-on disk HSBGs are also selected and studied using same procedure for comparisons. The results show that most of these HSBGs also have ages 1-5 Gyr, but with $\sim$3% having larger ages as 5-8 Gyrs, which are not much different from the LSBGs. However, the HSBGs are about 0.2 Gyr younger than the LSBGs, which may indicate that the HSBGs have more recent star forming activities than the LSBGs.
3\) Four sub-samples are further selected for checking the incompleteness effects: the volume-limited one selected from the $M_r-z$ plane ($z<0.1$ and those brighter ones than the corresponding $M_r$), the pure disc one with $fracDev_r$=0, the $r \leq 16$ one for testing 2MASS completeness, and further the pure disc with $r \leq 16$ one. All the four sub-samples show quite similar results to the whole sample, i.e., the derived ages of the sample galaxies are quite similar to each other, and only have small discrepancy, 0.02-0.13 Gyr.
4\) The similar ages between LSBGs and HSBGs could be consistent with the metallicity and stellar population analysis of them (Liang et al. 2010; Chen et al. 2010, in preparation). These suggest that LSBGs and HSBGs have not had dramatically different star formation and chemical enrichment histories.
We thank our referee for the very valuable comments and suggestions, and very efficient reviewing, which help in improving well our work. We also thank our Editor, Prof. Changbom Park, for his very efficient management on our paper. We thank Philippe Prugniel, Zhengyi Shao, Ruixiang Chang, Hector Flores, Myriam Rodrigues, Mathieu Puech, Chantal Balkowski, Rodney Delgado, Sylvain Fouquet, Jianling Wang and Xuhui Han for helpful discussions. This work was supported by the Natural Science Foundation of China (NSFC) Foundation under Nos.10933001, 10973006, 10973015, 10673002; the National Basic Research Program of China (973 Program) Nos.2007CB815404,06; and the Young Researcher Grant of National Astronomical Observatories, Chinese Academy of Sciences. We thank the wonderful SDSS, 2MASS and GALEX database, and the wonderful NYU-VAGC, CASJobs and MPA/JHU/SDSS.
[99]{}
Auld, R., de Blok, W. J. G., Bell, E., Davies, J. I., 2006, MNRAS, 366, 1475
Bell, E. F., Barnaby, D., Bower, R. G., et al. 2000, MNRAS, 312, 470
Bergvall, N., Zackrisson, E., Caldwell, B., 2009, MNRAS, 405, 2697
Blanton, M. R. et al. 2003, AJ, 125, 2348
Blanton, M. R. et al. 2005, AJ, 129, 2562
Blanton, M. R., & Roweis, S. 2007, AJ, 133, 734
Boissier, S. et al. 2003, MNRAS, 343, 653
Boissier, S. & Prantzos, N., 2000, MNRAS, 312, 398
Bothun G. D., Impey C., & McGaugh S., 1997, PASP, 109, 745
Caldwell, N., Rose, J. A., Concannon, K. D., 2003, AJ, 125, 2891
Chang, R. X., Shen, S. Y., Hou, J. L., Shu, C. G., Shao, Z. Y., 2006, MNRAS, 372, 199
Chen, X. Y. Liang, Y. C., Hammer, F. et al. 2009, A&A, 495, 457
Chen, X. Y. Liang, Y. C., Hammer, F. et al. 2010, A&A, 515, A101
de Blok, W. J. G., McGaugh, S. S., van der Hulst, J. M., 1996, MNRAS 283, 18
de Blok W. J. G., van der Hulst J. M., 1998, A&A, 335, 421
de Jong, R. S., 1996, A&A, 313, 377
Disney M. J., 1976, Nature, 263, 573
Fan, Z., Ma, J., de Grijs, R. et al. 2006, MNRAS, 371, 1648
Fioc, M., & Rocca-Volmerange, B. 1997, A&A, 326, 950
Freeman K. C., 1970, ApJ, 160, 811
Gavazzi, G. et al. 2002, ApJ, 576, 135
Gunn, J. E. et al. 1998, AJ, 125, 2348
Haberzettl, L., Bomans, D. J., & Dettmar, R.-J. 2005, in The Evolution of Starbursts: The 331st Wilhelm and Else Heraeus Seminar. AIP Conf. Proc., 783, 296
Hewett, P. C., Warren, S. J., Leggett, S. K., Hodgkin, S. T., 2006, MNRAS,367, 454
Hogg, D. W. et al., 2004, ApJ, 601, L29
Impey C. D. & Bothun G. D., 1997, ARA&A, 35, 267
Jarrett, T. H., Chester, T., Cutri, R., Schneider, S., Skrutskie, M., Huchra, J. P., 2000, AJ, 119, 2498
Jiang, Linhua, Ma, Jun, Zhou, Xu, Chen, Jiansheng, Wu, Hong, Jiang, Zhaoji, 2003, AJ, 125, 727
Jimenez, R., Padoan, P.,Matteucci F., & Heavens, A. F. 1998, MNRAS, 299, 123
Kaviraj, S., Rey, S.-C., Rich, R. M., Yoon, S.-J., Yi, S. K., 2007, MNRAS, 381L, 74
Kennicutt, R. C. 1983, ApJ, 272, 54
Kennicutt, R. C. 1992, ApJ, 388, 310
Kennicutt, R. C., Tamblyn, P., Congdon, C. W. 1994, ApJ, 435, 22
Kinman, T. D., Salim, S., Clewley, L., 2007, ApJ, 662, 111
Kurucz R.L. 1992, in: The stellar populations of galaxies (IAU Symp.149), Barbuy B., Renzini A. (eds.), Dordrecht, Kluwer
Lee, J. H., Lee, M. G., Park, C., Choi, Y. Y., 2010, MNRAS, 401, 1804
Lejeune, Th., Cuisinier, F., & Buser, R. 1997, A&AS, 125, 229
Li, Jiu-Li; Zhou, Xu; Ma, Jun; Chen, Jian-Sheng 2004a, ChJAA, 4, 143
Li, J., Ma, J., Zhou, X., Jiang, Z., Yang, Y., Chen, J., 2004b, A&A, 420, 89
Li, Z., Han, Z., Zhang, F., 2007, A&A, 464, 853
Liang, Y. C., Hammer, F., Yin, S. Y. et al. 2007, A&A, 473, 411
Liang, Y. C., Zhong, G. H., Hammer, F. et al. 2010, MNRAS, in press, arXiv: 1004.3683
Ma, J., Zhou, X., Chen, J. S. et al. 2002a, A&A, 385, 404
Ma, J., Zhou, X., Chen, J. S. et al. 2002b, AJ, 123, 3141
Ma, J., Yang, Y. B. Burstein, D. et al. 2007, ApJ, 659, 359
Ma, J., de Grijs, R., Fan, Z. et al. 2009a, RAA, 9, 641
Ma, J., Fan, Z., de Grijs, R. et al. 2009b, AJ, 137, 4884
MacArthur, L. A., Courteau, S., Bell, E., Holtzman, J. A., 2004, ApJS, 152, 175
Martin, D. C., et al. 2005, ApJ, 619, L1
Mattsson, L., Caldwell, B., Bergvall, N., 2008, ASPC, 396, 155 (arXiv:0712.0345)
McGaugh, S. S. 1994, ApJ, 426, 135
McGaugh, S. S., Schombert, J. M. & Bothun, G. D., 1995, AJ, 109, 2019
McIntosh, D. H., Bell, E. F., Weinberg, M. D., Katz, N., 2006, MNRAS, 373, 1321
Molla, M., Vilchez, J. M., Gavilan, M., Diaz, A. I., 2006, MNRAS, 372, 1069
Morrissey, P. et al., 2007, ApJS, 173, 682
O’Neil K., Bothun G. D., Schombert J., Cornell M. E., Impey C. D., 1997, AJ, 114, 2448
Padoan, P., Jimenez, R., & Antonuccio-Delogu, V. 1997, ApJ, 481, L27
Rocca-Volmerange B., & Guiderdoni B. 1988, A&AS 75, 93
Rey, S. -C. et al., 2007, ApJS, 173, 643
Ronnback, J., & Bergvall, N. 1994, A&AS 108, 193
Salpeter, E. E. 1955, ApJ, 121, 161
Sawicki, M., & Yee, H. K. C. 1998, AJ, 115, 1329
Schaerer, D., & de Koter, 1997, A&A, 322, 598
Schlegel, D. J., Finkbeiner, D. P., Davis, M., 1998, ApJ, 500, 525
Schombert, J. M., McCaugh, S. S., & Eder, J. A. 2001, AJ, 121, 2420
Shao, Z. Y., Xiao, Q. B., Shen, S. Y., Mo, H.J., Xia, X. Y., Deng, Z. G., 2007, ApJ, 659, 1159
Sprayberry, D, Impey, C. D., Bothun, G. D., Irwin, M. J., 1995, AJ, 109, 558
Strauss, M. et al., 2002, AJ, 124, 1810
Stoughton, C. et al., 2002, AJ, 123, 485
Terlevich, A. I., Forbes, D. A., 2002, MNRAS, 330, 547
van den Hoek, L. B., de Blok, W. J. G., van der Hulst, J. M., & de Jong, T. 2000, A&A 357, 397
van der Hulst, J. M., Skillman, E. D., Smith, T. R. et al. 1993, AJ, 106, 548
Vorobyov, E. I., Shchekinov, Yu., Bizyaev, D., Bomans, D., Dettmar, R. -J., 2009, A&A, 505, 483
Wang, S., Fan, Z., Ma, J. et al. 2010, AJ, 139, 1438
Worthey, G., 1994, ApJS, 95, 107
Wu, H., Shao, Z. Y., Mo, H. J., Xia, X. Y., Deng, Z. G., 2005, ApJ, 622, 244
York, D., et al. 2000, AJ, 120, 1579
Zackrisson, E., Bergvall, N., & Ostlin, G. 2005, A&A, 435, 29
Zhong G., Liang Y. C., Liu F. S. et al. 2008, MNRAS, 391, 986
Zwicky F., 1957, in Morphological Astronomy (New York, Springer-Verlag)
\[lastpage\]
[^1]: http://www.sdss.org
[^2]: http://www.ipac.caltech.edu/2mass/
[^3]: see http://irsa.ipac.caltech.edu/applications/Gator/
[^4]: http://galex.stsci.edu/GR4/
[^5]: http://mastweb.stsci.edu/gcasjobs/
[^6]: http://galex.stsci.edu/doc/CASJobsXTutorial.htm
[^7]: http://galexgi.gsfc.nasa.gov/docs/galex/Documents/GALEXPipelineDataGuide.pdf
[^8]: http://sdss.physics.nyu.edu/vagc/
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'In this work we investigate the impact of intrinsic charm on the prompt atmospheric neutrino flux. The color dipole approach to heavy quark production is generalized to include the contribution of processes initiated by charm quarks. The prompt neutrino flux is calculated assuming the presence of intrinsic charm in the wave function of the projetile hadron. The predictions are compared with previous color dipole results which were obtained taking into account only the process initiated by gluons. In addition, we estimate the atmospheric (conventional + prompt) neutrino flux and compare our predictions with the ICECUBE results for the astrophysical neutrino flux. Our results demonstrate that the contribution of the charm quark initiated process is non - negligible and that the prompt neutrino flux can be enhanced by a factor $\approx$ 2 at large neutrino energies if an intrinsic charm component is present in the proton wave function.'
author:
- 'A.V. Giannini$^\ddag$, V.P. Gonçalves$^\S$ and F.S. Navarra$^\ddag$'
title: Intrinsic charm contribution to the prompt atmospheric neutrino flux
---
Introduction
============
Astrophysical neutrinos detected by the IceCube Observatory mark the beginning of neutrino astronomy [@IceCube_Science; @Aartsen:2014gkd; @Aartsen:2016xlq], which allows us to study very high energy physical processes in the Universe [@review_icecube; @annual_mezsaros]. The atmospheric neutrino flux is produced in the atmosphere of the Earth (through cosmic-ray interactions with nuclei) and it is the main background in studies of cosmic neutrinos. During the last years, several neutrino observatories [@Abbasi:2010ie; @Aartsen:2012uu; @Adamson:2012gt; @Fukuda:1998ub] have studied the high-energy neutrino flux. The experimental data indicate that, at low energies ($E_{\nu}\lesssim 10^5$ GeV), the measured neutrino flux is dominated by atmospheric neutrinos which come from the decay of light mesons (pions and kaons). This flux is called [*conventional*]{} atmospheric neutrino flux [@Honda:2006qj; @Barr:2004br; @Gaisser:2014pda]. On the other hand, in the energy range 10$^{5}$ GeV $< E_{\nu} <$ 10$^{7}$ GeV, the [*prompt*]{} atmospheric neutrino flux (resulting from the decay of heavy quark hadrons) becomes important [@ingelman; @Martin:2003us; @sigl]. The precise knowledge of this contribution is crucial for the determination of the cosmic neutrino flux.
The calculation of the prompt atmospheric neutrino flux has been subject of intense activity [@enb_stasto1; @sigl; @rojo1; @enb_stasto2; @rojo2; @halzen; @laha; @prosa; @sigl2]. Since heavy quarks are an important source of neutrinos, it became necessary to describe their production with better accuracy. Different treatments (and approximations) of heavy quark production at high energies and of the QCD dynamics at small values of the Bjorken - $x$ variable were proposed. Although the LHC data on the prompt heavy quark cross sections (see e.g. Refs. [@Aaij:2013mga; @Aaij:2015bpa]) helped us to improve the description of heavy meson production at forward rapidities and significantly reduced some of the theoretical uncertainties, the predictions obtained by different groups can still differ by a factor $\ge 2$. This large uncertainty is due to the fact that the main contribution to the prompt neutrino flux comes from heavy quark production in a kinematical range which is not currently probed by the LHC. In Ref. [@vicantoni] the authors have presented a detailed analysis of the kinematical domains in which charm and prompt atmospheric neutrino production from cosmic rays are relevant for the IceCube experiment. They explored the sensitivity of the corresponding neutrino flux and of the charm cross section to the cuts on the maximal $pp$ c.m. energy, to the longitudinal momentum fraction in the target and projectile, to the Feynman $x_F$ and to $p_T$ values included in the calculation. They have demonstrated that in order to address the production of high-energy neutrinos ($E_{\nu} >$ 10$^7$ GeV) one needs to know the charm production cross section at energies larger than those available at the LHC as well as the parton/gluon distributions in the region 10$^{-8}$ $< x <$ 10$^{-5}$, which are not presently available in collider measurements. Consequently, the presence of new effects, which are expected to contribute to small values of $x$ and/or large values of the $x_F$, cannot be excluded by current data.
The production of heavy states at high energies is expected to be sensitive to non - linear effects of QCD dynamics [@kt_hq; @hq_sat; @vicmag; @cgn; @wata; @armesto], which are predicted to be enhanced at forward rapidites [@cgc]. The cross sections at forward rapidities are dominated by collisions of projectile partons with large light cone momentum fractions ($x_p \rightarrow 1$) with target partons carrying a very small momentum fraction ($x_t \ll 1$). Consequently, small-$x$ effects coming from the non-linear aspects of QCD and from the physics of the Color Glass Condensate (CGC) [@cgc] are expected to appear and the usual factorization formalism is expected to break down [@hq_sat]. Recent results [@dhj; @buw; @ptmedio.nois] indicate that the CGC formalism provides a satisfactory description of the experimental data on particle production at forward rapidities. Additionally, large - $x$ effects in the projectile are also expected to contribute to heavy quark production at forward rapidities. One of the possible new effects is the presence of intrinsic heavy quarks in the hadron wave function (For recent reviews see, e.g. Refs. [@review_brod_adv; @review_brod_prog]). Heavy quarks in the sea of the proton can be perturbatively generated by gluon splitting. Quarks generated in this way are usually denoted [*extrinsic*]{} heavy quarks. In contrast, the [*intrinsic*]{} heavy quarks have multiple connections to the valence quarks of the proton and thus are sensitive to its nonperturbative structure.
The existence of the intrinsic charm (IC) component was first proposed long ago in Ref. [@bhps] (see also Ref. [@hal]) and since then other models of IC have been discussed [@pnndb; @wally99]. Its existence implies a large enhancement of the charm distribution at large $x$ ($> 0.1$) in comparison to the extrinsic charm prediction. Moreover, due to the momentum sum rule, the gluon distribution is also modified by the inclusion of intrinsic charm. The intrinsic charm (IC) component of the proton wave function was included in recent global analyses, performed by different theoretical groups, in order to constrain the parton distributions. Although the resulting IC distributions are compatible with the world data, the amount of IC in the proton wave function is still subject of intense debate [@wally16; @brod_letter] and has motivated a large number of phenomenological studies (See e.g. Refs. [@russos; @bailas; @biw16]).
One of the most direct consequences of the intrinsic charm component is that it gives rise to heavy mesons with large fractional momenta relative to the beam particles, affecting the $x_F$ and rapidity distribution of charmed particles. This aspect was explored e.g. in Refs. [@ingelman; @kniehl; @npanos; @prdnos]. In particular, in Ref. [@prdnos] we studied $D$ - meson production at forward rapidities taking into account non - linear effects in the QCD dynamics and the intrinsic charm (IC) component of the proton wave function. The results show that, at the LHC, the intrinsic charm component changes the $D$ rapidity distributions in a region which is beyond the coverage of the LHCb detector. At higher energies, as those probed in neutrino observatories, our results indicated that the IC component dominates the rapidity and $x_F$ distributions, with the latter being enhanced by a factor 6 – 8 in the $0.2 \le x_F \le 0.8$ range. In Ref. [@prdnos] we have pointed out that one of the basic consequences of this enhancement is the modification of the prompt neutrino flux at large energies. The main purpose of this work is to estimate the impact of the intrinsic charm component on the prompt neutrino flux taking into account non - linear effects in the QCD dynamics. It is important to emphasize that the contribution associated to a heavy quark in the initial state was not taken into account in previous studies using the color dipole formalism performed e.g. in Refs. [@ers; @enb_stasto2]. The inclusion of heavy quarks in the initial state in the color dipole approach is done here for the first time. Additionally, previous estimates of the intrinsic charm contribution to the neutrino flux [@laha; @halzen] have made use of phenomenological models of the $x_F$ - distribution in $\Lambda_c$ and $D$ production, constrained only by the scarce low energy experimental data and then extrapolated to higher energies. In contrast, in our analysis the basic ingredients of the calculations, the charm/gluons PDF’s and the dipole - hadron amplitude, have been constrained by the most recent LHC data and non - linear effects in the QCD dynamics are taken into account.
This paper is organized as follows. In the next Section we present a brief review of the formalism to calculate the charm production at forward rapidities, as well as we discuss the impact of an intrinsic charm component in the Feynman $x_F$ - distribution. In Section \[res\] we present our results for the prompt neutrino flux. The contribution of the different components is analyzed and the impact of the intrinsic component is estimated. In particular, the predictions for the atmospheric (conventional + prompt) neutrino flux is compared with the recent ICECUBE results for the flux of the astrophysical neutrinos. Finally, in Section \[conc\] we summarize our main results and conclusions.
Formalism {#form}
=========
In the analysis performed in this paper, we will closely follow the procedure described in detail in Refs. [@vicantoni; @prdnos]. In what follows we will only present the main aspects of the formalism and refer the reader to Refs. [@vicantoni; @prdnos] for more details. As in Ref. [@vicantoni], we will calculate the prompt neutrino flux using the semi-analytical $Z$-moment approach, proposed many years ago in Ref. [@ingelman] and discussed in detail e.g. in Refs. [@ers; @rojo1]. In this approach, a set of coupled cascade equations for the nucleons, heavy mesons and leptons (and their antiparticles) fluxes is solved, with the equations being expressed in terms of the nucleon-to-hadron ($Z_{NH}$), nucleon-to-nucleon ($Z_{NN}$), hadron-to-hadron ($Z_{HH}$) and hadron-to-neutrino ($Z_{H\nu}$) $Z$-moments. These moments are inputs in the calculation of the prompt neutrino flux associated with the production of a heavy hadron $H$ and its decay into a neutrino $\nu$ in the low- and high-energy regimes. We will focus on vertical fluxes and will assume that the cosmic ray flux $\phi_N$ can be described by a broken power-law spectrum [@prs] or by the H3a spectrum proposed in Ref. [@gaisser], with the incident flux being represented by protons. Moreover, we will assume that the charmed hadron $Z$-moments can be expressed in terms of the charm $Z$-moment as follows: $Z_{pH} = f_H \times Z_{pc}$, where $f_H$ is the fraction of charmed particle which emerges as a hadron $H$. As in Ref. [@ers], we will assume that $f_{D^0} = 0.565$, $f_{D^+} = 0.246$, $f_{D_s^+} = 0.080$ and $f_{\Lambda_c} = 0.094$.
The charm $Z$-moment at high energies can be expressed by $$\begin{aligned}
Z_{pc} (E) = \int_0^1 \frac{dx_F}{x_F} \frac{\phi_p(E/x_F)}{\phi_p(E)}
\frac{1}{\sigma_{pA}(E)} \frac{d\sigma_{pA \rightarrow charm}(E/x_F)}{dx_F} \,\,,
\label{eq:zpc}\end{aligned}$$ where $E$ is the energy of the produced particle (charm), $x_F$ is the Feynman variable, $\sigma_{pA}$ is the inelastic proton-Air cross section, which we assume to be given as in Ref. [@sigl], and $d\sigma/dx_F$ is the differential cross section for the charm production, which we assume to be given by $d\sigma_{pA \rightarrow charm}/dx_F = 2
\, d\sigma_{pA \rightarrow c \bar{c}}/dx_F$. One of the main inputs in the $Z$-moment approach is the Feynman $x_F$ distribution of the heavy quarks produced in hadronic collisions. As discussed in Ref. [@vicantoni], the main contribution to the prompt neutrino flux comes from large values of $x_F$, that are associated with heavy quark production at forward rapidities. Moreover, as the production of neutrinos at a given neutrino energy, $E_{\nu}$, is determined by collisions of cosmic rays with nuclei in the atmosphere at energies that are a factor of order 100-1000 larger, the prompt neutrino flux measured in the kinematical range that is probed by the IceCube Observatory and future neutrino telescopes is directly associated with the treatment of the heavy quark production cross section at high energies. Following Ref. [@prdnos], we will use the approaches developed in Refs. [@nnz; @boris] and [@npanos] for the treatment of heavy quark production induced by gluon - gluon and charm - gluon interactions, respectively, and we will take into account non - linear effects in the QCD dynamics. The contribution of the $q \bar{q} \rightarrow c \bar{c}$ subprocess will be disregarded, since it is negligible at the energies and rapidities relevant for the prompt neutrino flux [@vicantoni].
The basic idea in Ref. [@prdnos] is that at forward rapidities, the projectile (dilute system) evolves according to the linear DGLAP dynamics and the target (dense system) is treated using the CGC formalism. In this approach, the charm $x_F$ - distribution will be determined by the contribution of the two diagrams presented in Fig. \[Fig:diagrama\], which are associated to the gluon and charm - initiated processes. The first contribution, represented in Fig. \[Fig:diagrama\] (a), can be estimated using the color dipole picture [@nnz; @boris], which implies that the rapidity distribution for the charm production in a $h_1 h_2$ collision can be expressed as follows $$\frac{d\sigma}{dy} = x_1g^{h_1}(x_1,\mu_F^2) \,
\sigma (gh_2 \rightarrow \{c\bar{c} \}X)\,\,,
\label{dsdy_gluons}$$ where $x_1g_{p}(x_1,\mu _F)$ is the projectile gluon distribution, the cross section $\sigma (gh_2 \rightarrow \{c\bar{c}\} X)$ describes charm production in a gluon - nucleon interaction, $y$ is the rapidity of the pair and $\mu_F$ is the factorization scale. Moreover, the cross section of the process $g + h_2 \rightarrow c \bar{c} X$ is given by: $$\sigma(g h_2 \rightarrow\{c\bar{c}\}X) = \int _0^1 d \alpha \int d^2{\mbox{\boldmath $r$}}\,\,
\vert \Psi _{g\rightarrow c\bar{c}} (\alpha,{\mbox{\boldmath $r$}})\vert ^2
\,\, \sigma^{h_2} _{c\bar{c}g}(\alpha , {\mbox{\boldmath $r$}})
\label{sec1}$$ where $\alpha$ ($\bar{\alpha} \equiv 1 - \alpha$) is the longitudinal momentum fraction carried by the quark (antiquark), ${{\mbox{\boldmath $r$}}}$ is the transverse separation of the pair, $\Psi _{g\rightarrow c\bar{c}}$ is the light-cone (LC) wave function of the transition $g \rightarrow c \bar{c} $ (which is calculable perturbatively and is proportional to $\alpha_s$) and $ \sigma^{h_2}_{c\bar{c}g}$ is the scattering cross section of a color neutral quark-antiquark-gluon system on the hadron target $h_2$ [@nnz; @boris]. The three - body cross section is given in terms of the dipole - nucleon cross section $\sigma _{c\bar{c}}$ as follows: $$\sigma^{h_2}_{c\bar{c}g}(\alpha , {\mbox{\boldmath $r$}}) = \frac{9}{8}[\sigma _{c\bar{c}}(\alpha {\mbox{\boldmath $r$}}) + \sigma _{c\bar{c}}(\bar{\alpha} {\mbox{\boldmath $r$}})]
- \frac{1}{8}\sigma _{c\bar{c}}({\mbox{\boldmath $r$}})\,\,.
\label{sec2}$$ Finally, the dipole - nucleon cross section can be expressed in terms of the forward scattering amplitude ${\cal{N}} (x,{\mbox{\boldmath $r$}})$, which is determined by the QCD dynamics, as follows: $$\sigma_{c \bar c}(x,{\mbox{\boldmath $r$}}) = \sigma_{0} {\cal N}(x,{\mbox{\boldmath $r$}})
\label{sigzero}$$ where $\sigma_0$ is a free parameter usually determined by a fit of the HERA data. On the other hand, the contribution of the charm initiated process, represented in Fig. \[Fig:diagrama\] (b), is given by [@npanos] $$\begin{aligned}
{d\sigma \over dy} =
{1 \over (2\pi)^2} \int d^2p_T
f_{c/h_1}(x_1,\mu_F^2)\,\, \sigma_0 \, \widetilde{\cal N} \left(x_2, {p_T}\right)
\label{dNdy_quarks}\,.\end{aligned}$$ where $x_{1,2}$ is defined by $x_{1,2} = p_{T} e^{\pm y}/\sqrt{s}$, $f_{c/h_1}$ represents the projectile charm distribution and $\widetilde{\cal N} (x, {p_T})$ is the Fourier transform of the scattering amplitude ${\cal N}(x,{\mbox{\boldmath $r$}})$.
In order to estimate the contributions of the gluon and charm - initiated processes, described by Eqs. (\[dsdy\_gluons\]) and (\[dNdy\_quarks\]), we should to assume a model to describe the dipole - nucleon scattering amplitude ${\cal N}(x_2,{\mbox{\boldmath $r$}})$, as well as a parametrization for gluon and charm distributions in the projectile. The the dipole - nucleon scattering amplitude ${\cal N}(x_2,{\mbox{\boldmath $r$}})$ describes the interaction of a color dipole of size $r$ with the nucleon and involves the QCD dynamics at high energies. Such quantity contains all the information about the initial state of the hadronic wavefunction and therefore about the non-linearities and quantum effects which are characteristic of a system such as the CGC (For reviews, see e.g. [@cgc]). Formally its evolution is usually described in the mean field approximation of the CGC formalism by the BK equation [@bk]. Its analytical solution is known only in some special cases. Advances have been made in solving the BK equation numerically [@la11]. Since the BK equation still lacks a formal solution in all kinematical space, several groups have constructed phenomenological models for the dipole scattering amplitude. These models have been used to fit the RHIC, LHC and HERA data [@dhj; @buw; @dips; @IIM; @Soyez]. As in Ref. [@prdnos], in our analysis we will use the BUW model for ${\cal N}$, originally proposed in Ref. [@buw], which assumes that ${\cal{N}}$ can be modelled through a simple Glauber-like formula, $$\begin{aligned}
{\cal{N}}(x,{\mbox{\boldmath $r$}}) = 1 - \exp\left[ -\frac{1}{4} ({\mbox{\boldmath $r$}}^2 Q_s^2(x))^{\gamma (x,{\mbox{\boldmath $r$}}^2)} \right] \,\,,
\label{ngeral}\end{aligned}$$ where $Q_s(x)$ is the saturation scale and $\gamma$ is the anomalous dimension of the target gluon distribution. The speed with which we move from the non-linear regime to the extended geometric scaling regime and then from the latter to the linear regime is what differs the BUW from other phenomenological models. This transition speed is dictated by the behavior of the anomalous dimension $\gamma (x,{\mbox{\boldmath $r$}}^2)$, which is assumed in the BUW [@buw] dipole model to be given by $$\label{dip_adimension}
\gamma(\omega = p_{T}/Q_{s})_{BUW} = \gamma_{s} + (1-\gamma_s)\frac{(\omega^a-1)}{(\omega^a-1)+b}
$$ where $a$, $b$ and $\gamma_s$ are free parameters to be fixed by fitting experimental data. In Ref. [@ptmedio.nois], the original parameters of the BUW model were updated in order to make this model compatible with all existing data. In particular, the recent LHC data on light hadron production at forward rapidity are satisfactorily reproduced by the updated model.
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ -- ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
![Contributions to charm production at high energies and forward rapidities. (a) Contribution from gluon - initiated processes. (b) Contribution from charm in the initial state.[]{data-label="Fig:diagrama"}](Diagrama_gluon_IC.eps "fig:"){width="30.00000%"} ![Contributions to charm production at high energies and forward rapidities. (a) Contribution from gluon - initiated processes. (b) Contribution from charm in the initial state.[]{data-label="Fig:diagrama"}](Diagrama_charm_IC.eps "fig:"){width="30.00000%"}
(a) (b)
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ -- ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- -- ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
![(a) Comparison between the BHPS and No IC predictions for the charm (lower red curves) and gluon (upper black curves) distributions. (b) Comparison between the BHPS and No IC predictions for the Feynman $x_F$ distributions considering $pp$ collisions at $\sqrt{s} = 13$ TeV.[]{data-label="Fig:intrinsic"}](xfx_CTEQ65_no_IC_BHPS.eps "fig:"){width="45.00000%"} ![(a) Comparison between the BHPS and No IC predictions for the charm (lower red curves) and gluon (upper black curves) distributions. (b) Comparison between the BHPS and No IC predictions for the Feynman $x_F$ distributions considering $pp$ collisions at $\sqrt{s} = 13$ TeV.[]{data-label="Fig:intrinsic"}](pp_xfdNdxf_13TeV.eps "fig:"){width="45.00000%"}
(a) (b)
-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- -- ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
In order to quantify the impact of the intrinsic charm component on the projectile wave function, we will use in our calculations the next - to - leading order CTEQ 6.5 parametrization for the parton distributions [@cteq]. In Ref. [@cteq], the CTEQ group determined the shape and normalization of the IC distribution in the same way as they do for other parton species. In fact, they find several IC distributions which were compatible with the world data. In what follows we will only consider the parametrization based on the BHPS model [@bhps], obtained assuming that the average longitudinal momentum fraction carried by the charm and anticharm is $\langle x_{c\bar{c}}\rangle = 2\%$ at the initial scale of the QCD evolution. The BHPS model assumes that the nucleon light cone wave function has higher Fock states, one of them being $|q q q c \overline{c}>$. The probability of finding the nucleon in this configuration is given by the inverse of the squared invariant mass of the system. Because of the heavy charm mass, this probability as a function of the quark fractional momentum, $P(x)$, is very hard, as compared to the one obtained through the DGLAP evolution. Such assumptions are used to describe the shape of the parametrization at the initial scale of the DGLAP evolution, which is evolved for larger scales in order to contrain, by a global data fit procedure, the parton distributions of the proton. In Fig. \[Fig:intrinsic\] (a) we compare the predictions of the BHPS model with those obtained disregarding the presence of an intrinsic component (denoted No-IC hereafter). In the case of the charm distribution (lower red curves), the BHPS model predict a large enhancement of the distribution at large $x$ ($> 0.1$). In Fig. \[Fig:intrinsic\] (a) we also present the corresponding gluon distributions (upper black curves). Due to the momentum sum rule, the gluon distribution is also modified by the inclusion of intrinsic charm. In particular, the BHPS model imply a suppression in the gluon distribution at large $x$. The impact of these different models on the Feynman $x_F$ distribution for the charm production in $pp$ collisions at $\sqrt{s} = 13$ TeV is presented in Fig. \[Fig:intrinsic\] (b). We present the sum of the gluon and charm contributions, denoted “total” in the figure, as well present separately the gluon contribution. We have that in the No IC case, the distribution is dominated by the gluon initiated process. In contrast, when intrinsic charm is included, the behavior of the distribution in the intermediate $x_F$ range ($0.2 \le x_F \le 0.8$) is strongly modified. It is important to emphasize that the contribution of the gluonic component decreases in this kinematical range, as expected from the analysis of the parton distributions presented in Fig. \[Fig:intrinsic\] (a). As we will show in the next Section, these modifications in the distribution, associated the presence of an intrinsic component, has important implications in the prompt neutrino flux.
Before to present our predictions for the prompt neutrino flux in the next Section, two comments are in order. In our analysis we will consider the next - to - leading order CTEQ 6.5C parametrization for the parton distributions [@cteq], which is provided by the CTEQ group. It is important to emphasize that the CTEQ-TEA group has also performed a global analysis of the recent experimental data including an intrinsic charm component, which is available in the CT14 parametrization [@cteq14]. However, this analysis has been performed at next-to-next-to-leading order. As demonstrated in Ref. [@rauf], the predictions of the dipole approach for the charm production agree with those obtained using the collinear formalism at NLO, in the kinematical range where the equivalence between the approaches is expected. Therefore, we believe that it is more consistent to use in our calculations PDFs obtained at NLO. Second, the CTEQ 6.5C parametrization is used in this paper for consistence with our previous study [@prdnos], where we have performed a detailed analysis of the $D$ - meson production at the LHC energies considering different models for the intrinsic charm component. In particular, in Ref. [@prdnos] we have compared the BHPS predictions with those obtained using the Meson Cloud model [@pnndb; @wally99]. Such model is not considered in the CTEQ 6.6C parametrization, which is an improved version of CTEQ 6.5C one. We have verified that the modifications in our predictions for the neutrino flux are negligible if the CTEQ 6.6C parametrization is used as input in our calculations.
Results {#res}
=======
In what follows we will present our estimates for the prompt atmospheric neutrino flux using the ingredients discussed above. In our analysis, we will assume $m_c = 1.5$ GeV and $\mu^2 = 4 m_c^2$ and compare the No IC predictions with the BHPS one. Moreover, in order to estimate the contribution associated to heavy quarks in the initial state, we also will compare our full predictions, estimated taking into account gluons and quarks in the initial state, with those derived disregarding the charm contribution, as was done in previous studies using the color dipole approach.
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ -- -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
![Energy dependence of the prompt neutrino flux, normalized by a factor $E_{\nu}^3$. (a) Comparison of our predictions with the results obtained in Refs. [@ers; @enb_stasto2] using the color dipole approach. The results derived in [@enb_stasto2] considering different phenomenological saturation models for the dipole - proton scattering amplitude are represented by the cyan band. (b) Comparison of the No IC and BHPS predictions considering the gluon and charm contributions and only the gluon one in the calculation of the $x_F$ distribution.[]{data-label="Fig:fluxE3"}](comps2_withenberg.eps "fig:"){width="45.00000%"} ![Energy dependence of the prompt neutrino flux, normalized by a factor $E_{\nu}^3$. (a) Comparison of our predictions with the results obtained in Refs. [@ers; @enb_stasto2] using the color dipole approach. The results derived in [@enb_stasto2] considering different phenomenological saturation models for the dipole - proton scattering amplitude are represented by the cyan band. (b) Comparison of the No IC and BHPS predictions considering the gluon and charm contributions and only the gluon one in the calculation of the $x_F$ distribution.[]{data-label="Fig:fluxE3"}](comps2.eps "fig:"){width="45.00000%"}
(a) (b)
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ -- -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Initially, let’s consider that the primary proton flux is described by a broken power-law spectrum. Our motivation to assume this model is associated to the fact that previous estimates for the prompt neutrino flux [@ers; @enb_stasto2], which we would like compare our results, has been obtained using this model. In Fig. \[Fig:fluxE3\] we present our predictions for the energy dependence of the prompt atmospheric neutrino flux, normalized by a factor $E_{\nu}^3$. In the left panel, we compare our predictions with those obtained in Refs. [@ers; @enb_stasto2] using the color dipole approach and different phenomenological models based on saturation physics. The predictions obtained in Ref. [@ers] are denoted by ERS and those derived in Ref. [@enb_stasto2] by the cyan band. As in Refs. [@ers; @enb_stasto2] only the processes initiated by gluon were taken into account, here, for the sake of comparison, we present our predictions derived considering only this channel as well. At low energies ($E_{\nu} \le 10^6$ GeV), the ERS and No IC predictions are very similar. On the other hand, at large energies ($E_{\nu} \ge 10^7$ GeV), the No IC predictions imply a larger prompt neutrino flux than that derived in Refs. [@ers; @enb_stasto2]. Such behavior can be attributed to the model of the dipole - proton scattering amplitude used in our calculations, which differs from those used in Refs. [@ers; @enb_stasto2]. As demonstrated e.g. in Ref. [@betemps], the BUW model predicts a slower transition between the linear and non - linear regimes of the QCD dynamics than e.g. the IIM-S one [@IIM; @Soyez] used in Refs. [@ers; @enb_stasto2]. Consequently, the BUW model implies a faster growth of the heavy quark cross section with the energy, which has direct impact on the prompt neutrino flux at high energies. On the other hand, if the gluon distribution associated to the BHPS parametrization is used as input in the calculations, we find that the resulting predictions are suppressed in comparison to the No IC one. As discussed before, in the BHPS parametrization the IC component is taken into account and this implies that a larger amount of the proton momentum is carried by the charm quarks. Due to the momentum sum rule, the amount carried by the other partons, in particular the gluons, will be reduced. Therefore, the gluon distributions associated to the parton parametrizations which include an intrinsic component are, in general, smaller than those derived disregarding this component \[See Fig. \[Fig:intrinsic\] (a)\]. Such reduction explains the result observed in the left panel of Fig. \[Fig:fluxE3\].
In Fig. \[Fig:fluxE3\] (b) we estimate the impact of the charm initiated processes on the energy dependence of the prompt neutrino flux. As expected, the inclusion of this new channel of charm production leads to an enhancement of the flux. The magnitude of this enhancement depends on the details of the model of the charm distribution. When the intrinsic charm component is disregarded \[the No IC (gluons + charm) curve in Fig. \[Fig:fluxE3\]\] the impact of the charm initiated processes is small. Such result is expected, since the magnitude of the extrinsic charm distribution for small values of $\mu^2$ is small. On the other hand, if the intrinsic component is taken into account, we have a large enhancement of the prompt flux.
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- -- -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
![(a) Comparison between the BPL and H3a models for the all - nucleon primary cosmic ray spectra. (b) Energy dependence of the prompt neutrino flux, normalized by a factor $E_{\nu}^3$, obtained using the BPL and H3a models as input in the calculations. []{data-label="Fig:h3a"}](comp_flux_primarios.eps "fig:"){width="45.00000%"} ![(a) Comparison between the BPL and H3a models for the all - nucleon primary cosmic ray spectra. (b) Energy dependence of the prompt neutrino flux, normalized by a factor $E_{\nu}^3$, obtained using the BPL and H3a models as input in the calculations. []{data-label="Fig:h3a"}](compIC_fluxes.eps "fig:"){width="45.00000%"}
(a) (b)
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- -- -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
One important question is the dependence of our results on the model assumed for the primary nucleon spectrum. Although the broken power - law (BPL) approximation of the cosmic ray nucleon flux has been used to evaluate the prompt neutrino flux in almost all early references, more recent publications also present the predictions obtained considering a modern description of the primary spectrum proposed by Gaisser in Ref. [@gaisser]. In what follows we also will estimate the prompt flux using the H3a spectrum, which assumes that the spectrum is given by a composition of 3 populations and 5 representative nuclei, with the set of parameters determined by a global fit of the cosmic ray data [@gaisser]. A comparison between the BPL and H3a spectra is performed in Fig. \[Fig:h3a\] (a). We have that for primary energies in the range $10^4 \lesssim E \lesssim 10^6$ GeV, the H3a spectrum is larger than the BPL one. On the other hand, for $E \gtrsim 10^6$ GeV, the H3a spectrum is smaller than the BPL one, and a structure associated to its composition is present. In Fig. \[Fig:h3a\] (b) we present our predictions for the energy dependence of the prompt neutrino flux obtained using the BPL and H3a models for the primary nucleon flux. As expected from Fig. \[Fig:h3a\] (a), the H3a predictions are smaller than the BPL one at large neutrino energies. However, our results indicate a large impact of the intrinsic charm also is present if the H3a spectrum is considered.
A more detailed estimate of the contribution of the charm initiated process and the intrinsic charm component is obtained by the analysis of Fig. \[Fig:ratios\]. In Fig. \[Fig:ratios\] (a) we present our results for the ratio between the flux calculated considering the gluon and charm channels ($\Phi_{g+c}$) and that derived assuming only the gluon channel ($\Phi_{g}$). The results associated to the BPL and H3a models are presented for comparison. When the IC component is absent, the impact of the charm channel is $\approx 10 \%$ and almost energy independent. In contrast, if an IC component is present, the impact increases with the neutrino energy and becomes larger than $200 \%$ at large energies, with the H3a predictions being slightly larger than BPL one for $E_{\nu} \lesssim 10^7$ GeV. Another way to estimate the impact of the IC component is to calculate the ratio between the flux derived assuming the presence of this component ($\Phi_{BHPS}$) and the flux calculated disregarding this component ($\Phi_{No IC}$). The results are presented in Fig. \[Fig:ratios\] (b). They indicate that if only the gluon channel is considered, the presence of the IC component implies a reduction of the prompt neutrino flux of $\approx 10 \%$. On the other hand, when the charm channel is included, we predict an enhancement of $\approx 160 \%$ in the flux. These results strongly suggest that a generalized treatment of heavy quark production, taking into account charm initiated processes, is crucial to obtain realistic predictions of the prompt flux using the color dipole approach, especially if an intrinsic charm component is present in the proton wave function.
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- -- ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
![Energy dependence of the ratios between prompt neutrino fluxes. (a) Ratio between the flux calculated considering the gluon and charm channels ($\Phi_{g+c}$) and that derived assuming only the gluon channel ($\Phi_{g}$). (b) Ratio between the flux derived assuming the presence of this component ($\Phi_{BHPS}$) with that calculated disregarding this component ($\Phi_{No IC}$). []{data-label="Fig:ratios"}](ratiogluonquark2.eps "fig:"){width="45.00000%"} ![Energy dependence of the ratios between prompt neutrino fluxes. (a) Ratio between the flux calculated considering the gluon and charm channels ($\Phi_{g+c}$) and that derived assuming only the gluon channel ($\Phi_{g}$). (b) Ratio between the flux derived assuming the presence of this component ($\Phi_{BHPS}$) with that calculated disregarding this component ($\Phi_{No IC}$). []{data-label="Fig:ratios"}](ratioBHPS2.eps "fig:"){width="45.00000%"}
(a) (b)
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- -- ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- -- ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
![Energy dependence of the prompt neutrino flux, normalized by a factor $E_{\nu}^2$, obtained considering the BPL model for the primary nucleon spectrum. (a) Comparison of our results with the predictions for the conventional neutrino flux derived in Ref. [@Honda:2006qj] and the astrophysical neutrino flux obtained in Ref. [@Aartsen:2016xlq], which is represented by the cyan band. (b) Comparison between the predictions for the atmospheric (conventional + prompt) neutrino fluxes and the astrophysical one.[]{data-label="Fig:fluxe2"}](fluxE2.eps "fig:"){width="45.00000%"} ![Energy dependence of the prompt neutrino flux, normalized by a factor $E_{\nu}^2$, obtained considering the BPL model for the primary nucleon spectrum. (a) Comparison of our results with the predictions for the conventional neutrino flux derived in Ref. [@Honda:2006qj] and the astrophysical neutrino flux obtained in Ref. [@Aartsen:2016xlq], which is represented by the cyan band. (b) Comparison between the predictions for the atmospheric (conventional + prompt) neutrino fluxes and the astrophysical one.[]{data-label="Fig:fluxe2"}](fluxE2_soma.eps "fig:"){width="45.00000%"}
(a) (b)
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- -- ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- -- ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
![(a) Comparison between the BPL and H3a predictions for the energy dependence of the atmospheric (conventional + prompt) neutrino fluxes, normalized by a factor $E_{\nu}^2$. (b) Comparison between the BHPS predictions for the atmospheric (conventional + prompt) neutrino flux with the results derived by Halzen and Willie (HW) [@halzen] and by Laha and Brodsky (LB) [@laha]. Results derived using the H3a spectrum. []{data-label="Fig:fluxe2h3a"}](fluxE2_soma_gaisser3.eps "fig:"){width="45.00000%"} ![(a) Comparison between the BPL and H3a predictions for the energy dependence of the atmospheric (conventional + prompt) neutrino fluxes, normalized by a factor $E_{\nu}^2$. (b) Comparison between the BHPS predictions for the atmospheric (conventional + prompt) neutrino flux with the results derived by Halzen and Willie (HW) [@halzen] and by Laha and Brodsky (LB) [@laha]. Results derived using the H3a spectrum. []{data-label="Fig:fluxe2h3a"}](fluxE2_soma_gaisser2_Comp2.eps "fig:"){width="45.00000%"}
(a) (b)
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- -- ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Using its six-year experimental data, the ICECUBE Collaboration has recently [@Aartsen:2016xlq] estimated the astrophysical neutrino spectrum with the help of an unbroken power - law model. In order to compare our predictions with the neutrino flux derived in Ref. [@Aartsen:2016xlq], in Fig. \[Fig:fluxe2\] we present our results for the neutrino flux, normalized by a factor $E_{\nu}^2$, obtained considering the BPL model for the primary nucleon spectrum. The results from Ref. [@Aartsen:2016xlq] are represented by the gray band. For the sake of comparison, we also present the predictions for the conventional atmospheric neutrino flux derived in Ref. [@Honda:2006qj]. In Fig. \[Fig:fluxe2\] (a) we compare our predictions for the prompt neutrino flux, considering different models and channels, with the conventional and astrophysical predictions. When the IC component and/or the charm channel is disregarded, the prompt contribution dominates for $E_{\nu} \gtrsim 6 \times 10^5$ GeV. On the other hand, if the IC component and the charm channel are taken into account, this transition occurs at $E_{\nu} \gtrsim 4.5 \times 10^5$ GeV. The most important aspect is that, although the astrophysical neutrino flux is dominant at large energies, the magnitude of the background associated to the prompt flux is strongly dependent on the presence or absence of the intrinsic component in the proton wave function, in agreement with the result obtained in Ref. [@laha]. This dependence is more visible in the right panel of Fig. \[Fig:fluxe2\], where we present our predictions for the sum of the conventional and prompt fluxes and compare them with the ICECUBE results for the astrophysical neutrino flux. We observe that the inclusion of the IC component leads to an enhancement of the atmospheric (conventional + prompt) neutrino flux of $\approx 45 \%$ at $E_{\nu} = 10^6$ GeV and of $\approx 200 \%$ at $E_{\nu} = 3.5 \times 10^6$ GeV. Such large values are in agreement with the expectation discussed in Ref. [@prdnos].
In Fig. \[Fig:fluxe2h3a\] (a) we compare the BPL predictions with those obtained using the H3a model for the primary nucleon spectrum. As expected from our previous analysis, the atmospheric (conventional + prompt) neutrino flux derived using the H3a model is smaller the BPL one at large neutrino energies and they are similar to $E_{\nu} \approx 10^5$ GeV. Finally, in Fig. \[Fig:fluxe2h3a\] (b) we compare the BHPS predictions, derived using the H3a spectrum, with the corresponding results obtained in Refs. [@halzen; @laha], considering different assumptions for the intrinsic charm component. In particular, in Ref. [@halzen] the authors have derived an upper limit for the neutrino spectrum assuming a maximum value for the contribution associated to the hadronization of the spectator charm with the valence quarks of projectile. This upper limit is denoted by HW in the figure. On the other, in Ref. [@laha] (denoted LB hereafter) the normalization of the intrinsic charm contribution for the $x_F$ - distribution is constrained by the data at low energies presented by the ISR experiments and LEBC - MPS Collaboration, with the energy dependence being that of the inelastic $pp$ cross section (See Ref. [@laha] for details). We have that our predictions are below from the BW one at large energies, with the difference between the results increasing with the neutrino energy. On the other hand, we predict a larger atmospheric neutrino flux than the LB model in the range probed by the IceCube. We believe that this difference is mainly associated to the energy dependence of the intrinsic contribution, which in our calculations is steeper than the energy dependence of the inelastic $pp$ cross section. As a consequence, our contribution of the charm initiated process increases faster with the center - of - mass energy than the LB one. The discrimination between these different predictions can, in principle, be feasible by future IceCube measurements dedicated to contrain the magnitude of the prompt contribution.
At this point, some comments are in order. In our calculations we have considered only the BHPS model of intrinsic charm, in which the amount of IC is maximal. From the results presented in Ref. [@prdnos], we can expect that if the Meson Cloud model is considered, we will get similar results for the enhancement associated to the IC component. However, if the amount of momentum carried by the IC component is reduced, this will change our predictions for the prompt neutrino flux. Therefore, our predictions should be considered as an upper limit for the impact of the IC component. Another important aspect that should be emphasized is that we have used the color dipole approach to estimate the prompt neutrino flux. As discussed in detail in Refs. [@enb_stasto2; @vicantoni], this model implies a $x_F$ - distribution that is larger at intermediate values of $x_F$ than those obtained with the collinear and with the $k_T$ - factorization formalisms. As explained in Refs. [@enb_stasto2; @vicantoni], such behavior is somewhat unexpected as this approach includes saturation effects that should lead rather to a reduction of the cross section compared to the collinear and the $k_T$ factorization approaches. The explanation of this difference is still an open question and theme of debate (For a more detailed discussion see Ref. [@vicantoni]). Consequently, the color dipole predictions should be considered as upper bounds for the prompt neutrino flux. The mentioned aspects imply that our predictions should be considered as upper bounds. However, we believe that our results strongly indicate that the inclusion of the charm channel in the color dipole approach for the heavy quark production is important to obtain realistic predictions and that the prompt neutrino flux is sensitive to the presence or absence of an IC component in the hadron wave function.
Summary {#conc}
=======
A complete knowledge of the partonic structure of hadrons is fundamental to make predictions for the Standard Model and beyond Standard Model processes observed at hadron colliders. In particular, the heavy quark component of the proton has a direct impact on the calculation of the prompt atmospheric neutrino flux, which is a background to the astrophysical neutrino flux measured by the ICECUBE Collaboration. One important (and not yet known) quantity for heavy quark production is the amount of the intrinsic component in the hadron wave function. It carries a large fraction of the hadron momentum and, consequently, is expected to modify the cross sections and associated distributions of the produced heavy quarks and heavy mesons at forward rapidities. In this paper we have estimated the impact of the intrinsic charm component on the prompt neutrino flux using the color dipole approach to compute the charm $x_F$ - distribution. Following Ref. [@prdnos], we have generalized previous color dipole calculations by taking into account the contribution of processes initiated by charm quarks. Moreover, differently from previous studies of IC effects, we have used in our calculations the parton distributions derived from the global analysis of a large set of experimental data, with evolution described by the DGLAP equations. Additionally, we have used as input in our calculations a model for the dipole - proton scattering amplitude that describes very well particle production at forward rapidities and LHC energies. Our results indicate that the inclusion of the channel initiated by charm quarks has a strong effect on the prompt neutrino flux. In particular, if an IC component is present in the hadron wave function, our results indicate that the flux is enhanced by a factor 2 at large neutrino energies. Furthermore, we find that the astrophysical neutrino flux becomes dominant at $E_{\nu} \approx 10^6$ GeV. However, the magnitude of the background is strongly sensitive to the description of the prompt neutrino flux. Consequently, in order to disentangle the magnitude of the astrophysical contribution to the neutrino flux, it is mandatory to have a better theoretical and experimental control of the prompt neutrino flux.
Acknowledgments {#acknowledgments .unnumbered}
===============
VPG thanks Rikard Enberg for provide the results obtained in Refs. [@ers; @enb_stasto2] and Diego Rossi Gratieri by useful discussions regarding the conventional neutrino flux. A.V.G. gratefully acknowledges the Brazilian Funding Agency FAPESP for financial support (contract:2017/14974-8). This work was partially financed by the Brazilian funding agencies CAPES, CNPq, FAPESP, FAPERGS and INCT-FNA (process number 464898/2014-5). Finally, the authors would like to thank the referee for the careful reading of the manuscript and valuable comments that contributed to the improvement of the paper.
[99]{}
M. G. Aartsen [*et al.*]{} \[IceCube Collaboration\], Science [**342**]{}, 1242856 (2013).
M. G. Aartsen [*et al.*]{} \[IceCube Collaboration\], Phys. Rev. Lett. [**113**]{}, 101101 (2014). M. G. Aartsen [*et al.*]{} \[IceCube Collaboration\], Astrophys. J. [**833**]{}, 3 (2016).
T. Gaisser and F. Halzen, Ann. Rev. Nucl. Part. Sci. [**64**]{}, 101 (2014).
P. Meszaros, Ann. Rev. Nucl. Part. Sci. [**67**]{}, 45 (2017)
R. Abbasi [*et al.*]{} \[IceCube Collaboration\], Phys. Rev. D [**83**]{}, 012001 (2011). M. G. Aartsen [*et al.*]{} \[IceCube Collaboration\], Phys. Rev. Lett. [**110**]{}, 151105 (2013). P. Adamson [*et al.*]{} \[MINOS Collaboration\], Phys. Rev. D [**86**]{}, 052007 (2012). Y. Fukuda [*et al.*]{} \[Super-Kamiokande Collaboration\], Phys. Lett. B [**436**]{}, 33 (1998). M. Honda, T. Kajita, K. Kasahara, S. Midorikawa and T. Sanuki, Phys. Rev. D [**75**]{}, 043006 (2007).
G. D. Barr, T. K. Gaisser, P. Lipari, S. Robbins and T. Stanev, Phys. Rev. D [**70**]{}, 023006 (2004). T. K. Gaisser and S. R. Klein, Astropart. Phys. [**64**]{}, 13 (2015). P. Gondolo, G. Ingelman and M. Thunman, Astropart. Phys. [**5**]{}, 309 (1996).
A. D. Martin, M. G. Ryskin and A. M. Stasto, Acta Phys. Polon. B [**34**]{}, 3273 (2003).
M. V. Garzelli, S. Moch and G. Sigl, JHEP [**1510**]{}, 115 (2015).
A. Bhattacharya, R. Enberg, M. H. Reno, I. Sarcevic and A. Stasto, JHEP [**1506**]{}, 110 (2015).
R. Gauld, J. Rojo, L. Rottoli and J. Talbert, JHEP [**1511**]{}, 009 (2015).
A. Bhattacharya, R. Enberg, Y. S. Jeong, C. S. Kim, M. H. Reno, I. Sarcevic and A. Stasto, JHEP [**1611**]{}, 167 (2016).
R. Gauld, J. Rojo, L. Rottoli, S. Sarkar and J. Talbert, JHEP [**1602**]{}, 130 (2016).
F. Halzen and L. Wille, Phys. Rev. D [**94**]{}, 014014 (2016).
R. Laha and S. J. Brodsky, Phys. Rev. D [**96**]{}, 123002 (2017) M. V. Garzelli [*et al.*]{} \[PROSA Collaboration\], JHEP [**1705**]{}, 004 (2017).
M. Benzke, M. V. Garzelli, B. Kniehl, G. Kramer, S. Moch and G. Sigl, JHEP [**1712**]{}, 021 (2017).
R. Aaij [*et al.*]{} \[LHCb Collaboration\], Nucl. Phys. B [**871**]{}, 1 (2013). R. Aaij [*et al.*]{} \[LHCb Collaboration\], JHEP [**1603**]{}, 159 (2016); Erratum: \[JHEP [**1609**]{}, 013 (2016)\]; Erratum: \[JHEP [**1705**]{}, 074 (2017)\].
V. P. Goncalves, R. Maciula, R. Pasechnik and A. Szczurek, Phys. Rev. D [**96**]{}, 094026 (2017)
D. Kharzeev and K. Tuchin, Nucl. Phys. A [**735**]{}, 248 (2004).
K. Tuchin, Phys. Lett. B [**593**]{}, 66 (2004); Nucl. Phys. A [**798**]{}, 61 (2008); Y. V. Kovchegov and K. Tuchin, Phys. Rev. D [**74**]{}, 054014 (2006); H. Fujii, F. Gelis and R. Venugopalan, J. Phys. G [**34**]{}, S937 (2007).
V. P. Goncalves and M. V. T. Machado, JHEP [**0704**]{}, 028 (2007).
E. R. Cazaroto, V. P. Goncalves and F. S. Navarra, Nucl. Phys. A [**872**]{}, 196 (2011).
H. Fujii and K. Watanabe, Nucl. Phys. A [**920**]{}, 78 (2013).
T. Altinoluk, N. Armesto, G. Beuf, A. Kovner and M. Lublinsky, Phys. Rev. D [**93**]{}, 054049 (2016).
F. Gelis, E. Iancu, J. Jalilian-Marian and R. Venugopalan, Ann. Rev. Nucl. Part. Sci. [**60**]{}, 463 (2010); E. Iancu and R. Venugopalan, arXiv:hep-ph/0303204; H. Weigert, Prog. Part. Nucl. Phys. [**55**]{}, 461 (2005); J. Jalilian-Marian and Y. V. Kovchegov, Prog. Part. Nucl. Phys. [**56**]{}, 104 (2006); J. L. Albacete and C. Marquet, Prog. Part. Nucl. Phys. [**76**]{}, 1 (2014).
A. Dumitru, A. Hayashigaki and J. Jalilian-Marian, Nucl. Phys. A [**765**]{}, 464 (2006); Nucl. Phys. A [**770**]{}, 57 (2006).
D. Boer, A. Utermann, E. Wessels, Phys. Rev. D [**77**]{}, 054014 (2008).
F. O. Durães, A. V. Giannini, V. P. Goncalves and F. S. Navarra, Phys. Rev. C [**94**]{}, 024917 (2016).
S. J. Brodsky, A. Kusina, F. Lyonnet, I. Schienbein, H. Spiesberger and R. Vogt, Adv. High Energy Phys. [**2015**]{}, 231547 (2015).
S. J. Brodsky, V. A. Bednyakov, G. I. Lykasov, J. Smiesko and S. Tokar, Prog. Part. Nucl. Phys. [**93**]{}, 108 (2017)
S. J. Brodsky, P. Hoyer, C. Peterson and N. Sakai, Phys. Lett. B [**93**]{}, 451 (1980).
V. D. Barger, F. Halzen and W. Y. Keung, Phys. Rev. D [**25**]{}, 112 (1982).
S. Paiva, M. Nielsen, F. S. Navarra, F. O. Duraes and L. L. Barz, Mod. Phys. Lett. A [**13**]{}, 2715 (1998); F. S. Navarra, M. Nielsen, C. A. A. Nunes and M. Teixeira, Phys. Rev. D [**54**]{}, 842 (1996).
F. M. Steffens, W. Melnitchouk and A. W. Thomas, Eur. Phys. J. C [**11**]{}, 673 (1999).
P. Jimenez-Delgado, T. J. Hobbs, J. T. Londergan and W. Melnitchouk, Phys. Rev. Lett. [**116**]{}, 019102 (2016).
S. J. Brodsky and S. Gardner, Phys. Rev. Lett. [**116**]{}, 019101 (2016).
V. A. Bednyakov, M. A. Demichev, G. I. Lykasov, T. Stavreva and M. Stockton, Phys. Lett. B [**728**]{}, 602 (2014); P. H. Beauchemin, V. A. Bednyakov, G. I. Lykasov and Y. Y. Stepanenko, Phys. Rev. D [**92**]{}, 034014 (2015).
G. Bailas and V. P. Goncalves, Eur. Phys. J. C [**76**]{}, 105 (2016).
T. Boettcher, P. Ilten and M. Williams, Phys. Rev. D [**93**]{}, 074008 (2016).
V. P. Goncalves, F. S. Navarra and T. Ullrich, Nucl. Phys. A [**842**]{}, 59 (2010).
B. A. Kniehl, G. Kramer, I. Schienbein and H. Spiesberger, Phys. Rev. D [**79**]{}, 094009 (2009).
F. Carvalho, A. V. Giannini, V. P. Goncalves and F. S. Navarra, Phys. Rev. D [**96**]{}, 094002 (2017)
R. Enberg, M. H. Reno and I. Sarcevic, Phys. Rev. D [**78**]{}, 043005 (2008)
L. Pasquali, M. H. Reno and I. Sarcevic, Phys. Rev. D [**59**]{}, 034020 (1999)
N. N. Nikolaev, G. Piller and B. G. Zakharov, J. Exp. Theor. Phys. [**81**]{}, 851 (1995) \[Zh. Eksp. Teor. Fiz. [**108**]{}, 1554 (1995)\]; Z. Phys. A [**354**]{}, 99 (1996).
B. Z. Kopeliovich and A. V. Tarasov, Nucl. Phys. A [**710**]{}, 180 (2002).
I. Balitsky, Nucl. Phys. B [**463**]{}, 99 (1996); Y. V. Kovchegov, Phys. Rev. D [**60**]{}, 034008 (1999); Phys. Rev. D [**61**]{}, 074018 (2000).
T. Lappi and H. Mantysaari, Phys. Rev. D [**91**]{}, 074016 (2015).
D. Kharzeev, Y.V. Kovchegov and K. Tuchin, Phys. Lett. [**B599**]{}, 23 (2004); J. Bartels, K. Golec-Biernat, H. Kowalski, Phys. Rev. D [**66**]{}, 014001 (2002); H. Kowalski and D. Teaney, Phys. Rev. D [**68**]{}, 114005 (2003); H. Kowalski, L. Motyka and G. Watt, Phys. Rev. D [**74**]{}, 074016(2006); V. P. Goncalves, M. S. Kugeratski, M. V. T. Machado and F. S. Navarra, Phys. Lett. B [**643**]{}, 273 (2006); C. Marquet, R. Peschanski and G. Soyez, Phys. Rev. D [**76**]{}, 034011 (2007); G. Watt and H. Kowalski, Phys. Rev. D [**78**]{}, 014016 (2008).
E. Iancu, K. Itakura, S. Munier, Phys. Lett. [**B590**]{}, 199 (2004)
G. Soyez, Phys. Lett. B [**655**]{}, 32 (2007)
J. Pumplin, H. L. Lai and W. K. Tung, Phys. Rev. D [**75**]{}, 054029 (2007).
M. A. Betemps, V. P. Goncalves and J. T. de Santana Amaral, Eur. Phys. J. C [**66**]{}, 137 (2010)
T. K. Gaisser, Astropart. Phys. [**35**]{}, 801 (2012)
S. Dulat, T.-J. Hou, J. Gao, J. Huston, J. Pumplin, C. Schmidt et al., Phys. Rev. D [**89**]{} 073004, (2014).
J. Raufeisen and J. C. Peng, Phys. Rev. D [**67**]{}, 054008 (2003)
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We prove an analytic version of the Kontsevich-Soibelman wall-crossing formula describing how the number of finite-length trajectories of a quadratic differential jumps as the differential is varied. We characterize certain maps appearing in this wall-crossing formula using Fock-Goncharov coordinates. As an application, we compute the Stokes automorphisms for Voros symbols in exact WKB analysis.'
author:
- 'Dylan G.L. Allegretti'
title: |
On the wall-crossing formula for\
quadratic differentials
---
Introduction
============
The concept of a quadratic differential is fundamental in several areas of low-dimensional geometry and dynamics. Given a Riemann surface $S$, a holomorphic quadratic differential $\phi$ on $S$ is defined as a holomorphic section of $\omega_S^{\otimes2}$ where $\omega_S$ is the holomorphic cotangent bundle of $S$. Such a differential $\phi$ determines a flat metric on $S$ with singularities at the zeros of $\phi$. Of particular importance in the theory are certain geodesics on $S$ with respect to the flat metric. These geodesics are the so called finite-length trajectories. Over the past four decades, the problem of counting finite-length trajectories on a surface equipped with a quadratic differential has led to many remarkable results [@Masur1; @Masur2; @Veech; @EskinMasur; @EskinMasurZorich; @EMM].
In a seemingly unrelated development, Kontsevich and Soibelman described an approach to Donaldson-Thomas theory that can be used to associate numerical invariants to moduli spaces of objects in a 3-Calabi-Yau triangulated category [@KontsevichSoibelman1; @KontsevichSoibelman2; @KontsevichSoibelman3]. These invariants are known as BPS invariants since they provide a mathematical approach to counting BPS states in physics. The BPS invariants depend on some extra data, namely a choice of Bridgeland stability condition [@Bridgeland07]. The set of all stability conditions has the structure of a complex manifold, and the BPS invariants remain constant as one varies the choice of stability condition within this manifold, except at certain real codimension one “walls” where they can change discontinuously. The celebrated Kontsevich-Soibelman wall-crossing formula describes precisely how the BPS invariants change as one crosses a wall in the space of stability conditions [@KontsevichSoibelman1].
As part of their work on the physics of four-dimensional supersymmetric quantum field theories, Gaiotto, Moore, and Neitzke proposed a relationship between Donaldson-Thomas theory and the theory of quadratic differentials [@GMN]. Their proposal, which was later proved mathematically by Bridgeland and Smith [@BridgelandSmith], implies that in some cases the BPS invariants defined by Kontsevich and Soibelman count finite-length trajectories of meromorphic quadratic differentials. In particular, the wall-crossing formula encodes how these integers counting finite-length trajectories change as one varies the quadratic differential.
The purpose of the present paper is to study the Kontsevich-Soibelman wall-crossing formula in the context of quadratic differentials. Although the wall-crossing formula is usually formulated as an identity between infinite products of automorphisms of a formal power series ring, we will show using Fock-Goncharov coordinates [@FockGoncharov1] that in our context these products can be viewed as birational transformations of a complex algebraic torus. As an application of our main result, we compute the Stokes automorphisms for the Voros symbols appearing in the theory of exact WKB analysis [@Allegretti19a; @IwakiNakanishi].
Basic setup
-----------
In order to apply results from Donaldson-Thomas theory, we will restrict attention in this paper to a special class of quadratic differentials studied by Gaiotto, Moore, and Neitzke [@GMN] and known as GMN differentials. Given a compact Riemann surface $S$, we will define a GMN differential on $S$ to be a meromorphic section of $\omega_S^{\otimes2}$ having simple zeros and satisfying some technical conditions concerning the orders of its critical points.
Let $\phi$ be a GMN differential on a compact Riemann surface $S$. Near any point of $S$ which is not a zero or pole of $\phi$, there is a local coordinate $w$, well defined up to transformations $w\mapsto\pm w+\text{constant}$, such that $\phi=dw\otimes dw$. By pulling back the Euclidean metric on $\mathbb{C}$ using these distinguished local coordinates, we get a flat metric on the complement of the zeros and poles. We will be interested in certain geodesics with respect to this flat metric. These geodesics are called finite-length trajectories and come in two types. The first is a saddle connection, which is roughly a geodesic connecting two zeros of $\phi$. A saddle connection is closed if the zeros at its endpoints coincide. The second type of finite-length trajectory is a closed geodesic, which forms a loop in the complement of all zeros and poles. A closed geodesic is always contained in an annulus foliated by homotopic closed geodesics on $S$. We refer to such an annulus as a cylinder for $\phi$. A cylinder is said to be degenerate if one of its boundaries is a pole of order two of $\phi$.
The differential $\phi$ determines a canonical branched double cover $\pi:\Sigma_\phi\rightarrow S$ on which the square root $\sqrt{\phi}$ is a well defined 1-form. Let $\Sigma_\phi^\circ$ denote the complement in $\Sigma_\phi$ of the preimages of all poles of $\phi$ of order $>1$, and let $\Gamma_\phi$ be the set of $\gamma\in H_1(\Sigma_\phi^\circ,\mathbb{Z})$ such that $\tau(\gamma)=-\gamma$ where $\tau$ is the involution exchanging the two sheets of the double cover $\Sigma_\phi$. Then $\Gamma_\phi$ is a lattice of finite rank, and we have a group homomorphism $$Z_\phi:\Gamma_\phi\rightarrow\mathbb{C}, \quad Z_\phi(\gamma)=\int_\gamma\sqrt{\phi},$$ called the period map.
We will see that any finite length trajectory $\alpha$ has a natural lift $\widehat{\alpha}\in\Gamma_\phi$. Moreover, any two closed geodesics in the same cylinder have the same lift. Thus there is a class in $\Gamma_\phi$ associated to any saddle connection or cylinder for $\phi$. We will define a notion of genericity for GMN differentials, and if $\phi$ is a generic GMN differential and $\gamma\in\Gamma_\phi$ is any class, we will consider the associated integer $$\begin{aligned}
\Omega_\phi(\gamma) &= |\{\text{non-closed saddle connections of class $\pm\gamma$}\}| \\
&\quad - 2\cdot|\{\text{nondegenerate cylinders of class $\pm\gamma$}\}|\end{aligned}$$ which “counts” the finite-length trajectories of $\phi$. By the work of Bridgeland and Smith [@BridgelandSmith], this integer coincides with the BPS invariant defined in a more general context by Kontsevich and Soibelman [@KontsevichSoibelman1].
The wall-crossing formula
-------------------------
If $\phi$ is any GMN differential, then an active ray is defined to be a ray in $\mathbb{C}^*$ of the form $\ell=\mathbb{R}_{>0}\cdot Z_\phi(\gamma)$ where $\gamma$ is the class of some finite-length trajectory. We will be interested in certain birational maps associated to the active rays in the case when the differential $\phi$ is generic.
To define these maps, we consider an object called the twisted torus, defined as the set $$\mathbb{T}_-=\{g:\Gamma_\phi\rightarrow\mathbb{C}^*:g(\gamma_1+\gamma_2)=(-1)^{\langle\gamma_1,\gamma_2\rangle}g(\gamma_1)g(\gamma_2)\}$$ where $\langle-,-\rangle$ is the intersection pairing on $\Gamma_\phi$. We will see that the twisted torus $\mathbb{T}_-$ has the natural structure of an algebraic variety whose coordinate ring is spanned as a vector space by the functions $x_\gamma:\mathbb{T}_-\rightarrow\mathbb{C}^*$ given by $x_\gamma(g)=g(\gamma)$. If $\phi$ is a generic GMN differential and $\ell\subset\mathbb{C}^*$ is any ray emanating from the origin, then there is a birational automorphism $\mathbf{S}_\phi(\ell)$ of $\mathbb{T}_-$ given on functions by $$\mathbf{S}_\phi(\ell)^*(x_\beta)=x_\beta\cdot\prod_{Z_\phi(\gamma)\in\ell}(1-x_\gamma)^{\Omega_\phi(\gamma)\cdot\langle\beta,\gamma\rangle}.$$ Note that this automorphism is the identity if $\ell$ is non-active.
We will extend this construction and describe, for any convex sector $\Delta\subset\mathbb{C}^*$, a partially defined automorphism $\mathbf{S}_\phi(\Delta)$ of $\mathbb{T}_-$. To do this, we define a notion of height for any ray in $\mathbb{C}^*$, and we consider the composition $$\mathbf{S}_{\phi,<H}(\Delta)=\mathbf{S}_\phi(\ell_1)\circ\mathbf{S}_\phi(\ell_2)\circ\dots\circ\mathbf{S}_\phi(\ell_k)$$ of the maps defined above where $\ell_1,\ell_2,\dots,\ell_k\subset\Delta$ are the rays of height $<H$ taken in the clockwise order. A result of Bridgeland [@Bridgeland19] says there is a nonempty analytic open subset of $\mathbb{T}_-$ on which the pointwise limit $$\mathbf{S}_\phi(\Delta)=\lim_{H\rightarrow\infty}\mathbf{S}_{\phi,<H}(\Delta)$$ exists and is holomorphic. This limiting function $\mathbf{S}_\phi(\Delta)$ is called the BPS automorphism associated to the sector $\Delta\subset\mathbb{C}^*$. Similar analytic maps have been studied by Kontsevich and Soibelman [@KontsevichSoibelman4], who described a general framework for the analytic study of wall-crossing formulas.
In our context, the wall-crossing formula is an identity that relates the BPS automorphisms $\mathbf{S}_\phi(\Delta)$ for different quadratic differentials $\phi$. To state this result, we need a suitable moduli space of GMN differentials. Following the approach of [@BridgelandSmith], our moduli space will be labeled by the combinatorial data of a marked bordered surface. This is defined as a compact oriented surface $\mathbb{S}$ together with a finite set $\mathbb{M}$ of marked points on $\mathbb{S}$ such that every boundary component of $\mathbb{S}$ contains at least one marked point. As we review below, any GMN differential determines an associated marked bordered surface. For a given $(\mathbb{S},\mathbb{M})$, there is a moduli space $\mathscr{Q}^\pm(\mathbb{S},\mathbb{M})$ parametrizing GMN differentials whose associated marked bordered surface is $(\mathbb{S},\mathbb{M})$.
In terms of this moduli space, we have the following analytic version of the Kontsevich-Soibelman wall-crossing formula.
\[thm:introWCF\] Let $(\mathbb{S},\mathbb{M})$ be a marked bordered surface which is not a closed surface with exactly one marked point, and let $\Delta\subset\mathbb{C}^*$ be a convex sector. Suppose $\phi_t$, $t\in[0,1]$, is a path in $\mathscr{Q}^\pm(\mathbb{S},\mathbb{M})$ with general endpoints such that the boundary rays of $\Delta$ are non-active for each differential $\phi_t$. Then $$\mathbf{S}_{\phi_0}(\Delta)=\mathbf{S}_{\phi_1}(\Delta).$$
Note that the BPS invariants for the differentials $\phi_0$ and $\phi_1$ will be completely different in general. In this way, the wall-crossing formula encodes how the finite length trajectories of a quadratic differential change as we vary the differential within the moduli space $\mathscr{Q}^\pm(\mathbb{S},\mathbb{M})$.
Fock-Goncharov coordinates
--------------------------
If $\phi$ is a general point in $\mathscr{Q}^\pm(\mathbb{S},\mathbb{M})$ and $\Delta\subset\mathbb{C}^*$ is a convex sector whose boundary rays are non-active, then Theorem \[thm:introWCF\] says that the BPS automorphism $\mathbf{S}_\phi(\Delta)$ is invariant under small deformations of the differential. Our main result will allow us to compute this invariant explicitly using Fock-Goncharov coordinates. As a consequence, we will also see that $\mathbf{S}_\phi(\Delta)$ is a birational automorphism of $\mathbb{T}_-$, a fact which is not at all obvious from the definition.
To explain this result in more detail, we consider triangulations of a marked bordered surface. Precisely, we define an ideal triangulation of $(\mathbb{S},\mathbb{M})$ to be a triangulation of $\mathbb{S}$ whose vertices are exactly the points of $\mathbb{M}$. For technical reasons, we will also consider a variant of this concept called a tagged triangulation [@FST]. Assuming $(\mathbb{S},\mathbb{M})$ is not a closed surface with exactly one marked point, any two tagged triangulations of $(\mathbb{S},\mathbb{M})$ are related by a sequence of elementary operations called flips. Roughly speaking, a flip is an operation that removes an arc of the triangulation and replaces it by the unique different arc that results in a new triangulation.
In their seminal paper on higher Teichmüller theory [@FockGoncharov1], Fock and Goncharov introduced a moduli space $\mathscr{X}(\mathbb{S},\mathbb{M})$ parametrizing flat $\operatorname{\operatorname{PGL}}_2(\mathbb{C})$-connections on the punctured surface $\mathbb{S}\setminus\mathbb{M}$ with additional framing data. If $\tau$ is any tagged triangulation of $(\mathbb{S},\mathbb{M})$, let us write $\Gamma_\tau\cong\mathbb{Z}^n$ for the lattice spanned by the arcs of $\tau$. Then there is a birational map $$X_\tau:\mathscr{X}(\mathbb{S},\mathbb{M})\dashrightarrow\mathbb{T}_\tau\coloneqq\operatorname{\operatorname{Hom}}_{\mathbb{Z}}(\Gamma_\tau,\mathbb{C}^*)$$ from the moduli space to a complex algebraic torus $\mathbb{T}_\tau\cong(\mathbb{C}^*)^n$. The components of this map corresponding to the arcs of $\tau$ are known as Fock-Goncharov coordinates.
Let $(\mathbb{S},\mathbb{M})$ be a marked bordered surface which is not a closed surface with exactly one marked point. If $\phi$ is a general point in the space $\mathscr{Q}^\pm(\mathbb{S},\mathbb{M})$ and $\Delta\subset\mathbb{C}^*$ is a convex sector whose boundary rays are non-active with phases $\theta_1$ and $\theta_2$, then we will see that each rotated differential $e^{-2i\theta_j}\cdot\phi$ determines a tagged triangulation $\tau_j$ of $(\mathbb{S},\mathbb{M})$. In this case the tori $\mathbb{T}_{\tau_j}$ are naturally identified with $\mathbb{T}_+=\operatorname{\operatorname{Hom}}_{\mathbb{Z}}(\Gamma_\phi,\mathbb{C}^*)$, and hence we can think of the Fock-Goncharov coordinates as providing birational maps $$\label{eqn:introsametarget}
X_{\tau_j}:\mathscr{X}(\mathbb{S},\mathbb{M})\dashrightarrow\mathbb{T}_+.$$ As we explain below, the twisted torus $\mathbb{T}_-$ is a torsor for $\mathbb{T}_+$, and hence these objects can be identified after choosing a basepoint in $\mathbb{T}_-$.
\[thm:intromain\] Take notation as in the last paragraph. Then
1. There is a distinguished basepoint $\xi\in\mathbb{T}_-$ such that $\xi(\gamma)=-1$ if $\gamma\in\Gamma_\phi$ is the class of a non-closed saddle connection and $\xi(\gamma)=+1$ if $\gamma$ is the class of a closed saddle connection.
2. $\mathbf{S}_\phi(\Delta)$ extends to a birational automorphism of $\mathbb{T}_-$. If we use the basepoint $\xi$ to identify $\mathbb{T}_-$ with $\mathbb{T}_+$, then this is the birational automorphism of $\mathbb{T}_+$ relating the maps .
We note that Theorem \[thm:main\] provides a means of computing the BPS automorphism $\mathbf{S}_\phi(\Delta)$. Indeed, the tagged triangulations $\tau_j$ are related by a sequence of flips. For each pair of tagged triangulations related by a flip, the associated Fock-Goncharov coordinates are related by a well known coordinate transformation. By composing these coordinate transformations, one obtains the transformation $\mathbf{S}_\phi(\Delta)$ explicitly.
Application to WKB analysis
---------------------------
Let us conclude this introduction by describing an application of the above results to exact WKB analysis. This material will not appear elsewhere in this paper and can be skipped. The ideas of this subsection are important in [@Allegretti19b] where they are used to solve a class of Riemann-Hilbert problems formulated in [@Bridgeland19].
The theory of exact WKB analysis is a tool for constructing exact solutions of Schrödinger’s equation for small values of the Planck constant $\hbar$. In exact WKB analysis, the Voros symbols are certain formal series in $\hbar$ which appear when explicitly calculating the monodromy group of Schrödinger’s equation. As explained in [@Allegretti19a], one can associate a Voros symbol to a general point $\phi\in\mathscr{Q}^\pm(\mathbb{S},\mathbb{M})$ and class $\gamma\in\Gamma_\phi$. This formal series has the form $$Y_{\phi,\gamma}(\hbar)=e^{-Z_\phi(\gamma)/\hbar}\cdot\sum_{k=0}^\infty\hbar^kf_k$$ for some coefficients $f_k$.
In general, the Voros symbol diverges and therefore does not define a holomorphic function of $\hbar$. To get around this difficulty, one must use Borel resummation. In the Borel resummation method, one starts with a (possibly divergent) formal power series $f(\hbar)=\sum_{k=0}^\infty\hbar^kf_k$ satisfying a condition called Borel summability. One then constructs a holomorphic function $\mathcal{S}[f]$ called the Borel sum, having $f$ as its asymptotic expansion near zero. The Borel sum is given by the expression $$\label{eqn:Borelsum}
\mathcal{S}[f](\hbar)=f_0+\int_0^\infty e^{-y/\hbar}f_B(y)dy$$ where $f_B(y)$ denotes the Borel transform of $f$ (see [@IwakiNakanishi] for details). The series appearing in the definition of the Voros symbol is known to be Borel summable [@IwakiNakanishi] provided there is no active ray of phase zero for the differential $\phi$. In this case, we can think of the Voros symbol as a holomorphic function given by $\mathcal{Y}_{\phi,\gamma}(\hbar)=e^{-Z_\phi(\gamma)/\hbar}\cdot\mathcal{S}[f](\hbar)$.
If there is an active ray of phase zero, then the Voros symbol may not be Borel summable. This happens because the Borel transform develops singularities along the positive real axis, making the integral in undefined. In this case, we can choose a small angle $\theta$, and define a modified Borel sum $\mathcal{S}_\theta[f]$ by the same formula where now the integral is taken along the ray $y=re^{i\theta}$, $r\geq0$ in $\mathbb{C}$. The resulting function $\mathcal{S}_\theta[f]$ is again holomorphic and has $f$ as its asymptotic expansion near zero. However, in this case an interesting Stokes phenomenon can occur: for sufficiently small angles $\theta>0$, the two functions $\mathcal{S}_{\pm\theta}[f]$ have the same asymptotic expansion as $\hbar\rightarrow0$ but are not the same function.
The Stokes phenomenon for Voros symbols was studied by Delabaere, Dillinger, and Pham [@DDP] who considered the situation where there is a unique horizontal finite-length trajectory (a finite-length trajectory whose class $\gamma$ satisfies $Z_\phi(\gamma)\in\mathbb{R}$), which moreover is a non-closed saddle connection. For sufficiently small $\theta>0$, the functions $\mathcal{Y}_{\phi,\beta}^{(\pm\theta)}(\hbar)=e^{-Z_\phi(\gamma)/\hbar}\cdot\mathcal{S}_{\pm\theta}[f](\hbar)$ are defined, and Delabaere, Dillinger, and Pham showed that $$\mathcal{Y}_{\phi,\beta}^{(-\theta)}=\mathcal{Y}_{\phi,\beta}^{(+\theta)}\cdot\left(1+\mathcal{Y}_{\phi,\gamma}^{(+\theta)}\right)^{\langle\beta,\gamma\rangle}$$ as analytic functions where $\gamma\in\Gamma_\phi$ is the class of the saddle connection. The transformation relating the functions $\mathcal{Y}_{\phi,\beta}^{(+\theta)}(\hbar)$ to $\mathcal{Y}_{\phi,\beta}^{(-\theta)}(\hbar)$ is known as a Stokes automorphism. Related results concerning Stokes automorphisms for Voros symbols have been obtained in [@AIT; @IwakiNakanishi].
Using Theorem \[thm:intromain\], we can give a completely general description of the Stokes automorphisms for Voros symbols. Indeed, suppose $\phi\in\mathscr{Q}^\pm(\mathbb{S},\mathbb{M})$ has an active ray of phase zero. For any $\varepsilon>0$, we can choose $0<\theta<\varepsilon$ so that the rays $r_\pm=\mathbb{R}_{>0}\cdot e^{\pm2i\theta}$ are non-active. Let $\tau_\pm$ be the corresponding tagged triangulations of $(\mathbb{S},\mathbb{M})$, and for any $\gamma\in\Gamma_\phi$, let $x_\gamma:\mathbb{T}_+\rightarrow\mathbb{C}^*$ be the character $x_\gamma(g)=g(\gamma)$. Then the main result of [@Allegretti19a] says that there exists a one parameter family of points $\mathcal{L}_\phi(\hbar)\in\mathscr{X}(\mathbb{S},\mathbb{M})$ such that $$\mathcal{Y}_{\phi,\gamma}^{\pm\theta}(\hbar)=x_\gamma\circ X_{\tau_\pm}(\mathcal{L}_\phi(\hbar)).$$ It follows from Theorem \[thm:intromain\] that after identifying $\mathbb{T}_+$ with $\mathbb{T}_-$ using the canonical basepoint $\xi$, the Stokes automorphism is precisely the BPS automorphism $\mathbf{S}_\phi(\Delta)$ where $\Delta\subset\mathbb{C}^*$ is the sector having boundary rays $r_\pm$. In particular, this Stokes automorphism is a product of (perhaps infinitely many) factors $\mathbf{S}_\phi(\ell)$ associated to active rays $\ell\subset\Delta$.
Acknowledgements. {#acknowledgements. .unnumbered}
-----------------
I thank Tom Bridgeland for encouraging me to write this paper and for suggesting a strategy for the proof of the main result. While working on this project, I benefited from conversations and correspondence with David Aulicino, Ben Davison, Kohei Iwaki, Sven Meinhardt, Andrew Neitzke, and Anton Zorich. This project was conceived while I was in residence at the Mathematical Sciences Research Institute in Berkeley, California, during the Fall 2019 semester, partially supported by National Science Foundation grant DMS-1440140.
Quadratic differentials on Riemann surfaces
===========================================
In this section, we review some basic definitions from the theory of quadratic differentials and describe our geometric setup. None of this material is new, and most of it can be found in [@BridgelandSmith].
GMN differentials
-----------------
Throughout this paper, we will write $S$ for a compact Riemann surface of genus $g\geq0$ and write $\omega_S$ for its holomorphic cotangent bundle. Then a meromorphic *quadratic differential* on $S$ is defined to be a meromorphic section of $\omega_S^{\otimes2}$. In terms of a local coordinate $z$ on $S$, such a section $\phi$ can be written $$\phi(z)=\varphi(z)dz^2$$ where $\varphi(z)$ is a meromorphic function in the local coordinate.
By a *critical point* of a quadratic differential $\phi$, we mean either a zero or pole of $\phi$. We will write $\operatorname{\operatorname{Crit}}(\phi)$ for the set of all critical points of $\phi$. Such a point is called a *finite critical point* if it is a zero or a simple pole and an *infinite critical point* otherwise. Thus the set of all critical points is a disjoint union $$\operatorname{\operatorname{Crit}}(\phi)=\operatorname{\operatorname{Crit}}_{<\infty}(\phi)\cup\operatorname{\operatorname{Crit}}_\infty(\phi)$$ where $\operatorname{\operatorname{Crit}}_{<\infty}(\phi)$ is the set of finite critical points and $\operatorname{\operatorname{Crit}}_\infty(\phi)$ is the set of infinite critical points.
In this paper, we will be concerned with quadratic differentials of the following special type.
\[def:GMNdifferential\] A *Gaiotto-Moore-Neitzke (GMN) differential* is a meromorphic quadratic differential $\phi$ on a compact, connected Riemann surface $S$ satisfying the following conditions:
1. $\phi$ has no zero of order $>1$.
2. $\phi$ has at least one pole.
3. $\phi$ has at least one finite critical point.
A GMN differential is said to be *complete* if in addition it has no simple poles so that every pole has order $\geq2$.
The canonical double cover
--------------------------
Let $\phi$ be a GMN differential on the compact Riemann surface $S$. Let $p_i\in S$ be the poles of $\phi$, and let $m_i$ be the order of the pole $p_i$. Then we can think of the differential $\phi$ as a holomorphic section $s_\phi$ of $\omega_S(E)^{\otimes2}$ where $E$ is the divisor $$E=\sum_i\left\lceil\frac{m_i}{2}\right\rceil p_i$$ and $s_\phi$ has simple zeros at both the zeros and odd order poles of $\phi$. We can then define the *canonical double cover* as the subspace $\Sigma_\phi$ of the total space of $\omega_S(E)$ cut out by the equation $$\lambda^2=s_\phi(p)$$ in the fiber over each point $p\in S$. The space $\Sigma_\phi$ defined in this way is a Riemann surface, and the obvious projection $\pi:\Sigma_\phi\rightarrow S$ is a covering map, branched precisely over the set of zeros and odd order poles of $\phi$. Since $\phi$ has at least one finite critical point, this covering map has at least one branch point and hence $\Sigma_\phi$ is connected.
By construction, the canonical double cover comes with a tautological section $\lambda$ of $\pi^*(\omega_S(E))$ satisfying $\pi^*(s_\phi)=\lambda\otimes\lambda$. It can be viewed alternatively as a meromorphic 1-form on $\Sigma_\phi$. Note that if $p$ is a simple zero of $\phi$ then the cover $\pi:\Sigma_\phi\rightarrow S$ is ramified at $p$, and therefore the preimage $\pi^{-1}(p)$ consists of a single point. Later, we will need to know the following simple fact concerning behavior of $\lambda$ at such a point.
\[lem:doublezeros\] Suppose $p\in S$ is a simple zero of $\phi$. Then the tautological 1-form $\lambda$ has a zero of order two at $\pi^{-1}(p)$.
By Theorem 6.1 of [@Strebel], there exists a local coordinate $z$ defined in a neighborhood of $p$ and satisfying $z(p)=0$ so that in this local coordinate we have $$\phi(z)=c\cdot zdz^2$$ for some $c\in\mathbb{C}^*$. Since $\pi:\Sigma_\phi\rightarrow S$ is ramified at $p$, we can choose a local coordinate $w$ in a neighborhood of $\pi^{-1}(p)$ so that the projection map is given by $\pi^*(z)=w^2$. Then $$(\pi^*\phi)(w)=c\cdot w^2d(w^2)^2=4c\cdot w^4dw^2,$$ and therefore $\lambda$ has a zero of order $\frac{1}{2}(4)=2$.
The period map
--------------
For convenience, we use the notation $$\Sigma_\phi^\circ=\Sigma_\phi\setminus\pi^{-1}\operatorname{\operatorname{Crit}}_\infty(\phi).$$ There is a natural involution $\tau:\Sigma_\phi\rightarrow\Sigma_\phi$ exchanging the two sheets of the double cover and commuting with the covering map $\pi$. We will write $$\Gamma_\phi=\left\{\sigma\in H_1(\Sigma_\phi^\circ,\mathbb{Z}):\tau(\sigma)=-\sigma\right\}$$ for the anti-invariant part of the first homology $H_1(\Sigma_\phi^\circ,\mathbb{Z})$ with respect to this covering involution. It is a free abelian group of finite rank. By integrating the canonical 1-form $\lambda$ around cycles in $\Gamma_\phi$, we obtain a group homomorphism $$Z_\phi:\Gamma_\phi\rightarrow\mathbb{C}, \quad Z_\phi(\gamma)=\int_\gamma\lambda,$$ called the *period map* for the differential $\phi$.
Trajectories
------------
Let $\phi$ be a meromorphic quadratic differential on a compact Riemann surface $S$. Near any point of $S\setminus\operatorname{\operatorname{Crit}}(\phi)$, there is a distinguished local coordinate $w$, unique up to transformations of the form $w\mapsto\pm w+\text{constant}$, with respect to which the quadratic differential $\phi$ is given by $$\phi(w)=dw\otimes dw.$$ Indeed, if we have $\phi(z)=\varphi(z)dz^2$ for some local coordinate $z$, then $w$ is given by $w=\int\sqrt{\varphi(z)}dz$ for some choice of the square root. These distinguished local coordinates determine two structures on $S\setminus\operatorname{\operatorname{Crit}}(\phi)$. The first is a flat metric defined by pulling back the Euclidean metric on $\mathbb{C}$ by the distinguished local coordinates. The other structure is the *horizontal foliation*, defined as the foliation the curves ${\operatorname{Im}}(w)=\text{constant}$.
By a *straight arc* in $S$, we mean a smooth path $\alpha:I\rightarrow S\setminus\operatorname{\operatorname{Crit}}(\phi)$, defined on an open interval $I\subset\mathbb{R}$, which makes a constant angle $\pi\theta$ with the leaves of the horizontal foliation. By convention, straight arcs will be parametrized by arc length in the flat metric induced by the distinguished local coordinates, and two straight arcs will be regarded as the same if they are related by a reparametrization of the form $t\mapsto\pm t+\text{constant}$. The phase $\theta$ of a straight arc is well defined in $\mathbb{R}/\mathbb{Z}$, and a straight arc of phase $\theta=0$ is said to be *horizontal*.
A straight arc is called a *trajectory* if it is not the restriction of a straight arc defined on a larger interval. Thus a horizontal trajectory is the same thing as a leaf of the horizontal foliation. A *saddle connection* is a trajectory of any phase whose domain of definition is a finite length interval. A saddle connection is said to be *closed* if its endpoints coincide. Note that if a trajectory intersects itself in $S\setminus\operatorname{\operatorname{Crit}}(\phi)$ then it must be periodic and have domain $I=\mathbb{R}$. In this case it is called a *closed trajectory*. By a *finite-length trajectory*, we mean either a saddle connection or a closed trajectory.
Any closed trajectory is contained in an annular region foliated by homotopic closed trajectories. Such a region will be called a *cylinder*. The boundary of a cylinder has two components. Typically, each component is composed of saddle connections, but it can also happen that the cylinder consists of closed trajectories encircling a single pole of order two, which forms one of the boundary components. In the latter case the cylinder is said to be *degenerate*.
Homology classes
----------------
Consider a GMN differential $\phi$ on a compact Riemann surface $S$. If $\alpha:I\rightarrow S$ is a finite-length trajectory for $\phi$ which is horizontal, then we can consider the preimage $\widehat{\alpha}=\pi^{-1}(\alpha)$ of this trajectory in the canonical double cover $\Sigma_\phi$. This preimage is a closed curve which may be disconnected if $\alpha$ is a closed trajectory. As we have seen, there is a canonical 1-form $\lambda$ on the double cover with the property that $\pi^*(\phi)=\lambda\otimes\lambda$. We can endow the closed curve $\widehat{\alpha}$ with a canonical orientation by requiring that $\lambda$ evaluated on a tangent vector to the oriented curve be real and positive.
Similarly, if $\alpha:I\rightarrow S$ is a finite-length trajectory with some nonzero phase $\theta$, then we can lift $\alpha$ to a closed curve $\widehat{\alpha}$ in the canonical double cover. We can once again endow this closed curve with an orientation, but in this case, we require that $\lambda$ evaluated on a tangent vector to the oriented curve have positive imaginary part.
Thus we associate, to any finite-length trajectory $\alpha$ of the differential, a corresponding cycle $\widehat{\alpha}$ in the canonical double cover. The covering involution reverses the orientation of this cycle, and so we obtain an anti-invariant class $\widehat{\alpha}\in\Gamma_\phi$ in homology, which we call the class of $\alpha$.
Finite-length trajectories
--------------------------
The wall-crossing formula considered in this paper describes the jumping behavior of a certain integer counting finite length trajectories with a given homology class in $\Gamma_\phi$. To define this integer, we will impose a genericity assumption on the differential $\phi$. Namely, we say that a GMN differential $\phi$ is *generic* if, for any two classes $\gamma_1$, $\gamma_2\in\Gamma_\phi$, we have $$\mathbb{R}\cdot Z_\phi(\gamma_1)=\mathbb{R}\cdot Z_\phi(\gamma_2) \implies \mathbb{Z}\cdot\gamma_1=\mathbb{Z}\cdot\gamma_2.$$ Let $\phi$ be a generic GMN differential. Note that a closed trajectory lies in a cylinder of trajectories of the same phase and any other closed trajectory in this cylinder has the same class in $\Gamma_\phi$. Therefore we can speak of the class of the cylinder.
The *BPS invariant* associated to a generic GMN differential $\phi$ and class $\gamma\in\Gamma_\phi$ is the integer $$\begin{aligned}
\Omega_\phi(\gamma) &= |\{\text{non-closed saddle connections of class $\pm\gamma$}\}| \\
&\quad - 2\cdot|\{\text{nondegenerate cylinders of class $\pm\gamma$}\}|.\end{aligned}$$
As we explain in Appendix \[sec:TheMotivicWallCrossingFormula\], this integer coincides with the BPS invariant defined in a general categorical setting in [@KontsevichSoibelman1]. It is for this reason that these integers $\Omega_\phi(\gamma)$ obey the wall-crossing formula.
From quadratic differentials to ideal triangulations {#sec:FromQuadraticDifferentialsToIdealTriangulations}
====================================================
In this section, we review some additional background material on quadratic differentials. We define a moduli space parametrizing GMN differentials and explain how a general point in this space determines an ideal triangulation of an associated marked bordered surface.
Critical points
---------------
Let $\phi$ be a GMN differential. Suppose that $p$ is a finite critical point of $\phi$, and let $k$ be the order of the singularity so that $k=1$ if $p$ is a simple zero and $k=-1$ if $p$ is a simple pole. By the results of [@Strebel], there is a local coordinate $t$ defined in a neighborhood of $p$ such that $$\phi(t)=\left(\frac{k+2}{2}\right)^2\cdot t^kdt^2.$$ Away from the point $p$, the function $w=t^{\frac{k+2}{2}}$ is a distinguished local coordinate. The horizontal trajectories in a neighborhood of $p$ are illustrated in Figure \[fig:finitecriticalpoint\].
$$\xy /l3pc/:
(1,0)*{}="O";
(-0.35,0.72)*{}="U";
(-0.75,-0.05)*{}="X1";
(-0.6,0.2)*{}="X2";
(-0.45,0.45)*{}="X3";
(-0.2,1)*{}="X4";
(0,1.25)*{}="X5";
(0.15,1.5)*{}="X6";
(1.85,1.5)*{}="Y1";
(2,1.25)*{}="Y2";
(2.2,1)*{}="Y3";
(2.45,0.45)*{}="Y4";
(2.6,0.2)*{}="Y5";
(2.75,-0.05)*{}="Y6";
(1.85,-1.5)*{}="Z1";
(1.55,-1.5)*{}="Z2";
(1.25,-1.5)*{}="Z3";
(0.75,-1.5)*{}="Z4";
(0.45,-1.5)*{}="Z5";
(0.15,-1.5)*{}="Z6";
(2.35,0.72)*{}="V";
(1,-1.5)*{}="W";
"O";"U" **\dir{-};
"O";"V" **\dir{-};
"O";"W" **\dir{-};
"X4";"Y3" **\crv{(0.9,0.2) & (1.1,0.2)};
"X5";"Y2" **\crv{(0.9,0.5) & (1.1,0.5)};
"X6";"Y1" **\crv{(0.9,0.8) & (1.1,0.8)};
"Y4";"Z3" **\crv{(1.35,0) & (1.15,0)};
"Y5";"Z2" **\crv{(1.5,-0.2) & (1.5,-0.3)};
"Y6";"Z1" **\crv{(1.65,-0.4) & (1.85,-0.6)};
"Z4";"X3" **\crv{(0.85,0) & (0.65,0)};
"Z5";"X2" **\crv{(0.5,-0.3) & (0.5,-0.2)};
"Z6";"X1" **\crv{(0.15,-0.6) & (0.35,-0.4)};
(1,0)*{\times};
(1,2)*{k=1};
\endxy
\quad
\xy /l3pc/:
(1,0)*{}="O";
(-0.75,0)*{}="U";
(-0.75,-0.75)*{}="U1";
(-0.75,-0.5)*{}="U2";
(-0.75,-0.25)*{}="U3";
(-0.75,0.25)*{}="U4";
(-0.75,0.5)*{}="U5";
(-0.75,0.75)*{}="U6";
(1,0.75)*{}="T6";
(1,0.5)*{}="T5";
(1,0.25)*{}="T4";
(1,-0.25)*{}="T3";
(1,-0.5)*{}="T2";
(1,-0.75)*{}="T1";
"O";"U" **\dir{-};
"T6";"U6" **\dir{-};
"T5";"U5" **\dir{-};
"T4";"U4" **\dir{-};
"T3";"U3" **\dir{-};
"T2";"U2" **\dir{-};
"T1";"U1" **\dir{-};
"T3";"T4" **\crv{(1.5,-0.25) & (1.5,0.25)};
"T2";"T5" **\crv{(1.9,-0.5) & (1.9,0.5)};
"T1";"T6" **\crv{(2.3,-0.75) & (2.3,0.75)};
(1,0)*{\bullet};
(0.5,1.25)*{k=-1};
\endxy$$
Similarly, if $p$ is a pole of $\phi$ of order two, then there is a local coordinate $t$ defined in a neighborhood of $p$ such that $$\phi(t)=\frac{r}{t^2}dt^2$$ for some well defined constant $r\in\mathbb{C}^*$. We define the *residue* of $\phi$ at $p$ to be the quantity $$\operatorname{\operatorname{Res}}_p(\phi)=\pm4\pi i\sqrt{r},$$ which is well defined up to a sign. Away from $p$, any branch of the function $w=\sqrt{r}\log(t)$ is a distinguished local coordinate. The horizontal foliation can exhibit three possible behaviors in the $t$-plane depending on the value of the residue at $p$:
1. If $\operatorname{\operatorname{Res}}_p(\phi)\in\mathbb{R}$, then the horizontal trajectories are concentric circles centered at the pole.
2. If $\operatorname{\operatorname{Res}}_p(\phi)\in i\mathbb{R}$, then the horizontal trajectories are radial arcs emanating from the pole.
3. If $\operatorname{\operatorname{Res}}_p(\phi)\not\in\mathbb{R}\cup i\mathbb{R}$, then the horizontal trajectories are logarithmic spirals that wrap around the pole.
Figure \[fig:doublepole\] illustrates the three types of foliations.
$$\xy /l1.5pc/:
(1,-3)*\xycircle(3,3){-};
(1,-2)*\xycircle(2,2){-};
(1,-1)*\xycircle(1,1){-};
(1,0)*{\bullet};
(1,4)*{\operatorname{\operatorname{Res}}_p(\phi)\in\mathbb{R}};
\endxy
\qquad
\qquad
\xy /l1.5pc/:
{\xypolygon12"A"{~:{(2,2):}~>{}}};
{(1,0)\PATH~={**@{-}}'"A1"};
{(1,0)\PATH~={**@{-}}'"A2"};
{(1,0)\PATH~={**@{-}}'"A3"};
{(1,0)\PATH~={**@{-}}'"A4"};
{(1,0)\PATH~={**@{-}}'"A5"};
{(1,0)\PATH~={**@{-}}'"A6"};
{(1,0)\PATH~={**@{-}}'"A7"};
{(1,0)\PATH~={**@{-}}'"A8"};
{(1,0)\PATH~={**@{-}}'"A9"};
{(1,0)\PATH~={**@{-}}'"A10"};
{(1,0)\PATH~={**@{-}}'"A11"};
{(1,0)\PATH~={**@{-}}'"A12"};
(1,0)*{\bullet};
(1,4)*{\operatorname{\operatorname{Res}}_p(\phi)\in i\mathbb{R}};
\endxy
\qquad
\xy /l1.5pc/:
(1,0)*{\bullet};
(1.13,0);(-2,0) **\crv{(1.5,0.5) & (0,1.5)};
(1,-0.13);(1,3) **\crv{(1.5,-0.5) & (2.5,1)};
(0.87,0);(4,0) **\crv{(0.5,-0.5) & (2,-1.5)};
(1,0.13);(1,-3) **\crv{(0.5,0.5) & (-0.5,-1)};
(1,4)*{\operatorname{\operatorname{Res}}_p(\phi)\not\in\mathbb{R}\cup i\mathbb{R}};
\endxy$$
Finally, suppose that $p$ is a pole of order $m\geq3$. By the results of [@Strebel], there is a local coordinate $t$ such that $$\phi(t)=\left(\frac{2-m}{2}t^{-m/2}+O(t^{-1})\right)^2dt^2 \quad\text{as $t\rightarrow0$}.$$ In this case, one can show that there are $m-2$ distinguished tangent directions at $p$. There is a neighborhood $U$ of $p$ such that any horizontal trajectory that enters $U$ eventually tends to $p$ along one of these tangent directions. We illustrate this for small values of $m$ in Figure \[fig:higherorderpole\].
$$\xy /l3pc/:
{\xypolygon3"T"{~:{(2,0):}~>{}}},
{\xypolygon3"S"{~:{(1.5,0):}~>{}}},
{\xypolygon3"R"{~:{(1,0):}~>{}}},
(1,0)*{}="O";
(-0.35,0.72)*{}="U";
(2.35,0.72)*{}="V";
(1,-1.5)*{}="W";
"O";"U" **\dir{-};
"O";"V" **\dir{-};
"O";"W" **\dir{-};
"O";"T1" **\crv{(2,0.75) & (2.25,1.85)};
"O";"T1" **\crv{(0,0.75) & (-0.25,1.85)};
"O";"T2" **\crv{(0,0.4) & (-1.2,0.25)};
"O";"T2" **\crv{(1,-1) & (0,-2.2)};
"O";"T3" **\crv{(1,-1) & (2,-2.2)};
"O";"T3" **\crv{(2,0.4) & (3.2,0.25)};
"O";"S1" **\crv{(1.75,0.56) & (2,1.5)};
"O";"S1" **\crv{(0.25,0.56) & (0,1.5)};
"O";"S2" **\crv{(0,0.3) & (-0.9,0.19)};
"O";"S2" **\crv{(0.9,-0.8) & (0.3,-1.7)};
"O";"S3" **\crv{(2,0.3) & (2.9,0.19)};
"O";"S3" **\crv{(1.1,-0.8) & (1.7,-1.7)};
"O";"R1" **\crv{(1.5,0.5) & (1.75,1)};
"O";"R1" **\crv{(0.5,0.5) & (0.25,1)};
"O";"R2" **\crv{(0.5,0.1) & (-0.3,0.25)};
"O";"R2" **\crv{(0.75,-0.8) & (0.5,-1)};
"O";"R3" **\crv{(1.5,0.1) & (2.3,0.25)};
"O";"R3" **\crv{(1.25,-0.8) & (1.5,-1)};
(1,0)*{\bullet};
(1,2.25)*{m=5};
\endxy
\qquad
\xy /l3pc/:
{\xypolygon4"A"{~:{(2,0):}~>{}}},
{\xypolygon4"B"{~:{(1.5,0):}~>{}}},
{\xypolygon4"C"{~:{(1,0):}~>{}}},
(1,0)*{}="O";
(1,-1.75)*{}="T";
(-0.75,0)*{}="U";
(1,1.75)*{}="V";
(2.75,0)*{}="W";
"O";"T" **\dir{-};
"O";"U" **\dir{-};
"O";"V" **\dir{-};
"O";"W" **\dir{-};
"O";"A1" **\crv{(1,1.5) & (1.5,2.25)};
"O";"A1" **\crv{(2.5,0) & (3.25,0.5)};
"O";"A2" **\crv{(1,1.5) & (0.5,2.25)};
"O";"A2" **\crv{(-0.5,0) & (-1.25,0.5)};
"O";"A3" **\crv{(1,-1.5) & (0.5,-2.25)};
"O";"A3" **\crv{(-0.5,0) & (-1.25,-0.5)};
"O";"A4" **\crv{(1,-1.5) & (1.5,-2.25)};
"O";"A4" **\crv{(2.5,0) & (3.25,-0.5)};
"O";"B1" **\crv{(1,1) & (1.5,1.6)};
"O";"B1" **\crv{(2,0) & (2.6,0.5)};
"O";"B2" **\crv{(1,1) & (0.5,1.6)};
"O";"B2" **\crv{(0,0) & (-0.6,0.5)};
"O";"B3" **\crv{(1,-1) & (0.5,-1.6)};
"O";"B3" **\crv{(0,0) & (-0.6,-0.5)};
"O";"B4" **\crv{(1,-1) & (1.5,-1.6)};
"O";"B4" **\crv{(2,0) & (2.6,-0.5)};
"O";"C1" **\crv{(1,0.75) & (1.25,1)};
"O";"C1" **\crv{(1.75,0) & (2,0.25)};
"O";"C2" **\crv{(1,0.75) & (0.75,1)};
"O";"C2" **\crv{(0.25,0) & (0,0.25)};
"O";"C3" **\crv{(1,-0.75) & (0.75,-1)};
"O";"C3" **\crv{(0.25,0) & (0,-0.25)};
"O";"C4" **\crv{(1,-0.75) & (1.25,-1)};
"O";"C4" **\crv{(1.75,0) & (2,-0.25)};
(1,0)*{\bullet};
(1,2.25)*{m=6};
\endxy
\qquad
\dots$$
Moduli spaces
-------------
By a *marked bordered surface*, we mean a compact, connected, oriented surface $\mathbb{S}$ with boundary together with a nonempty finite set $\mathbb{M}\subset\mathbb{S}$ of marked points such that every boundary component of $\mathbb{S}$ contains at least one marked point. A marked point in the interior of $\mathbb{S}$ is called a *puncture*, and the set of all punctures is denoted $\mathbb{P}\subset\mathbb{M}$. An isomorphism of marked bordered surfaces $(\mathbb{S}_1,\mathbb{M}_1)$ and $(\mathbb{S}_2,\mathbb{M}_2)$ is an orientation preserving diffeomorphism $f:\mathbb{S}_1\rightarrow\mathbb{S}_2$ which induces a bijection $\mathbb{M}_1\rightarrow\mathbb{M}_2$. Two such isomorphisms are said to be isotopic if they are related by an isotopy through isomorphisms.
A pair $(S,\phi)$ consisting of a compact Riemann surface $S$ and a GMN differential $\phi$ on $S$ determines an associated marked bordered surface $(\mathbb{S},\mathbb{M})$ by the following construction. To define the surface $\mathbb{S}$, we perform an oriented real blow up of the Riemann surface $S$ at each pole of $\phi$ of order $\geq3$. As we have seen, there are finitely many distinguished tangent directions at each pole of order $\geq3$. These determine points on the boundary of $\mathbb{S}$, and we define $\mathbb{M}$ to be the set consisting of these points together with the poles of order $\leq2$ regarded as punctures.
Let us now fix a marked bordered surface $(\mathbb{S},\mathbb{M})$. If $(S,\phi)$ is a pair consisting of a compact Riemann surface $S$ and a GMN differential $\phi$ on $S$, then we define a *marking* of the pair $(S,\phi)$ by $(\mathbb{S},\mathbb{M})$ to be an isotopy class of isomorphisms from $(\mathbb{S},\mathbb{M})$ to the marked bordered surface determined by $(S,\phi)$. A *marked GMN differential* is a triple $(S,\phi,\theta)$ where $S$ is a compact Riemann surface equipped with a GMN differential $\phi$ and $\theta$ is a marking of the pair $(S,\phi)$ by $(\mathbb{S},\mathbb{M})$. We will consider two such triples $(S_1,\phi_1,\theta_1)$ and $(S_2,\phi_2,\theta_2)$ to be equivalent if there is an isomorphism $f:S_1\rightarrow S_2$ of Riemann surfaces satisfying $f^*(\phi_2)=\phi_1$ and commuting with the markings $\theta_i$ in the obvious way. We denote by $\mathscr{Q}(\mathbb{S},\mathbb{M})$ the moduli space of equivalence classes of marked GMN differentials. Assuming $\mathbb{S}$ is not a genus zero surface with $|\mathbb{M}|\leq2$, Proposition 6.2 of [@Allegretti19b] implies that $\mathscr{Q}(\mathbb{S},\mathbb{M})$ has the structure of a complex manifold.
In fact, it will be important to enhance this construction slightly and consider a moduli space parametrizing marked GMN differentials with additional data. Recall that if $\phi$ is a GMN differential having a pole of order two at $p$, then the residue $\operatorname{\operatorname{Res}}_p(\phi)$ is well defined up to sign. We define a signing for $\phi$ to be a choice of sign for the residue at each pole of order two. There is a branched cover $$\mathscr{Q}^\pm(\mathbb{S},\mathbb{M})\rightarrow\mathscr{Q}(\mathbb{S},\mathbb{M})$$ of degree $2^{|\mathbb{P}|}$ obtained by choosing a signing for each differential in $\mathscr{Q}(\mathbb{S},\mathbb{M})$ having a pole of order two. It is branched precisely over the locus of differentials having simple poles. Assuming once again that $\mathbb{S}$ is not a genus zero surface with $|\mathbb{M}|\leq2$, Proposition 6.3 of [@Allegretti19b] gives $\mathscr{Q}^\pm(\mathbb{S},\mathbb{M})$ the structure of a complex manifold.
Horizontal strip decomposition
------------------------------
In general, if $\phi$ is a GMN differential on a compact Riemann surface $S$, then any horizontal trajectory of $\phi$ is necessarily one of the following (see [@Strebel], Sections 9–11):
1. A *saddle trajectory*, which connects two finite critical points of $\phi$.
2. A *separating trajectory*, which connects a finite and an infinite critical point of $\phi$.
3. A *generic trajectory*, which connects two infinite critical points of $\phi$.
4. A *closed trajectory*, which is a simple closed curve in $S\setminus\operatorname{\operatorname{Crit}}(\phi)$.
5. A *recurrent trajectory*, which has a limit set with nonempty interior in $S$.
We will be particularly interested in quadratic differentials having no saddle trajectories. Such a differential is said to be *saddle-free*.
Suppose $\phi$ is a saddle-free GMN differential with $\operatorname{\operatorname{Crit}}_\infty(\phi)\neq\emptyset$. Then according to Lemma 3.1 of [@BridgelandSmith], this differential $\phi$ has no closed or recurrent trajectories. Moreover, since $\phi$ has at most finitely many zeros, there can be at most finitely many separating trajectories. If we remove these separating trajectories from $S$, then the remaining open surface splits as a union of connected components, and each component is one of the following:
1. A *horizontal strip*, which is is a maximal domain in $S$ that corresponds, via the distinguished local coordinate, to a region $$\{w\in\mathbb{C}:a<{\operatorname{Im}}(w)<b\}\subset\mathbb{C}.$$ Every trajectory in a horizontal strip is generic, connecting two (not necessarily distinct) poles of $\phi$. Each component of the boundary is composed of separating trajectories.
2. A *half plane*, which is a maximal domain in $S$ that corresponds, via the distinguished local coordinate, to a region $$\{w\in\mathbb{C}:{\operatorname{Im}}(w)>0\}\subset\mathbb{C}.$$ The trajectories in a half plane are generic, connecting a fixed pole of order $>2$ to itself. The boundary is composed of separating trajectories.
This decomposition of the surface into horizontal strips and half planes is called the *horizontal strip decomposition*.
Ideal triangulations
--------------------
If $(\mathbb{S},\mathbb{M})$ is a marked bordered surface, then an *arc* on $(\mathbb{S},\mathbb{M})$ is defined as a smooth path $\gamma$ in $\mathbb{S}$ connecting points of $\mathbb{M}$ whose interior lies in $\mathbb{S}\setminus\mathbb{M}$ and which has no self-intersections in its interior. In addition, we require that $\gamma$ is not homotopic, relative to its endpoints, to a single point or to a path in $\partial\mathbb{S}$ whose interior contains no marked points. Two arcs are considered to be equivalent if they are homotopic relative to their endpoints, and they are *compatible* if there exist arcs in their respective equivalence classes which do not intersect in the interior of the surface $\mathbb{S}$.
An *ideal triangulation* of $(\mathbb{S},\mathbb{M})$ is defined to be a maximal set of pairwise compatible arcs, considered up to equivalence. When talking about an ideal triangulation $T$ of $(\mathbb{S},\mathbb{M})$, we always fix representatives for its arcs so that no two arcs intersect in the interior of $\mathbb{S}$. Then a *triangle* of $T$ is defined to be the closure in $\mathbb{S}$ of a component of the complement of the arcs of $T$. Any triangle is homeomorphic to a disk with two or three marked points. If a triangle contains only two marked points, it is said to be *self-folded*. In this case, it contains an arc in its interior called the *self-folded edge*. The boundary of a self-folded triangle is called the *encircling edge*.
If $\phi$ is a complete, saddle-free GMN differential on a compact Riemann surface $S$, then $\phi$ determines an ideal triangulation of the associated marked bordered surface $(\mathbb{S},\mathbb{M})$. Indeed, $\phi$ determines a horizontal strip decomposition of the underlying Riemann surface $S$, and we can choose a single generic trajectory within each of the horizontal strips of this decomposition. After passing to the real blow up $\mathbb{S}$, these trajectories become arcs of an ideal triangulation. This ideal triangulation is known as the *WKB triangulation* for $\phi$.
In general, if $(\mathbb{S},\mathbb{M})$ is a marked bordered surface, then a *signing* for $(\mathbb{S},\mathbb{M})$ is a function $\epsilon:\mathbb{P}\rightarrow\{\pm1\}$ associating a sign $\epsilon(p)=\pm1$ to every puncture $p\in\mathbb{P}\subset\mathbb{M}$. A *signed triangulation* of $(\mathbb{S},\mathbb{M})$ is a pair $(T,\epsilon)$ consisting of an ideal triangulation $T$ and a signing $\epsilon$ of $(\mathbb{S},\mathbb{M})$. Let us define the *valency* of a puncture $p\in\mathbb{P}$ to be the number of half arcs of $T$ that are incident to $p$. Then two signed triangulations $(T_1,\epsilon_1)$ and $(T_2,\epsilon_2)$ are considered to be equivalent if $T_1=T_2$ and the signings $\epsilon_1$ and $\epsilon_2$ differ at a puncture $p$ only if $p$ has valency one. An equivalence class of signed triangulations is called a *tagged triangulation*.
Suppose $T$ is the WKB triangulation for a complete, saddle-free differential $\phi\in\mathscr{Q}^\pm(\mathbb{S},\mathbb{M})$. If $p$ is a pole of order two with real residue, then $p$ forms one boundary component of a cylinder of horizontal trajectories. But then the other boundary component consists of saddle trajectories, contradicting the assumption that $\phi$ is saddle-free. It follows that the residue at a pole of order two cannot be real. The choice of the point $\phi\in\mathscr{Q}^\pm(\mathbb{S},\mathbb{M})$ includes a choice of sign for the residue for each pole $p$ of order two, and we can choose the sign $\epsilon(p)\in\{\pm1\}$ so that $$\epsilon(p)\cdot\operatorname{\operatorname{Res}}_p(\phi)\in\mathbb{H}$$ where $\mathbb{H}\subset\mathbb{C}$ is the upper half plane.
In this way, we associate a tagged triangulation to any complete, saddle-free differential $\phi\in\mathscr{Q}^\pm(\mathbb{S},\mathbb{M})$. Conversely, if $\tau$ is an arbitrary tagged triangulation of $(\mathbb{S},\mathbb{M})$, then there is an associated subset $\mathcal{C}_\tau\subset\mathscr{Q}^\pm(\mathbb{S},\mathbb{M})$ consisting of all points $\phi$ with associated tagged triangulation $\tau$. These subsets $\mathcal{C}_\tau$ are open and their union $\coprod_\tau\mathcal{C}_\tau\subset\mathscr{Q}^\pm(\mathbb{S},\mathbb{M})$ is the subset of all complete saddle-free differentials.
Spin structures and the twisted torus {#sec:SpinStructuresAndTheTwistedTorus}
=====================================
In this section, we discuss spin structures arising from GMN differentials. The idea of the construction comes from [@KontsevichZorich]. We discuss a related algebraic variety called the twisted torus, which appears in our formulation of the wall-crossing formula.
Spin structures
---------------
We begin by recalling the topological definition of a spin structure on a Riemann surface, following [@Atiyah; @KontsevichZorich]. Given a Riemann surface $\Sigma$, let $P_\Sigma\rightarrow\Sigma$ be the bundle whose fiber over $p\in\Sigma$ is the circle of nonzero tangent directions at $p$. Then a *spin structure* on $\Sigma$ is a double cover $Q\rightarrow P_\Sigma$ whose restriction to any fiber of $P_\Sigma$ is the standard double cover $S^1\rightarrow S^1$.
As shown in Section 3 of [@Atiyah], a line bundle $L$ on $\Sigma$ satisfying $L\otimes L\cong\omega_\Sigma$ determines a corresponding spin structure on $\Sigma$. Indeed, consider the map of line bundles $$\label{eqn:doublecover}
L^{-1}\rightarrow L^{-1}\otimes L^{-1}=\omega_{\Sigma}^{-1}$$ taking a section $s$ of $L^{-1}$ to the tensor product $s\otimes s$. Note that the sections $s$ and $-s$ have the same image so that induces a double cover $Q\rightarrow P_\Sigma$. It restricts to the standard double cover $S^1\rightarrow S^1$ on the fibers of $P_\Sigma$ and therefore defines a spin structure on $\Sigma$.
Now suppose we are given a GMN differential $\phi$ on a compact Riemann surface $S$. The differential $\phi$ determines the canonical double cover $\Sigma_\phi\rightarrow S$, and there is a canonical meromorphic 1-form $\lambda$ on $\Sigma_\phi$ which is holomorphic on the punctured surface $\Sigma_\phi^\circ$. This 1-form $\lambda$ can be regarded as a global section of the line bundle $\omega_{\Sigma_\phi^\circ}$. Therefore, if we write $D$ for the divisor of zeros of $\lambda$, then $\omega_{\Sigma_\phi^\circ}$ is isomorphic to the line bundle $[D]$ associated to $D$. By Lemma \[lem:doublezeros\], we can write $$D=2p_1+\dots+2p_s$$ where $p_1,\dots,p_s$ are the zeros of $\lambda$. We consider the line bundle $$L\coloneqq[p_1+\dots+p_s],$$ which has the property $L\otimes L\cong\omega_{\Sigma_\phi^\circ}$. Thus we have a canonical square root of the holomorphic cotangent bundle, and by the remarks above, there is an associated spin structure on $\Sigma_\phi^\circ$.
Associated quadratic forms
--------------------------
If $Q\rightarrow P_\Sigma$ is a spin structure for the Riemann surface $\Sigma$, then the group of deck transformations for the covering space $Q$ is isomorphic to $\mathbb{Z}_2$. Thus we have a group homomorphism $\pi_1(P_\Sigma)\rightarrow\mathbb{Z}_2$, and this homomorphism factors through $H_1(P_\Sigma;\mathbb{Z}_2)$. In this way, we see that spin structures on $\Sigma$ correspond bijectively to linear maps $H_1(P_\Sigma;\mathbb{Z}_2)\rightarrow\mathbb{Z}_2$ which are nonzero on the cycle represented by the $S^1$ fiber of $P_\Sigma$. In particular, spin structures can be viewed as elements of $H^1(P_\Sigma;\mathbb{Z}_2)$.
Note that if $\gamma$ is a smooth oriented simple closed curve on $\Sigma$, then there is an obvious lift $\vec{\gamma}$ of $\gamma$ to $P_\Sigma$ obtained by framing $\gamma$ with the unit tangent vector field. If $\beta$ is the oriented curve obtained from $\gamma$ by reversing the orientation, then the corresponding lift $\vec{\beta}$ is in fact homotopic to $\vec{\gamma}$; the homotopy is given by simultaneously rotating all framing vectors by an angle $\pi$.
Now suppose that $\gamma=\sum_{i=1}^m\gamma_i$ is a chain representing a class in $H_1(\Sigma;\mathbb{Z}_2)$ where each $\gamma_i$ is a smooth simple closed curve on $\Sigma$. Choosing an orientation for each of these curves $\gamma_i$, we obtain the lifts $\vec{\gamma}_i$ in $P_\Sigma$. Let $z$ denote the cycle in $P_\Sigma$ given by the $S^1$ fiber. Then $$\tilde{\gamma}=\sum_{i=1}^m\vec{\gamma}_i+mz$$ represents an element of $H_1(P_\Sigma;\mathbb{Z}_2)$. By Theorem 1A in [@Johnson], this element depends only on the class of $\gamma$ in $H_1(\Sigma,\mathbb{Z}_2)$ and not on its representation as a sum of smooth simple closed curves or on the choice of orientations for these curves. Therefore we have a map of sets $H_1(\Sigma;\mathbb{Z}_2)\rightarrow H_1(P_\Sigma;\mathbb{Z}_2)$ given by $\gamma\mapsto\tilde{\gamma}$. The following result explains precisely how this map fails to be a group homomorphism.
\[prop:nonhomom\] The mapping defined above satisfies $$\widetilde{\gamma_1+\gamma_2}=\tilde{\gamma_1}+\tilde{\gamma_2}+(\gamma_1\cdot\gamma_2)z$$ where $\gamma_1\cdot\gamma_2$ denotes the intersection pairing of $\gamma_1$, $\gamma_2\in H_1(\Sigma;\mathbb{Z}_2)$.
By a $\mathbb{Z}_2$-valued *quadratic form* on $H_1(\Sigma;\mathbb{Z}_2)$ with the associated bilinear form $(\gamma_1,\gamma_2)\mapsto\gamma_1\cdot\gamma_2$, we mean any function $Q:H_1(\Sigma;\mathbb{Z}_2)\rightarrow\mathbb{Z}_2$ such that $$\label{eqn:quadraticform}
Q(\gamma_1+\gamma_2)=Q(\gamma_1)+Q(\gamma_2)+\gamma_1\cdot\gamma_2$$ for $\gamma_1$, $\gamma_2\in H_1(\Sigma;\mathbb{Z}_2)$. It follows from Proposition \[prop:nonhomom\] that if $\omega\in H^1(P_\Sigma;\mathbb{Z}_2)$ is a spin structure, then there is an associated quadratic form $Q_\omega$ given by $$Q_\omega(\gamma)=\omega(\tilde{\gamma})$$ where the right hand side denotes the pairing of homology and cohomology.
Spin structures from differentials
----------------------------------
We have now seen that any spin structure on a Riemann surface gives rise to an associated quadratic form. In particular, if we have a GMN differential $\phi$, then there is an associated spin structure $\omega$ on the surface $\Sigma_\phi^\circ$, and this gives rise to a quadratic form $Q_\omega$ on $H_1(\Sigma_\phi^\circ;\mathbb{Z}_2)$. Given a class $\gamma\in\Gamma_\phi$, let us write $\gamma$ also for the image of this class under the map $\Gamma_\phi\rightarrow H_1(\Sigma_\phi^\circ;\mathbb{Z}_2)$ given by reduction modulo 2. We will be interested in the map $$\xi:\Gamma_\phi\rightarrow\mathbb{C}^*, \quad \xi(\gamma)=(-1)^{Q_\omega(\gamma)}.$$ By the identity , this map satisfies $$\label{eqn:basepoint}
\xi(\gamma_1+\gamma_2)=(-1)^{\langle\gamma_1,\gamma_2\rangle}\xi(\gamma_1)\xi(\gamma_2)$$ where $\langle\gamma_1,\gamma_2\rangle$ denotes the intersection pairing of $\gamma_1$, $\gamma_2\in\Gamma_\phi$. Below we will give an alternative description of this quadratic form, using ideas from [@KontsevichZorich] (see also [@GMN], Section 7.7).
Consider the differential $\lambda$ on the Riemann surface $\Sigma_\phi^\circ$. At any point of $\Sigma_\phi^\circ$ which is not a zero of $\lambda$, we can find a tangent vector $v$ such that $\lambda(v)=1$. These tangent vectors provide a nonvanishing section of the tangent bundle of $\Sigma_\phi^\circ$ over the complement of the zeros of $\lambda$, and therefore the tangent bundle is trivial over this set. Let $\gamma$ be a smooth oriented closed curve on $\Sigma_\phi^\circ$ which does not meet any zero of $\lambda$. Since every tangent space over $\gamma$ is canonically identified with $\mathbb{C}$, we get a map $G:\gamma\rightarrow S^1\subset\mathbb{C}$ sending a point $p$ on $\gamma$ to the tangent direction determined by the orientation of $\gamma$ at $p$. We define the *index* $\operatorname{\operatorname{ind}}_\gamma(\lambda)$ to be the integer such that $2\pi\cdot\operatorname{\operatorname{ind}}_\gamma(\lambda)$ is the total change in angle between $G(p)$ and $1\in S^1$ as $p$ goes around the curve $\gamma$.
\[prop:indexquadraticform\] Let $\omega\in H^1(P_\Sigma;\mathbb{Z}_2)$ be the spin structure on $\Sigma=\Sigma_\phi^\circ$ determined by the GMN differential $\phi$. If $\gamma$ is any smooth oriented simple closed curve on $\Sigma$ which does not meet a zero of $\lambda$, then $$Q_\omega(\gamma)=\operatorname{\operatorname{ind}}_\gamma(\lambda)+1\mod2.$$
Let $\vec{\gamma}$ be the lift of $\gamma$ to $P_\Sigma$ given by the unit tangent vector field. Since $\omega(\tilde{\gamma})=\omega(\vec{\gamma})+1$, we need to show that $$\omega(\vec{\gamma})=\operatorname{\operatorname{ind}}_\gamma(\lambda)\mod2.$$ First, note that if $z$ is the loop in $P_\Sigma$ given by the $S^1$ fiber of $P_\Sigma\rightarrow\Sigma$, then we have $\omega(z)=1\in\mathbb{Z}_2$ since $\omega$ is a spin structure. On the other hand, suppose $\alpha$ is a loop in $P_\Sigma$ which is obtained by framing a loop in $\Sigma$ so that the framing vector $v$ at any point satisfies $\lambda(v)=1$. In this case, the lift of $\alpha$ to the double cover induced by lies in a single sheet, so $\omega(\alpha)=0$. The result now follows since $\vec{\gamma}$ is homotopic to a concatenation of loops $z$ and $\alpha$ as above.
Using Proposition \[prop:indexquadraticform\], we can give an alternative description of the map $\xi$ introduced above. Indeed, if $\gamma=\sum_i\gamma_i$ is a class in $\Gamma_\phi$ which is a sum of mutually nonintersecting smooth oriented simple closed curves $\gamma_i$ that do not meet the zeros of $\lambda$, then we have $$\label{eqn:formulaforxi}
\xi(\gamma)=\prod_i(-1)^{\operatorname{\operatorname{ind}}_{\gamma_i}(\lambda)+1}$$ by Proposition \[prop:indexquadraticform\] and the fact that $\langle\gamma_i,\gamma_j\rangle=0$ for $i\neq j$. This formula provides a means of computing the value of the map $\xi$ on a given class in $\Gamma_\phi$.
The map $\xi$ satisfies $\xi(\gamma)=-1$ if $\gamma\in\Gamma_\phi$ is the class of a non-closed saddle connection and $\xi(\gamma)=+1$ if $\gamma$ is the class of a closed saddle connection.
If $\gamma\in\Gamma_\phi$ is the class of a non-closed saddle connection, then $\gamma$ is represented by a single smooth oriented closed curve of constant phase in $\Sigma_\phi^\circ$, while if $\gamma$ is the class of a closed saddle connection, then it is represented by a union of two disjoint curves of this type. The proposition therefore follows from .
The twisted torus
-----------------
Let $\phi$ be a GMN differential, and consider the associated lattice $\Gamma_\phi\cong\mathbb{Z}^n$ equipped with the skew form $\langle-,-\rangle$. This lattice defines an algebraic torus $$\mathbb{T}_+=\operatorname{\operatorname{Hom}}_{\mathbb{Z}}(\Gamma_\phi,\mathbb{C}^*)\cong(\mathbb{C}^*)^n.$$ We will be interested in a related object $$\mathbb{T}_-=\left\{g:\Gamma_\phi\rightarrow\mathbb{C}^*:g(\gamma_1+\gamma_2)=(-1)^{\langle\gamma_1,\gamma_2\rangle}g(\gamma_1)g(\gamma_2)\right\}$$ known as the *twisted torus* [@Bridgeland19].
There is an action of $\mathbb{T}_+$ on $\mathbb{T}_-$ given by $$(f\cdot g)(\gamma)=f(\gamma)g(\gamma)\in\mathbb{C}^*$$ for $f\in\mathbb{T}_+$ and $g\in\mathbb{T}_-$, and this action is free and transitive. Thus, after choosing a basepoint in the twisted torus $\mathbb{T}_-$, we get an identification of $\mathbb{T}_-$ and $\mathbb{T}_+$. This identification gives $\mathbb{T}_-$ the structure of an algebraic variety, and this variety structure is independent of the choice of basepoint since the transition maps on $\mathbb{T}_+$ are algebraic. The coordinate ring $\mathbb{C}[\mathbb{T}_-]$ of the twisted torus is spanned as a vector space by the functions $$x_\gamma:\mathbb{T}_-\rightarrow\mathbb{C}^*, \quad x_\gamma(g)=g(\gamma)\in\mathbb{C}^*,$$ which we call the *twisted characters*. The twisted torus $\mathbb{T}_-$ has a natural Poisson structure with the Poisson bracket given on the twisted characters by $$\{x_\alpha,x_\beta\}=\langle\alpha,\beta\rangle\cdot x_\alpha\cdot x_\beta$$ for $\alpha$, $\beta\in\Gamma_\phi$.
For any GMN differential $\phi$, we have constructed an associated map $\xi:\Gamma_\phi\rightarrow\mathbb{C}^*$. This map satisfies the relation and can therefore be considered as a point in the twisted torus $\mathbb{T}_-$. In the following, we will use $\xi$ as a canonical basepoint to identify $\mathbb{T}_-$ and $\mathbb{T}_+$. When there is no possibility of confusion, we will denote the twisted torus simply by $\mathbb{T}$.
The wall-crossing formula for quadratic differentials {#sec:TheWallCrossingFormulaForQuadraticDifferentials}
=====================================================
In this section, we formulate the Kontsevich-Soibelman wall-crossing formula as a statement about finite-length trajectories of quadratic differentials.
The ray diagram
---------------
Fix a GMN differential $\phi$. When we talk about a *ray* in $\mathbb{C}^*$, we will always mean a subset of the form $\ell=\mathbb{R}_{>0}\cdot e^{i\pi\theta}$ for some $\theta\in\mathbb{R}$. The ray $\ell$ is said to be *active* if there exists a finite-length trajectory for $\phi$ with phase $\theta$.
The *ray diagram* associated to the differential $\phi$ is defined as the union of all active rays in $\mathbb{C}^*$. An example is illustrated in Figure \[fig:raydiagram\]. Note that the phase of a finite-length trajectory is well defined up to addition of an integer, and therefore if $\ell$ is a ray in the ray diagram, then so is $-\ell$.
$$\xy /l1.5pc/:
{\xypolygon18"A"{~:{(2,2):}~>{}}};
(1,0)*{\circ}="a";
(3,-2.8)*{Z_\phi(\gamma_1)};
(-1,2.8)*{-Z_\phi(\gamma_1)};
(-0.5,-3)*{Z_\phi(\gamma_2)};
(2.5,3)*{-Z_\phi(\gamma_2)};
(-2.5,-1.5)*{Z_\phi(\gamma_3)};
(4.75,1.5)*{-Z_\phi(\gamma_3)};
\xygraph{
"a":"A2",
"a":"A11",
"a":"A9",
"a":"A18",
"a":"A5",
"a":"A14",
}
\endxy$$
Suppose now that the differential $\phi$ is generic. Then the *height* of a ray $\ell\subset\mathbb{C}^*$ is defined to be the number $$H(\ell)=\inf\{|Z_\phi(\gamma)|:\text{$\gamma\in\Gamma_\phi$ such that $Z_\phi(\gamma)\in\ell$ and $\Omega_\phi(\gamma)\neq0$}\},$$ whenever the set on the right hand side of this expression is non-empty. Otherwise, $\ell$ is considered to have infinite height. It follows from the remarks in [@Bridgeland19], Section 2.5, that for any $H>0$ one has at most finitely many rays of height $<H$.
BPS automorphisms {#sec:BPSautomorphisms}
-----------------
If $\phi$ is a generic GMN differential, then the *Donaldson-Thomas invariant* for $\gamma\in\Gamma$ is defined by the formula $$DT_\phi(\gamma)=\sum_{\gamma=m\alpha}\frac{1}{m^2}\Omega_\phi(\alpha)\in\mathbb{Q}$$ where the sum is over all integers $m>0$ such that $\gamma$ is divisible by $m$ in the lattice $\Gamma$. By the Möbius inversion formula, one can express the numbers $\Omega_\phi(\gamma)$ in terms of the $DT_\phi(\gamma)$, and so the BPS and Donaldson-Thomas invariants are equivalent data.
Given any ray $\ell\subset\mathbb{C}^*$, we can consider the associated formal generating series $$\label{eqn:generatingseries}
DT_\phi(\ell)=\sum_{Z_\phi(\gamma)\in\ell}DT_\phi(\gamma)\cdot x_\gamma$$ for Donaldson-Thomas invariants. We would like to view this generating series as a well defined holomorphic function on the twisted torus $\mathbb{T}$. To do this, we must specify the a suitable domain in $\mathbb{T}$. For any acute sector $\Delta\subset\mathbb{C}^*$ and real number $R>0$, we consider the open set $U_\Delta(R)\subset\mathbb{T}$ defined as the interior of $$\left\{g\in\mathbb{T}:Z_\phi(\gamma)\in\Delta\text{ and }\Omega_\phi(\gamma)\neq0\implies|g(\gamma)|<\exp(-R\|\gamma\|)\right\}\subset\mathbb{T}.$$ It is nonempty by Lemma B.2 of [@Bridgeland19]. Note that for each $\gamma\in\Gamma_\phi$, the corresponding BPS invariant satisfies $|\Omega_\phi(\gamma)|\leq2$ by Lemma 5.1 of [@BridgelandSmith]. Hence for sufficiently large $R>0$, we have $$\sum_{\gamma\in\Gamma}|\Omega_\phi(\gamma)|\cdot e^{-R|Z_\phi(\gamma)|}<\infty.$$ This shows that the data $(\Gamma_\phi,Z_\phi,\Omega_\phi)$ define a convergent BPS structure in the sense of [@Bridgeland19]. Hence we have the following.
\[prop:BPSautomorphism\] Let $\Delta\subset\mathbb{C}^*$ be a convex sector. Then for sufficiently large $R>0$, the following statements hold:
1. For each ray $\ell\subset\Delta$, the power series is absolutely convergent on $U_\Delta(R)$ and thus defines a holomorphic function $$DT_\phi(\ell):U_\Delta(R)\rightarrow\mathbb{C}.$$
2. The time-1 Hamiltonian flow $\exp\{DT_\phi(\ell),-\}$ of the function $DT_\phi(\ell)$ defines a holomorphic embedding $$\mathbf{S}_\phi(\ell):U_\Delta(R)\rightarrow\mathbb{T}.$$
3. For every $H>0$, the composition $$\mathbf{S}_{\phi,<H}(\Delta)=\mathbf{S}_\phi(\ell_1)\circ\mathbf{S}_\phi(\ell_2)\circ\dots\circ\mathbf{S}_\phi(\ell_k)$$ exists where $\ell_1,\ell_2,\dots,\ell_k\subset\Delta$ are the rays of height $<H$ in the sector $\Delta$ in the clockwise order, and the pointwise limit $$\mathbf{S}_\phi(\Delta)=\lim_{H\rightarrow\infty}\mathbf{S}_{\phi,<H}(\Delta):U_\Delta(R)\rightarrow\mathbb{T}$$ is a well defined holomorphic embedding.
We think of the map $\mathbf{S}_\phi(\Delta)$ defined by Proposition \[prop:BPSautomorphism\] as a partially defined automorphism of the twisted torus and call it the *BPS automorphism* associated to the sector $\Delta$. We note that similar analytic maps have been studied by Kontsevich and Soibelman [@KontsevichSoibelman4], who described a general framework for studying wall-crossing formulas in an analytic context.
For a generic GMN differential $\phi$, the data $(\Gamma_\phi,Z_\phi,\Omega_\phi)$ form a generic, integral, and ray-finite BPS structure in the sense of [@Bridgeland19]. Using this fact, we can give a more explicit description of the maps $\mathbf{S}_\phi(\ell)$.
\[prop:KSautomorphism\] If $\phi$ is a generic GMN differential, then for any ray $\ell\subset\mathbb{C}^*$, the holomorphic embedding $\mathbf{S}_\phi(\ell)$ extends to a birational automorphism of $\mathbb{T}$, whose action on the twisted characters is given by $$\mathbf{S}_\phi(\ell)^*(x_\beta)=x_\beta\cdot\prod_{Z_\phi(\gamma)\in\ell}(1-x_\gamma)^{\Omega_\phi(\gamma)\cdot\langle\beta,\gamma\rangle}.$$
The wall-crossing formula
-------------------------
We will now study the behavior of the BPS automorphisms $\mathbf{S}_\phi(\Delta)$ as we vary the differential $\phi$ in the moduli space $\mathscr{Q}^\pm(\mathbb{S},\mathbb{M})$ for a fixed marked bordered surface $(\mathbb{S},\mathbb{M})$. Note that if $\tau$ is any tagged triangulation of $(\mathbb{S},\mathbb{M})$, then the canonical double covers $\Sigma_\phi$ define a family of Riemann surfaces over the open set $\mathcal{C}_\tau\subset\mathscr{Q}(\mathbb{S},\mathbb{M})$. It follows that the lattices $\Gamma_\phi$ form a local system over $\mathcal{C}_\tau$ with flat connection given by the Gauss-Manin connection. Using this flat connection, we can identify the $\Gamma_\phi$ for $\phi\in\mathcal{C}_\tau$ with a single lattice. In particular, the associated twisted torus $\mathbb{T}$ is independent of $\phi\in\mathcal{C}_\tau$.
In Appendix \[sec:TheMotivicWallCrossingFormula\], we derive the following statement from the motivic wall-crossing formula. It describes implicitly how the BPS invariants $\Omega_\phi(\gamma)$ jump as $\phi$ varies.
\[thm:firstWCF\] Let $\tau$ be a tagged triangulation of a marked bordered surface $(\mathbb{S},\mathbb{M})$, and let $\Delta$ be a sector contained in the upper half plane. Suppose $\phi_t$, $t\in[0,1]$, is a path in $\mathcal{C}_\tau\subset\mathscr{Q}^\pm(\mathbb{S},\mathbb{M})$ with generic endpoints such that the boundary rays of $\Delta$ are non-active for each differential $\phi_t$. Then $$\mathbf{S}_{\phi_0}(\Delta)=\mathbf{S}_{\phi_1}(\Delta).$$
While this is the version of the wall-crossing formula that follows most readily from the one stated in the references [@Meinhardt; @DavisonMeinhardt1], our main result will ultimately allow us to drop the requirement that the path $\phi_t$ lies entirely in the domain $\mathcal{C}_\tau$.
Birationality of BPS automorphisms
==================================
In this section, we prove that the BPS automorphisms $\mathbf{S}_\phi(\Delta)$ introduced in the last section extend to birational automorphisms of the twisted torus $\mathbb{T}$. We do this by relating the BPS automorphisms to Fock-Goncharov coordinates on moduli spaces of flat $\operatorname{\operatorname{PGL}}_2(\mathbb{C})$-connections.
Fock-Goncharov coordinates
--------------------------
The birational transformations that we consider in this section arose from the work of Fock and Goncharov on an algebro-geometric approach to higher Teichmüller theory [@FockGoncharov1]. For any marked bordered surface $(\mathbb{S},\mathbb{M})$, Fock and Goncharov defined a moduli space denoted $\mathscr{X}(\mathbb{S},\mathbb{M})$. This moduli space parametrizes flat $\operatorname{\operatorname{PGL}}_2(\mathbb{C})$-connections on the punctured surface $\mathbb{S}\setminus\mathbb{M}$ with additional data associated to the marked points. We refer to [@Allegretti19b] for the details of this construction. What is important for us is that this moduli space $\mathscr{X}(\mathbb{S},\mathbb{M})$ has an interesting atlas of coordinate charts, which we will now describe.
Fix a marked bordered surface $(\mathbb{S},\mathbb{M})$, and assume this surface admits an ideal triangulation with $n>1$ arcs. Let $\tau$ be any tagged triangulation of $(\mathbb{S},\mathbb{M})$, and let $(T,\epsilon)$ be a signed triangulation representing $\tau$. We define a *tagged arc* of $\tau$ to be an arc of the ideal triangulation $T$. If $(T,\epsilon')$ is another signed triangulation where the signing $\epsilon'$ differs from $\epsilon$ at a single puncture $p$ of valency one with respect to $T$, then $p$ is incident to a unique arc $i$ which is the interior edge of a self-folded triangle with encircling edge $j$. In this case the tagged arc represented $j$ in $(T,\epsilon')$ is considered to be equivalent to the tagged arc represented by $i$ in $(T,\epsilon)$. In what follows, we will write $\Gamma_\tau\cong\mathbb{Z}^n$ for a lattice with basis $\{\gamma_j\}$ indexed by tagged arcs $j$ of $\tau$.
The lattice $\Gamma_\tau$ comes with a natural skew-form $\langle-,-\rangle:\Gamma_\tau\times\Gamma_\tau\rightarrow\mathbb{Z}$. To define it, let $(T,\epsilon)$ be a signed triangulation representing $\tau$. If $j$ is any arc of $T$, we consider the arc $\pi_T(j)$ defined as follows: If $j$ is the interior edge of a self-folded triangle, we let $\pi_T(j)$ be the encircling edge, and we let $\pi_T(j)=j$ otherwise. For each non-self-folded triangle $t$ of $T$, we define a number denoted $b_{ij}^t$ by the following rules:
1. $b_{ij}^t=+1$ if $\pi_T(i)$ and $\pi_T(j)$ are edges of $t$ with $\pi_T(j)$ following $\pi_T(i)$ in the clockwise order defined by the orientation.
2. $b_{ij}^t=-1$ if the same holds with the counterclockwise order.
3. $b_{ij}^t=0$ otherwise.
Finally, if $i$ and $j$ are tagged arcs of $\tau$, we define $$\langle\gamma_j,\gamma_i\rangle=\sum_tb_{ij}^t$$ where the sum is over all non-self-folded triangles in $T$. We extend this to a form on $\Gamma_\tau$ by bilinearity.
It follows from the work of Fock and Goncharov [@FockGoncharov1] that there exists a birational map $$X_\tau:\mathscr{X}(\mathbb{S},\mathbb{M})\dashrightarrow\mathbb{T}_\tau\coloneqq\operatorname{\operatorname{Hom}}_{\mathbb{Z}}(\Gamma_\tau,\mathbb{C}^*)$$ from the moduli space described above to an algebraic torus $\mathbb{T}_\tau\cong(\mathbb{C}^*)^n$. The components of this map corresponding to tagged arcs of $\tau$ are called *Fock-Goncharov coordinates*. We refer to [@Allegretti19b] for the detailed construction of this map $X_\tau$. Here we are mainly interested in the relationship between the Fock-Goncharov coordinates associated to different tagged triangulations.
Suppose $T$ is an ideal triangulation of $(\mathbb{S},\mathbb{M})$ and $k$ is an arc of $T$. We say that an ideal triangulation $T'$ is obtained from $T$ by a *flip* of $k$ if $T'\neq T$ and there is an arc $k'$ of $T'$ such that $T\setminus\{k\}=T'\setminus\{k'\}$. Similarly, suppose $\tau$ is a tagged triangulation and $k$ is a tagged arc of $\tau$. In this case, we say that a tagged triangulation $\tau'$ is obtained from $\tau$ by a flip of $k$ if $\tau$ and $\tau'$ are represented by signed triangulations $(T,\epsilon)$ and $(T',\epsilon')$, respectively, and $T'$ is obtained from $T$ by a flip of $k$. Note that while we cannot flip a self-folded edge in an ordinary ideal triangulation, we can flip any tagged arc of a tagged triangulation.
Let $\tau$ be a tagged triangulation and $\tau'$ the tagged triangulation obtained from $\tau$ by flipping the tagged arc $k$. Then the transition map $\mu_k=X_{\tau'}\circ X_\tau^{-1}$ is a birational map $$\mu_k:\mathbb{T}_\tau\dashrightarrow\mathbb{T}_{\tau'}$$ which can be described explicitly. In the following proposition, we will write $X_\gamma(f)\coloneqq f(\gamma)$ for $\gamma\in\Gamma_\tau$ and $f\in\mathbb{T}_\tau$, and write $X_\gamma'(f)\coloneqq f(\gamma)$ for $\gamma\in\Gamma_{\tau'}$ and $f\in\mathbb{T}_{\tau'}$. We will use the same notation for a tagged arc in $\tau$ and the corresponding tagged arc in the flipped triangulation $\tau'$.
\[prop:clustertransformation\] The rational map $\mu_k$ can be written as a composition $\mu_k=\iota_k\circ\kappa_k$ where
1. $\iota_k$ is the isomorphism $\mathbb{T}_\tau\rightarrow\mathbb{T}_{\tau'}$ given by $$\iota_k^*(X_{\gamma_j}')=
\begin{cases}
X_{\gamma_k}^{-1} & \text{if $j=k$} \\
X_{\gamma_j}X_{\gamma_k}^{[\langle\gamma_k,\gamma_j\rangle]_+} & \text{if $j\neq k$}
\end{cases}$$ where $[n]_+\coloneqq\max(n,0)$.
2. $\kappa_k$ is the birational automorphism $\mathbb{T}_\tau\dashrightarrow\mathbb{T}_\tau$ given by $$\kappa_k^*(X_\gamma)=X_\gamma\cdot(1+X_{\gamma_k})^{\langle\gamma,\gamma_k\rangle}.$$
If $(\mathbb{S},\mathbb{M})$ is not a closed surface with exactly one puncture, then any two tagged triangulations of $(\mathbb{S},\mathbb{M})$ are related by a sequence of flips ([@FST], Proposition 7.10). Thus Proposition \[prop:clustertransformation\] can be used to calculate the transition map $X_{\tau'}\circ X_\tau^{-1}$ for any tagged triangulations $\tau$ and $\tau'$ in this case.
The main result {#sec:TheMainResult}
---------------
Consider a complete, generic GMN differential $\phi$. We further assume that the associated marked bordered surface $(\mathbb{S},\mathbb{M})$ is not a closed surface with exactly one puncture. If $\Delta\subset\mathbb{C}^*$ is a convex sector whose boundary rays are non-active with phases $\theta_1$ and $\theta_2$, then each rotated differential $\phi_i=e^{-2i\theta_i}\cdot\phi$ is complete and saddle-free and hence determines a tagged WKB triangulation $\tau_i$. As we have seen, the Fock-Goncharov coordinates provide a birational map $$\label{eqn:twocharts}
X_{\tau_i}:\mathscr{X}(\mathbb{S},\mathbb{M})\dashrightarrow\operatorname{\operatorname{Hom}}_{\mathbb{Z}}(\Gamma_{\tau_i},\mathbb{C}^*)$$ where $\Gamma_{\tau_i}\cong\mathbb{Z}^n$ is the lattice spanned by the set of tagged arcs of $\tau_i$. By Lemma 10.3 of [@BridgelandSmith], the lattice $\Gamma_{\tau_i}$ is canonically isomorphic to $\Gamma_{\phi_i}$. There is a family of Riemann surfaces over $\mathbb{R}$ where the fiber over $\theta\in\mathbb{R}$ is the canonical double cover for the rotated differential $\phi_\theta=e^{-2i\theta}\cdot\phi$. The homology groups of these Riemann surfaces form a local system of lattices over $\mathbb{R}$ with flat connection given by the Gauss-Manin connection. Using this flat connection, we can identify the lattices $\Gamma_{\phi_i}$ with $\Gamma_\phi$. We can therefore think of the maps as taking values in the torus $\mathbb{T}_+=\operatorname{\operatorname{Hom}}_{\mathbb{Z}}(\Gamma_\phi,\mathbb{C}^*)$. The following is the main result of this paper.
\[thm:main\] Take notation as in the last paragraph. Then
1. There is a distinguished basepoint $\xi\in\mathbb{T}_-$ such that $\xi(\gamma)=-1$ if $\gamma\in\Gamma_\phi$ is the class of a non-closed saddle connection and $\xi(\gamma)=+1$ if $\gamma$ is the class of a closed saddle connection.
2. $\mathbf{S}_\phi(\Delta)$ extends to a birational automorphism of $\mathbb{T}_-$. If we use the basepoint $\xi$ to identify $\mathbb{T}_-$ with $\mathbb{T}_+$, then this is the birational automorphism of $\mathbb{T}_+$ relating the maps $$\label{eqn:sametarget}
X_{\tau_i}:\mathscr{X}(\mathbb{S},\mathbb{M})\dashrightarrow\mathbb{T}_+.$$
Theorem \[thm:main\] gives a way of computing the BPS automorphism $\mathbf{S}_\phi(\Delta)$ for a general sector $\Delta\subset\mathbb{C}^*$ and implies Theorem \[thm:introWCF\] from the introduction. Note that we have already proved part (1) in Section \[sec:SpinStructuresAndTheTwistedTorus\]. We will devote the remainder of the present section to the proof of part (2).
Stratification of moduli spaces
-------------------------------
We begin by describing some topological aspects of the moduli space of GMN differentials. Given a complete GMN differential $\phi\in\mathscr{Q}^\pm(\mathbb{S},\mathbb{M})$, let us write $r_\phi$ for the number of recurrent trajectories which approach a zero at one end and $s_\phi$ for the number of horizontal saddle connections. Following Section 5.2 of [@BridgelandSmith], we consider for every integer $p\geq0$ the subset $$B_p=\{\phi\in\mathscr{Q}^\pm(\mathbb{S},\mathbb{M}):\text{$\phi$ is complete and $r_\phi+2s_\phi\leq p$}\}.$$ Thus $B_0=B_1$ is the set of complete saddle-free differentials, and $B_2$ is the set of complete differentials having at most one horizontal saddle connection. As in [@BridgelandSmith], we define $F_0=B_0$ and $F_p=B_p\setminus B_{p-1}$ for $p\geq1$. Each of these sets $F_p$ is locally closed in $\mathscr{Q}^\pm(\mathbb{S},\mathbb{M})$ and the union $\coprod_{p=0}^\infty F_p\subset\mathscr{Q}^\pm(\mathbb{S},\mathbb{M})$ is the set of all complete GMN differentials. Thus, the $F_p$ provide a stratification of this set.
If $p\geq2$, then by Proposition 5.5 of [@BridgelandSmith], the set $F_p$ has codimension one in $B_p$, and hence one can think of $F_p$ as a wall in $B_p$, potentially separating two connected components of $B_{p-1}$. A crucial feature of the stratification by the sets $F_p$ is the “walls have ends” property proved in [@BridgelandSmith], Section 5.6:
\[prop:wallshaveends\] Assume $(\mathbb{S},\mathbb{M})$ is not a closed surface with exactly one puncture, and take $p>2$. Let $C$ be any connected component of $F_p\subset\mathscr{Q}^\pm(\mathbb{S},\mathbb{M})$. Then there is a point $\phi$ in the closure of $C$ such that for any neighborhood $\phi\in U\subset B_p$, we can find a smaller neighborhood $\phi\in V\subset U$ such that $V\cap B_{p-1}$ is connected.
Completing the proof
--------------------
In what follows, we will assume $(\mathbb{S},\mathbb{M})$ is not a closed surface with exactly one puncture, and we take $\phi\in\mathscr{Q}^\pm(\mathbb{S},\mathbb{M})$. We will say that the differential $\phi$ is *good* if there exists $\varepsilon>0$ such that if $-\varepsilon<\theta_1<0<\theta_2<\varepsilon$ and the rotated differentials $e^{-2i\theta_j}\cdot\phi$ are saddle-free, then the conclusion of part (2) of Theorem \[thm:main\] holds for the sector $\Delta\subset\mathbb{C}^*$ having boundary rays $\mathbb{R}_{>0}\cdot e^{i\theta_j}$.
\[lem:F0\] If $\phi$ is generic differential in $F_0$ then $\phi$ is good.
If $\phi\in F_0$ then $\phi$ is a saddle-free differential and we can find $\varepsilon>0$ such that if $-\varepsilon<\theta_1<0<\theta_2<\varepsilon$ then the rotated differentials $e^{-2i\theta_j}\cdot\phi$ are saddle-free and have the same associated tagged triangulation as $\phi$. Therefore the transformation relating the maps is the identity. If $\Delta\subset\mathbb{C}^*$ is the sector having boundary rays $\mathbb{R}_{>0}\cdot e^{i\theta_j}$, then $\Delta$ contains no active rays, so for generic $\phi$, the BPS automorphism $\mathbf{S}_\phi(\Delta)$ is defined and equals the identity. Hence $\phi$ is good.
\[lem:F1\] If $\phi$ is a generic differential in $F_2$ then $\phi$ is good.
If $\phi\in F_2$ then $\phi$ has a unique horizontal saddle connection. Let us assume for the time being that this is not a closed saddle connection. By Proposition 5.5 of [@BridgelandSmith], there exists $\varepsilon>0$ such that if $-\varepsilon<\theta_1<0<\theta_2<\varepsilon$ then the rotated differentials $e^{-2i\theta_j}\cdot\phi$ are saddle-free and the associated tagged triangulations are related by a flip. Let $\Delta\subset\mathbb{C}^*$ be the sector having boundary rays $\mathbb{R}_{>0}\cdot e^{i\theta_j}$. If $\phi$ is generic, then by Proposition \[prop:KSautomorphism\] the associated BPS automorphism is given by $$\mathbf{S}_\phi(\Delta)^*(x_\beta)=x_\beta\cdot(1-x_\gamma)^{\langle\beta,\gamma\rangle}$$ where $\gamma$ is the class of the unique horizontal saddle connection of $\phi$ and we have used the fact that $\Omega_\phi(\gamma)=1$. Let us write $\Gamma_{\tau_j}$ as in Section \[sec:TheMainResult\] for the lattice spanned by tagged arcs of the tagged triangulation $\tau_j$ determined by $e^{-2i\theta_j}\cdot\phi$. The Gauss-Manin connection gives an isomorphism $\Gamma_{\tau_1}\stackrel{\sim}{\rightarrow}\Gamma_{\tau_2}$, and by Proposition 10.4 of [@BridgelandSmith], the induced isomorphism $\mathbb{T}_{\tau_1}\rightarrow\mathbb{T}_{\tau_2}$ is precisely the isomorphism $\iota_k$ of Proposition \[prop:clustertransformation\] where $k$ is the edge being flipped. The transformation relating the maps is therefore the map $\kappa_k$ of Proposition \[prop:clustertransformation\]. If we identify $\mathbb{T}_-$ with $\mathbb{T}_+$ using the canonical basepoint $\xi$, then this agrees with $\mathbf{S}_\phi(\Delta)$ because $\xi(\gamma)=-1$.
If the unique horizontal saddle connection of $\phi\in F_2$ is closed, then the rotated differentials $e^{-2i\theta_j}\cdot\phi$ determine the same tagged triangulation, and therefore the transformation relating the maps is the identity. Under our assumptions, the class $\gamma$ of the saddle connection satisfies $\Omega_\phi(\gamma)=0$ so that $\mathbf{S}_\phi(\Delta)$ is the identity as well. Hence $\phi$ is good.
\[lem:Fp\] Every generic complete differential is good.
Since the union $\coprod_{p=0}^\infty F_p\subset\mathscr{Q}^\pm(\mathbb{S},\mathbb{M})$ is the set of all complete GMN differentials, we must show, for any $p\geq0$, that a generic differential in $F_p$ is good. The case $p=1$ is vacuous because $F_1=\emptyset$. The cases $p=0$ and $p=2$ were handled in Lemmas \[lem:F0\] and \[lem:F1\], respectively.
Let $p>2$ and assume inductively that all generic differentials in $F_{p-1}$ are good. Let $C$ denote a connected component of $F_p$. By Proposition 5.5 of [@BridgelandSmith], we can find, for any $\phi\in C$, an open neighborhood $\phi\in U_\phi\subset C$ and a constant $\varepsilon_\phi>0$ such that $e^{-2i\theta}\cdot q\in B_{p-1}$ when $0<|\theta|<\varepsilon_\phi$ and $q\in U_\phi$. By shrinking $U_\phi$ if necessary, we can assume further that there exist $\theta_j$ with $-\varepsilon_\phi<\theta_1<0<\theta_2<\varepsilon_\phi$ such that the differentials $e^{-2i\theta_j}\cdot q$ are saddle-free for all $q\in U_\phi$. If $q$, $q'\in U_\phi$ are generic differentials in this neighborhood, then it follows from Theorem \[thm:firstWCF\] that $q$ is good if and only if $q'$ is good. Since the sets $U_\phi$ form an open cover of $C$ and the generic differentials are dense in $F_p$, it will follow that all generic differentials in $C$ are good if we can prove that one such differential is good.
Let $\phi$ be a point in the closure of $C$ as in Proposition \[prop:wallshaveends\]. By Proposition 5.5 of [@BridgelandSmith], there exists a neighborhood $\phi\in U\subset B_p$ and a constant $\varepsilon>0$ such that $e^{-2i\theta}\cdot q\in B_{p-1}$ when $0<|\theta|<\varepsilon$ and $q\in U\cap F_{p}$. By shrinking this $U$ if necessary, we can assume there exist $\theta_j$ with $-\varepsilon<\theta_1<0<\theta_2<\varepsilon$ such that the differentials $e^{-2i\theta_j}\cdot q$ are saddle-free for all $q\in U$. Then by Proposition \[prop:wallshaveends\], there is a smaller neighborhood $\phi\in V\subset U$ for which $V\cap B_{p-1}$ is connected. Applying Theorem \[thm:firstWCF\] to paths in $V$, we see that any generic differential in $C\cap V$ is good. This completes the proof.
Now suppose $\phi\in\mathscr{Q}^\pm(\mathbb{S},\mathbb{M})$ is any complete and generic differential. Note that we have $$\mathbf{S}_\phi(\Delta_1\cup\Delta_2)=\mathbf{S}_\phi(\Delta_1)\circ\mathbf{S}_\phi(\Delta_2)$$ whenever $\Delta_1$, $\Delta_2\subset\mathbb{C}^*$ are adjacent sectors in the clockwise order whose boundary rays are non-active and whose union is a convex sector. We can use this fact to prove Theorem \[thm:main\] for a general convex sector $\Delta\subset\mathbb{C}^*$. Indeed, and let $t\in[\theta_1,\theta_2]$ where $\theta_1$ and $\theta_2$ are the phases of the boundary rays of $\Delta$. Then the rotated differential $e^{-2it}\cdot\phi$ is good by Lemma \[lem:Fp\]. Let $\varepsilon_t>0$ be the constant associated to $e^{-2it}\cdot\phi$ in the definition of a good differential. Then the open intervals $(t-\varepsilon_t,t+\varepsilon_t)$ form a cover of $[\theta_1,\theta_2]$. As this interval is compact, we can find finitely many $t_1,\dots,t_k\in[\theta_1,\theta_2]$ in increasing order such that the associated open intervals form a finite subcover. Choose $s_j\in[\theta_1,\theta_2]$ so that $$\theta_1=s_1<t_1<s_2<t_2<\dots<t_k<s_{k+1}=\theta_2$$ and so the differentials $e^{-2is_j}\cdot\phi$ are saddle-free and $t_j-s_j$, $s_{j+1}-t_j<\varepsilon_{t_j}$ for $j=1,\dots,k$. For each $j$, let $\Delta_j$ be the sector with boundary rays $\mathbb{R}_{>0}\cdot e^{is_j}$ and $\mathbb{R}_{>0}\cdot e^{is_{j+1}}$. Part (2) of Theorem \[thm:main\] now follows since $$\mathbf{S}_\phi(\Delta)=\mathbf{S}_\phi(\Delta_k)\circ\dots\circ\mathbf{S}_\phi(\Delta_1)$$ and each factor $\mathbf{S}_\phi(\Delta_j)$ on the right is identified with the birational transformation of $\mathbb{T}_+$ relating the Fock-Goncharov coordinates associated to $e^{-2is_j}\cdot\phi$ and $e^{-2is_{j+1}}\cdot\phi$.
The motivic wall-crossing formula {#sec:TheMotivicWallCrossingFormula}
=================================
In this appendix, we explain how the version of the wall-crossing formula considered in Section \[sec:TheWallCrossingFormulaForQuadraticDifferentials\] is derived from the usual version in motivic Donaldson-Thomas theory. Since we are mainly interested in the application to quadratic differentials, our discussion will be quite cursory. For other introductory accounts of this material, see the references [@KontsevichSoibelman2; @Reineke; @Keller; @Meinhardt; @Bridgeland18].
Quivers with potential
----------------------
Recall that a *quiver* is simply a directed graph. It consists of a finite set $Q_0$ (the set of *vertices*), a finite set $Q_1$ (the set of *arrows*), and maps $s:Q_1\rightarrow Q_0$ and $t:Q_1\rightarrow Q_0$ taking an arrow to its *source* and *target*, respectively. We often write $a:i\rightarrow j$ to mean that $a$ is an arrow with $s(a)=i$ and $t(a)=j$.
Let $Q=(Q_0,Q_1,s,t)$ be a quiver. Then a complex *representation* $M$ of $Q$ consists of a complex vector space $M_i$ for each vertex $i\in Q_0$ and a linear map $M_a:M_i\rightarrow M_j$ for each arrow $a:i\rightarrow j$ in $Q_1$. A representation is finite-dimensional if each of the vector spaces $M_i$ is finite-dimensional. In this case, we can define the *dimension vector* to be the vector $\underline{\dim}(M)=(\dim_\mathbb{C}M_i)_{i\in Q_0}$. The finite-dimensional representations of $Q$ form an abelian category where a morphism $M\rightarrow N$ of representations is a collection of linear maps $\eta_i:M_i\rightarrow N_i$ for $i\in Q_0$ satisfying the identity $\eta_{t(a)}\circ M_a=N_a\circ\eta_{s(a)}$ for every $a\in Q_1$.
Representations of a quiver can be viewed alternatively as modules over a certain noncommutative algebra spanned by paths in the quiver. By a *path* in $Q$, we mean a sequence of arrows $a_0,\dots,a_k$ such that $t(a_i)=s(a_{i-1})$ for $i=1,\dots,k$. Let us denote this path by $p=a_0\dots a_k$. We define its *source* by $s(p)=s(a_k)$ and its *target* by $t(p)=t(a_0)$. A path $p$ in $Q$ is *cyclic* if its source and target coincide. Two paths $p$ and $q$ are *composable* if $s(p)=t(q)$, and in this case their composition $pq$ is defined by juxtaposition. To any quiver $Q$, we associate a $\mathbb{C}$-algebra $\mathbb{C}Q$ called the *path algebra* of $Q$. It is spanned as a vector space by the set of all paths in $Q$. The product of two paths is defined to be their composition if the paths are composable and zero otherwise. Extending this operation linearly gives the multiplication on $\mathbb{C}Q$. The following fact is well known.
\[lem:equivalence\] The category of finite-dimensional representations of a quiver $Q$ is equivalent to the category of finite-dimensional left modules over the path algebra $\mathbb{C}Q$. Under this equivalence, a representation $M$ of $Q$ maps to the obvious module over $\mathbb{C}Q$ whose underlying vector space is the direct sum $\bigoplus_{i\in Q_0}M_i$.
It is often important to consider representations of a quiver $Q$ where the linear maps are required to satisfy certain relations. For us, these relations will always come from a *potential*, which is defined as a $\mathbb{C}$-linear combination of cyclic paths in $Q$. If $p=a_1\dots a_k$ is a cyclic path in $Q$ and $a\in Q_1$ is an arrow, then we can take the *cyclic derivative* $$\partial_a(p)=\sum_{i:a_i=a}a_{i+1}\dots a_ka_1\dots a_{i-1}\in\mathbb{C}Q.$$ Extending this operation linearly, we can define $\partial_a(W)$ for any potential $W$. Then the *Jacobian ideal* is the ideal $\mathfrak{a}\subset\mathbb{C}Q$ generated by all cyclic derivatives of $W$, and the *Jacobian algebra* is the quotient $J(Q,W)=\mathbb{C}Q/\mathfrak{a}$.
To relate quiver representations and quadratic differentials, we should consider a slightly modified setup. Namely, we should take the completion $\widehat{\mathbb{C}Q}$ of the path algebra $\mathbb{C}Q$ with respect to the ideal generated by the arrows. Then the *complete Jacobian algebra* is defined as the quotient of $\widehat{\mathbb{C}Q}$ by the closure of the Jacobian ideal. As shown in Section 10 of [@DWZ], a finite-dimensional module over the complete Jacobian algebra is the same thing as a nilpotent module over the usual Jacobian algebra, that is, a module annihilated by sufficiently long paths in $Q$.
Stability conditions
--------------------
The concept of a stability condition on a triangulated category was introduced by Bridgeland in [@Bridgeland07]. In this appendix, we will be concerned with a version of this concept for abelian categories of quiver representations. Let $\mathcal{A}=\mathcal{A}(Q,W)$ denote the category of nilpotent modules over the Jacobian algebra of a quiver with potential $(Q,W)$. Then a *stability condition* on $\mathcal{A}$ is a tuple $\zeta\in\mathbb{H}_+^{Q_0}$ of complex numbers in the domain $$\mathbb{H}_+=\{z=re^{i\pi\theta}:\text{$r>0$ and $0<\theta\leq1$}\}\subset\mathbb{C}.$$
Let $\Gamma=\mathbb{Z}^{Q_0}$ be the lattice spanned by vertices of the quiver $Q$. Given a stability condition $\zeta$ on $\mathcal{A}$, we define a group homomorphism $Z_\zeta:\Gamma\rightarrow\mathbb{C}$ called the *central charge* by the rule $$Z_\zeta(\gamma)=\sum_{i\in Q_0}\gamma_i\cdot\zeta_i\in\mathbb{C}.$$ We can then define the *slope* of a nonzero vector $\gamma\in\mathbb{N}^{Q_0}$ to be the real number $\frac{1}{\pi}\cdot\arg Z_\zeta(\gamma)\in(0,1]$. The slope $\mu(M)$ of an object $M\in\mathcal{A}$ is defined to be the slope of its dimension vector. A nonzero object $M\in\mathcal{A}$ is said to be *semistable* if we have $$\mu(N)\leq\mu(M)$$ for every proper nonzero submodule $N\subset M$.
For convenience, we define, for any real number $\mu\in(0,1]$, the subset $$\Lambda_\mu^\zeta=\{\gamma\in\mathbb{N}^{Q_0}:\text{$\gamma$ has slope $\mu$}\}\cup\{0\}\subset\Gamma.$$ We denote by $\langle-,-\rangle$ the skew-symmetrized *Euler pairing* on the lattice $\Gamma$ given by $\langle\alpha,\beta\rangle=(\alpha,\beta)-(\beta,\alpha)$ where $$(\alpha,\beta)=\sum_i\alpha_i\beta_i-\sum_{a:i\rightarrow j}\alpha_i\beta_j.$$ We say that a stability condition $\zeta$ is *generic* if, for each $\mu$, one has $\langle \alpha,\beta\rangle=0$ for $\alpha$, $\beta\in\Lambda_\mu^\zeta$.
Motivic theories
----------------
We now describe a general framework in which we can formulate the Kontsevich-Soibelman wall-crossing formula. Following [@Meinhardt], we define a *motivic theory* to be a rule associating to a scheme $X$ an abelian group $R(X)$. This rule is required to be functorial in two ways: If $u:X\rightarrow Y$ is any morphism of schemes, there is a homomorphism $u^*:R(Y)\rightarrow R(X)$ called the *pullback* along $u$, and if $u$ is of finite type, there is a homomorphism $u_!:R(X)\rightarrow R(Y)$ called the *pushforward* along $u$. In addition, there is an associative, symmetric operation $$\boxtimes:R(X)\otimes_{\mathbb{Z}}R(Y)\rightarrow R(X\times Y)$$ called the *exterior product* with unit element $1\in R(\operatorname{\operatorname{Spec}}\mathbb{C})$, and operations $$\sigma^n:R(X)\rightarrow R(\operatorname{\operatorname{Sym}}^n(X))$$ for each $n\in\mathbb{N}$ where $\operatorname{\operatorname{Sym}}(X)=X^n\sslash S_n$ is the $n$th symmetric product. These data are required to satisfy several axioms listed in Section 4 of [@Meinhardt].
A fundamental example of a motivic theory is the theory $R=\underline{\mathrm{K}}_0(\operatorname{\operatorname{Sch}})$, which associates to a connected scheme $X$ the group $R(X)=\mathrm{K}_0(\operatorname{\operatorname{Sch}}_X)$ generated by isomorphism classes $[V\rightarrow X]$ of schemes $V$ of finite type over $X$ subject to the relation $$[V\rightarrow X]=[Z\rightarrow X]+[V\setminus Z\rightarrow X]$$ for any closed subscheme $Z\subset V$. We extend this to nonconnected schemes $X$ by defining $R(X)=\prod_{X_i\in\pi_0(X)}K_0(\operatorname{\operatorname{Sch}}_{X_i})$. To get the structure of a motivic theory, we define the pullback by $$u^*\left([W\rightarrow Y]\right)=[X\times_Y W\rightarrow X]$$ for any morphism $u:X\rightarrow Y$ and define the pushforward by $$u_!\left([V\stackrel{v}{\rightarrow}X]\right)=[V\stackrel{u\circ v}{\longrightarrow}Y]$$ for any morphism $u:X\rightarrow Y$ of finite type. The exterior product is given by $$[V\stackrel{v}{\rightarrow}X]\boxtimes[W\stackrel{w}{\rightarrow}Y]=[V\times W\stackrel{v\times w}{\longrightarrow}X\times Y]$$ with unit element $1=[\operatorname{\operatorname{Spec}}\mathbb{C}\stackrel{\mathrm{id}}{\rightarrow}\operatorname{\operatorname{Spec}}\mathbb{C}]\in R(\operatorname{\operatorname{Spec}}\mathbb{C})$, and the $\sigma^n$ operations are given by $$\sigma^n\left([V\rightarrow X]\right)=[\operatorname{\operatorname{Sym}}^n(V)\rightarrow\operatorname{\operatorname{Sym}}^n(X)]$$ for each $n\in\mathbb{N}$.
Given any motivic theory $R$, we can define a product operation on $R(\operatorname{\operatorname{Spec}}\mathbb{C})$ by the rule $ab=+_!(a\boxtimes b)$ where $+:\operatorname{\operatorname{Spec}}\mathbb{C}\times\operatorname{\operatorname{Spec}}\mathbb{C}\rightarrow\operatorname{\operatorname{Spec}}\mathbb{C}$ is the obvious isomorphism. In this way, the abelian group $R(\operatorname{\operatorname{Spec}}\mathbb{C})$ becomes a ring with unit element 1. We can also associate to each scheme $X$, the element $[X]_R\coloneqq c_!c^*(1)\in R(\operatorname{\operatorname{Spec}}\mathbb{C})$ where $c:X\rightarrow\operatorname{\operatorname{Spec}}\mathbb{C}$ is the constant map. In particular, we define the element $\mathbb{L}_R\coloneqq[\mathbb{A}^1]_R$.
The identity $$[\mathrm{GL}(n)]_R=\prod_{i=0}^{n-1}(\mathbb{L}_R^n-\mathbb{L}_R^i)$$ holds in the ring $R(\operatorname{\operatorname{Spec}}\mathbb{C})$.
The notion of a motivic theory can be extended to the notion of a *stacky motivic theory*. This is an operation $\mathfrak{R}$ that assigns to every stack of the form $\mathfrak{X}=\coprod_iX_i/G_i$, where the $X_i$ are schemes and the $G_i$ are linear algebraic groups, an abelian group $\mathfrak{R}(\mathfrak{X})$. One has a pullback $u^*:\mathfrak{R}(\mathfrak{Y})\rightarrow\mathfrak{R}(\mathfrak{X})$ for any 1-morphism $u:\mathfrak{X}\rightarrow\mathfrak{Y}$, and a pushforward $u_!:\mathfrak{R}(\mathfrak{X})\rightarrow\mathfrak{R}(\mathfrak{Y})$ if $u$ is of finite type in the sense of [@Meinhardt]. One also has structures $\boxtimes$, $1\in\mathfrak{R}(\operatorname{\operatorname{Spec}}\mathbb{C})$, and $\sigma^n$ for $n\in\mathbb{N}$ satisfying various properties as in the definition of a motivic theory for schemes.
Let $R$ be any motivic theory satisfying the condition $\sigma^n(a\mathbb{L}_R)=\sigma^n(a)\mathbb{L}_R^n$ for any $a\in R(X)$ with $X$ any scheme and $n\in\mathbb{N}$. As explained in [@Meinhardt], Section 4.3, there is a functorial construction which associates to $R$ a stacky motivic theory $R^\mathrm{st}$ whose value on a connected scheme $X$ is the group $$R^\mathrm{st}(X)\coloneqq R(X)\left[[GL(n)]_R^{-1}:n\in\mathbb{N}\right]$$ defined using the $R(\operatorname{\operatorname{Spec}}\mathbb{C})$-module structure on $R(X)$ induced by the exterior product.
Motivic DT invariants
---------------------
Let us fix a motivic theory $R$ satisfying $\sigma^n(a\mathbb{L}_R)=\sigma^n(a)\mathbb{L}_R^n$ for any $a\in R(X)$ with $X$ any scheme and $n\in\mathbb{N}$. As in [@Meinhardt], Example 6.1, we can apply the results of [@DavisonMeinhardt1], Appendix B, to extend the $\sigma^n$ operations to $R(X)[\mathbb{L}_R^{1/2}]$ to get a new motivic theory and then pass to the associated stacky motivic theory $\mathfrak{R}=R[\mathbb{L}_R^{1/2}]^\mathrm{st}$. The constructions described below will depend on a choice of stacky vanishing cycle with values in $\mathfrak{R}$, which can be chosen canonically (see [@Meinhardt], Section 5 for the definition and construction of vanishing cycles).
Suppose we are given a stability condition $\zeta$ on $\mathcal{A}=\mathcal{A}(Q,W)$ and a real number $\mu\in(0,1]$. Let us write $\mathcal{S}(\mathcal{A})$ for the abelian group $\mathfrak{R}(\mathbb{N}^{Q_0})\cong\prod_{\gamma\in\mathbb{N}^{Q_0}}\mathfrak{R}(\operatorname{\operatorname{Spec}}\mathbb{C})\cong\mathfrak{R}(\operatorname{\operatorname{Spec}}\mathbb{C})\llbracket x_\gamma:\gamma\in\mathbb{N}^{Q_0}\rrbracket$. Then we can construct an element $$\mathcal{DT}(\mathcal{A})_\mu^\zeta\in\mathcal{S}(\mathcal{A})$$ called the *Donaldson-Thomas function*. The precise definition of this object is rather technical, and therefore we will only sketch the construction here, referring to [@Meinhardt] for further details. To begin, we equip the abelian group $\mathcal{S}(\mathcal{A})\cong\mathfrak{R}(\operatorname{\operatorname{Spec}}\mathbb{C})\llbracket x_\gamma:\gamma\in\mathbb{N}^{Q_0}\rrbracket$ with an auxiliary product $*$. It is the unique $\mathfrak{R}(\operatorname{\operatorname{Spec}}\mathbb{C})$-bilinear continuous product defined on generators by $$x_\alpha*x_\beta=\mathbb{L}_R^{\langle\alpha,\beta\rangle/2}\cdot x_{\alpha+\beta}.$$ Associated to the category $\mathcal{A}$, there is an associative algebra $\mathcal{H}(\mathcal{A})$ known as the *Ringel-Hall algebra*, and there is an algebra homomorphism $$\mathcal{I}:\mathcal{H}(\mathcal{A})\rightarrow\mathcal{S}(\mathcal{A})$$ known as the *integration map*. (To be somewhat more precise, the map $\mathcal{I}$ that we consider here is the composition of the integration map defined in [@Meinhardt] with the pushforward along the map $\underline{\dim}$ sending a $\mathbb{C}Q$-module to its dimension vector.) As explained in Section 6.3 of [@Meinhardt], there is an element $\delta_\mu^\zeta\in\mathcal{H}(\mathcal{A})$ naturally associated to each slope $\mu\in(0,1]$. We define $\mathcal{DT}(\mathcal{A})_\mu^\zeta$ to be the unique element of $\mathcal{S}(\mathcal{A})\cong\mathfrak{R}(\operatorname{\operatorname{Spec}}\mathbb{C})\llbracket x_\gamma:\gamma\in\mathbb{N}^{Q_0}\rrbracket$ with constant term zero such that $$\mathcal{I}(\delta_\mu^\zeta)=\operatorname{\operatorname{Sym}}\left(\mathcal{DT}(\mathcal{A})_\mu^\zeta\right)$$ where $\operatorname{\operatorname{Sym}}(a)\coloneqq\sum_{n\in\mathbb{N}}\operatorname{\operatorname{Sym}}^n(a)$. It follows from Lemma 6.3 of [@Meinhardt] that such an element $\mathcal{DT}(\mathcal{A})_\mu^\zeta$ is well defined.
The wall-crossing formula
-------------------------
We can now formulate a version of the wall-crossing formula for motivic DT invariants.
\[thm:motivicWCF\] Let $D\subset(0,1]$ be an interval. Then the product $$\prod_{\mu\in D}\mathcal{I}(\delta_\mu^\zeta)\in\mathcal{S}(\mathcal{A})$$ is constant as the stability condition $\zeta$ varies, provided there is no semistable object in $\mathcal{A}$ whose slope enters or leaves the interval $D$. Here the product is taken with respect to the multiplication $*$ in order of decreasing slopes.
The infinite product appearing in Theorem \[thm:motivicWCF\] is well defined because for every $\gamma\in\mathbb{N}^{Q_0}$, only finitely many factors contribute to its $\gamma$-component in $\mathcal{S}(\mathcal{A})\cong\prod_{\gamma\in\mathbb{N}^{Q_0}}\mathfrak{R}(\operatorname{\operatorname{Spec}}\mathbb{C})$.
Some remarks are in order as the statement of Theorem \[thm:motivicWCF\] differs slightly from the wall-crossing formula proved in [@DavisonMeinhardt1] and reviewed in [@Meinhardt]. First of all, these references consider arbitrary modules over the Jacobian algebra whereas we consider only nilpotent modules. In fact, the proof of the wall-crossing formula for nilpotent modules is essentially the same; one simply replaces all moduli stacks of quiver representations by the corresponding stacks of nilpotent representations.
Another difference between Theorem \[thm:motivicWCF\] and the wall-crossing formulas formulated in [@DavisonMeinhardt1] and [@Meinhardt] is that we take the product over slopes in an interval $D\subset(0,1]$ whereas [@DavisonMeinhardt1] and [@Meinhardt] take the product over all slopes. Again, the idea of the proof is the same. In the proof of Theorem \[thm:motivicWCF\], one considers objects of $\mathcal{A}$ whose Harder-Narasimhan factors have slopes contained in the interval $D$. (See [@Meinhardt] for details on the Harder-Narasimhan filtration and the proof of the wall-crossing formula.)
Quasi-classical limit
---------------------
As explained in [@Meinhardt], Example 5.7, there is a motivic theory $R$ that associates to a variety $X$ the Grothendieck group of the category $D^b(\mathrm{MMHM}(X))$ of “monodromic mixed Hodge modules” on $X$. We refer to [@DavisonMeinhardt2] for a more detailed description of this category. We can upgrade this $R$ to a stacky motivic theory $\mathfrak{R}$ as above and view the wall-crossing formula as an identity in the group $\mathcal{S}(\mathcal{A})=\mathfrak{R}(\mathbb{N}^{Q_0})$.
Let us write $\mathcal{T}(\mathcal{A})=\mathbb{Z}(\!(q^{1/2})\!)\llbracket x_i:i\in Q_0\rrbracket$. In Section 1.3 of [@DavisonMeinhardt2], the authors associate to each object $\mathcal{L}\in D^b(\mathrm{MMHM}(\mathbb{N}^{Q_0}))$, a formal power series $$\mathcal{Z}(\mathcal{L})=\sum_{\gamma\in\mathbb{N}^{Q_0}}\chi(\mathcal{L}_\gamma,q^{1/2})\cdot x_\gamma \in \mathcal{T}(\mathcal{A})$$ where for each $\gamma\in\mathbb{N}^{Q_0}$ we have $\chi(\mathcal{L}_\gamma,q^{1/2})\in\mathbb{Z}(\!(q^{1/2})\!)$ and $x_\gamma\coloneqq\prod_ix_i^{\gamma_i}$. This operation $\mathcal{Z}$ provides a map from a subgroup of $\mathcal{S}(\mathcal{A})$ to $\mathcal{T}(\mathcal{A})$. In the terminology of [@DavisonMeinhardt2], the category $\mathcal{A}=\mathcal{A}(Q,W)$ of nilpotent modules over the Jacobian algebra of the quiver with potential $(Q,W)$ is a Serre subcategory of the category of all finite-dimensional modules over the Jacobian algebra. The integrality theorem proved in [@DavisonMeinhardt2] therefore implies that for a generic stability condition $\zeta$ on $\mathcal{A}$, one has an identity $$\mathcal{Z}(\mathcal{I}(\delta_\mu^\zeta))=\operatorname{EXP}\left(\sum_{0\neq\gamma\in\Lambda_\mu^\zeta} \frac{\Omega_\gamma^\zeta(q^{-1/2})}{q^{1/2}-q^{-1/2}}\cdot x_\gamma\right).$$ Here $\Omega_\gamma^\zeta(q^{1/2})$ is a Laurent polynomial in $q^{1/2}$ called the *refined BPS invariant* for $\gamma$ and $\zeta$, and $\operatorname{EXP}$ denotes the *plethystic exponential*. The latter is a homomorphism from the additive group $\mathfrak{m}\subset\mathbb{Z}(q^{1/2})\llbracket x_i:i\in Q_0\rrbracket$ of series with constant term zero to the multiplicative group $1+\mathfrak{m}$ defined by the rule $$\operatorname{EXP}\left(f(q^{1/2}\right)\cdot x_\gamma)=\exp\left(\sum_{n=1}^\infty\frac{1}{n}f(q^{n/2})\cdot x_{n\gamma}\right)$$ for any $f(q^{1/2})\in\mathbb{Z}(q^{1/2})$ and $\gamma\in\mathbb{N}^{Q_0}$.
Just as we defined an auxiliary product $*$ on $\mathcal{S}(\mathcal{A})$, we can define an auxiliary product, also denoted $*$, on the ring $\mathcal{T}(\mathcal{A})$. It is the unique $\mathbb{Z}(\!(q^{1/2})\!)$-bilinear continuous product defined on basis elements by $$x_\alpha*x_\beta=(-q^{1/2})^{\langle\alpha,\beta\rangle}\cdot x_{\alpha+\beta}.$$ Then map $\mathcal{Z}$ preserves the $*$ products. Note that by skew-symmetry of the pairing $\langle-,-\rangle$, one has $x_\alpha*x_\beta=q^{\langle\alpha,\beta\rangle}\cdot x_\beta*x_\alpha$. There is an automorphism $\mathbf{S}_\mu^\zeta$ of the algebra $\mathcal{T}(\mathcal{A})$ with respect to the product $*$ given by conjugation with $\mathcal{Z}(\mathcal{I}(\delta_\mu^\zeta))$: $$\mathbf{S}_\mu^\zeta(a)=\mathcal{Z}(\mathcal{I}(\delta_\mu^\zeta))*a*\mathcal{Z}(\mathcal{I}(\delta_\mu^\zeta))^{-1}.$$ For a generic stability condition $\zeta$, it follows from the definition of the plethystic exponential and the commutation relation for $*$ that this automorphism acts on basis elements by $$\mathbf{S}_\mu^\zeta(x_\beta)=x_\beta*\operatorname{EXP}\left(-\sum_{0\neq\gamma\in\Lambda_\mu^\zeta}q^{-1/2}[\langle\beta,\gamma\rangle]_{q^{-1}}\,\Omega_\gamma^\zeta(q^{-1/2})\cdot x_\gamma\right)$$ where we have introduced the *quantum integer* $[n]_t\coloneqq\frac{t^n-1}{t-1}=1+t+\dots+t^{n-1}$. Using the definition, one can compute the plethystic exponential explicitly, and in the quasi-classical limit $q^{1/2}\rightarrow1$, one finds $$\mathbf{S}_\mu^\zeta(x_\beta)=x_\beta*\prod_{0\neq\gamma\in\Lambda_\mu^\zeta}(1-x_\gamma)^{\Omega_\zeta(\gamma)\cdot\langle\beta,\gamma\rangle}$$ where we have written $\Omega_\zeta(\gamma)=\Omega_\gamma^\zeta(1)$. Taking this quasi-classical limit in Theorem \[thm:motivicWCF\], we obtain the following version of the wall-crossing formula.
\[thm:numericalWCF\] Let $D\subset(0,1]$ be an interval. Suppose $\zeta_t$, $t\in[0,1]$, is a path in the space of stability conditions with generic endpoints and there is no semistable object with slope in the boundary of $D$ for any $\zeta_t$. Then we have $$\prod_{\mu\in D}\mathbf{S}_\mu^{\zeta_0}=\prod_{\mu\in D}\mathbf{S}_\mu^{\zeta_1}$$ in $\operatorname{\operatorname{Aut}}\mathbb{Z}\llbracket x_i:i\in Q_0\rrbracket$. Here the products are taken in order of decreasing slopes.
Link with quadratic differentials
---------------------------------
As explained in [@BridgelandSmith], a tagged triangulation $\tau$ of a marked bordered surface $(\mathbb{S},\mathbb{M})$ determines an associated quiver with potential $(Q(\tau),W(\tau))$, and the domain $\mathcal{C}_\tau\subset\mathscr{Q}^\pm(\mathbb{S},\mathbb{M})$ in the moduli space of quadratic differentials embeds naturally into the space of stability conditions on $\mathcal{A}=\mathcal{A}(Q(\tau),W(\tau))$. If $\phi\in\mathcal{C}_\tau$ is a quadratic differential and $\zeta$ is the corresponding stability condition on $\mathcal{A}$, then there is an isomorphism $\Gamma_\phi\cong\mathbb{Z}^{Q(\tau)_0}$, and the period map $Z_\phi$ coincides with the central charge $Z_\zeta$ under this isomorphism. In particular, if $\phi$ is generic then so is $\zeta$. By Theorem 1.4 of [@BridgelandSmith], the invariant $\Omega_\phi(\gamma)$ coincides with $\Omega_\zeta(\gamma)$ where we use the same notation for a class in $\gamma\in\Gamma_\phi$ and the corresponding element of $\mathbb{Z}^{Q(\tau)_0}$. If the boundary rays of a convex sector $\Delta\subset\mathbb{C}^*$ are non-active for the differential $\phi$ and $D\subset(0,1]$ is the set of phases of rays in $\Delta$, then one can show as in Lemma 11.4 of [@BridgelandSmith] that there is no semistable object for $\zeta$ with slope in the boundary of $D$. Hence Theorem \[thm:firstWCF\] follows from Proposition \[prop:KSautomorphism\] and Theorem \[thm:numericalWCF\].
[99]{}
Allegretti, D.G.L. (2019). Voros symbols as cluster coordinates. *Journal of Topology*, **12**(4), 1031–1068.
Allegretti, D.G.L. Stability conditions, cluster varieties, and Riemann-Hilbert problems from surfaces. `arXiv:1912.05938 [math.AG]`.
Aoki, T., Iwaki, K., and Takahashi, T. (2019). Exact WKB analysis of Schrödinger equations with a Stokes curve of loop type. *Funkcialaj Ekvacioj*, 62(1), 1–34.
Atiyah, M.F. (1971). Riemann surfaces and spin structures. *Annales Scientifiques de l’École Normale Supérieure*, **4**(1), 47–62.
Bridgeland, T. (2007). Stability conditions on triangulated categories. *Annals of Mathematics*, **166**(2), 317–345.
Bridgeland, T. (2018). Hall algebras and Donaldson-Thomas invariants. *Algebraic Geometry: Salt Lake City 2015*, 75–100.
Bridgeland, T. (2019). Riemann-Hilbert problems from Donaldson-Thomas theory. *Inventiones mathematicae*, **216**(1), 69–124.
Bridgeland, T. and Smith, I. (2015). Quadratic differentials as stability conditions. *Publications Mathématiques de l’Institut des Hautes Études Scientifiques*, **121**(1), 155–278.
Davison, B. and Meinhardt, S. (2015). Donaldson-Thomas theory for categories of homological dimension one with potential. `arXiv:1512.08898 [math.AG]`.
Davison, B. and Meinhardt, S. (2020). Cohomological Donaldson-Thomas theory of a quiver with potential and quantum enveloping algebras. *Inventiones mathematicae*, 1–95.
Delabaere, E., Dillinger, H., and Pham, F. (1993). Résurgence de Voros et périodes des courbes hyperelliptiques. *Annales de l’Institut Fourier*, **43**(1), 163–199.
Derksen, H., Weyman, J., and Zelevinsky, A. (2008). Quivers with potentials and their representations I: Mutations. *Selecta Mathematica*, **14**(1), 59–119.
Eskin, A. and Masur, H. (2001). Asymptotic formulas on flat surfaces. *Ergodic Theory and Dynamical Systems*, **21**(2), 443–478.
Eskin, A., Masur, H., and Zorich, A. (2003). Moduli spaces of Abelian differentials: the principal boundary, counting problems, and the Siegel-Veech constants. *Publications Mathématiques de l’Institut des Hautes Études Scientifiques*, **97**(1), 61–179.
Eskin, A., Mirzakhani, M., and Mohammadi, A. (2015). Isolation, equidistribution, and orbit closures for the $\mathrm{SL}(2,\mathbb{R})$ action on moduli space. *Annals of Mathematics*, 673–721.
Fock, V.V. and Goncharov, A.B. (2006). Moduli spaces of local systems and higher Teichmüller theory. *Publications Mathématiques de l’Institut des Hautes Études Scientifiques*, **103**(1), 1–211.
Fock, V.V. and Goncharov, A.B. (2009). Cluster ensembles, quantization and the dilogarithm. *Annales Scientifiques de l’École Normale Supérieure*, **42**(6), 865–930.
Fomin, S., Shapiro, M., and Thurston, D. (2008). Cluster algebras and triangulated surfaces. Part I: Cluster complexes. *Acta Mathematica*, **201**(1), 83–146.
Gaiotto, D., Moore, G.W., and Neitzke, A. (2013). Wall-crossing, Hitchin systems, and the WKB approximation. *Advances in Mathematics*, **234**(2013), 239–403.
Iwaki, K. and Nakanishi, T. (2014). Exact WKB analysis and cluster algebras. *Journal of Physics A: Mathematical and Theoretical*, **47**(47), 474009.
Johnson, D. (1980). Spin structures and quadratic forms on surfaces. *Journal of the London Mathematical Society*, **2**(2), 365–373.
Keller, B. (2011). On cluster theory and quantum dilogarithm identities. In *Representations of algebras and related topics*, 85–116. European Mathematical Society.
Kontsevich, M. and Soibelman, Y. (2008). Stability structures, motivic Donaldson-Thomas invariants, and cluster transformations. `arXiv:0811.2435 [math.AG]`.
Kontsevich, M. and Soibelman, Y. (2010). Motivic Donaldson-Thomas invariants: summary of results. *Mirror symmetry and tropical geometry*, **527**, 55–89.
Kontsevich, M. and Soibelman, Y. (2011). Cohomological Hall algebra, exponential Hodge structures and motivic Donaldson-Thomas invariants. *Communications in Number Theory and Physics*, **5**(2), 231–252.
Kontsevich, M. and Soibelman, Y. (2020). Analyticity and resurgence in wall-crossing formulas. `arXiv:2005.10651 [math.AG]`.
Kontsevich, M. and Zorich, A. (2003). Connected components of the moduli spaces of Abelian differentials with prescribed singularities. *Inventiones mathematicae*, **153**(3), 631–678.
Masur, H. (1988). Lower bounds for the number of saddle connections and closed trajectories of a quadratic differential. In Holomorphic Functions and Moduli I, 215–228. Springer.
Masur, H. (1990). The growth rate of trajectories of a quadratic differential. *Ergodic Theory and Dynamical Systems*, **10**(1), 151–176.
Meinhardt, S. (2017). An introduction to (motivic) Donaldson-Thomas theory. *Confluentes Mathematici*, **9**(2), 101–158.
Reineke, M. (2010). Poisson automorphisms and quiver moduli. *Journal of the Institute of Mathematics of Jussieu*, **9**(3), 653–667.
Strebel, K. (1984). *Quadratic differentials*. Springer-Verlag.
Veech, W.A. (1998). Siegel measures. *Annals of Mathematics*, **148**(3), 895–944.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We investigate the existence of ground states for functionals with nonhomogenous principal part. Roughly speaking, we show that the Nehari manifold method requires no homogeinity on the principal part of a functional. This result is motivated by some elliptic problems involving nonhomogeneous operators. As an application, we prove the existence of a ground state and infinitely many solutions for three classes of boundary value problems.'
address:
- 'Giovany M. Figueiredo Universidade Federal do Pará, Faculdade de Matemática, 66075-110, Belém-PA, Brazil'
- 'H. Ramos Quoirin Universidad de Santiago de Chile, Casilla 307, Correo 2, Santiago, Chile'
author:
- 'Giovany M. Figueiredo'
- Humberto Ramos Quoirin
title: Ground states of elliptic problems involving non homogeneous operators
---
[^1]
Introduction and main results
=============================
This article is concerned with a class of variational elliptic problems involving non-homogeneous operators. Our main goal is to provide a unified approach to obtain [*ground state*]{} solutions for these problems. This approach is based on the Nehari manifold method, which was introduced in [@N] and is by now a well-established and useful tool in finding solutions of problems with a variational structure, cf. [@Ad; @AC; @BB; @BWW; @BDH; @BZ; @BW; @CCN; @CZ; @Ta]. In an abstract setting, given a Banach space $X$ and a $\mathcal{C}^1$ functional $\Phi:X \rightarrow \Re$, a [*ground state*]{} of $\Phi$ is a solution $u_0$ of the problem $$\Phi'(u_0)=0, \quad \Phi(u_0)=\min\{ \Phi(u); \ u \text{ is a critical point of } \Phi \}.$$ When looking for such a solution, one may restrict $\Phi$ to the set $$\mathcal{N}=\{u \in X \setminus\{0\};\ \Phi'(u)u=0\},$$ which not only contains all nontrivial critical points of $\Phi$, but also turns out to have some useful properties. It is well-known that under some conditions on $\Phi$, $\mathcal{N}$ is a $\mathcal{C}^1$ submanifold of $X$ and critical points of the restriction of $\Phi$ to $\mathcal{N}$ are in fact critical points of $\Phi$. As an immediate consequence, one may obtain a [*ground state*]{} of $\Phi$ by minimizing $\Phi$ over $\mathcal{N}$. An overview of this procedure, as well as further developments and several applications of this method, are given in [@SW].
Throughout this article we assume that $\Omega \subset \Re^N$ ($N \geq 3$) is a bounded domain and $f:\Re \rightarrow
[0,\infty)$ is an odd $\mathcal{C}^{1}$ function, so that $F(t)=\displaystyle\int_{0}^{t}f(s)\ ds$, defined for $t \in \Re$, is an even function. A typical application of the Nehari manifold method provides the existence of a ground state for the prototype problem $$\label{p1}
-\Delta u=f(u), \quad u \in H_0^1(\Omega), \quad u \geq 0,$$ where $\Delta$ is the Laplace operator and $f$ is, in addition, subcritical and superlinear. More precisely, $f$ satisfies $$|f(s)|\leq C(1+|s|^{r-1}) \quad \forall s \in \Re$$ for some $C>0$ and $r \in (2,2^*)$, where $2^*=\frac{2N}{N-2}$, as well as $$\label{hf}
\lim_{s\to 0}\frac{f(s)}{s}=0, \quad \lim_{s \to \infty} \frac{f(s)}{s}=\infty, \quad \text{
and } \quad \frac{f(s)}{s} \text{ is increasing in } (0,\infty).$$ Then the functional $$\Phi(u):= \frac{1}{2}\|u\|^2- \int_\Omega F(u),$$ defined on $H_0^1(\Omega)$, has a non-negative and nontrivial [*ground state*]{} $u_0$, which is a classical solution of . This result has a natural extension to an abstract setting as follows:
Let $X$ be a uniformly convex Banach space and $\mathcal{S}$ be the unit sphere in $X$. Assume that $\| \cdot \|$ is a $\mathcal{C}^1$ functional on $X \setminus \{0\}$. Then the following result holds, cf. [@SW Theorem 13]:
[@SW]\[tsw\] Let $\Phi$ be such that $\Phi(0)=0$ and $\Phi=I_0-I$ where $I_0,I$ are $\mathcal{C}^1$ functionals on $X$ satisfying, for some $p>1$:
1. $I'(u)=o(\|u\|^{p-1})$ as $u \to 0$.
2. $s \mapsto \frac{I'(su)}{s^{p-1}}$ is strictly increasing in $(0,\infty)$ for every $u \neq 0$.
3. $\frac{I(su)}{s^p} \to \infty$ uniformly for $u$ on weakly compact subsets of $X \setminus \{0\}$ as $s \to \infty$.
4. $I'$ is completely continuous. i.e. if $u_n \rightharpoonup u$ in $X$ then $I'(u_n) \rightarrow I'(u)$ in $X'$.
5. $I_0$ is weakly lower semicontinuous, positively homogeneous of degree $p$, i.e. $I_0(su)=s^pI_0(u)$, and satisfies $$C_0\|u\|^p\leq I_0(u)\leq C_0^{-1}\|u\|^p$$ and $$\label{ip} \left(I_0'(v)-I_0'(w)\right)(v-w)\geq C_1\left( \|v\|^{p-1} - \|w\|^{p-1}\right)(\|v\|-\|w\|)$$ for some $C_0,C_1>0$ and every $u,v \in X$.
Then the equation $\Phi'(u)=0$ has a ground state solution. Moreover, if $\Phi$ is even then this equation has infinitely many pairs of solutions.
The above theorem is clearly motivated by problems involving the $p$-Laplace operator, namely, $$\label{p2}-\Delta_p u =f(u), \quad u \in W_0^{1,p}(\Omega),$$ where $\Delta_p u =div \left(|\nabla u|^{p-2} \nabla u\right)$ and $p>1$. In this case $X=W_0^{1,p}(\Omega)$ with $\|u\|=\left(\int_\Omega |\nabla u|^p\right)^{\frac{1}{p}}$ and $\Phi=I_0-I$, where $$I_0(u)=\frac{1}{p} \|u\|^p\quad
\text{and} \quad I(u)=\int_\Omega F(u).$$ Under conditions similar to , it can be shown that $I$ satisfies the assumptions of Theorem \[tsw\], so that has a ground state solution and infinitely many pairs of solutions.
The following geometrical properties of $\Phi$ are essential in the proof of Theorem \[tsw\]:
- For any $w \in X \setminus \{0\}$ the map $t \mapsto \Phi(tw)$, defined for $t>0$, has a unique critical point $t_w>0$ which satisfies $\dis \Phi(t_w w)=\max_{t>0} \Phi(tw)$.\
- $t_w$ is uniformly bounded away from zero for $w \in \mathcal{S}$, i.e. there exists $\delta>0$ such that $t_w\geq \delta$ for every $w \in \mathcal{S}$. Moreover, $t_w$ is bounded from above for $w$ in a compact subset of $\mathcal{S}$, i.e. given a compact set $\mathcal{W} \subset \mathcal{S}$ there exists $C_{\mathcal{W}}>0$ such that $t_w \leq C_{\mathcal{W}}$ for every $w \in \mathcal{W}$.
These properties are used to show that $\mathcal{S}$ is homeomorphic to $\mathcal{N}$ through the projection $w \mapsto t_w w$ and that one may carry out a critical point theory on $\mathcal{N}$, cf. [@SW Corollary 10]. We shall prove that (A2) and (A3) hold for a larger class of functionals, in particular, for $\Phi=I_0-I$, where $I_0$ is not positively homogeneous and $I(u)=\int_\Omega F(u)$. This situation is motivated by the following examples:\
1. $X=W_0^{1,p}(\Omega)$ and $$I_0(u)=\frac{1}{p}\int_{\Omega}A(|\nabla u|^p).$$ Here $A(s)=\int_0^s a(t) dt$ and $a:[0,\infty)\rightarrow
[0,\infty)$ is $\mathcal{C}^{1}$ in $(0,\infty)$ and satisfies $$k_0\left( 1+t^{\frac{q-p}{p}}\right) \leq a(t) \leq k_1\left( 1+t^{\frac{q-p}{p}}\right) \quad \forall t>0,$$ where $k_0,k_1$ are positive constants and $p\geq q>1$. The associated Euler-Lagrange equation is the quasilinear equation $$\label{quasi}
-div \left(a(|\nabla u|^p)|\nabla u|^{p-2}\nabla u\right) = f(u), \quad u \in W_0^{1,p}(\Omega).$$ This class of operators contains the $p$-Laplacian ($a(t) \equiv 1$), as well as the sum of the $p$-Laplacian and the $q$-Laplacian ($a(t)=1+t^{\frac{q-p}{p}}$). Problems involving this class of operators have been investigated, for instance, in [@BMV; @CI; @CD; @Fi2; @HP; @MP].\
2. $X=H_0^1(\Omega)$ and $$I_0(u)=\frac{1}{2}\hat{M}\left(\|u\|^2\right),$$ where $\|u\|=\left(\int_\Omega |\nabla u|^2\right)^{\frac{1}{2}}$, $\hat{M}(s)=\int_0^s M(t)dt$ and $M:[0,\infty)\rightarrow [0,\infty)$ is a continuous function. In this case, the corresponding Euler-Lagrange equation is the Kirchhoff type equation $$\label{kir}
-M\left(\int_{\Omega} |\nabla u|^2\right) \Delta u =f(u), \quad u \in H_0^1(\Omega),$$ which has been intensively investigated over the last years, specially for $M(t)=at+b$, with $a,b>0$, cf. [@ACF; @Ba; @Fi; @HZ; @Ma; @Na].
We shall prove that this equation has a ground state for a larger class of $M$, which includes, for instance, $$M(t) = m_{0} + \ln(1 + t)$$ or $$\displaystyle
M(t)=m_{0}+\displaystyle\sum_{i=1}^{k}b_{i}t^{\gamma_{i}}$$ where $b_{i}\geq 0$ and $\gamma_{i}\in (0,1]$ for $i=1,...,k$, with $b_i>0$ for at least one $i$.\
3. Let $\overrightarrow{p}=(p_1,p_2,...,p_n)$ with $p_i>1$ for $i=1,...,N$ and $\sum_{i=1}^N \frac{1}{p_i}>1$. Let $X=\mathcal{D}_0^{1,\overrightarrow{p}}(\Omega)$ be the completion of $\mathcal{C}_0^{\infty}(\Omega)$ with respect to the norm $\|u\|=\displaystyle \sum_{i=1}^N \|\partial_i u\|_{p_i}$. We set, for $u \in X$, $$I_0(u)=\sum_{i=1}^N \frac{1}{p_i}\int_\Omega |\partial_i u|^{p_i}.$$ The corresponding Euler-Lagrange equation is the anisotropic equation $$\label{anis}
-\sum_{i=1}^N \partial_i \left(|\partial_i u|^{p_i-2} \partial_i u\right) = f(u), \quad u \in \mathcal{D}_0^{1,\overrightarrow{p}}(\Omega).$$ For results on this class of problems, we refer to [@Alvesel; @Agnesi2; @FGK; @GP; @Vetois1] and references therein.
We shall establish an abstract result (in the same style as Theorem \[tsw\]) which applies to the problems above. In this sense, we shall prove that the Nehari manifold method applies to problems with nonhomogenous operators. To prove that $\Phi$ has a ground state, we follow a strategy slightly different from [@SW], since we do not prove that $\Phi$ satisfies the Palais-Smale condition at the ground state level. In doing so, we also get rid of the condition and the uniform convexity assumption on $X$. This approach, in a rather simple setting, can be found in [@BS]. Once we have proved that the infimum of $\Phi$ over $\mathcal{N}$ is achieved, we shall deduce that it is a critical value of $\Phi$ thanks to the results of [@SW], which apply to $\mathcal{C}^1$ functionals.
Finally, let us recall (as pointed out in [@SW]) that the Nehari manifold method also has the advantage of not requiring an Ambrosetti-Rabinowitz type condition on $f$, which is customary when dealing with the Palais-Smale condition.
We state now our main result:
\[tp\] Let $X$ be a reflexive Banach space such that $\| \cdot \|$ is a $\mathcal{C}^1$ functional on $X \setminus \{0\}$ and $\Phi: X \rightarrow \Re$ be a $\mathcal{C}^1$ functional such that $\Phi(0)=0$. In addition, we assume that there exist $p, r >1$ such that:
1. $\displaystyle \liminf_{u \to 0} \frac{\Phi'(u)u}{\|u\|^r}>0$
2. For every $u \in X$ we have $\Phi(u)\geq C_0\|u\|^r -I(u)$ where $C_0>0$ and $I$ is a weakly continuous functional on $X$.
3. $\displaystyle \lim_{t \to \infty} \frac{\Phi(tu)}{t^p}=-\infty$ uniformly for $u$ on weakly compact subsets of $X \setminus \{0\}$.
4. For every $u \in X \setminus \{0\}$ the maps $\dis t \mapsto \frac{\Phi'(tu)u}{t^{p-1}}$ and $ t \mapsto \Phi(tu)-\frac{1}{p}\Phi'(tu)tu$ are decreasing and increasing in $(0,\infty)$, respectively.
5. $u \mapsto \Phi'(u)u$ and $u \mapsto \Phi(u)-\frac{1}{p}\Phi'(u)u$ are weakly lower semicontinuous on $X$.
Then $c:=\inf_{\mathcal{N}} \Phi$ is positive and achieved by some $u_0 \neq 0$, i.e. $\Phi$ has a nontrivial ground state at a positive level. If, in addition, $\Phi$ is even then we may choose $u_0 \geq 0$.
The proof is divided in two steps: first we show that $c$ is achieved, and then we use the results of [@SW] to prove that $c$ is a critical value of $\Phi$.\
[**Step 1:**]{} $c$ is achieved\
Given $u \in X \setminus \{0\}$, we set $\gamma_u(t)=\Phi(tu)$ for $t>0$. From (1) and (3) it is clear that $\gamma_u(t)>0$ for $t$ sufficiently small and $\gamma_u(t)<0$ for $t$ sufficiently large. Consequently $\gamma_u$ has a global maximum point $t_u>0$, which is a critical point of $\gamma_u$. Since $t^{1-p}\gamma_u'(t)=t^{1-p}\Phi'(tu)u$, from (4) we infer that $t \mapsto t^{1-p}\gamma_u'(t)$ is decreasing. It follows that $\gamma_u'$ vanishes exactly once, i.e. $t_u$ is the unique critical point of $\gamma_u$. In particular, there holds $\Phi(u)>0$ for every $u \in \mathcal{N}$, so that $c \geq 0$.
We claim that $\mathcal{N}$ is bounded away from zero. Indeed, if $(u_n) \subset \mathcal{N}$ with $u_n \rightarrow 0$ in $X$ then $\frac{\Phi'(u_n)u_n}{\|u_n\|^r} =0$ for every $n$, which contradicts (1). Thus the claim is proved.
Let us prove now that if $(u_n) \subset \mathcal{N}$ is such that $(\Phi(u_n))$ is bounded from above then $(u_n)$ is bounded and, up to a subsequence, $u_n \rightharpoonup u_0$ with $u_0 \not \equiv 0$. Assume by contradiction that $(u_n) \subset \mathcal{N}$ is unbounded. Then we may assume that $\|u_n\| \to \infty$ and $v_n \rightharpoonup v_0$, where $v_n=\frac{u_n}{\|u_n\|}$. If $v_0 \equiv 0$ then, since $t=1$ is the global maximum point of $\gamma_{u_n}$, we have, using (2), $$\label{b1}\Phi(u_n)\geq \Phi(tv_n)\geq C_0t^r-I(tv_n) \rightarrow C_0t^r-I(0), \quad \forall t>0.$$ This contradicts the fact that $(\Phi(u_n))$ is bounded from above. Hence $v_0 \not \equiv 0$ and consequently, by (3), $$\frac{\Phi(u_n)}{\|u_n\|^p}= \frac{\Phi(\|u_n\|v_n)}{\|u_n\|^p} \rightarrow -\infty,$$ which contradicts the fact that $\Phi(u_n)>0$ for every $n$. Therefore $(u_n)$ must be bounded and, up to a subsequence, $u_n \rightharpoonup u_0$. If $u_0 \equiv 0$ then, repeating the argument used in the case $v_0 \equiv 0$, we get $$\Phi(u_n)\geq \Phi(tu_n)\geq C_0t^r\|u_n\|^r-I(tu_n) \geq D_0t^r -I(tu_n) \rightarrow D_0t^r-I(0),$$ where we used (2) and the fact that $\mathcal{N}$ is bounded away from zero. So we get another contradiction, which shows that $u_0 \not \equiv 0$. In particular, if $(u_n)$ is a minimizing sequence for $c$ then we may assume that $u_n \rightharpoonup u_0$ with $u_0 \not \equiv 0$. Let $t_0=t_{u_0}$, i.e. $t_0u_0 \in \mathcal{N}$. From (5), we infer that $$\Phi'(u_0)u_0 \leq \liminf \Phi'(u_n)u_n=0,$$ and, as a consequence, $t_0\leq 1$. We claim that $t_0=1$. Indeed, if $t_0<1$ then, using (4) and (5), we get $$\begin{aligned}
c&\leq& \Phi(t_0u_0)=\Phi(t_0u_0) -\frac{1}{p} \Phi'(t_0u_0)t_0u_0<\Phi(u_0) -\frac{1}{p} \Phi'(u_0)u_0\\ &\leq& \liminf \left( \Phi(u_n) -\frac{1}{p} \Phi'(u_n)u_n \right) =\lim \Phi(u_n)
=c,\end{aligned}$$ which is a contradiction. Therefore $t_0=1$, $u_0 \in \mathcal{N}$, and $\Phi(u_0)=c$. Finally, if $\Phi$ is even then $\Phi(u_0)=\Phi(|u_0|)$, so that $|u_0|$ achieves $c$.\
[**Step 2:**]{} $c$ is a critical value of $\Phi$\
From the previous step it is clear that $\Phi$ satisfies (A2) and (A3) from [@SW]. By [@SW Corollary 10], we deduce that $c=\dis \inf_{\mathcal{S}} \Psi$, where $\Psi$ is defined by $$\Psi(w)=\Phi(t_w w) \quad \text{for } w \in \mathcal{S}.$$ Moreover $\Psi$ is a $\mathcal{C}^1$ functional on $\mathcal{S}$, which is a $\mathcal{C}^1$ submanifold of $X$, and $w$ is a critical point of $\Psi$ if and only if $t_w w$ is a critical point of $\Phi$. This proves that $c$ is a critical value of $\Phi$.
\[r1\]
1. We may easily check that the proof of Theorem \[tp\] still can be carried out if instead of (2), the following conditions hold:\
1. For every $u\in X$ we have $\Phi(u)= I_0(u)-I(u)$, where $I$ is weakly continuous and $I_0$ is such that $\dis \lim_{t \to \infty} I_0(tu)=\infty$ uniformly for $u \in \mathcal{S}$.\
2. For every $u\in X$ we have $\Phi'(u)u=J_0(u)-J(u)$, where $J$ is weakly continuous and $J_0$ is such that $J_0(u_n) \rightarrow 0$ if and only if $u_n \to 0$.\
As a matter of fact, one may repeat and use (A) to get a contradiction if $(u_n) \subset \mathcal{N}$ is such that $(\Phi(u_n))$ is bounded from above and $v_n=\frac{u_n}{\|u_n\|} \rightharpoonup 0$. Moreover, if $(u_n) \subset \mathcal{N}$ and $u_n \rightharpoonup u_0$ then $u_0 \not \equiv 0$. Indeed, if $u_0 \equiv 0$ then, from $J_0(u_n)-J(u_n)=\Phi'(u_n)u_n=0$ and the weak continuity of $J$ we deduce that $J_0(u_n) \to 0$, so that, by (B), $u_n \to 0$, which contradicts the fact that $\mathcal{N}$ is bounded away from zero. The rest of the proof holds without further modifications.\
2. If $\Phi$ is weakly lower semi-continuous then, instead of (4) and (5), one may require only that for every $u \in X \setminus \{0\}$ the map $\dis t \mapsto \frac{\Phi'(tu)u}{t^{p-1}}$ is decreasing in $(0,\infty)$. Note indeed that one may still obtain a minimizing sequence for $c$ such that $u_n \rightharpoonup u_0$ and $u_0 \not \equiv 0$. From the weak lower semicontinuity of $\Phi$ and the fact that $\Phi(u_n)=\max_{t>0} \Phi(tu_n)$ we deduce that $$c \leq \Phi(t_0u_0) \leq \liminf \Phi(t_0 u_n) \leq \liminf \Phi(u_n)=c,$$ i.e. $c$ is achieved.\
3. Unlike [@SW], we don’t make use of the Palais-Smale condition of $\Phi$ to show that $c$ is achieved. Indeed, note that the proof of Theorem \[tp\] does not require the strong convergence of a minimizing sequence for $c$.\
4. From the proof of Theorem \[tp\], we shall highlight the following result: if $(u_n) \subset \mathcal{N}$ is such that $(\Phi(u_n))$ is bounded from above then $(u_n)$ is bounded.\
Following [@SW], we say that $\Phi$ satisfies the Palais-Smale condition on $\mathcal{N}$ if any Palais-Smale sequence of $\Phi$ which is moreover in $\mathcal{N}$ contains a convergent subsequence.
Combining Theorem \[tp\] above and Theorem 2 and Corollary 10 from [@SW], we get the following result:
\[cp\] Under the assumptions of Theorem \[tp\], assume in addition that $\Phi$ is even and satisfies the Palais-Smale condition on $\mathcal{N}$. Then $\Phi$ has infinitely many pairs of critical points.
Applications
============
We apply now Theorem \[tp\] and Corollary \[cp\] to the equations , and . Let us recall that $\Omega \subset \Re^N$ ($N \geq 3$) is a bounded domain, $f:\Re \rightarrow
[0,\infty)$ is an odd $\mathcal{C}^{1}$ function and $F(t)=\displaystyle\int_{0}^{t}f(s)\ ds$, for $t \in \Re$.
A quasilinear equation
----------------------
We assume that $a:[0,\infty)\rightarrow
[0,\infty)$ is $\mathcal{C}^{1}$ in $(0,\infty)$ and we set $A(t)=\displaystyle\int_{0}^{t}a(s)\ ds$ for $t \in \Re$.
\[c1\] Under the above assumptions on $a$, assume in addition that there exist $p\geq q>1$ such that:
1. $k_0\left( 1+t^{\frac{q-p}{p}}\right) \leq a(t) \leq k_1\left( 1+t^{\frac{q-p}{p}}\right), \quad \forall t>0$, where $k_0,k_1$ are positive constants.
2. $a$ is non-increasing.
3. $t\mapsto a(t^p)t^p$ and $t \mapsto A(t^p)- a(t^p)t^p$ are convex in $(0,\infty)$.
4. $\dis \lim_{t \rightarrow 0} \frac{f(t)}{t^{q-1}}=0$.
5. $\dis \lim_{t \rightarrow \infty} \frac{F(t)}{t^p}=\infty $.
6. $\dis \lim_{t \rightarrow \infty} \frac{f(t)}{t^{\alpha-1}}=0$ for some $\alpha \in (p,p^*)$.
7. $t\mapsto \frac{f(t)}{t^{p-1}}$ is increasing on $(0,\infty)$.
Then has a nontrivial and non-negative ground state.
First of all, note that $A(t)$ is well-defined in view of (1). Let $X=W_0^{1,p}(\Omega)$ with $\|u\|=\left(\int_\Omega |\nabla u|^p\right)^{\frac{1}{p}}$. We set, for $u \in X$, $$\label{ph1}
\Phi(u)=\frac{1}{p}\displaystyle\int_{\Omega}A(|\nabla u|^p)-\int_\Omega F(u).$$ From (1) we infer that $$\label{ina}
k_0\left(t^p+ t^q\right)\leq a(t^p)t^p \leq k_1 \left(t^p+ t^q\right), \quad \forall t>0$$ and $$\label{inag}
k_0\left(t^p+ \frac{p}{q} t^q\right)\leq A(t^p)\leq k_1 \left(t^p+ \frac{p}{q} t^q\right), \quad \forall t>0.$$ On the other hand, from (4) and (6) we infer that for any $\varepsilon>0$ there exists $C_{\varepsilon}>0$ such that $$\label{ef}
|f(t)| \leq \varepsilon |t|^{q-1} + C_\varepsilon |t|^{\alpha-1}, \quad \forall t \in \Re.$$ Since $1<q\leq p$, it follows that $\Phi$ is a $\mathcal{C}^1$ functional on $X$. From and the continuity of the embeddings $ X \subset L^{\alpha}(\Omega)$ and $W_0^{1,q}(\Omega) \subset L^q(\Omega)$, we infer that for any $\varepsilon>0$ there exists $C_{\varepsilon}>0$ such that $$\left|\int_\Omega f(u)u \right|\leq \varepsilon \int_\Omega |\nabla u|^q +C_\varepsilon \|u\|^{\alpha} \quad \forall u \in X.$$ Taking $\varepsilon>0$ sufficiently small and using , we get $$\Phi'(u)u \geq (k_0-\varepsilon) \int_\Omega |\nabla u|^q +k_0\|u\|^p - C_\varepsilon \|u\|^{\alpha} \quad \forall u \in X,$$ and consequently $$\liminf_{\|u\| \to 0} \frac{\Phi'(u)u}{\|u\|^p}>0.$$ From , note also that $$\Phi(u)\geq \frac{k_0}{p}\|u\|^p -\int_\Omega F(u)$$ and by the compact Sobolev embedding $X \subset L^{\alpha}(\Omega)$, the functional $u \mapsto \int_\Omega F(u)$ is weakly continuous on $X$. Still from , we have $$\frac{\Phi(tu)}{t^p}\leq \frac{k_1}{q}t^{q-p} \int_\Omega |\nabla u|^q+\frac{k_1}{p}\|u\|^p-\int_\Omega \frac{F(tu)}{t^p} \rightarrow -\infty,$$ uniformly for $u$ on weakly compact subsets of $X \setminus \{0\}$, by (5). From (2) and (7) we have that $$t \mapsto \frac{\Phi'(tu)u}{t^{p-1}}=\int_\Omega a\left(t^p|\nabla u|^p\right) |\nabla u|^p-\int_\Omega \frac{f(tu)}{t^{p-1}}u$$ is decreasing on $(0,\infty)$ for every $u \neq 0$. Furthermore, it is clear that $$t \mapsto A(t^p)-a(t^p)t^p \ \ \mbox{is
non-decreasing in } (0,\infty).$$ On the other hand, (7) provides that $$t \mapsto \frac{1}{p}f(t)t-F(t) \ \ \mbox{is
increasing in } (0,\infty).$$ Thus $t \mapsto \Phi(tu)-\frac{1}{p}\Phi'(tu)tu$ is increasing in $(0,\infty)$ for every $u \neq 0$.
Finally, (3) yields that $u \mapsto \Phi'(u)u$ and $u \mapsto \Phi(u)-\frac{1}{p}\Phi'(u)u$ are weakly lower semi-continuous on $X$. Therefore Theorem \[tp\] applies with $r=p$ and since $\Phi$ is even, we infer that $\Phi$ has a nontrivial and non-negative ground state.
Let $\Psi:X \rightarrow \Re$ be a $\mathcal{C}^1$ functional. Recall that $\Psi'$ belongs to the class $(S_+)$ condition if $$u_n \rightharpoonup u_0 \text{ in } X, \quad \limsup \Phi'(u_n)(u_n-u_0) \leq 0 \quad \Longrightarrow \quad u_n \rightarrow u_0 \text{ in } X.$$ Set $\Psi(u)=\int_\Omega A(|\nabla u|^p)$ for $u \in W_0^{1,p}(\Omega)$. It is known that if $t \mapsto A(t^p)$ is strictly convex and satisfies $$a_1t^p-b_1 \leq A(t^p)\leq a_2t^p+b_2, \quad \forall t>0$$ for some positive constants $a_1,a_2,b_1,b_2$, then $\Psi'$ belongs to the class $(S_+)$ (see [@DU] for a proof).
As a consequence of Remark \[r1\]-(4), we see that under the assumptions of Corollary \[c1\] and the condition $$\label{conv} t \mapsto A(t^p) \ \text{is strictly convex in} \ (0,\infty)$$ the functional $\Phi$ given by satisfies the Palais-Smale condition on $\mathcal{N}$. We infer then the following result:
\[c11\] Under the assumptions of Corollary \[c1\], assume in addition that holds. Then the problem has infinitely many pairs of solutions.
One may easily check that Corollaries \[c1\] and \[c11\] apply in particular to $a(t) \equiv 1$ and $a(t)=1+t^{\frac{q-p}{p}}$, with $p>q>1$, which correspond to the operators $-\Delta_p$ and $-\Delta_p-\Delta_q$, respectively.
A nonlocal equation
-------------------
Let $N=3$, so that $2^*=6$. We assume that $M:[0,\infty)\rightarrow
[0,\infty)$ is a $\mathcal{C}^{1}$ function and we set $\hat{M}(t)=\displaystyle\int_{0}^{t}M(s)\ ds$ for $t \in \Re$.
\[c2\] Under the above assumptions on $M$, assume in addition:
1. $M$ is increasing and $M(0):=m_0>0$.
2. $t\mapsto\displaystyle\frac{M(t)}{t}$ is decreasing.
3. $\dis \lim_{t \rightarrow 0} \frac{f(t)}{t}=0$.
4. $\dis \lim_{t \rightarrow \infty} \frac{F(t)}{t^4}=\infty $.
5. $\dis \lim_{t \rightarrow \infty} \frac{f(t)}{t^{\alpha-1}}=0$ for some $\alpha \in (4,6)$.
6. $t\mapsto \frac{f(t)}{t^{3}}$ is increasing.
Then has a nontrivial and non-negative ground state.
Let $X=H_0^1(\Omega)$ with $\|u\|=\left(\int_\Omega |\nabla u|^2\right)^{\frac{1}{2}}$. We set, for $u \in X$, $$\label{phk}
\Phi(u)=\frac{1}{2}\hat{M}\left(\|u\|^2\right)-\int_\Omega F(u).$$ Note that $\| \cdot \|$ is a $\mathcal{C}^1$ functional on $X$, so that $u \mapsto \hat{M}(\|u\|^2)$ is $\mathcal{C}^1$ as well. From (3) and (5) we have that for every $\varepsilon>0$ there exists $C_{\varepsilon}>0$ such that $$\label{ef2}
|f(t)|\leq \varepsilon |t| +C_\varepsilon |t|^{\alpha-1}, \quad \forall t \in \Re.$$ Thus $\Phi$ is a $\mathcal{C}^1$ functional. Using (1) and , we have $$\Phi'(u)u=M(\|u\|^2)\|u\|^2 - \int_\Omega f(u)u \geq m_0\|u\|^2- \int_\Omega f(u)u.$$ From and the continuous embedding $X \subset L^\alpha(\Omega)$, we get $$\left|\int_\Omega f(u)u \right|\leq \varepsilon \|u\|^2 +C_\varepsilon \|u\|^\alpha, \quad \forall u \in X.$$ Thus we have, for $u \in X$, $$\Phi'(u)u\geq m_0\|u\|^2-\varepsilon \|u\|^2 -C_\varepsilon \|u\|^\alpha.$$ Taking $\varepsilon>0$ sufficiently small, we get $$\liminf_{\|u\| \to 0} \frac{\Phi'(u)u}{\|u\|^2}>0.$$ Moreover, from (2) we infer that $$\hat{M}(t)=\int_0^t \frac{M(s)}{s} s\ ds\geq \frac{M(t)}{t}\int_0^t s\ ds=\frac{1}{2}M(t)t, \quad \forall t>0,$$ and consequently $$\Phi(u)\geq \frac{m_0}{2}\|u\|^2 -\int_\Omega F(u).$$ By and the compact embedding $X \subset L^\alpha(\Omega)$, the functional $u \mapsto \int_\Omega F(u)$ is weakly continuous on $X$. Now, from (2) we have $M(t)\leq M(1)t$ for $t\geq 1$, so that $$M(t)\leq M(1)t+C, \quad \forall t\geq 0$$ for some constant $C>0$. Consequently we have, for $u \in X$, $$\Phi(u)\leq C_1\|u\|^4+C_2\|u\|^2-\int_\Omega F(u),$$ for some $C_1,C_2>0$, so that, by (4), $\frac{\Phi(tu)}{t^4} \rightarrow -\infty$ uniformly on weakly compact subsets of $X \setminus \{0\}$. From (2) and (6), it follows that $$t \mapsto \frac{\Phi'(tu)u}{t^3}=\frac{1}{t^2}M(t^2\|u\|^2)\|u\|^2-\int_\Omega \frac{f(tu)}{t^3}u$$ is decreasing for every $u \neq 0$.
Furthermore, note that (2) yields $tM'(t)\leq M(t)$ for any $t>0$, and consequently $$t \mapsto \frac{1}{2}\hat{M}(t)-\frac{1}{4}M(t)t \quad \text{ is increasing}.$$ Hence $t \mapsto \Phi(tu)-\frac{1}{4}\Phi'(tu)tu$ is increasing in $(0,\infty)$ for every $u \neq 0$.
Finally, since $M$ and $t \mapsto \frac{1}{2}\hat{M}(t)-\frac{1}{4}M(t)t$ are increasing, the mappings $$u \mapsto M(\|u\|^2) \quad \text{and} \quad u \mapsto \frac{1}{2}\hat{M}(\|u\|^2)-\frac{1}{4}M(\|u\|^2)\|u\|^2$$ are weakly lower semicontinuous on $X$. Therefore Theorem \[tp\] applies with $r=2$ and $p=4$. Note also that $\Phi$ is even. The proof is now complete.
\[c22\] Under the assumptions of Corollary \[c2\], the problem has infinitely many pairs of solutions.
Let $(u_n) \subset \mathcal{N}$ be a Palais-Smale sequence for $\Phi$, defined in . By Remark \[r1\]-(4), we know that $(u_n)$ is bounded, so that, up to a subsequence, $u_n \rightharpoonup u_0$ in $X$. From , we know that $$\int_\Omega f(u_n)(u_n-u_0)=o(1).$$ Hence $$M(\|u_n\|^2) \int_\Omega \nabla u_n \nabla (u_n-u_0)=\Phi'(u_n)(u_n-u_0)+o(1)=o(1).$$ Since $M$ is continuous and $M(t)\geq m_0>0$ for all $t\geq 0$, we infer that $M(\|u_n\|^2)$ is bounded and bounded away from zero, so that $$\int_\Omega \nabla u_n \nabla (u_n-u_0)=o(1).$$ By the uniform convexity of $H_0^1(\Omega)$, we get $u_n \rightarrow u_0$.
Besides $M(t)=at+b$, with $a,b>0$, Corollaries \[c2\] and \[c22\] apply also to $M(t) = m_{0} + \ln(1 + t)$ and $
M(t)=m_{0}+\displaystyle\sum_{i=1}^{k}b_{i}t^{\gamma_{i}}
$, where $b_{i}\geq 0$ and $\gamma_{i}\in (0,1]$ for $i=1,...,k$, with $b_i>0$ for at least one $i$.
A anisotropic equation
----------------------
Let $1<p_1\leq p_2 \leq ...\leq p_N$ be such that $\sum_{i=1}^N \frac{1}{p_i}>1$ and $p_N < p^*$, where $p^*=\frac{N}{\left(\sum_{i=1}^N \frac{1}{p_i}\right)-1}$. If $r>1$, we denote by $\|v\|_r$ the norm of $v$ in $L^r(\Omega)$.
\[c3\] Under the above assumptions, assume in addition:
1. $\lim_{t \to 0^+} \frac{f(t)}{t^{p_1-1}}=0$.
2. $\lim_{t \to \infty} \frac{F(t)}{t^{p_N}}=\infty$.
3. $\lim_{t \to \infty} \frac{f(t)}{t^{\alpha-1}}=\infty$ for some $\alpha \in (p_N,p^*)$.
4. $t \mapsto \frac{f(t)}{t^{p_N-1}}$ is increasing.
Then has a nontrivial and non-negative ground state.
Let $X=\mathcal{D}_0^{1,\overrightarrow{p}}(\Omega)$ be the completion of $\mathcal{C}_0^{\infty}(\Omega)$ with respect to the norm $\|u\|=\displaystyle \sum_{i=1}^N \|\partial_i u\|_{p_i}$. It is known that $X$ is a reflexive Banach space which embedds continuously in $L^q(\Omega)$ if $q \in [1,p^*]$, and compactly if $q\in [1,p^*)$, cf [@FGK].
We set, for $u \in X$, $$\label{phan}
\Phi(u)=\sum_{i=1}^N \frac{1}{p_i}\int_\Omega |\partial_i u|^{p_i}-\int_\Omega F(u).$$ Note that if $\|u\|\leq 1$ then $\|\partial_i u\|_{p_i}<1$ for $i=1,..,N$, and since $p_i\leq p_N$ we get $\|\partial_i u\|_{p_i}^{p_i} \geq \|\partial_i u\|_{p_i}^{p_N}$ for $i=1,..,N$. Thus $$\sum_{i=1}^N \|\partial_i u\|_{p_i}^{p_i} \geq \sum_{i=1}^N \|\partial_i u\|_{p_i}^{p_N} \geq C \left(\sum_{i=1}^N \|\partial_i u\|_{p_i}\right)^{p_N} = C\|u\|^{p_N}.$$ Using (1), (3) and the continuous embedding $X \subset L^{\alpha}(\Omega)$, we have that for any $\varepsilon >0$ there exists $C_\varepsilon>0$ such that $$\left|\int_\Omega f(u)u \right|\leq \varepsilon \|\partial _1 u\|_{p_1}^{p_1} +C_\varepsilon \|u\|^\alpha, \quad \forall u \in X.$$ Hence $$\begin{aligned}
\Phi'(u)u&=&\sum_{i=1}^N \|\partial_i u\|_{p_i}^{p_i} - \int_\Omega f(u)u \geq \sum_{i=1}^N \|\partial_i u\|_{p_i}^{p_i} -\varepsilon \|\partial _1 u\|_{p_1}^{p_1}-C_\varepsilon \|u\|^\alpha\\
&\geq & (1-\varepsilon)\sum_{i=1}^N \|\partial_i u\|_{p_i}^{p_i} -C_\varepsilon \|u\|^\alpha \geq C(1-\varepsilon)\|u\|^{p_N} - C_\varepsilon \|u\|^\alpha.\end{aligned}$$ Taking $\varepsilon<1$ we get $$\liminf_{u \to 0} \frac{\Phi'(u)u}{\|u\|^{p_N}}>0.$$ Note also that $$\Phi(u) =I_0(u)-I(u)\quad \text{and} \quad \Phi'(u)u=J_0(u)-J(u),$$ where $$I_0(u)=\sum_{i=1}^N \frac{1}{p_i}\|\partial_i u\|_{p_i}^{p_i},\quad I(u)=\int_\Omega F(u),$$ $$J_0(u)=\sum_{i=1}^N \|\partial_i u\|_{p_i}^{p_i}\quad \text{and} \quad J(u)=\int_\Omega f(u)u.$$ From the compact embedding $X \subset L^{\alpha}(\Omega)$, it follows that $I$ and $J$ are weakly continuous. Moreover there exists $C>0$ such that $$I_0(tu)\geq \frac{t^{p_1}}{p_N} \sum_{i=1}^N \|\partial_i u\|_{p_i}^{p_i} \geq C \frac{t^{p_1}}{p_N} \|u\|^{p_N}$$ if $t>1$ and $\|u\| \leq 1$. In particular, if $u \in \mathcal{S}$ then $I_0(tu)\to \infty$ as $t \to \infty$. In addition, it is clear that $J_0(u_n) \to 0$ if and only if $u_n \to 0$ in $X$. From (2) we have $$\frac{\Phi(tu)}{t^{p_N}}= \sum_{i=1}^N \frac{t^{p_i-p_N}}{p_i}\|\partial_i u\|_{p_i}^{p_i} - \int_\Omega \frac{F(tu)}{t^{p_N}} \to -\infty,$$ whereas, from (4), it follows that $$t \mapsto t^{1-p_N}\Phi'(tu)u= \sum_{i=1}^N t^{p_i-p_N} \|\partial_i u\|_{p_i}^{p_i} - \int_\Omega \frac{f(tu)}{t^{p_N-1}}\ \ \text{is decreasing in } (0,\infty).$$ Still by (4), we have that $$t \mapsto \frac{1}{p_N}f(t)t-F(t)\ \ \text{is increasing in } (0,\infty).$$ Thus for any $u \neq 0$ $$t \mapsto \Phi(tu)-\frac{1}{p_N}\Phi'(tu)tu=\sum_{i=1}^N \left(\frac{1}{p_i}-\frac{1}{p_N} \right)t^{p_i-p_N}\| \partial_i u\|_{p_i}^{p_i}+\int_\Omega \left(\frac{1}{p_N}f(tu)t-F(tu)\right)$$ is increasing in $(0,\infty)$.
Finally, it is clear that $u \mapsto \Phi'(u)u$ and $\mapsto \Phi(u)-\frac{1}{p_N}\Phi'(u)u$ are weakly lower semicontinuous on $X$, and $\Phi$ is even.
In view of Theorem \[tp\] and Remark \[r1\]-(1), we infer that has a nontrivial and non-negative ground state.
Under the assumptions of Corollary \[c3\], the problem has infinitely many pairs of solutions.
Let $(u_n) \subset \mathcal{N}$ be a Palais-Smale sequence for $\Phi$, defined in . Since $(u_n)$ is bounded, passing to a subsequence, if necessary, we have $$u_{n}\rightharpoonup u \ \ \mbox{in} \ \
D_{0}^{1,\overrightarrow{p}}(\Omega), \ \
u_{n}\rightarrow u \ \ \mbox{in} \ \ L^{\sigma}(\Omega)
\mbox{ for} \ \ \sigma \in [1, p^{*}),\ \
\text{and} \ \
u_{n}\rightarrow u \ \ \mbox{a.e in} \ \ \Omega.$$ Since $|f(t)|\leq C(1+|t|^{\alpha-1})$ for every $t \in \Re$, we get $$\begin{aligned}
\label{converg1}
\displaystyle\int_{\Omega}f(u_{n})u_n -
\displaystyle\int_{\Omega}f(u_n)u = o(1).\end{aligned}$$ Now, from $u_{n}\rightharpoonup u$ in $D_{0}^{1,\overrightarrow{p}}(\Omega)$ we have $$\begin{aligned}
\label{converg2}
\displaystyle\sum^{N}_{i=1}\displaystyle\int_{\Omega}|\partial_i u|^{p_{i}-2}\partial_i u\ \partial_i u_n
-\displaystyle\sum^{N}_{i=1}\displaystyle\int_{\Omega}|\partial_i u|^{p_i}= o(1).\end{aligned}$$ We use now the standard inequality (cf. [@Si]) $$(|x|^{p-2}x-|y|^{p-2}y)(x-y) \geq \begin{cases} C_{p}|x-y|^{p}, & \text{if } p\geq 2\\ C_p\frac{|x-y|^{2}}{(|x|+|y|)^{2-p}}, & \text{if } 1<p<2, \end{cases}$$ which holds for some $C_p>0$. Note that if $1<p_i< 2$ then $$\begin{aligned}
\|\partial_i u_n- \partial_i u\|_{p_i}^{p_i} &=&\int_\Omega \frac{|\partial_i u_n- \partial_i u|^{p_i}}{\left(|\partial_i u_n|+|\partial_i u|\right)^{\frac{p_i(2-p_i)}{2}}}\left(|\partial_i u_n|+|\partial_i u|\right)^{\frac{p_i(2-p_i)}{2}} \\&\leq & \left(\int_\Omega \frac{|\partial_i u_n- \partial_i u|^2}{\left(|\partial_i u_n|+|\partial_i u|\right)^{2-p_i}}\right)^{\frac{p_i}{2}} \left(\int_\Omega \left(|\partial_i u_n|+|\partial_i u|\right)^{p_i}\right)^{\frac{2-p_i}{2}}\\ &\leq &
C \left(\int_{\Omega}\left(|\partial_i u_n|^{p_{i}-2}\partial_i u_n- |\partial_i u|^{p_{i}-2}\partial_i u\right)\left(\partial_i u_n-\partial_i u\right)\right)^{\frac{p_i}{2}}\end{aligned}$$ Thus $$\int_{\Omega}\left(|\partial_i u_n|^{p_{i}-2}\partial_i u_n- |\partial_i u|^{p_{i}-2}\partial_i u\right)\left(\partial_i u_n-\partial_i u\right)\geq
\begin{cases}
C\|\partial_i u_n- \partial_i u\|_{p_i}^{p_i}, & \text{if } p_i\geq2 \\
C\|\partial_i u_n- \partial_i u\|_{p_i}^2, & \text{if } 1<p_i<2.
\end{cases}$$ for some $C>0$, and consequently $$\begin{aligned}
\Phi'(u_n)(u_n-u)&=& \sum_{i=1}^N \int_{\Omega}|\partial_i u_n|^{p_{i}-2}\partial_i u_n\left(\partial_i u_n-\partial_i u\right)+o(1)\\
&=&
\sum_{i=1}^N \int_{\Omega}\left(|\partial_i u_n|^{p_{i}-2}\partial_i u_n- |\partial_i u|^{p_{i}-2}\partial_i u\right)\left(\partial_i u_n-\partial_i u\right)+o(1)\\
&\geq &C\left(\sum_{p_i\geq 2} \|\partial_i u_n- \partial_i u\|_{p_i}^{p_i} + \sum_{1<p_i<2} \|\partial_i u_n- \partial_i u\|_{p_i}^2\right)+o(1)\end{aligned}$$ Therefore $\|\partial_i u_n- \partial_i u\|_{p_i} \rightarrow 0$ for $i=1,...,N$, so that $u_{n}\rightarrow u$ in $D_{0}^{1,\overrightarrow{p}}(\Omega)$ and the proof is complete.
[99]{}
Adimurthi, Existence of positive solutions of the semilinear Dirichlet problem with critical growth for the n-Laplacian. Ann. Scuola Norm. Sup. Pisa Cl. Sci. (4) 17 (1990), no. 3, 393–413.
C. O. Alves, F. J. S. A. Corrêa, G. M. Figueiredo, On a class of nonlocal elliptic problems with critical growth. Differ. Equ. Appl. 2 (2010), no. 3, 409–-417.
C. O. Alves, A. El Hamidi, Existence of solution for a anisotropic equation with critical exponent, Differential Integral Equations, [**21**]{} (2008), 25-40.
A. Ambrosetti and E. Colorado, Standing waves of some coupled nonlinear Schrödinger equations, J. Lond. Math. Soc., 75 (2007), 67-–82.
M. Badiale, E. Serra, Semilinear elliptic equations for beginners. Existence results via the variational approach. Universitext. Springer, London, 2011.
A. Bahri and H. Berestycki, A perturbation method in critical point theory and applications, Trans. Amer. Math. Soc. 267 (1981), 1-–32.
T. Bartsch, Z.-Q. Wang, Zhi-Qiang, M. Willem, The Dirichlet problem for superlinear elliptic equations. Stationary partial differential equations. Vol. II, 1–55, Handb. Differ. Equ., Elsevier/North-Holland, Amsterdam, 2005.
C. J. Batkam, Ground state solution of a nonlocal boundary-value problem. Electron. J. Differential Equations 2013, No. 257, 8 pp.
V. Benci, A.M. Micheletti and D. Visetti, An eigenvalue problem for a quasilinear elliptic field equation, J. Differential Equations 184 (2002), 299–320.
P. A. Binding, P. Drabek, Y. X, Huang, On Neumann boundary value problems for some quasilinear elliptic equations, Vol. 1997(1997), No. 05. pp. 1-11.
K.J. Brown, Y. Zhang, [*The Nehari manifold for a semilinear elliptic equation with a sign-changing weight function.*]{} J. Differential Equations 193 (2003), no. 2, 481–499.
K.J. Brown, T-F. Wu, [*A fibering map approach to a semilinear elliptic boundary value problem.*]{} Electron. J. Differential Equations 2007, No. 69, 9 pp. (electronic)
A. Castro, J. Cossio and J. M. Neuberger, A sign-changing solution for a superlinear Dirichlet problem, Rocky Mountain J. Math. 27 (1997), 1041–-1053.
L. Cherfils and Y. Ilyasov, On the stationary solutions of generalized reaction diffusion equations with $p$&$q$-Laplacian, Commun. Pure Appl. Anal. 4 (2005), no. 1, 9–-22.
S. Cingolani, M. Degiovanni, Nontrivial solutions for $p$-Laplace equations with right-hand side having $p$-linear growth at infinity, Comm. Partial Differential Equations 30 (2005), no. 7-9, 1191–-1203.
C. V. Coffman and W. K. Ziemer, A prescribed mean curvature problem on domains without radial symmetry, SIAM J. Math. Anal. 22(4), 982–-990 (1991).
A. Di Castro, [*Existence and Regularity Results for Anisotropic Elliptic Problems*]{}, Adv. Nonlinear Stud., [**9**]{} (2009), 367-393.
J.M. Do Ó, P. Ubilla, A multiplicity result for a class of superquadratic Hamiltonian systems, Electron. J. Differential Equations 2003 (15) (2003) 1–14.
G. M. Figueiredo, Existence of a positive solution for a Kirchhoff problem type with critical growth via truncation argument. J. Math. Anal. Appl. 401 (2013), no. 2, 706-–713.
G. M. Figueiredo, Existence of positive solutions for a class of $p$&$q$ elliptic problems with critical growth on $R^N$, J. Math. Anal. Appl. 378 (2011), no. 2, 507-–518
I. Fragalà, F. Gazzola, B. Kawohl, Existence and nonexistence results for anisotropic quasilinear elliptic equations. Ann. Inst. H. Poincaré Anal. Non Linéaire 21 (2004), no. 5, 715-–734.
L. Gasinski, N. S. Papageorgiou, Anisotropic nonlinear Neumann problems. Calc. Var. Partial Differential Equations 42 (2011), no. 3-4, 323–354.
X. He, W. Zou, Ground states for nonlinear Kirchhoff equations with critical growth. Ann. Mat. Pura Appl. (4) 193 (2014), no. 2, 473–-500.
S.Hu, N.S. Papageorgiou, Solutions of nonlinear nonhomogeneous Neumann and Dirichlet problems. Commun. Pure Appl. Anal. 12 (2013), no. 6, 2889-–2922.
T. F. Ma, Remarks on an elliptic equation of Kirchhoff type, Nonlinear Anal., 63 (2005), e1967–e1977.
D. Mugnai, N. S. Papageorgiou, Wang’s multiplicity result for superlinear (p,q)-equations without the Ambrosetti-Rabinowitz condition. Trans. Amer. Math. Soc. 366 (2014), no. 9, 4919–-4937.
D. Naimen, The critical problem of Kirchhoff type elliptic equations in dimension four. J. Differential Equations 257 (2014), no. 4, 1168-–1193.
Z. Nehari, On a class of nonlinear second-order differential equations, Trans. Amer. Math. Soc. 95 (1960), 101–-123.
J. Simon, Regularité de la solution d’une equation non lineaire dans $\Re^N$, Lecture Notes in Math. No. 665 P. Benilan editors Springer Verlag, 1978.
A. Szulkin and T. Weth, The method of Nehari manifold. Handbook of nonconvex analysis and applications, 597–632, Int. Press, Somerville, MA, 2010.
G. Tarantello, On nonhomogeneous elliptic equations involving critical Sobolev exponent. Ann. Inst. H. Poincaré Anal. Non Linéaire 9 (1992), no. 3, 281–304.
J. Vétois, [*Existence and regularity for critical anisotropic equations with critical directions*]{}, Adv. Differential Equations, [**16**]{} (2011), 61-83.
[^1]: The first author was supported by CNPq/PQ 301242/2011-9. The second author was supported by FONDECYT 11121567
| {
"pile_set_name": "ArXiv"
} |
---
author:
-
bibliography:
- 'example.bib'
title: A Second Order Derivatives based Approach for Steganography
---
{#sec:introduction}
The objective of any *steganographic* approach is to dissimulate a message into another one in an imperceptible way. In the context of this work, the host message is an image in the spatial domain, *e.g.*, a raw image. A coarse steganographic technique consists in replacing the Least Significant Bit (LSB) of each pixel with the bits of the message to hide. On the contrary, the goal of a *steganalysis* approach is to decide whether a given content embeds or not a hidden message.
Steganographic schemes are evaluated according to their ability to face steganalyser tools. The efficiency of the former increases with the number of errors produced by the latter. An error is either a false positive decision or a false negative one. In the former case, the image is abusively declared to contain a hidden message whereas it is an original host. In the latter case, the image is abusively declared as free of hidden content while it embeds a message. The average error is thus the mean of these two ones. Let us select a security level expressed as a number in $[0,0.5]$, when developing a new steganographic scheme, the objective is to find an approach that maximizes the size of the message that can be embedded in any image with an average error larger than this security level.
Creating an efficient steganographic scheme aims at designing an accurate distortion function that associates to each pixel the ability of modifying it. This function indeed allows the extraction the set of pixels that can be modified with the smallest detectability. Highly Undetectable steGO (HUGO) [@DBLP:conf/ih/PevnyFB10], WOW [@conf/wifs/HolubF12], UNIWARD [@HFD14], STABYLO [@DBLP:journals/adt/CouchotCG15], EAI-LSBM [@5411758], and MVG [@FK2013] are some of the most efficient instances of such schemes. The next step, *i.e.*, the embedding process, is often common to all the steganographic schemes. For instance, this final step is the Syndrome-Trellis Code (STC) [@FillerJF11] in many steganographic schemes like the aforementioned ones.
The distortion function of HUGO evaluates for each pixel in $(x,y)$ the sum of the directional SPAM features of the cover and of the image after modifying its value $P(x,y)$. In STABYLO and EAI-LSBM, the distortion functions are based on edge detection. The higher the difference between two consecutive pixels is, the smaller its distortion value is. WOW (and similarly UNIWARD) distortion function is based on wavelet-based directional filters. These filters are applied twice to evaluate the cost of $\pm 1$ modification of the cover. In all these previously detailed schemes, the function is designed to focus on a specific area, namely textured or noisy regions where it is difficult to provide an accurate model. The distortion function of MVG, for its part, is based on minimizing the Kullback-Leibler divergence.
In all aforementioned schemes, the distortion function returns a large value in a easy-modelable smooth area and a small one in textured, a “chaotic” area, *i.e.*, where there is no model. In other words, these approaches assign a large value to pixels that are in a specific level curve: modifying this pixel leads to associating another level to this pixel. Conversely, when a pixel is not in a well defined level curve, its modification is hard to detect.
The mathematical tools that usually evaluate the level curves are first and second order derivatives. Level curves are indeed defined to be orthogonal to vectors of first order derivatives, *i.e.*, to *gradients*. Second order derivatives allow to detect whether these level curves are locally well defined or, on the contrary, change depending on neighborhood. Provided we succeed in defining a function $P$ that associates to each pixel $(x,y)$ its value $P(x,y)$, pixels such that all the second order derivatives having high values are good candidates to embed the message bits.
However, such a function $P$ is only known on pixels, *i.e.*, on a finite set of points. Its first and second derivatives cannot thus be mathematically computed. At most, one can provide approximate functions on the set of pixels. Even with such a function, ordering pixels according to the values of the Hessian matrix (*i.e.*, the matrix of second order derivatives) is not a natural task.
This work first explains how such first and second order approximations can be computed on numerical images (Section \[sec:gradient\]). Two proposals to compute second order derivatives are proposed and proven (Section \[sec:second\] and Section \[sec:poly\]). This is the main contribution of this work. An adaptation of an existing distortion function is studied in Section \[sec:distortion\]. A whole set of experiments is presented in Section \[sec:experiments\]. Concluding remarks and future work are presented in the last section.
Derivatives in an Image {#sec:gradient}
=======================
This section first recalls links between level curves, gradient, and Hessian matrix (Section \[sub:general\]). It next analyses them using kernels from signal theory (Section \[sub:class:1\] and Section \[sub:class:2\]).
Hessian Matrix {#sub:general}
--------------
Let us consider that an image can be seen as a numerical function $P$ that associates a value $P(x,y)$ to each pixel of coordinates $(x,y)$. The variations of this function in $(x_0,y_0)$ can be evaluated thanks to its gradient $\nabla{P}$, which is the vector whose two components are the partial derivatives in $x$ and in $y$ of $P$:
$$\nabla{P}(x_0,y_0) = \left(\frac{\partial P}{\partial x}(x_0,y_0),\frac{\partial P}{\partial y}(x_0,y_0)\right).$$
In the context of two variables, the gradient vector points to the direction where the function has the highest increase. Pixels with close values thus follow level curve that is orthogonal to the one of highest increase.
The variations of the gradient vector are expressed in the Hessian matrix $H$ of second-order partial derivatives of $P$.
$$H = \begin{bmatrix}
\dfrac{\partial^2 P}{\partial x^2} &
\dfrac{\partial^2 P}{\partial x \partial y} \\
\dfrac{\partial^2 P}{\partial y \partial x} &
\dfrac{\partial^2 P}{\partial y^2} \\
\end{bmatrix}.$$
In one pixel $(x_0,y_0)$, the larger the absolute values of this matrix are, the more the gradient is varying around $(x_0,y_0)$. We are then left to evaluate such an Hessian matrix.
This task is not as easy as it appears since natural images are not defined with differentiable functions from $\R^2$ to $\R$. Following subsections provide various approaches to compute these Hessian matrices.
Classical Gradient Image Approaches {#sub:class:1}
-----------------------------------
In the context of image values, the most used approaches to evaluate gradient vectors are the well-known “Sobel”, “Prewitt”, “Central Difference”, and “Intermediate Difference” ones.
Name Sobel Prewitt
-------- --------------------------------------------------------------------------------------------------- -----------------------------------------------------------------------------------------
Kernel $\textit{Ks}= \begin{bmatrix} -1 & 0 & +1 \\ -2 & 0 & +2 \\ -1 & 0 & +1 \end{bmatrix} $ $\textit{Kp}= \begin{bmatrix} -1 & 0 & +1 \\ -1 & 0 & +1 \\ -1 & 0 & +1 \end{bmatrix} $
Name Central Intermediate
Difference Difference
Kernel $\textit{Kc}= \begin{bmatrix} 0&0&0 \\ -\dfrac{1}{2} & 0 & +\dfrac{1}{2} \\ 0&0&0 \end{bmatrix} $ $\textit{Ki}= \begin{bmatrix} 0&0&0 \\ 0 & -1 & 1 \\ 0&0&0 \end{bmatrix} $
: Kernels of usual image gradient operators\[table:kernels:usual\]
Each of these approaches applies a convolution product $*$ between a kernel $K$ (recalled in Table \[table:kernels:usual\]) and a $3\times 3$ window of pixel values $A$. The result $A * K$ is an evaluation of the horizontal gradient, *i.e.*, $\dfrac{\partial P}{\partial x}$ expressed as a matrix in $\R$. Let $K\rl$ be the result of a $\pi/2$ rotation applied on $K$. The vertical gradient $\dfrac{\partial P}{\partial y}$ is similarly obtained by computing $A * K\rl$, which is again expressed as a matrix in $\R$.
The two elements of the first line of the Hessian matrix are the result of applying the horizontal gradient calculus first on $\dfrac{\partial P}{\partial x}$ and next on $\dfrac{\partial P}{\partial y}$. Let us study these Hessian matrices in the next section.
Hessian Matrices induced by Gradient Image Approaches {#sub:class:2}
-----------------------------------------------------
First of all, it is well known that $\dfrac{\partial^2 P}{\partial x \partial y} $ is equal to $\dfrac{\partial^2 P}{\partial y \partial x}$ if the approach that computes the gradient and the one which evaluates the Hessian matrix are the same. For instance, in the Sobel approach, it is easy to verify that the calculus of $\dfrac{\partial^2 P}{\partial x \partial y}$ and of $\dfrac{\partial^2 P}{\partial y \partial x}$ are both the result of a convolution product with the Kernel $\textit{Ks}''_{xy}$ given in Table \[table:hessian:usual\]. This one summarizes kernels $K_{x^2}''$ and $K_{xy}''$ that allow to respectively compute $\dfrac{\partial^2 P}{\partial x^2}$ and $\dfrac{\partial^2 P}{\partial x \partial y}$ with a convolution product for each of the usual image gradient operator.
----------------------------------------------------------------------------------------------
Sobel Prewitt
-------------------------------------------------------------- -------------------------------
$ $
\textit{Ks}_{x^2}''= \textit{Kp}_{x^2}''=
\begin{bmatrix} \begin{bmatrix}
1 & 0 & -2 & 0 & 1 \\ 1 & 0 & -2 & 0 & 1 \\
4 & 0 & -8 & 0 & 4 \\ 2 & 0 & -4 & 0 & 2 \\
6 & 0 & -12 & 0 & 6 \\ 3 & 0 & -6 & 0 & 3 \\
4 & 0 & -8 & 0 & 4 \\ 2 & 0 & -4 & 0 & 2 \\
1 & 0 & -2 & 0 & 1 1 & 0 & -2 & 0 & 1
\end{bmatrix} \end{bmatrix}
$ $
$ $
\textit{Ks}_{xy}''= \textit{Kp}_{xy}''=
\begin{bmatrix} \begin{bmatrix}
-1 & -2 & 0 & 2 & 1 \\ -1 & -1 & 0 & 1 & 1 \\
-2 & -4 & 0 & 4 & 2 \\ -1 & -1 & 0 & 1 & 1 \\
0 & 0 & 0 & 0 & 0 \\ 0 & 0 & 0 & 0 & 0 \\
2 & 4 & 0 & -4 & -2 \\ 1 & 1 & 0 & -1 & -1 \\
1 & 2 & 0 & -2 & -1 1 & 1 & 0 & -1 & -1
\end{bmatrix} \end{bmatrix}
$ $
Central Intermediate
Difference Difference
$ $
\textit{Kc}_{x^2}''= \textit{Ki}_{x^2}''=
\begin{bmatrix} \begin{bmatrix}
0 & 0 & 0 & 0 & 0 \\ 0 & 0 & 0 & 0 & 0 \\
0 & 0 & 0 & 0 & 0 \\ 0 & 0 & 0 & 0 & 0 \\
\dfrac{1}{4} & 0 & -\dfrac{1}{2} & 0 & \dfrac{1}{4} \\ 0 & 0 & 1 & -2 & 1 \\
0 & 0 & 0 & 0 & 0 \\ 0 & 0 & 0 & 0 & 0 \\
0 & 0 & 0 & 0 & 0 0 & 0 & 0 & 0 & 0
\end{bmatrix} \end{bmatrix}
$ $
$ $
\textit{Kc}_{xy}''= \textit{Ki}_{xy}''=
\begin{bmatrix} \begin{bmatrix}
-\dfrac{1}{4} & 0 & \dfrac{1}{4}\\ 0 & -1 & 1 \\
0 & 0 & 0 \\ 0 & 1 & -1 \\
\dfrac{1}{4} & 0 & -\dfrac{1}{4} 0 & 0 & 0
\end{bmatrix} \end{bmatrix}
$ $
----------------------------------------------------------------------------------------------
: Kernels of second order gradient operators\[table:hessian:usual\]
The Sobel kernel $\textit{Ks}_{x^2}''$ allows to detect whether the central pixel belongs to a “vertical” edge, even if this one is noisy, by considering its vertical neighbours. The introduction of these vertical neighbours in this kernel is meaningful in the context of finding edges, but not very accurate when the objective is to precisely find the level curves of the image. Moreover, all the pixels that are in the second and the fourth column in this kernel are ignored. The Prewitt Kernel has similar drawbacks in this context.
The Central Difference kernel $\textit{Kc}_{x^2}''$ is not influenced by the vertical neighbours of the central pixel and is thus more accurate here. However, the kernel $\textit{Kc}_{xy}''$ again looses the values of the pixels that are vertically and diagonally aligned with the central one.
Finally, the Intermediate Difference kernel $\textit{Ki}_{x^2}''$ shifts to the left the value of horizontal variations of $\dfrac{\partial P}{\partial x}$: the central pixel $(0,0)$ exactly receives the value $\dfrac{P(0,2)-P(0,1)}{1} - \dfrac{P(0,1)-P(0,0)}{1}$, which is an approximation of $\dfrac{\partial P}{\partial x}(0,1)$ and not of $\dfrac{\partial P}{\partial x}(0,0)$. Furthermore the Intermediate Difference kernel $\textit{Ki}_{xy}''$ only deals with pixels in the upper right corner, loosing all the other information.
Due to these drawbacks, we are then left to produce another approach to find the level curves with strong accuracy.
Second Order Kernels for Accurate Level Curves {#sec:second}
==============================================
This step aims at finding accurate level curve variations in an image. We do not restrict the kernel to have a fixed size (*e.g.*, $3\times3$ or $5 \times 5$ as in the aforementioned schemes). This step is thus defined with kernels of size $(2n+1)\times (2n+1)$, $n \in \{1,2,\dots,N\}$, where $N$ is a parameter of the approach.
The horizontal gradient variations are thus captured thanks to $(2n+1)\times (2n+1)$ square kernels
$$\arraycolsep=1.4pt
\def\arraystretch{1.4}
\def\arraystretch{1.4}
\textit{Ky}_{x^2}''=
\left(
\begin{array}{ccccccccc}
0 & & & & \dots& & & & 0 \\
\vdots & & & & & & & & \vdots \\
0 & & & & \dots & & & & 0 \\
\dfrac{1}{2n}& 0 & \dots & 0 & -\dfrac{2}{2n} & 0 & \dots & 0& \dfrac{1}{2n} \\
0 & & & & \dots& & & & 0 \\
\vdots & & & & & & & & \vdots \\
0 & & & & \dots & & & & 0
\end{array}
\right)$$
When the convolution product is applied on a $(2n+1)\times(2n+1)$ window, the result is $\dfrac{1}{2}\left(\dfrac{P(0,n)-P(0,0)}{n} - \dfrac{P(0,0)-P(0,-n)}{n}\right)$, which is indeed the variation between the gradient around the central pixel. This proves that this calculus is a correct approximation of $\dfrac{\partial^2 P}{\partial x^2}$.
When $n$ is 1, this kernel is a centered version of the horizontal Intermediate Difference kernel $\textit{Ki}_{x^2}''$ modulo a multiplication by $1/2$. When $n$ is 2, this kernel is equal to $\textit{Kc}_{x^2}''$.
The vertical gradient variations are again obtained by applying a $\pi/2$ rotation to each horizontal kernel $\textit{Ky}_{x^2}''$.
The diagonal gradient variations are obtained thanks to the $(2n+1)\times (2n+1)$ square kernels $\textit{Ky}_{xy}''$ defined by
$$\arraycolsep=1.4pt
\def\arraystretch{1.4}
\textit{Ky}_{xy}'' = \dfrac{1}{4}
\left(
\begin{array}{ccccccccc}
\frac{1}{n^2}& \dots & \frac{1}{2n} & \frac{1}{n}
& 0 &
-\frac{1}{n}&-\frac{1}{2n} & \dots & -\frac{1}{n^2}
\\
\vdots & 0 & &
& \dots &
& & 0 & \vdots
\\
\frac{1}{2n} & 0 & &
& \dots &
& & 0& -\frac{1}{2n}
\\
\frac{1}{n} & 0 & &
& \dots &
& & 0 & -\frac{1}{n}
\\
0 & & & & \dots& & & & 0 \\
-\frac{1}{n} & 0 & &
& \dots &
& &0 & \frac{1}{n}
\\
-\frac{1}{2n} & 0 & &
& \dots &
& & 0 & \frac{1}{2n}
\\
\vdots & 0 & &
& \dots &
& & 0& \vdots
\\
-\frac{1}{n^2}& \dots & -\frac{1}{2n} & -\frac{1}{n}
& 0 &
\frac{1}{n}& \frac{1}{2n} & \dots & \frac{1}{n^2}
\end{array}
\right).$$
When $n$ is 1, $\textit{Ky}_{xy}''$ is equal to the kernel $\textit{Kc}_{xy}''$, and the average vertical variations of the horizontal variations are $$\begin{array}{l}
\dfrac{1}{4}
\left[
\left((P(0,1)-P(0,0))-(P(1,1)-P(1,0))\right)+ \right.\\
\quad \left((P(-1,1)-P(-1,0))-(P(0,1)-P(0,0))\right)+\\
\quad \left((P(0,0)-P(0,-1))-(P(1,0)-P(1,-1))\right)+\\
\quad \left. \left((P(-1,0)-P(-1,-1))-(P(0,0)-P(0,-1))\right)
\right] \\
=\dfrac{1}{4}
\left[ P(1,-1) -P(1,1) - P(-1,-1) + P(-1,1)\right].
\end{array}$$ which is $\textit{Ky}_{xy}''$.
Let us now consider any number $n$, $1 \le n \le N$. Let us first investigate the vertical variations related to the horizontal vector ${\overrightarrow{P_{0,0}P_{0,1}}}$ (respectively ${\overrightarrow{P_{0,-1}P_{0,0}}}$) of length 1 that starts from (resp. that points to) $(0,0)$. As with the case $n=1$, there are 2 new vectors of length 1, namely ${\overrightarrow{P_{n,0}P_{n,1}}}$ and ${\overrightarrow{P_{-n,0}P_{-n,1}}}$ (resp. ${\overrightarrow{P_{n,-1}P_{n,0}}}$, and ${\overrightarrow{P_{-n,-1}P_{-n,0}}}$) that are vertically aligned with ${\overrightarrow{P_{0,0}P_{0,1}}}$ (resp. with ${\overrightarrow{P_{0,-1}P_{0,0}}}$).
The vertical variation is now equal to $n$. Following the case where $n$ is 1 to compute the average variation, the coefficients of the first and last line around the central vertical line are thus from left to right: $\dfrac{1}{4n}$, $\dfrac{-1}{4n}$, $\dfrac{-1}{4n}$, and $\dfrac{1}{4n}$.
Cases are similar with vectors ${\overrightarrow{P_{0,0}P_{0,1}}}$, … ${\overrightarrow{P_{0,0}P_{0,n}}}$ which respectively lead to coefficients $-\dfrac{1}{4 \times 2n}$, …, $-\dfrac{1}{4 \times n.n}$, and the proof is omitted. Finally, let us consider the vector ${\overrightarrow{P_{0,0}P_{0,1}}}$ and its vertical variations when $\delta y$ is $n-1$. As in the case where $n=1$, we thus obtain the coefficients $\dfrac{1}{4 \times (n-1)n}$ and $-\dfrac{1}{4 \times (n-1)n}$ (resp. $-\dfrac{1}{4 \times (n-1)n}$ and $\dfrac{1}{4 \times (n-1)n}$) in the second line (resp. in the penultimate line) since the vector has length $n$ and $\delta y$ is $n-1$. Coefficient in the other lines are similarly obtained and the proof is thus omitted.
We are then left to compute an approximation of the partial second order derivatives $\dfrac{\partial^2 P}{\partial x^2}$, $\dfrac{\partial^2 P}{\partial y^2}$, and $\dfrac{\partial^2 P}{\partial x \partial y}$ with the kernels, $\textit{Ky}_{x^2}''$, $\textit{Ky}_{y^2}''$, and $\textit{Ky}_{xy}''$ respectively. However, the size of each of these kernels is varying from $3\times3$ to $(2N+1)\times (2N+1)$. Let us explain the approach on the former partial derivative. The other can be immediately deduced.
Since the objective is to detect large variations, the second order derivative is approximated as the maximum of the approximations. More formally, let $n$, $1 \le n \le N$, be an integer number and $\dfrac{\partial^2 P}{\partial x^2}_n$ be the result of applying the Kernel $\textit{Ky}_{x^2}''$ of size $(2n+1)\times (2n+1)$. The derivative $\dfrac{\partial^2 P}{\partial x^2}$ is defined by
$$\dfrac{\partial^2 P}{\partial x^2}
= \max \left\{
\abs{\dfrac{\partial^2 P}{\partial x^2}_1}, \dots, \abs{\dfrac{\partial^2 P}{\partial x^2}_N}
\right \}.
\label{eq:d2p_dx2}$$
The same iterative approach is applied to compute approximations of $\dfrac{\partial^2 P}{\partial y \partial x}$ and of $\dfrac{\partial^2 P}{\partial y^2}$. Next section studies the suitability of approximating second order derivatives when considering an image as a polynomial.
Polynomial Interpolation of Images for Hessian Matrix Computation {#sec:poly}
=================================================================
Let $P(x,y)$ be the discrete value of the pixel $(x,y)$ in the image. Let $n$, $1 \le n \le N$, be an integer such that the objective is to find a polynomial interpolation on the $(2n+1)\times(2n+1)$ window where the central pixel has index $(0,0)$. There exists an unique polynomial $L : \R\times \R \to \R$ of degree $(2n+1)\times(2n+1)$ defined such that $L(x,y)=P(x,y)$ for each pixel $(x,y)$ in this window. Such a polynomial is defined by $$\begin{array}{l}
L(x,y) =
\sum_{i=-n}^{n}
\sum_{j=-n}^{n} \\
\quad P(i,j)
\left(
\prod_{\stackrel{-n\leq j'\leq n}{j'\neq j}}
\frac{x-j'}{i-j'}
\right)
\left(
\prod_{\stackrel{-n\leq i'\leq n}{i'\neq i}}
\frac{x-i'}{i-i'}
\right)
\end{array}$$
It is not hard to prove that the first order horizontal derivative of the polynomial $L(x,y)$ is $$\begin{array}{l}
\dfrac{\partial L}{\partial x} =
\sum_{i=-n}^{n}
\sum_{j=-n}^{n}
P(i,j)
\left(
\prod_{\stackrel{-n\leq j'\leq n}{j'\neq j}}
\frac{y-j'}{j-j'}
\right)\\
\quad
\left(
\sum_{\stackrel{-n\leq i'\leq n}{i'\neq i}}
\frac{1}{i-i'}
\prod_{\stackrel{-n\leq i''\leq n}{i''\neq i,i'}}
\frac{x-i''}{i-i''}
\right)
\end{array}$$ and thus to deduce that the second order ones are
$$\begin{array}{l}
\dfrac{\partial^2 L}{\partial x^2} =
\sum_{i=-n}^{n}
\sum_{j=-n}^{n}
P(i,j)
\left(
\prod_{\stackrel{-n\leq j'\leq n}{j'\neq j}}
\frac{y-j'}{j-j'}
\right)\\
\quad
\left(
\sum_{\stackrel{-n\leq i'\leq n}{i'\neq i}}
\frac{1}{i-i'}
\sum_{\stackrel{-n\leq i''\leq n}{i''\neq i,i'}}
\frac{1}{i-i''}
\prod_{\stackrel{-n\leq i'''\leq n}{i'''\neq i,i',i''}}
\frac{x-i'''}{i-i'''}
\right)
\end{array}
\label{eq:deriv:poly:x2}$$
$$\begin{array}{l}
\dfrac{\partial^2 L}{\partial y \partial x} =
\sum_{i=-n}^{n}
P(i,j) \\
\quad
\left(
\sum_{\stackrel{-n\leq j'\leq n}{j'\neq j}}
\frac{1}{j-j'}
\prod_{\stackrel{-n\leq j''\leq n}{j''\neq j, j'}}
\frac{y-j''}{j-j''}
\right)\\
\quad
\left(
\sum_{\stackrel{-n\leq i'\leq n}{i'\neq i}}
\frac{1}{i-i'}
\prod_{\stackrel{-n\leq i''\leq n}{i''\neq i, i'}}
\frac{x-i''}{i-i''}
\right)
\end{array}
\label{eq:deriv:poly:yx}$$
These second order derivatives are computed for each moving window and are associated to the central pixel, *i.e.*, to the pixel $(0,0)$ inside this one.
Let us first simplify $\dfrac{\partial^2 L}{\partial x^2}$ when $(x,y)=(0,0)$ defined in Equation (\[eq:deriv:poly:x2\]). If $j$ is not null, the index $j'$ is going to be null and the product $\left(
\prod_{\stackrel{-n\leq j'\leq n}{j'\neq j}}
\frac{-j'}{j-j'}
\right)$ is null too. In this equation, we thus only consider $j=0$. It is obvious that the product indexed with $j'$ is thus equal to 1. This equation can thus be simplified in:
$$\begin{array}{l}
\dfrac{\partial^2 L}{\partial x^2} =
\sum_{i=-n}^{n}
P(i,0)\\
\quad
\left(
\sum_{\stackrel{-n\leq i'\leq n}{i'\neq i}}
\frac{1}{i-i'}
\sum_{\stackrel{-n\leq i''\leq n}{i''\neq i,i'}}
\frac{1}{i-i''}
\prod_{\stackrel{-n\leq i'''\leq n}{i'''\neq i,i',i''}}
\frac{i'''}{i'''-i}
\right)
\end{array}
\label{eq:deriv:poly:x2:simpl}$$
and then in:
$$\begin{array}{l}
\dfrac{\partial^2 L}{\partial x^2} =
\sum_{i=-n}^{n}
P(i,0)\\
\quad
\left(
\sum_{\stackrel{-n\leq i' < i'' \le n}{i',i''\neq i}}
\frac{2}{(i-i')(i-i'')}
\prod_{\stackrel{-n\leq i'''\leq n}{i'''\neq i,i',i''}}
\frac{i'''}{i'''-i}
\right).
\end{array}
\label{eq:deriv:poly:x2:simpl:2}$$
From this equation, the kernel allowing to evaluate horizontal second order derivatives can be computed for any $n$. It is further denoted as $Ko''_{x^2}$. Instances of such matrix when $n=2$, $3$, and $4$ are given in Table \[table:sod:hori:poly\].
$n$ $Ko''_{x^2}$
----- -------------------------------------------------------------------------------------------------------------------------------------------------------------
$2$ $\left[\dfrac{-1}{12}, \dfrac{4}{3} , \dfrac{-5}{2}, \dfrac{4}{3} \dfrac{-1}{12}\right]$
$3$ $\left[\dfrac{1}{90}, \dfrac{-3}{20}, \dfrac{3}{2}, \dfrac{-49}{18}, \dfrac{3}{2}, \dfrac{-3}{20}, \dfrac{1}{90}\right]$
$4$ $\left[\dfrac{-1}{560}, \dfrac{8}{315}, \dfrac{-1}{5}, \dfrac{8}{5}, \dfrac{-205}{72}, \dfrac{8}{5}, \dfrac{-1}{5}, \dfrac{8}{315}, \dfrac{-1}{560}\right]$
: Kernels $Ko''_{x^2}$ for second order horizontal derivatives induced by polynomial interpolation
\[table:sod:hori:poly\]
-------------------------------------------------------------------------------
$n$ $Ko''_{xy}$
----- -------------------------------------------------------------------------
2 $
\begin{bmatrix}
\dfrac{1}{4} & 0 & \dfrac{-1}{4}\\
0 & 0 &0\\
\dfrac{-1}{4} & 0 & \dfrac{1}{4}\\
\end{bmatrix}
$
3 $\begin{bmatrix}
\dfrac{1}{144} & \dfrac{-1}{18} & 0 & \dfrac{1}{18} & \dfrac{-1}{144}\\
\dfrac{-1}{18} & \dfrac{4}{9} & 0 & \dfrac{-4}{9} & \dfrac{1}{18}\\
0 & 0 & 0 & 0 &0\\
\dfrac{1}{18} & \dfrac{-4}{9} & 0 & \dfrac{4}{9} & \dfrac{-1}{18}\\
\dfrac{-1}{144} & \dfrac{1}{18} & 0 & \dfrac{-1}{18} & \dfrac{1}{144}
\end{bmatrix}
$
-------------------------------------------------------------------------------
: Kernels for second order diagonal derivatives induced by polynomial interpolation \[table:sod:diag:poly\]
From Equation (\[eq:deriv:poly:yx\]), kernels allowing to evaluate diagonal second order derivatives (*i.e.*, $\dfrac{\partial^2 L}{\partial y \partial x}$) are computed. They are denoted as $Ko''_{xy}$. Table \[table:sod:diag:poly\] gives two examples of them when $n=1$ and $n=2$. Notice that for $n=1$, the kernel $Ko''_{xy}$ is equal to $Kc''_{xy}$.
Distortion Cost {#sec:distortion}
===============
The distortion function has to associate to each pixel $(i,j)$ the cost $\rho_{ij}$ of its modification by $\pm 1$.
The objective is to map a small value to a pixel when all its second order derivatives are high and a large value otherwise. In WOW and UNIWARD the distortion function is based on the Hölder norm with $$\rho_{ij}^w =
\left(
\abs{\xi_{ij}^h}^{p} +
\abs{\xi_{ij}^v}^{p} +
\abs{\xi_{ij}^d}^{p}
\right)^{-\frac{1}{p}}$$ where $p$ is a negative number and $\xi_{ij}^h$ (resp. $\xi_{ij}^v$ and $\xi_{ij}^d$) represents the horizontal (resp. vertical and diagonal) suitability. A small suitability in one direction means an inaccurate position to embed a message.
We propose here to adapt such a distortion cost as follows: $$\rho_{ij} =
\left(
\abs{\dfrac{\partial^2 P}{\partial x^2}(i,j)} +
\abs{\dfrac{\partial^2 P}{\partial y^2}(i,j)} +
\abs{\dfrac{\partial^2 P}{\partial y \partial x}(i,j)}
\right)^{-\frac{1}{p}}$$ It is not hard to check that such a function has large value when at least one of its derivatives is null. Otherwise, the larger the derivatives are, the smaller the returned value is.
Scheme Stego. content Changes with cover
--------------------- ------------------- ------------------------
$Ky$ based approach  
$Ko$ based approach  
Experiments {#sec:experiments}
===========
First of all, the whole steganographic approach code is available online[^1].
Figure \[fig:oneimage\] presents the results of embedding data in a cover image from the BOSS contest database [@Boss10] with respect to the two second order derivative schemes presented in this work. The $Ky$ based approach (resp. the $Ko$ based one) corresponds to the scheme detailed in Section \[sec:second\] (resp. in Section \[sec:poly\]). The payload $\alpha$ is set to 0.4 and kernels are computed with $N=4$. The central column outputs the embedding result whereas the right one displays differences between the cover image and the stego one. It can be observed that pixels in smooth area (the sky, the external access steps) and pixels in clean edges (the columns, the step borders) are not modified by the approach. On the contrary, an unpredictable area (a monument for example) concentrates pixel changes.
Choice of parameters
--------------------
The two methods proposed in Section \[sec:second\] and in Section \[sec:poly\] are based on kernels of size up to $(2N+1)\times(2N+1)$. This section aims at finding the value of the $N$ parameter that maximizes the security level. For each approach, we have built 1,000 stego images with $N=2$, $4$, $6$, $8$, $10$, $12$, and $14$ where the covers belong to the BOSS contest database. This set contains 10,000 grayscale $512\times 512$ images in a RAW format. The security of the approach has been evaluated thanks to the Ensemble Classifier [@DBLP:journals/tifs/KodovskyFH12] based steganalyser, which is considered as a state of the art steganalyser tool. This steganalysis process embeds the rich model (SRM) features [@DBLP:journals/tifs/FridrichK12] of size 34,671. For a payload $\alpha$, either equal to $0.1$ or to $0.4$, average testing errors (expressed in percentages) have been studied and are summarized in Table \[table:choice:parameter\].
------------------------ ------- ------ ------ ------ ------ ------ -------- --------
$2$ $4$ $6$ $8$ $10$ $12$ $14$
Average testing *0.1* 39 40.2 39.7 39.8 40.1 $39.9$ $39.8$
error for Kernel $K_y$ *0.4* 15 18.8 19.1 19.0 18.6 18.7 18.7
Average testing *0.1* 35.2 36.6 36.7 36.6 37.1 37.2 37.2
error for Kernel $K_o$ *0.4* 5.2 6.8 7.5 7.9 8.1 8.2 7.6
------------------------ ------- ------ ------ ------ ------ ------ -------- --------
: Average Testing Errors with respect to the the Kernel Size
\[table:choice:parameter\]
Thanks to these experiments, we observe that the size $N=4$ (respectively $N=12$) obtains sufficiently large average testing errors for the $Ky$ based approach (resp. for the $Ko$ based one). In what follows, these values are retained for these two methods.
Security Evaluation
-------------------
As in the previous section, the BOSS contest database has been retained. To achieve a complete comparison with other steganographic tools, the whole database of 10,000 images has been used. Ensemble Classifier with SRM features is again used to evaluate the security of the approach.
We have chosen 4 different payloads, 0.1, 0.2, 0.3, and 0.4, as in many steganographic evaluations. Three values are systematically given for each experiment: the area under the ROC curve (AUC), the average testing error (ATE), and the OOB error (OOB).
All the results are summarized in Table \[table:experiments:summary\]. Let us analyse these experimental results. The security approach is often lower than those observed with state of the art tools: for instance with payload $\alpha=0.1$, the most secure approach is WOW with an average testing error equal to 0.43 whereas our approach reaches 0.38. However these results are promising and for two reasons. First, our approaches give more resistance towards Ensemble Classifier (contrary to HUGO) for large payloads. Secondly, without any optimisation, our approach is not so far from state of the art steganographic tools. Finally, we explain the lack of security of the $Ko$ based approach with large payloads as follows: second order derivatives are indeed directly extracted from polynomial interpolation. This easy construction however induces large variations between the polynomial $L$ and the pixel function $P$.
Payload AUC ATE OOB
------------------------- --------- -------- -------- --------
[WOW]{} 0.1 0.6501 0.4304 0.3974
0.2 0.7583 0.3613 0.3169
0.3 0.8355 0.2982 0.2488
0.4 0.8876 0.2449 0.1978
[SUNIWARD]{} 0.1 0.6542 0.4212 0.3972
0.2 0.7607 0.3493 0.3170
0.3 0.8390 0.2863 0.2511
0.4 0.8916 0.2319 0.1977
[MVG]{} 0.1 0.6340 0.4310 0.4124
0.2 0.7271 0.3726 0.3399
0.3 0.7962 0.3185 0.2858
0.4 0.8486 0.2719 0.2353
[HUGO]{} 0.1 0.6967 0.3982 0.3626
0.2 0.8012 0.3197 0.2847
0.3 0.8720 0.2557 0.2212
0.4 0.9517 0.1472 0.1230
[$Ky$ based approach]{} 0.1 0.7378 0.3768 0.3306
0.2 0.8568 0.2839 0.2408
0.3 0.9176 0.2156 0.1710
0.4 0.9473 0.1638 0.1324
[$Ko$ based approach]{} 0.1 0.6831 0.3696 0.3450
0.2 0.8524 0.1302 0.2408
0.3 0.9132 0.1023 0.1045
0.4 0.9890 0.0880 0.0570
: Summary of experiments[]{data-label="table:experiments:summary"}
Conclusion
==========
The first contribution of this paper is to propose of a distortion function which is based on second order derivatives. These partial derivatives allow to accurately compute the level curves and thus to look favorably on pixels without clean level curves. Two approaches to build these derivatives have been proposed. The first one is based on revisiting kernels usually embedded in edge detection algorithms. The second one is based on the polynomial approximation of the bitmap image. These two methods have been completely implemented. The first experiments have shown that the security level is slightly inferior the one of the most stringent approaches. These first promising results encourage us to deeply investigate this research direction.
Future works aiming at improving the security of this proposal are planned as follows. The authors want first to focus on other approaches to provide second order derivatives with larger discrimination power. Then, the objective will be to deeply investigate whether the Hölder norm is optimal when the objective is to avoid null second order derivatives, and to give priority to the largest second order values.$\rightarrow$
{#section .unnumbered}
This work is partially funded by the Labex ACTION program (contract ANR-11-LABX-01-01). Computations presented in this article were realised on the supercomputing facilities provided by the Mésocentre de calcul de Franche-Comté.
[^1]: <https://github.com/stego-content/SOS>
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
The classical Fischer decomposition of spinor-valued polynomials is a key result on solutions of the Dirac equation in the Euclidean space $\bR^m.$ As is well-known, it can be understood as an irreducible decomposition with respect to the so-called $L$-action of the Pin group $Pin(m).$ But, on Clifford algebra valued polynomials, we can consider also the $H$-action of $Pin(m).$ In this paper, the corresponding Fischer decomposition for the $H$-action is obtained. It turns out that, in this case, basic building blocks are the spaces of homogeneous solutions to the Hodge-de Rham system. Moreover, it is shown that the Fischer decomposition for the $H$-action can be viewed even as a refinement of the classical one.
[**Keywords:**]{} Fischer decomposition, Clifford analysis, Hodge-de Rham equation, spherical monogenics
[**Mathematics Subject Classification (2000)**]{} 30G35, 58A10
author:
- 'Richard Delanghe, Roman L'' avička and Vladimír Souček'
title: 'The Fischer decomposition for Hodge-de Rham systems in Euclidean spaces'
---
Introduction
============
Clifford analysis is, from the very beginning, considered to be a refinement of harmonic analysis for Clifford algebra (or spinor) valued functions. The perfect description of this statement is the Fischer decomposition of spinor-valued polynomials.
Let us first recall the Fischer decomposition of the space $\cP$ of complex-valued polynomials in the Euclidean space $\bR^m.$ Denote by $\cH_k$ the space of $k$-homogeneous harmonic polynomials in $\bR^m.$ Then, under a natural action of the orthogonal group $O(m),$ the space $\cP$ has an irreducible (not multiplicity free) decomposition $$\label{Fischer}
\cP=\bigoplus_{k=0}^{\infty}\bigoplus_{p=0}^{\infty}r^{2p}\cH_k$$ where $r^2=x_1^2+\cdots +x_m^2$ for the vector variable $\underline{x}=(x_1,\ldots,x_m)\in\bR^m.$
For spinor-valued polynomials, there is a refinement of this decomposition. Let $\bC_m$ be the complex Clifford algebra generated by vectors of the standard basis $(e_1,\ldots,e_m)$ of $\bR^m.$ Recall that the $L$-action of the Pin group $Pin(m)$ on functions $f:\bR^m\to\bC_m$ is defined by $$\label{Laction}
[L(s)(f)](\underline{x}) = s\,f(s^{-1}\underline{x}s),\ s\in Pin(m)\text{\ \ and\ \ }\underline{x}=(x_1,\ldots,x_m)\in\bR^m.$$ Denote by $\bS$ a basic spinor representation for $Pin(m).$ As is well-known, the spinor space $\bS$ can be realized inside the Clifford algebra $\bC_m.$ Let us remark that, under the $L$-action, the space $\cP(\bS)=\cP\otimes\bS$ of spinor valued polynomials forms a $Pin(m)$-module. Denote by $\cM_k(\bS)$ the space of $k$-homogeneous polynomials $P\in\cP(\bS)$ which are (left) monogenic, i.e., which satisfy the Dirac equation $\underline{D}P=0$ where $$\underline{D}=e_1\pa_{x_1}+\cdots+e_m\pa_{x_m}.$$ Now we are ready to state the Fischer decomposition (sometimes called also Almansi decomposition) for this case. Namely, under the $L$-action, the space $\cP(\bS)$ has an irreducible (not multiplicity free) decomposition $$\label{FischerPin}
\cP(\bS)=\bigoplus_{k=0}^{\infty}\bigoplus_{p=0}^{\infty}\underline{x}^p\cM_k(\bS)$$ with $\underline{x}=e_1x_1+\cdots+e_mx_m.$ See [@BSES; @MR; @rya]. As $\cH_k\otimes\bS=\cM_k(\bS)\oplus\underline{x}\cM_{k-1}(\bS)$ and $\underline{x}^2=-r^2,$ it is easy to see that (\[FischerPin\]) is a real refinement of (\[Fischer\]).
The main aim of the underlying paper is to show that there exists a natural further refinement of the monogenic Fischer decomposition . It is quite surprising that such a finer Fischer decomposition was not described earlier. It was the study of special solutions of the Dirac equation which led to such a refinement. By special solutions we mean just solutions having their values in a chosen subspace $V$ of the Clifford algebra $\bC_m$. There are a lot of possibilities for a choice of $V,$ but it is clearly preferable to choose the subspace $V$ having some special properties.
Typical examples are solutions of the Dirac equation having values in spinor subspaces of the Clifford algebra. This case is closely related to the $L$-action . Indeed, it is well-known that the Clifford algebra $\bC_m$, considered as a $Pin(m)$-module by left multiplication, decomposes into many equivalent spinor submodules. Moreover, for every choice of the spinor submodule, the $Pin(m)$-module of spinor-valued solutions has quite analogous properties.
Another interesting example of special solutions of the Dirac equation is given by the so-called generalized Moisil-Théodoresco system (GMT system for short). A lot of interest has recently been paid to GMT systems (see [@DLS] and the references there). In this case, the space $V$ is supposed to be invariant under another (both side) action of the Pin group, namely the so-called $H$-action. The $H$-action on Clifford algebra valued functions $f:\bR^m\to\bC_m$ is given by $$\label{Haction}
[H(s)(f)](\underline{x}) = s\,f(s^{-1}\underline{x}s)s^{-1},\ s\in Pin(m)\text{\ \ and\ \ }\underline{x}\in\bR^m.$$ In what follows, we shall use the language of differential forms. Indeed, following [@BDS], we identify naturally the Clifford algebra $\bC_m$ with the Grassmann algebra $\La^*(\bC^m)$ and we study the space $\cP^*=\cP\otimes\La^*(\bC^m)$ of polynomial differential forms instead of Clifford algebra valued polynomials. Then the $H$-action translates into a natural action of the orthogonal group $O(m)$ on $\cP^*$ and the Dirac operator $\underline{D}$ corresponds to the operator $d+d^*.$ Here $d$ and $d^*$ are, respectively, the standard de Rham differential and its adjoint (see below). As an $O(m)$-module, the Grassmann algebra $\La^*(\bC^m)$ has a multiplicity free irreducible decomposition $$\Lambda^*(\bC^m)=\bigoplus_{s=0}^m\Lambda^s(\bC^m)$$ with $\Lambda^s(\bC^m)$ being the space of $s$-vectors over $\bC^m.$ A GMT system is then defined as the homogeneous system obtained by restricting the operator $d+d^*$ to functions having values in the space $$V=\bigoplus_{s\in S}\Lambda^s(\bC^m)$$ for some (suitable) subset $S\subset\{0,1,\ldots,m\},$ i.e. $$(d+d^*)P=0\mbox{\ \ \ for $V$-valued $P.$}$$ In particular, for $V=\Lambda^s(\bC^m),$ the corresponding GMT system coincide with the so-called Hodge-de Rham system $$\label{Hodge}
d P=0,\, d^* P=0.$$
Various versions of GMT systems of PDE’s were studied for a long time (in particular in low dimensions) and they were used in many different applications. Applications in numerical analysis and engineering sciences can be found in [@gs1; @gs2]. In a review paper [@spr], you can find various generalizations of the well-known Hodge-de Rham decomposition of smooth 1-forms, including decompositions for quaternionic and Clifford algebra valued functions and the Almansi (i.e., monogenic Fischer) decomposition. For applications in theory of electromagnetic fields we can refer to [@S_Maxwell].
The Fischer decomposition always played a key role in Clifford analysis. In [@lav_Fischer], the Fischer decomposition for the $H$-action has been recently applied to inframonogenic functions introduced in [@MPS]. Moreover, in [@lav_Fischer], the obtained results for the space $\cP^*$ are translated back into the framework of Clifford analysis. For yet another application, we can refer to [@lavSL3].
Recently, the Fischer decomposition (together with the Cauchy-Kovalevskaya extension) was used systematically for construction of orthogonal bases in the spaces of homogeneous polynomial solutions. In the classical Clifford analysis, it has a quite long history (see [@BGLS] for historical account, various results can be found in [@Bock2010c; @Bock2010a; @Bock2009; @BG; @CacGueMal; @CacGueBock; @BockCacGue; @cac; @CM06; @CM07; @CM08; @FCM; @FM; @Gurlebeck1999; @NGue2009; @lavSL2; @step2; @Malonek1987; @MS; @mor09; @som; @van]). Analogous results in Hermitean Clifford analysis are described in [@ckH; @kerH; @GTinH; @GT2H]. Finally, in [@DLS4], the Fischer decomposition for the $H$-action plays a key role in constructing orthogonal bases not only for the spaces $H^s_k$ of solutions to the Hodge-de Rham system but even for the spaces of homogeneous solutions of an arbitrary generalized Moisil-Théodoresco system.
In this paper, we will establish (using results from [@hom]) a form of the Fischer decomposition appropriate for the $H$-action. The theory of the Howe duality developed in [@how] shows us that we may expect a further refinement of the monogenic Fischer decomposition (\[FischerPin\]), see [@DLS3] for details. Indeed, this is visible from the form of invariants contained in the polynomial spaces considered. For scalar valued functions, invariant polynomials are generated by powers of $r^2,$ and the basic equation is the Laplace equation. For spinor-valued polynomials with the left action, we have to look for invariants with values in the space of endomorphisms of the spinor space (which is, basically, the corresponding Clifford algebra) and there is just a new invariant $\underline{x},$ acting as a refinement of $r^2.$ The corresponding basic equation is the Dirac equation.
In the case of the both side action, we deal with the space $\cP^*$ of $\La^*(\bC^m)$-valued polynomials. The space of invariants with values in the space of endomorphisms of the Grassmann algebra $\La^*(\bC^m)$ is now much richer. It is generated by two elements ${x}$ and ${x}^*$ which correspond to the differential operators $d^*$ and $d$ by the Fischer duality (see below). Consequently, the corresponding basic system of equations is the Hodge-de Rham system (\[Hodge\]) and the space of invariants consists of polynomials in $x$ and $x^*.$ Actually, due to the fact that ${x}^2=({x}^*)^2=0,$ such invariants are generated by the set $$\label{Omega}
\Omega=\{1, x, x^*, xx^*, x^*x, xx^*x, x^*xx^*, \ldots\}.$$ Moreover, denote by $\cP^*_k$ the space of $k$-homogeneous polynomial forms $P\in\cP^*$ and by $H^s_k$ the space of $\La^s(\bC^m)$-valued polynomial forms $P\in\cP^*_k$ which satisfy the Hodge-de Rham system (\[Hodge\]). Then, using results from [@hom], we shall deduce in Section 2 the corresponding Fischer decomposition for the $H$-action.
\[tFischer\] The space $\cP^*=\cP\otimes\La^*(\bC^m)$ decomposes as follows: $$\label{etFischer}
\cP^*=\cP^*_{(0,0)}\oplus\left(\bigoplus_{s=1}^{m-1}\bigoplus_{k=0}^{\infty}\cP^*_{(s,k)}\right)\oplus\cP^*_{(m,0)}\text{\ \ \ with\ \ \ }
\cP^*_{(s,k)}=\bigoplus_{w\in\Omega}wH^s_k.$$ Moreover, in (\[etFischer\]), all $O(m)$-modules $H^s_k$ are non-trivial, irreducible and mutually inequivalent and all $\cP^*_{(s,k)}$ are corresponding $O(m)$-isotypic components of $\cP^*.$
Now we show that the Fischer decomposition of the space $\cP^*$ given in Theorem \[tFischer\] is a refinement of the monogenic Fischer decomposition (\[FischerPin\]). Indeed, when we identify the Clifford algebra $\bC_m$ with the Grassmann algebra $\La^*(\bC^m)$ on the space $\cP^*$ we know that $$\underline{D}=d+d^*\text{\ \ \ and\ \ \ }-\underline{x}=x+x^*.$$ Consequently, the space of spherical monogenics of order $k$ is given by $$\cM_k=\{P\in\cP^*_k;\ (d+d^*)P=0\}.$$ Moreover, recall that the Laplace operator $\Delta$ is given by $\Delta=\sum_{j=1}^m\pa^2_{x_j}$ and put $$\Ker_k\Delta=\{P\in\cP^*_k;\ \Delta P=0\}.$$
As we mentioned before, under the $L$-action, $\bC_m\simeq\La^*(\bC^m)$ decomposes into many copies of basic spinor representations $\bS$ of $Pin(m)$ and so the whole space $\cM_k$ of spherical monogenics is not irreducible. Indeed, under the $L$-action, the space $\cM_k$ decomposes into many copies of irreducible modules $\cM_k(\bS)$. In particular, we have thus that $\Ker_k\Delta=\cM_k\oplus(x+x^*)\cM_{k-1}$ and, by (\[Fischer\]), we get easily the following decomposition of the space $\cP^*$ $$\label{FischerPin*}
\cP^*=\bigoplus_{k=0}^{\infty}\bigoplus_{p=0}^{\infty}r^{2p}(\cM_k\oplus(x+x^*)\cM_{k-1}).$$ In an obvious sense, the decomposition (\[FischerPin\*\]) is equivalent to (\[FischerPin\]).
In Section 3, we shall prove the following theorem which tells us that, under the $H$-action, the spaces $\cM_k$ and $(x+x^*)\cM_{k-1}$ decompose again into many irreducible pieces but, in this case, these pieces are not equivalent and they have a different representation character.
\[tdecompM\] For $k\geq 1,$ the following statements hold:
- $\cM_k=\left(\bigoplus_{s=0}^m H^s_k\right)\oplus\left(\bigoplus_{s=1}^{m-1}M_{s,k}\right)$
- where $M_{s,k}=[(k-1+m-s)x^*-(k-1+s)x]H^s_{k-1}.$
- $(x+x^*)\cM_{k-1}=\left(\bigoplus_{s=0}^m(x+x^*)H^s_{k-1}\right)\oplus\left(\bigoplus_{s=1}^{m-1}W^s_k\right)$
- where $W^s_k=[(k-2+m-s)xx^*-(k-2+s)x^*x]H^s_{k-2}.$
Using Theorem \[tdecompM\], we get from the decomposition directly a finer decomposition of the space $\cP^*$ which is irreducible with respect to the $H$-action.
The results stated in Theorems \[tFischer\] and \[tdecompM\] remain valid also for real valued polynomial forms, that is, in the case when the complex Grassmann algebra $\La^*(\bC^m)$ is replaced with the real one $\La^*(\bR^m).$ Indeed, it is sufficient to realize that irreducible $O(m)$-representations $\La^s(\bC^m)$ are all of real type, see [@GM p. 163].
A proof of the Fischer decomposition for the $H$-action {#stFischer}
=======================================================
In this section, we give a proof of Theorem \[tFischer\] stated in Introduction. Let $\cP_k$ stand for the space of $k$-homogeneous (complex-valued) polynomials of $\cP$ and let $\cP^s_k=\cP_k\otimes\Lambda^s(\bC^m).$ Then it is easy to see that $$\label{decompP}
\cP^*=\bigoplus_{s=0}^m\bigoplus_{k=0}^{\infty}\cP^s_k.$$ Let us remark that a polynomial form $P$ belongs to $\cP^s_k$ if and only if $$\label{skform}
P=\sum_I P_I\; dx_I$$ where the sum is taken over all finite strictly increasing sequences $I=\{i_j\}_{j=1}^s$ of numbers of the set $\{1,\cdots, m\},$ $P_I\in\cP_k$ and $dx_I=dx_{i_1}\wedge\cdots\wedge dx_{i_s}.$ The contraction $dx_j\krat$ is defined as $$dx_j\krat\; dx_I=\sum_{k=1}^s(-1)^{k-1}\delta_{ji_k}dx_{I\setminus\{i_k\}}\text{ and }
dx_j\krat\; P=\sum_I P_I\; dx_j\krat\; dx_I$$ for a polynomial form $P.$ Then we have that $$\label{dd^*}
d=\sum_{j=1}^m\; \pa_{x_j} dx_j\wedge\text{\ \ \ and\ \ \ }
d^*=-\sum_{j=1}^m\;\pa_{x_j} dx_j\krat,$$ $$\label{xx^*}
x=-\sum_{j=1}^m x_j\; dx_j\wedge\text{\ \ \ and\ \ \ }
x^*=\sum_{j=1}^m x_j\; dx_j\krat.$$ It is easy to see that $d,$ $d^*,$ $x$ and $x^*$ are $O(m)$-invariant operators on the space $\cP^*.$
Now we describe explicitly an irreducible decomposition of $O(m)$-modules $$\Ker_k^s\;\Delta=\{P\in\cP_k^s;\ \Delta P=0\}.$$ The following key result is obtained in [@hom].
[\[thforms\]]{} Given $0\leq s\leq m$ and $k\in\NN_0,$ we have that $$\Ker_k^s\Delta= H^s_k\oplus U^s_k\oplus V^s_k\oplus W^s_k$$ where $H^s_k,$ $U^s_k,$ $V^s_k$ and $W^s_k$ are irreducible $O(m)$-modules with the following properties:
- $H^s_k=\{P\in \cP^s_k;\ dP=0,\ d^*P=0\}$ and $\Ker_0^s\Delta= H^s_0=\cP_0^s.$
- In addition, $H^s_k=\{0\}$ for $s\in\{0,m\}$ and $k\geq 1.$ Otherwise, all $O(m)$ modules $H^s_k$ are non-trivial, irreducible and mutually inequivalent.
- $U^s_k=xH^{s-1}_{k-1}\simeq H^{s-1}_{k-1}$ for $1\leq s\leq m$ and $k\geq 1,$ and $U^s_k=\{0\}$ otherwise.
- $V^s_k=x^*H^{s+1}_{k-1}\simeq H^{s+1}_{k-1}$ for $0\leq s\leq m-1$ and $k\geq 1,$ and $V^s_k=\{0\}$ otherwise.
- $W^s_k=[(k-2+m-s)xx^*-(k-2+s)x^*x]H^s_{k-2}\simeq H^s_{k-2}$\
for $1\leq s\leq m-1$ and $k\geq 2,$ and $W^s_k=\{0\}$ otherwise.
Now we are ready to prove Theorem \[tFischer\].
As $\Ker^s_k\Delta=\cH_k\otimes\La^s(\bC^m)$ the Fischer decomposition (\[Fischer\]) yields $$\cP_k^s=\bigoplus_{p=0}^{[k/2]}r^{2p}\Ker^s_{k-2p}\Delta,$$ where for $y\in\bR,$ $[y]$ denotes the greatest integer not greater than $y.$ Consequently, by Lemma \[thforms\], we get the decomposition $$\label{decompPsk}
\cP_k^s= H^s_k\oplus
\bigoplus_{p=0}^{[k/2]}r^{2p}U^s_{k-2p}\oplus
\bigoplus_{p=0}^{[k/2]}r^{2p}V^s_{k-2p}\oplus
\bigoplus_{p=0}^{[k/2]}r^{2p}Z^s_{k-2p}$$ where $Z^s_k=r^2 H^s_{k-2}\oplus W^s_k.$ Since $r^2=-(xx^*+x^*x)$ Lemma \[thforms\] implies that, for $0\leq s\leq m$ and $k\geq 2,$ $$Z^s_k=(xx^*)H^s_{k-2}\oplus (x^*x)H^s_{k-2}.$$ Moreover, it is easy to see that $$r^{2p}U^s_k=(xx^*)^pxH^{s-1}_{k-1},\ \ \
r^{2p}V^s_k=(x^*x)^px^*H^{s+1}_{k-1}\text{\ \ and}$$ $$r^{2p}Z^s_k=(xx^*)^{p+1}H^s_{k-2}\oplus (x^*x)^{p+1}H^s_{k-2}.$$ Now to complete the proof it suffices to use the decompositions (\[decompP\]) and (\[decompPsk\]).
At the end of this section we collect the well-known relations we need later on. Put, for linear operators $T_1$ and $T_2$ on $\cP^*,$ $\{T_1,T_2\}=T_1T_2+T_2T_1$ and $[T_1,T_2]=T_1T_2-T_2T_1.$ Then we have that (see e.g. [@hom] or [@BDS])
\[lrels\] Let $E$ be the Euler operator and $\hat{E}$ be the skew Euler operator, i.e. $$E=\sum_{j=1}^m x_j\pa_{x_j}\text{\ \ \ and\ \ \ }
\hat{E}=\sum_{j=1}^m(dx_j\;\wedge)(dx_j\krat).$$ Then we have that $EP=kP$ and $\hat{E}P=sP$ for each $P\in\cP^s_k.$
Furthermore, the following relations hold: $$\begin{array}{lll}
\{x,x\}=0, &\{x^*,x^*\}=0, &\{x,x^*\}=-r^2,\medskip\\{}
\{d,d\}=0, &\{d^*,d^*\}=0, &\{d,d^*\}=-\Delta,\medskip\\{}
\{x^*,d\}=E+\hat{E}, &\{x,d^*\}=E-\hat{E}+m, &\{x^*,d^*\}=0=\{x,d\}.
\end{array}$$
Using Lemma \[lrels\], we may give, for example, an explicit description of the projections of the space $\Ker_k^s\Delta$ onto the pieces $H^s_k,$ $U^s_k,$ $V^s_k$ and $W^s_k.$
\[chforms\] Given $0\leq s\leq m$ and $k\in\NN_0,$ put $c_1=k-2+s$ and $c_2=k-2+m-s.$ Furthermore, let $\pi_1,$ $\pi_2,$ $\pi_3$ and $\pi_4$ be the projections of the space $\Ker_k^s\Delta$ onto the subspaces $H^s_k,$ $U^s_k,$ $V^s_k$ and $W^s_k,$ respectively. Then we have that $$%\label{pi4}
\pi_4=\left\{
\begin{array}{ll}
\frac{c_2xx^*-c_1x^*x}{c_1c_2(c_1+c_2+2)}\;dd^* &\text{for\ \ }1\leq s\leq m-1\text{\ and\ }k\geq 2,\medskip\\{}
0, &\text{otherwise.}
\end{array}
\right.$$ Moreover, denoting $\pi=1-\pi_4,$ we have that $$\begin{array}{ll}%\label{pi123}
&\pi_2 =\left\{
\begin{array}{ll}
\frac{1}{c_2+2}\;xd^*\pi &\text{for\ \ }1\leq s\leq m\ \text{\ and\ }k\geq 1,\medskip\\{}
0, &\text{otherwise;}
\end{array}
\right.\medskip\\{}
&\pi_3 =\left\{
\begin{array}{ll}
\frac{1}{c_1+2}\;x^*d\pi &\text{for\ \ }0\leq s\leq m-1\text{\ and\ }k\geq 1,\medskip\\{}
0, &\text{otherwise;}
\end{array}
\right.\medskip\\{}
&\pi_1 =1-\pi_2-\pi_3-\pi_4.
\end{array}$$
Let $P\in\Ker^s_k\Delta$ be given. Then, by Lemma \[thforms\], there are uniquely determined $P_1\in H^s_k,$ $P_2\in H^{s-1}_{k-1},$ $P_3\in H^{s+1}_{k-1}$ and $P_4\in H^s_{k-2}$ such that $$P=P_1+xP_2+x^*P_3+(c_2xx^*-c_1x^*x)P_4.$$ By Lemma \[lrels\], it is easy to see that $$dd^*P=c_1c_2(c_1+c_2+2)P_4=-d^*dP,$$ which easily implies the formula for $\pi_4.$
Moreover, $\pi(P)=P_1+xP_2+x^*P_3.$ By Lemma \[lrels\], we have that $$d^*\pi(P)=d^*xP_2=(c_2+2)P_2\text{\ \ \ and\ \ \ }d\pi(P)=dx^*P_3=(c_1+2)P_3,$$ from which the formulae for the projections $\pi_2$ and $\pi_3$ may be derived.
Decomposition of monogenic polynomial forms
===========================================
In this section, we give a proof of Theorem \[tdecompM\] stated in Introduction. To prove Theorem \[tdecompM\] we need some lemmas.
\[lMsk\] For $1\leq s\leq m-1$ and $k\geq 1,$ we have that $$\left(xH^s_{k-1}\oplus x^*H^s_{k-1}\right)\cap\cM_k=M_{s,k}.$$ Here $M_{s,k}=[(k-1+m-s)x^*-(k-1+s)x]H^s_{k-1}.$
Let $P_1,P_2\in H^s_{k-1}$ and put $P=xP_1+x^*P_2.$ It suffices to show that $(d+d^*)P=0$ if and only if $$P_1=-\frac{k-1+s}{k-1+m-s}P_2.$$ By virtue of Lemma \[lrels\], it is easy to see that $$(d+d^*)P=(E+m-\hat E)P_1+(E+\hat E)P_2=(k-1+m-s)P_1+(k-1+s)P_2,$$ which completes the proof.
\[ldUV\] For $1\leq s\leq m-1$ and $k\geq 1,$ we have that $$xH^s_{k-1}\oplus x^*H^s_{k-1}=(x+x^*)H^s_{k-1}\oplus M_{s,k}.$$
Obvious.
Put$$\tilde\cM_k=\left(\bigoplus_{s=0}^m H^s_k\right)\oplus\left(\bigoplus_{s=1}^{m-1}M_{s,k}\right).$$ Then, by Lemma \[lMsk\], it is easy to see that $\tilde\cM_k\subset\cM_k.$ Moreover, by Lemma \[thforms\], $W^s_k=(x+x^*)M_{s,k-1}.$ Finally, using Lemma \[ldUV\] and Lemma \[thforms\], we obtain that $$\Ker_k\Delta=\tilde\cM_k\oplus(x+x^*)\tilde\cM_{k-1}\subset\cM_k\oplus(x+x^*)\cM_{k-1}=\Ker_k\Delta,$$ which completes the proof.
Acknowledgments {#acknowledgments .unnumbered}
---------------
R. Lávička and V. Souček acknowledge the financial support from the grant GA 201/08/0397. This work is also a part of the research plan MSM 0021620839, which is financed by the Ministry of Education of the Czech Republic.
[99]{} S. Bock, K. Gürlebeck, R. L' avička and V. Souček, The Gelfand-Tsetlin bases for spherical monogenics in dimension 3, preprint. S. Bock, Orthogonal Appell bases in dimension 2,3 and 4. In Numerical Analysis and Applied Mathematics (T.E. Simos, G. Psihoyios, and Ch. Tsitouras, eds.), AIP Conference Proceedings, vol. 1281. American Institute of Physics: Melville, NY, 2010; 1447–1450. S. Bock, On a three dimensional analogue to the holomorphic $z$-powers: Power series and recurrence formulae, submitted, 2010. S. Bock, Über funktionentheoretische [M]{}ethoden in der räumlichen [E]{}lastizitätstheorie, PhD thesis, Bauhaus-University, Weimar, (url: http://e-pub.uni-weimar.de/frontdoor.php?source\_opus=1503, date: 07.04.2010), 2009. S. Bock and K. Gürlebeck, On a generalized Appell system and monogenic power series, Mathematical Methods in the Applied Sciences 33 (2010), 394–411. F. Brackx, R. Delanghe and F. Sommen, Differential forms and/or multi-vector functions, CUBO 7 (2005), 139-170. F. Brackx, H. De Schepper, D. Eelbode and V. Souček, The Howe dual pair in hermitian Clifford analysis, Rev. Mat. Iberoamericana 26 (2010)(2), 449-479. F. Brackx, H. De Schepper, R. Lávička, V. Souček, *The Cauchy-Kovalevskaya Extension Theorem in Hermitean Clifford Analysis*, preprint. F. Brackx, H. De Schepper, R. Lávička, V. Souček, *Fischer decompositions of kernels of Hermitean Dirac operators*, In: T.E. Simos, G. Psihoyios, Ch. Tsitouras, [*Numerical Analysis and Applied Mathematics*]{}, AIP Conference Proceedings, Rhodes, Greece (2010). F. Brackx, H. De Schepper, R. Lávička, V. Souček, *Gel’fand-Tsetlin procedure for the construction of orthogonal bases in Hermitean Clifford analysis*, In: T.E. Simos, G. Psihoyios, Ch. Tsitouras, [*Numerical Analysis and Applied Mathematics*]{}, AIP Conference Proceedings, Rhodes, Greece (2010). F. Brackx, H. De Schepper, R. Lávička, V. Souček, *Orthogonal basis of Hermitean monogenic polynomials: an explicit construction in complex dimension $2$*, In: T.E. Simos, G. Psihoyios, Ch. Tsitouras (eds.), [*Numerical Analysis and Applied Mathematics*]{}, AIP Conference Proceedings, Rhodes, Greece (2010). I. Cação, K. Gürlebeck, H.R. Malonek, Special monogenic polynomials and $L_2$-approximation. Advances in Applied Clifford Algebras 2001; 11(S2):47–60. I. Cação, K. Gürlebeck, S. Bock. Complete orthonormal systems of spherical monogenics - a constructive approach. In Methods of Complex and Clifford Analysis, Son LH, Tutschke W, Jain S (eds). Proceedings of ICAM, Hanoi, SAS International Publications, 2004. I. Cação, K. Gürlebeck, S. Bock, On derivatives of spherical monogenics. Complex Variables and Elliptic Equations 2006; 51(811):847–869. I. Cação and H. R. Malonek, Remarks on some properties of monogenic polynomials, ICNAAM 2006. International conference on numerical analysis and applied mathematics 2006 (T.E. Simos, G. Psihoyios, and Ch. Tsitouras, eds.), Wiley-VCH, Weinheim, 2006, pp. 596-599. I. Cação and H. R. Malonek, On a complete set of hypercomplex Appell polynomials, Proc. ICNAAM 2008, (T. E. Timos, G. Psihoyios, Ch. Tsitouras, Eds.), AIP Conference Proceedings 1048, 647-650. I. Cação, *Constructive approximation by monogenic polynomials*, PhD thesis, Univ. Aveiro, 2004. R. Delanghe, R. Lávička and V. Souček, On polynomial solutions of generalized Moisil-Théodoresco systems and Hodge-de Rham systems, to appear in Adv. appl. Clifford alg. (arXiv:0908.0842 \[math.CV\], 2009). R. Delanghe, R. Lávička and V. Souček, The Howe duality for Hodge systems, In: Proceedings of 18th International Conference on the Application of Computer Science and Mathematics in Architecture and Civil Engineering (ed. K. Gürlebeck and C. Könke), Bauhaus-Universität Weimar, Weimar, 2009. R. Delanghe, R. Lávička and V. Souček, The Gelfand-Tsetlin bases for Hodge-de Rham systems in Euclidean spaces, preprint. M. I. Falcão, J. F. Cruz and H. R. Malonek, Remarks on the generation of monogenic functions, Proc. of the 17-th International Conference on the Application of Computer Science and Mathematics in Architecture and Civil Engineering, ISSN 1611-4086 (K. Gürlebeck and C. Könke, eds.), Bauhaus-University Weimar, 2006. M. I. Falcão and H. R. Malonek, Generalized exponentials through Appell sets in $\bR^{n+1}$ and Bessel functions, Numerical Analysis and Applied Mathematics (T.E. Simos, G. Psihoyios, and Ch. Tsitouras, eds.), AIP Conference Proceedings, vol. 936, American Institute of Physics, 2007, pp. 750-753 (ISBN: 978-0-7354-0447-2).
L. Frappat, P. Sorba and A. Sciarrino, Dictionary on Lie superalgebras, arXiv:hep-th/960761v1, 1996. J. E. Gilbert and M. A. M. Murray, Clifford Algebras and Dirac Operators in Harmonic Analysis, Cambridge University Press, Cambridge, 1991. R. Goodman, Multiplicity-free spaces and Schur-Weyl-Howe duality, In: Representations of real and $p$-adic groups (E. C. Tan and C. B. Zhu eds.), Lecture note series - Institute for mathematical sciences, Vol. 2, World scientific, Singapore, 2004. R. Goodman and N. Wallach, Representations and invariants of the classical groups, Cambridge University Press, Cambridge, 1998. K. Gürlebeck, W. Sprössig, [*Quaternionic analysis and boundary value problems*]{}, Birkhäuser, Basel, 1990. K. Gürlebeck, W. Sprössig, [*Quaternionic and Clifford calculus for physicists and engineers*]{}, Wiley, Chichester, 1997. K. Gürlebeck and H. R. Malonek, A hypercomplex derivative of monogenic functions in $\mathbb{R}^{n+1}$ and its applications, Complex Variables 39 (1999), 199–228. N. Gürlebeck, On Appell Sets and the Fueter-Sce Mapping, Advances in Applied Clifford Algebras 19 (2009), 51-61. Y. Homma, Spinor-valued and Clifford algebra-valued harmonic polynomials, J. Geom. Phys. 37 (2001), 201-215. R. Howe, Remarks on classical invariant theory, Trans. Am. Math. Soc. 313 (1989) (2), 539-570. R. Lávička, The Fischer Decomposition for the H-action and Its Applications, arXiv:1002.0527v1 \[math.CV\], 2010, to appear. R. Lávička, Canonical bases for sl(2,C)-modules of spherical monogenics in dimension 3, arXiv:1003.5587v2 \[math.CV\], 2010, to appear in Arch. Math. R. Lávička, On the Structure of Monogenic Multi-Vector Valued Polynomials, In: ICNAAM 2009, Rethymno, Crete, Greece, 18-22 September 2009 (eds. T. E. simos, G. Psihoyios and Ch. Tsitouras), AIP Conf. Proc. 1168 (2009)(793), pp. 793-796. R. Lávička, V. Souček, P. Van Lancker, Spherical monogenics: step two branching, preprint. H. R. Malonek, M. I. Falcão, Special monogenic polynomials-properties and applications. In Numerical Analysis and Applied Mathematics (T.E. Simos, G. Psihoyios, and Ch. Tsitouras, eds.), AIP Conference Proceedings, vol. 936. American Institute of Physics: Melville, NY, 2007; 764–767. H. R. Malonek, [Z]{}um [H]{}olomorphiebegriff in höheren [D]{}imensionen, Habilitationsschrift. Pädagogische Hochschule Halle, 1987. H.R. Malonek, D. Peña Peña and F. Sommen, Fischer decomposition by inframonogenic functions, arXiv:0911.0070 \[math.CV\], 2009 (to appear in CUBO). H.R. Malonek and G. Ren, Almansi-type theorems in Clifford analysis, Math. Meth. Appl. Sci. 25 (2002), 1541-1552. I. M. Mitelman and M. V. Shapiro, Differentiation of the Martinelli–Bochner integrals and the notion of hyperderivability. [*Math. Nachr.*]{} 172: (1995), 211–238. J. Morais, Approximation by homogeneous polynomial solutions of the Riesz system in $\bR^3$, PhD thesis, Bauhaus-Univ., Weimar, 2009. J. Ryan, Iterated Dirac operators in $\bC^n,$ Z. Anal. Anwendungen 9 (1990), 385-401. F. Sommen, Spingroups and spherical means III, Rend. Circ. Mat. Palermo (2) Suppl. No 1 (1989), 295-323. W. Sprössig, Quaterionic analysis and Maxwell’s equations, CUBO A mathematical journal 7 (2005) (2), 57-67. W. Sprössig, On Helmholtz decompositions and their generalizations - An overview, Math. Meth. Appl. Sci. 33 (2010), 374-383. P. Van Lancker, Spherical Monogenics: An Algebraic Approach, Adv. appl. Clifford alg. 19 (2009), 467-496.
Richard Delanghe,\
Clifford Research Group, Department of Mathematical Analysis,\
Ghent University, Galglaan 2, B-9000 Gent, Belgium\
email: `[email protected]`
Roman Lávička and Vladimír Souček,\
Mathematical Institute, Charles University,\
Sokolovská 83, 186 75 Praha 8, Czech Republic\
email: `[email protected]` and `[email protected]`
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We found an easy and quick post-learning method named “Icing on the Cake” to enhance a classification performance in deep learning. The method is that we train only the final classifier again after an ordinary training is done.'
author:
- |
Tomohiko Konno[^1] and Michiaki Iwazume\
AI Science Research and Development Promotion Center\
National Institute of Information and Communications Technology, Tokyo Japan
bibliography:
- 'pseudo\_data\_generation.bib'
date: 18th May 2018
title: 'Icing on the Cake: An Easy and Quick Post-Learnig Method You Can Try After Deep Learning'
---
![The sketch of the proposed method.(“Icing on the Cake”). Left: Train a deep neural network as usual. Center: Extract features of input data as estimation from the layer before the final classifier, and then train the final classifier again by the extracted features. Right: Put the re-trained classifier back to the network. The accuracies in cifar100 in ResNet56 is improved from 66% to 73% in our experiment []{data-label="fig: network_ice"}](network_ice.pdf){width="0.7\hsize"}
Introduction and Related Works
==============================
We have seen a great success in deep learning [@goodfellow2016deep]. Now, deep leanings can do a lot of things including language translation, image classification, image generation [@goodfellow2014generative; @2018arXiv180401622J], image generation from text [@2018arXiv180105091H; @2018arXiv180208216S; @2018arXiv180209178Z; @2018arXiv180308495C], text generation from image [@2018arXiv180400861L; @2018arXiv180304376L], unsupervised machine translation [@2017arXiv171100043L; @2017arXiv171011041A; @2018arXiv180407755L], playing Go (reinforcement learning) [@silver2016mastering], and even network design of deep learning by themselves [@2018arXiv180203268P], to raise a few. The achievements are so many, that we cannot exhaustively enough at all.
Six years have passed since the advent of surprise to deep learning [@krizhevsky2012imagenet]. However, there is no such symptom that the progress in deep learning research would slow-down at all. To the contrary, we see that the research has been accelerated, and the realm that deep learning achieves has been spreading even at this moment. We, human, cannot catch up the progress. We do not still understand deep learning yet. In the present paper, we will show another phenomena, which we found by chance.
Our proposed method named “Icing on the Cake” enhances the accuracies of image classification problems in deep learnings by an easy and quick way. We do not fully understand the reason yet, and the phenomena is interesting to the community. It is another learning method as: early stopping [@prechelt1998early] that stops epochs early timing to prevent overfitting; dropout [@srivastava2014dropout] that deactivates some nodes when training; batch normalization [@ioffe2015batch]; group normalization [@2018arXiv180308494W]; He initialization [@2015arXiv150201852H]; Xavier initialization [@glorot2010understanding]; mix-up (between-class learning; Sample Parings) [@2017arXiv171009412Z; @2017arXiv171110284T; @2018arXiv180102929I] that mixes datas and labels to enhance learning; and so forth. We add another one to the existing list.
Proposed Method: Icing on the Cake
==================================
The method named “Icing on the Cake” consists of what follows. The sketch of the idea is illustrated in Fig. \[fig: network\_ice\].
Train a deep neural network as usual.
Extract features from the layer just before the final classifier as estimation.
Train only the final layer again by the extracted features.
Put the final classifier back to the original network.
After the procedure, accuracies increase.
An Easy Way to Implement
------------------------
In the implementation, we do not have to put the re-trained classifier back to the network. We have an easier equivalent method. First, we input test data, and then extract the features of the test data, and then input the features into the re-trained classifier. This is illustrated in Fig. \[fig: test\_ice\].
![An easy way to implement when estimate[]{data-label="fig: test_ice"}](test_mode_ice.pdf){width="0.4\hsize"}
We have two good things in “Icing on the Cake”.
1. It is easy to try.
2. It does not take time for re-training to finish.
It is not very rare that it takes a week or so to train deep neural networks. However, since “Icing on the Cake” is just to train the final layer only, it does not take time. It is done in about five minutes or so. It is really icing on the cake. We do not strongly insist that “Icing on the Cake” works all the cases. We cannot prove it for all the networks to hold as of now. The method is worth trying, and it does not hurt.
![Accuracies in ResNet with varying depth from 11 to 56 for 100-image classification in cifar100 before and after “Icing on the Cake”.[]{data-label="fig: resnet"}](ICK_10.png){width="0.5\hsize"}
Experiments
-----------
ResNet [@he2015deep] with varying depth from 11 to 56 for the100-image classification problem in Cifar100 [@krizhevsky2009learning] are investigated. Batch size is $128$, optimizer is Adam, epochs are $200$, and data augmentation is used. We use Keras [@chollet2015keras]. The result is summarized in Tab. \[tab: resnet-depth\], and illustrated in Fig. \[fig: resnet\]. The averages are taken over ten times. After “Icing on the Cake” the accuracies always get bigger than before, and the standard deviations get smaller.
In other experiments, ResNet56 and DenseNet[^2] [@DBLP:journals/corr/HuangLW16a] are investigated. The 100-image and 10-image classifications are done in Cifar100 and in Cifar10. The results are summarized in Tab. \[tab: result100\] and Tab. \[tab: result10\]. Batch size is $128$, optimizer is Adam, epochs are $200$, and data augmentation is used. “Icing on the Cake” is abbreviated as ICK in the tables. The results show that after “Icing on the Cake” the accuracies get clearly bigger than before. The results are the averages over ten data points for each. After “Icing on the Cake" the accuracies get bigger, and the standard deviations get smaller.
ResNet-Depth Before Icing on the Cake
-------------- -------------- -------------------
11 0.57 (0.010) 0.64 (0.003)
20 0.61 (0.014) 0.69 (0.005)
29 0.63 (0.021) 0.71 (0.005)
38 0.64 (0.016) 0.72 (0.004)
47 0.64 (0.020) 0.72 (0.003)
56 0.66 (0.015) 0.73 (0.003)
: The experiments in ResNet with varying depth from 11 to 56. Accuracies and stand deviations are written.[]{data-label="tab: resnet-depth"}
ResNet56 DenseNet
-------- -------------- --------------- -- --
Before 0.66 (0.015) 0.59 (0.023)
ICK 0.73 (0.003) 0.79 (0.0049)
: The classifications in Cifar10[]{data-label="tab: result10"}
ResNet56 DenseNet
-------- --------------- --------------- -- --
Before 0.885 (0.012) 0.86 (0.024)
ICK 0.92 (0.0025) 0.91 (0.0057)
: The classifications in Cifar10[]{data-label="tab: result10"}
Comments
--------
We do not fully understand the reason why “Icing on the Cake” works. We did experiments in an effort to find a counter example. However, we could not. Rather, we collected more positive evidences at last. The reason why “Icing on the Cake” works is not yet fully investigated, and is open to the community.
Conclusion
==========
We found that classification performance in deep learnings get improved with an easy and quick method named “Icing on the Cake”. The method is just to train the final classifier only again by the extracted features.
It is not that we strongly insist that every deep learning is improved by “Icing on the Cake”, but that there is a possibility that it works well. It is very easy to try, and it does not take time. It is just to train the last one layer. It is worth trying. It is really icing on the cake.
Other experiments in WideResNet
===============================
WideResNet–16–8 [@2016arXiv160507146Z] is investigated for 100-image classification problem in cifar100, and summarized in Tab. \[tab: wideresnet\]. The batch size is 128, optimizer is Adam, and the epochs are 200. The differences in Wide ResNets is not as big as in Resnet and DenseNet. However, after “Icing on the Cake”, the accuracies get always bigger than before as in the table.
Before Icing on the Cake
-------- -------------------
0.72 0.73
0.72 0.74
0.72 0.74
0.72 0.73
0.72 0.73
0.72 0.74
0.73 0.74
0.73 0.74
0.72 0.73
: The experiments in WideResNet–16–8[]{data-label="tab: wideresnet"}
[^1]: Corresponding Author:
[^2]: Depth=40, DenseBlock=3, and $k=12$
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'I study the possibility of directly detecting Ultra-high energy (UHE from now on) WIMPs by the IceCube experiment, via the WIMPs interaction with the nuclei in the ice. I evaluate galactic and extragalactic UHE WIMP and astrophysical and atmospheric neutrino event rates at energy range of 10 TeV - 10 PeV. I assume UHE WIMPs $\chi$ are only from the decay of superheavy dark matter $\phi$, that is $\phi\to\chi\bar{\chi}$. If the lifetime of superheavy dark matter $\tau_{\phi}$ is taken to be $5\times10^{21}$s, WIMPs can be detected at energies above O(40-100)TeV in this detection. Since UHE WIMP fluxes are actually depended on $\tau_{\phi}$, if superheavy dark matter can decay to standard model particles (that is $\tau_{\phi}$ is constrained to be larger than O($10^{26}$-$10^{29}$)s), UHE WIMPs could not be detected by IceCube.'
author:
- 'Ye Xu$^{1,2}$'
title: 'Directly search for Ultra-High Energy WIMPs at IceCube'
---
$^1$School of Information Science and Engineering, Fujian University of Technology, Fuzhou 350118, China
$^2$Research center for Microelectronics Technology, Fujian University of Technology, Fuzhou 350118, China
e-mail address: [email protected]
UHE WIMPs, Superheavy dark matter, Direct detection of dark matter
Introduction
============
It is indicated by the Planck data with measurements of the cosmic microwave background that $26.6\%$ of the overall energy density of the Universe is non-baryonic dark matter [@Planck2015]. Weakly Interacting Massive Particles (WIMPs from now on), predicted by extensions of the Standard Model of particle physics, are a class of candidates for dark matter[@GDJ]. They are distributed in a halo surrounding a galaxy. At present, one mainly searches for thermal WIMPs via direct and indirect detections[@CDMSII; @CDEX; @XENON1T; @LUX; @PANDAX; @AMS-02; @DAMPE; @fermi]. Because of the very small cross sections of the interactions between the thermal WIMPs and nucleus (maybe O(10$^{-47}$ cm$^2$))[@XENON1T; @PANDAX], so far one has not found dark matter yet.
It is a reasonable assumption that there exist various dark matter particles in the Universe. Then it is possible that this sector may comprise of non-thermal components. And these particles may also contain a small component which is relativistic. Although the fraction of these relativistic dark matter particles is small in the Universe, their large interaction cross sections (including between themselves and between them and the Standard Model (SM from now on) particles) make it possible to search for them. Due to the reasons mentioned above, one has to shift more attention to direct and indirect detection of UHE WIMPs. In fact, one has discussed the possibility of searching for UHE dark matter particles in Ref.[@ACNT; @BGA; @xu1].
There is a non-thermal dark sector generated by the early Universe with its bulk comprised of a very massive relic $\phi$ in the Universe. This superheavy dark matter[@KC87; @CKR98; @CKR99; @CGIT; @FKMY] decays to another much lighter WIMPs $\chi$ and its lifetime is greater than the age of the Universe. This lead to a small but significant flux of relativistic WIMPs[@LT; @BGA; @EIP; @BLS; @BKMTZ]. Actually, a few anomalous events may be UHE signatures of a beyond SM particle in the recent observations from the IceCube and ANITA experiments[@anita; @icecube2018].
In the present work, I only focus on direct detection of the UHE WIMPs $\chi$ induced by the decay of superheavy dark matter $\phi$ ($\phi\to\chi\bar{\chi}$). These relativistic WIMPs $\chi$, which pass through the Earth and ice and interact with nuclei, can be detected by a detector like IceCube. In this detection, the main contamination is from astrophysical and atmospheric neutrinos.
In what follows, I will estimate the UHE WIMP and neutrino event rates at IceCube. And it is discussed the possibility of direct detection of UHE WIMPs induced by the decay of superheavy dark matter at IceCube.
UHE WIMPs flux from the Galaxy and Extra Galaxy
===============================================
I consider a scenario where the dark matter sector is composed of at least two particle species in the Universe. One is a co-moving non-relativistic scalar species $\phi$, with mass $m_{\phi} >$ 10 TeV, the other is much lighter particle species $\chi$ ($m_{\chi} \ll m_{\phi}$), due to the decay of $\phi$, with a very large lifetime. And $\chi$ comprises the bulk of present-day dark matter. Since I assume the superheavy dark matter $\phi$ does not decay to SM particles, the constraint on the $\phi$ lifetime are only those based on cosmology[@IOT; @NS; @DMQ; @popolo], that is $\tau_{\phi}\geq10^{17}$s. In the present work, $\tau_{\phi}$ is taken to be $5\times10^{21}$s[@BGA].
The UHE WIMPs flux is composed of galactic and extragalactic components. So the total flux $\psi_{\chi}=\psi_{\chi}^G+\psi_{\chi}^{EG}$, where $\psi_{\chi}^G$ and $\psi_{\chi}^{EG}$ are the galactic and extragalactic fluxes, respectively. The UHE WIMPs flux from the Galaxy is obtained via the following equation[@BGA; @EIP]:
$$\psi_{\chi}^G=F^G\int_{E_{min}}^{E_{max}}\frac{dN_\chi}{dE_\chi}dE$$
with
$$F^G=1.7\times10^{-8}cm^{-2}s^{-1}\times\frac{10^{26}s}{\tau_{\phi}}\times\frac{1TeV}{m_{\phi}} cm^{-2}s^{-1}sr^{-1}.$$
Its zenith angle distribution is similar to the one of the galactic UHE neutrino flux at IceCube(see Fig.1 in Ref.[@CIMM]) in the present paper.
The UHE WIMPs flux from the extra galaxy is obtained via the following equation[@BGA; @EIP]:
$$\psi_{\chi}^{EG}=F^{EG}\int_{E_{min}}^{E_{max}}dE \int_0^{\infty}dz\frac{1}{\sqrt{\Omega_{\Lambda}+\Omega_m(1+z)^3}}\frac{dN_\chi}{dE_\chi}[(1+z)E_\chi]$$
with
$$F^{EG}=1.4\times10^{-8}cm^{-2}s^{-1}\times\frac{10^{26}s}{\tau_{\phi}}\times\frac{1TeV}{m_{\phi}} cm^{-2}s^{-1}sr^{-1}.$$
where z represents the red-shift of the source, $\Omega_{\Lambda}=0.685$ and $\Omega_m=0.315$ from the PLANCK experiment[@Planck2015]. $\displaystyle\frac{dN_{\chi}}{dE_{\chi}}=\delta(E_{\chi}-\displaystyle\frac{m_{\phi}}{2})$, where E$_{\chi}$ and N$_{\chi}$ are the energy and number of UHE WIMP, respectively.
UHE WIMP and neutrino interactions with nuclei
==============================================
In the present paper, I take a Z’ portal dark matter model[@APQ; @Hooper] for WIMPs to interact with nuclei within the IceCube zone (see Fig.1). In this model, a new Z’ gauge boson is considered as a simple and well-motivated extension of SM (see Fig.1(a) in Ref.[@BGA]). And the parameters in the model are taken to be the same as the ones in Ref.[@BGA], that is, the interaction vertexes ($\chi\chi$Z’ and qqZ’) are assumed to be vector-like, the coupling constant G ($G=g_{\chi\chi Z}g_{qqZ}$) is chosen to be 0.05 and the Z’ and $\chi$ masses are taken to be 5 TeV, 10 GeV, respectively. Theoretical models that encompass WIMP spectrum have been discussed in the literature in terms of Z or Z’ portal sectors with Z’ vector boson typically acquiring mass through the breaking of an additional U(1) gauge group at the high energies (see Ref.[@APQ; @Hooper]). The UHE WIMP interaction cross section with nucleus is obtained by the following function(see Fig.1(b) in Ref.[@BGA]):
$$\sigma_{\chi N}=6.13\times10^{-43} cm^2 \left(\frac{E_{\chi}}{1GeV}\right)^{0.518}$$
For neutrinos at the energy range of 10 TeV - 10 PeV, the interaction cross sections with nucleus are are given by simple power-law forms from Ref.[@BHM]:
$$\sigma_{\nu N}(CC)=1.17\times10^{-36} cm^2 \left(\frac{E_{\nu}}{1 GeV}\right)^{0.459}$$
$$\sigma_{\nu N}(NC)=3.78\times10^{-37} cm^2 \left(\frac{E_{\nu}}{1 GeV}\right)^{0.472}$$
where $E_{\nu}$ is the neutrino energy.
Then the above equations show that $\sigma_{\chi N}$ is smaller by 6-7 orders of magnitude, compared to $\sigma_{\nu N}$, at the same energies. The WIMP and neutrino interaction length can be obtained by
$$L_{\nu,\chi}=\frac{1}{N_A\rho\sigma_{\nu,\chi N}}$$
where $N_A$ is the Avogadro constant, and $\rho$ is the density of matter, which WIMPs and neutrinos interact with.
Evaluation of the numbers of UHE WIMPs and neutrinos detected by IceCube
========================================================================
IceCube is a km$^3$ neutrino telescope and can detect three flavour neutrinos via detecting the secondary particles, that in turn emit Cherenkov photons, produced by the interaction between neutrinos and the Antarctic ice[@icecube2018]. UHE WIMPs reach the Earth and pass through the Earth and ice, meanwhile these particles interact with matter of the Earth and ice. Cherenkov Photons are produced by cascades due to UHE WIMPs interaction with nuclei within the IceCube (see Fig. 1). A small part of these photons will be detected by the IceCube detector. Since the UHE WIMP interact with the nucleus in the ice and this is very similar to deep inelastic scattering, its secondary particles develop into a cascade at IceCube. The numbers of UHE WIMPs and neutrinos, (N$_{det}$)$_{\chi,\nu}$, detected by IceCube can be obtained by the following function:
$$(N_{det})_{\chi,\nu} = R\times T\times \int_{E_{min}}^{E_{max}} \int_{cos\theta_{max}}^1(A\Omega)_{eff}(cos\theta) \Phi_{\chi,\nu}(cos\theta,E)dcos\theta dE$$
where R and T are the duty cycle and lifetime of the IceCube experiment, respectively. $\theta$ is the polar angle for the Earth (see Fig. 1), and $\theta_{max}$ is the maximum of $\theta$. $\Phi_\chi=\displaystyle\frac{d\psi_\chi}{dE}$ and $(A\Omega)_{eff}(cos\theta)$ = the observational area $\times$ the effective solid angle $\times$ P(E,$D_e(cos\theta)$,D). Here
$$P(E,D_e(cos\theta),D)=exp\left(-\displaystyle\frac{D_e(cos\theta)}{L_{earth}}\right)\left[1-exp\left(-\displaystyle\frac{D}{L_{ice}}\right)\right]$$
where $P(E,D_e(cos\theta),D)$ is the probabilities that UHE WIMPs and neutrinos interact with ice after traveling a distance between $D_e(cos\theta)$ and $D_e(cos\theta)+D$, where D is the effective length in the IceCube detecting zone in the ice, $D_e(cos\theta)$ are the distances through the Earth, and $L_{earth,ice}$ are the UHE WIMP and neutrino interaction lengths with the Earth and ice, respectively. In the present paper, I assume that all UHE WIMPs and neutrinos detected by IceCube interact with the ice within its volume. In what follows, I will make a rough study of $(A\Omega)_{eff}(cos\theta)$ and then evaluate the numbers of UHE WIMPs detected by IceCube. Here I make an assumption that the observational area of IceCube is regarded as a point in the calculation of the effective solid angle $\Omega$. Under this approximation,
$$(A\Omega)_{eff}(cos\theta) \approx P(E,D_e(cos\theta),D)A\frac{2\pi {R_e}^2}{D_e(cos\theta)^2} .$$
where A is the observational area, $R_e$ is the radius of the Earth, $D_e(cos\theta)=\displaystyle\frac{R_e(1+cos\theta)}{cos\theta_z}$. $\theta_z$ is the zenith angle at IceCube and $\theta_z=\theta/2$. And the observational area $\approx 1 km^2$, D $\approx$ 1 km and R $\approx$ 100% at IceCube.
The background is mainly two sources: astrophysical and atmospheric neutrinos. The astrophysical neutrinos is estimated with a diffuse neutrino flux of $\Phi_{\nu}=0.9^{+0.30}_{-0.27}\times(E_{\nu}/100TeV)\times10^{-18}GeV^{-1} cm^{-2}s^{-1}sr^{-1}$[@icecube], where $\Phi_{\nu}$ represents the per-flavor flux, by the above method. And The atmospheric neutrinos is estimated with a flux of $lg E_{\nu}^2\Phi_{\nu}=\sum\limits_{k=0}^4\sum\limits_{n=0}^3a_{kn}x^ny^k$ by the same method[@SPS], where $\Phi_{\nu}$ represents the atmospheric neutrino flux, $y=lg(E_{\nu}/1GeV)$, $x=cos\theta_z$ and coefficients $a_{kn}$ are given in Table 1 in Ref.[@SPS].
There is the presence of the Glashow resonance(on-shell $W^-$ via $\bar{\nu}_e+e^- \to W^- \to anything$ and its cross section is $5.02\times10^{-31}cm^2$) at 6.3 PeV and its secondary particles develop into cascades at IceCube[@SLG; @BG1; @BG2; @BGRW]. So this effect has to be taken into account in the background estimation.
A mixture of all flavours of UHE neutrinos consists of two different event topologies(cascades and tracks) at IceCube. We can reject a part of neutrinos by distinguishing track-like events from cascade-like events. So the neutrino contamination is actually less than the results of the calculation in the present paper. As the conservative estimation of background, however, I didn’t consider this event topology identification in my calculation yet.
Results
=======
In my work, the UHE WIMP and neutrino event rates are evaluated at energy range of 10 TeV - 10 PeV at IceCube. Fig. 2 - 7 compare the galactic and extragalactic UHE WIMP event rates to the astrophysical and atmospheric neutrino event rates corresponding to different $\theta_{max}$ when $\tau_{\phi}=5\times10^{21}$s, respectively. We find that an energy threshold has to be set to reject neutrinos, since neutrinos are more dominant in the events detected by IceCube at low energy(see Fig. 2 - 7).
Fig. 2 shows the WIMP and neutrino event rates at $\theta_{max}=\displaystyle\frac{\pi}{6}$. In Fig. 2, we find that UHE WIMPs can be detected by IceCube at energies above about 40 TeV, and their event rates are $\sim$2 event/year and $\sim$2 events/10 years at 40 TeV and 10 PeV, respectively. Fig. 3 shows the WIMP and neutrino event rates at $\theta_{max}=\displaystyle\frac{\pi}{3}$. In Fig. 3, we find that UHE WIMPs can be detected by IceCube at energies above about 100 TeV, and their event rates are $\sim$10 events/year and $\sim$1 event/year at 100 TeV and 10 PeV, respectively. Fig. 4 shows the WIMP and neutrino event rates at $\theta_{max}=\displaystyle\frac{\pi}{2}$. In Fig. 4, we find that UHE WIMPs can be detected by IceCube at energies above about 110 TeV, and their event rates are $\sim$24 events/year and $\sim$3 events/year at 110 TeV and 10 PeV, respectively. Fig. 5 shows the WIMP and neutrino event rates at $\theta_{max}=\displaystyle\frac{2\pi}{3}$. In Fig. 5, we find that UHE WIMPs can be detected by IceCube at energies above about 130 PeV, and their event rates are $\sim$40 events/year and $\sim$5 events/year at 130 TeV and 10 PeV, respectively. Fig. 6 shows the WIMP and neutrino event rates at $\theta_{max}=\displaystyle\frac{5\pi}{6}$. In Fig. 6, we find that UHE WIMPs can be detected by IceCube at energies above about 160 TeV, and their event rates are $\sim$63 events/year and $\sim$9 events/year at 160 TeV and 10 PeV, respectively. Fig. 7 shows the WIMP and neutrino event rates at $\theta_{max}=\pi$. In Fig. 7, we find that UHE WIMPs can be detected by IceCube at energies above about 380 TeV, and their event rates are $\sim$925 events/year and $\sim$204 events/year at 380 TeV and 10 PeV, respectively.
Conclusion and Discussion
=========================
According to the results described above, it is possible that UHE WIMPs are directly detected with a detector like IceCube. Especially, the WIMPs with O(40-100)TeV from the other side of the earth (that is $\theta_{max} < \displaystyle\frac{\pi}{2}$) can be detected by IceCube when $\tau_{\phi}=5\times10^{21}$s.
Since the UHE WIMP flux $\Phi_{\chi}$ is proportional to $\displaystyle\frac{1}{\tau_{\phi}}$, WIMP event rates are actually depended on the lifetime of superheavy dark matter. If we assume the superheavy dark matter $\phi$ can decay to SM particles, $\tau_{\phi}$ is constrained to be larger than O($10^{26}$-$10^{29}$)s[@EIP; @MB; @RKP; @KKK] by diffuse gamma-ray and neutrino observations. Fig. 8-10 show the WIMP and neutrino event rates are evaluated at $\theta_{max}=\pi$ when $\tau_{\phi}=10^{17}$, $10^{25}$ and $10^{26}$s, respectively. In Fig. 8, we find UHE WIMPs can be detected at energies 10 TeV - 10 PeV at IceCube when $\tau_{\phi}=10^{17}$s. In Fig. 10, We find that WIMPs could not be detected at energies 10 TeV - 10 PeV since neutrinos are more dominant in the events by IceCube when $\tau_{\phi}$ is larger than $10^{26}$s.
In Fig. 9, we find the WIMPs can be detected by IceCube only at O(1PeV) and their event rate is O(1) events/10 years when $\tau_{\phi}=10^{25}$s after deducting the Glashow resonance contamination at near 6.3 PeV. This is consistent with the observations that there are three anomalous events at O(1PeV) in nine years of IceCube data[@icecube2018; @anita]. If they are actually UHE WIMPs, the lifetime of superheavy dark matter may be $\sim$10$^{25}$s. Fig. 11 shows the Signal to Background Rates(SBR) at different energy and different $\theta$ when $\tau_{\phi}=10^{25}$s. $SBR=\displaystyle\frac{N_{WIMP}}{N_{\nu}}$, where $N_{WIMP}$ and $N_{\nu}$ are the event rates of UHE WIMPs and neutrinos detected at IceCube, respectively. In Fig. 11, we find UHE WIMPs can be detected at energies above $\sim$2 PeV and the neutrino contamination is down to O($10^{-5}$) at low $\theta$ and high energy region(about $0^{\circ} < \theta < 65^{\circ}$ and 8 PeV < E < 10 PeV). It is significantly for the Glashow resonance to make an effect on this WIMP detection at near 6.3 PeV. There is the peak due to the UHE WIMP flux of the galactic center at about 56$^\circ$ in Fig. 11.
Acknowledgements
================
This work was supported by the National Natural Science Foundation of China (NSFC) under the contract No. 11235006, the Science Fund of Fujian University of Technology under the contract No. GY-Z14061 and the Natural Science Foundation of Fujian Province in China under the contract No. 2015J01577.
P.A.R. Ade, et al., Planck collaboration, A&A 594, A13, arXiv:1502.01589 G. Bertone, D. Hooper, and J. Silk, Physics Reports, 405, 279 (2005) R.Agnese, et al., SuperCDMS Collaboration, Phys. Rev. D 92, 072003 (2015) K.J. Kang, et al.,CDEX Collaboration, Front. Phys. 8(4),412-437 (2013) E.Aprile, et al., XENON1T Collaboration, Phys. Rev. Lett. 119, 181301 (2017) D.S.Akerib, et al., LUX Collaboration, Phys. Rev. Lett. 118, 251302 (2017) X.Y. Cui, et al., PandaX-II Collaboration, Phys. Rev. Lett. 119, 181302 (2017) M.Aguilar et al., AMS Collaboration, Phys. Rev. Lett. 113, 221102 (2014) G.Ambrosi, et al., DAMPE Collaboration, Nature, 552, 63-66 (2017) A.Albert, et al., Fermi-LAT, DES Collaborations, Astrophysical Journal 834, 110 (2017) K.Agashe, Y.Cui, L.Necib and J.Thaler, JCAP 10, 062 (2014), arXiv:1405.7370 A. Bhattacharya, R. Gandhi and A. Gupta, JCAP 03, 027 (2015), arXiv:1407.3280 Ye Xu, JCAP 05,055 (2018), arXiv: 1801.06781 M.Yu.Khlopov, V.M.Chechetkin, Sov. J. Part. Nucl 18, 267-288 (1987) D. J. Chung, E. W. Kolb, and A. Riotto, Phys.Rev.Lett. 81, 4048 (1998), arXiv: hep-ph/9805473 D. J. Chung, E. W. Kolb, and A. Riotto, Phys.Rev. D59, 023501 (1999), arXiv: hep-ph/9802238 L. Covi, M. Grefe, A. Ibarra, and D. Tran, JCAP 1004, 017 (2010), arXiv: 0912.3521 B. Feldstein, A. Kusenko, S. Matsumoto, and T. T. Yanagida, Phys. Rev. D 88, 015004 (2013), arXiv: 1303.7320 X.Q. Li and Z.J. Tao, Commun.Theor.Phys., 34, 123-128 (2000) arXiv:hep-ph/9710383 A. Esmaili, A. Ibarra, and O.L. Peres, JCAP, 1211, 034 (2012) arXiv:1205.5281 J.M.Carceller, J.I.Illana, M.Masip and D.Meloni, Astrophys.J. 852 no.1, 59 (2018) arXiv: 1703.10786 Y. Bai, R. Lu, and J. Salvado, JHEP, 01, 161 (2016) arXiv:1311.5864 P.S. Bhupal Dev, et al.,JCAP 1608, 034 (2016) arXiv:1606.04517 Derek B. Fox et al., The ANITA Anomalous Events as Signatures of a Beyond Standard Model Particle, and Supporting Observations from IceCube, arXiv: 1809.09615 M.G.Aartsen, et al., IceCube Collaboration, Phys. Rev. D 98, 062003 (2018), arXiv:1807.01820 M.G.AArtsen, et al., IceCube Collaboration, Astrophysical J. 833,3 (2016), arXiv:1607.08006 K. Ichiki, M. Oguri, and K. Takahashi, Phys.Rev.Lett. 93 (2004) 071302, arXiv:astro-ph/0403164 K. M. Nollett and G. Steigman, Phys. Rev. D 91, 083505 (2015), arXiv:1411.6005 P. Bhupal Dev, A. Mazumdar, and S. Qutub, Physics 2, 26 (2014), arXiv:1311.5297 A. Del Popolo, Astron.Rep. 51, 169¨C196 (2007), arXiv:0801.1091 A. Alves, S. Profumo, and F. S. Queiroz, JHEP 1404, 063 (2014), arXiv:1312.5281 D. Hooper, Phys. Rev. D 91, 035035, arXiv:1411.4079 Martin M. Block, Phuoc Ha, Douglas W. McKay, Phys. Rev. D 82, 077302 (2010), arXiv: 1008.4555 S.I.Sinegovsky, O.N.Petrova and T.S.Sinegovskaya, 32ND International Cosmic Ray Conference, BeiJing 2011, ICRC2011, Vol.4, 0487 (2011) S.L.Glashow, Phys. Rev. 118, 316-317 (1960) V. S. Berezinsky, A. Z. Gazizov, JETP Lett. 25, 254-256 (1977) V. S. Berezinsky, A. Z. Gazizov, Sov. J. Nucl. Phys. 33, 120-125 (1981) A.Bhattacharya, R.Gandi, W.Rodejohann and A.Watanabe, JCAP 1110, 017 (2011), arXiv: 1108.3163 K.Murase and J.F.Beacom, JCAP 1210 (2012), arXiv:1206.2595 C.Rott, K.Kohri and S.C.Park, Phys. Rev. D 92, 023529 (2015), arXiv:1408.4574 M.Kachelriess, O.E.Kalashev and M.Yu.Kuznetsov, Phys. Rev. D 98, 083016 (2018), arXiv: 1805.04500
![UHE WIMPs pass through the Earth and ice and can be detected by a detector like IceCube, via Cherenkov photons due to the development of cascades in the ice. $\theta$ is the polar angle for the Earth.$\theta_{max}$ is the maximum of $\theta$[]{data-label="fig:figure"}](figure){width="90.00000%"}
![The UHE WIMPs and neutrino event rates are evaluated at $\theta_{max} = \displaystyle\frac{\pi}{6}$ at IceCube[]{data-label="fig:30"}](30){width="90.00000%"}
![The UHE WIMPs and neutrino event rates are evaluated at $\theta_{max} = \displaystyle\frac{\pi}{3}$ at IceCube[]{data-label="fig:60"}](60){width="90.00000%"}
![The UHE WIMPs and neutrino event rates are evaluated at $\theta_{max} = \displaystyle\frac{\pi}{2}$ at IceCube[]{data-label="fig:90"}](90){width="90.00000%"}
![The UHE WIMPs and neutrino event rates are evaluated at $\theta_{max} = \displaystyle\frac{2\pi}{3}$ at IceCube[]{data-label="fig:120"}](120){width="90.00000%"}
![The UHE WIMPs and neutrino event rates are evaluated at $\theta_{max} = \displaystyle\frac{5\pi}{6}$ at IceCube[]{data-label="fig:150"}](150){width="90.00000%"}
![The UHE WIMPs and neutrino event rates are evaluated at $\theta_{max} = \pi$ at IceCube[]{data-label="fig:180"}](180){width="90.00000%"}
![The UHE WIMPs and neutrino event rates are evaluated at $\theta_{max} = \pi$ at IceCube, if $\tau_{\phi}=10^{17}s$.[]{data-label="fig:180_17"}](180_17){width="90.00000%"}
![The UHE WIMPs and neutrino event rates are evaluated at $\theta_{max} = \pi$ at IceCube, if $\tau_{\phi}=10^{25}s$ and the lifetime of the IceCube experiment is ten years.[]{data-label="fig:180_25_10y"}](180_25_10y){width="90.00000%"}
![The UHE WIMPs and neutrino event rates are evaluated at $\theta_{max} = \pi$ at IceCube, if $\tau_{\phi}=10^{26}s$ and the lifetime of the IceCube experiment is ten years[]{data-label="fig:180_26_10y"}](180_26_10y){width="90.00000%"}
![the Signal to Background Rates are evaluated at different energy and different $\theta$, if $\tau_{\phi}=10^{25}s$.[]{data-label="fig:sbr_25"}](sbr_25){width="90.00000%"}
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'In this paper, we develop necessary and sufficient conditions for the validity of a martingale approximation for the partial sums of a stationary process in terms of the maximum of consecutive errors. Such an approximation is useful for transferring the conditional functional central limit theorem from the martingale to the original process. The condition found is simple and well adapted to a variety of examples, leading to a better understanding of the structure of several stochastic processes and their asymptotic behaviors. The approximation brings together many disparate examples in probability theory. It is valid for classes of variables defined by familiar projection conditions such as the Maxwell–Woodroofe condition, various classes of mixing processes, including the large class of strongly mixing processes, and for additive functionals of Markov chains with normal or symmetric Markov operators.'
address:
- 'POMI (Saint Petersburg Department of the Steklov Institute of Mathematics), 27 Fontanka emb., Saint Petersburg 191023, Russia. '
- 'Department of Mathematical Sciences, University of Cincinnati, P.O. Box 210025, Cincinnati, OH 45221-0025, USA. '
author:
-
-
title: On the functional central limit theorem via martingale approximation
---
Introduction and results
========================
The objective of this paper is to find a characterization of stationary stochastic processes that can be studied via a martingale approximation in order to derive the functional central limit theorem for processes associated with partial sums.
There are several ways to present the results since stationary processes can be introduced in several equivalent ways. We assume that $(\xi
_{n})_{n\in\mathbb{Z}}$ denotes a stationary Markov chain defined on a probability space $(\Omega,\mathcal{F},P)$ with values in a measurable space $(S,\mathcal{A})$. The marginal distribution and the transition kernel are denoted by $\pi(A)=P(\xi_{0}\in A)$ and $Q(\xi_{0},A)=P(\xi_{1}\in A|
\xi
_{0})$, respectively. In addition, $Q$ denotes the operator [acting via $(Qf)(\xi)=\int
_{S}f(s)Q(\xi,\mathrm{d}s).$ Next, let $\mathbb{L}_{0}^{2}(\pi)$ be the set of functions on $S$ such that $\int f^{2}\,\mathrm{d}\pi<\infty$ and $\int f\,\mathrm{d}\pi=0.$ Denote by $\mathcal{F}_{k}$ the $\sigma$-field generated by $\xi_{i}$ with $i\leq k,$ $X_{i}=f(\xi_{i})$, $S_{n}=\sum _{i=0}^{n-1}X_{i}$ (i.e., $S_{0}=0,S_{1}=X_{0}$, $S_{2}=X_{0}+X_{1},\dots$). For any integrable variable $X$, we define $\mathbb{E}_{k}(X)=\mathbb{E}(X|\mathcal{F}_{k}).$ In our notation, $\mathbb{E}_{0}(X_{1})=Qf(\xi_{0})=\mathbb{E}(X_{1}|\xi
_{0}).$ ]{} We also set $\mathcal F_{-\infty}=\bigcap_{k \in\mathbb Z}
\mathcal F_{k} $ .
Throughout the paper, we assume $f\in\mathbb{L}_{0}^{2}(\pi)$; in other words, we assume that $\Vert X\Vert _{2}=(\mathbb{E}[X_{1}^{2}])^{1/2}<\infty$ and $\mathbb{E}[X_{1}]=0.$
Note that any stationary sequence $(Y_{k})_{k\in\mathbb{Z}}$ can be viewed as a function of a Markov process $\xi_{k}=(Y_{i};i\leq k)$ for the function $g(\xi_{k})=Y_{k}$.
The stationary stochastic processes may be also introduced in the following, alternative, way. Let $T\dvtx \Omega\mapsto\Omega$ be a bijective bimeasurable transformation preserving the probability. Let $\mathcal{F}_{0}$ be a sub-$\sigma$-algebra of $\mathcal{F}$ satisfying $\mathcal{F}_{0}\subseteq
T^{-1}(\mathcal{F}_{0})$. We then define the non-decreasing filtration $(\mathcal{F}_{i})_{i\in\mathbb{Z}}$ by $\mathcal{F}_{i}=T^{-i}(\mathcal
{F}_{0})$. Let $X_{0}$ be a random variable which is ${}\mathcal{F}_{0}$-measurable. We define the stationary sequence $(X_{i})_{i\in\mathbb
{Z}}$ by $X_{i}=X_{0}\circ T^{i}$.
In this paper, we shall use both frameworks.
In order to analyze the asymptotic behavior of the partial sums $S_{n}=\sum_{i=0}^{n-1}X_{i},$ Gordin, in [@g], proposed to decompose the sums related to the original stationary sequence into the sum $$S_{n}=M_{n}+R_{n} \label{martaprox}$$ of a square-integrable martingale $M_{n}=\sum_{i=0}^{n-1}D_{i}$ adapted to $\mathcal{F}_{n}$, whose martingale differences $(D_{i})$ are stationary, and a so-called coboundary $R_{n}$, that is, a telescoping sum of random variables with the basic property that $\sup_{n}\mathbb{E}(R_{n}^{2})<\infty$. More precisely, $X_{n}=D_{n}+Z_{n}-Z_{n-1},$ where $Z_{n}$ is another stationary sequence in $\mathbb{L}_{2}$. The limiting properties of the martingales can then be transported from the martingale to the general sequence. In the context of Markov chains, the existence of such a decomposition is equivalent to the solvability of the Poisson equation in $\mathbb{L}_{2}$.
For proving a central limit theorem for stationary sequences, a weaker form of martingale approximation has been pointed out by many authors (see, e.g., [@mpu2] for a survey). Recently, two interesting papers, one by Dedecker, Merlevède and Volný [@dmv] and the other by Zhao and Woodroofe [@Zw2], provided necessary and sufficient conditions for martingale approximation with an error term in (\[martaprox\]) satisfying $$\mathbb{E}\bigl((S_{n}-M_{n})^{2}\bigr)/n\rightarrow0. \label{MA}$$ This decomposition is strong enough for transporting the conditional central limit theorem from sums of stationary martingale differences in $\mathbb{L}_{2}$ to $S_{n}/\sqrt{n}.$ By conditional CLT, as discussed in [@DM02], we understand, in this context, that for any continuous function $f$ such that $|f(x)|/(1+x^{2})$ is bounded and for any $k \ge0,$ $$\biggl\Vert \mathbb{E}_{k}\bigl(f\bigl(S_{n}/\sqrt{n}\bigr)\bigr)- \int_{- \infty}^{\infty} f\bigl(x
\sqrt{\eta}\bigr) g(x)\,\mathrm{d}x\biggr\Vert _1 \mathop{\longrightarrow}_{n \to\infty} 0,
\label{CLT}$$ where $g$ is the standard normal density and $\eta\ge0$ is an invariant function satisfying $$\mathop{\lim}_{n \to\infty}\biggl\Vert\frac{\mathbb
{E}_{0}(S_n^2)}{n}-\eta
\biggr\Vert_1=0.$$ Here, and throughout the paper, we denote by $\Vert \cdot\Vert _{p}$ the norm in $\mathbb{L}_{p}$.
An important extension of this theory is to consider the conditional central limit theorem in its functional form. For $t\in\lbrack0,1]$, define $$S_n(t)=S_{[nt]}+(nt-[nt])X_{[nt]},$$ where $[x]$ denotes the integer part of $x$. Note that $S_n(\cdot)/\sqrt{n}$ is a random element of the space $C([0,1])$ endowed with the supremum norm $\Vert \cdot\Vert _{\infty}.$ Then, by the conditional CLT in the functional form (FCLT), we understand that for any continuous function $f\dvtx C([0,1]) \to\mathbb R$ such that $x \mapsto|f(x)|/(1+\Vert x\Vert _{\infty}^2)$ is bounded and for any $k\ge
0$, we have $$\biggl\Vert \mathbb{E}_{k}\bigl(f\bigl(S_n/\sqrt{n}\bigr)\bigr)- \int_{C([0,1])}\bigl(f\bigl(x
\sqrt{\eta}\bigr)\bigr)\,\mathrm{d}W(x)\biggr\Vert _1\mathop{\longrightarrow}_{n \to\infty} 0. \label{FCLT}$$ Here, $W$ is the standard Wiener measure on $C([0,1])$.
It is well known that a martingale with stationary differences in $\mathbb{L}_{2}$ satisfies this type of behavior with $\eta=
\lim_{n \to\infty} \sum_{l=0}^{n-1}D_{l}^2/n$ in $\mathbb
{L}_{1}$ – this is at the heart of many statistical procedures. This conditional form of the invariance principle is a stable type of convergence that makes possible the change of measure with another absolutely continuous measure, as discussed in [@b; @ru; @hh].
With such a result in mind, the question is now to find necessary and sufficient conditions for a martingale decomposition with the error term satisfying $$\mathbb{E}\Bigl(\max_{1\leq j\leq n}(S_{j}-M_{j})^{2}\Bigr)\big/n\rightarrow0 .
\label{maxcond}$$
In order to state our martingale approximation result, for fixed $m$, we consider the stationary sequence $$Y_{0}^{m}=\frac{1}{m}\mathbb{E}_{0}(X_{1}+\cdots+X_{m}),\qquad
Y_{k}^{m}=Y_{0}^{m}\circ T^{k}. \label{defY}$$ In the language of Markov operators, we then have $$Y_{0}^{m}=\frac{1}{m}(Qf+\cdots+Q^{m}f)(\xi_{0}) .$$ It is convenient to introduce a seminorm notation, namely,$$\Vert Z\Vert _{M^{+}}=\lim\sup_{n\rightarrow\infty}\frac{1}{\sqrt{n}}\Biggl\Vert \max_{1\leq k\leq n}\Biggl| \sum_{j=1}^{k} Z \circ T^j
\Biggr| \Biggr\Vert_{2}$$ on the space of all $Z \in L^2_0$ with $\Vert Z\Vert _{M^{+}} < \infty.$
\[T\]\[maxgencopy(2)\]Assume that $(X_{k})_{k\in Z}$ is a stationary sequence of centered square-integrable random variables. Then$$\Vert Y_{0}^{m}\Vert _{M^{+}}\rightarrow0\qquad\mbox{as }m\rightarrow\infty
\label{MAX}$$ if and only if there exists a martingale with stationary increments satisfying [(\[maxcond\])]{}. Such a martingale is unique if it exists. In particular, [(\[MAX\])]{} implies [(\[FCLT\])]{}.
As a consequence of the proof of Theorem \[T\], we also obtain the following result that adds a new equivalent condition to the characterizations by Dedecker, Merlevède and Volný [@dmv] and Zhao and Woodroofe [@Zw2]. With $(Y_{k}^{m})_{k\in Z}$ defined by [(\[defY\])]{} and the seminorm notation $$\Vert Y_{0}^{m}\Vert _{+}=\lim\sup_{n\rightarrow\infty}\frac{1}{\sqrt{n}}\Biggl\Vert \sum_{j=1}^{n}Y_{j}^{m}\Biggr\Vert _{2}$$ we have the following characterization.
\[T2\]\[maxgencopy(1)\]\[Tcopy(1)\]\[maxgen\]Assume that $(X_{k})_{k\in Z}$ is as in Theorem \[T\]. Then $$\Vert Y_{0}^{m}\Vert _{+}\rightarrow0\qquad\mbox{as }m\rightarrow\infty
\label{MA1}$$ if and only if there exists a stationary martingale satisfying [(\[MA\])]{}. Such a martingale is unique if it exists. In particular, [(\[MA1\])]{} implies [(\[CLT\])]{}.
Our approach is constructive. If the stationary sequence is supposed to be ergodic, then the constructed martingale differences are also ergodic and therefore the conditional theorems (\[CLT\]) and (\[FCLT\]) can be easily transported to the original processes satisfying (\[MA1\]) and (\[MAX\]), respectively, with $\eta=\Vert D_{0}\Vert _{2}$.
A natural and useful question is to provide classes of stochastic processes that have a martingale decomposition with an error term satisfying [(\[maxcond\])]{}, in other words, to provide sharp sufficient conditions for such a decomposition. Obviously, a maximal inequality is needed in order to verify this condition. We shall combine our approach with several maximal inequalities. One is due to Rio [@rio], formula (3.9), page 53; for related inequalities, see [@Pel] and [@DR].
- For any stationary process with centered variables in $\mathbb{L }_{2}
$, $$\mathbb{E}\Bigl(\max_{1\leq i\leq n}S_{i}^{2}\Bigr)\leq8n\mathbb{E}(X_{0}^{2})+16
\sum_{k=2}^{n}\mathbb{E}|X_{0}\mathbb{E}_{0}(S_{k}-S_1)|.
\label{Rio}$$
Another inequality comes from [@PU1], Proposition (2.3); see also [@puw], Theorem 1, for the inequality in $\mathbb{L}_{p}$.
- For any stationary process with centered variables in $\mathbb{L}_{2}$, $$\begin{aligned}
\label{PU}
\mathbb{E}\Bigl(\max_{1\leq i\leq n}S_{i}^{2}\Bigr) & \leq& n\Biggl(2\Vert
X_{0}\Vert_{2}+3\sum_{j=0}^{r-1}\frac{\Vert \mathbb
{E}_{0}(S_{2^{j}})\Vert _{2}}{2^{j/2}}\Biggr)^{2}
\nonumber
\\[-8pt]
\\[-8pt]
\nonumber
& \leq& n\Biggl(2\Vert X_{0}\Vert_{2}+80\sum_{j=1}^{n}\frac{\Vert \mathbb{E}_{0}(S_{j})\Vert _{2}}{j^{3/2}}\Biggr)^{2},\end{aligned}$$ where $2^{r-1} < n\leq2^{r}.$
The following maximal inequality is a particular case of Dedecker and Merlevède [@DM02], Proposition 6; see [@Wu], Theorem 1, for the inequality in $\mathbb{L}_{p}$.
- For any stationary process with centered variables in $\mathbb{L}_{2}$ such that $\mathbb E(X_{0}|\mathcal{F}_{-\infty})=0$ almost surely, we have$$\mathbb{E}\Bigl(\max_{1\leq i\leq n}S_{i}^{2}\Bigr)\leq4n\Biggl(\sum
_{i=0}^{\infty}\Vert
\mathbb{E}_{-i}(X_{0})-\mathbb{E}_{-i-1}(X_{0})\Vert_{2}\Biggr)^{2}
.\label{DM}$$
Another inequality we use for additive functionals of stationary reversible Markov chains is a consequence of Wu [@wu], Corollary 2.7 and relation (2.5) in the same paper (note that there is a typographical error in this relation, namely, a square should be added to the norm); see also [@svy]:
- Assume $(\xi_{n})_{n\in\mathbb{Z}}$ is a stationary, reversible Markov chain and $X_n=f(\xi_n)$ with $f\in\mathbb
{L}_{0}^{2}(\pi)$. Then, for every $n\geq1$,$$\mathbb{E}\Bigl(\max_{1\leq i\leq n}S_{i}^{2}\Bigr)\leq
(24n+3)\sum _{n=0}^{\infty}\mathbb E(X_{0}X_{n}),
\label{LW}$$ provided the series on the right-hand side is convergent.
This inequality, originally stated for the ergodic case, extends without changes to the general case.
By combining the martingale decomposition in Theorem \[T\] with these maximal inequalities, we point out various classes of stochastic processes for which a conditional functional limit theorem holds. These include mixing processes and classes of Markov chains.
Proof of Theorem \[T\]
======================
The proof of this theorem has several steps.
*Step Construction of the approximating martingale*.
The construction of the martingale decomposition is based on averages. It was introduced by Wu and Woodroofe [@W] (see their definition (6) on page 1677) and further developed in [@Zw2], extending the construction in [@Heyde] and [@gl]; see also [@1994], Theorem 8.1, and [@KV]. We give the martingale construction here for completeness.
We introduce a parameter $m\geq1$ (kept fixed for the moment) and define the following stationary sequence of random variables:$$\theta_{0}^{m}=\frac{1}{m}\sum_{i=1}^{m}\mathbb{E}_{0}(S_{i}),\theta_{k}^{m}=\theta_{0}^{m}\circ T^{k}.$$ Set $$D_{k}^{m}=\theta_{k+1}^{m}-\mathbb{E}_{k}(\theta_{k+1}^{m}),\qquad
M_{n}^{m}=\sum_{k=0}^{n-1}D_{k}^{m}. \label{martd}$$ Then $(D_{k}^{m})_{k\in\mathbb{Z}}$ is a stationary martingale difference sequence and $(M_{n}^{m})_{n\geq0}$ is a martingale. Thus, we have$$X_{k}=D_{k}^{m}+\theta_{k}^{m}-\theta_{k+1}^{m}+\frac{1}{m}\mathbb{E}_{k}(S_{k+m+1}-S_{k+1})$$ and therefore$$\begin{aligned}
\label{martdec}
S_{k}& =&M_{k}^{m}+\theta_{0}^{m}-\theta_{k}^{m}+\sum
_{j=1}^{k}\frac{1}{m}\mathbb{E}_{j-1}(S_{j+m}-S_{j})
\nonumber
\\[-8pt]
\\[-8pt]
\nonumber
& =&M_{k}^{m}+\theta_{0}^{m}-\theta_{k}^{m}+\overline{R}_{k}^{ m},\end{aligned}$$ where we have made use of the notation$$\overline{R}_{k}^{ m}=\sum _{j=1}^{k}\frac{1}{m}\mathbb{E}_{j-1}(S_{j+m}-S_{j}).$$ Observe that $$\overline{R}_{k}^{ m}=\sum_{j=0}^{k-1}Y_{j}^{m}.\label{rests}$$ With the notation $$R_{k}^{m}=\theta_{0}^{m}-\theta_{k}^{m}+\overline{R}_{k}^{ m},
\label{rest}$$ we have $$S_{k}=M_{k}^{m}+R_{k}^{m}. \label{deco}$$
*Step Sufficiency*.
We show that $\Vert Y_{0}^{m}\Vert _{M^{+}}\rightarrow0$ as $m\rightarrow\infty
$ is sufficient for (\[maxcond\]).
The starting point is the construction of the martingale differences, as in (\[martd\]). By the martingale property and (\[deco\]), for all positive integers $m^{\prime}
$ and $m^{\prime\prime}$, we have $$\Vert D_{0}^{m^{\prime}}-D_{0}^{m^{\prime\prime}}\Vert _{2}=\frac{1}{\sqrt
{n}}\Vert M_{n}^{m^{\prime}}-M_{n}^{m^{\prime\prime}}\Vert _{2}=\frac{1}{\sqrt
{n}}\Vert R_{n}^{m^{\prime}}-R_{n}^{m^{\prime\prime}}\Vert _{2} .$$ We now let $n\rightarrow\infty.$ By relation (\[rest\]) and stationarity, it follows that $$\begin{aligned}
\lim\sup_{n\rightarrow\infty}\frac{1}{\sqrt{n}}\Vert R_{n}^{m^{\prime
}}-R_{n}^{m^{\prime\prime}}\Vert _{2}&=&\lim\sup_{n\rightarrow\infty
}\frac{1}{\sqrt{n}}\Vert \overline{R}_{n}^{ m^{\prime}}-\overline{R}_{n}^{
m^{\prime
\prime}}\Vert _{2} \\
&\leq&\lim\sup_{n\rightarrow\infty}\frac{1}{\sqrt{n}}(\Vert \overline{R}_{n}^{ m^{\prime}}\Vert _{2}+\Vert \overline{R}_{n}^{ m^{\prime\prime
}}\Vert _{2}).\end{aligned}$$ By (\[MAX\])$,$ the limit when $m^{\prime}$ and $m^{\prime\prime}$ both tend to $\infty$ is then $0$, giving that $(D_{0}^{m})$ is Cauchy in $\mathbb{L}_{2}$ and therefore convergent. Denote its limit by $D_{0}$. Then $M_{n}=\sum _{k=0}^{n-1}D_{k}$ is a martingale with the desired properties. To see this, we start from the decomposition in relation ([martdec]{}) and obtain $$|S_{k}-M_{k}|\leq|M_{k}^{m}-M_{k}|+|\theta_{k}^{m}-\theta_{0}^{m}|+|\overline{R}_{k}^{m}|.$$ Then$$\begin{aligned}
\frac{{1}}{\sqrt{n}}\Bigl\Vert \max_{1\leq k\leq n}|S_{k}-M_{k}|
\Bigr\Vert_{2}&\leq&
\frac{{1}}{\sqrt{n}}\Bigl\Vert \max_{1\leq k\leq n}|M_{k}^{m}-M_{k}|
\Bigr\Vert _{2} \\
&&{}+\frac{{1}}{\sqrt{n}}\Vert \theta_{0}^{m}\Vert _{2}+\frac{{1}}{\sqrt{n}}\Bigl\Vert \max_{1\leq k\leq n}|\theta_{k}^{m}| \Bigr\Vert_{2}+\frac{{1}}{\sqrt{n}}\Bigl\Vert
\max_{1\leq k\leq n}|\overline{R}_{k}^{ m}| \Bigr\Vert_{2}.\end{aligned}$$ By Doob’s maximal inequality for martingales and by stationarity, we conclude that $$\frac{{1}}{\sqrt{n}}\Bigl\Vert \max_{1\leq k\leq n}|M_{k}^{m}-M_{k}|
\Bigr\Vert _{2}\leq
\Vert D_{0}^{m}-D_{0}\Vert _{2}.$$ For $m$ fixed, since $(\theta_{k}^{m})_{k\in Z}$ is a stationary sequence of square-integrable random variables, for any $A>0$, we have$$\begin{aligned}
\frac{{1}}{n}\mathbb{E}\Bigl[\max_{1\leq k\leq n}|\theta_{k}^{m}|^{2}\Bigr]
&\leq&\frac{A^{2}}{n}+\frac{1}{n}\sum_{k=1}^{n}\mathbb{E[}|\theta
_{k}^{m}|^{2}I(|\theta_{k}^{m}|>A)] \\
&=&\frac{A^{2}}{n}+\mathbb{E[}|\theta_{0}^{m}|^{2}I(|\theta_{0}^{m}|>A)]\end{aligned}$$ and then, clearly,$$\lim_{n\rightarrow\infty}\frac{{1}}{n}\mathbb{E}\Bigl[\max_{1\leq k\leq
n}|\theta_{k}^{m}|^{2}\Bigr]=0. \label{maxnegl}$$ Then, taking into account [(\[rests\])]{}, we easily obtain$$\lim\sup_{n\rightarrow\infty}\frac{{1}}{\sqrt{n}}\Bigl\Vert \max_{1\leq
k\leq
n}|S_{k}-M_{k}| \Bigr\Vert_{2}\leq\Vert D_{0}^{m}-D_{0}\Vert _{2}+\Vert Y_{0}^{m}\Vert _{M^{+}}$$ and the result follows by letting $m\rightarrow\infty$, from the fact that $D_{0}^{m}\rightarrow D_{0}$ in $\mathbb{L}_{2}$. It is easy to see that the martingale is unique.
*Step Necessity*.
Assume that the martingale approximation (\[maxcond\]) holds. With the notation $R_{n}=S_{n}-M_{n},$ we then have $$\lim_{n\rightarrow\infty}\frac{{1}}{\sqrt{n}}\Bigl\Vert \max_{1\leq k\leq
n}|R_{k}| \Bigr\Vert _{2}=0.$$ In particular, this approximation implies that $$\lim_{n\rightarrow\infty}\frac{1}{\sqrt{n}}\max_{1\leq k\leq
n}\Vert \mathbb E(S_{k}|\mathcal{F}_{0})\Vert _{2}=0. \label{natural}$$ From $$\Vert \overline{R}_{n}^{ n}\Vert _{2}\leq\Vert \mathbb
E(S_{n}|\mathcal{F}_{0})\Vert _{2},$$ we deduce that $$\Vert R_{n}^{n}\Vert _{2}=\Vert \theta_{0}^{n}-\theta_{n}^{n}+\overline{R}_{n}^{ n}\Vert _{2}\leq2\Vert \theta_{0}^{n}\Vert _{2}+\Vert \overline{R}_{n}^{ n}\Vert _{2}\leq3\max_{1\leq k\leq n}\Vert \mathbb E(S_{k}|\mathcal
{F}_{0})\Vert _{2},$$ whence, by (\[natural\]), it follows that $$\lim_{n\rightarrow\infty}\frac{\Vert R_{n}^{n}\Vert _{2}}{\sqrt{n}}=0.$$
As a consequence, we obtain$$\mathbb{E}(D_{0}^{n}-D_{0})^{2}=\frac{\mathbb
{E}(M_{n}^{n}-M_{n})^{2}}{n}=\frac{\mathbb{E}(R_{n}^{n}-R_{n})^{2}}{n}\rightarrow0\qquad\mbox{as }
n\rightarrow\infty.$$ This shows that $D_{0}^{n}\rightarrow D_{0}$ in $\mathbb{L}_{2}.$ By the triangle inequality, followed by Doob’s inequality, for any positive integer $m$, we have $$\begin{aligned}
\frac{{1}}{\sqrt{n}}\Bigl\Vert \max_{1\leq k\leq n}|R_{k}^{m}| \Bigr\Vert _{2}
&\leq&
\frac{{1}}{\sqrt{n}}\Bigl\Vert \max_{1\leq k\leq n}|R_{k}| \Bigr\Vert _{2}+\frac
{{1}}{\sqrt{n}}\Bigl\Vert \max_{1\leq k\leq n}|M_{k}^{m}-M_{k}| \Bigr\Vert_{2} \\
&\leq&\frac{{1}}{\sqrt{n}}\Bigl\Vert \max_{1\leq k\leq n}|R_{k}| \Bigr\Vert _{2}+\Vert D_{0}^{m}-D_{0}\Vert .\end{aligned}$$ Now, letting $n\rightarrow\infty$ followed by $m\rightarrow\infty,$ we obtain $$\lim_{m\rightarrow\infty}\lim\sup_{n\rightarrow\infty}\frac
{{1}}{\sqrt{n}}\Bigl\Vert \max_{1\leq k\leq n}|R_{k}^{m}| \Bigr\Vert _{2}=0. \label{rest1}$$ Now, observe that by [(\[rest\])]{}, $R_{n}^{m}-\overline{R}_{n}^{
m}=\theta
_{0}^{m}-\theta_{n}^{m}$. Then, for every fixed $m,$ by (\[maxnegl\]), we have $$\frac{{1}}{\sqrt{n}}\Bigl\Vert \max_{1\leq k\leq n}|\theta_{0}^{m}-\theta
_{k}^{m}| \Bigr\Vert _{2}\mathop{\rightarrow}_{n\rightarrow\infty}0.$$ Thus, we conclude from [(\[rest1\])]{} that $$\lim_{m\rightarrow\infty}{\Vert }Y_{0}^{ m}{\Vert }_{M^{+}}=0$$ and the necessity follows.
Applications
============
Applications using projective criteria
--------------------------------------
The first application involves the class of variables satisfying the Maxwell–Woodroofe condition [@mw].
\[underMW\] Assume that $$\Delta(X_{0})=\sum_{k=1}^{\infty}\frac{\Vert \mathbb
{E}_{0}(S_{k})\Vert _{2}}{k^{3/2}}<\infty. \label{MW}$$ The martingale approximation [(\[maxcond\])]{} then holds.
In order to verify condition (\[MAX\]) of Theorem \[T\], we apply inequality (\[PU\]) to the stationary sequence $(Y_{k}^{m})_{k\in
\mathbb{Z}}$ defined by (\[defY\]). Then $$\Biggl\Vert \max_{1\leq j\leq n}\Biggl|\sum_{k=0}^{j-1}Y_{k}^{m}\Biggr|
\Biggr\Vert_{2}\leq n^{1/2}\bigl(2\Vert Y_{0}^{m}\Vert _{2}+80\Delta(Y_{0}^{m})\bigr).$$ First, note that by Peligrad and Utev [@PU1], Proposition 2.5, we know that condition (\[MW\]) implies that $\Vert Y_{0}^{m}\Vert _{2}\rightarrow
0.$ We complete the proof by showing that $$\Delta(Y_{0}^{m})\mathop{\longrightarrow}_{m\rightarrow\infty}0.$$ Since $\Vert Y_{0}^{m}\Vert _{2}\rightarrow0,$ by the triangle inequality and stationarity, every term of the series on the right-hand side of the equality $$\Delta(Y_{0}^{m})=\sum_{k=1}^{\infty}\frac{1}{k^{3/2}}\Vert \mathbb{E}_{0}(Y_{0}^{m}+\cdots+Y_{k-1}^{m})\Vert _{2}$$ tends to $0$ as $m\rightarrow\infty.$ Furthermore, because $$\begin{aligned}
\Vert \mathbb{E}_{0}(Y_{0}^{m}+\cdots +Y_{k-1}^{m})\Vert _{2}&=&\Biggl\Vert\mathbb{E}_{0}\Biggl(\frac{1}{m}\sum_{l=1}^{m}\sum_{i=0}^{k-1}\mathbb
{E}_{i}(X_{i+l})\Biggr)\Biggr\Vert _{2} \\
&\leq&\Vert \mathbb{E}_{0}(X_{0}+\cdots+X_{k-1})\Vert _{2},\end{aligned}$$ each term in $\Delta(Y_{0}^{m})$ is dominated by the corresponding term in $\Delta(X_{0})$, the latter being independent of $m$. The result follows from the above considerations, along with the Lebesgue dominated convergence theorem for the counting measure.
For the sake of applications, we give the following corollary.
Assume that$$\sum_{n=1}^{\infty}\frac{1}{\sqrt{n}}\Vert\mathbb{E}_{0}(X_{n})\Vert
_{2}<\infty. \label{mixg1}$$ The martingale representation [(\[maxcond\])]{} then holds.
The fact that (\[mixg1\]) implies (\[MW\]) was observed in Maxwell and Woodroofe [@mw].
We shall now combine Theorem \[T\] with Rio’s maximal inequality (\[Rio\]) to obtain the following proposition.
\[underRio\]Assume that for any $j\geq0$,$$\Gamma_{j}=\sum_{k\geq j}\Vert X_{j}\mathbb{E}_{0}(X_{k})\Vert _{1}<\infty\quad \mbox{and}\quad \frac{1}{m}\sum_{j=0}^{m-1}\Gamma_{j}\rightarrow0
\qquad\mbox{as } m \rightarrow\infty. \label{condRio}$$ The martingale representation [(\[maxcond\])]{} then holds.
In order to verify condition (\[MAX\]), we now apply the maximal inequality (\[Rio\]) to $(Y_{k}^{m})_{k\geq1}$ defined by (\[defY\]). We conclude that for $n\geq m$, $$\begin{aligned}
\Biggl\Vert \max_{1\leq j\leq n}\Biggl|\sum_{k=0}^{j-1}Y_{k}^{m}\Biggr|\Biggr \Vert
_{2}^{2}&\leq&
8n\Vert Y_{0}^{m}\Vert _{2}^{2}+16\sum_{j=1}^{n-1}\Vert Y_{0}^{m}\mathbb{E}_{0}(Y_{1}^{m}+\cdots+Y_{j}^{m})\Vert _{1} \\
&\leq&8n(12m+1)\Vert Y_{0}^{m}\Vert _{2}^{2}+16\sum
_{j=m+1}^{n-1}\Vert Y_{0}^{m}\mathbb{E}_{0}(Y_{m+1}^{m}+\cdots+Y_{j}^{m})\Vert _{1},\end{aligned}$$ where, in the last sum, we have implemented a decomposition into two terms to deal with overlapping blocks. So, for an absolute constant $C$, $$\frac{1}{n}\Biggl\Vert \max_{1\leq j\leq n}\Biggl|\sum_{k=0}^{j-1}Y_{k}^{m}\Biggr| \Biggr\Vert _{2}^{2}\leq C\Biggl(\frac{\Vert \mathbb{E}_{0}(S_{m})\Vert _{2}^{2}}{m}+\frac
{1}{n}\sum_{l=m+1}^{n-1}\Vert Y_{0}^{m}\mathbb{E}_{0}(Y_{m+1}^{m}+\cdots
+Y_{l}^{m})\Vert _{1}\Biggr).$$ Since, for any $l>m$, $$\begin{aligned}
\Vert Y_{0}^{m}\mathbb{E}_{0}(Y_{m+1}^{m}+\cdots+Y_{l}^{m})\Vert _{1}& \leq&\frac
{1}{m}\sum_{j=1}^{m}\sup_{i>m}\Vert (\mathbb{E}_{0}(X_{j}))\mathbb{E}_{0}(X_{i}+\cdots+X_{i+l})\Vert _{1} \\
&\leq&\frac{1}{m}\sum_{j=1}^{m}\sum_{k\geq m}\Vert \mathbb
{E}_{0}(X_{j})\mathbb{E}_{0}(X_{k})\Vert _{1}\end{aligned}$$ and also $$\Vert \mathbb{E}_{0}(S_{m})\Vert _{2}^{2}\leq
2\sum _{j=0}^{m-1}\sum _{k=j}^{m-1}\Vert \mathbb{E}_{0}(X_{j})\mathbb{E}_{0}(X_{k})\Vert _{1},$$ we then obtain, by the properties of conditional expectations, that for a certain absolute constant $C^{\prime}$, $$\frac{1}{n}\Biggl\Vert \max_{1\leq j\leq n}\Biggl|\sum
_{k=0}^{j-1}Y_{k}^{m}\Biggr| \Biggr\Vert _{2}^{2} \leq \frac{C^{\prime}}{m}\sum
_{j=0}^{m}\sum_{k\geq j}\Vert X_{j}\mathbb{E}_{0}(X_{k})\Vert _{1}$$ and the result follows from condition (\[condRio\]), by first letting $n\rightarrow\infty$, followed by $m\rightarrow\infty$.
The projective criteria in the next proposition were studied in [@hh; @HA; @GG], among others.
Assume $$\mathbb E(X_{0}|\mathcal{F}_{-\infty})=0\quad \mbox{almost surely
and}\quad
\sum_{i=1}^{\infty}\Vert\mathbb{E}_{-i}(X_{0})-\mathbb{E}_{-i-1}(X_{0})\Vert_{2}<\infty. \label{P}$$ The martingale approximation [(\[maxcond\])]{} then holds.
The validity of this proposition easily follows by verifying condition (\[MAX\]) via maximal inequality (\[DM\]) applied to $(Y_{k}^{m})_{k\geq1}$ defined by (\[defY\]). Indeed, by (\[DM\]), the triangle inequality and stationarity, we have $$\begin{aligned}
\frac{1}{\sqrt{n}}\Biggl\Vert \max_{1\leq j\leq n}\Biggl|\sum_{k=0}^{j-1}Y_{k}^{m}\Biggr|\Biggr\Vert_{2}&\leq&2\sum_{i=0}^{\infty}\Vert\mathbb{E}_{-i}(Y_{0}^{m})-\mathbb
{E}_{-i-1}(Y_{0}^{m})\Vert_{2} \\
&\leq&\frac{2}{m}\sum_{i=0}^{\infty}\sum_{k=1}^{m}\Vert\mathbb
{E}_{-i}(X_{k})-\mathbb{E}_{-i-1}(X_{k})\Vert_{2}.\end{aligned}$$ Now, by stationarity, change of order of summation and change of variable, $$\frac{1}{\sqrt{n}}\Biggl\Vert\max_{1\leq j\leq n}\Biggl|\sum_{k=0}^{j-1}Y_{k}^{m}\Biggr|
\Biggr\Vert_{2}\leq\frac{2}{m}\sum_{k=1}^{m}\sum_{j=k}^{\infty}\Vert\mathbb{E}_{-j}(X_{0})-\mathbb{E}_{-j-1}(X_{0})\Vert_{2}.$$ To verify condition (\[MAX\]), we let $n\rightarrow\infty$ followed by $m\rightarrow\infty.$ Note that the term on the right-hand side of the previous inequality tends to $0$ as $m\rightarrow\infty$, by (\[P\]).
Application to mixing sequences
-------------------------------
The results in the previous section can be immediately applied to mixing sequences, leading to the sharpest possible results and providing additional information about the structures of these processes. Examples include various classes of Markov chains and Gaussian processes.
We shall introduce the following mixing coefficients: for any two $\sigma$-algebras $\mathcal{A}$ and $\mathcal{B}$, define the strong mixing coefficient $\alpha(\mathcal{A},\mathcal{B)}$, $$\alpha(\mathcal{A},\mathcal{B)=}\sup\{|\mathbb{P}(A\cap B)-\mathbb{P}(A)
\mathbb{P}(B)|;A\in\mathcal{A},B\in\mathcal{B\}},$$ and the $\rho$-mixing coefficient, known also as the maximal coefficient of correlation $\rho(\mathcal{A},\mathcal{B})$, $$\rho(\mathcal{A},\mathcal{B})=\sup\{\operatorname{Cov}(X,Y)/\Vert X\Vert
_{2}\Vert
Y\Vert_{2}\dvtx X\in\mathbb{L}_{2}(\mathcal{A}),Y\in\mathbb{L}_{2}(
\mathcal{B})\}.$$
For the stationary sequence of random variables $(X_{k})_{k\in
\mathbb{Z}},$ we also define $\mathcal{F}_{m}^{n}$, the $\sigma$-field generated by $X_{i}$ with indices $m\leq i\leq n$. $\mathcal{F}^{n}$ denotes the $
\sigma$-field generated by $X_{i}$ with indices $i\geq n$ and $\mathcal
{F}_{m}$ denotes the $\sigma$-field generated by $X_{i}$ with indices $i\leq m.
$ The sequences of coefficients $\alpha(n)$ and $\rho(n)$ are then defined by $$\alpha(n)=\alpha(\mathcal{F}_{0},\mathcal{F}_{n}^{n})\quad \mbox{and}\quad \rho(n)=\rho(\mathcal{F}_{0},\mathcal{F}^{n}).$$ Equivalently (see [@rick], Chapter 4), $$\rho(n)=\sup\{\Vert\mathbb{E}(Y|\mathcal{F}_{0})\Vert_{2}/\Vert Y\Vert
_{2}\dvtx Y\in\mathbb{L}_{2}(\mathcal{F}^{n}),\mbox{ }\mathbb{E}(Y)=0\}.$$ Finally, we say that the stationary sequence is strongly mixing if $\alpha(n)\rightarrow0$ as $n\rightarrow\infty$ and $\rho$-mixing if $\rho(n)\rightarrow0$ as $n\rightarrow\infty$.
An interesting application of Proposition \[underMW\] is to $\rho$-mixing sequences. It is well known that the central limit theorem and its invariance principle hold for stationary centered sequences with finite second moments under the assumption $$\sum_{k=1}^{\infty}\rho(2^{k})<\infty, \label{condrho}$$ where $\rho(n)=\rho(\mathcal{F}_{0},\mathcal{F}^{n}\mathcal{)}$. Let us recall that the central limit theorem is due to [@II], while the invariance principle is found in [@pm; @shao; @u89; @u91]. The fact that condition (\[condrho\]) is sharp in this context is due to [@rick], Volume 1, page 367, and Volume 3, Theorem 34.13. Bradley’s example shows that if (\[condrho\]) fails, then $S_{n}/ \Vert S_{n}\Vert _2$ might have non-degenerate non-normal distributions as weak limit points.
As a corollary of Proposition \[underMW\], we obtain the conditional invariance principle for $\rho$-mixing sequences.
Assume $\sum_{k=1}^{\infty}\rho(2^{k})<\infty.$ The martingale representation [(\[maxcond\])]{} then holds.
As in [@mpu2], for a positive constant $C$, we have $$\sum_{r=0}^{\infty}\frac{\Vert
\mathbb{E}(S_{2^{r}}|\mathcal{F}_{0})\Vert_{2}}{2^{r/2}}\leq C
\Vert X_0\Vert _2\sum_{j=0}^{\infty}\rho(2^{j}).$$
To obtain sharp results for strongly mixing sequences, we shall use Proposition \[underRio\].
According to Doukhan, Massart and Rio [@DMR], a condition that is optimal for CLT or the invariance principle for strongly mixing sequences is $$\label{condalha}
\sum _{k\geq1}{\mathbb{E}}X_{0}^{2}I\bigl(|X_{0}|\geq
Q_{|X_{0}|}(2\alpha_{k})\bigr)<\infty,$$ where $Q_{|X_{0}|}$ denotes the cadlag inverse of the function $t\rightarrow P(|X_{0}|>t).$ Also under this condition, we add the additional information given by Theorem \[T\].
Assume that condition [(\[condalha\])]{} is satisfied. The martingale representation [(\[maxcond\])]{} then holds.
We shall just verify the condition of Proposition \[underRio\]. Note that on the set $[0,{\ P}(|Y|>0)]$, the function $H_{Y}\dvtx x\rightarrow
\int_{0}^{x}Q_{Y}(u)\,\mathrm{d}u$ is an absolutely continuous and increasing function with values in $[0,{\ E}|Y|]$. Denote by $G_{Y}$ the inverse of $H_{Y}$. With this notation, by Merlevède and Peligrad [@mp], relation (4.84), we have $$\Vert X_{j}\mathbb{E}(X_{k}|\mathcal{F}_{0})\Vert _{1}\leq3\int_{0}^{\Vert
{\mathbb{E}}(X_{k}|\mathcal{F}_{0})\Vert_{1}}Q_{|X_{0}|}\circ G(u)\,\mathrm{d}u$$ and we then majorize the right-hand side in the previous inequality by Dedecker and Doukhan [@dd], Proposition 1, to obtain $$\Vert X_{j}\mathbb{E}(X_{k}|\mathcal{F}_{0})\Vert _{1}\leq6\int_{0}^{2\alpha
(k)}Q_{|X_{0}|}^{2}\,\mathrm{d}u .$$ Therefore,$$\begin{aligned}
\sum_{k\geq j}\Vert X_{j}\mathbb{E}_{0}(X_{k})\Vert _{1}&\leq&6\sum_{k\geq
j}\int_{0}^{2\alpha(k)}Q_{|X_{0}|}^{2}\,\mathrm{d}u \\
&\leq&6\sum_{k\geq j}{\mathbb{E}}X_{0}^{2}I\bigl(|X_{0}|\geq Q_{|X_{0}|}(2\alpha
_{k})\bigr) \rightarrow0\qquad \mbox{as }j\rightarrow\infty
.\end{aligned}$$
Note that the coefficient $\alpha(k)$ is defined by using only one variable in the future. Moreover, by the Cauchy–Schwarz inequality, condition (\[condalha\]) is satisfied if the variables have finite moments of order $2+\delta$ for a $\delta>0$ and $$\sum_{k\geq1}\alpha(k)^{\delta/(2+\delta)}<\infty.$$ An excellent source of information for classes of mixing sequences and classes of Markov chains satisfying mixing conditions is the book by Bradley [@rick]. Further applications can be obtained by using the coupling coefficients in [@dp].
Application to additive functionals of reversible Markov chains
---------------------------------------------------------------
For reversible Markov processes (i.e., $Q=Q^{\ast}$), the invariance principle under an optimal condition is known since Kipnis and Varadhan [@KV]. The following is a formulation in terms of martingale approximation.
Let $(\xi_{i})_{i\in\mathbb{Z}}$ be a stationary reversible Markov chain and $f\in\mathbb{L}_{0}^{2}(\pi)$ with the property $$\lim_{n\rightarrow\infty}\frac{\operatorname{var}(S_{n})}{n}\rightarrow\sigma
_{f}^{2}<\infty.
\label{revcondCLT}$$ The martingale approximation satisfying [(\[maxcond\])]{} then holds.
We have to verify condition [(\[MAX\])]{}. Denote by $\rho_{f}$ the spectral measure of $f$ corresponding to the self-adjoint operator $Q$ on $\mathbb{L}_{2}(\pi)$. It is well known that the assumption [(\[revcondCLT\])]{} for $f \in\mathbb L^2_0$ implies that $ \int _{-1}^{1}(1-t)^{-1}\rho_{f}(\mathrm{d}t)<\infty$ (see [@KV]). Define $Y_{0}^{m}$ by [(\[defY\])]{}. By the maximal inequality [(\[LW\])]{}, we have$$\frac{1}{n}\mathbb{E}\Biggl(\max_{1\leq j\leq n}\Biggl|\sum
_{k=0}^{j-1}Y_{k}^{m}\Biggr|\Biggr)^{2}\leq27\sum_{k\geq0}\mathbb{E(}Y_{0}^{m}Y_{k}^{m}),$$ provided that the sum on the right-hand side is finite. To prove it, by using spectral calculus for the self-adjoint operator $Q$, we obtain $$\sum_{k\geq0}\mathbb{E(}Y_{0}^{m}Y_{k}^{m})\leq\frac{1}{m^{2}}\int _{-1}^{1}\frac{(1+t+\cdots+t^{m-1})^{2}}{(1-t)}\rho_{f}(\mathrm{d}t)$$ and therefore, for every positive integer $m>0,$$$\Vert Y_{0}^{m}\Vert _{M^{+}}^{2}\leq27\int _{-1}^{1}\frac{(1+t+\cdots+t^{m-1})^{2}}{m^{2}(1-t)}\rho_{f}(\mathrm{d}t).$$ Since $\int _{-1}^{1}(1-t)^{-1}\rho_{f}(\mathrm{d}t)<\infty$, the right-hand side is finite and, by the dominated convergence theorem,$$\lim_{m\rightarrow\infty}\Vert Y_{0}^{m}\Vert _{M^{+}}^{2}=0.$$
Similar results are expected to hold for other classes of stationary and ergodic Markov chains when $Q~$is not necessarily self-adjoint, but instead satisfies a quasi-symmetry or strong sector condition, or is symmetrized. See [@wu] and [@svy] for these related processes.
Application to additive functionals of normal Markov chains
-----------------------------------------------------------
For additive functionals of normal Markov chains $(QQ^{\ast}=Q^{\ast
}Q),$ the central limit theorem below is a result of Gordin and Lifshitz [@gl]. As an application of Theorem \[T2\], we give an alternative proof.
Let $\rho_{f}$ be the spectral measure on the closed unit disk $D
\subset\mathbb C$ corresponding to the function $f\in\mathbb{L}_{0}^{2}(\pi)$.
\[normal\] Let $(\xi_{i})_{i\in\mathbb{Z}}$ be a stationary normal Markov chain and a function $f\in\mathbb{L}_{0}^{2}(\pi),$ satisfying the condition $$\int _{D}\frac{1}{|1-z|}\rho_{f}(\mathrm{d}z)<\infty.
\label{normcond}$$ The martingale approximation [(\[MA\])]{} then holds.
According to Theorem \[T2\], we have to verify condition (\[MA1\]). By using spectral calculus as in [@1994], Chapter 4, after some computations, we get$$\lim\sup_{n\rightarrow\infty}\frac{1}{n}\Biggl\Vert \sum_{k=0}^{n-1}Y_{k}^{m}\Biggr\Vert _{2}^{2}\leq4\int _{D}\frac{|1+z+\cdots+z^{m-1}|^{2}}{m^{2}|1-z|}\rho_{f}(\mathrm{d}z)$$ and condition (\[MA1\]) is therefore satisfied by condition (\[normcond\]) and the dominated convergence theorem.
Condition [(\[normcond\])]{} has an interesting equivalent formulation in terms of conditional moments that is in the spirit of (and which implies) the Mawxell–Woodroofe condition [(\[MW\])]{}.
Condition [(\[normcond\])]{} is equivalent to $$\sum_{k=1}^{\infty}\frac{\Vert \mathbb
{E}_{0}(S_{k})\Vert _{2}^{2}}{k^{2}}<\infty. \label{Gap}$$
Condition [(\[Gap\])]{} is further implied by $$\sum_{k=1}^{\infty}\Vert \mathbb{E}_{0}(X_{k})\Vert _{2}^{2}<\infty.
\label{cor2}$$
The equivalence in the above remark can be found in [@Cuny], Lemma 2.1. The fact that [(\[cor2\])]{} implies [(\[Gap\])]{} is easily established, much like the proof that [(\[mixg1\])]{} implies [(\[MW\])]{}.
Acknowledgements {#acknowledgements .unnumbered}
================
Mikhail Gordin was supported in part by a Charles Phelps Taft Memorial Fund grant and RFBR Grant 10-01-00242\_a. Magda Peligrad was supported in part by a Charles Phelps Taft Memorial Fund grant and NSA Grant H98230-09-1-0005. The authors are grateful to the referees for carefully reading the paper and for numerous suggestions that improved the presentation of the paper.
[99]{} Billingsley, P. (1968). *Convergence of Probability Measures*. New York: Wiley.
Bradley, R.C. (2007). *Introduction to Strong Mixing Conditions*, Vols 1–3. Heber City, UT: Kendrick Press.
Borodin, A.N. and Ibragimov, I.A. (1994). Limit theorems for functionals of random walks. *Trudy Mat. Inst. Steklov.* **195** 286. Transl. in English: *Proc. Steklov Inst. Math.* **195** (1995).
Cuny, C. (2009). Pointwise ergodic theorems with rate and application to limit theorems for stationary processes. Available at [arXiv:0904.0185v1](http://arXiv.org/abs/0904.0185v1).
Dedecker, J. and Doukhan, P. (2003). A new covariance inequality and applications. *Stochastic Process. Appl.* **106** 63–80.
Dedecker, J. and Merlevède, F. (2002). Necessary and sufficient conditions for the conditional central limit theorem. *Ann. Probab.* **30** 1044–1081.
Dedecker, J., Merlevède, F. and Volný, D. (2007). On the weak invariance principle for non-adapted stationary sequences under projective criteria. *J. Theoret. Probab.* **20** 971–1004.
Dedecker, J. and Prieur, C. (2005). New dependence coefficients. Examples and applications to statistics. *Probab. Theory Related Fields* **132** 203–236.
Dedecker, J. and Rio, E. (2000). On the functional central limit theorem for stationary processes. *Ann. Inst. H. Poincaré Probab. Statist.* **36** 1–34.
Doukhan, P., Massart, P. and Rio, E. (1994). The functional central limit theorem for strongly mixing processes. *Ann. Inst. H. Poincaré Probab. Statist.* **30** 63–82.
Hall, P. and Heyde, C.C. (1980). *Martingale Limit Theory and Its Application*. New York: Academic Press.
Heyde, C.C. (1974). On the central limit theorem for stationary processes. *Z. Wahrsch. Verw. Gebiete.* **30** 315–320.
Hannan, E.J. (1979). The central limit theorem for time series regression. *Stochastic Process. Appl.* **9** 281–289.
Ibragimov, I.A. (1975). A note on the central limit theorem for dependent variables. *Theory Probab. Appl.* **20** 135–140.
Gordin, M.I. (1969). The central limit theorem for stationary processes. *Soviet. Math. Dokl.* **10** 1174–1176.
Gordin, M.I. (2004). A remark on the martingale method for proving the central limit theorem for stationary sequences. *Zap. Nauchn. Sem. S.-Peterburg. Otdel. Mat. Inst. Steklov (POMI)* **311**. *Veroyatn. i Stat.* **7** 124–132, 299–300. Transl.: *J. Math. Sci. (N.Y.)* **133** (2006) 1277–1281.
Gordin, M.I. and Lifshitz, B. (1981). A remark about a Markov process with normal transition operator. In *Third Vilnius Conf. Probab. Stat.* **1** 147–148. Vilnius: Akad. Nauk Litovsk.
Kipnis, C. and Varadhan, S.R.S. (1986). Central limit theorem for additive functionals of reversible Markov processes and applications to simple exclusions. *Comm. Math. Phys.* **104** 1–19.
Maxwell, M. and Woodroofe, M. (2000). Central limit theorems for additive functionals of Markov chains. *Ann. Probab.* **28** 713–724.
Merlevède, F. and Peligrad, M. (2006). On the weak invariance principle for stationary sequences under projective criteria. *J. Theoret. Probab.* **19** 647–689.
Merlevède, F., Peligrad, M. and Utev, S. (2006). Recent advances in invariance principles for stationary sequences. *Probab. Surv.* **3** 1–36.
Peligrad, M. (1982). Invariance principle for mixing sequences of random variables. *Ann. Probab.* **10** 968–981.
Peligrad, M. (1999). Convergence of stopped sums of weakly dependent random variables. *Electron. J. Probab.* **4** 1–13.
Peligrad, M. and Utev, S. (2005). A new maximal inequality and invariance principle for stationary sequences. *Ann. Probab.* **33** 798–815.
Peligrad, M., Utev, S. and Wu, W.B. (2007). A maximal $L_p$-inequality for stationary sequences and its applications. *Proc. Amer. Math. Soc.* **135** 541–550.
Rio, E. (2000). *Théorie asymptotique des processus aléatoires faiblement dépendants*, Mathématiques & Applications **31**. Berlin: Springer.
Rootzén, H. (1976). Fluctuations of sequences which converge in distribution. *Ann. Probab.* **4** 456–463.
Sethuraman, S., Varadhan, S.R.S. and Yau, H.T. (2000). Difusive limit of a tagged particle in asymmetric simple exclusion processes. *Comm. Pure Appl. Math.* **53** 972–1006.
Shao, Q. (1989). On the invariance principle for stationary $\rho$-mixing sequences of random variables. *Chinese Ann. Math. Ser. B* **10B** 427–433.
Utev, S.A. (1989). Sums of random variables with $\varphi$-mixing. *Trudy Inst. Mat.* **1** 78–100.
Utev, S.A. (1991). Sums of random variables with $\varphi$-mixing. *Siberian Adv. Math.* **1** 124–155.
Zhao, O. and Woodroofe, M. (2008). On martingale approximations. *Ann. Appl. Probab.* **18** 1831–1847.
Wu, L. (1999). Forward–backward martingale decomposition and compactness results for additive functionals of stationary ergodic Markov processes. *Ann. Inst. H. Poincaré Probab. Statist.* **35** 121–141.
Wu, W.B. (2007). Strong invariance principles for dependent random variables. *Ann. Probab.* **35** 2294–2320.
Wu, W.B. and Woodroofe, M. (2004). Martingale approximations for sums of stationary processes. *Ann. Probab.* **32** 1674–1690.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Light field imaging is limited in its computational processing demands of high sampling for both spatial and angular dimensions. Single-shot light field cameras sacrifice spatial resolution to sample angular viewpoints, typically by multiplexing incoming rays onto a 2D sensor array. While this resolution can be recovered using compressive sensing, these iterative solutions are slow in processing a light field. We present a deep learning approach using a new, two branch network architecture, consisting jointly of an autoencoder and a 4D CNN, to recover a high resolution 4D light field from a single coded 2D image. This network decreases reconstruction time significantly while achieving average PSNR values of 26-32 dB on a variety of light fields. In particular, reconstruction time is decreased from 35 minutes to 6.7 minutes as compared to the dictionary method for equivalent visual quality. These reconstructions are performed at small sampling/compression ratios as low as 8%, allowing for cheaper coded light field cameras. We test our network reconstructions on synthetic light fields, simulated coded measurements of real light fields captured from a Lytro Illum camera, and real coded images from a custom CMOS diffractive light field camera. The combination of compressive light field capture with deep learning allows the potential for real-time light field video acquisition systems in the future.'
author:
- |
Mayank Gupta\
Arizona State University\
- |
Arjun Jauhari\
Cornell University\
- |
Kuldeep Kulkarni\
Arizona State University\
- |
Suren Jayasuriya\
Carnegie Mellon University\
- |
Alyosha Molnar\
Cornell University\
- |
Pavan Turaga\
Arizona State University\
bibliography:
- 'lfreconnet\_review.bib'
title: Compressive Light Field Reconstructions using Deep Learning
---
Introduction
============
Related Work
============
Light Field Photography
=======================
Deep Learning for Light Field Reconstruction
============================================
Experimental Results
====================
Discussion
==========
**Acknowledgements:** The authors would like to thank the anonymous reviewers for their detailed feedback, Siva Sankalp for running some experiments, and Mark Buckler for GPU computing support. AJ was supported by a gift from Qualcomm. KK and PT were partially supported by NSF CAREER grant 1451263. SJ was supported by a NSF Graduate Research Fellowship and a Qualcomm Innovation Fellowship.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'A test particle of mass $\mu$ on a bound geodesic of a Kerr black hole of mass $M \gg \mu$ will slowly inspiral as gravitational radiation extracts energy and angular momentum from its orbit. This inspiral can be considered [*adiabatic*]{} when the orbital period is much shorter than the timescale on which energy is radiated, and [*quasicircular*]{} when the radial velocity is much less than the azimuthal velocity. Although the inspiral always remains adiabatic provided $\mu \ll M$, the quasicircular approximation breaks down as the particle approaches the innermost stable circular orbit (ISCO). In this paper, we relax the quasicircular approximation and solve the radial equation of motion explicitly near the ISCO. We use the requirement that the test particle’s 4-velocity remains properly normalized to calculate a new contribution to the difference between its energy and angular momentum. This difference determines how a black hole’s spin changes following a test-particle merger, and can be extrapolated to help predict the mass and spin of the final black hole produced in finite-mass-ratio black-hole mergers. Our new contribution is particularly important for nearly maximally spinning black holes, as it can affect whether a merger produces a naked singularity.'
author:
- Michael Kesden
date: January 2011
title: Transition from adiabatic inspiral to plunge into a spinning black hole
---
Introduction {#S:intro}
============
Supermassive black holes (SBHs) with masses $10^6 M_\odot \lesssim M
\lesssim 10^{10} M_\odot$ reside at the centers of most large galaxies [@Kormendy:1995er]. These SBHs will be surrounded by dense cusps of stars [@Bahcall:1976aa], some fraction of which will consist of compact objects (white dwarfs, neutron stars, or stellar-mass black holes) of mass $\mu \sim 1 - 10~M_\odot$. Compact objects whose orbital velocities lie within a “loss cone” about the radial direction [@Frank:1976uy] will inspiral into the SBHs under the influence of gravitational radiation [@Sigurdsson:1996uz]. These extreme-mass-ratio inspirals (EMRIs) will be an important source of gravitational waves (GWs) for the proposed space-based GW detector LISA [@LISA]. LISA will rely on matched filtering to detect EMRIs, requiring templates that accurately track the phase of the GWs when their fundamental frequency $f$ is in the range $10^{-4} \lesssim
f \lesssim 10^{-1}$ Hz. These GW frequencies correspond to orbital frequencies $\Omega = \pi f$ at the innermost stable circular orbits (ISCOs) of SBHs with masses $10^5 M_\odot \lesssim M \lesssim 10^7
M_\odot$. To maximize the number of EMRIs detected by LISA, it is therefore essential to understand the trajectories of compact objects as they inspiral all the way to the ISCO.
In the absence of gravitational radiation, test particles travel on geodesics of the Kerr metric $g_{\mu\nu}$ [@Kerr:1963ud] that describes the spacetime of spinning black holes. Kerr geodesics are characterized by four constants of motion: the particle’s rest mass $\mu$, energy $E$, $z$ component of angular momentum $L_z$, and Carter constant $Q$ [@Carter:1968rr]. These constants can be determined from the particle’s 4-momentum $p^\mu$ and the Kerr metric’s timelike Killing vector $T_\mu$, azimuthal Killing vector $\Phi_\mu$, and Killing tensor $Q_{\mu\nu}$ [@Walker:1970un]:
\[E:4con\] $$\begin{aligned}
\label{E:masscon}
\mu &=& \sqrt{-g_{\mu\nu} p^\mu p^\nu}~, \\ \label{E:Econ}
E &=& -T_\mu p^\mu~, \\ \label{E:Lcon}
L_z &=& \Phi_\mu p^\mu~, \\ \label{E:Qcon}
Q &=& Q_{\mu\nu} p^\mu p^\nu~.
\end{aligned}$$
Equatorial geodesics have $Q = 0$; for computational simplicity we will restrict our attention to equatorial orbits for the remainder of this paper.
In the test-particle limit $\eta \equiv \mu/M \ll 1$, the stress-energy tensor of a test particle moving on a Kerr geodesic sources gravitational radiation that can be calculated using black-hole perturbation theory [@Teukolsky:1973ha]. This radiation extracts energy and angular momentum from the orbit, causing the particle to migrate through the phase space $\{ E, L_z \}$. The GW energy flux $\dot{E}_{\rm GW}$ and angular momentum flux $\dot{L_z}_{\rm GW}$ (an overdot symbolizes a derivative with respect to Boyer-Lindquist coordinate time) are proportional to $\eta^2$, implying that the timescale $t_{\rm GW} \sim E/\dot{E}_{\rm GW}$ is proportional to $\eta^{-1}$. The orbital period $t_{\rm orb} \sim
\Omega^{-1}$ is independent of $\eta$ to lowest order, implying that the inspiral will always be [*adiabatic*]{} ($t_{\rm GW} \gg t_{\rm
orb}$) for sufficiently small $\eta$. Gravitational radiation circularizes eccentric orbits in the post-Newtonian regime [@Peters:1963ux], implying that the early inspiral will also be [*quasicircular*]{} ($\dot{r} \ll r\Omega$). During this adiabatic, quasicircular portion of the inspiral, the trajectory of the test particle is well described by a sequence of circular geodesics of ever decreasing Boyer-Lindquist radius $r$. In the quasicircular approximation, the radial velocity is given by $$\label{E:adRV}
\dot{r} = -\frac{\dot{E}_{\rm GW}}{dE/dr}~,$$ where $dE/dr$ is the derivative of the energy of a circular equatorial geodesic with respect to its Boyer-Lindquist radius. This radial velocity is proportional to $\eta$, and approaches negative infinity as $r$ approaches $r_{\rm ISCO}$ where $dE/dr$ vanishes. This behavior is unphysical, and reflects the breakdown of the quasicircular approximation as the test particle approaches the ISCO.
In the vicinity of the ISCO, the test particle transitions from the quasicircular inspiral described above to a “captured” plunge that crosses the event horizon [@Bardeen:1972fi]. This [*transition regime*]{} has been investigated in the test-particle limit by Ori and Thorne [@Ori:2000zn] (hereafter OT) and for nonspinning black holes or arbitrary mass ratio by Buonanno and Damour [@Buonanno:2000ef]. Sundararajan [@Sundararajan:2008bw] generalized the approach of OT to inclined and eccentric orbits, deriving trajectories that served as sources for the gravitational waveforms calculated in Sundararajan, Khanna, and Hughes [@Sundararajan:2010sr]. We will adopt the notation of OT throughout this paper to facilitate comparisons between our results and theirs. In Sec. \[S:prev\], we will review the previous treatment of the transition regime and the discovery of scaling relations that can be used to apply a universal dimensionless trajectory to mergers with arbitrary mass ratio and black-hole spin. In Sec. \[S:4v\], we show that the test particle’s 4-momentum $p^\mu$ is not properly normalized according to Eq. (\[E:masscon\]) in this model of the transition. We develop a new model that satisfies this requirement by allowing the particle’s energy and angular momentum to vary independently. We then explore this new model’s predictions for the behavior of the transition regime for nearly maximal black-hole spins in Sec. \[S:maxspins\]. A summary of our principle findings and their implications is given in Sec. \[S:disc\].
Previous Treatment {#S:prev}
==================
Geodesic Motion {#SS:geo}
---------------
We begin by reviewing the motion of test particles on Kerr geodesics. In the equatorial plane ($\theta = \pi/2$), the Kerr metric for a black hole of mass $M$ and spin $a$ can be written in Boyer-Lindquist coordinates [@Boyer:1966qh] as $$\begin{gathered}
\label{E:KerrEq}
ds^2 = -\left( 1 - \frac{2M}{r} \right) dt^2
+ \left( 1 - \frac{2M}{r} + \frac{a^2}{r^2} \right)^{-1} dr^2 \\
+ r^2 d\theta^2
+ \left (r^2 + a^2 + \frac{2Ma^2}{r} \right) d\phi^2
- \frac{4Ma}{r} dt d\phi~.\end{gathered}$$ Here and throughout this paper we use units in which Newton’s constant $G$ and the speed of light $c$ are unity. We can also define a dimensionless radius ${\Tilde{r}}\equiv r/M$, coordinate time ${\Tilde{t}}\equiv
t/M$, and spin ${\Tilde{a}}\equiv a/M$. In these coordinates, the three constants of motion given in Eqs. (\[E:masscon\]) through (\[E:Lcon\]) provide three equations for the evolution of ${\Tilde{t}}$, ${\Tilde{r}}$, and $\phi$ as functions of the dimensionless proper time ${\Tilde{\tau}}\equiv \tau/M$ along the particle’s worldline. Eq. (\[E:masscon\]) can be rewritten as $$\begin{aligned}
\label{E:4norm}
1 &=& -g_{\mu\nu} \frac{d{\Tilde{x}}^\mu}{d{\Tilde{\tau}}} \frac{d{\Tilde{x}}^\nu}{d{\Tilde{\tau}}}
\nonumber \\
&=& \left( 1 - \frac{2}{{\Tilde{r}}} \right) \left( \frac{d{\Tilde{t}}}{d{\Tilde{\tau}}} \right)^2
- \left( 1 - \frac{2}{{\Tilde{r}}} + \frac{{\Tilde{a}}^2}{{\Tilde{r}}^2} \right)^{-1}
\left( \frac{d{\Tilde{r}}}{d{\Tilde{\tau}}} \right)^2 \nonumber \\
&& - \left ({\Tilde{r}}^2 + {\Tilde{a}}^2 + \frac{2{\Tilde{a}}^2}{{\Tilde{r}}} \right)
\left( \frac{d\phi}{d{\Tilde{\tau}}} \right)^2 + \frac{4{\Tilde{a}}}{{\Tilde{r}}}
\frac{d{\Tilde{t}}}{d{\Tilde{\tau}}} \frac{d\phi}{d{\Tilde{\tau}}}~,\end{aligned}$$ while according to Eqs. (\[E:Econ\]) and (\[E:Lcon\]), the dimensionless energy ${\Tilde{E}}\equiv E/\mu$, and angular momentum ${\Tilde{L}}\equiv L_z/\mu M$ are given by
\[E:DEL\] $$\begin{aligned}
\label{E:DE}
{\Tilde{E}}&=& \left( 1 - \frac{2}{{\Tilde{r}}} \right) \frac{d{\Tilde{t}}}{d{\Tilde{\tau}}}
+ \frac{2{\Tilde{a}}}{{\Tilde{r}}} \frac{d\phi}{d{\Tilde{\tau}}}~, \\ \label{E:DL}
{\Tilde{L}}&=& \left ({\Tilde{r}}^2 + {\Tilde{a}}^2 + \frac{2{\Tilde{a}}^2}{{\Tilde{r}}} \right)
\frac{d\phi}{d{\Tilde{\tau}}} - \frac{2{\Tilde{a}}}{{\Tilde{r}}} \frac{d{\Tilde{t}}}{d{\Tilde{\tau}}}~.
\end{aligned}$$
Solving Eqs. (\[E:DE\]) and (\[E:DL\]) for $d{\Tilde{t}}/d{\Tilde{\tau}}$ and $d\phi/d{\Tilde{\tau}}$, inserting the result into Eq. (\[E:4norm\]), then solving for $(d{\Tilde{r}}/d{\Tilde{\tau}})^2$ yields $$\label{E:RV2}
\left( \frac{d{\Tilde{r}}}{d{\Tilde{\tau}}} \right)^2 = {\Tilde{E}}^2 - V({\Tilde{r}}, {\Tilde{E}}, {\Tilde{L}})~,$$ where $$\label{E:pot}
V({\Tilde{r}}, {\Tilde{E}}, {\Tilde{L}}) \equiv 1 - \frac{2}{{\Tilde{r}}}
+ \frac{{\Tilde{L}}^2 + {\Tilde{a}}^2 - {\Tilde{E}}^2 {\Tilde{a}}^2}{{\Tilde{r}}^2} - \frac{2({\Tilde{L}}- {\Tilde{E}}{\Tilde{a}})^2}{{\Tilde{r}}^3}$$ is the effective potential.
Geodesic motion is alternatively described by the geodesic equations $$\label{E:GE}
\frac{d^2{\Tilde{x}}^\mu}{d{\Tilde{\tau}}^2} + \Gamma^{\mu}_{\alpha\beta}
\frac{d{\Tilde{x}}^\alpha}{d{\Tilde{\tau}}} \frac{d{\Tilde{x}}^\beta}{d{\Tilde{\tau}}} = 0~,$$ where $\Gamma^{\mu}_{\alpha\beta}$ are the Christoffel symbols for the Kerr metric. When ${\Tilde{E}}$ and ${\Tilde{L}}$ are constants, the second of these equations (${\Tilde{x}}^1 = {\Tilde{r}}$) is equivalent to the derivative of Eq. (\[E:RV2\]) with respect to ${\Tilde{\tau}}$ $$\label{E:RA}
\frac{d^2{\Tilde{r}}}{d{\Tilde{\tau}}^2} = -\frac{1}{2} \frac{\partial V}{\partial {\Tilde{r}}}~.$$
Near the ISCO {#SS:nIsco}
-------------
In the vicinity of the ISCO, Eq. (\[E:RV2\]) can be Taylor expanded about the ISCO values of ${\Tilde{r}}$, ${\Tilde{E}}$, and ${\Tilde{L}}$. Expanding in the small variables
\[E:pert\] $$\begin{aligned}
\label{E:rpert}
R &\equiv& {\Tilde{r}}- {\Tilde{r}}_{\rm ISCO}~, \\ \label{E:Epert}
\chi &\equiv& {\Tilde{\Omega}}^{-1}_{\rm ISCO} ({\Tilde{E}}- {\Tilde{E}}_{\rm ISCO})~,
\\ \label{E:Lpert}
\xi &\equiv& {\Tilde{L}}- {\Tilde{L}}_{\rm ISCO}~,
\end{aligned}$$
where ${\Tilde{\Omega}}_{\rm ISCO}$ is the orbital frequency at the ISCO, yields $$\begin{aligned}
\label{E:1OT}
\left( \frac{dR}{d{\Tilde{\tau}}} \right)^2 &=& - \frac{2\alpha}{3} R^3
+ 2\beta R \xi + \frac{\partial V}{\partial {\Tilde{L}}} (\chi - \xi)
\nonumber \\
&& - {\Tilde{\Omega}}\frac{\partial^2 V}{\partial {\Tilde{E}}\partial {\Tilde{r}}} (\chi - \xi)R
+~...\end{aligned}$$ where the ellipsis denotes higher-order terms. Following the OT notation we define
\[E:OTco\] $$\begin{aligned}
\label{E:OTalpha}
\alpha &\equiv& \frac{1}{4} \left(
\frac{\partial^3 V}{\partial {\Tilde{r}}^3} \right)_{\rm ISCO}~,
\\ \label{E:OTbeta}
\beta &\equiv& -\frac{1}{2} \left(
\frac{\partial^2 V}{\partial {\Tilde{L}}\partial {\Tilde{r}}} +
{\Tilde{\Omega}}\frac{\partial^2 V}{\partial {\Tilde{E}}\partial {\Tilde{r}}}
\right)_{\rm ISCO}~,
\end{aligned}$$
and we have made use of the relation $$\label{E:con1}
{\Tilde{E}}- \frac{1}{2} \frac{\partial V}{\partial {\Tilde{E}}} - \frac{1}{2}
{\Tilde{\Omega}}^{-1} \frac{\partial V}{\partial {\Tilde{L}}} = 0$$ which holds at extrema of the effective potential. Again assuming that ${\Tilde{E}}$ and ${\Tilde{L}}$ are constant, Eq. (\[E:1OT\]) is equivalent to $$\label{E:2OT}
\frac{d^2R}{d{\Tilde{\tau}}^2} = - \alpha R^2 + \beta \xi
- \frac{1}{2} {\Tilde{\Omega}}\frac{\partial^2 V}{\partial {\Tilde{E}}\partial {\Tilde{r}}}
(\chi - \xi)~.$$ Terms of $\mathcal{O}(R^3)$, $\mathcal{O}(\xi^2, \chi^2)$ and higher order have been neglected in Eq. (\[E:2OT\]).
Radiation Reaction {#SS:RR}
------------------
We have so far neglected the effects of radiation reaction on the test particle’s motion. The proper way to include radiation reaction would be to calculate the conservative and dissipative parts of the self-force [@Mino:1996nk], and add these additional terms to the right-hand side of Eq. (\[E:GE\]). Although self-force calculations have progressed rapidly in recent years (for recent reviews see [@PoissLiv; @Barack:2009ux]), the full self-force is not yet available for test particles on Kerr geodesics near the ISCO. Without this self-force, the effects of radiation reaction can be approximated by making the energy and angular momentum time dependent in Eqs. (\[E:RV2\]) and (\[E:RA\]).
For particles on circular orbits, the rates at which energy and angular momentum are radiated are related by $$\label{E:circrad}
\frac{d{\Tilde{E}}}{d{\Tilde{\tau}}} = {\Tilde{\Omega}}\frac{d{\Tilde{L}}}{d{\Tilde{\tau}}}~.$$ OT assume that this relation holds throughout the transition, implying that $$\label{E:xicon}
\chi = \xi = -\eta \kappa {\Tilde{\tau}}~,$$ where $$\label{E:kappadef}
\kappa \equiv \left( {\Tilde{\Omega}}^{-1} \frac{d{\Tilde{E}}}{d{\Tilde{t}}} \frac{d{\Tilde{t}}}{d{\Tilde{\tau}}}
\right)_{\rm ISCO}~.$$ We will later relax this assumption in Sec. \[S:4v\]. Inserting Eq. (\[E:xicon\]) into Eq. (\[E:2OT\]) yields $$\label{E:2OTc}
\frac{d^2R}{d{\Tilde{\tau}}^2} = - \alpha R^2 - \eta \beta \kappa {\Tilde{\tau}}~,$$ identical to Eq. (3.15) of OT.
Dimensionless Equation of Motion {#SS:dimless}
--------------------------------
![The dimensionless radius $X$ as a function of dimensionless time $T$ during the transition from adiabatic inspiral to plunge. The solid black curve shows the numerical solution to Eq. (\[E:dimEOM\]), while the dashed blue and dotted red curves show the approximate analytic solutions at early and late times given by Eqs. (\[E:diminsp\]) and (\[E:dimplunge\]) respectively. This figure is a reproduction of Fig. 2 in OT.[]{data-label="F:dimtraj"}](Figure1){width="3.5in"}
Previous studies [@Ori:2000zn; @Buonanno:2000ef] noticed that Eq. (\[E:2OTc\]) can be converted into dimensionless form by defining
\[E:dimvar\] $$\begin{aligned}
\label{E:dimR}
R &\equiv& \eta^{2/5} R_0 X~, \\ \label{E:dimtau}
{\Tilde{\tau}}&\equiv& \eta^{-1/5} \tau_0 T~,
\end{aligned}$$
where
\[E:scal\] $$\begin{aligned}
\label{E:Rscal}
R_0 &=& (\beta \kappa)^{2/5} \alpha^{-3/5}~, \\ \label{E:tauscal}
\tau_0 &=& (\alpha \beta \kappa)^{-1/5}~.
\end{aligned}$$
In these variables, Eq. (\[E:2OTc\]) becomes $$\label{E:dimEOM}
\frac{d^2X}{dT^2} = - X^2 - T~.$$ At early times, the particle’s radial velocity and acceleration approach zero, suggesting that the correct solution to Eq. (\[E:dimEOM\]) asymptotes to $$\label{E:diminsp}
X = \sqrt{-T} \quad \quad T \to -\infty~.$$ At late times, the particle plunges into the horizon ($X = -\infty$) in a finite proper time ${\Tilde{\tau}}$. The second term on the right-hand of Eq. (\[E:dimEOM\]) can then be neglected, yielding the approximate solution $$\label{E:dimplunge}
X = \frac{-6}{({T_{\rm plunge}}- T)^2} \quad \quad T \to {T_{\rm plunge}}~.$$ Numerically, ${T_{\rm plunge}}= 3.412$. The numerical solution to Eq. (\[E:dimEOM\]) and approximate analytic solutions of Eqs. (\[E:diminsp\]) and (\[E:dimplunge\]) are given by the solid black, dashed blue, and dotted red curves in Fig. \[F:dimtraj\].
In addition to the approximate radial trajectory $X(T)$, this model also provides an estimate of the energy and angular momentum radiated during the transition. Assuming that energy and angular momentum are radiated at the ISCO rate throughout the transition, Eq. (\[E:dimplunge\]) implies that the test particle will radiate an additional amount
\[E:final\] $$\begin{aligned}
\label{E:Efinal}
\Delta {\Tilde{E}}_{\rm tr} &\equiv& {\Tilde{E}}_{\rm ISCO} - {\Tilde{E}}_{\rm final} =
\eta^{4/5} {\Tilde{\Omega}}_{\rm ISCO} \kappa \tau_0 {T_{\rm plunge}}\quad \\ \label{E:Lfinal}
\Delta {\Tilde{L}}_{\rm tr} &\equiv& {\Tilde{L}}_{\rm ISCO} - {\Tilde{L}}_{\rm final} =
\eta^{4/5} \kappa \tau_0 {T_{\rm plunge}}\end{aligned}$$
beyond that calculated in the adiabatic approximation. Since $\Delta
{\Tilde{E}}_{\rm tr}$ and $\Delta {\Tilde{L}}_{\rm tr}$ scale as $\eta^{4/5}$, in the test-particle limit $\eta \to 0$ the specific energy ${\Tilde{E}}$ and angular momentum ${\Tilde{L}}$ radiated during the transition should exceed that radiated during the ringdown which scales as $\eta$. Given the success of efforts to predict black-hole spins by extrapolating from the test-particle limit [@Hughes:2002ei; @Buonanno:2007sv; @Kesden:2008ga; @Kesden:2009ds], a closer examination of the energy and angular momentum radiated during the transition is worthwhile.
Normalization of 4-velocity {#S:4v}
===========================
![The square of the dimensionless radial velocity $dX/dT$ as a function of dimensionless time $T$ during the transition from adiabatic inspiral to plunge. The solid black curve shows the left-hand side of Eq. (\[E:dimKE\]) for the numerical solution to Eq. (\[E:dimEOM\]), while the dashed blue curve shows the right-hand side of Eq. (\[E:dimKE\]) for this [*same*]{} numerical solution. The two curves clearly differ, and the dashed blue curve becomes negative when it falls below the dotted red line. Since $(dX/dT)^2$ is positive definite, this indicates that the numerical solution occupies a forbidden portion of $\{ X, T \}$ parameter space.[]{data-label="F:dimKE"}](Figure2){width="3.5in"}
We noted in Sec. \[SS:geo\] that the normalization of the 4-velocity (\[E:4norm\]) and the geodesic equation (\[E:GE\]) provide alternative equations of radial motion. While they are equivalent for geodesics, they differ once ${\Tilde{E}}$ and ${\Tilde{L}}$ become time dependent. Ori and Thorne [@Ori:2000zn] solve the geodesic equation under the assumption that ${\Tilde{E}}$ and ${\Tilde{L}}$ vary with proper time according to Eq. (\[E:xicon\]). What does this assumption imply for the norm of the 4-velocity? Expressed in the dimensionless variables of Eq. (\[E:dimvar\]), Eq. (\[E:1OT\]) becomes $$\label{E:dimKE}
\left( \frac{dX}{dT} \right)^2 = -\frac{2}{3} X^3 - 2XT~.$$ The left and right-hand sides of Eq. (\[E:dimKE\]) for the numerical solution $X(T)$ of Eq. (\[E:dimEOM\]) are plotted in Fig. \[F:dimKE\]. The solid black curve, showing the left-hand side $(dX/dT)^2$, is positive definite as one would expect. The dashed blue curve shows the right-hand side. If Eq. (\[E:dimKE\]) was satisfied, the two curves would be identical and the right-hand side would be positive definite as well. This is clearly not the case.
What does this mean physically? According to Eq. (\[E:xicon\]), the dimensionless variable $T$ is simultaneously proportional to the proper time ${\Tilde{\tau}}$, the energy $\chi$, and the angular momentum $\xi$. Values of $X(T)$ for which the right-hand side of Eq. (\[E:dimKE\]) is negative correspond to values of $\{ {\Tilde{r}}, {\Tilde{E}},
{\Tilde{L}}\}$ for which the effective potential $V({\Tilde{r}}, {\Tilde{E}}, {\Tilde{L}})$ exceeds the square of the energy ${\Tilde{E}}^2$ in violation of Eq. (\[E:RV2\]). For example, at $T = 0$ the energy and angular momentum equal their ISCO values, yet $X(0) > 0$ as seen in Fig. \[F:dimtraj\]. Since $V({\Tilde{r}}_{\rm ISCO}, {\Tilde{E}}_{\rm ISCO},
{\Tilde{L}}_{\rm ISCO}) = {\Tilde{E}}_{\rm ISCO}^2$ and the effective potential $V$ is a monotonically increasing function of $X$ for $T = 0$, the right-hand side of Eq. (\[E:dimKE\]) must be negative at $T = 0$ as seen in Fig. \[F:dimKE\]. The discrepancy between the left and right-hand sides of Eq. (\[E:dimKE\]) is not a consequence of a breakdown in the Taylor expansion of the effective potential for $R, \xi \gtrsim
1$. Figure \[F:dimKE\] shows that Eq. (\[E:dimKE\]) is violated even for $X, T \simeq 1$, which corresponds to $R, \xi \ll 1$ when $\eta \ll 1$.
![The solid black, short-dashed blue, and dotted red curves in this plot are identical to those in Fig. \[F:dimKE\], but we have also added the dimensionless difference between the energy and angular momentum $Y(T)$ which is given by the long-dashed green curve. The short-dashed blue curve shows the first two terms on the right-hand side of Eq. (\[E:dimKEnew\]), while the long-dashed green curve shows our new third term. Their sum equals the solid black curve, implying that Eq. (\[E:dimKEnew\]) is now satisfied and the test particle’s 4-velocity is properly normalized.[]{data-label="F:Y"}](Figure3){width="3.5in"}
How can we reconcile the normalization of the 4-velocity (\[E:4norm\]) and the geodesic equation (\[E:GE\]) in the presence of radiation reaction? Although Eqs. (\[E:circrad\]) and (\[E:xicon\]) correctly specify the energy, angular momentum, and their derivatives on a [*circular*]{} orbit at ${\Tilde{r}}_{\rm ISCO}$, there is no reason to expect them to hold exactly when the radial velocity is nonzero as it is during the transition. If we relax the requirement that $\chi = \xi$, we recover the third and fourth terms on the right-hand side of Eq. (\[E:1OT\]) that vanished in the previous treatment. We can retain the dimensionless form of the equations of motion by defining $$\label{E:Ydef}
\chi - \xi \equiv \eta^{6/5} (\chi - \xi)_0 Y~,$$ where $$\label{E:Yscale}
(\chi - \xi)_0 = \alpha^{-4/5} (\beta\kappa)^{6/5} \left(
\frac{\partial V}{\partial {\Tilde{L}}} \right)^{-1}~.$$ The second-order equation (\[E:dimEOM\]) remains unchanged to lowest order in $\eta$, while Eq. (\[E:dimKE\]) gains an additional term to become $$\label{E:dimKEnew}
\left( \frac{dX}{dT} \right)^2 = -\frac{2}{3} X^3 - 2XT + Y~.$$ Instead of Eq. (\[E:circrad\]), which would imply that $Y$ is constant throughout the transition, we evolve $Y$ according to $$\label{E:dYdT}
\frac{dY}{dT} = 2X~.$$ This is precisely what is required to restore the consistency between Eqs. (\[E:dimEOM\]) and (\[E:dimKEnew\]) that exists for geodesic motion. Since Eq. (\[E:dimEOM\]) remains unchanged, its solution $X(T)$ remains unchanged as well.
We can solve Eq. (\[E:dYdT\]) by inserting $X(T)$ into the right-hand side and choosing the correct initial condition. This initial condition can be found by matching to the quasicircular inspiral at early times. During the inspiral, ${\Tilde{E}}$ and ${\Tilde{L}}$ are given by [@Bardeen:1972fi]
\[E:ELgeo\] $$\begin{aligned}
\label{E:Egeo}
{\Tilde{E}}_c({\Tilde{r}}) &=& \frac{1 - 2/{\Tilde{r}}+ {\Tilde{a}}/{\Tilde{r}}^{3/2}}{\sqrt{1 -
3/{\Tilde{r}}+2{\Tilde{a}}/{\Tilde{r}}^{3/2}}}~, \\ \label{E:Lgeo}
{\Tilde{L}}_c({\Tilde{r}}) &=& {\Tilde{r}}^{1/2} \frac{1 - 2{\Tilde{a}}/{\Tilde{r}}^{3/2} +
{\Tilde{a}}^2/{\Tilde{r}}^2}{\sqrt{1 - 3/{\Tilde{r}}+2{\Tilde{a}}/{\Tilde{r}}^{3/2}}}~.
\end{aligned}$$
Taylor expanding about the ISCO, $$\begin{aligned}
\label{E:CXearly}
\chi - \xi &\equiv& {\Tilde{\Omega}}_{\rm ISCO}^{-1}({\Tilde{E}}_c(r) - {\Tilde{E}}_{\rm ISCO})
- ({\Tilde{L}}_c(r) - {\Tilde{L}}_{\rm ISCO})
\nonumber \\
&\simeq& \frac{1}{6} \left( {\Tilde{\Omega}}^{-1} \frac{d^3{\Tilde{E}}_c}{d{\Tilde{r}}^3} -
\frac{d^3{\Tilde{L}}_c}{d{\Tilde{r}}^3} \right)_{\rm ISCO} R^3~.\end{aligned}$$ Using the definitions of $X$ and $Y$ in Eqs. (\[E:dimR\]) and (\[E:Ydef\]), this implies $$\begin{aligned}
\label{E:Yearly}
Y &=& \frac{1}{6} \alpha^{-1} \left( {\Tilde{\Omega}}^{-1} \frac{d^3{\Tilde{E}}_c}{d{\Tilde{r}}^3}
- \frac{d^3{\Tilde{L}}_c}{d{\Tilde{r}}^3} \right)_{\rm ISCO} \left(
\frac{\partial V}{\partial {\Tilde{L}}} \right)_{\rm ISCO} X^3
\nonumber \\
&=& -\frac{4}{3} X^3 \quad \quad T \to -\infty~.\end{aligned}$$ Inserting Eqs. (\[E:diminsp\]) and (\[E:Yearly\]) into the right-hand side of Eq. (\[E:dimKEnew\]) shows that $dX/dT$ vanishes as $T \to -\infty$ as required by Eq. (\[E:diminsp\]). The numerical solution $Y(T)$ with this initial condition is shown by the long-dashed green curve in Fig. \[F:Y\]. With the addition of this new term to the right-hand side, Eq. (\[E:dimKEnew\]) is now satisfied ensuring that the 4-velocity is normalized according to Eq. (\[E:masscon\]).
Our new term $Y(T)$ is important for more than just mathematical consistency. Equation (\[E:Ydef\]) suggests that the physical quantity $\chi - \xi$ is proportional to $\eta^{6/5}$ during the transition, making it higher order than self-force corrections that scale as $\eta$. However, as the test particle plunges into the horizon, the asymptotic solution (\[E:dimplunge\]) and Eq. (\[E:dYdT\]) imply $$\begin{aligned}
\label{E:YplungeT}
Y(T) &=& Y(0) + \int_{0}^{T} \frac{dY}{dT^\prime} dT^\prime
\nonumber \\
&\simeq& Y(0) - \int_{0}^{T} \frac{12}{({T_{\rm plunge}}- T^\prime)^2}
dT^\prime \nonumber \\
&\simeq& -\frac{12}{{T_{\rm plunge}}- T}\end{aligned}$$ which diverges as $T \to {T_{\rm plunge}}$. This divergence can be seen from the limiting behavior of the three curves in Fig. \[F:Y\]. Although the solid black and short-dashed blue curves appear to converge as $T \to
{T_{\rm plunge}}$, their difference $Y(T)$ shown by the long-dashed green curve in fact diverges in accordance with Eq. (\[E:YplungeT\]). Does the divergence of the dimensionless $Y(T)$ imply a similar divergence in the physical quantity $\chi - \xi$ as the test particle plunges into the black hole?
To answer this question, we must examine the validity of the asymptotic solution (\[E:dimplunge\]) as $T \to {T_{\rm plunge}}$. Footnote 3 of OT notes that the divergence of $X(T)$ results from a breakdown in the dimensionless equation of motion (\[E:dimEOM\]) when higher-order terms in the Taylor expansion about the ISCO become important. This breakdown occurs at a coordinate radius $R_{\rm break}$, which corresponds to a dimensionless radius $X_{\rm break} \propto
\eta^{-2/5}$ according to Eq. (\[E:dimR\]). This in turn implies that ${T_{\rm plunge}}- T_{\rm break} \propto \eta^{1/5}$ by Eq. (\[E:dimplunge\]) and $Y_{\rm break} \propto \eta^{-1/5}$ from Eq. (\[E:YplungeT\]). According to Eq. (\[E:Ydef\]), the test particle’s energy during the transition will receive a correction $$\label{E:DelEcorr}
\Delta {\Tilde{E}}_{\rm norm} = \eta^{6/5} {\Tilde{\Omega}}_{\rm ISCO} (\chi - \xi)_0 Y~.$$ When $Y \sim Y_{\rm break}$, this correction will be linearly proportional to $\eta$ just like self-force corrections. It will be instructive to compare future self-force and time-domain perturbation theory calculations with this analytic result.
Maximal Spins {#S:maxspins}
=============
Kerr black holes [@Kerr:1963ud] have spins ${\Tilde{a}}< 1$; objects with larger spins are “naked singularities” unclothed by an event horizon. Penrose [@Penrose:1969pc] proposed that a cosmic censor protects general relativity by preventing the formation of such naked singularities, but this cosmic censorship conjecture has never been proven in full generality. Wald [@Wald:1974tp] determined that maximally spinning black holes could not accrete [*test*]{} particles that would drive them over the Kerr limit ${\Tilde{a}}= 1$, but Jacobson and Sotiriou [@Jacobson:2009kt] recently showed that black holes with spins $\delta \equiv 1 - {\Tilde{a}}\ll 1$ could be spun above this limit by accreting particles with a [*finite*]{} energy $E = \eta M {\Tilde{E}}$. Barausse [*et al.*]{} [@Barausse:2010ka] showed that gravitational radiation could not prevent some of these finite-mass-ratio mergers from producing a naked singularity, but that self-force corrections could be significant. Given that finite-mass-ratio effects may determine whether naked singularities can exist, it is worthwhile to examine what happens to the transition region in the maximally spinning limit.
The previous treatment of the transition from quasicircular inspiral to plunge assumes that the effective potential $V$ and fluxes $d{\Tilde{E}}/d{\Tilde{\tau}}$ and $d{\Tilde{L}}/d{\Tilde{\tau}}$ can be Taylor expanded in a neighborhood of the ISCO. However, as the black hole’s spin approaches the Kerr limit $a \to M$, $r_{\rm ISCO}$ and the outer horizon $r_+$ both approach $M$ in Boyer-Lindquist coordinates. This suggests that it may not be possible to construct a neighborhood of the ISCO that does not include the horizon. To determine whether or not such a neighborhood exists, we must compare $R_+ \equiv {\Tilde{r}}_{\rm
ISCO} - {\Tilde{r}}_+$ with $R_0$ in Eq. (\[E:Rscal\]) as $\delta \to 0$. In this limit [@Bardeen:1972fi],
\[E:Rlim\] $$\begin{aligned}
\label{E:ISCOlim}
{\Tilde{r}}_{\rm ISCO} &\to& 1 + (4\delta)^{1/3}~, \\ \label{E:EHlim}
{\Tilde{r}}_+ &\to& 1 + (2\delta)^{1/2}~,
\end{aligned}$$
implying $R_+ \propto \delta^{1/3}$.
![The general-relativistic correction ${\dot{\mathcal{E}}}$ to the Newtonian quadrupole-moment formula for the GW luminosity as a function of black-hole spin $\delta \equiv 1 - {\Tilde{a}}$. The points have been calculated using the black-hole perturbation theory code GREMLIN [@Hughes:1999bq] and fitted with a power law ${\dot{\mathcal{E}}}= A \delta^m$.[]{data-label="F:RC"}](Figure4){width="3.5in"}
The behavior of $R_0$ as $\delta \to 0$ depends on the behavior of the fluxes $d{\Tilde{E}}/d{\Tilde{\tau}}$ and $d{\Tilde{L}}/d{\Tilde{\tau}}$. These in turn depend on ${\dot{\mathcal{E}}}$, defined by $$\label{E:RC}
\dot{E}_{\rm GW} = \frac{32}{5} \eta^2 {\Tilde{\Omega}}^{10/3} {\dot{\mathcal{E}}}$$ as the general-relativistic correction to the Newtonian quadrupole-moment formula for the GW luminosity [@Finn:2000sy]. The ISCO values of this correction ${\dot{\mathcal{E}}}$ were calculated for spins $-0.99 \leq {\Tilde{a}}\leq 0.999$ in [@Finn:2000sy], and have been calculated down to $\delta = 10^{-4}$ [@HughesComm] using the GREMLIN (Gravitational Radiation in the Extreme Mass Ratio Limit) code presented in [@Hughes:1999bq]. We have fitted these calculated values to a power law ${\dot{\mathcal{E}}}= A \delta^m$ as shown in Fig. \[F:RC\]. The best-fit parameters for this power law are $A = 1.80$, $m = 0.317$. The summation over spheroidal harmonics of the Weyl scalar $\psi_4$ needed to compute ${\dot{\mathcal{E}}}$ converges very slowly in the limit $\delta \to 0$, so our best-fit parameters should be regarded with caution until they can be confirmed by a technique better suited to this limit. Chrzanowski [@Chrzanowski:1976jy] estimated that $$\label{E:Chrz}
\dot{E}_{\rm GW} \sim \eta^2 R_+~,$$ which also suggests that $m \sim 1/3$. The rough agreement between our numerical fit and Chrzanowski’s estimate gives us some confidence that $m \simeq 1/3$ is close to the correct value.
The scale $R_0$ of the transition region depends on the relativistic correction ${\dot{\mathcal{E}}}$ through $\kappa$, which was defined in Eq. (\[E:kappadef\]) and can be expressed as $$\label{E:kappa2}
\kappa = \frac{32}{5} \left( {\Tilde{\Omega}}^{7/3} \frac{d{\Tilde{t}}}{d{\Tilde{\tau}}}
{\dot{\mathcal{E}}}\right)_{\rm ISCO}~.$$ For circular equatorial Kerr geodesics, as $\delta \to 0$,
\[E:max1\] $$\begin{aligned}
\label{E:Omax}
{\Tilde{\Omega}}&\to& \frac{1}{2} \left[ 1 - \frac{3}{4} (4\delta)^{1/3} \right]~,
\\ \label{E:dtdtaumax}
\frac{d{\Tilde{t}}}{d{\Tilde{\tau}}} &\to& \frac{4}{\sqrt{3}} (4\delta)^{-1/3}~,
\\ \label{E:kappamax}
\kappa &\to& \frac{16}{5\sqrt{3}} A \delta^{m - 1/3}~,
\\ \label{E:alphamax}
\alpha &\to& 1~, \\ \label{E:betamax}
\beta &\to& \frac{\sqrt{3}}{2} (4\delta)^{1/3}~, \\ \label{E:dVdLmax}
\frac{\partial V}{ \partial {\Tilde{L}}} &\to&
\frac{4}{\sqrt{3}} (4\delta)^{1/3}~,
\end{aligned}$$
implying
\[E:max2\] $$\begin{aligned}
\label{E:R0max}
R_0 &\propto& \delta^{2m/5}~, \\ \label{E:tau0max}
\tau_0 &\propto& \delta^{-m/5}~, \\ \label{E:chi0max}
(\chi - \xi)_0 &\propto& \delta^{6m/5 - 1/3}~,
\end{aligned}$$
from the definitions in Eqs. (\[E:scal\]) and (\[E:Yscale\]).
According to Eq. (\[E:dimR\]), the transition region will be cut off by the horizon at a dimensionless radius $$\label{E:Xcut}
X_+ = -\eta^{-2/5} \frac{R_+}{R_0} \propto -\eta^{-2/5} \delta^{1/3 - 2m/5}$$ as $\delta \to 0$. We plot $X_+(\delta)$ for several mass ratios and our best-fit value of $m$ in Fig. \[F:EH\]. Since $m < 5/6$, $X_+
\to 0$ in the limit of large spins. One should not expect GW emission beyond $X_+$, so Eq. (\[E:final\]) overestimates the additional energy and angular momentum radiated during the transition. In the limit of large spins, $\Delta {\Tilde{E}}_{\rm tr}$ and $\Delta {\Tilde{L}}_{\rm tr}$ will be suppressed by a factor $T_0/{T_{\rm plunge}}\simeq 0.21$, where $X(T_0) = 0$.
![The dimensionless horizon radius $X_+ = -\eta^{-2/5}R_+/R_0$ as a function of $\delta \equiv 1 - {\Tilde{a}}$. The dotted red, dashed blue, and solid black curves correspond to mass ratios $\eta =
10^{-4}$, $10^{-5}$, and $10^{-6}$ respectively. $X_+ \to 0$ as $\delta \to 0$ because the relativistic correction ${\dot{\mathcal{E}}}$ to the energy radiated in GWs has a power-law index $m < 5/6$.[]{data-label="F:EH"}](Figure5){width="3.5in"}
Despite this suppression, the extra energy and angular momentum radiated during the transition remain proportional to $\kappa \tau_0
\propto \delta^{4m/5 - 1/3}$, which diverges for $m < 5/12$. Our numerical fit shown in Fig. \[F:RC\] suggests that $m$ satisfies this inequality. However, this divergence might not be physical; it could merely reflect a breakdown of the Taylor expansion of the equation of motion given in Eq. (\[E:2OT\]). The first two terms on the right-hand side of this equation are proportional to $\eta^{4/5}$, while Eq. (\[E:Ydef\]) shows that the third term is proportional to $\eta^{6/5}$. This higher-order dependence on $\eta$ justifies neglecting this third term for modest spins, but we must also consider how each term depends on $\delta$ in the maximally spinning limit. In this limit, the first two terms are proportional to $\delta^{4m/5}$ according to Eqs. (\[E:max1\]) and (\[E:max2\]) while the third term is proportional to $\delta^{6m/5 - 1/3}$. For $m < 5/6$, this third term will dominate over the first two terms, indicating a breakdown of the original dimensionless equation of motion (\[E:dimEOM\]). For completeness, we note that we have neglected an additional term $$\label{E:4term}
-\frac{1}{12} \left( \frac{\partial^4 V}{\partial {\Tilde{r}}^4} \right)_{\rm ISCO} R^3$$ at $\mathcal{O}(\eta^{6/5})$ on the right-hand side Eq. (\[E:2OT\]). This term is proportional to $\delta^{6m/5}$ as $\delta \to 0$ and thus can always be neglected.
Including the third term in Eq. (\[E:2OT\]) on the right-hand side of Eq. (\[E:dimEOM\]) yields $$\label{E:dim3T}
\frac{d^2X}{dT^2} = - X^2 - T + \eta^{2/5} CY~,$$ where $$\label{E:3TC}
C \equiv -\frac{1}{2} \alpha^{-3/5} (\beta \kappa)^{2/5} {\Tilde{\Omega}}\frac{\partial^2 V}{\partial {\Tilde{E}}\partial {\Tilde{r}}} \left(
\frac{\partial V}{\partial {\Tilde{L}}} \right)^{-1} \propto \delta^{2m/5 - 1/3}$$ diverges as $\delta \to 0$ for $m < 5/6$. $Y$ must evolve according to $$\label{E:Y3T}
\frac{dY}{dT} = 2X + 2\eta^{2/5} CY \frac{dX}{dT}$$ to preserve the normalization the 4-velocity given by Eq. (\[E:dimKEnew\]). The product $\eta^{2/5}C$ thus measures the deviation of $X(T)$ from the OT solution in the limit of large spins. We plot our new solutions $X(T,\eta^{2/5}C)$ for a mass ratio $\eta =
10^{-3}$ and spins $\delta = 10^{-2}$, $10^{-4}$ and $10^{-6}$ in Fig. \[F:3T\]. Although our solutions diverge from the OT solution as $\eta^{2/5}C$ increases, the test particle crosses the ISCO $X = 0$ (shown by the horizontal dotted line) at a [*later*]{} dimensionless time $T$. This implies that more energy and angular momentum are radiated and the divergence in $\Delta {\Tilde{E}}_{\rm tr}$ and $\Delta
{\Tilde{L}}_{\rm tr}$ as $\delta \to 0$ cannot be avoided. One must look to finite-mass ratio effects beyond the scope of this paper to eliminate this unphysical result.
One such effect that might provide a solution to this problem is the spinning down of the black hole due to the extraction of angular momentum by the superradiant scattering of GWs earlier in the inspiral [@Kesden:2009ds]. This effect implies that even an initially maximally spinning black hole will have $\delta \propto \eta$ by the time the test particle reaches the transition region. The additional energy and angular momentum $\Delta {\Tilde{E}}_{\rm tr}, {\Tilde{L}}_{\rm tr} \propto
\eta^{4/5} \delta^{4m/5 - 1/3} \propto \eta^{4m/5 + 7/15}$ would then remain finite even if $\delta = 0$ initially.
![[*Upper panel:*]{} The dimensionless radius $X$ as a function of dimensionless time $T$ in the limit $\delta \equiv 1 - {\Tilde{a}}\to
0$. The solid black curve is the OT solution, while the long-dashed blue, short-dashed green, and dot-dashed red curves show our solutions with $\eta = 10^{-3}$ and $\delta = 10^{-2}$, $10^{-4}$ and $10^{-6}$ respectively. [*Lower panel:*]{} The dimensionless difference $Y$ between the energy and angular momentum for the solutions shown in the upper panel.[]{data-label="F:3T"}](Figure6){width="3.5in"}
How does the energy and angular momentum radiated during the transition affect the spin of the final black hole produced in the merger? If we assume that energy and angular momentum are conserved after the end of the transition, the final spin will be $$\begin{aligned}
\label{E:af}
{\Tilde{a}}_f &=& \frac{{\Tilde{a}}+ \eta({\Tilde{L}}_{\rm ISCO} - \Delta {\Tilde{L}}_{\rm tr})}
{[1 + \eta({\Tilde{E}}_{\rm ISCO} - \Delta {\Tilde{E}}_{\rm tr} +
\Delta {\Tilde{E}}_{\rm norm})]^2}
\nonumber \\
&\simeq& {\Tilde{a}}+ \eta (\Delta {\Tilde{a}}_{\rm ISCO} + \Delta {\Tilde{a}}_{\rm tr}
+ \Delta {\Tilde{a}}_{\rm norm})\end{aligned}$$ for $\eta \ll 1$, where in the limit $\delta \to 0$
\[E:max3\] $$\begin{aligned}
\label{E:ISCOmax}
\Delta {\Tilde{a}}_{\rm ISCO} &\equiv& {\Tilde{L}}_{\rm ISCO} - 2{\Tilde{E}}_{\rm ISCO}
\to \frac{\sqrt{3}}{2} (4\delta)^{2/3},
\\ \label{E:Trmax}
\Delta {\Tilde{a}}_{\rm tr} &\equiv& -(1 - 2{\Tilde{\Omega}}_{\rm ISCO})\Delta {\Tilde{L}}_{\rm tr}
\propto \eta^{4/5} \delta^{4m/5},
\\ \label{E:Normmax}
\Delta {\Tilde{a}}_{\rm norm} &\equiv& -2\Delta {\Tilde{E}}_{\rm norm}
\propto \eta^{6/5} \delta^{6m/5 - 1/3}.
\end{aligned}$$
Although our new correction $\Delta {\Tilde{a}}_{\rm norm}$ to the black hole’s final spin is subdominant in $\eta$, for $m < 5/6$ it becomes the dominant correction as $\delta \to 0$. Since $\Delta {\Tilde{E}}_{\rm
norm} > 0$ when the particle crosses the horizon in the limit $\delta
\to 0$, our analysis suggests that gravitational radiation in the transition region will [*not*]{} promote the formation of naked singularities. This supports our earlier result in [@Kesden:2009ds], where we showed that the superradiant scattering of GWs emitted at ${\Tilde{r}}> {\Tilde{r}}_{\rm ISCO}$ would also reduce the final spin for ${\Tilde{a}}\gtrsim 0.998$.
Discussion {#S:disc}
==========
The existence of an ISCO is one of the most distinctive features of a black hole. Although the presence of an ISCO indirectly affects the luminosity and spectra of accreting black holes, it is more cleanly probed by the GWs emitted as test particles (compact objects like white dwarfs, neutron stars, and stellar-mass black holes) plunge into supermassive black holes. Such GWs are a primary source for the proposed space-based GW detector LISA. The LISA detection strategy relies on convolving observations with theoretically determined GW templates. Understanding how a test particle’s position, energy, and angular momentum evolve near the ISCO is an important first step toward constructing these templates.
Previous studies [@Ori:2000zn; @Buonanno:2000ef] identified a transition region near the ISCO where neither the quasicircular approximation nor the assumption of geodesic motion are valid. They solved the radial equation of motion in this region by Taylor expanding it about the ISCO. Ori and Thorne (2000) [@Ori:2000zn] fixed the energy and angular momentum fluxes to their ISCO values, while Buonanno and Damour (2000) [@Buonanno:2000ef] worked with the Hamilton equations of motion and did not need to specify an explicit energy flux. The dimensionless solution $X(T)$ obtained by both groups can be rescaled to determine the evolution of the test particle’s position, energy, and angular momentum for arbitrary mass ratios and spins. The simplicity of their approach and the universality of their solution are highly appealing, but a closer examination reveals that their solution does not properly normalize the particle’s 4-momentum.
We undertook this study to see whether this technical problem could be easily remedied, or was a symptom of a more serious flaw in their approach. We found that by introducing a correction $\Delta {\Tilde{E}}_{\rm
norm}$ to the particle’s energy at a higher order in the mass ratio $\eta \ll 1$, we could properly normalize the 4-momentum without altering their universal solution $X(T)$ at lowest order. This relatively modest correction increases our confidence in this particle trajectory, and validates its use as a source for the construction of GW waveforms as in [@Sundararajan:2010sr].
In addition to its role in constructing GW templates, gravitational radiation during the transition also affects whether a test-particle merger can produce a naked singularity by increasing a black hole’s spin above the Kerr limit ${\Tilde{a}}= 1$. Our calculation of the energy and angular momentum flux at the ISCO as $\delta \equiv 1 - {\Tilde{a}}\to 0$ suggests that the total energy $\Delta {\Tilde{E}}_{\rm tr}$ and angular momentum $\Delta {\Tilde{L}}_{\rm tr}$ radiated during the transition diverge in this limit. This divergence cannot be physical, and must therefore be moderated by high-spin corrections beyond the scope of this paper. Despite this divergence, the change $\Delta {\Tilde{a}}_{\rm tr}$ in the black hole’s spin due to this radiation remains finite in magnitude and negative in sign. It therefore reduces the likelihood that a naked singularity will be produced, as does our new correction $\Delta
{\Tilde{a}}_{\rm norm}$ which is also negative and becomes dominant as $\delta
\to 0$.
Throughout this paper, we have neglected self-force corrections to the particle’s energy even though they are formally lower order in $\eta$ than our correction $\Delta {\Tilde{E}}_{\rm norm} \propto \eta^{6/5}$. Far from the ISCO, where $|Y| \gg 1$, our correction should dominate self-force effects despite its scaling with $\eta$. Near the ISCO however, self-force corrections should be significant. Reliable calculations of the self-force do not yet exist near the ISCOs of highly spinning black holes. Comparing such self-force calculations with our analysis as they become available is a subject for future work. Another important future test of our analysis would be to use time-domain perturbation theory to calculate the energy and angular momentum radiated by a particle whose position is given by an appropriate rescaling of the solution $X(T)$. Agreement with our predicted solution $Y(T)$ would strongly support our conjecture that proper normalization of the 4-momentum can be used to predict GW emission. Comparison with numerical relativity may soon be possible as well, since simulations with mass ratios as small as $\eta = 0.01$ have recently been performed [@Lousto:2010ut]. Our predictions could be even further improved by including the gravitational radiation emitted near the horizon [@Mino:2008at]. We look forward to comparing our calculation to these alternative theoretical approaches in the near future, and comparing all these predictions to GW observations in the hopefully not too distant future.
[**Acknowledgements.**]{} I would like to thank Sterl Phinney for his advice during the initial stages of this project, and Scott Hughes for insight into the emission of gravitational radiation in the limit of large spins. Jeandrew Brink, Marc Favata, Chris Hirata, Guglielmo Lockhart, Samaya Nissanke, and David Tsang also offered helpful comments on this work.
[99]{}
J. Kormendy and D. Richstone, Ann. Rev. Astron. Astrophys. [**33**]{}, 581 (1995). J. N. Bahcall and R. A. Wolf, Astrophys. J. [**209**]{}, 214 (1976). J. Frank and M. J. Rees, Mon. Not. Roy. Astron. Soc. [**176**]{}, 633 (1976). S. Sigurdsson and M. J. Rees, Mon. Not. Roy. Astron. Soc. [**284**]{}, 318 (1997). LISA, http://lisa.nasa.gov/.
R. P. Kerr, Phys. Rev. Lett. [**11**]{}, 237 (1963). B. Carter, Phys. Rev. [**174**]{}, 1559 (1968). M. Walker and R. Penrose, Commun. Math. Phys. [**18**]{}, 265 (1970). S. A. Teukolsky, Astrophys. J. [**185**]{}, 635 (1973). P. C. Peters and J. Mathews, Phys. Rev. [**131**]{}, 435 (1963). J. M. Bardeen, W. H. Press and S. A. Teukolsky, Astrophys. J. [**178**]{}, 347 (1972). A. Ori and K. S. Thorne, Phys. Rev. D [**62**]{}, 124022 (2000). A. Buonanno and T. Damour, Phys. Rev. D [**62**]{}, 064015 (2000). P. A. Sundararajan, Phys. Rev. D [**77**]{}, 124050 (2008). P. A. Sundararajan, G. Khanna and S. A. Hughes, Phys. Rev. D [**81**]{}, 104009 (2010). R. H. Boyer and R. W. Lindquist, J. Math. Phys. [**8**]{}, 265 (1967). Y. Mino, M. Sasaki and T. Tanaka, Phys. Rev. D [**55**]{}, 3457 (1997). E. Poisson, Living Rev. Relativity [**7**]{}, 6 (2004) \[http://www.livingreviews.org/lrr-2004-6\].
L. Barack, Class. Quant. Grav. [**26**]{}, 213001 (2009). S. A. Hughes and R. D. Blandford, Astrophys. J. [**585**]{}, L101 (2003). A. Buonanno, L. E. Kidder and L. Lehner, Phys. Rev. D [**77**]{}, 026004 (2008). M. Kesden, Phys. Rev. D [**78**]{}, 084030 (2008). M. Kesden, G. Lockhart, E. S. Phinney, Phys. Rev. [**D82**]{}, 124045 (2010). R. Penrose, Riv. Nuovo Cim. [**1**]{}, 252 (1969), reprinted in \[Gen. Rel. Grav. [**34**]{}, 1141 (2002)\]. R.M. Wald, Ann. Phys. [**82**]{}, 548 (1974).
T. Jacobson, T. P. Sotiriou, Phys. Rev. Lett. [**103**]{}, 141101 (2009). E. Barausse, V. Cardoso, G. Khanna, Phys. Rev. Lett. [**105**]{}, 261102 (2010). L. S. Finn, K. S. Thorne, Phys. Rev. [**D62**]{}, 124021 (2000). S. A. Hughes, private communication.
S. A. Hughes, Phys. Rev. [**D61**]{}, 084004 (2000). P. L. Chrzanowski, Phys. Rev. [**D13**]{}, 806-818 (1976).
C. O. Lousto, Y. Zlochower, Phys. Rev. Lett. [**106**]{}, 041101 (2011). Y. Mino, J. Brink, Phys. Rev. [**D78**]{}, 124015 (2008).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'A host of spatially extended systems, both in physics and in other disciplines, are well described at a coarse-grained scale by a Langevin equation with multiplicative-noise. Such systems may exhibit non-equilibrium phase transitions, which can be classified into universality classes. Here we study in detail one of such classes that can be mapped into a Kardar-Parisi-Zhang (KPZ) interface equation with a positive (negative) non-linearity in the presence of a bounding lower (upper) wall. The wall limits the possible values taken by the height variable, introducing a lower (upper) cut-off, and induce a phase transition between a pinned (active) and a depinned (absorbing) phase. This transition is studied here using mean field and field theoretical arguments, as well as from a numerical point of view. Its main properties and critical features, as well as some challenging theoretical difficulties, are reported. The differences with other multiplicative noise and bounded-KPZ universality classes are stressed, and the effects caused by the introduction of “attractive” walls, relevant in some physical contexts, are also analyzed.'
author:
- 'Miguel A. Mu\~ noz'
- Francisco de los Santos
- Abdelfattah Achahbar
date: today
title: 'Critical behavior of a bounded Kardar-Parisi-Zhang equation'
---
Introduction
============
Non-equilibrium phase transitions occurring in systems amenable to be described by Langevin equations including a multiplicative noise (MN) term are the subject of current intense studies. This embraces a broad variety of systems both in physics and in other disciplines. A phenomenology much richer and complex than that appearing in equilibrium systems, including counterintuitive behaviors, has been reported to appear in these, typically non-equilibrium, situations. See [@Review; @Sancho] for detailed introductions to this growing field, including many different realizations.
The interest in MN problems is enlarged even further, because of the existing mappings between them and other prototypical non-equilibrium problems [@Review]. A well known instance is the Kardar-Parisi-Zhang (KPZ) equation, describing the kinetic roughening transition of generic interfaces under non-equilibrium conditions [@KPZ; @Barabasi; @HZ], which can be mapped onto a MN Langevin equation, by performing the so called Cole-Hopf transformation linking the interface height at each point with the activity field of the MN equation. If an interface under consideration is described by the KPZ equation and it is physically limited by a wall, [*i.e.*]{} if the heights cannot be larger of smaller than a certain value, then this problem, [*bounded KPZ*]{}, can be mapped into a multiplicative noise equation by employing the abovementioned transformation [@MN1; @MN2; @Review]. The bounded KPZ equation may experience, as parameters are varied, a phase transition from a depinned phase in which the interface escapes with probability one from the wall, to a pinned phase characterized by a finite expectation value of the stationary averaged height (measured from the wall). In the MN language ([*i.e.*]{} after employing the Cole-Hopf transformation) the pinning-depinning transition corresponds to a critical point separating an [*absorbing phase*]{} in which the order parameter goes exponentially to zero (depinned phase) to an [*active phase*]{} in which the order parameter takes a non-vanishing average value.
Surprisingly enough, it was shown a few years ago that the introduction of “upper” or “lower” walls into a given KPZ equation (with a fixed non-linearity sign) lead to quite different phenomenologies. The origin of this can be tracked down to the fact that the KPZ equation is not invariant upon inverting the height (see [@MN3] or [@Review] for a more detailed explanation). Taking, for instance, the sign of the coefficient of the KPZ non-linearity to be positive, the introduction of an upper wall leads to a (well established by now) set of critical exponents characterizing the, so called, multiplicative noise 1 (MN1) universality class [@MN1; @MN2; @MN3; @MN4]. On the other hand, a wall limiting negative values of the interface height (lower wall) leads to a different type of phase transition as shown by T. Hwa and one of us some years back [@MN3]. In what follows, and following the nomenclature introduced in [@Review], we will use the term MN2 to name this class.
It can be easily shown that a KPZ equation with positive non-linearity and a lower wall is completely equivalent to a KPZ with a negative non-linearity coefficient and an upper wall [@Review]: one just have to change the sign of the height variable in the KPZ-like equation to verify this.
The MN2 class is of great importance in the context of alignment of DNA and other biological sequences. It has been argued by Hwa and collaborators, that the phase transition appearing upon changing the, so called, scoring parameter in the commonly used alignment algorithm can be mapped into the MN2 critical point [@Hwa]. It is also related to some instances of non-equilibrium wetting [@Haye1]. For some other applications and physical instances within this class see [@Review] and references therein.
While the MN1 class has been extensively studied, specially after its connections with non-equilibrium wetting [@Haye1; @Haye2; @Lisboa] and with the problem of synchronization in extended systems were established [@synchro; @Review], the MN2 class remains poorly studied. Furthermore, recent numerical analysis have revealed that the preliminary critical exponent values reported in [@Hwa] might be far from their true asymptotic values.
Aimed at clarifying these issues it is the purpose of this paper to analyze the MN2 phase transition in one dimensional systems using: i) mean-field and field-theoretical techniques and ii) numerical (Monte Carlo) analysis of different models claimed to belong to this class.
Finally let us stress that if the wall becomes attractive, rather than simply bounding, new phenomenology might appear. This is particularly interesting in the context of synchronization [@Review; @synchro]. This possibility will also be discussed along the paper.
The MN2 class
=============
Let us consider a KPZ equation with a positive non-linearity coefficient, $\lambda >0$, in the presence of a lower wall, $$\partial_t h(x,t) = a + b ~e^{-ph} + D \nabla^2 h + \lambda (\nabla h)^2
+ \sigma \eta(x,t).
\label{BKPZ}$$ where $h$ is a height variable, $a$ represents a constant drift while $ b ~e^{-ph}$ is a bounding wall. The parameter $p>0$ controls the wall penetrability, the limit $p \to \infty$ corresponding to a perfectly rigid (impenetrable) wall. It has been shown previously that for both the MN1 and the MN2 classes the magnitude of $p$ does not influence the asymptotic properties at criticality [@MN1; @MN2] (its sign, however, is important, as it determines whether the wall is a lower or an upper one). The same property applies to equilibrium systems ([*i.e.*]{}, for $\lambda =0$) [@lipowskyfisher]. $\eta$ is a stochastic white noise with $\langle \eta \rangle =0$ and $\langle \eta(x,t) \eta(x',t') \rangle = 2 \delta(t-t') \delta (x-x')$, where $\langle \cdot \rangle$ denotes an average over the distribution of the noise.
For a fixed value of $b$ the interface experiences a pinning-depinning transition at some value $a=a_c$.
Some remarks concerning the connection of the previous equation with wetting problems follow. When $h(x,t)$ is viewed as the distance separating a liquid-gas interface from a solid wall, Eq. (\[BKPZ\]) can then be interpreted as a dynamic model for nonequilibrium wetting. Under this perspective $a$ is the chemical potential difference between the liquid and the gas phases, $\langle h \rangle$ is the thickness of the wetting layer, and the wall is a rigid, physical substrate. At [*bulk phase coexistence*]{}, [*i.e.*]{} for the value of $a=a_c$ for which in the absence of the wall the interface does not move on average, $\langle h \rangle$ diverges at all temperatures above certain wetting temperature, $T_W$, while for $a \not= a_c$ the thickness of the liquid film can be big, but finite (pinned interface). The temperature here is controlled by the parameter $b$, which vanishes linearly with the mean-field wetting temperature as $T-T_W$. Thus, on approaching coexistence for $T>T_W$ ($b>0$ at the mean-field level), $\langle h \rangle$ diverges as $\langle h \rangle \sim |a-a_c|^{\beta_h}$. This transition, termed [*complete wetting*]{}, is always continuous and the value of the $\beta_h$ exponent depends on the nature of the forces between the particles in the fluid phases and the wall. In this paper only short-range, exponentially decaying interactions between all the particles and the substrate (as described by Eq. (\[BKPZ\])) are considered.
The change of variables $n=\exp(-h)$ transforms Eq. (\[BKPZ\]) into the MN2 Langevin equation: $${\partial}_{t} n(x,t) =
{\nabla}^{2} n - 2 {(\nabla n)^2 \over n}
-(a+1)n-bn^{p+1} + n \eta,
\label{mnminus}$$ where for the sake of simplicity we have set $\lambda=D=\sigma=1$ and Ito calculus [@Ito] has been used (different coefficients for the Laplacian and the KPZ non-linear term could be reabsorbed using $n=\exp(-\alpha h)$ with a proper choice of $\alpha$). This transformation maps the depinning from the wall $\langle h \rangle \to \infty$ to a transition into an absorbing state $\n \to 0$. The physical equivalence between the cases $\lambda>0$ with a lower-wall ($p>0$), and $\lambda<0$ with an upper wall ($p<0$), reflects in the fact that the same equation is obtained using $n=\exp(-h)$ and $n=\exp h$, respectively.
Observe that Eq.(\[mnminus\]) is identical to the MN equation describing the MN1 class [@MN1; @MN2; @Review], except for the presence of an extra term $ (\nabla n)^2 /n $. This term can also be written as $(\nabla n) \cdot (\nabla \ln(n)) = (\nabla n ) \cdot (\nabla h)$ suggesting that the interface language is the natural one for this class. In fact, except for the factor 2 in front of $(\nabla n)^2 /n$ Eq. (\[mnminus\]) coincides with the Cole-Hopf transform of $$\partial_t h (x,t) = \nabla^2 h + a +b e^{-ph} + \eta(x,t)$$ that describes the growth of wetting layers toward their equilibrium state [@lipowsky] (observe that this is just the equilibrium, Edwards-Wilkinson model, in the presence of a bounding wall). Note also that the factor $2$ in Eq. (\[mnminus\]) cannot be readsorbed by reparametrizing.
Finally, let us underline that in the regime where $b<0$ the wall becomes attractive (which might be necessary to describe some physical situations as, for instance, synchronization problems as said in the introduction) and a new term, say, $c \exp(-2h)$ (equivalently $c n^{2p+1}$) with $c>0$ has to be added to stabilize the equation.
Having presented the equations defining the model, in the forthcoming sections we study the associated physics by using i) mean field approaches, ii) field theory, and iii) numerical, Monte Carlo simulations combined with scaling arguments.
Mean-field approaches
=====================
Mean-field approaches to Eq. (\[BKPZ\]) can be implemented with several degrees of sophistication. A crude approximation consists of ignoring the noise and spatial variations. At this level one trivially gets that the order parameter $\langle h \rangle$ vanishes as $a \to 0$ with an exponent $\beta_h=0$. When applied to Eq.(\[mnminus\]), this approximation yields a shifted critical point at $a_c=-1$ and the usual result for the order parameter critical exponent, $\beta_n=1/p$. But, as experience with other systems with multiplicative-noise dictates, neglecting completely the noise is a too crude approximation, that eliminates most of the characteristic traits of MN physics.
Now the effect of allowing a spatially varying order parameter and taking the noise into consideration are examined. The Laplacian is discretized as $$\nabla^2 n_i ={1 \over 2d} \sum_{j} (n_j-n_i) \approx \n
-n_i,$$ where the sum runs over the nearest neighbors of $i$ and a large system dimensionality has been assumed. Similarly, the square gradient term can be written as $${(\nabla n)_i^2 \over n_i} \approx {\langle n^2 \rangle \over n_i}
-2\n + n_i.$$ The one-site stationary probability distribution is then readily obtained from the associated Fokker-Planck equation, $$\begin{aligned}
P_{st}\Big(n,\n\Big) &\propto& {1 \over n^2} \exp \int^n {
(\nabla x)_i^2-2(\nabla x)_i^2/ x_i -(a+1)x_i-bx_i^{p+1} \over x_i^2} \ dx_i,
\nonumber \\
&\approx &{1 \over n^{a+6}} \exp \Bigg(-5{\n \over n}-b {n^p\over p}
+{\langle n^2 \rangle \over n^2}\Bigg).
\label{pmnminus}\end{aligned}$$ where $\n$ and $\langle n^2 \rangle$ have to be calculated self-consistently. For $a< a_c=-5$, $P_{st}$ is not normalizable what means that the stationary state is the absorbing phase $\n =0$. For $\n \not=0$, however, the non-analyticity of $\exp[\langle n^2 \rangle / n^2]$ at $n=0$ again renders $P_{st}$ non-normalizable. As a result, there is no well-defined active phase at this mean-field level for Eq. (\[pmnminus\]). Similar problems are found if the same type of approach is applied to Eq. (\[BKPZ\]) instead of Eq. (\[mnminus\]).
To avoid the presence of the square gradient term in Eq. (\[mnminus\]), and the complications of its, somewhat arbitrary, discretization, we resort to a different change of variables. After $n=\exp h$, $$\begin{aligned}
\partial_t n &=& \nabla^2 n +(a+1)n +bn^{3-p} +n \eta, \nonumber \\
P_{st} &\sim& {1 \over n^{2-a}} \exp\bigg[{b n^{2-p} \over 2-p}-{\n\over n}\bigg].
\label{mnplus}\end{aligned}$$ These equations are simpler than (\[mnminus\]) and (\[pmnminus\]), but at the cost that $\n$ is no longer an order parameter as, at the transition point $\n \to \infty$ rather than going to $0$. Furthermore, Landau expansions only make sense when the order parameter vanishes at the critical point, making the whole approach inconsistent. One possible way to circumvent this problem is to monitor $m \equiv 1/n$, and study $\tilde{P}_{st}(m) dm =P_{st}(n) dn$, but given the non-Gaussian nature of the probability distribution the substitution $\n = 1/\langle m \rangle$ is likely to be incorrect, and there seem to be no safe way to proceed.
Summing up, non-trivial mean-field approaches detect some problems with the model under consideration, and are not able to predict a phase correct phase diagram. A sound mean-field approximation, needs therefore to be found. Notice that none of these problems occur in the case of a negative KPZ non-linearity, [*i.e.*]{} in the MN1 class, where a standard mean-field approximation yields qualitatively correct results (see [@Review; @Lisboa] and references therein).
Field theoretical considerations
================================
The Langevin equation for MN1 is known to be super-renormalizable, [*i.e.*]{} Feynman diagrams can be computed to all orders and resummated. This does not imply that critical exponents can be computed in all the cases, as the renormalization group flow-equation has runaway trajectories supposed to converge to a [*strong coupling fixed point*]{}. But at least, the correct phase diagram, including strong and weak coupling fixed points can be obtained.
For the MN2 the situation is far more complicated, as can be [ *a priori*]{} anticipated given the failure of mean-field approaches. The extra term, $ (\nabla n)^2 /n $, being singular in $n$ precludes the use of perturbative expansions around $n=0$. Given the lack of a non-perturbative approach to the KPZ and MN strong coupling fixed points, there is not much we can add to this section, except that there is a promising attempt to tackle this and related KPZ-like problems. There is a formalism, developed by Fogedby, aimed at developing a strong coupling theory for KPZ based on a semiclassical or WKB approximation applied upon the Martin-Siggia-Rose generating functional [@fogedby]. Its main advantage is that it does not involve expansions around classical noiseless solutions, but around classical (extremal) noisy solutions. It would be very interesting to extend these ideas to KPZ problems in the presence walls, namely, to the multiplicative noise universality classes MN1 and MN2.
Numerical results
=================
Owing to the failure of standard (non-trivial) mean-field approaches and lacking so far of an alternative analytical route, numerical methods are required to glean insight into the system properties. We have carried out simulations of two surface growth models. Both of them, in the absence of walls, are known to belong to the KPZ universality class. An extra rule is then added to generate a bounding wall, as described above.
Model 1
-------
Our first model was introduced in [@krugs] and is defined as follows: the surface position at time $t$ above a site $x$ on a one-dimensional lattice of size $L$ is given by a continuous height variable $h_t(x)$. A new height configuration is then generated in a three-step process.
1. Each lattice site $h_t(x)$ is updated according to $h'_t(x)=h_t(x)+a+\eta_t(x)$. $\eta_t(x)$ is a random number uniformly distributed in \[0,1\] and $a$ is a constant drift term analogous to that of (\[BKPZ\]).
2. The configuration is changed to $h_{t+1}(x)=\min[h'_t(x\pm 1)+\gamma,h'_t(x)]$, where $\gamma$ is a constant whose precise value is not essential for the final results. We have set without loss of generality $\gamma=0.1$ as in previous numerical analyses.
3. A hard wall at $h=0$ is introduced by way of the additional rule $h_t(x)=\min[h_t(x),0]$ [@MN3].
Finally, periodic boundary conditions are imposed and $h_t(x)$ is initially set to $0$.
The continuum counterpart of this model is known to be a KPZ equation [@krugs] with $\lambda<0$ in the presence of an upper wall [@note] which, as remarked before, is equivalent to the case $\lambda >0$ and a lower wall, and corresponds to the MN2 class. Numerical results for this model were first presented in [@MN3] and seemed to be consistent with mean-field ([*i.e.*]{} single-site) like behavior. However, a similar problem recently studied in the context of synchronization has revealed inconsistencies probably due to insufficient statistics [@Review; @synchro]. In this subsection improved simulation results are provided upon revisiting the analysis reported in [@MN3] for larger system sizes and longer sampling times. The case of an attractive wall, not included in [@MN3], is also considered.
First, we take up the case of a simple (non-attractive) wall, corresponding to $b>0$ in Eqs. (\[BKPZ\]) and (\[mnminus\]). Figure (1) shows how the steady order parameter $\n$ changes with the system size $L$. Within the active phase, it saturates to a constant value, while in the absorbing one it bends down and decays exponentially. Our best estimate for the critical point is $a_c=1.57433(2)$ and from the slope of the curve we get $\beta_n/\nu_\bot \approx 0.33(2)$ [@note2]. This value of $a_c$ corresponds to the point where a free interface (far from the wall) has zero average velocity. The time evolution of $\n$ at the critical point for different system sizes (Figure (2)) behaves like $\langle n(t) \rangle \sim t^{-\theta_n}$, with $\theta_n =0.215(15)$ or, equivalently, $\langle h(t) \rangle \sim t^{\theta_h}$, with $\theta_h =0.355(15)$. As for the exponents $\beta$, which govern the saturation of the order parameter within the active phase, they have been computed using the largest available system sizes ($L=1600$ and $3200$). The best fit to $ \langle X \rangle \sim |a-a_c|^{\pm \beta_X}$ yields $\beta_n \approx 0.32 (3)$ (for $\langle n \rangle$) the and $\beta_h =0.52(2)$ (for $\langle h\rangle$), respectively. The error margin is typically larger here than for other exponents due to the sensitivity to the uncertainty in the determination of the critical point.
It was proved in [@MN1; @MN2] that these exponents must satisfy the scaling relation $z= \beta/(\nu_\bot \theta)$, where $z$ is the dynamic exponent of the KPZ equation. The exact value for $z$ in $d=1$ is $3/2$, thereby $z = 0.33 / 0.215 = 1.5(1)$ in agreement with the prediction. In addition, also from [@MN1; @MN2], $\nu_\bot =1$, and from our direct measurements we get $(\beta_n/\nu_\bot)/ \beta_n = 0.33/0.32$, implying $\nu_\bot \approx 1$. In terms of $h$ and assuming $\nu_\bot=1$, $z=0.52/.355= 1.5(1)$ which, again, is compatible with $3/2$ within error bars.
The two alternative, but equivalent, mathematical descriptions of the MN2 class in terms of $h$ and $n=\exp(-h)$ can be related noting that the latter is essentially the density of sites at zero hight, $n(x,t)=\delta_{h(x,t),0}$ [@Haye3]. We have verified that $n$ and $\delta_{h(x,t),0}$ exhibit the same asymptotic scaling behavior.
{width="8cm"} \[betanusin\]
{width="8cm"} \[theta\]
{width="8cm"} \[betanucon\]
Attractive walls can also be simulated within this model by simply substituting $a$ by $a-b\delta_{h,0}$, where $b<0$ and the sign convention is chosen to keep the analogy with Eq. (\[BKPZ\]). This means that wherever the interface in attached to the wall it experiences an additional (“sticky”) force pushing it against the wall. Extensive Monte Carlo simulations for $b=-0.3$ show that the previous results, in what respect universal features, carry over without change, the only difference being that the approach to asymptotics is slower. Upon increasing the attractiveness of the wall the transients become longer. The estimate for the critical point is the same one as before (this is due to the fact that the free interface is not affected by variations of the attractiveness parameter). Again, our best estimates for the critical exponents are $\beta_n/\nu_\bot =0.32$ and $\theta_n=0.215$ (see Figure (3)). For $b=-0.4$ we still observe a second-order phase transition with a crossover to the mentioned exponents. For $b=-0.5$ transition becomes first-order but it still occurs at $a=a_c$. Within the active phase all the sites are closed to the wall and the order parameter is $1$, but it suddenly changes to $0$ upon decreasing $a$ and hysteresis is observed for slightly subcritical values $a$ (the interface is pinned for up to long times). Therefore, a tricritical point must exist between $0.4$ and $0.5$. We have identified it at $b=-0.42(1)$. The tricritical behavior has not been analyzed.
Let us stress that, contrarily to what happens for the MN1 class, where the presence of an attractive wall induces a new and rich phenomenology (including a broad region of phase coexistence and directed-percolation type of transitions [@Review]), the addition of “attractiveness” has a very mild effect here. Basically, it just shifts the position of the critical point and induces a first-order phase transition for very strong attractions.
------------- ------------ ------------- --------------------- --------------- ------------- --------------
$\beta_n$ $\nu_\bot$ $\beta_n/\nu_\perp$ $z$ $\theta_n$ $\eta$
Model 1 $0.32(2)$ $0.97(5)$ $0.34(2)$ $1.55(5)$ $0.215(15)$ not measured
Model 2 $0.325(5)$ $\approx 1$ $0.33(2)$ $\approx 1.5$ $0.215(5)$ $\eta = 0.8$
------------- ------------ ------------- --------------------- --------------- ------------- --------------
: Critical exponents for the MN2 class in $d=1$.
Model 2
-------
In order to verify the robustness and eventual universality of the previous results, we have performed a second study of a different model. It is a restricted solid-on-solid (RSOS) model, a variant of the single-step model introduced in [@krugs], in the presence of a wall. A similar model has been recently studied in the context of synchronization transitions [@Ahlers], to study MN1 type of transitions. Initially the wall is located at $h_w =0$ and a grooved interface is placed beneath it, [*i.e.*]{} the interface has negative height at all the positions.
The dynamics proceeds as follows: At each time step, a site is randomly picked from a one-dimensional lattice of length $L$ and its height decreased two units, $h(i) \to h(i)-2$, provided that $h(i)>h(i+1)$ and $h(i)>h(i-1)$, [*i.e.*]{} provided that it is a local maximum. Should the the RSOS constraint be violated, the trial is discarded and repeated. Every $2(L-1)/[1-2\delta v (1-L^{-1})]$ steps the wall retreats one unit and, simultaneously, the interface is moved downwards by two units wherever it lies above the wall [@Ahlers]. The difference between the wall and interface velocities, $\delta v$, acts as the control parameter: if $\delta v$ is negative, the interface eventually depins from the wall, while for $\delta v>0$ it remains pinned). It can be easily shown, using random walk arguments that for the chosen wall velocity the system seats at its critical point. By varying it, we have a control parameter. The possibility of computing analytically the critical point largely simplifies the numerical analysis, and makes of this an efficient discrete model.
The quantities monitored are $\langle \exp(h_w-h) \rangle$ and $h_w-h$. We have measured the exponents $\beta_n$ and $\theta_n$ for a system of $L=2^{20}$ sites. Our results lead to $\beta_n=0.325(5)$ and $\theta_n=0.215(5)$, in excellent agreement with those of model 1 (Figure (4)). Once more, assuming $\nu_\bot=1$, we get $z \approx 1.51$ in good agreement with the scaling laws. In addition, we have also measured the spreading exponent, $\eta$, that characterizes the number of pinned sites. It is computed averaging over all the runs and starting with an initial condition with a single point attached to the wall. Our best estimate is $\eta =0.80(2)$ (Figure (5)). Measuring the surviving probability, and therefore the exponents $\delta$ and $\zeta'$, is a delicate technical point because it is hard to decide when the activity of a run has ceased. We have not tackled this problem here. Lastly, we obtain $\theta_h =0.34(5)$, which is again in good agreement with the value reported for our Model 1.
{width="8cm"} \[sidebyside\]
{width="6cm"} \[fattaheta\]
We have also considered different variations of the model in which the interface can penetrate the wall at some points, [*i.e.*]{} the wall is not perfectly rigid. None of the universal features seem to be affected by this change.
In conclusion, all these results support strongly the existence of robust universality in the MN2 class.
As a matter of consistency, we have modified the algorithm of the model to simulate a lower wall, and therefore a case expected to be in the MN1 class. We obtain the set of exponents $ \beta_n=1.69, \theta_n= 1.19$ and $\eta = -0.4$, all of them in agreement we previously reported results and showing that the upper and lower problem belong to different universality classes [@Review; @MN2; @MN4].
Numeric integration of stochastic differential equations
--------------------------------------------------------
As a further test for universality, we have numerically integrated Eq. (\[BKPZ\]) using a Milshtein’s algorithm [@Maxi]. A system size of $L=2000$ was considered and the time step and mesh size were set to $\Delta t =0.001$ and $\Delta x=1$, respectively. For simulation times up to $t=10^{6}$ ($10^9$ trials per site) our results lie far from the asymptotic regime. We do not discard numerical instabilities in the integration scheme, as it is known that results from numerical integrations may not agree with the predictions from the continuum KPZ [@KPZproblem]. The Cole-Hopf transform is a standard way to account numerically for the integration of bounded KPZ equations and, indeed, Langevin equations with MN are by far more stable than their interface-language KPZ-like counterparts [@MN2; @MN4]. Nevertheless, we have also found numerical problems when integrating (\[mnminus\]), either when the extra term is written in logarithmic form or averaging for smoothing the gradient. All numerical attempts are unstable nearby the absorbing state, owing to the presence of the extra singular term. We leave, therefore, the numerical integration of a continuous Langevin equation, representative of the MN2 class as an open, challenging problem.
Discussion
==========
We have characterized the MN2 universality class, or analogously its bounded KPZ counterpart, which, as commented above, accommodates different physical phenomena. We have studied it from, somehow deceptive mean-field and field theoretical approaches, as well as by numerical studies. None of the analytical methods provides a satisfactory description of the phase transition present in this class. On the other hand, Monte Carlo simulations of two different discrete interface models, argued to belong to this universality class, give a firm evidence for the existence of a robust universality class. Contrarily to what previous simulations seemed to indicate [@MN3], our results are not a simple extension of the ones obtained for one-site, implying that spatial correlations play an important role. Table 1 gathers the values of the critical exponents in terms of $n$ for the two discrete models considered in this paper. From the Monte Carlo estimates, it cannot be discarded that they adopt the rational values $\beta_n=1/3$ and $\theta_n=2/9$, which combined with the exact values derived in [@MN1; @MN2], $\nu_\bot=1$ and $z=3/2$, would lead to $\beta_n/\nu_\bot= 1/3$. Note that our results do not compare well with those of the nonequilibrium wetting model reported in [@Haye2]. We believe this is probably due to the extremely long transients known to be present in that model. We have verified that the transition point is located at the same value of the control parameter for any value of the “attractiveness” parameter ($b$ in Model 1), either representing an attractive or a non attractive wall. For strong enough attractive walls, [*i.e.*]{} $b$ sufficiently negative, the transition becomes first order as in [@Haye2], while if the wall is weakly attractive then it remains in the MN2 class.
The problem of reaching a satisfactory analytical understanding, and even that of obtaining sound results from numerical integrations of the continuous Langevin-equation (in either the interface or the density language) representative of this class remains an open challenge.
Summing up, even though strong evidence is provided confirming the existence of a universality class ([*i.e.* ]{} the corresponding critical exponents are computed with good precision in one dimension and they are universal in two different discrete models), its theoretical description in terms of Langevin equations, contrarily to what happens for the closely related MN1 class, is far from satisfactory. In particular, the Langevin equation does not seem to admit sound mean-field solutions, nor is amenable to be treated by means of standard perturbative field theoretical tools, nor it admits a stable numerical integration. Identifying the physical causes at the root of these difficulties is a challenge for future research.
**ACKNOWLEDGMENTS**
F.S. acknowledges financial support from the Fundação para a Ciência e a Tecnologia, contract SFRH/BPD/5654/2001. Financial support from the Spanish MCyT (FEDER) under project BFM2001-2841, and from the AECI, are also acknowledged.
See J. Garc[í]{}a-Ojalvo, and J. M. Sancho, [*Noise in Spatially Extended Systems*]{}, Springer, New York, 1999; and references therein. See also, J. M. Sancho and J. Garc[í]{}a-Ojalvo, in Lecture Notes in Physics [**557**]{}, p.235, ed. J. A. Freund and T. Pöschel, Springer-Verlag, Berlin (2000).
M.A. Muñoz, preprint 2003, cond-mat/0303650.
M. Kardar, G. Parisi and Y. C. Zhang, Phys. Rev. Lett. [**56**]{}, 889 (1986).
T. Halpin-Healy and Y.-C. Zhang, Phys. Rep. [**254**]{}, 215 (1995); and references therein.
A. L. Barabási, H. E. Stanley, [*Fractal Concepts in Surface Growth*]{} Cambridge University Press, Cambridge, 1995; and references therein.
G. Grinstein, M.A. Muñoz, and Y. Tu Phys. Rev. Lett. [**76**]{}, 4376 (1996).
Y. Tu, G. Grinstein and M.A. Muñoz, Phys. Rev. Lett. [**78**]{}, 274 (1997).
M.A. Muñoz and T. Hwa, Europhys. Lett. [**41**]{}, 147 (1998).
W. Genovese and M.A. Muñoz, Phys. Rev. E [**60**]{}, 69 (1999).
H. Hinrichsen, R. Livi, D. Mukamel, and A. Politi, Phys. Rev. Lett. [**79**]{}, 2710 (1997).
H. Hinrichsen, R. Livi, D. Mukamel, and A. Politi, Phys. Rev. E [**61**]{}, R1032 (2000).
F. de los Santos, M.M. Telo da Gama, and M.A. Muñoz, Europhys. Lett. [**57**]{}, 803 (2002); Phys. Rev. E [**67**]{}, 021607 (2003); Proceedings of the 7th Granada Seminar on Computational Physics. Ed. J. Marro and P. L. Garrido; Am. Inst. of Phys. 661 (2003). Cond-mat/0211124.
M.A. Mu[ñ]{}oz and R. Pastor Satorras, Preprint. Cond-mat/0301059.
T. Hwa and M. Lassig, Phys. Rev. Lett. [**76**]{}, 2591 (1996). See also, T. Hwa and M. Lassig, “Optimal Detection of Sequence Similarity by Local Alignment“ in Proceedings of the Second Annual Int. Conf. on Computational Molecular Biology (RECOMB98), S. Istrail, P. Pevzner, and M.S. Waterman eds, 109-116 (ACM Press, 1998); and references therein. R. Olsen, T. Hwa and M. Lassig, ”Optimizing Smith-Waterman Alignments" in Pacific Symposium on Biocomputing 4, 302-313 (1999).
R. Lipowsky and M.E. Fisher, Phys. Rev. B [**36**]{}, 2126 (1987).
The sole difference between utilizing the Ito or the Stratonovich conventions, in this case, is a trivial shift in $a$ [@VK; @Gardiner].
N.G. van Kampen, [*Stochastic Processes in Physics and Chemistry*]{}, North Holland, Amsterdam, 1981.
C.W. Gardiner, [*Handbook of Stochastic Methods*]{}, Springer Verlag, Berlin and Heidelberg, 1985.
R. Lipowsky, J. Phys. A [**18**]{}, L585 (1985).
H. D. Fogedby, Phys. Rev. E. [**57**]{}, 4943 (1998). H. D. Fogedby, Cond-mat/0303632.
J. Krug, Adv. in Phys. [**46**]{}, 139 (1997). J. Krug and H. Spohn, in [*Solids far from equilibrium*]{}, Ed. C. Godrèche, Cambridge University Press, (1991).
Observe that the presence of the function “$\min$” is the second step of the algorithm induces a negative average velocity, implying that in the continuum counterpart $\lambda$ has to be negative. On the other hand the last step $\min{h(x,t),0}$ obviously generates an upper wall.
We use the subscript $_n$ to denote exponents related to the order parameter $n$ and $_h$ for exponents associated with the average height.
H. Hinrichsen, cond-mat/0302381.
V. Ahlers and A. Pikovsky, Phys. Rev. Lett. [**88**]{}, 254101 (2002). V. Ahlers, Ph. D. thesis. http:// www.stat.physik.uni-potsdam.de/ volker/publ.html. F. Ginelli, [*et al.*]{}, cond-mat/0302588.
M. San Miguel and R. Toral, [*Stochastic Effects in Physical Systems*]{}, to be published in [*Instabilities and Nonequilibrium Structures*]{}, VI, E. Tirapegui and W. Zeller, eds. Kluwer Academic Pub. (1997). (Cond-mat/9707147).
T. J. Newman and A. J. Bray, J. Phys. A [**29**]{}, 7917 (1996). C.H. Lam and F. G. Shin, Phys. Rev. E [**58**]{}, 5592 (1998).
| {
"pile_set_name": "ArXiv"
} |
---
author:
- 'A. Lazarian'
- 'G. Brunetti'
title: ' Turbulence, Reconnection and Cosmic Rays in Galaxy Clusters '
---
Guide to the review
===================
Mergers between galaxy clusters are the most energetic events in the present day Universe. During these mergers a fraction of the gravitational energy can be converted into fluid motions, i.e. shocks and turbulence, that generate magnetic fields and, through a variety of processes, accelerate relativistic protons and electrons [e.g., @Ryu2003; @CB2005; @BL2007; @Hoeft2007; @Pfrommer2008; @Skillman2008; @B2009; @Vazza2009]. In this short review we address some of the basic processes involved, namely, magnetic turbulence in galaxy clusters and the possibility of its observational studies (Sect. 2), magnetic reconnection (Sect. 3), as well as various ways of accelerating cosmic rays (Sect. 4). Our summary is presented in Sect. 5.
Turbulence in clusters of galaxies
==================================
Astrophysical fluids are characterized by high Reynolds numbers and are known to be turbulent [e.g., @Armstrong1995; @ChL2010; @Elmegreen2004; @McKee2007]. As properties of turbulent magnetized fluids are very different from laminar ones, the [*correct*]{} description of the particle acceleration requires taking into account the fundamental properties of magnetic turbulence as well as the mutual feedback of magnetic fields and cosmic rays in the turbulent fluids.
Properties of intracluster plasmas: instabilities induced by turbulence and effective collisions
------------------------------------------------------------------------------------------------
Turbulence in galaxy clusters is magnetized. A very important question is whether the MHD description of turbulence is applicable. When Coulomb collisions in the rarefied inter-galactic medium (IGM) are considered one has to conclude that the plasma is collisionless. This strongly affects the proparation of compressible modes, cosmic ray acceleration etc (see [@BL2007] and ref. therein). In what follows we argue that the degree of collisionality of astrophysical plasmas is underestimated when only Coloumb collisions are taken into account (see [@L2010; @BL2011a]).
It is well known that the mean free path of thermal protons due to Coulomb collisions in the hot IGM is very large, ten to hundred kpc [e.g., @Sarazin1986]. Fluids in such a collisionless regime can be very different from their collisional counterparts [@Schekochihin2005; @Schekochihin2006; @Schekochihin2010]. Several instabilities (e.g. firehose, mirror, gyroresonance etc) can be generated in the IGM in the presence of turbulence, leading to a transfer of the energy of large-scale compressions to perturbations on smaller scales.
Many instabilities have growth rate which peaks at scales near the particle gyroradius, making very large the scale separation between the energy injection scale and the scale where this energy is being deposited. The scattering induced by instabilities dramatically [*reduces the effective mean free path*]{} of thermal ions [*decreasing the effective viscosity*]{} of the IGM and making plasmas [*effectively collisional*]{} on smaller scales. Indeed, charged particles can be randomized if they interact with perturbed magnetic field. If this field is a result of plasma instabilities, [*the process can be viewed as the collective interaction of an individual ion with the rest of the plasma*]{}, which is the process mediated by magnetic field. As a result, the fluid would behave as collisional on scales less that the Coulomb mean free path. This issue has been addressed in [@LBe2006] for the case of a collisionless fluid subject to the gyroresonance instability that is driven by the anisotropy of the particle distribution in the momentum space that arises from magnetic field compression; the larger the magnetic field compression, the higher the anisotropy induced and the higher is the instability growth rate. They found that the turbulent magnetic compressions on the scale of the mean free path and less are the most effective for inducing the instability[^1]. As the scattering happens on magnetic perturbations induced by the instability, the mean free path of particles decreases as a result of the operation of the instability. This results in the process being self-regulating, i.e. the stronger the turbulence at the scale of injection, the smaller is the mean free path of plasma particles and the larger is the span of scales over which the fluid behaves as essentially collisional.
This induces an interesting picture where the mean free path of plasma protons depends on the level of compressions induced by turbulence and the mean free path is determined not by Columb collisions, but scattering on magnetic field inhomogeneities at the Larmor radius of thermal protons. The peculiar feature of this picture is that the aforementioned magnetic field perturbations are not part of the normal turbulent cascade, but results of compressible turbulent motions at much larger scales. Thermal protons do not scatter each other through electric interactions, but participate in non-local interactions mediated by the perturbed magnetic field. [*The higher the level of comressible turbulence, the better is MHD description of the IGM*]{}.
MHD turbulence: brief summary of theory and main properties of turbulence in the IGM
------------------------------------------------------------------------------------
The last decade has been marked by substantial advances in understanding of magnetic turbulence in the MHD regime [e.g., @GS1995; @LV1999; @Cho2000; @Muller2000; @Lithwick2001; @Cho2002; @ChoL2002; @ChoL2003; @BeL2010; @Kowal2010].
The presence of a magnetic field makes MHD turbulence anisotropic [@Montgomery1981; @Matthaeus1983; @Higdon1984; @Oughton2003]. The relative importance of hydrodynamic and magnetic forces changes with scale, so the anisotropy of MHD turbulence does too. A landmark event in this was a seminal work by [@GS1995] (GS95) which contained ideas that radically changed the further development of the subject. The corner stone of this model was the so-called [*critical balance*]{} idea which provided the analytical relation between the fluctuations parallel and perpendicular to the magnetic field. It also contains prophetic statements about mode coupling, providing guidelines for generalization of the model from the incompressible to compressible MHD.
The original model was improved in the subsequent publications. For instance, GS95 uses the closure relations that employ in the global system of reference related to the mean field, which, in fact, is an incorrect system to be used for the critical balance description. In [@LV1999] and later publications [e.g., @Cho2000; @Maron2001; @Cho2002] the importance of the [*local system of reference*]{}, which is defined by the local direction of the magnetic field of a wave packet, was reviled. The local system of reference was employed in the successful testing of the GS95 model. In addition, [@LV1999] generalized the GS95 model for the case when the turbulent injection velocity at the injection scale is less than the Alfvenic velocity.
The predictions of the GS95 model are in rough agreement with numerical simulations [e.g., @Cho2000; @Maron2001; @Cho2002; @BeL2006], although some disagreement in terms of the measured spectral slope was noted. This disagreement produced a flow of papers with suggestions to improve the GS95 model by including additional effects like dynamical alignment [@Boldyrev2005; @Boldyrev2006], polarization intermittency [@BeL2006], non-locality [@Gogoberidze2007]. More recent studies in [@BeL2009; @BeL2010] indicate that numerical simulations do not have sufficiently extended inertial range to get the actual spectral slope[^2] and therefore worries about the “inconsistency” of the GS95 model are premature. Evidence of the GS95 spectrum for the MHD incompressible turbulence was recently obtained by [@Be2011].
We shall add parenthetically that in a number of applications the empirical so-called composite 2D/slab model of magnetic fluctuations is used. In the latter model, which is also known as [*two-component model*]{}, it is assumed that fluctuations can be described as a superposition of fluctuations with wave vectors parallel to the ambient large-scale magnetic field (so-called [*slab modes*]{}) and perpendicular to the mean field (so-called [*two-dimensional modes*]{}). It results in a [*maltese cross*]{} structure of magnetic correlations. This model was developed to account for the solar wind observations, which it does well by adjusting the intensity of the two components [e.g. @Matthaeus1990]. This theory of 2D fluctuations is consistent with the theory of weak Alfvenic turbulence [e.g., @Ng1996; @LV1999; @Galtier2000] but it can describe Alfvenic turbulence only over a limited range of scales. It may be treated as a parameterization of a particular type of magnetic perturbation dominated by the peculiarities of driving, but recent simulations by Gosh (2011, private communication) show that, at best, the model represents a special transient state of a not fully developed turbulence. In addition, slab modes do not arise naturally in turbulence with large-scale driving, as shown by MHD numerical simulations [@ChoL2002; @ChoL2003]. Thus we do not consider this model for clusters of galaxies.
The GS95 model of turbulence can be adopted to describe the Alfvenic part of MHD turbulent fluctuations in galaxy clusters. The model can be generalized also to compressible turbulence and even for supersonic motions numerical calculations show that the Alfvenic perturbations exhibit GS95 scaling [@ChoL2002; @ChoL2003; @Kowal2010]. We note that we consider MHD turbulence where the flows of energy in the opposite directions are balanced. When this is not true, i.e. when the turbulence has non-zero cross-helicity, the properties of turbulence depart substantially from the GS95 model[^3]. Solar wind presents a system with imbalanced turbulence. However, the degree of imbalance of turbulence in clusters of galaxies is unclear and we know that in compressible media the imbalance decreases due to reflecting of waves from pre-existing density fluctuations and due to the development of parametric instabililites [@DelZanna2001]. Similarly, we shall not discuss MHD turbulence at high magnetic Prandtl numbers, when the viscosity is much larger than resistivity [e.g., @Cho2002; @Cho2003].
The GS95 model of turbulence combined with several considerations on the macro- and micro-physics of the IGM allows for a basic picture of the properties of turbulence in galaxy clusters [e.g., @BL2007].
Cosmological numerical simulations show that large-scale turbulent motions are generated during the process of cluster formation ([@Dolag2005; @Iapichino2008; @Vazza2011], see also Nagai 2011, Iapichino 2011, Vazza 2011, this conference). These motions, injected at large scales $L_o \sim 300-500$ kpc, are believed to provide the driver for turbulence at smaller scales. The typical velocity of the turbulent eddies at the injection scale is expected to be around $V_L \sim 500-700$ km/s which makes turbulence sub–sonic, but strongly super–Alfvénic. Turbulence at large scales is thus essentially hydrodynamic and – most likely – made of a mix of compressive and incompressive eddies. The cascading of compressive (magnetosonic) modes may indeed couple with that of solenoidal motions (Kolmogorov eddies).
Viscosity in a turbulent and magnetised IGM is strongly suppressed due to the effect of the bending of magnetic field lines and of the perturbations of the magnetic field induced by plasma instabilities (e.g., Sect. 2.1). The important consequence is that an inertial range in the IGM is established – for both solenoidal and compressive modes – down to collisionless scales where a fraction of the turbulent energy is channelled into acceleration/heating of CR and thermal plasma (see [@BL2007; @BL2011a] for discussion). At small scales – in the inertial range – the velocity of turbulent eddies becomes sub-Alfvenic and turbulence is described in the MHD regime. At these scales the coupling between Alfvén and compressible modes gets changed and only slow modes are cascaded by Alfvénic modes [e.g., @GS1995; @Lithwick2001; @ChoL2002]. The cascading of fast modes is not particularly sensitive to the presence of the other modes, fast modes remain isotropic while the spectrum of other modes becomes anisotropic.
Spectroscopic ways of turbulence studies
----------------------------------------
Recent observational advances to constrain turbulence in the IGM focussed on the broadening of lines in the X-ray spectra of galaxy clusters and provide interesting limits in the case of cool-core clusters [e.g., @Sanders2011].
Turbulence in clusters of galaxies can be studied in future using Doppler broderned emission. Here we briefly review techniques originally developed for studies of Doppler broderned emission and absorption lines in the interstellar medium research. These techniques, Velocity Channel Analysis (VCA) and Velocity Correlation Spectrum (VCS) have been developed by [@LP2000; @LP2004; @LP2006; @LP2008] (henceforth LP00, LP04, LP06, LP08, respectively) and successfully used for studying turbulence in diffuse and molecular gas [@Lazarian2009; @Padoan2009; @Chepurnov2010]. These techniques can be applied – at some extent – to the case of the IGM and future X-ray telescopes with very good spectroscopic capabilities (eg ASTRO-H) can be used for the studies.
The difference between the VCA and the VCS is how the data is being handled.
In the VCA technique the Position-Position-Velocity data cubes available through spectroscopic observations are analysed by taking spectrum of the velocity slice of the cube. The spectrum of the fluctuations is analysed while changing the thickness of the velocity slice and the analytical description of the statistics of the fluctuations in the PPV slices described in LP00 and LP04 is used to obtain both the spectrum of velocity and the spectrum of density fluctuations.
A different approach is used in the VCS technique, where fluctuations are analysed along the velocity coordinate. For the VCS technique one does not require good coverage of the Position-Position plane and a few spectral lines are sufficient to get the spectra of velocity and density (see Figure \[abs10\]).
![Illustration of VCS absorption studies of turbulence. [*Upper Panel*]{}: Schematic of measuring turbulence with absorption lines from point sources, e.g. stars, and an extended source, e.g. a galaxy. [*Lower panel*]{}: Velocity Coordinate Spectrum obtained using sampling of a turbulent volume along 10 lines of sight. The solid line corresponds to the theoretical expectations. Readapted from [@ChLazarian2009]. []{data-label="abs10"}](lazarian_fig1a.ps "fig:"){width=".5\textwidth"} ![Illustration of VCS absorption studies of turbulence. [*Upper Panel*]{}: Schematic of measuring turbulence with absorption lines from point sources, e.g. stars, and an extended source, e.g. a galaxy. [*Lower panel*]{}: Velocity Coordinate Spectrum obtained using sampling of a turbulent volume along 10 lines of sight. The solid line corresponds to the theoretical expectations. Readapted from [@ChLazarian2009]. []{data-label="abs10"}](lazarian_fig1b.ps "fig:"){width=".4\textwidth"}
New effects arise when strong absorption lines, which are in a saturated regime, are studied. The procedure for studying of the saturated lines is presented in LP08.
Our study of the effect of finite temperatures for the technique reveals that, unlike the VCA, the temperature broadening does not prevent the turbulence spectrum from being recovered from observations. Indeed, in VCA, gas temperature acts in the same way as the width of a channel. Within the VCS the term with temperature gets factorized and it influences the amplitude of fluctuations (LP06). One can correct for this term[^4], which also allows for a new way of estimating the interstellar gas temperature.
Another advantage of the VCS compared to the VCA is that it reveals the spectrum of turbulence directly, while within the VCA the slope of the spectrum should be inferred from varying the thickness of the channel. As the thermal line width acts in a similar way as the channel thickness, additional care (see LP04) should be exercised not to confuse the channel that is still thick due to thermal velocity broadening with the channel that shows the thin slice asymptotics. A simultaneous use of the VCA and the VCS makes the turbulence spectrum identification more reliable.
Both VCA and VCS are applicable to studies of not only emission, but also absorption lines. We note, that while dealing with emission lines we may face additional complications. For instance, Lazarian & Pogosyan (see LP00, LP04, LP06, LP08) treated the emissivities proportional to the density to the first power. Therefore, in terms of scalings, the emissivities and densities were interchangeable. This is not true, however, when the emissivities are proportional to $\rho^2$, as is the case of the recombination lines in plasma. The latter regime modifies the analysis. In particular, for the shallow spectrum of density, [@ChL2006] showed that the power spectrum of density $P_{\rho}\sim k^{-\alpha}$ has a shallow spectral index $\alpha<3$ emissivity spectrum $P_{\epsilon}\sim k^{\alpha_{\epsilon}}$ is $\alpha_{\epsilon}=2\alpha-3$ and this index should be used in all the expressions obtained of the VCA and VCS techniques. For the steep power law index of density, the power law indexes of the emissivity and density coinside for sufficiently large wavenumbers $k$.
Reconnection and Reconnection Diffusion
=======================================
It is generally believed that magnetic field embedded in a highly conductive fluid preserves its topology for all time due to magnetic fields being frozen-in [e.g., @Alfven1942; @Parker1979]. Although ionized astrophysical objects are almost perfectly conducting, they show indications of changes in topology, “magnetic reconnection”, on dynamical time scales [e.g., @Lovelace1976; @Priest2000]. Reconnection can be observed directly in the solar corona [e.g., @Yokoyama1995; @Masuda1994], but can also be inferred from the existence of large-scale dynamo activity inside stellar interiors [e.g., @Parker1993]. Also Solar flares [@Sturrock1966] and $\gamma$-ray bursts [e.g., @Fox2005; @Galama1998] are usually associated with magnetic reconnection.
To understand the difference between reconnection in astrophysical situations and in numerical simulations, one should recall that the dimensionless combination that controls the reconnection rate is the Lundquist number[^5], defined as $S = L_xV_A / \lambda$, where $L_x$ is the length of the reconnection layer, $V_A$ is the Alfvén velocity, and $\lambda=\eta c^2/4\pi$ is Ohmic diffusivity. Because of the huge astrophysical length-scales $L_x$ involved, the astrophysical Lundquist numbers are also huge, e.g. for the IGM they can be as high as $10^{20}$, while present-day MHD simulations correspond to $S<10^4$. As the numerical efforts scale as $L_x^4$, where $L_x$ is the size of the box, it is feasible neither at present nor in the foreseeable future to have simulations with sufficiently high Lundquist numbers.
Observations have always been suggestive that magnetic reconnection can happen at high speed, in spite of theoretical difficulties to explain the effect. At the same, the phenomenon of solar flares was suggestive that magnetic reconnection may be slow in order to ensure the accumulation of magnetic flux and suddenly gets fast to explain the observed fast release of energy. A model that can naturally explain this and other observational manifestations of magnetic reconnection was proposed in [@LV1999] (LV99). The model appeals to the ubiquitous astrophysical turbulence as a universal trigger and controller of fast reconnection.
To deal with strong, dynamically important magnetic fields, LV99 proposed a model of fast reconnection in the presence of sub-Alfvénic turbulence (see Figure \[LV99\]). They identified stochastic wandering of magnetic field-lines as the most critical property of MHD turbulence which permits fast reconnection. As we discuss more below, this line-wandering widens the outflow region and alleviates the controlling constraint of mass conservation. The LV99 model has been successfully tested recently in [@Kowal2009] (see also higher resolution results in [@L2010]). The model is radically different from its predecessors which also appealed to the effects of turbulence. For instance, unlike [@Speiser1970] and [@Jacobson1984] the model does not appeal to changes of microscopic properties of plasma[^6].
The LV99 model justifies the notion of turbulent mixing perpendicular to magnetic field lines. Indeed, LV99 showed that the GS95 model gets self-consistent only in the presence of the turbulence-induced reconnection with the rates predicted in LV99. Otherwise, the formation of the magnetic knots would change the character of the turbulent interactions.
The understanding of fast magnetic reconnection in the presence of turbulence induced the notion of “reconnection diffusion” that was described in [@L2005] and later used for describing different phenomena from star formation to heating of magnetic filaments in IGM [e.g., @Santos2010; @L2010]. The same concept was implicitly used earlier in [@Cho2003] where it was claimed that the heat conductivity of the IGM is influenced by the heat advection by turbulent eddies. Explicit calculations done by [@L2006] show that the heat conduction by turbulent eddies mixing magnetic field perpendicular to the local direction of magnetic field is the dominant way of heat transport in clusters of galaxies. The effect of reduced mean free path of thermal electrons induced by turbulence that we discussed above (Sect. 2) increases the relative importance of thermal transfer via reconnection diffusion. Rigorous arguments justifying the concept of reconnection diffusion can be found in [@Eyink2011].
Cosmic ray acceleration
=======================
Radio observations of galaxy clusters probe particle acceleration by shocks and turbulence in the IGM (Brunetti 2011, this conference for review on physics of cosmic rays (CR) in the IGM). In this Section we briefly discuss the importance of turbulence in the acceleration of CR and the connected issue of CR acceleration induced by magnetic reconnection.
Acceleration by magnetic turbulence
-----------------------------------
The interaction of turbulence and cosmic rays is a vital component of models of CR propagation and acceleration. It has been a concern from the very beginning of CR research [e.g., @Ginzburg1966; @Jokipii1966; @Wentzel1969]. To account for the interaction properly, one must know both the scaling of turbulence, the changes with time of turbulence spectrum due to the damping processes (e.g. with CR), and the interactions of turbulence with various waves produced by CRs.
Clusters of galaxies present magnetic fields of the largest extend and they are also considered on the role of the accelerators of the ultra high energy CR. The acceleration of particles in large (Mpc) regions in galaxy clusters is generally believed to happen via the second order Fermi process as a result of the interaction of particle–turbulence interactions [e.g., @BL2007; @Petrosian2008; @Brunetti2008]. Similarly, acceleration by magnetic turbulence is a very robust process that is likely be important for Solar flares, gamma ray bursts and many other astrophysical environments [e.g., @Hamilton1992; @Miller1996; @Schlickeiser2000; @Dermer2001].
MHD turbulence is the most important for the acceleration of particles of largest energies and it is vital to use the theoretically justified and numerically tested relations in the studies of particle acceleration. From the start of the work in this direction [e.g., @Chandran2000] it became clear that the earlier models for the acceleration and propagation of energetic particles that were based on ad hoc representation of turbulence are in error of many orders of magnitude as far as Alfvenic perturbations are concerned. [@YanL2002; @YanL2004] identified compressible fast modes as the principal agent for CR acceleration by MHD turbulence. As the aforementioned modes, unlike Alfvenic ones, are subject to rather strong damping, the description of the acceleration gets more complicated. In [@BL2007] we derived a comprehensive picture of compressible turbulence in galaxy clusters and studies CR acceleration considering all the relevant damping processes, with the results providing good correspondence with observations. More recently we extended this formalism to the case of the reacceleration of CR and of the secondary particles generated in the IGM via pp collisions [@BL2011b].
In addition, the accuracy of the particle acceleration using analytical theory has been improved by extending the quasi-linear theory to the regime of substantial perturbations of magnetic field and applied to the case of Solar flares [@YanL2008; @Yan2008]. The improved theory has been successfully tested with direct tracing of CR trajectories in data cubes obtained with results of direct MHD simulations of turbulence [@Beresnyak2011]. Future applications of these extensions to the case of galaxy clusters will be important.
Compressible turbulence interacts both with CR and with thermal particles. This interaction may also induce magnetic field perturbations (trough plasma instabilities, e.g. Sect.2) that may further come into play in the particle acceleration process. First attempts in this direction suggest that the fraction of turbulence that goes into CR acceleration increases when turbulent-induced instabilities are taken into account [@BL2011a].
Shock acceleration and turbulence
---------------------------------
Here we focus on the importance of turbulence in shock acceleration mechanisms. Shock acceleration is thought to be one of the principal accepted mechanisms of energetic particle acceleration. The shock induces compression and particles trapped between magnetic fluctuations ahead and behind shocks fill the acceleration every time they bounce back and forth between converging fluctuations. This is an efficient way of accelerating particles which results in the energy gain per bouncing to increase as the first power of the ratio of the particle velocity to that of light, i.e. $v/c$, making this process known as the first order Fermi acceleration.
Shock acceleration in galaxy clusters is believed to contribute the most of the CR (protons), while shock acceleration of CR electrons is the most popular model to explain the origin of radio relics ([@Ensslin1998], Ryu 2011, Brüggen 2011, this conference for review).
The necessity of particles to bounce back and forth limits the efficiency of the acceleration of high energy particles through a requirement that the energetic particle should have the Larmor radius less than size of the magnetic fluctuations that they bounce off. Therefore to increase the energy of the accelerated particles one should have strong magnetic field and strong magnetic fluctuations both in the preshock and postshock regions. The situation with the postshock region is relatively simple. Gas passing through the shocks is known to create turbulence [e.g., @Giacalone2007]. The turbulence is known to increase the magnetic field energy, enabling particles to scatter efficiently and return to the shock region for further acceleration. For the preshock region, most work was concentrated on instabilities that can enhance magnetic field. The most commonly discussed is the so-called Bell instability [@Bell2004] which is a non-resonant current driven instability, that can increase magnetic field in front of the shock. In [@BeJL2009] we proposed that a turbulent generation of magnetic field is happening in front of the shock, in the region which is called precursor. The properties of precursor and its formation in front of the shock are described in the literature [e.g., @Malkov2001]. As the precursor interacts with the density inhomogeneities preexisting in the medium in front of the shock, it gets perturbed, creating vorticity and turbulence. New studies of turbulent amplification of magnetic field [e.g., @Cho2009] provide the rates of magnetic field amplification by turbulence. These rates were made use of in [@BeJL2009] to obtain the values of the turbulent magnetic field that is generated in front of the shock. The corresponding estimates show that the preshock magnetic fields produced via this process are larger than those arising from the Bell instability and that they account for cosmic ray acceleration in galactic supernovae shock up to $10^{15}$ eV, the so-called “knee” of the cosmic ray spectrum.
Further development of this direction presents a very promising avenue of the cosmic ray acceleration research. Interesting boot strap processes are likely to be at work as generation of magnetic fluctuations in front of the shock increases the efficiency of the acceleration, contributing to the development of the precursor.
Acceleration induced by magnetic reconnection
---------------------------------------------
An important consequence of fast reconnection of turbulent magnetic fields that we discussed in Sect. 3 is the formation of a thick volume filled with reconnected magnetic flux loops. These 3D loops contract, presenting favorable conditions for energetic particle acceleration. This process of first order Fermi acceleration of energetic particles in reconnection regions has been described in [@deGouveia2005] (see also Fig. \[f5\]) for the situation when there is no back reaction of the accelerated particles on the reconnected magnetic flux. [@Drake2006] appealed to a similar process within their preferred model of collisionless reconnection and proposed that firehose instability can play a role of the feedback for the accelerated particles.
More recently, the acceleration in reconnection regions has obtained observational support. It was suggested in [@LazarianO2009] that anomalous CR measured by Voyagers are, in fact accelerated in the reconnection regions of magnetopose (see also [@Drake2010]). Such a model explains why Voyagers did not see any signatures of acceleration passing the Solar system termination shock. In a separate development, [@LazarianD2010] appealed to the energetic particle acceleration in the wake produced as the Solar system moves through interstellar gas to explain the excess of cosmic rays of the range of both sub-Tev and multi-TeV energies in the direction of the magnetotail Magnetic reconnection is ubiquitous in astrophysical circumstances and therefore it is expected to induce acceleration of particles in a wide range of astrophysical environments. For instance, the process has been already discussed for the acceleration of particles in gamma ray bursts [@L2003; @Zhang2011] and microquasars [@deGouveia2005]. We expect the process to be important for the acceleration of protons and electrons in galaxy clusters.
Numerical 2D simulations presented in [@Drake2010] confirmed high efficiency of particle acceleration in regions of magnetic reconnection. However, results in [@L2010] show that the process of acceleration happens rather differently in 2D and 3D situations. The 3D geometry shows a wider variety of acceleration regimes and this calls for much more detailed studies of the acceleration.
Summary
=======
The main points of our review can be summarized as follows
- Turbulence is essential for understanding of the IGM. On large scale the description of turbulence obtained in MHD can be used. Compressions induced by turbulence induce instabilities in the IGM, changing the mean free path of thermal ions. This should extend the range over which the MHD description of turbulence is applicable.
- Studies of turbulence in the IGM can get a boost if Doppler-broaderned spectral emission and absorption lines are used. The techniques originally developed and successfully used in the interstellar research, namely Velocity Channel Analysis (VCA) and Velocity Correlation Spectrum (VCS) are promissing for studing of turbulence in the IGM.
- Magnetic reconnection happens fast in turbulent media, which makes the models of MHD turbulence self-consistent. Fast magnetic reconnection makes MHD turbulence somewhat similar to hydrodynamic if one considers turbulent motions perpendicular to the local direction of magnetic field. Such motions can induce a process of “reconnection diffusion” which efficient heat transfer in the IGM.
- Magnetic turbulence is very important for particle acceleration in clusters of galaxies. It can accelerate particles through direct interactions with turbulent fluctuations. However, it can also modify shocks, inducing magnetic field generation in shock precursors and increasing the efficiency of high energy particle acceleration by shocks. In addition, it can enable fast magnetic reconnection which can accelerate particles within the thick reconnection regions.
AL thanks the NSF grant AST 0808118, NASA grant NNX09AH78G and the support of the Center for Magnetic Self Organization. GB acknowledge partial support from PRIN-INAF 2008, 2009.
Alfvén, H. 1942, Ark. Mat., Astron. o. Fys., 29B, 1
Armstrong, J. W., Rickett, B. J., & Spangler, S. R. 1995, , 443, 209
Bell, A. R. 2004, , 353, 550
Beresnyak, A. 2011, Phys. Rev. Lett., 106, 075001
Beresnyak, A., & Lazarian, A. 2006, , 640, L175
Beresnyak, A., & Lazarian, A. 2008, , 682, 1070
Beresnyak, A., & Lazarian, A. 2009, , 702, 1190
Beresnyak, A., Jones T., & Lazarian, A. 2009, , 707, 1541
Beresnyak, A., & Lazarian, A. 2010, , 722, L110
Beresnyak, A., Yan, H., & Lazarian, A. 2011, ApJ, 728, 60
Boldyrev, S. 2005, , 626, L37
Boldyrev, S. 2006, Phys. Rev. Lett., 96, 115002
Brunetti G., Lazarian A., 2007, , 378, 245
Brunetti, G., et al. 2008, , 455, 944
Brunetti, G., et al., 2009, , 507, 661
Brunetti G., Lazarian A., 2011a, , 412, 817
Brunetti G., Lazarian A., 2011b, , 410, 127
Cassano R., Brunetti G., 2005, , 357, 1313
Chandran B.D.G., 2000, Phys. Rev. Lett., 85 (22), 4656
Chandran, B. D. G. 2008, , 685, 646
Chepurnov A. & Lazarian, A. 2006, arXiv:0611465
Chepurnov, A., & Lazarian, A. 2009, , 693, 1074
Chepurnov, A., & Lazarian, A. 2010, , 710, 853
Chepurnov, A., et al., 2010, , 714, 1398
Cho, J., & Vishniac, E. T. 2000, , 539, 273
Cho, J., & Lazarian, A. 2002, Physical Review Letters, 88, 245001
Cho, J., Lazarian, A., & Vishniac, E. T. 2002, , 564, 291
Cho, J., & Lazarian, A. 2003, , 345, 325
Cho, J., Lazarian, A., & Vishniac, E. T. 2003, Turbulence and Magnetic Fields in Astrophysics, Lecture notes Physics, 614, 56
Cho, J., et al., 2008, , 693, 1449
de Gouveia Dal Pino, E. & Lazarian, A. 2005, A&A, 441, 845
Del Zanna, L., Velli, M., & Londrillo, P. 2001, , 367, 705
Dermer, C. D., & Humi, M. 2001, , 556, 479
Dolag, K., et al., 2005, , 364, 753
Drake, J. F., et al., 2006, , 443, 553
Drake, J. F., et al., 2010, , 709, 963
Eyink, G., Lazarian, A. & Vishniac, E. 2011, ApJ, submitted, arXiv:1103.1882
Elmegreen, B. G., & Scalo, J. 2004, , 42, 211
Enßlin T.A., et al., 1998, A&A, 333, 47
Fox, D. B., et al. 2005, , 437, 845
Galama, T. J., et al. 1998, , 395, 670
Galtier, S., Nazarenko, S. V., Newell, A. C. & Pouquet, A. 2000, J. Plasma Phys., 63, 447
Giacalone, J., & Jokipii, J. R. 2007, ApJL, 663, L41
Ginzbirg, V. 1966, Sov. Astr., 9, 877
Gogoberidze, G., Rogava, A., & Poedts, S. 2007, , 664, 549
Goldreich, P. & Sridhar, S. 1995, , 438, 763
Hamilton, R. J., & Petrosian, V. 1992, , 398, 350
Higdon, J.C., 1984, , 285, 109
Hoeft, M. & Brueggen, M. 2007, , 375, 77
Jacobson, A. R., & Moses, R. W. 1984, , 29, 3335
Jokipii, R. 1966, ApJ, 146, 480
Kowal, G., et al., 2009, , 700, 63
Kowal, G., & Lazarian, A. 2010, , 720, 742
Iapichino, L., & Niemeyer, J. C. 2008, , 388, 1089
Lazarian, A. 2005, Magnetic Fields in the Universe: From Laboratory and Stars to Primordial Structures., AIPC, 784, 42
Lazarian, A. 2006, , 645, L25
Lazarian, A. 2009, , 143, 357
Lazarian, A., & Vishniac, E. T. 1999, , 517, 700
Lazarian, A. & Pogosyan, D., 2000, ApJ, 537, 720 (LP00)
Lazarian, A., Petrosian, V., Yan, H., & Cho, J. 2003, arXiv:astro-ph/0301181
Lazarian, A., Vishniac, E. T., & Cho, J. 2004, , 603, 180
Lazarian, A. & Pogosyan, D., 2004, ApJ, 616, 943 (LP04)
Lazarian A., Beresnyak A., 2006, , 373, 1195
Lazarian, A. & Pogosyan, D., 2006, ApJ, 652, 1348, (LP06)
Lazarian, A. & Pogosyan, D., 2008, ApJ, 686, 650 (LP08)
Lazarian A., & Opher M., 2009, , 703, L8
Lazarian, A., & Desiati, P. 2010, , 722, 188
Lazarian, A., et al., 2010, Space and Planetary Science, doi:10.1016/j.pss.2010.07.020
Lithwick, Y., & Goldreich, P. 2001, , 562, 279
Lithwick, Y., Goldreich, P., & Sridhar, S. 2007, , 655, 269
Lovelace, R. V. E. 1976, , 262, 649
Malkov, M. A., & O’C Drury, L. 2001, Rep. Prog. in Physics, 64, 429
Maron, J., & Goldreich, P. 2001, , 554, 1175
Masuda, S., et al., 1994,, 371, 495
Matthaeus, W. H., Montgomery, D. C., & Goldstein, M. L. 1983, Physical Review Letters, 51, 1484
Matthaeus, W. H. & Lamkin, S. L. 1985, Phys. Fluids, 28, 303
Matthaeus, W. H. & Lamkin, S. L. 1986, Phys. Fluids, 29, 2513
Matthaeus, W. H., Goldstein, M. L., & Roberts, D. A. 1990, , 95, 20673
McKee, C. F., & Ostriker, E. C. 2007, , 45, 565
Miller, J. A., Larosa, T. N., & Moore, R. L. 1996, , 461, 445
Montgomery, D., & Turner, L. 1981, Physics of Fluids, 24, 825
Müller, W-C., Biskamp, D., 2000, PhRvL, 84, 475
Ng, C. S., & Bhattacharjee, A. 1996, , 465, 845
Oughton, S. 2003, Solar Wind Ten, 679, 421
Padoan, P., et al., 2009, , 707, L153
Parker, E. N. 1979, Oxford, Clarendon Press; New York, Oxford University Press, 1979
Parker, E. N. 1993, , 408, 707
Perez, J. C., & Boldyrev, S. 2009, Physical Review Letters, 102, 025003
Petrosian, V., & East, W. E. 2008, ApJ, 682, 175
Pfrommer C., Ensslin T.A., Springel V., 2008, , 385, 1211
Priest, E. & Forbes, T. 2000, in: Magnetic Reconnection: MHD theory and applications, Eds. E. Priest & T. Forbes, pp. 612, Cambridge, UK: Cambridge University Press
Ryu, D., et al., 2003, , 593, 599
Sanders, J. S., Fabian, A. C., & Smith, R. K. 2011, , 410, 1797
Santos-Lima, R., et al., 2010, , 714, 442
Sarazin C.L., 1986, Rev. Mod. Phys. 58, 1
Schekochihin A.A., et al., 2005, , 629, 139
Schekochihin A.A., Cowley S.C., 2006, PhPl 13, 6501
Schekochihin A.A., et al., 2010, , 405, 291
Schlickeiser, R., & Dermer, C. D. 2000, , 360, 789
Skillman S.W., et al., 2008, , 689, 1063
Speiser, T. W. 1970, , 18, 613
Sturrock, P. A. 1966, , 211, 695
Vazza F., Brunetti G., Gheller C., 2009, , 395, 1333
Vazza, F., et al., 2011, , 529, A17
Yan, H., & Lazarian, A. 2002, Physical Review Letters, 89, 1102
Yan, H., & Lazarian, A. 2004, , 614, 757
Yan, H., & Lazarian, A. 2008, , 673, 942
Yan, H., Lazarian, A., & Petrosian, V. 2008, , 684, 1461
Yan, H., & Lazarian, A. 2011, , 731, 35
Yokoyama, T., & Shibata, K. 1995, , 375, 42
Wentzel, D. G. 1969, , 156, 303
Zhang, B., & Yan, H. 2011, , 726, 90
[^1]: The larger scale compressions do still induce the instability, but their effect is reduced due to their reduced ability to induce large changes of $B$ over the time scale between scattering. The model is further elaborated and improved in [@YanL2011].
[^2]: [@BeL2010] noticed that the magnetic turbulence is less local compared with the hydrodynamic one and therefore one requires a substantially larger resolution to distinguish the actual spectral slope from the slope affected by the bottleneck.
[^3]: Among the existing theories of imbalanced turbulence [e.g., @Lithwick2007; @BeL2008; @Chandran2008; @Perez2009], all, but [@BeL2008] contradict to numerical testing in [@BeL2009; @BeL2010]
[^4]: To do this, one may attempt to fit for the temperature that would remove the exponential fall off in the spectrum of fluctuations along the velocity coordinate [@ChL2006]
[^5]: The magnetic Reynolds number, which is the ratio of the magnetic field decay time to the eddy turnover time, is defined using the injection velocity $v_l$ as a characteristic speed instead of the Alfvén speed $V_A$, which is taken in the Lundquist number.
[^6]: The nearest progenitor to LV99 was the work of [@Matthaeus1985; @Matthaeus1986], who studied the problem numerically in 2D MHD and who suggested that magnetic reconnection may be fast due to a number of turbulence effects, e.g. multiple X points and turbulent EMF. However, [@Matthaeus1985; @Matthaeus1986] did not observe the important role of magnetic field-line wandering, and did not obtain a quantitative prediction for the reconnection rate, as did LV99.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We show that the existence of a pair of zero-energy modes bound to a vortex carrying a $\pi$-flux is a generic feature of the topologically non-trivial phase of the $M-B$ model, which was introduced to describe the topological band insulator in HgTe quantum wells. We explicitly find the form of the zero-energy states of the corresponding Dirac equation, which contains a novel momentum-dependent mass term and describes a generic topological transition in a band insulator. The obtained modes are exponentially localized in the vortex-core, with the dependence of characteristic length on the parameters of the model matching the dependence extracted from a lattice version of the model. We consider in full generality the short-distance regularization of the vector potential of the vortex, and show that a particular choice yields the modes localized and simultaneously regular at the origin. Finally, we also discuss a realization of two-dimensional spin-charge separation through the vortex zero-modes.'
address:
- 'Department of Physics, Boston College, Chestnut Hill, Massachusetts 02467, USA'
- 'Instituut-Lorentz for Theoretical Physics, Universiteit Leiden, P.O. Box 9506, 2300 RA Leiden, The Netherlands'
author:
- Andrej Mesaros
- 'Robert-Jan Slager'
- Jan Zaanen
- 'Vladimir Juriči'' c'
title: 'Zero-energy states bound to a magnetic $\pi$-flux vortex in a two-dimensional topological insulator'
---
Introduction {#Intro}
============
Topological band insulators (TBIs) have recently opened a new frontier in both theoretical and experimental condensed-matter physics due to their peculiar properties (extensive reviews are in Refs. [@TIreview; @hasan-review]). These stem from the fundamental fact that TBIs are described by topological field theories [@TFTTI], thereby taking the interest in studying them far outside the standard condensed-matter domain. The novelty of TBIs is their protection by time-reversal symmetry (TRS), which leads to a non-trivial topological ${\mathbb Z}_2$ invariant [@fu-kane] of their free electron crystalline band structure, and to a description in terms of the topological $BF$ theory in (2+1)D and (3+1)D [@BF]. Striking consequences of their topological nature are the remarkable effective field theories describing the responses of TBIs, for instance, axion electrodynamics in (3+1)D [@response] and gravitational Chern-Simons in thermal response [@response; @gravanomaly]. Such theories are governed by anomalies, and the possibility of their direct study, e.g. through the Witten effect of axion electrodynamics [@Witteneffect], is of great interest and potential for both the high-energy and condensed matter communities. Further beyond the non-interacting case, there has been a proposal of (3+1)D fractional TBIs [@fracTI], having TRS and an axion angle different from $0$ or $\pi$; these have a description in the form of deconfined non-Abelian gauge fields explicitly realized using holography [@holographyTI].
The salient feature of TBIs, crucial for their characterization and detection, is that they are fully gapped in the bulk while possessing on their boundary gapless propagating modes protected by TRS. [@TIreview; @hasan-review] The presence of TRS however limits, fundamentally and especially experimentally, the availability of robust probes of such bulk-boundary correspondence, aforementioned anomalous responses, and ${\mathbb Z}_2$ topological order itself. For instance, in (2+1)D TBIs, on which we are focusing from now on, the charge Hall response vanishes, and instead a much more involved TRS invariant quantum spin Hall (QSH) effect characterizes the topological phase. It has been understood through numeric studies that a $\pi$-flux vortex, which actually preserves TRS, can play exactly the role of a ${\mathbb Z}_2$ probe in a QSH insulator through appearance of topologically protected zero-modes. [@BF; @TopProbes; @dung-hai1; @qi-zhang] In this paper, we analytically study the properties of $\pi$-flux vortex in presence and absence of ${\mathbb Z}_2$ order. The general way by which we achieve this is using the $M-B$ model initially constructed to describe the HgTe quantum well QSH insulator. [@bernevig; @konig] The salient and universal feature of the low-energy (continuum) version of this model is that it describes a topological phase transition between a trivial and non-trivial ${\mathbb Z}_2$ topological phase, through a massive Dirac-Schr" odinger theory. This field theory, especially in the presence of a $U(1)$ vortex, has not been widely studied for its own sake. A peculiar property of this theory is that the presence of both linear and quadratic kinetic terms together with the ordinary Dirac mass term allows for a gap-closing transition which changes the Chern number of the bands and the ${\mathbb Z}_2$ invariant. The same theory turns out to harbor analytically solvable zero-modes tied to $\pi$-flux vortex, but only in the non-trivial phase.
The relationship of the $\pi$-flux modes to the QSH phase and the question of their protection are general problems in the context of zero-energy fermionic modes bound to a topological defect. Namely, as Aharonov and Casher have shown in Ref. [@aharonov-casher], when non-relativistic (Schr" odinger) fermions are coupled to a magnetic flux carrying $n$ flux quanta there are precisely $n$ zero-energy modes in the spectrum of the Hamiltonian. Later, Jackiw in Ref. [@jackiw-prd84] has demonstrated that a magnetic flux with $n$ flux quanta hosts precisely $n$ zero-energy modes when coupled to relativistic Dirac fermions, and they are related to an index theorem for the Dirac Hamiltonian defined on a compact space.[@ansourian] On the other hand, the existence of fermionic zero-modes bound to a vortex in the complex scalar order parameter has actually been established in one and two spatial dimensions in the pioneering works by Jackiw and Rebbi[@jackiw-rebbi] and Jackiw and Rossi[@jackiw-rossi], respectively. Their existence is, at the deep mathematical level, tied to an index theorem that relates the spectral asymmetry of the corresponding Hamiltonian defined on an open space and a topological invariant of the background scalar fields.[@weinberg] At the same time, the results of Jackiw and Rebbi have been applied to the polyacetilene system through the Su-Schrieffer-Heeger model.[@SSH] However, the connection between Jackiw-Rossi and the $M-B$ model, relevant for the quantum spin Hall system, has been shown only very recently in Ref. [@chamon]. Namely, these authors have demonstrated that the Jackiw-Rossi model in the limit when Zeeman coupling and the chemical potential are large reduces to the $M-B$ model, and therefore the results presented in this work are also relevant for this problem. Moreover, a Hamiltonian of the same form as the $M-B$ Hamiltonian describes non-relativistic $p_x+ip_y$ superconductor, and the results of our work are thus relevant for this system as well. In particular, our solution for the zero-energy mode bound to a $\pi$-flux vortex in the quantum spin Hall state corresponds to the Majorana state in the core of a $\pi$-flux vortex in this topological superconductor.
The flux-carrying vortex is by its definition singular in real space. It is well known that a Hamiltonian with singular potential (e.g. Aharonov-Bohm flux vortex [@Persson:2006p260; @melikyan], Coulomb potential [@Fulop:2007p3442; @KhalilovSAE; @ParkSAE], delta function potential [@Jackiw:1991p1161]), once made Hermitian through a self-adjoint extension [@Weidmann], can exhibit finite or zero-energy bound states, even if the original Hamiltonian was scale-free. Therefore, the question of regularizing the vortex singularity, and thereby completely defining a Hermitian fermionic theory, becomes physically relevant. It fixes the real space profile of zero-modes at the vortex and the scattering phase shift there. This problem has been considered for (2+1)D [@Persson:2006p260; @2dDiracSAE], and (3+1)D [@3dDiracSAE] Dirac particles, as well as quasiparticles in superconductors [@melikyan]. The Dirac-Schr" odinger theory of the $M-B$ model has not been studied in this context before, and we find here the general form of the regularized theory in presence of a magnetic $\pi$ flux vortex. Furthermore, we show that a regularization corresponding to a thin solenoid limit surprisingly leads to localized zero-modes that vanish at the origin. In this paper, we first explicitly introduce the tight-binding and continuum $M-B$ models. Then we use the continuum $M-B$ model to analytically show that the $\pi$-flux vortex hosts precisely a pair of exponentially localized zero-energy modes, and therefore the states found numerically in Refs. [@dung-hai2; @qi-zhang] are indeed a generic feature of the $M-B$ model. Moreover, the relationship of these midgap states with the topological properties of the quantum spin Hall state are also considered. Namely, we show that these modes, in fact, exist in the entire range of parameters describing topologically non-trivial phase in the $M-B$ model with the gap opening near the zero momentum ($\Gamma$ point) in the Brillouin zone. For obtaining the explicit form of the midgap states, a short-distance regularization of the Hamiltonian is necessary due to the singularity of the vortex potential. We show that a particular choice, corresponding to a limit of infinitely thin solenoid, yields the modes non-singular at the origin. We then consider in full generality the problem of consistent regularization of the vortex potential, and show that the corresponding self-adjoint extensions are given in terms of parameters described by $U(2)$ matrices. Next, we test the prediction for the localization length of the zero-modes obtained from the continuum theory as a function of the transition driving parameter by comparing it to the results from a lattice regularization of the model. Finally, we discuss the quantum numbers of the obtained zero-energy states.
The $M-B$ tight-binding model on the square lattice
===================================================
We begin by considering a minimal tight-binding model proposed to describe a two-band quantum spin Hall insulator [@bernevig] $$\label{eq:tight-binding}
{\mathcal H}=\sum_{{\bf k}}\Psi^\dagger({\bf k})\left(\begin{array}{cc}H({\bf k})& 0\\
0 & H^*(-{\bf k})\end{array}\right)\Psi({\bf k})$$ where $\Psi^\top=(u_\uparrow,v_\uparrow,u_\downarrow,v_\downarrow)\equiv(\Psi_\uparrow,\Psi_\downarrow)$, with $u$ and $v$ representing two low-energy orbitals. The upper and the lower blocks in the Hamiltonian are related by time-reversal symmetry, and $H({\bf k})$ acting in the orbital space has the form $$\label{eq:upper-ham}
H({\bf k})=\sigma_\mu d_\mu({\bf k}),$$ where $\sigma_\mu$, $\mu=1,2,3$, are the Pauli matrices, $d_{1,2}({\bf k})=A \sin k_{x,y}$, and $d_3=M-2B(2-\cos k_x-\cos k_y)$, the length unit is set by lattice constant $a=1$, and summation over repeated indices is assumed hereafter. We also set $\hbar=c=e=1$ in the following, unless otherwise stated. Since the above Hamiltonian has spectrum $E({\bf k})=\sqrt{d_\mu d_\mu}$ doubly degenerate in spin space, the band gap closes at the $\Gamma$-point (${\bf k}=0$) in the Brillouin zone when the value of model parameters is $M/B=0$. We will only consider the range of parameters $0<M/B<4$ in what follows, so that the spectrum is gapped. This lattice model then describes a topologically nontrivial state with a Kramers’ pair of counterpropagating modes on the edge of the system leading to a quantized spin Hall conductance[@konig] $\sigma_{xy}^{S}=2\frac{e^2}{h}$. For negative values of $M/B$, the model describes a trivial insulator ($\sigma_{xy}^S=0$).
The Dirac-Schr" odinger continuum theory
========================================
By taking the continuum, i.e. large wavelength limit ($|{\bf k}|\ll 1$) of , we arrive at a Dirac Hamiltonian which besides the ordinary Dirac mass term ($M$) contains a Schr" odinger kinetic term ($B$) $$\label{eq:cont-ham}
H_{\rm eff}({\bf k})=i\gamma_0\gamma_i k_i+(M-B{\bf k}^2)\gamma_0,$$ where the four-dimensional $\gamma$-matrices are given by $\gamma_0=\sigma_3\otimes\tau_0$, $\gamma_1=\sigma_2\otimes\tau_3$, and $\gamma_2=-\sigma_1\otimes\tau_0$. Here, Pauli matrices $\{\tau_0,\tau_\mu\}$ act in spin space, with $\sigma_0,\tau_0$ as the $2\times 2$ identity matrices. The $\gamma$-matrices satisfy canonical anticommutation relations $\{\gamma_\alpha,\gamma_\beta\}=2\delta_{\alpha\beta}$, with $\alpha,\beta=0,1,2$. Notice that it is enough to focus on a single spin projection, since the two spin projections are related by the time-reversal operator $T=-i\tau_2K$, with $K$ as the complex conjugation. For convenience, we have divided through Eq. by the energy scale $A$ (the lattice intersite hopping energy) and by the length-scale $a$, before redefining $M/(A a)\rightarrow M$, $B a/A\rightarrow B$, so that the continuum theory Eq. has a dimensionless parameter $MB$ and parameters $B$ and $\sqrt{B/M}$ with dimension of length (we revert to lattice units for comparison to the tight binding model in Section \[sec:tb\]).
Zero-energy states
==================
Let us now consider the effect of the magnetic $\pi$-flux inserted into the system. As usually, the vector potential is coupled to the electronic degrees of freedom through the minimal substitution, ${\bf k}\rightarrow{\bf k}+{\bf A}$, and the Hamiltonian (\[eq:cont-ham\]) for spin up electrons assumes the form $$\label{eq:ham-A}
H_{\rm eff}({\bf k},{\bf A})=\sigma_i (k_i+A_i)+[M-B({\bf k}+{\bf A})^2]\sigma_3.$$ The vector potential $$\label{eq:A}
{\bf A}=\frac{-y{\bf e}_x+x{\bf e}_y}{2r^2}$$ represents the magnetic vortex carrying the flux $\Phi=\pi$. Notice that the spin down electrons are coupled to the $\pi$-flux with the opposite sign because of the time-reversal symmetry, and in that respect $\pi$-flux thus acts on the spin components as a pseudomagnetic vortex in graphene does on the two valley degrees of freedom.[@graphBerry; @graphcones; @fullerene] Of course, the time-reversal invariance of the Hamiltonian (\[eq:ham-A\]) is present only when the flux corresponding to the vector potential ${\bf A}$ is equal to $\pi$ or $0$.
We now show that the above Hamiltonian possesses precisely one bulk zero-energy state with spin up. Time-reversal symmetry then implies the existence of the zero-energy state for electrons with spin down. Expressing the Hamiltonian (\[eq:ham-A\]) in polar coordinates $(r,\varphi)$, taking into account that ${\bf A}=(1/2r){\bf e}_\varphi$, we obtain $$\begin{aligned}
\label{eq:ham-A-explicit}
H_{\rm eff}&=&-i{ e}^{-i\varphi}\left[\partial_r-\frac{i}{r}{\tilde\partial}_\varphi\right]\sigma_ {+}-i{ e}^{i\varphi}\left[\partial_r+\frac{i}{r}{\tilde\partial}_\varphi\right]\sigma_{-}\nonumber\\
&+&\left[M+B\left(\partial_r^2+\frac{1}{r}\partial_r+\frac{1}{r^2}{\tilde\partial}_\varphi^2\right)\right]\sigma_3,\end{aligned}$$ where ${\tilde\partial}_\varphi\equiv\partial_\varphi+i/2$, and $\sigma_\pm\equiv(\sigma_1\pm i\sigma_2)/2$. It is easy to see that in case of an arbitrary flux $\Phi$, the Hamiltonian (\[eq:ham-A\]) also acquires the form (\[eq:ham-A-explicit\]), but with the operator ${\tilde\partial}_\varphi=\partial_\varphi+i\Phi/2\pi$.
In the presence of a vortex carrying a $\pi$-flux, we seek the zero-energy modes of the form $$\label{eq:zero-energy-spinor}
\Psi(r,\varphi)=\left(\begin{array}{cc}e^{i(l-1)\varphi}u_{l-1}(r)\\
e^{il\varphi}v_l(r)\end{array} \right),$$ where $l\in \mathbb{Z}$ is the angular momentum quantum number, and the functions $u,v$ are the solutions of the following equations $$\begin{aligned}
&&\Delta_{l-\frac{1}{2}}u_{l-1}(r)-i\left(\partial_r+\frac{l+\frac{1}{2}}{r}\right)v_l(r)=0\label{eq:zero-1}\\
&&i\left(\partial_r-\frac{l-\frac{1}{2}}{r}\right)u_{l-1}(r)+\Delta_{l+\frac{1}{2}}v_l(r)=0\label{eq:zero-2}.\end{aligned}$$ Here the operator $\Delta_l$ is defined as $$\label{eq:Delta-def}
\Delta_l\equiv M+B\left(\partial_r^2+\frac{1}{r}\partial_r-\frac{l^2}{r^2}\right)\equiv M+B{\cal O}_l.$$ Acting on Eq. (\[eq:zero-1\]) with the operator $\Delta_{l+\frac{1}{2}}$, and using the identity $$[\Delta_l,\partial_r+\frac{l}{r}]=-(2l-1)\frac{B}{r^2}\left(\partial_r+\frac{l}{r}\right),$$ we can eliminate the function $v_l(r)$ from the same equation to obtain $$\left(\Delta_{l+\frac{1}{2}}\Delta_{l-\frac{1}{2}}-{\cal O}_{l-\frac{1}{2}}+\frac{2Bl}{r^2}\Delta_{l-\frac{1}{2}}\right)u_{l-1}(r)=0.$$ After some algebra, the above equation may be rewritten as $$\label{eq:zero-mode-u}
\left[M^2+(2MB-1){\cal O}_{l-\frac{1}{2}}+B^2{\cal O}_{l-\frac{1}{2}}^2\right]u_{l-1}(r)=0.$$ This result may also be obtained by noting that if the spinor in Eq. (\[eq:zero-energy-spinor\]) is an eigenstate with the zero eigenvalue of the Hamiltonian (\[eq:ham-A\]), then it is also an eigenstate with the same eigenvalue of the square of this Hamiltonian. Using Eq. (\[eq:ham-A\]), one then readily obtains $$H_{\rm eff}({\bf k},{\bf A})^2=B^2 ({\tilde{\bf k}}^2)^2+(1-2MB){\tilde{\bf k}}^2+M^2,$$ with ${\tilde{\bf k}}\equiv{\bf k}+{\bf A}$, and the operator ${\tilde{\bf k}}^2$ after acting on the angular part of the upper component of the spinor (\[eq:zero-energy-spinor\]) yields Eq. (\[eq:zero-mode-u\]). Similarly, it may be shown that the function $v_l(r)$ in the spinor given by Eq. (\[eq:zero-energy-spinor\]) obeys an equation of the same form as (\[eq:zero-mode-u\]) with $l\rightarrow l+1$. From Eq. (\[eq:zero-mode-u\]) we conclude that the function $u_{l-1}(r)$ is an eigenfunction of the operator ${\cal O}_{l-1/2}$ with a [*positive*]{} eigenvalue $$\label{eq:u}
{\cal O}_{l-\frac{1}{2}}u_{l-1}(r)=\lambda^2 u_{l-1}(r),$$ since the operator ${\tilde{\bf k}}^2$ when acting on a function with the angular momentum $l$ is equal to $-{\cal O}_{l+1/2}$, and the eigenstates of the operator ${\tilde{\bf k}}^2$ with a [*negative*]{} eigenvalue are localized. Eqs. (\[eq:zero-mode-u\]) and (\[eq:u\]) then imply $$\label{eq:lambda-square}
\lambda_\pm=\frac{1\pm\sqrt{1-4MB}}{2B},$$ and the function $u_l(r)\sim I_{l-\frac{1}{2}}(\lambda r)$ with $I_l(x)$ as the modified Bessel function of the first kind. However, from the above solutions the only square-integrable ones are in the zero angular-momentum channel since $I_l(x)\sim x^{-|l|}$ as $x\rightarrow0$. Furthermore, for $l=0$ only the linear combination $I_{1/2}(x)-I_{-1/2}(x)\sim x^{-1/2}e^{-x}$ has the asymptotic behavior at infinity consistent with a finite norm of the state. In the above equation we should distinguish two regimes of parameters, $0<MB<1/4$ and $MB>1/4$, for which the argument of the square-root is positive and negative, respectively.
For $0<MB<1/4$, since the argument of the square-root in the above equation is positive, we obtain two zero-energy solutions $$\label{eq:zero-energy-states1}
\Psi_\pm({\bf r})=\frac{e^{-\lambda_\pm r}}{\sqrt{2\pi\lambda_\pm^{-1} r}}\left(\begin{array}{cc}e^{-i\varphi}\\ i\end{array}\right),$$ and, of course, $\lambda_\pm>0$ because of the square-integrability. On the other hand, when $MB>1/4$, up to a normalization constant, the solutions have the form $$\begin{aligned}
\label{eq:zero-energy-states2}
\Psi_1({\bf r})&=&\frac{e^{-r\sqrt{\frac{M}{B}}\cos\theta}\cos\left(r\sqrt{\frac{M}{B}}\cos\theta\right)}{\sqrt{r}}\left(\begin{array}{cc}e^{-i\varphi}\\ i\end{array}\right),\nonumber\\
\Psi_2({\bf r})&=&\frac{e^{-r\sqrt{\frac{M}{B}}\cos\theta}\sin\left(r\sqrt{\frac{M}{B}}\cos\theta\right)}{\sqrt{r}}\left(\begin{array}{cc}e^{-i\varphi}\\ i\end{array}\right),\end{aligned}$$ where $$\theta=\frac{1}{2}\arctan\frac{\sqrt{|1-4MB|}}{1-2MB}.$$ However, since the identity $$\label{eq:13}
\sqrt{x}\cos{\left(\frac{1}{2}\arctan\frac{\sqrt{|1-4x|}}{1-2x}\right)}=\frac{1}{2}$$ holds for $1/4<x<4$, the localization length of the zero-modes for $MB>1/4$ is actually independent of $M$, namely Eqs. become $$\begin{aligned}
\label{eq:zero-energy-states2final}
\Psi_1({\bf r})&=&\frac{e^{-\frac{r}{2B}}\cos\left(\frac{r}{2B}\right)}{\sqrt{r}}\left(\begin{array}{cc}e^{-i\varphi}\\ i\end{array}\right),\nonumber\\
\Psi_2({\bf r})&=&\frac{e^{-\frac{r}{2B}}\sin\left(\frac{r}{2B}\right)}{\sqrt{r}}\left(\begin{array}{cc}e^{-i\varphi}\\ i\end{array}\right).\end{aligned}$$ Therefore, we can conclude that the Hamiltonian (\[eq:ham-A\]) possesses zero-energy modes in the entire range of parameters $M$ and $B$ for which the system is in the topologically non-trivially phase, $0<M/B<4$. In particular, as it can be seen from Eq. (\[eq:lambda-square\]), when $4MB<1$ zero-energy states are purely exponentially localized, while for $4MB>1$ the exponentially localized solutions have an oscillatory part with a characteristic length-scale exactly equal to the localization length. Notice also that in the regime when $0<MB<1/4$, there are two characteristic length scales associated with the midgap modes, $\xi_\pm\sim\lambda_\pm^{-1}$. Of course, after a short-distance regularization is imposed, only a linear combination of the two states survives. The physical interpretation of the two length scales depends on the form of the superposition of the state after the regularization has been imposed, as it may be easily seen from the form of the states . In the regime $MB>1/4$, the zero-energy states are characterized by a single length-scale $\xi_{\rm loc}\sim 2B$, which is at the same time the localization length and characterizes the oscillations of the exponentially decaying state. Therefore, the appearance of the zero-energy states bound to a $\pi$-flux vortex is a generic feature of the Hamiltonian (\[eq:cont-ham\]) describing the quantum spin Hall system. Furthermore, in the vortex-free system, it may be shown by imposing open boundary conditions on the wave-function at one of the edges of the system, for instance the one perpendicular to the $x$-axis, and at infinity, $\Psi(x,y=0)=\Psi(x,y=\infty)=0$, that the Hamiltonian gives rise to gapless edge modes with the penetration depth given by exactly the same expression as the localization length for the zero-energy modes bound to the $\pi$-flux vortex. The bulk-boundary correspondence may be thus probed by inserting a $\pi$-flux vortex in the quantum spin Hall system.
Thin solenoid regularization of vortex
======================================
The zero-energy modes, given by Eqs. (\[eq:zero-energy-states1\]) and (\[eq:zero-energy-states2final\]), form an overcomplete basis in the zero angular-momentum channel, because the Hamiltonian (\[eq:ham-A\]) is not self-adjoint, which is due to the singularity of the vortex vector potential (\[eq:A\]) at the origin. Thus the gauge potential has to be regularized.
A possible regularization is provided by considering the vortex with the flux concentrated in a thin annulus of a radius $R$. Let us first consider the Hamiltonian in the range of parameters $0<MB<1/4$. The zero-energy state of the Hamiltonian outside the annulus is then a linear combination of the modes $\Psi_\pm$ given by Eq. (\[eq:zero-energy-states1\]). Inside the annulus the vector potential ${\bf A}=0$, and the zero-energy modes are $$\label{eq:vortex-free}
\Psi_{<}({\bf r})=C_1\left(\begin{array}{cc}e^{-i\varphi}I_1(\lambda_+r)\\ iI_0(\lambda_+r)\end{array}\right)+C_2\left(\begin{array}{cc}e^{-i\varphi}I_1(\lambda_{-}r)\\ iI_0(\lambda_-r)\end{array}\right),$$ with $\lambda_\pm$ given by Eq. (\[eq:lambda-square\]), and $C_{1,2}$ being complex constants. By matching these solutions at $r=R$, and taking $R\rightarrow0$, we obtain, up to a normalization constant, the zero-energy state of the form $$\label{eq:zero-energy-annulus}
\Psi({\bf r})=\frac{e^{-\lambda_+r}-e^{-\lambda_-r}}{\sqrt{r}}\left(\begin{array}{cc}e^{-i\varphi}\\ i\end{array}\right).$$ Notice that this zero-energy state is regular at the origin which is a consequence of the regularity at the origin of the solutions (\[eq:vortex-free\]) of the vortex-free problem. Similarly, one may show that when $MB>1/4$ the zero-energy mode is given by the spinor $\Psi_2$ in Eq. (\[eq:zero-energy-states2\]) also regular at the origin and behaving $\sim r^{1/2}$ when $r\rightarrow0$.
Self-adjoint extension of the Hamiltonian {#sae}
=========================================
Although the above regularization results in concrete solutions to the problem, we should consider the self-adjoint extension of the corresponding Hamiltonian in a more general manner by specifying the proper Hilbert space. That way, a family of Hermitian Hamiltonians is obtained, depending on free physical parameters that determine the scattering at the vortex core and the detailed profile of the single zero-mode (per spin). In the last section, we will comment on the regularization provided by the tight-binding version of the model, Eq. .
The application of the standard theory of self-adjoint extensions (SAE) [@Weidmann; @Thaller; @Fulop:2007p3442; @Jackiw:1991p1161] prescribes that we need to ensure that the massive Dirac Hamiltonian (a differential operator) becomes Hermitian (self-adjoint) only after choosing the proper Hilbert space (i.e. domain of functions) on which it is allowed to act. Instead of analyzing the imaginary spectrum (which needs to be removed), we implement von Neumann’s construction by looking directly at the conditions under which the Hamiltonian is Hermitian when acting on arbitrary functions that are square integrable (but might diverge at the origin, due to diverging potential there). This will effectively determine the coefficients of the linear combination $C_1\Psi_++C_2\psi_-$ in angular momentum channel $l=0$ and thereby fix the zero-mode.
Using symmetry, we start from the radial part of the operator, $H^l(r)$, which acts in the subspace of angular momentum $l$ spanned by functions of the form $$\label{eq:4}
\psi_l(r)\equiv e^{i l\varphi}{\left(\begin{matrix}e^{-i\varphi} u_l\\#2\end{matrix}\right)},$$ completely determined by ${\left(\begin{smallmatrix}u_l\\#2\end{smallmatrix}\right)}$. Recall that the zero-energy states of the $M-B$ model in presence of a $\pi$-flux vortex come in the form of Kramers pairs $${\Psi}_{\uparrow}(x,y)=
\begin{pmatrix}
{\psi}(r,\varphi)\\
{0}
\end{pmatrix},
\qquad
{\Psi}_{\downarrow}(r,\varphi)=
\begin{pmatrix}
{0}\\
{\psi}(r,\varphi)^*
\end{pmatrix}$$ where ${\psi}(r,\varphi)$ is exactly of the form in Eq. .
We also implement the standard change of scalar product in $r$-space by rescaling $\psi_l(r)=\frac{1}{\sqrt{r}}\tilde{\psi}_l(r)$, $\partial\psi_l(r)=\frac{1}{\sqrt{r}}\tilde{\partial}\tilde{\psi}_l(r)$ ($\partial$ always denotes $d/dr$), where $\tilde{\partial}\equiv\partial-\frac{1}{2r}$, after which $H^l$ takes the form $$\label{eq:5} \tilde{H}^l(r)={\left(\begin{smallmatrix}M+B\left(\partial^2-\frac{l(l-1)}{r^2}\right) & -i\left(\partial+\frac{l}{r}\right)\\#3 & -M-B\left(\partial^2-\frac{l(l+1)}{r^2}\right)\end{smallmatrix}\right)}.$$ We have reverted to the standard derivative ($\partial$) here. Since the $\pi$-flux enters through $l\rightarrow l_{eff}$, we omit it here. The non-derivative terms will not play any role in the following analysis since the standard ’centrifugal force’ provided by $l\neq 0$ does not lead to singularities. However, the gauge potential will have the chance to provide us with the boundary condition exactly when $l=0$.
Now, for two arbitrary wavefunctions $\phi$, $\psi$ which are determined by $\tilde{F}\equiv{\left(\begin{smallmatrix}f\\#2\end{smallmatrix}\right)}$, $\tilde{U}\equiv{\left(\begin{smallmatrix}u\\#2\end{smallmatrix}\right)}$ (we dropped index $l$), respectively, the condition of hermiticity of $\tilde{H}^l(r)$ becomes (note the change in $r\textrm{d}r$): $$\begin{aligned}
\label{eq:6}
{\left\langle\tilde{\phi}\right|}\tilde{H}-\tilde{H}^\dagger{\left|\tilde{\psi}\right\rangle}&= \int\!\textrm{d}r\; \tilde{\phi}(r)^*\tilde{H}^l(r)\tilde{\psi}(r)- \left(\int\!\textrm{d}r\; \tilde{\psi}(r)^*\tilde{H}^l(r) \tilde{\phi}(r)\right)^*=\notag\\
&=B\left.\left\{\tilde{f}^*{\partial_r}\tilde{u}-{\partial_r}\tilde{f}^*\tilde{u}-\tilde{g}^*{\partial_r}\tilde{v}+{\partial_r}\tilde{g}^*\tilde{v}\right\}\right|_0^\infty-\left.i\left\{\tilde{f}^*\tilde{v}+\tilde{g}^*\tilde{u}\right\}\right|_0^\infty=\notag\\
&=B\left[\tilde{F}^*\sigma_3{\partial_r}\tilde{U}-{\partial_r}\tilde{F}^*\sigma_3 \tilde{U}-i \tilde{F}^*\sigma_1 \tilde{U}\right](0)\notag\\
&\equiv 0,\end{aligned}$$ where the $\sigma$ Pauli matrices act on the two component functions, which vanish at infinity, and are evaluated at the origin (point $r=0$) in the next-to-last line.
There is a continuous family of restrictions on the behavior of square-integrable functions at the origin, such that is satisfied, leading to the Hamiltonian which is Hermitian on such a chosen domain. The proper parametrization of the most general restriction on the allowed domains is achieved by using the linearity of . Namely, we define two linear operators $\Gamma_1$, $\Gamma_2$ which map arbitrary functions, i.e. the domain of $\tilde{H}^\dagger$, onto their value at the boundary, i.e. the space of complex two component vectors: $$\label{eq:15}
\Gamma_i:\tilde{\psi}(r)\rightarrow{\left(\begin{smallmatrix}\tilde{u}(0)\\#2\end{smallmatrix}\right)}.$$ These operators are defined by : $$\label{eq:7}
B\left[\tilde{F}^*\sigma_3{\partial_r}\tilde{U}-{\partial_r}\tilde{F}^*\sigma_3 \tilde{U}-i \tilde{F}^*\sigma_1 \tilde{U}\right](0)\equiv \langle\Gamma_2\tilde{F},\Gamma_1\tilde{U}\rangle-\langle\Gamma_1\tilde{F},\Gamma_2\tilde{U}\rangle.$$ Notice that this form can always be achieved due to the original form of the subtraction between $\tilde{H}$ and $\tilde{H}^\dagger$. We can choose in particular, without loss of generality, $$\begin{aligned}
\label{eq:8}
\Gamma_1 \tilde{U}&=B\sigma_3{\partial_r}\tilde{U}(0)-i\frac{\sigma_1}{2}\tilde{U}(0),\\
\Gamma_2 \tilde{U}&=\tilde{U}(0).\end{aligned}$$ Any vector in the boundary space, i.e. ${\left(\begin{smallmatrix}\chi_1\\#2\end{smallmatrix}\right)}\in\mathbb{C}^2$, is an image by $\Gamma_i$ of some wavefunction, i.e. of some ${\left(\begin{smallmatrix}\tilde{u}\\#2\end{smallmatrix}\right)}$. Since $\tilde{U}(0)={\left(\begin{smallmatrix}\tilde{u}(0)\\#2\end{smallmatrix}\right)}$ and ${\partial_r}\tilde{U}(0)={\left(\begin{smallmatrix}{\partial_r}\tilde{u}(0)\\#2\end{smallmatrix}\right)}$ take on arbitrary values, this means that the boundary space indeed is $\mathcal{H}_b=\mathbb{C}^2$.
The most general relation that has to be satisfied by a wavefunction such that will hold is now parametrized by unitary mappings $S$ in $\mathcal{H}_b$: $$\label{eq:9}
\mathcal{D}(\tilde{H}_U)=\{\psi|(S-\sigma_0)\Gamma_1\psi+i(S+\sigma_0)\Gamma_2\psi=0\},$$ with $\sigma_0$ the 2x2 identity matrix, and $\mathcal{D}$ denoting the domain of operator. One can directly understand from that forcing arbitrary linear combinations of a general $\tilde{U}(0)$ and ${\partial_r}\tilde{U}(0)$ to zero will still preserve the condition , due to the linearity and the antisymmetric nature of the form of this expression. Eq. is giving us a precise recipe and parametrization of the fact that this is the most general restriction that needs to be made on the wavefunctions $\tilde{U}(r)$, i.e. on the $\psi(r)$.
We now proceed to use the form of $\Gamma_i$ to explore the allowed boundary conditions on the wavefunctions, in particular determining whether there is a self-adjoint extension $\tilde{H}_U$ with the previously found zero energy states in its domain. For concreteness we focus on the case $MB<1/4$.
Since $\mathcal{H}_b$ is $\mathbb{C}^2$, our mappings $S\in U(2)$, in contrast to the same problem in the case of an ordinary massive Dirac Hamiltonian necessitating a $U(1)$ parametrization [@2dDiracSAE]. The mappings $S\in U(2)$ can be parametrized by $$\label{eq:1}
S=\frac{1}{d}\sum_\mu m_\mu\sigma_\mu,\quad m_\mu\in\mathbb{R},\quad\sum_\mu m_\mu^2=1,\quad d\equiv {\left(\begin{matrix}e^{i\eta} & 0\\#3 & 1\end{matrix}\right)},\quad \eta\in[0,2\pi),$$ with the quaternion basis $\sigma_\mu=(\sigma_0,i \vec{\sigma})$. We will use the notation $[m_0,m_1,m_2,m_3]$ to represent the quaternion $\sum_\mu m_\mu\sigma_\mu$, while to label the boundary states of the zero energy wavefunctions $\psi_0\equiv C_1\Psi_++C_2\psi_-$ we will use $$\begin{aligned}
\label{eq:16}
\phi &\equiv\tilde{\psi_0}(0)\equiv\left.\sqrt{r}\psi_0\right|_{r=0}={\left(\begin{matrix}C_1+C_2\\#2\end{matrix}\right)},\\ \phi' &\equiv{\partial_r}\tilde{\psi_0}(0)\equiv\left.{\partial_r}(\sqrt{r}\psi_0)\right|_{r=0}=-{\left(\begin{matrix}C_1\lambda_1+C_2\lambda_2\\#2\end{matrix}\right)}.\end{aligned}$$ Due to normalization, both $\phi$ and $\phi'$ depend only on the vector ${\left(\begin{smallmatrix}1\\#2\end{smallmatrix}\right)}$, where $$\label{eq:17}
x\equiv\frac{C_1}{C_2}.$$ It turns out that the case with $\eta=0$ is special, and so we examine it first in detail. Only this subclass contains the extension with regular zero-modes.
Extensions described by $SU(2)$, $\eta=0$
-----------------------------------------
Eqs. , lead to the following condition on two quaternions $P,Q$: $$\begin{aligned}
\label{eq:2}
B\cdot P\phi'&=Q\phi,\\
Q&=[\frac{m_1}{2}+i(1+m_0),\frac{1-m_0}{2}+i m_1,\frac{m_3}{2}+i m_2,-\frac{m_2}{2}+i m_3], \notag\\
P&=[-i m_3, -i m_2, i m_1, i(m_0-1)].\end{aligned}$$ Once the values of $M,B$ (and therefore also $\lambda_{1/2}\equiv\lambda_\pm$) are given, this equation determines $x$ as function of the particular SAE $m_\mu$ (if a solution for $x$ exists), and $x$ then determines the specific linear combination of $\lambda_{1/2}$ decaying functions in the zero mode, influencing also its regularity properties at the origin. Since $\mathrm{det}(P)=2(m_0-1)$, we first consider:
### Extension with $\eta=0$, $m_0=1$ {#saeregular}
According to definition this immediately implies $\vec{m}=0$, and $$\begin{aligned}
\label{eq:18}
\sigma_0\phi=0&\Longleftrightarrow \sqrt{r}\psi_0|_{r=0}=0,\text{ i.e.}\\
C_1&=-C_2.\end{aligned}$$ Such a wavefunction is regular at the origin and localized on the scale $1/\textrm{min}(\lambda_i)$, no matter the values $M,B$ (or $\lambda_{1,2}$). We therefore see that such a wavefunction is allowed when the physical SAE is given by $S=[1,0,0,0]$, $\eta=0$. This will turn out to be the only extension allowing $x=-1$, see Fig. \[fig:1\], essentially because it is the only extension for which the matrix on the left-hand side of Eq. (\[eq:2\]) vanishes.
### Extension with $\eta=0$, $m_0<1$ {#extension-with-eta0-m_01}
We must now consider Eq. as a vector equation, treating $C_{1,2}$ as unknown variables. We get $$\begin{aligned}
\label{eq:3}
\hat{Y}{\left(\begin{matrix}C_1\\#2\end{matrix}\right)}&=0\\
\hat{Y}&={\left(\begin{matrix}Ba\lambda_1+p & Ba\lambda_2+p\\#3 & Bb\lambda_2+q\end{matrix}\right)}\\
\textrm{det}(\hat{Y})&=B(\lambda_1-\lambda_2)(aq-bp),\end{aligned}$$ where $a\equiv P_{11}+i P_{12},\ b\equiv P_{21}+i P_{22},\ p\equiv Q_{11}+i Q_{12},\ q\equiv Q_{21}+i Q_{22}$. There are solutions only when $$\label{eq:19}
aq=bp\Leftrightarrow m_2=0.$$ So in the case of $m_0\neq 1$, the zero energy mode is allowed in principle only for Hamiltonians with $m_2=0$. The value of $x$ in such a system is given by $$\label{eq:10}
x=-\frac{g(\lambda_2)}{g(\lambda_1)}, \quad g(\lambda)\equiv B a\lambda+p=B b\lambda+q,$$ where $$\label{eq:11}
g(\lambda)=B(1-m_0-m_1-i m_3)\lambda+\left(\frac{1}{2}-i\right)(m_1-1-i m_3)+\left(\frac{1}{2}+i\right)m_0$$ is a linear function of $\lambda$. From this expression it seems that another way for $x=-1$ (and the zero mode becoming regular at the origin for all model parameters $M,B$), is if $m_1=1$, where we assume $\lambda_1\neq\lambda_2$. However, this is not the case, because when $m_1=1$ only one component of the vector equation Eq. (\[eq:3\]) can be satisfied, and not both component equations at the same time. Note that the extensions identified here by having $m_2=0$ include the case from previous subsection, $m_0=1$. In fact, demanding that $x=-1$ directly from Eq. (\[eq:3\]) implies that the columns of $Y$ must be the same, which reduces to the demand $a=b=0\Rightarrow P=[0,0,0,0]$ since $\lambda_{1,2}$ are non-degenerate. It is easy to see, using the definition Eq. (\[eq:1\]), that this is only possible in the above considered case $m_0=1$.
Dependence on $MB$
------------------
To analyze possible values for $x$ for given model parameters $M,B$, we parametrize the SAE using hyperspherical coordinates $$\label{eq:14}
m_\mu=(\cos{\psi},\sin{\psi}\cos{\theta},\sin{\psi}\sin{\theta}\cos{\varphi},\sin{\psi}\sin{\theta}\sin{\varphi}),$$ where the hypersphere $SU(2)\simeq S^3$ is properly covered by $\psi\in[0,\pi]$, $\theta\in[0,\pi]$, and $\varphi\in[0,2\pi)$. The condition $m_2=0$, Eq. (\[eq:19\]), means $\varphi=\pi/2$, and also $\psi\neq 0$ ($m_0\neq 1$), $(\psi,\theta)\neq(\frac{\pi}{2},0)$ ($m_1\neq 1$) in this subsection. This parametrization can be plugged into the function $g(\lambda)$ from Eq. together with the values of $\lambda_{1/2}=\frac{1\pm\sqrt{1-4k}}{2}$, where $k\equiv MB$, so that one can deduce $x$ from the model parameters $M,B$ and the boundary condition determined by $\psi,\theta$. Analysis shows that $x(\psi,\theta)$ weakly depends on the value of $k\equiv MB$, and so Fig. \[fig:1\] presents the typical behavior for fixed $k$. On the $(\psi,\theta)$ plane, the function $x(\psi,\theta)$ is strongly localized and a sharply peaked dipole. One can use the real and imaginary parts of $x$ to identify structure in the scattering phase shift coming from the regularized vortex.
From $SU(2)$ to $U(2)$ extensions, $\eta\neq 0$
-----------------------------------------------
The introduction of $\eta\neq 0$ in Eq. (\[eq:1\]) does not change the analysis method of previous subsections.
To start, the expressions for quaternions $P,Q$ in Eq. (\[eq:2\]) are slightly more complicated, so that demanding $x=-1$ through the vanishing of left-hand side of Eq. (\[eq:2\]), as in Eq. (\[eq:18\]), becomes $$\begin{aligned}
\label{eq:20}
\mathrm{det}(P)&\equiv 0,\text{ i.e.}\\\notag
\exp(i\eta)&=\frac{m_3+i(1-m_0)}{m_3-i(1-m_0)}.\end{aligned}$$ However, even when the determinant vanishes, the matrix $P$ can not become zero matrix, unless $\cos(\eta)=1\Rightarrow\eta=0$, which actually just takes us back to a SAE considered in previous subsections.
Next, considering Eqs. (\[eq:2\]) by treating $C_1,C_2$ as unknowns, as in Eq. (\[eq:3\]), implies that zero-modes can exist only for a SAE that has $$\begin{aligned}
\label{eq:21}
\mathrm{det}(Y)&=0\Longrightarrow\\
\exp(i\eta)&=\frac{1+m_1-im_2}{1+m_1+im_2}\Longleftrightarrow\\
\tan\left(\frac{\eta+n\pi}{2}\right)&=\frac{\sin(\psi) \sin(\theta) \cos(\phi)}{1+\sin(\psi) \cos(\theta)},\end{aligned}$$ where in the last line we used the parametrization from Eq. (\[eq:14\]).
We emphasize that here, as for $\eta=0$ case, demanding a zero-mode with $x=-1$ implies from Eq. (\[eq:3\]) that the columns of $Y$ have to be the same. Since $\lambda_{1,2}$ are not degenerate for physical values of $MB$, this gives $a=b=0$, i.e. the already considered demand that the matrix $P$ vanishes, which is impossible when $\eta\neq 0$.
Following the derivation of Eq. (\[eq:10\]), the $x$ in general is again given by a ratio $$\label{eq:22}
x=-\frac{\tilde{g}(m_\mu,\eta,\lambda_2)}{\tilde{g}(m_\mu,\eta,\lambda_1)},$$ where we omit the lengthy explicit expression for $\tilde{g}$. Notice that even when this ratio approaches a singular limit by $\tilde{g}\rightarrow 0\text{ or }\infty$, the previous paragraph ensures that $x$ will not reach the value $-1$.
To summarize, the SAE analysis of the $M-B$ model shows that a $U(2)$ parametrization which completes this model allows the existence of zero modesfor a subclass of Hamiltonians described by three free angle parameters, e.g. $\eta,\theta,\psi$ in Eq. . These determine a matrix relating the value of the wavefunction spinor to its derivative, both taken at the origin, according to Eq. . There is only a single SAE, Section \[saeregular\], determined by $\eta=\psi=0$, which allows for the existence of a zero energy state which is *both* localized and regular (vanishing) at the origin.
![ \[fig:1\] Vortex singularity regularization. The value of $x=C_1/C_2$, where the zero-mode wavefunction $\Psi_0\sim r^{-1/2}(x e^{-\lambda_1 r}+e^{-\lambda_2 r})$, is presented as a function of the SAE parametrized by two angles $(\psi,\theta)$, and $\eta=0$ (see Section \[sae\]). The wavefunction becomes regular at the origin for $x=-1$. Notice that Re$(x)$ approaches $-1$ for $\psi=0$, where also Im$(x)=0$. The plots are evaluated for $MB=0.15$, but are representative for all topologically non-trivial model parameters $0<MB<1/4$.](MesarosFig1.png){width="100.00000%"}
Comparison to the tight-binding $M-B$ model {#sec:tb}
===========================================
![ \[fig:2\] a) Comparison of the vortex-bound zero-mode localization lengths $\xi_\pm$ (black lines) predicted by the present $M-B$ model (see text after Eq. ), and measured in the tight-binding version of the model on a 31x31 lattice. Inset shows in detail the excellent agreement in the $MB<1/4$ regime. The lattice model has a phase transition at $M_{latt}/B_{latt}=4$ which is absent in the continuum model, since the gap closes at finite momentum. b) The complex constant $x=C_1/C_2$ determines the form of the zero-mode wavefunction. When $x=-1$ the mode is regular at the $\pi$-flux, and this situation is realized throughout the regime $MB<1/4$ (dots in the legend mark the values of $MB$ of the plotted points).](MesarosFig2.pdf){width="100.00000%"}
In this Section we present the results from a tight-binding $M-B$ model, which in momentum space has the limit Eqs. , . We numerically study this model on a 31x31 sized square lattice with periodic boundary conditions and a $\pi$-flux—anti-$\pi$-flux pair positioned at maximal distance. The details of this tight-binding model written in real space, i.e. on the lattice, are presented in Ref. [@TopProbes].
We first isolate the zero energy mode localized on a single $\pi$-flux, using the nearly degenerate inversion-symmetric and anti-symmetric zero modes on the finite lattice. The lattice symmetries guarantee the $(\exp{(-i\phi)},i)^T$ form of the wavefunction spinor, with $\phi$ the polar angle, just as in the continuum. By fitting the radial envelope of the spinor using Eq. (\[eq:zero-energy-states1\]) and the angle-averaged wavefunction, we extract the two localization lengths $\xi^\pm_{latt}$ in units of the lattice constant. The results are shown in Fig. \[fig:2\](a). The agreement with the present continuum model prediction, see text after Eq. and Eq. , is excellent in the $MB<1/4$ regime (inset of Fig. \[fig:2\](a)). For $MB>1/4$ the oscillatory part of the wavefunction (see Eq. ) makes the numerical fitting less reliable, and the agreement is only qualitative.
One should note that the lattice tight-binding model has a natural length-scale, the lattice constant $a$, beside its dimensionless constants $M_{latt},B_{latt}$ (in units of inter-site hopping energy $A$), see Eq. . In the present continuum model, Eq. , the parameters $M,B$ determine two length-scales, $B$ and $B^{1/2}M^{-1/2}$, where the former acts as a length-scale in the zero-modes. The lattice constant is simply absorbed by $M_{latt},B_{latt}$ to give $M,B$ of the continuum model (see Eq. ), and that is why the agreement in Fig. \[fig:2\](a) is quantitatively precise. (In the lattice calculation we actually set $B_{latt}\equiv 1$ and vary only $M_{latt}$, driving therefore both the parameter $MB=M_{latt}B_{latt}$ and the lattice topological transition parameter $M_{latt}/B_{latt}$.) To determine the SAE which is realized in the lattice model, we calculate the complex constant $x=C_1/C_2$, which was defined in Eq. (\[eq:17\]), and represents the ratio of contributions of two singular functions in the zero-mode wavefunction, Eqs. (\[eq:zero-energy-states1\]), (\[eq:zero-energy-states2final\]). Fig. \[fig:2\](b) shows that throughout the regime $MB<1/4$, $x$ keeps near the value $x=-1$, which is the special case of zero-mode being regular at the $\pi$-flux position. As shown above, only a single SAE allows this, and this SAE is also realized in a thin-solenoid regularization of the continuum model.
Quantum numbers of the zero-energy modes
========================================
We will now show that these zero-energy modes carry non-trivial charge or spin quantum number depending on their occupation. For that purpose, we will write the continuum $4\times 4$ Hamiltonian (\[eq:cont-ham\]) coupled to a $U(1)$ vector potential as $$H=i\gamma_0\gamma_i(k_i+A_i)+(M-B({\bf k}+{\bf A})^2)\gamma_0$$ with the vector potential ${\bf A}$ given by Eq. (\[eq:A\]). Note that the unitary matrices $\gamma_3=\sigma_2\otimes\tau_2$ and $\gamma_5=\sigma_2\otimes\tau_1$ anticommute with the gamma-matrices $\gamma_\alpha$, $\alpha=0,1,2$. Therefore, the Hamiltonian anticommutes with the matrices $\Gamma_3\equiv i\gamma_0\gamma_3$ and $\Gamma_5\equiv i\gamma_0\gamma_5$ which then generate chiral (spectral) symmetry relating states with positive and negative energies, i.e., if $H|E\rangle=E|E\rangle$, then, for instance, $\Gamma_5|E\rangle=|-E\rangle$, and the matrix $\Gamma_5$ reduces in the zero-energy subspace of the Hamiltonian. These two properties then imply that in the ground-state the expectation value of a traceless Hermitian operator $Q$ is given in terms of the zero-energy states of the Hamiltonian $$\langle Q\rangle=\frac{1}{2}\left(\sum_{{\rm occupied}}\Psi^\dagger Q \Psi-\sum_{{\rm unoccupied}}\Psi^\dagger Q \Psi\right).$$ Therefore, in the case of a quantum spin Hall insulator threaded by a $\pi$-flux vortex, depending on the occupation of a pair of zero-energy states, there are four possibilities for the ground state quantum numbers. Namely, when both states are occupied or empty, according to the above expression, the charge is $+e$ or $-e$ and the spin quantum number is zero. On the other hand, when one of the states is occupied, the spin quantum number is $+1/2$ or $-1/2$, while the charge is zero. In that way, the spin-charge separation, characteristic for one-dimensional systems[@SSH], appears also in a two-dimensional system [@hou-chamon], and is tied to a topologically non-trivial nature of the quantum spin Hall state [@dung-hai1; @qi-zhang]. Similarly, zero-energy modes bound to the vortex core in an antiferromagnetic state on a honeycomb lattice lead to the phenomenon of spin fragmentation [@igor-spin-fragmentation]; see Ref. [@igor-Dirac-isospin] for a general discussion of this class of problems in the context of Dirac systems.
Finally, let us note that the topological stability of $\pi$-flux zero-modes, i.e. stability under smooth deformations of the vector potential, has been strictly proved for both Dirac and Schr" odinger Hamiltonians [@aharonov-casher; @jackiw-prd84], but for the $M-B$ model, which is a sum of both, we are not aware of an analogous proof based on an index theorem.
Conclusions
===========
In conclusion, we have shown that a pair of zero-modes bound to a magnetic $\pi$-flux are a generic feature of the $M-B$ model in the topologically non-trivial phase. The continuum Hamiltonian of that model has both Dirac and Schr" odinger kinetic terms and a mass, and has not been studied in detail previously. We have analytically found the zero-modes in the presence of the $\pi$-flux vortex in the entire range of parameters describing a topologically non-trivial phase with the bandgap opening at the zero momentum. These modes are exponentially localized around the vortex core, and a particular regularization of the vector potential corresponding to the vortex yields the modes regular at the origin, but in general, as we have shown, the form of the solution depends on the short-distance regularization of the vortex. Vortex zero modes obtained within a lattice tight-binding model match the ones found for a particular self-adjoint extension of the continuum Hamiltonian. Finally, we discussed a realization of the two-dimensional spin-charge separation through the vortex zero-modes.
Acknowledgements {#acknowledgements .unnumbered}
================
The authors gladly thank Oskar Vafek and Igor Herbut for useful discussions. This work was supported by Dutch Foundation for Fundamental Research on Matter (FOM). V. J. acknowledges the support of the Netherlands Organization for Scientific Research (NWO).
[00]{} X.-L. Qi and S.-C. Zhang, Rev. Mod. Phys. [**83**]{}, 1057 (2011). M. Z. Hasan and C. L. Kane, Rev. Mod. Phys. [**82**]{}, 3045 (2010). X.-L. Qi, T. L. Taylor and S.-C. Zhang, Phys. Rev. B [**78**]{}, 195424 (2008). L. Fu and C. L. Kane, Phys. Rev. B [**74**]{}, 195312 (2006). G. Y. Cho and J. E. Moore, Ann. of Phys. [**326**]{}, 1515 (2011). S. Ryu, J. E. Moore and A. W. W. Ludwig, Phys. Rev. B [**85**]{}, 045104 (2012). M. Stone, arXiv:1201.4095 (2012). G. Rosenberg and M. Franz, Phys. Rev. B [**82**]{}, 035105 (2010). J. Maciejko, X.-L. Qi, A. Karch and S.-C. Zhang, Phys. Rev. Lett. [**105**]{}, 246809 (2010). C. Hoyos, K. Jensen and A. Karch, Phys. Rev. D [**82**]{}, 086001 (2010). V. Juričić, A. Mesaros, R. J. Slager and J. Zaanen, Phys. Rev. Lett. [**108**]{}, 106403 (2012). Y. Ran, A. Vishwanath, and D.-H. Lee, Phys. Rev. Lett. [**101**]{}, 086801 (2008). X.-L. Qi and S.-C. Zhang, Phys. Rev. Lett. [**101**]{}, 086802 (2008). B. A. Bernevig, T. A. Hughes, and S. C. Zhang, Science [**314**]{}, 1757 (2006). M. K" onig [*et al.*]{}, J. Phys. Soc. Jpn. [**77**]{}, 031007 (2008). Y. Aharonov and A. Casher, Phys. Rev. A [**19**]{}, 2461 (1979). R. Jackiw, Phys. Rev. D [**29**]{}, 2375 (1984). M. Ansourian, Phys. Lett. [**70B**]{}, 301 (1977). R. Jackiw and C. Rebbi, Phys. Rev. D [**13**]{}, 3398 (1976). R. Jackiw and P. Rossi, Nucl. Phys. B [**190**]{}, 681 (1981). E. J. Weinberg, Phys. Rev. D [**24**]{}, 2669 (1981). W. P. Su, J. R. Schrieffer, and A. J. Heeger, Phys. Rev. Lett. [**42**]{}, 1698 (1979). Y. Nishida, L. Santos, and C. Chamon, Phys. Rev. B [**82**]{}, 144513 (2010). M. Persson, [Letters in Mathematical Physics]{} [**78**]{}, 139 (2006). A. Melikyan and Z. Tešanovi' c, Phys. Rev. B [**76**]{}, 094509 (2007). T. F[ü]{}l[ö]{}p, [Symmetry, Integrability and Geometry: Methods and Applications]{} [**3**]{}, 107 (2007). V. R. Khalilov and K. E. Lee, J. Phys. A: Math. Theor. [**44**]{}, 205303 (2011). D. K. Park and J. G. Oh, Phys. Rev. D [**50**]{}, 7715 (1994). R. Jackiw, [B[é]{}g Memorial Volume]{}, p. 1 (1991). J. Weidmann, “[Spectral Theory of Ordinary Differential Operators]{}”, Springer-Verlag (1987). Ph. de Sousa Gerbert, Phys. Rev. D [**40**]{}, 1346 (1989). P. R. Giri, Modern Physics Letters A [**23**]{}, 2177 (2008). D.-H. Lee, G.-M. Zhang, and T. Xiang, Phys. Rev. Lett. [**99**]{}, 196805 (2007). A. Mesaros, D. Sadri and J. Zaanen, Phys. Rev. B [**79**]{}, 155111 (2009). P. E. Lammert and V. H. Crespi, Phys. Rev. Lett. [**85**]{}, 5190 (2000). J. Gonzalez, F. Guinea and M. A. H. Vozmediano, Nucl. Phys. B [**406**]{}, 771 (1993). B. Thaller, “[The Dirac Equation]{}”, Springer-Verlag (1992). C.-Y. Hou, C. Chamon, and C. Mudry, Phys. Rev. Lett. [**98**]{}, 186809 (2007). I. F. Herbut, Phys. Rev. Lett. [**99**]{}, 206404 (2007). I. F. Herbut, Phys. Rev B [**85**]{}, 085304 (2012).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'I review the insights emerging from recent large kinematic surveys of galaxies at low redshift, with particular reference to the SAMI, CALIFA and MaNGA surveys. These new observations provide a more comprehensive picture of the angular momentum properties of galaxies over wide ranges in mass, morphology and environment in the present-day universe. I focus on the distribution of angular momentum within galaxies of various types and the relationship between mass, morphology and specific angular momentum. I discuss the implications of the new results for models of galaxy assembly.'
author:
- Matthew Colless$^1$
title: |
Emerging angular momentum\
physics from kinematic surveys
---
Introduction
============
This brief review focusses on recent integral field spectroscopy surveys of the stellar kinematics in large samples of galaxies at low redshifts. It does not cover radio HI surveys of the neutral gas in low-redshift galaxies (which are important for understanding the kinematics at large radius) nor does it extend to surveys at high redshifts (which explore the origin and evolution of galaxy kinematics). What [*local*]{} surveys of stellar kinematics can tell us about angular momentum in galaxies is its dependence on mass, morphology and other properties (if sample selection is understood) and its dependence on environment (if the sample is embedded in a fairly complete redshift survey); such dependencies can provide [*indirect*]{} evidence for the origin and evolution of angular momentum.
It is immediately apparent that all current kinematic surveys have weaknesses relating to the trade-offs demanded by instrumental constraints: firstly, between spatial resolution and spatial coverage (also between spectral resolution and spectral coverage) and, secondly, between this per-galaxy information and sample size (also sample volume and completeness). The lack of radial coverage is a serious problem for late-type disk galaxies having exponential mass profiles (i.e. having Sersic index $n \approx 1$), for which $M/M_{\rm tot}=0.5$,0.8 at $R/R_e \approx 1.0$,1.8 and $j/j_{\rm tot} = 0.5$,0.8 at $R/R_e \approx 1.0$,2.2. But it is a much worse problem for early-type spheroidal galaxies with deVaucouleurs profiles ($n \approx 4$), for which $M/M_{\rm tot}=0.5$,0.8 at $R/R_e \approx 1.0$,3.2 and $j/j_{\rm tot} = 0.5$,0.8 at $R/R_e \approx 4.4$,$>$9 (see Figure 1a). This problem is compounded by the necessary instrumental trade-off between radial coverage (field of view) and spatial resolution (spaxel scale) of integral field units (IFUs) due to constraints imposed by the limited available detector area. For example, in the SAMI sample the median major axis is $R_e=4.4$arcsec (10–-90% range spans 1.8-–9.4arcsec) which means that the SAMI IFUs only sample out to a median radius of 1.7$R_e$ (see Figure 1b).
![(a) Left panel: the fraction of mass ($M$), angular momentum ($J$), and specific angular momentum ($j=J/M$) as functions of radius (in units of the effective radius, $R_e$) for both an exponential disk profile (Sersic index $n=1$; top panel) and a deVaucouleurs spheroid profile (Sersic index $n=4$; bottom panel) \[[@Romanowsky+Fall2012], Fig.3\]. (b) Right panel: the distribution of effective radius $R_e$ (in arcsec) for the SAMI galaxy sample, showing those parts of the sample for which the integral field unit covers $<$1$R_e$, $>$1$R_e$ and $>$2$R_e$ \[based on [@Green+2015], Fig.1\].[]{data-label="fig1"}](fig1a.pdf "fig:"){width="40.00000%"} ![(a) Left panel: the fraction of mass ($M$), angular momentum ($J$), and specific angular momentum ($j=J/M$) as functions of radius (in units of the effective radius, $R_e$) for both an exponential disk profile (Sersic index $n=1$; top panel) and a deVaucouleurs spheroid profile (Sersic index $n=4$; bottom panel) \[[@Romanowsky+Fall2012], Fig.3\]. (b) Right panel: the distribution of effective radius $R_e$ (in arcsec) for the SAMI galaxy sample, showing those parts of the sample for which the integral field unit covers $<$1$R_e$, $>$1$R_e$ and $>$2$R_e$ \[based on [@Green+2015], Fig.1\].[]{data-label="fig1"}](fig1b.pdf "fig:"){width="55.00000%"}
Surveys
=======
SAMI
----
SAMI is the Sydney-AAO Multi-IFU instrument on the 3.9m Anglo-Australian Telescope (AAT). It has 13 IFUs that can be positioned over a 1degree field at the telescope’s prime focus. Each hexabundle IFU has 61$\times$1.6arcsec fibres covering a 15arcsec diameter field of view. SAMI feeds the AAOmega spectrograph, which gives spectra over 375–575nm at $R$$\approx$1800 (70kms$^{-1}$) and 630–740nm at $R$$\approx$4300 (30kms$^{-1}$). The SAMI Second Data Release (DR2) includes 1559 galaxies (about half the full sample) covering 0.004$<$z$<$0.113 and 7.5$<$$\log(M_*/M_\odot)$$<$11.6. The core data products for each galaxy are two primary spectral cubes (blue and red), three spatially binned spectral cubes, and a set of standardised aperture spectra. For each core data product there are a set of value-added data products, including aperture and resolved stellar kinematics, aperture emission line properties, and Lick indices and stellar population parameters. The data release is available online through AAO Data Central ([datacentral.org.au](datacentral.org.au)).
CALIFA
------
CALIFA is the Calar Alto Legacy Integral Field survey, consisting of integral field spectroscopy for 667 galaxies obtained with PMAS/PPak on the Calar Alto 3.5m telescope. There are three different spectral setups: 375–750nm at 0.6nm FWHM resolution for 646 galaxies, 365–484nm at 0.23nm FWHM resolution for 484 galaxies, and a combination of these over 370–750nm at 0.6nm FWHM resolution for 446 galaxies. The CALIFA Main Sample spans 0.005$<$$z$$<$0.03 and the colour-magnitude diagram, with a wide range of stellar masses, ionization conditions and morphological types; the CALIFA Extension Sample includes rare types of galaxies that are scarce or absent in the Main Sample.
![Upper left: table of key parameters of the MaNGA, SAMI, CALIFA and Atlas3D surveys. Lower left: the wavelength coverage of the MaNGA (black), SAMI (grey), CALIFA (red & green) and Atlas3D (blue) surveys. Right: illustration of the relative fields of view covered by the IFUs used in each survey. \[Based on [@Sanchez+2015], Table 1 & Figure 1.\][]{data-label="fig2"}](fig2.pdf){width="\textwidth"}
MaNGA
-----
MaNGA is the Mapping Nearby Galaxies at Apache Point Observatory survey (part of SDSS-IV). It is studying the internal kinematic structure and composition of gas and stars in 10,000 nearby galaxies. It employs 17 fibre-bundle IFUs varying in diameter from 12arcsec (19 fibres) to 32arcsec (127 fibers) that feed two dual-channel spectrographs covering 360–1030nm at $R$$\approx$2000. The targets have $M_*$$>$$10^9$$M_\odot$ based on SDSS-I redshifts and $i$-band luminosities. The MaNGA sample is designed to approximate uniform radial coverage in terms of $R_e$, a flat stellar mass distribution, and a wide range of environments. SDSS Data Release 14 (DR14) includes MaNGA data cubes for 2812 galaxies.
Comparison
----------
Figure 2 provides a tabular and graphical summary of the parameters of these three surveys (and also the earlier Atlas3D survey), which helps to understand their various relative strengths and weaknesses, and consequently their complementarities. A few kinematic surveys of small samples offer greater radial coverage and higher velocity resolution: SLUGGS surveyed kinematics of 25 early-type galaxies to $\sim$3$R_e$ from stars and to $\sim$10$R_e$ using globular clusters ([@Bellstedt+2018 Bellstedt 2018]); PN.S surveyed the kinematics of 33 early-type galaxies to $\sim$10$R_e$ using planetary nebulae ([@Pulsoni+2018 Pulsoni 2018]).
Results
=======
Role of angular momentum
------------------------
After mass, angular momentum is the most important driver of galaxy properties, with a key role in the formation of structure and morphology. For regular oblate rotators, angular momentum can be derived from dynamical models as well as direct estimates of projected angular momentum. Surveys can determine population variations in the total angular momentum and its distribution with radius, exploring dependencies on mass, morphology, ellipticity and other properties. These relations can provide insights on the assembly histories of galaxies for comparison with simulations.
Angular momentum & spin profiles
--------------------------------
SAMI, CALIFA and MaNGA together now provide angular momentum profiles (or, alternatively, spin proxy, $\lambda_R$, as a function of $R/R_e$) for thousands of galaxies to $R/R_e$$\sim$1 and for hundreds of galaxies to $R/R_e$$\sim$2. These samples are large enough to be useful when split by mass, morphology or environment. Figure 3 shows spin profiles for galaxies from the CALIFA survey ([@Falcon-Barroso2016 Falcón-Barroso 2016]) and the MaNGA survey ([@Greene+2018 Greene 2018]); [@Foster+2018] give similar results from the SAMI survey.
![Left: Galaxy spin profiles from CALIFA, showing the variation with Hubble type \[[@Falcon-Barroso2016], Fig.2\]. Right: Galaxy spin profiles from MaNGA, showing the variation with mass for early and late-type galaxies \[[@Greene+2018], Fig.3\].[]{data-label="fig3"}](fig3a.pdf "fig:"){width="40.00000%"} ![Left: Galaxy spin profiles from CALIFA, showing the variation with Hubble type \[[@Falcon-Barroso2016], Fig.2\]. Right: Galaxy spin profiles from MaNGA, showing the variation with mass for early and late-type galaxies \[[@Greene+2018], Fig.3\].[]{data-label="fig3"}](fig3b.pdf "fig:"){width="59.00000%"}
Spin, morphology & ellipticity
------------------------------
Typical galaxies lie on a plane relating mass $M$, $j$ and stellar distribution (quantified by, e.g., Sersic index $n$ or photometric concentration index), with overall morphologies regulated by their mass and dynamical state (see, e.g., [@Cortese+2016 Cortese 2016]). The correlation shown in the left panel of Figure 4 between the offset from the mass–angular momentum ($M$-–$j$) relation and spin parameter $\lambda_R$ shows that at fixed $M$ the contribution of ordered motions to dynamical support varies by more than a factor of three. The right panel of Figure 4 shows that $\lambda_R$ correlates strongly with morphology and concentration index (especially if slow-rotators are removed), suggesting that late-type galaxies and early-type fast-rotators form a continuous class in terms of their kinematic properties.
![Left: Specific angular momentum versus stellar mass for SAMI galaxies, colour-coded by spin parameter $\lambda_R$. Right: Galaxy spin parameter versus $r$-band concentration for SAMI galaxies, colour-coded by morphology. \[Based on [@Cortese+2016], Figs 6 & 7\].[]{data-label="fig4"}](fig4a.pdf "fig:"){width="52.00000%"} ![Left: Specific angular momentum versus stellar mass for SAMI galaxies, colour-coded by spin parameter $\lambda_R$. Right: Galaxy spin parameter versus $r$-band concentration for SAMI galaxies, colour-coded by morphology. \[Based on [@Cortese+2016], Figs 6 & 7\].[]{data-label="fig4"}](fig4b.pdf "fig:"){width="46.00000%"}
The spin–ellipticity ($\lambda_R$–$\epsilon$) diagram is a particularly revealing frame for understanding relations between kinematic and morphological properties of galaxies. This is illustrated in Figure 5, from the work of [@Graham+2018] using the MaNGA survey. The left panel shows the strong correlation between the mass of a galaxy and its position in this diagram, with more massive galaxies tending to have lower spin and ellipticity. The central panel shows the areas of the diagram occupied by various morphological types: elliptical galaxies occupy the low-$\lambda_R$, low-$\epsilon$ region, while lenticular and spiral galaxies largely overlap, covering the full range of $\epsilon$ at $\lambda_R$$>$0.5. The right panel shows how galaxies belonging to different kinematic classes are distributed: spirals generally lie in the region consistent with rotationally-dominated kinematics, while regular (fast-rotating) early-type galaxies occupy a wider range of $\lambda_R$ at given $\epsilon$, with lower $\lambda_R$ corresponding to systems with more pressure-support; slowly-rotating (‘non-rotating’) early-type galaxies mainly occupy the region with $\lambda_R$$<$0.15 and 0$<$$\epsilon$$<$0.4.
![Distributions of galaxy properties in the spin–ellipticity ($\lambda_R$–$\epsilon$) diagram: left—stellar mass; centre—visual morphology; right—kinematic class. \[[@Graham+2018], Figs 5, 8 & 9.\][]{data-label="fig5"}](fig5.pdf){width="\textwidth"}
![Left/middle panels: The ratio of ordered to random motions ($V/\sigma$) versus the apparent ellipticity for early/late-type galaxies. Right panel: assuming galaxies are oblate rotators, the derifed distribution of intrinsic ellipticity as a function of apparent inclination. \[[@vandeSande+2018], Figs 3 & 4.\][]{data-label="fig6"}](fig6.pdf){width="\textwidth"}
There is also an strong correlation between a galaxy’s spin parameter and its intrinsic ellipticity, as demonstrated using the SAMI survey by [@vandeSande+2018]. Figure 6 shows the distribution of the ratio of rotation velocity to velocity dispersion ($V/\sigma$) with apparent ellipticity ($\epsilon$) for early-type and late-type galaxies, together with the inferred distribution of intrinsic ellipticity ($\epsilon_{\rm int}$). This is derived using the theoretical model predictions for rotating, oblate, axisymmetric spheroids with varying intrinsic shape and anisotropy shown by the dashed and dotted lines in the left two panels. The galaxies are colour-coded by the luminosity-weighted age of their stellar populations, and the righthand panel shows the clear trend of age with intrinsic ellipticity. As [@vandeSande+2018] discuss in detail, this newly discovered relation extends beyond the general notion that ‘disks are young’ and ‘bulges are old’.
The mass–angular momentum relation
----------------------------------
The mass–angular momentum ($M$-–$j$) relation is discussed in detail elsewhere in these proceedings. However, it is worth noting the opportunties for studying this key relation that follow from large surveys providing kinematics for many galaxies. Some prelminary results from the SAMI survey are shown in Figure 7 (D’Eugenio , in prep.), using hundreds of galaxies with masses and angular momenta derived from self-consistent dynamical models—in this case, Jeans anisotropic mass (JAM) models. This permits the study of the $M$-–$j$ relation for subsets of the population, such as different morphological types. While the results shown here are too preliminary to allow conclusions to be drawn, the opportunities are clear.
![The relation between mass and specific angular momentum, both derived from Jeans anisotropic mass (JAM) models fitted to the SAMI kinematic data, for elliptical (E), lenticular (S0), early-spiral (Sa-Sb) and late-spiral/irregular (Sc-Irr) galaxies. \[D’Eugenio , in prep.\][]{data-label="fig7"}](fig7.pdf){width="56.00000%"}
Summary
=======
This is a golden age for studying galaxy angular momentum. Large kinematic surveys using integral field spectrographs are vastly increasing the amount and richness of the available information Sample sizes are now beginning to allow studies of the dependence on multiple simultaneous influences (mass/morphology/environment...) The main limitations remain instrumental trade-offs between spatial resolution and radial coverage, and challenges in spatial resolution and surface brightness at higher redshift.
, 2018, *MNRAS*, 476, 4543 \[[DOI: 10.1093/mnras/sty456](https://doi.org/10.1093/mnras/sty456)\]
, 2016, *MNRAS*, 463, 170 \[[DOI: 10.1093/mnras/stw1891](https://doi.org/10.1093/mnras/stw1891)\]
, 2016, *Astronomical Surveys and Big Data*, ASP Conf. Series, 505, 133 \[[ADS: 2016ASPC..505..133F](https://ui.adsabs.harvard.edu/#abs/2016ASPC..505..133F)\]
, 2018, *MNRAS*, 480, 3105 \[[DOI: 10.1093/mnras/sty2059](https://doi.org/10.1093/mnras/sty2059)\]
, 2018, *MNRAS*, 477, 4711 \[[DOI: 10.1093/mnras/sty504](https://doi.org/10.1093/mnras/sty504)\]
, 2018, *ApJ*, 852, 36 \[[DOI: 10.3847/1538-4357/aa9bde](https://doi.org/10.3847/1538-4357/aa9bde)\]
, 2018, *A&A*, 618, 94 \[[DOI: 10.1051/0004-6361/201732473](https://doi.org/10.1051/0004-6361/201732473)\]
, 2015, *Galaxies in 3D across the Universe*, IAU Symposium 309, pp85-92 \[[DOI: 10.1017/S1743921314009375](https://doi.org/10.1017/S1743921314009375)\]
, 2018, *Nature Astronomy*, 2, 483 \[[DOI: 10.1038/s41550-018-0436-x](https://doi.org/10.1038/s41550-018-0436-x)\]
| {
"pile_set_name": "ArXiv"
} |
---
author:
- 'R. Smits'
- 'D.R. Lorimer'
- 'M. Kramer'
- 'R. Manchester'
- 'B. Stappers'
- 'C.J. Jin'
- 'R.D. Nan'
- 'D. Li'
bibliography:
- '11939.bib'
date: 'Received / Accepted '
nocite: '[@bdp+03; @lbk+04]'
title: Pulsar science with the Five hundred metre Aperture Spherical Telescope
---
Introduction
============
FAST, the Five-hundred-metre Aperture Spherical Telescope, is an Arecibo-style telescope currently under construction in China. FAST will be located in a karst depression in Guizhou province at a latitude of about $+26\degr$ and is planned to be completed around 2014 [@Nanb06; @Nan08]. With an illuminated aperture of 300m diameter, it will be an order of magnitude more sensitive than the steerable 100-metre telescopes at Effelsberg and Green Bank and about twice as sensitive as Arecibo. Moreover, FAST will cover a frequency range of 70MHz to 3GHz and it will be able to observe at a zenith angle of 40 degrees without a significant loss in gain [@jin08]. By illuminating only part of the 300-metre aperture the zenith angle can be extended even further; at a zenith angle of 60 degrees about half the aperture is still illuminated. With its sensitivity and its ability to see the majority of the sky, including a large part of the Galactic plane, FAST has excellent potential for finding radio pulsars [see also @nan06].
Radio pulsars provide insights into a rich variety of physics and astrophysics. Applications to date [see e.g. @lk05] include the study of the Milky Way, globular clusters, the evolution and collapse of massive stars, the formation and evolution of binary systems, the properties of super-dense matter, extreme plasma physics, tests of theories of gravity and as cosmological probes. A goal for the coming decade is the direct detection of low-frequency gravitational waves via high-precision timing of an array of millisecond pulsars (MSPs) [@Jenet05]. Such an array would be greatly facilitated by further discoveries and timing of MSPs with FAST.
The aim of this paper is to investigate the possibilities of FAST for finding and timing radio pulsars. In §2 we present a simulation of a FAST survey of the pulsar sky, where we look into the survey speed and the number of pulsars that can be detected for different receivers. In §3 we present calculations to estimate the data rates and computational requirements for a pulsar survey with FAST. In §4 we look into the possibility for FAST to participate in a global pulsar timing array effort. §5 contains a discussion of the results, as well as a comparison between FAST and the Square Kilometre Array.
Survey simulation
=================
In the past, the southern hemisphere has been the subject of extensive pulsar surveys [@mld+96; @lml+98; @mlc+01]. FAST will be able to complement these surveys in the northern hemisphere with an even greater sensitivity. Initially, FAST will have a 19-beam receiver with a frequency range of 1.1–1.5GHz and a system temperature of 20K (not including the sky temperature). A possible future phased array feed (PAF) is planned to have over 100 beams [@jin08] with the same frequency range and a system temperature of 30K. Table \[tab:parameters\] shows the parameters involved in the survey simulation. The number of beams will determine the speed of the survey and the dwell time per pointing. We therefore perform the simulations as a function of both frequency and field of view (FoV).
Parameter Value
-------------------------------------- ----------------
Gain 16.5KJy$^{-1}$
System temperature 19-beam receiver 20K
System temperature 100-beam receiver 30K
Centre frequency 1315MHz
Bandwidth 400MHz
Number of frequency channels 9500
No. of polarisations 2
Beam FWHM 3$^\prime$.4
No. of beams 19 or 100
: Expected system parameters of FAST and the pulsar survey.
\[tab:parameters\]
Simulation method
-----------------
We performed Monte Carlo simulations following @Lorimer2006 using the psrpop[^1] package [cf. @Smits09]. In their study, @Lorimer2006 used the results from recent surveys with the Parkes Multibeam system to derive an underlying population of pulsars with an optimal set of probability density functions for pulsar period ($P$), 1400-MHz radio luminosity ($L$), Galactocentric radius ($R$) and height above the Galactic plane ($z$). We make use of these results in our simulations described below which use as a starting point model C$^\prime$ from @Lorimer2006.
Our simulation procedure begins by generating a population of normal pulsars which beam towards the Earth. Each pulsar is assigned a value of $P$, $L$, $R$ and $z$ based on the assumed probability density functions. For the distributions in $P$, $R$ and $z$, we use the distributions from C$^\prime$ of @Lorimer2006. To compute $L$, we adopt the log-normal distribution found by @Faucher2006. This has the advantage of not requiring a specific lower bound in $L$, as is the case for the power-law luminosity models considered by @Lorimer2006.
We compute intrinsic pulse widths using the following self-consistent approach. Assuming a simple geometry with circular beams of radius $\rho$, the pulse width can be found from the inclination angle between the pulsar spin and magnetic axis, $\alpha$, and the impact parameter between the magnetic axis and the line of sight, $\beta$. Following Kramer et al. (1998), we use the empirical relationship between $\rho$ and spin period $P$ (s) defined as follows: $$\rho_{\rm model} = \left\{
\begin{array}{ll}
5.4^{\circ} P^{-1/2}
& \mbox{if $P>30$~ms} \\
31.2^{\circ} & \mbox{otherwise},
\end{array}
\right.$$ where the angle of $31.2^{\circ}$ is the value of the above expression evaluated for $P=30$ ms. To model the scatter about this relationship (see, e.g., Fig. 3.5 from Lorimer & Kramer 2005), we use a Monte Carlo approach to find a dithered beam radius $$\rho = 10^{\log_{10} (\rho_{\rm model}) + p},$$ where $p$ is a random number drawn from a flat distribution in the range $-0.15 \leq p \leq 0.15$. This, somewhat arbitrary, scheme dithers the beam radius logarithmically by a third of a decade and provides a good match to the observed scatter. The pulsar is assumed to be beaming towards us and the impact parameter $\beta$ is chosen from a simple flat distribution in the range $-\rho \leq \beta \leq \rho$. Finally, we assume that the magnetic inclination angle $\alpha$ is randomly distributed, that is $$\alpha = \arccos q,$$ where $q$ is a random number drawn from a flat distribution in the range $0<q<1$. With these quantities defined, the observed pulse width $W$ is found using the following geometrical relationship (see, e.g., Gil et al. 1984): $$\sin^2 \left( \frac{W}{4} \right) =
\frac{ \sin^2 \left( \frac{\rho}{2} \right) -
\sin^2 \left( \frac{\beta}{2} \right) }{
\sin\alpha \cdot \sin ( \alpha + \beta ) }.$$
To compute the expected DM and scatter broadening effects on each pulse, we use the NE2001 electron density model [@NE2001]. Note that, since we are primarily concerned with the distant population of highly dispersed pulsars in these simulations, we do not attempt to account for interstellar scintillation. Finally, to allow us to extrapolate the 1400-MHz luminosities to other survey frequencies in the next step, we make the reasonable assumption that pulsar spectra can be approximated as a power law [@Lorimer95] and assign each pulsar a spectral index drawn from a normal distribution with mean of –1.6 and standard deviation 0.35.
As described by @Lorimer2006, when generating the normal pulsar population according to the above criteria, we keep track of the number of pulsars detectable by the detailed model of the Parkes Multibeam Pulsar Surveys of the Galactic plane [@mlc+01] and at high latitudes [@bjd+06]. Our simulations terminate when the numbers of model detectable normal pulsars matches the 1005 pulsars found by these surveys. This results in an underlying sample of 120,000 model pulsars.
To model the MSP population, we follow the same approach and assumptions as for the normal pulsars described above, but with two exceptions: (i) model $z$ heights are chosen from an exponential distribution with a mean of 500 pc [@Cordes97]; (ii) the underlying period distribution used is taken from a recent study of the MSP population (Lorimer et al. 2009; in preparation). Note that implicit in this approach is the notion that the luminosity functions for normal pulsars and MSPs are identical; this is consistent with earlier results [@lml+98].
In addition to the Parkes Multibeam Surveys considered above for the normal pulsars, we also make use of the Parkes 70-cm pulsar survey of the southern sky [@mld+96; @lml+98] and two intermediate latitude surveys carried out by the Parkes multibeam system [@ebsb01; @jbo+07]. Our simulations are normalized such that they terminate when the total of 49 MSPs detected by these surveys is reached. This results in an underlying sample of 23000 model MSPs.
Once the pulsar population was determined, we performed simulations to find the number of pulsars FAST would detect as a function of frequency. For this purpose, we allowed the centre frequency to range from 400MHz to 1.4GHz. The bandwidth was kept at one third of the centre frequency, the observation time per pointing was kept constant at 600s and the gain was kept constant at 16.5KJy$^{-1}$. The frequency-dependent sky temperature was added to the system temperature. Considering the beamwidth of about 3.4 arcminutes at 1.4GHz, we limited the surveys to Galactic latitudes of either between $\pm 5^\circ$ or between $\pm 10^\circ$. Further, we simulated the number of pulsars FAST will detect as a function of observation time. All detected pulsars include already known pulsars.
Simulation Results
------------------
Fig. \[fig:Frequency\] shows the number of normal pulsars and MSPs that were detected in the simulation of the Galactic plane as a function of centre frequency. The figure shows that for both normal pulsars and MSPs, the number of detected pulsars is reasonably stable from centre frequencies of 1 to 1.4GHz. Similar results are found for observation times of 120 and 1800 seconds.
The number of detected pulsars as a function of observation time is shown in Fig. \[fig:ObservationTime\]. This figure shows that, while the number of detected pulsars continues to climb with increasing observation time, the rate of increase slows significantly after times of about 600 s.
The 19 beams in the multibeam receiver of FAST have a total FoV of about 0.061deg$^2$. A possible future PAF with over 100 beams will have a total FoV of 0.32deg$^2$. Table \[tab:speed\] shows the total survey time for different observation times per pointing, for different survey regions for both a 19-beam and a 100-beam receiver. The total survey time is given in days assuming eight hours of observation time per day.
Parameter Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7
------------------------------------------ ------------------------ ------------------------- ------------------------ ------------------------- ------------------------ ------------------------- -------------
Survey region $|\rm{b}|$$<$$5^\circ$ $|\rm{b}|$$<$$10^\circ$ $|\rm{b}|$$<$$5^\circ$ $|\rm{b}|$$<$$10^\circ$ $|\rm{b}|$$<$$5^\circ$ $|\rm{b}|$$<$$10^\circ$ Visible Sky
Pointing time (s) 600 600 600 600 1800 1800 120
Number of beams 19 19 100 100 100 100 100
Number of normal pulsars detected 5100 6300 4200 5200 5700 7000 4500
Number of MSPs detected 470 670 380 550 550 770 500
Currently known normal pulsars in region 352 418 352 418 352 418 662
Currently known MSPs in region 14 20 14 20 14 20 62
Total survey time (eight hour days) 229 454 44 86 131 259 296
\[tab:speed\]
With the initial 19-beam receiver about 5200 previously unknown pulsars, including 460 MSPs, can be found by searching the region with $20^\circ<\mathrm{l}<90^\circ$ and $|\mathrm{b}|<5^\circ$, using 10-minute pointings. This survey would take just over 200 days to complete. Using the 100-beam receiver to survey the same region with 30-minute pointings, 5900 previously unknown pulsars can be found in 130 days, including about 540 MSPs. Doubling the survey area to $|\mathrm{b}|<10^\circ$ will add about 1100 pulsars using 10-minute pointings and about 1450 pulsars using 30-minute pointing. The 100-beam receiver will also enable an all-sky survey. Using 2-minute pointings, 4300 previously unknown pulsars can be found in 300 days, including about 440 MSP’s.
\
\
Fig. \[fig:Scatter\] shows the distribution of detected pulsars for all the models. The top two plots show the normal pulsars that are detected in the models that are limited to the Galactic plane. The plots directly below show the millisecond pulsars that are detected in the same models. The bottom plot shows the all-sky survey.
Survey of M31 and M33
---------------------
The great sensitivity of FAST will permit searches for pulsars in other galaxies. We can estimate the number of pulsars that FAST can detect in a survey of M31 and M33, both visible for about six hours per day from FAST. Table \[tab:M31M33\] shows the parameters of these galaxies.
Galaxy Declination Distance (kpc) Mass (M$_\odot$) Angular extent
-------- ------------------------ ------------------------- ----------------------------------------- ----------------
M31 41$^\circ$ 16$\arcmin$ 772$\pm$44 [@Ribas05] $8.2\times 10^{11}$ [@Seigar08] 3deg$^2$
M33 30$^\circ$ 39$\arcmin$ 964$\pm$54 [@Bonanos06] $\gtrsim 5\times 10^{10}$ [@Corbelli00] 0.8deg$^2$
\[tab:M31M33\]
The minimum detectable flux density of a pulsar in a survey is given by: $$S_{\rm min}=\frac{{\rm S/N}\,(T_{\rm sys}+T_{\rm sky})}{G\sqrt{n_{\rm p}t_{\rm obs}B}}\sqrt\frac{W}{P-W} {\rm mJy},$$ where S/N is the signal-to-noise ratio required for a detection, $T_{\rm sys}$ is the system noise temperature (K), $T_{\rm sky}$ is the sky temperature (K), $G$ is the gain of the telescope (KJy$^{-1}$), $n_{\rm p}$ is the number of polarisations, $t_{\rm
obs}$ is the observation length (s) per pointing, $B$ is the observation bandwidth (MHz), $P$ is the period of the pulsar (s) and $W$ is the equivalent width (s) [@lk05]. For an observation frequency of 1.315GHz and S/N = 9, $T_{\rm sys}$ = 20K, $T_{\rm sky}$ = 1.4K for M31 and $T_{\rm sky}$ = 1.1K for M33, $G$ = 16.5KJy$^{-1}$, $n_{\rm p}$ = 2, $t_{\rm obs}$ = 6 hours, $B$ = 400MHz and assuming an equivalent width of 5% of the period we obtain a minimum detectable flux density of 600nJy. This leads to a minimum detectable luminosity of 360 and 560mJykpc$^2$ for a pulsar in M31 and M33, respectively. There are at least 55 pulsars known in our Galaxy with a higher luminosity at 1.4GHz than 360mJykpc$^2$ and 32 pulsars higher than 560mJykpc$^2$. If we correct for the mass ratio between our Galaxy and the galaxies M31 and M33, then we can expect of order 80 pulsars to be detected by a FAST survey of M31 and about 3 pulsars in M33. However, the supernova rate per unit mass in our Galaxy is thought to be higher than in M31 and considerably lower than in M33 [@Berkhuijsen84; @Gordon98]. Also, since the angular extent of M33 is about 1 quarter that of M31, the number of pulsars found per pointing might not differ much between M31 and M33. We estimate that the total number of pulsars found in a FAST survey of both M31 and M33 would lie between 50 and 100. Finding such a number of pulsars would allow studies of the pulsar population in these galaxies and would provide a valuable probe of the inter-galactic medium. The combined angular extent of M31 and M33 is 3.8deg$^2$, which means about 1500 beams are required. With a 19-beam receiver, this would require 79 pointings each of duration six hours, leading to a total of 470 hours of observing time.
Data processing
===============
The total amount of data from a FAST pulsar survey is given by $${\cal D}_{total} = Bt_{obs}\frac{\Omega_{survey}}{\rm{FoV_{beam}}}N_{pol}\frac{N_{bits}}{8}\rm{\,Bytes},
\label{eq:Data}$$ where $B$ is the bandwidth, $t_{obs}$ is the observation time per pointing, FoV$_{\rm beam}$ is the FoV of 1 beam, $\Omega_{\rm survey}$ is the total field of the survey, $N_{\rm pol}$ is the number of polarisations and $N_{\rm bits}$ is the number of bits used in the digitisation.
Figure \[fig:Data\] shows the total amount of data from a FAST pulsar survey as a function of the total survey field for different observation times. We have assumed a bandwidth of 400MHz, a FoV$_{beam}$ of 0.0025 deg$^2$, 1 polarisation and 2 bits per sample. Table \[tab:data\] shows the total amount of data for the seven different survey models from table \[tab:speed\].
Parameter Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7
----------------------------------- ------------------------ ------------------------- ------------------------ ------------------------- ------------------------ ------------------------- -------------
Survey region $|\rm{b}|$$<$$5^\circ$ $|\rm{b}|$$<$$10^\circ$ $|\rm{b}|$$<$$5^\circ$ $|\rm{b}|$$<$$10^\circ$ $|\rm{b}|$$<$$5^\circ$ $|\rm{b}|$$<$$10^\circ$ Visible Sky
Pointing time (s) 600 600 600 600 1800 1800 120
Total amount of data (peta-bytes) 16.7 33.0 16.7 33.0 50.0 99.0 113
\[tab:data\]
The data acquisition rate from a pulsar survey sampling the detected outputs of a filterbank of total bandwidth $B$ and channel bandwidth $\Delta\nu$ is given by: $${\cal D}_{rate} = N_{\rm beams}\frac{1}{t_{\rm samp}}\frac{B}{\Delta\nu}\frac{N_{\rm bits}}{8}\mathrm{\,Bps},
\label{eq:DR}$$ where $N_{\rm beams}$ is the number of beams in the receiver, $t_{\rm
samp}$ is the sampling time and $N_{\rm bits}$ is the number of bits used in the digitisation. $\Delta\nu$ can be estimated by demanding that the dispersion smearing within the frequency channel does not exceed the effective time resolution, given by the sampling time and the scattering time: $$\Delta\nu(\mathrm{GHz}) \le \frac{\sqrt{t_{\rm
samp}^2(\mu\mathrm{s})+t_{\rm scatt}^2(\mu\rm{s})}\nu_{\rm
min}^3(\mathrm{GHz})} {8.3\times10^3{\rm DM}},
\label{eq:deltanu}$$ where $\nu_{\rm min}$ is the minimum (lowest) frequency in the observation frequency band, DM is the dispersion measure and $t_{\rm
scatt}$ is the scattering time. Following @Bhat04, we relate $t_{\rm scatt}$ to the DM and observing frequency as follows: $$\begin{aligned}
\lefteqn{ \log_{10}(t_{\rm scatt}(\rm{ms})) = -6.46 + 0.154\log_{10}({\rm DM}) + \phantom{1.07(\log_{10}{\rm DM})} }\nonumber \\
& & \qquad\qquad\qquad1.07(\log_{10}{\rm DM})^2 - 3.86 \log_{10}(\nu(\rm{GHz})).
\label{eq:tscatt}\end{aligned}$$ In practice, however, the scattering time can differ from this by a factor up to 100 [see @Bhat04]. A smaller scattering time leads to the requirement of smaller – and thus more – frequency channels. To be on the safe side, we apply the highest frequency in the frequency band to Eq \[eq:tscatt\] and divide the resulting scattering time by 100. Substituting $t_{\rm scatt}$ from Eq. \[eq:tscatt\] in Eq. \[eq:deltanu\] and taking $\nu_{\rm min}$ = 1.115GHz, leads to a relationship between $\Delta\nu$ and DM with a minimum of 42kHz (at DM = 428). This means that having a constant frequency channel width of 42kHz over the frequency band will be sufficient for the pulsar survey. This leads to 9500 frequency channels for $B$=400MHz.
Figure \[fig:Datarate\] shows the data rate from a FAST pulsar survey as a function of FoV. We have assumed 100$\mu$s sampling time, a frequency range of 1.115 to 1.515GHz and 2 bits per sample. For the initial 19-beam receiver, the data rate is just under 0.5 GB per second. For 100 beams it is about 2.4 GB per second.
The number of operations required to search these data for normal pulsars, millisecond pulsars and accelerated pulsars in binary systems is approximately $$N_{\rm oa}=N_{\rm DM}N_{\rm acc}\times5N_{\rm samp}\log_2(N_{\rm samp}),$$ where $N_{\rm DM}$ is the number of DM-trial values and $N_{\rm
acc}$ is the number of trial accelerations. Fig. \[fig:Analysis\] shows the number of operations per second required to perform a real time analysis of a FAST pulsar survey, assuming 100 trial accelerations, a sampling time of 100$\mu$s, an observation time of 600s and DM$_{\rm max}$ = 1000cm$^{-3}$pc. $N_{\rm DM}$ was obtained numerically by demanding that the total time delay within the frequency band between two consecutive DM-trial values is equal to twice the effective time resolution, given by the sampling time and the scattering time of Eq. \[eq:tscatt\]. This leads to $N_{\rm DM}$ = 4280. For the initial 19-beam receiver, the required computation power for real time analysis is 9 tera-ops. For 100 beams it becomes 48 tera-ops. Contributions from de-dispersion, harmonic folding and possibly other processes would increase these values by a factor less than two. However, it is not unreasonable to store all the data from a FAST pulsar survey and to perform the analysis off-line. This would reduce the required computation power significantly.
Timing
======
In addition to being a very powerful search machine, FAST will also excel in the high-precision timing of pulsars. For a given pulsar, the timing precision roughly scales with the S/N of the observed pulse profile [@lk05], and hence with telescope sensitivity. Until the Square Kilometre Array comes on-line (§5) FAST will have the largest sensitivity of all existing radio telescopes and will also be able to see a larger fraction of the sky than the Arecibo telescope. Consequently, FAST will be the telescope of choice for a number of high-precision timing experiments. While the Northern location of FAST means that it is unlikely to contribute significantly to the timing of, say, the Double Pulsar (PSR J0737$-$3039A/B, Burgay et al. 2003, Lyne et al. 2004), a major contribution of FAST will be the participation in the global Pulsar Timing Array efforts.
High-precision timing of a network of pulsars allows detection of a gravitational wave background by correlating the timing residuals of the pulsars [e.g. @Jenet06]. Current efforts include PPTA (Parkes Pulsar Timing Array), EPTA (European Pulsar Timing Array) and NANOGrav (North American Nanohertz Observatory for Gravitational Waves). In terms of combination of sky coverage and sensitivity, FAST will only be matched by the European LEAP efforts (Large European Array for Pulsars), where the European 100-m class telescopes are combined to obtain Arecibo-like sensitivity for pulsar timing. This effort, however, has to overcome the challenge of coherently combining the signals between telescopes up to 2000 kilometres apart.
Pulsar Timing Array experiment
------------------------------
Following earlier work by @det79 and @Kaspi94, pulsar timing is now routinely used to constrain the amplitude of a gravitational wave (GW) background. Currently the most stringent limits have been placed by @Jenet06 from observations of seven MSPs with the Parkes 64-m radio telescope [cf. @Manchester08] which constrain $\Omega_{\rm GW}$, the ratio of the energy density of the GW background to the closure density to be $\Omega_{\rm GW}h^2 <
1.9\times 10^{-8}$, where $h$ is the dimensionless Hubble constant defined as $H_0\equiv h\times 100$kms$^{-1}$Mpc$^{-1}$. The sensitivity that is needed to achieve a GW detection depends on details of the Galaxy merger rate, since the GW background generated by an ensemble of super-massive black holes distributed throughout the Universe is expected to be the strongest signal to be detected. The signal originating from such sources is expected at a level of $\Omega_{\rm GW}h^2 \sim 2\times 10^{-10}$ [@Jaffe03; @Wyithe03; @Enoki04]. Simulations suggest that the timing of 20 pulsars with a timing precision of 100ns over five years would lead to a sensitivity of $\Omega_{\rm GW}h^2 < 6.6\times 10^{-11}$ and hence to a first direct detection of a GW background [@Jenet06]. While this is difficult with current technology, FAST will have the sensitivity and sky coverage to make a major contribution to achieving this goal.
Here we estimate the amount of time needed to time the brightest MSPs, found by FAST, to high precision. For this simulation we choose model 2 from Table \[tab:speed\], which provides us with 670 MSPs. We take the 50 brightest of these and calculate the amount of time required for a S/N of 500, taking into account a conservative profile stabilisation time of five minutes. This leads to 24 hours to time all 50 pulsars once. These numbers do not include accompanying low-frequency observations to allow correction for variations in interstellar dispersion. As expected, we found that the FoV provided by 19 beams had no impact on the timing performance, as there was no instance where two or more pulsars could be timed simultaneously.
FAST versus the SKA
===================
The Square Kilometre Array (SKA) is a planned multi-purpose radio telescope with a collecting area approaching 1 million square metres. Although the exact design is not yet determined, it is likely to consist of aperture arrays to cover frequencies from 70MHz up to 500 or even 800MHz and 15-metre dishes for the higher frequencies. About half of the elements will be placed within a 5-kilometre core of the SKA, the rest will be placed on spiral arms extending several thousands of kilometres outwards. @Smits09 have studied the performance of the SKA concerning pulsar surveys and timing. Here we will discuss the differences between these findings and those of the current paper.
A pulsar survey with FAST is straightforward compared to an SKA pulsar survey. With the initial receiver, no beam-forming is required. The future PAF does require beam-forming, but it does not need the vast computational power of several peta-ops required for beam-forming the SKA. Another benefit for FAST is that the data do not need to be transported over long distances; the data processing can be performed close to the dish. Also, the data rates and required computation power for a real time acceleration search are much lower for the FAST survey. This is mostly due to the smaller FoV of FAST, which is 10 to 50 times less than the FoV for an SKA pulsar survey, but also because these data rates and computational power scale with the square of the telescope diameter. The diameter of FAST is given by the illuminated aperture of 300 metres. For the SKA it is determined by the largest baselines between the elements used in the survey. @Smits09 estimate that the limits in computational power will restrict these baselines to 1 kilometre. Because of this limitation, the telescope sensitivity of a pulsar survey with the SKA is about 2000m$^2$K$^{-1}$ which is equal to that of FAST. However, the larger FoV of the SKA will not only enable a survey of a much larger portion of the sky, but also allow for a longer dwell time per pointing. The actual FoV of the SKA depends on the final design, but at the very least it will be equal to the FoV of the 15-metre dishes which at 1.4GHz is about 0.64deg$^2$. For comparison, a reasonable assumption would be that an SKA survey has a dwell time of 40 minutes versus 10 minutes for a FAST survey. This makes the survey sensitivity of the SKA twice that of FAST. Compared to the initial 19-beam receiver of FAST, a pulsar survey with the SKA would be about 2.5 times faster, but with the 100-beam PAF, the survey speed of FAST would be twice that of the SKA. Moreover, because the elements of the SKA are fully steerable, the SKA will be able to survey a much larger part of the sky and with its latitude of about –30$^\circ$ this includes almost the entire Galactic plane.
Timing pulsars with the SKA can be performed with a considerable fraction of the full collecting area, providing a sensitivity close to 10000m$^2$K$^{-1}$. Also, the large FoV of the SKA allows timing of many pulsars simultaneously (but depending on the SKA design, this might not be the case when using the SKA for a pulsar timing array). Even the future 100-beam PAF of FAST only has a quarter of the FoV of 15-metre dishes and should the SKA be equipped with PAFs or mid-frequency aperture arrays, its FoV increases significantly to 20deg$^2$ or even 250deg$^2$.
Given that FAST will be operational well ahead of the full SKA, it is likely to be the most powerful telescope to perform pulsar science in the next decade and will remain an outstanding instrument for pulsar science, complementing the SKA once it is operational.
Summary
=======
The FAST pulsar survey simulation in §2 shows that, despite the small natural beam-size and limitations in zenith angle, FAST will be a formidable instrument for finding pulsars.
A tradeoff between the number of pulsars and total observation time can be determined from Table \[tab:speed\]. With the initial 19-beam receiver about 5200 previously unknown pulsars, including 460 MSPs, can be found in a survey of about 230 days of the visible part of the Galactic plane with $|{\rm b}|<5^\circ$. With a 100-beam receiver 5900 previously unknown pulsars can be found in the same region in 130 days, including about 540 previously unknown MSPs. The 100-beam receiver will also enable an all-sky survey yielding 4300 previously unknown pulsars in 300 days. Further, we estimate that a 470-hour survey of M31 and M33 would yield between 50 and 100 extra-Galactic pulsars.
A pulsar survey with the 19-beam receiver will produce just under 500 MB of data per second. Depending on the survey, this leads to a total of 17 or 33 peta-bytes. The major part of a real-time acceleration search of these data would require 9 tera-ops. After 2014, such data rates and processing power should be very feasible.
FAST will also be able to contribute to the existing efforts to detect a gravitational wave background by timing a large number of MSPs to high precision. Although the visible sky of FAST is limited to 58% of the entire sky, those pulsars that are visible to FAST can be timed to great precision very quickly. A simulation in §4 suggests that the 50 brightest MSPs, visible to FAST can be timed to a S/N of 500 in just 24 hours.
Compared to a pulsar survey with the SKA, FAST will match the sensitivity of the SKA, but the large FoV of the SKA will allow for longer dwell times and make the survey faster, at least for the initial 19-beam configuration of FAST. Also, the SKA will be able to survey a much larger part of the sky. However, having a single dish rather than many elements spread over a large area significantly reduces the required data rates and computational power. When timing pulsars for a pulsar timing array, the SKA will have much more sensitivity than FAST, since the SKA can use almost the full collecting area in this case. Also, the larger FoV of the SKA will allow timing of many pulsars simultaneously, although this will probably not benefit the timing of the limited number of pulsars from the timing array. Given that FAST will be operational well ahead of the full SKA, it will provide the best prospects for pulsar science in the next decade.
[^1]: <http://psrpop.sourceforge.net>.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Generalized Berwald manifolds are Finsler manifolds admitting linear connections such that the parallel transports preserve the Finslerian length of tangent vectors. By the fundamental result of the theory [@V5] such a linear connection must be metrical with respect to the averaged Riemannian metric given by integration of the Riemann-Finsler metric on the indicatrix hypersurfaces. Therefore the linear connection is uniquely determined by its torsion tensor. If the torsion is zero then we have a classical Berwald manifolds. Otherwise the torsion is a strange data we need to express in terms of quantities of the Finsler manifold. In the paper we are going to give explicit formulas for the linear connections with totally anti-symmetric torsion tensor of three-dimensional generalized Berwald manifolds. The results are based on averaging of (intrinsic) Finslerian quantities by integration over the indicatrix surfaces. They imply some consequences for the base manifold as a Riemannian space with respect to the averaged Riemannian metric. The possible cases are Riemannian spaces of constant zero curvature, constant positive curvature or Riemannian spaces admitting Killing vector fields of constant Riemannian length.'
address: |
Inst. of Math., Univ. of Debrecen\
H-4010 Debrecen, P.O.Box 12\
Hungary
author:
- 'Cs. Vincze'
title: 'On compatible linear connections with totally anti-symmetric torsion tensor of three-dimensional generalized Berwald manifolds'
---
Introduction {#introduction .unnumbered}
============
The notion of generalized Berwald manifolds goes back to V. Wagner [@Wag1]. They are Finsler manifolds admitting linear connections such that the parallel transports preserve the Finslerian length of tangent vectors (compatibility condition). The basic questions of the theory are the unicity of the compatible linear connection and its expression in terms of the canonical data of the Finsler manifold (intrinsic characterization). In case of a classical Berwald manifold admitting a compatible linear connection with zero torsion, the intrinsic characterization is the vanishing of the mixed curvature tensor of the canonical horizontal distribution. In general the intrinsic characterization of the compatible linear connection is based on the so-called *averaged Riemannian metric* given by integration of the Riemann-Finsler metric on the indicatrix hypersurfaces. By the fundamental result of the theory [@V5] such a linear connection must be metrical with respect to the averaged Riemannian metric. Therefore the linear connection is uniquely determined by its torsion tensor. Following Agricola-Friedrich [@AF] consider the decomposition $$T(X,Y):=T_1(X,Y)+T_2(X,Y),\ \ \textrm{where}\ \ \displaystyle{T_1(X,Y):=T(X,Y)-\frac{1}{n-1}\big(\tilde{T}(X)Y-\tilde{T}(Y)X\big)},$$ $\tilde{T}$ is the trace tensor of the torsion and $$\label{trace}
T_2(X,Y):=\frac{1}{n-1}\left(\tilde{T}(X)Y-\tilde{T}(Y)X\right).$$ In case of 2D the torsion tensor is automatically of the form (\[trace\]); see [@V13]. If the dimension is at least three then the trace-less part can be divided into two further components $$T_1(X,Y)=A_1(X,Y)+S_1(X,Y)\ \ \Rightarrow\ \ T(X,Y)=A_1(X,Y)+S_1(X,Y)+T_2(X,Y)$$ by separating the totally anti-symmetric/axial part $A_1$. Therefore we have eight possible classes of generalized Berwald manifolds depending on the surviving terms such as classical Berwald manifolds admitting torsion-free compatible linear connections [@Szab1] (we have no surviving terms) or Finsler manifolds admitting semi-symmetric compatible linear connections (we have no trace-less part) [@V6], [@V9], [@V10] and [@V11]. In the paper we are going to give explicit formulas for the linear connections with totally anti-symmetric torsion preserving the Finslerian length of tangent vectors in case of three-dimensional Finsler manifolds. The results are based on averaging of (intrinsic) Finslerian quantities by integration over the indicatrix surfaces. They imply some consequences for the base manifold as a Riemannian space with respect to the averaged Riemannian metric. The possible cases are Riemannian space forms of constant zero curvature, constant positive curvature or Riemannian spaces admitting Killing vector fields of constant Riemannian length.
Notations and terminology
=========================
Let $M$ be a differentiable manifold with local coordinates $u^1, \ldots, u^n.$ The induced coordinate system of the tangent manifold $TM$ consists of the functions $x^1, \ldots, x^n$ and $y^1, \ldots, y^n$. For any $v\in T_pM$, $x^i(v):=u^i\circ \pi (v)=p$ and $y^i(v)=v(u^i)$, where $i=1, \ldots, n$ and $\pi \colon TM \to M$ is the canonical projection.
Finsler metrics
---------------
A Finsler metric is a continuous function $F\colon TM\to \mathbb{R}$ satisfying the following conditions:
- $\displaystyle{F}$ is smooth on the complement of the zero section (regularity),
- $\displaystyle{F(tv)=tF(v)}$ for all $\displaystyle{t> 0}$ (positive homogenity),
- the Hessian $$g_{ij}=\frac{\partial^2 E}{\partial y^i \partial y^j}, \ \ \textrm{where} \ \ E=\frac{1}{2}F^2$$ is positive definite at all nonzero elements $\displaystyle{v\in T_pM}$ (strong convexity).
The so-called *Riemann-Finsler metric* $g$ is constituted by the components $g_{ij}$. It is defined on the complement of the zero section because the second order partial differentiability of the energy function at the origin does not follow automatically: if $E$ is of class $C^2$ on the entire tangent manifold $TM$ then, by the positively homogenity of degree two, it follows that $E$ is quadratic on the tangent spaces, i.e. the space is Riemannian. The Riemann-Finsler metric makes each tangent space (except at the origin) a Riemannian manifold with standard canonical objects such as the [*volume form*]{} $\displaystyle{d\mu=\sqrt{\det g_{ij}}\ dy^1\wedge \ldots \wedge dy^n}$, the *Liouville vector field* $\displaystyle{C:=y^1\partial /\partial y^1 +\ldots +y^n\partial / \partial y^n}$ and the [*induced volume form*]{} $$\mu=\sqrt{\det g_{ij}}\ \sum_{i=1}^n (-1)^{i-1} \frac{y^i}{F} dy^1\wedge\ldots\wedge dy^{i-1}\wedge dy^{i+1}\ldots \wedge dy^n.$$ on the indicatrix hypersurface $\displaystyle{\partial K_p:=F^{-1}(1)\cap T_pM\ \ (p\in M)}$.
Generalized Berwald manifolds
-----------------------------
A linear connection $\nabla$ on the base manifold $M$ is called *compatible* to the Finslerian metric if the parallel transports with respect to $\nabla$ preserve the Finslerian length of tangent vectors. Finsler manifolds admitting compatible linear connections are called generalized Berwald manifolds.
Suppose that the parallel transports with respect to $\nabla$ (a linear connection on the base manifold) preserve the Finslerian length of tangent vectors and let $X_t$ be a parallel vector field along the curve $c\colon [0,1]\to M$. We have that $$\label{paralleldiff}
(x^k\circ X_t)'={c^k}'\ \ \textrm{and}\ \ (y^k \circ X_t)'={X_t^k}'=-{c^i}' X_t^j \Gamma_{ij}^k\circ c$$ because of the differential equation for parallel vector fields. If $F$ is the Finslerian fundamental function then $$\label{eq:5}
(F \circ X_t)'=(x^k\circ X_t)'{\frac{\partial F}{\partial x^k}}\circ X_t+(y^k \circ X_t)'{\frac{\partial F}{\partial y^k}}\circ X_t$$ and, by formula (\[paralleldiff\]), $$\label{eq:55}
(F \circ X_t)'={c^i}'\bigg(\frac{\partial F}{\partial x^i}-y^j {\Gamma}_{ij}^{k}\circ \pi \frac{\partial F}{\partial y^k}\bigg)\circ X_t.$$ This means that the parallel transports with respect to $\nabla$ preserve the Finslerian length of tangent vectors (compatibility condition) if and only if $$\label{cond1}
\frac{\partial F}{\partial x^i}-y^j {\Gamma}^k_{ij}\circ \pi \frac{\partial F}{\partial y^k}=0\ \ (i=1, \ldots,n),$$ where the vector fields of type $$\label{eq:6}
X_i^{h^{\nabla}}:=\frac{\partial}{\partial x^i}-y^j {\Gamma}^k_{ij}\circ \pi \frac{\partial}{\partial y^k}$$ span the associated horizontal distribution belonging to $\nabla$.
\[heritage\] *[@V5]* If a linear connection on the base manifold is compatible to the Finslerian metric function then it must be metrical with respect to the averaged Riemannian metric $$\label{averagemetric1}
\gamma_p (v,w):=\int_{\partial K_p} g(v, w)\, \mu=v^i w^j \int_{\partial K_p} g_{ij}\, \mu \ \ (v, w\in T_p M, p\in U).$$
Three-dimensional Finsler manifolds admitting compatible linear connections with totally anti-symmetric torsion tensor
======================================================================================================================
Suppose that $\nabla$ is a compatible linear connection of a three-dimensional generalized Berwald manifold. By Theorem \[heritage\], such a linear connection must be metrical with respect to the averaged Riemannian metric (\[averagemetric1\]) given by integration of the Riemann-Finsler metric on the indicatrix hypersurfaces. Therefore $\nabla$ is uniquely determined by its torsion tensor. Taking vector fields with pairwise vanishing Lie brackets on the neighbourhood $U$ of the base manifold, the Christoffel process implies that $$X\gamma(Y,Z)+Y\gamma(X,Z)-Z\gamma(X,Y)=$$ $$2\gamma(\nabla_X Y, Z)+\gamma(X, T(Y,Z))+\gamma(Y, T(X,Z))-\gamma(Z, T(X,Y))$$ and, consequently, $$\label{Cproc}
\gamma(\nabla^*_X Y,Z)=\gamma(\nabla_X Y, Z)+\frac{1}{2}\left(\gamma(X, T(Y,Z))+\gamma(Y, T(X,Z))-\gamma(Z, T(X,Y))\right),$$ where $\nabla^*$ denotes the Lévi-Civita connection of the averaged Riemannian metric.
The torsion tensor is totally anti-symmetric if its lowered tensor $$T_{\flat}(X,Y,Z):=\gamma (T(X,Y),Z)$$ belongs to $\wedge^3 M$.
If $\nabla$ is a metrical linear connection with totally anti-symmetric torsion then $$\label{totanti}
\nabla^*_X Y=\nabla_X Y-\frac{1}{2}T(X,Y)$$ and the geodesics of $\ \nabla^*$ and $\nabla$ coincide.
If $\dim M=3$ then $\dim \wedge^3 M=1$ and, consequently, $$T_{\flat}\left(\frac{\partial}{\partial u^i}, \frac{\partial}{\partial u^j}, \frac{\partial}{\partial u^k}\right)=f \gamma \left(\frac{\partial}{\partial u^i} \times_{\gamma} \frac{\partial}{\partial u^j}, \frac{\partial}{\partial u^k}\right)=f\det \gamma_{ij}$$ for some local function $f\colon U\to \mathbb{R}$, where the orientation is choosen such that the coordinate vector fields represent a positive basis. This means that $$\label{formulamain06}
\nabla^*_X Y=\nabla_X Y-\frac{f}{2}X \times_{\gamma} Y.$$ Taking the Riemannian energy $\displaystyle{E^*(v):=\gamma(v,v)/2}$, the Riemann-Finsler metric is $\displaystyle{g^*_{ij}=\gamma_{ij}\circ \pi}$ and the cross product of vertical vector fields is defined by $$g^* \left(\frac{\partial}{\partial y^i}\times_{g^*}\frac{\partial}{\partial y^j}, \frac{\partial}{\partial y^k}\right)=\det g^*_{ij}=\det \gamma_{ij}\circ \pi$$ with bilinear extension. Formula (\[formulamain06\]) can be written in terms of the induced horizontal structures as follows. Since the horizontal distributions induced by $\nabla^*$ and $\nabla$ are spanned by the vector fields $$X_i^{h^{*}}=\frac{\partial }{\partial x^i}-y^j {\Gamma^*}_{ij}^l\circ \pi\frac{\partial}{\partial y^l}\ \ \textrm{and}\ \ X_i^{h^{\nabla}}=\frac{\partial }{\partial x^i}-y^j {\Gamma}_{ij}^l\circ \pi\frac{\partial}{\partial y^l},$$ respectively, we have, by formula (\[formulamain06\]), that $$\label{comparison}
X_i^{h^*}=X_i^{h^{\nabla}}+f\circ \pi V_i, \ \ \textrm{where}\ \ V_i=\frac{1}{2} \frac{\partial}{\partial y^i}\times_{g^*} C \ \ (i=1,2,3).$$
Three-dimensional Finsler manifolds admitting compatible linear connections with totally anti-symmetric torsion tensor
----------------------------------------------------------------------------------------------------------------------
Let $M$ be a three-dimensional Finsler manifold admitting a compatible linear connection with totally anti-symmetric torsion tensor. Using the comparison formula (\[comparison\]), the compatibility condition (\[cond1\]) gives that $$\label{key1}
X_i^{h^*}E=f\circ \pi V_i E \ \ (i=1, 2, 3) \ \ \Rightarrow \ \ VE=f\circ \pi \sum_{i=1}^3 (V_i E)^2,$$ where the vector field $V$ is defined by the formula $\displaystyle{V:=\sum_{i=1}^3 (V_i E) X_i^{h^*}}$.
\[allorno\] If $\displaystyle{\sum_{i=1}^3 (V_i E)_v^2 = 0}$ for any $v\in T_pM$ then the Finslerian indicatrix $\partial K_p$ is a sphere with respect to the averaged Riemannian metric.
Since the infinitesimal rotation represented by the matrix $$\left(\begin{matrix}
0 & -v^1 & \ \ v^2 \\
\ \ v^1 & 0 & -v^3\\
-v^2 & \ \ v^3 & 0
\end{matrix}\right)$$ is of rank $2$, $\displaystyle{\sum_{i=1}^3 (V_i E)_v^2 = 0}$ implies that the vector fields $V_1$, $V_2$ and $V_3$ span the tangent plane to the Finslerian indicatrix at any $v\in \partial K_p$. Therefore its Euclidean normal vector field (with respect to $g^{*}$) is proportional to $C$. Taking a curve $c\colon [0,1]\to \partial K_p$ we have $$0=g^*_{c(t)}\left(C\circ c(t), c'(t)\right)=\gamma_p(c(t), c'(t))=\frac{1}{2}\gamma(c,c)'(t),$$ i.e. the Euclidean norm of $c(t)$ is constant. Since the Finslerian indicatrix surface is arcwise connected this means that it is a sphere with respect to the averaged Riemannian metric.
For a three-dimensional non-Riemannian Finsler manifold, the compatible linear connection with totally anti-symmetric torsion tensor must be of the form $$\label{comparison01}
\nabla_X Y=\nabla^*_{X} Y+\frac{f}{2} X \times_{\gamma} Y,$$ where $\nabla^*$ is the Lévi-Civita connection of the averaged Riemannian metric $\gamma$ and the function $f$ is given by $$f(p)=\frac{1}{\sigma(p)}\int_{\partial K_p} VE\, \mu,$$ where $$V=\sum_{i=1}^3 (V_i E) X_i^{h^*},\ \ \sigma(p)=\sum_{i=1}^3 \int_{\partial K_p} (V_i E)^2 \, \mu, \ \ V_i=\frac{1}{2} \frac{\partial}{\partial y^i}\times_{g^*} C.$$
Since $\displaystyle{VE=f\circ \pi \sum_{i=1}^3 (V_i E)^2}$ it follows that $$\label{bubu}
\int_{\partial K_p} VE \, \mu=f(p)\sum_{i=1}^3 \int_{\partial K_p} (V_i E)^2\, \mu$$ If the integrand on the right hand side is zero then, by Lemma \[allorno\], $\partial K_p$ is a Euclidean sphere in $T_pM$ with respect to $\gamma$. In case of a generalized Berwald manifolds we have linear parallel transports between the tangent spaces. Since the translates of a quadratic surface are quadratic, this means that the manifold is Riemannian. Otherwise we can divide equation (\[bubu\]) to express the function $f$.
Curvature properties
====================
Let a point $p\in M$ be fixed and consider the subgroup $G$ of orthogonal transformations with respect to the averaged inner product leaving the indicatrix $\partial K_p$ invariant in $T_pM$. Such a group is obviously closed in $O(3)$ and, consequently, it is compact. If we have a generalized Berwald manifold then the group $G$ is essentially independent of the choice of $p$ because the parallel translations with respect to the compatible linear connection $\nabla$ makes them isomorphic provided that the manifold is connected. On the other hand $G$ must be finite or reducible unless the manifold is Riemannian; see [@V12 Remark 5]. According to Theorem \[heritage\] it follows that $\textrm{Hol} \ \nabla\subset G$, i.e. the holonomy group of a compatible linear connection is finite or reducible in case of a a non-Riemannian generalized Berwald manifold. Using vector fields $X$, $Y$ and $Z$ with pairwise vanishing Lie-brackets on a neighbourhood $U\subset M$, the comparison formula (\[comparison01\]) says that $$\label{curv}
R(X,Y)Z=R^*(X,Y)Z+\frac{1}{2} \left((Xf)Y-(Yf)X\right)\times_{\gamma} Z+\frac{f^2}{4} \left(X\times_{\gamma} Y\right)\times_{\gamma} Z$$ because of the Jacobi identity $$\left(X\times_{\gamma} Y\right)\times_{\gamma} Z+\left(Z\times_{\gamma} X\right)\times_{\gamma} Y+\left(Y\times_{\gamma} Z\right)\times_{\gamma} X=0$$ and the product rule $$\nabla_X^* (Y\times_\gamma Z)=\left(\nabla_X^* Y\right)\times_\gamma Z+Y\times_{\gamma} \left(\nabla_X^* Z\right).$$ Using the vector triple product extension formula $$X\times_{\gamma} \left(Y\times_{\gamma} Z\right)=\gamma(X,Z)Y-\gamma(X,Y)Z$$ it follows that $$\label{sect}
\gamma(R(X,Y)Y,X)=\gamma(R^*(X,Y)Y,X)-\frac{f^2}{4} \det \left(\begin{matrix}
\gamma(X,X) & \gamma(X,Y)\\
\gamma(X,Y) & \gamma(Y,Y)\end{matrix}\right).$$
The curvature tensor of $\nabla$ obviously satisfies the curvature property $$\label{curv1}
R(X,Y)Z=-R(Y,X)Z.$$ Property $$\label{curv2}
\gamma(R(X,Y)Z,W)=-\gamma(R(X,Y)W,Z)$$ also holds because $\nabla$ is metrical with respect to the averaged Riemannian metric $\gamma$. We are going to investigate the Jacobi identity and the block symmetry.
\[ccp\] The curvature tensor of $\nabla$ satisfies the Jacobi identity $$\label{curv3}
R(X,Y)Z+R(Z,X)Y+R(Y,Z)X=0$$ if and only if the function $f$ is constant.
As a straightforward calculation shows $$\label{jacobi}
R(X,Y)Z+R(Z,X)Y+R(Y,Z)X\stackrel{(\ref{curv})}{=}(Xf) Y\times_\gamma Z+(Zf) X\times_\gamma Y+(Yf) Z\times_\gamma X.$$ Taking the inner product of both sides with $Z$ (for example) it can be easily seen that the left hand side is zero if and only if $Zf=0$ for any vector field $Z$ on the base manifold.
*To complete the list of the classical curvature properties we need to investigate the so-called block-symmetry $$\label{curv4}
\gamma(R(X,Y)Z,W)=\gamma(R(Z,W)X,Y);$$ especially, if the dimension is $3$, then properties (\[curv3\]) and (\[curv4\]) are equivalent to each other for any curvature tensor satisfying (\[curv2\]). It follows by a pure algebraic way [@V11 Remark 3.11].*
The case of finite holonomy group
---------------------------------
Suppose that $G$ is finite and, consequently, the holonomy group of the compatible linear connection is also finite, i.e. its curvature is zero.
If $M$ is a connected three-dimensional non-Riemannian Finsler manifold admitting a compatible flat linear connection with totally anti-symmetric torsion tensor then
- $M$ is a classical Berwald manifold of constant zero sectional curvature with respect to the averaged Riemannian metric or
- $M$ is a proper generalized Berwald manifold of constant positive sectional curvature with respect to the averaged Riemannian metric.
If we have a compatible linear connection with zero curvature then the classical curvature properties (\[curv1\]) - (\[curv4\]) are all satisfied and $f$ must be a constant function due to Lemma \[ccp\]. The result comes from the comparison formula (\[sect\]): if the (constant) function $f$ is identically zero then we have a Riemannian space form of constant zero curvature. Otherwise it is of constant positive curvature.
[*If $M$ is complete then, by the Killing-Hopf theorem of Riemannian geometry, it follows that the universal cover of $M$ (as a Riemannian space with respect to the averaged Riemannian metric) is isometric to $\mathbb{R}^3$ or the Euclidean unit sphere $S^3\subset \mathbb{R}^4$. Otherwise the manifold (as a non-Riemannian Finsler space) does not admit a compatible flat linear connection with totally anti-symmetric torsion tensor.*]{}
The case of non-finite reducible holonomy group
-----------------------------------------------
\[thm:4\] If $M$ is a connected three-dimensional non-Riemannian Finsler manifold admitting a compatible non-flat linear connection $\nabla$ with totally anti-symmetric torsion tensor then there exists a one-dimensional distribution $\mathcal{D}$ such that
- any local section of constant length is a covariant constant vector field with respect to $\nabla$,
- any local section of constant length is a Killing vector field of constant length with respect to the averaged Riemannian metric.
Suppose that $R_p(X,Y)\neq 0$ for some point $p\in M$, i.e. the holonomy group of the compatible linear connection contains a one-parameter subgroup of orthogonal transformations with respect to the averaged Riemannian metric leaving the indicatrix $\partial K_p$ invariant. Since $M$ is a non-Riemannian generalized Berwald manifold it follows that $G$ containing all orthogonal transfromations leaving the indicatrix $\partial K_p$ invariant can not be transitive or finite because $G\supset \textrm{Hol} \ \nabla$. Therefore it is a reducible group containing a one-parameter subgroup of rotations; see [@V12 Remark 5]. The distribution is constituted by the one-dimensional invariant subspace of $G$ as the point of the manifold is varying. The local sections of constant length must be constructed as follows: if $\beta_p$ generates the one-dimensional invariant subspace (the common axis of the rotations) of $G$ in $T_pM$ then, by parallel transports with respect to $\nabla$, we can extend it to a local covariant constant section $\beta$, i.e. $\displaystyle{\nabla_{X_q} \beta =0}$ for any vector field $X$ on $M$ and $q\in U$, where $U$ is a local neighbourhood around $p\in M$. Since $\displaystyle{\nabla_X \beta=0}$ it follows that $$\nabla^*_X \beta\stackrel{(\ref{comparison01})}{=}-\frac{f}{2} X \times_{\gamma} \beta$$ and, consequently, $$\left(\mathcal{L}_{\beta} \gamma\right)(X,Y)=\beta \gamma(X,Y)-\gamma([\beta, X],Y)-\gamma([\beta,Y],X),$$ where $$[\beta,X]=\nabla^*_{\beta}X-\nabla^*_X \beta=\nabla^*_{\beta}X+\frac{f}{2} X \times_{\gamma} \beta=\nabla^*_{\beta}X -\frac{f}{2} \beta \times_{\gamma} X.$$ Therefore $$\left(\mathcal{L}_{\beta} \gamma\right)(X,Y)=\beta \gamma(X,Y)-\gamma(\nabla^*_{\beta}X,Y)-\gamma(\nabla^*_{\beta}Y,X)+\frac{f}{2}\gamma (\beta \times_{\gamma} X,Y)+\frac{f}{2}\gamma (\beta \times_{\gamma} Y,X)=$$ $$\beta \gamma(X,Y)-\gamma(\nabla^*_{\beta}X,Y)-\gamma(\nabla^*_{\beta}Y,X)=0$$ as was to be proved.
The following example can be considered as a converse of the global version of Theorem \[thm:4\].
An example
----------
Suppose that $M$ is a connected Riemannian manifold admitting a Killing vector field $\beta$ of unit length. An easy direct computation shows that $$\left(\mathcal{L}_{\beta} \gamma\right)(X,Y)=\gamma(\nabla^*_X \beta,Y)+\gamma(\nabla^*_Y \beta ,X)=0,$$ i.e. the Hesse form $\gamma(\nabla^*_X \beta,Y)$ is anti-symmetric. On the other hand $\gamma(\nabla^*_X \beta,\beta)=0$ because $\beta$ is of constant length. Therefore $$\gamma(\nabla^*_X \beta,Y)=-\frac{f}{2} \gamma(\beta \times_{\gamma} X, Y)$$ for some function $f$, i.e. $$\nabla^*_X \beta=-\frac{f}{2} \beta \times_{\gamma} X.$$ Taking the metric connection $\nabla$ with torsion $T(X,Y)=f X\times_{\gamma} Y$ it follows, by the comparison formula (\[comparison01\]), that $\beta$ is a covariant constant vecor field with respect to $\nabla$. This means that for any vector fields $X$ and $Y$ $$R(X,Y)\beta=0, \ \ \textrm{i.e.} \ \ R(X,Y)Z=r(X,Y) \beta\times_{\gamma} Z$$ for some anti-symmetric scalar-valued form $r\in \wedge^2 M$ because $\beta$ is the vector invariant of the anti-symmetric mapping $\displaystyle{Z \mapsto R(X,Y)Z}$. By the Ambrose-Singer theorem, the unit component of the holonomy group (the so-called restricted holonomy group) of $\nabla$ is the one-parameter rotational group generated by $\beta$ at each point of the manifold. Taking a point $p\in M$ let us choose a non-qadratic convex revolution surface around the axis of $\beta_p$. For an explicite example consider a trifocal ellipsoid (it is a kind of generalized conics instead of the classical conics of the Riemannian geometry) body defined by the equation $$\label{trifocal}
\|v+\beta_p\|+ \|v\|+\|v-\beta_p\|\leq \textrm{const.};$$ the focal set consists of $\pm \beta_p$, ${\bf 0}$ and the constant is large enough to contain the focal points in the interior of the body to avoid singularities. Using parallel transports with respect to $\nabla$ we can extend (\[trifocal\]) to each point of the manifold. Note that $\pm \beta_p$ in the focal set provide that (\[trifocal\]) is invariant under not only the restricted holonomy group but the entire one including possibly reflections about the two-dimensional invariant subspace. Such a smoothly varying family of convex bodies induces a (non-Riemannian) fundamental function $F$ such that it is invariant under the parallel transport with respect to $\nabla$.
*Killing vector fields of constant length naturally appear in different geometric constructions such as $K$-contact and Sasakian manifolds [@Blair], [@BMS] and [@BG]. There are many restrictions to the existence of Killing vector fields of constant length on a Riemannian manifold; for a comprehensive survey see [@BN]: for example, if a compact Riemannian manifold admits such a vector field then its Euler characteristic must be zero in the sense of a Theorem due to H. Hopf [@BN Section 1].*
[99]{}
I. Agricola and T. Friedrich, *On the holonomy of connections with skew-symmetric torsion*, Math. Ann. [**328**]{} (4) (2004), pp. 711-748.
D. Blair, *Contact manifolds in Riemannian geometry*, Springer Lectures Notes in Math., V. [**509**]{}, Springer Verlag, Berlin and New York, 1976.
F. Belgun, A. Moroianu, and U. Semmelmann, *Symmetries of contact metric manifolds*, Geom. Dedic. [**101**]{} (1) (2003), pp. 203-216.
V.N. Beretovskii and Yu. G. Nikonorov, *Killing vector fields of constant length on Riemannian manifolds*, Siberian Mathematical Journal [**49**]{} (3) (2008), pp 395-407.
C. Boyer and K. Galicki, *On Sasakian-Einstein geometry*, Internat. J. Math. [**11**]{} (7) (2000), pp. 873-909.
Z. I. Szabó, *Positive definite Berwald spaces. Structure theorems on Berwald spaces*, Tensor (N. S.) [**35**]{} (1) (1981), pp. 25-39.
Cs. Vincze, *A new proof of Szabó’ s theorem on the Riemann-metrizability of Berwald manifolds*, J. AMAPN, 21 (2005), 199-204.
Cs. Vincze, *On a scale function for testing the conformality of Finsler manifolds to a Berwald manifold*, Journal of Geometry and Physics. 54 (2005), 454-475.
Cs. Vincze, *On Berwald and Wagner manifolds*, J. AMAPN, 24 (2008) 169-178.
Cs. Vincze, *Generalized Berwald manifolds with semi-symmetric linear connections*, Publ. Math. Debrecen 83 (4) (2013), pp. 741-755.
Cs. Vincze, *On a special type of generalized Berwald manifolds: semi-symmetric linear connections preserving the Finslerian length of tangent vectors*, “Finsler geometry: New methods and Perspectives”, European Journal of Mathematics, December 2017, Volume 3, Issue 4, pp 1098 - 1171 .
Cs. Vincze, *Lazy orbits: an optimization problem on the sphere*, Journal of Geometry and Physics Volume 124, January 2018, Pages 180-198.
Cs. Vincze, T. Khosdhani, Z. Mehdizadeh and M. Oláh, *Intrinsic characterizations of compatible linear connections of two-dimensional generalized Berwald manifolds*, submitted for piblication to J. of Diff. Geom. and Its Appl.
V. Wagner, *On generalized Berwald spaces*, CR Dokl. Acad. Sci. USSR (N.S.) [**39**]{} (1943), pp. 3-5.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We show that the automorphism group of a linking system associated to a saturated fusion system ${\mathcal{F}}$ depends only on ${\mathcal{F}}$ as long as the object set of the linking system is ${\mathrm{Aut}}({\mathcal{F}})$-invariant. This was known to be true for linking systems in Oliver’s definition, but we demonstrate that the result holds also for linking systems in the considerably more general definition introduced previously by the author of this paper. A similar result is proved for linking localities, which are group-like structures corresponding to linking systems. Our argument builds on a general lemma about the existence of an extension of a homomorphism between localities. This lemma is also used to reprove a theorem of Chermak showing that there is a natural bijection between the sets of partial normal subgroups of two possibly different linking localities over the same fusion system.'
address: 'Institute of Mathematics, University of Aberdeen, Fraser Noble Building, Aberdeen AB24 3UE, U.K.'
author:
- Ellen Henke
bibliography:
- 'my.bib'
title: Extensions of homomorphisms between localities
---
Introduction
============
Given a finite group $G$ with a Sylow $p$-subgroup $S$, the fusion system ${\mathcal{F}}_S(G)$ is the category whose objects are all subgroups of $S$ and whose morphisms are the injective group homomorphisms induced by conjugation in $G$. It turns out that the fusion system ${\mathcal{F}}_S(G)$ determines the $p$-completed classifying space $BG^\wedge_p$ up to homotopy; this statement is known as the Martino–Priddy conjecture and was first proved by Oliver [@Oliver2004; @Oliver2006]. Fusion systems also play an important role in many other contexts, for example in a program announced by Aschbacher to revisit the classification of finite simple groups. The concept of a saturated fusion system generalizes the properties of ${\mathcal{F}}_S(G)$. In particular, a saturated fusion system is a category ${\mathcal{F}}$ which comes equipped with a $p$-group $S$ such that the objects of ${\mathcal{F}}$ are all the subgroups of $S$ and the morphism sets consist of injective group homomorphisms subject to certain axioms.
For the purposes of homotopy theory, Broto, Levi and Oliver [@BLO2] defined *centric linking systems* associated to saturated fusion systems. A category $\L_S^c(G)$, which is a centric linking system associated to ${\mathcal{F}}_S(G)$, can be constructed directly from the group $G$. The $p$-completed classifying space $BG^\wedge_p$ of $G$ is homotopy equivalent to the $p$-completed nerve of the category $\L_S^c(G)$. This fact played an important role in the proof of the Martino-Priddy conjecture. In the abstract context, there is an essentially unique centric linking system associated to every saturated fusion system. This longstanding conjecture was proved by Chermak [@Ch] and subsequently by Oliver [@Oliver:2013]. Both proofs depend a priori on the classification of finite simple groups, but work of Glauberman–Lynd [@GL2016] removes the dependence of Oliver’s proof on the classification.
Linking systems form not only the algebraic foundation for defining $p$-completed classifying spaces of fusion systems, but they are also important when studying extensions of fusion systems. The object set of a centric linking system associated to a fusion system ${\mathcal{F}}$ over $S$ is a certain set of subgroups of $S$ determined by ${\mathcal{F}}$. When studying extensions, one often wants to choose the object sets of linking systems more flexibly. At least partly for that reason, a more general notion of linking systems was introduced by Oliver [@OliverExtensions] (building on earlier work of Broto, Castellana, Grodal, Levi and Oliver [@controlling]). Linking systems are special cases of transporter systems as defined by Oliver and Ventura [@OV]. Extensions of linking systems and transporter systems were studied for example in [@AOV1], [@BCGLO2007], [@OV] and [@OliverExtensions].
Chermak [@Ch] introduced with *localities* group-like structures that correspond to transporter systems in a certain way. A locality consists more precisely of a “partial group” $\L$ (i.e. a set $\L$ with a “product” defined on some tuples of elements of $\L$ subject to group-like axioms), a “Sylow $p$-subgroup” $S$ of $\L$, and a set $\Delta$ of subgroups of $S$ (cf. Definitions \[partial\] and \[locality\]). Here $\Delta$ is called the set of objects of $\L$ and turns out to be the object set of the transporter system corresponding to $(\L,\Delta,S)$. A rich theory of localities akin to the local theory of finite groups was developed by Chermak [@loc1; @loc2; @loc3]. Extensions of partial groups and localities were studied by Gonzalez [@Gonzalez] and are also the subject of work in progress of Valentina Grazian and the author of this paper. At least with the currently known conceptual framework, it seems in fact that there are some advantages to studying extensions of localities rather than extensions of linking systems or transporter systems. For example, for partial groups, there are natural notions of *homomorphisms* and of *partial normal subgroups* such that the kernels of the homomorphisms from a locality $\L$ are precisely the partial normal subgroups of $\L$.
The author of this paper [@subcentric] suggested a definition of a linking system which is significantly more general than the previously existing notion, and this leads to the corresponding concept of a *linking locality* (by Chermak [@loc2; @loc3] also called a proper locality). It is one of the purposes of this paper to prove in this more general context some results which are known to hold for linking systems in Oliver’s definition [@OliverExtensions Definition 3]. Another purpose of this paper is to prove a Lemma about homomorphisms between localities (Lemma \[L:Main\]) and to reprove in Theorem \[T:MainChermakII\] a result of Chermak [@loc2 Theorem A2]. Both Lemma \[L:Main\] and Theorem \[T:MainChermakII\] are used in joint work of Chermak and the author of this paper to show that there is a one-to-one correspondence between the normal subsystems of a fusion system and the partial normal subgroups of an associated linking locality; see [@normal] for a preliminary version of this work.
When studying extensions of linking systems or linking localities, their automorphism groups play an important role. Thus, it is of interest to see that different linking systems or linking localities associated to the same fusion system ${\mathcal{F}}$ have the same automorphism group. This is indeed the case if we consider linking systems and linking localities with ${\mathrm{Aut}}({\mathcal{F}})$-invariant object sets, as for example the typically used sets of ${\mathcal{F}}$-centric, ${\mathcal{F}}$-quasicentric or ${\mathcal{F}}$-subcentric subgroups. For linking localities, we prove the following theorem. In Theorem \[T:MainIso\] and Theorem \[T:MainAut\] we also prove some more general statements about isomorphisms and automorphisms of linking localities.
[A.1]{}\[T:A1\] Let ${\mathcal{F}}$ be a saturated fusion system over $S$. If $(\L,\Delta,S)$ and $(\L^+,\Delta^+,S)$ are linking localities over ${\mathcal{F}}$ such that $\Delta$ and $\Delta^+$ are ${\mathrm{Aut}}({\mathcal{F}})$-invariant, then ${\mathrm{Aut}}(\L,\Delta,S)\cong {\mathrm{Aut}}(\L^+,\Delta^+,S)$. In the case that $\Delta\subseteq\Delta^+$ and $\L=\L^+|_\Delta$, a group isomorphism is given by $${\mathrm{Aut}}(\L^+,\Delta^+,S)\longrightarrow {\mathrm{Aut}}(\L,\Delta,S),\;\alpha\mapsto \alpha|_\L.$$
The reader is referred to Definition \[D:RestrictionLocality\] for the definition of the “restriction” $\L^+|_\Delta$. The above mentioned correspondence between transporter systems and localities (which we outline in Subsection \[SS:TransLoc\]) leads to a correspondence between linking systems and linking localities. Taking the restriction $\L^+|_\Delta$ corresponds in the world of transporter systems to taking the full subcategory with object set $\Delta$. Thus, we obtain the following theorem for linking systems.
[A.2]{}\[T:A2\] Suppose ${\mathcal{F}}$ is a saturated fusion system. If ${\mathcal{T}}$ and ${\mathcal{T}}^+$ are linking systems associated to the same saturated fusion system ${\mathcal{F}}$ such that the object sets of ${\mathcal{T}}$ and ${\mathcal{T}}^+$ are ${\mathrm{Aut}}({\mathcal{F}})$-invariant, then ${\mathrm{Aut}}({\mathcal{T}})\cong{\mathrm{Aut}}({\mathcal{T}}^+)$. In the case that ${\mathcal{T}}$ is a full subcategory of ${\mathcal{T}}^+$, a group isomorphism ${\mathrm{Aut}}({\mathcal{T}}^+)\longrightarrow {\mathrm{Aut}}({\mathcal{T}})$ is given by restriction.
By ${\mathrm{Aut}}({\mathcal{T}})$ we mean here the group of isotypical self-equivalences of ${\mathcal{T}}$ which send inclusions to inclusions; see Definition \[D:TransIso\]. In the literature, ${\mathrm{Aut}}({\mathcal{T}})$ is often denoted by ${\mathrm{Aut}}_{\operatorname{typ}}^I({\mathcal{T}})$. We emphasize also that the term linking system refers to a linking system in the general sense of [@subcentric] (cf. Definition \[D:LinkingSystem\]). A version of Theorem \[T:A2\] was proved before by Andersen, Oliver and Ventura [@AOV1 Lemma 1.17] for linking systems in Oliver’s definition, i.e. for linking systems whose objects are quasicentric subgroups. The precise statement is actually given for outer automorphism groups of linking systems. We formulate a similar result in Theorem \[T:A2Outer\]. For this purpose, we state in Lemma \[L:ExactSequence\] that, for any linking system ${\mathcal{T}}$ associated to a saturated fusion system ${\mathcal{F}}$, there is an exact sequence $$1\longrightarrow Z({\mathcal{F}})\xrightarrow{\;\;\delta_S\;\;} {\mathrm{Aut}}_{\mathcal{T}}(S)\longrightarrow {\mathrm{Aut}}({\mathcal{T}})\longrightarrow {\mathrm{Out}}_{{{\operatorname{typ}}}}({\mathcal{T}})\longrightarrow 1.$$ Again, this was known to be true for linking systems in Oliver’s definition (cf. [@AOV1 Lemma 1.14(a)]) and the proof of the more general statement is given by similar arguments.
Theorem \[T:A2Outer\] allows us to prove the following theorem from the corresponding statement for centric linking systems which was shown by Broto, Levi and Oliver [@BLO2 Theorem 8.1]. The statement was also known before for linking systems in Oliver’s definition; see [@AKO Theorem 4.22]. For any space $X$, ${\mathrm{Out}}(X)$ denotes the group of homotopy classes of self-equivalences of $X$.
[B]{}\[C:B\] Let ${\mathcal{T}}$ be a linking system associated to a saturated fusion system ${\mathcal{F}}$ such that ${\mathrm{Ob}}({\mathcal{T}})$ is ${\mathrm{Aut}}({\mathcal{F}})$-invariant. Then there is an isomorphism $${\mathrm{Out}}_{\operatorname{typ}}({\mathcal{T}})\xrightarrow{\;\;\cong\;\;} {\mathrm{Out}}(|{\mathcal{T}}|^\wedge_p)$$ which sends the class of $\alpha\in{\mathrm{Aut}}({\mathcal{T}})$ to $|\alpha|^\wedge_p\colon |{\mathcal{T}}|^\wedge_p\rightarrow |{\mathcal{T}}|^\wedge_p$.
We show Theorem \[T:A2\] and some more general theorems about isomorphisms and automorphisms of linking systems (Theorems \[T:MainIsoT\] and \[T:MainAutTrans\]) from the corresponding statements for linking localities via the one-to-one correspondence between localities and transporter systems. However, in Remark \[R:AOV\], we outline how a direct proof could be given via similar arguments as in [@AOV1 Lemma 1.17]. The crucial point in each of the proofs of Theorems \[T:A1\] and \[T:A2\] is to show that the appropriate restriction map is surjective. The necessary argument for localities is similar to the argument for transporter systems in [@AOV1 Lemma 1.17], but it can be formulated in a very general way such that it becomes also useful in other contexts. Namely, in Lemma \[L:Main\] we show that, under certain assumptions, a homomorphism from a locality $(\L,\Delta,S)$ can be extended to a homomorphism from a locality $(\L^+,\Delta^+,S)$ with $\L^+|_\Delta=\L$. We use Lemma \[L:Main\] to give a new proof of [@loc2 Theorem A2] (stated as Theorem \[T:MainChermakII\](a) below) in Section \[S:PartialNormal\]. Moreover, both Lemma \[L:Main\] and Theorem \[T:MainChermakII\] will be used in a new version of [@normal]. For any partial group $\L$, we denote by ${{\mathfrak{N}}}(\L)$ the set of partial normal subgroups of $\L$.
[C]{}\[T:MainChermakII\] If $(\L,\Delta,S)$ and $(\L^+,\Delta^+,S)$ are linking localities over the same fusion system ${\mathcal{F}}$ with $\Delta\subseteq\Delta^+$ and $\L=\L^+|_\Delta$, then the following hold:
- The map $$\Phi_{\L^+,\L}\colon{{\mathfrak{N}}}(\L^+)\longrightarrow {{\mathfrak{N}}}(\L),\;{\mathcal{N}}^+\mapsto {\mathcal{N}}^+\cap\L$$ is well-defined and bijective. Both $\Phi_{\L^+,\L}$ and $\Phi_{\L^+,\L}^{-1}$ are inclusion-preserving.
- Given ${\mathcal{N}}^+\in{{\mathfrak{N}}}(\L)$ and ${\mathcal{N}}:={\mathcal{N}}^+\cap\L\in{{\mathfrak{N}}}(\L)$ such that ${\mathcal{F}}_{S\cap{\mathcal{N}}}({\mathcal{N}})$ is ${\mathcal{F}}_S(\L)$-invariant, we have ${\mathcal{F}}_{S\cap {\mathcal{N}}^+}({\mathcal{N}}^+)={\mathcal{F}}_{S\cap{\mathcal{N}}}({\mathcal{N}})$.
- Let ${\mathcal{N}}^+,{\mathcal{K}}^+\in{{\mathfrak{N}}}(\L)$, set ${\mathcal{N}}:={\mathcal{N}}^+\cap\L$, ${\mathcal{K}}:={\mathcal{K}}^+\cap\L$ and $T:={\mathcal{N}}^+\cap S={\mathcal{N}}\cap S$. Then ${\mathcal{N}}={\mathcal{K}}T$ if and only if ${\mathcal{N}}^+={\mathcal{K}}^+ T$.
The statement in part (a) of the above theorem that $\Phi_{\L,\L^+}$ and its inverse are inclusion-preserving is equivalent to saying that every ${\mathcal{N}}^+\in{{\mathfrak{N}}}(\L^+)$ is the smallest partial normal subgroup of $\L^+$ containing ${\mathcal{N}}^+\cap\L$. As a corollary to Theorem \[T:MainChermakII\](a) one can also show that any two linking localities over the same fusion system have the same number of partial normal subgroups; see Corollary \[C:PartialNormal\].
In his original proof of Theorem \[T:MainChermakII\](a), Chermak goes into the (somewhat complicated) details of the construction of elementary expansions as introduced in [@Ch Section 5]. Applying Lemma \[L:Main\] makes this unnecessary in our new proof. We do however use [@Ch Theorem 5.14], which is proved via elementary expansions. Theorem \[T:MainChermakII\](c) fills in a small gap in the proof of [@loc2 Lemma 7.3].
*Organization of the paper.* After introducing some background in Section \[S:LocFus\], we prove Lemma \[L:Main\], which is used in the proofs of our main results. Theorems \[T:A1\] and \[T:A2\] together with some more general theorems and with Theorem \[C:B\] are proved in Section \[S:Iso\]. In preparation for that, in Section \[S:Trans\], we define automorphisms and isomorphisms of transporter systems (cf. Definition \[D:TransIso\]). Moreover, we explain the correspondence between localities and transporter systems, which is then used in Section \[S:Iso\] to prove theorems about linking systems from corresponding statements about linking localities. Finally, in Section \[S:PartialNormal\], we prove Theorem \[T:MainChermakII\]. The proof of Theorem \[T:MainChermakII\] is independent of the results stated and proved in Sections \[S:Trans\] and \[S:Iso\].
Localities and fusion systems {#S:LocFus}
=============================
In this section we will introduce some basic definitions and show some lemmas needed in the proofs of our main theorems. The reader is referred to [@AKO] for background on fusion systems and to [@Ch] and [@loc1] for a more comprehensive introduction to localities. We will however summarize the most important definitions and results concerning localities. In particular we will recall the definitions of homomorphisms, projections, isomorphisms and automorphisms of localities in Subsection \[SS:LocalityHomomorphism\]. Some background on morphisms of fusion systems is also provided in Subsection \[SS:FusionMorphisms\].
Partial groups
--------------
For any set ${\mathcal{M}}$, write ${\bold{W}}({\mathcal{M}})$ for the set of words in ${\mathcal{M}}$. If $u,v\in{\bold{W}}({\mathcal{M}})$, then $u \circ v$ denotes the concatenation of the two words. The empty word will be denoted by $\emptyset$.
\[partial\] Let $\L$ be a non-empty set, let ${\mathbf{D}}$ be a subset of ${\bold{W}}(\L)$, let $\Pi \colon {\mathbf{D}}\longrightarrow \L$ be a map and let $(-)^{-1} \colon \L \longrightarrow \L$ be an involutory bijection, which we extend to a map $$(-)^{-1} \colon {\bold{W}}(\L) \longrightarrow {\bold{W}}(\L), w = (g_1, \dots, g_k) \mapsto w^{-1} = (g_k^{-1}, \dots, g_1^{-1})).$$ We say that $\L$ is a *partial group* with product $\Pi$ and inversion $(-)^{-1}$ if the following hold:
- $\L \subseteq {\mathbf{D}}$ (i.e. ${\mathbf{D}}$ contains all words of length 1), and $$u \circ v \in {\mathbf{D}}\Rightarrow u,v \in {\mathbf{D}};$$ (So in particular, $\emptyset\in{\mathbf{D}}$.)
- $\Pi$ restricts to the identity map on $\L$;
- $u \circ v \circ w \in {\mathbf{D}}\Rightarrow u \circ (\Pi(v)) \circ w \in {\mathbf{D}}$, and $\Pi(u \circ v \circ w) = \Pi(u \circ (\Pi(v)) \circ w)$;
- $w \in {\mathbf{D}}\Rightarrow w^{-1} \circ w\in {\mathbf{D}}$ and $\Pi(w^{-1} \circ w) = {\operatorname{\mathbf{1}}}$ where ${\operatorname{\mathbf{1}}}:=\Pi(\emptyset)$.
Note that any group $G$ can be regarded as a partial group with product defined on ${\mathbf{D}}={\bold{W}}(G)$ by extending the “binary” product to a map $$\Pi_G\colon{\bold{W}}(G)\longrightarrow G,(g_1,g_2,\dots,g_n)\mapsto g_1g_2\cdots g_n.$$ If $\L$ is a partial group with product $\Pi\colon{\mathbf{D}}\longrightarrow\L$ and $u=(f_1,f_2,\dots,f_n)\in{\mathbf{D}}$, then we write also $f_1f_2\cdots f_n$ for $\Pi(u)$.
\[L:PartialGroup\] Let $\L$ be a partial group with product $\Pi\colon{\mathbf{D}}\longrightarrow\L$.
- If $u,v\in{\bold{W}}(\L)$ with $u\circ ({\operatorname{\mathbf{1}}})\circ v\in{\mathbf{D}}$, then $u\circ v\in{\mathbf{D}}$ and $\Pi(u\circ ({\operatorname{\mathbf{1}}})\circ v)=\Pi(u\circ v)$.
- Given $u,v,w\in{\bold{W}}(\L)$ such that $u\circ v\circ v^{-1}\circ w\in{\mathbf{D}}$, we have $u\circ w\in{\mathbf{D}}$ and $\Pi(u\circ v\circ v^{-1}\circ w)=\Pi(u\circ w)$.
Let $u,v$ as in (a). If $u=v=\emptyset$, then by axiom (PG1) $u\circ v=\emptyset\in{\mathbf{D}}$, and by axiom (PG2) and the definition of ${\operatorname{\mathbf{1}}}$, we have $\Pi(u\circ v)=\Pi(\emptyset)={\operatorname{\mathbf{1}}}=\Pi({\operatorname{\mathbf{1}}})=\Pi(u\circ ({\operatorname{\mathbf{1}}})\circ v)$. So to prove (a), we may assume that $u\neq\emptyset$ or $v\neq \emptyset$.
For any element $f\in\L$, axiom (PG1) gives $f=(f)\circ\emptyset\in{\mathbf{D}}$. So by axioms (PG2) and (PG3) we have $(f,{\operatorname{\mathbf{1}}})=(f)\circ (\Pi(\emptyset))\in{\mathbf{D}}$ and $f=\Pi(f)=\Pi((f)\circ(\Pi(\emptyset)))=\Pi(f,{\operatorname{\mathbf{1}}})$. So if $u=(f_1,\dots,f_n)\neq \emptyset$, then $u\circ ({\operatorname{\mathbf{1}}})\circ v=(f_1,\dots,f_{n-1})\circ (f_n,{\operatorname{\mathbf{1}}})\circ v\in{\mathbf{D}}$ implies by axiom (PG3) that $u\circ v=(f_1,\dots,f_{n-1})\circ (\Pi(f_n,{\operatorname{\mathbf{1}}}))\circ v\in{\mathbf{D}}$ and $\Pi(u\circ v)=\Pi(u\circ ({\operatorname{\mathbf{1}}})\circ v)$. So (a) holds in this case. A similar argument show (a) in the case that $v\neq\emptyset$.
For the proof of (b), let now $u,v,w\in{\bold{W}}(\L)$ be arbitrary such that $u\circ v\circ v^{-1}\circ w\in{\mathbf{D}}$. Then by axiom (PG3), we have $u\circ ({\operatorname{\mathbf{1}}})\circ w=u\circ (\Pi(v\circ v^{-1}))\circ w\in{\mathbf{D}}$ and $\Pi(u\circ v\circ v^{-1}\circ w)=\Pi(u\circ ({\operatorname{\mathbf{1}}})\circ w)$. Hence, (b) follows from (a).
Let $\L$ be a partial group with product $\Pi\colon{\mathbf{D}}\longrightarrow\L$.
- For every $g\in \L$ we define $${\mathbf{D}}(g) = \{ x\in \L \mid (g^{-1}, x, g) \in {\mathbf{D}}\}.$$ The map $c_g \colon {\mathbf{D}}(g) \longrightarrow \L$, $x \mapsto x^g = \Pi(g^{-1}, x, g)$ is the *conjugation map* by $g$.
- If $\H$ is a subset of $\L$ and $\H \subseteq {\mathbf{D}}(g)$, then we set $$\H^g = \{ h^g \mid h \in \H\}.$$
- If $P\subseteq \L$, then $N_\L(P)$ is the set of all $g\in\L$ such that $P\subseteq{\mathbf{D}}(g)$ and $P^g=P$. Similarly, if $P$ and $Q$ are subsets of $\L$, we write $N_\L(P,Q)$ for the set of all $g\in\L$ such that $P\subseteq{\mathbf{D}}(g)$ and $P^g\subseteq Q$.
- A *partial subgroup* is a subset $\H\subseteq\L$ such that $h^{-1}\in\H$ for all $h\in\H$, and $\Pi(w)\in\H$ for all $w\in{\mathbf{D}}(\L)\cap{\bold{W}}(\H)$. A partial subgroup $\H$ of $\L$ is a called a *subgroup* of $\L$ if ${\bold{W}}(\H)\subseteq{\mathbf{D}}(\L)$.
- If ${\mathcal{N}}$ is a partial subgroup of $\L$, then ${\mathcal{N}}$ is called a *partial normal subgroup* if $n^f\in{\mathcal{N}}$ for all $f\in\L$ and all $n\in{\mathcal{N}}\cap{\mathbf{D}}(f)$.
We remark that a subgroup $\H$ of $\L$ is always a group in the usual sense with the group multiplication defined by $hg=\Pi(h,g)$ for all $h,g\in\H$. In particular, we can talk about $p$-subgroups of partial groups, meaning subgroups whose number of elements is a power of $p$.
Localities
----------
Roughly speaking, localities are partial groups with some some extra structure, in particular with a “Sylow $p$-subgroup” and a set $\Delta$ of “objects” which in a sense determines the domain of the product. Crucial is the following definition.
\[D:Sf\] Let $\L$ be a partial group.
- If $\Delta$ is a collection of subgroups of $\L$, define ${\mathbf{D}}_\Delta$ to be the set of words $w=(g_1, \dots, g_k) \in {\bold{W}}(\L)$ such that there exist $P_0, \dots ,P_k \in \Delta$ with $$P_{i-1} \subseteq {\mathbf{D}}(g_i)\mbox{ and }P_{i-1}^{g_i} = P_i\mbox{ for all }1 \leq i \leq k.$$ If such $P_0,\dots,P_k$ are given, then we say also that $w\in{\mathbf{D}}_\Delta$ via $P_0,P_1,\dots,P_k$, or just that $w\in{\mathbf{D}}_\Delta$ via $P_0$. In situations where we wish to stress the dependence of ${\mathbf{D}}_\Delta$ on $\L$ and on the product $\Pi\colon{\mathbf{D}}\longrightarrow\L$, we write ${\mathbf{D}}_\Delta(\L,\Pi)$ for ${\mathbf{D}}_\Delta$.
- Given a $p$-subgroup $S$ of $\L$ and $f\in\L$ set $$S_f:=\{x\in S\colon x\in{\mathbf{D}}(f)\mbox{ and }x^f\in S\}.$$ If we want to stress the dependence of $S_f$ on $\L$ and on the partial product and inversion on $\L$, then we write $S_f^\L$ for $S_f$.
\[locality\] Let $\L$ be a finite partial group, let $S$ be a $p$-subgroup of $\L$ and let $\Delta$ be a non-empty set of subgroups of $S$. We say that $(\L, \Delta, S)$ is a *locality* if the following hold:
1. $S$ is maximal with respect to inclusion among the $p$-subgroups of $\L$;
2. ${\mathbf{D}}= {\mathbf{D}}_\Delta$;
3. $\Delta$ is closed under taking $\L$-conjugates and overgroups in $S$; i.e. if $P \in \Delta$ then $P^g\in \Delta$ for every $g\in \L$ with $P \subseteq S_g$, and every subgroup of $S$ containing an element of $\Delta$ is an element of $\Delta$.
We remark that the above definition of a locality is a reformulation of the one given by Chermak [@loc1 Definition 2.8]. As argued in [@subcentric Remark 5.2], the two definitions are equivalent.
\[E:LGammaM\] Let $M$ be a finite group, $S\in{\mathrm{Syl}}_p(M)$ and ${\mathcal{F}}={\mathcal{F}}_S(M)$. Let $\Gamma$ be a non-empty ${\mathcal{F}}$-closed collection of subgroups of $S$. Set $$\L_\Gamma(M):=\{g\in G\colon S\cap S^g\in \Gamma\}=\{g\in G\colon \mbox{There exists } P\in\Gamma\mbox{ with }P^g\leq S\}$$ and let ${\mathbf{D}}$ be the set of tuples $(g_1,\dots,g_n)\in{\bold{W}}(M)$ such that there exist $P_0,P_1,\dots,P_n\in\Gamma$ with $P_{i-1}^{g_i}=P_i$. Then $\L_\Gamma(M)$ forms a partial group whose product is the restriction of the multivariable product in $M$ to ${\mathbf{D}}$, and whose inversion map is the restriction of the inversion map on the group $M$ to $\L_\Gamma(M)$. Moreover, $(\L_\Gamma(M),\Gamma,S)$ forms a locality. See [@Ch Example/Lemma 2.10] for a proof.
\[L:LocalitiesProp\] If $(\L,\Delta,S)$ is a locality, then the following hold:
- $N_\L(P)$ is a subgroup of $\L$ for every $P\in\Delta$.
- Let $P\in\Delta$ and $g\in\L$ with $P\subseteq S_g$. Then $Q:=P^g\in\Delta$, $N_\L(P)\subseteq {\mathbf{D}}(g)$ and $$c_g\colon N_\L(P)\longrightarrow N_\L(Q)$$ is an isomorphism of groups.
- Let $w=(g_1,\dots,g_n)\in{\mathbf{D}}$ via $(X_0,\dots,X_n)$. Then $$c_{g_1}\circ \dots \circ c_{g_n}=c_{\Pi(w)}$$ is a group isomorphism $N_\L(X_0)\longrightarrow N_\L(X_n)$.
- For every $g\in\L$, $S_g\in\Delta$. In particular, $S_g$ is a subgroup of $S$. Moreover, $S_g^g=S_{g^{-1}}$ and $c_g\colon S_g\longrightarrow S$ is an injective group homomorphism.
- For every $g\in\L$, $c_g\colon {\mathbf{D}}(g)\longrightarrow {\mathbf{D}}(g^{-1})$ is a bijection with inverse map $c_{g^{-1}}$.
- For every $w\in{\bold{W}}(\L)$, $S_w$ is a subgroup of $S_{\Pi(w)}$, and $S_w\in\Delta$ if and only if $w\in{\mathbf{D}}$.
Properties (a),(b) and (c) correspond to the statements in [@loc1 Lemma 2.3] except for the fact stated in (b) that $Q\in\Delta$, which is however clearly true if one uses our definition of a locality. Property (d) holds by [@loc1 Proposition 2.6(a),(b)] and property (e) is stated in [@Ch Lemma 2.5(c)]. Property (f) is [@loc1 Corollary 2.7].
\[L:NLSbiset\] If $(\L,\Delta,S)$ is a locality, $r\in N_\L(S)$ and $f\in\L$, then $(r,f)$, $(f,r)$ and $(r^{-1},f,r)$ are words in ${\mathbf{D}}$. Moreover, $$S_{(f,r)}=S_{fr}=S_f,\;S_{(r,f)}=S_{rf}=S_f^{r^{-1}}\mbox{ and }S_{f^r}=S_f^r.$$
We will use Lemma \[L:LocalitiesProp\](f) frequently in this proof without further reference. As $S_f^r\leq S$, we have $S_f\subseteq S_{(f,r)}$. In particular, since $S_f\in\Delta$, we have $S_{(f,r)}\in\Delta$ and $(f,r)\in{\mathbf{D}}$. So by [@Chermak:2015 Lemma 1.4(d)], $(f,r,r^{-1})\in{\mathbf{D}}$ and $f=\Pi(f,r,r^{-1})=(fr)r^{-1}$. Applying the first stated property with $(fr,r^{-1})$ in place of $(f,r)$, we also get $S_{fr}\subseteq S_{(fr,r^{-1})}$. We see now that $$S_f\subseteq S_{(f,r)}\subseteq S_{fr}\subseteq S_{(fr,r^{-1})}\subseteq S_{(fr,r^{-1})}\leq S_{(fr)r^{-1}}=S_f.$$ Hence, all the inclusions above are equalities and $S_f=S_{(f,r)}=S_{fr}$.
Similarly, as conjugation by $r$ takes $S_f^{r^{-1}}\leq S$ to $S_f$, we have $S_f^{r^{-1}}\leq S_{(r,f)}\in\Delta$ and $(r,f)\in{\mathbf{D}}$. So by [@Chermak:2015 Lemma 1.4(d)], $(r^{-1},r,f)\in{\mathbf{D}}$ and $f=r^{-1}(rf)$. Similarly, we have $S_{rf}^r\leq S_{(r^{-1},rf)}\leq S_{r^{-1}(rf)}=S_f$ and thus $S_{rf}\leq S_f^{r^{-1}}$. Hence $$S_f^{r^{-1}}\subseteq S_{(r,f)}\subseteq S_{rf}\subseteq S_f^{r^{-1}}$$ and equality holds everywhere above, i.e. $S_f^{r^{-1}}=S_{(r,f)}=S_{rf}$.
Note that $(r^{-1},f,r)\in{\mathbf{D}}$ via $S_f^r$. Using the properties proved above, we see now that $S_{f^r}=S_{(r^{-1}f)r}=S_{r^{-1}f}=S_f^r$.
Fusion systems of localities
----------------------------
Similarly as we we can attach to a finite group a fusion system over a Sylow $p$-subgroup, we can attach a fusion system to a locality.
Let $(\L,\Delta,S)$ be a locality.
- For all $P,Q\in\Delta$ set $${\mathrm{Hom}}_\L(P,Q):=\{c_g|_P\colon g\in N_\L(P,Q)\}.$$
- We write ${\mathcal{F}}_S(\L)$ for the smallest fusion system over $S$ containing all the conjugation maps $c_f\colon S_f\longrightarrow S$ with $f\in\L$, or equivalently for the fusion system generated by the sets ${\mathrm{Hom}}_\L(P,Q)$, where $P,Q$ are elements of $\Delta$.
- We say that $(\L,\Delta,S)$ is a locality over ${\mathcal{F}}$ to indicate that ${\mathcal{F}}={\mathcal{F}}_S(\L)$.
\[L:LocalityoverFFnatural\] If $(\L,\Delta,S)$ is a locality over ${\mathcal{F}}$ and $P\in\Delta$, then the following hold:
- For every morphism ${\varphi}\in{\mathrm{Hom}}_{\mathcal{F}}(P,S)$, then there exists $f\in\L$ such that $P\leq S_f$ and ${\varphi}=c_f|_P$.
- The subgroup $P$ is fully ${\mathcal{F}}$-normalized if and only if $N_S(P)\in{\mathrm{Syl}}_p(N_\L(P))$. Moreover, if so then for every $Q\in P^{\mathcal{F}}$, there exists $g\in N_\L(N_S(Q),S)$ such that $Q^g=P$.
- We have $N_{\mathcal{F}}(P)={\mathcal{F}}_{N_S(P)}(N_\L(P))$.
For (a) see [@subcentric Lemma 5.6]. Part (c) follows easily from (a). For the proof of (b) assume first that $P$ is fully normalized. As $N_S(P)$ is a $p$-subgroup of $N_\L(P)$, we can pick a Sylow $p$-subgroup $T$ of $N_\L(P)$ such that $N_S(P)\leq T$. By [@Ch Proposition 2.22(b)], there exists $x\in\L$ such that $T\subseteq{\mathbf{D}}(x)$ and $T^x\leq S$. Then in particular, $P\leq N_S(P)\leq S_x$ and by Lemma \[L:LocalitiesProp\](b), $T^x\leq N_S(P^x)$. By (a), we have $P^x\in P^{\mathcal{F}}$ and thus, as $P$ is fully normalized, $|N_S(P^x)|\leq |N_S(P)|$. On the other hand, $|N_S(P)|\leq |T|=|T^x|\leq |N_S(P^x)|$. Hence equality holds and thus $N_S(P)=T$ is a Sylow $p$-subgroup of $N_\L(P)$.
Suppose now on the other hand that $N_S(P)\in{\mathrm{Syl}}_p(N_\L(P))$. Take $Q\in P^{\mathcal{F}}$. By (a), there exists $f\in \L$ such that $Q^f=P$, and by Lemma \[L:LocalitiesProp\](b), the map $c_f\colon N_\L(Q)\longrightarrow N_\L(P)$ is an isomorphism of groups. Hence, $N_S(Q)^f$ is a $p$-subgroup of $N_\L(P)$. As $N_S(P)\in{\mathrm{Syl}}_p(N_\L(P))$, by Sylow’s theorem, there exists $a\in N_\L(P)$ such that $N_S(Q)^{fa}=(N_S(Q)^f)^a\leq N_S(P)$, where the equality uses Lemma \[L:LocalitiesProp\](c). Then $g:=fa\in N_\L(N_S(Q),S)$ with $Q^g=(Q^f)^a=P^a=P$. Moreover, $|N_S(Q)|=|N_S(Q)^g|\leq |N_S(P)|$. Because $Q\in P^{\mathcal{F}}$ was arbitrary, this shows that $P$ is fully normalized. Hence, (b) holds.
If $(\L,\Delta,S)$ is a locality and ${\mathcal{F}}={\mathcal{F}}_S(\L)$, then notice that $\Delta$ is ${\mathcal{F}}$-closed as defined next.
Let ${\mathcal{F}}$ be a fusion system over $S$, and let $\Delta$ be a set of subgroups of $S$.
- The set $\Delta$ is *closed under ${\mathcal{F}}$-conjugacy* if $P^{\mathcal{F}}\subseteq \Delta$ for every $P\in\Delta$.
- We call $\Delta$ *${\mathcal{F}}$-closed* if $\Delta$ is both closed under ${\mathcal{F}}$-conjugacy and overgroup closed in $S$.
Important examples of ${\mathcal{F}}$-closed collections are the set ${\mathcal{F}}^c$ of ${\mathcal{F}}$-centric subgroups (cf. [@AKO Definition 3.1]), the set ${\mathcal{F}}^q$ of ${\mathcal{F}}$-quasicentric subgroups (cf. Definition 4.5 and Lemma 4.6(d) in [@AKO]) and the set ${\mathcal{F}}^s$ of subcentric subgroups (cf. Definition 1 and Proposition 3.3 in [@subcentric]).
Morphisms of fusion systems {#SS:FusionMorphisms}
---------------------------
Throughout this subsection let ${\mathcal{F}}$ and ${{\widetilde{{\mathcal{F}}}}}$ be fusion systems over $S$ and ${{\widetilde{S}}}$ respectively.
We say that a group homomorphism $\alpha\colon S\longrightarrow {{\widetilde{S}}}$ *induces a morphism* from ${\mathcal{F}}$ to ${{\widetilde{{\mathcal{F}}}}}$ if, for each ${\varphi}\in{\mathrm{Hom}}_{\mathcal{F}}(P,Q)$, there exists $\psi\in{\mathrm{Hom}}_{{{\widetilde{{\mathcal{F}}}}}}(P\alpha,Q\alpha)$ such that $(\alpha|_P)\psi={\varphi}(\alpha|_Q)$.
Note that, for any ${\varphi}\in{\mathrm{Hom}}_{\mathcal{F}}(P,Q)$, a map $\psi\in{\mathrm{Hom}}_{{{\widetilde{{\mathcal{F}}}}}}(P\alpha,Q\alpha)$ as in the above definition is uniquely determined. So if $\alpha$ induces a morphism from ${\mathcal{F}}$ to ${{\widetilde{{\mathcal{F}}}}}$, then $\alpha$ induces a map $$\alpha_{P,Q}\colon{\mathrm{Hom}}_{\mathcal{F}}(P,Q)\longrightarrow {\mathrm{Hom}}_{{{\widetilde{{\mathcal{F}}}}}}(P\alpha,Q\alpha).$$ Together with the map $P\mapsto P\alpha$ from the set of objects of ${\mathcal{F}}$ to the set of objects of ${{\widetilde{{\mathcal{F}}}}}$ this gives a functor from ${\mathcal{F}}$ to ${{\widetilde{{\mathcal{F}}}}}$. Moreover, $\alpha$ together with the maps $\alpha_{P,Q}$ ($P,Q\leq S$) is a morphism of fusion systems in the sense of [@AKO Definition II.2.2]. We call $(\alpha,\alpha_{P,Q}\colon P,Q\leq S)$ the *morphism induced by $\alpha$*.
Suppose $\alpha\colon S\longrightarrow {{\widetilde{S}}}$ induces a morphism from ${\mathcal{F}}$ to ${{\widetilde{{\mathcal{F}}}}}$. We say that $\alpha$ *induces an epimorphism* from ${\mathcal{F}}$ to ${{\widetilde{{\mathcal{F}}}}}$ if the induced morphism $(\alpha,\alpha_{P,Q}\colon P,Q\leq S)$ is a surjective morphism of fusion systems. This means that $\alpha$ is surjective as a map $S\longrightarrow {{\widetilde{S}}}$ and ${\mathcal{F}}\alpha^*={{\widetilde{{\mathcal{F}}}}}$, i.e. for all $P,Q\leq S$ with $\ker(\alpha)\leq P\cap Q$, the map $\alpha_{P,Q}$ is surjective. If $\alpha$ is in addition injective, then we say that $\alpha$ *induces an isomorphism* from ${\mathcal{F}}$ to ${{\widetilde{{\mathcal{F}}}}}$. If $\alpha\in{\mathrm{Aut}}(S)$ and $\alpha$ induces a morphism from ${\mathcal{F}}$ to ${\mathcal{F}}$, then we say that $\alpha$ *induces an automorphism* of ${\mathcal{F}}$. We will write ${\mathrm{Aut}}({\mathcal{F}})$ for the set of automorphisms of $S$ which induce an automorphism of ${\mathcal{F}}$.
If $\alpha\colon S\longrightarrow {{\widetilde{S}}}$ is an isomorphism of groups, then it is easy to see that $\alpha$ induces an isomorphism from ${\mathcal{F}}$ to ${{\widetilde{{\mathcal{F}}}}}$ if and only if, for all $P,Q\leq S$ and every group homomorphism ${\varphi}\colon P\longrightarrow Q$, we have ${\varphi}\in{\mathrm{Hom}}_{\mathcal{F}}(P,Q)$ if and only if $\alpha^{-1}{\varphi}\alpha\in{\mathrm{Hom}}_{{{\widetilde{{\mathcal{F}}}}}}(P\alpha,Q\alpha)$; if so then the map $\alpha_{P,Q}$ as above is given by ${\varphi}\mapsto\alpha^{-1}{\varphi}\alpha$. It follows from this observation that, if $\alpha$ induces an isomorphism from ${\mathcal{F}}$ to ${{\widetilde{{\mathcal{F}}}}}$, then the inverse map $\alpha^{-1}$ induces an isomorphism from ${{\widetilde{{\mathcal{F}}}}}$ to ${\mathcal{F}}$.
\[L:FusionEpi\] Suppose $\alpha\colon S\longrightarrow {{\widetilde{S}}}$ induces an epimorphism from ${\mathcal{F}}$ to ${{\widetilde{{\mathcal{F}}}}}$. Let $\ker(\alpha)\leq R\leq S$. Then the following hold:
- We have $(R\alpha)^{{{\widetilde{{\mathcal{F}}}}}}=\{R_0\alpha\colon R_0\in R^{\mathcal{F}}\}$.
- The subgroup $R$ is fully normalized if and only if $R\alpha$ is fully normalized.
- The group homomorphism $\alpha|_{N_S(R)}\colon N_S(R)\longrightarrow N_{{{\widetilde{S}}}}(R\alpha)$ induces an epimorphism from $N_{\mathcal{F}}(R)$ to $N_{{{\widetilde{{\mathcal{F}}}}}}(R\alpha)$.
Property (a) is elementary to check, and property (b) follows from (a), since $N_S(R_0)\alpha=N_{{{\widetilde{S}}}}(R_0\alpha)$ has order $|N_S(R_0)|/|\ker(\alpha)|$ for all $R_0\in R^{\mathcal{F}}$.
For the proof of (c) let $P,Q\leq N_S(R)$ with $\ker(\alpha)\leq P\cap Q$, ${\varphi}\in{\mathrm{Hom}}_{\mathcal{F}}(P,Q)$ and $\psi={\varphi}\alpha_{P,Q}\in{\mathrm{Hom}}_{{{\widetilde{{\mathcal{F}}}}}}(P\alpha,Q\alpha)$. We have then $\alpha|_P\psi={\varphi}\alpha|_Q$. Moreover, if $R\leq P$, then $\ker(\alpha)\leq R{\varphi}$ as $\ker(\alpha)$ is strongly closed. Hence, we have $R\leq P\cap Q$ and $R{\varphi}=R$ if and only if $R\alpha\leq P\alpha\cap Q\alpha$ and $(R\alpha)\psi=(R{\varphi})\alpha=R\alpha$. This implies (c).
\[L:GroupEpiFusionEpi\] Let $\alpha\colon G\longrightarrow \tilde{G}$ be an epimorphism from a group $G$ to a group $\tilde{G}$. Let $S\in{\mathrm{Syl}}_p(G)$ and ${{\widetilde{S}}}=S\alpha\in{\mathrm{Syl}}_p(\tilde{G})$. Then $\alpha|_S$ induces an epimorphism from ${\mathcal{F}}_S(G)$ to ${\mathcal{F}}_{{{\widetilde{S}}}}(\tilde{G})$.
Let $P,Q$ be subgroups of $S$. If $g\in G$ with $P^g\leq Q\leq S$, then $(P\alpha)^{g\alpha}=P^g\alpha\leq Q\alpha\leq{{\widetilde{S}}}$ and $(\alpha|_P) (c_{g\alpha}|_{P\alpha})=(c_g|_P)(\alpha|_Q)$. So $\alpha|_S$ is fusion preserving and the corresponding morphism of fusion systems takes $c_g|_P$ to $c_{g\alpha}|_{P\alpha}$. To show that $\alpha|_S$ induces an epimorphism, assume now that $\ker(\alpha|_S)\leq P\cap Q$ and fix $h\in\tilde{G}$ with $(P\alpha)^h\leq Q\alpha$. Since $\alpha$ is an epimorphism, there exists $g\in G$ with $g\alpha=h$. We have then $P^g\alpha=(P\alpha)^h\leq Q\alpha$. As $\ker(\alpha|_S)=\ker(\alpha)\cap S\leq Q$, the group $Q$ is a Sylow $p$-subgroup of $\ker(\alpha)Q$, which is the preimage of $Q\alpha$ in $G$. Thus, by Sylow’s theorem, there exists $n\in\ker(\alpha)$ with $P^{gn}\leq Q$. Replacing $g$ by $gn$, we may assume that $P^g\leq Q$. As seen at the beginning, this means that $c_h|_{P\alpha}\in{\mathrm{Hom}}_{{\mathcal{F}}_{{{\widetilde{S}}}}(\tilde{G})}(P\alpha,Q\alpha)$ is the image of $c_g|_P\in{\mathrm{Hom}}_{{\mathcal{F}}_S(G)}(P,Q)$ under the morphism induced by $\alpha$.
Homomorphisms of partial groups. {#SS:LocalityHomomorphism}
--------------------------------
In this subsection, we will introduce natural notions of homomorphisms, projections, isomorphisms and automorphisms of partial groups and of localities. We state moreover a few simple results needed in the proofs of our main theorems.
\[N:PartialHomWordMap\] If $\L$ and ${{\widetilde{\mathcal{L}}}}$ are sets and $\alpha\colon\L\longrightarrow{{\widetilde{\mathcal{L}}}},f\mapsto f\alpha$ is a map, then we denote by $\alpha^*$ the induced map on words $${\bold{W}}(\L) \longrightarrow {\bold{W}}({{\widetilde{\mathcal{L}}}}),\quad w=(f_1,\dots,f_n)\mapsto w\alpha^*=(f_1\alpha,\dots,f_n\alpha).$$ If ${\mathbf{D}}\subseteq{\bold{W}}(\L)$, set ${\mathbf{D}}\alpha^*:=\{w\alpha^*\colon w\in{\mathbf{D}}\}$.
For the remainder of this subsection let $\L$ and ${{\widetilde{\mathcal{L}}}}$ be partial groups with products $\Pi\colon{\mathbf{D}}\longrightarrow\L$ and ${\widetilde{\Pi}}\colon{\widetilde{\mathbf{D}}}\longrightarrow{{\widetilde{\mathcal{L}}}}$ respectively.
\[D:PartialHom\] A map $\alpha\colon\L\longrightarrow{{\widetilde{\mathcal{L}}}}$ is called a *homomorphism of partial groups* if
1. ${\mathbf{D}}\alpha^* \subseteq {\widetilde{\mathbf{D}}}$; and
2. $\Pi(w)\alpha = {\widetilde{\Pi}}(w\alpha^*)$ for every $w \in {\mathbf{D}}$.
If moreover ${\mathbf{D}}\alpha^* = {\widetilde{\mathbf{D}}}$, then we say that $\alpha$ is a *projection* of partial groups. If $\alpha$ is injective and ${\mathbf{D}}\alpha^*={\widetilde{\mathbf{D}}}$, then $\alpha$ is called an *isomorphism*. The isomorphisms of partial groups from $\L$ to itself are called *automorphisms* and the set of these automorphisms is denoted by ${\mathrm{Aut}}(\L)$.
For any homomorphism $\alpha\colon\L\longrightarrow{{\widetilde{\mathcal{L}}}}$, we call $\ker(\alpha)=\{f\in\L\colon f\alpha={\operatorname{\mathbf{1}}}\}$ the *kernel* of $\alpha$.
Notice that every projection $\L\longrightarrow{{\widetilde{\mathcal{L}}}}$ is surjective, as ${\widetilde{\mathbf{D}}}$ contains all the words of length one. In particular, every isomorphism is a bijection. In fact, we have the following characterization of isomorphisms.
\[L:PartialIso\] A map $\alpha\colon \L\longrightarrow{{\widetilde{\mathcal{L}}}}$ is an isomorphism of partial groups if and only if $\alpha$ is bijective and $\alpha$ and $\alpha^{-1}$ are both homomorphisms of partial groups.
If $\alpha$ is bijective and $\alpha$ and $\alpha^{-1}$ are both homomorphisms of partial groups, then ${\mathbf{D}}\alpha^*\subseteq{\widetilde{\mathbf{D}}}$ and ${\widetilde{\mathbf{D}}}(\alpha^{-1})^*\subseteq {\mathbf{D}}$, with the latter inclusion implying ${\widetilde{\mathbf{D}}}\subseteq{\mathbf{D}}\alpha^*$. Thus, we get ${\mathbf{D}}\alpha^*={\widetilde{\mathbf{D}}}$. As $\alpha$ is an injective homomorphism of partial groups, this yields that $\alpha$ is an isomorphism of partial groups.
Assume now that $\alpha$ is an isomorphism of partial groups. Then $\alpha$ is a bijection. Moreover, ${\mathbf{D}}\alpha^*={\widetilde{\mathbf{D}}}$ and thus ${\widetilde{\mathbf{D}}}(\alpha^{-1})^*={\mathbf{D}}$. Given $w\in{\widetilde{\mathbf{D}}}$, it remains to show that ${\widetilde{\Pi}}(w)\alpha^{-1}=\Pi(w(\alpha^{-1})^*)$. Note that $w(\alpha^{-1})^*\in{\mathbf{D}}$ and thus, as $\alpha$ is a homomorphism of partial groups, $\Pi(w(\alpha^{-1})^*)\alpha={\widetilde{\Pi}}(w(\alpha^{-1})^*\alpha^*)={\widetilde{\Pi}}(w)$. This implies the required equality.
\[L:PartialProj\] Suppose $\alpha\colon \L\longrightarrow {{\widetilde{\mathcal{L}}}}$ is a homomorphism of partial groups. If $M$ is a subgroup of $\L$, then $M\alpha$ is a subgroup of ${{\widetilde{\mathcal{L}}}}$ and $\alpha$ restricts to a group homomorphism $M\longrightarrow M\alpha$.
If $w=(f_1,\dots,f_n)\in {\bold{W}}(M\alpha)$, then for $i=1,\dots,n$, there exists $g_i\in M$ such that $f_i=g_i\alpha$. It follows $u:=(g_1,\dots,g_n)\in{\bold{W}}(M)\subseteq{\mathbf{D}}$ and $w=u\alpha^*\in{\widetilde{\mathbf{D}}}$. Moreover, ${\widetilde{\Pi}}(w)={\widetilde{\Pi}}(u\alpha^*)=\Pi(u)\alpha\in M\alpha$ as $M$ is a subgroup. Hence, $M\alpha$ is a subgroup of ${{\widetilde{\mathcal{L}}}}$. The assertion follows since $(gh)\alpha=\Pi(v)\alpha={\widetilde{\Pi}}(v\alpha^*)=(g\alpha)(h\alpha)$ for every word $v=(g,h)\in{\bold{W}}(M)$ of length two.
We now turn attention to maps between localities.
\[D:LocalityHomomorphism\] Let $(\L, \Delta, S)$ and $({{\widetilde{\mathcal{L}}}}, {\widetilde{\Delta}},{{\widetilde{S}}})$ be localities and let $\alpha \colon \L \longrightarrow {{\widetilde{\mathcal{L}}}}$ be a projection of partial groups.
- For any set $\Gamma$ of subgroups of $\L$, set $$\Gamma\alpha:=\{P\alpha\colon P\in\Gamma\}.$$
- We say that $\alpha$ is a *projection of localities* from $(\L,\Delta,S)$ to $({{\widetilde{\mathcal{L}}}},{\widetilde{\Delta}},{{\widetilde{S}}})$ if $\Delta\alpha= {\widetilde{\Delta}}$.
- If $\alpha$ is a projection of localities which is injective (and thus an isomorphism of partial groups), then $\alpha$ is a called an *isomorphism* of localities. We write ${\mathrm{Iso}}((\L,\Delta,S),({{\widetilde{\mathcal{L}}}},{\widetilde{\Delta}},{{\widetilde{S}}}))$ for the set of isomorphisms from $(\L,\Delta,S)$ to $({{\widetilde{\mathcal{L}}}},{\widetilde{\Delta}},{{\widetilde{S}}})$ (which may be empty).
- Given a set $\Gamma$ of subgroups of $S$ and a set ${\widetilde{\Gamma}}$ of subgroups of ${{\widetilde{S}}}$, we write $${\mathrm{Iso}}((\L,\Delta,S),({{\widetilde{\mathcal{L}}}},{\widetilde{\Delta}},{{\widetilde{S}}}))_{\Gamma,{\widetilde{\Gamma}}}$$ for the set of isomorphisms $\alpha$ from $(\L,\Delta,S)$ to $({{\widetilde{\mathcal{L}}}},{\widetilde{\Delta}},{{\widetilde{S}}})$ with $\Gamma\alpha={\widetilde{\Gamma}}$.
- An isomorphism from $(\L,\Delta,S)$ to itself is called an *automorphism*. We write ${\mathrm{Aut}}(\L,\Delta,S)$ for the group of automorphisms of $(\L,\Delta,S)$. If $\Gamma$ is a set of subgroups of $S$, then ${\mathrm{Aut}}(\L,\Delta,S)_{\Gamma}$ denotes the set of automorphisms $\alpha$ of $(\L,\Delta,S)$ with $\Gamma\alpha=\Gamma$.
- An automorphism of $(\L,\Delta,S)$ is called *rigid*, if it restricts to the identity on $S$.
If $\alpha$ is a projection of localities from $(\L,\Delta,S)$ to $({{\widetilde{\mathcal{L}}}},{\widetilde{\Delta}},{{\widetilde{S}}})$, then notice that $\alpha$ maps $S$ to ${{\widetilde{S}}}$, as $S$ and ${{\widetilde{S}}}$ are the unique maximal elements of $\Delta$ and ${\widetilde{\Delta}}$ respectively. In particular, ${\mathrm{Aut}}(\L,\Delta,S)$ acts on $S$ for every locality $(\L,\Delta,S)$.
\[L:ProjectionLocalityMorphismFusionSystem\] Suppose $\alpha\colon\L\longrightarrow{{\widetilde{\mathcal{L}}}}$ is a projection from a locality $(\L,\Delta,S)$ to a locality $({{\widetilde{\mathcal{L}}}},{\widetilde{\Delta}},{{\widetilde{S}}})$. Then the following hold:
- We have $N_\L(R)\alpha=N_{{{\widetilde{\mathcal{L}}}}}(R\alpha)$ for every $R\leq S$ with $S\cap\ker(\alpha)\leq R$.
- The map $\alpha|_S\colon S\longrightarrow{{\widetilde{S}}}$ induces an epimorphism of fusion systems from ${\mathcal{F}}_S(\L)$ to ${\mathcal{F}}_{{{\widetilde{S}}}}({{\widetilde{\mathcal{L}}}})$.
- If $\alpha$ is an isomorphism, then ${{\widetilde{S}}}_{f\alpha}=S_f\alpha$ for every $f\in\L$.
For the proof of (a) let $T:=S\cap\ker(\alpha)\leq R\leq S$. By [@loc1 Lemma 3.1(a)], $T$ is strongly closed in ${\mathcal{F}}_S(\L)$. Clearly, $N_\L(R)\alpha\subseteq N_{{{\widetilde{\mathcal{L}}}}}(R\alpha)$. Let ${{\tilde{f}}}\in N_{{{\widetilde{\mathcal{L}}}}}(R\alpha)$ and write $P$ for the full preimage of ${{\widetilde{S}}}_{{{\tilde{f}}}}$ in $S$. Then $T\leq R\leq P$ and ${{\tilde{f}}}\in N_{{{\widetilde{\mathcal{L}}}}}(P\alpha,{{\widetilde{S}}})$. Hence, by [@loc1 Theorem 4.3(c)], we may choose $f\in N_\L(P,S)$ with $f\alpha={{\tilde{f}}}$. Then $R^f\leq S$ and $R^f\alpha=(R\alpha)^{{{\tilde{f}}}}=R\alpha$. So $R^f=R$ as $T=T^f\leq R\cap R^f$. Hence, we have shown that $f\in N_\L(R)$ and thus that $N_{\ov{\L}}(\ov{R})\subseteq N_\L(R)\alpha$. This proves (a).
The fusion system ${\mathcal{F}}_S(\L)$ is generated by maps of the form $c_f\colon P\longrightarrow Q$, where $P,Q\in\Delta$ and $f\in N_\L(P,Q)$. Similarly, ${\mathcal{F}}_{{{\widetilde{S}}}}({{\widetilde{\mathcal{L}}}})$ is generated by maps of the form $c_{{{\tilde{f}}}}\colon P\alpha\longrightarrow Q\alpha$ where $P,Q\in\Delta$ and ${{\tilde{f}}}\in N_{{{\widetilde{\mathcal{L}}}}}(P\alpha,Q\alpha)$. Fixing $P,Q\in\Delta$, by [@loc1 Theorem 4.3(c)], $\alpha$ induces a surjection $N_\L(P,Q)\longrightarrow N_{{{\widetilde{\mathcal{L}}}}}(P\alpha,Q\alpha)$. Moreover, if $f\in N_\L(P,Q)$, then $(c_f|_P)(\alpha|_Q)=(\alpha|_P)(c_{f\alpha}|_{P\alpha})$. This implies (b).
For the proof of (c) let $f\in\L$ be arbitrary and suppose $\alpha$ is an isomorphism. Using that $\alpha$ maps $S$ isomorphically to ${{\widetilde{S}}}$ and that $(f^{-1})\alpha=(f\alpha)^{-1}$ by [@loc1 Lemma 1.13], we see $$\begin{aligned}
S_f\alpha&=&\{s\alpha\colon s\in S,\;(f^{-1},s,f)\in{\mathbf{D}},\;s^f\in S\}\\
&=&\{s\alpha\colon s\in S,\;((f\alpha)^{-1},s\alpha,f\alpha)\in{\widetilde{\mathbf{D}}},\;(s\alpha)^{f\alpha}\in {{\widetilde{S}}}\}\\
&=&\{t\in {{\widetilde{S}}}\colon ((f\alpha)^{-1},t,f\alpha)\in{\widetilde{\mathbf{D}}},\;t^{f\alpha}\in {{\widetilde{S}}}\}\\
&=& ({{\widetilde{S}}})_{f\alpha}.\end{aligned}$$
Restrictions of localities {#SS:Restrictions}
--------------------------
\[D:RestrictionLocality\] Let $(\L^+,\Delta^+,S)$ be a locality with partial product $\Pi^+\colon{\mathbf{D}}^+\longrightarrow\L^+$, and let $\Delta\subseteq \Delta^+$ be closed with respect to taking $\L$-conjugates and overgroups in $S$. Suppose $\Delta$ is non-empty. Then we set $$\L^+|_{\Delta}:=\{f\in\L^+\colon S_f\in\Delta\}.$$ Note that ${\mathbf{D}}:={\mathbf{D}}_{\Delta}(\L^+,\Pi^+)\subseteq{\mathbf{D}}^+\cap{\bold{W}}(\L^+|_{\Delta})$ and, by Lemma \[L:LocalitiesProp\](c), $\Pi^+(w)\in\L|_{\Delta}$ for all $w\in{\mathbf{D}}$. We call $\L:=\L^+|_{\Delta}$ together with the partial product $\Pi^+|_{{\mathbf{D}}}\colon{\mathbf{D}}\longrightarrow \L$ and the restriction of the inversion map on $\L^+$ to $\L$ the *restriction* of $\L^+$ to $\Delta$.
The properties of the restriction $\L^+|_\Delta$ are summarized in the following lemma, which we will use throughout, most of the time without reference.
\[L:RestrictionProp\] Let $(\L^+,\Delta^+,S)$ be a locality with partial product $\Pi^+\colon{\mathbf{D}}^+\longrightarrow\L^+$, and let $\Delta\subseteq \Delta^+$ be non-empty and closed with respect to taking $\L^+$-conjugates and overgroups in $S$. Set $\L:=\L^+|_\Delta$, ${\mathbf{D}}:={\mathbf{D}}_\Delta(\L^+,\Pi^+)$ and $\Pi:=\Pi^+|_{\mathbf{D}}$.
- $\L$ together with $\Pi\colon{\mathbf{D}}\longrightarrow \L$ and the restriction of the inversion map on $\L^+$ to $\L$ forms a partial group.
- If $f\in\L$, then it does not matter whether we form $S_f$ inside of $\L$ or of $\L^+$, i.e. using the notation introduced in Definition \[D:Sf\], we have $S_f^\L=S_f^{\L^+}$.
- The triple $(\L,\Delta,S)$ is a locality.
Part (a) is straightforward to check. Let $f\in\L$. As ${\mathbf{D}}\subseteq{\mathbf{D}}^+$ and $\Pi=\Pi^+|_{\mathbf{D}}$, we have $S_f^\L\subseteq S_f^{\L^+}$. Setting $P:=S_f^{\L^+}$, by definition of $\L|_\Delta$, we have $P\in\Delta$. Moreover, the conjugate $P^f$ is defined in $\L^+$ and an element of $\Delta$, as $\Delta$ is closed under taking $\L$-conjugates. Now for $a\in P$, we have $(f^{-1},a,f)\in{\mathbf{D}}={\mathbf{D}}_\Delta(\L^+,\Pi^+)$ via $P^f,P,P,P^f$. Hence, $P^f$ is defined in $\L$, which implies $P\subseteq S_f^{\L}$. This shows (b).
The proof of (c) is given in [@Ch Lemma 2.21(a)], but we repeat the argument here in detail, since we feel that there is a small gap in the proof: Note that $S\in\Delta$ and so ${\bold{W}}(S)\subseteq {\bold{W}}(N_\L(S))\subseteq {\mathbf{D}}={\mathbf{D}}_\Delta(\L^+,\Pi^+)$. Hence, $S$ is a $p$-subgroup of $\L$. As ${\mathbf{D}}\subseteq{\mathbf{D}}^+$ and $\Pi=\Pi^+|_{\mathbf{D}}$, every $p$-subgroup of $\L$ is also a $p$-subgroup of $\L^+$. Therefore, $S$ is a maximal $p$-subgroup of $\L$, since it is a maximal $p$-subgroup of $\L^+$. By assumption, $\Delta$ is closed under taking $\L^+$-conjugates and overgroups in $S$, so it is in particular closed under taking $\L$-conjugates in $S$. Thus, it remains to show that ${\mathbf{D}}_\Delta(\L,\Pi)={\mathbf{D}}$. Clearly, ${\mathbf{D}}_\Delta(\L,\Pi)\subseteq{\mathbf{D}}:={\mathbf{D}}_\Delta(\L^+,\Pi^+)$. If $w=(f_1,\dots,f_n)\in{\mathbf{D}}:={\mathbf{D}}_\Delta(\L^+,\Pi^+)$ via $P_0,\dots,P_n\in\Delta$, this means that the conjugate $P_{i-1}^{f_i}$ is defined in $\L^+$ and equal to $P_i$ for $i=1,\dots,n$. Then $P_{i-1}\subseteq S_{f_i}^{\L^+}=S_{f_i}^\L$ by (b). Hence, $w\in{\mathbf{D}}_\Delta(\L,\Pi)$ via $P_0,P_1,\dots,P_n$. This proves (c).
\[L:IsoRestrictionLocality\] Let $(\L^+,\Delta^+,S)$ and $({{\widetilde{\mathcal{L}}}}^+,{\widetilde{\Delta}}^+,{{\widetilde{S}}})$ be localities. Let $\emptyset\neq\Delta\subseteq\Delta^+$ and $\emptyset\neq{\widetilde{\Delta}}\subseteq{\widetilde{\Delta}}^+$ such that $\Delta$ is closed under taking $\L^+$-conjugates and overgroups in $S$ and ${\widetilde{\Delta}}$ is closed under taking ${{\widetilde{\mathcal{L}}}}$-conjugates and overgroups in ${{\widetilde{S}}}$. Set $\L:=\L^+|_{\Delta}$ and ${{\widetilde{\mathcal{L}}}}:={{\widetilde{\mathcal{L}}}}^+|_{{\widetilde{\Delta}}}$. Then for every $\gamma\in{\mathrm{Iso}}((\L^+,\Delta^+,S),({{\widetilde{\mathcal{L}}}}^+,{\widetilde{\Delta}}^+,{{\widetilde{S}}}))_{\Delta,{\widetilde{\Delta}}}$ we have $\gamma|_{\L}\in{\mathrm{Iso}}((\L,\Delta,S),({{\widetilde{\mathcal{L}}}},{\widetilde{\Delta}},{{\widetilde{S}}}))_{\Delta^+,{\widetilde{\Delta}}^+}$.
If $f\in\L^+$, then by Lemma \[L:ProjectionLocalityMorphismFusionSystem\](c), we have ${{\widetilde{S}}}_{f\gamma}=S_f\gamma$. As $\Delta\gamma={\widetilde{\Delta}}$ and $\gamma$ is bijective, this means $S_f\in\Delta$ if and only if ${{\widetilde{S}}}_{f\gamma}\in{\widetilde{\Delta}}$. Hence, $f\in\L$ if and only if $f\gamma\in{{\widetilde{\mathcal{L}}}}$, i.e. $\gamma|_{\L}\colon\L\longrightarrow{{\widetilde{\mathcal{L}}}}$ is well-defined and surjective. Clearly, $\gamma|_\L$ is injective.
Write $\Pi\colon{\mathbf{D}}\longrightarrow \L$ and ${\widetilde{\Pi}}\colon {\widetilde{\mathbf{D}}}\rightarrow{{\widetilde{\mathcal{L}}}}$ for the products on $\L$ and ${{\widetilde{\mathcal{L}}}}$ respectively. Let $w=(f_1,\dots,f_n)\in{\mathbf{D}}$ via $P_0,\dots,P_n\in\Delta$, i.e. $P_{i-1}\leq S_{f_i}$ and $P_{i-1}^{f_i}=P_i$ for $i=1,\dots,n$. Then $P_{i-1}\gamma\leq (S_{f_i})\gamma={{\widetilde{S}}}_{f_i\gamma}$ and, as $\gamma$ is a homomorphism of partial groups, $(P_{i-1}\gamma)^{f_i\gamma}=(P_{i-1}^{f_i})\gamma=P_i\gamma$. Since $\Delta\gamma={\widetilde{\Delta}}$, this shows that $w\gamma^*=(f_1\gamma,\dots,f_n\gamma)\in{\widetilde{\mathbf{D}}}$ via $P_0\gamma,\dots,P_n\gamma\in{\widetilde{\Delta}}$. Hence, ${\mathbf{D}}\gamma^*\subseteq{\widetilde{\mathbf{D}}}$. As $\gamma^{-1}$ is an isomorphism from $({{\widetilde{\mathcal{L}}}}^+,{\widetilde{\Delta}}^+,S)$ to $(\L^+,\Delta^+,S)$ by Lemma \[L:PartialIso\], a symmetric argument shows that ${\widetilde{\mathbf{D}}}(\gamma^{-1})^*\subseteq{\mathbf{D}}$ and thus ${\widetilde{\mathbf{D}}}\subseteq{\mathbf{D}}\gamma^*$. This proves ${\mathbf{D}}\gamma^*={\widetilde{\mathbf{D}}}$. As $\gamma\colon\L^+\longrightarrow{{\widetilde{\mathcal{L}}}}^+$ is a homomorphism of partial groups and since $\Pi$ and ${\widetilde{\Pi}}$ are restrictions of the products on $\L^+$ and ${{\widetilde{\mathcal{L}}}}^+$ respectively, we have ${\widetilde{\Pi}}(w\gamma^*)=\Pi(w)\gamma$ for all $w\in{\mathbf{D}}$. So $\gamma|_{\L}$ is an isomorphism of partial groups from $\L$ to ${{\widetilde{\mathcal{L}}}}$ and the assertion follows.
Linking localities {#SS:LinkingLoc}
------------------
\[D:LinkingLoc\]
- A finite group $G$ is said to be of *characteristic $p$* of $C_G(O_p(G))\leq O_p(G)$.
- A locality $(\L,\Delta,S)$ is called a *linking locality* if ${\mathcal{F}}_S(\L)$ is saturated, $N_\L(P)$ is of characteristic $p$ for every $P \in \Delta$, and ${\mathcal{F}}_S(\L)^{cr}\subseteq \Delta$.
- If ${\mathcal{F}}$ is a saturated fusion system over a $p$-group $S$, then a subgroup $P\leq S$ is called *${\mathcal{F}}$-subcentric* if $N_{\mathcal{F}}(Q)$ is constrained for every fully ${\mathcal{F}}$-normalized ${\mathcal{F}}$-conjugate $Q$ of $P$. Write ${\mathcal{F}}^s$ for the set of ${\mathcal{F}}$-subcentric subgroups of $S$.
- A *subcentric linking locality* is a linking locality $(\L,\Delta,S)$ such that $\Delta={\mathcal{F}}_S(\L)^s$.
Linking localities are closely related to linking systems. We provide some more details on that in Subsection \[SS:LinkingSystems\]. Given a saturated fusion system ${\mathcal{F}}$, it is elementary to show that the object set $\Delta$ of a linking locality over ${\mathcal{F}}$ is always contained in ${\mathcal{F}}^s$. On the other hand, using the existence and uniqueness of centric linking systems, it is shown in [@subcentric Theorem A] that, for every ${\mathcal{F}}$-closed set $\Delta$ with ${\mathcal{F}}^{cr}\subseteq\Delta\subseteq{\mathcal{F}}^s$, there exists a linking locality $(\L,\Delta,S)$ over ${\mathcal{F}}$ which is unique up to rigid isomorphism. Moreover, it is proved that the set ${\mathcal{F}}^s$ is ${\mathcal{F}}$-closed and thus there exists a subcentric linking locality over ${\mathcal{F}}$ which is unique up to rigid isomorphism.
We will need the following slightly technical lemma.
\[L:LinkingLocN(R)\] Suppose $(\L,\Delta,S)$ and $(\L^+,\Delta^+,S)$ are linking localities over the same fusion system ${\mathcal{F}}$ such that $\Delta\subseteq\Delta^+$ and $\L=\L^+|_\Delta$. Let $R\in\Delta^+\backslash\Delta$ such that $R$ is fully normalized and every proper overgroup of $R$ is in $\Delta$. Then $N_\L(R)=N_{\L^+}(R)$ is a subgroup of $R$. Moreover, $R^*=O_p(N_{\L^+}(R))\in\Delta$ and $N_\L(R)=N_{N_\L(R^*)}(R)$.
As $R\not\in\Delta$ and ${\mathcal{F}}^{cr}\subseteq\Delta$, we have $R\not\in{\mathcal{F}}^{cr}$. By [@subcentric Lemma 6.2], this implies $R<R^*:=O_p(N_{\L^+}(R))$ and so $R^*\in\Delta$. Hence, using Lemma \[L:RestrictionProp\](b), we see that $N_{\L^+}(R)\subseteq N_{\L^+}(R^*)=N_\L(R^*)$ and $N_{\L^+}(R)=N_{N_{\L^+}(R^*)}(R)=N_{N_\L(R^*)}(R)=N_\L(R)$ is a subgroup of $\L$.
A crucial lemma
===============
In this section we prove the following lemma, on which the proofs of Theorem \[T:A1\] and Theorem \[T:MainChermakII\] will be based on. It is also used in [@normal] to show that there is a one-to-one correspondence between the normal subsystems of a saturated fusion system and the partial normal subgroups of an associated linking locality.
\[L:Main\] Let $(\L^+,\Delta^+,S)$ and $({{\widetilde{\mathcal{L}}}},{\widetilde{\Delta}},{{\widetilde{S}}})$ be localities, and let $\Delta$ be a subset of $\Delta^+$ which is ${\mathcal{F}}_S(\L^+)$-closed. Set $\L:=\L^+|_\Delta$. Assume that every proper overgroup of an element of $\Delta^+\backslash \Delta$ is in $\Delta$, or assume more generally $N_S(P)\in\Delta$ for every $P\in\Delta^+\backslash\Delta$. Suppose we are given
- a homomorphism of partial groups $\alpha\colon \L\longrightarrow {{\widetilde{\mathcal{L}}}}$ with $\Delta^+\alpha\subseteq{\widetilde{\Delta}}$;
- a set $\Gamma_0\subseteq\Delta^+\backslash\Delta$ of fully ${\mathcal{F}}_S(\L^+)$-normalized representatives of the ${\mathcal{F}}_S(\L^+)$-conjugacy classes of the subgroups in $\Delta^+\backslash\Delta$; and
- for each $Q\in\Gamma_0$ a homomorphism of groups $\alpha_Q\colon N_{\L^+}(Q)\longrightarrow N_{{{\widetilde{\mathcal{L}}}}}(Q\alpha)$ with $\alpha_Q|_{N_{\L}(Q)}=\alpha|_{N_\L(Q)}$.
Then there exists a unique homomorphism of partial group $\gamma\colon \L^+\longrightarrow {{\widetilde{\mathcal{L}}}}$ with $\gamma|_\L=\alpha$ and $\gamma|_{N_{\L^+}(Q)}=\alpha_Q$ for every $Q\in\Gamma_0$.
Set ${\mathcal{F}}:={\mathcal{F}}_S(\L^+)$. Write $\Pi^+\colon{\mathbf{D}}^+\longrightarrow \L^+$, $\Pi\colon{\mathbf{D}}\longrightarrow\L$ and ${\widetilde{\Pi}}\colon{\widetilde{\mathbf{D}}}\longrightarrow{{\widetilde{\mathcal{L}}}}$ for the partial products on $\L^+$, $\L$ and ${{\widetilde{\mathcal{L}}}}$ respectively. Recall from Definition \[D:RestrictionLocality\] that then ${\mathbf{D}}\subseteq{\mathbf{D}}^+$ and $\Pi=\Pi^+|_{\mathbf{D}}$. As $\Gamma_0$ is a set of representatives of the ${\mathcal{F}}$-conjugacy classes of subgroups in $\Delta^+\backslash\Delta$, for every $P\in\Delta^+\backslash \Delta$, there is a unique element of $\Gamma_0\cap P^{\mathcal{F}}$, which we denote by $Q_P$. By Lemma \[L:LocalityoverFFnatural\](b), for every $P\in\Delta^+\backslash\Delta$, we may moreover pick $h_P\in N_{\L^+}(N_S(P),S)$ with $P^{h_P}=Q_P$. As $S\in\Delta$, we have $P\neq S$ and thus $P<N_S(P)$. So by assumption, in any case $N_S(P)\in\Delta$ and thus $h_P\in N_\L(N_S(P),S)$. Moreover, the conjugate $P^{h_P}$ is defined in $\L$.
We define now first a map $\gamma\colon\L^+\longrightarrow {{\widetilde{\mathcal{L}}}}$ and show then that it has the required properties. If $f\in\L$, then we set $f\gamma=f\alpha$. Suppose now that $f\in\L^+\backslash\L$ so that $P:=S_f\in\Delta^+\backslash\Delta$. Notice that $P$ and $P^f$ are ${\mathcal{F}}$-conjugate and thus $Q:=Q_P=Q_{P^f}$. As $u_f:=(h_P,h_P^{-1},f,h_{P^f},h_{P^f}^{-1})\in{\mathbf{D}}^+$ via $P,Q,P,P^f,Q,P^f$, we have $f=\Pi^+(u_f)=\Pi^+(h_P,g,h_{P^f}^{-1})$, where $g:=\Pi^+(h_P^{-1},f,h_{P^f})\in N_{\L^+}(Q)$. Observe that $P^{h_P}=Q$ and $Q^{h_{P^f}^{-1}}=P^f$ in $\L$. Moreover, $$g\alpha_Q\in N_{\L^+}(Q)\alpha_Q\subseteq N_{{{\widetilde{\mathcal{L}}}}}(Q\alpha).$$ Since $\Delta^+\alpha\subseteq{\widetilde{\Delta}}$ and $\alpha$ is a homomorphism of partial groups, we conclude that $$(h_P\alpha,g\alpha_Q,h_{P^f}^{-1}\alpha)\in {\widetilde{\mathbf{D}}}\mbox{ via }P\alpha,Q\alpha,Q\alpha,P^f\alpha.$$ So for every $f\in\L^+\backslash\L$, setting $P:=S_f$ and $Q:=Q_P$, we may define $f\gamma$ via $$\label{fgamma}
f\gamma={\widetilde{\Pi}}(h_P\alpha,g\alpha_{Q},h_{P^f}^{-1}\alpha)\mbox{ where }g=\Pi^+(h_P^{-1},f,h_{P^f})\in N_{\L^+}(Q).$$ If $\beta\colon\L^+\longrightarrow {{\widetilde{\mathcal{L}}}}$ is a homomorphism of partial groups with $\beta|_\L=\alpha$ and $\beta|_{N_{\L^+}(Q)}=\alpha_{Q}$ for all $Q\in\Gamma_0$, then with $f$, $g$, $P$ and $Q$ as above, we see that $$f\beta=\Pi^+(h_P,g,h_{P^f}^{-1})\beta={\widetilde{\Pi}}(h_P\beta,g\beta,h_{P^f}^{-1}\beta)={\widetilde{\Pi}}(h_P\alpha,g\alpha_{Q},h_{P^f}^{-1}\alpha)=f\gamma.$$ Hence, we have in this case $\beta=\gamma$. So to prove (a), it is sufficient to show that $\gamma$ is a homomorphism of partial groups with $\gamma|_{N_{\L^+}(Q)}=\alpha_Q$ for all $Q\in\Gamma_0$.
*Step 1:* Given $f\in\L^+$ and $P\in\Delta^+\backslash\Delta$ with $P\leq S_f$, we show that holds for $Q:=Q_P$. If $P=S_f$, then $f\not\in\L$ and the equation holds by the definition of $\gamma$. So assume now $P<S_f$ and thus $P<R:=N_{S_f}(P)\in\Delta$. Notice that $R\leq N_S(P)\leq S_{h_P}$ and $R^f\leq N_S(P^f)\leq S_{h_{P^f}}$. Hence, $(h_P,h_P^{-1},f,h_{P^f},h_{P^f}^{-1})\in{\mathbf{D}}$ via $R$, and so $f=\Pi(h_P,h_P^{-1},f,h_{P^f},h_{P^f}^{-1})=\Pi(h_P,g,h_{P^f}^{-1})$, where $g:=\Pi(h_P^{-1},f,h_{P^f})=\Pi^+(h_P^{-1},f,h_{P^f})\in N_\L(Q)$. As $\alpha$ is a homomorphism of partial groups with $\alpha|_{N_{\L}(Q)}=\alpha_Q|_{N_\L(Q)}$, it follows $f\gamma=f\alpha={\widetilde{\Pi}}(h_P\alpha,g\alpha,h_{P^f}^{-1}\alpha)={\widetilde{\Pi}}(h_P\alpha,g\alpha_Q,h_{P^f}^{-1}\alpha)$. So holds.
*Step 2:* We show that $\gamma|_{N_{\L^+}(Q)}=\alpha_Q$ for every $Q\in\Gamma_0$. To prove this fix $Q\in\Gamma_0$ and $f\in N_{\L^+}(Q)$. Observe that $h_Q\in N_{\L^+}(Q)$ and $h_Q\alpha_Q\in N_{{{\widetilde{\mathcal{L}}}}}(Q\alpha)$. By Step 1, we have $f\gamma={\widetilde{\Pi}}(h_Q\alpha,g\alpha_{Q},h_Q^{-1}\alpha)$ where $g=\Pi^+(h_Q^{-1},f,h_Q)\in N_{\L^+}(Q)$. Moreover, $u:=(h_Q\alpha,h_Q^{-1}\alpha,f\alpha_Q,h_Q\alpha,h_Q^{-1}\alpha)\in{\mathbf{D}}^+$ via $Q\alpha$. So $$f\alpha_Q={\widetilde{\Pi}}(u)={\widetilde{\Pi}}(h_Q\alpha,{\widetilde{\Pi}}(h_Q^{-1}\alpha,f\alpha_Q,h_Q\alpha),h_Q^{-1}\alpha)={\widetilde{\Pi}}(h_Q\alpha,g\alpha_Q,h_Q^{-1}\alpha)=f\gamma,$$ where the third equality uses that $\alpha_Q$ is a homomorphism of groups with $h_Q\alpha_Q=h_Q\alpha$ and $h_Q^{-1}\alpha_Q=h_Q^{-1}\alpha$, and the last equality uses Step 1.
*Step 3:* We show that $\gamma$ is a homomorphism of partial groups and thus the assertion holds by Step 2. For the proof let $w=(f_1,\dots,f_n)\in{\mathbf{D}}^+$. If $w\in{\mathbf{D}}$, then $\Pi^+(w)\gamma=\Pi(w)\alpha={\widetilde{\Pi}}(w\alpha^*)={\widetilde{\Pi}}(w\gamma^*)$ as $\gamma|_\L=\alpha$ is assumed to be a homomorphism of partial groups. Thus, we may assume $w\not\in{\mathbf{D}}$. Then $w\in{\mathbf{D}}^+$ via $P_0,P_1,\dots,P_n\in\Delta^+\backslash\Delta$. Notice that $P_0,P_1,\dots,P_n$ are all ${\mathcal{F}}$-conjugate and so $Q:=Q_{P_0}=Q_{P_i}$ for $i=1,\dots,n$. Set $h_i:=h_{P_i}$ for $i=0,1,\dots,n$. By Step 1, we have $$\label{E:fi}
f_i\gamma={\widetilde{\Pi}}(h_{i-1}\alpha,g_i\alpha_Q,h_i^{-1}\alpha)\mbox{ where }g_i=\Pi^+(h_{i-1}^{-1},f_i,h_i)\in N_{\L^+}(Q)$$ for $i=1,\dots,n$. Set $$u:=(h_0,g_1,h_1^{-1},h_1,g_2,h_2^{-1},\dots,h_{n-1},g_n,h_n^{-1})\mbox{ and }g:=\Pi^+(g_1,g_2,\dots,g_n).$$ Using Step 2, we see that $u\gamma^*=(h_0\alpha,g_1\alpha_Q,h_1^{-1}\alpha,h_1\alpha,g_2\alpha_Q,h_2^{-1}\alpha,\dots,h_{n-1}\alpha,g_n\alpha_Q,h_n^{-1}\alpha)$. Notice that $u\in{\mathbf{D}}^+$ via $P_0,Q,Q,P_1,\dots,P_{n-1},Q,Q,P_n$. Similarly, as $\alpha$ is a homomorphism of partial groups and $g_i\alpha_Q\in N_{{\widetilde{\mathcal{L}}}}(Q\alpha)$ for $i=1,\dots,n$, we have $u\gamma^*\in{\widetilde{\mathbf{D}}}$ via $P_0\alpha,Q\alpha,Q\alpha,P_1\alpha,\dots,P_{n-1}\alpha,Q\alpha,Q\alpha,P_n\alpha$. Using and applying axiom (PG3) of a partial group and Lemma \[L:PartialGroup\](b) several times, we get that $w\gamma^*=(f_1\gamma,\dots,f_n\gamma)\in{\widetilde{\mathbf{D}}}$ and $$\begin{aligned}
{\widetilde{\Pi}}(w\gamma^*)&=&{\widetilde{\Pi}}(u\gamma^*)\\
&=&{\widetilde{\Pi}}(h_0\alpha,g_1\alpha_Q,\dots,g_n\alpha_Q,h_n^{-1}\alpha)\\
&=&{\widetilde{\Pi}}(h_0\alpha,{\widetilde{\Pi}}(g_1\alpha_Q,\dots,g_n\alpha_Q),h_n^{-1}\alpha)\\
&=&{\widetilde{\Pi}}(h_0\alpha,g\alpha_Q,h_n^{-1}\alpha).\end{aligned}$$ Observe also that $f_i=\Pi^+(h_{i-1},h_{i-1}^{-1},f_i,h_i,h_i^{-1})=\Pi^+(h_{i-1},g_i,h_i^{-1})$ for $i=1,\dots,n$. So similarly, again using axiom (PG3) and Lemma \[L:PartialGroup\](b) repeatedly, we see that $$f:=\Pi^+(w)=\Pi^+(u)=\Pi^+(h_0,g_1,\dots,g_n,h_n^{-1})=\Pi^+(h_0,g,h_n^{-1})\in N_{\L^+}(P_0,P_n).$$ As $g=\Pi^+(h_0^{-1},h_0,g,h_n^{-1},h_n)=\Pi^+(h_0^{-1},f,h_n)$, it follows from Step 1 that $$f\gamma={\widetilde{\Pi}}(h_0\alpha,g\alpha_Q,h_n^{-1}\alpha).$$ Putting everything together, we get $\Pi^+(w)\gamma=f\gamma={\widetilde{\Pi}}(w\gamma^*)$ and thus $\gamma$ is a homomorphism of partial groups. This completes Step 3 and the proof of the assertion.
In the proofs of our main theorems, Lemma \[L:Main\] will be used in the form of the following two corollaries. Corollary \[C:MainCor\] is actually a special case of Corollary \[C:MainCor2\]. We formulate it separately, since it is shorter to prove and is all we need in the proof of Theorem \[T:MainChermakII\].
\[C:MainCor\] Let $(\L^+,\Delta^+,S)$ and $(\L,\Delta,S)$ be linking localities such that ${\mathcal{F}}_S(\L)={\mathcal{F}}_S(\L^+)$, $\Delta\subseteq\Delta^+$, $\L=\L^+|_\Delta$, and every proper overgroup of an element of $\Delta^+\backslash \Delta$ is in $\Delta$. Suppose we are moreover given a locality $({{\widetilde{\mathcal{L}}}},{\widetilde{\Delta}},{{\widetilde{S}}})$ and a homomorphism of partial groups $\alpha\colon \L\longrightarrow {{\widetilde{\mathcal{L}}}}$ with $\Delta^+\alpha\subseteq{\widetilde{\Delta}}$. Then there exists a unique homomorphism of partial groups $\gamma\colon \L^+\longrightarrow {{\widetilde{\mathcal{L}}}}$ with $\gamma|_\L=\alpha$.
Notice that $\Delta^+\backslash\Delta$ is closed under ${\mathcal{F}}_S(\L^+)$-conjugacy, as $\Delta$ and $\Delta^+$ are closed under ${\mathcal{F}}_S(\L^+)$-conjugacy. Thus, we may choose a set $\Gamma_0\subseteq\Delta^+\backslash\Delta$ of fully ${\mathcal{F}}_S(\L^+)$-normalized representatives of the ${\mathcal{F}}_S(\L^+)$-conjugacy classes of the elements in $\Delta^+\backslash\Delta$. By Lemma \[L:LinkingLocN(R)\], for every $Q\in\Gamma_0$, the normalizer $N_\L(Q)=N_{\L^+}(Q)$ is a subgroup of $\L$. As $\alpha$ is a homomorphism of partial groups, we have $N_\L(Q)\alpha\subseteq N_{{{\widetilde{\mathcal{L}}}}}(Q\alpha)$. Since we assume that $Q\alpha\in\Delta^+\alpha\subseteq{\widetilde{\Delta}}$, the normalizer $N_{{{\widetilde{\mathcal{L}}}}}(Q\alpha)$ is a subgroup of ${{\widetilde{\mathcal{L}}}}$. So $\alpha_Q:=\alpha|_{N_\L(Q)}$ is a homomorphism of groups from $N_\L(Q)=N_{\L^+}(Q)$ to $N_{{{\widetilde{\mathcal{L}}}}}(Q\alpha)$. Now the assertion follows from Lemma \[L:Main\].
\[C:MainCor2\] Let $(\L^+,\Delta^+,S)$ and $(\L,\Delta,S)$ be linking localities over the same fusion system ${\mathcal{F}}$ such that $\Delta\subseteq\Delta^+$ and $\L=\L^+|_\Delta$. Suppose we are given a locality $({{\widetilde{\mathcal{L}}}},{\widetilde{\Delta}},{{\widetilde{S}}})$ and a homomorphism of partial groups $\alpha\colon \L\longrightarrow {{\widetilde{\mathcal{L}}}}$ with $\Delta^+\alpha\subseteq{\widetilde{\Delta}}$. Then there exists a unique homomorphism of partial groups $\gamma\colon \L^+\longrightarrow {{\widetilde{\mathcal{L}}}}$ with $\gamma|_\L=\alpha$.
Let $\Gamma$ be the set of the elements in $\Delta^+\backslash\Delta$ of maximal order. Then $\Gamma$ is closed under ${\mathcal{F}}$-conjugacy, since $\Delta$ and $\Delta^+$ are closed under ${\mathcal{F}}$-conjugacy. Moreover, as $\Delta^+$ is overgroup closed in $S$, every proper overgroup of an element of $\Gamma$ is in $\Delta$. In particular, as $\Delta$ is ${\mathcal{F}}$-closed, the set $\Delta^*:=\Delta\cup\Gamma$ is ${\mathcal{F}}$-closed and $\L^*:=\L^+|_{\Delta}$ is well-defined. Then $(\L^*,\Delta^*,S)$ is a locality with ${\mathcal{F}}^{cr}\subseteq\Delta\subseteq\Delta^*$ and $\L^*|_\Delta=\L$. Moreover, $N_{\L^*}(P)=N_{\L^+}(P)$ is of characteristic $p$ for every $P\in\Delta^*$. Hence, using Alperin’s fusion theorem [@AKO Theorem I.3.6], we conclude that $(\L^*,\Delta^*,S)$ is a linking locality over ${\mathcal{F}}$. So by Corollary \[C:MainCor\], there exists a unique homomorphism of partial groups $\gamma^*\colon\L^*\longrightarrow{{\widetilde{\mathcal{L}}}}$ with $\gamma^*|_\L=\alpha$. Now by induction on $|\Delta^+\backslash\Delta|$, there exists a unique homomorphism of partial groups $\gamma\colon\L^+\longrightarrow {{\widetilde{\mathcal{L}}}}$ with $\gamma|_{\L^*}=\gamma^*$. Then $\gamma|_\L=\gamma^*|_\L=\alpha$. Moreover, if $\beta\colon\L^+\longrightarrow{{\widetilde{\mathcal{L}}}}$ with $\beta|_\L=\alpha$, then $\beta^*:=\beta|_{\L^*}$ is a homomorphism of partial groups from $\L^*$ to ${{\widetilde{\mathcal{L}}}}$ with $\beta^*|_\L=\alpha$. As $\gamma^*$ is unique, it follows $\beta|_{\L^*}=\beta^*=\gamma^*$, and then the unique choice of $\gamma$ implies $\beta=\gamma$. This proves the assertion.
Transporter systems {#S:Trans}
===================
Transporter systems are certain categories associated to fusion systems which were introduced by Oliver and Ventura [@OV]. As shown by Chermak [@Ch Appendix A], there is a one-to-one correspondence between localities and transporter systems, which we will outline in Subsection \[SS:TransLoc\]. We will moreover introduce isomorphisms between transporter systems in Subsection \[SS:IsoTrans\] and linking systems in Subsection \[SS:LinkingSystems\].
Since the literature on transporter systems is mainly written in “left hand notation”, in this section we will write functions on the left side of the argument. Similarly, we will conjugate from the left. Given a group $G$, we set ${}^g\!x:=gxg^{-1}$ and ${}^g\!P=gPg^{-1}$ for all $x,g\in G$ and $P\subseteq G$. So conjugation by $g$ from the left corresponds to conjugation by $g^{-1}$ from the right.
Isomorphisms between transporter systems {#SS:IsoTrans}
----------------------------------------
If $\Delta$ is a set of subgroups of $G$, write ${\mathcal{T}}_\Delta(G)$ for the category whose object set is $\Delta$, and whose morphism set between two subgroups $P,Q\in\Delta$ is $N_G^l(P,Q):=\{g\in G\colon {}^g\!P\leq Q\}$ or, more precisely, the set of triples $(P,Q,g)$ with $g\in N_G^l(P,Q)$. Here composition of morphisms corresponds to the group multiplication.
A *transporter system* associated to a fusion system ${\mathcal{F}}$ over $S$ is a category ${\mathcal{T}}$ whose set $\Delta$ of objects is an ${\mathcal{F}}$-closed collection of subgroups of $S$, together with functors $${\mathcal{T}}_\Delta(S)\xrightarrow{\;\;\;\delta\;\;\;}{\mathcal{T}}\xrightarrow{\;\;\;\pi\;\;\;}{\mathcal{F}}$$ subject to certain axioms. For example, $\delta$ is the identity on objects and injective on morphism sets, and $\pi$ is the inclusion on objects and surjective on morphism sets; see [@OV Definition 3.1] for details. If we want to be more precise, we say that $({\mathcal{T}},\delta,\pi)$ is a transporter system. By [@OV Lemma A.6], if $P\in\Delta$, then every element of ${\mathrm{Aut}}_{\mathcal{T}}(P):={\mathrm{Mor}}_{\mathcal{T}}(P,P)$ is an isomorphism and so ${\mathrm{Aut}}_{\mathcal{T}}(P)$ is a group. We set $\delta_P:=\delta_{P,P}$ and $\pi_P:=\pi_{P,P}$ for every $P\in\Delta$. Similarly, we set $\alpha_P:=\alpha_{P,P}$ for every functor $\alpha$ from ${\mathcal{T}}$.
If ${\mathcal{T}}$ is a transporter system and $P,Q\in{\mathrm{Ob}}({\mathcal{T}})$, then $\delta_{P,Q}(1)$ should be thought of as the inclusion map. Given $P,Q,P_0,Q_0\in{\mathrm{Ob}}({\mathcal{T}})$ and $\psi\in{\mathrm{Mor}}_{\mathcal{T}}(P,Q)$ with $P_0\leq P$, $Q_0\leq Q$ and $\pi(\psi)(P_0)\leq Q_0$, by [@OV Lemma 3.2(c)], there is a unique morphism $\psi_0\in{\mathrm{Mor}}_{\mathcal{T}}(P_0,Q_0)$ such that $\delta_{Q_0,Q}(1)\circ \psi_0=\psi\circ \delta_{P_0,P}(1)$. The morphism $\psi_0$ is then denoted by $\psi|_{P_0,Q_0}$ and called a restriction of $\psi$. On the other hand, if $\psi_0$ is given then, since every morphism in ${\mathcal{T}}$ is an epimorphism by [@OV Lemma 3.2(d)], the “extension” $\psi$ is uniquely determined if it exists.
\[D:TransIso\] Let $({\mathcal{T}},\delta,\pi)$ and $({{\widetilde{{\mathcal{T}}}}},{{\tilde{\delta}}},{{\tilde{\pi}}})$ be transporter systems associated to fusion systems ${\mathcal{F}}$ and ${{\widetilde{{\mathcal{F}}}}}$ over $p$-groups $S$ and ${{\widetilde{S}}}$ respectively.
- An equivalence of categories $\alpha\colon {\mathcal{T}}\longrightarrow{{\widetilde{{\mathcal{T}}}}}$ is called an *isomorphism* if
- $\alpha$ is *isotypical*, i.e. $\alpha_P(\delta_P(P))={{\tilde{\delta}}}_{\alpha(P)}(\alpha(P))$ for every $P\in{\mathrm{Ob}}({\mathcal{T}})$; and
- $\alpha$ *sends inclusions to inclusions*, i.e. $\alpha_{P,Q}(\delta_{P,Q}(1))={{\tilde{\delta}}}_{\alpha(P),\alpha(Q)}(1)$ for all $P,Q\in{\mathrm{Ob}}({\mathcal{T}})$.
We write ${\mathrm{Iso}}({\mathcal{T}},{{\widetilde{{\mathcal{T}}}}})$ for the set of isomorphisms from ${\mathcal{T}}$ to ${{\widetilde{{\mathcal{T}}}}}$.
- If $\Gamma$ and ${\widetilde{\Gamma}}$ are sets of subgroups of $S$ and ${{\widetilde{S}}}$ respectively, then we write ${\mathrm{Iso}}({\mathcal{T}},{{\widetilde{{\mathcal{T}}}}})_{\Gamma,{\widetilde{\Gamma}}}$ for the set of isomorphisms $\alpha\colon{\mathcal{T}}\longrightarrow{{\widetilde{{\mathcal{T}}}}}$ with $\{\alpha_S(\delta_S(P))\colon P\in\Gamma\}=\{\delta_{{\widetilde{S}}}(Q)\colon Q\in{\widetilde{\Gamma}}\}$.
- An isomorphism $\alpha\colon{\mathcal{T}}\rightarrow{{\widetilde{{\mathcal{T}}}}}$ is called *rigid* if $S={{\widetilde{S}}}$ and $\alpha_S\circ \delta_S={{\tilde{\delta}}}_{{\widetilde{S}}}$.
- An *automorphism* of ${\mathcal{T}}$ is an isomorphism ${\mathcal{T}}\longrightarrow{\mathcal{T}}$. Set ${\mathrm{Aut}}({\mathcal{T}}):={\mathrm{Iso}}({\mathcal{T}},{\mathcal{T}})$ and, for any set $\Gamma$ of subgroups of $S$, set ${\mathrm{Aut}}({\mathcal{T}})_\Gamma:={\mathrm{Iso}}({\mathcal{T}},{\mathcal{T}})_{\Gamma,\Gamma}$.
- If $\gamma\in{\mathrm{Aut}}_{\mathcal{T}}(S)$, then an automorphism $c_\gamma\in{\mathrm{Aut}}({\mathcal{T}})$ is defined on objects via $P\mapsto c_\gamma(P):=\pi(\gamma)(P)$ and on morphisms ${\varphi}\in{\mathrm{Hom}}_{\mathcal{T}}(P,Q)$ by sending ${\varphi}$ to $$c_\gamma({\varphi}):=\gamma|_{Q,c_\gamma(Q)} \circ {\varphi}\circ (\gamma|_{P,c_\gamma(P)})^{-1}\in{\mathrm{Hom}}_{\mathcal{T}}(c_\gamma(P),c_\gamma(Q)).$$ We will refer to $c_\gamma$ as the automorphism of ${\mathcal{T}}$ induced by *conjugation by $\gamma$*. The group of automorphisms of ${\mathcal{T}}$ of the form $c_\gamma$ with $\gamma\in{\mathrm{Aut}}_{\mathcal{T}}(S)$ is called the group of *inner automorphisms* of ${\mathcal{T}}$.
- Let ${\mathrm{Out}}_{{\operatorname{typ}}}({\mathcal{T}})$ be the set of natural isomorphism classes of isotypical self-equivalences of $\L$.
It should be noted that ${\mathrm{Out}}_{{\operatorname{typ}}}({\mathcal{T}})$ is a submonoid of the (finite) group of natural isomorphism classes of self-equivalences of $\L$, and thus forms a finite group. In Lemma \[L:ExactSequence\] below, we will see that ${\mathrm{Out}}_{{\operatorname{typ}}}({\mathcal{T}})$ is actually the image of ${\mathrm{Aut}}({\mathcal{T}})$ under a homomorphism whose kernel is the group of inner automorphisms.
The above definition of isomorphisms and rigid isomorphisms of transporter systems follows Glauberman–Lynd [@GL2020 Definition 2.3]. The definition of isomorphisms used previously in the literature (e.g. in [@BLO2 p.799], [@AKO p.146] and [@Ch Definition A.2]) is different. In the situations we care about (in particular when we consider linking systems later on), it agrees with the definition of a rigid isomorphism as we explain in Remark \[R:Transporter\](d) below. The group ${\mathrm{Aut}}({\mathcal{T}})$ is also often denoted by ${\mathrm{Aut}}_{{\operatorname{typ}}}^I({\mathcal{T}})$; see e.g. [@AKO p.153].
\[R:Transporter\] Suppose $({\mathcal{T}},\delta,S)$ and $({{\widetilde{{\mathcal{T}}}}},{{\tilde{\delta}}},{{\tilde{\pi}}})$ are transporter systems over fusion systems ${\mathcal{F}}$ and ${{\widetilde{{\mathcal{F}}}}}$ respectively, and let $\alpha\colon{\mathcal{T}}\longrightarrow{{\widetilde{{\mathcal{T}}}}}$ be an isomorphism of transporter systems.
- It follows from the axioms of a transporter system that $\delta_S\colon S\longrightarrow {\mathrm{Aut}}_{\mathcal{T}}(S)$ is a group homomorphism whose image $\delta_S(S)$ is a normal Sylow $p$-subgroup of ${\mathrm{Aut}}_{\mathcal{T}}(S)$. Similarly, ${{\tilde{\delta}}}_{{\widetilde{S}}}({{\widetilde{S}}})$ is a normal Sylow $p$-subgroup of ${\mathrm{Aut}}_{{\widetilde{{\mathcal{T}}}}}({{\widetilde{S}}})$. In particular $$\alpha_S\colon {\mathrm{Aut}}_{\mathcal{T}}(S)\longrightarrow{\mathrm{Aut}}_{{\widetilde{{\mathcal{T}}}}}({{\widetilde{S}}})$$ is an isomorphism of groups which takes $\delta_S(S)$ to ${{\tilde{\delta}}}_{{\widetilde{S}}}({{\widetilde{S}}})$. So writing ${{\tilde{\delta}}}_{{\widetilde{S}}}^{-1}$ for the inverse of the map ${{\widetilde{S}}}\rightarrow {{\tilde{\delta}}}({{\widetilde{S}}})$ induced by ${{\tilde{\delta}}}_{{\widetilde{S}}}$, the map $\beta:={{\tilde{\delta}}}_{{\widetilde{S}}}^{-1}\circ \alpha_S\circ \delta_S$ is a group isomorphism from $S$ to ${{\widetilde{S}}}$. If $\Gamma$ and ${\widetilde{\Gamma}}$ are sets of subgroups of $S$ and ${{\widetilde{S}}}$ respectively, notice that $\alpha\in{\mathrm{Iso}}({\mathcal{T}},{{\widetilde{{\mathcal{T}}}}})_\Gamma$ if and only if $\beta(\Gamma):=\{\beta(P)\colon P\in\Gamma\}$ equals ${\widetilde{\Gamma}}$.
- As can be seen from the proof of [@GL2020 Proposition 2.5], if ${\mathcal{F}}^{cr}\subseteq{\mathrm{Ob}}({\mathcal{T}})$ and ${{\widetilde{{\mathcal{F}}}}}^{cr}\subseteq{\mathrm{Ob}}({{\widetilde{{\mathcal{T}}}}})$, then the isomorphism $\beta\colon S\rightarrow{{\widetilde{S}}}$ from (a) induces an isomorphism of fusion systems from ${\mathcal{F}}$ to ${{\widetilde{{\mathcal{F}}}}}$. In particular, if ${\mathcal{F}}^{cr}\subseteq{\mathrm{Ob}}({\mathcal{T}})$ and $\Gamma$ is an ${\mathrm{Aut}}({\mathcal{F}})$-invariant set of subgroups of $S$, then ${\mathrm{Aut}}({\mathcal{T}})_\Gamma={\mathrm{Aut}}({\mathcal{T}})$.
- Since $\alpha$ maps inclusions to inclusions, $\alpha$ commutes with taking restrictions and extensions. If $P\in{\mathrm{Ob}}({\mathcal{T}})$, then observe that $\delta_S(x)|_{P,P}=\delta_P(x)$ for every $x\in P$ and similarly, ${{\tilde{\delta}}}_{{\widetilde{S}}}(y)|_{\alpha(P),\alpha(P)}={{\tilde{\delta}}}_{\alpha(P)}(y)$ for every $y\in\alpha(P)$. Hence, as $\alpha_P(\delta_P(P))={{\tilde{\delta}}}_{\alpha(P)}(\alpha(P))$, we have $\alpha_S(\delta_S(P))={{\tilde{\delta}}}_{{\widetilde{S}}}(\alpha(P))$ for every $P\in{\mathrm{Ob}}({\mathcal{T}})$. So if $\beta\colon S\rightarrow{{\widetilde{S}}}$ is as in (a), we see that $\alpha(P)=\beta(P)$ for every $P\in{\mathrm{Ob}}({\mathcal{T}})$. Hence, $\alpha$ is a bijection on objects and thus an isomorphism of categories. In particular, ${\mathrm{Aut}}({\mathcal{T}})$ is a group. Moreover, if $\Gamma\subseteq{\mathrm{Ob}}({\mathcal{T}})$ and ${\widetilde{\Gamma}}\subseteq{\mathrm{Ob}}({{\widetilde{{\mathcal{T}}}}})$, we have $\alpha\in{\mathrm{Iso}}({\mathcal{T}},{{\widetilde{{\mathcal{T}}}}})_{\Gamma,{\widetilde{\Gamma}}}$ if and only if $\{\alpha(P)\colon P\in\Gamma\}={\widetilde{\Gamma}}$.
- If $\alpha$ is a rigid isomorphism, then $S={{\widetilde{S}}}$ and the isomorphism $\beta\colon S\rightarrow S$ from (a) is the identity. In particular, by (c), we have then $\alpha(P)=P$ for all $P\in{\mathrm{Ob}}({\mathcal{T}})$. If ${\mathcal{F}}^{cr}\subseteq{\mathcal{T}}$ and ${{\widetilde{{\mathcal{F}}}}}^{cr}\subseteq{{\widetilde{{\mathcal{T}}}}}$, then it follows from [@GL2020 Proposition 2.5] that $\alpha\circ \delta={{\tilde{\delta}}}$ and ${{\tilde{\pi}}}\circ \alpha=\pi$. Thus, in this case a rigid isomorphism ${\mathcal{T}}\longrightarrow{{\widetilde{{\mathcal{T}}}}}$ is the same as an isomorphism of transporter systems in the sense of [@Ch Definition A.2] (which extends the definition of an isomorphism of linking systems in [@BLO2 p. 799] and [@AKO p. 146]).
The correspondence between transporter systems and localities {#SS:TransLoc}
-------------------------------------------------------------
If $(\L,\Delta,S)$ is a locality, then one can easily define a transporter system ${\mathcal{T}}_\Delta(\L)$ over ${\mathcal{F}}_S(\L)$ similarly as for groups; the object set of ${\mathcal{T}}_\Delta(\L)$ is $\Delta$, and the morphism set between two objects $P$ and $Q$ consists of the triples $(P,Q,g)$, where $g\in\L$ with $P\subseteq{\mathbf{D}}(g^{-1})$ and ${}^g\!P:=P^{g^{-1}}\leq Q$. We will outline now how one can construct a locality from a transporter system.
Let $({\mathcal{T}},\delta,\pi)$ be a transporter system associated to a fusion system ${\mathcal{F}}$. By $\Delta$ denote the set of objects of ${\mathcal{T}}$, write ${\mathrm{Iso}}_{\mathcal{T}}(P,Q)$ for the set of isomorphisms between two objects $P$ and $Q$, and ${\mathrm{Iso}}({\mathcal{T}})$ for the set of all isomorphisms in ${\mathcal{T}}$. By [@Ch], there is a partial order $\uparrow_{\mathcal{T}}$ defined on ${\mathrm{Iso}}({\mathcal{T}})$ via ${\varphi}_0\uparrow_{\mathcal{T}}{\varphi}$ if ${\varphi}_0\in{\mathrm{Iso}}_{\mathcal{T}}(P_0,Q_0)$, ${\varphi}\in{\mathrm{Iso}}_{\mathcal{T}}(P,Q)$, $P_0\leq P$, $Q_0\leq Q$ and $${\varphi}\circ\delta_{P_0,P}(1)=\delta_{Q_0,Q}(1)\circ{\varphi}_0.$$ Note that the latter condition means that ${\varphi}_0={\varphi}|_{P_0,Q_0}$.
Let $\L_\Delta({\mathcal{T}})$ be the set of equivalence classes of the elements of ${\mathrm{Iso}}({\mathcal{T}})$ with respect to the smallest equivalence relation on ${\mathrm{Iso}}({\mathcal{T}})$ containing $\uparrow_{\mathcal{T}}$. If ${\varphi}\in{\mathrm{Iso}}({\mathcal{T}})$, write $[{\varphi}]$ for the equivalence class of ${\varphi}$ in $\L_\Delta({\mathcal{T}})$. By ${\mathbf{D}}$ denote the set of tuples $w=(f_k,f_{k-1},\dots,f_1)\in{\bold{W}}(\L_\Delta({\mathcal{T}}))$ for which there exist ${\varphi}_i\in f_i$ for $i=1,\dots,k$ such that the composition ${\varphi}_k\circ{\varphi}_{k-1}\circ\cdots\circ {\varphi}_1$ is defined in the category ${\mathcal{T}}$. Moreover, given such $w$ and ${\varphi}_i$, set $\Pi(w):=[{\varphi}_k\circ{\varphi}_{k-1}\circ\cdots\circ {\varphi}_1]$.
The map $\Pi\colon {\mathbf{D}}\longrightarrow\L_\Delta({\mathcal{T}})$ defined above is well-defined. Together with $\Pi$ and the map $\L\longrightarrow\L,[{\varphi}]\mapsto[{\varphi}^{-1}]$ (which is also well-defined), the set $\L$ forms a partial group by [@Ch Proposition A.9]. Moreover, the map $$S\longrightarrow\L_\Delta({\mathcal{T}}),\;x\mapsto [\delta_S(x)]$$ is an injective homomorphism of partial groups, and its image $[S]$ is a subgroup of $\L_\Delta({\mathcal{T}})$. Most of the time, we will identify $x\in S$ with $[\delta_S(x)]\in [S]$. With this identification, by [@Ch Proposition A.13], $(\L_\Delta({\mathcal{T}}),\Delta,S)$ is a locality.
\[L:BasicLDeltaT\] Let $({\mathcal{T}},\delta,\pi)$ be a transporter system associated to a fusion system ${\mathcal{F}}$, $\Delta:={\mathrm{Ob}}({\mathcal{T}})$ and $\L:=\L_\Delta({\mathcal{T}})$. As above write $[{\varphi}]$ for the equivalence class of ${\varphi}\in{\mathrm{Iso}}({\mathcal{T}})$ in $\L$. Then the following hold:
- If $P,Q\in\Delta$, ${\varphi}\in{\mathrm{Iso}}_{\mathcal{T}}(P,Q)$ and $f=[{\varphi}]$, then $P\subseteq S_{f^{-1}}$, ${}^f\!P=Q$ and $c_{f^{-1}}|_P=\pi({\varphi})$.
- We have ${\mathrm{Aut}}_{\mathcal{T}}(P)\cong N_\L(P)$.
- ${\mathcal{F}}_S(\L)$ is the subsystem of ${\mathcal{F}}$ generated by all the sets ${\mathrm{Hom}}_{\mathcal{F}}(P,Q)$ with $P,Q\in\Delta$.
Let $P$, $Q$, ${\varphi}$ and $f$ be as in (a), fix $x\in P$ and set $y:=\pi({\varphi})(x)\in Q$. Observe that, via the usual identification of the elements of $S$ with elements of $\L$, we have $x=[\delta_S(x)]=[\delta_P(x)]$, since $\delta_S(x)\delta_{P,S}(1)=\delta_{P,S}(x)=\delta_{P,S}(1)\delta_P(x)$. Similarly, we have $y=[\delta_S(y)]=[\delta_Q(y)]$. Notice that the composition ${\varphi}\circ \delta_P(x)\circ {\varphi}^{-1}$ is defined in $\L$. Moreover, it follows from the definition of a transporter system (axiom (C) in [@OV Definition 3.1]) that ${\varphi}\circ \delta_P(x)\circ {\varphi}^{-1}=\delta_Q(y)$. Hence, $(f,x,f^{-1})=([{\varphi}],[\delta_P(x)],[{\varphi}^{-1}])\in{\mathbf{D}}$ and $x^{f^{-1}}=\Pi(f,x,f^{-1})=[{\varphi}\circ \delta_P(x)\circ {\varphi}^{-1}]=[\delta_Q(y)]=y$. This shows (a).
Property (a) yields in particular that the map $\alpha\colon {\mathrm{Aut}}_{\mathcal{T}}(P)\longrightarrow N_\L(P),{\varphi}\mapsto[{\varphi}]$ is well-defined. Moreover, $\alpha$ is surjective by [@Ch Corollary A.11] and injective by [@Ch Lemma A.8(b)]. For all ${\varphi},\psi\in{\mathrm{Aut}}({\mathcal{T}})$, we have $\alpha({\varphi}\circ\psi)=[{\varphi}\circ\psi]=\Pi([{\varphi}],[\psi])=\Pi(\alpha({\varphi}),\alpha(\psi))$. Hence, $\alpha$ is an isomorphism of groups and (b) holds.
To prove (c) notice that ${\mathcal{F}}_S(\L)$ is generated by all the maps of the form $c_{f^{-1}}\colon P\longrightarrow Q$, where $P,Q\in\Delta$, $P\leq S_{f^{-1}}$ and ${}^f\!P=Q$. For such $P,Q,f$, by [@Ch Corollary A.11], there exists always ${\varphi}\in{\mathrm{Iso}}_{\mathcal{T}}(P,Q)$ with $f=[{\varphi}]$. Moreover, the fusion system generated by the sets ${\mathrm{Hom}}_{\mathcal{F}}(P,Q)$ with $P,Q\in\Delta$ is actually generated by the sets ${\mathrm{Iso}}_{\mathcal{F}}(P,Q)$ with $P,Q\in\Delta$. Since $\pi$ is surjective on morphism sets and, by [@OV Lemma A.6], the preimages of isomorphisms in ${\mathcal{F}}$ under $\pi$ are isomorphisms in ${\mathcal{T}}$, property (c) follows from (a).
If ${\mathcal{C}}$ is a small category and $\Gamma\subseteq{\mathrm{Ob}}({\mathcal{C}})$, we will write ${\mathcal{C}}|_\Gamma$ for the full subcategory of ${\mathcal{C}}$ with object set $\Gamma$.
\[L:Iota\] Let ${\mathcal{T}}^+$ be a transporter system associated to a fusion system ${\mathcal{F}}$ and let $\Delta\subseteq\Delta^+:={\mathrm{Ob}}({\mathcal{T}}^+)$ such that $\Delta$ is ${\mathcal{F}}$-closed. Then ${\mathcal{T}}:={\mathcal{T}}^+|_\Delta$ is a transporter system associated to ${\mathcal{F}}$. Moreover, writing $[{\varphi}]_+$ for the equivalence class of ${\varphi}\in{\mathrm{Iso}}({\mathcal{T}}^+)$ in $\L_{\Delta^+}({\mathcal{T}}^+)$, and $[{\varphi}]$ for the equivalence class of ${\varphi}\in{\mathrm{Iso}}({\mathcal{T}})$ in $\L_\Delta({\mathcal{T}})$, the map $$\iota\colon \L_\Delta({\mathcal{T}})\longrightarrow \L_{\Delta^+}({\mathcal{T}}^+)|_\Delta,\;[{\varphi}]\mapsto [{\varphi}]_+\mbox{ for all }{\varphi}\in{\mathrm{Iso}}({\mathcal{T}})$$ is well-defined and an isomorphism of partial groups, which restricts to the identity on $S$ (if one identifies the elements of $S$ with elements of $\L_\Delta({\mathcal{T}})$ and $\L_{\Delta^+}({\mathcal{T}}^+)$ as usual).
As $\Delta$ is ${\mathcal{F}}$-closed, it is immediate from the axioms of a transporter system that ${\mathcal{T}}:={\mathcal{T}}^+|_\Delta$ together with the restriction of $\delta$ to ${\mathcal{T}}_\Delta(S)$ and the restriction of $\pi$ to ${\mathcal{T}}$ is a transporter system. Set $\L:=\L_\Delta({\mathcal{T}})$ and $\L^+:=\L_{\Delta^+}({\mathcal{T}}^+)$. Write ${\mathbf{D}}$ for the domain of the partial product on $\L$ and ${\mathbf{D}}':={\mathbf{D}}_\Delta(\L^+)$ for the domain of the partial product on $\L^+|_\Delta$.
If ${\varphi},\psi\in{\mathrm{Iso}}({\mathcal{T}})\subseteq{\mathrm{Iso}}({\mathcal{T}}^+)$, then ${\varphi}\uparrow_{\mathcal{T}}\psi$ implies ${\varphi}\uparrow_{{\mathcal{T}}^+}\psi$ and so $[{\varphi}]=[\psi]$ yields $[{\varphi}]_+=[\psi]_+$. Hence, the map $$\iota'\colon \L\longrightarrow \L^+,\;[{\varphi}]\mapsto [{\varphi}]_+\mbox{ for all }{\varphi}\in{\mathrm{Iso}}({\mathcal{T}})$$ is well-defined. It follows from the construction of the partial products on $\L=\L_\Delta({\mathcal{T}})$ and $\L^+=\L_{\Delta^+}({\mathcal{T}}^+)$ that $\iota$ is a homomorphism of partial groups. Moreover, since we identify every element $x\in S$ with $[\delta_S(x)]\in\L$ and with $[\delta_S(x)]_+\in\L^+$, the map $\iota'$ restricts to the identity on $S$. In particular, as $(\L,\Delta,S)$ is a locality, it follows $\iota'(\L)\subseteq\L^+|_\Delta$ and $(\iota')^*({\mathbf{D}})\subseteq{\mathbf{D}}'={\mathbf{D}}_\Delta(\L^+)$ (with $(\iota')^*$ defined as in Definition \[N:PartialHomWordMap\], but written on the left). Hence, $\iota$ is well-defined and a homomorphism of partial groups. It remains to show that $\iota$ is injective and ${\mathbf{D}}'\subseteq\iota^*({\mathbf{D}})$.
If ${\varphi}\uparrow_{{\mathcal{T}}^+}\chi$ for some ${\varphi}\in{\mathrm{Iso}}({\mathcal{T}})$ and $\chi\in{\mathrm{Iso}}({\mathcal{T}}^+)$, then the assumption that $\Delta$ is overgroup closed in $S$ implies $\chi\in{\mathrm{Iso}}({\mathcal{T}})$ and ${\varphi}\uparrow_{\mathcal{T}}\chi$. By [@Ch Lemma A.8(a)], every element of $\L^+$ contains a unique maximal element with respect to the partial order $\uparrow_{{\mathcal{T}}^+}$. So if $[{\varphi}]_+=[\psi]_+$ for some ${\varphi},\psi\in{\mathrm{Iso}}({\mathcal{T}})$, then for the $\uparrow_{{\mathcal{T}}^+}$-maximal element $\chi$ of $[{\varphi}]_+=[\psi]_+$, we have $\chi\in{\mathrm{Iso}}({\mathcal{T}})$, ${\varphi}\uparrow_{\mathcal{T}}\chi$ and $\psi\uparrow_{\mathcal{T}}\chi$. Hence, $[{\varphi}]=[\psi]$ proving that $\iota$ is injective.
If $P,Q\in\Delta$ and $f\in\L^+$ with ${}^f\!P=Q$, then by [@Ch Corollary A.11], there exists $\psi\in{\mathrm{Iso}}_{{\mathcal{T}}^+}(P,Q)$ with $f=[\psi]_+$. For such $\psi$, we have $\psi\in{\mathrm{Iso}}_{\mathcal{T}}(P,Q)$ and $\iota([\psi])=f$. From this property, the definition of ${\mathbf{D}}$ and the definition of ${\mathbf{D}}'={\mathbf{D}}_\Delta(\L^+)$, one sees that ${\mathbf{D}}'\subseteq\iota^*({\mathbf{D}})$. Hence the assertion holds.
\[L:Lambda\] Let $({\mathcal{T}},\delta,\pi)$ and $({{\widetilde{{\mathcal{T}}}}},{{\tilde{\delta}}},{{\tilde{\pi}}})$ be transporter systems associated to fusion systems ${\mathcal{F}}$ and ${{\widetilde{{\mathcal{F}}}}}$ over $p$-groups $S$ and ${{\widetilde{S}}}$ respectively. Set $\Delta:={\mathrm{Ob}}({\mathcal{T}})$, $\L:=\L_\Delta({\mathcal{T}})$, ${\widetilde{\Delta}}={\mathrm{Ob}}({{\widetilde{{\mathcal{T}}}}})$ and ${{\widetilde{\mathcal{L}}}}:=\L_{{\widetilde{\Delta}}}({{\widetilde{{\mathcal{T}}}}})$. For $\alpha\in{\mathrm{Iso}}({\mathcal{T}},{{\widetilde{{\mathcal{T}}}}})$ define $\Lambda(\alpha)\colon \L\longrightarrow {{\widetilde{\mathcal{L}}}}$ to be the map which, for all $P,Q\in\Delta$ and all ${\varphi}\in{\mathrm{Iso}}_{{\mathcal{T}}}(P,Q)$, sends the class $[{\varphi}]\in\L$ to the class $[\alpha_{P,Q}({\varphi})]\in{{\widetilde{\mathcal{L}}}}$. Then this defines a bijection $$\Lambda\colon {\mathrm{Iso}}({\mathcal{T}},{{\widetilde{{\mathcal{T}}}}})\longrightarrow {\mathrm{Iso}}((\L,\Delta,S),({{\widetilde{\mathcal{L}}}},{\widetilde{\Delta}},{{\widetilde{S}}})),\;\alpha\mapsto\Lambda(\alpha).$$ Moreover, if $\Gamma$ and ${\widetilde{\Gamma}}$ are sets of subgroups of $S$ and ${{\widetilde{S}}}$ respectively, then $\Lambda$ induces a bijection $${\mathrm{Iso}}({\mathcal{T}},{{\widetilde{{\mathcal{T}}}}})_{\Gamma,{\widetilde{\Gamma}}}\longrightarrow {\mathrm{Iso}}((\L,\Delta,S),({{\widetilde{\mathcal{L}}}},{\widetilde{\Delta}},{{\widetilde{S}}}))_{\Gamma,{\widetilde{\Gamma}}}.$$
By [@GL2020 Theorem 2.11] and the proof of this result, there is an equivalence $\Lambda'$ from the category of transporter systems with isomorphisms to the category of localities with isomorphisms, which is defined on objects by sending a transporter system ${\mathcal{T}}$ to $\L_{{\mathrm{Ob}}({\mathcal{T}})}({\mathcal{T}})$, and on morphisms by sending an isomorphism $\alpha\in{\mathrm{Iso}}({\mathcal{T}},{{\widetilde{{\mathcal{T}}}}})$ to $\Lambda(\alpha)$ as defined in the theorem. In particular, $\Lambda=\Lambda'_{{\mathcal{T}},{{\widetilde{{\mathcal{T}}}}}}$ is a bijection ${\mathrm{Iso}}({\mathcal{T}},{{\widetilde{{\mathcal{T}}}}})\longrightarrow{\mathrm{Iso}}((\L,\Delta,S),({{\widetilde{\mathcal{L}}}},{\widetilde{\Delta}},{{\widetilde{S}}}))$.
Let now $P\leq S$ and $Q\leq {{\widetilde{S}}}$. Via the usual identifications of the elements of $S$ and ${{\widetilde{S}}}$ with elements of $\L$ and ${{\widetilde{\mathcal{L}}}}$, we have $$P=\{[\delta_S(x)]\colon x\in P\}\mbox{ and }Q=\{[{{\tilde{\delta}}}_{{\widetilde{S}}}(y)]\colon y\in Q\}.$$ So given $\alpha\in{\mathrm{Iso}}({\mathcal{T}},{{\widetilde{{\mathcal{T}}}}})$, we have $$\Lambda(\alpha)(P)=\{\Lambda(\alpha)([\delta_S(x)])\colon x\in P\}=\{[\alpha_S(\delta_S(x))]\colon x\in P\}.$$ As the map ${\mathrm{Aut}}_{{{\widetilde{{\mathcal{T}}}}}}(S)\longrightarrow{{\widetilde{\mathcal{L}}}},{\varphi}\mapsto[{\varphi}]$ is by [@Ch Lemma A.8(b)] injective, it follows that $\Lambda(\alpha)(P)=Q$ if and only if $\alpha_S(\delta_S(P))={{\tilde{\delta}}}_{{\widetilde{S}}}(Q)$. So $\Lambda(\alpha)(\Gamma):=\{\Lambda(\alpha)(P)\colon P\in\Gamma\}$ equals ${\widetilde{\Gamma}}$ if and only if $\{\alpha_S(\delta_S(P))\colon P\in\Gamma\}=\{{{\tilde{\delta}}}_{{\widetilde{S}}}(Q)\colon Q\in{\widetilde{\Gamma}}\}$. Equivalently, $\Lambda(\alpha)\in{\mathrm{Iso}}((\L,\Delta,S),({{\widetilde{\mathcal{L}}}},{\widetilde{\Delta}},{{\widetilde{S}}}))_{\Gamma,{\widetilde{\Gamma}}}$ if and only if $\alpha\in{\mathrm{Iso}}({\mathcal{T}},{{\widetilde{{\mathcal{T}}}}})_{\Gamma,{\widetilde{\Gamma}}}$. (Unlike in Definition \[D:LocalityHomomorphism\], we write maps here on the right.)
Linking systems {#SS:LinkingSystems}
---------------
In this paper we work with the following definition of a linking system, which is slightly non-standard, but fits well with the earlier given definition of a linking locality (Definition \[D:LinkingLoc\]).
\[D:LinkingSystem\] If ${\mathcal{F}}$ is a saturated fusion system, then a *linking system* associated to ${\mathcal{F}}$ is a transporter system ${\mathcal{T}}$ associated to ${\mathcal{F}}$ such that ${\mathcal{F}}^{cr}\subseteq{\mathrm{Ob}}({\mathcal{T}})$ and ${\mathrm{Aut}}_{\mathcal{T}}(P)$ is of characteristic $p$ for every $P\in{\mathrm{Ob}}({\mathcal{T}})$. If ${\mathrm{Ob}}({\mathcal{T}})={\mathcal{F}}^s$, then ${\mathcal{T}}$ is called a *subcentric linking system* associated to ${\mathcal{F}}$.
The original definitions of linking systems in [@BLO2], [@controlling] and [@OliverExtensions] are not based on the definition of a transporter systems. A linking system in either of these definitions is a linking system in the above definition, while the converse does not hold in general. Historically, *centric linking systems*, i.e. linking systems over ${\mathcal{F}}$ whose object sets are the sets of ${\mathcal{F}}$-centric subgroups, were studied first. The longstanding conjecture that there is a centric linking system associated to every saturated fusion system, and that such a centric linking system is unique up to a rigid isomorphism was shown by Chermak [@Ch], and subsequently by Oliver [@Oliver:2013]. Originally, these proofs use the classification of finite simple groups, but the dependence on the classification of the proof in [@Oliver:2013] was removed by Glauberman–Lynd [@GL2016].
If ${\mathcal{T}}$ is a linking system associated to a saturated fusion system ${\mathcal{F}}$ over $S$, then ${\mathrm{Ob}}({\mathcal{T}})\subseteq{\mathcal{F}}^s$. On the other hand, if ${\mathcal{F}}^{cr}\subseteq\Delta\subseteq{\mathcal{F}}^s$ such that $\Delta$ is ${\mathcal{F}}$-closed, then it is stated in [@subcentric Theorem A] that there is a linking system ${\mathcal{T}}$ with object set $\Delta$ associated to ${\mathcal{F}}$; moreover, such ${\mathcal{T}}$ is unique up to rigid isomorphism. The proof relies heavily on the existence and uniqueness of centric linking systems. Formally, [@subcentric Theorem A] is proved as a consequence of the corresponding statement about linking localities which is summarized in Subsection \[SS:LinkingLoc\]. We use this opportunity to point out that a precise argument that ${\mathcal{T}}$ is unique up to a rigid isomorphism is actually missing in [@subcentric]. However, the uniqueness of ${\mathcal{T}}$ follows from [@GL2020 Theorem 2.11] (or [@Ch Lemma A.14, Lemma A.15]) and from Lemma \[L:LinkingSystemLinkingLoc\] below.)
If $(\L,\Delta,S)$ is a locality over ${\mathcal{F}}$, then it is easy to see that the corresponding transporter system ${\mathcal{T}}_\Delta(\L)$ is a linking system associated to ${\mathcal{F}}$ if and only if $(\L,\Delta,S)$ is a linking locality. Moreover, we have the following lemma.
\[L:LinkingSystemLinkingLoc\] If ${\mathcal{T}}$ is a linking system associated to a saturated fusion system ${\mathcal{F}}$, then for $\Delta:={\mathrm{Ob}}({\mathcal{T}})$, the locality $(\L_\Delta({\mathcal{T}}),\Delta,S)$ is a linking locality over ${\mathcal{F}}$.
As ${\mathcal{F}}^{cr}\subseteq\Delta$, it follows from Alperin’s fusion theorem [@AKO Theorem I.3.6] and Lemma \[L:BasicLDeltaT\](c) that ${\mathcal{F}}_S(\L)={\mathcal{F}}$. In particular, ${\mathcal{F}}_S(\L)$ is saturated and ${\mathcal{F}}_S(\L)^{cr}\subseteq\Delta$. Moreover, by Lemma \[L:BasicLDeltaT\](b), $N_\L(P)\cong{\mathrm{Aut}}_{\mathcal{T}}(P)$ is of characteristic $p$ for every $P\in\Delta$.
\[L:LinkingSystemsElementary\] If $({\mathcal{T}},\delta,\pi)$ is a linking system associated to a saturated fusion system ${\mathcal{F}}$ over $S$, then the following hold:
- We have $\ker(\pi_S)=\delta_S(Z(S))$.
- For every $P\in{\mathrm{Ob}}({\mathcal{T}})$, we have $O_p({\mathrm{Aut}}_{\mathcal{T}}(P))=\delta_P(P)$ if and only if $P\in{\mathcal{F}}^{cr}$.
- *(Alperin’s Fusion Theorem for linking systems)* Each morphism in ${\mathcal{T}}$ is the composite of restrictions of elements in the automorphism groups ${\mathrm{Aut}}_{\mathcal{T}}(P)$, where $P\in{\mathcal{F}}^{cr}$ is fully ${\mathcal{F}}$-normalized.
If $\gamma\in{\mathrm{Aut}}_{\mathcal{T}}(S)$, then for all $g\in S$, $\pi_S(\gamma)(g)=g$ if and only if $\gamma$ commutes with $\delta_S(g)$ by Axiom C in [@OV Definition 3.1] and [@OV Lemma 3.3]. Hence, $\ker(\pi_S)=C_{{\mathrm{Aut}}_{\mathcal{T}}(S)}(\delta_S(S))$. As ${\mathrm{Aut}}_{\mathcal{T}}(S)$ is of characteristic $p$ and $\delta_S(S)$ is a normal Sylow $p$-subgroup of ${\mathrm{Aut}}_{\mathcal{T}}(S)$, we have $\gamma\in C_{{\mathrm{Aut}}_{\mathcal{T}}(S)}(\delta_S(S))=Z(\delta_S(S))=\delta_S(Z(S))$ showing (a).
Property (b) follows from [@subcentric Lemma 6.2] and Lemma \[L:LinkingSystemLinkingLoc\], or alternatively this property can be shown by reformulating the argument in the proof of [@subcentric Lemma 6.2] for transporter systems. Property (c) follows from (b) and [@OV Proposition 3.9].
\[L:ExactSequence\] If $({\mathcal{T}},\delta,\pi)$ is a linking system associated to a saturated fusion system ${\mathcal{F}}$ over $S$, then the sequence $$1\longrightarrow Z({\mathcal{F}})\xrightarrow{\;\;\delta_S\;\;} {\mathrm{Aut}}_{\mathcal{T}}(S)\xrightarrow{\;\gamma\mapsto c_\gamma\;} {\mathrm{Aut}}({\mathcal{T}})\longrightarrow {\mathrm{Out}}_{{{\operatorname{typ}}}}({\mathcal{T}})\longrightarrow 1$$ is exact.
The statement was shown in [@AOV1 Lemma 1.14(a)] for linking systems in Oliver’s definition, i.e. for linking systems whose objects are quasicentric subgroups. The argument can be repeated verbatim (with $\L$ replaced by ${\mathcal{T}}$) to prove exactness in ${\mathrm{Aut}}({\mathcal{T}})$ and in ${\mathrm{Out}}_{{\operatorname{typ}}}({\mathcal{T}})$ and to show that $c_\gamma={\operatorname{id}}_{\mathcal{T}}$ implies $\gamma\in\delta_S(Z({\mathcal{F}}))$; here only the reference to [@AOV1 Lemma 1.11(b’)] needs to be replaced by a reference to [@OV Lemma 3.2(c)], the reference to axiom (A) needs to be replaced by a reference to Lemma \[L:LinkingSystemsElementary\](a), and the reference to [@AOV1 Lemma 1.11(e)] needs to be replaced by a reference to axiom (II) in the definition of a transporter system [@OV Definition 3.1]. On the other hand, by [@AKO Proposition 4.5], we have $Z({\mathcal{F}})\leq P$ for all $P\in{\mathcal{F}}^{cr}$. So if $a\in Z({\mathcal{F}})$, then Lemma \[L:LinkingSystemsElementary\](c) yields that any morphism $\psi\in{\mathrm{Mor}}_\L(P,Q)$ extends to a morphism $\ov{\psi}\in{\mathrm{Mor}}_\L({\langle}P,a{\rangle},{\langle}Q,a{\rangle})$ with $\pi(\ov{\psi})(a)=a$. Such $\ov{\psi}$ commutes with $\gamma=\delta_S(a)$ by axiom (C) again. So $\gamma$ commutes with $\psi$ and thus $c_\gamma={\operatorname{id}}_\L$. This shows exactness in ${\mathrm{Aut}}_{\mathcal{T}}(S)$.
Isomorphisms between linking localities and linking systems {#S:Iso}
===========================================================
In this section we prove Theorems \[T:A1\] and \[T:A2\]. Moreover, we show considerably more general versions of these theorems, where for each result we formulate a version for linking localities and a version for linking systems. Theorem \[T:A2\] leads naturally to a statement about outer automorphism groups (Theorem \[T:A2Outer\]), and building on this we prove Theorem \[C:B\].
\[T:MainIso\] Let $(\L,\Delta,S)$, $(\L^+,\Delta^+,S)$, $({{\widetilde{\mathcal{L}}}},{\widetilde{\Delta}},{{\widetilde{S}}})$ and $({{\widetilde{\mathcal{L}}}}^+,{\widetilde{\Delta}}^+,{{\widetilde{S}}})$ be linking localities such that
- ${\mathcal{F}}_S(\L^+)={\mathcal{F}}_S(\L)$, $\Delta\subseteq\Delta^+$, $\L=\L^+|_\Delta$, and
- ${\mathcal{F}}_{{{\widetilde{S}}}}({{\widetilde{\mathcal{L}}}}^+)={\mathcal{F}}_{{{\widetilde{S}}}}({{\widetilde{\mathcal{L}}}})$, ${\widetilde{\Delta}}\subseteq{\widetilde{\Delta}}^+$, ${{\widetilde{\mathcal{L}}}}={{\widetilde{\mathcal{L}}}}^+|_{{\widetilde{\Delta}}}$.
Then the map $$\Psi\colon {\mathrm{Iso}}((\L^+,\Delta^+,S),({{\widetilde{\mathcal{L}}}}^+,{\widetilde{\Delta}}^+,{{\widetilde{S}}}))_{\Delta,{\widetilde{\Delta}}}\longrightarrow {\mathrm{Iso}}((\L,\Delta,S),({{\widetilde{\mathcal{L}}}},{\widetilde{\Delta}},{{\widetilde{S}}}))_{\Delta^+,{\widetilde{\Delta}}^+}$$ with $\Psi(\gamma)=\gamma|_{\L}$ is well-defined and a bijection.
By Lemma \[L:IsoRestrictionLocality\], the map $\Psi$ is well-defined. If $\alpha\in {\mathrm{Iso}}((\L,\Delta,S),({{\widetilde{\mathcal{L}}}},{\widetilde{\Delta}},{{\widetilde{S}}}))_{\Delta^+,{\widetilde{\Delta}}^+}$, then $\alpha$ regarded as a map $\L\longrightarrow{{\widetilde{\mathcal{L}}}}^+$ is a homomorphism of partial groups. Thus, by Corollary \[C:MainCor2\], $\alpha$ extends to a unique homomorphism of partial groups $\gamma\colon \L^+\longrightarrow{{\widetilde{\mathcal{L}}}}^+$. By Lemma \[L:PartialIso\], $\alpha^{-1}$ is a homomorphism of partial groups from ${{\widetilde{\mathcal{L}}}}$ to $\L$, which can be regarded as a homomorphism of partial groups ${{\widetilde{\mathcal{L}}}}\longrightarrow\L^+$. So again by Corollary \[C:MainCor\], $\alpha^{-1}$ extends to a homomorphism of partial groups $\hat{\gamma}\colon{{\widetilde{\mathcal{L}}}}^+\longrightarrow\L^+$. Then $\gamma\hat{\gamma}\colon\L^+\longrightarrow\L^+$ and $\hat{\gamma}\gamma\colon{{\widetilde{\mathcal{L}}}}^+\longrightarrow{{\widetilde{\mathcal{L}}}}^+$ are homomorphisms of partial groups with $(\gamma\hat{\gamma})|_{\L}=\alpha\alpha^{-1}={\operatorname{id}}_{\L}$ and $(\hat{\gamma}\gamma)|_{{{\widetilde{\mathcal{L}}}}}=\alpha^{-1}\alpha={\operatorname{id}}_{{{\widetilde{\mathcal{L}}}}}$. It follows from Corollary \[C:MainCor\] applied with ${\operatorname{id}}_{\L}$ in place of $\alpha$ that there is a unique homomorphism of partial groups $\L^+\longrightarrow\L^+$ which restricts to the identity on $\L$. Thus, any such homomorphism equals ${\operatorname{id}}_{\L^+}$. Similarly, any homomorphism of partial groups ${{\widetilde{\mathcal{L}}}}^+\longrightarrow{{\widetilde{\mathcal{L}}}}^+$ which restricts to the identity on ${{\widetilde{\mathcal{L}}}}$ equals ${\operatorname{id}}_{{{\widetilde{\mathcal{L}}}}^+}$. This shows $\gamma\hat{\gamma}={\operatorname{id}}_{\L^+}$ and $\hat{\gamma}\gamma={\operatorname{id}}_{{{\widetilde{\mathcal{L}}}}^+}$, i.e. $\gamma$ is bijective with inverse map $\hat{\gamma}$. So $\gamma\colon\L^+\longrightarrow{{\widetilde{\mathcal{L}}}}^+$ is an isomorphism of partial groups by Lemma \[L:PartialIso\]. As $\Delta^+\gamma=\Delta^+\alpha={\widetilde{\Delta}}^+$ and $\Delta\gamma=\Delta\alpha={\widetilde{\Delta}}$, it follows that $\gamma\in {\mathrm{Iso}}((\L^+,\Delta^+,S),({{\widetilde{\mathcal{L}}}}^+,{\widetilde{\Delta}}^+,{{\widetilde{S}}}))_{\Delta,{\widetilde{\Delta}}}$ with $\Psi(\gamma)=\gamma|_{\L}=\alpha$. This shows that $\Psi$ is surjective. As $\gamma$ is the unique homomorphism of partial groups $\L^+\longrightarrow{{\widetilde{\mathcal{L}}}}^+$ which restricts to $\alpha$, the maps $\Psi$ is also injective.
\[T:MainIsoT\] Suppose ${\mathcal{T}}$, ${\mathcal{T}}^+$, ${{\widetilde{{\mathcal{T}}}}}$ and ${{\widetilde{{\mathcal{T}}}}}^+$ are linking systems, and ${\mathcal{F}}$ and ${{\widetilde{{\mathcal{F}}}}}$ are saturated fusion systems such that
- ${\mathcal{T}}$ and ${\mathcal{T}}^+$ are linking systems associated to ${\mathcal{F}}$, ${\mathrm{Ob}}({\mathcal{T}})\subseteq{\mathrm{Ob}}({\mathcal{T}}^+)$, ${\mathcal{T}}={\mathcal{T}}^+|_{{\mathrm{Ob}}({\mathcal{T}})}$;
- ${{\widetilde{{\mathcal{T}}}}}$ and ${{\widetilde{{\mathcal{T}}}}}^+$ are linking systems associated to ${{\widetilde{{\mathcal{F}}}}}$, ${\mathrm{Ob}}({{\widetilde{{\mathcal{T}}}}})\subseteq{\mathrm{Ob}}({{\widetilde{{\mathcal{T}}}}}^+)$, ${{\widetilde{{\mathcal{T}}}}}={{\widetilde{{\mathcal{T}}}}}^+|_{{\mathrm{Ob}}({{\widetilde{{\mathcal{T}}}}})}$.
Then the map $${\mathrm{Iso}}({\mathcal{T}}^+,{{\widetilde{{\mathcal{T}}}}}^+)_{{\mathrm{Ob}}({\mathcal{T}}),{\mathrm{Ob}}({{\widetilde{{\mathcal{T}}}}})}\longrightarrow {\mathrm{Iso}}({\mathcal{T}},{{\widetilde{{\mathcal{T}}}}})_{{\mathrm{Ob}}({\mathcal{T}}^+),{\mathrm{Ob}}({{\widetilde{{\mathcal{T}}}}}^+)},\;\alpha\mapsto\alpha|_{\mathcal{T}}$$ is a bijection.
As we are dealing with transporter systems, in this proof, we will again write functions from the left. Set $\Delta:={\mathrm{Ob}}({\mathcal{T}})$, $\Delta^+:={\mathrm{Ob}}({\mathcal{T}}^+)$, ${\widetilde{\Delta}}:={\mathrm{Ob}}({{\widetilde{{\mathcal{T}}}}})$ and ${\widetilde{\Delta}}^+:={\mathrm{Ob}}({{\widetilde{{\mathcal{T}}}}}^+)$, $$\L:=\L_\Delta({\mathcal{T}}),\;\L^+:=\L_{\Delta^+}({\mathcal{T}}^+),\;{{\widetilde{\mathcal{L}}}}:=\L_{{\widetilde{\Delta}}}({{\widetilde{{\mathcal{T}}}}})\mbox{ and }{{\widetilde{\mathcal{L}}}}^+:=\L_{{\widetilde{\Delta}}^+}({{\widetilde{{\mathcal{T}}}}}^+).$$ By $[{\varphi}]$ we denote the equivalence class of ${\varphi}$ in $\L$ if ${\varphi}\in{\mathrm{Iso}}({\mathcal{T}})$, and the equivalence class of ${\varphi}$ in ${{\widetilde{\mathcal{L}}}}$ if ${\varphi}\in{\mathrm{Iso}}({{\widetilde{{\mathcal{T}}}}})$. Similarly, $[{\varphi}]_+$ denotes the equivalence class of ${\varphi}$ in $\L^+$ if ${\varphi}\in{\mathrm{Iso}}({\mathcal{T}}^+)$ and the equivalence class of ${\varphi}$ in ${{\widetilde{\mathcal{L}}}}^+$ if ${\varphi}\in{\mathrm{Iso}}({{\widetilde{{\mathcal{T}}}}}^+)$. By Lemma \[L:Iota\], the maps $\iota\colon \L\longrightarrow \L^+|_\Delta,[{\varphi}]\mapsto [{\varphi}]_+$ and ${{\tilde{\iota}}}\colon{{\widetilde{\mathcal{L}}}}\longrightarrow{{\widetilde{\mathcal{L}}}}^+|_{{\widetilde{\Delta}}},[{\varphi}]\mapsto[{\varphi}]_+$ are isomorphisms of localities which restrict to the identity on $S$. In particular, the map $$\Phi\colon {\mathrm{Iso}}((\L,\Delta,S),({{\widetilde{\mathcal{L}}}},{\widetilde{\Delta}},{{\widetilde{S}}}))_{\Delta^+,{\widetilde{\Delta}}^+}\longrightarrow {\mathrm{Iso}}((\L^+|_\Delta,\Delta,S),({{\widetilde{\mathcal{L}}}}^+|_{{\widetilde{\Delta}}},{\widetilde{\Delta}},{{\widetilde{S}}}))_{\Delta^+,{\widetilde{\Delta}}^+},\;\beta\mapsto {{\tilde{\iota}}}\circ\beta\circ \iota^{-1}$$ is a bijection. By Lemma \[L:IsoRestrictionLocality\], there is also a bijection $$\Psi\colon {\mathrm{Iso}}((\L^+,\Delta^+,S),({{\widetilde{\mathcal{L}}}}^+,{\widetilde{\Delta}}^+,{{\widetilde{S}}}))_{\Delta,{\widetilde{\Delta}}}\longrightarrow {\mathrm{Iso}}((\L^+|_\Delta,\Delta,S),({{\widetilde{\mathcal{L}}}}^+|_{{\widetilde{\Delta}}},{\widetilde{\Delta}},{{\widetilde{S}}}))_{\Delta^+,{\widetilde{\Delta}}^+}$$ given by restriction. By Lemma \[L:Lambda\], there are moreover bijections $$\Lambda\colon {\mathrm{Iso}}({\mathcal{T}},{{\widetilde{{\mathcal{T}}}}})_{\Delta^+,{\widetilde{\Delta}}^+}\longrightarrow {\mathrm{Iso}}((\L,\Delta,S),({{\widetilde{\mathcal{L}}}},{\widetilde{\Delta}},{{\widetilde{S}}}))_{\Delta^+,{\widetilde{\Delta}}^+}$$ and $$\Lambda^+\colon {\mathrm{Iso}}({\mathcal{T}}^+,{{\widetilde{{\mathcal{T}}}}}^+)_{\Delta,{\widetilde{\Delta}}}\longrightarrow {\mathrm{Iso}}((\L^+,\Delta^+,S),({{\widetilde{\mathcal{L}}}}^+,{\widetilde{\Delta}}^+,{{\widetilde{S}}}))_{\Delta,{\widetilde{\Delta}}}.$$ Here $\Lambda$ is defined by $$\Lambda(\alpha)([{\varphi}])=[\alpha_{P,Q}({\varphi})]$$ for all $\alpha\in{\mathrm{Iso}}({\mathcal{T}},{{\widetilde{{\mathcal{T}}}}})_{\Delta^+,{\widetilde{\Delta}}^+}$, all $P,Q\in\Delta$ and all ${\varphi}\in{\mathrm{Iso}}_{\mathcal{T}}(P,Q)$, and $\Lambda^+$ is defined by $$\Lambda^+(\alpha)([{\varphi}]_+)=[\alpha_{P,Q}({\varphi})]_+$$ for all $\alpha\in{\mathrm{Iso}}({\mathcal{T}}^+,{{\widetilde{{\mathcal{T}}}}}^+)_{\Delta,{\widetilde{\Delta}}}$, all $P,Q\in\Delta^+$ and all ${\varphi}\in{\mathrm{Iso}}_{{\mathcal{T}}^+}(P,Q)$.
Now $\Psi\circ\Lambda^+$ is a bijection from ${\mathrm{Iso}}({\mathcal{T}}^+,{{\widetilde{{\mathcal{T}}}}}^+)_{\Delta,{\widetilde{\Delta}}}$ to ${\mathrm{Iso}}((\L^+|_\Delta,\Delta,S),({{\widetilde{\mathcal{L}}}}^+|_{{\widetilde{\Delta}}},{\widetilde{\Delta}},{{\widetilde{S}}}))_{\Delta^+,{\widetilde{\Delta}}^+}$ and $\Phi\circ\Lambda$ is a bijection from ${\mathrm{Iso}}({\mathcal{T}},{{\widetilde{{\mathcal{T}}}}})_{\Delta^+,{\widetilde{\Delta}}^+}$ to ${\mathrm{Iso}}((\L^+|_\Delta,\Delta,S),({{\widetilde{\mathcal{L}}}}^+|_{{\widetilde{\Delta}}},{\widetilde{\Delta}},{{\widetilde{S}}}))_{\Delta^+,{\widetilde{\Delta}}^+}$. Hence, $\Theta:=(\Phi\circ\Lambda)^{-1}\circ (\Psi\circ \Lambda^+)$ is a bijection from ${\mathrm{Iso}}({\mathcal{T}}^+,{{\widetilde{{\mathcal{T}}}}}^+)_{\Delta,{\widetilde{\Delta}}}$ to ${\mathrm{Iso}}({\mathcal{T}},{{\widetilde{{\mathcal{T}}}}})_{\Delta^+,{\widetilde{\Delta}}^+}$. Fixing $\alpha\in{\mathrm{Iso}}({\mathcal{T}}^+,{{\widetilde{{\mathcal{T}}}}}^+)_{\Delta,{\widetilde{\Delta}}}$, it only remains to show that $\Theta(\alpha)=\alpha|_{\mathcal{T}}$, or equivalently, $\Psi(\Lambda^+(\alpha))=\Phi(\Lambda(\alpha|_{\mathcal{T}}))$. To prove the latter equality, recall that $\iota\colon \L\longrightarrow\L^+|_\Delta$ is bijective. So every element of $\L^+|_\Delta$ is of the form $\iota([{\varphi}])=[{\varphi}]_+$ for some $P,Q\in\Delta$ and ${\varphi}\in{\mathrm{Iso}}_{\mathcal{T}}(P,Q)$. We compute then $$\Psi(\Lambda^+(\alpha))[{\varphi}]_+=\Lambda^+(\alpha)[{\varphi}]_+=[\alpha_{P,Q}({\varphi})]_+$$ and $$\begin{aligned}
\Phi(\Lambda(\alpha|_{\mathcal{T}}))[{\varphi}]_+ &=& ({{\tilde{\iota}}}\circ\Lambda(\alpha|_{\mathcal{T}})\circ \iota^{-1})[{\varphi}]_+\\
&=& ({{\tilde{\iota}}}\circ\Lambda(\alpha|_{\mathcal{T}}))[{\varphi}]\\
&=&{{\tilde{\iota}}}([\alpha_{P,Q}({\varphi})])=[\alpha_{P,Q}({\varphi})]_+. \end{aligned}$$ This proves the assertion.
The two preceding theorems seem most important in situations where we consider automorphisms of linking localities or linking systems. In the next two theorems we state the results for automorphisms explicitly.
\[T:MainAut\] Let $(\L,\Delta,S)$ and $(\L^+,\Delta^+,S)$ be linking localities over the same fusion system ${\mathcal{F}}$ such that $\Delta\subseteq\Delta^+$ and $\L=\L^+|_\Delta$. Then the map $$\Psi\colon{\mathrm{Aut}}(\L^+,\Delta^+,S)_\Delta\longrightarrow {\mathrm{Aut}}(\L,\Delta,S)_{\Delta^+},\;\gamma\mapsto\gamma|_\L$$ is well-defined and an isomorphism of groups. In particular, if $\Delta$ and $\Delta^+$ are ${\mathrm{Aut}}({\mathcal{F}})$-invariant, then the map ${\mathrm{Aut}}(\L^+,\Delta^+,S)\longrightarrow{\mathrm{Aut}}(\L,\Delta,S),\;\gamma\mapsto\gamma|_\L$ is an isomorphism of groups.
By Theorem \[T:MainIso\], the map $\Psi$ is well-defined and a bijection. Moreover, if $\beta,\gamma\in{\mathrm{Aut}}(\L^+,\Delta^+,S)_\Delta$, then $\Psi(\beta\gamma)=(\beta\gamma)|_\L=(\beta|_\L)(\gamma|_\L)=\Psi(\beta)\Psi(\gamma)$. Hence, $\Psi$ is an isomorphism of groups. By Lemma \[L:ProjectionLocalityMorphismFusionSystem\], if $\alpha$ is an element of ${\mathrm{Aut}}(\L,\Delta,S)$ or of ${\mathrm{Aut}}(\L^+,\Delta^+,S)$, then $\alpha|_S\in{\mathrm{Aut}}({\mathcal{F}})$. Hence, if $\Delta$ and $\Delta^+$ are ${\mathrm{Aut}}({\mathcal{F}})$-invariant, then ${\mathrm{Aut}}(\L^+,\Delta^+,S)_\Delta={\mathrm{Aut}}(\L^+,\Delta^+,S)$ and ${\mathrm{Aut}}(\L,\Delta,S)_{\Delta^+}={\mathrm{Aut}}(\L,\Delta,S)$. This yields the assertion.
\[T:MainAutTrans\] Let ${\mathcal{T}}$ and ${\mathcal{T}}^+$ be linking systems associated to the same fusion system ${\mathcal{F}}$ such that ${\mathrm{Ob}}({\mathcal{T}})\subseteq{\mathrm{Ob}}({\mathcal{T}}^+)$ and ${\mathcal{T}}={\mathcal{T}}^+|_{{\mathrm{Ob}}({\mathcal{T}})}$. Then the map $$\Theta\colon {\mathrm{Aut}}({\mathcal{T}}^+)_{{\mathrm{Ob}}({\mathcal{T}})}\longrightarrow{\mathrm{Aut}}({\mathcal{T}})_{{\mathrm{Ob}}({\mathcal{T}}^+)},\;\alpha\mapsto \alpha|_{\mathcal{T}}$$ is an isomorphism of groups. In particular, if ${\mathrm{Ob}}({\mathcal{T}})$ and ${\mathrm{Ob}}({\mathcal{T}}^+)$ are ${\mathrm{Aut}}({\mathcal{F}})$-invariant, then the map ${\mathrm{Aut}}({\mathcal{T}}^+)\longrightarrow{\mathrm{Aut}}({\mathcal{T}}),\;\alpha\mapsto\alpha|_{\mathcal{T}}$ is an isomorphism of groups.
By Theorem \[T:MainIsoT\], $\Theta$ is a bijection, and it is easy to to see that $\Theta$ is an isomorphism of groups. As explained in Remark \[R:Transporter\](b), if ${\mathrm{Ob}}({\mathcal{T}})$ and ${\mathrm{Ob}}({\mathcal{T}}^+)$ are ${\mathrm{Aut}}({\mathcal{F}})$-invariant, then ${\mathrm{Aut}}({\mathcal{T}})_{{\mathrm{Ob}}({\mathcal{T}}^+)}={\mathrm{Aut}}({\mathcal{T}})$ and ${\mathrm{Aut}}({\mathcal{T}}^+)_{{\mathrm{Ob}}({\mathcal{T}})}={\mathrm{Aut}}({\mathcal{T}}^+)$.
Let $(\L,\Delta,S)$ and $(\L^+,\Delta^+,S)$ be linking localities over the same fusion system ${\mathcal{F}}$ such that $\Delta$ and $\Delta^+$ are ${\mathrm{Aut}}({\mathcal{F}})$-invariant. By [@subcentric Theorem A(b)], there exists a subcentric linking locality $(\L^s,{\mathcal{F}}^s,S)$ over ${\mathcal{F}}$. As ${\mathcal{F}}^{cr}\subseteq\Delta$ and ${\mathcal{F}}^{cr}\subseteq\Delta^+$, it follows that $(\L^s|_\Delta,\Delta,S)$ and $(\L^s|_{\Delta^+},\Delta^+,S)$ are linking localities over ${\mathcal{F}}$. Hence, by [@subcentric Theorem A(a)], there exist rigid isomorphisms from $(\L^s|_\Delta,\Delta,S)$ to $(\L,\Delta,S)$ and from $(\L^s|_{\Delta^+},\Delta^+,S)$ to $(\L^+,\Delta^+,S)$. By [@subcentric Lemma 3.6], ${\mathcal{F}}^s$ is ${\mathrm{Aut}}({\mathcal{F}})$-invariant. Hence, applying Theorem \[T:MainAut\] twice, we obtain $${\mathrm{Aut}}(\L,\Delta,S)\cong{\mathrm{Aut}}(\L^s|_\Delta,\Delta,S)\cong{\mathrm{Aut}}(\L^s,{\mathcal{F}}^s,S)\cong{\mathrm{Aut}}(\L^s|_{\Delta^+},\Delta^+,S)\cong {\mathrm{Aut}}(\L^+,\Delta^+,S).$$ So Theorem \[T:A1\] follows from Theorem \[T:MainAut\].
We will prove Theorem \[T:A2\] together with the following similar statement about outer automorphism groups, which is a generalization of [@AOV1 Lemma 1.17].
\[T:A2Outer\] If ${\mathcal{T}}$ and ${\mathcal{T}}^+$ are linking system associated to ${\mathcal{F}}$ such that ${\mathrm{Ob}}({\mathcal{T}})$ and ${\mathrm{Ob}}({\mathcal{T}}^+)$ are ${\mathrm{Aut}}({\mathcal{F}})$-invariant, then $${\mathrm{Out}}_{{\operatorname{typ}}}({\mathcal{T}}^+)\cong {\mathrm{Out}}_{{\operatorname{typ}}}({\mathcal{T}}).$$ If ${\mathrm{Ob}}({\mathcal{T}})\subseteq{\mathrm{Ob}}({\mathcal{T}}^+)$ and ${\mathcal{T}}={\mathcal{T}}^+|_{{\mathrm{Ob}}({\mathcal{T}})}$, then an isomorphism ${\mathrm{Out}}_{{\operatorname{typ}}}({\mathcal{T}}^+)\xrightarrow{\;\;\cong\;\;}{\mathrm{Out}}_{{\operatorname{typ}}}({\mathcal{T}})$ is given by sending the class of $\alpha\in{\mathrm{Aut}}({\mathcal{T}}^+)$ to the class of $\alpha|_{\mathcal{T}}\in{\mathrm{Aut}}({\mathcal{T}})$.
Let ${\mathcal{T}}$ and ${\mathcal{T}}^+$ be transporter systems over the same fusion system ${\mathcal{F}}$ such that ${\mathrm{Ob}}({\mathcal{T}})$ and ${\mathrm{Ob}}({\mathcal{T}}^+)$ are ${\mathrm{Aut}}({\mathcal{F}})$-invariant. As usual when dealing with transporter systems, we write maps on the left side of the argument.
If ${\mathrm{Ob}}({\mathcal{T}})\subseteq{\mathrm{Ob}}({\mathcal{T}}^+)$ and ${\mathcal{T}}={\mathcal{T}}^+|_{{\mathrm{Ob}}({\mathcal{T}})}$, then by Theorem \[T:MainAutTrans\], the map ${\mathrm{Aut}}({\mathcal{T}}^+)\rightarrow{\mathrm{Aut}}({\mathcal{T}}),\gamma\mapsto\gamma|_{\mathcal{T}}$ is a group isomorphism. As ${\mathrm{Aut}}_{\mathcal{T}}(S)={\mathrm{Aut}}_{{\mathcal{T}}^+}(S)$, one easily observes that it induces an isomorphism between the group of inner automorphisms of ${\mathcal{T}}^+$ and the group of inner automorphisms of ${\mathcal{T}}$. Hence, by Lemma \[L:ExactSequence\], it induces an isomorphism ${\mathrm{Out}}_{{\operatorname{typ}}}({\mathcal{T}}^+)\rightarrow {\mathrm{Out}}_{{\operatorname{typ}}}({\mathcal{T}})$ which takes the class of $\alpha\in{\mathrm{Aut}}({\mathcal{T}}^+)$ to the class of $\alpha|_{\mathcal{T}}$.
Suppose now that ${\mathcal{T}}$ and ${\mathcal{T}}^+$ are arbitrary. By [@subcentric Theorem A], there exists a subcentric linking system ${\mathcal{T}}^s$ over ${\mathcal{F}}$; moreover, ${\mathcal{T}}$ is rigidly isomorphic to ${\mathcal{T}}^s|_{{\mathrm{Ob}}({\mathcal{T}})}$, and ${\mathcal{T}}^+$ is rigidly isomorphic to ${\mathcal{T}}^s|_{{\mathrm{Ob}}({\mathcal{T}}^+)}$. If $\alpha\colon{\mathcal{T}}\rightarrow{\mathcal{T}}^s|_{{\mathrm{Ob}}({\mathcal{T}})}$ is a rigid isomorphism, then the map $\Phi\colon{\mathrm{Aut}}({\mathcal{T}})\rightarrow{\mathrm{Aut}}({\mathcal{T}}^s|_{{\mathrm{Ob}}({\mathcal{T}})}),\beta\mapsto \alpha\circ\beta\circ\alpha^{-1}$ is an isomorphism of groups and so ${\mathrm{Aut}}({\mathcal{T}})\cong {\mathrm{Aut}}({\mathcal{T}}^s|_{{\mathrm{Ob}}({\mathcal{T}})})$. One can check now that, for any $\gamma\in{\mathrm{Aut}}({\mathcal{T}})$, we have $\alpha\circ c_\gamma\circ\alpha^{-1}=c_{\alpha_S(\gamma)}$; to see that $\alpha\circ c_\gamma\circ\alpha^{-1}$ and $c_{\alpha_S(\gamma)}$ agree on objects, one uses that ${{\tilde{\pi}}}\circ \alpha=\pi$ (cf. Remark \[R:Transporter\](d)), and to see that the two functors agree on morphisms, one uses that $\alpha$ takes inclusions to inclusions and thus commutes with taking restrictions. So $\Phi$ induces an isomorphism between the group of inner automorphisms of ${\mathcal{T}}$ and the group of inner automorphisms of ${\mathcal{T}}^s|_{{\mathrm{Ob}}({\mathcal{T}})}$. Thus, by Lemma \[L:ExactSequence\], ${\mathrm{Out}}_{{\operatorname{typ}}}({\mathcal{T}})\cong {\mathrm{Out}}_{{\operatorname{typ}}}({\mathcal{T}}^s|_{{\mathrm{Ob}}({\mathcal{T}})})$. Similarly, one shows that ${\mathrm{Aut}}({\mathcal{T}}^+)\cong {\mathrm{Aut}}({\mathcal{T}}^s|_{{\mathrm{Ob}}({\mathcal{T}}^+)})$ and ${\mathrm{Out}}_{{\operatorname{typ}}}({\mathcal{T}}^+)\cong{\mathrm{Out}}_{{\operatorname{typ}}}({\mathcal{T}}^s|_{{\mathrm{Ob}}({\mathcal{T}}^+)})$. So using Theorem \[T:MainAutTrans\] twice, we can conclude that $${\mathrm{Aut}}({\mathcal{T}})\cong {\mathrm{Aut}}({\mathcal{T}}^s|_{{\mathrm{Ob}}({\mathcal{T}})})\cong {\mathrm{Aut}}({\mathcal{T}}^s)\cong{\mathrm{Aut}}({\mathcal{T}}^s|_{{\mathrm{Ob}}({\mathcal{T}}^+)})\cong {\mathrm{Aut}}({\mathcal{T}}^+).$$ and similarly $${\mathrm{Out}}_{{\operatorname{typ}}}({\mathcal{T}})\cong {\mathrm{Out}}_{{\operatorname{typ}}}({\mathcal{T}}^s|_{{\mathrm{Ob}}({\mathcal{T}})})\cong {\mathrm{Out}}_{{\operatorname{typ}}}({\mathcal{T}}^s)\cong{\mathrm{Out}}_{{\operatorname{typ}}}({\mathcal{T}}^s|_{{\mathrm{Ob}}({\mathcal{T}}^+)})\cong {\mathrm{Out}}_{{\operatorname{typ}}}({\mathcal{T}}^+).$$
\[R:AOV\] Theorem \[T:A2\] and Theorem \[T:A2Outer\] were shown for linking systems whose objects are quasicentric in [@AOV1 Lemma 1.17] and its proof via more direct arguments. As we will briefly indicate now, the proof could be adapted to give a proof of Theorems \[T:A2\] and \[T:A2Outer\], which does not use linking localities:
- Using the notation in the proof of [@AOV1 Lemma 1.17], Lemma \[L:LinkingSystemsElementary\](b) is needed to conclude that $P$ is properly contained in $\hat{P}$. The reference to [@AOV1 Theorem 1.12] needs to be replaced by a reference to Lemma \[L:LinkingSystemsElementary\](c).
- The references to Proposition 1.11(b),(b’) and Proposition 1.11(d) in [@AOV1] need to be replaced by references to Lemma 3.2(c) and Proposition 3.4(a) in [@OV] respectively.
- References to [@AOV1 Proposition 1.11(e)] could be replaced by references to Axiom II in the definition of a transporter system [@OV Definition 3.1] and to [@OV Lemma 3.3].
- The reference to [@AOV1 Lemma 1.15] can be replaced by a reference to [@GL2020 Proposition 2.5] (cf. Remark \[R:Transporter\](d)).
It seems that the arguments could also be adapted to give direct proofs of the more general Theorems \[T:MainIsoT\] and \[T:MainAutTrans\].
If ${\mathcal{T}}$ and ${\mathcal{T}}^+$ are linking systems associated to the same saturated fusion system ${\mathcal{F}}$ such that ${\mathrm{Ob}}({\mathcal{T}})\subseteq{\mathrm{Ob}}({\mathcal{T}}^+)$ and ${\mathcal{T}}={\mathcal{T}}^+|_{{\mathrm{Ob}}({\mathcal{T}})}$, then by [@subcentric Theorem A], the inclusion map $\iota\colon {\mathcal{T}}\hookrightarrow{\mathcal{T}}^+$ induces a homotopy equivalence $|\iota|\colon |{\mathcal{T}}|\rightarrow |{\mathcal{T}}^+|$ and thus a homotopy equivalence $|\iota|^\wedge_p\colon |{\mathcal{T}}|^\wedge_p\rightarrow |{\mathcal{T}}^+|^\wedge_p$. Moreover, if ${\mathrm{Ob}}({\mathcal{T}})$ and ${\mathrm{Ob}}({\mathcal{T}}^+)$ are ${\mathrm{Aut}}({\mathcal{F}})$-invariant and $\gamma\in{\mathrm{Aut}}({\mathcal{T}}^+)$, then the commutative square $$\xymatrix{
{\mathcal{T}}^+ \ar[r]^{\gamma} & {\mathcal{T}}^+\\
{\mathcal{T}}\ar@{^{(}->}[u]^{\iota} \ar[r]^{\gamma|_{\mathcal{T}}} & {\mathcal{T}}\ar@{^{(}->}[u]_{\iota}
}$$ induces a commutative square after applying the functor $|\cdot|^\wedge_p$. Thus, if ${\mathrm{Ob}}({\mathcal{T}})$ and ${\mathrm{Ob}}({\mathcal{T}}^+)$ are ${\mathrm{Aut}}({\mathcal{F}})$-invariant, then by Theorem \[T:A2Outer\], the conclusion of Corollary \[C:B\] is true for ${\mathcal{T}}$ if and only if it is true with ${\mathcal{T}}^+$ in place of ${\mathcal{T}}$.
Suppose now ${\mathcal{T}}$ is an arbitrary linking system associated to ${\mathcal{F}}$ such that ${\mathrm{Ob}}({\mathcal{T}})$ is ${\mathrm{Aut}}({\mathcal{F}})$-invariant. By [@subcentric Theorem A], there exists a subcentric linking system ${\mathcal{T}}^s$ associated to ${\mathcal{F}}$ such that ${\mathcal{T}}^s|_{{\mathrm{Ob}}({\mathcal{T}})}={\mathcal{T}}$; moreover, ${\mathcal{T}}^c:={\mathcal{T}}^s|_{{\mathcal{F}}^c}$ is a centric linking system associated to ${\mathcal{F}}$. By [@BLO2 Theorem 8.1] and its proof, the statement in Theorem \[C:B\] is true for ${\mathcal{T}}^c$ in place of ${\mathcal{T}}$. The object sets ${\mathrm{Ob}}({\mathcal{T}}^c)={\mathcal{F}}^c$ and ${\mathrm{Ob}}({\mathcal{T}}^s)={\mathcal{F}}^s$ are ${\mathrm{Aut}}({\mathcal{F}})$-invariant (cf. [@subcentric Lemma 3.6]). Hence, as remarked above, the conclusion of Theorem \[C:B\] is true for ${\mathcal{T}}^s$ and thus also for ${\mathcal{T}}$.
Partial normal subgroups {#S:PartialNormal}
========================
This section is mainly devoted to the proof of Theorem \[T:MainChermakII\]. We will however start in the first subsection with some background on partial normal subgroups of localities. Most importantly, we prove with Lemma \[L:PartialNormalAlperin\] a result which seems to be of general interest and can be considered as a version of Alperin’s Fusion Theorem for partial normal subgroups. This lemma is also applied in [@normal]. Using Lemma \[L:PartialNormalAlperin\] we will then prove Theorem \[T:MainChermakII\] and a corollary in Subsection \[SS:PartialNormalLinking\].
General results {#SS:PartialNormal}
---------------
If $\alpha\colon \L\longrightarrow{{\widetilde{\mathcal{L}}}}$ is a homomorphism of partial groups, then by [@loc1 Lemma 1.14], $\ker(\alpha)$ is a partial normal subgroup of $\L$. The other way around, if $(\L,\Delta,S)$ is a locality and ${\mathcal{N}}$ is a partial normal subgroup of $\L$, then one can construct a partial group $\L/{\mathcal{N}}$ and a projection of partial groups $$\alpha\colon\L\longrightarrow \L/{\mathcal{N}}$$ with $\ker(\alpha)={\mathcal{N}}$. We refer the reader to Lemma 3.16 and the preceding explanations in [@loc1 Section 3] for details of the construction. We will often adopt a “bar notation” similarly as for groups. This means that, setting $\ov{\L}=\L/{\mathcal{N}}$, for every element or subset $X$ of $\L$, we write $\ov{X}$ for the image of $X$ in $\ov{\L}$ under $\alpha$. Moreover, for any set $\Gamma$ of subgroups of $\L$, we set $\ov{\Gamma}:=\{\ov{P}\colon P\in\Gamma\}$. By [@loc1 Corollay 4.5], $(\ov{\L},\ov{\Delta},\ov{S})$ is a locality and $\alpha$ is a projection of localities from $(\L,\Delta,S)$ to $(\ov{\L},\ov{\Delta},\ov{S})$.
\[ProductPrepare\] Let $(\L,\Delta,S)$ be a locality with a partial normal subgroup ${\mathcal{N}}$. Then the following hold:
- The triple $({\mathcal{N}}S,\Delta,S)$ is a locality.
- For every $P\in\Delta$, we have $O^p(N_{{\mathcal{N}}S}(P))=O^p(N_{\mathcal{N}}(P))$.
Part (a) is true by [@loc1 Lemma 4.1]. In particular, $N_{{\mathcal{N}}S}(P)$ is a subgroup of $\L$. Moreover, we may consider the canonical projection $\alpha\colon {\mathcal{N}}S\longrightarrow \ov{{\mathcal{N}}S}:={\mathcal{N}}S/{\mathcal{N}}$. Then $\ov{{\mathcal{N}}S}=\ov{S}$ is a $p$-group and $\alpha$ induces a group homomorphism $\alpha|_{N_{{\mathcal{N}}S}(P)}\colon N_{{\mathcal{N}}S}(P)\longrightarrow \ov{S}$. Thus, $O^p(N_{{\mathcal{N}}S}(P))\leq \ker(\alpha|_{N_{{\mathcal{N}}S}(P)})=\ker(\alpha)\cap N_{{\mathcal{N}}S}(P)=N_{\mathcal{N}}(P)$. This implies (b).
\[L:PartialNormalAlperin\] Let $(\L,\Delta,S)$ be a locality. If ${\mathcal{N}}$ is a partial normal subgroup of $\L$ and $n\in{\mathcal{N}}$, then there exist $k\in\mathbb{N}$, $R_1,R_2,\dots,R_k\in \Delta$ and $(t,n_1,n_2,\dots,n_k)\in{\mathbf{D}}$ such that the following hold:
- $S_n=S_{(t,n_1,\dots,n_k)}$ and $n=tn_1n_2\cdots n_k$;
- $n_i\in O^p(N_{\mathcal{N}}(R_i))$, $S_{n_i}=R_i$, $O_p(N_{{\mathcal{N}}S}(R_i))=R_i$ and $N_S(R_i)\in{\mathrm{Syl}}_p(N_{{\mathcal{N}}S}(R_i))$ for all $i=1,\dots,k$; and
- $t\in T$.
By Lemma \[ProductPrepare\](a), $({\mathcal{N}}S,\Delta,S)$ is a locality. So by Alperin’s fusion theorem for localities [@Molinier:2016 Theorem 2.5], there exist $k\in\mathbb{N}$, $Q_1,Q_2,\dots,Q_k\in \Delta$ and $(g_1,g_2,\dots,g_k)\in{\mathbf{D}}$ such that the following hold:
- $S_n=S_{(g_1,\dots,g_k)}$ and $n=g_1g_2\cdots g_k$;
- $g_i\in N_{{\mathcal{N}}S}(Q_i)$, $S_{g_i}=Q_i$, $O_p(N_{{\mathcal{N}}S}(Q_i))=Q_i$ and $N_S(Q_i)\in{\mathrm{Syl}}_p(N_{{\mathcal{N}}S}(Q_i))$ for all $i=1,\dots,k$.
As $N_S(Q_i)\in{\mathrm{Syl}}_p(N_{{\mathcal{N}}S}(Q_i))$, it follows from Lemma \[ProductPrepare\](b) that $N_{{\mathcal{N}}S}(Q_i)=N_S(Q_i)O^p(N_{\mathcal{N}}(Q_i))$. So for all $i=1,\dots,n$, we can write $g_i=s_im_i$ with $s_i\in N_S(Q_i)$ and $m_i\in O^p(N_{\mathcal{N}}(Q_i))$. By Lemma \[L:NLSbiset\], we have $Q_i=S_{g_i}=S_{(s_i,m_i)}$ for all $i=1,\dots,k$. In particular, $S_n=S_{(g_1,g_2,\dots,g_k)}=S_{(s_1,m_1,s_2,m_2,\dots,s_k,m_k)}$ and $$w:=(s_1,m_1,s_2,m_2,\dots,s_k,m_k)\in{\mathbf{D}}\mbox{ via }S_n.$$ Note also $n=\Pi(g_1,\dots,g_n)=\Pi(w)$. Set $$x_i:=s_{i}s_{i+1}\dots s_k\mbox{ for all }1\leq i\leq k$$ and $$n_i:=m_i^{x_{i+1}}\mbox{ for all }1\leq i<k\mbox{ and }n_k=m_k.$$ Notice that $$x_ix_{i+1}^{-1}=s_i\mbox{ for }1\leq i<k\mbox{ and }x_k=s_k.$$ Since $w\in{\mathbf{D}}$ via $S_n$, it follows that $$v:=(x_1,x_2^{-1},m_1,x_2,x_3^{-1},m_2,x_3,\dots,x_k^{-1},m_{k-1},x_k,m_k)\in{\mathbf{D}}\mbox{ via }S_n$$ and so, setting $t:=x_1$, also $$u:=(t,n_1,n_2,\dots,n_{k-1},n_k)\in{\mathbf{D}}\mbox{ via }S_n.$$ Observe moreover that, by axiom (PG3), we have $$n=\Pi(w)=\Pi(v)=\Pi(u).$$ Using Lemma \[L:LocalitiesProp\](f), it follows $S_n=S_u$. So (i) holds. Set $$R_i:=Q_i^{x_{i+1}}\mbox{ for }1\leq i<k\mbox{ and }R_k:=Q_k.$$ As $m_i\in O^p(N_{\mathcal{N}}(Q_i))$, $O_p(N_{{\mathcal{N}}S}(Q_i))=Q_i$ and $N_S(Q_i)\in{\mathrm{Syl}}_p(N_{{\mathcal{N}}S}(Q_i))$, Lemma \[L:LocalitiesProp\](b) gives that $n_i\in O^p(N_{\mathcal{N}}(R_i))$, $O_p(N_{{\mathcal{N}}S}(R_i))=R_i$ and $N_S(R_i)\in{\mathrm{Syl}}_p(N_{{\mathcal{N}}S}(R_i))$ for $i=1,\dots,k$. Moreover, since $Q_i=S_{m_i}$ and $(m_i,Q_i)$ is conjugate to $(n_i,R_i)$ under $x_{i+1}\in S$, we have $R_i=S_{n_i}$ by Lemma \[L:NLSbiset\]. So (ii) holds. Note now that $u':=(t^{-1},t,n_1,\dots,n_k,n^{-1})\in{\mathbf{D}}$ via $S_n^t$ and $t^{-1}=\Pi(t^{-1},n,n^{-1})=\Pi(u')=\Pi(n_1,\dots,n_k,n^{-1})\in S\cap{\mathcal{N}}=T$. Hence, $t\in T$ and the proof is complete.
Partial normal subgroups of linking localities {#SS:PartialNormalLinking}
----------------------------------------------
In this subsection, we will first prove Theorem \[T:MainChermakII\]. Afterwards, we prove as a corollary that any two linking localities over the same fusion system have the same number of partial normal subgroups. Against our usual convention, we will use the left hand notation for the map $\Phi_{\L^+,\L}$ from Theorem \[T:MainChermakII\]. Recall that ${{\mathfrak{N}}}(\L)$ denotes the set of partial normal subgroups of a partial groups $\L$. We first show the following lemma.
\[L:ChermakIIHelp\] Let $(\L,\Delta,S)$ and $(\L^+,\Delta^+,S)$ be linking localities over the same fusion system ${\mathcal{F}}$ such that $\Delta\subseteq\Delta^+$ and $\L=\L^+|_\Delta$. Assume that every proper overgroup of an element of $\Delta^+\backslash \Delta$ is in $\Delta$. Let ${\mathcal{N}}^+\in{{\mathfrak{N}}}(\L^+)$, ${\mathcal{N}}:={\mathcal{N}}^+\cap \L\in{{\mathfrak{N}}}(\L)$ and set $T:={\mathcal{N}}^+\cap S={\mathcal{N}}\cap S$. Then the following hold.
- We have ${\mathcal{N}}^+={\langle}{\mathcal{N}}^{\L^+}{\rangle}$, where ${\langle}{\mathcal{N}}^{\L^+}{\rangle}$ denotes the smallest partial subgroup of $\L^+$ containing all the elements of the form $n^f$ with $f\in\L^+$ and $n\in{\mathcal{N}}\cap{\mathbf{D}}^+(f)$.
- If ${\mathcal{F}}_{S\cap{\mathcal{N}}}({\mathcal{N}})$ is ${\mathcal{F}}$-invariant, then ${\mathcal{F}}_T({\mathcal{N}})={\mathcal{F}}_T({\mathcal{N}}^+)$.
- Let ${\mathcal{K}}^+\in{{\mathfrak{N}}}(\L^+)$ and ${\mathcal{K}}:={\mathcal{K}}^+\cap\L\in{{\mathfrak{N}}}(\L)$. Then ${\mathcal{K}}^+T={\mathcal{N}}^+$ if and only if ${\mathcal{K}}T={\mathcal{N}}$.
Observe that, for any $k\in \L^+$ and $t\in S$, we have $S_{kt}=S_{(k,t)}=S_k$. Hence, $kt\in\L$ if and only if $k\in\L$. With ${\mathcal{K}}^+$ and ${\mathcal{K}}$ as in (c), it follows ${\mathcal{K}}^+T\cap\L=({\mathcal{K}}^+\cap\L)T={\mathcal{K}}T$. Hence, if ${\mathcal{K}}^+ T={\mathcal{N}}^+$, then ${\mathcal{N}}={\mathcal{N}}^+\cap\L=({\mathcal{K}}^+ T)\cap\L={\mathcal{K}}T$. On the other hand, if ${\mathcal{K}}T={\mathcal{N}}$ and (a) holds, then ${\mathcal{K}}\subseteq{\mathcal{N}}$ and so ${\mathcal{K}}^+={\langle}{\mathcal{K}}^{\L^+}{\rangle}\subseteq{\langle}{\mathcal{N}}^{\L^+}{\rangle}={\mathcal{N}}^+$. Thus, since ${\mathcal{N}}^+$ is a partial subgroup, it follows in this case ${\mathcal{K}}^+T\subseteq{\mathcal{N}}^+$. Hence, it remains to prove (a), (b) and the following property: $$\label{E:Provec}
\mbox{If ${\mathcal{K}}^+$ and ${\mathcal{K}}$ are as in (c) and ${\mathcal{K}}T={\mathcal{N}}$, then }{\mathcal{N}}^+\subseteq {\mathcal{K}}^+ T.$$ Set ${\mathcal{E}}:={\mathcal{F}}_T({\mathcal{N}})$. As ${\mathcal{N}}\subseteq{\mathcal{N}}^+\unlhd\L^+$, we have ${\langle}{\mathcal{N}}^{\L^+}{\rangle}\subseteq{\mathcal{N}}^+$. Moreover, clearly ${\mathcal{E}}={\mathcal{F}}_T({\mathcal{N}})\subseteq{\mathcal{F}}_T({\mathcal{N}}^+)$. So fixing $n\in{\mathcal{N}}^+$, we need to show that $n\in{\langle}{\mathcal{N}}^{\L^+}{\rangle}$ and, if ${\mathcal{E}}$ is ${\mathcal{F}}$-invariant, then $c_n\colon S_n\cap T\longrightarrow T$ is a morphism in ${\mathcal{E}}$. Furthermore, fixing ${\mathcal{K}}^+$ and ${\mathcal{K}}$ are as in (c) such that ${\mathcal{K}}T={\mathcal{N}}$, we need to show $n\in {\mathcal{K}}^+ T$.
As $S_{(k,t)}=S_{kt}$ for all $k\in{\mathcal{K}}^+$ and $t\in T$, using the Frattini calculus [@loc1 Lemma 3.4], one sees that ${\mathcal{K}}^+ T=T{\mathcal{K}}^+$ is a subgroup of $\L^+$. So by Lemma \[L:PartialNormalAlperin\] applied with $\L^+$ and ${\mathcal{N}}^+$ in place of $\L$ and ${\mathcal{N}}$, we may assume that $P=S_n\in\Delta^+$ and $n\in O^p(N_{{\mathcal{N}}^+}(P))$. If $P\in\Delta$, then $N_{\L^+}(P)=N_\L(P)$ and $n\in N_{{\mathcal{N}}^+}(P)=N_{\mathcal{N}}(P)$ since ${\mathcal{N}}^+\cap\L={\mathcal{N}}$. So in this case, $n\in{\langle}{\mathcal{N}}^{\L^+}{\rangle}$, the conjugation homomorphism $c_n|_{P\cap T}$ is a morphism in ${\mathcal{E}}$, and $n\in {\mathcal{N}}={\mathcal{K}}T\subseteq {\mathcal{K}}^+ T$.
Suppose now that $P\in \Delta^+\backslash\Delta$. Then by Lemma \[L:LocalityoverFFnatural\](a), there exists $f\in\L^+$ such that $P\leq S_f$ and $R:=P^f$ is fully ${\mathcal{F}}$-normalized. By Lemma \[L:LocalitiesProp\](b), the conjugation map $c_f\colon N_{\L^+}(P)\longrightarrow N_{\L^+}(R)$ is defined and an isomorphism of groups. In particular, $n^f\in O^p(N_{{\mathcal{N}}^+}(P))^f\subseteq O^p(N_{{\mathcal{N}}^+}(R))=O^p(N_{\mathcal{N}}(R))$, where the last equality uses $\L\cap{\mathcal{N}}^+={\mathcal{N}}$ and $N_{\L^+}(R)=N_\L(R)$ by Lemma \[L:LinkingLocN(R)\]. Hence, using Lemma \[L:LocalitiesProp\](c), we see that $n=(n^f)^{f^{-1}}\subseteq N_{\mathcal{N}}(R)^{f^{-1}}\subseteq{\langle}{\mathcal{N}}^{\L^+}{\rangle}$ proving (a).
As ${\mathcal{N}}={\mathcal{K}}T\subseteq{\mathcal{K}}S$, Lemma \[ProductPrepare\](b) applied with ${\mathcal{K}}$ in place of ${\mathcal{N}}$ gives that $n^f\in O^p(N_{\mathcal{N}}(R))\subseteq O^p(N_{{\mathcal{K}}S}(R))=O^p(N_{\mathcal{K}}(R))$. Hence, we have $n=(n^f)^{f^{-1}}\in {\langle}{\mathcal{K}}^{\L^+}{\rangle}\subseteq{\mathcal{K}}^+\subseteq{\mathcal{K}}^+ T$ proving and thus (c).
For the proof of (b) note that $c_{n^f}|_{R\cap T}\in{\mathrm{Aut}}_{\mathcal{E}}(R\cap T)$. Define ${\varphi}:=c_f|_{P\cap T}\in{\mathrm{Hom}}_{\mathcal{F}}(P\cap T,R\cap T)$. For every $x\in R$, we have $(f^{-1},n,f,x,f^{-1},n,f)\in{\mathbf{D}}^+$ via $R$, and so $x^{n^f}=((x^{f^{-1}})^n)^f$. Hence, ${\varphi}^{-1}(c_n|_{P\cap T}){\varphi}=c_{n^f}|_{R\cap T}\in{\mathrm{Aut}}_{\mathcal{E}}(R\cap T)$. If ${\mathcal{E}}$ is ${\mathcal{F}}$-invariant, using the characterization of ${\mathcal{F}}$-invariant subsystems given in [@AKO Proposition I.6.4(d)], we can conclude that $c_n|_{P\cap T}={\varphi}(c_{n^f}|_{R\cap T}) {\varphi}^{-1}\in{\mathrm{Aut}}_{\mathcal{E}}(P\cap T)$. This shows (b) and completes the proof.
For every partial normal subgroup ${\mathcal{N}}^+$ of $\L^+$, it is easy to see that the intersection ${\mathcal{N}}^+\cap\L$ is a partial normal subgroup of $\L$. Hence, the map $$\Phi_{\L^+,\L}\colon{{\mathfrak{N}}}(\L^+)\longrightarrow {{\mathfrak{N}}}(\L),\;{\mathcal{N}}^+\mapsto {\mathcal{N}}^+\cap\L$$ is well-defined. Moreover, this map is clearly inclusion preserving.
Without loss of generality, assume that $\Delta\neq\Delta^+$. Let $R\in\Delta^+\backslash \Delta$ be of maximal order. As $\Delta^+$ and $\Delta$ are closed under ${\mathcal{F}}$-conjugacy, $\Delta^*:=\Delta\cup R^{\mathcal{F}}$ is closed under ${\mathcal{F}}$-conjugacy and contained in $\Delta^+$. If $P$ is a proper overgroup of an element of $R^{\mathcal{F}}$, then $P\in\Delta^+$ as $\Delta^+$ is overgroup closed, so the maximality of $|R|$ yields $P\in\Delta$. Since $\Delta$ is overgroup closed, this shows that $\Delta^*$ is ${\mathcal{F}}$-closed and $\L^*:=\L^+|_{\Delta^*}$ is well-defined. Notice that ${\mathcal{F}}^{cr}\subseteq\Delta\subseteq \Delta^*$ and $N_{\L^*}(P)=N_{\L^+}(P)$ is of characteristic $p$ for every $P\in\Delta^*$. Therefore, $(\L^*,\Delta^*,S)$ is a linking locality over ${\mathcal{F}}$. So similarly, we have maps $$\Phi_{\L^+,\L^*}\colon {{\mathfrak{N}}}(\L^+)\longrightarrow{{\mathfrak{N}}}(\L^*),\;{\mathcal{N}}^+\mapsto {\mathcal{N}}^+\cap\L^*$$ and $$\Phi_{\L^*,\L}\colon {{\mathfrak{N}}}(\L^*)\longrightarrow{{\mathfrak{N}}}(\L),\;{\mathcal{N}}^*\mapsto {\mathcal{N}}^*\cap\L$$ defined. Notice that $\Phi_{\L^+,\L}=\Phi_{\L^*,\L}\circ \Phi_{\L^+,\L^*}$. By induction on $|\Delta^+\backslash\Delta|$, we may assume that the assertion is true with $(\L^*,\Delta^*,S)$ in place of $(\L,\Delta,S)$. That means that $\Phi_{\L^+,\L^*}$ is a bijection such that $\Phi_{\L^+,\L^*}^{-1}$ is inclusion preserving; moreover, given ${\mathcal{N}}^+\unlhd \L^+$ and ${\mathcal{N}}^*={\mathcal{N}}^+\cap\L^*\unlhd\L^*$ such that ${\mathcal{F}}_{S\cap{\mathcal{N}}^*}({\mathcal{N}}^*)$ is normal in ${\mathcal{F}}$, we have ${\mathcal{F}}_{S\cap{\mathcal{N}}^*}({\mathcal{N}}^*)={\mathcal{F}}_{S\cap{\mathcal{N}}^+}({\mathcal{N}}^+)$; also, if ${\mathcal{N}}^+,{\mathcal{K}}^+\in{{\mathfrak{N}}}(\L^+)$, ${\mathcal{K}}^*={\mathcal{K}}^+\cap\L^*$, ${\mathcal{N}}^*={\mathcal{N}}^+\cap\L^*$ and $T=S\cap{\mathcal{N}}^+=S\cap{\mathcal{N}}^*$, we have ${\mathcal{K}}^+T={\mathcal{N}}^+$ if and only if ${\mathcal{K}}^*T={\mathcal{N}}^*$.
As noted above, every proper overgroup of an element of $\Delta^*\backslash\Delta=R^{\mathcal{F}}$ is in $\Delta$. Hence, by Lemma \[L:ChermakIIHelp\](b),(c), properties (b) and (c) hold with $(\L^*,\Delta^*,S)$ in place of $(\L^+,\Delta^+,S)$. Suppose now that ${\mathcal{N}}^+$ is a partial normal subgroup of $\L^+$ and ${\mathcal{N}}:={\mathcal{N}}^+\cap \L\unlhd\L$ such that ${\mathcal{F}}_{S\cap{\mathcal{N}}}({\mathcal{N}})$ is ${\mathcal{F}}$-invariant. Then ${\mathcal{N}}^*:={\mathcal{N}}^+\cap\L^*\unlhd\L^*$ with ${\mathcal{N}}^*\cap\L={\mathcal{N}}^+\cap\L={\mathcal{N}}$. Since (b) is true with $(\L^*,\Delta^*,S)$ in place of $(\L^+,\Delta^+,S)$, it follows that ${\mathcal{F}}_{S\cap{\mathcal{N}}^*}({\mathcal{N}}^*)={\mathcal{F}}_{S\cap{\mathcal{N}}}({\mathcal{N}})$ and in particular, ${\mathcal{F}}_{S\cap{\mathcal{N}}^*}({\mathcal{N}}^*)$ is ${\mathcal{F}}$-invariant. So ${\mathcal{F}}_{S\cap{\mathcal{N}}^+}({\mathcal{N}}^+)={\mathcal{F}}_{S\cap{\mathcal{N}}^*}({\mathcal{N}}^*)={\mathcal{F}}_{S\cap{\mathcal{N}}}({\mathcal{N}})$. This proves (b).
If ${\mathcal{N}}^+,{\mathcal{K}}^+\in{{\mathfrak{N}}}(\L^+)$ are arbitrary, ${\mathcal{N}}^*:=\L^*\cap{\mathcal{N}}^+$, ${\mathcal{K}}^*:=\L^*\cap{\mathcal{N}}^+$, ${\mathcal{N}}:=\L\cap{\mathcal{N}}^+$, ${\mathcal{K}}:=\L\cap{\mathcal{K}}^+$ and $T:=S\cap{\mathcal{N}}$, then we see similarly that $${\mathcal{K}}^+T={\mathcal{N}}^+\Longleftrightarrow {\mathcal{K}}^*T={\mathcal{N}}^*\Longleftrightarrow {\mathcal{K}}T={\mathcal{N}}$$ and (c) holds. Hence, it remains to prove (a).
If (a) is true with $(\L^*,\Delta^*,S)$ in place of $(\L^+,\Delta^+,S)$, then $\Phi_{\L^*,\L}$ is a bijection and $\Phi_{\L^*,\L}^{-1}$ is inclusion preserving. Hence, $\Phi_{\L^+,\L}=\Phi_{\L^*,\L}\circ \Phi_{\L^+,\L^*}$ is a bijection and $\Phi_{\L^+,\L}=\Phi_{\L^+,\L^*}^{-1}\circ \Phi_{\L^*,\L}^{-1}$ is inclusion preserving. Thus, replacing $(\L^+,\Delta^+,S)$ by $(\L^*,\Delta^*,S)$, we may assume from now on that $$\Delta^+=\Delta\cup R^{\mathcal{F}}.$$ In particular, we have then that every proper overgroup of an element of $\Delta^+\backslash\Delta=R^{\mathcal{F}}$ is an element of $\Delta$. So by Lemma \[L:LinkingLocN(R)\], $N_\L(R)=N_{\L^+}(R)$ is a subgroup of $\L$.
Note that Lemma \[L:ChermakIIHelp\](a) implies that $\Phi_{\L^+,\L}$ is injective. Moreover, if ${\mathcal{M}}^+$ and ${\mathcal{N}}^+$ are partial normal subgroups of $\L^+$ with ${\mathcal{M}}^+\cap\L\subseteq{\mathcal{N}}^+\cap\L$, then Lemma \[L:ChermakIIHelp\](a) gives that ${\mathcal{M}}^+={\langle}({\mathcal{M}}^+\cap\L)^{\L^+}{\rangle}\subseteq {\mathcal{N}}^+={\langle}({\mathcal{N}}^+\cap\L)^{\L^+}{\rangle}$. So if $\Phi_{\L^+,\L}$ is a bijection, then $\Phi_{\L^+,\L}^{-1}$ is inclusion preserving. Hence, it remains to show that $\Phi_{\L^+,\L}$ is surjective.
For the remainder of this proof let ${\mathcal{N}}$ be a partial normal subgroup of $\L$ and set $T:=S\cap{\mathcal{N}}$. We will show that there exists ${\mathcal{N}}^+\unlhd\L^+$ with ${\mathcal{N}}^+\cap\L={\mathcal{N}}$. For the proof we set $\ov{\L}:=\L/{\mathcal{N}}$ and consider the natural projection $$\alpha\colon \L\longrightarrow \ov{\L}.$$ By [@loc1 Corollary 4.5], using the “bar notation”, the triple $(\ov{\L},\ov{\Delta},\ov{S})$ is a locality. Observe also that $N_\L(R)\alpha\subseteq N_{\ov{\L}}(\ov{R})$. We consider two cases now.
*Case 1:* The subgroup $T$ is not contained in $R$. As $T$ is strongly closed in ${\mathcal{F}}$ by [@loc1 Lemma 3.1], it follows that $T\not\leq Q$ for every $Q\in R^{\mathcal{F}}$. Thus, for any such $Q$, we have $QT\in\Delta$ and $Q\alpha=\ov{Q}=\ov{QT}\in\ov{\Delta}$. This proves $\Delta^+\alpha\subseteq\ov{\Delta}$. Applying Corollary \[C:MainCor\] with $(\ov{\L},\ov{\Delta},\ov{S})$ in place of $({{\widetilde{\mathcal{L}}}},{\widetilde{\Delta}},{{\widetilde{S}}})$, we conclude that there exists a homomorphism of partial group $\gamma\colon \L^+\longrightarrow \ov{\L}$ with $\gamma|_\L=\alpha$. By [@Ch Lemma 3.3], ${\mathcal{N}}^+:=\ker(\gamma)$ is a partial normal subgroup of $\L^+$. Moreover, ${\mathcal{N}}^+\cap \L=\ker(\alpha)={\mathcal{N}}$.
*Case 2:* We have $T\leq R$. In this case, by Lemma \[L:ProjectionLocalityMorphismFusionSystem\](a), $N_\L(R)\alpha=N_{\ov{\L}}(\ov{R})$. As $N_\L(R)=N_{\L^+}(R)$ is a subgroup of $\L$, it follows now from Lemma \[L:PartialProj\] that $M:=N_{\ov{\L}}(\ov{R})$ is a subgroup of $\ov{\L}$ and $\alpha|_{N_\L(R)}\colon N_\L(R)\longrightarrow M$ is a surjective group homomorphism. As $N_S(R)\in{\mathrm{Syl}}_p(N_\L(R))$ by Lemma \[L:LocalityoverFFnatural\](b), this yields that $N_{\ov{S}}(\ov{R})=N_S(R)\alpha$ is a Sylow $p$-subgroup of $M$. Moreover, since $N_{\mathcal{F}}(R)={\mathcal{F}}_{N_S(R)}(N_{\L^+}(R))={\mathcal{F}}_{N_S(R)}(N_\L(R))$ by Lemma \[L:LocalityoverFFnatural\](c), it follows from Lemma \[L:GroupEpiFusionEpi\] that $\alpha|_{N_S(R)}$ induces and epimorphism from $N_{\mathcal{F}}(R)$ to ${\mathcal{F}}_{N_{\ov{S}}(\ov{R})}(M)$.
By Lemma \[L:ProjectionLocalityMorphismFusionSystem\](b), $\alpha|_S$ induces an epimorphism from ${\mathcal{F}}={\mathcal{F}}_S(\L)$ to $\ov{{\mathcal{F}}}:={\mathcal{F}}_{\ov{S}}(\ov{\L})$. Hence, by Lemma \[L:FusionEpi\], we have $\Delta^+\alpha=\ov{\Delta}\cup \ov{R}^{\ov{{\mathcal{F}}}}$, the subgroup $\ov{R}$ is fully $\ov{{\mathcal{F}}}$-normalized, and $\alpha|_{N_S(R)}$ induces an epimorphism from $N_{\mathcal{F}}(R)$ to $N_{\ov{{\mathcal{F}}}}(\ov{R})$. The latter fact implies that $N_{\ov{{\mathcal{F}}}}(\ov{R})={\mathcal{F}}_{N_{\ov{S}}(\ov{R})}(M)$. By Lemma \[L:LinkingLocN(R)\], we have $R^*:=O_p(N_{\L^+}(R))\in\Delta$ and $R^*\unlhd N_\L(R)$. Hence, setting $\ov{\Delta}_{\ov{R}}=\{\ov{P}\in\ov{\Delta}\colon \ov{R}\unlhd\ov{P}\}$, we have $\ov{R^*}\in\ov{\Delta}_{\ov{R}}$ and $\ov{R^*}\unlhd M$, which implies $\L_{\ov{\Delta}_{\ov{R}}}(M)=M=N_{\ov{\L}}(\ov{R})$. Now [@Ch Hypothesis 5.3] holds with $\ov{{\mathcal{F}}}$, $(\ov{\L},\ov{\Delta},\ov{S})$, $\ov{R}$ and ${\operatorname{id}}_M$ in place of ${\mathcal{F}}$, $(\L,\Delta,S)$, $T$ and $\lambda$. So by [@Ch Theorem 5.14], setting ${\widetilde{\Delta}}:=\Delta^+\alpha$, there exists a locality $({{\widetilde{\mathcal{L}}}},{\widetilde{\Delta}},\ov{S})$ such that $\ov{\L}\subseteq {{\widetilde{\mathcal{L}}}}$, $N_{{{\widetilde{\mathcal{L}}}}}(R)=M$, and the inclusion map $\ov{\L}\hookrightarrow{{\widetilde{\mathcal{L}}}}$ is a homomorphism of partial groups. Hence, $\alpha$ regarded as a map $\L\longrightarrow{{\widetilde{\mathcal{L}}}}$ is a homomorphism of partial groups, which by Corollary \[C:MainCor\] extends to a homomorphism $\gamma\colon \L^+\longrightarrow {{\widetilde{\mathcal{L}}}}$ of partial groups. Then ${\mathcal{N}}^+:=\ker(\gamma)\unlhd \L^+$ and ${\mathcal{N}}^+\cap\L=\ker(\alpha)={\mathcal{N}}$. This proves the assertion.
\[C:PartialNormal\] Let $(\L,\Delta,S)$ and $(\L^+,\Delta^+,S)$ be linking localities over the same fusion system ${\mathcal{F}}$. Then $|{{\mathfrak{N}}}(\L)|=|{{\mathfrak{N}}}(\L^+)|$.
Suppose $(\L,\Delta,S)$ and $(\L^+,\Delta^+,S)$ are linking localities over the same fusion system ${\mathcal{F}}$. By Proposition 3.3 and Theorem 7.2(a) in [@subcentric], there exist subcentric linking localities $(\hat{\L},{\mathcal{F}}^s,S)$ and $(\hat{\L}^+,{\mathcal{F}}^s,S)$ over ${\mathcal{F}}$ such that $\hat{\L}|_\Delta=\L$ and $\hat{\L}^+|_{\Delta^+}=\L^+$. Moreover, by [@subcentric Theorem A(b)], there exists a rigid isomorphism $\alpha\colon \hat{\L}\longrightarrow\hat{\L}^+$. Then $\alpha$ induces a bijection ${{\mathfrak{N}}}(\hat{\L})\longrightarrow {{\mathfrak{N}}}(\hat{\L}^+),{\mathcal{N}}\mapsto {\mathcal{N}}\alpha$. So by Theorem \[T:MainChermakII\] (applied twice), we have $|{{\mathfrak{N}}}(\L)|=|{{\mathfrak{N}}}(\hat{\L})|=|{{\mathfrak{N}}}(\hat{\L}^+)|=|{{\mathfrak{N}}}(\L^+)|$. This shows the assertion.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'This paper proves Hölder continuity of viscosity solutions to certain nonlocal parabolic equations that involve a generalized fractional time derivative of Marchaud or Caputo type. As a necessary and preliminary result, this paper first shows that viscosity solutions to certain nonlinear ordinary differential equations involving the generalized fractional time derivative are Hölder continuous.'
address: 'Department of Mathematics, Brigham Young University, Provo, UT 84604'
author:
- Mark Allen
bibliography:
- 'refmarchaud.bib'
title: Hölder regularity for nondivergence nonlocal parabolic equations
---
@text
Introduction
============
This paper studies nonlocal parabolic equations of nondivergence type involving a generalized fractional time derivative. Specifically, we study $$\label{e:main}
\begin{aligned}
f(x,t) &= {\mathcal{D}}_t^{\alpha} u(x,t) - \mathcal{I} u(x,t) \\
&:= \int_{-\infty}^t [u(x,t)-u(x,s)]\mathcal{K}(t,s) \ ds \\
& \quad - \sup_{i}\inf_{j}\left( \int_{{{\mathbb R}}^n}[u(x+y,t)+ u(x-y,t)-2u(x,t)]K_{ij}(x,y,t) \right).
\end{aligned}$$ The main result is that viscosity solutions are Hölder continuous. Before stating the exact assumptions made for the kernels ${\mathcal{K}}$ and $K_{ij}$, we first describe the recent history of showing Hölder continuity for parabolic equations of nondivergence type as well as the motivation for studying a parabolic equation involving a generalized fractional time derivative.
For local linear parabolic equations of nondivergence type, Krylov and Safonov [@ks80] proved Hölder continuity of solutions without requiring any regularity assumptions on the coefficients of the equation. This method was later adapted to study regularity properties for fully nonlinear elliptic equations [@cc95]. L. Wang [@w92] adapted the methods for nonlinear elliptic equations to prove the continuity of solutions to fully nonlinear parabolic equations. In recent years nonlocal equations have gained interest due to the applications in the physical sciences [@mk00; @z02]. Of particular interest are nonlocal elliptic operators of integro-differential type $$\label{e:ellnon}
\mathcal{L} u(x) = \int_{{{\mathbb R}}^n} [u(x)-u(y)]K(x,y).$$ When $K(x,y)=c_{n,\sigma}|x-y|^{-n-2\sigma}$, then $\mathcal{L}$ is simply the fractional Laplacian $(-\Delta)^{\sigma}$. An assumption such as $K(x,y)=K(x,-y)$ will make $\mathcal{L}$ a nondivergence type operator whereas an assumption such as $K(x,y)=K(y,x)$ will give an operator of divergence type. If $\mathcal{L}$ is of nondivergence type, then $\mathcal{L}$ is the nonlocal analogue of the linear elliptic operator $$Lu = a_{ij} u_i u_j.$$ A typical ellipticity assumption on $K(x,y)$ is that $\lambda |x-y|^{-n-2\sigma} \leq K(x,y) \leq \Lambda|x-y|^{-n-2\sigma}$, and is analogous to the assumption for local equations that $\lambda |\xi|^2 \leq a_{ij}\xi_i\xi_j \leq \Lambda |\xi|^2$. Although regularity estimates were known for operators of type , these estimates were not uniform as the order $\sigma$ of the operator went to $1$. Caffarelli and Silvestere [@cs09] adapted the techniques for fully nonlinear local equations [@cc95] to the nonlocal setting, and obtained uniform estimates as the order of the operator $\sigma$ went to $1$. Chang Lara and Davila [@cl141] then proved regularity estimates for parabolic equations of the type $$\partial_t u(x,t) - \mathcal{I}u(x,t) = 0,$$ where $\mathcal{I}$ is a nonlocal nonlinear elliptic operator. The estimates in [@cl141] are uniform as the order of the operator $\sigma$ approaches $1$, so that the results in [@cl141] are not only a nonlocal analogue, but also recover many of the regularity results for local parabolic equations.
The equation of interest in this paper is motivated by the following equation $$\label{e:plasma}
\partial_t^{\alpha} u - D_{|x|}^{\beta} u = f,$$ introduced in [@dcl04; @dcl05] to model plasma transport. The function $u$ is the probability density function for tracer particles in the plasma which represents the probability of finding a particle at time $t$ and position $x$. The right hand side $f$ is a source term. The nonlocal diffusion operator $D_{|x|}^{\beta}$ is one-dimensional and accounts for avalanche-like transport. The fractional derivative $\partial_t^{\alpha}$ is the Caputo derivative, and in the context of this model, the Caputo derivative accounts for the trapping of the trace particles in turbulent eddies. Although is an equation for one spatial dimension, we may consider the following nonlocal parabolic equation in higher dimensions $$\label{e:firstdive}
\partial_t^{\alpha}u - \mathcal{L}u = f$$ where $\mathcal{L}$ is nonlocal elliptic operator of type , and $\partial_t^{\alpha}$ is the Caputo derivative. As mentioned earlier, certain assumptions made on the kernel $K(x,y)$ lead to an operator of either divergence or nondivergence type. The author in [@a16] studied weak solutions of of nondivergence type. The assumptions made on the kernel $K(x,y,t)$ were that $$\label{e:elliplamb}
\frac{\lambda}{|x-y|^{n+\sigma}}\leq K(x,y,t) \leq \frac{\Lambda}{|x-y|^{n+\sigma}},$$ for fixed $0<\lambda \leq \Lambda$, and that $K(x,y,t)=K(x,-y,t)$. For divergence form equations, the author in [@acv16] studied weak solutions of of divergence type and proved a De Giorgi-Nash-Moser type theorem which gives Hölder continuity of solutions. The results in [@acv16] assumed and $K(x,y,t)=K(y,x,t)$. The latter assumption leads to being an equation of divergence type. The author showed in [@2acv16] that the methods utilized in [@acv16] can be applied to an equation of type with a much more general fractional time derivative. The main aim of this paper is to extend the results for nondivergence form equations in [@a16] to equations involving this more general fractional time derivative which we now describe.
The Marchaud and Caputo Derivatives
-----------------------------------
The Caputo derivative is useful in modeling phenomena that take into account interactions in the past. In this manner one can think of the equation has having “memory”. This is in contrast to parabolic equations such as the heat operator $\partial_t - \Delta$ that do not account for the past. The Caputo derivative is defined as $$_a\partial_t^{\alpha} f(t) := \frac{1}{\Gamma(1-\alpha)} \int_{a}^t \frac{f'(s)}{(t-s)^{1+\alpha}} \ ds.$$ For $C^1$ functions one may use integration by parts to show the equivalent formula $$_a\partial_t^{\alpha} f(t) = \frac{1}{\Gamma(1-\alpha)} \frac{f(t)-f(a)}{(t-a)^{\alpha}} + {\frac{\alpha}{\Gamma(1-\alpha)}}\int_a^t \frac{u(t)-u(s)}{(t-s)^{1+\alpha}} \ ds.$$ If we define $f(t)=f(a)$ for $t<a$ then as in [@acv16] we have the equivalent formulation $$\label{e:fracform}
\partial_t^{\alpha} f(t) := {\frac{\alpha}{\Gamma(1-\alpha)}}\int_{-\infty}^t \frac{f(t)-f(s)}{(t-s)^{1+\alpha}} \ ds.$$ This one-sided nonlocal derivative is known as the Marchaud derivative [@skm93], and was recently studied in [@bmst16]. The formulation in is very useful. It is no longer essential to know, and therefore label, the initial point $a$. Another useful feature of this formulation in is rather than assigning initial data as simply $u(a)=c$, one may assign more general “initial” data as $u(t)=\phi(t)$ for $t\leq a$ with $\phi(x)$ not necessarily differentiable or even continuous. The formulation in will be particularly useful for the notion of viscosity solutions in the context of nondivergence solutions later described in Section \[s:viscosity\]. The formulation in looks similar to the one-dimensional fractional Laplacian except that the integration occurs from only one side. We may then apply many of the techniques developed for nonlocal elliptic operators like the fractional Laplacian to equations involving the Caputo derivative as was done in [@acv16; @2acv16]. Finally, this formulation in allows for a different type of generalization of the Caputo derivative. Rather than generalizing as $$\frac{d}{dt} (k \ast (f-f(a))) \quad \text{ or } \quad \int_a^t f'(s)K(t-s) \ ds,$$ one may generalize as $$\label{e:maintime}
{\mathcal{D}}_t^{\alpha}f(t):= \int_{-\infty}^t [f(t)-f(s)]{\mathcal{K}}(t,s) \ ds.$$ The proof of Hölder continuity in [@acv16] for the linear divergence equation works for the more general fractional time derivative $$\int_{-\infty}^t [f(t)-f(s)]{\mathcal{K}}(t,s,x),$$ provided that the kernel satisfies $$\label{e:timediv}
{\mathcal{K}}(t,t-s,x) = {\mathcal{K}}(t+s,t,x)$$ and $$\label{e:alphabound}
\frac{\alpha}{\Gamma(1-\alpha)}\frac{\lambda}{(t-s)^{1+\alpha}}
\leq {\mathcal{K}}(t,s,x)
\leq \frac{\alpha}{\Gamma(1-\alpha)} \frac{\Lambda}{(t-s)^{1+\alpha}}.$$ See for instance [@2acv16] where a kernel ${\mathcal{K}}(t,s,x)$ satisfying only and is utilized. The assumption is analogous to the assumption for the kernel $K(x,y,t)$ of $\mathcal{L}$. Likewise, the condition is analogous to $K(x,y,t)=K(y,x,t)$ and was necessary in [@2acv16] because the equation was of divergence form.
Main Results
------------
We will assume that the kernels $K_{ij}$ satisfy and $K_{ij}(x,y,t)=K_{ij}(x,-y,t)$. We will also assume that the kernel ${\mathcal{K}}(t,s)$ satisfies .
In order to prove Hölder continuity of solutions to nondivergence parabolic equations involving the Caputo derivative, the author in [@a16] followed the idea in [@s11] to solve an ordinary differential equation in order to capture information backwards in time for a solution. In the context of solutions to this requires solving $$\label{e:odeode}
\begin{cases}
m(t)= 0 &\text{ if } t\leq -2 \\
{\mathcal{D}}_t^{\alpha} m(t)= c_0|\{x \in B_1 : u(x,t)\leq 0\}| - C_1m(t) &\text{ if } t \in (-2,0).
\end{cases}$$ Several complications arise when considering solutions of . First, we need to prove existence of solutions. Second, the right hand side is not necessarily continuous. Third, a solution of may not be regular enough to utilize as a part of a test function for the notion of viscosity solution. Fourth, we need to show roughly that if $|\{x \in B_1\times (-2,-1): u(x,t)\leq 0\}|\geq \mu_1$, then $m(t)\geq \mu_2$ if $t \in (-1,0)$. When ${\mathcal{D}}_t^{\alpha}=\partial_t^{\alpha}$ the author in [@a16] utilized explicit representation formulas [@d04] for solutions to to obtain the necessary properties. Since no such formulas are available for solutions to in general, we overcome the four mentioned complications in a different way. We first show a priori Hölder continuity estimates for such ordinary differential equations. Our class of weak solutions will be considered in the viscosity sense as described in Section \[s:viscosity\]. Our first main result is
\[t:main\] Let $f$ be a continuous function on $[-1,0]$, and let $u$ be a bounded and a viscosity solution to $$\sup_{k} \inf_{l} \left( [u(x,t)-u(x,s)]{\mathcal{K}}^{kl}(t,s) \ ds \right) = f(t)$$ on $(-1,0]$ with $f \in L^{\infty}$. Assume also that the kernels ${\mathcal{K}}^{kl}(t,s)$ satisfy . Then if $0< \alpha \leq 1$ there exists two constants $0<\beta\leq 1$ and $C>0$ depending on $ \alpha,\lambda,\Lambda$ and $\displaystyle \| u \|_{L^{\infty}((-\infty,-1))}$ but uniform as $\alpha \to 1$ such that $$\label{e:fholder}
\| u \|_{C^{0,\beta}([-1/2,0])} \leq C \| f \|_{L^{\infty}}.$$
We utilize Theorem \[t:main\] to prove existence of solutions to . To accommodate the second and third complications we show in Section \[s:barrier\] that we may approximate uniformly from below by Lipschitz subsolutions to which will be sufficient for the purposes of this paper. The fourth complication, however, remains. The Hölder estimate in depending on an $L^{\infty}$ norm of $f$ is insufficient to overcome the fourth complication. We therefore consider a subclass of kernels ${\mathcal{K}}(t,s)$ which we assume also satisfy . This last assumption allows us to prove our second main result.
\[t:main2\] Let $u$ be a bounded, continuous viscosity solution to in $B_2 \times [-2,0]$. Assume the kernels $K_{ij}(x,y,t)$ satisfy $K_{ij}(x,y,t)=K_{ij}(x,-y,t)$ and . Assume the kernel ${\mathcal{K}}(t,s)$ satisfies and . Then there exists three constants $C,\kappa,{\epsilon}_0>0$ depending only on $\Lambda,\lambda, n,\alpha,\sigma$, but uniform as $\alpha \to 1$ such that $u$ is Hölder continuous in $B_1 \times [-1,0]$, and for $(x,t),(y,s) \in B_1 \times [-1,0]$ the following estimates holds $$|u(x,t)-u(y,s)| \leq C (\| u\|_{L^{\infty}} + \epsilon_0^{-1} \| f\|_{L^{\infty}})|x-y|^{\kappa}+|t-s|^{\kappa \alpha/(2\sigma)}.$$
We note that the estimates in Theorem \[t:main2\] remain uniform as the order $\alpha$ of the fractional time derivative approaches $1$. However, the estimates do not remain uniform as the order of the elliptic operator $\sigma$ approaches $1$. Having obtained the necessary theory for solutions to ordinary differential equations involving ${\mathcal{D}}_t^{\alpha}$, the author plans in a future work to utilize the techniques of Chang Lara and Davila in [@cl141] to prove Theorem \[t:main2\] with estimates uniform as $\sigma$ approaches $1$. What would also be of interest is to prove Hölder regularity to solutions of an equation of the form $$\label{e:nontime}
\begin{aligned}
&\sup_{k}\inf_{l}\left( \int_{-\infty}^t [u(x,t)-u(x,s)]{\mathcal{K}}^{kl}(t,s,x) \right) \\
& \quad - \sup_{i}\inf_{j}\left( \int_{{{\mathbb R}}^n}u(x+y,t)-u(x,t)K^{ij}(t,x,y) \right) = f(t,x).
\end{aligned}$$ To do so it appears necessary to prove Theorem \[t:main\] with the Hölder estimate depending on the $L^p$ norm of the right hand side $f$.
Notation
--------
We here define notation that will be consistent thoughout the paper.
- $\partial_t^{\alpha}$ - the Caputo derivative as defined in .
- ${\mathcal{D}}_t^{\alpha}$ - the generalized Marchaud derivative as defined in
- $\alpha$ - will always denote the order of the fractional time derivative.
- $\sigma$ - will always denote the order of the nonlocal elliptic spatial operator.
- $t,s$ - will always be variables reserved as time variables.
- $K(x,y,t)$ - the kernel for the elliptic operator $\mathcal{L}$ as defined in .
- ${\mathcal{K}}(t,s)$ - the kernel for ${\mathcal{D}}_t^{\alpha}$ as defined in .
- $M_{\sigma}^{\pm}$ - Pucci’s extremal operators (defined in Section \[s:viscosity\]) for the elliptic spatial operators.
- $M_{\alpha}^{\pm}$ - Pucci’s extremal operators (defined in Section \[s:viscosity\]) for the fractional time derivatives.
- $\lambda, \Lambda$ - Ellipticity constants as appearing in and .
- $Q_r(x_0,t_0)$ - the space-time cylinder $B_r(x_0) \times (t_0 - r^{2\sigma/\alpha} , t_0)$.
- $Q_r$ - the cylinder centered at the origin $Q_r(0,0)$.
Outline
-------
The outline of our paper is as follows: In Section \[s:viscosity\] we explain the notion of viscosity solution that will be used in the paper. In Section \[s:unique\] we use a standard method to show the comparison principle and uniqueness for ordinary differential equations involving ${\mathcal{D}}_t^{\alpha}$. In Section \[s:holder\] we prove our first main result that solutions to certain ordinary differential equations involving ${\mathcal{D}}_t^{\alpha}$ are Hölder continuous. In Section \[s:barrier\] we establish the necessary properties for solutions of . In Section \[s:holder2\] we prove our second main result that solutions to parabolic equations of type are Hölder continuous.
Viscosity Solutions and Pucci’s Extremal Operators {#s:viscosity}
==================================================
In order to study weak solutions of equations of type ${\mathcal{D}}_t^{\alpha} u - \mathcal{L} u=f$ in nondivergence form we will utilize the notion of viscosity solution. For the elliptic operator $\mathcal{L}$ we recall the notion of Pucci’s extremal operators introduced in [@cs09]. For fixed time $t$ we denote the second order difference $\delta(u,x,y,t):= u(x+y,t)+u(x-y,t)-2u(x,t)$. We fix two constants $0 < \lambda \leq \Lambda$ and define $$\begin{aligned}
M_\sigma^+ u(x,t) &:= \int_{{{\mathbb R}}^n} \frac{\Lambda \delta(u,x,y,t)_+ - \lambda\delta(u,x,y,t)_-}{|y|^{n+2\sigma}} \\
M_\sigma^- u(x,t) &:= \int_{{{\mathbb R}}^n} \frac{\lambda \delta(u,x,y,t)_+ - \Lambda\delta(u,x,y,t)_-}{|y|^{n+2\sigma}}.
\end{aligned}$$ We now define a Pucci-type extremal operator for fractional derivatives of type . For fixed $x \in {{\mathbb R}}^n$, $$\begin{aligned}
M_{\alpha}^+ u(x,t) &:= \frac{\alpha}{\Gamma(1-\alpha)} \int_{-\infty}^t \frac{\Lambda (u(x,t)-u(x,s))_+ - \lambda (u(x,t)-u(x,s))_- }{(t-s)^{1+\alpha}} \\
M_{\alpha}^- u(x,t) &:= \frac{\alpha}{\Gamma(1-\alpha)} \int_{-\infty}^t \frac{\lambda (u(x,t)-u(x,s))_+ - \Lambda (u(x,t)-u(x,s))_- }{(t-s)^{1+\alpha}}.
\end{aligned}$$ Since $\alpha$ is reserved for ${\mathcal{D}}_t^{\alpha}$ and $\sigma$ is reserved for the kernel of $\mathcal{L}$, there should be no confusion between $M_{\alpha}^{\pm}$ and $M_{\sigma}^{\pm}$. These operators give rise to the equations $$\begin{aligned}
&M_{\alpha}^- u(x,t) - M^+ u(x,t) &\leq f(x,t) \\
&M_{\alpha}^+ u(x,t) - M^- u(x,t) &\geq f(x,t).
\end{aligned}$$ Since in this paper we show regularity for the parabolic equation when the kernels ${\mathcal{K}}(t,s)$ satisfy , we will only consider solutions to $$\begin{aligned}
{2}
&{\mathcal{D}}_t^{\alpha} u(x,t) - M^+ u(x,t) &\leq f(x,t) \label{e:p3} \\
&{\mathcal{D}}_t^{\alpha} u(x,t) - M^- u(x,t) &\geq f(x,t). \label{e:p4}
\end{aligned}$$ When proving regularity for ordinary differential equations we will not assume ${\mathcal{K}}(t,s)$ satisfies , so we consider solutions to $$\begin{aligned}
{2}
&M_{\alpha}^- u(t)) &\leq f(t) \label{e:p5} \\
&M_{\alpha}^+ u(t) &\geq f(t). \label{e:p6}
\end{aligned}$$
As in [@cc95] we have the following properties for Pucci’s extremal operators.
\[p:basics\] For fixed $u,v$ evaluated at fixed $(x,t)$ we have the following properties where $M^{\pm}$ denote either $M_{\alpha}^{\pm}$ or $M_{\sigma}^{\pm}$. $$\begin{aligned}
(i) \quad &M^- u \leq M^+ u. \\
(ii) \quad &M^- u = -M^+ (-u). \\
(iii) \quad &M^{\pm} cu = c M^{\pm} u \text{ if } c\geq0. \\
(iv) \quad &M^+ u + M^- v \leq M^+(u+v) \leq M^+ u + M^+ v.\\
(v) \quad &M^- u + M^- v \leq M^- (u+v) \leq M^-u + M^+ v.
\end{aligned}$$
We now define a viscosity solution. We say that an upper semi-continuous function $u$ is a viscosity subsolution of (or a solution of ) if whenever a $C^{2,1}$ function satisfies $\phi \geq u$ on $[t_0-r,t_0] \times B_{\rho}(x_0)$ and $\phi(x_0,t_0)=u(x_0,t_0)$ and if $v$ is defined as $$v(x,t) :=
\begin{cases}
\phi(x,t) & \text{if } (x,t) \in[t_0-r,t_0]\times B_{\rho}(x_0) \\
u(x,t) & \text{ otherwise } ,\\
\end{cases}$$ Then ${\mathcal{D}}_t^{\alpha} v(x_0,t_0) - \mathcal{I}v(x_0,t_0) \leq f(x_0,t_0)$ (or $v$ is a solution to at $(x_0,t_0)$). A viscosity supersolution of (or a solution of ) for lower semi-continuous functions is defined similarly. We point out that a viscosity subsolution (supersolution) of is a viscosity solution of (). A solution is both a subsolution and supersolution, and consequently a continuous function.
The notion of viscosity solutions and supersolutions for ${\mathcal{D}}_t^{\alpha} u = f$ (or solutions to and ) are similarly defined. We note that we may extend our class of test functions that touch from above or below to functions $\phi$ that are $C^2$ in the $x$-variable for fixed $t$ and Lipschitz in time for fixed $x$. It is clear that if function ${\mathcal{D}}_t^{\alpha} u$ can be evaluated classically and solves ${\mathcal{D}}_t^{\alpha} u=f$ then $u$ is a solution in the viscosity sense. This is made clear in the following two Propositions.
\[p:classic\] Let $u$ be a continuous bounded function. Let $\phi \in C^{0,\gamma}$ with $\alpha < \gamma \leq 1$. If $\phi \geq (\leq )u$ on $[t_0 - {\epsilon}, t_0]$ and $\phi(t_0)=u(t_0)$, then the integral $$\int_{-\infty}^{t_0} [u(t_0)-u(s)] {\mathcal{K}}(t,s)\ ds$$ is well defined and possibly $\infty$ $\ ( -\infty )$ so that ${\mathcal{D}}_t^{\alpha} u (t_0)$ is well defined.
Without loss of generality we assume that $\phi \geq u$ on $[t_0 - {\epsilon}, t_0]$. Then $$\begin{aligned}
&\int_{-\infty}^{t_0} [u(t_0) - u(s)] {\mathcal{K}}(t,s) \ ds \\
&\quad =\int_{-\infty}^{t_0 - {\epsilon}} [u(t_0) - u(s)] {\mathcal{K}}(t,s) \ ds + \int_{t_0 - {\epsilon}}^{t_0} [u(t_0) - u(s)] {\mathcal{K}}(t,s) \ ds\\
&\quad\geq \int_{-\infty}^{t_0 - {\epsilon}} [u(t_0) - u(s)] {\mathcal{K}}(t,s) \ ds + \int_{t_0 - {\epsilon}}^{t_0} [u(t_0) - \phi(s)] {\mathcal{K}}(t,s) \ ds\\
&\quad= \int_{-\infty}^{t_0 - {\epsilon}} [u(t_0) - u(s)] {\mathcal{K}}(t,s) \ ds + \int_{t_0 - {\epsilon}}^{t_0} [\phi(t_0) - \phi(s)] {\mathcal{K}}(t,s) \ ds\\
&\quad\geq \int_{-\infty}^{t_0 - {\epsilon}} [u(t_0) - u(s)] {\mathcal{K}}(t,s) \ ds - \Lambda \| \phi \|_{C^{0,\gamma}} \int_{t_0 - {\epsilon}}^{t_0} (t_0 - s)^{\gamma-1-\alpha} \ ds\\
&\quad\geq \int_{-\infty}^{t_0 - {\epsilon}} [u(t_0) - u(s)] {\mathcal{K}}(t,s) \ ds - C_1 \\
&\quad \geq -2\Lambda \| u \|_{L^{\infty}} \int_{-\infty}^{t_0 - {\epsilon}} (t_0-s)^{-1-\alpha} \ ds - C_1 \\
&\quad \geq -C_2.
\end{aligned}$$ Therefore, the integral is well defined and possibly $\infty$.
\[p:classic2\] Let $u$ be a continuous bounded function on $(-\infty,T]$ and assume that for some $t \in (-\infty,T]$ there exists a Lipschitz function touching $u$ by below (above) at $t$. From Proposition \[p:classic\], the term ${\mathcal{D}}_t^{\alpha} u(t)$ is well defined and we have that $$\label{e:pclassic}
\int_{-\infty}^{t} [u(t)-u(s)] {\mathcal{K}}(t,s)\ ds \geq (\leq) f(t)$$ if and only if ${\mathcal{D}}_t^{\alpha} u(t)\leq (\geq) f(t)$ in the viscosity sense.
Assume the inequality in . If $\phi$ touches $u$ from below in $[t-{\epsilon},t]$, then $$\begin{aligned}
f(t)&\leq \int_{-\infty}^{t} [u(t) - u(s)] {\mathcal{K}}(t,s) \ ds \\
&\quad =\int_{-\infty}^{t - {\epsilon}} [u(t) - u(s)] {\mathcal{K}}(t,s) \ ds + \int_{t - {\epsilon}}^{t} [u(t) - u(s)] {\mathcal{K}}(t,s) \ ds\\
&\quad\leq \int_{-\infty}^{t - {\epsilon}} [u(t) - u(s)] {\mathcal{K}}(t,s) \ ds + \int_{t - {\epsilon}}^{t} [u(t) - \phi(s)] {\mathcal{K}}(t,s) \ ds\\
&\quad= \int_{-\infty}^{t - {\epsilon}} [u(t) - u(s)] {\mathcal{K}}(t,s) \ ds + \int_{t - {\epsilon}}^{t} [\phi(t) - \phi(s)] {\mathcal{K}}(t,s) \ ds.
\end{aligned}$$ and so ${\mathcal{D}}_t^{\alpha} u \geq f(t)$ in the viscosity sense. Assume now that ${\mathcal{D}}_t^{\alpha} u \geq f(t)$ in the viscosity sense. From the assumption, we may touch $u$ from below by a Lipschitz function $\phi$ in some neighborhood $[t-{\epsilon},t]$. We may then find Lipschitz $\phi_k$ converging uniformly to $u$ in $[t - {\epsilon},t]$ with $\phi_k \geq u$ in $[t-{\epsilon},t]$. Since the integral in is well defined we have from Lebesgue’s dominated convergence theorem $$\begin{aligned}
f(t) &\leq \lim_{k \to \infty}\int_{-\infty}^{t - {\epsilon}} [u(t) - u(s)] {\mathcal{K}}(t,s) \ ds + \int_{t - {\epsilon}}^{t} [\phi_k(t) - \phi_k(s)] {\mathcal{K}}(t,s) \ ds \\
&= \lim_{k \to \infty} \int_{-\infty}^{t - {\epsilon}} [u(t) - u(s)] {\mathcal{K}}(t,s) \ ds + \int_{t - {\epsilon}}^{t} [u(t) - \phi_k(s)] {\mathcal{K}}(t,s) \ ds \\
&= \int_{-\infty}^{t - {\epsilon}} [u(t) - u(s)] {\mathcal{K}}(t,s) \ ds + \int_{t - {\epsilon}}^{t} [u(t) - u(s)] {\mathcal{K}}(t,s) \ ds \\
&= \int_{-\infty}^{t} [u(t) - u(s)] {\mathcal{K}}(t,s) \ ds.
\end{aligned}$$
The notion of continuity is important for viscosity solutions. If we let $$\begin{cases}
u(t) = t^{\alpha - 1} &\text{ if } t>0 \\
u(t) =0 &\text{ if } t\leq 0,
\end{cases}$$ then one may explicitly compute that $\partial_t^{\alpha} u(t)=0$ for any $t \in {{\mathbb R}}$. However, $u$ is not upper semi-continuous, and therefore not a viscosity subsolution.
Viscosity solutions are closed under appropriate limits.
\[l:limit\] Let $u_k$ be a sequence of continuous bounded viscosity solutions to (or ) in $B_{R_1} \times [-R_2,0]$, iconverging in ${{\mathbb R}}^{n} \times (-\infty,T]$ to $u_0$ bounded and continuous. Then $u_0$ is a viscosity solution to (or ) in $B_{R_1} \times [-R_2,0]$.
The proof is standard and straightforward from the definition of viscosity solutions (see [@cc95]).
Approximating Solutions {#s:unique}
=======================
In order to show uniqueness for an ordinary differential equation such as ${\mathcal{D}}_t^{\alpha} u =f$ we use the notion of sup- and inf-convolution. For a bounded and upper-semicontinuous function $u$ on $(-\infty,t_0]$, for $t\leq t_0$ we define $$\label{e:jensenup}
u^{{\epsilon}}(t):= \sup_{s\leq t} \{u(s) +\frac{1}{{\epsilon}}(s-t) + {\epsilon}\}.$$ If $u$ is bounded and lower-semicontinuous on $(-\infty,t_0]$, for $t\leq t_0$ we define $$\label{e:jensendown}
u_{{\epsilon}}(t):= \inf_{s\leq t} \{u(s) -\frac{1}{{\epsilon}}(s-t) - {\epsilon}\}.$$
We have the following properties
\[p:jensen1\] For $u^{{\epsilon}}$ as defined in and $t\leq t_0$ the following hold $$\begin{aligned}
{2}
&(1) \quad \text{there exists } t^* \leq t \text{ such that } u^{{\epsilon}}(t)=u(t^*) + \frac{1}{{\epsilon}}(t^*-t) + {\epsilon}. \\
&(2) \quad u^{{\epsilon}}(t)\geq u(t)+ {\epsilon}. \\
&(3) \quad u^{{\epsilon}}(t_2)-u^{{\epsilon}}(t_1) \geq {\epsilon}^{-1}(t_2-t_1) \text{ for } t_1 < t_2. \\
&(4) \quad 0<{\epsilon}_1<{\epsilon}_2 \Rightarrow u^{{\epsilon}_1}(t) \leq u^{{\epsilon}_2}(t). \\
&(5) \quad t-t^* \leq 2{\epsilon}\sup |u|. \\
&(6) \quad 0<u^{{\epsilon}}(t)-u(t) \leq u(t^*) - u(t) + {\epsilon}.
\end{aligned}$$
All properties except for $(3)$ are as in Lemma 5.2 in [@cc95]. For property $(3)$ we note that for $t\leq t_1 \leq t_2$ we have $$\begin{aligned}
u^{{\epsilon}}(t_2) &\geq u(t) + \frac{1}{{\epsilon}}(t-t_2) + {\epsilon}\\
&= u(t) +\frac{1}{{\epsilon}}(t-t_1) + {\epsilon}+ \frac{1}{{\epsilon}}(t_1 - t_2).
\end{aligned}$$ Taking the supremum over $t\leq t_1$ we obtain $$u^{{\epsilon}}(t_2) - u^{{\epsilon}}(t_1) \geq -{\epsilon}^{-1}(t_2 - t_1).$$
Using the properties listed in Proposition \[p:jensen1\], it is standard to show the following Proposition which is analogous to Theorem 5.1 in [@cc95] and Propositions 5.4 and 5.5 in [@cs09].
\[p:jensen2\] If $u$ is bounded and lower-semicontinuous (upper-semicontinuous) in $(-\infty,T]$, then $u_{{\epsilon}} (-u^{{\epsilon}}) \ \Gamma$-converges to $u (-u)$. If $u$ is continuous then $u^{{\epsilon}},u_{{\epsilon}}$ converge uniformly to $u$. Furthermore, if ${\mathcal{D}}_t^{\alpha} u \geq (\leq)f$, then there exists $d_{{\epsilon}} \to 0$ as ${\epsilon}\to 0$ such that ${\mathcal{D}}_t^{\alpha} u_{{\epsilon}} \geq f- d_{{\epsilon}} \ ({\mathcal{D}}_t^{\alpha}u^{{\epsilon}} \leq f + d_{{\epsilon}})$.
\[l:jensen\] Let $u$ be bounded and upper-semicontinuous and $v$ be bounded and lower semi-continuous on $(-\infty,T]$. Let $f,g$ be continuous functions. If ${\mathcal{D}}_t^{\alpha} u \leq f$ and ${\mathcal{D}}_t^{\alpha} v \geq g$ in the viscosity sense, then ${\mathcal{D}}_t^{\alpha} (v-u) \geq g-f$ in the viscosity sense.
We take the approximating solutions $u^{{\epsilon}},v_{{\epsilon}}$ with ${\mathcal{D}}_t^{\alpha} u^{{\epsilon}}\leq f + d_{{\epsilon}}$ and ${\mathcal{D}}_t^{\alpha} v_{{\epsilon}} \geq f - d_{e}$. From property $(3)$ in Proposition \[p:jensen1\] we have that at every point $u^{{\epsilon}}$ can be touched from above by a Lipschitz function and $v_{{\epsilon}}$ can be touched from below. Then also $v_{{\epsilon}}-u^{{\epsilon}}$ can be touched from below by a Lipschitz function at every point. From Propositions \[p:classic\] and \[p:classic2\], the terms ${\mathcal{D}}_t^{\alpha} u^{{\epsilon}}, {\mathcal{D}}_t^{\alpha} v_{{\epsilon}}, D_t^{\alpha} (v_{{\epsilon}}-u^{{\epsilon}})$ are well defined and at any point $t$ $${\mathcal{D}}_{t}^{\alpha} (v_{{\epsilon}} - u^{{\epsilon}}(t)) = {\mathcal{D}}_t^{\alpha} v_{{\epsilon}}(t) - {\mathcal{D}}_t^{\alpha} u^{{\epsilon}}(t) \geq g -f - 2d_{{\epsilon}}.$$ Then from Proposition \[p:classic2\] we conclude that ${\mathcal{D}}_{t}^{\alpha} (v_{{\epsilon}} - u^{{\epsilon}}) \geq g-f-2d_{{\epsilon}}$. Letting ${\epsilon}\to 0$ we obtain from Lemma \[l:limit\] that ${\mathcal{D}}_{t}^{\alpha} (v - u) \geq g-f$ in the viscosity sense.
We are now able to prove a comparison principle.
\[t:comparison\] Let $u$ be bounded and upper-semicontinuous and $v$ be bounded and lower semi-continuous on $(-\infty,T_2]$. Let $f$ be a continuous function and assume that ${\mathcal{D}}_t^{\alpha} u \leq f \leq {\mathcal{D}}_t^{\alpha} v$ on $(T_1,T_2]$ with $u \leq v$ on $(-\infty,T_1]$. Then $u\leq v$ on $(-\infty,T_2]$, and if $u(t_0)=v(t_0)$ for some $t_0 \in (T_1,T_2]$, then $u(t)=v(t)$ for all $t \leq t_0$.
Suppose there exists $t_1 \in (T_1,T_2]$ such that $v(t_1)-u(t_1)\leq 0$. Let $v-u$ achieve its minimum in $[T_1,T_2]$ at $t_0 \in (T_1,T_2]$. From Lemma \[l:jensen\], we have ${\mathcal{D}}_{t}^{\alpha} (v-u)\geq 0$ in the viscosity sense. Since $v-u \geq 0$ for $t\leq T_1$, then $(v-u)$ is touched from below by the constant function $v(t_0)-u(t_0)$ at $t_0$. From Propositions \[p:classic\] and \[p:classic2\] the term ${\mathcal{D}}_t^{\alpha} (v-u)(t_0)$ is well defined and $$\int_{-\infty}^{t_0} [(v-u)(t_0) - (v-u)(s)]{\mathcal{K}}(t,s) \ ds \geq 0.$$ Since $(v-u)$ achieves a minimum over $(-\infty,t_0]$ at $t_0$ we also have $$0 \geq \int_{-\infty}^{t_0} [(v-u)(t_0) - (v-u)(s)]{\mathcal{K}}(t,s) \ ds.$$ Then if $(v-u)(t_0) \leq 0$ we have that $(v-u)(t)=0$ for all $t\leq t_0$. Then either $(v-u)(t_0)>0$ or $(v-u)(t)=0$ for all $t\leq t_0$.
With a comparison principle available, we may use Perron’s method to prove existence of solutions.
\[t:exist\] Let $\phi(t)$ be continuous on $(-\infty,T_1]$ and $f$ continuous on $[T_1,T_2]$. There exists a unique viscosity solution $u$ to $$\begin{cases}
{\mathcal{D}}_t^{\alpha} u(t) = f(t) &\text{ for } t\in [T_1,T_2] \\
u(t)=\phi(t) &\text{ if } t\leq T_1
\end{cases}$$ on $(T_1,T_2]$.
The proof of Theorem \[t:exist\] is standard. We forgo the proof since the ideas and methods are later used in the proof of Lemma \[l:zorn\]. One may also show existence for solutions of ; however, since the main focus of this paper is the regularity of solutions we do not include the result here.
Later in the paper we will need that the continuous divergence solutions constructed in [@acv16; @2acv16] are also viscosity solutions.
\[t:divnondiv\] Let $f$ be a continuous function on $[-2,0]$ with $f(-2)=0$. Assume the kernel ${\mathcal{K}}(t,s)$ of ${\mathcal{D}}_t^{\alpha}$ satisfies and . Then the weak divergence solution $v$ constructed in [@acv16] of $$\label{e:divint1}
\begin{cases}
{\mathcal{D}}_t^{\alpha} v = f &\text{ in } (-\infty,0]. \\
v=0 &\text{ for } t\leq -2,
\end{cases}$$ is also a viscosity solution and hence the unique viscosity solution.
In [@acv16; @2acv16] the solution $v$ is constructed with an ${\epsilon}$ approximation. Via recursion we find the solution $v_{{\epsilon}}$ to $$\label{e:discrete}
{\epsilon}\sum_{i<j} [v_{{\epsilon}}({\epsilon}j) - v_{{\epsilon}}({\epsilon}i)]{\mathcal{K}}({\epsilon}j, {\epsilon}i) = f({\epsilon}j),$$ where $i,j \in {{\mathbb Z}}$. From [@acv16; @2acv16], $v_{{\epsilon}} \to v$ where $v$ solves the divergence form equation $$\label{e:divlimit}
\begin{aligned}
&\int_{-\infty}^T \int_{-\infty}^t [v(t)-v(s)][\phi(t)-\phi(s)]{\mathcal{K}}(t,s) \ ds \ dt \\
& \ + \int_{-\infty}^T \int_{-\infty}^{2t-T}v(t)\phi(t){\mathcal{K}}(t,s) \ ds \ dt
- \int_{-\infty}^T v(t){\mathcal{D}}_t^{\alpha} \phi(t) \ dt \\
&= \int_{-\infty}^T f(t)\phi(t) \ dt, \\
\end{aligned}$$ for all $T\leq 0$ and $\phi$ bounded and Lipschitz on $(-\infty,0]$. We extend $v_{{\epsilon}}$ to all of $(-\infty,0]$ by $v_{{\epsilon}}(t)=v({\epsilon}i)$ where ${\epsilon}(i-1) < t \leq {\epsilon}j$. We now show that $v$ is a viscosity solution. The main idea in the following computations is that since is a discrete equation, then $v_{{\epsilon}}$ is also a viscosity-type discretized solution, and so the limit will also be a viscosity solution. We now let $\psi$ be a Lipschitz function touching $v$ strictly from above at $t_0 \leq 0$, that is $v(t)<\psi(t)$ for $t<t_0$ and $v(t_0)=\psi(t_0)$. We let $\delta_{{\epsilon}}$ be such that $\psi+\delta_{{\epsilon}}$ touches $v_{{\epsilon}}$ from above at $t_{{\epsilon}} \leq t_0$. Since $\psi$ touches $v$ strictly from above at $t_0$, then $t_{{\epsilon}} \to t_0$ as ${\epsilon}\to 0$. Now If ${\epsilon}(j_{{\epsilon}}-1)< t_{{\epsilon}} \leq {\epsilon}j_{{\epsilon}}$, then since $\psi + \delta_{{\epsilon}}$ touches $v_{{\epsilon}}$ from above and is a discrete equation, we have that $${\epsilon}\sum_{i<j_{{\epsilon}}} [\psi(t_{{\epsilon}}) - \psi({\epsilon}i)]{\mathcal{K}}({\epsilon}j_{{\epsilon}}, {\epsilon}i) \geq f({\epsilon}j_{{\epsilon}}).$$ Now $$\begin{aligned}
{\epsilon}\sum_{i<j_{{\epsilon}}} [\psi(t_{{\epsilon}}) - \psi({\epsilon}i)]{\mathcal{K}}({\epsilon}j_{{\epsilon}}, {\epsilon}i) &=
{\epsilon}\sum_{i<j_{{\epsilon}}} [\psi(t_{{\epsilon}}) - \psi({\epsilon}j_{{\epsilon}})]{\mathcal{K}}({\epsilon}j_{{\epsilon}}, {\epsilon}i)\\
&\quad + {\epsilon}\sum_{i<j_{{\epsilon}}} [\psi({\epsilon}j_{{\epsilon}}) - \psi({\epsilon}i)]{\mathcal{K}}({\epsilon}j_{{\epsilon}}, {\epsilon}i) \\
&= (I) + (II).
\end{aligned}$$ Since $\psi$ is Lipschitz we have $$\begin{aligned}
|(I)| &= \left| {\epsilon}\sum_{i<j_{{\epsilon}}} [\psi(t_{{\epsilon}}) - \psi({\epsilon}j_{{\epsilon}})]{\mathcal{K}}({\epsilon}j_{{\epsilon}}, {\epsilon}i) \right| \\
& \leq {\epsilon}\sum_{i<j_{{\epsilon}}} [C {\epsilon}]{\mathcal{K}}({\epsilon}j_{{\epsilon}}, {\epsilon}i) \\
& \leq {\epsilon}^2 \sum_{i<j_{{\epsilon}}}\Lambda {\epsilon}^{-1-\alpha} (j_{{\epsilon}}-i)^{-1-\alpha} \\
& = {\epsilon}^{1-\alpha} \Lambda \sum_{i=1}^{\infty} i^{-1-\alpha}.
\end{aligned}$$ Then $|(I)| \to 0$ as ${\epsilon}\to 0$. Since $\psi$ is Lipschitz continuous it follow that $(II) \to {\mathcal{D}}_t^{\alpha}(t_0)$ as ${\epsilon}\to 0$. Since $f$ is continuous and $t_{{\epsilon}} \to t_0$ as ${\epsilon}\to 0$, we have that $f({\epsilon}j_{{\epsilon}}) \to f(t_0)$. Then ${\mathcal{D}}_t^{\alpha} \psi(t_0) \geq f(t_0)$. The proof if $\psi$ touches from below is similar. Then $v$ is a viscosity solution.
This next Corollary will be useful when we want to show a limit is not equivalently zero.
\[c:divint\] Let $v$ be the viscosity solution to as in Theorem \[t:divnondiv\]. Assume also that $v \geq 0$. Then for $-2 <t\leq 0$ we have $$\label{e:divint2}
\frac{\lambda\alpha}{\Gamma(1-\alpha)} \int_{-2}^{t} \frac{v(s)}{(t-s)^{\alpha}} \ ds
\leq \int_{-2}^{t} f(s) \ ds \leq \frac{\Lambda\alpha}{\Gamma(1-\alpha)} \int_{-2}^{t} \frac{v(s)}{(t-s)^{\alpha}} \ ds.$$
We take $\phi \equiv 1$ in . Then $$\begin{aligned}
\int_{-2}^{T} f(t) \ dt &= \int_{-\infty}^{T}\int_{-\infty}^{2t-T} v(t) K(t,s) \ ds \ dt \\
&\leq \frac{\alpha \Lambda}{\Gamma(1-\alpha)} \int_{-\infty}^{2t-T} \int_{-\infty}^t \frac{v(t)}{(t-s)^{1+\alpha}} \ ds \ dt \\
&= \frac{\alpha\Lambda}{\Gamma(1-\alpha)} \int_{-2}^{T} \frac{v(t)}{(T-t)^{\alpha}} \ dt.
\end{aligned}$$ The bound from the other side is shown in the same manner.
Hölder continuity for the time derivative {#s:holder}
=========================================
We will follow the ideas in [@cs09] to prove the Hölder continuity. The outline and statements of the Lemmas are intentionally similar to those in Sections $8$-$12$ in [@cs09] so that the reader may compare and contrast properties of the operator ${\mathcal{D}}_t^{\alpha}$ with properties of $\mathcal{L}$. Rather than work with the concave envelope we will work with a different envelope. $${\mathcal{M}}_{u}(t) := \sup_{s\leq t} u(s).$$
\[l:meas\] Assume $M_{\alpha}^- u(t) \leq f(t)$ and ${\mathcal{M}}_u(t_0)=u(t_0)$. Let $r_k :=2^{-1/(1-\alpha)}2^{-k}$ and $R_k(t):= [t-r_k,t-r_{k+1}]$. There is a constant $C_0$ depending on $\lambda,\Lambda$ but not $\alpha$ such that for any $M>0$, there exists $k$ such that $$\label{e:upbound}
|R_k(t_0) \cap \{u(s)<u(t_0)-Mr_k\}|\leq C_0 \frac{f(t_0)}{M} |R_k(t_0)|.$$
We first note that $|R_k(t_0)|=r_{k+1}=r_k/2$. Since ${\mathcal{M}}_u(t_0)=u(t_0)$, then $u$ is touched from above by the constant ${\mathcal{M}}_u(t_0)$. Then from Proposition \[p:classic2\] we have $$\int_{-\infty}^{t_0} [u(t_0)-u(s)]K(t_0,s) ds \leq f(t_0).$$ Suppose that the conclusion is not true. It then follows that $$\begin{aligned}
f(t_0) \geq M_{\alpha}^- u(t_0) &\geq \frac{\alpha \lambda}{\Gamma(1-\alpha)} \int_{-\infty}^{t_0} \frac{u(t_0)-u(s)}{(t_0 - s)^{1+\alpha}} ds \\
&\geq \frac{\alpha \lambda}{\Gamma(1-\alpha)} \int_{t_0-2^{-1/(1-\alpha)}}^{t_0} \frac{u(t_0)-u(s)}{(t_0 - s)^{1+\alpha}} ds \\
&\geq \frac{\alpha \lambda}{\Gamma(1-\alpha)} \sum_{k=0}^{\infty} \int_{R_k(t_0)} \frac{u(t_0)-u(s)}{r_{k+1}^{1+\alpha}} ds \\
&\geq \frac{\alpha \lambda}{\Gamma(1-\alpha)} M r_k \frac{C_0f(t_0)}{M} \frac{|R_k(t_0)|}{r_{k+1}^{1+\alpha}} \\
&= 2^\alpha \frac{\alpha \lambda}{\Gamma(1-\alpha)} C_0 f(t_0) \sum_{k=0}^{\infty} r_k^{1-\alpha}.
\end{aligned}$$ Now $$\sum_{k=0}^{\infty} r_k^{1-\alpha} = \frac{1}{2}\sum_{k=0}^\infty\left(\frac{1}{2^{1-\alpha}} \right)^k = \frac{1}{2}\left[\frac{1}{1-2^{-(1-\alpha)}} \right].$$ Then $$f(t_0) \geq \frac{\alpha \lambda}{\Gamma(1-\alpha)} \frac{C_0 f(t_0)}{2^{1-\alpha}-1}.$$ From the following relation on the Gamma function $ \Gamma(1-z)\Gamma(z)=\frac{\pi}{\sin(\pi z)} $ it follows that for $\alpha_0<\alpha\leq 1$, the right hand side of the above equation is bounded below. Then for $C_0$ large enough we obtain a contradiction.
\[r:alpha\] In the above proof it is clear that by choosing $C_0$ larger (say for instance $2C_0$) that for fixed $\alpha$ and $M$ bounded from above we may choose $k \leq N$ for some large $N$. This $N$ will depend on $\alpha$, but this gives a bound from below on $r_k$ such that holds.
\[c:estimate\] For any ${\epsilon}_0>0$, there exists $C$ such that for $u$ as defined in Lemma \[l:meas\], there exists $r \in (0,2^{-1/(1-\alpha)})$ such that if ${\mathcal{M}}_{u}(t)=u(t)$ and $M_{\alpha}^- u(t)\leq f(t)$, then there exists a constant $c_\alpha$ (depending on $\alpha$) and $r \in (c_{\alpha},2^{-1/(1-\alpha)})$ such that $$\label{e:pull}
\frac{|\{s \in [t-r,t-r/2]\}: u(s)<u(t)-Cf(t)r|}{|[t-r,t-r/2]|} \leq {\epsilon}_0$$
By choosing $M=Cf(t){\epsilon}_0^{-1}$ we obtain from Lemma \[l:meas\] and Remark \[r:alpha\].
\[l:abp\] Let $u(t) \leq 0$ for $t<-2$ and $M_{\alpha}^- u \leq f$ on $[-2,T]$. There exists a constant $C$ depending on $\lambda, \alpha_0$ and there exists finitely many intervals $I_j$ with disjoint interiors and with length between $c_\alpha$ and $2^{-(1-\alpha)}$ such that $$\label{e:abp}
\sup u^+ \leq C \sum_{j} \max_{I_j}f^+(t) |I_j|.$$ Furthermore, if $I_j=[t_j-r_j,t_j]$ is such an interval, and $[t_j-2r_j,t_j]\subset [-2,T]$, then $$\label{e:abp2}
|\{s \in [t_j-2r_j,t_j-r_j]:u(s)\geq u(t_j)-Cr_j\}|\geq \mu r_j.$$
Notice that as $\alpha \to 1$ the inequality becomes $$\sup u \leq C \int_{\{u={\mathcal{M}}_u\}} f^+(s) ds.$$
We will choose the finite sequence of intervals inductively. Let $u^+$ achieve its maximum on $[-2,T]$ at the point $t_1$. From Corollary \[c:estimate\] there exists $r_1$ with $c_{\alpha}\leq r_1\leq 2^{-{1-\alpha}}$ such that holds for $I_1 = [t_1 - r_1,t_1]$. $${\mathcal{M}}_{u^+}(t_1)-{\mathcal{M}}_{u^+}(t_1-r_1) \leq u(t_1) - u(s) \text{ for any } s \in [t_1-2r_1,t_1-r_1].$$ Since $[t_1-2r_1,t_1-r_1]$ satisfies , we then have that $${\mathcal{M}}_{u^+}(t_1) - {\mathcal{M}}_{u^+}(t_1-r_1) \leq Cf(t_1)r_1 = Cf(t_1)|I_1|.$$ Proceeding inductively, supposing $I_{j-1}$ has been chosen, we choose $$t_{j} := \max \{-2\leq s \leq t_{j-1}-r_{j-1}: u(s)={\mathcal{M}}_{u}(s)\}.$$ If no such $t_j$ exists then the process terminates. Otherwise, by Corollary \[c:estimate\] we choose $r_j \in (c_{\alpha},2^{-1/(1-\alpha)})$ such that $[t_j-2r_j,t_j-r_j]$ satisfies . We label $I_j = [t_j-r_j,t_j]$. As in the case of $I_1$ we obtain $${\mathcal{M}}_{u^+}(t_j)-{\mathcal{M}}_{u^+}(t_j-r_j) \leq Cf(t_j)|I_j|.$$ Since each $r>c_{\alpha}$ the process will eventually terminate. Now $$\{u = {\mathcal{M}}_{u^+}\} \subseteq \cup_{j} I_j,$$ and so $$\sup u^+ \leq \sum_{j} |{\mathcal{M}}_{u^+}(t_j) - {\mathcal{M}}_{u^+}(t_j - r_j) \leq C\sum_{j} \max_{I_j} f^+(t) |I_j|.$$ Finally, from how each $I_j$ is chosen we have holds for each $[t_j-2r_j,t_j-r_j]$ and also $\mathring{I}_j \cap \mathring{I}_k = \emptyset$ for $j\neq k$.
For this next Lemma for an interval $I_j = [t_j - r_j, t_j]$ we define $$2I_j := [t_j - 2r_j, t_j].$$
\[l:cover\] Let $A \subset [0,1]$ and let $c_1,c_2$ be two positive constants. Let $\{I_j: j \in \mathcal{I}\}$ be a collection of closed intervals with nonempty interior satisfying the following for every $j \in \mathcal{I}$: $$\begin{aligned}
&(i) \quad 2I_j \subset [0,1] \\
&(ii) \quad \dot{I}_i \cap \dot{I}_j = \emptyset \text{ for every } i \neq j \\
&(iii) \quad \sum_{j \in \mathcal{I}} |I_j| > c_1 \\
&(iv) \quad |A \cap 2I_j| \geq c_2 |2I_j|,
\end{aligned}$$ then $|A| \geq (c_1 \cdot c_2)/3$.
We first assume the collection $\{I_j\}$ is finite. We will choose a subset $\mathcal{J} \subseteq \mathcal{I}$. Recall that each interval has nonempty interior and the intersection of the interiors is empty. We will pick a subcollection of intervals as follows: We choose and label $I_1$ as the interval with the farthest right end point. If there is another interval $I_k$ for some $k \in \mathcal{I}$ such that $$\begin{aligned}
&(1) \quad 2I_k \cap 2I_1 \neq \emptyset \text{ and } \\
&(2) \quad 2I_k \setminus (2I_k \cap 2I_1) \neq \emptyset,
\end{aligned}$$ then we choose $I_2$ to be the interval satisfying $(1)$ and $(2)$ such that $2I_k$ has the farthest left endpoint. If no $2I_k$ satisfies $(1)$ and $(2)$ then we choose $I_2$ to be the interval whose right end point is closest to the left end point of $2I_1$. We then choose all remaining intervals in the same manner. It is clear from the construction that each interval from the collection $\{2I_j\}$ for $j \in \mathcal{J}$ can intersect at most two other intervals from the same collection. It is also clear that $$\label{e:covered}
\bigcup_{i \in \mathcal{I}} I_i \subset \bigcup_{j \in \mathcal{J}} 2I_j,$$ Then $$\begin{aligned}
|A| &\geq \frac{1}{2} \sum_{j \in \mathcal{J}} |A \cap 2I_j| \\
&\geq \frac{c_2}{3} \sum_{j \in \mathcal{J}} |2I_j| \\
&\geq \frac{c_2}{3} \sum_{i \in \mathcal{I}} |I_i| \\
&\geq \frac{c_1 c_2}{3}.
\end{aligned}$$ If the collection $\{I_i\}$ is infinite, then we simply choose a finite subcollection of intervals whose measure is within ${\epsilon}$, and then let ${\epsilon}\to 0$.
We now prove the most important Lemma.
\[l:krylov\] Let $0<\alpha_0\leq \alpha \leq 1$. There exists ${\epsilon}_0 >0$, $0<\mu<1$, and $M>1$ all depending on $\alpha_0, \lambda, \Lambda$ such that if $$\begin{aligned}
(i) \quad &u \geq 0 \text{ in } (-\infty,1] \\
(ii) \quad &u(1) \leq 1/2 \\
(iii) \quad &M_{\alpha}^+ u \geq -{\epsilon}_0 \text{ in } [-1,1],
\end{aligned}$$ then $$|\{u \leq M\} \cap [0,1]|>\mu.$$
We utilize the function $$\Phi(t):=
\begin{cases}
0 &\text{ for } t\leq 3/4 \\
4(t-3/4) &\text{ for } t \geq 3/4.
\end{cases}$$ Notice that $M_{\alpha}^- u \leq 4\lambda$ for all $0<\alpha\leq 1$. We now apply Lemma \[l:abp\] to $v:= \Phi - u$. Since $u(1)\leq 1/2$ we have that $v(1)\geq 1/2$. We have From estimate we have $$\begin{aligned}
1/2 \leq \sup v &\leq C \sum_{j} \max_{I_j} M_{\alpha}^- (\Phi - u) |I_j| &\text{ from } \eqref{e:abp}\\
&\leq C({\epsilon}_0 + 4\lambda C_1)\sum_{j} |I_j| &\text{ from Proposition } \ref{p:basics} \\
&\leq C{\epsilon}_0 + 4\lambda C_1\sum_{j} |I_j|.
\end{aligned}$$ Then for ${\epsilon}_0$ small enough we obtain $$\frac{1}{4}\leq C \sum_{j} |I_j|.$$ Now $\Phi$ is supported in $[3/4,1]$. Then if $I_j=[t_j-r_j/4, t_j]$, then $t_j \in [3/4,1]$ and so $2I_j :=[t_j - r_j,t_j]\subset [0,1]$. From $$|\{s\in 2I_j : v(s)\geq v(t_j) - Cr_j \}| \geq \mu |I_j|. $$ We recall that $r_j \leq 2^{-(1-\alpha)}$ and $v(t_j) \geq 0$, so that $$|\{s\in 2I_j : \Phi(s)- u(s) \geq - C \}| \geq \mu |I_j|.$$ Furthermore, $\Phi(s)\leq 1$, and so $$|\{s\in 2I_j : 1+ C \geq u(s) \}| \geq \mu |I_j|.$$ We now use Lemma \[l:cover\] to conclude that $$|\{s \in [0,1] : 1+C \geq u(s) \}| \geq \mu_1.$$
Using Lemma \[l:krylov\] with Lemma 4.2 from [@cc95] one proves the following two Lemmas just as in [@cc95].
\[l:krylov2\] Let $u$ be as in Lemma \[l:krylov\], then $$|\{u > M^k\} \cap [0,1]|\leq (1-\mu)^k.$$
\[l:krylov3\] Let $u$ be as in Lemma \[l:krylov\], then $$|\{u > t\} \cap [0,1]|\leq dt^{-{\epsilon}}$$ where $d,{\epsilon}>0$ depend only on $\lambda, \Lambda, \alpha_0$.
We now have
\[t:Lepsilon\] Let $u \geq 0$ in $(-\infty,0]$ and $M_{\alpha}^+ u \geq -C_0$ in $[-2r,0]$. Assume $\alpha \geq \alpha_0>0$. Then $$|\{u >t\}\cap [-r,0]|\leq dr(2u(0)+C_0r^{\alpha}{\epsilon}_0^{-1})^{{\epsilon}} t^{-{\epsilon}} \text{ for every } t$$ where the constants $d,{\epsilon}$ depend only on $\lambda, \Lambda,$ and $\alpha_0$ as in Lemma \[l:krylov3\].
We rescale and let $$v(t):= \frac{u(rt)}{2u(0)+C_0r^{\alpha} {\epsilon}_0^{-1}}.$$ Then $M_{\alpha}^+ v \geq -{\epsilon}_0$ on $[-2,0]$ and $v(0)\leq 1/2$, so we may apply Lemma \[l:krylov3\] to $v$ and conclude $$|\{v > t\} \cap [-1,0]|\leq dt^{-{\epsilon}},$$ which written in terms of $u$ becomes $$|\{u>t\} \cap [-r,0]| \leq dr[2u(0)+C_0r^{\alpha}{\epsilon}_0^{-1}]^{{\epsilon}}t^{-{\epsilon}}.$$
Theorem \[t:main\] will follow from the following
\[t:regular\] Let $\alpha_0 < \alpha <1$ for some $\alpha_0>0$. Let $u$ be a bounded in $(-\infty,0]$ and continuous on $[-1,0]$. If $$\begin{aligned}
M_{\alpha}^+ u &\geq - C_0 \text{ in } (-1,0], \\
M_{\alpha}^- u &\leq C_0 \text{ in } (-1,0],
\end{aligned}$$ then there is a $\beta>0$ and $C>0$ both depending only on $\lambda, \Lambda, \alpha_0$ such that $u \in C^{0,\beta}([-1/2,0])$ and $$\| u \|_{C^{0,\beta}([-1/2,0])} \leq C(\| u \|_{L^{\infty}} + C_0).$$
Theorem \[t:regular\] follows from Theorem \[t:Lepsilon\] and the following Lemma
\[l:main\] Let $-1/2 \leq u \leq 1/2$ in $(-\infty,0]$. There exists $\delta_0$ depending only on $\lambda, \Lambda$ and $0<\alpha_0<1$ such that if $$\begin{aligned}
&M_{\alpha}^+ u \geq -\delta_0 &\text{ in } [-1,0]\\
&M_{\alpha}^- u \leq \delta_0 &\text{ in } [-1,0],
\end{aligned}$$ then there is a $\beta$ depending only on $\lambda, \Lambda,$ and $\alpha_0$ such that if $1>\alpha\geq \alpha_0$ $$\label{e:beta1}
|u(t)-u(0)| \leq C|t|^\beta$$ For some constant $C$.
The proof is nearly the same as in [@cs09]. We give the details. We will construct a sequence $m_k, M_k$ with $m_k \leq u \leq M_k$ and $$\begin{aligned}
\label{e:beta}
m_k \leq u \leq M_k \text{ in } [-4^{-k},0] \\
m_k \leq m_{k+1} \text{ and } M_k \geq M_{k+1} \\
M_k - m_k = 4^{-\beta k}.
\end{aligned}$$ This will show with constant $C=2^{\beta}$. For $k=0$ we choose $m_0=-1/2$ and $M_0=1/2$ and by assumption we have . We proceed by induction. Assume holds up to $k$. We have either $$\begin{gathered}
\left |\{u \geq (M_k + m_k)/2\} \cap [-3\cdot 4^{-(k+3)},-4^{-(k+2)}] \right| \geq 4^{-(k+4)} \\
\text{ or } \\
\left |\{u \leq (M_k + m_k)/2\} \cap [-3\cdot4^{-(k+3)},-4^{-(k+2)}] \right| \geq 4^{-(k+4)}.
\end{gathered}$$ If we assume the first then we define $$v(t):= \frac{u(4^{-k}t)-m_k}{(M_k - m_k)/2}.$$ We have that $v \geq 0$ in $[-1,0]$ and $|\{v\geq 1\} \cap [-1/2,-1/4]| \geq 1/16$. We also have that $$M_{\alpha}^+ v \geq \frac{-4^{-k\alpha}\delta_0}{(M_k - m_k)/2} = -2\delta_0 4^{k(\beta-\alpha)} \geq -2\delta_0$$ as long as $\beta <\alpha$. From the inductive hypothesis, for any $0 \leq j < k$ we have $$v \geq \frac{m_{k-j} - m_k}{2(M_k-m_k)} \geq \frac{m_{k-j}-M_{k-j} + M_{k}- m_k}{2(M_k-m_k)} \geq 2(1-4^{\beta j}),$$ and so $v \geq -2(|4t|^{\beta}-1)$ outside $[-1,0]$. We define $w:= \max\{v,0\}$. For any $t \in [-3/4,0]$ we have $$\begin{aligned}
\label{e:remark}
M_{\alpha}^+ w(t) - M_{\alpha}^+v(t) &= \frac{\alpha}{\Gamma(1-\alpha)}\int_{\{v<0\}\cap \{t<-1\}} \frac{\Lambda v(s)}{(t-s)^{1+\alpha}} \\
&\geq \frac{\alpha \Lambda}{\Gamma(1-\alpha)}\int_{-\infty}^{-1} \frac{-2(|4s|^\beta-1)}{(t-s)^{1+\alpha}}.
\end{aligned}$$ Thus $M_{\alpha}^+ w - M_{\alpha}^+ v \geq -2 \delta_0$ for $\beta$ small enough. We then have $M_{\alpha}^+ w \geq -4\delta_0$, and so for any $t_0 \in [-1/4,0]$ we may apply Theorem \[t:Lepsilon\] to $w$ to obtain $$\begin{aligned}
\frac{1}{16} \leq |\{w>1\}\cap[-3/8,-1/4]| &\leq |\{w>1\}\cap [-3/8,t_0]| \\
&\leq d(1/8)(2u(t_0) + 4(1/8)^{\alpha} \delta_0 e_0^{-1})^{{\epsilon}}.
\end{aligned}$$ Thus, for $\delta_0$ small enough we have $u(t_0)\geq \theta>0$ for any $t_0 \in [-1/4,0]$. We let $M_{k+1}=M_k$ and $m_{k+1}=m_k + \theta(M_k-m_k)/2$. Then, $M_{k+1}-m_{k+1}=(1-\theta/2)4^{-\beta k}$. Choosing $\beta$ and $\theta$ small with $(1-\theta/2)=4^{-\beta}$ we have $M_{k+1}-m_{k+1}= 4^{-\beta (k+1)}$.
If on the other hand $|\{u\leq (M_k +m_k)/2 \cap [-3\cdot 4^{-(k+3)},-4^{-(k+2)}]\}| \geq 4^{-(k+4)}$, then we define $$v:= \frac{M_k - u(4^{-k}t)}{(M_k - m_k)/2}$$ and use that $M_{\alpha}^- u \leq \delta_0$.
A Barrier {#s:barrier}
=========
In this section we construct a subsolution to an ordinary differential equation that will allow us to prove the Hölder continuity. We begin with the following
\[l:cont\] Let $u \in C^{0,\beta}((-\infty,0])$ with $ \alpha< \beta <1$. Then $M_{\alpha}^{\pm} u$ is continuous on $(-\infty,0]$.
We will show that for fixed $t$ and any $\epsilon >0$, there exists $h_0$ such that if $0\leq h <h_0$, then $$\label{e:lipcont}
\left|M_{\alpha}^{\pm} u(t+h) - M_{\alpha}^{\pm}u(t) \right| < \epsilon.$$ For fixed $\xi<t$ and for $0\leq h$, we have $$\begin{aligned}
&\left| \frac{\alpha}{\Gamma(1-\alpha)} \int_{\xi}^{t+h} \frac{\Lambda [u(t+h)-u(s)]_+ -\lambda[u(t+h)-u(s)]_-}{(t+h-s)^{1+\alpha}} \ ds \right| \\
&\quad \leq \frac{\alpha}{\Gamma(1-\alpha)} \Lambda \| u\|_{C^{0,\beta}} \frac{(t+h-\xi)^{\beta-\alpha}}{\beta-\alpha}.
\end{aligned}$$ We may choose $\xi$ close enough to $t$ and choose $h$ small enough, so that the above inequality is less than $\epsilon/2$. Now since $u$ is continuous, if $t>\xi$, then both $$\left|\int_{-\infty}^{\xi} \frac{ [u(t+h)-u(s)]_-}{(t+h -s)^{1+\alpha}}
-\frac{[u(t)-u(s)]_-}{(t-s)^{1+\alpha}} \ ds \right| \to 0$$ and $$\left| \int_{-\infty}^{\xi} \frac{ [u(t+h)-u(s)]_+}{(t+h -s)^{1+\alpha}}
-\frac{[u(t)-u(s)]_+}{(t-s)^{1+\alpha}} \ ds \right| \to 0$$ as $h \to 0$. Then we may choose $h$ small enough, so that holds.
For the next Lemma we will require a solution to an ordinary differential equation. We recall that from Theorem \[t:exist\] if $f$ is continuous on $[-2,0]$ with $f(-2)=0$. Then exists a viscosity solution $u$ to the differential equation $$\begin{cases}
u(t)=0 &\text{ for } t\leq -2\\
{\mathcal{D}}_t^{\alpha} u = f(t) &\text{ on } (-\infty,0].
\end{cases}$$ That $u$ is a viscosity solution on $(-2,0]$ is a direct result from Theorem \[t:exist\]. Since $f(t)=u(t)=0$ for $t\leq -2$ it is immediate that $u$ is also a solution on $(-\infty,-2]$ as well. From Theorem \[t:main\] the solution $u$ is Hölder continuous. In the next Lemma we will also utilize the following bump function. Let $\eta \geq 0$ with support in $[0,1]$. Let $\eta' \geq 0$ for $t \leq 1/2$ and $\eta' \leq 0$ for $t\geq 1/2$. We note that ${\mathcal{D}}_t^{\alpha} \eta \leq C$ for some $C$ independent of $\alpha$ and ${\mathcal{D}}_t^{\alpha} \eta(t) <0 $ for $t>1$. We will use $$\eta_{\epsilon} := \epsilon \eta(t/\epsilon).$$
\[l:zorn\] Let $f$ be smooth on $[-2,0]$ with $f(-2)=0$. Let $u$ be the viscosity solution to $$\label{e:subbelow}
\begin{cases}
{\mathcal{D}}_t^{\alpha} u(t) = f &\text{ on } (-\infty,0] \\
u(t)=0 &\text{ for } t\leq -2.
\end{cases}$$ Then there exists a sequence of Lipschitz subsolutions $\{u_{k}\}$ to the above equation with $u_{k}\leq u$ and $u_{k}\to u$ uniformly on $[-2,0]$.
We consider for fixed $M> \sup f$ the set $$\mathfrak{K}_1 := \{w: -M \leq D_t^{\alpha} w \leq f \text{ on } [0,2], \text{ and } w \text{ is Lipschitz}\}.$$ We define $$\mathfrak{K}_2 := \{v : -M \leq {\mathcal{D}}_t^{\alpha} v \leq f \text{ on } [0,2] \text{ and } u_k \Rightarrow v \text{ with } u_k \leq v \text{ and } u_k \in \mathfrak{K}_2 \}.$$ We note that any $v \in \mathfrak{K}_2$ is continuous since it is the uniform limit on a compact set of Lipschitz continuous functions. We now show that $\mathfrak{K}_1$ is nonempty, and hence $\mathfrak{K}_2$ is also nonempty. From the theory of ordinary differential equations [@d04], we solve $$\partial_t^{\alpha} g(t) = h(t)/\lambda \leq f(t) /\lambda,$$ with $h$ smooth and strictly decreasing. Then $g$ is strictly decreasing, and so $${\mathcal{D}}_t^{\alpha} g \leq M_{\alpha}^+ g(t) = \lambda \partial_t^{\alpha} g(t) = h(t) \leq f(t).$$ Thus $g$ is a smooth subsolution to and $g$ can be approximated from below by itself. We choose $-M< \inf h/\lambda$. Thus $\mathfrak{K}_2$ is nonempty.
We now assign a partial ordering to $\mathfrak{K}_2$ with the natural assignment that $v_1 \leq v_2$ if $v_1\leq v_2$ everywhere on $(-\infty,0]$. From the comparison principle we have that $u$ as in is an upper bound for $\mathfrak{K_2}$.
We will now show that $u \in \mathfrak{K_2}$. By Zorn’s Lemma there exists a maximal element $w$. We will show that $w \equiv u$. If ${\mathcal{D}}_t^{\alpha} w \equiv {\mathcal{D}}_t^{\alpha} u$, then by comparison and uniqueness, $w \equiv u$. Suppose by way of contradiction that $w$ is not identically $u$. Then $w$ is not a supersolution of , and so there exists a $t_0 \in [0,2]$ and a Lipschitz function $\psi$ with $\psi(t_0)=w(t_0)$ and $\psi \leq w$ on $[t_0 - \delta,t]$ for some $\delta >0$ such that ${\mathcal{D}}_t^{\alpha} \phi(t_0) < f(t_0)$ where $$\phi(t) :=
\begin{cases}
\psi(t) \text { if } t \in (t_0 - \delta],\\
w(t) \text{ if } t < t_0 - \delta.
\end{cases}$$ Then from Proposition \[p:classic2\] we can evaluate $D_t^{\alpha} w(t_0)$ classically and ${\mathcal{D}}_t^{\alpha} w(t_0) < f(t_0)$. Because $w \in \mathfrak{K}_2$ there exists Lipschitz subsolutions $w_{k} \to w$ uniformly from below. Then there exists $k$ large enough so that $$\phi_k(t) :=
\begin{cases}
\psi(t) \text { if } t \in (t_0 - \delta],\\
w_k(t) \text{ if } t < t_0 - \delta.
\end{cases}$$ satisfies $M_{\alpha}^+ \phi_k < f$ on $[t_0 - \delta_1,t_0]$ for some $0<\delta_1 < \delta$. We now consider two different situations. If $w_k(t_0)=w(t_0)$, then for $k$ large enough, ${\mathcal{D}}_t^{\alpha}w_k(t_0) < f(t_0)$ because ${\mathcal{D}}_t^{\alpha} w(t_0)< f(t_0)$ classically. Since $w_k$ is Lipschitz, it follows from Lemma \[l:cont\] that ${\mathcal{D}}_t^{\alpha} w_k < f$ in $[t_0 - \delta_2, t_0 + \delta_2]$. Then since ${\mathcal{D}}_t^{\alpha} (w_k + \eta_{{\epsilon}}) = {\mathcal{D}}_t^{\alpha} w_k + {\mathcal{D}}_t^{\alpha} \eta_{{\epsilon}}$, then $w_k + \eta_{{\epsilon}}$ is a Lipschitz subsolution for $\epsilon$ small enough, and $w_k(t_0) + \eta_{{\epsilon}}(t_0) > w(t_0)$. Now the $\max\{w_k +\eta_{{\epsilon}},w\} \in \mathfrak{K}_2$, and this contradicts the maximality of $w$.
If in the second situation, $w_k(t_0) < w(t_0)$, then we extend $\phi_k$ to the right of $t_0$ by $\phi_k(t_0) = -M_1 (t_0 -t) + w(t_0)$. From Lemma \[l:cont\], for $M_1$ large enough, ${\mathcal{D}}_t^{\alpha} \phi_k \leq f$ for all $t \in [-2,0]$, and ${\mathcal{D}}_t^{\alpha} \phi_k < f$ in $[t_0 - \delta_3, t_0 + \delta_3]$. We let $\tilde{\phi}_k := \max \{\phi_k, w_k\}$ and note that $\tilde{\phi}_k \in \mathfrak{K}_1$ and Lipschitz, and ${\mathcal{D}}_t^{\alpha} \phi_k < f$ in $[t_0 - \delta_4, t_0 + \delta_4]$ for some $\delta_4 >0$ with $\tilde{\phi}_k(t_0)=w(t_0)$. Then as before we may take $\tilde{\phi}_k + \eta_{\epsilon}$ with $\epsilon$ small enough and obtain a contradiction.
In order to prove Hölder continuity of solutions to we will follow the method presented in [@s11]. One of the main ingredients is to solve an ordinary differential equation in time. We begin with the following.
\[l:timeexist\] Let $C_1$ be a fixed constant. Let $g(t)$ be a continuous function on $[-2,0]$. There exists a continuous viscosity solution $m(t)$ in $[-2,0]$ to $${\mathcal{D}}_t^{\alpha} m(t)= C_1m(t) + g(t),$$ with $m(t)=0$ for $t\leq -2$.
Let $\beta_1 < \beta$ for $\beta$ as in Theorem \[t:main\]. From Theorem \[t:exist\], for $v \in C^{\beta_1}([0,2])$ there exists a solution $h(t)$ to $${\mathcal{D}}_t^{\alpha} h(t)= C_1v(t) + g(t),$$ with $h(t)=0$ for $t\leq -2$. From Theorem \[t:main\] we have that $$\| h \|_{C^{0,\beta}} \leq C \| C_1 v + g \|_{L^{\infty}}.$$ Since $C^{0,\beta_1}([-2,0])$ is compactly contained in $C^{0,\beta}([-2,0])$ we have a compact mapping $M: v \to h$ from $C^{0,\beta_1}([-2,0])$ into itself. From Corollary 11.2 in [@gt01] it follows that there is a fixed point $m(t)$ which is a viscosity solution from Lemma \[l:limit\].
As in [@s11] we will utilize an ordinary differential equation to capture information backwards in time. We consider the fractional ordinary differential equation $$\label{e:fode}
\begin{cases}
m(-2)=0 & \text{ for } t\leq -2 \\
{\mathcal{D}}_t^{\alpha} m(t) = c_0 |\{x \in B_1: u(x,t)< 0\}| - C_1m(t) &\text{ for } -2<t\leq 0.
\end{cases}$$ We would like to use $m$ as a test function for a viscosity solution. However, since the right hand side is not continuous, we cannot apply Lemma \[l:timeexist\] to obtain the existence of $m$. Furthermore, the solution $m$ may not be Lipschitz and therefore not a valid test function. To overcome these two issues we obtain a Lipschitz subsolution to . We consider $|\{x \in B_1: u(x,t)< 0\}|$ rather than $|\{x \in B_1: u(x,t)\leq 0\}|$ because we may easily approximate the former from below by smooth functions. We accomplish this by considering $g(x,t,{\epsilon}):=\min\{\epsilon^{-1}\max\{0,-u\} , 1\}$. We then let $$G(t,{\epsilon}) = \int_{B_1}g(x,t,{\epsilon}) \ dx.$$ Now $G(t,{\epsilon})$ is continuous in $t$, and $0\leq G(t,{\epsilon})\leq |\{x \in B_1: u(x,t)< 0\}|$, and $G(t,{\epsilon}) \to |\{x \in B_1: u(x,t)< 0\}|$ as ${\epsilon}\to 0$. Since $G(t,{\epsilon})$ is continuous, from Lemma \[l:timeexist\], we may solve $$\begin{cases}
m(-2)=0 & \text{ for } t\leq -2 \\
{\mathcal{D}}_t^{\alpha} m(t) = c_0 G(t,{\epsilon}) - C_1m(t) &\text{ for } -2<t\leq 0.
\end{cases}$$
\[l:mubound\] Let $\alpha_0 \leq \alpha <1$. Assume that the kernel for ${\mathcal{D}}_t^{\alpha}$ satisfies . Let $m$ be a solution to ${\mathcal{D}}_t^{\alpha} m = c_0 f(t)-C_1 m$ with $m(t)=0$ for $t\leq -2$, $0\leq f\leq |B_1|$, and $$\label{e:muf}
\int_{-2}^{-1} f(t) \geq \mu.$$ Then there exists two moduli of continuity $\omega_1(c_0 \mu), \omega_2(C_1^{-1})$ with $\omega_i$ increasing and $\omega_i(s)>0$ for $s>0$ and depending only on $\lambda, \Lambda, |B_1|, \alpha_0$ so that $$\label{e:expbound}
m(t) \geq \omega_1 (c_0 \mu) \omega_2 (C_1^{-1}) \quad \text{for } -1\leq t \leq 0.$$
From Lemma \[l:main\] any such solution $m$ is Hölder continuous. We claim that $m>0$ for $t \in [-1,0]$. Suppose by way of contradiction that there exists $t_0 \in [-1,0]$ such that $m(t_0)\leq 0$. Let $m$ achieve its minimum at $t_1$. Then $m$ is touched from below at $t_1$ by the constant function $m(t_1)$, and so by Propositions \[p:classic\] and \[p:classic2\] we may evaluate ${\mathcal{D}}_t^{\alpha} m$ at $t_1$ and $$\int_{-\infty}^{t_1} [m(t_1) - m(s)] {\mathcal{K}}(t,s) \ ds \geq f(t_1) \geq 0.$$ But since $m(t_1)\leq 0$ and $m$ achieves a minimum at $t_1$ we have $$0 \geq \int_{-\infty}^{t_1} [m(t_1) - m(s)] {\mathcal{K}}(t,s) \ ds.$$ Then $m(t)=0$ for $t\leq t_1$. But then since $f(t)=0$ for $t\leq t_1$. But this contradicts the assumption . Then $m(t)>0$ for $-1\leq t \leq 0$.
Let us fix $c_0,C_1, \mu >0$. Suppose by way of contradiction that there exists $m_k, \alpha_k, f_k$ all satisfying the assumptions of the Lemma, but $$\inf_{[-1,0]} m_k \to 0$$ as $k \to \infty$. From Lemma \[l:main\], we have $m_k \to m_0$ in $C^{0,\beta_1}([-2,0])$ for any $\beta_1 < \beta$. Then there exists $t_1 \in [-1,0]$ such that $m_0(t_1)=0$.
We now claim that $m_0(t_0)>0$ for some $t_0 \in [-2,-1]$. Suppose by way of contradiction that $m_0 \equiv 0$ in $[-2,-1]$. From Corollary \[c:divint\], we have that $$\begin{aligned}
C \Lambda \sup_{[-2,-1]} m_k &\geq
\frac{\Lambda}{\Gamma(1-\alpha_k)} \int_{-2}^{-1} \frac{m_k(t)}{(-1-t)^{\alpha}} \ dt \\
&\geq \int_{-2}^{-1} c_0 f_k(t)- C_1 m_k \ dt\\
&\geq c_0 \mu - \int_{-2}^{-1}C_1 m_k \ dt.
\end{aligned}$$ Letting $k \to \infty$ we obtain that $0 \geq \mu$. Which is a contradiction, and therefore the claim that $m_0$ is not identically zero is true.
We now consider two different cases. First assume that for a subsequence $\alpha_k \to \alpha_1 <1$. Let $t_2$ be the first point after $t_0$ such that $m_0(t_2)=0$. Because $m_k \to m_0$ uniformly, and since $m_0(t_0)>0$, we may choose $\psi \geq 0$ smooth with $\psi(t)=0$ in a neighborhood of $t_2$ and $\psi(t_0)>0$ and also satisfying $m_k \geq \psi$ for all $k$. Now $M_{\alpha}^+ \psi(t) \leq -\delta_1$ for $t \in [t_2 - \delta_2]$ and $\alpha \in (\alpha_1 - \delta_3, \alpha + \delta_3)$. We let ${\epsilon}_k$ be such that $\psi + {\epsilon}_k \leq m_k$ on $[t_2 -\delta_2,t_2]$ and $\psi(t_k) + {\epsilon}_k = m_k(t_k)$ for some $t_k \in [t_2 - \delta_2, t_2]$. We define $$\phi_k := \begin{cases}
\psi + {\epsilon}_k &\text{ if } t \geq t_2 - \delta_2 \\
m_k &\text{ if } t < t_2 - \delta_2.
\end{cases}$$ Then $$M_{\alpha}^{+} \phi_k(t_k) \geq {\mathcal{D}}_t^{\alpha_k} \phi_k (t_k) \geq -C_1 m_k(t_k).$$ Since $m_0(t)>0$ for $t\in (t_2 - \delta_2, t_2)$, as $k \to \infty$, we have $t_k \to t_2$ and ${\epsilon}\to 0$. Then $$\lim_{k \to \infty} M_{\alpha_k}^{+} \phi_k(t_k) \geq \lim_{k \to \infty} - C_1 u_{k}(t_k) =0.$$ But we also have $$\lim_{k \to \infty} M_{\alpha_k}^{+} \phi_k(t_k) \leq
M_{\alpha_1}^+ \psi(t_2) \leq - \delta_1 <0.$$ This is a contradiction to the first case.
We now consider the case in which $\alpha_k \to 1$. For a further subsequence there exists $f_0$ such that $f_k$ converges to $f_0$ in weak star $L^{\infty}$, so that $$\int_{-2}^{-1} f_0(t) \geq \mu.$$ Then from Corollary \[c:divint\] we have that $$\frac{\lambda\alpha_k}{\Gamma(1-\alpha_k)} \int_{-2}^{t} m_k(s) \ ds \leq
\int_{-2}^t f_k(s) - C_1m_k(s) \ ds
\leq \frac{\Lambda\alpha_k}{\Gamma(1-\alpha_k)} \int_{-2}^{t} m_k(s) \ ds.$$ Since $m_k \to m_0$ uniformly and since [@d04] for any continuous function $h$ we have $$\frac{\alpha_k}{\Gamma(1-\alpha_k)} \int_{-2}^t \frac{h(s)}{(t-s)^{\alpha_k}} \ ds \to h(t),$$ as $\alpha_k \to 1$, then we obtain as $k \to \infty$ the inequality $$\lambda m_0(t) \leq \int_{-2}^t f_0(s) - C_1 m_0(s) \ ds \leq \Lambda m_0(t).$$ Then there exists $\lambda \leq g(t) \leq \Lambda $ such that $$g(t)m_0(t) = \int_{-2}^t f_0(s) - C_1 m_0(s) \ ds.$$ Then $gm_0$ is Lipschitz continuous since $f_0 $ is bounded. Furthermore, we have that $$g(t)m_0(t) \geq \int_{-2}^t f_0(s) - C_1 \lambda^{-1}g(s)m_0(s) \ ds.$$ Then $gm_0$ is a supersolution and from the theory of ordinary differential equations, $gm_0 \geq \tilde{g}$ with $\tilde{g}$ solving $$\tilde{g}(t) = \int_{-2}^t f_0(s) - C_1 \lambda^{-1}\tilde{g}(s) \ ds.$$ Since $f_0$ is not identically zero, then $\tilde{g}>0$ on $[-1,0]$. It follows that $gm_0>0$ and hence $m_0>0$ on $[-1,0]$ as well. This is a contradiction to the second case. Then for fixed $c_0,C_1, \mu$, there exists $\delta_4>0$ depending on $c_0,C_1, \mu, \lambda, \Lambda, \alpha_0$ such that for any solution satisfying the assumptions in the statement of the lemma, we have that $u \geq \delta_4$ in $[-1,0]$. We then obtain a modulus of continuity as stated in the Lemma.
Hölder Continuity {#s:holder2}
=================
In this section we follow the method used in [@s11] to prove our main result. We will need the following Proposition to account for the growth in the tails.
\[l:decrease\] Let $u$ be a continuous function, $u\leq 1$ in $({{\mathbb R}}^n \times [-2,0])\cup (B_2 \times [-\infty,0])$, which satisfies the inequality in the viscosity sense in $B_2 \times [-2,0]$ $$\label{e:assume1}
{\mathcal{D}}_t^{\alpha} u - M^+ u \leq \epsilon_0.$$ with $\alpha_0 \leq \alpha <1$. Assume also that $$\label{e:assume2}
| \{u\leq 0\} \cap (B_1 \times [-2,-1])| \geq \mu.$$ Then if $\epsilon_0$ is small enough there exists $\theta>0$ such that $u \leq 1-\theta$ in $B_1 \times [-1,0]$. The maximum value of $\epsilon_0$ as well as $\theta$ depend only on $\alpha_0, \lambda, \Lambda,n$ and $\sigma$.
We first mention that it is sufficient to prove the Lemma under the assumption $$| \{u< 0\} \cap (B_1 \times [-2,-1])| \geq \mu.$$ For if $u$ satisfies and , then $u-c$ for any positive constant $c$ will satisfy as well as the inequality above. Then $u-c \leq 1 -\theta$ in $B_1 \times [-1,0]$ independent of $c$ and so letting $c \to 0$ one obtains $u\leq 1 -\theta$.
We consider the fractional ordinary differential equation $m: (-\infty,0] \to {{\mathbb R}}$ $$\label{e:fode}
\begin{aligned}
m(t)&=0 \quad &\text{ for } t\leq -2\\
{\mathcal{D}}_t^{\alpha} m(t) &= c_0 f(t) - C_1m(t) \quad &\text{ for } t>-2,
\end{aligned}$$ Where $f(t)$ is a smooth approximation from below of $|\{x \in B_1: u(x,t)<0\}|$. From the hypothesis and Lemma \[l:mubound\], we can choose an approximation $f$ such that $$m(t) \geq \omega_1(c_0\mu) \omega_2(C_1^{-1})/2 >0,$$ for $t \in [-1,0]$. By Lemma \[l:zorn\] we can approximate $m$ uniformly from below by a Lipschitz function $g$ such that $$\begin{aligned}
g(-2)&=0 \\
{\mathcal{D}}_t^{\alpha} g(t) &\leq c_0 f(t) - C_1m(t),
\end{aligned}$$ and $$g(t) \geq \omega_1(c_0\mu) \omega_2(C_1^{-1})/4 >0,$$ for $t \in [-1,0]$. We utilize the function $g(t)$ which is not just a viscosity solution but also a classical solution since it is Lipschitz. We can then calculate ${\mathcal{D}}_t^{\alpha} g$ everywhere classically. Furthermore, $g$ is allowed as a test function for touching above or below for viscosity solutions.
We want to show that $u \leq 1- g(t) + \epsilon_0 c_{\alpha}2^{\alpha}$ if $c_0$ is small and $C_1$ is large. We can then set $\theta = \omega_1(c_0\mu) \omega_2(C_1^{-1})/4$ for $\epsilon_0$ small to obtain the result of the Lemma. We pick the constant $c_{\alpha}$ such that $\partial_t^{\alpha} c_{\alpha}(2+t)_+^{\alpha} = 1$ for $t>-2$, and note [@d04] that $c_{\alpha}$ is uniform as $\alpha \to 1$. Let $\beta:{{\mathbb R}}\to {{\mathbb R}}$ be a fixed smooth nonincreasing function such that $\beta(x)=1$ if $x\leq 1$ and $\beta(x)=0$ if $x\geq 2$. Let $b(x)= \beta(|x|)$. Where $b=0$ we have $M^- b>0$. Since $b$ is smooth $M^-b$ is continuous and it remains positive for $b$ small enough ([@s11]). Thus there exists $\beta_1$ such that $M^- b \geq 0$ if $b \leq \beta_1$.
Assume that there exists some point $(x,t) \in B_1 \times [-1,0]$ such that $$u(x,t) > 1 -g(t)+\epsilon_0 \lambda^{-1} c_{\alpha} (2+t)_+^{\alpha}.$$ We will arrive at a contradiction by looking at the maximum of the function $$w(x,t) =u(x,t) +g(t)b(x) - \epsilon_0 \lambda^{-1} c_{\alpha}(2+t)_+^{\alpha}$$ We assume there exists a point in $B_1 \times [-1,0]$ where $w(x,t)>1$. Let $(x_0,t_0)$ be the point that realizes the maximum of $w$: $$w(x_0,t_0) = \max_{{{\mathbb R}}^n \times (-\infty,0]}w(x,t).$$ This maximum is larger than 1, and so it must be achieved when $t> -2$ and $|x|<2$.
Let $\phi(x,t):= w(x_0,t_0)-g(t)b(x)+{\epsilon}_0 \lambda^{-1} c_{\alpha}(2+t)_+^{\alpha}$. We remark that $\phi(x,t)=\phi(x,-2)$ for $t\leq -2$, and $\phi$ touches $u$ from above at the point $(x_0,t_0)$. We define $$v(x,t) :=
\begin{cases}
\phi(x,t) & \text{if } x \in B_r \\
u(x,t) & \text{if } x \notin B_r.\\
\end{cases}$$ Then from the definition of viscosity solution we have $$\label{e:viscos}
\partial_t^{\alpha}v - M^+v \leq \epsilon_0 \quad \text{ at } \quad (x_0,t_0).$$ We have that $$\begin{aligned}
{\mathcal{D}}_t^{\alpha} v(x_0,t_0) &= {\mathcal{D}}_t^{\alpha} \left( -g(t_0)b(x_0) +\epsilon_0\lambda^{-1}c_{\alpha}(2+t)_+^{\alpha} \right)\\
&\geq (C_1 m(t) - c_0 f(t))b(x_0) + {\epsilon}_0 \lambda^{-1} M_{\alpha}^- (2+t)_+^{\alpha} \\
&= (C_1 m(t) - c_0 f(t))b(x_0) + {\epsilon}_0 \lambda^{-1} \lambda\partial_t^{\alpha} (2+t)_+^{\alpha} \\
&=(C_1 m(t) - c_0f(t))b(x_0) + {\epsilon}_0 \\
&\geq(C_1 m(t) - c_0|\{x \in B_1: u(x,t_0)<0\}|)b(x_0) + {\epsilon}_0 \\
&\geq (C_1 m(t) - c_0|\{x \in B_1: u(x,t_0)\leq 0\}|)b(x_0) + {\epsilon}_0. \\
\end{aligned}$$ Then $$\begin{aligned}
{\epsilon}_0 &\geq {\mathcal{D}}_t^{\alpha} v(x_0,t_0) - M^+v(x_0,t_0) \\
&\geq (C_1 m(t) - c_0|\{x \in B_1: u(x,t_0)\leq 0\}|)b(x_0) + {\epsilon}_0 - M^+v(x_0,t_0),
\end{aligned}$$ or $$\label{e:timepart1}
0 \geq (C_1 m(t) - c_0|\{x \in B_1: u(x,t_0)\leq 0\}|)b(x_0) - M^+v(x_0,t_0).$$ Now exactly as in [@s11] we obtain the following bound for $G := \{x \in B_1 \mid u(x,t_0)\leq 0 \}$, $$\label{e:silv}
M^+ v(x_0,t_0) \leq -m(t_0) M^-b(x_0,t_0) -c_0|G \setminus B_r|$$ for some universal constant $c_0$. This is how we choose $c_0$ in the fractional ordinary differntial equation. We now look at two different cases and obtain a contradiction in both. Suppose $b(x_0)\leq\beta_1$. Then $M^-b(x_0)\geq0$, and so from $$M^+v(x_0,t_0)\leq -c_0|G\setminus B_r|.$$ Combining the above inequality with , we obtain $$0 \geq \left(-c_0 |\{x \in B_1: u(x,t)\leq 0\}| + C_1m(t)\right)b(x_0) + c_0|G\setminus B_r|.$$ For any $C_1>0$ this will be a contradiction by taking $r$ small enough.
Now suppose $b(x_0)>\beta_1$. Since $b$ is a smooth compactly supported function, there exists $C$ such that $|M^-b|\leq C$. We then have from the bound $$M^+ v(x_0,t_0) \leq Cm(t_0) -c_0|G \setminus B_r|$$ and inserting this in we have $$0 \geq \left(-c_0 |\{x \in B_1: u(x,t)\leq 0\}| + C_1m(t)\right)b(x_0) - Cm(t_0) +c_0|G \setminus B_r|$$ Letting $r \to 0$ we obtain $$\begin{aligned}
0 &\geq c_0(1-b(x_0))|G| + (C_1b(x_0)-C)m(t_0) \\
&\geq c_0(1-b(x_0))|G| + (C_1\beta_1-C)m(t_0).
\end{aligned}$$ Choosing $C_1$ large enough we obtain a contradiction.
We now define $$Q_r := B_r \times [-r^{2\sigma/\alpha},0],$$ and note the rescaling property that if $v(x,t)=u(rx,r^{2\sigma/\alpha}t)$, then $$\tilde{{\mathcal{D}}}_t^{\alpha}v(x,t) - M_{\sigma}^{\pm}v(x,t) =
r^{2\sigma} \left({\mathcal{D}}_t^{\alpha}u(rx,r^{2\sigma/\alpha}t) - M_{\sigma}^{\pm}u(rx,r^{2\sigma/\alpha}t) \right)$$ where if ${\mathcal{K}}(t,s)$ is the kernel for ${\mathcal{D}}_t^{\alpha}$, then $\tilde{{\mathcal{D}}}_t^{\alpha}$ has kernel $$\frac{{\mathcal{K}}(rt,rs)}{r^{1+\alpha}}$$ which will also satisfy and . For the next three results we fix $r=\min\{4^{-1},4^{-\alpha/2\sigma}\}$. We will need the following Proposition to bound the tails.
\[p:nubound\] Let $h(t)= \max{\{2|rt|^{\nu}-1,0\}}$ with $r= \min\{4^{-1},4^{-\alpha/2\sigma}\}$. If $t_1\leq 0$ and $\nu <\alpha$ then $$0 \geq {\mathcal{D}}_t^{\alpha} h(t_1) \geq - \lambda c_{\alpha, \nu}$$ where $c_{\alpha,\nu}$ is a constant depending only on $\alpha$ and $\nu$ but for fixed $\nu$ remains uniform as $\alpha \to 1$.
Now $$0 \geq {\mathcal{D}}_t^{\alpha} h(t_1) \geq M_{\alpha}^- h(t_1) = \lambda \partial_t^{\alpha} h(t_1).$$ From [@a16] we have $$\partial_t^{\alpha} h(t_1) \geq c_{\alpha, \nu},$$ which for fixed $\nu$ remains uniform as $\alpha \to 1$. Combining the above two inequalities, the Proposition is proven.
\[l:down\] Let $u$ be a bounded continuous function which satisfies the following two inequalities in the viscosity sense in $Q_1$ $$\label{e:pucci}
\begin{aligned}
&{\mathcal{D}}_t^{\alpha} u - M^+ u \leq \epsilon_0/2, \\
&{\mathcal{D}}_t^{\alpha} u - M^- u \geq -\epsilon_0/2.
\end{aligned}$$ Let the kernel ${\mathcal{K}}(t,s)$ of ${\mathcal{D}}_t^{\alpha}$ satisfy and with $0<\alpha_0\leq \alpha <1$. Then there are univeral constants $\theta>0$ and $\nu>0$ depending only on $n,\sigma,\Lambda,\lambda,\alpha_0$ such that if $$\begin{aligned}
|u| \leq 1 &\quad \text{in} \quad B_1 \times [-1,0] \\
|u(x,t)| \leq 2|rx|^{\nu}-1 &\quad \text{in} \quad ({{\mathbb R}}^n\setminus B_1) \times [-1,0] \\
|u(x,t)| \leq 2|rt|^{\nu}-1 &\quad \text{in} \quad B_1 \times (-\infty,-1],
\end{aligned}$$ with $r=\min\{4^{-1},4^{-\alpha/2\sigma}\}$, then $$\text{osc}_{Q_{r}} u \leq 1-\theta$$
We consider the rescaled version $$\tilde{u}(x,t) := u(r^{-1}x,r^{-2\sigma/\alpha} t).$$ The function $\tilde{u}$ will stay either positive or negative in half of the points in $B_1 \times [-2,-1]$. Let us assume that $\{\tilde{u} \leq 0\}\cap (B_1 \times [-2,-1])\geq |B_1|/2$. Otherwise we can repeat the proof for $-\tilde{u}$. We would like to apply Lemma \[l:decrease\]. To do so we would need $\tilde{u}\leq 1$. We consider $v:= \min\{1,\tilde{u}\}$. Inside $Q_{r^{-1}}$ we have $v=\tilde{u}$. The error comes only from the tails in the computations. Exactly as in [@s11] we obtain for $\kappa$ small enough $$-M^+ v \leq -M^+ \tilde{u} + \epsilon_0/4.$$ From Proposition \[p:nubound\] we have for small enough $\kappa$ that $${\mathcal{D}}_t^{\alpha} v \leq {\mathcal{D}}_t^{\alpha} \tilde{u} + \epsilon_0/4.$$ Thus $${\mathcal{D}}_t^{\alpha} v - M^+ v \leq \epsilon_0.$$ We now apply Lemma \[l:decrease\] to $v$ and rescale back to conclude the proof.
We are now able to give the proof of our main result.
We first choose $\kappa<\nu$ for $\nu$ as in Lemma \[l:down\] Let $(x_0,t_0) \in Q_1$. We consider the rescaled function $$v(x,t) = \frac{u(x_0 + x,t+t_0}{\| u\|_{L^{\infty}} + \epsilon_0^{-1} \| f\|_{L^{\infty}}},$$ and note that $|v|\leq 1$ and $v$ is a solution to $$\begin{aligned}
{\mathcal{D}}_t^{\alpha} v - M_{\sigma}^+v &\leq {\epsilon}_0 \\
{\mathcal{D}}_t^{\alpha} v - M_{\sigma}^-v &\geq -{\epsilon}_0.
\end{aligned}$$ in $B_2 \times [-1,0]$. We let $r=\min\{4^{-1},4^{-\alpha/2\sigma}\}$ and the estimate will follow as soon as we show $$\label{e:osc2}
\text{osc}_{Q_{r_k}} v \leq 2 r^{\kappa k}.$$ Estimate will be proven by constructing two sequences $a_k \leq v \leq b_k$ in $Q_{r_k}$, $b_k - a_k= 2r^{\kappa k}$ with $a_k$ nondecreasing and $b_k$ nonincreasing. The sequence is constructed inductively.
Since $|v|\leq 1$ everywhere, we can start by choosing some $a_0 \leq \inf v$ and $b_0 \geq \sup v$ so that $b_0 - a_0=2$. Assuming now that the sequences have been constructed up to the value $k$ we scale $$w(x,t) = (v(r^k x,r^{2k\sigma /\alpha}t) -(a_k+b_k)/2)r^{-\kappa k}.$$ We then have $$\begin{aligned}
|w| \leq 1 &\quad \text{in } \quad Q_1 \\
|w| \leq 2r^{-\kappa k} -1 &\quad \text{in } \quad Q_{r^{-k}}.
\end{aligned}$$ and so $$\begin{aligned}
|w(x,t)|\leq 2|x|^{\nu}-1 &\quad \text{for } \quad (x,t) \in B_1^c \times [-1,0] \\
|w(x,t)|\leq 2|t|^{\nu}-1 &\quad \text{for } \quad (x,t) \in B_1 \times (-\infty,-1).
\end{aligned}$$ Notice also that $w$ has new right hand side bounded by $$\epsilon_0 r^{k(\kappa-2\sigma)}$$ which is strictly smaller than $\epsilon_0$ for $\kappa<2\sigma$. For $\kappa$ small enough we can apply Lemma \[l:down\] to obtain $$\text{osc}_{ Q_r} w \leq 1-\theta.$$ Then if $\kappa$ is chosen smaller than the $\kappa$ in Lemma \[l:down\] and also so that $1-\theta \leq r^{\kappa}$, then this implies $$\text{osc}_{ Q_{r^{k+1}}} w \leq r^{\kappa(k+1)}$$ so we can find $a_{k+1}$ and $b_{k+1}$ and this finishes the proof.
| {
"pile_set_name": "ArXiv"
} |
---
author:
- Ugo Dal Lago
- Davide Sangiorgi
- Michele Alberti
bibliography:
- 'biblio.bib'
title: |
On Coinductive Equivalences\
for Higher-Order Probabilistic Functional Programs
---
Introduction
============
Preliminaries {#sect:p}
=============
Probabilistic Applicative Bisimulation and Howe’s technique {#sect:pasht}
===========================================================
The Discriminating Power of Probabilistic Contexts {#sect:dppc}
==================================================
Coupled Logical Bisimulation
============================
Beyond Call-by-Name Reduction {#sect:beyond}
=============================
A Comparison with Nondeterminism {#sect:comparison}
================================
Conclusions
===========
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
We prove a technical result, showing that the existence of a closable unbounded dual system in the sense of Voiculescu is equivalent to the finiteness of free Fisher information. This approach allows one to give a purely operator-algebraic proof of the computation of the non-microstates free entropy dimension for generators of groups carried out in an earlier joint work with I. Mineyev [@shlyakht-mineyev:freedim]. The same technique also works for finite-dimensional algebras.
We also show that Voiculescu’s question of semi-continuity of free entropy dimension, as stated, admits a counterexample. We state a modified version of the question, which avoids the counterexample, but answering which in the affirmative would still imply the non-isomorphism of free group factors.
author:
- 'Dimitri Shlyakhtenko [^1]'
title: 'Remarks on Free Entropy Dimension.'
---
Introduction. {#introduction. .unnumbered}
=============
Free entropy dimension was introduced by Voiculescu [@dvv:entropy2; @dvv:entropy3; @dvv:entropy5] both in the context of his microstates and non-microstates free entropy. We refer the reader to the survey [@dvv:entropysurvey] for a list of properties as well as applications of this quantity in the theory of von Neumann algebras.
The purpose of this note is to discuss several technical aspects related to estimates for free entropy dimension.
The first deals with existence of “Dual Systems of operators”, which were considered by Voiculescu [@dvv:entropy5] in connection with the properties of the difference quotient derivation, which is at the heart of the non-microstates definition of free entropy. We prove that if one considers dual systems of closed unbounded operators (as opposed to bounded operators as in [@dvv:entropy5]), then existence of a dual system becomes equivalent to finiteness of free Fisher information. Using these ideas allows one to give a purely operator-algebraic proof of the expression for the free entropy dimension of a set of generators of a group algebra in terms of the $L^{2}$ Betti numbers of the group [@shlyakht-mineyev:freedim], clarifying the reason for why the equality holds in the group case. We also point out that for the same reason one is able to express the non-microstates free entropy dimension of an $n$-tuple of generators of a finite-dimensional von Neumann algebra in terms of its $L^{2}$ Betti numbers. In particular, the microstates and non-microstates free entropy dimension is the same in this case.
The second aspect deals with the question of semi-continuity of free entropy dimension, as formulated by Voiculescu in [@dvv:entropy2; @dvv:entropy3]. We point out that a counterexample exists to the question of semi-continuity, as stated. However, the possibility that the free entropy dimension is independent of the choice of generators of a von Neumann algebra is not ruled out by the counterexample.
Unbounded Dual Systems and Derivations.
=======================================
Non-commutative difference quotients and dual systems.
------------------------------------------------------
Let $X_{1},\ldots,X_{n}$ be an $n$-tuple of self-adjoint elements in a tracial von Neumann algebra $M$. In [@dvv:entropy5], Voiculescu considered the densely defined derivations $\partial_{j}$ defined on the polynomial algebra $\mathbb{C}(X_{1},\ldots,X_{n})$ generated by $X_{1},\ldots,X_{n}$ and with values in $L^{2}(M)\otimes L^{2}(M)\cong HS(L^{2}(M))$, the space of Hilbert-Schmidt operators on $L^{2}(M)$. If we denote by $P_{1}:L^{2}(M)\to L^{2}(M)$ the orthogonal projection onto the trace vector $1$, then the derivations $\partial_{j}$ are determined by the requirement that $\partial_{j}(X_{i})=\delta_{ij}P_{1}$.
Voiculescu showed that if $\partial_{j}^{*}(P_{1})$ exists, then $\partial_{j}$ is closable. This is of interest because the existence of $\partial_{j}^{*}(P_{1})$, $j=1,\ldots,n$ is equivalent to finiteness of the free Fisher information of $X_{1},\ldots,X_{n}$ [@dvv:entropy5].
Also in [@dvv:entropy5], Voiculescu introduced the notion of a “dual system” to $X_{1},\ldots,X_{n}$. In his definition, such a dual system consists of an $n$ tuple of operators $Y_{1},\ldots,Y_{n}$, so that $[Y_{i},X_{j}]=\delta_{ij}P_{1}$, where Although Voiculescu required that the operators $Y_{j}$ be anti-self-adjoint, it will be more convenient to drop this requirement. However, this is not a big difference, since if $(Y'_{1},\ldots,Y'_{n})$ is another dual system, then $[Y_{i}-Y'_{i},X_{j}]=0$ for all $i,j$, and so $Y_{i}-Y'_{i}$ belongs to the commutant of $W^{*}(X_{1},\ldots,X_{n})$.
Note that the existence of a dual system is equivalent to the requirement that the derivations $\partial_{j}:\mathbb{C}(X_{1},\ldots,X_{n})\to HS\subset B(L^{2}(M))$ are inner as derivations into $B(L^{2}(M))$. In particular, Voiculescu showed that if a dual system exists, then $\partial_{j}:L^{2}(M)\to HS$ are actually closable, and $\partial_{j}^{*}(P_{1})$ is given by $(Y_{j}-JY_{j}^{*}J)1$. However, the existence of a dual system is a stronger requirement than the existence of $\partial_{j}^{*}(P_{1})$.
Dual systems of unbounded operators.
------------------------------------
More generally, given an $n$-tuple $T=(T_{1},\ldots,T_{n})\in HS^{n}$, we may consider a derivation $\partial_{T}:\mathbb{C}(X_{1},\ldots,X_{n})\to HS$ determined by $\partial_{T}(X_{j})=T_{j}$ [@shlyakht:qdim]. The particular case of $\partial_{j}$ corresponds to $T=(0,\ldots,P_{1},\ldots,0)$ ($P_{1}$ in $j$-th place).
\[thm:unboubdedDual\]Let $T\in HS^{n}$ and assume that $M=W^{*}(X_{1},\ldots,X_{n})$. The following are equivalent:\
(a) $\partial_{T}^{*}(P_{1})$ exists;\
(b) There exists a closable unbounded operator $Y:L^{2}(M)\to L^{2}(M)$, whose domain includes $\mathbb{C}(X_{1},\ldots,X_{n})$, so that $Y1=0$ and $1$ belongs to the domain of $Y^{*}$, and so that $[Y,X_{j}]=T_{j}$.
Assume first that (b) holds. Let $\xi=(Y-JY^{*}J)1=JY^{*}1$, which by assumptions on $Y$ makes sense. Then for any polynomial $Q\in\mathbb{C}(X_{1},\ldots,X_{n})$,$$\begin{aligned}
\langle\xi,P\rangle & = & \langle(Y-JY^{*}J)1,Q1\rangle\\
& = & \langle[Y,Q]1,1\rangle\\
& = & Tr(P_{1}[Y,Q])\\
& = & \langle P_{1},[Y,Q]\rangle_{HS}\\
& = & \langle P_{1},\partial_{T}(Q)\rangle_{HS},\end{aligned}$$ since the derivations $Q\mapsto\partial_{T}(Q)$ and $Q\mapsto[Y,Q]$ have the same values on generators and hence are equal on $\mathbb{C}(X_{1},\ldots,X_{n})$. But this means that $\xi=\partial_{T}^{*}(P_{1})$.
Assume now that (a) holds. If we assume that $Y1=0$, then the equation $[Y,X_{j}]=T_{j}$ determines an operator $Y:\mathbb{C}(X_{1},\ldots,X_{n})\to L^{2}(M)$. Indeed, if $Q$ is a polynomial in $X_{1},\ldots,X_{n}$, then we have$$Y(Q\cdot1)=[Y,Q]\cdot1-Q(Y\cdot1)=[Y,Q]\cdot1=\partial_{T}(Q)\cdot1.$$ To show that the operator $Y$ that we have thus defined is closable, it is sufficient to prove that a formal adjoint can be defined on a dense subset. We define $Y^{*}$ on $Q\in\mathbb{C}(X_{1},\ldots,X_{n})$ by$$Y^{*}(Q\cdot1)=-(\partial_{T}(Q^{*}))^{*}\cdot1+\partial_{T}^{*}(P_{1}).$$ Hence $Y^{*}\cdot1=\partial_{T}^{*}(P_{1})$ and $Y^{*}$ satisfies $[Y^{*},Q]=-(\partial_{T}(Q^{*}))^{*}=-[Y,Q^{*}]^{*}.$
It remains to check that $\langle YQ\cdot1,R\cdot1\rangle=\langle Q\cdot1,Y^{*}R\cdot1\rangle$, for all $Q,R\in\mathbb{C}(X_{1},\ldots,X_{n})$. We have:$$\begin{aligned}
\langle YQ\cdot1,R\cdot1\rangle & = & \langle[Y,Q]\cdot1,R\cdot1\rangle\\
& = & \langle1,-[Y^{*},Q^{*}]R\cdot1\rangle\\
& = & \langle1,Q^{*}Y^{*}R\cdot1\rangle-\langle1,Y^{*}Q^{*}R\cdot1\rangle\\
& = & \langle Q\cdot1,Y^{*}R\cdot1\rangle-\langle1,Y^{*}Q^{*}R\cdot1\rangle.\end{aligned}$$ Hence it remains to prove that $\langle1,Y^{*}Q^{*}R\cdot1\rangle=0$. To this end we write$$\begin{aligned}
\langle1,Y^{*}Q^{*}R\cdot1\rangle & = & \langle1,[Y^{*},Q^{*}R]\cdot1\rangle-\langle1,Q^{*}RY^{*}\cdot1\rangle\\
& = & \langle[R^{*}Q,Y]\cdot1,1\rangle-\langle R^{*}Q\cdot1,Y^{*}\cdot1\rangle\\
& = & Tr([R^{*}Q,Y]P_{1})-\langle R^{*}Q\cdot1,\partial_{T}^{*}(P_{1})\rangle\\
& = & \langle\partial_{T}(R^{*}Q),P_{1}\rangle_{HS}-\langle\partial_{T}(R^{*}Q),P_{1}\rangle_{HS}=0.\end{aligned}$$ Thus $Y$ is closable.
Let $M=W^{*}(X_{1},\ldots,X_{n})$. Then $\Phi^{*}(X_{1},\ldots,X_{n})<+\infty$ if and only if there exist unbounded essentially anti-symmetric operators $Y_{1},\ldots,Y_{n}:L^{2}(M)\to L^{2}(M)$ whose domain includes $\mathbb{C}(X_{1},\ldots,X_{n})$, and which satisfy $[Y_{j},X_{i}]=\delta_{ji}P_{1}$.
A slight modification of the first part of the proof of Theorem \[thm:unboubdedDual\] gives that if $Y_{1},\ldots,Y_{n}$ exist, then $\partial_{j}^{*}(P_{1})=(Y_{j}-JY_{j}J)1$ and hence $\Phi^{*}(X_{1},\ldots,X_{n})$ (which is by definition $\sum_{j}\Vert\partial_{j}^{*}(P_{1})\Vert_{2}^{2}$) is finite.
Conversely, if $\Phi^{*}(X_{1},\ldots,X_{n})<+\infty$, then $\partial_{j}^{*}(P_{1})$ exists for all $j$. Hence by Theorem \[thm:unboubdedDual\], we obtain non-self adjoint closable unbounded operators $Y_{1},\ldots,Y_{n}$ so that the domains of $Y_{j}$and $Y_{j}^{*}$ include $\mathbb{C}(X_{1},\ldots,X_{n})$, and so that $[Y_{j},X_{i}]=\delta_{ji}P_{1}$. Now since $X_{j}=X_{j}^{*}$ we also have $[Y_{j}^{*},X_{i}]=-\delta_{ji}P_{1}^{*}=-\delta_{ji}P_{1}$. Hence if we set $\tilde{Y}_{j}=\frac{1}{2}(Y_{j}-Y_{j}^{*})$, we obtain the desired operators.
Dual systems and $L^{2}$ cohomology.
====================================
Let as before $X_{1},\ldots,X_{n}\in(M,\tau)$ be a family of self-adjoint elements.
In conjunctions with estimates on free entropy dimension [@shlyakht:qdim; @shlyakht-mineyev:freedim] and $L^{2}$ cohomology [@connes-shlyakht:l2betti], it is interesting to consider the following spaces:$$\begin{aligned}
H_{0} & = & {\operatorname{cl}}{\{ T=(T_{1},\ldots,T_{n})\in HS^{n}:} \\
& & \qquad \exists Y\in B(L^{2}(M))\ \ [Y,X_{j}]=T_{j}\}.\end{aligned}$$ Here ${\operatorname{cl}}$ refers to closure in the Hilbert-Schmidt topology. We also consider $$\begin{aligned}
H_{1} & = & \textrm{span}\ {\operatorname{cl}}{\{ T=(T_{1},\ldots,T_{n})\in HS^{n}:\exists
Y=Y^*\ \textrm{unbounded densely defined}}\\
& & \qquad \textrm{with $1$ in the domain of }Y,\ [Y,X_{j}]=T_{j},\ j=1,\ldots,n\}.\end{aligned}$$ Note that in particular, $H_{0}\subset H_{1}$.
One has the following estimates [@shlyakht:qdim; @connes-shlyakht:l2betti]:$$\dim_{M\bar{\otimes}M^{o}}H_{0}\leq\delta^{*}(X_{1},\ldots,X_{n})\leq\delta^{\star}(X_{1},\ldots,X_{n})\leq\Delta(X_{1},\ldots,X_{n}).$$ The main result of this section is the following theorem, whose proof has similarities with the Sauvageot’s theory of quantum Dirichlet forms [@sauvageot:dirichlet]:
\[thm:H1isH2\]$H_{0}=H_{1}$.
It is sufficient to prove that $H_{0}$ is dense in $H_{1}$.
Let $T=(T_{1},\ldots,T_{n})\in HS^{n}$ be such that $T_{j}^{*}=T_{j}=[\I A,X_{j}]$, $j=1,\ldots,n$, with $A=A^*$ a closed unbounded operator and $1$ in the domain of $A$.
For each $0<R<\infty$, let now $f_{R}:\mathbb{R}\to\mathbb{R}$ be a $C^{\infty}$ function, so that
1. $f_{R}(x)=x$ for all $-R\leq x\leq R$;
2. $|f_{R}(x)|\leq R+1$ for all $x$;
3. the difference quotient $g_{R}(s,t)=\frac{f_{R}(s)-f_{R}(t)}{s-t}$ is uniformly bounded by $2$.
Let $A_{R}=f_{R}(A)$ and let $T_{j}^{(R)}=[\I A_{R},X_{j}]$. Note that for each $R$, $T^{(R)}=(T_{1}^{(R)},\ldots,T_{n}^{(R)})\in H_{0}$. Hence it is sufficient to prove that $T^{(R)}\to T$ in Hilbert-Schmidt norm as $R\to\infty$.
Let $\mathcal{A}\cong L^{\infty}(\mathbb{R},\mu)$ be the von Neumann algebra generated by the spectral projections of $A$; hence $A_{R}\in\mathcal{A}$ for all $R$. If we regard $L^{2}(M)$ as a module over $A$, then $HS=L^{2}(M)\bar{\otimes}L^{2}(M)$ is a bimodule over $\mathcal{A}$, and hence a module over $\mathcal{A}\bar{\otimes}\mathcal{A}\cong L^{\infty}(\mathbb{R}^{2},\mu\times\mu)$ in such a way that if $s,t$ are coordinates on $\mathbb{R}^{2}$, and $Q\in HS$, then $sQ=AQ$ and $tQ=QA$ (more precisely, for any bounded measurable function $f$, $f(s)Q=f(A)Q$ and $f(t)Q=Qf(A)$). In particular, we can identify, up to multiplicity, $HS$ with $L^{2}(\mathbb{R}^{2},\mu\times\mu)$.
It is not hard to see that then$$[f(A),X_{j}]=g\cdot[A,X_{j}],$$ where $g$ is the difference quotient $g(s,t)=(f(s)-f(t))/(s-t)$. Indeed, it is sufficient to verify this equation on vectors in $\mathbb{C}[X_{1},\ldots,X_{n}]$ for $f$ a polynomial in $A$, in which case the result reduces to$$[A^{n},X_{j}]=\sum_{k=0}^{n-1}A^{k}[A,X_{j}]A^{n-k-1}=\frac{s^{n}-t^{n}}{s-t}\cdot[A,X_{j}].$$ It follows that$$T_{j}^{(R)}=[A_{R},X_{j}]=[f_{R}(A),X_{j}]=g_{R}(A)\cdot[A,X_{j}]=g_{R}(A)\cdot T_{j}.$$ Now, since $g_{R}(A)$ are bounded and $g_{R}(A)=1$ on the square $-R\leq s,t\leq R$, it follows that multiplication operators $g_{R}(A)$ converge to $1$ ultra-strongly as $R\to\infty$. Since $HS$ is a multiple of $L^{2}(\mathbb{R}^{2},\mu\times\mu)$, it follows that $g_{R}(A)T_{j}\to T_{j}$ in Hilbert-Schmidt norm. Hence $T^{(R)}\to T$ as $R\to\infty$.
As a corollary, we re-derive the main result of [@shlyakht-mineyev:freedim] (the difference is that we use Theorem \[thm:H1isH2\] instead of the more combinatorial argument [@shlyakht-mineyev:freedim]; we sketch the proof to emphasize the exact point at which the fact that we are dealing with a group algebra becomes completely clear):
Let $X_{1},\ldots,X_{n}$ be generators of the group algebra $\mathbb{C}\Gamma$. Then $\delta^{\star}(X_{1},\ldots,X_{n})=\delta^{*}(X_{1},\ldots,X_{n})=\Delta(X_{1},\ldots,X_{n})=\beta_{1}^{(2)}(\Gamma)-\beta_{0}^{(2)}(\Gamma)+1$, where $\beta_{j}^{(2)}(\Gamma)$ are the $L^{2}$-Betti numbers of $\Gamma$.
(Sketch). We first point out that in the preceding we could have worked with self-adjoint families $F=(X_{1},\ldots,X_{n})$ rather than self-adjoint elements (all we ever needed was that $X\in F\Rightarrow X^{*}\in F$).
By [@shlyakht-mineyev:freedim], we may assume that $X_{j}\in\Gamma\subset\mathbb{C}\Gamma$, since the dimension of $H_{0}$ depends only on the pair $\mathbb{C}(X_{1},\ldots,X_{n})$ and its trace.
Recall [@connes-shlyakht:l2betti] that $\Delta(X_{1},\ldots,X_{n})=\dim_{M\bar{\otimes}M^{o}}H_{2}$, where$$H_{2}=\{(T_{1},\ldots,T_{n})\in HS:\exists Y^{(k)}\in HS\ \textrm{s.t. }[Y^{(k)},X_{j}]\to T_{j}\ \textrm{weakly}\}.$$ By [@connes-shlyakht:l2betti], $\Delta(X_{1},\ldots,X_{n})=\beta_{1}^{(2)}(\Gamma)-\beta_{0}^{(2)}(\Gamma)+1$; moreover, from the proof we see that in the group case,$$H_{2}={\operatorname{cl}}\{MXM\},$$ where $$\begin{aligned}
X&=&\{(T_{1}X_{1},\ldots,T_{n}X_{n}):T_{j}\in\ell^{2}(\Gamma),\\
& & \qquad T_{j}\textrm{ is the value of some }\ell^{2}\ \textrm{group cocycle on }X_{j}\},\end{aligned}$$ and where we think of $\ell^{2}(\Gamma)\subset HS$ as “diagonal operators” by sending a sequence $(a_{\gamma})_{\gamma\in\Gamma}\in\ell^{2}(\Gamma)$ to the Hilbert-Schmidt operator $\sum a_{\gamma}P_{\gamma}$, where $P_{\gamma}$ is the rank $1$ projection onto the subspace spanned by the delta function supported on $\gamma$.
Let $\mathfrak{F}$ be the space of all functions on $\Gamma$. Since the group cohomology $H^{1}(\Gamma;\mathfrak{F}(\Gamma))$ is clearly trivial, it follows that if $c$ is any $\ell^{2}$-cocycle on $\Gamma$, then $c(X_{j})=f(X_{j})-f(e)$, for some $f\in\mathfrak{F}$. Hence$$X=\{([f,X_{1}],\ldots,[f,X_{n}]):f\in\mathfrak{F}\}\cap HS^{n}.$$ Since *every element of $\mathfrak{F}$ is automatically an essentially self-adjoint operator on $\ell^{2}(\Gamma)$*, whose domain includes $\mathbb{C}\Gamma$ we obtain that$$MXM\subset H_{1}.$$ In particular, $H_{2}\subset H_{1}$. Hence $$\dim_{M\bar{\otimes}M^{o}}H_{2}=\beta_{1}^{(2)}(\Gamma)-\beta_{0}^{(2)}(\Gamma)+1\leq\dim_{M\bar{\otimes}M^{o}}H_{1}\leq\dim_{M\bar{\otimes}M_{0}}H_{2},$$ which forces $H_{1}=H_{2}$. Since $H_{0}=H_{1}$, we get that in the following equation$$\dim_{M\bar{\otimes}M^{o}}H_{0}\leq\delta^{*}\leq\delta^{\star}\leq\dim_{M\bar{\otimes}M^{o}}H_{2}=\beta_{1}^{(2)}(\Gamma)-\beta_{0}^{(1)}(\Gamma)+1$$ all inequalities are forced to be equalities, which gives the result.
Let $(M,\tau)$ be a finite-dimensional algebra, and let $X_{1},\ldots,X_{n}$ be any of its self-adjoint generators. Then $\delta^{*}(X_{1},\ldots,X_{n})=\delta^{\star}(X_{1},\ldots,X_{n})=\Delta(X_{1},\ldots,X_{n})=1-\beta_{0}(M,\tau)=\delta_{0}(X_{1},\ldots,X_{n})$.
As in the proof of the last corollary, we have the inequalities$$\dim_{M\bar{\otimes}M^{o}}H_{0}\leq\delta^{*}\leq\delta^{\star}\leq\dim_{M\bar{\otimes}M^{o}}H_{2},$$ where$$H_{2}=\{(T_{1},\ldots,T_{n})\in HS:\exists Y^{(k)}\in HS\ \textrm{s.t. }[Y^{(k)},X_{j}]\to T_{j}\ \textrm{weakly}\}.$$ Since $L^{2}(M)$ is finite-dimensional, there is no difference between weak and norm convergence; hence $H_{2}$ is in the (norm) closure of $\{(T_{1},\ldots,T_{n}):\exists Y\in HS\ \textrm{s.t. }T_{j}=[Y,X_{j}]\}\subset H_{0}$; since $H_{0}$ is closed, we get that $H_{0}=H_{1}$ and so all inequalities become equalities. Moreover,$$\dim_{M\bar{\otimes}M^{o}}H_{2}=\Delta(X_{1},\ldots,X_{n})=1-\beta_{0}(M,\tau)$$ (see [@connes-shlyakht:l2betti]).
Comparing the values of $1-\beta_{0}(M,\tau)$ with the computations in [@jung-freexentropy] gives $\delta_{0}(X_{1},\ldots,X_{n})=\delta^{*}(X_{1},\ldots,X_{n})$.
Some Remarks on Semi-Continuity of Free Dimension.
==================================================
In [@dvv:entropy2; @dvv:entropy3], Voiculescu asked the question of whether the free dimension $\delta$ satisfies the following semi-continuity property. Let $X_{j}^{(k)},X_{j}\in(M,\tau)$ be self-adjoint variables, $j=1,\ldots,n$, $k=1,2,\ldots$, and assume that $X_{j}^{(k)}\to X_{j}$ strongly, $\sup_{k}\Vert X_{j}^{(k)}\Vert<\infty$. Then is it true that$$\liminf_{k}\delta(X_{1}^{(k)},\ldots,X_{n}^{(k)})\geq\delta(X_{1},\ldots,X_{n})?$$ As shown in [@dvv:entropy2; @dvv:entropy3], a positive answer to this question (or a number of related questions, where $\delta$ is replaced by some modification, such as $\delta_{0}$, $\delta^{*}$, etc.) implies non-isomorphism of free group factors. In the case of $\delta_{0}$. a positive answer would imply that the value of $\delta_{0}$ is independent of the choice of generators of a von Neumann algebra.
Although this question is very natural from the geometric standpoint, we give a counterexample, which shows that some additional assumptions on the sequence $X_{j}^{(k)}$ are necessary. Fortunately, the kinds of properties of $\delta$ that would be required to prove the non-isomorphism of free group factors are not ruled out by this counterexample (see Question \[pro:semicontinuity\]).
We first need a lemma.
\[lem:genFn\]Let $X_{1},\ldots,X_{n}$ be any generators of the group algebra of the free group $\mathbb{F}_{k}$. Then $\delta_{0}(X_{1},\ldots,X_{n})=\delta(X_{1},\ldots,X_{n})=\delta^{*}(X_{1},\ldots,X_{n})=\delta^{\star}(X_{1},\ldots,X_{n})=k$.
Note that by [@guionnet-biane-capitaine:largedeviations] we always have$$\delta_{0}(X_{1},\ldots,X_{n})\leq\delta(X_{1},\ldots,X_{n})\leq\delta^{*}(X_{1},\ldots,X_{n})\leq\delta^{\star}(X_{1},\ldots,X_{n});$$ furthermore, by [@shlyakht-mineyev:freedim], $\delta^{\star}(X_{1},\ldots,X_{n})=k$. Since $\delta_{0}$ is an algebraic invariant [@dvv:improvedrandom], $\delta_{0}(X_{1},\ldots,X_{n})=\delta_{0}(U_{1},\ldots,U_{k})$, where $U_{1},\ldots,U_{k}$ are the free group generators. Then by [@dvv:entropy3], $\delta_{0}(U_{1},\ldots,U_{k})=k$. This forces equalities throughout.
Let $u,v$ be two free generators of $\mathbb{F}_{2}$, and consider the map $\phi:\mathbb{F}_{2}\to\mathbb{Z}/2\mathbb{Z}=\{0,1\}$ given by $\phi(u)=\phi(v)=1$. The kernel of this map is a subgroup $\Gamma$ of $\mathbb{F}_{2}$, which is isomorphic to $\mathbb{F}_{3}$, having as free generators, e.g. $u^{2}$, $v^{2}$ and $uv$. Let $X_{1}^{(k)}={\operatorname{Re\,}}u^{2}$, $X_{2}^{(k)}={\operatorname{Im\,}}u^{2}$, $Y_{1}^{(k)}={\operatorname{Re\,}}v^{2}$, $Y_{2}^{(k)}={\operatorname{Im\,}}v^{2}$, $Z_{1}^{(k)}={\operatorname{Re\,}}uv$, $Z_{2}^{(k)}={\operatorname{Im\,}}uv$, $W_{1}^{(k)}=\frac{1}{k}{\operatorname{Re\,}}u$, $W_{2}^{(k)}=\frac{1}{k}{\operatorname{Im\,}}v$.
Thus if $X_{1}={\operatorname{Re\,}}u^{2}$, $X_{2}={\operatorname{Im\,}}u^{2}$, $Y_{1}={\operatorname{Re\,}}v^{2}$, $Y_{2}={\operatorname{Im\,}}v^{2}$, $Z_{1}={\operatorname{Re\,}}uv$, $Z_{2}={\operatorname{Im\,}}uv$, $W_{1}=0$, $W_{2}=0$, then $X_{j}^{(k)}\to X_{j}$, $Y_{j}^{(k)}\to Y_{j}$, $Z_{j}^{(k)}\to Z_{j}$ and $W_{j}^{(k)}\to W_{j}$ (in norm, hence strongly).
Note finally that for $k$ finite, $X_{1}^{(k)},X_{2}^{(k)},Y_{1}^{(k)},Y_{2}^{(k)},Z_{1}^{(k)},Z_{2}^{(k)},W_{1}^{(k_{,})},W_{2}^{(k)}$ generate the same algebra as $u^{2},v^{2},uv,\frac{1}{k}v$, which is the same as the algebra generated by $u$ and $v$, i.e., the entire group algebra of $\mathbb{F}_{2}$. Hence $\delta(X_{1}^{(k)},X_{2}^{(k)},Y_{1}^{(k)},Y_{2}^{(k)},Z_{1}^{(k)},Z_{2}^{(k)},W_{1}^{(k_{,})},W_{2}^{(k)})=2$ by Lemma \[lem:genFn\]. Hence$$\liminf_{k}\delta(X_{1}^{(k)},X_{2}^{(k)},Y_{1}^{(k)},Y_{2}^{(k)},Z_{1}^{(k)},Z_{2}^{(k)},W_{1}^{(k_{,})},W_{2}^{(k)})=2$$ On the other hand, $X_{1},X_{2},Y_{1},Y_{2},Z_{1,}Z_{2},W_{1},W_{2}$ generate the same algebra as $u^{2},v^{2},uv,0$, i.e., the group algebra of $\Gamma\cong\mathbb{F}_{3}.$ Hence$$\delta(X_{1},X_{2},Y_{1},Y_{2},Z_{1},Z_{2},W_{1},W_{2})=3,$$ which is the desired counterexample.
The same example (in view of Lemma \[lem:genFn\]) also works for $\delta_{0}$, $\delta^{*}$ and $\delta^{\star}$.
The following two versions of the question are not ruled out by the counterexample. If either version were to have a positive answer, it would still be sufficient to prove non-isomorphism of free group factors:
\[pro:semicontinuity\](a) Let $X_{j}^{(k)},X_{j}\in(M,\tau)$ be self-adjoint variables, $j=1,\ldots,n$, $k=1,2,\ldots$, and assume that $X_{j}^{(k)}\to X_{j}$ strongly, $\sup_{k}\Vert X_{j}^{(k)}\Vert<\infty$. Assume that $X_{1},\ldots,X_{n}$ generate $M$ and that for each $k$, $X_{1}^{(k)},\ldots,X_{n}^{(k)}$ also generate $M$. Then is it true that$$\liminf_{k}\delta(X_{1}^{(k)},\ldots,X_{n}^{(k)})\geq\delta(X_{1},\ldots,X_{n})?$$ (b) A weaker form of the question is the following. Let $X_{j}^{(k)},X_{j},Y_{j}\in(M,\tau)$ be self-adjoint variables, $j=1,\ldots,n$, $k=1,2,\ldots$, and assume that $X_{j}^{(k)}\to X_{j}$ strongly, $\sup_{k}\Vert X_{j}^{(k)}\Vert<\infty$. Assume that $Y_{1},\ldots,Y_{m}$ generate $M$. Then is it true that$$\liminf_{k}\delta(X_{1}^{(k)},\ldots,X_{n}^{(k)},Y_{1},\ldots,Y_{m})\geq\delta(X_{1},\ldots,X_{n},Y_{1},\ldots,Y_{m})?$$
We point out that in the case of $\delta_0$, these questions are actually equivalent to each other and to the statement that $\delta_0(Z_1,\ldots,Z_n)$ only depends on the von Neumann algebra generated by $Z_1,\ldots,Z_n$.
Indeed, it is clear that (a) implies (b).
On the other hand, if we assume that (b) holds, then we can choose $X_{j}^{(k)}$ to be polynomials in $Y_{1},\ldots,Y_{m}$, so that $\delta_{0}(X_{1}^{(k)},\ldots,X_{n}^{(k)},Y_{1},\ldots,Y_{m})=\delta_{0}(Y_{1},\ldots,Y_{m})$ by [@dvv:improvedrandom]. Hence $\delta_{0}(Y_{1},\ldots,Y_{m})\geq\delta_{0}(X_{1},\ldots,X_{n},Y_{1},\ldots,Y_{m})\geq\delta_{0}(Y_{1},\ldots,Y_{m})$, where the first inequality is by (b) and the second inequality is proved in [@dvv:entropy3]. Hence if $W^{*}(X_{1},\ldots,X_{n})=W^{*}(Y_{1},\ldots,Y_{m})$, then one has $\delta_{0}(X_{1},\ldots,X_{n})=\delta_{0}(X_{1},\ldots,X_{n},Y_{1},\ldots,Y_{m})=\delta_{0}(Y_{1},\ldots,Y_{m})$. Hence (b) implies that $\delta_0$ is the same on any generators of $M$.
Lastrly, if we assume that $\delta_{0}$ is an invariant of the von Neumann algebra, then (a) clearly holds, since the value of $\delta_{0}(X_{1}^{(k)},\ldots,X_{n}^{(k)})$ is then independent of $k$ and is equal to $\delta_{0}(X_{1},\ldots,X_{n},Y_{1},\ldots,Y_{m})$.
[10]{}
P. Biane, M. Capitaine, and A. Guionnet. Large deviation bounds for matrix [B]{}rownian motion. , 152(2):433–459, 2003.
A. Connes and D. Shlyakhtenko. . Preprint math.OA/0309343, to appear in J. Reine Angew. Math.
Kenley Jung. The free entropy dimension of hyperfinite von [N]{}eumann algebras. , 355(12):5053–5089 (electronic), 2003.
I. Mineyev and D. Shlyakhtenko. Non-microstates free entropy dimension for groups. Preprint, math.OA/0312242, to appear in GAFA.
J.-L. Sauvageot. Quantum dirichlet forms, differential calculus, and semigroups. In [*Quantum Probability and Applications V*]{}, volume 1442 of [ *Lecture Notes in Mathematics*]{}, pages 531–536. Springer-Verlag, 1990.
D. Shlyakhtenko. Some estimates for non-microstates free entropy dimension, with applications to $q$-semicircular families. , 51:2757–2772, 2004.
D.-V. Voiculescu. The analogues of entropy and of [Fisher’s]{} information measure in free probability theory [II]{}. , 118:411–440, 1994.
D.-V. Voiculescu. The analogues of entropy and of [Fisher]{}’s information measure in free probability theory, [III]{}. , 6:172–199, 1996.
D.-V. Voiculescu. The analogues of entropy and of [Fisher]{}’s information measure in free probabilility, [V]{}. , 132:189–227, 1998.
D.-V. Voiculescu. A strengthened asymptotic freeness result for random matrices with applications to free entropy. , 1:41 – 64, 1998.
D.-V. Voiculescu. Free entropy. , 34(3):257–278, 2002.
[^1]: Research supported by the National Science Foundation.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'AGN with the so-called ‘double-double’ radio structure have been interpreted as restarted AGN where the inner structure is a manifestation of a new phase of activity which happened to begin before the outer radio lobes resulting from the previous one had faded completely. The radio galaxy 1245+676 is an extreme example of such a double-double object — its outer structure, measuring $970\,h^{-1}$kpc, is five orders of magnitude larger than the $9.6\,h^{-1}$pc inner one. We present a series of VLBI observations of the core of 1245+676 which appears to be a compact symmetric object (CSO). We have detected the motion of the CSO’s lobes, measured its velocity, and inferred the kinematic age of that structure.'
author:
- 'Andrzej Marecki$^{1}$, Peter D. Barthel$^{2}$, Antonis Polatidis$^{3}$'
- 'and Izabela Owsianik$^{3}$'
title: '1245+676 — A CSO/GPS Source being an Extreme Case of a Double-Double Structure'
---
=2em
=15.5cm =22.6 cm =-1.0 cm =0.5cm =0.5cm
[$^1$ Toruń Centre for Astronomy, Nicholas Copernicus University, Toruń, Poland\
$^2$ Kapteyn Institute, Department of Astronomy, Groningen University, Netherlands\
$^3$ Max-Planck-Institut für Radioastronomie, Bonn, Germany\
]{}
[**Keywords:**]{} galaxies: active — galaxies: individual (1245+676)
Objects with the ‘Double-Double’ Radio Structure
================================================
Radio galaxies which are not beamed toward the observer are normally perceived as double structures. The majority of them are large scale objects (LSOs) with angular sizes ranging from several arcseconds to several arcminutes. These angular sizes translate to large linear sizes of the order of $10^5-10^6$pc; galaxies with sizes $>1$Mpc are labelled ‘giant radio galaxies’ (GRGs). A very interesting exception to this (relatively simple) picture exists however: a few LSOs, which are also double at first sight, turn out to have so-called double-double structure, radio galaxy 1450+333 (Schoenmakers et al. 2000) being a prime example.
LSO radio lobes are powered by central engines for a maximum of approximately $10^7$ years (Alexander & Leahy 1987; Liu, Pooley, & Riley 1992). If the nuclear energy supply stops, the extended radio structure will stop growing, its luminosity drops and the spectrum gradually gets steeper and steeper because of radiation and expansion losses. Komissarov & Gubanov (1994) estimate fade-away time scales of several $10^7$ years which is comparable to the timescale of the activity itself.
After that time the radio structure should in principle disappear, however there are a number of known mechanisms of restarting activity. For example, Hatziminaoglou, Siemiginowska, & Elvis (2001) elaborated a theory of thermal–viscous instabilities in the accretion disks of supermassive black holes (SMBHs). It predicts that the activity is recurrent and the length of the activity phase as well as the timescale of the activity re-occurrence is controlled by the mass of the SMBH. In favourable circumstances a new phase of activity can start before the radio lobes resulting from the previous one have faded completely. This would lead to double-double radio structure.
Inner and Outer Radio Structures of 1245+676
============================================
1245+676 is a classical double-lobed radio galaxy redshifted to $z=0.1073$. Its angular size amounts to $12\farcm4$ (Lara et al. 2001) which is equivalent to $970\,h^{-1}$kpc (assuming $q_0=0.5$). Figure \[NVSS\] shows the NVSS map of 1245+676. The central component is noticeably strong. Indeed, according to Lara et al. (2001) $2/3$ of the whole flux at 1.4GHz is emitted by the core. Moreover the core has a gigahertz-peaked spectrum (GPS) — see for example the spectrum shown in Marecki et al. (1999).
A number of VLBI observations of the core of 1245+676 have been conducted in different epochs. More technical details about these observations will be given elsewhere. In Figure \[6cm\] we present a selection of resulting 5GHz images; in Figure \[hifreq\] we show the 8.4GHz map made in 1991 and the most detailed VLBI map of the 1245+676 core observed in 1998 at 15GHz. These maps clearly show that the ‘core’ is actually a compact symmetric object (CSO) with a double-lobed inner structure. Its size is only $9.6\,h^{-1}$pc. This means 1245+676 is an extreme case of a double-double object — its outer structure is five orders of magnitude larger than the inner one.
We are confident that the core of 1245+676 is indeed a CSO because based on our dual frequency (5 and 15GHz) quasi-simultaneous VLBA observations made on 6 September 1998 we were able to calculate the spectral indices of the milliarcsecond scale components. They amount to $\alpha_{North}=-1.03$ and $\alpha_{South}=-1.23$ so they are both steep and roughly equal. We can therefore safely rule out a possibility they are not lobes but, for example, a core and a jet. (This approach is similar to that adopted by Peck & Taylor 2000.)
We note a modest misalignment between the parsec scale ($PA=-19\degr$) and the megaparsec scale ($PA=-50\degr$) which may be attributed to precession of SMBH and the accretion disk.
Expansion Velocity and the Kinematic Age
========================================
Given the substantial time baseline between the 5GHz observations we attempted the measurement of the lobes’ expansion velocity. We used different software tools from AIPS and Difmap packages, and the final expansion figure is an average of the measurements obtained. It amounts to $0.0349\pm0.0018\,{\mathrm
mas}\,{\mathrm yr^{-1}}$, which is equivalent to $v_{exp}=0.164\pm0.008\,h^{-1}c$. The inferred kinematic age of the CSO structure is 191 years.
References {#references .unnumbered}
==========
Alexander, P., & Leahy, J.P. 1987, MNRAS, 225, 1 Hatziminaoglou, E., Siemiginowska, A., & Elvis, M.2001, ApJ, 547, 90
Komissarov, S.S., & Gubanov, A.G. 1994, A&A, 285, 27
Lara, L., Cotton, W.D., Feretti, L., Giovannini, G., Marcaide, J.M., Márquez, I., & Venturi, T. 2001, A&A, 370, 409
Liu, R., Pooley, G.G., & Riley, J.M. 1992, MNRAS, 257, 545
Marecki, A., Falcke, H., Niezgoda, J., Garrington, S.T., & Patnaik, A.R. 1999, A&AS, 135, 273
Peck, A.B., & Taylor, G.B. 2000, ApJ, 534, 90
Schoenmakers, A.P., de Bruyn, A.G., Röttgering, H.J.A., van der Laan, H., & Kaiser, C.R. 2000, MNRAS, 315, 371
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'In this paper we study the quantitative recurrence properties of self-conformal sets $X$ equipped with the map $T:X\to X$ induced by the left shift. In particular, given a function $\varphi:{\ensuremath{\mathbb{N}}}\to(0,\infty),$ we study the metric properties of the set $$R(T,\varphi)=\left\{x\in X:|T^nx-x|<\varphi(n)\textrm{ for infinitely many }n\in {\ensuremath{\mathbb{N}}}\right\}.$$ Our main result is that under the open set condition, for the natural measure $\mu$ supported on $X$, $\mu$-almost every $x\in X$ is contained in $R(T,\varphi)$ when an appropriate volume sum diverges. We also prove a complementary result which states that for self-similar sets satisfying the open set condition, when the volume sum converges, then $\mu$-almost every $x\in X$ is not contained in $R(T,\varphi).$'
address:
- 'Simon Baker: School of Mathematics, University of Birmingham, Birmingham, B15 2TT, UK'
- 'Michael Farmer: Mathematics institute, University of Warwick, Coventry, CV4 7AL, UK'
author:
- Simon Baker and Michael Farmer
title: 'Quantitative recurrence properties for self-conformal sets'
---
Introduction {#sec:1}
============
The notion of recurrence is of central importance within Dynamical Systems and Ergodic Theory. A well known theorem due to Poincaré states that if $(X,\mathcal{B},\mu)$ is a probability space, and $T:X\to X$ is a measure preserving transformation, then for any $E\in \mathcal{B}$ we have $$\mu\left(\{x\in E:T^nx\in E \textrm{ for infinitely many }n\in {\ensuremath{\mathbb{N}}}\}\right)=\mu(E).$$ If $X$ is endowed with a metric $d$ so that $(X,d)$ is a separable metric space, and $\mathcal{B}$ is the Borel $\sigma$-algebra, then Poincaré’s theorem allows us to conclude the following topological statement: $$\label{Topological reccurrence}
\liminf_{n\to\infty}d(T^nx,x)=0$$for $\mu$-a.e. $x\in X$. We call $(X,\mathcal{B},\mu,d,T)$ a metric measure-preserving system or an m.m.p.s. The information provided by is qualitative in nature. It tells us nothing about the speed at which an orbit can recur upon its initial point. One of the first general quantitative recurrence results was proved by Boshernitzan in [@Bos].
Let $(X,\mathcal{B},\mu,d,T)$ be a m.m.p.s. Assume that for some $\alpha>0$ the $\alpha$-dimensional Hausdorff measure $\mathcal{H}^{\alpha}$ is $\sigma$-finite on $(X,d)$. Then for $\mu$-a.e $x\in X$ we have $$\liminf_{n\to\infty} n^{1/\alpha}d(T^nx,x)<\infty.$$ Moreover, if $\mathcal{H}^{\alpha}(X)=0$ then for $\mu$-a.e $x\in X$ $$\liminf_{n\to\infty} n^{1/\alpha}d(T^nx,x)=0.$$
Building upon the work of Boshernitzan, Barreira and Saussol in [@BarSol] showed how the lower local dimension of a measure can be used to obtain quantitative recurrence results.
If $T:X\to X$ is a Borel measurable map on $X\subset {\ensuremath{\mathbb{R}}}^d$, and $\mu$ is a $T$-invariant Borel probability measure on $X,$ then for $\mu$-almost every $x\in X,$ we have $$\liminf_{n\to\infty}d(T^nx,x)=0\textrm{ for any }\alpha>\liminf_{r\to 0}\frac{\log \mu(B(x,r))}{\log r}.$$
A suitable framework for describing recurrence quantitatively is the following. Given $(X,\mathcal{B},\mu,d,T)$ an m.m.p.s. and $\varphi:{\ensuremath{\mathbb{N}}}\times X\to (0,\infty)$, let $$R(T,\varphi):=\left\{x\in X:d(T^nx,x)<\varphi(n,x) \textrm{ for infinitely many }n\in {\ensuremath{\mathbb{N}}}\right\}.$$ Typically one is interested in determining the metric properties of $R(T,\varphi)$ and relating these properties to $T$ and $\varphi$. This was the line of research pursued by Tan and Wang in [@TanWang] where they calculated the Hausdorff dimension of $R(T,\varphi)$ when $T$ is the $\beta$-transformation. Later Seuret and Wang proved similar results for self-conformal sets in [@SeuretWang]. As remarked upon by Chang et al in [@ChangWuWu], very few results exist on the Hausdorff measure of $R(T,\varphi)$. In this paper we continue the line of research instigated in [@ChangWuWu] and obtain results on the Hausdorff measure of $R(T,\varphi)$ when $X$ is a self-conformal set and $T$ is the natural map induced by the left shift. Before introducing our problem formally, we would like to mention a related topic and include some references.
The shrinking target problem is concerned with determining the speed at which the orbit of a $\mu$-typical point accumulates on a fixed point $x_0$. The shrinking target problem and the problem of obtaining quantitative recurrence results have many common features. One can define a suitable analogue of the set $R(T,\varphi)$ and ask what are its metric properties. For the shrinking target problem much more is known about the Hausdorff measure of this set see [@CheKle; @FMP], for results on the Hausdorff dimension of this set see [@HV; @LWWX; @WW].
Statement of results
--------------------
Let $V\subset {\ensuremath{\mathbb{R}}}^d$ be an open set, a $C^{1}$ map $\phi:V\to {\ensuremath{\mathbb{R}}}^{d}$ is a conformal mapping if it preserves angles. Equivalently $\phi$ is a conformal mapping if the differential $\phi'$ satisfies $|\phi'(x)y|=|\phi'(x)||y|$ for all $x\in V$ and $y\in {\ensuremath{\mathbb{R}}}^{d}$. Let $\Phi=\{\phi_{i}\}_{i\in {\mathcal{D}}}$ be a finite set of contractions on a compact set $Y\subset {\ensuremath{\mathbb{R}}}^d$, i.e. there exists $r\in(0,1)$ such that $|\phi_i(x)-\phi_i(y)|\leq r|x-y|$ for all $x,y\in Y$. We say that $\Phi$ is a conformal iterated function system if each $\phi_i$ can be extended to an injective conformal contraction on some open connected neighbourhood $V$ that contains $Y$ and $0<\inf_{x\in V} |\phi_i'(x)|\leq \sup_{x\in V}|\phi_i'(x)|<1$. Throughout this paper we will assume that the differentials are Hölder continuous. This means there exists $\alpha>0$ and $c>0$ such that $$||\phi_i'(x)|-|\phi_i'(y)||\leq c|x-y|^\alpha$$ for all $x,y\in V$. A well known result due to Hutchinson [@Hut] implies that for any conformal iterated function system there exists a unique non-empty compact set $X\subset {\ensuremath{\mathbb{R}}}^d$ such that $$X=\bigcup_{i\in {\mathcal{D}}}\phi_{i}(X).$$ We call the set $X$ the self-conformal set of $\Phi$. The simplest self-conformal sets are those that arise from similarities. We say that a contraction $\phi:Y\to Y$ is a similarity if $\phi=r\cdot O+t$ for some $r\in(0,1)$, some $d\times d$ orthogonal matrix $O$, and some $t\in {\ensuremath{\mathbb{R}}}^d$. Clearly any similarity is a conformal mapping. When $\Phi$ consists solely of similarities we say that $X$ is a self-similar set.
In what follows, if $I=(i_1,\ldots,i_n)$ then we let $\phi_{I}=\phi_{i_1}\circ \cdots \circ \phi_{i_n},$ $X_{I}=\phi_{I}(X),$ and $|I|$ will denote the length of $I$. We may refer to the set $X_{I}$ as a cylinder or a cylinder set. For a word $I\in \cup_n{\mathcal{D}}^n$ we let $I^{\infty}$ denote the element of ${\mathcal{D}}^{\mathbb{N}}$ obtained by repeating $I$ indefinitely. Similarly for $k\geq 1$ we let $I^k$ denote the word $I$ repeated $k$ times. When $\Phi$ consists of similarities then we let $O_{I}=O_{i_1}\circ \cdots\circ O_{i_n}$ and $r_{I}=\prod_{m=1}^n r_{i_m}.$
We say that a conformal iterated function system $\Phi$ satisfies the open set condition if there exists an open set $O\subset \mathbb{R}^d$ such that $\phi_{i}(O)\subseteq O$ for all $i\in {\mathcal{D}}$ and $\phi_{i}(O)\cap \phi_{j}(O)=\emptyset$ for $i\neq j$. Under the open set condition, the Hausdorff dimension of the self-conformal set $X$ is the unique solution to $P(s)=0$ where $$P(s):=\lim_{n\to\infty}\frac{1}{n}\log \inf_{x\in X} \sum_{I\in {\mathcal{D}}^{n}}|\phi_{I}'(x)|^{s}=\lim_{n\to\infty}\frac{1}{n}\log \sup_{x\in X} \sum_{I\in {\mathcal{D}}^{n}}|\phi_{I}'(x)|^{s}.$$ When $\Phi$ consists of similarities then the unique solution to $P(s)=0$ can be expressed more succinctly as the unique solution to $$\label{similarity dimension}
\sum_{i\in {\mathcal{D}}}r_i^s=1.$$ In what follows we denote the Hausdorff dimension of a self-conformal set satisfying the open set condition by $\gamma$, i.e. $\gamma$ is the unique solution to $P(s)=0$. The conformal iterated system we are referring to should be clear from the context. In addition to $\gamma$ being given by the zero of some appropriate pressure equation, under the open set condition it is known that $\mathcal{H}^{\gamma}(X)$ is positive and finite. Moreover under the open set condition $X$ is Ahlfors regular, this means that there exists $C>1$ such that $$\label{Ahlfors regular}\frac{r^{\gamma}}{C}\leq\mathcal{H}^{\gamma}(X\cap B(x,r))\leq Cr^{\gamma}$$ for all $x\in X$ and $0<r<Diam(X)$. These results are well known and date back to the work of Ruelle [@Rue]. For a proof see [@Fal].
One can encode elements of a self-conformal set using sequences in ${\mathcal{D}}^{{\ensuremath{\mathbb{N}}}}$ as follows. Let $\pi:{\mathcal{D}}^{{\ensuremath{\mathbb{N}}}}\to X$ be given by $$\pi((i_m))=\lim_{n\to\infty}(\phi_{i_1}\circ\cdots \circ \phi_{i_n})(0).$$ The map $\pi$ is surjective. Moreover, equipping ${\mathcal{D}}^{{\ensuremath{\mathbb{N}}}}$ with the product topology it can be shown that $\pi$ is continuous. For any $x\in X$, we call any sequence $(i_m)\in {\mathcal{D}}^{\mathbb{N}}$ such that $\pi((i_m))=x$ a coding of $x$. Without any separation hypothesis on the IFS, it is possible that a typical $x\in X$ will have multiple, possibly infinitely many, distinct codings. However, assuming the open set condition, $\mathcal{H}^{\gamma}$-almost every $x\in X$ has a unique coding. This follows from Theorem 3.7. from [@Kae]. With this fact in mind we now define our map $T:X\to X$ induced by the left shift on ${\mathcal{D}}^{\mathbb{N}}$. Let $Tx=\pi((i_{m+1}))$ where $(i_m)$ is an arbitrary choice of coding for $x$. Since $\mathcal{H}^{\gamma}$-almost every $x\in X$ has a unique coding, it follows from the definition of $T$ that $T^{n}x=\pi((i_{m+n}))$ for $\mathcal{H}^{\gamma}$-almost every $x\in X$ for any $n\in {\ensuremath{\mathbb{N}}}$. We will only be interested in statements which hold for $\mathcal{H}^{\gamma}$-almost every $x\in X.$ As such we can effectively ignore those points with multiple codings and assume that $T$ maps $x$ to the point whose coding is the unique coding of $x$ with the first digit removed.
Recalling the definition of $R(T,\varphi)$ from our introduction, and taking $d$ to be the usual Euclidean metric, we may now state our main result.
\[Main theorem\] Let $\Phi$ be a conformal iterated function system satisfying the open set condition and $\varphi:\mathbb{N}\to (0,\infty)$. Then the following statements are true:
1. If $\Phi$ consists of similarities and $\sum_{n=1}^{\infty}\varphi(n)^{\gamma}<\infty$ then $\mathcal{H}^{\gamma}(R(T,\varphi))=0$.
2. If $\sum_{n=1}^{\infty}\varphi(n)^{\gamma}=\infty$ then $\mathcal{H}^{\gamma}(R(T,\varphi))=\mathcal{H}^{\gamma}(X)$.
Theorem \[Main theorem\] was proved in [@ChangWuWu] by Chang et al for homogeneous self-similar sets in ${\ensuremath{\mathbb{R}}}$ satisfying the strong separation condition (i.e. $\phi_{i}(X)\cap \phi_{j}(X)=\emptyset$ for all $i\neq j$). As such Theorem \[Main theorem\] significantly improves upon [@ChangWuWu] as it allows for a more general class of iterated function systems and has weaker separation hypothesis.\
**Notation.** Given two positive real valued functions $f$ and $g$ defined on some set $S,$ we write $f\preceq g$ if there exists a positive constant $C$ such that $f(x)\leq C g(x)$ for all $x\in S$. Similarly we write $f\succeq g$ if $g\preceq f$. We write $f\asymp g$ if $f\preceq g$ and $f\succeq g.$
Proof of Theorem \[Main theorem\]
=================================
Proof of Theorem \[Main theorem\].1. (Convergence part)
-------------------------------------------------------
Throughout this section we assume that $\Phi$ consists of similarities. The proof of the convergence part of Theorem \[Main theorem\] will be a consequence of the following lemma.
\[covering lemma\] For $n$ sufficiently large, for each $I\in {\mathcal{D}}^n$ there exists a point $x_{I}\in {\ensuremath{\mathbb{R}}}^d$ such that $$\{x\in X:|T^nx-x|<\varphi(n)\}\subseteq \bigcup_{I\in {\mathcal{D}}^n}B(x_{I},2r_{I}\varphi(n)).$$
Let $x\in X$ and suppose that $I\in {\mathcal{D}}^n$ is such that $x=\phi_{I}(T^nx).$ Then we have $$\label{Step1}
x=r_{I}O_{I}T^nx + s_{I},$$ where $s_{I}\in {\ensuremath{\mathbb{R}}}^d$ depends solely upon $I.$ Rearranging we have $$\label{Step2}
T^nx=O_{I}^{-1}\left(\frac{x-s_{I}}{r_{I}}\right).$$If we assume $x$ satisfies $|T^nx-x|<\varphi(n),$ then implies $$\label{Step3}
\left|(O_I^{-1}-r_{I}\mathcal{I}_d)x-O_{I}^{-1}s_{I}\right|<r_{I}\varphi(n).$$ Here $\mathcal{I}_d$ is the $d\times d$ identity matrix. For $n$ sufficiently large the linear map $O_{I}^{-1}-r_{I}\mathcal{I}_d$ is invertible for any $I\in {\mathcal{D}}^n$. Equation therefore tells us that for $n$ sufficiently large $$\label{half job}
\{x\in X:|T^nx-x|<\varphi(n)\}\subseteq \bigcup_{I\in {\mathcal{D}}^n}(O_{I}^{-1}-r_{I}\mathcal{I}_d)^{-1}(B(O_{I}^{-1}s_{I},r_{I}\varphi(n))).$$ The image of a unit ball under the linear map $(O_{I}^{-1}-r_{I}\mathcal{I}_d)^{-1}$ is an ellipse whose semi-axes have lengths equal to the singular values of $(O_{I}^{-1}-r_{I}\mathcal{I}_d)^{-1}.$ Importantly the singular values of a matrix when interpreted as a function from ${\ensuremath{\mathbb{R}}}^{d^2}\to {\ensuremath{\mathbb{R}}}^d$ is continuous. Moreover, as the group of $d\times d$ orthogonal matrices is compact, the singular value function restricted to some small neighbourhood of the set of $d\times d$ orthogonal matrices is uniformly continuous. Since the singular values of an orthogonal matrix are all equal to $1$, it follows from uniform continuity that for $n$ sufficiently large we have $$(O_{I}^{-1}-r_{I}\mathcal{I}_d)^{-1}(B(O_{I}^{-1}s_{I},r_{I}\varphi(n))\subseteq B(x_{I},2r_{I}\varphi(n)),$$ for any $I\in {\mathcal{D}}^n$ where $x_{I}:=(O_{I}^{-1}-r_{I}\mathcal{I}_d)^{-1}(O_{I}^{-1}s_{I})$. Combining this fact with we have shown that $$\{x\in X:|T^nx-x|<\varphi(n)\}\subseteq \bigcup_{I\in {\mathcal{D}}^n}B(x_{I},2r_{I}\varphi(n))$$ for $n$ sufficiently large.
Assume $\varphi$ is such that $\sum_{n=1}^{\infty}\varphi(n)^{\gamma}<\infty.$ By Lemma \[covering lemma\] and the definition of Hausdorff measure (see [@Fal2]) we have the following: $$\begin{aligned}
{\mathcal{H}}^{\gamma}(R(T,\varphi))&\leq \liminf_{N\to\infty}\sum_{n=N}^{\infty}\sum_{I\in{\mathcal{D}}^n}Diam(B(x_I, 2r_{I}\varphi(n))^{\gamma}\\
&= \liminf_{N\to\infty}\sum_{n=N}^{\infty}\sum_{I\in{\mathcal{D}}^n}(4r_{I}\varphi(n))^\gamma\\
&=\liminf_{N\to\infty}4^{\gamma}\sum_{n=N}^{\infty}\varphi(n)^\gamma\sum_{I\in{\mathcal{D}}^n}r_{I}^{\gamma}\\
&\stackrel{\eqref{similarity dimension}}{=}\liminf_{N\to\infty}4^{\gamma}\sum_{n=N}^{\infty}\varphi(n)^\gamma\\
&=0.\end{aligned}$$ In the last line we used our assumption $\sum_{n=1}^{\infty}\varphi(n)^\gamma<\infty.$
Proof of Theorem \[Main theorem\].2. (Divergence part)
------------------------------------------------------
Our proof of the divergence part of Theorem \[Main theorem\] is based upon the proof of Theorem 1.4. from [@Bak]. Before diving into our proof we state some properties of self-conformal sets.
Suppose $\Phi$ is a conformal iterated function system satisfying the open set condition. For any $I\in \cup_n {\mathcal{D}}^n$ we let $$\widetilde{X_{I}}:=\{x\in X_I: x \textrm{ has a unique coding }\}.$$ Since $\mathcal{H}^{\gamma}$-almost every $x\in X$ has a unique coding, we have $$\label{Samemeasure}
\mathcal{H}^{\gamma}(X_I)=\mathcal{H}^{\gamma}(\widetilde{X}_I)$$for any $I\in \cup_n{\mathcal{D}}^n$. Let $\mu:=\mathcal{H}^{\gamma}|_{X}$ be the $\gamma$-dimensional Hausdorff measure restricted to $X.$ The properties stated below are well known for cylinders without the unique coding restriction. These properties still hold for the sets $\widetilde{X_{I}}$ because of .
We have the following:
- For any $n\in\mathbb{N}$ and $I,J\in {\mathcal{D}}^{n}$ such that $I\neq J,$ we have $$\label{measure zero intersection}
\mu(\widetilde{X}_{I}\cap \widetilde{X}_{J})=0.$$
- For any $I,J\in \cup_n{\mathcal{D}}^n$ $$\label{Weak Bernoulli measure}
\mu(\widetilde{X}_{IJ})\asymp \mu(\widetilde{X}_{I})\mu(\widetilde{X}_{J}).$$
- For any $I\in \cup_n{\mathcal{D}}^n$ $$\label{Measure and diameter}
\mu(\widetilde{X}_{I})\asymp Diam(X_{I})^{\gamma}.$$
- There exists $\kappa\in(0,1)$ such that for any $I\in \cup_n {\mathcal{D}}^n$ $$\label{measure decay}
\mu(\widetilde{X}_{I})\preceq \kappa^{|I|}.$$
- Let $x\in X$ and $(i_m)\in {\mathcal{D}}^{{\ensuremath{\mathbb{N}}}}$ be a coding of $x$. For any $0<r<Diam(X)$ there exists $N\in\mathbb{N}$ such that $$\label{cylinder approx}
X_{i_{1},\ldots,i_{N}}\subset B(x,r) \textrm{ and } Diam(X_{i_{1},\ldots,i_{N}})\asymp r.$$
- For any $n\in {\ensuremath{\mathbb{N}}}$ we have $$\label{Sum prefixes}
\sum_{I\in {\mathcal{D}}^n}\mu(\widetilde{X}_I)=\mu(X).$$ Similarly, for any $J\in \cup_{n}{\mathcal{D}}^n$ and $n\geq |J|$ we have $$\label{Sum prefixes2}\sum_{\stackrel{I\in {\mathcal{D}}^n}{J\textrm{ is a prefix of }I}}\mu(\widetilde{X}_{I})=\mu(\widetilde{X}_J).$$
In the above we have denoted the concatenation of two words $I$ and $J$ by $IJ$. Property follows from Theorem 3.7. from [@Kae]. For a proof of the remaining properties see [@Fal] and [@Rue]. Properties , , and are essentially a consequence of the fact that $\mu$ is equivalent to the pushforward of a suitably defined Gibbs measure for a particular Hölder continuous potential. The proof of is standard. Properties and are a consequence of and the fact $X=\cup_{I\in {\mathcal{D}}^n}\phi_{I}(X)$ for any $n\in {\ensuremath{\mathbb{N}}}$.
We will also need the following two lemmas.
\[Density lemma\] Let $X$ be a compact set in ${\ensuremath{\mathbb{R}}}^{d}$ and let $\mu$ be a finite doubling measure on X such that any open set is $\mu$ measurable. Let $E$ be a Borel subset of $X$. Assume that there are constants $r_0,c > 0$ such that for any ball $B$ with radius less than $r_0$ and centre in $X$ we have $$\mu(E \cap B) > c \mu(B).$$ Then $\mu(X \setminus E) = 0.$
For a proof of Lemma \[Density lemma\] see [@BDV §8]. Note that a measure $\mu$ supported on a compact set $X$ is doubling if there exists a constant $C>1$ such that for any $x\in X$ and $r>0$ we have $$\mu(B(x,2r))\leq C \mu(B(x,r)).$$ Since $X$ is Ahlfors regular it follows from that $\mu$ is automatically a doubling measure.
\[Erdos lemma\] Let $X$ be a compact set in ${\ensuremath{\mathbb{R}}}^{d}$ and let $\mu$ be a finite measure on $X$. Also, let $E_n$ be a sequence of $\mu$-measurable sets such that $\sum_{n=1}^{\infty}\mu(E_n)=\infty.$ Then $$\mu(\limsup_{n\to\infty} E_{n})\geq \limsup_{Q\to\infty}\frac{(\sum_{n=1}^{Q}\mu(E_{n}))^{2}}{\sum_{n,m=1}^{Q}\mu(E_{n}\cap E_m)}.$$
For a proof of Lemma \[Erdos lemma\] see [@Spr Lemma 5].
We may now proceed with our proof of the divergence part of Theorem \[Main theorem\]. Let $I=(i_1,\ldots,i_n)\in{\mathcal{D}}^n$ and consider the ball $B(\pi(I^{\infty}),\varphi(n)/2)$. Applying we know that there exists $k_{I}\geq 0$ and $1\leq s_{I}\leq n-1$ such that $$X_{I^{k_I}(i_1,\ldots,i_{s_I})}\subseteq B(\pi(I^{\infty}),\varphi(n)/2)$$ and $$\label{inclusion}
Diam(X_{I^{k_I}(i_1,\ldots,i_{s_I})})\asymp \frac{\varphi(n)}{2}.$$ Now consider the set $\widetilde{X}_{I^{k_I+1}(i_1,\ldots,i_{s_I})}.$ For any $x\in \widetilde{X}_{I^{k_I+1}(i_1,\ldots,i_{s_I})}$ we have $T^nx\in \widetilde{X}_{I^{k_I}(i_1,\ldots,i_{s_I})}.$ Moreover, since $$\widetilde{X}_{I^{k_I+1}(i_1,\ldots,i_{s_I})}\subseteq X_{I^{k_I}(i_1,\ldots,i_{s_I})}\subseteq B(\pi(I^{\infty}),\varphi(n)/2),$$ we may conclude by the triangle inequality that if $x\in \widetilde{X}_{I^{k_I+1}(i_1,\ldots,i_{s_I})}$ then $|T^nx-x|<\varphi(n)$. So if we let $$E_n'=\bigcup_{I\in {\mathcal{D}}^n}\widetilde{X}_{I^{k_I+1}(i_1,\ldots,i_{s_I})}$$ then $$\limsup_{n\to\infty} E_n'\subseteq R(T,\varphi).$$ To prove the divergence part of Theorem \[Main theorem\] it suffices to show that $\mu(\limsup_{n\to\infty} E_n')=\mu(X).$ To do this we will apply Lemma \[Density lemma\]. As such let us fix an arbitrary ball $B$ with centre in $X$ and radius less then $Diam(X)$. Applying we know that there exists $J\in \cup_n D^n$ such that $X_{J}\subseteq B$ and $Diam(X_J)\asymp Radius(B)$. By and we know that $\mu(X_J)\asymp \mu(B).$ Therefore to prove the divergence part of Theorem \[Main theorem\], instead of proving that there exists $c>0$ such that $\mu(\limsup_{n\to\infty} E_n' \cap B)>c\mu(B)$, it suffices to show that there exists $c>0$ such that $\mu(\limsup_{n\to\infty} E_n' \cap X_J)>c\mu(X_J).$
For $n\geq |J|$ if we let $$E_n=\bigcup_{\stackrel{I\in {\mathcal{D}}^n}{J\textrm{ is a prefix of }I}}\widetilde{X}_{I^{k_I+1}(i_1,\ldots,i_{s_I})}$$ then $\limsup_{n\to\infty} E_n\subset \limsup_{n\to\infty} E_n'\cap X_J$. Therefore to prove the divergence part of Theorem \[Main theorem\] it suffices to show that there exists $c>0$ such that $$\label{Sufficestoshow}
\limsup_{n\to\infty} E_n>c\mu(X_J).$$ To do this we will use Lemma \[Erdos lemma\]. To use this lemma we first have to check $\sum_{n=|J|}^{\infty}\mu(E_n)=\infty.$
\[Divergence lemma\] We have $\sum_{n=|J|}^{Q}\mu(E_n)\asymp \mu(X_J)\sum_{n=|J|}^{\infty}\varphi(n)^{\gamma}.$
The following holds $$\begin{aligned}
\sum_{n=|J|}^{Q}\mu(E_n)&=\sum_{n=|J|}^{Q}\mu\left(\bigcup_{\stackrel{I\in {\mathcal{D}}^n}{J\textrm{ is a prefix of }I}}\widetilde{X}_{I^{k_I+1}(i_1,\ldots,i_{s_I})}\right)\\
&\stackrel{\eqref{measure zero intersection}}{=}\sum_{n=|J|}^{Q}\sum_{\stackrel{I\in {\mathcal{D}}^n}{J\textrm{ is a prefix of }I}}\mu(\widetilde{X}_{I^{k_I+1}(i_1,\ldots,i_{s_I})})\\
&\stackrel{\eqref{Weak Bernoulli measure}}{\asymp}\sum_{n=|J|}^{Q}\sum_{\stackrel{I\in {\mathcal{D}}^n}{J\textrm{ is a prefix of }I}}\mu(\widetilde{X}_I)\mu(\widetilde{X}_{I^{k_I}(i_1,\ldots,i_{s_I})})\\
&\stackrel{\eqref{Measure and diameter},\eqref{inclusion}}{\asymp}\sum_{n=|J|}^{Q}\varphi(n)^{\gamma}\sum_{\stackrel{I\in {\mathcal{D}}^n}{J\textrm{ is a prefix of }I}}\mu(\widetilde{X}_I)\\
&\stackrel{\eqref{Sum prefixes2}}{=} \mu(\widetilde{X}_J)\sum_{n=|J|}^{Q}\varphi(n)^{\gamma}\\
&\stackrel{\eqref{Samemeasure}}{=} \mu(X_J)\sum_{n=|J|}^{Q}\varphi(n)^{\gamma}.\end{aligned}$$
Lemma \[Divergence lemma\] shows that when $\sum_{n=1}^{\infty}\varphi(n)^{\gamma}=\infty$ the sequence $(E_n)$ satisfies the hypothesis of Lemma \[Erdos lemma\]. It will also be used in some of our later calculations.
\[Near independence\] Let $I\in {\mathcal{D}}^n$ be such that $J$ is a prefix of $I.$ Then for $m>n$ we have $$\begin{aligned}
\mu(\widetilde{X}_{I^{k_{I}+1}(i_1,\ldots,i_{s_I})}\cap E_m)&\preceq \mu(\widetilde{X}_{J})\mu(\widetilde{X}_{i_{|J|+1},\ldots,i_n})\kappa^{m-n}\varphi(m)^{\gamma}\\
&+\mu(\widetilde{X}_{J})\mu(\widetilde{X}_{i_{|J|+1},\ldots,i_n}))\varphi(m)^{\gamma}\varphi(n)^{\gamma}.
\end{aligned}$$
Fix $I\in D^n$ such that $J$ is a prefix of $I$ and let $m>n$. It is useful to consider two separate cases. We first consider the case where $m\leq (k_{I}+1)n+s_{I}.$
If $m\leq (k_{I}+1)n+s_{I}$ then there exists at most one $\tilde{I}\in {\mathcal{D}}^m$ such that $$\mu(\widetilde{X}_{I^{k_I+1}(i_1,\ldots,i_{s_I})}\cap \widetilde{X}_{\tilde{I}^{k_{\tilde{I}}+1}(\tilde{i}_1,\ldots,\tilde{i}_{s_{\tilde{I}}})})>0.$$ Therefore $$\begin{aligned}
\mu(\widetilde{X}_{I^{k_I+1}(i_1,\ldots,i_{s_I})}\cap E_m)&=\mu(\widetilde{X}_{I^{k_I+1}(i_1,\ldots,i_{s_I})}\cap \widetilde{X}_{\tilde{I}^{k_{\tilde{I}}+1}(\tilde{i}_1,\ldots,\tilde{i}_{s_{\tilde{I}}})})\\
&\leq \mu( \widetilde{X}_{\tilde{I}^{k_{\tilde{I}}+1}(\tilde{i}_1,\ldots,\tilde{i}_{s_{\tilde{I}}})})\\
&\stackrel{\eqref{Weak Bernoulli measure}}{\asymp} \mu(\widetilde{X}_{\tilde{I}})\mu( \widetilde{X}_{\tilde{I}^{k_{\tilde{I}}}(\tilde{i}_1,\ldots,\tilde{i}_{s_{\tilde{I}}})})\\
&\stackrel{\eqref{Weak Bernoulli measure}}{\asymp} \mu(\widetilde{X}_{J})\mu(\widetilde{X}_{\tilde{i}_{|J|+1},\ldots,\tilde{i}_m})\mu( \widetilde{X}_{\tilde{I}^{k_{\tilde{I}}}(\tilde{i}_1,\ldots,\tilde{i}_{s_{\tilde{I}}})})\\
&\stackrel{\eqref{Weak Bernoulli measure}}{\asymp}\mu(\widetilde{X}_{J})\mu(\widetilde{X}_{i_{|J|+1},\ldots,i_n})\mu(\widetilde{X}_{\tilde{i}_{n+1},\ldots,\tilde{i}_m})\mu( \widetilde{X}_{\tilde{I}^{k_{\tilde{I}}}(\tilde{i}_1,\ldots,\tilde{i}_{s_{\tilde{I}}})})\\
&\stackrel{\eqref{measure decay}}{\preceq}\mu(\widetilde{X}_{J})\mu(\widetilde{X}_{i_{|J|+1},\ldots,i_n})\kappa^{m-n}\mu( \widetilde{X}_{\tilde{I}^{k_{\tilde{I}}}(\tilde{i}_1,\ldots,\tilde{i}_{s_{\tilde{I}}})})\\
&\stackrel{\eqref{Measure and diameter},\eqref{inclusion}}{\asymp} \mu(\widetilde{X}_{J})\mu(\widetilde{X}_{i_{|J|+1},\ldots,i_n})\kappa^{m-n}\varphi(m)^{\gamma}.\end{aligned}$$ Summarising the above, we have shown that if $n<m\leq (k_{I}+1)n+s_{I}$ then $$\label{Bound1}
\mu(\widetilde{X}_{I^{k_I+1}(i_1,\ldots,i_{s_I})}\cap E_m)\preceq \mu(\widetilde{X}_{J})\mu(\widetilde{X}_{i_{|J|+1},\ldots,i_n})\kappa^{m-n}\varphi(m)^{\gamma}.$$ Now suppose $m>(k_I+1)n+s_I.$ Then $$\begin{aligned}
\mu(\widetilde{X}_{I^{k_I+1}(i_1,\ldots,i_{s_I})}\cap E_m)&=\sum_{\stackrel{\tilde{I}\in {\mathcal{D}}^m}{I^{k_I+1}(i_1,\ldots,i_{s_{I}})\textrm{ is a prefix of }\tilde{I}}}\mu(\widetilde{X}_{\tilde{I}^{k+1}(\tilde{i}_1,\ldots,\tilde{i}_{s_{\tilde{I}}})})\\
&\stackrel{\eqref{Weak Bernoulli measure}}{\asymp} \sum_{\stackrel{\tilde{I}\in {\mathcal{D}}^m}{I^{k_I+1}(i_1,\ldots,i_{s_{I}})\textrm{ is a prefix of }\tilde{I}}}\mu(\widetilde{X}_{\tilde{I}})\mu(\widetilde{X}_{\tilde{I}^{k}(\tilde{i}_1,\ldots,\tilde{i}_{s_{\tilde{I}}})})\\
&\stackrel{\eqref{Measure and diameter},\eqref{inclusion}}{\asymp} \sum_{\stackrel{\tilde{I}\in {\mathcal{D}}^m}{I^{k_I+1}(i_1,\ldots,i_{s_{I}})\textrm{ is a prefix of }\tilde{I}}}\mu(\widetilde{X}_{\tilde{I}})\varphi(m)^{\gamma}\\
&\stackrel{\eqref{Sum prefixes2}}{=}\mu(\widetilde{X}_{I^{k_I+1}(i_1,\ldots,i_{s_{I}})})\varphi(m)^{\gamma}\\
&\stackrel{\eqref{Weak Bernoulli measure}}{\asymp} \mu(\widetilde{X}_{I})\mu(\widetilde{X}_{I^{k_I}(i_1,\ldots,i_{s_{I}})})\varphi(m)^{\gamma}\\
&\stackrel{\eqref{Weak Bernoulli measure}}{\asymp} \mu(\widetilde{X}_{J})\mu(\widetilde{X}_{i_{|J|+1},\ldots,i_n})\mu(\widetilde{X}_{I^{k_I}(i_1,\ldots,i_{s_{I}})})\varphi(m)^{\gamma}\\
&\stackrel{\eqref{Measure and diameter},\eqref{inclusion}}{\asymp} \mu(\widetilde{X}_{J})\mu(\widetilde{X}_{i_{|J|+1},\ldots,i_n})\varphi(n)^{\gamma}\varphi(m)^{\gamma}.\end{aligned}$$Summarising the above, we have shown that if $m>(k_{I}+1)n+s_{I}$ then $$\label{Bound2}
\mu(\widetilde{X}_{I^{k_I+1}(i_1,\ldots,i_{s_I})}\cap E_m)\asymp \mu(\widetilde{X}_{J})\mu(\widetilde{X}_{i_{|J|+1},\ldots,i_n})\varphi(m)^{\gamma}\varphi(n)^{\gamma}.$$Adding together the bounds provided by and , we see that for any $m>n$ we have $$\begin{aligned}
\mu(\widetilde{X}_{I^{k_{I}+1}(i_1,\ldots,i_{s_I})}\cap E_m)&\preceq \mu(\widetilde{X}_{J})\mu(\widetilde{X}_{i_{|J|+1},\ldots,i_n})\kappa^{m-n}\varphi(m)^{\gamma}\\
&+\mu(\widetilde{X}_{J})\mu(\widetilde{X}_{i_{|J|+1},\ldots,i_n}))\varphi(m)^{\gamma}\varphi(n)^{\gamma}.\end{aligned}$$This completes our proof.
\[Quasi independence prop\] $$\sum_{n,m=|J|}^{Q}\mu(E_n\cap E_m)\preceq \mu(X_J)\left(\sum_{n=|J|}^{Q}\varphi(n)^{\gamma}+\left(\sum_{n=|J|}^{Q}\varphi(n)^{\gamma}\right)^2\right).$$
We have the following $$\begin{aligned}
\sum_{n,m=|J|}^{Q}\mu(E_n\cap E_m)&=\sum_{n=|J|}^{Q}\mu(E_n)+2\sum_{n=|J|}^{Q-1}\sum_{m=n+1}^{Q}\mu(E_n\cap E_m)\\
&=\sum_{n=|J|}^{Q}\mu(E_n)+2\sum_{n=|J|}^{Q-1}\sum_{\stackrel{I\in {\mathcal{D}}^n}{J\textrm{ is a prefix of }I}}\sum_{m=n+1}^{Q}\mu(\widetilde{X}_{I^{k_{I}+1}(i_1,\ldots,i_{s_{I}})}\cap E_m).\end{aligned}$$ Applying Lemma to the above we obtain $$\begin{aligned}
\label{Triple split}
\sum_{n,m=|J|}^{Q}\mu(E_n\cap E_m)&\preceq\sum_{n=|J|}^{Q}\mu(E_n)+\sum_{n=|J|}^{Q-1}\sum_{\stackrel{I\in {\mathcal{D}}^n}{J\textrm{ is a prefix of }I}}\sum_{m=n+1}^{Q}\mu(\widetilde{X}_{J})\mu(\widetilde{X}_{i_{|J|+1},\ldots,i_n})\kappa^{m-n}\varphi(m)^{\gamma}\\
&+\sum_{n=|J|}^{Q-1}\sum_{\stackrel{I\in {\mathcal{D}}^n}{J\textrm{ is a prefix of }I}}\sum_{m=n+1}^{Q}\mu(\widetilde{X}_{J})\mu(\widetilde{X}_{i_{|J|+1},\ldots,i_n}))\varphi(m)^{\gamma}\varphi(n)^{\gamma}\nonumber.\end{aligned}$$We focus on the three terms on the right hand side of individually. By Lemma \[Divergence lemma\] we know that the following holds for the first term $$\label{BOUND1}
\sum_{n=|J|}^{Q}\mu(E_n)\asymp \mu(X_{J})\sum_{n=|J|}^{Q}\varphi(n)^{\gamma}.$$Now let us focus on the second term in . We have $$\begin{aligned}
\label{BOUND2}
&\sum_{n=|J|}^{Q-1}\sum_{\stackrel{I\in {\mathcal{D}}^n}{J\textrm{ is a prefix of }I}}\sum_{m=n+1}^{Q}\mu(\widetilde{X}_{J})\mu(\widetilde{X}_{i_{|J|+1},\ldots,i_n})\kappa^{m-n}\varphi(m)^{\gamma}\nonumber\\
= &\mu(\widetilde{X}_{J})\sum_{n=|J|}^{Q-1}\sum_{\stackrel{I\in {\mathcal{D}}^n}{J\textrm{ is a prefix of }I}}\mu(\widetilde{X}_{i_{|J|+1},\ldots,i_n})\sum_{m=n+1}^{Q}\kappa^{m-n}\varphi(m)^{\gamma}\nonumber\\
= &\mu(\widetilde{X}_{J})\sum_{n=|J|}^{Q-1}\sum_{I'\in {\mathcal{D}}^{n-|J|}}\mu(\widetilde{X}_{I'})\sum_{m=n+1}^{Q}\kappa^{m-n}\varphi(m)^{\gamma}\nonumber\\
\stackrel{\eqref{Sum prefixes}}{\preceq}&\mu(\widetilde{X}_{J})\sum_{n=|J|}^{Q-1}\sum_{m=n+1}^{Q}\kappa^{m-n}\varphi(m)^{\gamma}\nonumber\\
=&\mu(\widetilde{X}_{J})\sum_{m=|J|+1}^Q\sum_{n=|J|}^{m-1}\kappa^{m-n}\varphi(m)^{\gamma}\nonumber\\
\stackrel{(\kappa<1)}{\preceq}&\mu(\widetilde{X}_{J})\sum_{m=|J|+1}^Q\varphi(m)^{\gamma}\nonumber\\
\stackrel{\eqref{Samemeasure}}{=}&\mu(X_{J})\sum_{m=|J|+1}^Q\varphi(m)^{\gamma}.
$$It remains to bound the third term: $$\begin{aligned}
\label{BOUND3}
&\sum_{n=|J|}^{Q-1}\sum_{\stackrel{I\in {\mathcal{D}}^n}{J\textrm{ is a prefix of }I}}\sum_{m=n+1}^{Q}\mu(\widetilde{X}_{J})\mu(\widetilde{X}_{i_{|J|+1},\ldots,i_n}))\varphi(m)^{\gamma}\varphi(n)^{\gamma}\nonumber\\
=&\mu(\widetilde{X}_J)\sum_{n=|J|}^{Q-1}\varphi(n)^{\gamma}\sum_{\stackrel{I\in {\mathcal{D}}^n}{J\textrm{ is a prefix of }I}}\mu(\widetilde{X}_{i_{|J|+1},\ldots,i_n}))\sum_{m=n+1}^{Q}\varphi(m)^{\gamma}\nonumber\\
=&\mu(\widetilde{X}_J)\sum_{n=|J|}^{Q-1}\varphi(n)^{\gamma}\sum_{I'\in {\mathcal{D}}^{n-|J|}}\mu(\widetilde{X}_{I'})\sum_{m=n+1}^{Q}\varphi(m)^{\gamma}\nonumber\\
\stackrel{\eqref{Sum prefixes}}{\preceq}&\mu(\widetilde{X}_J)\sum_{n=|J|}^{Q-1}\varphi(n)^{\gamma}\sum_{m=n+1}^{Q}\varphi(m)^{\gamma}\nonumber\\
\stackrel{\eqref{Samemeasure}}{=}&\mu(X_J)\sum_{n=|J|}^{Q-1}\varphi(n)^{\gamma}\sum_{m=n+1}^{Q}\varphi(m)^{\gamma}\nonumber\\
\preceq&\mu(X_J)\left(\sum_{n=|J|}^{Q}\varphi(n)^{\gamma}\right)^2.\end{aligned}$$ Collecting the bounds provided by , , and , and substituting them into completes our proof.
With Proposition \[Quasi independence prop\] we can now complete our proof of Theorem \[Main theorem\].
By to prove the divergence part of Theorem \[Main theorem\] it suffices to show that there exists $c>0$ such that $$\label{Proofsufficestoshow}
\mu(\limsup_{n\to\infty} E_n)>c\mu(X_J).$$ By Lemma \[Divergence lemma\] we know that $\sum_{n=|J|}^{\infty}\mu(E_n)=\infty$. Therefore Lemma \[Erdos lemma\] and Lemma \[Divergence lemma\] combined tell us that $$\label{nearlydone}\mu(\limsup_{n\to\infty} E_n)\geq \limsup_{Q\to\infty}\frac{\left(\sum_{n=|J|}^{Q}\mu(E_n)\right)^2}{\sum_{n,m=|J|}^{Q}\mu(E_n\cap E_m)}\succeq \limsup_{Q\to\infty}\frac{\mu(\widetilde{X}_J)^2\left(\sum_{n=|J|}^{Q}\varphi(n)^{\gamma}\right)^2}{\sum_{n,m=|J|}^{Q}\mu(E_n\cap E_m)}.$$Since $\sum_{n=1}^{\infty}\varphi(n)^{\gamma}=\infty$, we know that for any $Q$ sufficiently large we have $$\label{Square dominates}
\sum_{n=|J|}^{Q}\varphi(n)^{\gamma}<\left(\sum_{n=|J|}^{Q}\varphi(n)^{\gamma}\right)^2.$$ Combining this observation with and Proposition \[Quasi independence prop\], it follows that $$\mu(\limsup_{n\to\infty} E_n)\succeq \mu(X_J),$$ i.e. there exists $c>0$ such that $\mu(\limsup_{n\to\infty} E_n)>c\mu(X_J).$ So we have shown that holds. This completes our proof.
**Acknowledgements.** This second author was supported by the University of Warwick Undergraduate Research Support Scheme.
[1]{} S. Baker, *An analogue of Khintchine’s theorem for self-conformal sets,* Mathematical Proceedings of the Cambridge Philosophical Society, 1–31. doi:10.1017/S030500411800049X L. Barreira, B. Saussol, *Hausdorff dimension of measures via Poincaré recurrence,* Comm. Math. Phys. 219 (2001), no. 2, 443–-463. V. Beresnevich, D. Dickinson, S. Velani, *Measure theoretic laws for lim sup sets,* Mem. Amer. Math. Soc. 179 (2006), no. 846, x+91 pp. M. Boshernitzan *Quantitative recurrence results,* Invent. Math. 113 (1993), no. 3, 617–-631. Y. Chang, M. Wu, W. Wu, *Quantitative recurrence properties and homogeneous self-similar sets,* Proc. Amer. Math. Soc. 147 (2019), no. 4, 1453-–1465. N. Chernov, D. Kleinbock *Dynamical Borel-Cantelli lemmas for Gibbs measures,* Israel J. Math. 122 (2001), 1–27. K. Falconer, *Fractal geometry. Mathematical foundations and applications,* Third edition. John Wiley & Sons, Ltd., Chichester, 2014. xxx+368 pp. ISBN: 978-1-119-94239-9 K. Falconer, *Techniques in fractal geometry,* John Wiley & Sons, Ltd., Chichester, 1997. xviii+256 pp. ISBN: 0-471-95724-0 J. L. Fernández, M. V. Melián, and D. Pestana, *Quantitative mixing results and inner functions,* Math. Ann. 337 (2007), no. 1, 233-–251. R. Hill, S. Velani, *The ergodic theory of shrinking targets,* Invent. Math. 119 (1995), no. 1, 175-–198. J. Hutchinson, *Fractals and self-similarity*, Indiana Univ. Math. J. 30 (1981), no. 5, 713–-747. A. Käenmäki, *On natural invariant measures on generalised iterated function systems*, Ann. Acad. Sci. Fenn. Math. 29 (2004), no. 2, 419–458. B. Li, B.-W. Wang, J. Wu, J. Xu, *The shrinking target problem in the dynamical system of continued fractions,* Proc. Lond. Math. Soc. (3) 108 (2014), no. 1, 159–186. D. Ruelle, *Repellers for real analytic maps,* Ergodic Theory Dynamical Systems 2 (1982), no. 1, 99-–107. S. Seuret, B.-W. Wang, *Quantitative recurrence properties in conformal iterated function systems,* Adv. Math. 280 (2015), 472-–505. V. G. Sprindžuk, *A proof of Mahler’s conjecture on the measure of the set of S-numbers,* Izv. Akad. Nauk SSSR Ser. Mat. 29 1965 379–-436. B. Tan, B.-W. Wang, *Quantitative recurrence properties of beta dynamical systems,* Adv. Math. 228 (2011) 2071-–2097. B.-W. Wang, J. Wu, *A Survey on the dimension theory in dynamical Diophantine approximation,* Recent Developments in Fractals and Related Fields Conference on Fractals and Related Fields III, ile de Porquerolles, France, 2015, edited by Julien Barral and Stéphane Seuret.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'In this work animations of the random walk movement using a freeware Algodoo were done in order to support teaching the concepts of Brownian Motion. The random walk movement were simulate considering elastic collision between the particles in suspension in a fluid, and the particles which constitute the fluid. The intensity of velocities where defined in an arbitrary range, and we have a random distribution of the velocity directions. Using two methods, the distribution histogram of displacements (DHD) and the mean-square-displacement ${\langle{\Delta r^{2}}\rangle}$ (MSD), it was possible to measure the diffusion coefficient of the system, and determine the regions where the system presents ballistic regime or diffusive transport regime. The ballistic regime was observed graphically when the MSD has a parabolic dependence with time, which differing from the typical diffusive regime where MSD has a linear dependence. The didactical strategy for combining analytical approaches as graphic analysis, and animations in software’s with easy implementation supports the teaching and learning processes, especially in Physics were we want to explain experimental results within theoretical models.'
address: |
$^{a}$Instituto Federal do Espírito Santo, campus Vitória, 29040-780 Vitória, Brazil.\
$^{b}$Instituto Federal Fluminense, campus Bom Jesus do Itabapoana, 28360-000 Bom Jesus do Itabapoana, Brazil.\
$^{c}$Universidade Tecnológica Federal do Paraná, campus Curitiba, DAFIS, 80230-901. Curitiba, Brazil.\
$^{d}$Instituto Federal Fluminense, campus Campos Centro, 28030-130 Campos dos Goytacazes, Brazil.\
$^{*}[email protected]
author:
- |
Samir L. da Silva$^{a}$$^{, *}$; Judismar T. Guaitolini Junior$^{a}$;\
Rodrigo L. da Silva$^{b}$; Emilson. R. Viana$^{c}$ and Fábio F. Leal$^{d}$
title: An alternative for teaching and learning the simple diffusion process using Algodoo animations
---
Brownian motion; Random Walk; Diffusion Coefficient; Animation; Algodoo; Teaching of Physics.
Introduction {#sec1}
============
The diffusion is one of the most studied and spread processes in science. The Einstein’s description about the erratic motion of small particles on fluid surfaces, the historical Brownian Motion (BM) [@r01; @r02; @r03; @r04], lead to a proof that matter is constituted by atoms and molecules in constant motion in according to the kinetic-theory. Other numerous physical systems exhibit some diffusion process as the spreading of dengue by migration of infected individual or mosquitoes [@r05], the diffusion-limited aggregation applied to growth phenomena [@r06], the analysis of financial data on stock market [@r07], the model of protein folding [@r08], the fixational eye movements [@r09] and many more. Many of these stochastic processes can be studied using the classical random walk (RW).
It is not easy to obtain an intuitive comprehension of the stochastic phenomena, because this usually requires advanced mathematical tools. Actually, several teachers presenting a good pedagogical approach by using numerical experiments of low computational cost, and that are available to students with access to a computer. Also, many other numeric-experiments can be used to complement real experiences. The use of computational technologies is explored by many educators that propose strategies to elaborate good tools for supporting the cognitive development of the students. Such technologies have been widely used in the physics teaching [@r10; @r11; @r12]. To minimize the difficulties of teaching and learning physical phenomena, some teachers has added to your classes some softwares able to determine the evolution of the equations, in order to create simulations and animations of present phenomena, or even, to automatization of the experimental data acquisition, to modeling and interacting with virtual physical environments [@r13; @r14; @r15; @r16; @r17; @r18; @r19].
Therefore, in this paper, the simulation of the two-dimensional random walk movement of particles on fluid surface, in order to study the Brownian motion, was done by using computational animations. To accomplish this task, we use the free software for 2D simulation known as Algodoo of Algoryx Simulation AB$\textsuperscript{\textregistered}$[@r20]. This software has an interactive environment, allowing creation of experimental scenarios, as animated movies, but with having as feedback, the equations and physical properties imposed to the simulations. The Algodoo is an easy manipulation tool, and does not require specific knowledge about computer programming or training to realize tasks with the software, which allow easy learning by students. In a recent work of our research group [@r21] we present the potentiality of that software as a tool for teaching and learning physics by considering the launching projectiles animations.
We believe that the complementation of mathematical description of the Brownian motion by using the animations with Algodoo will provide to the students a better learning about all techniques involved in this stochastic model. With this project we can also create the basis for the theoretical instrumentation for studying others diffusive process when we use the Algodoo’s animations for the Brownian motion.
This paper is organized as follows; firstly, we will introduce some of the basic properties of random walks in two-dimension for an active particle with random direction velocity in a two-dimensional homogeneous environment. For convenience, we will use the familiar picture of one diffusing particle. Then the animations were built by using the Algodoo software. Next, from the animations, we will show how to calculate the diffusion coefficient by using two methods: the mean-square-displacement and the displacement histogram of the Brownian particle. And finally, we show that the random walk animations provide a clear understanding of transition between the ballistic behavior and the diffusive behavior of Brownian motion.
The Brownian motion and the procedure for building animations {#sec2}
=============================================================
The Brownian system is constituted by a suspended particle in a fluid with random motion resulting for their collision with the atoms and molecules of the fluid. Since, is not an easily task to simulate a fluid, because this simulation involve to determine the solutions of the Euler and Navier-Stokes equations. So, in our animations the fluid is formed by small blue disks moving randomly.
The manufacture of this fluid consist in to create a little set of identical disks and to attribute to them velocity with different magnitude and direction randomly. To increase the number of disks of the fluid the process are repeated several times until we get a desired concentration. The null friction and maximum elasticity in the collision between the disks is considered, to maintain constant the mean kinetic energy of system. These characteristics can be established selecting all disks and to edit in the item “*Material*" of the Algodoo software. A red disk can be introduced in any region of the system, and represents our “Brownian” particle. The particles of the fluid (red disks) has a diameter defined much bigger than the particles in suspension (blue disks) in order to enhance the contrast.
For the present animation was estabilished, by convinience, a flat rectangular region of ${575.0\;}$ $m^{2}$, where were uniformly distributed 7.2$\times$$10^3$ blue disks forming the system fluid. Each blue disk has a mass of ${25\;}$ g and occupies an area of 12.0$\times$$10^{-2}$${\;}$${m^2}$. Intensities of velocities were distributed between 0.1 m/s e 5.5 m/s with random directions. The Brownian particles, i.e., the red disk, has a mass of 70 g and area of 3.0$\times$$10^{-1}$${\;}$ ${m^2}$. These specification were choose arbitrary and do not represent any material or specific system. This software has a limitation to assign small dimensions and mass in the drawn objects. Our aim is to create fictitious animations of the random walk movement to study the Brownian motion properties.
![(Color online) Illustration of Brownian motion animation. The small discs represent the fluid and the larger disk is a Brownian particle that will perform the randomic motion because of collisions between the particles of the fluid.(See video submitted)[]{data-label="Figure1"}](fig1.eps){width="1.0\linewidth"}
In the fig. 1 (a) we present the illustration of the environment built to observe the Brownian motion. When we start the animations in the Algodoo environment, the simulate dynamics of the blue disks a random motion of the molecules in a fluid (fig. 1 (b)). To see trajectories of the red disk we use the tool “*Tracer*". This tool draws a line by all the trajectory of the selected object. The trajectory of the motion can be observed in the fig. 1 (b) by a path illustrated in a red line.
Theoretically, the motion in the *x* and *y* axes obeys the classical random walk with the two-dimensional mean displacement $\langle{\Delta r}\rangle$ null and the mean square displacement (MSD) $\langle{\Delta r^{2}}\rangle$ is given by:
$$\langle{\Delta r^{2}}\rangle = \langle{\Delta x^{2}}\rangle + \langle{\Delta y^{2}}\rangle = 4Dt
.\label{eq01}$$
where $\langle{\Delta x^{2}}\rangle$ and $\langle{\Delta y^{2}}\rangle$ are the one-dimensional MSD, *D* is the diffusion coefficient and *t* the time between the points of the data [@r01; @r02; @r03; @r04; @r22; @r23]. Thus, the MSD depends of the diffusion coefficient and time. The root mean square (RMS) displacement $\langle{\Delta r^{2}}\rangle$$_{rms}$= $\sqrt{\langle{\Delta r^{2}}\rangle}$ is proportional to $\sqrt{t}$, allowing that the red disk visits quickly *(t$\downarrow$)* the surrounding area of the starting point and that requires a long times *(t$\uparrow$)* to get long distances [@r23].
The BM presents a transition between a smooth ballistic behavior to the diffusive behavior. The scaling-time of the transition is given by the relaxation time $\tau$. In real physical systems, the relaxation time $\tau$ is ordinarily very short, typically in the order of some microseconds or nanoseconds [@r23; @r24; @r25; @r26]. For a short interval of time ${(\Delta t<\tau}$), ${\langle{\Delta r^{2}}\rangle}$ has a quadratic time dependence, and for long times (${\Delta t\gg\tau}$) this dependence becomes linear [@r23; @r24; @r25; @r26]. Even the theory was proposed in 1905 [@r02], the transition between ballistic and diffusive behavior was experimentally proofed only in 2010 [@r24], when was possible to measure the position of the particle to study the instantaneous velocity and the transition of the ballistic to the diffusive behavior.
![(Color online) Algodoo graph which presents the time series for *x* and *y* positions for Brownian disc.[]{data-label="Figure2"}](fig2.eps){width="1.0\linewidth"}
To study the BM by animation with Algodoo we use the graphical tool “*Show Plot*". With this tool we can chose one of the variables used in the animation and present them in a graphic, as illustrated in the fig. 2. In the case of BM, the time evolution of the horizontal *x(t)* and the vertical *y(t)* positions of the Brownian disk, was treated as time series. The option “*Show Plot*" exhibits the *x* and *y* positions of the disk as a function of time, that can be observed in the fig. 2 items (*A*) and (*B*) respectively. With this tool is possible to save a spreadsheet in an extension .CSV (item (*C*) of the fig. 2) with a frequency of 60 points per second. Was used one hundred Brownian disks and monitored the time evoluting of each disk during ten seconds. Since any disk presents a different initial condition in the one hundred CSV files were produced containing the time series of the Brownian disks, and the time series were used to analyze the BM.
In our animations, the diffusion coefficient *D* obtained from two methods. The first method is the evaluation of the graphic of the probability-distribution of the particle position. Using this graphic we can fit the gaussian curve by using:
$$P(x) = \frac{1}{{\sigma \sqrt {2\pi } }}e^{{{ - \left( {x - \langle{\Delta x}\rangle } \right)^2 } \mathord{\left/ {\vphantom {{ - \left( {x - \mu } \right)^2 } {2\sigma ^2 }}} \right. \kern-\nulldelimiterspace} {2\sigma ^2 }}}
.\label{eq03}$$
the variance $\sigma^2$ in the Brownian motion is equal to the MSD ${\langle{\Delta x^{2}}\rangle}$, since the mean displacement ${\langle{\Delta x}\rangle}$ is null in this case.
$$\sigma^2 = \langle{\Delta x^{2}}\rangle - \langle{\Delta x}\rangle^2
.\label{eq04}$$
We can also use the variance $\sigma^2$ from the displacements distribution ${\langle{\Delta x^{2}}\rangle}$ to calculate *D* by equation of one-dimensional random walk:
$$\sigma^2 = \langle{\Delta x^{2}}\rangle =2Dt
.\label{eq05}$$
Using any software that analyze mathematically the spreadsheet, as the Microsoft Excell, Open Office or Origin, we can do the mathematical treatment of the time series, as desired. By calculating the horizontals and verticals displacements $\Delta x_i=x_i(t)-x_i(t_0)$ and $\Delta y_i=y_i(t)-y_i(t_0)$ it is possible to create a list and to produce the graphic of the probability distribution of the particle displacements, which will be fitted using equation 3.
The second method consists in calculate the two-dimensional MSD for each instant of time t by:
$$\langle{\Delta x^{2}(t)}\rangle = \frac{1}{N} \sum_{i=1}^{N} \left[\left(x_i(t)-x_0(t_0)\right)^2+\left(y_i(t)-y_0(t_0)\right)^2\right]
.\label{eq06}$$
where the cartesian pair ${\left(x_i(t_0),y_i(t_0)\right)}$ represents the initial position of the particle in each sample and *N* is the number of samples. The diffusion coefficient *D* is the slope of the linear fit of the curve ${\langle{\Delta x^{2}(t)}\rangle}$ versus *t*, according to equation 1.
To demonstrate the transition between the ballistic and diffusive behavior we use the MSD curve for a short interval of time, generated by the second method. From the graphic of $\langle{\Delta r^{2}(t)}\rangle$ versus *t* we can demonstrate the quadratic and linear temporal behavior of $\langle{\Delta r^{2}(t)}\rangle$ and from this graphic the relaxation time $\tau$ is obtained.
In laboratories classes for undergraduate students, an experiment about a typical BM requires that the student register the motion of at least ten particles for measure the diffusion coefficient and/or the relaxation time. Experiments of BM in advanced stages became significant, but the cost of preparation and the quantification of these experiments are difficult, usually producing deviations about 10$\%$ to 15$\%$ or even more [@r22].
In practical terms, the study of the Brownian motion by computational experiments minimize largely the experimental problems. In this scenario, the Algodoo software is a good option and it stand out by easier manipulation, and does not requiring a specific programming knowledge. We do not propose here the elimination of BM experiments, but we present a new tool to support the teaching/learning of physics by animations built with Algodoo.
Results and Discussions {#sec3}
=======================
![In method 1, the diffusion coefficient is found by making a histogram of the *x* and *y* displacements (gray) and a Gaussian fit (black line) for the data; the variance is used to calculate the diffusion coefficient *D* = 0.221 $m^2$/s.[]{data-label="Figure3"}](fig3.eps){width="1.0\linewidth"}
With the animations done, teachers can establish with their students a great opportunity for discussions about the excitement of the small disks associating qualitatively to thermal excitement of the hypothetical fluid. Such thermal excitement produces continuous collisions with the red disk located in that region, causing an erratic motion, characterizing the Brownian motion (fig. 1 (a) e (b)). Due to the fact that this software can be easy manipulate, teachers can pause, increase or decrease the speed of the animations, or also modify the environment characteristic to enrich the qualitative presentation of the motion.
As said in previous section, the time series of the Brownian disk positions is saved, by “*Show Plot*" tool, in an extension .CSV where data are analyzed by worksheet software. We have executed one hundred independently Brownian motions for 10 s. During this interval of time we have exclude all initial points till 1.5 s. Until approximately 0.8 s we observe that the animations do not get the random walk state and consequently it can disturb the analysis of the motion. These observations are done for the animation with the initial conditions described in the previous section. If those conditions change, can one realize new analysis of the initial time interval. Therefore, we have established a 1.5 s, for convenience.
By using method $\#$1, we have fitted a Gaussian curve (equation 3) on the displacement distribution graphics in the fig. 3. The histogram is built from the independently one-dimensional displacements $\Delta x$ and $\Delta y$ together. The variance of the adjusted Gaussian curve is equal to the mean-square of one-dimensional displacement $\langle{\Delta x^{2}}\rangle$, and we can use this to calculate the diffusion coefficient by using equation 5. The obtained value from the displacement histogram is *D* = 0.221 m${^2}$/s.
![(Color online) (Gray) The square displacement of 100 samples. (Black) Mean square displacement versus time. Inside the linear fit for MSD with *D* = 0.205 $m^2$/s.[]{data-label="Figure3"}](fig4.eps){width="1.0\linewidth"}
After the square position displacements were calculated for all Brownian disks in the animation (fig. 4 – grey curves), we calculate the average over all samples, i.e., MSD (fig. 4 – black curve), by equation 6. The diffusion coefficient is determined by the slope of MSD as function of time by using equation 1, where we obtain *D* = 0.205 m${^2}$/s (using method $\#$2). We found a difference of about 8$\%$ between the methods.
For a short time interval (${\Delta t < \tau}$) the dynamics of the Brownian disk is governed by the translacional inertia and the motion is in ballistic regime. Is demonstrated in figure 5 that MSD has a parabolic time dependence (curve with red points) differing from the typical diffusive motion (black points), where MSD has a linear dependence. The relaxation time was availed at $\tau$ = 0.15 s. We have adjusted a curve with the form $\langle{\Delta r^{2}}\rangle$(t) = C$t^2$+Bt+A on the parabolic region and we obtain C = 1.9 $m^2$/$s^2$, B = 3.5${\times}$${10^{-2}}$ ${m^2}$/s and A = -1.9${\times}$${10^{-4}}$ ${m^2}$.
Changing the animation of the BM it’s possible to generate others diffusion regimes, that can be used to teach other diffusive phenomena, as electrical diffusive motion on semiconductor materials.
![(Color online) MSD is ballistic for (${\Delta t < \tau}$) with $\langle{\Delta r^{2}}\rangle$ $\propto$ $t^2$ - Adjust (red curve) with . (Black) For long times (${\Delta t \gg \tau}$) $\langle{\Delta r^{2}}\rangle$ becomes linear. The relaxation time was estimated at $\tau$ = 0.15 s.[]{data-label="Figure5"}](fig5.eps){width="1.0\linewidth"}
These animations can be used in several levels in the school. For the high school level, the animations of the BM can be used to illustrate the random motion of a particle suspended on a fluid. For undergraduate levels, the animations can be also a tool to facilitate comprehension of mathematics procedures involved in the BM studying. From the data collected by Algodoo, the teacher can elaborate scripts to teach the students how to obtain physical quantities from the time series, the same way as is done in laboratory experiments. Also, it would be an interesting introduction to the concepts of statistical physics, normal distribution, Gaussian curve, and so on which is very important in many areas of physics, from thermodynamics to quantum mechanics.
For higher levels, students can learn scientific procedures for formulation of models, collecting and data treatment. This basic knowledge supports students, beginning in the science world, by development of research projects.
Conclusions {#sec4}
===========
In this paper, we showed a computational tool of great potential for teaching and learning physics, using the freeware Algodoo. Using an animation based on the two-dimensional random walk movement of particles on fluid surface we can study the basic concepts of the Brownian motion as a great support-tool for teaching/learning of statistical physics.
The animations presented in this paper provide to teachers and students a simple tool for quantitative and quantitative analyzes of the BM. It is possible to observe and to discuss the the random walk movement of the small disks resulting in an erratic motion, and to associate ones qualitatively to the thermal excitement of a hypothetical fluid, which characterize the BM.
The diffusion coefficient was calculated from the animations by two methods: Method $\#$1 – fitting the graphic of the distribution of the independently one-dimensional displacements $\Delta x$ and $\Delta y$ together by a Gaussian curve. And Method $\#$2 – by the fitting the graphic MSD as function of the time, using equation 1. It was found a difference of approximately 8$\%$ between these two methods.
We have demonstrated that for a short interval of time (${\Delta t < \tau}$), $\langle{\Delta r^{2}}\rangle$ is proportional to *t${^2}$* and the motion presents a ballistic behavior, were we have a movement of the particle without suffering collisions in the path. For long times ${\Delta t \gg \tau}$, $\langle{\Delta r^{2}}\rangle$ is proportional to *t*, and the motion is diffusive, movement suffering collisions along the path. The time of relaxation was availed as $\tau$ = 0.15 s.
The Algodoo is easy to manipulate and does not require any specific knowledge in programming or training to realize the tasks in the software. Through this environment, educators and students can explore all potentialities of the studied theme, proposing modifications: in the size of the particles, in the intensity of the velocities, imposing boundary conditions; to the system and discussing their consequences: enhance of the relaxation time, reduction of the diffusive coefficient, and so on.
Many diffusive systems can be explored by these initial proposal of an animation of the Brownian motion. The didactic strategy for combining analytical approaches and animations, supports teaching and learning processes. In this way Algodoo also show one as a support-tool [@r20] that becomes the teaching/learning process of physics simpler and fruitful when compared to others software of animations and learning [@r10; @r11; @r12; @r13; @r14; @r15; @r16; @r17; @r18; @r19].
Acknowledgments {#sec4}
===============
The authors thanks FAPERJ for the finantial support of this work.
[10]{} Salinas, S.R.A. \[2005\] “Einstein e a teoria do movimento browniano,” [*Rev. Bras. Ens. Fis.*]{} [**27**]{}, 263 - 269.
Einstein, A. \[1905\] “On the movement of small particles suspended in stationary liquids required by the molecular-kinetic theory of heat,” [*Ann. Phys. (Berlin)*]{} [**15**]{}, 549.
E. Nelson [*Dynamical Theories of Brownian Motion*]{} (Princeton U.P., Princeton, NJ) \[1967\]
Silva, J.M. e Lima, J.A.S. \[2007\] “Quatro abordagens para o movimento browniano,” [*Rev. Bras. Ens. Fis.*]{} [**29**]{}, 25-35.
Silva, S.L. , Ferreira, J.A. and Martins, M.L. \[2007\] “ Epidemic spreading in a scale-free network of regular lattices,” [*Physica A*]{} [**377**]{}, 689-697.
Braga, F.L. and Ribeiro, M.S. \[2011\] “Diffusion Limited Aggregation: Algorithm optimization revisited,” [*Comput. Phys. Commun.*]{} [**182**]{}, 1602-1605.
Nakamura, T. and Small, M. \[2007\] “Tests of the random walk hypothesis for financial data,” [*Physica A*]{} [**377**]{}, 599-615.
Huang, K. \[2007\] “Conditioned self-avoiding walk (CSAW): stochastic approach to protein folding,” [*Biophysical Rev. and Lett.*]{} [**2**]{}, 139–154.
Makarava, N., Bettenbuhl, M., Engbert, R. and Holschneider, M. \[2012\] “Bayesian estimation of the scaling parameter of fixational eye movements,” [*EPL-Europhys Lett.*]{} [**100**]{}, 40003.
Rodrigues, M. & Carvalho, P. S. “Teaching physics with Angry Birds: exploring the kinematics and dynamics of the game,” [*Phys. Educ.*]{} [**48**]{},\[2013\] 431-437.
Hurkała, J., Gall, M., Kutner, R. and Maciejczyk, M. \[2005\] “ Real-time numerical simulation of the Carnot cycle,” [*Eur. J. Phys.*]{} [**26**]{}, 673-680.
Wee, L. K. \[2012\] “One-dimensional collision carts computer model and its design ideas for productive experiential learning,” [*Phys. Educ.*]{} [**47**]{}, 301-308.
Fernandes, J. C., Ferraz, A. & Rogalski, M. S. \[2010\] “Computer-assisted experiments with oscillatory circuits,” [*Eur. J. Phys.*]{} [**31**]{}, 299-306.
García-March, M. A., Córdoba, P. F., Urcheguía, J. F. & Monsoriu, J. A. \[2007\] “Introductory quantum physics courses using a LabVIEW multimedia module, ” [*Comput. Appl. Eng. Educ.*]{} [**15**]{}, 124-133.
Forjan, M., Marhl, M. and Grubelnik, V. \[2014\] “Mathematical modelling of the electrostatic pendulum in school and undergraduate education,” [*Eur. J. Phys.*]{} [**28**]{}, 015022.
Zachariadou, K., Yiasemides, K. & Trougkakos, N. \[2012\] “A low-cost computer-controlled Arduino-based educational laboratory system for teaching the fundamentals of photovoltaic cells,” [*Eur. J. Phys.*]{} [**33**]{}, 1599-1610.
De Souza, A. R., Paixão, D. D., Uzêda, A. C., Dias, M. A., Duarte, S. & Amorim, H. S. \[2011\] “The Arduino board: a low cost option for physics experiments assisted by PC,” [*Rev. Bras. Ens. Fis.*]{} [**33**]{}, 1702.
Teodoro, V. D. and Neves, R. G. \[2011\] “Mathematical modelling in science and mathematics education,” [*Comput. Phys. Commun.*]{} [**182**]{}, 8-10.
Martinez, E., Carbonell, V., Florez, M. & Amaya, J. \[2010\] “Simulations as a new physics teaching tool,” [*Comput. Appl. Eng. Educ.*]{} [**18**]{}, 757-761.
Algodoo “(online),” [ URL: https://www.algodoo.com \[December 2014\]]{}.
da Silva, S. L., da Silva, R. L., Guaitolin Junior, J. T., Gonçalves, E., Viana,E. R. and Wyatt, J. B. L. \[2014\] “Animation with Algodoo: a simple tool for teaching and learning physics,” [Exatas Online]{} [**5**]{}, 28-39.
Catipovic, M. A., Tyler, P. M., Trapani, J. G. and Cartera, A. R. \[2013\] “Improving the quantification of Brownian motion,” [*Am. J. Phys.*]{} [**81**]{}, 485-491.
Volpe, G. and Volpe, G. \[2013\] “Simulation of a Brownian particle in an optical trap,” [*Am. J. Phys.*]{} [**81**]{}, 224.
Li, T., Kheifets, S., Medellin, D. and Raizen, M. G. \[2010\] “Measurement of the instantaneous velocity of a brownian particle,” [*Science*]{} [**328**]{}, 1673-1675.
Li, T. and Raizen, M. G. \[2013\] “Brownian motion at short time scales,” [*Ann. Phys.*]{} [**525**]{}, 281–295.
Kheifets, S., Simha, A., Melin, K., Li, T. and Raizen, M. G. \[2014\] “Observation of brownian motion in liquids at short times,” [*Science*]{} [**343**]{}, 1493.
| {
"pile_set_name": "ArXiv"
} |
Sur le théorème des trois distances
et la construction des gammes
— École Nationale de Musique de Villeurbanne —
------------------------------------------------------------------------
Introduction {#introduction .unnumbered}
============
Ce mémoire trouve son origine dans une discussion informelle avec Guillaume Hanrot, professeur d’informatique à l’École Normale Supérieure de Lyon et musicien amateur. Il a remarqué qu’un théorème, démontré dans les année 1950 par la mathématicienne Vera Sós et connu sous le nom de [“]{}théorème des trois distances[”]{}, permet d’éclairer certains aspects de la théorie des gammes musicales. Après quelques recherches infructueuses sur internet, j’ai cru que ce lien n’avait pas encore été découvert par les musicologues. Mais une fois décidé le sujet de ce mémoire, des recherches plus poussées m’ont conduit à la thèse de Norman Carey, soutenue en 1998 à l’université de Rochester aux Etats-Unis, et qui porte essentiellement sur les propriétés formelles des gammes et en particulier, leur lien avec le théorème sus-mentionné. Je me trouvais d’une part conforté dans l’idée que le sujet était digne d’intérêt, et d’autre part un peu déçu d’avoir été précédé. De ce fait, on ne trouvera ici rien ou presque qui ne soit déjà présent dans la littérature.
On s’intéresse donc ici à l’une des multiples facettes du lien entre mathématiques et musique. Ce lien a fait couler beaucoup d’encre depuis des siècles, et c’est aujourd’hui encore un domaine actif de recherche. De par ma formation (j’ai étudié les mathématiques, et longuement), je me trouvais entraîné comme par une pente vers ce type de sujets, mais les pentes étant plus avantageusement remontées que descendues, j’aurais souhaité travailler sur autre chose. J’ajouterais que je ne regarde pas sans un certain scepticisme le lien entre mathématiques et musique. Pour réel et profond qu’il soit, son étude ne me semble pas toujours produire la meilleure des musiques, ni les meilleures des mathématiques. En outre, l’éloignement et la technicité des deux sujets qu’il se propose de rapprocher limitent le nombre des contradicteurs, ce qui est propice à diverses impostures, et paradoxalement à une certaine nonchalance. Bref, sans ce point de départ intrigant, à savoir ce magnifique théorème des trois distances, et sans l’enthousiasme de Claire Lapalu (professeur de formation musicale et responsable des mémoires de 3 cycle à l’ENM), j’aurais sûrement choisi de travailler sur un autre sujet, plus proche de mes intérêts musicaux, sans doute sur la chanson ou les musiques vocales.
La première partie de ce mémoire explique le problème général de la construction de gammes. La deuxième partie donne quelques rappels indispensables sur les logarithmes. Dans la troisième partie, le théorème des trois distances est expliqué et appliqué à la construction de gammes. Nous montrons comment il permet de retrouver naturellement les gammes pythagoricienne, pentatonique, diatonique et chromatique, ainsi que les demi-tons diatoniques et chromatiques, puis nous expliquons un procédé général de construction de gammes provenant de la thèse de Norman Carey.
Remarques préalables {#remarques-préalables .unnumbered}
--------------------
Avant de poursuivre, précisons que ce travail est conçu pour être lu par des musiciens ayant un bagage mathématique minimal (disons lycée ou fin de collège). Les notions de solfège comme [“]{}intervalle[”]{}, [“]{}quinte[”]{}, [“]{}transposition[”]{}, etc sont donc supposées connues du lecteur, au contraire des notions mathématiques, notamment les logarithmes, qui sont expliquées en détail. Notons que les logarithmes sont inutiles pour énoncer, comprendre, ou même prouver le théorème des trois distances. C’est son utilisation en musique qui nécessite les logarithmes.
Dans ce travail, on distingue la construction des gammes du problème bien connu des musiciens sous le nom de [“]{}tempérament[”]{}. Expliquons ce point. Une fois une gamme plus ou moins fixée (par exemple do-ré-mi-fa-sol-la-si-do), le tempérament consiste en l’ajustement exact des hauteurs des notes en fonction de différentes contraintes (consonnance des intervalles, possibilité de transposer une mélodie ou de moduler, fidélité à la musique ancienne, recherche d’une couleur, facture et accord d’instruments, etc). Le problème du tempérament est selon moi bien plus musical que mathématique.
Le problème de la construction de gammes est plus général: on ne présuppose aucune échelle, même approximative. En particulier le nombre de notes de la gamme n’est pas connu d’avance: il n’y a pas [*a priori*]{} 5 notes (gamme pentatonique), 7 notes (gamme diatonique) ou 12 notes (gamme chromatique). Le problème de la construction de gammes ne se pose pas de manière urgente aux musiciens qui ne semblent pas réclamer instamment de nouvelles gammes (alors qu’ils réclament que leurs instruments soient accordés, ce qui implique le choix d’un tempérament). Ainsi n’y a-t-il pas d’inconvénient majeur à considérer la question sous un angle exclusivement mathématique. Le problème est au demeurant surtout retrospectif, en ce qu’il éclaire le pourquoi des gammes aujourd’hui utilisées. Dans ses aspects exploratoires, il relève de la pure spéculation, avec tout ce que cela implique de promesse vague et de risque.
On se donne donc certaines contraintes musicales exprimées mathématiquement, et on cherche (mathématiquement encore) l’ensemble des sons permettant au mieux de les satisfaire. Une des conclusions de ce travail est que les nombres 5, 7, et 12 qu’on rencontre dans de nombreuses gammes ne proviennent pas seulement de contingences culturelles, mais ont des propriétés mathématiques remarquables qui expliquent [*a posteriori*]{} leur présence. Ce fait est connu depuis longtemps, mais le théorème des trois distances l’illustre de manière claire et saisissante. Cependant, le théorème des trois distances n’éclaire pas le problème du tempérament tel que défini ci-dessus.
Sur la bibliographie {#sur-la-bibliographie .unnumbered}
--------------------
Sur toutes les généralités concernant la construction des gammes, notamment le procédé pythagoricien et la définition des différents commas, Wikipédia est une source remarquablement claire (et fiable pour autant que je puisse en juger). La plupart des informations générales de ce mémoire en proviennent. Sur le tempérament, le livre de Pierre-Yves Asselin [@asselin:mt] m’a paru tout à fait remarquable. Il est principalement centré sur la résolution pratique de véritables problèmes musicaux, comme l’accordage des clavecins. Sa lecture est donc un peu frustrante dans la mesure où elle ne donne pas à *entendre* l’effet musical des différents tempéraments évoqués. Dans le livre de Gareth Loy [@Loy:musimathicsV1], on trouve une bonne synthèse sur les mathématiques et la musique, qui explique bien les bases de l’accoustique et de la construction des gammes. Une preuve du théorème des trois distances peut être consultée dans le livre de Jean-Paul Allouche et Jeffrey Shallit [@DBLP:books/daglib/0025558]. On peut aussi consulter la synthèse de Pascal Alessandri et Valérie Berthé [@alesandriniBe:98]. Sur le lien entre ce théorème et la musique, les seules sources que j’ai trouvées sont la thèse de Norman Carey [@carey:these] et l’article qu’il a co-signé avec David Clampitt [@careyCl:2012].
Construction de gammes
======================
Un *son* est la perception que l’oreille humaine a d’une oscillation périodique de la pression de l’air, appelée *onde sonore*. La *hauteur* d’un son s’exprime mathématiquement par sa fréquence, exprimées en hertz, c’est-à-dire par le nombre d’oscillations par seconde de l’onde qui le produit. Pour fixer les idées concrètement, un son de $440$ hertz est un La obtenu juste à droite de la serrure d’un piano. L’oreille humaine humaine peut entendre des sons d’environ 20Hz (le plus grave) à 20000Hz (le plus aigu). Entre ces bornes extrêmes, l’ensemble des hauteurs possibles est un continuum, c’est-à-dire que n’importe quel nombre réel[^1] est la fréquence d’un son. Il y a donc potentiellement une infinité de hauteurs possibles.
Pour des raisons culturelles et pratiques (touches d’un clavier, frettes d’un manche, trous d’une flûte…), les musiciens préfèrent limiter les hauteurs de sons à un ensemble fini. Construire une gamme, c’est choisir les fréquences des sons auxquels on se restreint. Le problème peut paraître abstrait, mais il s’est posé concrètement, bien qu’en des termes différents, aux tout premiers facteurs et accordeurs d’instruments, sans doute à des facteurs de flûtes de la préhistoire (la page Wikipédia [“]{}instrument de musique[”]{} mentionne les restes d’une flûte en os fabriquée 45000 ans avant notre ère). La théorie des ondes sonores date seulement du XVII siècle avec les travaux de Mersenne et Galilée, mais il se trouve que la fréquence d’un son émis par une corde grattée est inversement proportionnelle à la longueur de la corde, ce qui a permis sans théorie ni instrument de mesure sophistiqué de quantifier précisemment la hauteur des sons dès la plus haute antiquité. Les aspects mathématiques du problème de la construction des gammes sont donc reconnus depuis au moins 2500 ans, et les Grecs anciens attribuent cette découverte à l’école de Pythagore. On en vient donc au premier principe de la construction de gammes (que l’on choisit d’exprimer mathématiquement).
[**Principe 1:**]{} une gamme est un ensemble fini (et en pratique assez petit) de nombres, qui sont les fréquences des différents sons de la gamme.
Les pythagoriciens, ou peut-être Pythagore lui-même (on manque de sources sur ces temps reculés), ont remarqué qu’en doublant la fréquence d’un son, on obtient un deuxième son très consonnant avec le premier, à tel point que, dans certains contextes musicaux, ces deux sons peuvent être considérés comme équivalents (dans notre système musical, il s’agit d’une note et de la même note à l’octave supérieure). Ceci mène au deuxième principe de la construction des gammes, le principe de l’identité des octaves, explicité pour la première fois par Jean-Philippe Rameau dans son [*Traité de l’harmonie réduite à ses principes naturels*]{}, mais bien évidemment utilisé implicitement par les musiciens l’ayant précédé:
[**Principe 2:**]{} si un nombre est dans la gamme, le nombre obtenu en le doublant est considéré comme la fréquence d’un son équivalent musicalement.
De même, en mulpliant la fréquence d’un son par d’autres nombres rationnels simples[^2], comme par exemple $3/2$ ou $4/3$, $5/4$ ou $6/5$, on obtient un nouveau son formant un intervalle assez consonnant avec le premier, et perçu par l’oreille humaine comme musicalement intéressant. Par exemple, en multipliant par $3/2$, on obtient ce qu’on appelle dans notre système musical une quinte pure, par $4/3$ une quarte pure, par $5/4$ une tierce majeure pure et par $6/5$ une tierce mineure pure. Notons toutefois qu’à ce stade, les mots [“]{}tierce[”]{}, [“]{}quarte[”]{} et [“]{}quinte[”]{} sont fallacieux puisqu’ils présupposent un écart de $3$, $4$ ou $5$ notes entre les deux sons, alors qu’aucune gamme (et aucun nombre de notes) n’est présupposé. Il s’agit simplement d’intervalles consonnants, qu’il faudrait peut-être nommer autrement.
Les auteurs se perdent en conjectures sur l’origine de cette perception de consonnance, à tel point qu’on peut se demander si la conclusion de tel ou tel n’en dit pas plus sur lui-même que sur l’origine réelle de la consonnance. Est-elle physiologique, acoustique, culturelle, numérologique, mystique ou mathématique? Y a-t-il d’ailleurs une origine [“]{}réelle[”]{} de la consonnance? Pour la commodité de l’écriture de ce mémoire, nous tenons la consonnance des intervalles provenant de rapports rationnels simples pour un fait brut que nous ne cherchons pas à expliquer plus avant. On en vient au troisième principe de la construction des gammes:
[**Principe 3:**]{} si un nombre est dans la gamme, on souhaite que les nombres obtenus par multiplication par des rapports simples, comme $2$, $3/2$ ou $4/3$ (correspondant à des intervalles consonnants ascendants), et leurs inverses $1/2$, $2/3$ ou $3/4$ (correspondant aux mêmes intervalles descendants), soient aussi présents dans la gamme.
Pris ensemble, les trois principes sont mathématiquement contradictoires. On peut prouver qu’il est impossible de construire un ensemble fini de nombres réels clots par multiplication par $1/2$, $2$, $3/2$ et $2/3$. La preuve n’est pas très compliquée, mais son intérêt pour ce mémoire étant limité nous l’omettons. Indiquons qu’elle se ramène entre autres au fait qu’en multipliant indéfiniment des $3$, on n’obtiendra jamais un multiple de $2$. L’impossibilité de construire une gamme pleinement satisfaisante vient donc d’une impossibilité mathématique fondamentale, et non d’un manque d’habileté des mathématiciens ou des musiciens. C’est de là que la construction de gammes est réellement un *problème*.
Devant l’impossibilité de construire une gamme répondant pleinement aux trois principes, plusieurs stratégies sont envisageables: on peut s’affranchir partiellement du premier principe en acceptant qu’il y ait un nombre de sons possibles, non pas infini, mais s’en approchant, c’est-à-dire très grand (cette méthode est par exemple utilisée dans le tempérament à 53 intervalles égaux de Holder qu’on rencontrera dans la troisième partie de ce mémoire); on peut se contenter d’une approximation des rapports simples du deuxième principe (c’est la solution actuellement retenue dans ce qu’on appelle le tempérament égal); on peut garder des rapports exacts pour un maximum d’intervalles, en en [“]{}sacrifiant[”]{} certains autres qui deviennent inutilisables musicalement (c’est la solution retenue dans de nombreux systèmes anciens de tempérament, où une quinte est [“]{}sacrifiée[”]{}, la fameuse quinte du loup).
Logarithmes
===========
Il est hélas difficile de poursuivre sans un peu de bagage mathématique. L’une des difficultés techniques pour comprendre le lien entre le théorème des trois distances et la construction des gammes est que l’accoustique suit une logique multiplicative (on obtient un intervalle de quinte pure en *multipliant* par $3/2 = 1.5$), tandis que les musiciens pensent plutôt les intervalles de manière additive, ce en quoi ils ont parfaitement raison (on obtient une quinte en *additionnant* une tierce majeure et une tierce mineure, sûrement pas en les multipliant). Ce passage d’une logique multiplicative à une logique additive est tout à fait légitime. Les mathématiciens l’appellent un [“]{}passage au logarithme[”]{}, et je n’ai malheureusement pas trouvé de moyen de me passer de cette notion (le mathématicien Henri Cartan disait : [“]{}Il est plus facile d’apprendre les mathématiques que d’apprendre à s’en passer[”]{}).
Le logarithme est une notion tout à fait concrète, disons aussi concrète que l’addition ou la multiplication. Tout d’abord, le logarithme est une *fonction*, c’est-à-dire que c’est un procédé qui transforme un nombre en un autre nombre. Par exemple, le logarithme de 1 vaut 0, ce qui se note $\log(1) = 0$. Le logarithme de $12.7$ est un nombre assez difficile à calculer à la main; on ne peut même pas donner de niveau scolaire correspondant à cette tâche, puisque le calcul numérique a presque entièrement disparu des programmes de l’enseignement secondaire. Mais la plupart des calculatrices, utilisées à partir du collège et responsables de cette disparition, ont pour se la faire pardonner une touche log qui permet de trouver que le logarithme de $12.7$ vaut environ $1.1038$, ce qui s’écrit $\log(12.7) = 1.1038$.
Voilà donc un problème de résolu: grâce à une calculatrice, on peut tout à fait concrètement calculer le logarithme de n’importe quel nombre. Mais que signifie le logarithme? Pour s’en faire une première idée approximative, il suffit de compter les chiffres du nombre, puis d’enlever $1$. Par exemple, $103$ s’écrit avec $3$ chiffres (un, zéro et trois), son logarithme est donc à peu près $3-1=2$, en fait $2.0128$, soit $2$ [“]{}et des poussières[”]{} comme on peut le vérifier à la calculatrice. Si on s’approche de nombres à $4$ chiffres (donc de $1000$), l’approximation devient de moins moins bonne, par exemple $\log(998) = 2.9991$, ce qui est assez [“]{}éloigné[”]{} de $2$, et presque égal à $3$ (998 a [“]{}presque[”]{} 4 chiffres). Le nombre $367\,987\,434$ a $9$ chiffres, son logarithme sera environ de $9-1=8$ (et encore des poussières). Cette définition approximative (car on ne dit pas comment calculer les [“]{}poussières[”]{}, qui seront pourtant essentielles par la suite) permet de saisir quelques propriétés importantes du logarithme. Par exemple, qu’un nombre très grand a un logarithme assez petit: un nombre dont le logarithme serait $50$ doit être vraiment gigantesque, puisqu’il s’écrit avec $51$ chiffres. Et aussi, que quand on multiplie un nombre par $10$, on lui ajoute un chiffre, ce qui fait que son logarithme augmente de $1$.
Pour les nombres $1$, $10$, $100$, etc, (ce qu’on appelle les puissances de $10$) la méthode de calcul du logarithme donnée ci-dessus est en fait exacte, et le logarithme est précisement le nombre de chiffres moins $1$, ou ce qui revient au même, le nombre de zéros. Et il est bien connu que pour multiplier des puissances de $10$, il suffit d’additionner le nombre de zéros (par exemple, $100$ a $2$ zéros, $1000$ a $3$ zéros, et $100\times 1000 = 100000$ a $2+3 = 5$ zéros). Ainsi, quand on mulptiplie des nombres, leurs logarithmes s’additionnent (du moins pour les puissances de $10$). Cela s’écrit $\log(a \times b) = \log(a) + \log(b)$. Cette formule ci-dessus est en fait vraie pour tous nombres réels strictement positifs $a$ et $b$. La définition approximative du logarithme donnée ici ne permet pas de prouver cette relation. En effet, calculer ou définir précisement le logarithme est difficile (il faut ajuster ces fameuses [“]{}poussières[”]{} en s’arrangeant pour que la formule soit vraie), et l’existence même des logarithmes est un fait mathématique tout à fait remarquable découvert seulement au XVII siècle, c’est-à-dire bien après la théorie du tempérament qui date de l’Antiquité. Les logarithmes peuvent être vus comme un dictionnaire qui traduit les multiplications en additions. Ce fut d’ailleurs jusqu’à l’avènement des calculatrices l’une de leurs applications: pour mutiplier deux nombres (opération fastidieuse), on calculait leur logarithme grâce à une table établie une fois pour toute, on additionnait les résultats (opération facile), puis on regardait à nouveau dans la table pour voir de quel nombre le résultat était le logarithme. Ce nombre est le résultat de la multiplication. L’addition pouvait d’ailleurs être effectuée analogiquement, c’est-à-dire en mettant bout-à-bout deux réglettes graduées de longueur égale aux nombres à additionner, et en trouvant le résultat par une mesure de la longueur totale: c’est le principe de la fameuse règle à calcul, aujourd’hui tombée en désuétude.
Dans le cadre de ce travail, nous admettons simplement qu’il existe une fonction notée $\log$, dont l’existence est matérialisée par la touche [“]{}log[”]{} des calculatrices, et telle que pour tous nombres $a$ et $b$, on a: $\log(a\times b) = \log(a) + \log(b)$.
Pour ne pas compliquer la discussion, j’ai omis jusque-là de préciser qu’il y a plusieurs fonctions logarithmes différentes. Celle que j’ai approximativement définie ici est ce que les mathématiciens appellent le logarithme en base $10$. Il a la propriété que lorsqu’on multiplie un nombre par $10$, son logarithme augmente de $1$. Pour la théorie des gammes, la multiplication par $10$ n’a pas grand intérêt, alors que la multiplication par $2$ est riche de sens (elle fournit l’intervalle d’octave). On utilisera donc plutôt le logarithme en base $2$. Il peut être obtenu avec une calculatrice en calculant le logarithme en base $10$, puis en le multipliant par une constante à peu près égale à $3.32$. Une propriété intéressante du logarithme en base $2$ est que si un nombre est multiplié par $2$, alors son logarithme augmente de $1$.
À partir de maintenant, $\log(x)$ désigne le logarithme en base $2$ de $x$. Il est très important de remarquer que $\log(2) = 1$. Et la formule $\log(a\times b) = \log(a) + \log(b)$ reste vraie.
Nous sommes maintenant mûrs pour reformuler le troisième principe de la construction des gammes. Comme la multiplication est une opération plus difficile que l’addition, il est plus simple de rechercher non pas les fréquences des sons, mais leur logarithme. On choisit certains intervalles simples, comme la quinte pure, ou la tierce pure, sous la forme de rapports rationnels, $3/2$ et $5/4$ comme déjà expliqué. On souhaite que, si une note est dans la gamme, alors en additionnant (et non plus en multipliant) un nombre correspondant à l’octave ou à la quinte pure par exemple, on reste dans la gamme. Ces nombres à additionner ne sont autres que les logarithmes des rapports précédents. L’élévation d’une octave correspond donc à l’ajout de 1 (car $\log(2)=1$). On le verra dans la troisième partie, le nombre $0.58496$ joue un grand rôle dans la construction de la gamme pythagoricienne. C’est qu’il n’est autre que le logarithme de $3/2$.
Ce qu’on peut retenir de ce chapitre, c’est qu’ajouter $0.58496$ à la fréquence d’un son (ou plutôt à son logarithme) est une opération naturelle musicalement (il s’agit de s’élever d’une quinte). Il est important de comprendre que ce nombre n’a rien d’arbitraire, du moins n’est-il pas plus arbitraire que la quinte elle-même. Il ne provient ni des mathématiques (même si les mathématiques participent à une étape purement technique de son élaboration), ni d’une quelconque mystique numérologique, comme si l’on avait choisi $\pi$, ou la racine carrée de l’âge du chef d’orchestre, ou le nombre d’or (quoiqu’étudier le nombre d’or dans ce contexte ne serait peut-être pas si absurde étant données les propriétés remarquables de son développement en fraction continue, mais cela sort du cadre de ce travail).
Le théorème des trois distances
===============================
Le théorème des trois distances concerne l’addition répétitive d’un même nombre, ce qui est intéressant musicalement, puisque comme on l’a vu, cela peut être interprété comme l’ajout successif de quintes ou d’autre intervalles. Il ne provient pas de recherches sur la musique et les sources à propos de son application en musicologie ne remontent pas au-delà de la fin des années 1990. J’estime toutefois probable que certains mathématiciens bien au fait du procédé pythagoricien de constructions des gammes ont pu (à l’instar de Guillaume Hanrot) deviner le lien. Il me semble plus improbable que des musicologues professionnels aient deviné ce lien, car si le procédé pythagoricien est assez largement connu (y compris de nombreux mathématiciens qui pratiquent la musique), le théorème des trois distances n’est quant à lui connu que de mathématiciens spécialisés. Mais énonçons le théorème.
Postulons un cercle de longueur 1 avec un point de départ. Par exemple, la piste circulaire d’un stade longue de 1 kilomètre, mais peu importe l’unité de longueur utilisée:
{width="1cm"}
Un marcheur avance à partir du point de départ d’une longueur fixée, qu’on appelle la [“]{}longueur de son pas[”]{}, par exemple $0.58496$ (nombre intéressant on l’a vu). Le marcheur arrive à un certain endroit:
{width="1cm"}
Il refait un pas de même longueur, puis encore un, et ainsi de suite, un grand nombre de fois, appelé le [“]{}nombre de pas[”]{}. On garde trace sur le cercle de toutes ses stations, mais on ne tient aucun compte de l’ordre dans lequel les stations ont été visitées. On se retrouve donc avec un cercle et un certain nombre de points sur ce cercle. Ci-dessous, les cercles successifs après douze pas:
{width=".9cm"} {width=".9cm"} {width=".9cm"} {width=".9cm"} {width=".9cm"} {width=".9cm"} {width=".9cm"} {width=".9cm"} {width=".9cm"} {width=".9cm"} {width=".9cm"} {width=".9cm"}
On regarde maintenant un deuxième marcheur, qui souhaiterait parcourir les différentes stations du premier, mais cette fois dans leur ordre naturel sur le cercle. Il devra faire des pas assez petits (et ce d’autant plus que le premier marcheur a créé beaucoup de stations, qui [“]{}encombrent[”]{} le cercle), et pas nécessairement tous de la même longueur sur le cercle. Le théorème des trois distances affirme:
[**Théorème:**]{} Pour toute longueur de pas choisie au départ, et pour tout nombre de pas du premier marcheur, le deuxième marcheur rencontrera au maximum trois longueurs de pas différentes lors de son parcours.
Un lecteur méticuleux pourra vérifier la validité de l’énoncé sur chacun des douze cercles de la figure ci-dessus. Le théorème concerne la distance curviligne (c’est-à-dire le long du cercle) entre deux points, mais celle-ci étant en correspondance bi-univoque avec la longueur de la corde, un double-décimètre permet de procéder pratiquement aux vérifications. Parfois, il y a bien trois distances différentes, mais parfois seulement deux, nous y reviendrons (le théorème dit bien trois *au maximum*). Bien qu’assez simple, cet énoncé n’est pas évident à démontrer rigoureusement. Il a d’abord été conjecturé par Hugo Steinhaus, puis démontré par Vera Sós dans les années 1950, avant que plusieurs mathématiciens n’en donnent diverses preuves. Il est appelé de différentes manières, ce qui ne facilite pas les premières recherches à son sujet: théorème de Steinhaus, conjecture de Steinhaus, théorème des trois longueurs, ou [“]{}three gap theorem[”]{} en anglais. À son sujet, on peut consulter le livre de Allouche et Shalit, dont la lecture nécessite de bonnes connaissances mathématiques (disons bac+2 ou 3).
Le cercle de longueur $1$ représente toutes les fréquences des sons possibles (ou plutôt, leur logarithme). Le point de départ représente une note choisie arbitrairement comme début de la gamme. Ce peut être un Do, ou le La 440. Ma préférence pour établir un lien avec le solfège classique va au Fa, parce que c’est le point de départ du cycle des quintes dans le fameux ordre des dièses. Sur les figures suivantes, je pars donc du Fa. La forme circulaire est pertinente en raison du principe d’identité des octaves: quand on monte d’une octave, c’est-à-dire quand on multiplie par $2$ une fréquence, ou qu’en logarithme on ajoute $1$ (puisque $\log(2) = 1$), on revient à son point de départ. Le cercle a cette même propriété: si on avance de $1$, qui est justement la circonférence du cercle, on revient à son point de départ.
Monter d’une quinte pure, c’est faire un pas de longueur $\log(3/2) = 0.58496$. Si le premier marcheur est un pythagoricien, il fait donc douze pas, qui correspondent à douze élévations consécutives d’une quinte, parfois réduite à l’octave (chaque fois qu’on repasse par le départ). Le second marcheur lui parcourra les douze sons dans l’ordre des aiguilles d’une montre, qui forment ce qu’on appelle la gamme chromatique pythagoricienne. Un analyse fine des longueurs de pas du second marcheur montre qu’il parcourt des pas correspondant à des demi-tons, mais de deux longueurs différentes. Comme on le voit sur la figure suivante, chaque ton se divise en deux demi-ton inégaux, et les demi-tons diatoniques sont bien inférieurs aux demi-tons chromatiques:
{width="3cm"}
Le musicien trouvera là une origine à la distinction entre demi-ton diatonique et demi-ton chromatique. Le mathématicien quant à lui remarquera que le théorème des trois distances est bien pessimiste: il prévoit trois distances au maximum pour le second marcheur, or nous n’en observons que deux, ce qui est souhaitable (deux sortes de demi-ton, c’est déjà compliqué, trois ce serait bien pire). En fait cette situation désirable se rencontre un certain nombre de fois lors du parcours du premier marcheur. On la rencontre après $2$ pas, après $5$ pas, après $7$ pas, puis enfin après $12$ pas, comme le lecteur peut le vérifier sur les quatre figures suivantes :
{height="2cm"} {height="2cm"} {height="2cm"} {height="2cm"}
Voilà une explication remarquable à la prédominance des gammes à $5$, $7$ ou $12$ sons dans plusieurs systèmes musicaux. Cette explication n’a rien de neuf au demeurant, elle est mathématiquement équivalente à des considérations bien connues (quoique assez techniques) sur le développement en fractions continues de $\log(3/2)$, mais le cercle et le théorème des trois distances en donnent une belle illustration. Le procédé est général: on peut partir d’un intervalle quelconque jugé intéressant, on calcule son logarithme qui donne une longueur de pas, puis on fait marcher le premier marcheur un certain nombre de fois, en espérant rencontrer une situation favorable où le second marcheur n’a que deux distances à parcourir. Ce procédé fait jouer un rôle central à l’octave (car on utilise le logarithme en base $2$), mais on peut par exemple faire des pas égaux à une tierce majeure, et calculer un logarithme en [“]{}base de quinte[”]{}, c’est-à-dire en base $3/2$ (ce qui revient à considérer que deux notes séparées d’une quinte sont [“]{}équivalentes[”]{}), ou dans la base que l’on souhaite. Norman Carey a étudié dans sa thèse ce principe général de construction de gammes; il a montré qu’il conduisait toujours à ce qu’il appelle des gammes [“]{}bien formées[”]{} (well-formed scales), un concept qu’il a développé à l’origine indépendamment du théorème des trois distances. Une propriété centrale d’une gamme bien formée est que l’intervalle qui sert à la générer (ici la quinte), se divise toujours en un nombre égal de [“]{}petites distances[”]{} du second marcheur, ici sept demi-tons. Cette propriété, évidente pour le musicien habitué à la gamme chromatique, est vraie pour toute gamme obtenue selon le procédé décrit ci-dessus, ce qui n’est pas évident [*a priori*]{}. Ceci est une propriété supplémentaire du théorème des trois distances, démontrée par Norman Carey (où l’on voit que des questions posées par des musiciens peuvent faire avancer les mathématiques).
Que se passe-t-il si on ajoute un 13 pas ? On tombe sur un son très proche du Fa (il est bien connu que 12 quintes pures correspondent à peu près à 7 octaves):
{height="2cm"}
L’écart entre le Fa et ce 13 son est ce qu’on appelle le comma pythagoricien. Mathématiquement, après ce 13 pas, le deuxième marcheur se retrouve de nouveau avec trois distances sur le cercle (dont une minuscule). En fait, si on retourne à la figure du cercle de la gamme chromatique avec les noms de notes, on voit que la quinte qui va du La\# au Fa (ou plutôt de l’enharmonique Sib de La\#, jusqu’au Fa, pour bien avoir une quinte au sens du solfège, et non une sixte diminuée), cette quinte donc est inférieure à toutes les autres quintes du cercle: c’est la seule à contenir $5$ demi-tons diatoniques au lieu de $4$, et aussi la seule à ne pas être pure: elle ne provient pas d’une multiplication par $3/2$. C’est la fameuse quinte du loup des tempéraments anciens.
Mais que se passe-t-il si le premier marcheur marche suffisamment longtemps, au-delà de ce treizième pas? Se peut-il qu’on se retrouve de nouveau dans le cas favorable de seulement deux distances? La réponse est oui, c’est une autre conclusion du théorème des trois distances: il est toujours possible de faire marcher plus longtemps le premier marcheur, jusqu’à un point où le nombre de distances différentes du deuxième marcheur est au maximum 2. Le procédé général fonctionne donc à tous les coups, ce qui n’avait rien de certain jusque-là, si l’on accepte toutefois de marcher assez longtemps. Pour retrouver une telle situation favorable au-delà de 12 pas (dans le cas de l’élévation par quintes), il faut marcher au total 53 pas. Ceci mène à la gamme à 53 intervalles de Holder, dont le principe est en fait connu depuis l’Antiquité (selon Wikipédia, il a été découvert par le théoricien chinois Ching Fang (78-37 avant JC)). On conçoit aisément qu’une gamme à 53 micro-intervalles (égaux au comma de Holder, qui divise le demi-ton chromatique en 5, et le demi-ton diatonique en 4) ne soit pas très pratique et que la gamme de Holder reste cantonnée à la théorie.
On voit donc que la gamme à douze sons est tout à fait remarquable. Elle a de nombreuses autres propriétés qui ne découlent pas du théorème des trois distances pour autant qu’on sache. Par exemple le rapport entre do et mi (une fois débarrassé des logarithmes, et donc vu à nouveau multiplicativement) est remarquablement proche de 5/4 (l’intervalle de tierce majeure pure) qui est [“]{}naturellement[”]{}, ou au moins culturellement, consonnant. Le rapport de ces deux fréquences est appelé le comma syntonique. Il vaut $81/80$, ce qui est remarquablement proche de 1, et je n’ai trouvé nulle part d’explication à ce phénomène, qui semble une heureuse coïncidence mathématique. Autre coïncidence, le rapport entre la tierce mineure pythagoricienne et 6/5 (tierce mineure pure) est par chance lui aussi égal au comma syntonique (sans quoi il y aurait un comma syntonique mineur et un comma syntonique majeur).
Comme on l’a déjà dit, rien n’empêche d’utiliser ce même principe de construction des gammes en choisissant d’autres valeurs comme point de départ. Par exemple, on peut choisir de baser toute la construction sur l’intervalle de tierce majeure pure 5/4 et les octaves. Sans préjuger du résultat, cela peut être motivé par l’harmonie dans la musique occidentale. Quels sont les nombres de pas conduisant à des situations favorables ? Voici les 29 premières étapes de la construction:
{width="1.11cm"} {width="1.11cm"} {width="1.11cm"} {width="1.11cm"} {width="1.11cm"} {width="1.11cm"} {width="1.11cm"} {width="1.11cm"} {width="1.11cm"} {width="1.11cm"} {width="1.11cm"} {width="1.11cm"} {width="1.11cm"} {width="1.11cm"} {width="1.11cm"} {width="1.11cm"} {width="1.11cm"} {width="1.11cm"} {width="1.11cm"} {width="1.11cm"} {width="1.11cm"} {width="1.11cm"} {width="1.11cm"} {width="1.11cm"} {width="1.11cm"} {width="1.11cm"} {width="1.11cm"} {width="1.11cm"} {width="1.11cm"}
On constate que la situation favorable des deux distances se rencontre à la troisième étape (ça n’est pas une surprise pour le musicien: trois tierces majeures valent une octave dans le tempérament égal), et ensuite à la 28 étape seulement. On obtient une gamme de $28$ micro-intervalles, $3$ longs et 25 petits comme on le voit sur la figure, presbytie à part. On pourrait calculer différents commas, évaluer la fausseté de la dernière tierce, une sorte de [“]{}tierce du loup[”]{}, mesurer à quel point la gamme permet de s’approcher des quintes, etc, mais je stoppe là faute de place et de temps.
Conclusion {#conclusion .unnumbered}
==========
Le théorème des trois distances explique bien certains aspects des constructions classiques de gammes. Les figures circulaires de ce mémoire sont selon moi plus faciles à comprendre que les considérations classiques sur le développement en fraction continue du logarithme de $3/2$. Il suffit d’accepter que l’idée d’additionner $0.58496$ est intéressante, par paresse, ou comme un argument d’autorité, ou en s’efforçant de comprendre les logarithmes, chacun s’y retrouvera selon sa personnalité. Ensuite, avec simplement quelques figures, on trouve une explication aux gammes pentatoniques, diatoniques et chromatiques, à la distinction entre les demi-tons chromatiques et diatoniques, et à d’autres choses que je n’ai peut-être pas vues, ce qui n’est déjà pas si mal.
Une nouvelle manière de comprendre les constructions du passé pourrait être un point de départ de celles du futur. J’ai à peine esquissé cet aspect en proposant une construction basée sur l’intervalle de tierce majeure pure. Cette construction, et toutes celles qu’on pourrait faire sur le même principe, ne nécessitent pas vraiment le théorème des trois distances (encore une fois, c’est un cadre commode de représentation) et les plus naturelles, ou disons les moins artificielles, sont très probablement déjà connues, bien que je n’aie pas trouvé de sources à leur sujet. Poursuivre dans cette voie et en aborder le côté réellement musical se heurtera nécessairement aux habitudes culturelles: des mouvements conjoints dans des gammes étranges sonneront bizarrement à nos oreilles, il y a peu de risque à le parier.
Évidemment, il faudrait être bien naïf pour penser que l’application de quelques formules simples, même issues de mathématiques récentes et enrobées d’un vocabulaire abscons, produira un résultat artistique intéressant, surtout après des millénaires de culture musicale de plusieurs civilisations. Ça n’en reste pas moins amusant, et la naïveté est parfois une qualité chez le scientifique ou l’artiste, alors pourquoi ne pas essayer? Le risque est surtout de parvenir à la conclusion que le système classique n’a pas survécu jusqu’à nous sans de bonnes raisons et possède des qualités intrinsèques difficiles à dépasser, ou même à imiter. Pour illustrer ce type de circularités vaines, je me permets pour conclure de citer la cinquième œuvre au catalogue des écrits de Pierre Ménard, l’immortel auteur du Quichotte, telle que relatée par Jorge Luis Borges:
[*[“]{}Un article technique sur la possibilité d’enrichir le jeu d’échecs en éliminant un des pions de la tour. Ménard propose, recommande, discute et finit par rejeter cette innovation[”]{}.*]{}
Remerciements {#remerciements .unnumbered}
=============
Je remercie Guillaume Hanrot pour m’avoir donné l’idée de ce travail, et Claire Lapalu de l’ENM pour l’avoir validée. Merci à Valérie Berthé qui m’a donné de nombreuses pistes de travail m’ayant finalement conduit à la thèse de Norman Carey. Je remercie ce dernier pour m’avoir aimablement communiqué sa thèse. Et merci à ma chère épouse Christelle Petit d’avoir relu le document.
[1]{} P. Alessandri and V. Berth[é]{}, *Three distance theorems and combinatorics on words*, L’Enseignement Mathématique. Revue Internationale. IIe Série **44** (1998), no. 1–2, 103–132.
J.[-]{}P. Allouche and J.O. Shallit, *Automatic sequences – theory, applications, generalizations*, Cambridge University Press, 2003.
P.-Y. Asselin, *Musique et tempérament*, Jobert, 2007.
N. Carey, *Distribution modulo 1 and musical scales*, Ph.D. thesis, Rochester University, 1998.
N. Carey and D. Clampitt, *Two theorems concerning rational approximations*, Journal of Mathematics and Music **6** (2012), no. 1, 61–66.
G. Loy, *Musimathics—the mathematical foundations of music. [V]{}olume 1*, MIT Press, Cambridge, MA, 2006, With a foreword by Max Mathews, \[2011\] reprint of the 2006 original.
[^1]: Un nombre réel est un nombre [“]{}avec ce qu’on veut après la virgule[”]{}.
[^2]: Un nombre rationnel est le résultat de la division de deux nombres entiers ([“]{}entier[”]{} signifie sans rien après la virgule).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Let $H$ be a separable complex Hilbert space. Denote by ${\mathcal G}_{\infty}(H)$ the Grassmannian consisting of closed linear subspaces with infinite dimension and codimension. This Grassmannian is partially ordered by the inclusion relation. We show that every continuous order preserving bijective transformation of ${\mathcal G}_{\infty}(H)$ is induced by an invertible bounded semi-linear operator.'
address: 'Department of Mathematics and Information Technology, University of Warmia and Mazury, [. Z]{}olnierska 14A, 10-561 Olsztyn, Poland'
author:
- Mark Pankov
title: 'Order preserving transformations of the Hilbert grassmannian: complex case'
---
Introduction
============
Let $H$ be a separable (real or complex) Hilbert space. We write ${\mathcal G}(H)$ for the lattice of closed linear subspaces of $H$. The group of invertible bounded linear operators acts on ${\mathcal G}(H)$ and the orbits of this action is called [*Grassmannians*]{}:
1. ${\mathcal G}_{k}(H)$ consists of $k$-dimensional linear subspaces,
2. ${\mathcal G}^{k}(H)$ consists of closed linear subspaces with codimension $k$,
3. ${\mathcal G}_{\infty}(H)$ consists of closed linear subspaces with infinite dimension and codimension.
Every closed linear subspace $S\subset H$ can be identified with the orthogonal (self-adjoint) projection $p_{S}:H\to S$, then ${\mathcal G}(H)$ and all our Grassmannian are closed subsets in the Banach algebra with the standard operator norm. In other words, these are complete metric spaces with the distance $||P_{S}-P_{U}||$ (we refer [@Shubin] for some elementary topological properties of the Grassmannians).
Let us consider the Grassmannian ${\mathcal G}_{\infty}(H)$ which is partially ordered by the inclusion relation.
In the real case, Mackey’s result [@Mackey] states that every automorphism of the lattice ${\mathcal G}(H)$ is induced by an invertible bounded linear operator, and it was shown by author [@Pankov] that every order preserving bijective transformation of ${\mathcal G}_{\infty}(H)$ can be extended to an automorphism of the lattice. In the present paper complex versions of these results will be given.
Now suppose that our Hilbert space is complex. We say that $A:H\to H$ is a [*semi-linear operator*]{} if $$A(x+y)=A(x)+A(y)$$ for all $x,y\in H$ and there exists an automorphism $\sigma: \mathbb{C}\to \mathbb{C}$ such that $$A(ax)=\sigma(a)A(x)$$ for all $a\in \mathbb{C}$ and $x\in H$. If $A$ is a non-zero bounded semi-linear operator then the associated automorphism $\sigma: \mathbb{C}\to \mathbb{C}$ is continuous. Since $\sigma(q)=q$ for every $q\in \mathbb{Q}$ and $\sigma(i)=\pm i$, one of the following possibilities is realized:
1. $\sigma$ is identical and $A$ is linear,
2. $\sigma$ is the complex conjugate mapping and $A(ax)=\bar{a}A(x)$ for all $a\in \mathbb{C}$, $x\in H$.
Note that non-continuous automorphisms of $\mathbb{C}$ exist.
Every invertible bounded semi-linear operator induces a continuous order preserving bijective transformation of ${\mathcal G}_{\infty}(H)$. Conversely, the following statement will be proved.
Every continuous order preserving bijective transformation of ${\mathcal G}_{\infty}(H)$ is induced by an invertible bounded semi-linear operator.
In the real case [@Pankov], we do not require continuity.
Hilbert projective spaces
=========================
Let $H$ be, as in the previous section, a separable (real or complex) Hilbert space. Consider the associated [*Hilbert projective space*]{} $\Pi_{H}$ whose points are $1$-dimensional linear subspaces and whose lines are defined by $2$-dimensional linear subspaces of $H$. It is clear that lines are closed and all closed subspaces of $\Pi_{H}$ are induced by closed linear subspaces of $H$. The projective space whose points are closed hyperplanes and whose lines are defined by closed linear subspaces of codimension $2$ will be called the [*dual Hilbert projective space*]{} and denoted by $\Pi^{*}_{H}$. The mapping $S\to S^{\perp}$ ($S^{\perp}$ is the orthogonal complement) induces an isometry between ${\mathcal G}_{k}(H)$ and ${\mathcal G}^{k}(H)$. By this mapping, $\Pi_{H}$ and $\Pi^{*}_{H}$ are canonically isomorphic as metric projective spaces.
It was proved in [@Mackey] that every collineation of $\Pi_{H}$ preserving the class of closed hyperplanes (“closed” collineation) is induced by an invertible bounded linear operator if $H$ is real. If our Hilbert space is complex then the following “weak” version of Mackey’s result is true.
Every continuous collineation of $\Pi_{H}$ is induced by an invertible bounded semi-linear operator.
To prove Theorem 2 we use the following lemma.
Let $A:H\to H$ be an invertible semi-linear operator such that the associated automorphism of $\mathbb{C}$ is identical or the complex conjugate mapping. If $A$ preserves ${\mathcal G}^{1}(H)$ then it is bounded.
Similar to the proof of Lemma B in [@Mackey].
Let $f:{\mathcal G}_{1}(H)\to {\mathcal G}_{1}(H)$ be a continuous collineation of $\Pi_{H}$. By the Fundamental Theorem of Projective Geometry [@Baer], there exists an invertible semi-linear operator $A:H\to H$ such that $$f(P)=A(P)$$ for all $P\in {\mathcal G}_{1}(H)$. For the associated automorphism $\sigma:{\mathbb C}\to {\mathbb C}$ one of the following possibilities is realized:
1. the restriction of $\sigma$ to ${\mathbb Q} + {\mathbb Q}i$ is identical,
2. $\sigma$ transfers $p+qi$ to $p-qi$ for each $p,q\in {\mathbb Q}$.
Suppose that $x,y\in H$ are linearly independent and $g$ is the continuous collineation of $\Pi_{H}$ induced by an invertible bounded semi-linear operator transferring $A(x)$ and $A(y)$ to $x$ and $y$, respectively; we require that the associated automorphism $\varphi$ is identical or the complex conjugate mapping in the case (1) or (2), respectively; in other words, the restriction of $\varphi\sigma$ to ${\mathbb Q} + {\mathbb Q}i$ is identical. Then $gf$ preserves the line of $\Pi_{H}$ joining $\langle x \rangle$ and $\langle y \rangle$, and it transfers $\langle x+ay \rangle$ to $\langle x+\varphi\sigma(a)y \rangle$. Since $$\langle x+by \rangle,\;\;b\in{\mathbb Q} + {\mathbb Q}i$$ form an everywhere dense subset in the line, the restriction of $gf$ to the line is identical. Hence $\sigma=\varphi$. We apply Lemma 1 and finish the proof.
[It follows from Theorem 2 that every continuous automorphism of the lattice ${\mathcal G}(H)$ is induced by an invertible bounded semi-linear operator. ]{}
Proof of Theorem 1
==================
We will use the following trivial remark concerning the automorphisms of ${\mathcal G}(H)$ induced by invertible bounded semi-linear operators. If $A:H\to H$ is an invertible bounded semi-linear operator then the transformation $$S\to (A(S^{\perp}))^{\perp}$$ is induced by the operator $(A^{*})^{-1}$, where $A^{*}$ is the adjoint operator. Recall that $A^{*}:H\to H$ is defined by the formula $$(A^{*}x, y)=(x,Ay)\;\;\;\;\;\forall\;x,y\in H$$ or $$(A^{*}x, y)=\overline{(x,Ay)}\;\;\;\;\;\forall\;x,y\in H$$ if the associated automorphism of $\mathbb{C}$ is identical or the complex conjugate mapping, respectively.
Let $f:{\mathcal G}_{\infty}(H)\to{\mathcal G}_{\infty}(H)$ be a continuous order preserving bijection.
Let $S\in {\mathcal G}_{\infty}(H)$ and ${\mathcal X}$ be the set of all elements of ${\mathcal G}_{\infty}(H)$ contained in $S$. There exists an invertible bounded semi-linear operator $A:S\to f(S)$ such that $$f(U)=A(U)\;\;\;\;\;\forall\;U\in {\mathcal X}.$$
We suppose that $f(S)=S$, since in the general case we can take an invertible bounded linear operator $C:H\to H$ which sends $f(S)$ to $S$ and consider the transformation $U\to C(f(U))$. The restriction of $f$ to ${\mathcal G}^{1}(S)$ is a continuous collineation of $\Pi^{*}_{S}$. Then $$P\to (f(P^{\perp}\cap S))^{\perp}\cap S\;\;\;\;\;\forall\;P\in {\mathcal G}_{1}(S)$$ is a continuous collineation of $\Pi_{S}$. By Theorem 2, it is induced by an invertible bounded semi-linear operator $B:S\to S$. Hence the restriction of $f$ to ${\mathcal G}^{1}(S)$ is induced by the operator $$A:=(B^{*})^{-1}.$$ Let $[U]^{1}$ be the set of all elements of ${\mathcal G}^{1}(S)$ containing $U\in {\mathcal X}$. Since $f$ is order preserving, $$f([U]^{1})=[f(U)]^{1};$$ on the other hand, $$f([U]^{1})=A([U]^{1})=[A(U)]^{1}.$$ Thus $f(U)=A(U)$ and $A$ is as required.
[In the real case, an order preserving transformation is not assumed to be continuous and the proof of the analogous lemma (Lemma 3 in [@Pankov]) is more complicated. ]{}
Let $S$ and $U$ be elements of ${\mathcal G}_{\infty}(H)$ such that $S\dotplus U$ [(]{}the minimal closed linear subspace containing $S+U$[[)]{}]{} belongs to ${\mathcal G}_{\infty}(H)$. Then $$\dim(S\cap U)=\dim(f(S)\cap f(U)).$$
Lemma 4 in [@Pankov].
If $S$ and $U$ are closed linear subspaces of finite dimension and codimension (respectively) then we write $[S]$ and $[U]$ for the sets of all elements of ${\mathcal G}_{\infty}(H)$ containing $S$ and contained in $U$, respectively. Using Lemma 3, we establish the following.
For every $S\in {\mathcal G}_{k}(H)$ and $U\in {\mathcal G}^{k}(H)$ there exist $S'\in {\mathcal G}_{k}(H)$ and $U'\in {\mathcal G}^{k}(H)$ such that $$f([S])=[S']\;\mbox{ and }\;f([U])=[U'].$$
Lemmas 5 and 6 in [@Pankov].
Therefore $f$ can be extended to an automorphisms of ${\mathcal G}(H)$. By the Fundamental Theorem of Projective Geometry, it is induced by an invertible semi-linear operator. Lemma 2 guarantees that the associated automorphism of ${\mathbb C}$ is identical or the complex conjugate mapping. Since our operator preserves ${\mathcal G}^{1}(H)$, Lemma 1 gives the claim.
Remark
======
Suppose that $A:H\to H$ is an invertible bounded semi-linear operator such that $A$ and $A^{*}$ give the same transformation of ${\mathcal G}_{\infty}(H)$ which will be denoted by $f$. In other words, $A^{*}=aA$ for a certain scalar $a$. If $A$ is linear then $$A= A^{**}=(aA)^{*}=\bar{a}A^{*}=|a|^{2}A$$ ($a=\pm 1$ in the real case) in the real and the complex case, respectively; thus $|a|=1$. If $A$ is over the complex conjugate mapping then $(aA)^{*}=aA^{*}$ and we get $a=\pm 1$. It is easy to see that $$\label{eq-1}
\forall\;S,U\in {\mathcal G}_{\infty}(H)\;\;\;\;\;S\perp f(U)\;\Longleftrightarrow\; f(S)\perp U.$$ Now we prove the following.
If a bijective transformation $f$ of ${\mathcal G}_{\infty}(H)$ satisfies then it is order preserving.
Since a linear subspace is contained in $S\in {\mathcal G}_{\infty}(H)$ if and only it is orthogonal to $S^{\perp}$, the condition guarantees that $f$ sends the set of all elements of ${\mathcal G}_{\infty}(H)$ contained in $S$ to the set of all elements of ${\mathcal G}_{\infty}(H)$ contained in the subspace $$f'(S):=f^{-1}(S^{\perp})^{\perp}.$$ This means that the bijection $f'$ is order preserving. Then $f^{-1}$ is order preserving and we get the claim.
Let $f$ be as in Lemma 5. If $H$ is real then it follows from the main result of [@Pankov] that $f$ is induced by an invertible bounded linear operator $A:H\to H$. In the complex case, Theorem 1 implies that our transformation $f$ is induced by an invertible bounded semilinear operator $A:H\to H$ if it is continuous. In each of these cases, shows that $$A(T)=A^{-1}(T^{\perp})^{\perp}$$ for every $T\in {\mathcal G}_{\infty}(H)$; hence $A$ and $A^{*}$ gives the same transformation of ${\mathcal G}_{\infty}(H)$.
[99]{}
Baer R., [*Projective Geometry and Linear Algebra*]{}, Academic Press, New York 1952.
Mackey G. W., [*Isomorphisms of normed linear spaces*]{}, Ann. of Math. 43(1942), 244–260.
Pankov M., [*Order preserving transformations of the Hilbert grassmannian*]{}, accepted to Arch. Math.
Shubin M. A., [*Remarks on topology of the Hilbert grassmannian*]{}, Contemporary mathematical physics, 191–198, Amer. Math. Soc. Transl. Ser. 2, 175, Amer. Math. Soc., Providence, RI, 1996.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'In this paper, we analyze a virtual element method (VEM) for solving a non-selfadjoint fourth-order eigenvalue problem derived from the transmission eigenvalue problem. We write a variational formulation and propose a $C^1$-conforming discretization by means of the VEM. We use the classical approximation theory for compact non-selfadjoint operators to obtain optimal order error estimates for the eigenfunctions and a double order for the eigenvalues. Finally, we present some numerical experiments illustrating the behavior of the virtual scheme on different families of meshes.'
address:
- 'GIMNAP, Departamento de Matemática, Universidad del Bío-Bío, Casilla 5-C, Concepción, Chile.'
- 'Centro de Investigación en Ingeniería Matemática (CI$^2$MA), Universidad de Concepción, Concepción, Chile.'
- 'Departamento de Ingeniería Matemática, Universidad de Concepción, Concepción, Chile.'
author:
- David Mora
- Iván Velásquez
title: A Virtual Element Method for the Transmission Eigenvalue Problem
---
Virtual element method ,transmission eigenvalue ,spectral problem ,error estimates. 65N25 ,65N30 ,65N21 ,78A46
Introduction {#SEC:INTR}
============
In this work, we study a Virtual Element Method for an eigenvalue problem arising in scattering theory. The [*Virtual Element Method*]{} (VEM), introduced in [@BBCMMR2013; @BBMR2014], is a generalization of the Finite Element Method which is characterized by the capability of dealing with very general polygonal/polyhedral meshes, and it also permits to easily implement highly regular discrete spaces. Indeed, by avoiding the explicit construction of the local basis functions, the VEM can easily handle general polygons/polyhedrons without complex integrations on the element (see [@BBMR2014] for details on the coding aspects of the method). The VEM has been developed and analyzed for many problems, see for instance [@ABMV2014; @ABSV2016; @BBM; @BLM2015; @BLV-M2AN; @BMR2016; @BBBPS2016-0; @BGS17; @BM12; @CG2017; @CGS17; @CMS2016; @ChM-camwa; @FS-M2AN18; @PPR15; @V-m3as18]. Regarding VEM for spectral problems, we mention [@BMRR; @GMV-Arxiv2018; @GV-IMA2017; @MR2017; @MRR2015; @MRV]. We note that there are other methods that can make use of arbitrarily shaped polygonal/polyhedral meshes, we cite as a minimal sample of them [@BLMbook2014; @CGH14; @DPECMAME2015; @TPPM10].
Due to their important role in many application areas, there has been a growing interest in recent years towards developing numerical schemes for spectral problems (see [@Boffi]). In particular, we are going to analyze a virtual element approximation of the transmission eigenvalue problem. The motivation for considering this problem is that it plays an important role in inverse scattering theory [@CakColMonkSun2010; @ColtonKress2013]. This is due to the fact that transmission eigenvalues can be determined from the far-field data of the scattered wave and used to obtain estimates for the material properties of the scattering object [@CakCayCol2008; @CakGinHad2010].
In recent years various numerical methods have been proposed to solve this eigenvalue problem; see for example the following references [@CakMonkSun2014; @CRV2018; @ChenGZZ2016; @ColtonMonkSun2010; @GengJiSunXu2016; @HYBi2017NonConf; @HYBi2017; @Sun2011]. In particular, the transmission eigenvalue problem is often solved by reformulating it as a fourth-order eigenvalue problem. In [@CakMonkSun2014], a $C^1$ finite element method using Argyris elements has been proposed, a complete analysis of the method including error estimates are proved using the theory for compact non-self-adjoint operators. However, the construction of conforming finite elements for $H^2(\O)$ is difficult in general, since they usually involve a large number of degrees of freedom (see [@ciarlet]). More recently, in [@GengJiSunXu2016] a discontinuous Galerkin method has been proposed and analyzed to solve the fourth-order transmission eigenvalue problem; moreover, in [@ChenGZZ2016] a $C^0$ linear finite element method has been introduced to solve the spectral problem.
The purpose of the present paper is to introduce and analyze a $C^1$-VEM for solving a fourth-order spectral problem derived from the transmission eigenvalue problem. We consider a variational formulation of the problem written in $H^{2}(\O)\times H^1(\O)$ as in [@CakMonkSun2014; @GengJiSunXu2016], where an auxiliary variable is introduced to transform the problem into a linear eigenvalue problem. Here, we exploit the capability of VEM to build highly regular discrete spaces (see [@BM13; @BM12]) and propose a conforming $H^{2}(\O)\times H^1(\O)$ discrete formulation, which makes use of a very simple set of degrees of freedom, namely 4 degrees of freedom per vertex of the mesh. Then, we use the classical spectral theory for non-selfadjoint compact operators (see [@BO; @Osborn1975]) to deal with the continuous and discrete solution operators, which appear as the solution of the continuous and discrete source problems, and whose spectra are related with the solutions of the transmission eigenvalue problem. Under rather mild assumptions on the polygonal meshes (made by possibly non-convex elements), we establish that the resulting VEM scheme provides a correct approximation of the spectrum and prove optimal-order error estimates for the eigenfunctions and a double order for the eigenvalues. Finally, we note that, differently from the FEM where building globally conforming $H^2(\O)$ approximation is complicated, here the virtual space can be built with a rather simple construction due to the flexibility of the VEM. In a summary, the advantages of the present virtual element discretization are the possibility to use general polygonal meshes and to build conforming $H^2(\O)$ approximations.
The remainder of this paper is structured as follows: In Section \[SEC:SpProCont\], we introduce the variational formulation of the transmission eigenvalue problem, define a solution operator and establish its spectral characterization. In Section \[SEC:DISCR\_PROB\], we introduce the virtual element discrete formulation, describe the spectrum of a discrete solution operator and establish some auxiliary results. In Section \[SEC:approximation\], we prove that the numerical scheme provides a correct spectral approximation and establish optimal order error estimates for the eigenvalues and eigenfunctions using the standard theory for compact and non-selfadjoint operators. Finally, we report some numerical tests that confirm the theoretical analysis developed in Section \[SEC:NumRes\].
In this article, we will employ standard notations for Sobolev spaces, norms and seminorms. In addition, we will denote by $C$ a generic constant independent of the mesh parameter $h$, which may take different values in different occurrences.
The transmission eigenvalue problem {#SEC:SpProCont}
===================================
Let $\O\subset\R^2$ be a bounded domain with polygonal boundary $\partial \O$. We denote by $\nu$ the outward unit normal vector to $\partial \O$ and by $\partial_\nu$ the normal derivative. Let $n$ be a real value function in $L^{\infty}(\O)$ such that $n-1$ is strictly positive (or strictly negative) almost everywhere in $\O$. The transmission eigenvalue problem reads as follows:
Find the so-called transmission eigenvalue $k\in \mathbb{C}$ and a non-trivial pair of functions $(w_1,w_2)\in L^2(\O)\times L^2(\O)$, such that $(w_1-w_2)\in H^2(\O)$ satisfying $$\begin{aligned}
\Delta w_1+k^2n(x)w_1=0&\quad \mbox{in }\O,\label{eq1}\\
\Delta w_2+k^2w_2=0&\quad \mbox{in }\O,\label{eq2}\\
w_1=w_2&\quad \mbox{on }\partial \O,\label{eq3}\\
\partial_\nu w_1=\partial_\nu w_2&\quad \mbox{on }\partial \O.\label{eq4}\end{aligned}$$ Now, we rewrite problem above in the following equivalent form for $u:=(w_1-w_2)\in H_0^2(\O)$ (see [@CakMonkSun2014]):
Find $(k,u)\in\mathbb{C}\times H_0^2(\O)$ such that $$\begin{aligned}
\label{eqfourt-order}
(\Delta +k^2n)\frac{1}{n-1}(\Delta+k^2)u=0 &\quad \mbox{in }\O.\end{aligned}$$
The variational formulation of problem can be stated as: Find $(k,u)\in\mathbb{C}\times H_0^2(\O)$, $u\ne0$ such that $$\begin{aligned}
\label{eqfourt-order-weak}
\int_{\O} \frac{1}{n-1}(\Delta u +k^2 u)(\Delta \overline{v}+k^2n\overline{v})=0& \quad \forall v\in H_0^2(\O),\end{aligned}$$ where $\overline{v}$ denotes the complex conjugate of $v$. Now, expanding the previous expression we obtain the following quadratic eigenvalue problem: $$\begin{aligned}
\label{eqfourt-ord-expand}
\int_\O \frac{1}{n-1}\Delta u\Delta \overline{v}+\tau\int_\O \frac{1}{n-1} u\Delta\overline{v}
+\tau\int_\O \frac{1}{n-1} \Delta u \overline{nv}+\tau^2 \int_\O \frac{1}{n-1} u\overline{nv}=0
\quad \forall v\in H_0^2(\O),\end{aligned}$$ where $\tau:=k^2$. It is easy to show that $k = 0$ is not an eigenvalue of the problem (see [@CakMonkSun2014]). Moreover, for the sake of simplicity, we will assume that the index of refraction function $n(x)$ as a real constant. Nevertheless, this assumption do not affect the generality of the forthcoming analysis.
For the theoretical analysis it is convenient to transform problem into a linear eigenvalue problem. With this aim, let $\phi$ be the solution of the problem: Find $\phi\in \Hcu$ such that $$\begin{aligned}
&\Delta \phi=\tau \frac{n}{n-1}u \quad &\mbox{in }&\O,\label{prob-aux1}\\
&\phi=0 \quad &\mbox{on }&\partial\Omega.\label{prob-aux2}\end{aligned}$$
Therefore, by testing problem - with functions in $H_0^1(\O)$, we arrive at the following weak formulation of the problem:
\[P1\] Find $(\lambda,u,\phi)\in \mathbb{C}\times H_0^2(\O)\times H_0^1(\O)$ with $(u,\phi)\neq 0$ such that $$\begin{aligned}
a((u,\phi),(v,\psi))=\lambda b((u,\phi),(v,\psi))\quad \forall (v,\psi)\in \HdoO\times\Hcu,\end{aligned}$$
where $\lambda=-\tau$ and the sesquilinear forms $a(\cdot,\cdot)$ and $b(\cdot,\cdot)$ are defined by $$\begin{aligned}
a((u,\phi),(v,\psi)):=&\frac{1}{n-1}\int_\O D^2 u:D^2 \overline{v}
+\int_\O \nabla \phi \cdot \nabla \overline{\psi},\nonumber\\[2ex]
b((u,\phi),(v,\psi)):=&\frac{n}{n-1} \int_\O \Delta u \overline{v}+\frac{1}{n-1}\int_\O u\Delta\overline{v}
-\int_\O \nabla \phi \cdot\nabla \overline{v}+ \frac{n}{n-1}\int_\O u\overline{\psi},\nonumber\end{aligned}$$ for all $(u,\phi),(v,\psi)\in\HdoO\times\Hcu$. Moreover, $":"$ denotes the usual scalar product of $2\times2$-matrices, $D^2 u:=(\partial_{ij}u)_{1\le i,j\le2}$ denotes the Hessian matrix of $u$.
We endow $\HdoO\times\Hcu$ with the corresponding product norm, which we will simply denote $\Vert(\cdot,\cdot)\Vert$.
Now, we note that the sesquilinear forms $a(\cdot,\cdot)$ and $b(\cdot,\cdot)$ are bounded forms. Moreover, we have that $a(\cdot,\cdot)$ is elliptic.
\[ha-elipt\] There exists a constant $\alpha_{0}>0$, depending on $\O$, such that $$a((v,\psi),(v,\psi))
\ge\alpha_{0}\left\|(v,\psi)\right\|^2
\qquad\forall (v,\psi)\in \HdoO\times\Hcu.$$
The result follows immediately from the fact that $\{\Vert D^2 v\Vert_{0,\O}^2 + \Vert \nabla\psi\Vert_{0,\O}^2\}^{1/2}$ is a norm on $\HdoO\times \Hcu$, equivalent with the usual norm.
We define the solution operator associated with Problem \[P1\]: $$\begin{array}{clll}
T: \HdoO\times \Hcu & \longrightarrow& \HdoO\times\Hcu\\
(f,g)&\longmapsto &T(f,g)=(\tilde{u},\tilde{\phi})
\end{array}$$ as the unique solution (as a consequence of Lemma \[ha-elipt\]) of the corresponding source problem: $$\label{source}
a((\tilde{u},\tilde{\phi}),(v,\psi))=b((f,g),(v,\psi))\quad \forall (v,\psi)\in \HdoO\times\Hcu.$$
The linear operator $T$ is then well defined and bounded. Notice that $(\lambda,u,\phi)\in\mathbb{C}\times\HdoO\times\Hcu$ solves Problem \[P1\] if and only if $(\mu,u,\phi)$, with $\mu:=\frac1{\l}$, is an eigenpair of $T$, i.e., $T(u,\phi)=\mu (u,\phi)$.
We observe that no spurious eigenvalues are introduced into the problem since if $\mu\ne0$, $(0,\phi)$ is not an eigenfunction of the problem.
The following is an additional regularity result for the solution of the source problem and consequently, for the generalized eigenfunctions of $T$.
\[lem\_regul\] There exist $s,t\in(1/2,1]$ and $C>0$ such that, for all $(f,g)\in \HdoO\times\Hcu$, the solution $(\tilde{u},\tilde{\phi})$ of problem satisfies $\tilde{u}\in H^{2+s}(\Omega)$, $\tilde{\phi}\in H^{1+t}(\Omega)$, and $$\Vert\tilde{u}\Vert_{2+s,\Omega}+\Vert \tilde{\phi}\Vert_{1+t,\Omega}\le C\Vert(f,g)\Vert.$$
The estimate for $\tilde{\phi}$ follows from the classical regularity result for the Laplace problem with its right-hand side in $\LO$. The estimate for $\tilde{u}$ follows from the classical regularity result for the biharmonic problem with its right-hand side in $H^{-1}(\Omega)$ (cf. [@G]).
\[rem\_regul\] The constant $s$ in the lemma above is the Sobolev regularity for the biharmonic equation with the right-hand side in $H^{-1}(\Omega)$ and homogeneous Dirichlet boundary conditions. The constant $t$ is the Sobolev exponent for the Laplace problem with homogeneous Dirichlet boundary conditions. These constants only depend on the domain $\Omega$. If $\Omega$ is convex, then $s=t=1$. Otherwise, the lemma holds for all $s < s_0$ and $t < t_0$, where $s_0,t_0\in(1/2,1]$ depend on the largest reentrant angle of $\Omega$.
Hence, because of the compact inclusions $ H^{2+s}(\O) \hookrightarrow \HdoO$ and $H^{1+t}(\O)\hookrightarrow \Hcu$, we can conclude that $T$ is a compact operator. So, we obtain the following spectral characterization result.
The spectrum of $T$ satisfies $\sp(T)=\{0\}\cup\{\mu_k\}_{k\in \N}$, where $\{\mu_k\}_{k\in \N}$ is a sequence of complex eigenvalues which converges to 0 and their corresponding eigenspaces lie in $H^{2+s}(\O)\times H^{1+t}(\O)$. In addition $\mu=0$ is an infinite multiplicity eigenvalue of $T$.
The proof is obtained from the compactness of $T$ and Lemma \[lem\_regul\].
The virtual element discretization {#SEC:DISCR_PROB}
==================================
In this section, we will write the $C^1$-VEM discretization of Problem \[P1\]. With this aim, we start with the mesh construction and the assumptions considered to introduce the discrete virtual element spaces.
Let $\left\{\CT_h\right\}_h$ be a sequence of decompositions of $\O$ into polygons $\E$ we will denote by $h_\E$ the diameter of the element $\E$ and $h$ the maximum of the diameters of all the elements of the mesh, i.e., $h:=\max_{\E\in\CT_h}h_\E$. In what follows, we denote by $N_\E$ the number of vertices of $\E$, by $e$ a generic edge of $\left\{\CT_h\right\}_h$ and for all $e\in\partial\E$, we define a unit normal vector $\nu_\E^e$ that points outside of $\E$.
In addition, we will make the following assumptions as in [@BBCMMR2013; @BMRR]: there exists a positive real number $C_{\CT}$ such that, for every $h$ and every $\E\in \CT_h$,
- the ratio between the shortest edge and the diameter $h_\E$ of $\E$ is larger than $C_{\CT}$;
- $\E\in\CT_h$ is star-shaped with respect to every point of a ball of radius $C_{\CT}h_\E$.
In order to introduce the method, we first define two preliminary discrete spaces as follows: For each polygon $\E\in \mathcal{T}_h$ (meaning open simply connected set whose boundary is a non-intersecting line made of a finite number of straight line segments) we define the following finite dimensional spaces, $$\begin{aligned}
\widetilde{W}_h^K
:=\left\{v_h\in \HdoK : \Delta^2v_h\in\P_{2}(\E), v_h|_{\partial\E}\in C^0(\partial\E),
v_h|_e\in\P_3(e)\,\,\forall e\in\partial\E,\right.\\
\left.\nabla v_h|_{\partial\E}\in C^0(\partial\E)^2,
\partial_\nu v_h|_e\in\P_1(e)\,\,\forall e\in\partial\E\right\},\end{aligned}$$ and $$\begin{aligned}
\widetilde{V}_h^K:=\{\psi_h\in \HuE :\Delta \psi_h \in \mathbb{P}_1(K),
\psi_h|_{\partial K}\in C^0(\partial K), \psi_h|_{e}\in \mathbb{P}_1(e) \ \forall e \in \partial K \},\end{aligned}$$ where $\Delta^2$ represents the biharmonic operator and we have denoted by $\mathbb{P}_k(S)$ the space of polynomials of degree up to $k$ defined on the subset $S\subseteq\R^2$.
The following conditions hold:
- for any $v_h\in\widetilde{W}_h^\E$ the trace on the boundary of $\E$ is continuous and on each edge is a polynomial of degree 3;
- for any $v_h\in\widetilde{W}_h^\E$ the gradient on the boundary is continuous and on each edge its normal (respectively tangential) component is a polynomial of degree 1 (respectively 2);
- for any $\psi_h\in\widetilde{V}_h^\E$ the trace on the boundary of $\E$ is continuous and on each edge is a polynomial of degree 1;
- $\mathbb{P}_2(\E)\times \mathbb{P}_1(\E)\subseteq \widetilde{W}_h^\E\times \widetilde{V}_h^\E$.
Next, with the aim to choose the degrees of freedom for both spaces, we will introduce three sets ${\bf D_1}$, ${\bf D_2}$ and ${\bf D_3}$. The first two sets (${\bf D_1}, {\bf D_2}$) are provided by linear operators from $\widetilde{W}_h^K$ into $\R$ and the set ${\bf D_3}$ by linear operators from $\widetilde{V}_h^K$ into $\R$. For all $(v_{h},\psi_h)\in\widetilde{W}_h^K\times \widetilde{V}_h^K$ they are defined as follows:
- ${\bf D_1}$ contains linear operators evaluating $v_{h}$ at the $N_{\E}$ vertices of $\E$,
- ${\bf D_2}$ contains linear operators evaluating $\nabla v_h$ at the $N_{\E}$ vertices of $\E$,
- ${\bf D_3}$ contains linear operators evaluating $\psi_{h}$ at the $N_{\E}$ vertices of $\E$.
Note that, as a consequence of definition of the discrete spaces, the output values of the three sets of operators ${\bf D_1}$, ${\bf D_2}$ and ${\bf D_3}$, are sufficient to uniquely determine $v_h$ and $\nabla v_h$ on the boundary of $\E$, and $\psi_h$ on the boundary of $\E$, respectively.
In order to construct the discrete scheme, we need some preliminary definitions. First, we split the forms $a(\cdot,\cdot)$ and $b(\cdot,\cdot)$, introduced in the previous section, as follows:
$$a((u,\phi),(v,\psi))=\sum_{\E\in\CT_h}a^{\Delta}_{\E}(u,v)+a^{\nabla}_{\E}(\phi,\psi),
\qquad (u,\phi),(v,\phi)\in\HdoO\times\Hcu,$$ $$b((u,\phi),(v,\psi))=\sum_{\E\in\CT_h}b_{\E}((u,\phi),(v,\psi)),
\qquad (u,\phi),(v,\phi)\in\HdoO\times\Hcu,$$ with $$\label{alocal}
a^{\Delta}_{\E}(u,v)
:=\int_{\E}D^2 u:\,D^2 \overline{v},
\qquad u,v\in\HdoK,$$ $$\label{alocalnabla}
a^{\nabla}_{\E}(\phi,\psi)
:=\int_{\E}\nabla \phi\cdot\nabla\overline{\psi},
\qquad \phi,\psi\in H^1(K),$$ and for all $(u,\phi),(v,\phi)\in H^2(K)\times H^1(K)$, $$\label{blocal}
b_{\E}((u,\phi),(v,\psi))
:=\frac{n}{n-1} \int_\E \Delta u \overline{v}+\frac{1}{n-1}\int_\E u\Delta\overline{v}
-\int_\E \nabla \phi \cdot\nabla \overline{v}+ \frac{n}{n-1}\int_\E u\overline{\psi}.$$ Now, we define the projector $\Pi_{2}^{\Delta}:\ H^2(\E)\longrightarrow\P_2(\E)\subseteq\widetilde{W}_h^K$ for each $v\in H^2(\E)$ as the solution of
$$\begin{aligned}
a^\Delta_{\E}\big(\Pi_{2}^{\Delta} v,q\big)
& =a^\Delta_{\E}(v,q)
\qquad\forall q\in\P_2(\E),
\label{numero}
\\
((\Pi_{2}^{\Delta} v,q))_{\E}
&=((v,q))_{\E} \qquad\forall q\in\P_1(\E),
\label{numeroo}\end{aligned}$$
where $((\cdot,\cdot))_{\E}$ is defined as follows: $$((u,v))_{\E}=\sum_{i=1}^{N_{\E}}u(P_i)v(P_i)\qquad\forall u,v\in C^0(\partial\E),$$ where $P_i, 1\le i\le N_{\E}$, are the vertices of $\E$. We note that the bilinear form $a^\Delta_{\E}(\cdot,\cdot)$ has a non-trivial kernel, given by $\P_1(\E)$. Hence, the role of condition is to select an element of the kernel of the operator. We observe that operator $\Pi_{2}^{\Delta}$ is well defined on $\widetilde{W}_h^K$ and, most important, for all $v\in\widetilde{W}_h^K$ the polynomial $\Pi_{2}^{\Delta} v$ can be computed using only the values of the operators ${\bf D_1}$ and ${\bf D_2}$ calculated on $v$. This follows easily with an integration by parts (see [@ABSV2016]).
In a similar way, we define the projector $\Pi_1^{\nabla}:\ H^1(\E)\longrightarrow\P_1(\E)\subseteq\widetilde{V}_h^K$ for each $\psi\in H^1(\E)$ as the solution of
$$\begin{aligned}
a^\nabla_{\E}\big(\Pi^{\nabla}_1 \psi,q\big)
& =a^\nabla_{\E}(\psi,q)
\qquad\forall q\in\P_1(\E),
\label{numero1}
\\
(\Pi^{\nabla}_1\psi,1)_{\partial\E}
&=(\psi,1)_{\partial\E}.
\label{numeroo1}\end{aligned}$$
We observe that operator $\Pi^{\nabla}_1$ is well defined on $\widetilde{V}_h^K$ and, as before, for all $\psi\in\widetilde{V}_h^K$ the polynomial $\Pi^{\nabla}_1\psi$ can be computed using only the values of the operators ${\bf D_3}$ calculated on $\psi$, which follows by an integration by parts (see [@AABMR13]).
Now, we introduce our local virtual spaces: $$\begin{aligned}
\WK:=\left\{v_h\in\widetilde{W}_h^K : \int_{\E}(\Pi_2^{\Delta} v_h)q=\int_{\E}v_hq\qquad\forall q\in\P_{2}(\E)\right\},\end{aligned}$$ and $$\begin{aligned}
\VK:=\left\{ \psi_h \in \widetilde{V}_h^K: \int_K (\Pi_1^{\nabla}\psi_h)q=\int_K \psi_h q
\qquad\forall q\in \P_1(\E) \right\}.\end{aligned}$$
It is clear that $W_h^K\times V_h^K\subseteq \widetilde{W}_h^K\times \widetilde{V}_h^K$. Thus, the linear operators $\Pi_2^{\Delta}$ and $\Pi_1^{\nabla}$ are well defined on $W_h^K$ and $V_h^K$, respectively.
In [@ABSV2016 Lemma 2.1] has been established that the sets of operators ${\bf D_1}$ and ${\bf D_2}$ constitutes a set of degrees of freedom for the space $\WK$. Moreover, the set of operators ${\bf D_3}$ constitutes a set of degrees of freedom for the space $\VK$ (see [@AABMR13]).
We also have that $\mathbb{P}_2(K)\times\mathbb{P}_1(K)\subseteq\WK\times\VK$. This will guarantee the good approximation properties for the spaces.
To continue the construction of the discrete scheme, we will need to consider new projectors: First, we define the projector $\Pi_2^{\nabla}:\ H^2(\E)\longrightarrow\P_2(\E)$ for each $w\in H^2(\E)$ as the solution of
$$\begin{aligned}
a^\nabla_{\E}\big(\Pi^{\nabla}_2 w,q\big)
& =a^\nabla_{\E}(w,q)
\qquad\forall q\in\P_2(\E),
\label{numero2}
\\
(\Pi^{\nabla}_2 w,1)_{0,\E}
&=(w,1)_{0,\E}.
\label{numeroo2}\end{aligned}$$
Moreover, we consider the $\LO$ orthogonal projectors onto $\P_{l}(\E)$, $l=1,2$ as follows: we define $\Pi_{l}^{0}:\LO\to\P_{l}(\E)$ for each $p \in \LO$ by $$\label{fff}
\int_{\E}(\Pi_{l}^{0} p)q=\int_{\E}pq\qquad\forall q\in\P_{l}(\E).$$
Now, due to the particular property appearing in definition of the space $\WK$, it can be seen that the right hand side in is computable using $\Pi_{2}^{\Delta} v_h$, and thus $\Pi_{2}^{0} v_h$ depends only on the values of the degrees of freedom for $v_h$ and $\nabla v_h$. Actually, it is easy to check that on the space $\WK$ the projectors $\Pi_{2}^{0}v_h$ and $\Pi_{2}^{\Delta} v_h$ are the same operator. In fact: $$\label{L2proj}
\int_{\E}(\Pi_{2}^{0} v_h)q=\int_{\E}v_hq=
\int_{\E}(\Pi_{2}^{\Delta} v_h)q\qquad\forall q\in\P_{2}(\E).$$ Repeating the arguments, it can be proved that $\Pi_{1}^{0}\phi_h$ and $\Pi_{1}^{\nabla}\phi_h$ are the same operator in $\VK$.
Now, for every decomposition $\mathcal{T}_h$ of $\Omega$ into simple polygons $K$, we introduce our the global virtual space denoted by $Z_h$ as follow: $$Z_h:=W_h\times V_h,$$ where $$\begin{aligned}
W_h:=\{v_h\in \Hcd: v_h|_{K}\in \WK\} \quad \mbox{and}\quad
V_h:= \{\psi_h\in \Hcu: \psi_h|_{K}\in \VK\}.\end{aligned}$$
A set of degrees of freedom for $Z_h$ is given by all pointwise values of $v_h$ and $\psi_h$ on all vertices of $\mathcal{T}_h$ together with all pointwise values of $\nabla v_h$ on all vertices of $\mathcal{T}_h$, excluding the vertices on $\partial\Omega$ (where the values vanishes). Thus, the dimension of $Z_h$ is four times the number of interior vertices of $\mathcal{T}_h$.
In what follows, we discuss the construction of the discrete version of the local forms. With this aim, we consider $s_{K}^{\Delta}(\cdot,\cdot)$ and $s_{K}^{\nabla}(\cdot,\cdot)$ any symmetric positive definite forms satisfying: $$\begin{aligned}
c_0a_K^{\Delta}(v_h,v_h)\leq s_\E^{\Delta}(v_h,v_h)\leq c_1
a_K^{\Delta}(v_h,v_h)&\quad \forall v_h \in W_h^K\quad \mbox{with }\quad
\Pi_2^{\Delta} v_h =0 \label{term-stab-SK},\\[1ex]
c_2a_K^{\nabla}(\psi_h,\psi_h)\leq s_\E^{\nabla}(\psi_h,\psi_h)\leq
c_3 a_K^{\nabla}(\psi_h,\psi_h)&\quad \forall \psi_h \in V_h^K\quad
\mbox{with }\quad \Pi_1^{\nabla} \psi_h =0. \label{term-stab-SK0}\end{aligned}$$
We define the discrete sesquilinear forms $a_h(\cdot,\cdot):Z_h\times Z_h\to \mathbb{C} $ and $b_h(\cdot,\cdot):Z_h\times Z_h\to \mathbb{C}$ by $$\begin{aligned}
a_h((u_h,\phi_h),(v_h,\psi_h))&
:=\sum_{\E\in\CT_h} a_{h,\E}^{\Delta}(u_h,v_h)+a_{h,\E}^{\nabla}(\phi_h,\psi_h)
\qquad \forall (u_h,\phi_h),(v_h,\psi_h)\in Z_h,\\
b_h((u_h,\phi_h),(v_h,\psi_h))&
:=\sum_{\E\in\CT_h}b_{h,\E}((u_h,\phi_h),(v_h,\psi_h))
\qquad \forall (u_h,\phi_h),(v_h,\psi_h)\in Z_h,\end{aligned}$$ where $a_{h,\E}^{\Delta}(\cdot,\cdot)$, $a_{h,\E}^{\nabla}(\cdot,\cdot)$ and $b_{h,\E}(\cdot,\cdot)$ are local forms on $\WK\times\WK$ and $\VK\times\VK$ defined by $$\begin{aligned}
a_{h,\E}^{\Delta}(u_h,v_h)
:=a_{\E}^{\Delta}\big(\Pi_2^{\Delta} u_h,\Pi_2^{\Delta} v_h\big)
+s_{\E}^{\Delta}\big(u_h-\Pi_2^{\Delta} u_h,v_h-\Pi_2^{\Delta} v_h\big),
\qquad \forall u_h,v_h\in\WK,\nonumber\\
a_{h,\E}^{\nabla}(\phi_h,\psi_h)
:=a_{\E}^{\nabla}\big(\Pi_1^{\nabla} \phi_h,\Pi_1^{\nabla} \psi_h\big)
+s_{\E}^{\nabla}\big(\phi_h-\Pi_1^{\nabla} \phi_h,\psi_h-\Pi_1^{\nabla} \psi_h\big),
\qquad \forall \phi_h, \psi_h\in V_h^K,\nonumber\end{aligned}$$ $$\begin{aligned}
b_{h,\E}((u_h,\phi_h),(v_h,\psi_h)):&= \frac{n}{n-1}\int_K \Pi_2^{0}(\Delta u_h) \Pi_2^{0} v_h
+\frac{1}{n-1}\int_K\Pi_2^{0} u_h \Pi_2^{0}(\Delta v_h)
- \int_K \nabla \Pi_1^{\nabla}\phi_h\cdot \nabla \Pi_2^{\nabla} v_h\\
&+ \frac{n}{n-1}\int_K \Pi_2^{0} u_h \Pi_1^{0}\psi_h
\qquad \forall (u_h,\phi_h),(v_h,\psi_h)\in\WK\times\VK.\end{aligned}$$
The construction of the local sesquilinear forms guarantees the usual consistency and stability properties, as is stated in the proposition below. Since the proof follows standard arguments in the VEM literature, it is omitted.
The local forms $a_{h,\E}^{\Delta}(\cdot,\cdot)$ and $a_{h,\E}^{\nabla}(\cdot,\cdot)$ on each element $\E$ satisfy
- *Consistency*: for all $h > 0$ and for all $\E\in\CT_h$ we have that $$\begin{aligned}
a_{h,\E}^{\Delta}(v_h,q)=a_{\E}^{\Delta}(v_h,q)\qquad\forall q \in\P_2(\E),
\quad\forall v_h\in \WK,\label{consis-a1}\\
a_{h,\E}^{\nabla}(\psi_h,q)=a_{\E}^{\nabla}(\psi_h,q)\qquad\forall q \in\P_1(\E),
\quad\forall \psi_h\in \VK.\label{consis-a2}\end{aligned}$$
- *Stability and boundedness*: There exist positive constants $\alpha_i, i=1,2,3,4,$ independent of $\E$, such that: $$\begin{aligned}
\alpha_1 a_{\E}^{\Delta}(v_h,v_h)\leq a_{h,\E}^{\Delta}(v_h,v_h)
\leq\alpha_2 a_{\E}^{\Delta}(v_h,v_h)\qquad\forall v_h\in\WK,\label{stab-a1}\\
\alpha_3 a_{\E}^{\nabla}(\psi_h,\psi_h)\leq a_{h,\E}^{\nabla}(\psi_h,\psi_h)
\leq\alpha_4 a_{\E}^{\nabla}(\psi_h,\psi_h)\qquad\forall \psi_h\in\VK.\label{stab-a2}\end{aligned}$$
Now, we are in a position to write the virtual element discretization of Problem \[P1\].
\[P2\] Find $(\lambda_h,u_h,\psi_h)\in\mathbb{C}\times Z_h$, $(u_h,\phi_h)\ne0$ such that $$\begin{aligned}
a_h((u_h,\phi_h),(v_h,\psi_h))=\lambda_hb_h((u_h,\phi_h),(v_h,\psi_h)).\label{probspecdiscr}\end{aligned}$$
It is clear that by virtue of and the sesquilinear form $a_{h}(\cdot,\cdot)$ is bounded. Moreover, we will show in the following lemma that $a_{h}(\cdot,\cdot)$ is also uniformly elliptic.
\[ha-elipt-disc\] There exists a constant $\beta>0$, independent of $h$, such that $$a_{h}((v_h,\psi_h),(v_h,\psi_h))
\ge\beta \Vert(v_h,\psi_h)\Vert^2\qquad\forall (v_h,\psi_h)\in Z_h.$$
The result is deduced from Lemma \[ha-elipt\], and .
Now, we introduce the discrete solution operator $T_h$ which is given by $$\begin{array}{cccll}
T_h:& \Hcd\times \Hcu & \longrightarrow & \Hcd\times \Hcu&\\
&(f,g)& \longmapsto & T_h(f,g)=(\tilde{u}_h,\tilde{\phi}_h)&
\end{array}$$ where $(\tilde{u}_h,\tilde{\phi}_h)\in Z_h$ is the unique solution of the corresponding discrete source problem $$\label{source-disc}
a_h((\tilde{u}_h,\tilde{\phi}_h),(v_h,\psi_h))=b_{h}((f,g),(v_h,\psi_h))
\qquad\forall (v_h,\psi_h)\in Z_h.$$
Because of Lemma \[ha-elipt-disc\], the linear operator $T_h$ is well defined and bounded uniformly with respect to $h$. Once more, as in the continuous case, $(\lambda_h,u_h,\phi_h)\in\mathbb{C}\times Z_h$ solves Problem \[P2\] if and only if $(\mu_h,u_h,\phi_h)$, with $\mu_h:=\frac1{\l_h}$, is an eigenpair of $T_h$, i.e., $T_h(u_h,\phi_h)=\mu_h (u_h,\phi_h)$.
Spectral approximation and error estimates {#SEC:approximation}
==========================================
To prove that $T_h$ provides a correct spectral approximation of $T$, we will resort to the classical theory for compact operators (see [@BO]). With this aim, we first recall the following approximation result which is derived by interpolation between Sobolev spaces (see for instance [@GR Theorem I.1.4] from the analogous result for integer values of $s$). In its turn, the result for integer values is stated in [@BBCMMR2013 Proposition 4.2] and follows from the classical Scott-Dupont theory (see [@BS-2008] and [@ABSV2016 Proposition 3.1]):
\[app1\] There exists a constant $C>0$, such that for every $v\in H^{\delta}(\E)$ there exists $v_{\pi}\in\P_k(\E)$, $k\geq 0$ such that $$\begin{aligned}
\vert v-v_{\pi}\vert_{\ell,\E}\leq C h_\E^{\delta-\ell}|v|_{\delta,\E}\quad 0\leq\delta\leq
k+1, \ell=0,\ldots,[\delta],\end{aligned}$$ with $[\delta]$ denoting largest integer equal or smaller than $\delta \in {\mathbb R}$.
For the analysis we will introduce some broken seminorms: $$|\psi|_{1,h}^{2}:=\sum_{\E\in\CT_h}|\psi|_{1,\E}^{2} \qquad \mbox{and } \quad
|v|_{2,h}^{2}:=\sum_{\E\in\CT_h}|v|_{2,\E}^{2},$$ which are well defined for every $(\psi,v)\in [L^{2}(\O)]^2$ such that $(\psi,v)|_{\E}\in H^{1}(\E)\times H^{2}(\E)$ for all polygon $\E\in \CT_{h}$.
In what follows, we derive several auxiliary results which will be used in the following to prove convergence and error estimates for the spectral approximation.
\[app0\] Assume [**A1**–**A2**]{} are satisfied, let $\psi\in\HutO$ with $t\in(0,1]$. Then, there exist $\psi_{I}\in\Vh$ and $C>0$ such that $$\Vert\psi-\psi_{I}\Vert_{1,\O}\le Ch^t\vert\psi\vert_{1+t,\O}.$$
This result has been proved in [@CGPS Theorem 11] (see also [@MRR2015 Proposition 4.2]).
\[app2\] Assume [**A1**–**A2**]{} are satisfied, let $v\in\HdsO$ with $s\in(0,1]$. Then, there exist $v_{I}\in\Wh$ and $C>0$ such that $$\Vert v-v_{I}\Vert_{2,\O}\le Ch^s\vert v\vert_{2+s,\O}.$$
This result has been establish in [@ABSV2016 Proposition 3.1].
Now, we establish a result which will be useful to prove the convergence of the operator $T_h$ to $T$ as $h$ goes to zero.
\[lemcotste\] There exists $C>0$ independent of $h$ such that for all $(f,g)\in \Hcd\times\Hcu$, if $(\tilde{u},\tilde{\phi}):=T(f,g)$ and $(\tilde{u}_h,\tilde{\phi}_h):=T_h(f,g)$, then $$\begin{aligned}
\left\|\left(T-T_h\right)(f,g)\right\|\leq C h \Vert(f,g)\Vert + \vert\tilde{u}-\tilde{u}_{I}\vert_{2,\O}
+\vert \tilde{u}-\tilde{u}_{\pi} \vert_{2,h} + \vert\tilde{\phi} - \tilde{\phi}_{I}\vert_{1,\O}
+\vert \tilde{\phi} - \tilde{\phi}_{\pi}\vert_{1,h},\end{aligned}$$ for all $(\tilde{u}_I,\tilde{\phi}_I)\in Z_h$ and for all $(\tilde{u}_{\pi},\tilde{\phi}_{\pi})\in[\LO]^2$ such that $(\tilde{u}_{\pi},\tilde{\phi}_{\pi})|_K\in \mathbb{P}_2(K)\times \mathbb{P}_1(K)$.
Let $(f,g)\in \Hcd\times\Hcu$, for any $(\tilde{u}_I,\tilde{\phi}_I)\in\Wh\times\Vh$, we have, $$\begin{aligned}
\Vert(T-T_h)(f,g)\Vert \leq \Vert(\tilde{u},\tilde{\phi})-(\tilde{u}_I,\tilde{\phi}_I)\Vert
+\Vert(\tilde{u}_I,\tilde{\phi}_I)-(\tilde{u}_h,\tilde{\phi}_h)\Vert.\label{estim-oper}\end{aligned}$$ Now, we define $(v_h,\psi_h)=(\tilde{u}_h-\tilde{u}_I,\tilde{\phi}_h-\tilde{\phi}_I)\in Z_h$, then from the ellipticity of $a_h(\cdot,\cdot)$ and the definition of $T$ and $T_h$, we have $$\begin{aligned}
&\beta\Vert(v_h,\psi_h)\Vert^2 \leq a_h((v_h,\psi_h),(v_h,\psi_h))
=a_h((\tilde{u}_h,\tilde{\phi}_h),(v_h,\psi_h))-a_h((\tilde{u}_I,\tilde{\phi}_I),(v_h,\psi_h))\nonumber\\[1ex]
&=b_h((f,g),(v_h,\psi_h))-\sumkth\Big\{a_{h,K}^{\Delta}(\tilde{u}_I,v_h)
+a_{h,K}^{\nabla}(\tilde{\phi}_I,\psi_h)\Big\}\nonumber\\[1ex]
&=b_h((f,g),(v_h,\psi_h))-\sumkth\Big\{a_{h,K}^{\Delta}(\tilde{u}_I-\tilde{u}_{\pi},v_h)+a_{h,K}^{\Delta}(\tilde{u}_{\pi},v_h)
+a_{h,K}^{\nabla}(\tilde{\phi}_I-\tilde{\phi}_{\pi},\psi_h)+a_{h,K}^{\nabla}(\tilde{\phi}_{\pi},\psi_h) \Big\}\nonumber\\[1ex]
&=b_h((f,g),(v_h,\psi_h))-\sumkth\Big\{a_{h,K}^{\Delta}(\tilde{u}_I-\tilde{u}_{\pi},v_h)
+a_{K}^{\Delta}(\tilde{u}_{\pi}-\tilde{u},v_h)+a_{K}^{\Delta}(\tilde{u},v_h)\nonumber\\[1ex]
&\hspace{5.0cm}+a_{h,K}^{\nabla}(\tilde{\phi}_I-\tilde{\phi}_{\pi},\psi_h)
+a_{K}^{\nabla}(\tilde{\phi}_{\pi}-\tilde{\phi},\psi_h)+a_{K}^{\nabla}(\tilde{\phi},\psi_h) \Big\}\nonumber\\[1ex]
&=\underbrace{\sumkth \Big\{ b_{h,K}((f,g ),(v_h,\psi_h))-b_K((f,g),(v_h,\psi_h))\Big\}}_{E_1}
-\underbrace{\sumkth\Big\{a_{h,K}^{\Delta}(\tilde{u}_I-\tilde{u}_{\pi},v_h)
+a_{K}^{\Delta}(\tilde{u}_{\pi}-\tilde{u},v_h)\Big\}}_{E_2}\nonumber \\[1ex]
&- \underbrace{\sumkth \Big\{ a_{h,K}^{\nabla}(\tilde{\phi}_I-\tilde{\phi}_{\pi},\psi_h)
+a_{K}^{\nabla}(\tilde{\phi}_{\pi}-\tilde{\phi},\psi_h) \Big\}}_{E_3}\label{cotconv},\end{aligned}$$ where we have used the consistency properties -. We now bound each term $E_i|_K$, $i = 1,2,3$.
First, the term $E_1|_K$ can be written as follows: $$\begin{aligned}
& b_{h,K}((f,g),(v_h,\psi_h))-b_K((f,g),(v_h,\psi_h))\nonumber\\[1ex]
&=\frac{n}{n-1}\Big\{\underbrace{\int_K \Pi_2^{0}(\Delta f)\Pi_2^{0}v_h-\int_K \Delta f \overline{v_h}}_{E_{11}} \Big\}
+\frac{1}{n-1}\Big\{\underbrace{\int_K \Pi_2^{0}f \Pi_2^{0}(\Delta v_h)
-\int_K f \Delta \overline{v_h}}_{E_{12}} \Big\}\nonumber\\[1ex]
& -\Big\{\underbrace{\int_K \nabla \Pi_1^{\nabla}g \cdot \nabla \Pi_2^{\nabla}v_h
- \int_K \nabla g \cdot \nabla \overline{v_h}}_{E_{13}} \Big\}
+ \frac{n}{n-1}\Big\{\underbrace{\int_K (\Pi_2^{0}f)(\Pi_1^{0}\psi_h)
-\int_K f \overline{\psi_h}}_{E_{14}} \Big\}.\label{exp-E_1}\end{aligned}$$
Now, we will bound each term $E_{1i}|_K \ i=1,2,3,4$. The term $E_{11}$ can be bounded as follows: Using the definition of $\Pi_2^{0}$ and Proposition \[app1\], we have $$\begin{aligned}
E_{11}&=\int_K \Delta f(\overline{v_h}-\Pi_2^{0}v_h)\leq |f |_{2,K} ||v_h-\Pi_2^{0}v_h||_{0,K}\\
&=|f |_{2,K} \inf_{q\in \mathbb{P}_2(K)}||v_h-q||_{0,K}\leq Ch_\E^{2}|f|_{2,K}|v_h|_{2,K}.\end{aligned}$$
For the term $E_{12}$, we repeat the same arguments to obtain: $$\begin{aligned}
E_{12}\leq Ch_K^2|f |_{2,K}|v_h|_{2,K}.\end{aligned}$$
Now, we bound $E_{13}$. From the definition of $\Pi_2^{\nabla}$, we have $$\begin{aligned}
E_{13}&=\int_K \nabla \Pi_1^{\nabla}g \cdot \nabla \overline{v_h}
- \int_K \nabla g \cdot \nabla \overline{v_h}=\int_K \nabla(\Pi_1^{\nabla}g -g )\cdot \nabla \overline{v_h}\\[1ex]
&=\int_K\nabla(\Pi_1^{\nabla}g -g )\cdot \nabla (\overline{v_h} -\tilde{v}_{\pi})\leq
|\Pi_1^{\nabla}g -g |_{1,K}|v_h -\tilde{v}_{\pi} |_{1,K}\\[1ex]
&\leq Ch_K|g |_{1,K} |v_h |_{2,K},\end{aligned}$$ where we have used the definition and the stability of $\Pi_1^{\nabla}$ with $\tilde{v}_{\pi}\in \mathbb{P}_1(K)$ such that Proposition \[app1\] holds true.
For the term $E_{14}$, we first use the definition of $\Pi_2^{0}$, the definition and the stability of $\Pi_1^{0}$ with respect to $\hat{f}_{\pi}\in \mathbb{P}_1(K)$ such that Proposition \[app1\] holds true, thus, we have $$\begin{aligned}
E_{14}&=\int_K f \Pi_1^{0}\psi_h-\int_K f\overline{\psi_h}=\int_K(f-\hat{f}_{\pi})(\Pi_1^{0}\psi_h-\overline{\psi_h})\\[1ex]
&\leq Ch_K^{2}|f |_{2,K}||\Pi_1^{0}\psi_h -\psi_h ||_{0,K}
\leq Ch_K^{2} | f|_{2,K}\Vert\psi_h\Vert_{0,\E}.\end{aligned}$$
Therefore, using the Cauchy-Schwarz inequality, we can deduce from that $$\begin{aligned}
E_1\leq Ch||(f,g ) || ||(v_h,\psi_h) ||.\end{aligned}$$
Finally, from we have $$\begin{aligned}
\beta ||(v_h,\psi_h) ||\leq C \Big\{ h||(f,g ) || +| u-u_I|_{2,\O} + | u-u_{\pi}|_{2,h}
+| \phi-\phi_I|_{1,\O} + |\phi-\phi_{\pi} |_{1,h} \Big\}.\end{aligned}$$ Therefore, the proof follows from and the previous inequality.
For the convergence and error analysis of the proposed virtual element scheme for the transmission eigenvalue problem, we first establish that $T_h\to T$ in norm as $h\to0$. Then, we prove a similar convergence result for the adjoint operators $T^*$ and $T_h^*$ of $T$ and $T_h$, respectively.
\[propP1\] There exist $C>0$ and $\tilde{s}\in(0,1]$, independent of $h$, such that $$\|T-T_h\|\le Ch^{\tilde{s}}.$$
Let $(f,g)\in\Hcd\times\Hcu$ such that $||(f,g)||=1$, let $(\tilde{u},\tilde{\phi})$ and $(\tilde{u}_h,\tilde{\phi}_h)$ be the solution of problems and , respectively, so that $(\tilde{u},\tilde{\phi}):=T(f,g)$ and $(\tilde{u}_h,\tilde{\phi}_h):=T_h(f,g)$. From Lemma \[lemcotste\], we have $$\begin{aligned}
\left\|\left(T-T_h\right)(f,g )\right\|& \leq C h ||(f,g )|| + \Vert u-u_{I}\Vert_{2,\O}
+\vert u-u_{\pi} \vert_{2,h} + \Vert \phi - \phi_{I}\Vert_{1,\O}
+\vert \phi - \phi_{\pi}\vert_{1,h}\\[1ex]
&\leq C\left( h||(f,g ) ||+ h^s||f||_{2,\O} + h^t||g ||_{1,\O}\right)\\
&\le Ch^{\tilde{s}}|| (f,g )||\nonumber\end{aligned}$$ where we have used the Propositions \[app1\], \[app0\] and \[app2\], and Lemma \[lem\_regul\], with $\tilde{s}:=\min\{s,t\}$. Thus, we conclude the proof.
Let $T^*$ and $T_h^*:\Hcd\times\Hcu\to\Hcd\times\Hcu$ the adjoint operators of $T$ and $T_h$, respectively, defined by $T^*(f,g):=(\tilde{u}^*,\tilde{\phi}^*)$ and $T^*_h(f,g):=(\tilde{u}_h^*,\tilde{\phi}_h^*)$, where $(\tilde{u}^*,\tilde{\phi}^*)$ and $(\tilde{u}_h^*,\tilde{\phi}_h^*)$ are the unique solutions of the following problems: $$\begin{aligned}
& a((v,\psi),(\tilde{u}^*,\tilde{\phi}^*))=b((v,\psi),(f,g))\qquad \forall
(v,\psi)\in \Hcd\times\Hcu,\label{AdjCont}\\[1ex]
& a_h((v_h,\psi_h),(\tilde{u}_h^*,\tilde{\phi}_h^*))=b_h((v_h,\psi_h),(f,g))\qquad
\forall (v_h,\psi_h)\in Z_h.\label{AdjDisc} \end{aligned}$$
It is simple to prove that if $\mu$ is an eigenvalue of $T$ with multiplicity $m$, $\overline{\mu}$ is an eigenvalue of $T^{*}$ with the same multiplicity $m$.
Now, we will study the convergence in norm $T_h^{*}$ to $T^{*}$ as $h$ goes to zero. With this aim, first we establish an additional regularity result for the solution $(\tilde{u}^*,\tilde{\phi}^*)$ of problem .
\[lem\_regul\_adj\] There exist $s,t\in(1/2,1]$ and $C>0$ such that, for all $(f,g)\in \HdoO\times\Hcu$, the solution $(\tilde{u}^*,\tilde{\phi}^*)$ of problem satisfies $\tilde{u}^*\in H^{2+s}(\Omega)$, $\tilde{\phi}^*\in H^{1+t}(\Omega)$, and $$\Vert\tilde{u}^*\Vert_{2+s,\Omega}+\Vert\tilde{\phi}^*\Vert_{1+t,\Omega}\le C\Vert(f,g)\Vert.$$
The result follows repeating the same arguments used in the proof of Lemma \[lem\_regul\].
We note that the constants $s$ and $t$ in Lemma \[lem\_regul\_adj\] are the same as in Lemma \[lem\_regul\].
Now, we are in a position to establish the following result.
\[propP1adj\] There exist $C>0$ and $\tilde{s}\in(0,1]$, independent of $h$, such that $$\|T^*-T_h^*\|\le Ch^{\tilde{s}}.$$
It is essentially identical to that of Lemma \[lemcotste\].
Our final goal is to show convergence and obtain error estimates. With this aim, we will apply to our problem the theory from [@BO; @Osborn1975] for non-selfadjoint compact operators.
We first recall the definition of spectral projectors. Let $\mu$ be a nonzero eigenvalue of $T$ with algebraic multiplicity $m$ and let $\Gamma$ be an open disk in the complex plane centered at $\mu$, such that $\mu$ is the only eigenvalue of $T$ lying in $\Gamma$ and $\partial \Gamma\cap\sp(T)=\emptyset$. The spectral projectors $E$ and $E^{*}$ are defined as follows:
- The spectral projector of $T$ relative to $\mu$: $E:=(2\pi i)^{-1}\int_{\partial \Gamma} (z-T)^{-1}dz$;
- The spectral projector of $T^{*}$ relative to $\overline{\mu}$: $E^*:=(2\pi i)^{-1}\int_{\partial \Gamma} (z-T^*)^{-1}dz.$
$E$ and $E^{*}$ are projections onto the space of generalized eigenvectors $R(E)$ and $R(E^*)$, respectively. It is simple to prove that $R(E),R(E^*)\in H^{2+s}(\Omega)\times H^{1+t}(\Omega)$.
Now, since $T_h\to T$ in norm, there exist $m$ eigenvalues (which lie in $\Gamma$) $\mu_h^{(1)},\ldots,\mu_h^{(m)}$ of $T_h$ (repeated according to their respective multiplicities) will converge to $\mu$ as $h$ goes to zero.
In a similar way, we introduce the following spectral projector $E_h:=(2\pi i)^{-1}\int_{\partial \Gamma} (z-T_h)^{-1}dz$, which is a projector onto the invariant subspace $R(E_h)$ of $T_h$ spanned by the generalized eigenvectors of $T_h$ corresponding to $\mu_h^{(1)},\ldots,\mu_h^{(m)}$.
We recall the definition of the *gap* $\hdel$ between two closed subspaces $\mathcal{X}$ and $\mathcal{Y}$ of a Hilbert space $\mathcal{V}$: $$\hdel(\mathcal{X},\mathcal{Y})
:=\max\left\{\delta(\mathcal{X},\mathcal{Y}),\delta(\mathcal{Y},\mathcal{X})\right\},$$ where $$\delta(\mathcal{X},\mathcal{Y})
:=\sup_{\mathbf{x}\in\mathcal{X}:\
\left\|x\right\|_{\mathcal{V}}=1}\delta(x,\mathcal{Y}),
\quad\text{with }\delta(x,\mathcal{Y}):=
\inf_{y\in\mathcal{Y}}\|x-y\|_{\mathcal{V}}.$$
Let $\mathcal{P}_h:=\mathcal{P}_h^2\times \mathcal{P}_h^1:\Hcd\times\Hcu\to
Z_h\subseteq \Hcd\times\Hcu$ be the projector defined by $$a(\mathcal{P}_h(u,\phi)-(u,\phi),(v_h,\psi_h))=a^{\Delta}(\mathcal{P}_h^2u-u,v_h)
+a^{\nabla}(\mathcal{P}_h^1\phi-\phi,\psi_h)=0\qquad \forall (v_h,\psi_h)\in Z_h.$$ We note that the form $a(\cdot,\cdot)$ is the inner product of $\Hcd\times\Hcu$. Therefore, we have $$\begin{aligned}
|(u,\phi)-\mathcal{P}(u,\phi)|_{\Hcd\times\Hcu}=\inf\limits_{(v_h,\psi_h)\in Z_h}|(u,\phi)
-(v_h,\psi_h)|_{\Hcd\times\Hcu},\label{best_appr_of_P}\end{aligned}$$ and $$\begin{aligned}
|\mathcal{P}(u,\phi)|_{\Hcd\times\Hcu}\leq |(u,\phi)|_{\Hcd\times\Hcu}\qquad
\forall (u,\phi)\in \Hcd\times\Hcu.\label{acot_de_P}\end{aligned}$$
The following error estimates for the approximation of eigenvalues and eigenfunctions hold true.
\[gap\] There exists a strictly positive constant $C$ such that $$\begin{aligned}
\hdel(R(E),R(E_h))
& \le C h^{\min\{s,t\}},\label{bound1}
\\
\left|\mu-\hat{\mu}_h\right|
& \le Ch^{2\min\{s,t\}},\label{bound2}\end{aligned}$$ where $\hat{\mu}_h:=\frac{1}{m}\sum\limits_{k=1}^{m}\mu_h^{(k)}$ and with the constants $s$ and $t$ as in Lemmas \[lem\_regul\] and \[lem\_regul\_adj\] (see also Remark \[rem\_regul\]).
As a consequence of Lemma \[propP1\], $T_h$ converges in norm to $T$ as $h$ goes to zero. Then, the proof of follows as a direct consequence of Theorem 7.1 from [@BO] and the fact that, for $(f,g)\in R(E)$, $\Vert(f,g)\Vert_{H^{2+s}(\Omega)\times H^{1+t}(\Omega)}\le \Vert(f,g)\Vert$, because of Lemma \[lem\_regul\].
In what follows we will prove : assume that $T(u_k,\phi_k)=\mu(u_k,\phi_k)$, $k=1,\ldots,m$. Since $a(\cdot,\cdot)$ is an inner product in $\Hcd\times\Hcu$, we can choose a dual basis for $R(E^*)$ denoted by $(u_k^*,\phi_k^*)\in\Hcd\times\Hcu$ satisfying $$a((u_k,\phi_k),(u_l^*,\phi_l^*))=\delta_{k,l}.$$ Now, from [@BO Theorem 7.2], we have that $$|\mu-\hat{\mu}_h|\leq \frac{1}{m}\sum\limits_{k=1}^{m}|\langle(T-T_h)(u_k,\phi_k),(u_k^*,\phi_k^*)\rangle|
+C||(T-T_h)|_{R(E)}|| ||(T^*-T_h^*)|_{R(E^*)}||,$$ where $\langle\cdot,\cdot\rangle$ denotes the corresponding duality pairing.
Thus, in order to obtain , we need to bound the two terms on the right hand side above.
The second term can be easily bounded from Lemmas \[propP1\] and \[propP1adj\]. In fact, we have $$\label{est1_conv_autv}
||(T-T_h)|_{R(E)}|| ||(T^*-T_h^*)|_{R(E^*)}||\leq Ch^{2\min\{s,t\}}.$$
Next, we manipulate the first term as follows: adding and subtracting $(v_h,\psi_h)\in Z_h$ and using the definition of $T$ and $T_h$, we obtain, $$\begin{aligned}
&\langle(T-T_h)(u_k,\phi_k),(u_k^*,\phi_k^*)\rangle
=a((T-T_h)(u_k,\phi_k),(u_k^*,\phi_k^*))\nonumber\\
&=a((T-T_h)(u_k,\phi_k),(u_k^*,\phi_k^*)-(v_h,\psi_h))+a(T(u_k,\phi_k),(v_h,\psi_h))
-a(T_h(u_k,\phi_k),(v_h,\psi_h))\nonumber\\
&=a((T-T_h)(u_k,\phi_k),(u_k^*,\phi_k^*)-(v_h,\psi_h))+b((u_k,\phi_k),(v_h,\psi_h))
-a(T_h(u_k,\phi_k),(v_h,\psi_h))\nonumber\\
&+a_h(T_h(u_k,\phi_k),(v_h,\psi_h)) -b_h((u_k,\phi_k),(v_h,\psi_h))\nonumber\\
&=\Big\{a((T-T_h)(u_k,\phi_k),(u_k^*,\phi_k^*)-(v_h,\psi_h))\Big\}
+\Big\{b((u_k,\phi_k),(v_h,\psi_h))-b_h((u_k,\phi_k),(v_h,\psi_h))\Big\}\nonumber\\
&+\Big\{a_h(T_h(u_k,\phi_k),(v_h,\psi_h))- a(T_h(u_k,\phi_k),(v_h,\psi_h))\Big\}
\qquad \forall (v_h,\psi_h)\in Z_h.\label{exp_Rod}\end{aligned}$$ Now, we estimate each bracket in separately. First, to bound the second bracket, we use the additional regularity of $(u_k,\phi_k)\in R(E)\subset H^{2+s}(\O)\times H^{1+t}(\O)$ and repeating the same steps used to derive (in this case with $(u_k,\phi_k)$ instead of $(f,g)$), we have $$\begin{aligned}
& b_{h,K}((u_k,\phi_k),(v_h,\psi_h))-b_{\E}((u_k,\phi_k),(v_h,\psi_h))
=E_{11}+E_{12}+E_{13}+E_{14}.\nonumber\end{aligned}$$
Now, we will bound each term $E_{1i} \ i=1,2,3,4$, as in the proof of Lemma \[lemcotste\], but in this case exploiting the additional regularity and the estimates in Lemmas \[lem\_regul\] and \[lem\_regul\_adj\] for $(u_k,\phi_k)\in R(E)$ and $(u_k^{*},\phi_k^{*})\in R(E^{*})$, respectively.
In particular, the terms $E_{11},E_{12}$ and $E_{14}$ can be bound exactly as in the proof of Lemma \[lemcotste\]. However, for the term $E_{13}$, we proceed as follows:
$$\begin{aligned}
E_{13}&=\int_K \nabla \Pi_1^{\nabla}\phi_k \cdot \nabla \overline{v_h} - \int_K \nabla \phi_k
\cdot \nabla \overline{v_h}=\int_K \nabla(\Pi_1^{\nabla}\phi_k -\phi_k )\cdot \nabla \overline{v_h}\\[1ex]
&=\int_K\nabla(\Pi_1^{\nabla}\phi_k -\phi_k )\cdot \nabla (\overline{v_h}
-\tilde{v}_{h}^{\pi})\leq |\Pi_1^{\nabla}\phi_k -\phi_k |_{1,K}|v_h -\tilde{v}_{h}^{\pi} |_{1,K}\\[1ex]
&=\inf_{q_h\in\mathbb{P}_1(K)}|\phi_k-q_h|_{1,K}|v_h -\tilde{v}_{h}^{\pi} |_{1,K}
\leq Ch_K^{1+t}|\phi_k |_{1+t,K}|v_h |_{2,K}\\
&\leq Ch_K^{2\min\{s,t\}}|\phi_k |_{1+t,K}|v_h |_{2,K},\end{aligned}$$
where we have used the definition of $\Pi_1^{\nabla}$ with $\tilde{v}_{h}^{\pi}\in\mathbb{P}_1(K)$ such that Proposition \[app1\] holds true and the fact that $\phi_k\in H^{1+t}(\O)$ together with Proposition \[app1\] again.
Therefore taking sum and using the additional regularity for $\phi_k$, together with Lemma \[lem\_regul\], we obtain $$\begin{aligned}
&\Big\{b((u_k,\phi_k),(v_h,\psi_h))-b_h((u_k,\phi_k),(v_h,\psi_h))\Big\}
\leq Ch^{2\min\{s,t\}}||(u_k,\phi_k )||||(v_h,\psi_h) ||
\qquad \forall (v_h,\psi_h) \in Z_h.\label{eq_Est_Rod0}\end{aligned}$$
Now, we estimate the third bracket in . Let $(w_h,\xi_h):=T_h(u_k,\phi_k)$ and $\Pi_h^K$ be defined by $(\Pi_h^K(v,\psi))|_K:=(\Pi_2^{\Delta}v,\Pi_1^{\nabla}\psi)$ for all $K\in\CT_h$ and for all $(v,\psi)\in\Hcd\times\Hcu$, where $\Pi_2^{\Delta}$ and $\Pi_1^{\nabla}$ have been defined in - and -, respectively. Hence, we have $$\begin{aligned}
&a_h((w_h,\xi_h),(v_h,\psi_h))-a((w_h,\xi_h),(v_h,\psi_h))=\sumkth
\Big\{ a_{h,K}((w_h,\xi_h),(v_h,\psi_h))-a_\E((w_h,\xi_h),(v_h,\psi_h)) \Big\}\nonumber\\
&=\sumkth \Big\{ a_{h,K}((w_h,\xi_h)-(\Pi_2^{\Delta}w_h,\Pi_1^{\nabla}\xi_h),(v_h,\psi_h))
+a_K((\Pi_2^{\Delta}w_h,\Pi_1^{\nabla}\xi_h)-(w_h,\xi_h),(v_h,\psi_h))\Big\}\nonumber\\
&=\sumkth \Big\{ a_{h,K}((w_h,\xi_h)-(\Pi_2^{\Delta}w_h,\Pi_1^{\nabla}\xi_h),(v_h,\psi_h)
-(\Pi_2^{\Delta}v_h,\Pi_1^{\nabla}\psi_h))\nonumber\\
&+a_K((\Pi_2^{\Delta}w_h,\Pi_1^{\nabla}\xi_h)-(w_h,\xi_h),(v_h,\psi_h)-(\Pi_2^{\Delta}v_h,\Pi_1^{\nabla}\psi_h)) \Big\}\nonumber\\
&\leq C\sumkth \Big\{|(w_h,\xi_h)-(\Pi_2^{\Delta}w_h,\Pi_1^{\nabla}\xi_h)|_{H^2(K)\times H^{1}(K)}
|(v_h,\psi_h)-(\Pi_2^{\Delta}v_h,\Pi_1^{\nabla}\psi_h)|_{H^2(K)\times H^{1}(K)}\Big\}\nonumber\\
&= C\sumkth \Big\{|T_h(u_k,\phi_k)-\Pi_h^K T_h(u_k,\phi_k)|_{H^2(K)\times H^{1}(K)}
|(v_h,\psi_h)-\Pi_h^K(v_h,\psi_h)|_{H^2(K)\times H^{1}(K)}\Big\},\label{eq_Est_Rod1}\end{aligned}$$ for all $(v_h,\psi_h)\in Z_h$, where we have used -, Cauchy-Schwarz inequality and -. Now, using the triangular inequality, we have that $$\begin{aligned}
|T_h(u_k,\phi_k)-\Pi_h^K T_h(u_k,\phi_k)|_{H^{2}(K)\times H^{1}(K)}&\leq
|T_h(u_k,\phi_k)-T(u_k,\phi_k)|_{H^{2}(K)\times H^{1}(K)}\nonumber\\
&+|\Pi_h^K T_h(u_k,\phi_k)-\Pi_h^K T(u_k,\phi_k)|_{H^{2}(K)\times H^{1}(K)}\nonumber\\
&+|\Pi_h^K T(u_k,\phi_k)-T(u_k,\phi_k)|_{H^{2}(K)\times H^{1}(K)}.\nonumber\end{aligned}$$ Thus, from , the above estimate, the stability of $\Pi_h^K$ and the additional regularity for $(u_k,\phi_k)$ together with Lemma \[lem\_regul\], we have $$\begin{aligned}
&a_h(T_h(u_k,\phi_k),(v_h,\psi_h))- a(T_h(u_k,\phi_k),(v_h,\psi_h))\nonumber\\
&\leq Ch^{\min\{s,t \}}||(u_k,\phi_k )||\sumkth
|(v_h,\psi_h)-\Pi_h^K(v_h,\psi_h)|_{H^2(K)\times H^{1}(K)}\qquad \forall (v_h,\psi_h)\in Z_h. \label{eq_Est_Rod2}\end{aligned}$$
Finally, we take $(v_h,\psi_h):=\mathcal{P}(u_k^*,\phi_k^*)\in Z_h$ in . Thus, on the one hand, we bound the first bracket in as follows, $$\begin{aligned}
&a((T-T_h)(u_k,\phi_k),(u_k^*,\phi_k^*)-(v_h,\psi_h))=a((T-T_h)(u_k,\phi_k),(u_k^*,\phi_k^*)
-\mathcal{P}(u_k^*,\phi_k^*))\nonumber\\
&\leq |(T-T_h)(u_k,\phi_k)|_{\Hcd\times\Hcu}|(u_k^*,\phi_k^*)-\mathcal{P}(u_k^*,\phi_k^*)|_{\Hcd\times\Hcu}\nonumber\\
&= |(T-T_h)(u_k,\phi_k)|_{\Hcd\times\Hcu}\inf\limits_{(r_h,s_h)\in Z_h}|(u_k^*,\phi_k^*)
-(r_h,s_h)|_{\Hcd\times\Hcu}\nonumber\\
&\leq |(T-T_h)(u_k,\phi_k)|_{\Hcd\times\Hcu}|(u_k^*,\phi_k^*)-((u_k^*)_I,(\phi_k^*)_I)|_{\Hcd\times\Hcu}\nonumber\\
&\leq Ch^{2{\min\{s,t \}}}||(u_k^*,\phi_k^*)||,\nonumber\end{aligned}$$ where we have used , Propositions \[app0\] and \[app2\], the additional regularity for $(u_k^*,\phi_k^*)$, Lemma \[lem\_regul\_adj\] and Lemma \[propP1\].
On the other hand, from we have that $$\begin{aligned}
|(v_h,\psi_h)-\Pi_h^K(v_h,\psi_h)|&_{H^2(K)\times H^{1}(K)}=|\mathcal{P}(u_k^*,\phi_k^*)-\Pi_h^K\mathcal{P}(u_k^*,\phi_k^*)|_{H^2(K)\times H^{1}(K)}\nonumber\\
&\leq |\mathcal{P}(u_k^*,\phi_k^*)-(u_k^*,\phi_k^*) |_{H^2(K)\times H^{1}(K)}+|(u_k^*,\phi_k^*)-\Pi_h^K(u_k^*,\phi_k^*) |_{H^2(K)\times H^{1}(K)}\nonumber\\
&+|\Pi_h^K((u_k^*,\phi_k^*)-\mathcal{P}(u_k^*,\phi_k^*))|_{H^2(K)\times H^{1}(K)}.\nonumber\end{aligned}$$ Then, using again , Propositions \[app0\] and \[app2\], the additional regularity for $(u_k^*,\phi_k^*)$, Lemma \[lem\_regul\_adj\] and Lemma \[propP1\], we obtain from that $$\begin{aligned}
&a_h(T_h(u_k,\phi_k),(v_h,\psi_h))- a(T_h(u_k,\phi_k),(v_h,\psi_h))\leq
Ch^{2{\min\{s,t \}}}||(u_k,\phi_k )||||(u_k^*,\phi_k^*)||. \label{eq_Est_Rod3}\end{aligned}$$ Thus, from , and , we obtain $$\begin{aligned}
|\langle (T-T_h)(u_k,\phi_k),(u_k^*,\phi_k^*) \rangle|\leq C h^{2\min\{s,t \}} \label{preend}.\end{aligned}$$ Therefore, the proof follows from estimates and .
The error estimate for the eigenvalue $\mu$ of $T$ yield analogous estimate for the approximation of the eigenvalue $\lambda = 1/\mu$ of Problem \[P1\] by means of $\hat{\lambda}_h:=\frac{1}{m}\sum\limits_{k=1}^{m}\lambda_h^{(k)}$, where $\lambda_h^{(k)}=1/\mu_h^{(k)}$.
Numerical results {#SEC:NumRes}
=================
In this section we present a series of numerical experiments to solve the transmission eigenvalue problem with the Virtual Element scheme . However, to complete the choice of the VEM, we had to fix the forms $s_{K}^{\Delta}(\cdot,\cdot)$ and $s_{K}^{\nabla}(\cdot,\cdot)$ satisfying and , respectively. For $s_{K}^{\Delta}(\cdot,\cdot)$, we consider the same definition as in [@MRV]: $$\begin{aligned}
s_\E^{\Delta}(u_h,v_h):=\sigma_{\E}\sum\limits_{i=1}^{N_\E}[u_h(P_i)v_h(P_i)
+h_{P_i}^2\nabla u_h(P_i)\cdot\nabla v_h(P_i)] & \quad \forall u_h,v_h\in W_h^{K},\end{aligned}$$ where $P_1,\ldots,P_{N_{\E}}$ are the vertices of $\E$, $h_{P_i}$ corresponds to the maximum diameter of the elements with $P_i$ as a vertex and $\sigma_\E>0$ is a multiplicative factor to take into account the magnitude of the parameter and the $h$-scaling, for instance, in the numerical tests we have picked $\sigma_\E>0$ as the mean value of the eigenvalues of the local matrix $a_{\E}^{\Delta}\big(\Pi_2^{\Delta} u_h,\Pi_2^{\Delta} v_h\big)$. This ensures that the stabilizing term scales as $a_{\E}^{\Delta}(v_h,v_h)$. Now, a choice for $s_{K}^{\nabla}(\cdot,\cdot)$ is given by $$\begin{aligned}
s_\E^{\nabla}(\phi_h,\psi_h):=\sum\limits_{i=1}^{N_\E}\phi_h(P_i)\psi_h(P_i)
& \quad \forall \phi_h,\psi_h\in V_h^{K},\end{aligned}$$ which corresponds to the identity matrix of dimension $N_\E$. A proof of and for the above choices could be derived following the arguments in [@BLR2017]. Finally, we mention that the previous definitions are in accordance with the analysis presented in [@MRR2015; @MRV] in order to avoid spectral pollution.
We have implemented in a MATLAB code the proposed VEM on arbitrary polygonal meshes, by following the ideas presented in [@BBMR2014]. Moreover, we compare our results with those existing in the literature, for example [@CakMonkSun2014; @ChenGZZ2016; @ColtonMonkSun2010; @HYBi2017]. We have considered three different domains, namely: square domain, a circular domain centered at the origin and an L-shaped domain.
Test 1: Square domain
---------------------
In this test, we have taken $\O:=(0,1)^2$ and index of refraction $n=4$ and $n=16$. We have tested the method by using different families of meshes (see Figure \[FIG:VM1\]):
- $\CT_h^1$: triangular meshes;
- $\CT_h^2$: rectangular meshes;
- $\CT_h^3$: hexagonal meshes;
- $\CT_h^4$: non-structured hexagonal meshes made of convex hexagons.
The refinement parameter $N$ used to label each mesh is the number of elements on each edge of the domain.
![ Sample meshes: $\CT_h^1$ (top left), $\CT_h^2$ (top right), $\CT_h^3$ (bottom left) and $\CT_h^{4}$ (bottom right), for $N=8$.[]{data-label="FIG:VM1"}](trian.eps){height="6.3cm" width="6.3cm"}
![ Sample meshes: $\CT_h^1$ (top left), $\CT_h^2$ (top right), $\CT_h^3$ (bottom left) and $\CT_h^{4}$ (bottom right), for $N=8$.[]{data-label="FIG:VM1"}](cuad8.eps){height="6.3cm" width="6.3cm"}
![ Sample meshes: $\CT_h^1$ (top left), $\CT_h^2$ (top right), $\CT_h^3$ (bottom left) and $\CT_h^{4}$ (bottom right), for $N=8$.[]{data-label="FIG:VM1"}](exagonos.eps){height="6.3cm" width="6.3cm"}
![ Sample meshes: $\CT_h^1$ (top left), $\CT_h^2$ (top right), $\CT_h^3$ (bottom left) and $\CT_h^{4}$ (bottom right), for $N=8$.[]{data-label="FIG:VM1"}](hex_irreg8.eps){height="6.3cm" width="6.3cm"}
We report in Tables \[TAB:Sq01n=16\] and \[TAB:Sq01n=4\] the lowest transmission eigenvalues $k_{ih}$, $i=1,2,3,4$ computed by our method with two different families of meshes and $N=32,64,128,$ and with the index of refraction $n=16$ and $n=4$, respectively. The tables include computed orders of convergence, as well as more accurate values extrapolated by means of a least-squares fitting. Moreover, we compare the performance of the proposed method with those presented in [@ColtonMonkSun2010; @HYBi2017]. With this aim, we include in the last row of Tables \[TAB:Sq01n=16\] and \[TAB:Sq01n=4\] the results reported in that references, for the same problem.
Meshes $k_{ih}$ $k_{1h}$ $k_{2h}$ $ k_{3h}$ $k_{4h}$
----------- ---------------------------------------- ---------- ---------- ------------ ---------- --
$N=32$ 1.8805 2.4467 2.4467 2.8691
$N=64$ 1.8798 2.4449 2.4449 2.8671
$\CT_h^1$ $N=128$ 1.8796 2.4444 2.4444 2.8666
Order 2.01 2.00 2.00 2.01
Extrapolated 1.8796 2.4442 2.4442 2.8664
$N=32$ 1.8764 2.4318 2.4318 2.8645
$N=64$ 1.8788 2.4410 2.4410 2.8658
$\CT_h^2$ $N=128$ 1.8794 2.4434 2.4434 2.8663
Order 1.95 1.95 1.95 1.61
Extrapolated 1.8796 2.4443 2.4443 2.8665
[@ColtonMonkSun2010 Argyris method] 1.8651 2.4255 2.4271 2.8178
[@ColtonMonkSun2010 Continuous method] 1.9094 2.5032 2.5032 2.9679
[@ColtonMonkSun2010 Mixed method] 1.8954 2.4644 2.4658 2.8918
[@HYBi2017] 1.8796 2.4442 2.4442 2.8664
: Test 1: Lowest transmission eigenvalues $k_{ih}$, $i=1,2,3,4$ computed on different meshes and with index of refraction $n=16$.[]{data-label="TAB:Sq01n=16"}
Meshes $k_{ih}$ $k_{1h}$ $k_{2h}$ $ k_{3h}$ $k_{4h}$
----------- -------------- ---------------- ---------------- ------------ ---------- --
$N=32$ 4.2835-1.1367 4.2835+1.1367 5.3373 5.4172
$N=64$ 4.2745-1.1446 4.2745+1.1446 5.4375 5.4599
$\CT_h^3$ $N=128$ 4.2724-1.1467 4.2724+1.1467 5.4661 5.4719
Order 2.10& 1.89 2.10& 1.89 1.81 1.84
Extrapolated 4.2717-1.1475 4.2717+1.1475 5.4775 5.4765
$N=32$ 4.2870-1.1341 4.2870+1.1341 5.3245 5.4178
$N=64$ 4.2753-1.1438 4.2753+1.1438 5.4329 5.4602
$\CT_h^4$ $N=128$ 4.2726-1.1465 4.2726+1.1465 5.4647 5.4719
Order 2.12 &1.86 2.12&1.86 1.77 1.85
Extrapolated 4.2718-1.1475 4.2718+1.1475 5.4779 5.4765
[@HYBi2017] 4.2717-1.1474i 4.2717+1.1474i 5.4761 5.4761
: Test 1: Lowest transmission eigenvalues $k_{ih}$, $i=1,2,3,4$ computed on different meshes and with index of refraction $n=4$.[]{data-label="TAB:Sq01n=4"}
It can be seen from Tables \[TAB:Sq01n=16\] and \[TAB:Sq01n=4\] that the eigenvalue approximation order of our method is quadratic (as predicted by the theory for convex domains) and that the results obtained by the two methods agree perfectly well.
Test 2: Circular domain
-----------------------
In this test, we have taken as domain the circle $\O:=\{(x,y)\in \R^2: x^2+y^2<1/2\}$. We have used polygonal meshes created with PolyMesher [@malla] (see Figure \[FIG:CirclePolymesherLap\]). The refinement parameter $N$ is the number of elements intersecting the boundary.
We report in Table \[TAB:3transp\] the five lowest transmission eigenvalues computed with the virtual element method analyzed in this paper. The table includes orders of convergence, as well as accurate values extrapolated by means of a least-squares fitting. Once again, the last rows show the values obtained by extrapolating those computed with different methods presented in [@CakMonkSun2014; @ChenGZZ2016; @ColtonMonkSun2010].
$k_{1h}$ $k_{2h}$ $k_{3h}$ $k_{4h}$ $k_{5h}$
---------------------------------------- ---------- ---------- ---------- ---------- ----------
$N=32$ 1.9835 2.6032 2.6037 3.2115 3.2117
$N=64$ 1.9869 2.6105 2.6106 3.2225 3.2227
$N=128$ 1.9877 2.6123 2.6123 3.2255 3.2256
Order 1.98 1.97 2.01 1.86 1.90
Extrapolated 1.9880 2.6129 2.6129 3.2267 3.2267
[@CakMonkSun2014] 1.9881 - - - -
[@ChenGZZ2016] 1.9879 2.6124 2.6124 3.2255 3.2255
[@ColtonMonkSun2010 Argyris method] 2.0076 2.6382 2.6396 3.2580 3.2598
[@ColtonMonkSun2010 Continuous method] 2.0301 2.6937 2.6974 3.3744 3.3777
[@ColtonMonkSun2010 Mixed method] 1.9912 2.6218 2.6234 3.2308 3.2397
: Test 2: Computed lowest transmission eigenvalues $k_{ih}$, $i=1,2,3,4,5$ with index of refraction $n=16$.[]{data-label="TAB:3transp"}
Once more, a quadratic order of convergence can be clearly appreciated from Table \[TAB:3transp\].
We show in Figure \[FIG:CirclePolymesherLap\] the eigenfunctions corresponding to the four lowest transmission eigenvalues.
![Test 2. Eigenfunctions $u_{1h}$ (top left), $u_{2h}$ (top right), $u_{3h}$ (bottom left) and $u_{4h}$ (bottom right).[]{data-label="FIG:CirclePolymesherLap"}](modo1_u_n16_TH_circle_radio05_Poly.eps){height="6cm" width="6.9cm"}
![Test 2. Eigenfunctions $u_{1h}$ (top left), $u_{2h}$ (top right), $u_{3h}$ (bottom left) and $u_{4h}$ (bottom right).[]{data-label="FIG:CirclePolymesherLap"}](modo2_u_n16_TH_circle_radio05_Poly.eps){height="6cm" width="6.9cm"}
![Test 2. Eigenfunctions $u_{1h}$ (top left), $u_{2h}$ (top right), $u_{3h}$ (bottom left) and $u_{4h}$ (bottom right).[]{data-label="FIG:CirclePolymesherLap"}](modo3_u_n16_TH_circle_radio05_Poly.eps){height="6cm" width="6.9cm"}
![Test 2. Eigenfunctions $u_{1h}$ (top left), $u_{2h}$ (top right), $u_{3h}$ (bottom left) and $u_{4h}$ (bottom right).[]{data-label="FIG:CirclePolymesherLap"}](modo4_u_n16_TH_circle_radio05_Poly.eps){height="6cm" width="6.9cm"}
Test 3: L-shaped domain
-----------------------
Finally, we have considered an L-shaped domain: $\O:=(-1/2,1/2)^2\backslash([0,1/2]\times[-1/2,0])$. We have used uniform triangular meshes as those shown in Figure \[FIG:Square01\_n=16\_Lshape\_unif\]. The meaning of the refinement parameter $N$ is the number of elements on each edge.
We report in Table \[TAB:LShape05\] the four lowest transmission eigenvalues computed with the virtual scheme analyzed in this paper. The table includes orders of convergence, as well as accurate values extrapolated by means of a least-squares fitting. Once again, we compare the performance of the proposed virtual scheme with the one presented in [@CakMonkSun2014] for the same problem, using triangular meshes.
$k_{ih}$ $k_{1h}$ $k_{2h}$ $k_{3h}$ $k_{4h}$
------------------- ---------- ---------- ---------- ----------
$N=32$ 2.9690 3.1480 3.4216 3.5744
$N=64$ 2.9590 3.1417 3.4136 3.5683
$N=128$ 2.9551 3.1400 3.4113 3.5667
Order 1.37 1.94 1.76 2.00
Extrapolated 2.9527 3.1395 3.4103 3.5662
[@CakMonkSun2014] 2.9553 - - -
: Test 3: Computed lowest transmission eigenvalues $k_{ih}$, $i=1,2,3,4$ with index of refraction $n=16$.[]{data-label="TAB:LShape05"}
We notice that for the first transmission eigenvalue, the method converges with order close to $\min\{1.089,1.333\}$, which corresponds to the Sobolev regularity of the domain for the biharmonic equation and Laplace equation and with homogeneous Dirichlet boundary conditions, respectively (see [@G]). Moreover, the method converges with larger orders for the rest of the transmission eigenvalues.
Finally, Figure \[FIG:Square01\_n=16\_Lshape\_unif\] shows the eigenfunctions corresponding to the four lowest transmission eigenvalues with index of refraction $n=16$.
![Test 3. Eigenfunctions $u_{1h}$ (top left), $u_{2h}$ (top right), $u_{3h}$ (bottom left) and $u_{4h}$ (bottom right).[]{data-label="FIG:Square01_n=16_Lshape_unif"}](modo1_u_n16_TH11.eps){height="6cm" width="6.9cm"}
![Test 3. Eigenfunctions $u_{1h}$ (top left), $u_{2h}$ (top right), $u_{3h}$ (bottom left) and $u_{4h}$ (bottom right).[]{data-label="FIG:Square01_n=16_Lshape_unif"}](modo2_u_n16_TH11.eps){height="6cm" width="6.9cm"}
![Test 3. Eigenfunctions $u_{1h}$ (top left), $u_{2h}$ (top right), $u_{3h}$ (bottom left) and $u_{4h}$ (bottom right).[]{data-label="FIG:Square01_n=16_Lshape_unif"}](modo3_u_n16_TH11.eps){height="6cm" width="6.9cm"}
![Test 3. Eigenfunctions $u_{1h}$ (top left), $u_{2h}$ (top right), $u_{3h}$ (bottom left) and $u_{4h}$ (bottom right).[]{data-label="FIG:Square01_n=16_Lshape_unif"}](modo4_u_n16_TH11.eps){height="6cm" width="6.9cm"}
Acknowledgments {#acknowledgments .unnumbered}
===============
The authors are deeply grateful to Prof. Carlo Lovadina (Università degli Studi di Milano, Italy) and Prof. Rodolfo Rodríguez (Universidad de Concepción, Chile) for the fruitful discussions. The First author was partially supported by CONICYT (Chile) through FONDECYT project 1180913 and by DIUBB through project 171508 GI/VC Universidad del Bío-Bío, Chile. The second author was partially supported by a CONICYT (Chile) fellowship.
[99]{}
<span style="font-variant:small-caps;">B. Ahmad, A. Alsaedi, F. Brezzi, L. D. Marini and A. Russo</span>, *Equivalent projectors for virtual element methods*, Comput. Math. Appl., [**66**]{}, (2013), pp. 376–391.
<span style="font-variant:small-caps;">P.F. Antonietti, L. Beirão da Veiga, D. Mora and M. Verani</span>, *A stream virtual element formulation of the Stokes problem on polygonal meshes*, SIAM J. Numer. Anal., [**52(1)**]{}, (2014), pp. 386–404.
<span style="font-variant:small-caps;">P.F. Antonietti, L. Beirão da Veiga, S. Scacchi and M. Verani</span>, *A $C^1$ virtual element method for the Cahn–Hilliard equation with polygonal meshes*, SIAM J. Numer. Anal., [**54(1)**]{}, (2016), pp. 36–56.
<span style="font-variant:small-caps;">I. Babuška and J. Osborn</span>, *Eigenvalue problems*, in *Handbook of Numerical Analysis*, Vol. II, P.G. Ciarlet and J.L. Lions, eds., North-Holland, Amsterdam, 1991, pp. 641–787.
<span style="font-variant:small-caps;">L. Beirão da Veiga, F. Brezzi, A. Cangiani, G. Manzini, L.D. Marini and A. Russo</span>, *Basic principles of virtual element methods*, Math. Models Methods Appl. Sci., [**23**]{}, (2013), pp. 199–214.
<span style="font-variant:small-caps;">L. Beirão da Veiga, F. Brezzi and L.D. Marini</span>, *Virtual elements for linear elasticity problems*, SIAM J. Numer. Anal., [**51**]{} (2013), pp. 794–812.
<span style="font-variant:small-caps;">L. Beirão da Veiga, F. Brezzi, L.D. Marini and A. Russo</span>, *The hitchhiker’s guide to the virtual element method*, Math. Models Methods Appl. Sci., [**24**]{}, (2014), pp. 1541–1573.
<span style="font-variant:small-caps;">L. Beirão da Veiga, K. Lipnikov and G. Manzini</span>, *The Mimetic Finite Difference Method for Elliptic Problems*, Springer, MS&A, vol. [**11**]{}, 2014.
<span style="font-variant:small-caps;">L. Beirão da Veiga, C. Lovadina and D. Mora</span>, *A virtual element method for elastic and inelastic problems on polytope meshes*, Comput. Methods Appl. Mech. Engrg., [**295**]{}, (2015) pp. 327–346.
<span style="font-variant:small-caps;">L. Beirão da Veiga, C. Lovadina and A. Russo</span>, *Stability analysis for the virtual element method*, Math. Models Methods Appl. Sci., [**27**]{}, (2017) pp. 2527–2594.
<span style="font-variant:small-caps;">L. Beirão da Veiga, C. Lovadina and G. Vacca</span>, *Divergence free virtual elements for the Stokes problem on polygonal meshes*, ESAIM Math. Model. Numer. Anal., [**51(2)**]{}, (2017) pp. 509–535.
<span style="font-variant:small-caps;">L. Beirão da Veiga and G. Manzini</span>, *A virtual element method with arbitrary regularity*, IMA J. Numer. Anal., [**34**]{}, (2014), pp. 759–781.
<span style="font-variant:small-caps;">L. Beirão da Veiga, D. Mora and G. Rivera</span>, *Virtual elements for a shear-deflection formulation of Reissner-Mindlin plates*, Math. Comp., DOI: https://doi.org/10.1090/mcom/3331 (2017).
<span style="font-variant:small-caps;">L. Beirão da Veiga, D. Mora, G. Rivera and R. Rodríguez</span>, *A virtual element method for the acoustic vibration problem*, Numer. Math., [**136(3)**]{}, (2017), pp. 725–763.
<span style="font-variant:small-caps;">M.F. Benedetto, S. Berrone, A. Borio, S. Pieraccini and S. Scialò</span>, *Order preserving SUPG stabilization for the virtual element formulation of advection–diffusion problems*, Comput. Methods Appl. Mech. Engrg., [**311**]{}, (2016), pp. 18–40.
<span style="font-variant:small-caps;">D. Boffi</span>, *Finite element approximation of eigenvalue problems*, Acta Numerica, [**19**]{} (2010), pp. 1–120.
<span style="font-variant:small-caps;">S.C. Brenner, Q. Guan and L.-Y Sung</span>, *Some estimates for virtual element methods*, Comput. Methods Appl. Math., [**17(4)**]{}, (2017), pp. 553–574.
<span style="font-variant:small-caps;">S.C. Brenner and R.L. Scott</span>, *The Mathematical Theory of Finite Element Methods*, Springer, New York, 2008.
<span style="font-variant:small-caps;">F. Brezzi and L.D. Marini</span>, *Virtual elements for plate bending problems*, Comput. Methods Appl. Mech. Engrg., [**253**]{}, (2012), pp. 455–462.
<span style="font-variant:small-caps;">E. Cáceres and G.N. Gatica</span>, *A mixed virtual element method for the pseudostress-velocity formulation of the Stokes problem*, IMA J. Numer. Anal., [**37(1)**]{}, 2017, pp. 296–331.
<span style="font-variant:small-caps;">E. Cáceres, G.N. Gatica and F. Sequeira</span>, *A mixed virtual element method for the Brinkman problem*, Math. Models Methods Appl. Sci., [**27(4)**]{}, (2017) pp. 707–743.
<span style="font-variant:small-caps;">F. Cakoni, M. Çayören and D. Colton</span>, *Transmission eigenvalues and the nondestructive testing of dielectrics*, Inverse Problems, [**24(6)**]{}, (2008), 065016.
<span style="font-variant:small-caps;">F. Cakoni, D. Colton, P. Monk and J. Sun</span>, *The inverse electromagnetic scattering problem for anisotropic media*, Inverse Problems, [**26(7)**]{}, (2010), 074004.
<span style="font-variant:small-caps;">F. Cakoni, D. Gintides and H. Haddar</span>, *The existence of an infinite discrete set of transmission eigenvalue*, SIAM J. Math. Anal., [**42**]{}, (2010), pp. 237–255.
<span style="font-variant:small-caps;">F. Cakoni, P. Monk and J. Sun</span>, *Error analysis for the finite element approximation of transmission eigenvalues*, Comput. Methods Appl. Math., [**14(4)**]{}, (2014), pp. 419–427.
<span style="font-variant:small-caps;">J. Camaño, R. Rodríguez and P. Venegas</span>, *Convergence of a lowest-order finite element method for the transmission eigenvalue problem*, CI$^2$MA Preprint 2018–06, Universidad de Concepción.
<span style="font-variant:small-caps;">A. Cangiani, E.H. Georgoulis and P. Houston</span>, *$hp$-version discontinuous Galerkin methods on polygonal and polyhedral meshes*, Math. Models Methods Appl. Sci., [**24(10)**]{}, (2014), pp. 2009–2041.
<span style="font-variant:small-caps;">A. Cangiani, E.H. Georgoulis, T. Pryer and O.J. Sutton</span>, *A posteriori error estimates for the virtual element method*, Numer. Math., [**137(4)**]{}, (2017) pp. 857–893.
<span style="font-variant:small-caps;">A. Cangiani, G. Manzini and O.J. Sutton</span>, *Conforming and nonconforming virtual element methods for elliptic problems*, IMA J. Numer. Anal., [**37(3)**]{}, (2017), pp. 1317–1354.
<span style="font-variant:small-caps;">H. Chen, H. Guo, Z. Zhang and Q. Zou</span>, *A $C^0$ linear finite element method for two fourth-order eigenvalue problems*, IMA J. Numer. Anal., [**37(4)**]{}, (2017), pp. 2120–2138.
<span style="font-variant:small-caps;">C. Chinosi and L.D. Marini</span>, *Virtual element method for fourth order problems: $L^2$-estimates*, Comput. Math. Appl., [**72(8)**]{}, (2016), pp. 1959–1967.
<span style="font-variant:small-caps;">P.G. Ciarlet</span>, *The Finite Element Method for Elliptic Problems*, SIAM, Philadelphia, 2002.
<span style="font-variant:small-caps;">D. Colton and R. Kress</span>, *Inverse Acoustic and Electromagnetic Scattering Theory*, 3rd ed., Springer, New York, 2013.
<span style="font-variant:small-caps;">D. Colton, P. Monk and J. Sun</span>, *Analytical and computational methods for transmission eigenvalues*, Inverse Problems, [**26(4)**]{}, (2010), 045011.
<span style="font-variant:small-caps;">D. Di Pietro and A. Ern</span>, *A hybrid high-order locking-free method for linear elasticity on general meshes*, Comput. Methods Appl. Mech. Eng., [**283**]{}, (2015), pp. 1–21.
<span style="font-variant:small-caps;">M. Frittelli and I. Sgura</span>, *Virtual element method for the Laplace–Beltrami equation on surfaces*, ESAIM Math. Model. Numer. Anal., DOI: https://doi.org/10.1051/m2an/2017040 (2017).
<span style="font-variant:small-caps;">F. Gardini, G. Manzini and G. Vacca</span>, *The nonconforming virtual element method for eigenvalue problems*, arXiv:1802.02942 \[math.NA\], (2018).
<span style="font-variant:small-caps;">F. Gardini and G. Vacca</span>, *Virtual element method for second order elliptic eigenvalue problems*, IMA J. Numer. Anal., DOI: https://doi.org/10.1093/imanum/drx063 (2017).
<span style="font-variant:small-caps;">H. Geng, X. Ji, J. Sun and L. Xu</span>, *C$^0$IP methods for the transmission eigenvalue problem*, J. Sci. Comput., [**68(1)**]{}, (2016), pp. 326–338.
<span style="font-variant:small-caps;">V. Girault and P.A. Raviart</span>, *Finite Element Methods for Navier-Stokes Equations*, Springer-Verlag, Berlin, 1986.
<span style="font-variant:small-caps;">P. Grisvard</span>, *Elliptic Problems in Non-Smooth Domains*, Pitman, Boston, 1985.
<span style="font-variant:small-caps;">J. Han, Y. Yang and H. Bi</span>, *Non-conforming finite element methodos for a transmission eigenvalue problem*, Comput. Methods Appl. Mech. Engrg., [**307**]{}, (2016), pp. 144–163.
<span style="font-variant:small-caps;">J. Han, Y. Yang and H. Bi</span>, *A new multigrid finite element method for the transmission eigenvalue problems*, Appl. Math. Comput., [**292**]{}, (2017), pp. 96–106.
<span style="font-variant:small-caps;">D. Mora and G. Rivera</span>, *A priori and a posteriori error estimates for a virtual element spectral analysis for the elasticity equations*, arXiv:1712.06441 \[math.NA\], (2017).
<span style="font-variant:small-caps;">D. Mora, G. Rivera and R. Rodríguez</span>, *A virtual element method for the Steklov eigenvalue problem*, Math. Models Methods Appl. Sci., [**25(8)**]{}, (2015), pp. 1421–1445.
<span style="font-variant:small-caps;">D. Mora, G. Rivera and I. Velásquez</span>, *A virtual element method for the vibration problem of Kirchhoff plates*, ESAIM Math. Model. Numer. Anal., DOI: https://doi.org/10.1051/m2an/2017041 (2017).
<span style="font-variant:small-caps;">J. Osborn</span>, *Spectral approximation for compact operators*, Math. Comp., [**29**]{}, (1975), pp 712–725.
<span style="font-variant:small-caps;">I. Perugia, P. Pietra and A. Russo</span>, *A plane wave virtual element method for the Helmholtz problem*, ESAIM Math. Model. Numer. Anal., [**50(3)**]{}, (2016), pp. 783–808.
<span style="font-variant:small-caps;">J. Sun</span>, *Iterative methods for transmission eigenvalues*, SIAM J. Numer. Anal., [**49(5)**]{}, (2011), pp. 1860–1874.
<span style="font-variant:small-caps;">C. Talischi, G.H. Paulino, A. Pereira and I.F.M. Menezes</span>, *Polygonal finite elements for topology optimization: A unifying paradigm*, Internat. J. Numer. Methods Engrg., [**82(6)**]{}, (2010), pp. 671–698.
<span style="font-variant:small-caps;">C. Talischi, G.H. Paulino, A. Pereira and I.F.M. Menezes</span>, *PolyMesher: a general-purpose mesh generator for poygonal elements written in Matlab*, Struct. Multidisc. Optim., [**45**]{}, (2012), pp. 309–328.
<span style="font-variant:small-caps;">G. Vacca</span>, *An $H^1$-conforming virtual element for Darcy and Brinkman equations*, Math. Models Methods Appl. Sci., [**28(1)**]{}, (2018), pp. 159–194.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The problem of radar detection in compound Gaussian clutter when a radar signature is not completely known has not been considered yet and is addressed in this paper. We proposed a robust technique to detect, based on the generalized likelihood ratio test, a point-like target embedded in compound Gaussian clutter. Employing an array of antennas, we assume that the actual steering vector departs from the nominal one, but lies in a known interval. The detection is then secured by employing a semi-definite programming. It is confirmed via simulation that the proposed detector experiences a negligible detection loss compared to an adaptive normalized matched filter in a perfectly matched case, but outperforms in cases of mismatched signal. Remarkably, the proposed detector possesses constant false alarm rate with respect to the clutter covariance matrix.'
author:
- 'Mai P. T. Nguyen and Iickho Song, [^1]'
title: Robust Radar Detection of a Mismatched Steering Vector Embedded in Compound Gaussian Clutter
---
Generalized likelihood ratio test, compound Gaussian clutter, semi-definite programming
Introduction
============
Problem Formulation and Proposed Detector
=========================================
Numerical Results
=================
Conclusion
==========
This paper has addressed the problem of detecting a mismatched signal embedded in compound Gaussian noise. Specifically, phase shifting of the actual steering vector departs from that of the nominal one but belongs to a known interval. The proposed detector is shown to be more robust to mismatched signals than the adaptive NMF, and even achieves reasonable detection probabilities when the signal to detect lying out of the designed interval. Remarkably, the $\theta$-MLE detector has CFAR w.r.t all statistic of noise. A drawback of the proposed detector is that the likelihood ratio has no explicit form, for which it is difficult to gain a deeper insight into the performance of the detector. Another drawback is the complexity associated with the SDP. Though proposed scheme can detect a seriously mismatched signal, it does not include effects of possible interference, which might be a topic for a further research.
Acknowledgment {#acknowledgment .unnumbered}
==============
The authors would like to thank the anonymous reviewers for their valuable comments and suggestions to improve the paper.
[^1]: The authors are with the School of Electrical Engineering, Korean Advanced Institute of Science and Technology, Daejeon 34141, Korea.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'In recent years there has been increased interest in extremal problems for “counting” parameters of graphs. For example, the Kahn-Zhao theorem gives an upper bound on the number of independent sets in a $d$-regular graph. In the same spirit, the Upper Matching Conjecture claims an upper bound on the number of $k$-matchings in a $d$-regular graph. Here we consider both matchings and matchings of fixed sizes in graphs with a given number vertices and edges. We prove that the graph with the fewest matchings is either the lex or the colex graph. Similarly, for fixed $k$, the graph with the fewest $k$-matchings is either the lex or the colex graph. To prove these results we first prove that the lex bipartite graph has the fewest matchings of all sizes among bipartite graphs with fixed part sizes and a given number of edges.'
author:
- 'L. Keough'
- 'A.J. Radcliffe'
bibliography:
- 'Matchings.bib'
title: Graphs with the Fewest Matchings
---
Introduction and statement of results
=====================================
In recent years there has been increased interest in extremal problems for “counting” parameters of graphs. A classic example is the Kahn-Zhao theorem, proved initially by Kahn [@Kahn] in the bipartite case, and then extended to the general case by Zhao [@Zhao].
\[thm:KahnZhao\] If $G$ is a $d$-regular graph then $\operatorname{ind}(G)$, the number of independent sets in $G$, satisfies $$\operatorname{ind}(G) \le \left(2^{d+1}-1\right)^{\frac{n}{2d}} = (\operatorname{ind}(K_{d,d}))^{\frac{n}{2d}}.$$
Such problems have been studied (extensively) for the number of perfect matchings in a graph: see for instance [@Alon; @KahnCuckler; @EntropyKahnLovasz; @Gross] and others. Usually these results concern regular graphs or graphs with a given degree sequence. In another vein there has been work on determining which graph with given numbers of vertices and edges maximizes or minimizes a given counting parameter. (We write ${\mathcal{G}}_{n,e}$ for the class of graphs having $n$ vertices and $e$ edges.) For instance Cutler and Radcliffe [@WidomRowlinson; @TheFox] determined which graphs have the largest number of homomorphisms in various fixed images graphs $H$. They also showed that the Kruskal-Katona Theorem implies that the lex graph has the greatest number of independent sets among graphs in ${\mathcal{G}}_{n,e}$. A classic paper of Ahlswede and Katona [@Ahlswede] determines the minimum number of pairs in non-incident edges a graph in ${\mathcal{G}}_{n,e}$ can have. (Which is of course the same as maximizing the number of pairs of incident edges.)
In this paper we solve the problem of determining which graph with $n$ vertices and $e$ edges has the fewest matchings. (A *matching* in a graph is simply a set of pairwise non-incident edges.) We denote the set of matchings in $G$ by ${\mathcal{M}}(G)$ and the set of $k$-matchings in $G$ by ${\mathcal{M}}_k(G)$. Also we write $m(G)=|{\mathcal{M}}(G)|$ and $m_k(G)= |{\mathcal{M}}_k(G)|$. In general our notation is standard and follows that of Bollobás [@MGT]. It turns out that, following the general approach of Ahlswede and Katona, we need to consider the class ${\mathcal{B}}_{\ell,r,e}$ of all bipartite graphs with $\ell$ vertices in the left part, $r$ vertices on the right, and having $e$ edges. We determine the graph in ${\mathcal{B}}_{\ell,r,e}$ having the fewest matchings. Our techniques also allow us to determine the graphs minimizing the number of matchings of size $k$, for all values of $k$, in ${\mathcal{B}}_{\ell,r,e}$ and in ${\mathcal{G}}_{n,e}$. In order to state our results we need to describe three collections of graphs: lex graphs, colex graphs, and lex bipartite graphs.
The *lexicographic order*, $<_L$, on finite subsets of ${\mathbb{N}}$ is defined by $A<_L B$ if $\min(A\Delta B)\in A$. The *colexigraphic order*, $<_C$, is defined by $A<_C B$ if $\max(A\Delta B) \in B$.
Restricting these orderings to $2$-subsets of $[n]$ results in the lex and colex orderings on $E(K_n)$. The first few edges in the lex ordering on $E(K_n)$ are $$\set{1,2},\set{1,3},\dots,\set{1,n},\set{2,3},\set{2,4},\dots,\set{2,n},\set{3,4},\dots$$ and the first few edges in the colex ordering on $E(K_n)$ are $$\set{1,2},\set{1,3},\set{2,3},\set{1,4},\set{2,4},\set{3,4},\set{1,5},\set{2,5},\dots$$
The *lex graph* ${\mathcal{L}}(n,e)$ is the graph with vertex set $[n]$ and edge set consisting of the first $e$ edges in the lex order on $E(K_n)$. Similarly, the *colex graph* ${\mathcal{C}}(n,e)$ is the graph with vertex set $[n]$ and edge set consisting of the first $e$ edges in the colex order on $E(K_n)$.
The graph below is ${\mathcal{L}}(7,8)$.
\(0) at (0,0) ; (1) at (1,0) ; (2) at (2,0) ; (3) at (3,0) ; (4) at (4,0) ; (5) at (5,0) ; (6) at (6,0) ;
\(0) circle (3pt) ; (1) circle (3pt) ; (2) circle (3pt) ; (3) circle (3pt) ; (4) circle (3pt) ; (5) circle (3pt) ; (6) circle (3pt) ;
=\[bend left\]
(0)(1)
(0)(2)
(0)(3)
(0)(4) (0)(5) (0)(6)
(1)(2)
(1)(3)
The graph below is ${\mathcal{C}}(7,8)$.
\(0) at (0,0) ; (1) at (1,0) ; (2) at (2,0) ; (3) at (3,0) ; (4) at (4,0) ; (5) at (5,0) ; (6) at (6,0) ;
\(0) circle (3pt) ; (1) circle (3pt) ; (2) circle (3pt) ; (3) circle (3pt) ; (4) circle (3pt) ; (5) circle (3pt) ; (6) circle (3pt) ;
=\[bend left\]
(0)(1)
(0)(2)
(1)(2)
(0)(3) (1)(3) (2)(3)
(0)(4)
(1)(4)
Our main result is that $m(G)$ and $m_k(G)$ are each minimized by either the lex graph or the colex graph.
\[general\] For all graphs $G$ with $n$ vertices and $e$ edges, and for all $k$, $$m(G) \geq \min\{ m({\mathcal{L}}(n,e)), m({\mathcal{C}}(n,e)\}$$ and $$m_k(G) \geq \min\{ m_k({\mathcal{L}}(n,e)), m_k({\mathcal{C}}(n,e)\}.$$
Suppose $n = \ell+ r$ with $\ell\leq r$ and $e\leq \ell r$. Write $e = qr+c$ where $0\leq c < r$. The *lex bipartite graph* with $e$ edges and partite sets $L$ and $R$ of size $\ell$ and $r$ respectively is the bipartite graph in which $q$ vertices in $L$ have degree $r$ and one vertex in $L$ has degree $c$. We will denote this graph by $L_{\ell,r}(e)$. Note that the lex bipartite graph contains the first $e$ edges of $E(L,R)$ in lex order.
A core part of the proof of Theorem \[general\] is to establish that, for all $k\geq 0$, $L_{\ell,r}(e)$ minimizes $m_k(G)$ in the class ${\mathcal{B}}_{\ell,r,e}$. It is key that there is a unique minimizer in the bipartite case.
\[bipartite\] Suppose $1\leq k\le \ell \leq r$ and $B\in {\mathcal{B}}_{\ell,r,e}$. Then $$m(B)\geq m(L_{\ell,r}(e))$$ and $$m_k(B)\geq m_k(L_{\ell,r}(e)).$$
To prove Theorem \[general\], we will first show that the graph attaining the minimum number of matchings is threshold. In Section \[Threshold\] we discuss the results we need concerning threshold graphs. In Section 3 we prove Theorem \[bipartite\], i.e., that there is a bipartite graph that simultaneously minimizes the number of matchings of each size. Finally, we use the bipartite case to show that the lex or colex graph minimizes $m(G)$ and $m_k(G)$ in the family ${\mathcal{G}}_{n,e}$ in Section 4.
Threshold Graphs {#Threshold}
================
Threshold graphs appear as an answer to many extremal questions, especially those in which it is advantageous to have all the edges “bunched together”. For instance in the class ${\mathcal{G}}_{n,e}$ threshold graphs maximize the number of independent sets and minimize homomorphisms into the Widom-Rowlinson graph [@WidomRowlinson].
There are many equivalent definitions of threshold graphs. The one that gives them their name is as follows.
A simple graph $G$ is a *threshold graph* if there exists a function $w: V(G) \to {\mathbb{R}}$ and a threshold $t\in {\mathbb{R}}$ such that $xy\in E(G)$ if and only if $w(x)+w(y)\geq t$.
The following lemma describes an alternative characterization that we will use in the proof of Theorem \[general\].
A graph $G$ is a threshold graph if and only if $V(G)$ can be partitioned into a clique and an independent set and moreover there exists a labeling $i_1,\dots, i_k$ of the vertices in the independent set such that $N(i_1) \supseteq N(i_2) \supseteq \dots \supseteq N(i_k)$. $\hfill$ $\blacksquare$
Note that it is immediate in this definition to see that the lex and colex graphs are threshold. In the colex case only (at most) one vertex in the independent set has any neighbors at all. In the lex graph case the clique consists (with one possible exception) of dominant vertices, so all the vertices in the independent set are joined either to all the vertices in the clique or all but one (and the one missing vertex is the same in each case). The following equivalent characterization of threshold graphs will help us define a way to move a graph towards being threshold.
\[thresholdnbhd\] A graph $G$ is threshold if and only if for all $x,y\in V(G)$ we have $N[x] \subseteq N[y]$ or $N[y]\subseteq N[x]$. $\hfill$ $\blacksquare$
We now define a compression move that makes a graph “more threshold". We will use this move later to show that we can find a graph that minimizes $m_k(G)$ in ${\mathcal{G}}_{n,e}$ that is threshold.
Let $G$ be a graph and $x$ and $y$ two vertices in $G$. Define $$N_G(x,{\overline}{y}) =\{v\in V(G)\setminus \{x,y\}: v\sim x,v\not\sim y\}.$$ Let $G_{x\to y}$ be the graph formed by deleting all edges between $x$ and $N_G(x,{\overline}{y})$ and adding all edges from $y$ to $N_G(x,{\overline}{y})$. This is called the *compression of $G$ from $x$ to $y$*. It is clear that $G_{x\to y}$ has the same number of edges as $G$.
We will begin our proof of Theorem \[general\] by repeatedly compressing a graph that minimizes the number of matchings. The following lemma will allow us to be sure that we are making progress, and not compressing round and round in a circle. The variance of the degree sequence, or (essentially equivalently) the quantity $$d_2(G) = \sum_{v\in V(G)} d(v)^2$$ strictly increases whenever we do a non-trivial compression.
\[lem:allcompressed\] Given $x,y\in V(G)$, $e(G_{x\to y}) = e(G)$ and $d_2(G_{x\to y})\geq d_2(G)$. If $N[x]\not\subseteq N[y]$ and $N[y] \not\subseteq N[x]$ then $d_2(G_{x\to y}) > d_2(G)$. $\hfill$ $\blacksquare$
\[cor:allcompressed2\] Suppose that ${\mathcal{G}}$ is a family of graphs on a fixed vertex set $V$ such that for any $G'\in{\mathcal{G}}$ and $x,y\in V$ we also have $G'_{x\to y}\in {\mathcal{G}}$. In addition suppose that $G$ satisfies $$d_2(G) = \max{\left\{d_2(G')\;:\;G'\in {\mathcal{G}}\right\}}.$$ Then $G$ is threshold.
Suppose $x,y\in V$ and $N_G[x]$ and $N_G[y]$ are incomparable. By hypothesis, $G_{x\to y}\in {\mathcal{G}}$ and by Lemma \[lem:allcompressed\] we know $d_2(G_{x\to y}) > d_2(G)$. This contradicts the assumption that $G$ attains the maximum value of $d_2$ in ${\mathcal{G}}$. Thus, $N_G[x] \subseteq N_G[y]$ or $N_G[y]\subseteq N_G[x]$ and so $G$ is threshold by Lemma \[thresholdnbhd\].
We will use the closely related topic of threshold bipartite graphs to prove Theorem \[bipartite\]. In [@OrangeBook] threshold bipartite graphs are defined with a similar vertex weighting.
\[thresholdbipartite\] A graph $G=(V,E)$ is said to be *threshold bipartite* if there exists a threshold $t$ and a function $w: V(G) \to {\mathbb{R}}$ such that $\abs{w(v)}<t$ for all $v\in V$ and distinct vertices $u$ and $v$ are adjacent if and only if $\abs{w(u)-w(v)}\geq t$.
Threshold bipartite graphs are called difference graphs in [@OrangeBook] and chain graphs in [@Yannakakis]. As with threshold graphs, there are many equivalent definition of threshold bipartite graphs. The following lemma describes the definition that will be most useful to us.
\[bipartiteinclusion\] A graph is threshold bipartite if and only if $G$ is bipartite and the neighborhoods of vertices in one of the partite sets can be linearly ordered by inclusion. $\hfill$ $\blacksquare$
Note that a threshold graph can be obtained from a threshold bipartite graph by adding all possible edges in one of the partite sets (on either side).
\[bipartited2\] Let $G$ be a bipartite graph with bipartition $(X,Y)$. Given $u,v\in X$, $e(G_{u\to v}) = e(G)$ and $d_2(G_{u\to v})\geq d_2(G)$. If $N(u)\not\subseteq N(v)$ and $N(u) \not\subseteq N(v)$ then $d_2(G_{u\to v}) > d_2(G)$.
Same calculations as in the proof of Lemma \[lem:allcompressed\].
\[bipartitecompress\] Suppose that ${\mathcal{G}}$ is a family of bipartite graphs on a fixed vertex set $V$ with a fixed bipartition $(X,Y)$ such that for any $G'\in{\mathcal{G}}$ and $u,v\in X$ we also have $G'_{u\to v}\in {\mathcal{G}}$. In addition suppose that $G$ satisfies $$d_2(G) = \max\{d_2(G'): G'\in {\mathcal{G}}\}.$$ Then $G$ is bipartite threshold.
Suppose that $u,v\in X$ such that $N_G(u) \not\subseteq N_G(v)$ and $N_G(v) \not\subseteq N_G(u)$. Then $G_{u\to v} \in {\mathcal{G}}$ by assumption and by Lemma \[bipartited2\] $d_2(G_{u\to v}) > d_2(G)$, a contradiction. Thus, $N_G(u) \subseteq N_G(v)$ or $N_G(v) \subseteq N_G(u)$ and so by Lemma \[bipartiteinclusion\] the graph $G$ is threshold bipartite.
Partitions
----------
We can relate threshold bipartite graphs to partitions and matchings in threshold bipartite graphs to rook placements in the Young diagram of the partition. We will make use of this connection heavily in the proof of Theorem \[bipartite\].
Given $(\ell,r,e)$, threshold bipartite graphs $G$ with vertex classes of size $\ell$ and $r$ and $\abs{E(G)} = e$ are in bijective correspondence with partitions of the integer $e$ with at most $\ell$ parts each of size at most $r$. From a partition $\lambda$ we can construct the associated bipartite graph $G_{\lambda}$ by letting $E(G_{\lambda}) = \{x_iy_j: j\leq \lambda_i\}$. Given a threshold bipartite graph we get a partition of $\abs{E(G)}$ by letting the degree of each vertex on the left be the size of a part. We will use this correspondence to represent threshold bipartite graphs as Young diagrams.
Let $B$ be a subset of $[\ell]\times [r]$. If $(i,j)\in B$ we call $(i,j)$ a *box* of $B$. We call $B$ a *Young diagram* if for all $(i,j)\in B$ with $i>1$ we have $(i-1,j)\in B$ and for all $(i,j)\in B$ with $j>1$ we have $(i,j-1)\in B$. We call $[\ell]
\times [r]$ the *frame* of the Young diagram and we’ll say that the Young diagram has dimensions $\ell\times r$. A *matching* in a Young diagram $B$ is a subset $M$ of $B$ such that for all $(a_1,b_1),(a_2,b_2)\in M$ we have $a_1\neq a_2$ and $b_1\neq b_2$.
Equivalently, a matching is a placement of non-attacking rooks on $B$. (A placement of non-attacking rooks is a placement of rooks such that no two rooks are in the same row or column.) The total number of ways to place non-attacking rooks is called the rook number. There is extensive literature on rook numbers, for example see [@Rooknumber]. We will use the language of rook placements in some of the proofs.
Matchings in a Young diagram $B$ correspond to a matchings in the bipartite graph associated with $B$ by equating the box $(i,j)\in B$ with the edge $x_iy_j$ in the associated bipartite graph.
Let the set of $k$-matchings in a Young diagram $B$ be denoted ${\mathcal{M}}_k(B)$ and let ${\mathcal{M}}(B) = \bigcup_{k\geq 0} {\mathcal{M}}_k(B)$. Also define $m_k(B)= \abs{{\mathcal{M}}_k(B)}$ and $m(B) = \abs{{\mathcal{M}}(B)}$.
Figure \[YD\] is an example of a Young diagram in a $4\times 6$ frame and Figure 2 is a matching of size 3 in that Young diagram. In all our diagrams we label rows and columns using matrix numbering.
[2]{}
(0,2) grid (6,3); (0,1) grid (5,2); (0,0) grid (3,1); (0,-1) grid (2,0); (0,-1) rectangle (6,3); (0.5,3.5) node [$1$]{}; (1.5,3.5) node [$2$]{}; (2.5,3.5) node [$3$]{}; (3.5,3.5) node [$4$]{}; (4.5,3.5) node [$5$]{}; (5.5,3.5) node [$6$]{}; (-0.5,2.5) node [$1$]{}; (-0.5,1.5) node [$2$]{}; (-0.5,0.5) node [$3$]{}; (-0.5,-0.5) node [$4$]{};
(0,2) grid (6,3); (0,1) grid (5,2); (0,0) grid (3,1); (0,-1) grid (2,0); (.5,.5) circle (5pt); (2.5,2.5) circle (5pt); (4.5,1.5) circle (5pt); (0,-1) rectangle (6,3); (0.5,3.5) node [$1$]{}; (1.5,3.5) node [$2$]{}; (2.5,3.5) node [$3$]{}; (3.5,3.5) node [$4$]{}; (4.5,3.5) node [$5$]{}; (5.5,3.5) node [$6$]{}; (-0.5,2.5) node [$1$]{}; (-0.5,1.5) node [$2$]{}; (-0.5,0.5) node [$3$]{}; (-0.5,-0.5) node [$4$]{};
Bipartite Minimizer
===================
In this section we prove Theorem \[bipartite\]. We are motivated to look at the bipartite case by the following lemma.
\[minimize\] Given a threshold graph $G$ with vertex set $V$, let $V = K \cup (V\setminus K)$ be a partition of the vertex set such that $G[K]$ forms a clique and $G[V\setminus K]$ forms an independent set. Suppose $\abs{K} = s$. Let $B$ be the bipartite graph with partite sets $K$ and $V\setminus K$ and edge set $E(K, V\setminus K)$. Then $$m(G) = \sum_{k\geq 0} m_k(B) \cdot m(K_{s-k}).$$
Fixing a matching $M$ in $B$ of size $k$, there are exactly $m(K_{s-k})$ matchings in $G$ that contain $M$. So there are $m_k(B) \cdot m(K_{s-k})$ matchings in $G$ that contain a $k$-matching in $B$. Summing over all possible sizes of matchings in $B$ we count every matching in $G$ exactly once.
Theorem \[bipartite\] states that for a given $k$ there is a graph that simultaneously minimizes every $m_k(G)$ in ${\mathcal{B}}_{\ell,r,e}$. We will first show that there is a threshold bipartite graph that minimizes $m_k(G)$ in ${\mathcal{B}}_{\ell,r,e}$. We will then use moves on the associated Young diagrams to show that the lex bipartite graph $L_{\ell,r}(e)$ minimizes $m_k(G)$ for all $k$.
\[differencegraph\] Suppose $G$ is a bipartite graph with bipartition $(X,Y)$. Let $u,v \in X$. Then for every $k\geq 0$ the graph $H:= G_{u \to v}$ has at most as many $k$-matchings as $G$.
We will construct an injection from ${\mathcal{M}}(H)\setminus {\mathcal{M}}(G)$ to ${\mathcal{M}}(G)\setminus {\mathcal{M}}(H)$ that preserves size. It then follows that $m_k(H)\leq m_k(G)$ for all $k$. We define a replacement function $r: E(H) \to E(G)$ by $$r(e) = \begin{cases}
uy &\text{ if } e=vy \text{ for } y \in N_G(u,{\overline}{v})\\
vz &\text{ if } e=uz\\
e &\text{otherwise}
\end{cases}.$$
Given $e$ in the edge set of $H$, we claim that $r(e)$ is an edge in $G$. If $y\in N_G(u,{\overline}{v})$, then $uy\in E(G)$. Also, if $uz\in H$ then $z\in N_G(u)\cap N_G(v)$ and so $vz\in E(G)$. Finally, if $e\neq vy$ for $y\in N_G(u{\overline}{v})$ then $e\in E(G)\cap E(H)$.
Now define $\phi: {\mathcal{M}}(H)\setminus{\mathcal{M}}(G) \to {\mathcal{M}}(G)\setminus{\mathcal{M}}(H)$ by $$\phi(M) = \{r(e): e\in M\}$$
Given $M\in {\mathcal{M}}(H)\setminus{\mathcal{M}}(G)$ note that $\phi(M) \subseteq E(G)$ since $r(e)\in E(G)$ for all $e\in E(H)$. We claim that in fact $\phi(M)\in {\mathcal{M}}(G)\setminus{\mathcal{M}}(H)$. For the first case, suppose that $u\in r(e) \cap r(f)$. Note that if $u\in r(e)$ then $v\in e$ since edges in $H$ containing $u$ are replaced by edges in $G$ containing $v$ instead. So if $r(e) \cap r(f) = u$ then $e\cap f = v$, a contradiction since $e,f\in M$, a matching.
Now suppose that $e, f\in M$ and $r(e) \cap r(f) = v$. There are two possible ways for $v$ to be in $r(e)$. The first is for $u$ to be in $e$ and the second is for $e=va$ for some $a\notin
N_G(u,{\overline}{v})$. However, since $M\in {\mathcal{M}}(H)\setminus{\mathcal{M}}(G)$ it must be the case that $vy\in M$ for some $y \in N_G(u,{\overline}{v})$. Since $M$ is a matching, $vy$ is the only edge incident to $v$. Therefore, $e$ and $f$ must both contain $u$, a contradiction since $M$ is a matching.
Finally suppose that $z\in r(e) \cap r(f)$ for some $z\neq u,v$. Then $z\in e\cap f$, a contradiction. Thus each vertex has at most one incident edge and $\phi(M)$ is a matching. Note that $\phi(M)\notin {\mathcal{M}}(G)$ since $uy\in \phi(M)$ for some $y\in N_G(u,{\overline}{v})$ and no such edge is in $E(G)$. So $\phi(M) \in {\mathcal{M}}(H)\setminus {\mathcal{M}}(G)$.\
To finish the proof of the lemma we need only show that $\phi$ is an injection. We’ll show that $\phi$ has a left inverse defined similarly to $\phi$. Consider $r': E(G) \to E(H)$ defined by $$r'(e)= \begin{cases}
vy &\text{ if } e = uy \text{ for any } y \in N_G(u,{\overline}{v})\\
uz &\text{ if } e = vz\\
e &\text{otherwise}
\end{cases}.$$ Define $\phi': {\mathcal{M}}(G)\setminus {\mathcal{M}}(H) \to {\mathcal{M}}(H)\setminus {\mathcal{M}}(G)$ by $\phi'(M) = \{r'(e): e\in M\}$. It is straightforward to check that $\phi'(\phi(M))= M$. Thus $\phi$ has a left inverse and so $\phi$ is injective.
By Corollary \[bipartitecompress\] and Lemma \[differencegraph\], a bipartite graph minimizing the total number of matchings can be found among the threshold bipartite graphs.
Say $P$ is an *out-corner* of a Young diagram $B$ if $P \in B$ and there is no box in the diagram to its right or beneath it. Say $Q$ is an *in-corner* if a box can be added there to create an out-corner.
\[corners\]
(0,2) grid (6,3); (0,1) grid (5,2); (0,0) grid (3,1); (0,-1) grid (2,0); (0,3) rectangle (6,-2); (5.5,2.5) node [O]{}; (4.5,1.5) node[O]{}; (2.5,.5) node[O]{}; (1.5,-.5) node[O]{}; (5.5, 1.5) node [I]{}; (3.5,.5) node [I]{}; (2.5,-.5) node [I]{}; (.5,-1.5) node [I]{};
In Figure \[corners\] the in-corners are labeled with $I$ and the out-corners are labeled with $O$. Notice that the dimensions of the frame matter; for example, in a $4\times 6$ frame $(5,1)$ would not be an in-corner.
We now define a move that removes a box that is an out-corner and puts a box at an in-corner that is “further out". Let $s(P)$ be the sum of the coordinates of $P$.
\[oneblock\] Let $B$ be a Young diagram in an $m\times n$ frame. Suppose $P$ is an out-corner of $B$ and $P'$ is an in-corner of $B$. If $s(P)<
s(P')$ then $B':=B+P'-P$ is a Young diagram in an $m\times n$ frame and has at most as many $k$-matchings as $B$ for all $k\geq 0$. Moreover, $m(B')<m(B)$.
Suppose $P=(i,j)$ and $P'=(i',j')$. The hypothesis states that $i+j<i'+j'$. Define $B^+ = B+P' = B'+P$. Note that $B^+$ is a Young diagram since $P'$ is an in-corner. Showing that $B'$ has fewer matchings than $B$ is equivalent to showing that there are fewer matchings in $B^+$ that contain $P'$ and not $P$ than matchings in $B^+$ that contain $P$ and not $P'$. To do this we will define an injection from the collection of $k$-matchings of $B^+$ that contain $P'$ to the collection of $k$-matchings of $B^+$ that contain $P$.
Define $S:={\left\{(a,j)\;:\;i'<a<i\right\}}$ and $T:={\left\{(i',b)\;:\; j<b<j'\right\}}$. The injection will be defined in two parts: firstly on those matchings of $B$ that don’t intersect $S$ and secondly on those that do.
(-1,0) rectangle (10,6);
(-1.5, 1.75) node [$i$]{}; (1.75 , 6.5) node [$j$]{}; (-1.5, 4.25) node [$i'$]{}; (6.75, 6.5) node [$j'$]{};
(2,4) grid (3,4.5); (2,4) – (3,4); (2,4) – (2,4.5); (4,4.25) – (5,4.25); (6,4) grid (7,4.5); (6,4) – (6,4.5); (6,4) – (7,4); (2.5,4.5) – (6.5,4.5) node \[black,midway,yshift=.5cm\][$T$]{}; (6.75,4.25) node [$P'$]{};
(1.5,3.5) grid (2,4); (1.5, 3.5) – (1.5,4); (1.5,3.5) – (2,3.5); (1.75,2.75) – (1.75,3.25); (1.5,1.5) grid (2,2.5); (1.5,1.5) – (1.5,2.5); (1.5,1.5) – (2,1.5); (1.5,2) – (1.5,4) node \[black,midway,xshift=-0.6cm\][$S$]{}; (1.75,1.75) node [$P$]{};
(-1,.5) – (-.5,.5); (-.5,.5) – (-.5,1); (-.5,1) – (0,1); (0.2,1) – (.8,1.2); (1,1.5) – (1.5,1.5); (1,1.25) – (1,1.5);
(2,2) – (3,2); (3,2) – (3,2.5); (3.5,2.7) – (4.7,3.2); (6,3.5) – (6,4); (5.5,3.5) – (6,3.5);
(7,4.5) – (7.5,4.5); (7.5,4.5) – (7.5,5); (7.7, 5.2) – (8.2,5.4);
In Figure \[boardB+\] we have drawn $P'$ up and to the right of $P$. It is also possible that $P'$ is down and to the left of $P$. In this case the same proof will work by using the transpose of $B^+$.
Let $$A_{{\overline}{S}}: = \{M\in M(B^+): P'\in M, P \notin M, S\cap M=\emptyset\},$$ $$A_{{\overline}{T}}: = \{M\in M(B^+): P\in M, P'\notin M T\cap M=\emptyset\},$$ $$A_{S}: = \{M\in M(B^+): P'\in M, P\notin M S\cap M\neq\emptyset\},$$ $$A_{T}: = \{M\in M(B^+): P\in M, P'\notin M, T\cap M\neq\emptyset\},$$ We first define a bijection between $A_{{\overline}{S}}$ and $A_{{\overline}{T}}$ and then we define an injection from $A_S$ to $A_T$. For each case, we will define a replacement function $r_i$ on the blocks of $B^+$ and then an injection $f_i$ on the appropriate matchings. In the end we will have $$A_{{\overline}{S}}\overset{f_1}{\longleftrightarrow}A_{{\overline}{T}}$$ $$A_{S} \overset{f_2}{\hookrightarrow} A_T.$$ Since each map will send matchings of size $k$ to matchings of size $k$, we conclude $m_k(B')\leq m_k(B)$ for all $k \geq 0$.
*Case 1:* Suppose $M\in A_{{\overline}{S}}$, that is, $M$ is a matching in $B^+$ such that $P'\in M$, $P\notin M$, and $S\cap M =\emptyset$. Define $r_1: B^+\to B^+$ by $$r_1(a,b):=\begin{cases}
(i,j) &\text{ if } a = i', b = j'\\
(i',b) &\text{ if } a = i\\
(a,j') &\text{ if } b = j\\
(a,b) &\text{ otherwise}
\end{cases}.$$
We can think of $r_1$ as sending the rook at $P'$ to $P$, and then projecting rooks in row $i$ and column $j$ onto row $i'$ and column $j'$ respectively. We claim $r_1(a,b)\in B^+$ for all $(a,b)\in B^+$. Since $P = (i,j)$ is an out-corner of $B$, if $(i,b)\in B^+$ then $b\leq j<j'$. Because $B^+$ is a Young diagram and $P' = (i',j')\in B^+$, we have $r_1(i,b) = (i',b)\in B^+$. In this case $M\cap S = \emptyset$ so a rook of the form $(a,j)$ with $a\neq i$ must have $a<i'$. Since $B^+$ is a Young diagram and $P' = (i',j')\in B^+$, we know $r_1(a,j) = (a,j')$ is in $B^+$. Therefore, $r_1(a,b)\in B^+$ for all $(a,b)\in B^+$.
Define $f_1: A_{{\overline}{S}} \to A_{{\overline}{T}}$ by $$f_1(M) = \{r_1(a,b): (a,b)\in M\}.$$ First, we show that given $M\in A_{{\overline}{S}}$, $f_1(M)$ is in fact in $A_{{\overline}{T}}$. Sending the rook in $P'$ to $P$ causes conflicts only for rooks in row $i$ and column $j$. We have solved the problem of a rook in row $i$ since we changed the row of this rook to $i'$. Note that row $i'$ is otherwise unoccupied since $P'\in M$ and $M$ is a matching. Similarly, we have solved the problem of a rook in column $j$ since we changed the column of this rook to $j'$. This column is otherwise unoccupied since $M$ is a matching and $P'\in M$. Thus, $f_1(M)$ is a matching in $B$. Moreover, $f_1(M)$ has no rooks in $T$ and so $f_1(M) \in A_{{\overline}{T}}$.
In addition, $f_1$ is injective. Define $r_1': B^+ \to B^+$ by $$r_1'(a,b):=\begin{cases}
(i',j') &\text{ if } (a,b) = (i,j)\\
(i,b) &\text{ if } a = i', b\neq j\\
(a,j) &\text{ if } a\neq i, b = j'\\
(a,b) &\text{otherwise}
\end{cases}$$ and define $f_1': A_{{\overline}{T}} \to A_{{\overline}{S}}$ by $$f_1'(M) = \{r_1'(a,b): (a,b)\in M\}.$$ It is straightforward to check that $f_1'(f_1(M))= M$ for all $M\in A_{{\overline}{S}}$ and $f_1(f_1'(M))= M$ for all $M\in A_{{\overline}{T}}$. Thus, there is a bijection between the matchings in $B^+$ with $P'$ and no rooks in $S$ and matchings in $B^+$ with $P$ and no rooks in $T$.
*Case 2:* Suppose $M\in A_S$. That is, $M$ is a matching in $B^+$ with a rook in $P'$ and a rook in $S$. Define $E:= \{(a,b): (a,b)\in B^+, a>i', b>j\}$. In Figure \[boardB+\] this is collection of blocks that are both to the right of $P$ and below $P'$. Let $S^*(M)\subset \{i'+1,\dots,i-1\}$ be the rows of blocks in $S$ that do not share a row with any rooks of $M$ that are in $E$. Similarly, let $T^*(M)\subset\{j+1,\dots,j'-1\}$ be the columns of blocks in $T$ that do not share a column with any rooks of $M$ that are in $E$. We claim that $\abs{S^*(M)}<\abs{T^*(M)}$. Note that $\abs{S} = i-i'-2$ and $\abs{T} = j'-j-2$. Since $i+j< i'+j'$ we have $i-i'< j-j'$ and so $\abs{S}< \abs{T}$. Letting $a$ be the number of rooks in $E$, then $\abs{S^*(M)} = \abs{S} - a$ and $\abs{T^*(M)} = \abs{T} - a$. So $\abs{S^*(M)}< \abs{T^*(M)}$. Thus, there is an injection $s: S^*(M)\to T^*(M)$. We fix some arbitrary injection $s$ and let $(a_0,j)$ be the location of the rook in $S$. Define $r_2: B^+ \to B^+$ by $$r_2(a,b):=
\begin{cases}
(i,j) &\text{ if } (a,b) = (i',j')\\
(i',s(a)) &\text{ if } (a,b)\in S\\
(a_0,b) &\text{ if } a=i\\
(a,j') &\text{ if } b=s(a_0)\\
(a,b) &\text{otherwise}
\end{cases}.$$
In Figure \[r2sketch\] the gray boxes are the images of the black boxes under the map $r_2$. We can think of $r_2$ as sending the rook in $P'$ to $P$ (arrow 1), sending the rook in $S$ to a place in $T$ via $s$ (arrow 2), and then projecting rooks in conflicting rows and columns to rows and columns that are known to be unoccupied (arrows 3 and 4).
\(P) at (0,-1); (P’) at (2,0); (S) at (0,-.5); (S’) at (1,0); (V) at (-1,-1); (V’) at (-1,-.5); (H) at (1,.5); (H’) at (2,.5);
(-.1,-1.3) node [$P$]{}; (P’) node [$P'$]{}; (-2,-.1) node [$i'$]{}; (-2,-1.1) node [$i$]{};
(1.9,-.2) to (0,-1.15); (1,-.75) node [$1$]{}; (0,-.35) to (.7,-.2); (.35, -.4) node [$2$]{}; (-1.3,-1) to (-1.3,-.4); (-1.4,-.7) node[$3$]{}; (.6,.6) rectangle (.8,.8); (.8,.7) to (1.8,.7); (1.3,.8) node [$4$]{};
(-.2,-1) grid (0,-.8); (-.1,-.7) – (-.1,-.5); (-.2,-.4) rectangle (0,-.2); (-.2,-.4) grid (0,0); (-.2,0) grid (0,.4); (-.1,.5) – (-.1,.7); (-.2,.8) grid (0,1); (-.2,.8) – (0,.8);
(1.8,0) grid (2,.2); (1.9,.3) –(1.9,.5); (1.8,.6) rectangle (2,.8); (1.8,.6) grid (2,1); (1.8,.6) – (1.8,1); (1.8,.6) – (2,.6); (1.8,0) – (1.8,.2);
(-2,-.2) grid (-1.2,0); (-1.1,-.1) – (-.9,-.1); (-.8,-.2) grid (0,0); (.6,-.2) rectangle (.8,0); (0,-.2) grid (.8,0); (1,-.1) – (1.2,-.1); (1.8,-.2) rectangle (2,0); (1.4,-.2) grid (2,0); (1.4,-.2) – (1.4,0); (-1.4,-.4) rectangle (-1.2 ,-.2); (-1.4,-.4) rectangle (-1.2 ,-.2);
(-1.4,-1.2) rectangle (-1.2,-1); (-2,-1.2) grid (-1.2,-1); (-1,-1.1) – (-.8,-1.1); (-.2,-1.2) rectangle (0,-1); (-.6,-1.2) grid (0,-1);
Again we need to show that $r_2(a,b) \in B^+$ for all $(a,b)\in B^+$. If $(a,b)\in S$ then $(i',s(a))\in B^+$ as $(i',s(a))\in T$. If the rook $(a,b)$ is in row $i$ then $r_2(a,b) = (a_0,b)\in B^+$ as $a_0 <i$ and $(i,b)\in B^+$. Finally, if $(a,b) = (a,s(a_0))$, then $r_2(a,b) = (a,j') \in B$ since $a<i'$ and $(i',j')$ is in $B^+$.
For $M\in A_S$, define $$f_2(M):=\{r_2(a,b): (a,b)\in M\}.$$ We claim that if $M$ is a matching in $A_S$ then $f_2(M)$ is a matching in $A_T$. First we will show that no two rooks are in the same row. There are only three rooks that change rows when we apply $r_2$. These rooks originally have rows $i'$, $i$ and $a_0$. After applying $r_2$ these rooks are in rows $i$, $a_0$ and $i'$ respectively. Thus the rooks of $f_2(M)$ occupy the same collection of rows as those of $M$. Now we must show that no two rooks occupy the same column. Similarly, there are only three rooks that change columns. These rooks originally occupy columns $j',j,$ and $s(a_0)$. After applying $r_2$, these rooks occupy $j,s(a_0),$ and $j'$ respectively. So the rooks of $f_2(M)$ occupy the same collection of columns as those of $M$ and so no two rooks are in the same column. Finally, $T\cap f_2(M) \neq \emptyset$ since the rook in $S$ got sent to a rook in $T$. Thus $f_2(M)$ is a matching in $A_T$.
Using $f_1$ and $f_2$ we know where to send all matchings in $B^+$ that have a rook in $P'$ and not $P$. Moreover, the images of $f_1$ and $f_2$ are disjoint and thus there is an injection from matchings in $B^+$ with a rook in $P'$ and not $P$ to matchings in $B^+$ that have a rook in $P$ and not $P'$. Since this injection preserves the size of the matching, $m_k(B') \leq m_k(B)$.
It remains to prove that $B'$ has strictly fewer matchings than $B$. Suppose that there are no rooks of $M$ that are in $E$ and fix an injection $s: S^*(M)\to T^*(M)$ Since $\abs{S^*(M)}<\abs{T^*(M)}$ there exists $Q\in T^*(M)$ such that $Q\neq s(R)$ for any $R\in S^*(M)$. Then $\{Q,P\}\in A_T$ and there does not exist $M\in A_S$ such that $f_2(M) = \{Q,P\}$. Thus, $m(B')<m(B)$.
Given a Young diagram $B$ with an out-corner $P$ and an in-corner $P'$ with $s(P)<s(P')$ we call the move that results in $B-P+P'$ an *out-block move*.
There are examples of Young diagrams that are not the Young diagrams associated to $L_{\ell,r}(e)$ that have no out-block moves. For example, see Figure \[YDNOB\]. For this reason we are forced to introduce an additional move. It is clear that taking the transpose of a Young diagram preserves $m_k(B)$ for all $k$. The following definition describes a way that we can transpose a piece of a Young diagram.
Let $B$ be a Young diagram in an $\ell\times r$ frame and let $(i,j)\in B$. Define the *transpose of $B$ at $P = (i,j)$* to be the diagram $$B_{P}^* := \{(a,b)\in B: a<i \text{ or } b<j\} \cup \{ (a,b)^*: (a,b)\in B, a\geq i, \text{ and } b\geq j\}$$ where $(a,b)^* = (b-j+i,a-i+j)$. Call transposing at $P$ [*legal*]{} if $B_{P}^*$ is a Young diagram in an $\ell\times r$ frame.
Note that the definition of $(a,b)^*$ depends on $(i,j)$, the place we are transposing, but we suppress this in the notation.
Figures \[YDNOB\] and \[Transposed\] are two Young diagrams in a $4\times 5$ frame. To get the Young diagram in Figure \[Transposed\] from the Young diagram in Figure \[YDNOB\] we transpose at $P= (1,2)$
[2]{}
(0,0) rectangle (5,4); (0,0) grid (3,4); (1.5,3.5) circle (3pt);
(0,0) rectangle (5,4); (0,2) grid (5,4); (0,0) grid (1,2);
We will now prove that legally transposing at $P\in B$ preserves $m_k$ for all $k\geq 0$.
\[transpose\] Let $B$ be a Young diagram and let $P\in B$. Performing a legal transpose at $P$ preserves the number of matchings of all sizes.
Suppose that the sub-board to be transposed has dimensions $x\times y$. Write $P = (i,j)$. Without loss of generality, let $\max\{x,y\}=x$. We single out the following pieces of $B$ as sketched in Figure \[SixPieces\] : $$\begin{aligned}
T &= {\left\{(a,b)\;:\;(a,b)\in B, a\geq i, b\geq j\right\}} &&\text{the \bf{T}ransposed portion}\\
U &= {\left\{(a,b)\;:\; a<i,j\leq b \leq j+x\right\}} &&\text{the portion \bf{U}p from $T$}\\
L &= {\left\{(a,b)\;:\;i\leq a \leq i+x, b<j\right\}} &&\text{the portion to the \bf{L}eft of $T$}\\\end{aligned}$$
(0,0) rectangle (8,6);
(0,.5) parabola (3,2); (3,2) parabola (4.5,4); (5,4) parabola (8,5); (0,.5) – (0,6); (0,6) – (8,6); (8,6) – (8,5); (4.5,4) – (5,4);
(0,2) – (3,2); (3,2) – (3,6); (0,4) – (5,4); (5,4) – (5,6);
(3,2) – (5,2); (4,1.5) node [$x$]{}; (5,2) – (5,4); (5.5,3) node [$x$]{};
(4,5) node [$U$]{}; (1.5,3) node [$L$]{}; (3.75,3.25) node [$T$]{}; (-.5,4) node [$i$]{}; (3,6.5) node [$j$]{};
(3,4) circle (2pt);
Fix a matching in $T$, call it $M_T$. Define $T^*:=\{(a,b)^*: (a,b)\in T\}$ and let $M_T^*$ be the matching in $T^*$ obtained by transposing the location of each of the rooks. We will define an injection from matchings in $B$ that contain $M_T$ to matchings in $B^*_P$ that contain $M_T^*$. Doing this for every matching in $T$ will show that there are at most as many matchings in $B$ as in $B_{P}^*$. A similarly defined injection will work to conclude that $B_{P}^*$ has at most as many matchings as $B$. Since these injections will preserve the size of each matching we will conclude that $m_k(B) = m_k(B_P^*)$.
Let $U_1\subseteq \{j,\dots, j+r\}$ be the set of column indices between $j$ and $j+r$ that are unoccupied by a rook in $M_T$ and similarly let $U_2\subseteq\{j,\dots, j+x\}$ be the set of column indices between $j$ and $j+x$ that are unoccupied by a rook in $M_T^*$. If $\abs{M_T}=t$ then $\abs{U_1} = \abs{U_2} = x+1 - t$. Thus, there is a bijection $u: U_1\to U_2$.
Similarly, let $L_1\subseteq \{i,\dots,i+x\}$ be the set of row indices between $i$ and $i+x$ that are not occupied by a rook in $M_T$ and let $L_2\subseteq\{i,\dots, i+x\}$ be the set of row indices between $i$ and $i+x$ of rows that are not occupied by a rook in $M_T^*$. Again, $\abs{L_1}=\abs{L_2} = x+1 - t$ where $t$ is the size of $M_T$ and hence there is a bijection $l: L_1 \to L_2$.
Define $r: B \to B^*_P$ by $$r(a,b):=
\begin{cases}
(a,b)^* &\text{ if } (a,b) \in T\\
(l(a),b) &\text{ if } (a,b)\in L, a \in L_1\\
(a,u(b)) &\text{ if } (a,b) \in U, b \in U_2\\
(a,b) &\text{ otherwise}
\end{cases}.$$
Define $f$ from matchings in $B$ containing $M_T$ to matchings in $B^*_P$ containing $M_T^*$ by $$f(M) = \{ r(a,b): (a,b) \in M\}.$$ First we note that given $M\in {\mathcal{M}}(B)$ we have $f(M) \in {\mathcal{M}}(B^*_P)$. We know $f(M) \subset B^*_P$ since $r(a,b)\in B^*_P$ for all $(a,b) \in B$. For each rook, $(a,b)\in M_T$ we send $(a,b)$ to $(a,b)^*$. Conflicts are only caused in rows $i,i+1,\dots, i+r$ and columns $j,j+1,\dots j+r$. These conflicts are resolved using the injections $l$ and $u$ which send all rooks in $L$ and $U$ to rows and columns unoccupied by $M_T^*$. This causes no additional conflicts since no additional rows or columns are changed. Thus $f(M) \in {\mathcal{M}}(B^*_P)$. Moreover, for a matching $M\in {\mathcal{M}}(B)$ containing $M_T$, $f(M)$ contains $M_T^*$.
We claim that $f$ is a bijection. Define $$r'(a,b):=
\begin{cases}
(a,b)^* &\text{ if } (a,b) \in T\\
(l^{-1}(a),b) &\text{ if } (a,b)\in L, a \in L_2\\
(a,u^{-1}(b)) &\text{ if } (a,b) \in U, b \in U_2\\
(a,b) &\text{ otherwise}
\end{cases}.$$ Define $f'$ from matchings in $B^*_P$ containing $M_T$ to matchings in $B$ containing $M_{T}^*$ by $$f'(M) = \{r'(a,b): (a,b)\in M\}.$$ It is straightforward to check that $f'$ is actually the inverse of $f$. Thus, $f$ is a bijection and $m_k(B) = m_k(B^*)$.
The next lemma shows how we piece together the out-block move and the transpose move. First we define the lex order on Young diagrams.
To define the lex order on Young diagrams, we first define an ordering on ordered pairs. Say $(a,b)\lesssim(c,d)$ if $a<c$ or $a=c$ and $b<d$. Then the *lex order* $<_L$ on Young diagrams is defined by $B<_L B'$ if and only if $\min_{\lesssim}(B\Delta B')\in B$.
Note that by this definition, $L_{\ell,r}(e)$ is least among all Young diagrams in $\ell\times r$ frames. The next lemma states that if we can’t find an out-block move we can find a legal transpose that moves a board to one earlier in lex order.
\[transposeupinlex\] Consider a Young diagram $B$ in an $\ell\times r$ frame where $\ell\leq r$ that has no out-block moves and is not the Young diagram of $L_{\ell,r}(e)$. There exists $P\in B$ such that the transpose at $P$ is legal and $B^*_P<_L B$.
First we set up some notation. For $P = (i,j)\in B$, let $\rho(P) = i$ (so $\rho(P)$ is the row of $P$) and let $c(P) = j$ (so $c(P)$ is the column of $P$). For $P,Q\in B$, define $v(P,Q) = \abs{\rho(P) - \rho(Q)}$, so $v(P,Q)$ is the vertical distance between $P$ and $Q$. In the Young diagram $B$ let $P$ be the out-corner with $\rho(P)$ greatest among all out-corners and $Q$ be the in-corner of $B$ with $\rho(Q)$ least among all in-corners. If legal, transpose at $S=(\rho(Q),c(P))$. If transposing at $S$ is not legal then we “count back" from the right hand limit of the partition so that the vertical distance between $P$ and $Q$ will fit. In more detail, transpose at $S' = (\rho(Q), r - v(P,Q))$ whenever transposing at $S$ is not legal. We claim that this gives a place to transpose legally and that it results in a board that is earlier in lex order than $B$. Call the result of the performed transpose $B^*$.
First suppose that transposing at $S$ is legal. Note that $S\in B$ as $B$ is a Young diagram and $P\in B$. If we transpose at $S$ then it is because the transpose at $S$ is legal. The first row where $B^*$ is different from $B$ is $\rho(Q)$. In $B$ this row has $c(Q)-1$ blocks. In $B^*$ this row has $c(P) + v(P,Q) = c(P) + \rho(P) - \rho(Q)$ blocks. (Since $B$ is not the Young diagram of $L_{\ell,r}(e)$ we know $\rho(P)>\rho(Q)$ and so may remove the absolute value signs in $v(P,Q)$.) Since $B$ has no out-block moves, we know $s(P)>s(Q)$, i.e, $\rho(P) + c(P) > \rho(Q) + c(Q)$. Thus $c(Q) < c(P) + \rho(Q) - \rho(Q)$ and so $B^*>_L B$.
Now suppose that transposing at $S$ is not legal. We claim that that $S' =(\rho(Q), r- v(P,Q))\in B$. Because transposing at $S$ is not legal, we know $c(Q) + v(P,Q) >r$. Thus, $r-v(P,Q)< c(Q)$. Moreover $r-v(P,Q)\geq 1$ as $\ell\leq r$ and $v(P,Q)<\ell$. Using that $Q$ is an in-corner of $B$, we get that $S'\in B$. We need to show that the transpose is legal. Intuitively, the transposed piece fits horizontally because we “counted back" far enough to make it so. Algebraically, the length of the row $\rho(Q)$ in $B^*$ is $r-v(P,Q) + v(P,Q) = r$ which is exactly the length that can fit. We’ll show that the transposed piece fits vertically after we show that $B^*_S>_L B$. Note the first place $B$ and $B^*$ differ is in row $\rho(Q)$ and row $\rho(Q)$ in $B^*$ has $r$ blocks while row $\rho(Q)$ in $B$ has strictly fewer than $r$ blocks since $Q$ is an in-corner. In particular, this means, if the dimensions of the transpose section are $s\times t$, we have $s\geq t$. Thus the transposed piece will fit vertically. Therefore, transposing at one of $S$ or $S'$ will result in a legal transpose that moves $B$ earlier in lex order.
We are now ready to prove Theorem \[bipartite\]. The proof follows easily from Lemmas \[differencegraph\] and \[transposeupinlex\].
Suppose $G$ is a bipartite graph that is not equal to $L_{\ell,r}(e)$. If $G$ is not bipartite threshold then we can apply Lemma \[differencegraph\] to get another graph with the same number of vertices and edges, but at most as many matchings. If $G$ is bipartite threshold, consider the associated Young diagram $B_G$. If $B_G$ has no out-block moves and $G\neq L_{l,r}(e)$ then by Lemma \[transposeupinlex\] there is a legal transpose at $(i,j)$ that moves the associated board earlier in lex order. This move preserves the number of matchings by Lemma \[transpose\]. Thus, $L_{l,r}(e)$ attains the minimum number of matchings of all sizes. This concludes the proof of Theorem \[bipartite\].
Proof of Theorem \[general\]
============================
Recall that our main result states that either the lex or colex graph minimizes the number of matchings (of size $k$) in ${\mathcal{G}}_{n,m}$. In this section we will prove this. First we will show that there is a graph attaining the minimum that is threshold. Finally, we will use the bipartite case to complete the proof.
\[compressing\] For all graphs $G$ and all $x,y\in V(G)$ $$m_k(G_{x\to y}) \leq m_k(G).$$
Let $H:=G_{x\to y}$. As in the proof of Lemma \[differencegraph\] we will construct an injection $\phi$ from ${\mathcal{M}}_k(H)\setminus {\mathcal{M}}_k(G)$ to ${\mathcal{M}}_k(G)\setminus {\mathcal{M}}_k(H)$ from which it follows that $m(H)\leq m(G)$. Let $A = E(x,N_G(x,{\overline}{y}))\subset E(G)$ and $B=E(y,N_G(x,{\overline}{y}))\subset E(H)$. Then $H=G-A+B$. So $${\mathcal{M}}_k(H)\setminus {\mathcal{M}}_k(G)=\{M\in {\mathcal{M}}_k(H): M\cap B\neq \emptyset\}$$ and similarly $${\mathcal{M}}_k(G)\setminus {\mathcal{M}}_k(H)=\{M\in {\mathcal{M}}_k(G): M\cap A\neq \emptyset\}$$ Define a replacement function $r:E(H)\to E(G)$. Let $$r(e):=\begin{cases}
e \Delta\{x,y\} &\text{ if } e\in B \\
yz &\text{ if } e=xz, z\neq y\\
e &\text{ otherwise }
\end{cases}.$$
For each $e\in E(H)$ note that $r(e)\in E(G)$. If $e\in B$ then $r(e) = e\Delta \{x,y\}\in A\subset E(G)$. If $e=xz\in H$ and $z\neq y$ then $z\in N_G(x)\cap N_G(y)$ and so $\phi(e) = yz\in E(G)$. Finally, if $e\notin B$ then $e\in E(G)\cap E(H)$.
Now define $\phi:{\mathcal{M}}_k(H)\setminus {\mathcal{M}}_k(G)\to {\mathcal{M}}_k(G)\setminus {\mathcal{M}}_k(H)$ by $$\phi(M):= \{r(e): e\in M\}.$$
Suppose $M\in {\mathcal{M}}_k(H)\setminus {\mathcal{M}}_k(G)$. We claim $\phi(M)\in {\mathcal{M}}_k(G)\setminus {\mathcal{M}}_k(H)$. Since $r:E(H)\to E(G)$ we know $\phi(M)\subset E(G)$. To show that $\phi(M)\in {\mathcal{M}}_k(G)$ we suppose to a contradiction that $r(e)$ is incident to $r(f)$ for some $e,f\in M$. For the first case, suppose that $r(e) \cap r(f) = x$. Note that $x \in r(e)$ for any edge $e$ if and only if $e \in B$ or $e = xy$. Since $M \cap B \neq \emptyset$ we know that $xy\notin M$. Thus, both $e$ and $f$ are in $B$ and $e\cap f = y$, a contradiction since $e$ and $f$ are in the matching $M$.
Next suppose that $r(e) \cap r(f) = y$. In $H$ we know neither $e$ nor $f$ are incident to $y$ since $M\cap B \neq \emptyset$ and if $e$ or $f$ are in $B$ then we replaced them with edges incident to $x$. Thus, it must be the case that both $e$ and $f$ are incident to $x$ in $M$, a contradiction since $M$ is a matching. Finally, if $r(e) \cap r(f) = z$ for $z\neq x$ and $z\neq y$ then $r$ acted as the identity on $e$ and $f$. So $e\cap f = z$, a contradiction. Thus, $\phi(M) \in {\mathcal{M}}_k(G)$. Finally, $\phi(M)\notin {\mathcal{M}}_k(H)$ since $\phi(M)$ has an edge from $A$ and $A\cap E(H) = \emptyset$.
To complete the proof we show $\phi$ is an injection. Define $r':E(G)\to E(H)$ by $$r'(e):=
\begin{cases}
e\Delta\{x,y\} &\text{ if } e\in A\\
xz &\text{ if } e=yz, z\neq x\\
e &\text{otherwise}
\end{cases}.$$ Define $\phi': \operatorname{im}(\phi)\to {\mathcal{M}}_k(H)$ by $\phi'(M) = \{r'(e): e\in M\}$. Given $M\in{\mathcal{M}}_k(H)\setminus{\mathcal{M}}_k(G)$ it is easy to check that $\phi'(\phi(M)) = M$. Therefore $\phi$ has a left inverse and so $\phi$ is injective.
Thus we may assume that the graph that minimizes $m_k$ is threshold. In fact, we can find a graph that minimizes $m_k$ that has even more structure.
For a threshold graph $G$ write $V(G) = K\cup I$ where $G[K]$ is a clique and $G[I]$ is an independent set. Let $B$ be the bipartite graph with partite sets $K$ and $I$ and edge set $E(K,I)$. If $E(K,I)$ is ${\mathcal{L}}_{|K|,|I|}(e)$ for some $e$, we will say that $G$ is *lex-across*.
We will now prove a lemma that states that if some parameter is either maximized or minimized by a lex-across graph, then that parameter is maximized or minimized by the lex or colex graph.
\[lem:lexacross\] If a parameter $P$ is such that for all $n,e$ there is a $P$-optimal graph in ${\mathcal{G}}_{n,e}$ that is lex-across, then either the lex graph ${\mathcal{L}}(n,e)$ or the colex graph ${\mathcal{C}}(n,e)$ is $P$-optimal.
Given $n,e$ let ${\mathcal{O}}$ be the collection of all triples $(G,K,I)$ such that $V(G) = K\cup I$, $G[K]$ is complete, $G[I]$ is independent, and $G$ is a $P$-optimal lex-across graph. Define ${\overline}{s}$ to be the maximum size of $K$ among all triples in ${\mathcal{O}}$ and let ${\underline}{s}$ be the minimum size of $K$ among all triples in ${\mathcal{O}}$. We consider two cases: when ${\overline}{s}\geq \frac{n}{2}$ and when ${\overline}{s}< \frac{n}{2}$.
Suppose first that ${\overline}{s}\geq \frac{n}{2}$. Let $(G,K,I)\in {\mathcal{O}}$ such that $\abs{K} = {\overline}{s}$. In this case $K$ has at least as many vertices as $I$. If $v\in I$ has $N(v) = K$ then $(G, K\cup\{v\}, I\setminus\{v\}) \in {\mathcal{O}}$ and $\abs{K\cup\{v\}}\geq {\overline}{s}$, a contradiction. Since $(G,K,I)\in {\mathcal{O}}$ we know $G$ is lex-across, and so it must be the case that some vertex in $I$ has fewer than ${\overline}{s}$ neighbors and all other vertices in $I$ are isolated. Thus $G$ is the colex graph ${\mathcal{C}}(n,e)$.
Suppose now that ${\overline}{s}< \frac{n}{2}$. In this case, ${\underline}{s}<\frac{n}{2}$ as well. Let $(G,K,I) \in {\mathcal{O}}$ such that $\abs{K} = {\underline}{s}$. Here $\abs{K}<\abs{I}$. Suppose that there is a vertex $k\in K$ such that $N(k) \cap I = \emptyset$. Then $(G,K\setminus\{k\}, I\cup \{k\})\in {\mathcal{O}}$ and $\abs{K\setminus\{k\}}<{\underline}{s}$, a contradiction. So every vertex in $K$ is adjacent to some vertex in $I$. Since $G$ is lex across all vertices in $K$ except for one possible exception have closed neighborhood all of $G$. Thus $G$ is the lex graph ${\mathcal{L}}(n,e)$.
We are now ready to prove our main theorem, Theorem \[general\].
Let $G\in {\mathcal{G}}_{n,e}$ be a graph that minimizes $m$ (respectively $m_k$). By Corollary \[cor:allcompressed2\] and Lemma \[compressing\] we can assume that $G$ is threshold. By Theorem \[bipartite\] the lex bipartite graph minimizes $m_k$ for all $k\geq 0$ in ${\mathcal{B}}_{\ell,r,e}$, so by Lemma \[minimize\] we can assume $G$ is lex-across. By Lemma \[lem:lexacross\] the lex or colex graph minimizes $m$ (respectively $m_k$) in ${\mathcal{G}}_{n,e}$ which completes the proof of Theorem \[general\].
Further Directions
==================
There are many open problems remaining in this area. For instance the Upper Matching Conjecture of Friedland, Krop, and Markström [@FKM] claims that for all $d$-regular graphs $G$ on $2n$-vertices such that $d$ divides $n$ we have $$m_k(G) \le m_k\bigl(\frac{n}d K_{d,d}\bigr)$$ for all $k$.
We know that the lex or colex graph doesn’t necessarily minimize $m_k(G)$ for all $k$ simultaneously. For example, consider the family ${\mathcal{G}}_{18,87}$. Then $$\begin{array}{c | c | c}
&m_2 &m_7\\
\hline
{\mathcal{L}}(18,87) &2745 &0\\
\hline
{\mathcal{C}}(18,87) &2739 &93,555
\end{array}$$ While $m_2({\mathcal{C}}(18,87)) < m_2({\mathcal{L}}(18,87))$, the lex graph has no $7$-matchings and the colex graph has many. This indicates that it is a non-trivial problem to determine the graph $G\in {\mathcal{G}}_{n,e}$ that minimizes the matching polynomial $$m_G(\lambda) = \sum_{k\ge 0} m_k(G) \lambda^k$$ for a given value of $\lambda>0$. Theorem \[general\] includes case $\lambda=1$. By Lemma \[compressing\] the extremal graph can be taken to be threshold.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
We present a numerical method for solving the time-independent thermal radiative transfer (TRT) equation or the neutron transport (NT) equation when the opacity or cross-section varies rapidly in energy (frequency) on the microscale $\varepsilon$; $\varepsilon$ corresponds to the characteristic spacing between absorption lines or resonances, and is much smaller than the macroscopic energy (frequency) variation of interest. The approach is based on a rigorous homogenization of the TRT/NT equation in the energy (frequency) variable. Discretization of the homogenized TRT/NT equation results in a multigroup-type system, and can therefore be solved by standard methods.
We demonstrate the accuracy and efficiency of the approach on three model problems. First we consider the Elsasser band model with constant temperature and a line spacing $\varepsilon=10^{-4}$. Second, we consider a neutron transport application for fast neutrons incident on iron, where the characteristic resonance spacing $\varepsilon$ necessitates $\approx16,000$ energy discretization parameters if Planck-weighted cross sections are used. Third, we consider an atmospheric TRT problem for an opacity corresponding to water vapor over a frequency range $1000-2000\,\mathrm{cm}^{-1}$, where we take $12$ homogeneous layers between $1\,$km - $15\,$km, and temperature/pressure values in each layer from the standard US atmosphere. For all three problems, we demonstrate that we can achieve between $0.1$ and $1$ percent relative error in the solution, and with several orders of magnitude fewer parameters than a standard multigroup formulation using Planck-weighted opacities for a comparable accuracy.
author:
- |
T. S. Haut and C. Ahrens and A. Jonko and R. Lowrie and A. Till\
Los Alamos National Laboratory\
Los Alamos, NM 87545
bibliography:
- 'opacity\_averaging.bib'
title: 'A new multigroup method for cross-sections that vary rapidly in energy'
---
Background
==========
Thermal radiative transfer (TRT) plays a key role in a number of scientific and engineering disciplines. For example, resolving the radiation field in three-dimensional cloudy atmospheres is key to understanding a number of atmospheric science and remote sensing problems [@MAR-DAV:2005]. In many TRT problems, there is rapid variation in the opacity with energy (or frequency) due to bound-bound and bound-free transitions. In fact, for broad-band TRT problems there can be hundreds of thousands of absorption lines, whose widths are many times smaller than the overall energy range of interest. This fine scale structure in the opacities, coupled with discretizing the spatial and angular variables, places large demands on computational resources. Thus, researchers have sought methods to “average” or “homogenize” the opacities and derive so-called “grey” or frequency independent approximations, thereby reducing the complexity of solving the full TRT problem. Many opacity homogenization techniques have been developed over the years. Here we mention only those most closely related with the method developed in this paper.
A commonly used averaging technique (the multigroup or picket fence) starts by integrating the transport equation over the energy interval $\left[E_{g},E_{g+1}\right]$. This formerly results in a transport equation for the group averaged intensity, $$\psi_{g}\left(\mathbf{x},\boldsymbol{\Omega}\right)=\int_{E_{g}}^{E_{g+1}}\psi\left(\mathbf{x},\boldsymbol{\Omega},E\right)dE.$$ However, the group averaged opacities $\sigma_{g}$ in the resulting multigroup equations depend on the unknown solution $\psi\left(\mathbf{x},\boldsymbol{\Omega},E\right)$, and an approximation is therefore needed to close the system. Typically, either a Rosseland or Plank mean opacity is used; generally, these closures are only accurate when $\left[E_{g},E_{g+1}\right]$ is small relative to the variation of $\sigma\left(E\right)$ or in certain limiting physical regimes (e.g. the optically thick limit). See, e.g., [@POMRAN:1971] for details. We remark that the averaging method developed in this paper does not require one to postulate such a closure relationship.
The multiband method [@CUL-POM:1980] is another approach for averaging the transport equation. The formulation results in a multigroup-type system for $$\psi_{g,b}\left(\mathbf{x},\boldsymbol{\Omega}\right)=\int_{\sigma_{g,b}}^{\sigma_{g,b+1}}\int_{E_{g}}^{E_{g+1}}\delta\left(\sigma\left(E\right)-\xi\right)\psi\left(\mathbf{x},\boldsymbol{\Omega},E\right)dEd\xi.\label{eq:multiband average}$$ The multiband equations depend on an averaged opacity $\sigma_{g,b}$ that again involves the unknown solution $\psi\left(\mathbf{x},\boldsymbol{\Omega},E\right)$. Like the multigroup method, this requires one to make some approximation in order to close the system. We note that the group averaged solution is recovered from (\[eq:multiband average\]) via summing over bands $b$ within each group $g$.
In the context of atmospheric TRT calculations, the correlated k-Distribution method [@ARK-GRO:1972] and the multigroup k-Distribution method [@MOD-ZHA:2001] have been shown to drastically reduce the computational cost of direct line-by-line calculations, and have similarities to the current homogenization approach. Another method that is related to the current homogenization approach is the so-called Opacity Distribution Function (ODF) method [@STR-KUR:1966], which also takes a statistical approach toward coarse-graining the TRT equation in energy. See [@MODEST:2013] for a clear overview of these, and related, methods.
Finally, let us mention that the homogenization in energy of transport equations in the absence of scattering can be analyzed using techniques developed by Tartar [@TARTAR:1990]. Unlike when the opacity has rapid spatial variation (see [@DUM-GOL:2000]), the homogenized version of the transport equation *is not* simply another transport equation with a homogenized opacity. In fact, the homogenized equation is an integro-differential equation, where the integral equation is nonlocal in the spatial variable. An alternative homogenization approach for solving the time-dependent TRT equation, in the absence of scattering, has also been pursued in [@ET-GO-SA:2010] and [@MAT-SAL:2013], where the authors develop homogenized equations on an enlarged phase space. This approach is similar in spirit to that taken in the current paper; however, one advantage of that given here is the ability to handle scattering (in angle and energy), as well as to reduce the numerical computation to a standard multigroup formulation. Finally, we remark that, when the opacity is of the form $\sigma_{\varepsilon}\left(E,T\right)=\sigma_{0}\left(E/\varepsilon,E,T\right)$, where $\sigma_{0}\left(\kappa,E,T\right)$ is almost periodic in $\kappa$, then (\[eq:transport equation on extended phase space, intro\]) and (\[eq:psi\_0 as weighting against Young measure, intro\]) is analogous to the two-scale homogenization theory developed in [@ALLAIR:1992].
Outline of the new multigroup method
====================================
Here we outline a computational method for the transport equation $$\begin{aligned}
\boldsymbol{\Omega}\cdot\nabla_{\mathbf{x}}\psi_{\varepsilon}+\sigma_{\varepsilon}\left(T\left(\mathbf{x}\right),E\right)\psi_{\varepsilon} & = & \sigma_{\varepsilon}^{a}\left(T\left(\mathbf{x}\right),E\right)S\left(\mathbf{x},E\right)+Q_{0}\left(\mathbf{x},\boldsymbol{\Omega},E\right)+\nonumber \\
& & \int_{\mathbb{S}^{2}}\int_{0}^{\infty}\Sigma^{s}\left(\mathbf{x},E,E',\boldsymbol{\Omega}\cdot\boldsymbol{\Omega}'\right)\psi_{\varepsilon}dE'd\boldsymbol{\Omega}',\label{eq:transport equation, intro}\end{aligned}$$ where $$\Sigma^{s}\left(\mathbf{x},E,E',\boldsymbol{\Omega}\cdot\boldsymbol{\Omega}'\right)=\sigma^{s}\left(T\left(\mathbf{x}\right),E'\right)K\left(E,E',\boldsymbol{\Omega}\cdot\boldsymbol{\Omega}'\right),$$ $$\sigma_{\varepsilon}\left(T\left(\mathbf{x}\right),E\right)=\sigma_{\varepsilon}^{a}\left(T\left(\mathbf{x}\right),E\right)+\sigma^{s}\left(T\left(\mathbf{x}\right),E\right),$$ and the absorption opacity (cross-section) $\sigma_{\varepsilon}^{a}\left(T\left(\mathbf{x}\right),E\right)$ rapidly varies in energy $E$ on the microscale $0<\varepsilon\ll1$; here $\varepsilon$ denotes the characteristic spacing between spectral lines or resonances. For simplicity, our discussion ignores density and pressure dependence in the opacity $\sigma_{\varepsilon}$, but its incorporation into the proposed algorithm is straightforward; in fact, see Section \[sub:An-atmospheric-TRT\] for an atmospheric TRT example where the pressure and density dependence are included. We note that, for neutron problems, $\sigma^{s}$ can vary rapidly on the micro-scale as well, but we assume her for simplicity that $\sigma^{s}\left(T,E\right)$ smoothly depends on $E$ (i.e., is independent of $\varepsilon$).
The method is based on a rigorous homogenization theory for (\[eq:transport equation, intro\]), and has a computational cost that scales independently of the microscale parameter $\varepsilon$, aside from a pre-computation step analogous to computing Planck-weighted or Rosseland-weighted opacities. In this framework, the fine-scale solution $\psi_{\varepsilon}$ converges to the homogenized solution $\psi_{0}$ in the sense that $$\lim_{\varepsilon_{k}\rightarrow0}\int_{E_{0}}^{E_{1}}\psi_{\varepsilon_{k}}\left(\mathbf{x},\boldsymbol{\Omega},E\right)dE=\int_{E_{0}}^{E_{1}}\psi_{0}\left(\mathbf{x},\boldsymbol{\Omega},E\right)dE,$$ for arbitrary energy values $E_{0}$ and $E_{1}$ and some sequence $\varepsilon_{k}\rightarrow0$. Our derivation assumes that the total opacity $\sigma_{\varepsilon}\left(E,T\right)$ depends on a free parameter $\varepsilon>0$ (governing the characteristic spacing between lines), and that $\left\Vert \sigma_{\varepsilon}\left(\cdot,T\right)\right\Vert _{\infty}$ is uniformly bounded in $\varepsilon>0$ for each $T$. In practice, the final algorithm only makes use of a single opacity $\sigma_{\varepsilon}\left(T,E\right)$ at some fixed characteristic microscale $\varepsilon$ that is much smaller than the macroscopic scale we are interested in capturing.
As a key tool, we use the Young measure $\lambda_{E}^{T}$ associated with $\sigma_{\varepsilon}\left(E,T\right)$ (cf. [@BALL:1989]). Roughly speaking, $\lambda_{E}^{T}$ gives the probability distribution of values $\sigma_{\varepsilon}\left(E,T\right)$ in a vanishingly small neighborhood of $E$ as $\varepsilon\rightarrow0$. The key property that the Young measure $\lambda_{E}^{T}$ satisfies (see [@BALL:1989] for a proof) is that, for any continuous function $F\left(E,\xi\right)$ defined for $E\geq0$ and $\xi\geq0$, there is a sequence $\varepsilon_{k}\rightarrow0$ such that $$\lim_{\varepsilon_{k}\rightarrow0}\int_{E_{0}}^{E_{1}}F\left(E,\sigma_{\varepsilon_{k}}\left(E,T\right)\right)dE=\int_{E_{0}}^{E_{1}}\left(\int_{0}^{\infty}F\left(E,\xi\right)\lambda_{E}^{T}\left(d\xi\right)\right)dE.\label{eq:Young measure, key rep-1}$$ Intuitively, for small $\varepsilon$ and for all $E'\in\left[E-\Delta E/2,E+\Delta E/2\right]$ in a neighborhood of $E$ with $0<\varepsilon\ll\Delta E$, the values $F\left(E',\sigma_{\varepsilon}\left(E'\right)\right)\approx F\left(E,\sigma_{\varepsilon}\left(E'\right)\right)$ can be wildly varying since $\sigma_{\varepsilon}\left(E'\right)$ can rapidly oscillate for $E'\in\left[E-\Delta E/2,E+\Delta E/2\right]$; however, the average value of $F\left(E,\sigma_{\varepsilon}\left(E'\right)\right)$ for $E'\in\left[E-\Delta E/2,E+\Delta E/2\right]$ is given by weighting $F\left(E,\xi\right)$ against the probability of $\sigma_{\varepsilon}\left(E\right)\in\left[\xi-d\xi/2,\xi+d\xi/2\right]$, $$\int_{0}^{\infty}F\left(E,\xi\right)\lambda_{E}^{T}\left(d\xi\right).\label{eq:expected value}$$ A simple but instructive example is the Elsasser band model [@ELSAS:1938], $$\sigma_{\varepsilon}\left(E\right)=\frac{\cosh\left(\beta\right)+1}{\cosh\left(\beta\right)-\cos\left(2\pi E/\varepsilon\right)},\label{eq:Elsasser band model, intro}$$ $\beta>0$, which models an infinite number of Lorenz lines with equal spacing of $\varepsilon$ and uniform strength; in Section \[sub:Example-1:-the\], we compute $\lambda_{E}\left(\xi\right)$ for (\[eq:Elsasser band model, intro\]) analytically. The Young measure for $\sigma_{\varepsilon}\left(E,T\right)$–and the associated homogenized solution—exists if $\left\Vert \sigma_{\varepsilon}\left(\cdot,T\right)\right\Vert _{\infty}$ is uniformly bounded in $\varepsilon>0$ for each $T$ [@BALL:1989]. Our derivation will be a direct application of (\[eq:Young measure, key rep-1\]).
In order to derive the homogenized equations from (\[eq:Young measure, key rep-1\]), we assume that the absorption opacity $\sigma_{\varepsilon}^{a}\left(E,T\right)$ is of the form $$\sigma_{\varepsilon}^{a}\left(E,T\right)=\chi\left(\sigma_{\varepsilon}\left(E\right),E,T\right),\label{eq:basic opacity assumption}$$ where $\chi\left(\sigma,E,T\right)$ is continuous in its first argument and $\sigma_{\varepsilon}\left(E\right)$ is an appropriate function (that may or may not be directly related to the original opacity $\sigma_{\varepsilon}^{a}\left(E,T\right)$). This assumption, in particular, is a generalization of the commonly assumed assumption in atmospheric TRT calculations that $\sigma_{\varepsilon}^{a}\left(E,T\right)=\chi\left(\sigma_{\varepsilon}\left(E,T_{0}\right),T\right)$, where $T_{0}$ is a fixed reference temperature (see [@QIA-LIO:1992] for its use in the correlated k-Distribution method). In Section \[sub:Relaxation-of-the\], we relax the assumption in (\[eq:basic opacity assumption\]) of an exact equality, and discuss how to numerically compute an approximation $\sigma_{\varepsilon}^{a}\left(E,T\right)\approx\chi\left(\sigma_{\varepsilon}\left(E,T_{0}\right),E,T\right)$, for a given reference temperature $T_{0}$, that is optimal in a certain sense.
Assume that $\sigma_{\varepsilon}^{a}\left(E,T\right)=\chi\left(\sigma_{\varepsilon}\left(E\right),E,T\right)$. Then from the solution of the transport equation for $\Psi\left(\mathbf{x},\boldsymbol{\Omega},E,\xi\right)$, parameterized by the real number $\xi\geq0$, $$\begin{aligned}
\boldsymbol{\Omega}\cdot\nabla_{\mathbf{x}}\Psi+\left[\chi\left(\xi,E,T\left(\mathbf{x}\right)\right)+\sigma^{s}\left(E,T\left(\mathbf{x}\right)\right)\right]\Psi & = & \chi\left(\xi,E,T\left(\mathbf{x}\right)\right)S\left(T\left(\mathbf{x}\right),E\right)+Q_{0}\left(\mathbf{x},\boldsymbol{\Omega},E\right)+\nonumber \\
& & \int_{\mathbb{S}^{2}}\int_{0}^{\infty}\int_{0}^{\infty}\Sigma^{s}\left(\mathbf{x},E,E',\boldsymbol{\Omega}\cdot\boldsymbol{\Omega}'\right)\Psi\,\lambda_{E'}\left(d\xi\right)dE'd\boldsymbol{\Omega}',\label{eq:transport equation on extended phase space, intro}\end{aligned}$$ we can exactly recover the homogenized solution by weighting against the Young measure of $\sigma_{\varepsilon}\left(E\right)$, $$\psi_{0}\left(\mathbf{x},\boldsymbol{\Omega},E\right)=\int_{0}^{\infty}\Psi\left(\mathbf{x},\boldsymbol{\Omega},E,\xi\right)\lambda_{E}\left(d\xi\right).\label{eq:psi_0 as weighting against Young measure, intro}$$ Heuristically, $\Psi\left(\mathbf{x},\boldsymbol{\Omega},E,\xi\right)d\lambda_{E}\left(\xi\right)$ weights the solution of the transport equation (\[eq:transport equation on extended phase space, intro\]) by the probability that $\sigma_{\varepsilon}\left(E\right)=\xi$ when $\varepsilon\ll1$, and integration over all possible values $\sigma$ yields the homogenized solution at $\left(\mathbf{x},\boldsymbol{\Omega},E\right)$ in the limit of $\varepsilon\rightarrow0$. The proof of this is a direct application of (\[eq:Young measure, key rep-1\]), and is given in Section \[sec:The-homogenized-transport\]. In the proof, we assume that (\[eq:transport equation on extended phase space, intro\]) has a unique solution that is continuous in its last argument $\xi$.
By discretizing equations (\[eq:transport equation on extended phase space, intro\]) and (\[eq:psi\_0 as weighting against Young measure, intro\]), we obtain an algorithm that is analogous to the multiband method (cf. [@CUL-POM:1980]). In particular, we choose a coarse number $m_{g}$ of energy groups $\left[E_{i},E_{i+1}\right]$ and a coarse number $m_{\sigma}$ opacity bands $\left[\sigma_{j},\sigma_{j+1}\right]$, and use the theory in [@BALL:1989] to construct a discrete approximation $$\lambda_{E}\left(\xi\right)\approx\sum_{j=1}^{m}p_{i,j}\delta\left(\xi-\sigma_{i,j}\right),\,\,\,\, E\in\left[E_{i},E_{i+1}\right],\label{eq:discrete approx of Young measure, intro}$$ where $p_{i,j}$ gives the probability that $\sigma_{j}\leq\sigma_{\varepsilon}\left(E\right)\leq\sigma_{j+1}$ for $E\in\left[E_{i},E_{i+1}\right]$ (see Section \[sub:Discrete-approximation-of\] for more details). Convergence of (\[eq:discrete approx of Young measure, intro\]) to $\lambda_{E}$ is made precise in [@BALL:1989]. The key point of this construction is that, for realistic opacities, the number parameters $\sigma_{i,j}$ and $p_{i,j}$ is typically a small constant independent of the size $\varepsilon$ of the microscale; this is provably so when the opacity is multiscale, e.g. of the form $\sigma_{0}\left(T,E,E/\varepsilon\right)$, where, e.g., $\sigma_{0}\left(T,E,\kappa\right)$ is almost periodic in $\kappa$. In general, computing (\[eq:discrete approx of Young measure, intro\]) scales linearly in $\varepsilon^{-1}$, but needs to be performed only once for a fine enough energy grid in order for interpolation to be accurate; this is analogous to the pre-computation of Planck-weighted or Rosseland-weighted opacities for use in multigroup transport codes, which also scales linearly in $\varepsilon^{-1}$.
Now define $\kappa_{ij}\left(T\right)=\chi\left(\sigma_{i,j},E_{i},T\right)$ and $$\Psi_{ij}\left(\mathbf{x},\boldsymbol{\Omega}\right)=\int_{E_{i}}^{E_{i+1}}\Psi\left(\mathbf{x},\boldsymbol{\Omega},E,\sigma_{i,j}\right)dE,\,\,\,\,\,\,\,\, S_{i}\left(E,T\right)=\int_{E_{i}}^{E_{i+1}}S\left(E,T\right)dE.$$ Then using the representation (\[eq:discrete approx of Young measure, intro\]) in (\[eq:transport equation on extended phase space, intro\]), we obtain the multigroup-type equations $$\begin{aligned}
\boldsymbol{\Omega}\cdot\nabla_{\mathbf{x}}\Psi_{ij}+\kappa_{ij}\left(T\left(\mathbf{x}\right)\right)\Psi_{ij} & = & \kappa_{ij}\left(T\left(\mathbf{x}\right)\right)S_{i}\left(x\right)+\nonumber \\
& & \int_{\mathbb{S}^{2}}\Sigma^{s}\left(\mathbf{x},E\rightarrow E',\boldsymbol{\Omega}\cdot\boldsymbol{\Omega}'\right)\Psi_{ij}d\boldsymbol{\Omega}',\label{eq:transport multiband-multigroup, intro}\end{aligned}$$ This corresponds to a standard multigroup-type approximation of (\[eq:transport equation on extended phase space, intro\]) at each band parameter $\xi=\sigma_{i,j}$. The homogenized solution at $I_{0}\left(\mathbf{x},\boldsymbol{\Omega},E\right)$ in each group interval $\left[E_{i},E_{i+1}\right]$ is then given by weighting against the discrete approximation (\[eq:discrete approx of Young measure, intro\]) of the Young measure $\lambda_{E}$, $$\psi_{0}\left(\mathbf{x},\boldsymbol{\Omega},E\right)\approx\sum_{j=1}^{m}p_{i,j}\Psi_{ij}\left(\mathbf{x},\boldsymbol{\Omega}\right),\,\,\,\, E\in\left[E_{i},E_{i+1}\right].\label{eq:formula for psi_o, discretized Young measures}$$ Here we used the discrete approximation (\[eq:discrete approx of Young measure, intro\]) of the Young measure in (\[eq:psi\_0 as weighting against Young measure, intro\]). We remark that the values of $p_{i,j}$ can be obtained via interpolation from a pre-computed table, as explained in Section \[sub:Discrete-approximation-of\].
In our numerical experiments, we choose the band parameters $\sigma_{i,j}$ to be either equally spaced or logarithmically spaced between the minimum and maximum opacity values in each group, $$\min_{E_{i}\leq E\leq E_{i+1}}\sigma_{\varepsilon}\left(E\right),\,\,\,\,\max_{E_{i}\leq E\leq E_{i+1}}\sigma_{\varepsilon}\left(E\right).$$ This choice for the band parameters $\sigma_{i,j}$ is justified in Section \[sub:Discrete-approximation-of\].
To summarize: we solve the multigroup-type equations (\[eq:transport multiband-multigroup, intro\]) for each group interval $\left[E_{i},E_{i+1}\right]$ and for each opacity band $\left[\sigma_{i,j},\sigma_{i,j+1}\right]$, and then average with respect to the discrete Young measure, (\[eq:formula for psi\_o, discretized Young measures\]). The opacity bands $\left[\sigma_{i,j},\sigma_{i,j+1}\right]$ can be chosen to be equally spaced or log-spaced within the range of $\sigma_{\varepsilon}\left(E\right)$, $E\in\left[E_{i},E_{i+1}\right]$.
Equations (\[eq:transport multiband-multigroup, intro\]) and (\[eq:formula for psi\_o, discretized Young measures\]) are analogous to the multiband method, but are derived from within a homogenization framework. Unlike the multiband method, however, no closure assumption (i.e., weighting spectrum) is needed to compute the multiband parameters $\sigma_{i,j}$. In addition, the group average (\[eq:formula for psi\_o, discretized Young measures\]) is not a direct sum as in the multiband method, but instead uses the discrete Young measure to weight the multiband solutions within each group. As previously remarked, the homogenization approach is also related to the correlated k-Distribution method [@ARK-GRO:1972] and to the multigroup k-Distribution method [@MOD-ZHA:2001]. We point out that the current approach is able to explicitly handle scattering in energy, which to the best of our knowledge has not been explored with k-Distribution methods.
The remainder of this paper is organized as follows. In Section \[sec:The-homogenized-transport\], we derive the homogenized equations, and their discrete approximation. We then apply we apply the above methodology to three examples in Section \[sec:Examples\]. For simplicity, we neglect scattering in all examples. In Section \[sub:Example-1:-the\], we consider the Elsasser band model (\[eq:Elsasser band model, intro\]) with $\varepsilon=10^{-4}$ (see also [@POMRAN:1971]). For the Elsasser band model, we also analytically compute the Young measure $\lambda_{E}\left(\xi\right)$ and compare it with its discrete approximation (\[eq:discrete approx of Young measure, intro\]). In the second example, we consider a neutron transport example using an absoption cross section $\sigma_{\varepsilon}\left(E\right)$ for iron at room temperature and a Watt fission spectrum; note that, in this case, the subscript $\varepsilon$ of $\sigma_{\varepsilon}\left(E\right)$ is formerly retained in order to denote the characteristic resonance spacing, but is not an actual free parameter. Finally, in our last example, we consider an atmospheric TRT calculation; we use $12$ homogenous atmoshperic layers between $1\,$km-$12\,$km (the temperature and pressure in each layer come from the 1976 US standard atmosphere), and consider the absorption opacity corresponding to water vapor. For all three examples, we demonstrate a small number of energy discretization parameters can capture the solution with between $0.1$ and $1$ percent accuracy, and using orders of magnitude fewer parameters than the standard multigroup formulation with Planck-weighted opacities for comparable accuracies. We note that, for these examples, we expect that the correlated k-Distribution method can yield the same accuracy with a comparable number of parameters.
The homogenized transport equation and its discrete approximation\[sec:The-homogenized-transport\]
==================================================================================================
In this Section \[sub:Derivation-of-homogenized\], we first derive the homogenized equations (\[eq:transport equation on extended phase space, intro\]) and (\[eq:psi\_0 as weighting against Young measure, intro\]). We then discuss the discrete approximation of the Young measure in Section \[sub:Discrete-approximation-of\], and derive the discrete approximation (\[eq:transport multiband-multigroup, intro\]) and (\[eq:formula for psi\_o, discretized Young measures\]) in Section \[sub:Derivation-of-discrete\]. Finally, we conclude this section with a heuristic derivation of the homogenized system for the time-dependent TRT equations.
Derivation of the homogenized system \[sub:Derivation-of-homogenized\]
----------------------------------------------------------------------
To derive (\[eq:transport equation on extended phase space, intro\]) and (\[eq:psi\_0 as weighting against Young measure, intro\]), first assume that the scattering kernel only depends on angle, i.e. $\Sigma^{s}=\Sigma^{s}\left(\mathbf{x},\boldsymbol{\Omega}\cdot\boldsymbol{\Omega}'\right)$. Consider the solution $\Psi\left(\mathbf{x},\boldsymbol{\Omega},E,\xi\right)$ of (\[eq:transport equation on extended phase space, intro\]). Then $\Psi\left(\mathbf{x},\boldsymbol{\Omega},E,\sigma_{\varepsilon}\left(E\right)\right)$ satisfies the transport equation (\[eq:transport equation, intro\]). Therefore, since (\[eq:transport equation, intro\]) has a unique solution, $\psi_{\varepsilon}\left(\mathbf{x},\boldsymbol{\Omega},E\right)=\Psi\left(\mathbf{x},\boldsymbol{\Omega},E,\sigma_{\varepsilon}\left(E\right)\right)$. It follows from (\[eq:Young measure, key rep-1\]) that, for any $0<E_{0}<E_{1}$, $$\begin{aligned}
\lim_{\varepsilon\rightarrow0}\int_{E_{0}}^{E_{1}}\psi_{\varepsilon}\left(\mathbf{x},\boldsymbol{\Omega},E\right)dE & = & \lim_{\varepsilon\rightarrow0}\int_{E_{0}}^{E_{1}}\Psi\left(\mathbf{x},\boldsymbol{\Omega},E,\sigma_{\varepsilon}\left(E\right)\right)dE.\\
& = & \int_{E_{0}}^{E_{1}}\int_{0}^{\infty}\Psi\left(\mathbf{x},\boldsymbol{\Omega},E,\xi\right)d\lambda_{E}\left(\xi\right)dE\end{aligned}$$ Since $E_{0}$ and $E_{1}$ are arbitrary, we finally conclude that $$\psi_{0}\left(\mathbf{x},\boldsymbol{\Omega},E\right)=\int_{0}^{\infty}\Psi\left(\mathbf{x},\boldsymbol{\Omega},E,\xi\right)\lambda_{E}\left(d\xi\right).$$
Now suppose that the scattering kernel $\Sigma^{s}=\Sigma^{s}\left(\mathbf{x},E,E',\boldsymbol{\Omega}\cdot\boldsymbol{\Omega}'\right)$ depends on both energy and angle. We again argue that (\[eq:transport equation on extended phase space, intro\]) and (\[eq:psi\_0 as weighting against Young measure, intro\]) are the appropriate homogenized equations. To do so, define $\tilde{\psi}_{\varepsilon}\left(\mathbf{x},\boldsymbol{\Omega},E\right)\equiv\Psi\left(\mathbf{x},\boldsymbol{\Omega},E,\sigma_{\varepsilon}\left(E\right)\right)$. Then $$\boldsymbol{\Omega}\cdot\nabla_{\mathbf{x}}\tilde{\psi}_{\varepsilon}+\sigma_{\varepsilon}\tilde{\psi}_{\varepsilon}=\sigma_{\varepsilon}^{a}S+Q_{0}+\int_{\mathbb{S}^{2}}\left(\int_{0}^{\infty}\Sigma^{s}\left(\mathbf{x},E,E',\boldsymbol{\Omega}\cdot\boldsymbol{\Omega}'\right)\tilde{\psi}_{\varepsilon}dE'd\boldsymbol{\Omega}'+\mathcal{E}_{\varepsilon}\right)d\boldsymbol{\Omega}',$$ where the residual term $\mathcal{E}_{\varepsilon}$ is given by $$\mathcal{E}_{\varepsilon}\left(\mathbf{x},\boldsymbol{\Omega}',E\right)=\int_{0}^{\infty}\Sigma^{s}\left(\mathbf{x},E,E',\boldsymbol{\Omega}\cdot\boldsymbol{\Omega}'\right)\left(\tilde{\psi}_{\varepsilon}\left(\mathbf{x},\boldsymbol{\Omega}',E'\right)-\int_{0}^{\infty}\Psi\left(\mathbf{x},\boldsymbol{\Omega}',E',\xi\right)\lambda_{E'}\left(d\xi\right)\right)dE'.$$ From the property (\[eq:Young measure, key rep-1\]) and $\tilde{\psi}_{\varepsilon}\left(\mathbf{x},\boldsymbol{\Omega},E'\right)=\Psi\left(\mathbf{x},\boldsymbol{\Omega},E',\sigma_{\varepsilon}\left(E'\right)\right)$, we see that $\mathcal{E}_{\varepsilon}\left(\mathbf{x},\boldsymbol{\Omega},E\right)\rightarrow0$ for $\varepsilon\rightarrow0$. In addition, the energy dependence in $\mathcal{E}_{\varepsilon}\left(\mathbf{x},\boldsymbol{\Omega},E\right)$ is only through $\Sigma^{s}\left(\mathbf{x},E,E',\boldsymbol{\Omega}\cdot\boldsymbol{\Omega}'\right)$, and is therefore slow (i.e., it is independent of the small parameter $\varepsilon$). It follows that for $\varepsilon\ll1$, $$\boldsymbol{\Omega}\cdot\nabla_{\mathbf{x}}\tilde{\psi}_{\varepsilon}+\sigma_{\varepsilon}\tilde{\psi}_{\varepsilon}-\sigma_{\varepsilon}^{a}S+Q_{0}+\int_{0}^{\infty}\Sigma^{s}\left(\mathbf{x},E,E',\boldsymbol{\Omega}\cdot\boldsymbol{\Omega}'\right)\tilde{\psi}_{\varepsilon}dE\approx0.$$ Since the transport equation is well-posed, $\tilde{\psi}_{\varepsilon}\approx\psi_{\varepsilon}$. Finally, invoking (\[eq:Young measure, key rep-1\]) again, $$\lim_{\varepsilon\rightarrow0}\int_{E_{0}}^{E_{1}}\tilde{\psi}_{\varepsilon}\left(\mathbf{x},\boldsymbol{\Omega},E\right)dE=\int_{E_{0}}^{E_{1}}\int_{0}^{\infty}\Psi\left(\mathbf{x},\boldsymbol{\Omega},E,\xi\right)\lambda_{E}\left(d\xi\right)dE,$$ and we see that (\[eq:transport equation on extended phase space, intro\]) and (\[eq:psi\_0 as weighting against Young measure, intro\]) are the appropriate homogenized equations.
Discrete approximation of the Young measures\[sub:Discrete-approximation-of\]
-----------------------------------------------------------------------------
Here we discuss the derivation of the discrete approximation (\[eq:discrete approx of Young measure, intro\]), from which equations (\[eq:transport multiband-multigroup, intro\]) and (\[eq:formula for psi\_o, discretized Young measures\]) follow from the homogenized equations (\[eq:transport equation on extended phase space, intro\]) and (\[eq:psi\_0 as weighting against Young measure, intro\]). As we will see, the number of bands $\left[\sigma_{j},\sigma_{j+1}\right]$ needed in the approximation to determined by the particular function $F\left(\xi\right)$ used in the fundamental representation (\[eq:Young measure, key rep-1\]) for the Young measure; in our applications, $F\left(\xi\right)$ is very smooth, and a small number of bands are required, independent of the scale $\varepsilon$ at which $\sigma_{\varepsilon}\left(E\right)$ varies in energy.
We will construct a discrete approximation to the Young measure via the theory developed in [@BALL:1989]; for notational simplicity, in this section we drop the temperature dependence in the notation. In particular, the measure $\lambda_{E}$ is entirely determined by its action on continuous functions $f$ via $$\left\langle \lambda_{E},f\right\rangle =\lim_{\Delta E\rightarrow0}\lim_{\varepsilon\rightarrow0}\frac{1}{\Delta E}\int_{E-\Delta E/2}^{E+\Delta E/2}f\left(E',\sigma_{\varepsilon}\left(E'\right)\right)dE'.$$ Note the order of the limits intuitively corresponds to choosing a scale $\delta$ that is large relative to the microscopic behavior but small relative to the macroscopic behavior, $0<\varepsilon\ll\Delta E$. Then $$\left\langle \lambda_{E},f\right\rangle \approx\frac{1}{\Delta E}\int_{E-\Delta E/2}^{E+\Delta E/2}f\left(E,\sigma_{\varepsilon}\left(E'\right)\right)dE',\label{eq:Ball, approximation of Young measure}$$ and this becomes precise by first letting $\varepsilon\rightarrow0$ and then $\Delta E\rightarrow0$. In (\[eq:Ball, approximation of Young measure\]), approximated $f\left(E',\sigma_{\varepsilon}\left(E'\right)\right)\approx f\left(E,\sigma_{\varepsilon}\left(E'\right)\right)$, which is valid over $\left[E-\Delta E/2,E+\Delta E/2\right]$ since the variation of $f$ in its first argument does not depend on the fast scale $\varepsilon$.
Given fixed $\sigma_{1}<\dots\sigma_{j}<\sigma_{j+1}<\ldots$, define the characteristic functions $\zeta_{j}\left(\xi\right)$, $$\zeta_{j}\left(\xi\right)=\begin{cases}
1, & \,\sigma_{j}\leq\xi\leq\sigma_{j+1}\\
0, & \,\text{else}.
\end{cases}.$$ Consider the collection of step functions $$f\left(E,\xi\right)=\sum_{j}f\left(E,\sigma_{j}\right)\zeta_{j}\left(\xi\right).\label{eq:step function}$$ Although $\zeta_{j}$ are not continuous, any continuous function can be approximated by such functions. Now, for a general step function (\[eq:step function\]) and using (\[eq:Ball, approximation of Young measure\]), $$\begin{aligned}
\left\langle \lambda_{E},f\right\rangle & \approx & \frac{1}{\Delta E}\int_{E-\Delta E/2}^{E+\Delta E/2}\left(\sum_{j}f\left(E',\sigma_{j}\right)\zeta_{j}\left(\sigma_{\varepsilon}\left(E'\right)\right)\right)dE'\\
& \approx & \sum_{j}f\left(E,\sigma_{j}\right)\left(\frac{1}{\Delta E}\int_{E-\Delta E/2}^{E+\Delta E/2}\zeta_{j}\left(\sigma_{\varepsilon}\left(E'\right)\right)dE'\right)\\
& = & \sum_{j}f\left(E,\sigma_{j}\right)p_{j}\left(E\right).\end{aligned}$$ In the last equality, the probability $p_{j}\left(E\right)$ is given by $$p_{j}\left(E\right)=\frac{\lambda\left(E'\in\left[E-\Delta E/2,E+\Delta E/2\right]\mid\sigma_{j}\leq\sigma_{\varepsilon}\left(E'\right)\leq\sigma_{j+1}\right)}{\Delta E},\label{eq:definition of p_j(E)}$$ with $\lambda$ in (\[eq:definition of p\_j(E)\]) denoting the Lebesque measure. Approximating a general continuous function $f\left(\xi\right)$ (defined for $\xi\geq0$) by a step function, we have that $$\begin{aligned}
\left\langle \lambda_{E},f\right\rangle & \approx & \sum_{j}p_{j}\left(E\right)f\left(E,\sigma_{j}\right)\\
& = & \int f\left(E,\xi\right)\left(\sum_{j}p_{j}\left(E\right)\delta\left(\xi-\sigma_{j}\right)\right)d\xi.\end{aligned}$$ Thus, a discrete approximation to the Young measure is given by $$\lambda_{E}\left(\xi\right)\approx\sum_{j}p_{j}\left(E\right)\delta\left(\xi-\sigma_{j}\right),$$ where $p_{j}\left(E\right)$ is defined via (\[eq:definition of p\_j(E)\]).
The function $p_{j}\left(E\right)$ give the probability that $\sigma_{j}\leq\sigma_{\varepsilon}\left(E'\right)\leq\sigma_{j+1}$ for $E'\in\left[E-\Delta E/2,E+\Delta E/2\right]$, where $0<\varepsilon\ll\Delta E$; in particular, $p_{j}\left(E\right)$ may be computed by uniformly sampling $E'\in\left[E-\Delta E/2,E+\Delta E/2\right]$ and counting how many times $\sigma_{j}\leq\sigma_{\varepsilon}\left(E'\right)\leq\sigma_{j+1}$ for each band $\left[\sigma_{j},\sigma_{j+1}\right]$. We remark that $p_{j}\left(E\right)$ can be precomputed on a fine energy grid, and evaluated at other energy points via interpolation.
Derivation of the discrete homogenized system\[sub:Derivation-of-discrete\]
---------------------------------------------------------------------------
To derive (\[eq:transport multiband-multigroup, intro\]), evaluate (\[eq:transport equation on extended phase space, intro\]) at $\xi=\sigma_{i,j}$ and integrate in $E$ over $\left[E_{i},E_{i+1}\right]$. Now, approximate the integral $$\int_{E_{i}}^{E_{i+1}}\chi\left(\sigma_{i,j},E,T\left(\mathbf{x}\right)\right)\Psi\left(\mathbf{x},\boldsymbol{\Omega},E,\sigma_{i,j}\right)dE$$ by $$\chi\left(\sigma_{i,j},E_{i},T\left(\mathbf{x}\right)\right)\int_{E_{i}}^{E_{i+1}}\Psi\left(\mathbf{x},\boldsymbol{\Omega},E,\sigma_{i,j}\right)\Psi dE=\kappa_{ij}\left(T\left(\mathbf{x}\right)\right)\Psi_{ij}\left(\mathbf{x},\boldsymbol{\Omega},E\right).$$ This step is accurate since, by assumption, $\chi\left(\xi,E,T\left(\mathbf{x}\right)\right)$ smoothly varies in energy $E$. We perform a similar calculation for the integral of $\chi\left(\sigma_{i,j},E,T\left(\mathbf{x}\right)\right)S\left(T\left(\mathbf{x}\right),E\right)$ over $\left[E_{i},E_{i+1}\right]$, and obtain the multigroup-type system (\[eq:transport multiband-multigroup, intro\]).
To derive (\[eq:formula for psi\_o, discretized Young measures\]), we use the discrete approximation (\[eq:discrete approx of Young measure, intro\]) in (\[eq:psi\_0 as weighting against Young measure, intro\]), $$\begin{aligned}
\psi_{0}\left(\mathbf{x},\boldsymbol{\Omega},E\right) & = & \int_{0}^{\infty}\Psi\left(\mathbf{x},\boldsymbol{\Omega},E,\xi\right)\lambda_{E}\left(d\xi\right)\\
& \approx & \int_{0}^{\infty}\Psi\left(\mathbf{x},\boldsymbol{\Omega},E,\xi\right)\left(\sum_{j=1}^{m}p_{i,j}\delta\left(\xi-\sigma_{i,j}\right)\right)d\xi\\
& = & \sum_{j=1}^{m}p_{i,j}\Psi\left(\mathbf{x},\boldsymbol{\Omega},E,\sigma_{i,j}\right).\end{aligned}$$ Finally, integrating both sides of $$\psi_{0}\left(\mathbf{x},\boldsymbol{\Omega},E\right)=\sum_{j=1}^{m}p_{i,j}\Psi\left(\mathbf{x},\boldsymbol{\Omega},E,\sigma_{i,j}\right)$$ over $\left[E_{i},E_{i+1}\right]$, we obtain (\[eq:formula for psi\_o, discretized Young measures\]).
Relaxation of the correlated opacity assumption\[sub:Relaxation-of-the\]
------------------------------------------------------------------------
Our derivation of the homogenized system assumed that the correlated assumption (\[eq:basic opacity assumption\]) is an equality for appropriate functions $\chi\left(x,E,T\right)$ and $\sigma_{\varepsilon}\left(E\right)$. In this section, we assume instead that $$\sigma_{\varepsilon}\left(E,T\right)\approx\chi\left(\sigma_{\varepsilon}\left(E,T_{0}\right),E,T\right),\label{eq:correlated assumption, relaxation section}$$ *approximately* holds, where $T_{0}$ is some appropriate reference temperature. That is, we approximate the opacity at a general temperature $T$ as functionally related to the opacity as some reference temperature $T_{0}$. As discussed below, we also include the possibility of slow energy variation in the functional relationship (\[eq:correlated assumption, relaxation section\]), which arises naturally in our following discussion. In general, the correlated assumption (\[eq:correlated assumption, relaxation section\]) does not hold. We therefore discuss in this section how to compute a function $\chi\left(x,E,T\right)$ that best approximates $\sigma_{\varepsilon}\left(E,T\right)\approx\chi\left(\sigma_{\varepsilon}\left(E,T_{0}\right),E,T\right)$.
To motivate the basic idea, consider two random variables $X\geq0$ and $Y\geq0$ and the associated joint probability density $p\left(x,y\right)=\mathbb{P}\left(X=x,Y=y\right)$. Then it is a well-known fact that the conditional expected value, $$\chi\left(x\right)\equiv\int_{0}^{\infty}yp\left(x,y\right)dy,\label{eq:conditional expected value}$$ minimizes the mean squared error $$\mathbb{E}\left(Y-\chi\left(X\right)\right)^{2}=\min_{g}\mathbb{E}\left(Y-g\left(X\right)\right)^{2}\equiv\int_{0}^{\infty}\int_{0}^{\infty}\left(y-g\left(x\right)\right)^{2}p\left(x,y\right)dxdy,$$ among all ($X$-measurable) functions $g$. In other words, $\chi\left(X\right)$ is the best functional fit to $Y$ in the sense of minimizing the average mean squared error.
This naturally leads us to consider the joint Young measure $\lambda_{E,T}\left(\xi_{1},\xi_{2}\right)$ associated with the pair of functions $\left(\sigma_{\varepsilon}\left(E,T_{0}\right),\sigma_{\varepsilon}\left(E,T\right)\right)$ (we explicitly include the temperature $T$ in the notation $\lambda_{E,T}$ to emphasize this dependence). Heuristically, $\lambda_{E,T}\left(\xi_{1},\xi_{2}\right)$ gives the probability density that $\sigma_{\varepsilon}\left(E',T_{0}\right)=\xi_{1}$ and $\sigma_{\varepsilon}\left(E',T\right)=\xi_{2}$ for $E'$ in a small neighborhood of $E-\delta\leq E'\leq E+\delta$, where $\delta$ is large relative to the characteristic line spacing $\varepsilon$ but small relative to the macroscopic variation of interest (i.e., $\varepsilon\ll\delta\ll1$). For small $\varepsilon$, we then have from (\[eq:conditional expected value\]) that the conditional expected value, $$\chi\left(x,E,T\right)\equiv\int_{0}^{\infty}\xi_{2}\lambda_{E,T}\left(x,d\xi_{2}\right),\label{eq: chi, conditional expected value}$$ approximately minimizes the average error, $$\mathbb{E}\left(\sigma_{\varepsilon}\left(E,T\right)-\chi\left(\sigma_{\varepsilon}\left(E,T_{0}\right),E,T\right)\right)^{2}\equiv\int_{0}^{\infty}\int_{0}^{\infty}\left(\xi_{2}-\chi\left(\xi_{1}\right)\right)^{2}\lambda_{E,T}\left(d\xi_{1},d\xi_{2}\right),$$ in the limit of small $\varepsilon$. Note that, if the assumption (\[eq:correlated assumption, relaxation section\]) exactly holds, then the joint probability measure $\lambda_{E}\left(\xi_{1},\xi_{2}\right)$ is supported on the curve $\xi_{2}=\chi\left(\xi_{1}\right)$.
To numerically approximate (\[eq: chi, conditional expected value\]), suppose that $\Delta E$ is chosen so that $0<\varepsilon\ll\Delta E$. Divide the range of $\sigma_{\varepsilon}\left(E',T_{0}\right)$ and $\sigma_{\varepsilon}\left(E',T\right)$, $E'\in\left[E-\Delta E/2,E+\Delta E/2\right]$, into temperature-dependant and energy-dependant “bands” $\sigma_{j}\left(E,T_{0}\right)$ and $\sigma_{j'}\left(E,T\right)$. For example, in Section \[sub:An-atmospheric-TRT\], we take logarithmically spaced bands between the minimum and maximum opacity values in $\left[E-\Delta E/2,E+\Delta E/2\right]$. Now define the discrete probabilities $$p_{j,k}\left(E,T\right)\approx\frac{\lambda\left(E'\in\left[E-\Delta E/2,E+\delta/2\right]\mid\sigma_{j}\left(E,T\right)\leq\sigma_{\varepsilon}\left(E',T\right)\leq\sigma_{j+1}\left(E,T\right),\sigma_{j'}\left(E,T_{0}\right)\leq\sigma_{\varepsilon}\left(E',T_{0}\right)\leq\sigma_{j'+1}\left(E,T_{0}\right)\right)}{\delta},\label{eq:p_=00007Bi,j=00007D(E,T)}$$ where $\lambda$ again denotes the Lebesgue measure. Then using the same reasoning as in Section \[sub:Discrete-approximation-of\], we approximate $\chi\left(x,E,T\right)$ as $$\chi\left(x,E,T\right)\approx\sum_{j'}p_{j,j'}\left(E,T\right)\sigma_{j'}\left(E,T\right),\,\,\,\,\text{if}\,\,\,\,\sigma_{j}\left(E,T_{0}\right)\leq x<\sigma_{j+1}\left(E,T_{0}\right).$$ In practice, we compute the discrete probabilities $p_{j,j'}\left(E,T\right)$ by uniformly sampling energy values $E'$ in $\left[E-\Delta E/2,E+\Delta E/2\right]$ and counting the number of samples for which $\sigma_{j}\left(E,T\right)\leq\sigma_{\varepsilon}\left(E',T\right)\leq\sigma_{j+1}\left(E,T\right)$ and $\sigma_{j'}\left(E,T_{0}\right)\leq\sigma_{\varepsilon}\left(E',T_{0}\right)\leq\sigma_{j'+1}\left(E,T_{0}\right)$.
We approximate the joint Young measure by $$\lambda_{E,T}\left(\xi_{1},\xi_{2}\right)\approx\sum_{j,j'}p_{j,j'}\left(E,T\right)\delta\left(\xi_{1}-\sigma_{j}\left(E,T_{0}\right)\right)\delta\left(\xi_{2}-\sigma_{j'}\left(E,T\right)\right).$$ Then from (\[eq: chi, conditional expected value\]), $$\begin{aligned}
\chi\left(x,E,T\right) & \approx & \int_{0}^{\infty}\xi_{2}\left(\sum_{j,j'}p_{j,j'}\left(E,T\right)\delta\left(x-\sigma_{j}\left(E,T_{0}\right)\right)\delta\left(\xi_{2}-\sigma_{j'}\left(E,T\right)\right)\right)d\xi_{2}\\
& = & \sum_{j,j'}\sigma_{j'}\left(E,T\right)p_{j,j'}\left(E,T\right)\delta\left(x-\sigma_{j}\left(E,T_{0}\right)\right).\end{aligned}$$ Integrating $x$ from $\sigma_{j}\left(E,T_{0}\right)$ to $\sigma_{j+1}\left(E,T_{0}\right)$, $$\int_{\sigma_{j}\left(E,T_{0}\right)}^{\sigma_{j+1}\left(E,T_{0}\right)}\chi\left(x,E,T\right)dx\approx\sum_{j'}\sigma_{j'}\left(E,T\right)p_{j,j'}\left(E,T\right).$$
To summarize: in the discrete version of the homogenized system (\[eq:transport multiband-multigroup, intro\]), we take $$\kappa_{ij}\left(T\right)=\chi\left(\sigma_{i,j},E_{i},T\right)\approx\sum_{j'}p_{j,j'}\left(E_{i},T\right)\sigma_{j'}\left(E_{i},T\right),\label{eq:kappa_=00007Bi,j=00007D}$$ where $E_{i}$ denotes the left end point of the $i$th coarse group, the temperature-dependent probabilities $p_{j,k}\left(E_{i},T\right)$ are defined by (\[eq:p\_=00007Bi,j=00007D(E,T)\]), and the temperature-dependent bands $\sigma_{k}\left(E_{i},T\right)$ are, e.g., logarithmically spaced between $\min_{E\in\left[E_{i},E_{i+1}\right]}\sigma_{\varepsilon}\left(E,T\right)$ and $\max_{E\in\left[E_{i},E_{i+1}\right]}\sigma_{\varepsilon}\left(E,T\right)$. Notice that the probabilities $p_{j,j'}\left(E,T\right)$ may be pre-computed on a fine energy and temperature grid and evaluated at arbitrary energy and temperature values via interpolation.
Numerical Examples\[sec:Examples\]
==================================
We apply this methodology to three examples. We first consider in Section \[sub:Example-1:-the\] the radiative transfer equation at constant temperature and using the Elsasser band opacity (\[eq:Elsasser band model, intro\]), where we take the line spacing $\varepsilon=10^{-4}$; this example is also considered in [@POMRAN:1971]. For this simple but instructive example, we can compute the Young measure analytically and compare it to its discrete approximation.
In our second example, we consider a neutron transport problem using the absorption cross section for iron at room temperature and a Watt fission spectrum for our source. Whereas opacities contain lines, nuclear cross sections for neutron applications contain resonances, which are similar to lines. The cross sections in natural iron contain thousands of fine resonances much like previous examples contained many lines.
Our final example is an atmospheric TRT calculation using $12$ homogeneous atmospheric layers from $0-15$ km, and taking a cross-section corresponding to water vapor over the frequency interval $1000\leq\nu\leq2000$ (in units of 1/cm); the cross-section for water vapor exhibits thousands of lines in this frequency range.
Let us discuss the approximation scheme first for Sections \[sub:Example-1:-the\] and \[sub:Example-2:-iron\], since the discretization schemes are essentially identical; we discuss the approximation scheme for the atmospheric problem in more detail in Section \[sub:An-atmospheric-TRT\].
We consider a transport equation of the form $$\mu\partial_{x}\psi_{\varepsilon}\left(x,\mu,E\right)+\sigma_{\varepsilon}\left(E\right)\psi_{\varepsilon}\left(x,\mu,E\right)=S\left(E\right).\label{eq:transport in slab geometry}$$ In Section \[sub:Example-1:-the\], $\sigma_{\varepsilon}\left(E\right)$ denotes the Elasser band opacity (\[eq:Elsasser band model, intro\]) and $S\left(E\right)$ denotes $\sigma_{\varepsilon}\left(E\right)B\left(E,T\right)$, with $B\left(E,T\right)$ denoting the Planck function at constant temperature; in Section \[sub:Example-2:-iron\], $\sigma_{\varepsilon}\left(E\right)$ denotes the cross-section for iron at room temperature and $S\left(E\right)$ denotes a Watts fission spectrum.
To compute the discrete approximation of the Young measure $\lambda_{E}^{T}$, we approximate for each energy group $\left[E_{i},E_{i+1}\right]$, $$\mu_{i}^{T}\left(\sigma\right)\approx\sum_{j=1}^{m}p_{i,j}\delta\left(\sigma-\sigma_{i,j}\right),\label{eq:discrete rep of Young measure, example}$$ where $p_{i,j}$ is proportional to the probability that $\sigma_{i,j}\leq\sigma_{\varepsilon}\left(E\right)\leq\sigma_{i,j+1}$ for $E_{i}\leq E\leq E_{i+1}$; $p_{i,j}$ is computed by uniformly sampling random numbers from $\left[E_{i},E_{i+1}\right]$ (the number of samples is chosen to be much larger than the number of energy values needed to resolve $\sigma_{\varepsilon}\left(E\right)$ in $\left[E_{i},E_{i+1}\right]$), counting how many times $ $$\sigma_{i,j}\leq\sigma_{\varepsilon}\left(E\right)\leq\sigma_{i,j+1}$ for each band $\left[\sigma_{i,j},\sigma_{i,j+1}\right]$, and normalizing by the total number of samples. We emphasize that, although evaluating $p_{i,j}$ scales linearly in $\varepsilon^{-1}$, this is a pre-computation and need only be done once; this pre-computation is analogous to computing Planck-weighted or Rosseland-weighted opacities.
Using the discrete representation (\[eq:discrete rep of Young measure, example\]), we have that for $\mu>0$, $$\begin{aligned}
\lim_{\varepsilon\rightarrow0}\int_{E_{i}}^{E_{i+1}}\psi_{\varepsilon}\left(x,\mu,E\right)dE & = & \int_{E_{i}}^{E_{i+1}}S\left(E\right)\left(\int_{0}^{\mathbb{R}}\frac{\left(1-e^{-x\left(\sigma/\mu\right)}\right)}{\sigma}d\lambda_{E}^{T}\left(\sigma\right)\right)dE\nonumber \\
& \approx & \int_{E_{i}}^{E_{i+1}}S\left(E\right)\left(\sum_{j=1}^{m}p_{i,j}\frac{\left(1-e^{-x\left(\sigma_{i,j}/\mu\right)}\right)}{\sigma_{i,j}}\right)dE\nonumber \\
& = & S_{i}\sum_{j=1}^{m}p_{i,j}\frac{\left(1-e^{-x\left(\sigma_{i,j}/\mu\right)}\right)}{\sigma_{i,j}},\label{eq:mu > 0, homogenized}\end{aligned}$$ where $$S_{i}=\int_{E_{i}}^{E_{i+1}}S\left(E\right)dE.$$ Similarly, for $\mu<0$, $$\lim_{\varepsilon\rightarrow0}\int_{E_{i}}^{E_{i+1}}\psi_{\varepsilon}\left(s,E\right)dE\approx S_{i}\sum_{j=1}^{m}p_{i,j}\left(1-e^{-\left(x-1\right)\left(\sigma_{i,j}/\mu\right)}\right).\label{eq:mu < 0, homogenized}$$
In Sections \[sub:Example-1:-the\] and \[sub:Example-2:-iron\], we compare the discrete approximation, (\[eq:mu > 0, homogenized\])-(\[eq:mu < 0, homogenized\]), to the exact solution integrated over $\left[E_{i},E_{i+1}\right]$, $$\int_{E_{i}}^{E_{i+1}}\psi_{\varepsilon}\left(x,\mu,E\right)dE=\int_{E_{i}}^{E_{i+1}}S\left(E\right)\frac{\left(1-e^{-\left(\sigma_{\varepsilon}\left(E\right)/\mu\right)x}\right)}{\sigma_{\varepsilon}\left(E\right)}dE,$$ for both the Elsasser band (\[eq:Elsasser band model, intro\]) and for iron opacity as generated via the NJoy [@NJOY] program. More precisely, we compute the exact energy-integrated solution, $$\int_{E_{i}}^{E_{i+1}}\psi_{\varepsilon}\left(x_{i},\mu_{j},E\right)dE,$$ for $10$ equispaced spatial points $x_{i}\in\left[0,1\right]$ and for $8$ Gauss-Legendre nodes $\mu_{j}$, as well as the homogenized energy-integrated solution. We compare the “exact” scalar flux (that is, exact to within angular discretization errors), $$\varphi_{\varepsilon}\left(x_{k}\right)=\sum_{j}\sum_{i}\int_{E_{i}}^{E_{i+1}}\psi_{\varepsilon}\left(x_{k},\mu_{j},E\right)dE,\label{eq:discrete scalar flux, exact}$$ against its homogenized version $$\varphi_{0}\left(x_{k}\right)=\sum_{j}\sum_{i}\int_{E_{i}}^{E_{i+1}}\psi_{0}\left(x_{k},\mu_{j},E\right)dE.\label{eq:discrete scalar flux, approx}$$
We compare the results to the standard multigroup method using Planck-weighted and Rosseland-weighted opacities. In particular, we integrate (\[eq:transport equation, intro\]) over $\left[E_{i},E_{i+1}\right]$, $$\mu\partial_{x}\int_{E_{g}}^{E_{g+1}}\psi_{\varepsilon}\left(x,\mu,E\right)dE=\int_{E_{g}}^{E_{g+1}}\sigma_{\varepsilon}\left(E,T\right)\left(\frac{S\left(E\right)}{\sigma_{\varepsilon}\left(E\right)}-\psi_{\varepsilon}\left(x,\mu,E\right)\right)dE.$$ We write $$\int_{E_{g}}^{E_{g+1}}\sigma_{\varepsilon}\left(E\right)\left(\frac{S\left(E\right)}{\sigma_{\varepsilon}\left(E\right)}-\psi_{\varepsilon}\left(x,\mu,E\right)\right)dE\approx\sigma_{g}\left(\int_{E_{g}}^{E_{g+1}}\frac{S\left(E\right)}{\sigma_{\varepsilon}\left(E\right)}dE-\int_{E_{g}}^{E_{g+1}}\psi_{\varepsilon}\left(x,\mu,E\right)dE\right),$$ where $$\sigma_{g}=\frac{\int_{E_{g}}^{E_{g+1}}S\left(E\right)dE}{\int_{E_{g}}^{E_{g+1}}S\left(E\right)/\sigma_{\varepsilon}\left(E\right)dE}.$$ Thus, we need to solve $$\mu\partial_{x}\psi_{g}\left(x,\mu\right)+\sigma_{g}\psi_{g}\left(x,\mu\right)=S_{g}\left(T\right),$$ where $$S_{g}\left(T\right)=\int_{E_{g}}^{E_{g+1}}\frac{S\left(E\right)}{\sigma_{\varepsilon}\left(E\right)}dE,\,\,\,\,\sigma_{g}=\frac{\int_{E_{g}}^{E_{g+1}}S\left(E\right)dE}{\int_{E_{g}}^{E_{g+1}}S\left(E\right)/\sigma_{\varepsilon}\left(E\right)dE}.$$
A regular band model example\[sub:Example-1:-the\]
--------------------------------------------------
In order to assess the accuracy of the discrete approximation of the Young measure $\lambda_{E}$ associated with (\[eq:Elsasser band model, intro\]), we first compute $\lambda_{E}$ analytically (note that, since (\[eq:Elsasser band model, intro\]) does not depend on temperature $T$, we drop the superscript $T$ on **$\lambda_{E}^{T}$**).
Using that $\sigma_{\varepsilon}\left(E\right)=\sigma_{1}\left(E/\varepsilon\right)$, with $\sigma_{1}\left(E\right)$ a $1$-periodic function, it is a standard result (see e.g. [@BESICO:1954]) that $$\lim_{\varepsilon\rightarrow0}\int_{E_{0}}^{E_{1}}f\left(\sigma_{\varepsilon}\left(E'\right)\right)dE'=\int_{0}^{1}f\left(\sigma_{1}\left(E'\right)\right)dE',$$ for any values $0\leq E<E_{1}\leq1$. Now, consider the change of variables, $$\xi=\frac{\cosh\left(\beta\right)+1}{\cosh\left(\beta\right)-\cos\left(2\pi E\right)}.$$ Then $$d\xi=-\frac{2\pi}{\cosh\left(\beta\right)+1}\xi^{2}\sqrt{1-\frac{\left(\xi\cosh\left(\beta\right)-\cosh\left(\beta\right)-1\right)^{2}}{\xi^{2}}}dE.$$ It follows that, with $c_{\beta}=\left(\cosh\left(\beta\right)+1\right)/\left(\cosh\left(\beta\right)-1\right)$, $$\int_{0}^{1}f\left(\sigma_{1}\left(E'\right)\right)dE'=\frac{\cosh\left(\beta\right)+1}{2\pi}\int_{1}^{c_{\beta}}\frac{f\left(\xi\right)}{\xi^{2}\sqrt{1-\xi^{-2}\left(\sigma\cosh\left(\beta\right)-\cosh\left(\beta\right)-1\right)^{2}}}d\xi.$$ Therefore, the Young measure is given by $$d\lambda_{E}\left(\xi\right)=\frac{1}{\xi^{2}\sqrt{1-\xi^{-2}\left(\sigma\cosh\left(\beta\right)-\cosh\left(\beta\right)-1\right)^{2}}}\chi_{\left[1,c_{\beta}\right]}\left(\xi\right)d\xi.\label{eq:analytic Young measure}$$ As expected, $\lambda_{E}\left(\xi\right)$ is independent of energy $E$.
We follow the discussion in (\[sub:Discrete-approximation-of\]) and approximate $$\lambda_{E}\left(\xi\right)\approx\sum_{j=1}^{m}p_{j}\delta\left(\xi-\xi_{j}\right).\label{eq:discrete approx of Young measure, Elsasser}$$ For this example, $p_{j}$ and $\xi$ are independent of $E$. In Figure \[fig:young-measure\], we compare the Young measure (\[eq:analytic Young measure\]) associated with (\[eq:Elsasser band model, intro\]) against its discrete approximation (\[eq:discrete approx of Young measure, Elsasser\]); in this example, we take $\beta=1$ and, in (\[eq:discrete approx of Young measure, Elsasser\]), $m=30$ equispaced values $\sigma_{j}$ between $1$ and $c_{\beta}$, $\beta=1$. Note that the error in the discrete approximation (\[eq:discrete approx of Young measure, Elsasser\]) necessarily rises at the end points, since the density $d\lambda_{E}\left(\xi\right)$ is infinite at $\sigma=1$ and $\sigma=c_{\beta}$.
Using (\[eq:discrete approx of Young measure, Elsasser\]) in (\[eq:mu > 0, homogenized\])-(\[eq:mu < 0, homogenized\]), we then obtain the (approximate) homogenized solution $\psi_{0}\left(x,\mu,E\right)$. In Figure \[fig:transport comparison, Elsasser\], we compare the “exact” scalar flux (\[eq:discrete scalar flux, exact\]) against its homogenized version (\[eq:discrete scalar flux, approx\]), where $\psi_{0}$ is computed from (\[eq:discrete approx of Young measure, Elsasser\]) and (\[eq:mu > 0, homogenized\])-(\[eq:mu < 0, homogenized\]). We use $m=30$ equispaced values $\sigma_{j}$ between $1$ and $c_{\beta}$, $\beta=1$, and a single group $\left[E_{0},E_{1}\right]=\left[0,1\right]$.
![Comparison of the Young measure (\[eq:analytic Young measure\]) and its discrete approximation (\[eq:discrete approx of Young measure, Elsasser\]). We use $m=30$ “band boundaries” $\sigma_{j}$ in (\[eq:discrete approx of Young measure, Elsasser\]). \[fig:young-measure\]](young_measure_and_discrete_approx)
![Comparison of the exact scalar flux (\[eq:discrete scalar flux, exact\]) against its discrete approximation (\[eq:discrete scalar flux, approx\]), using $m=30$ equispaced values $\sigma_{j}$ between $1$ and $c_{\beta}$, $\beta=1$, and a single group $\left[E_{0},E_{1}\right]=\left[0,1\right]$. \[fig:transport comparison, Elsasser\]](scalar_flux_comparison_Elsasser)
A neutron transport example with absorption cross section for iron \[sub:Example-2:-iron\]
------------------------------------------------------------------------------------------
![Absorption cross section for iron.\[fig:opacity for iron\]](opacity_iron)
(a)
![Comparison of the exact scalar flux (\[eq:discrete scalar flux, exact\]) against its homogenized approximation (\[eq:discrete scalar flux, approx\]) and the multigroup method with Planck-weighted cross sections, using (a) $1,2024$, (b) $4,096$, (c) $16384$, and (d) $65,536$ equally spaced groups. Here the cross section is for iron as shown in Figure \[fig:opacity for iron\]. \[fig:comparison of scalar fluxes, iron\]](scalar_flux_comparison_iron_10 "fig:")
(b)
![Comparison of the exact scalar flux (\[eq:discrete scalar flux, exact\]) against its homogenized approximation (\[eq:discrete scalar flux, approx\]) and the multigroup method with Planck-weighted cross sections, using (a) $1,2024$, (b) $4,096$, (c) $16384$, and (d) $65,536$ equally spaced groups. Here the cross section is for iron as shown in Figure \[fig:opacity for iron\]. \[fig:comparison of scalar fluxes, iron\]](scalar_flux_comparison_iron_14)
(c)
![Comparison of the exact scalar flux (\[eq:discrete scalar flux, exact\]) against its homogenized approximation (\[eq:discrete scalar flux, approx\]) and the multigroup method with Planck-weighted cross sections, using (a) $1,2024$, (b) $4,096$, (c) $16384$, and (d) $65,536$ equally spaced groups. Here the cross section is for iron as shown in Figure \[fig:opacity for iron\]. \[fig:comparison of scalar fluxes, iron\]](scalar_flux_comparison_iron_16)
![Relative error of the homogenized scalar flux (\[eq:discrete scalar flux, approx\]), as compared to the exact scalar flux (\[eq:discrete scalar flux, exact\]), using the cross section for iron shown in Figure \[fig:opacity for iron\]. \[fig:error in scalar flux, iron\]](error_scalar_flux_comparison_iron)
We consider a neutron transport example in slab geometry, which describes the uncollided neutron flux and is useful in shielding applications. In particular, in equation (\[eq:transport in slab geometry\]), we take $\sigma_{\varepsilon}\left(E\right)$ to correspond to the absorption cross section of iron at room temperature over the energy range $50$ keV to $10$ MeV, as displayed in Figure \[fig:opacity for iron\]; energy/cross-section pairs $\left(E_{j},\sigma_{j}\right)$ for iron at were generated via the NJoy problem and the cross section $\sigma_{\varepsilon}\left(E\right)$ at a general value $E$ is evaluated via linear interpolation. The subscript $\varepsilon$ on $\sigma_{\varepsilon}\left(E\right)$ is retained for notational consistency, and represents the characteristic resonance spacing; note, however, that $\varepsilon$ is not a free parameter. We use a source term $S\left(E\right)$ corresponding to a Watt fission spectrum,
$$\begin{aligned}S(E) & =c\exp\left(-E/a\right)\sinh\sqrt{bE}\end{aligned}
,$$ with $a=0.988$ MeV, $b=2.2249$ MeV$^{-1}$, and $c=\exp\left(-ab/4\right)/\sqrt{\left(\pi a^{3}b/4\right)}$ MeV$^{-1}$ a normalization constant.
As in Section \[sub:Example-1:-the\], we compare the exact scalar flux (\[eq:discrete scalar flux, exact\]) (that is, exact to within angular discretization errors) against its homogenized version (\[eq:discrete scalar flux, approx\]). We take $4$ energy groups $\left[E_{i},E_{i+1}\right]$, where the group boundaries $E_{i}$ are equally spaced between $E_{\min}=50.002$ (KeV) and $E_{\max}=10^{4}$ (KeV) Within each energy group $\left[E_{i},E_{i+1}\right]$, we use $m=40$ equispaced values $\sigma_{j}$ between $$\sigma_{\min,i}=\min_{E_{i}\leq E\leq E_{i+1}}\sigma_{\varepsilon}\left(E\right),$$ and $$\sigma_{\max,i}=\max_{E_{i}\leq E\leq E_{i+1}}\sigma_{\varepsilon}\left(E\right).$$ In Figure \[fig:comparison of scalar fluxes, iron\], we compare the homogenized scalar flux with the exact scalar flux, as well as with the scalar flux obtained using the multigroup method with $1,024$, $ $ $16,384$, and $65,536$ equally groups. Figure \[fig:error in scalar flux, iron\] displays the relative errors in the scalar for in the homogenized scalar flux and the scalar flux using the Planck-weighted opacities; we see from this figure that the homogenized scalar flux and the multigroup method $16,384$, achieves about a $.1$ percent relative error in comparison to the exact scalar flux. Recall that the homogenization approach used $4\times40=160$ energy discretization parameters, and so for this accuracy this translates into orders of magnitude fewer parameters. We note that, for a larger error, the efficiency gain of the homogenized approach in this example is much less significant.
An atmospheric TRT example with water vapor\[sub:An-atmospheric-TRT\]
---------------------------------------------------------------------
![Cross-section $\kappa\left(\nu,T,p\right)$ (kg/m\^2) for water vapor as a function of frequency (1/cm), with $T=288.15$ (K) and $p=8.9874\times10^{4}$ (Pa).\[fig:Cross-section–(kg/m\^2)\]](cross-section_water_vapor)
Height (km) Temperature (K) Pressure (Pa) volume fraction (unitless)
------------- ----------------- --------------- ----------------------------
0-1 281.65 8.98746E+4 .0081
1-2 275.15 7.94952E+4 .0077
2-3 268.65 7.01085E+4 .0059
3-4 262.15 6.16402E+4 .0028
4-5 255.65 5.40199E+4 .0016
5-6 249.15 4.71810E+4 .0008
6-7 242.65 4.10607E+4 .0003
7-8 236.15 3.55998E+4 7.96E-5
8-9 229.65 3.07425E+4 3.21E-5
9-10 223.15 2.64363E+4 1.78E-5
10-12 216.65 1.93304E+4 6.94E-6
12-15 216.65 1.20446E+4 3.84E-6
: Temperature and pressure values in each homogeneous layer.\[tab:Temperature-and-pressure\]
(a)
![Comparison of the outgoing flux using solutions of the exact and homogenized equations (\[eq:psi, atmosphere\]) and (\[eq:Psi, atmosphere\]). In solving (\[eq:psi, atmosphere\]) using the homogenization approach, we use $10$ energy groups and $5$ bands. We also solve (\[eq:psi, atmosphere\]) using $2500$ and $5000$ Planck-weighted opacities. Plot (a) shows the exact $F_{\varepsilon}^{+}\left(x\right)$ and homogenized outgoing fluxes $F_{0}^{+}\left(x\right)$ (homogenized) and $F_{P}^{+}\left(x\right)$ (Planck). Plot (b) shows the relative errors in the homogenized solution and the Planck solutions. \[fig:comparison of line-by-line and homogenized outgoing fluxes\]](outgoing_fluxes)
(b)
![Comparison of the outgoing flux using solutions of the exact and homogenized equations (\[eq:psi, atmosphere\]) and (\[eq:Psi, atmosphere\]). In solving (\[eq:psi, atmosphere\]) using the homogenization approach, we use $10$ energy groups and $5$ bands. We also solve (\[eq:psi, atmosphere\]) using $2500$ and $5000$ Planck-weighted opacities. Plot (a) shows the exact $F_{\varepsilon}^{+}\left(x\right)$ and homogenized outgoing fluxes $F_{0}^{+}\left(x\right)$ (homogenized) and $F_{P}^{+}\left(x\right)$ (Planck). Plot (b) shows the relative errors in the homogenized solution and the Planck solutions. \[fig:comparison of line-by-line and homogenized outgoing fluxes\]](errors_in_outgoing_flux_all)
We consider an atmospheric model problem consisting of $12$ homogeneous layers of constant temperature $T\left(x\right)=T_{k}$ and pressure $p\left(x\right)=p_{k}$ for $x_{k}\leq x\leq x_{k+1}$ (see e.g. [@G-W-C-C:1989]); for consistency with traditional notation, we consider the frequency variable $\nu$ in place of the energy variable $E$. We solve the TRT equations using cross-sections $\hat{\sigma}\left(\nu,T,p\right)$, corresponding to water vapor, over the frequency range $1000-2000$ (1/cm). As standard for infrared atmospheric TRT calculations, we neglect scattering and assume slab (i.e., plane parallel) geometry. We also use the 1976 U. S. Standard Atmosphere values for temperature and pressure in each layer, as displayed in Table \[tab:Temperature-and-pressure\]. In addition, the volume fraction of water vapor in each layer is also shown in Table \[tab:Temperature-and-pressure\].
The cross-sections $\hat{\sigma}\left(\nu,T,p\right)$ (in units of kg/m\^3) are computed using the HITRAN database and assuming a Lorenz shape profile; in our calculations, we use the Julia module by J. Bloch-Johnson, https://github.com/jsbj/Jultran.jl/blob/master/src/Jultran.jl, for processing the HITRAN files in order to generate the opacities. Here $\log_{10}\left(\hat{\sigma}\left(\nu,T,p\right)\right)$ is plotted in Figure \[fig:Cross-section–(kg/m\^2)\] for $T=288.15$ (K) and $p=8.9874\times10^{4}$ (Pa).
We now describe the problem in more detail. In each homogeneous layer $x\in\left[x_{k},x_{k+1}\right]$, $1\leq k\leq12$, the intensity $\psi_{\varepsilon}\left(x,\mu,\nu\right)$ satisfies the transport equation $$\begin{aligned}
\mu\partial_{x}\psi_{\varepsilon}^{k}\left(x,\mu,\nu\right)+\sigma_{\varepsilon}\left(\nu,T_{k},p_{k}\right)\psi_{\varepsilon}^{k}\left(x,\mu,\nu\right) & = & \sigma_{\varepsilon}\left(\nu,T_{k},p_{k}\right)B\left(\nu,T_{k}\right),\\
\psi_{\varepsilon}^{k}\left(x_{k},\mu,\nu\right) & = & \psi_{\varepsilon}^{k-1}\left(x_{k},\mu,\nu\right),\,\,\,\,\,\,\,\,\mu>0,\\
\psi_{\varepsilon}^{k}\left(x_{k+1},\mu,\nu\right) & = & \psi_{\varepsilon}^{k+1}\left(x_{k+1},\mu,\nu\right),\,\,\,\,\,\,\,\,\mu<0.\label{eq:psi, atmosphere}\end{aligned}$$ Here $B\left(\nu,T\right)$ denotes the Planck function; $\sigma_{\varepsilon}\left(\nu,T_{k},p_{k}\right)=r_{k}\rho_{k}\hat{\sigma}_{\varepsilon}\left(\nu,T_{k},p_{k}\right)$, where $\rho_{k}$ and $r_{k}$ denote the density of air (kg/m\^3) and the fraction of water vapor in the $k$th layer and $\hat{\sigma}_{\varepsilon}\left(\nu,T_{k},p_{k}\right)$ denotes the cross-section corresponding to water vapor. We also take boundary conditions $$\begin{aligned}
\psi_{\varepsilon}\left(x_{0},\mu,\nu\right) & = & B\left(\nu,T_{0}\right),\,\,\,\,\mu>0,\\
\psi_{\varepsilon}\left(x_{K+1},\mu,\nu\right) & = & 0,\,\,\,\,\mu<0.\end{aligned}$$
For the homogenized transport equation (\[eq:transport multiband-multigroup, intro\]), we use in each spatial interval $\left[x_{k},x_{k+1}\right]$, $k=1,\ldots,12$, seven logarithmically spaced values $\kappa_{ij}\left(T_{k_{0}}\right)$, $j=1,\ldots,7$, for each each frequency group $\nu\in\left[\nu_{i},\nu_{i+1}\right]$ and for the reference temperature $T_{k_{0}}$ for layer $k_{0}=6$ (we find that the error is relatively insensitive to this choice); that is, for each frequency group $\left[\nu_{i},\nu_{i+1}\right]$, the bands $\log\left(\kappa_{ij}\left(T_{k_{0}}\right)\right)$, $j=1,\ldots,5$, are equally spaced between the minimum $\min_{\nu_{i}\leq\nu\leq\nu_{i+1}}\log\left(\sigma_{\varepsilon}\left(\nu,T_{k_{0}},p_{k_{0}}\right)\right)$ and the maximum $\max_{\nu_{i}\leq\nu\leq\nu_{i+1}}\log\left(\sigma_{\varepsilon}\left(\nu,T_{k_{0}},p_{k_{0}}\right)\right)$. For values of $k\neq k_{0}$ (i.e., other spatial intervals), we compute values $\kappa_{ij}\left(T_{k}\right)$ using equation (\[eq:kappa\_=00007Bi,j=00007D\]) and the techniques discussed in Section \[sub:Relaxation-of-the\].
From $\kappa_{ij}\left(T_{k}\right)$, we solve $$\begin{aligned}
\mu\partial_{x}\Psi_{ij}^{k}\left(x,\mu,\nu\right)+\kappa_{ij}\left(T_{k}\right)\Psi_{ij}^{k}\left(x,\mu,\nu\right) & = & \kappa_{ij}\left(T_{k}\right)B_{i}\left(T_{k}\right),\\
\Psi_{ij}^{k}\left(x_{k},\mu,\nu\right) & = & \Psi_{ij}^{k}\left(x_{k-1},\mu,\nu\right),\,\,\,\,\,\,\,\,\mu>0,\label{eq:Psi, atmosphere}\\
\Psi_{ij}^{k}\left(x_{k},\mu,\nu\right) & = & \Psi_{ij}^{k}\left(x_{k+1},\mu,\nu\right),\,\,\,\,\,\,\,\,\mu<0,\end{aligned}$$ where $$B_{i}\left(T_{k}\right)=\int_{\nu_{i}}^{\nu_{i+1}}B\left(\nu,T\right).$$ Given $\Psi_{ij}^{k}\left(x,\mu,\nu\right)$ the homogenized solution $\psi_{0}$ is computed via $$\psi_{0}\left(x,\mu,\nu\right)=\sum_{j}p_{ij}\Psi_{ij}^{k}\left(x,\mu,\nu\right),\,\,\,\, x\in\left[x_{k},x_{k+1}\right],\,\,\,\nu\in\left[\nu_{i},\nu_{i+1}\right],$$ where $p_{ij}$ denotes the probability that $\sigma_{j}\leq\sigma_{\varepsilon}\left(\nu,T_{k_{0}},p_{k_{0}}\right)\leq\sigma_{j+1}$ for $\nu_{i}\leq\nu\leq\nu_{i+1}$ (recall that $k_{0}=6$ is taken for the reference layer).
In the line-by-line solution of (\[eq:psi, atmosphere\]), we discretize in angle using $n_{\mu}=8$ Gaussian quadrature nodes $\mu_{p}$ and weights $w_{p}$, and $n_{\nu}=200,001$ equally spaced frequency points in $1000\leq\nu\leq2000$. In each spatial interval $\left[x_{k},x_{k+1}\right]$, we directly evaluate the analytic solution using $n_{x}=100$ equally spaced spatial points. For example, for $\mu>0$, we proceed from the first spatial interval $\left[x_{1},x_{2}\right]$ to the last spatial interval $\left[x_{11},x_{12}\right]$ and directly evaluate the analytic solution $$\psi_{\varepsilon}^{k}\left(x,\mu,\nu\right)=e^{-\left(\sigma_{\varepsilon}\left(\nu,T_{k},p_{k}\right)/\mu\right)\left(x-x_{k}\right)}\psi_{\varepsilon}^{k-1}\left(x_{k},\mu,\nu\right)+\left(1-e^{-\left(\sigma_{\varepsilon}\left(\nu,T_{k},p_{k}\right)/\mu\right)\left(x-x_{k}\right)}\right)B\left(\nu,T_{k}\right),$$ at $n_{x}$ equally spaced points $x_{k}\leq x_{q}^{k}\leq x_{k+1}$, with $x_{1}^{k}=x_{k}$ and $x_{n_{x}}^{k}=x_{k+1}$.
Similarly, in the solution of (\[eq:Psi, atmosphere\]), we discretize in angle using $n_{\mu}=8$ Gaussian quadrature nodes $\mu_{p}$ and weights $w_{p}$, and $10$ frequency groups $\left[\nu_{i},\nu_{i+1}\right]$. In each spatial interval $\left[x_{k},x_{k+1}\right]$ and each energy group $\left[\left[\nu_{i},\nu_{i+1}\right]\right]$, we use $7$ bands $\kappa_{ij}\left(T_{k}\right)$, that are equally spaced on a logarithmic scale. As in the solution of (\[eq:psi, atmosphere\]), we directly evaluate the analytic solution using $n_{x}=100$ equally spaced spatial points. For example, for $\mu>0$, we proceed from the first spatial interval $\left[x_{1},x_{2}\right]$ to the last spatial interval $\left[x_{12},x_{13}\right]$ and directly evaluate the analytic solution $$\Psi_{ij}^{k}\left(x,\mu\right)=e^{-\left(\kappa_{ij}\left(T_{k}\right)/\mu\right)\left(x-x_{k}\right)}\Psi_{ij}^{k-1}\left(x_{k},\mu\right)+\left(1-e^{-\left(\kappa_{ij}\left(T_{k}\right)/\mu\right)\left(x-x_{k}\right)}\right)B_{i}\left(T_{k}\right),$$ at $n_{x}$ equally spaced points $x_{k}\leq x_{q}^{k}\leq x_{k+1}$, with $x_{1}^{k}=x_{k}$ and $x_{n_{x}}^{k}=x_{k+1}$.
In Figure \[fig:comparison of line-by-line and homogenized outgoing fluxes\], we compare the line-by-line outgoing flux $$F_{\varepsilon}\left(x\right)=\sum_{p=1}^{n_{\mu}}\sum_{i=1}^{n_{\nu}}w_{p}\mu_{p}\left(\nu_{i+1}-\nu_{i}\right)\psi_{\varepsilon}\left(x,\mu_{p},\nu_{i}\right),$$ against its homogenized version $$F_{0}\left(x\right)=\sum_{p=1}^{n_{\mu}}\sum_{i=1}^{n_{G}}w_{p}\mu_{p}\psi_{0}\left(x,\mu_{p},\nu_{i}\right).$$ We also compare the exact outgoing flux that against that obtained using $5000$ Planck-weighted opacities. Plot (a) shows the line-by-line solution, the solution using $2500$ and $5000$ Planck-weighted frequency groups, and the homogenized solution using $10$ frequency groups and $7$ bands . We see from plot (b) in Figure \[fig:comparison of line-by-line and homogenized outgoing fluxes\] that, for a comparable error, the homogenized solution requires about $70\times$ fewer parameters than the solution obtained via Planck-weighted frequency groups.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'In spite of its ubiquity in strongly correlated systems, the competition of paired and nematic ground states remains poorly understood. Recently such a competition was reported in the two-dimensional electron gas at filling factor $\nu=5/2$. At this filling factor a pressure-induced quantum phase transition was observed from the paired fractional quantum Hall state to the quantum Hall nematic. Here we show that the pressure induced paired-to-nematic transition also develops at $\nu=7/2$, demonstrating therefore this transition in both spin branches of the second orbital Landau level. However, we find that pressure is not the only parameter controlling this transition. Indeed, ground states consistent with those observed under pressure also develop in a sample measured at ambient pressure, but in which the electron-electron interaction was tuned close to its value at the quantum critical point. Our experiments suggest that electron-electron interactions play a critical role in driving the paired-to-nematic transition.'
author:
- 'K.A. Schreiber'
- 'N. Samkharadze'
- 'G.C. Gardner'
- 'Y. Lyanda-Geller'
- 'M.J. Manfra'
- 'L.N. Pfeiffer'
- 'K.W. West'
- 'G.A. Csáthy'
title: ' Electron-Electron Interactions and the Paired-to-Nematic Quantum Phase Transition in the Second Landau Level'
---
[**Introduction.**]{}
Nematicity is of interest in various strongly correlated electron systems [@kivelson; @lilly; @du; @fogler; @moessner; @kiv; @fradkin]. It is generally accepted that nematicity originates from competing interactions on different length scales. However, the interplay of nematicity with other phases, such as with superconductivity in the cuprates [@high1; @high2], is not understood. For example, the influence of the nematic fluctuations on pairing in the superconductive phase is actively debated [@fluct1; @fluct2; @fluct3; @fluct4; @kim17].
Nematic and paired ground states also develop in half-filled Landau levels of the two-dimensional electron gas confined to high quality GaAs/AlGaAs structures. Indeed, a strong resistance anisotropy at the Landau level filling factors $\nu=9/2, 11/2, 13/2, ...$ signals a ground state state with broken rotational symmetry [@lilly; @du]. There are two distinct ground states consistent with such an anisotropy: the smectic and nematic phases [@kiv; @fradkin; @fogler; @moessner; @nem7]. The difference between these two is that the former has unidirectional translational order, whereas the latter does not. Since electrons are buried deep within the GaAs crystal, a direct detection of translational order remains elusive. In addition, disorder is expected to destroy translational order and therefore it favors nematicity. In lack of certainty about the translational order, these anisotropic phases are often referred to as the quantum Hall nematic, or simply the nematic. Henceforth we adopt this terminology.
In contrast to the ground states at filling factors $\nu=9/2, 11/2, 13/2,...$, those at $\nu=5/2$ and $7/2$ are isotropic fractional quantum Hall states (FQHSs) [@firstfivehalf; @fivehalf; @sevenhalf; @MooreRead]. Since in GaAs each orbital Landau level has two spin branches, $\nu=5/2$ and $7/2$ describe two half-filled spin banches of the second orbital Landau level. FQHSs are incompressible and possess topological order. Topological order in the FQHSs at $\nu=5/2$ and $7/2$ is still under active investigation. However, within the framework of the composite fermion theory [@jain; @halperin], these FQHSs are due to pairing of the composite fermions, hence the paired FQHS terminology [@pair1; @pair2; @pair3; @pair4]. A schematic representation of the ordered ground states at half-filling is seen in Fig.1.
![[**Schematic of the ordered phases at even denominator filling factors.**]{} [**a**]{} The paired FQHS consists of Cooper pairs of composite fermions and posesses edge states [@pair1; @pair2; @pair3; @pair4]. Composite fermions are depicted as electrons with two magnetic fluxlines attached [@jain]. [**b**]{} The quantum Hall nematic is a filamentary electronic phase which breaks rotational symmetry [@kiv]. []{data-label="fig-1"}](fig-1){width="0.95\columnwidth"}
{width="95.00000%"}
At a given orbital quantum number, the ordered ground state at half-filled Landau level is either the nematic or the FQHS, but a transition between them did not seem possible in the absence of a symmetry breaking field favoring the nematic. This was surprising, since in the second orbital Landau level at $\nu=5/2$ and $7/2$ tilted field experiments suggested that the two ground states are close in energy [@tilt1; @tilt2]. Additionally, incipient nematicity was seen at $\nu=7/2$ [@Pan72]. However, a phase transition from the FQHS to the nematic in the absence of an in-plane symmetry breaking magnetic field was only recently observed [@Samkharadze]. In these experiments the transition occurred at $\nu=5/2$ and it was driven by pressure. Because of the hydrostatic nature of the applied pressure, the rotational symmetry in these experiments was not explicitly broken.
Our understanding of the paired-to-nematic phase transition and the associated quantum critical point remains lacunar. Tuning the Haldane pseudopotentials in the second Landau level induces a transition from the paired FQHS to the nematic [@pair4]. However, the very nature of this transition remains unknown. Recent theories find that the nematic phase is stabilized by a Pomeranchuk instability of the Fermi sea of composite fermions [@frad; @rez18]. A paired-to-nematic transition is compatible with these theories, but details have not yet been worked out. In another work, the influence of the nematic fluctuations on the paired FQHS has been examined, with the assumption that a paired-to-nematic transition exists [@kim17]. Ref.[@inti17] captures a paired-to-nematic transition by tuning the mass anisotropy of the carriers. However, there is no evidence that such a mass anisotropy plays a significant role in the electron gas hosted in GaAs. What determines the quantum critical point? Can the transition be induced using a parameter other than pressure?
Guided by these questions, here we investigate the ground state of the two-dimensional electron gas in a wider phase space. We establish that the paired-to-nematic transition also occurs at filling factor $\nu=7/2$, the particle-hole conjugate of $\nu=5/2$. However, this transition is not observed outside the second Landau level nor at unpaired FQHSs forming in the second Landau level. This finding highlights the importance of pairing in the transition from a FQHS to the nematic and establishes the presence of the paired-to-nematic transition and the associated quantum critical point in both spin branches of the half-filled second orbital Landau level. We observe that the critical pressure of the transition at $\nu=7/2$ is much reduced when compared to that at $\nu=5/2$. In contrast, we find that the transition occurs at nearly the same magnetic field. This observation allows us to conclude that pressure is not a primary driver of the transition, but the electron-electron interaction is. To demonstrate this, we show that ground states consistent with those at high pressures also develop in at [*ambient pressure*]{}, but in which the electron-electron interaction is engineered to be close to its critical value.
[**Results.**]{}
[**Samples.**]{} We measured two samples. Sample A is a 30 nm quantum well sample with an as-grown density of $29.0 \times 10^{10}$ cm$^{-2}$, which was investigated under hydrostatic pressure. The mobility of this sample in the ambient is $20 \times10^{6}$ cm$^2$V$^{-1}$s$^{-1}$. Sample B is also a 30 nm quantum well sample, but with an as-grown density of $10.9 \times 10^{10}$ cm$^{-2}$. Sample B was measured only at ambient pressure and has a mobility of $18 \times10^{6}$ cm$^2$V$^{-1}$s$^{-1}$.
{width="90.00000%"}
[**Terminology.**]{} The energy spectrum of a two-dimensional electron gas of density $n$ in a magnetic field $B$ at large enough fields consists of spin-split Landau levels. The number of filled energy levels is given by the Landau level filling factor $\nu=h n / e B$, where $e$ is the electron charge and $h$ is Planck’s constant. In the absence of the valley degree of freedom, the second orbital Landau level in GaAs corresponds to the $2< \nu < 4$ range. Of this range, the $2< \nu < 3$ is the lower spin branch, while the $3< \nu < 4$ range the upper spin branch. Therefore at $\nu=5/2$ and $\nu=7/2$ the system has half-filled Landau levels with the same orbital quantum number, but different spin quantum numbers.
[**Pressure dependent magnetoresistance at low temperatures.**]{} Figure 2 highlights the evolution of the magnetoresistance in the two spin branches of the second orbital Landau level at the lowest temperature of $T \approx 12$ mK reached in our pressure cell. Traces are measured along two mutually perpendicular directions: $R_{xx}$ along the $[1\bar{1}0]$ and $R_{yy}$ along the $[110]$ crystal axis of GaAs. These traces show several features which can be associated with known ground states of the electron gas at ambient pressure [@Xia; @Ethan]; in the following we focus our attention to $\nu=5/2$ and $\nu=7/2$. The magnetoresistance at $\nu=5/2$ is isotropic and vanishing at $3.26$ and $7.22$ kbar, signaling a FQHS [@firstfivehalf; @fivehalf]. The magnetoresistance at $\nu=5/2$ is strongly anisotropic at $9.26$ kbar and has very little anisotropy at $10.54$ kbar, exhibiting therefore nematic behavior [@lilly; @du]. This behavior with increasing pressure is consistent with a FQHS, quantum Hall nematic, isotropic Fermi fluid sequence of ground states [@Samkharadze].
The magnetoresistance trend at $\nu=7/2$ shown in Fig.2 is qualitatively similar to that at $\nu=5/2$ as it evolves from isotropic and nearly vanishing at $3.26$ kbar, to strongly anisotropic at $7.22$ and $9.26$ kbar, to weakly anisotropic at $10.54$ kbar. This behavior at $\nu=7/2$ suggests the same sequence of ground states as at $\nu=5/2$ and hints at the existence of a paired-to-nematic transition at $\nu=7/2$. A FQHS at at $\nu=7/2$ and at $3.26$ kbar is supported by observation of Hall resistance quantization, shown in Supplementary Figure 1. Furthermore, as demonstrated by temperature-dependent measurements shown in Supplementary Figure 2, the nematic observed at $\nu=7/2$ is a compressible ground state similar to the one observed at $\nu=9/2$ of the third Landau level in samples measured at ambient pressure [@lilly; @du].
At certain pressures, Fig.2 shows the same type of ground states at both $\nu=5/2$ and $7/2$. Indeed, at $P=3.26$ kbar we observe two FQHSs, while at $P=9.26$ and $10.54$ kbar we observe two nematic phases. This arrangement of similar ground states at different half-filled spin branches of a given orbital Landau level is typical for samples in the ambient. For example, ground states at both $\nu=5/2$ and $7/2$ in the second Landau level are FQHSs [@sevenhalf] and those at $\nu=9/2$ and $11/2$ in the third Landau level are nematic states [@lilly; @du]. At $P=7.22$ kbar, however, we observe an exception to such an arrangement. Indeed, at this pressure the ground state at $\nu=5/2$ is a FQHS, while that at $\nu=7/2$ is the nematic. This asymmetry implies that the nematic at $\nu=7/2$ is stabilized at a lower pressure than that at $\nu=5/2$.
{width="90.00000%"}
[**Temperature dependence and the stability diagram.**]{} In order to understand the evolution of phases with pressure, we turn to finite temperature measurements. We extract a characteristic energy scale of each ordered phase. We define the onset temperature for the nematic $T_{onset}$ as the temperature at which $R_{xx}=2R_{yy}$ and the energy gap $\Delta$ of a FQHS by fitting the magnetoresistance to an activated expression $ e^{-\Delta/2k_BT}$. The obtained values are summarized in Supplementary Tables 1 and 2. By plotting these two quantities against pressure, we obtain the stability diagrams in $P$-$T$ space shown in Fig.3. The stability diagram at $\nu=5/2$ has three regions [@Kate]. At low pressures, we observe a fractional quantum Hall ground state at $T=0$ and thermally excited quasiparticles at finite $T$; the energy gap of the FQHS decreases with an increasing pressure. At higher pressures we observe nematicity under a dome-like region. At even higher pressures the nematic is destroyed into a featureless Fermi fluid. In our earlier work we argued that the simplest explanation for the sequence of the phases and of the stability diagram at $\nu=5/2$ is the existence of two quantum phase transitions in the limit of $T=0$: one from a paired FQHS to the nematic occurring at $P_c$, and another from the nematic to an isotropic Fermi fluid at $\tilde{P}_c$ [@Samkharadze; @Kate]. Fig.3 reproduces this earlier result at $\nu=5/2$ in a sample of similar structure and of similar density, but cut from a different wafer [@Kate]. Furthermore, the stability diagram at $\nu=7/2$, also shown in Fig.3, is qualitatively similar to that at $\nu=5/2$ as it also exhibits the same phases and the same two quantum critical points.
Our observation of competition of the FQHS and the nematic near the quantum critical point highlights the importance of pairing in our experiments. Of the large number of FQHSs forming in the second Landau level [@firstfivehalf; @fivehalf; @sevenhalf; @Xia; @Ethan] only the paired FQHSs at $\nu=5/2$ and $7/2$ show the pressure induced transition to the nematic. Indeed, the nematic in our pressurized samples does not develop at well-known filling factors, such as the ones at $\nu=7/3$, $8/3$, $11/5$ or $14/5$, at which the ground state in the ambient are FQHSs lacking pairing. Furthermore, in the parameter space accessed in our experiment, we did not observe a paired-to-nematic quantum phase transition at any other half-filled Landau levels, such as at $\nu=9/2$ in the third Landau level or at $\nu=3/2$ in the lowest Landau level. Taken together, these results establish the independence on the spin branch of the stability diagram and of the paired-to-nematic quantum phase transition in the second orbital Landau level. In the following we focus on the critical point of the paired-to-nematic quantum phase transition. We estimate the critical pressure of the paired-to-nematic transition to be half way between the highest pressure for the FQHS and the lowest pressure for the nematic. We obtain $P_c^{5/2}=8.2 \pm 0.5$ kbar and $P_c^{7/2}=5.9 \pm 0.6$ kbar; these critical points are marked in Fig.3 by green squares. The critical pressure at $\nu=5/2$ is consistent with $7.8$ kbar, the value found in a similar sample [@Samkharadze; @Kate]. We attribute the difference of the two pressures to the 3% difference in the density of the two samples and to variations due to room temperature cycling of the sample described in Methods. Strikingly, the critical pressure $P_c^{7/2}=5.9$ kbar at $\nu=7/2$ is much reduced from its value at $\nu=5/2$. We notice that in our sample the ratio of the critical pressures $P_c^{5/2}/ P_c^{7/2}=8.2/5.9 \approx 1.4$ is equal to the ratio of the two filling factors $7/5=1.4$. This result suggests that pressure is not a primary driving parameter of the transition, but there may be other ways to induce the same quantum phase transition. This hypothesis is not unreasonable since pressure tunes all band parameters, some of which are discussed in Supplementary Note 1. The quantity changing the most dramatically with pressure is the electron density: it decreases linearly with pressure, reaching at 10 kbar nearly 20% of its value in the ambient [@p1; @Samkharadze; @Kate]. In Fig.4 we explore the premise of other driving parameters by plotting the nematic onset temperature against pressure, electron density, and magnetic field. Fig.4c is particularly significant, showing that in Sample A the paired-to-nematic critical point at the two different filling factors is at nearly the same magnetic field: $B_c^{5/2}=1.91$ T and $B_c^{7/2}=1.94$ T.
![ [**Sampling of the $\kappa$-$w/l_B$ space at half-filling.**]{} Parameters are calculated at $\nu=5/2$ (panel [**a**]{}) and $\nu=7/2$ (panel [**b**]{}). Open symbols represent fractional quantum Hall states, closed ones nematics, and crosses Fermi fluid ground states. Squares are quantum critical points for our pressurized Sample A. Stars show parameters for Sample B, measured at ambient pressure. []{data-label="fig-5"}](fig-5){width="0.9\columnwidth"}
The competition of the FQHS and of the nematic hinges on a delicate energy balance of these phases near the quantum critical point. We propose that this phase competition is driven by the electron-electron interaction, which in Sample A is tuned by pressure. The role of the electron-electron interaction in stabilizing different ground states of the two-dimensional electron gas is well known [@pair4; @wang]. In a realistic sample the electron-electron interaction is modified from its Coulomb expression by the structure of the Landau levels [@llm0; @llm1; @llm2; @llm2b; @llm3; @llm3b; @llm4; @llm5; @llm6; @Kennett] and the finite thickness of the electron layer in the direction perpendicular to the plane of the electrons $w$ [@pair4; @width1; @width2; @width3]. These effects are encoded in two adimensional quantities: the Landau level mixing parameter $\kappa=E_C/ \hbar \omega$ and the adimensional width of the electron layer $w/l_B$. Here $E_C = e^2/( 4 \pi \epsilon l_B)$ is the Coulomb energy, $\hbar \omega$ is the cyclotron energy, and $l_B = \sqrt{\hbar /e B}$ the magnetic length. The Landau level mixing parameter scales as $\kappa \propto m/\epsilon \sqrt{B}$, where $m$ is the effective mass of electrons. Thus, in a given orbital Landau level and at fixed $m$, $\epsilon$, and $w$, both $\kappa$ and $w/l_B$ are functions of the magnetic field only. Under such constraints, therefore, the electron-electron interaction depends only on the magnetic field. We conclude that the observation of a paired-to-nematic quantum critical point at both $\nu=5/2$ and $\nu=7/2$ at the same critical magnetic field may indeed be due to the tuning of the electron-electron interaction. We think that in Sample A this interaction is tuned by the pressure through changing the electron density. As we tune the pressure, in the $\kappa$-$w/l_B$ space we sample the curves shown in Fig.5. At the critical pressure of the paired-to-nematic transition we find $\kappa_c^{5/2}=1.95$, $w/l_{B,c}^{5/2}=1.62$ and $\kappa_c^{7/2}=1.90$, $w/l_{B,c}^{7/2}=1.63$, nearly independent of the filling factor. Here we took into account the pressure dependence of the effective mass and dielectric constant [@p1]. It is tempting to think of Fig.5 as a phase diagram. However, phase boundaries in this figure are expected to be significantly affected by disorder and by the lowest temperature reached. Nonetheless, Fig.5 may serve as a guide to place constraints on the ordered phases. An expanded version of this figure, that includes published data obtained in samples in the ambient, is shown in Supplementary Figure 3.
![ [**Magnetoresistance in the second Landau level of Sample B in the ambient.**]{} Traces are measured at ambient pressure and at $T \approx 4.5$ mK. The strong resistance anisotropy at $\nu=7/2$ shows a nematic ground state, while at $\nu = 5/2$ we observe a FQHS. The nature of the ground state at these two filling factors is marked by text. []{data-label="fig-6"}](fig-6){width="0.95\columnwidth"}
[**Measurements of a sample in the ambient.**]{} To test the relevance of electron-electron interactions, we investigate Sample B to be measured at ambient pressure, but in which the electron-electron interaction was tuned near its value at the quantum critical point. Sample B has the same width of the quantum well as Sample A, but it has a reduced density. By design, the density was picked in such a way that the parameters $\kappa$ and $w/l_B$ calculated at $\nu=7/2$ fall in the range of the nematic (shown as a red star in Fig.5). We note that data points for Sample B in Fig.5 are slightly off the curve for Sample A since pressure corrections of the mass and dielectric strength are no longer needed. Magnetoresistance traces for this sample, as measured with the sample mounted in a $^3$He immersion cell [@Imm-cell], are shown in Fig.6. At $\nu=7/2$ we indeed observe an extremely large resistance anisotropy. Furthermore, at $\nu=5/2$ we observe a weak FQHS, consistent with the $\kappa$ and $w/l_B$ parameters being just outside the range for the nematic. Hall resistance at $\nu=5/2$, shown in Supplementary Figure 4, is consistent with a FQHS. Taken together, there is compelling evidence that the nematic phase is stabilized in the second orbital Landau level at ambient pressure, when the electron-electron interaction is tuned via the parameters $\kappa$ and $w/l_B$, to the stability range of the nematic. We emphasize that, according to our findings, the numerical values of the critical parameters of the paired-to-nematic transition are valid only for $\nu=5/2$ and $7/2$ in the second orbital Landau level and are dependent on parameters such as the width of the quantum well.
[**Discussion.**]{}
It is interesting to note that in Sample A the nematic develops at $\nu=5/2$ for pressures for which the electron density is in the range of $10.6-6.3 \times 10^{10}$ cm$^{-2}$. Such densities have already been accessed, but the nematic at $\nu=5/2$ was not observed [@ld1; @ld2; @llm3; @Pan72]. Since samples from Refs.[@ld2; @Pan72] had a wider quantum well than our samples, the nematic in them either does not develop or it forms at a yet unknown critical $\kappa$ and $w/l_B$ parameters. The other two samples, however, had quantum wells of the same width as our samples [@ld1; @llm3]. In one of these samples densities necessary for the nematic, lower than $10.6 \times 10^{10}$ cm$^{-2}$, have not been studied [@ld1]. In the other $30$ nm quantum well sample the FQHS at $\nu=5/2$ is seen down to a density $12.5 \times 10^{10}$ cm$^{-2}$, but the nematic at $\nu=5/2$ was not seen at $9.5 \times 10^{10}$ cm$^{-2}$ [@llm3]. Possible reasons for the absence of the nematic in Ref.[@llm3] are disorder effects or effects due to the asymmetric shape of the wavefunction in the direction perpendicular to the plane of the electrons in gated samples. Resistance anisotropy at $\nu=7/2$ was, however, observed in 60 nm quantum well sample having a density of $5 \times 10^{10}$cm$^{-2}$, providing an important clue on the influence of the width of the quantum well [@Pan72]. No data is available at $\nu=7/2$ in Refs.[@ld1; @llm3].
Interest in paired FQHSs has been recently rekindled by the discovery of FQHSs at even denominators in electron gases confined to ZnO [@zno] and bilayer graphene hosts [@gr1; @gr2]. However, in contrast to the GaAs system, in these hosts there is no evidence of the nematic. The reason for the absence of the nematic in ZnO and bilayer graphene is not currently known; disorder effects, a different crystal symmetry and/or a different electron-electron interaction may be at play. However, we cannot rule out future observations of the nematic in these hosts. We will next compare the electron-electron interaction in these systems as parametrized by $\kappa$ and $w/l_B$. The dielectric environment of the bilayer graphene encapsulated in boron nitride is not well characterized; we will use $\epsilon \simeq 3.5$ and $m=0.05m_0$. For this host the parameters for the densities accessed fall in the $\kappa \simeq 1.5 - 2.8$ and $w/l_B \simeq 0.03 - 0.06$ range [@gr2]. For the strongest $\nu=7/2$ FQHS developing in ZnO [@zno], we find $\kappa \simeq 15$ and $w/l_B \simeq 0.3$. It is interesting to note that, in comparison to the GaAs system [@nod], the even denominator FQHSs in bilayer graphene develop at similar values of $\kappa$, but at much reduced value of $w/l_B$. This is in sharp contrast with ZnO, in which the even denominator states develop at an extremely large values of $\kappa$. We think that this opens the possibility that the nature of the even denominator FQHSs in ZnO may be fundamentally different from those developing in GaAs or bilayer graphene.
Enhanced quantum fluctuations may have observable consequences close to the critical point. A recent theory has examined the influence of the nematic fluctuations on the paired FQHS [@kim17]. Nematic fluctuations may also influence the nematic phase itself in a description beyond the mean field [@fogler; @moessner]. Our data show several anomalies close to the quantum critical point which may be related to fluctuation effects. One anomaly, shown in Fig.2c, is the resistance anisotropy at $\nu=7/2$ exceeds that at $5/2$. At fixed density and fixed temperature, a larger anisotropy typically develops in the lower spin branch. For example, in the third orbital Landau level the anisotropy observed at $\nu=9/2$ is larger than that at $\nu=11/2$ [@lilly; @du]. Other anomalies develop in Sample B, shown in Fig.6. The resistance near $\nu=5/2$ is not isotropic in the vicinity of $\nu=5/2$ and data at $\nu \approx 2.42$ suggests a nematic which is not centered at half-filling. Furthermore, resistance anisotropy in the upper spin branch is not exactly centered to $\nu=7/2$. Since the mean field approach predicts a nematic centered at half-filling [@fogler; @moessner], we think that this approach is insufficient to describe the anomalies we see and that fluctuations are most likely at play. Fluctuation effects stemming from the proximity to the paired-to-nematic quantum critical point warrant further investigations.
In Fig.3 there is a second quantum phase transition at high pressures, from the nematic to an isotropic Fermi fluid. The critical pressures of this transition, $\tilde{P}_c^{5/2}=11.0 $ kbar and $\tilde{P}_c^{7/2}=11.4$ kbar, are estimated by linear extrapolation to $T=0$ of the nematic onset temperatures forming at the two highest pressures. These critical points are marked in Fig.3 by orange squares. When comparing the critical values of different parameters at $\nu=5/2$ and $7/2$ which may drive the nematic-to-Fermi fluid transition we find that, in contrast to the paired-to-nematic transition, this transition occurs at nearly the same pressure, at values of the electron density close to each other $\tilde{n}_c^{5/2} = 5.2 \times 10^{10}$ cm$^{-2}$ and $\tilde{n}_c^{7/2} = 4.5 \times 10^{10}$ cm$^{-2}$, but at very different magnetic fields. The nematic onset temperature as function of these parameters is seen in Fig.4. As discussed in Supplementary Note 2, at such low electron densities we expect that disorder effects do not permit nematic order. We thus think that the destruction of the nematic both at $\nu=5/2$ and $\nu=7/2$ at similar electron densities is an indication that disorder became dominant. This idea is further supported by data in Supplementary Figure 5, which depicts the suppression of the nematic at high pressures in both the second and third Landau levels.
In summary, the observation of the pressure-driven quantum phase transition from a paired FQHS to the nematic at both $\nu=5/2$ and $\nu=7/2$ Landau level filling factors assures the independence of the spin branch of this transition in the second orbital Landau level. Furthermore, by observing phases consistent with those at high pressure in a sample in the ambient, we have shown that pressure is not the only driving parameter of this transition. Our observations suggest that tuning the electron-electron interactions, as parametrized by Landau level mixing and adimensional width of the quantum well, play a critical role in driving the paired-to nematic phase transition. These results invite further investigations of the effects of fluctuations developing near the quantum critical point on both the paired FQHS and on the nematic.
[**Methods.**]{}
[**Cryogenic measurements.**]{} Measurements were performed in a dilution refrigerator, using a standard low frequency lockin technique. Magnetic fields up to 10 T were applied perpendicularly to the plane of the electron gas. Before cooling to low temperatures, samples were illuminated at 10 K using a red light emitting diode.
[**Details of the pressure cell and sample illumination.**]{} Sample A was cut to a $2 \times 2$ mm$^2$ size and was mounted in a pressure cell [@pcell]. The pressure transmitting fluid was an equal mixture of pentane and isopentane. In order to change pressure, the sample was warmed up to room temperature. After each room temperature cycling, the same illumination technique was used. We estimate the lowest electronic temperature reached in this pressure cell is about 12 mK.
[**Details of measurements under ambient pressure.**]{} Sample B was cut to a $4 \times 4$ mm$^2$ size and was measured in a $^3$He immersion cell [@Imm-cell]. Using this cell we can thermalize electrons to temperatures below 5 mK. Details of the immersion cell setup are found in Supplementary Note 3.
[**Acknowledgements.**]{} Research at Purdue was supported by the US Department of Energy, Office of Basic Energy Sciences, Division of Materials Sciences and Engineering under the awards DE-SC0006671 (G.A.C. and M.J.M.) and DE-SC0010544 (Y.L.G.). K.A.S. acknowledges the Purdue Cagiantas Graduate Research Fellowship in Science. L.N.P. and K.W.W. of Princeton University acknowledge the Gordon and Betty Moore Foundation Grant No. GBMF 4420, and the National Science Foundation MRSEC Grant No. DMR-1420541.
[l]{}
Fradkin, E., Kivelson, S.A., Lawler, M.J., Eisenstein, J.P. & Mackenzie, A.P. Nematic Fermi fluids in condensed matter physics. *Annu. Rev. Conden. Matter Phys.* **1**, 153-178 (2010). Lilly, M.P., Cooper, K.B., Eisenstein, J.P., Pfeiffer, L.N. & West, K.W. Evidence for an anisotropic state of two-dimensional electrons in high Landau levels. *Phys. Rev. Lett*. **82**, 394-397 (1999). Du, R.R., Tsui, D.C., Stormer, H.L., Pfeiffer, L.N., Baldwin, K.W. & West, K.W. Strongly anisotropic transport in higher two-dimensional Landau levels. *Solid State Commun.* **109**, 389-394 (1999). Koulakov, A.A., Fogler, M.M. & Shklovskii, B.I. Charge density wave in two-dimensional electron liquid in weak magnetic field. *Phys. Rev. Lett*. **76**, 499-502 (1996). Moesnner, R. & Chalker, J.T. Exact results for interacting electrons in high Landau levels. *Phys. Rev. B* **54**, 5006-5015 (1996). Kivelson, S.A., Fradkin, E., & Emery, V.J. Electronic liquid-crystal phases of a doped Mott insulator *Nature* **393**, 550-553 (1998). Fradkin, E. & Kivelson S.A. Liquid-crystal phases of quantum Hall systems. *Phys. Rev. B* **59**, 8065-8072 (1999).
Keimer, B., Kivelson, S.A., Norman, M.R., Uchida, S. & Zaanen, J. From quantum matter to high-temperature superconductivity in copper oxides. *Nature* **518**, 179-186. (2015). Fradkin, E., Kivelson, S.A. & Tranquada, J.M. *Colloquium:* Theory of intertwined orders in high temperature superconductors. *Rev. Mod. Phys.* **87**, 457-482. (2015).
Metlitski, M.A, Mross, D.F., Sachdev, S. & Senthil, T. Cooper pairing in non-Fermi liquids. *Phys. Rev. B* **91**, 115111 (2015). Lederer, S., Schattner, Y., Berg, E., & Kivelson S.A. Enhancement of superconductivity near a nematic quantum critical point. *Phys. Rev. Lett.* **114**, 097001 (2015). Schattner, Y., Lederer, S., Kivelson, S.A., & Berg, E. Ising nematic quantum critical point in a metal: a Monte Carlo study. *Phys. Rev. X* **6**, 031028 (2016). Dumitrescu, P.T., Serbyn, M., Scalettar R.T., & Vishwanath, A. Superconductivity and nematic fluctuations in a model of doped FeSe monolayers: determinant quantum Monte Carlo study. *Phys. Rev. B* **94**, 155127 (2016). Mesaros A., Lawler M. J., & Kim, E-A. Nematic fluctuations balancing the zoo of phases in half-filled quantum Hall systems. *Phys. Rev. B* **95**, 125127 (2017).
Qian, Q., Nakamura, J., Fallahi, S., Gardner, G.C., & Manfra, M.J.Possible nematic to smectic phase transition in a two-dimensional electron gas at half-filling. *Nature Commun.* **8**, 1536 (2017).
Willett, R., Eisenstein, J.P., Stormer, H.L., Tsui, D.C., Gossard, A.C., & English, J.H. Observation of an even-denominator quantum number in the fractional quantum Hall effect. *Phys. Rev. Lett.* **59**, 1776-1779 (1987). Pan, W, Xia, J.-S., Shvarts, V., Adams, D.E., Stormer, H.L., Tsui, D.C., Pfeiffer, L.N., Baldwin, K.W., & West, K.W. Exact quantization of the even-denominator fractional quantum Hall state at $\nu = 5/2$ Landau level filling factor. *Phys. Rev. Lett.* **83**, 3530-3533 (1999). Eisenstein, J.P., Cooper, K.B., Pfeiffer, L.N., & West, K.W. Insulating and fractional quantum Hall tates in the first excited Landau level. *Phys. Rev. Lett.* **88**, 076801 (2002). Moore, G., & Read, N. Nonabelions in the fractional quantum Hall effect. *Nucl. Phys. B* **360**, 362- 396 (1991).
Jain, J.K. Composite-fermion approach for the fractional quantum Hall effect. *Phys. Rev. Lett.* **63**, 199-202 (1989). Halperin, B.I., Lee, P.A., & Read, N. Theory of the half-filled Landau level. *Phys. Rev. B* **47**, 7312-7343 (1993).
Greiter, M., Wen, X.-G., & Wilczek, F. Paired Hall state at half filling. *Phys. Rev. Lett.* **66**, 3205-3208 (1991). Park K., Melik-Alaverdian, V., Bonesteel, N.E., & Jain, J.K. Possibility of $p$-wave pairing of composite fermions at $\nu =1/2$. *Phys. Rev. B* **58**, 10167-10170 (1998). Scarola, V.W., Park, K., & Jain, J.K. Cooper instability of composite fermions. *Nature* (London) **406**, 863-865 (2000). Rezayi, E.H. & Haldane, F.D.M. Incompressible paired Hall state, stripe order, and the composite Fermion liquid phase in half-filled Landau levels. *Phys. Rev. Lett.* **84**, 4685-4688 (2000). Pan, W., Du, R.R., Stormer, H.L., Tsui, D.C., Pfeiffer, L.N., Baldwin, K.W., & West, K.W. Strongly anisotropic electronic transport at Landau level filling factor $\nu =9/2$ and $\nu=5/2$ under a tilted magnetic field. *Phys. Rev. Lett.* **83**, 820-823 (1999). Lilly, M.P., Cooper, K.B., Eisenstein, J.P., Pfeiffer, L.N., & West, K.W. Anisotropic states of two-dimensional electron systems in high Landau levels: effect of an in-plane magnetic field. *Phys. Rev. Lett.* **83**, 824-827 (1999). Pan, W., Serafin, A., Xia, J.S., Yin, L., Sullivan, N.S., Baldwin, K.W., West, K.W., Pfeiffer, L.N., & Tsui, D.C. Competing quantum Hall phases in the second Landau level in the low-density limit. *Phys. Rev. B.* **89**, 241302 (2014). Samkharadze, N., Schreiber, K.A., Gardner, G.C., Manfra, M.J., Fradkin, E., & Csáthy, G.A. Observation of a transition from a topologically ordered to a spontaneously broken symmetry phase. *Nature Phys.* **12**, 191-195 (2016).
You, Y., Cho, G.Y., & Fradkin, E. Nematic quantum phase transition of composite Fermi liquids in half-filled Landau levels and their geometric response. *Phys. Rev. B* **93**, 205401 (2016). Zhu, Z, Sodemann, I., Sheng, D.N., & Fu, L. Anisotropy-driven transition from the Moore-Read state to quantum Hall stripes. *Phys. Rev. B* **95**, 201116 (2017).
Xia, J.S., Pan, W., Vicente, C.L., Adams, E.D., Sullivan, N.S., Stormer, H.L., Tsui, D.C., Pfeiffer, L.N., Baldwin, K.W., & West, K.W. Electron correlation in the second Landau level: a competition between many nearly degenerate quantum phases. *Phys. Rev. Lett.* **93**, 176809 (2004). Lee, K., Shao, J., Kim, E.-A., Haldane, F.D.M., & Rezayi, E.H. Pomeranchuk instability of composite Fermi liquid. Preprint at https://arxiv.org/abs/1802.08261 (2018). Kleinbaum, E., Kumar, A., Pfeiffer, L.N., West, K.W., & Csáthy, G.A. Gap reversal at filling factors $3+1/3$ and $3+1/5$: towards novel topological order in the fractional quantum Hall regime *Phys. Rev. Lett.* **114**, 076801 (2015).
Schreiber, K.A., Samkharadze, N., Gardner, G.C., Biswas, R.R., Manfra, M.J., & Csáthy, G.A. Onset of quantum criticality in the topological-to-nematic transition in a two-dimensional electron gas at filling factor $\nu =5/2$. *Phys. Rev. B*, **96**, 041107 (2017).
Dmowski, L., & Portal, J.C. Magnetotransport in 2D semiconductor systems under pressure. *Semicond. Sci. Technol.* **4**, 211-217 (1989). Wang, H., Sheng, D.N., & Haldane, F.D.M. Particle-hole symmetry breaking and the $\nu=5/2$ fractional quantum Hall effect *Phys. Rev. B* **80**, 241311 (2009).
Yoshioka, D. Excitation energies of the fractional quantum Hall effect. *J. Phys. Soc. Jpn.*, **55**, 885-896 (1986). Wójs, A. & Quinn, J.J. Landau level mixing in the $\nu=5/2$ fractional quantum Hall state. *Phys. Rev. B* **74**, 235319 (2006). Wójs, A., Tőke, C. & Jain, J.K. Landau-level mixing and the emergence of Pfaffian excitations for the 5/2 fractional quantum Hall effect. *Phys. Rev. Lett.* **105**, 096802 (2010). Rezayi, E.H. & Simon, S.H. Breaking of particle-hole symmetry by Landau level mixing in the $\nu =5/2$ quantized Hall state. *Phys. Rev. Lett.* **106**, 116801 (2011). Nuebler, J., Umansky, V., Morf, R., Heiblum, M., von Klitzing, K. & Smet, J. Density dependence of the $\nu=5/2$ energy gap: experiment and theory. *Phys. Rev. B* **81**, 035316 (2010). Zaletel, M.P., Mong, R.S.K., Pollmann, F. & Rezayi, E.H. Infinite density matrix renormalization group for multicomponent quantum Hall systems. *Phys. Rev. B* **91**, 045115 (2015). Pakrouski, K., Peterson, M.J. Jolicoeur, Th., Scarola, V.W., Nayak, C. & Troyer, M. Phase diagram of the $\nu=5/2$ fractional quantum Hall effect: effects of Landau-level mixing and nonzero width. *Phys. Rev. X* **5**, 021004 (2015). Tylan-Tyler, A. & Lyanda-Geller, Y. Phase diagram and edge states of the $\nu=5/2$ fractional quantum Hall state with Landau level mixing and finite well thickness. *Phys. Rev. B* **91**, 205404 (2015). Rezayi, E.H. Landau level mixing and the ground state of the $\nu=5/2$ quantum Hall effect. *Phys. Rev. Lett.* **119**, 026801 (2017). Smith, P.M. & Kennett, M.P. Tuning the effects of Landau level mixing on anisotropic transport in quantum Hall systems. *J. Phys. Condens. Matter* **24**, 055601 (2012). Peterson, M.R., Jolicoeur, Th. & Das Sarma, S. Finite-layer thickness stabilizes the Pfaffian state for the 5/2 fractional quantum Hall effect: wave function overlap and topological degeneracy. *Phys. Rev. Lett.* **101**, 016807 (2008). Papić, Z., Regnault, N. & Das Sarma, S. Interaction-tuned compressible-to-incompressible phase transitions in quantum Hall systems. *Phys. Rev. B* **80**, 201303 (2009). Jeong, J.-S. & Park, K. Bilayer mapping of the paired quantum Hall state: instability toward anisotropic pairing. *Phys. Rev. B* **91**, 195119 (2015). Samkharadze, N., Kumar, A., Manfra, M.J., Pfeiffer, L.N., West, K.W., & Csáthy, G.A. Integrated electronic transport and thermometry at milliKelvin temperatures and in strong magnetic fields. *Rev. Sci. Instrum.* **82**, 053902 (2011).
Watson, J.D., Csáthy, G.A., & Manfra, M.J. Impact of heterostructure design on transport properties in the second Landau level of *in situ* back-gated two-dimensional electron gases. *Phys. Rev. Appl.* **3**, 064004 (2015). Samkharadze, N., Ro, D., Pfeiffer, L.N., West, K.W., & Csáthy, G.A. Observation of an anomalous density-dependent energy gap of the $\nu=5/2$ fractional quantum Hall state in the low-density regime. *Phys. Rev. B* **96**, 085105 (2017).
Falson, J., Maryenko, D., Friess, B., Zhang, D., Kozuka, Y., Tsukazaki, A., Smet, J.H. & Kawasaki, M., Even-denominator fractional quantum Hall physics in ZnO, *Nature Physics* **11**, 347–351 (2015). Zibrov, A.A., Kometter, C., Zhou, H., Spanton, E.M., Taniguchi, T., Watanabe, K., Zaletel, M.P., & Young, A.F. Tunable interacting composite fermion phases in a half-filled bilayer graphene Landau level. *Nature* **549** 360-364, (2017). Li, J.I.A., Tan, C., Chen, S., Zeng, Y., Taniguchi, T., Watanabe, K., Hone, J.& Dean, C.R. Even denominator fractional quantum Hall states in bilayer graphene. *Science* **358**, 648-652 (2017).
Samkharadze, N., Watson, J.D., Gardner, G. Manfra, M.J., Pfeiffer, L.N., West, K.W., & Csáthy, G.A. Quantitative analysis of the disorder broadening and the intrinsic gap for the $\nu=5/2$ fractional quantum Hall state. *Phys. Rev. B* **84**, 121305 (2011).
Almax easyLab Technologies Ltd., model Pcell 30.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
We discuss the fundamental constants of Physics in the Standard Model and possible changes of these constants on the cosmological time scale. The Grand Unification of the strong, electromagnetic and weak interactions implies relations between the time variation of the finestructure constant $\alpha$ and of the QCD scale $\Lambda_c$. A change of $\alpha $ by $10^{-15}$ / year, as seen by an astrophysics experiment, implies thus a time variation of $\Lambda_c$ of at least $10^{-15}$ / year. An experiment in Quantum Optics at the MPQ in Munich, which was designed to look for a time variation of $\Lambda_c$, is discussed.\
\
1. Introduction
2. Fundamental Constants and the Standard Model
3. The Time Variation of the Finestructure Constant
4. Grand Unification and Time Variation
5. Results from Quantum Optics
6. Conclusions and Outlook
author:
- |
H. Fritzsch\
University of Munich, Physics Department\
Munich, Germany
title: The Fundamental Constants in Physics
---
The Standard Model
==================
The Standard Model consists of
1. the gauge theory of the strong interactions: Quantum Chromodynamics (QCD)[@Fritzsch],
2. the gauge theory of the electroweak interactions, based on the gauge group $SU(2) \times U(1)$[@Glash].
QCD is an unbroken gauge theory, based on the gauge group $SU(3)$, acting in the internal space of ”color“. The basic fermions of the theory are the six quarks, which form color triplets. The gluons, the eight massless gauge bosons, are $SU(3)$– octets. The interactions of the quarks and gluons are dictated by the gauge properties of the theory. The quarks and gluons interact through the vertex $g_s \cdot \bar q \gamma_{\mu} \frac{\lambda_i}{2} q \cdot A^{\mu}_i$, where $q$ are the quark fields and $A^{\mu}_i$ the eight gluon fields. The eight $SU(3)$–matrices are denoted by $\lambda_i$. The strength of the coupling constant is given by $g_s$.
QCD is a non–Abelian gauge theory. There is a direct coupling of the gluons among each other. There is also a trilinear coupling, proportional to $g_s$, and a quadrilinear coupling, proportional to $g^2_s$. It is assumed, that the QCD interaction leads to a confinement of all colored quanta, in particular of the quarks and the gluons. But this is thus far not proven. Replacing the continuous space–time continuum by a lattice, one can solve the QCD field equations with the computer. The results confirm the confinement hypothesis.
The experimental data are in very good agreement with QCD [@Rev]. Quantum Chromodynamics has the property of asymptotic freedom. The strength of the quark-gluon-interaction converges to zero on a logarithmic scale at high energies. At low energies the interaction strength is large. Thus the confinement property of QCD might indeed be true.
The equations, describing the renormalization of the coupling constant, give for $\alpha_s = \frac{g_s \, ^2}{4 \pi}$: $$\begin{aligned}
\mu \cdot \frac{\partial \alpha_s}{\partial_{\mu}} & = &
- \frac{\beta_0}{2 \pi} \alpha_s^2
- \frac{\beta_1}{4 \pi^2} \alpha_s^3 - \ldots \nonumber \\
\beta_0 & = & 11 - \frac{2}{3} n_f \nonumber \\
\beta_1 & = & 51 - \frac{19}{3} n_f\end{aligned}$$ ($n_f$: number of relevant quark flavors)
Since the interaction is weak at high energies, the quarks and gluons appear nearly as pointlike objects at small distances. This has been observed in the experiments of deep inelastic scattering of electrons, myons and neutrinos off nuclear targets.
The strong coupling constant at high energies is small, but not zero. Therefore one expects violations of the scaling behaviour of the cross-sections. This has been seen in many experiments. The value of the QCD coupling constant $\alpha_s = \frac{g_s \, ^2}{4 \pi}$ depends on the energy. One has found in the analysis of scaling violations[@Rev]: $$\alpha_s \left( M_z^2 \right) \approx 0.1187 \pm 0.002$$ ($M_z$: mass of the $Z$–boson, $M_z \cong 91.2 \, \,\, {\rm GeV}$).
We can express $\alpha_s (\mu)$ as a function of the scale parameter of QCD $\Lambda_c$: $$\begin{aligned}
\alpha_s (\mu )^{-1} & \approx & \left( \frac{\beta_0}{4 \pi} \right)
ln
\left( \frac{\mu^2}{\Lambda_c^2} \right) \nonumber \\
\beta_0 & = & \left( 11 - \frac{2}{3} n_f \right)\end{aligned}$$ The experiments give the following value: $$\Lambda_c \approx 217^{+25}_{-23} \, \, \, {\rm MeV} \, .$$
The electroweak gauge theory is based on the gauge group $SU(2) \times U(1)$. Thus there are three $W$–bosons, related to the $SU(2)$ group, and a $B$–boson, related to the $U(1)$–group. The lefthanded quarks and leptons are $SU(2)$–doublets, the righthanded leptons and quarks are singlets. Parity is violated in a maximal way.
The gauge invariance of the $SU(2) \times U(1)$–model is broken by the ”Higgs“–mechanism[@Higgs]. The masses of the gauge bosons are generated by a spontaneous symmetry breaking. Goldstone bosons appear as longitudinal components of the gauge bosons. In the standard ”Higgs“ mechanism there exists a self–interacting complex doublet of scalar fields. In the process of symmetry breaking the neutral component of the scalar doublet acquires a vacuum expectation value $v $, which is determined by the Fermi constant of the weak interactions. Therefore the vacuum expectation value is known from the experiments, if the theory is correct: $$v \cong 246 \, \, \, {\rm GeV}$$
This energy sets the energy scale for the electroweak symmetry breaking. Three massless Goldstone bosons are generated, but they are absorbed to give masses to the $W^+, W^-$ and $Z$–bosons. One component of the complex doublet is not absorbed. This is the ”Higgs“–boson, thus far a hypothetical particle. It would be the only elementary scalar boson in the Standard Model. One hopes to find this particle with the new accelerator LHC at CERN (start in 2009).
In the electroweak model one has two neutral gauge bosons, which are mixtures of $W_3$ and $B$, the $Z$–boson and the photon. The associated electroweak mixing angle $\Theta_w$ is a fundamental parameter which has to be fixed by experiment. It is given by the $Z$–mass, the Fermi constant and the fine structure constant $\alpha$:
$$sin^2 \Theta_w \cdot cos^2 \Theta_w =
\frac{\pi \alpha \left( M_z \right)}{\sqrt{2} \cdot G_F \cdot M_Z^2} \,
.$$
In the experiments one finds $ sin^2 \Theta_w \approx 0.231.$
Note that the electroweak mixing angle is also related to the mass ratio $M_W / M_Z$. If one neglects radiative corrections, one finds:
$$\begin{aligned}
sin^2 \Theta_w & = & 1 - M_W^2 / M_Z^2 \nonumber \\
M_Z & = & M_W / cos \Theta_w \, .\end{aligned}$$
In the Standard Model the interactions depend on 28 fundamental constants. These are:\
the constant of gravitiy $G$,\
the finestructure constant $\alpha $,\
the coupling constant $g_w$ of the weak interactions,\
the coupling constant $g_s$ of the strong interactions,\
the mass of the W-boson,\
the mass of the ”Higgs“–boson,\
the masses of the three charged leptons, $m_e, m_{\mu}, m_{\tau}$,\
the neutrino masses $m(\nu_1), m(\nu_2), m(\nu_3)$,\
the masses of the six quarks $m_u, m_d, m_c, m_s, m_t, m_b$,\
the four parameters, describing the flavor mixing of the quarks,\
and the six parameters, describing the flavor mixing of the leptons, measured by the neutrino oscillations.
In physics we are dealing with the laws of nature, but little thought is given to the boundary condition of the universe, related directly to the Big Bang. We do not know at the moment, what role is played by the fundamental costants, but these constants could form a bridge between the boundary conditions and the local laws of nature. Thus they would be accidental relics of the Big Bang.
Some physicists believe that at least some of the fundamental constants are just cosmic accidents, fixed by the dynamics of the Big Bang. Thus the constants are arbitrary, depending on details of the Big Bang. Obviously in this case there is no way to calculate the fundamental constants.
Some fundamental constants might be cosmic accidents, but it is unlikely, that this is the case for all fundamental constants. New interactions, discovered e. g. with the new LHC–accelerator at CERN, might offer a way to calculate at least some of the fundamental constants.
We also do not understand, why the fundamental constants are constant in time. Small time variations are indeed possible and even suggested by astrophysical experiments. In the theory of superstrings one expects time variations of the fundamental constants, in particular of the finestructure constant, of the QCD scale parameter $\Lambda_c$, and of the weak interaction coupling constant[@Tay; @Witt].
If one finds that the fundamental constants are changing in time, then they are not just numbers, but dynamical quantities which change according to some deeper laws that we have to understand. These laws would be truly fundamental and may even point the way to a unified theory including gravity.
Fundamental Constants in the Standard Model
===========================================
The Standard Model of particle physics is the theory of the observed particle physics phenomena. However it depends on 28 fundamental constants. Within the Standard Model there is no way to calculate these constants.
The most famous fundamental constant is the finestructure constant $\alpha$, introduced in 1916 by Arnold Sommerfeld: $$\alpha = \frac{e^2}{\hbar c} \, .$$ In this constant the electromagnetic coupling $e$ enters, as well as the constant of the quantum physics h, and the speed of light $c$. Sommerfeld realized that $\alpha$ is a dimensionless number, close to the inverse of the prime number 137. The experiments give the following value for $\alpha^{-1}$: 137,03599911(46)[@Rev].
Werner Heisenberg proposed in 1936 the relation: $$\alpha = 2^{-4} \, 3^{-3} \pi \, ,$$ which gives $\alpha^{-1} = 137,51$. In 1971 Wyler[@Wyler] published the following expression for $\alpha$: $$\alpha = \frac{9}{8 \pi ^4} \left( \frac{\pi ^ 5}{2^4 \cdot 5!}
\right)^{1/4}
\, ,$$ which gives $\alpha^{-1} = 137,03608$.
Richard P. Feynman wrote about the finestructure constant[@Feyn]: ”It has been a mystery ever since it was discussed more than fifty years ago, and all good theoretical physicists put this number up on their wall and worry about it. Immediately you would like to know where this number for a coupling comes from: is it related to $\pi $ or perhaps to the base of the natural logarithms? Nobody knows. It’s one of the greatest mysteries of physics: a magic number that comes to us with no understanding by man …”.
In quantum field theory the strength of an interaction is not a fixed constant, but a function of the energy involved. The groundstate of a system is filled with virtual pairs of quanta, e.g. with $e^+e^-$–pairs in QED. Thus an electron is surrountded by $e^+e^-$–pairs. The virtual electrons are repelled by the electrons, the virtual positrons are attracted. The electron charge is partially shielded by the virtual positrons. At relatively large distances the electron charge is smaller than at distances less than $\lambda_c$. The dependence on the energy is described by the renormalization group equations of Murray Gell–Mann and Francis Low[@Gell]: $$\frac{d}{d \, ln \left(q/M \right)} e(q) = \beta(e) \, ,$$ where $$\beta(e) = \frac{e^3}{12 \pi ^ 2} + \, \, \, {\rm higher \, \, \, order
\, \, \, terms} \, \, .$$
In QED one has to include not only virtual $e^+e^-$–pairs, but also the $\mu^+ \mu^-$– and $\tau^+ \tau^-$–pairs, as well as the quark–antiquark–pairs. One finds that the finestructure constant $\alpha$ at the mass of the $Z$–boson should be the inverse of 128, in good agreement with the experimental data taken with the LEP–accelerator[@Rev].
Another fundamental parameter of the Standard Model is the mass of the proton. In QCD the proton mass is a parameter, which can be calculated as a function of the QCD scale parameter $\Lambda_c$ and of the light quark masses. The QCD scale parameter has been determined many experiments:
$$\Lambda_c = 217 \pm 25 \, \,\, {\rm MeV} \, .$$
($\Lambda_c$ is defined in the modified minimal subtraction $\left( \bar{MS} \right)$ scheme for five quark flavors).
The QCD theory gives a very clear picture of the mass generation. In the limit, where the quark masses are neglected, the nucleon mass is the confined field energy of the gluons and quarks. It can be written as:
$$M(Nucleon) = const. \cdot \Lambda_c \, .$$
The [*const.*]{} has been calculated using the lattice approach to QCD. It is about 3,9, predicting a nucleon mass in the limit $m_q = 0$ of about 860 MeV. The observed nucleon mass (about 940 MeV) is higher, due to the contributions of the mass terms of the light quarks $u, d, s$, which in reality are not massless.
The mass of the proton can be decomposed as follows: $$\begin{aligned}
M_p & = & \, \, \, {\rm const.} \, \, \, \Lambda_c + \\
& & < p \mid m_u \bar u u \mid p > + < p \mid m_d \bar dd \mid p >
+ < p \mid m_s \bar s s \mid p > + c_{\rm elm} \, \, \, \cdot \Lambda_c
\, . \nonumber\end{aligned}$$ The last term describes the electromagnetic self–energy. It is proportional to the QCD–scale $\Lambda $. Calculations give[@Leut]: $$c_{\rm elm} \cdot \Lambda_c \approx 2.0 \, \, \, {\rm MeV} \, .$$ The up–quark mass term contributes about 20 MeV to the proton mass, the $d$–quark mass term about 19 MeV. Thus the $d$–contribution to the proton mass is about as large as the $u$–contribution, although there are two $u$–quarks in the proton, and only one $d$–quark. This is due to the fact that the $d$–mass is larger than the $u$–mass.
In chiral perturbation theory the $u$– and $d$–masses can be estimated[@Leut1]: $$\begin{aligned}
m_u & \approx & 3 \pm 1 \, \, \, {\rm MeV} \nonumber \\
m_d & \approx & 6 \pm 1.5 \, \, \, {\rm MeV} \, .\end{aligned}$$ These masses are normalized at the scale $\mu = 2 \, \, \,
{\rm GeV}$. Note that quark masses are not the masses of free particles, but of dynamical quantities. They depend on the energy scale $\mu $, relevant for the discussion.
The mass of the strange quark can also be estimated in the chiral perturbation theory[@Leut1]. One finds at $\mu = 2 \, \, \, {\rm GeV}$: $$m_s \approx 103 \pm 20 \, \, \, {\rm MeV} \, .$$ The mass of the strange quark is about 20 times larger than the $d$– mass. Although there are no valence $s$–quarks in the proton, the $\bar ss$–pairs contribute about 35 MeV to the proton mass, i. e. more than the $\bar uu$– or $\bar dd$–pairs, due to the large ratio $m_s /
m_d$. Heavy quarks, e. g. $c$–quarks, contribute at most $ \sim $ 1 MeV to the nucleon mass[@Shif].
We can decompose the proton mass as follows, leaving out the contribution of the heavy quarks: $$\begin{aligned}
M_p & = & 938 \, \, \, {\rm MeV} \nonumber \\
& = & (862 \qquad + \qquad 20 \qquad + \qquad 19 \qquad + \qquad 35
\qquad
+ \qquad 2) \, \, \, {\rm MeV} \nonumber \\
& & \, \, \, \, \uparrow \, \, \, \, \, \, \, \, \hspace{2cm} \uparrow
\, \, \, \, \, \hspace{2cm} \uparrow \, \, \, \, \, \hspace{2cm}
\uparrow \, \, \, \hspace{2cm} \uparrow \nonumber \\
& & QCD \, \, \hspace{1cm} u-quarks \, \, \, \hspace{0.5cm}
d-quarks \, \, \, \hspace{0.4cm} s-quarks \, \, \, \, \, \hspace{0.2cm}
QED\end{aligned}$$ The masses of the heavy quarks $c$ and $b$ can be estimated by considering the spectra of the particles, containing $c$– or $b$–quarks, e. g. the charm–mesons or the $B$–mesons. One finds[@Rev]: $$\begin{aligned}
m_c: 1.15 & \ldots & 1.35 \, \, \, {\rm GeV} \, \, \, (\bar{MS}-mass)
\nonumber\\
m_b: \, 4.1 & \ldots & 4.4 \, \, \, {\rm GeV} \, \, \, (\bar{MS}-mass.\end{aligned}$$ The dark corner of the Standard Model is the sector of the fermion masses. There are the six quark masses, three charged fermion masses, three neutrino masses, four flavor mixing parameters of the quarks and six flavor mixing parameters of the leptons (if neutrinos are Majorana particles). These parameters make up 22 of the 28 fundamental constants.
What are the fermion masses? We do not know. They might also be due to a confined field energy, but in this case the quarks and leptons would have to have a finite radius, as in composite models. The masses would be generated by a new interaction. The experiments give a limit on the internal radius of the leptons and quarks, which is of the order of $10^{-17}$ cm[@Rev].
In the Standard Model the masses of the leptons and quarks are generated spontaneously, like the $W$ and $Z$–masses. Each fermion couples with a certain strength to the scalar ”Higgs“–boson via a Yukawa coupling. A fermion mass is then given by: $$m(fermion) = g \cdot V \, ,$$ where $V$ is the vacuum expectation value of the ”Higgs“–field. For the electron this Yukawa coupling constant must be very small, since $V$ is about 246 GeV: $$g(electron) = 0,00000208 \, .$$ Nobody understands, why this coupling constant is so small. The problem of fermion masses remains to be solved. It seems to be the most fundamental problem we are facing at the present time. New experiments at the LHC and at the International Linear Collider (ILC) might clarify the issue.
If one is interested only in stable matter, as e. g. in solid state physics, only seven fundamental constants enter: $$G, \Lambda, \alpha, m_e, m_u, m_d, m_s \, .$$ The mass of the $s$–quark has been included, since the $\left( \bar ss \right)$–pairs contribute to the nucleon mass about 40 MeV. These seven constants describe the atoms and molecules.
It is possible, that there exist relations between the fundamental constants. Relations, which seem to work very well, are the relations between the flavor mixing angles and the quark masses, which where predicted some time ago[@Frit]: $$\begin{aligned}
\Theta_u & = & \sqrt{m_u / m_c} \nonumber \\
\Theta_d & = & \sqrt{m_d / m_s} \, .\end{aligned}$$ Similar relations ca be derived for the neutrino masses and the associated mixing angles[@Frixi].
These relations are obtained if both for the $u$–type and for the $d$–type quarks the following mass matrices are relevant (texture 0 matrices): $$M = \left( \begin{array}{lll}
0 & A & 0 \\
A^* & C & B \\
0 & B^* & D
\end{array} \right) \, .$$ It would be interesting to know whether such mass matrices are indeed realized in nature.
Does the Finestructure Constant depend on Time?
===============================================
Recent observations in astrophysics[@Webb] indicate that the finestructure constant $\alpha$ depends on the cosmic time. Billions of years ago it was smaller than today. A group of researchers from Australia, the UK and the USA analysed the spectra of distant quasars, using the Keck telescope in Hawaii. They studied about 150 quasars, some of them about 11 billion lightyears away. The redshifts of these objects varied between 0.5 and 3.5. This corresponds to ages varying between 23% and 87% of the age of our universe.
They studied the spectral lines of iron, nickel, magnesium, zinc and aluminium. It was found that $\alpha$ is not constant: $$\frac{\Delta \alpha}{\alpha} = (- 0.72 \pm 0.18 ) \cdot 10^{-5} \, .$$ Taking into account the ages of the observed quasars, one concludes that in a linear approximation the absolute magnitude of the relative change of $\alpha $ must be: $$\left| \frac{d \alpha / dt }{\alpha} \right| \approx 1.2 \cdot 10^{-15}
/ year \, .$$ But recent observations of quasar spectra, performed by different groups, seem to rule out a time variation of $\alpha $ at the level given above[@Chasid; @Chaud].
The idea that the fundamental constants have a cosmological time dependence, is not new. In the 1930s P. Dirac[@Dirac] discussed a time variation of Newtons constant $G$. Dirac argued that the gravity constant should vary by about a factor of two during the lifetime of the universe. The present limit on the time variation of $G$ is: $\dot{G}/G \le 10^{-11}
year^{-1}$[@Damour]. According to Dirac’s hypothesis the time variation of $G$ should be about $10^{10} / year$, in conflict with the quoted limit. In the 1950s L. Landau discussed a possible time variation of the finestructure constant $\alpha $ in connection with the renormalization of the electric charge[@Land].
French nuclear physicists discovered that about 1.8 billion years ago a natural reactor existed in Gabon, West–Africa, close to the river Oklo. About 2 billion years ago uranium -235 was more abundant than today (about 3,7%). Today it is only 0,72%. The water of the river Oklo served as a moderator for the reactor. The natural reactor operated for about 100 million years.
The isotopes of the rare earths, for example the element Samarium, were produced by the fission of uranium. The observed distribution of the isotopes today is consistent with the calculation, assuming that the isotopes were exposed to a strong neutron flux.
Especially the reaction of Samarium with neutrons is interesting[@Dam]: $$Sm(149) + n \rightarrow Sm (150) + \gamma \, .$$ The very large cross–section for this reaction (about $60 \ldots 90$ kb) is due to a nuclear resonance just above threshold. The energy of this resonance is very small: $E = 0.0973 \, \, \, {\rm eV}$. The position of this resonance cannot have changed in the past 2 billion years by more than 0.1 eV. Suppose $\alpha $ has changed during this time. The energy of the resonance depends in particular on the strength of the electromagnetic interaction. Nuclear physics calculations give: $$\frac{ \alpha \left( Oklo) \right) - \alpha (now)}{\alpha (now)} <
10^{-7}
\, .$$ The relative change of $\alpha $ per year must be less than $10^{-16}$ per year, as estimated by T. Damour and F. Dyson[@Dam]. This conclusion is correct only if no other fundamental parameters changed in the past two billion years. If other parameters, like the strong interaction coupling constant, changed also, the constraint mentioned above does not apply.
The Oklo constraint for $\alpha $ is not consistent with the astrophysical observation for the relative changes of $\alpha $ of order $10^{-15}$ per year. However, if other parameters also changed in time there will be a rather complicated constraint for a combination of these parameters, but there is no inconsistency.
Recently one has also found a time change of the mass ratio $$\mu = \frac{M(proton)}{m(electron)} \, .$$ One observed the light from a pair of quasars, which are 12 billion light years away from the earth[@Rein]. This light was emitted, when the universe was only 1.7 billion years old. The study of the spectra revealed, that the mass ratio $\mu $ has changed in time: $$\frac{\Delta \mu}{\mu} \approx \left( 2 \pm 0.6 \right) \cdot 10^{-5}
\, .$$ Taking into account the lifetime of 12 billion years, the change of $\mu $ per year would be $10^{-15}$ / year.
Grand Unification
=================
In the Standard Model we have three basic coupling constants. The gauge group of the Standard Model is $SU(3)_c \times SU(2) \times U(1)$.
\[Show Quoted Text - 660 lines\]\[Hide Quoted Text\] The three gauge interactions are independent of each other.
Since 1974 the idea is discussed that the gauge group of the Standard Model is a subgroup of a larger simple group. The three gauge interactions are embedded in a Grand Unified Theory (GUT). A Grand Unification implies that $\alpha_3, \alpha_2$ and $\alpha_1$ are related. They can be expressed in terms of the unified coupling constant $\alpha_{un}$ and the energy scale of the unification $\Lambda_u$.
The simplest theory of Grand Unification is based on the gauge group $SU(5)$[@Geor]. The quarks and leptons of one generation can be described by two $SU(5)$–representations. Let us consider the 5–representation of $SU(5)$. After the breakdown of $SU(5)$ to $SU(3) \times SU(2) \times U(1)$ one obtains: $$\begin{aligned}
5 & \rightarrow & (3,1) + (1,2) \nonumber \\
\bar 5 & \rightarrow & (\bar 3,1) + (1,2) \, .\end{aligned}$$ The 5–representation contains a color triplet, which is a singlet under $SU(2)$, and a color singlet ($SU(2)$–doublet): $$(\bar 5) = \left( \begin{array}{l}
\bar d_r\\
\bar d_g\\
\bar d_b\\
\nu_e\\
e^- \end{array} \right) \, .$$ The representation with the next higher dimension is the 10–representation, which is an antisymmetric second–rank tensor. The 10–representation decomposes after as follows: $$(10) \rightarrow (3,2) + (\bar 3,1) + (1,1)$$ In terms of the lepton and quark fields of the first generation we can write the 10–representation (an antisymmetric $5 \times 5$–matrix) as follows: $$(10) = \frac{1}{\sqrt{2}} \left( \begin{array}{ccccc}
0 & \bar u_b & - \bar u_g & - \bar u_r & - \bar d_r \\
- \bar u_b & 0 & \bar u_r & \- \bar u_g & - \bar d_g
\\
\bar u_g & - \bar u_r & 0 & - \bar u_b & - \bar d_b
\\
u_r & u_g & u_b & 0 & e^+\\
d_r & d_g & d_b & -e^+ & 0 \, .
\end{array} \right) \, .$$ Combining these two representations, one finds the lepton and quarks of one generation: $$\bar 5 + 10 \rightarrow (3,2) + 2 \left( \bar 3,1 \right) + \left(1,2
\right)
+ (1,1) \, .$$ For the first generation we have: $$\bar 5 + 10 \rightarrow \left( {u \atop d} \right)_L + \bar u_L + \bar
d_L
+ \left( {\nu_e \atop e^-} \right)_L + e^+_L \, .$$ The second and third generation are analogous. The unification based on the gauge gruop $SU(5)$ has a number of interesting features:
1. The electric charge is quantized. $$t r Q = O \rightarrow \, Q(d) = \frac{1}{3} \, Q \left( e^- \right)$$
2. At some high mass scale $\Lambda_{un}$ the gauge group of the Standard Model turns into the group $SU(5)$, and there is only one single gauge coupling. The three coupling constants $g_3, g_2, g_1$ for $SU(3), SU(2)$ and $U(1)$ must be of the same order of magnitude, related to each other by algebraic constants.
The rather different values of the coupling constants $g_3, g_2, g_1$ at low energies must be due to renormalization effects. This would also explain why the strong interactions are strong and the weak interactions are weak. It is related to the size of the corresponding group.
Apart from normalization constants the three coupling constants $g_3, g_2$ and $g_1$, are equal at the unification mass $\Lambda_{un}$. Thus the $SU(2) \times U(1)$ mixing angle, given by $tan \Theta_w = \frac{g_1}{g_2}$, is fixed at or above $\Lambda_{un}$: $$sin^2 \Theta_w = tr T_3^2 / tr Q^2 = \frac{3}{8} \, .$$ At an energy scale $\mu << \Lambda_{un}$ the parameter $sin^2 \Theta$ changes along with the three coupling constants: $$\begin{aligned}
\frac{sin^2 \Theta_w}{\alpha} - \frac{1}{\alpha_s} & = & \frac{11}{6
\pi}
\, ln \, \left( \frac{M}{\mu} \right) \nonumber \\
\alpha / \alpha_s & = & \frac{3}{10} \left( 6 sin^2 \Theta_w - 1
\right) \, .\end{aligned}$$ At $\mu = M_z$ the electroweak mixing angle has been measured: $sin^2\Theta_w = 0.2312$. Note that above the unification energy $\alpha$ and $\alpha_s$ are related: $$\alpha / \alpha_s = 3/8 \, .$$ This relation can be checked by experiment. In order to get an agreement between the observed values for $g_3, g_2$ and $g_1$ and the values predicted by the $SU(5)$ theory, one can easily see that the unification scale must be very high. Note that $$\begin{aligned}
ln \left( \frac{M}{\mu}\right) & = & \frac{6 \pi}{11}
\left( \frac{sin^2 \Theta_w}{\alpha} - \frac{1}{\alpha_s} \right)
\nonumber \\
\mu & = & M_Z \nonumber \\
ln \left( M / M_Z \right) & \cong & 39,9 \nonumber \\
M & \approx & 2 \cdot 10^{15} \, \, \, {\rm GeV} \, .\end{aligned}$$ The precise values of the three coupling constants, determined by the LEP–experiments[@Rev], disagree with the $SU(5)$ prediction. The three coupling constants do not converge to a single coupling constant $\alpha_{un}$[@Amal]. A convergence takes place, if supersymmetric particles are added above the energy of 1 TeV. Supersymmetry implies that for each fermion a boson is introduced (s–leptons, s-quarks), and for each boson a new fermion is introduced (photino, etc.). These new particles are not observed in the experiments. It is assumed that they have a mass of about 1 TeV.
The new particles contribute to the renormalization of the gauge coupling constants at high energies (about 1 TeV). A convergence of the three coupling constants taken place. Therefore a supersymmetric version of the $SU(5)$–theory is consistent with the experiments[@Amal].
In theories of Grand Unification like the $SU(5)$–theory one has quarks, antiquarks and leptons in one fermion representation. Thus the proton can decay, e. g. $p \rightarrow e^+ \pi^0$. The lifetime depends on the mass scale for the unification. For $\Lambda_{un} = 5 \cdot 10^{14} \, \, \, {\rm GeV}$ in the $SU(5)$–theory without supersymmetry one finds $10^{30}$ years for the proton lifetime. The experimental lower limit is about $10^{33}$ years.
There is a natural embedding of a group $SU(n)$ into $SO(2n)$, due to the fact that $n$ complex numbers can be represented by $2n$ real numbers. One may consider to use the gauge group $SO(10)$ instead of $SU(5)$. This was discussed in 1975 by P. Minkowski and the author[@Frimin]. The fermions of one generation are described by a 16–dimensional spinor representation of $SO(10)$.
Since $SU(5)$ is a subgroup of $SO(10)$, one has the following decomposition: $$16 \rightarrow \bar 5 + 10 + 1 \, .$$ The fermions of the $SU(5)$–theory are obtained, plus one additional fermion (per family). This state is an $SU(5)$–singlet and describes a lefthanded antineutrino field. Using the leptons and quarks of the first generation we can write the 16–representation as follows in terms of lefthanded fields: $$(16) = \left( \begin{array}{lllllllll}
\bar \nu_e & \bar u_r & \bar u_g & \bar u_b & \vdots & u_r & u_g & u_b
&\nu_e \\
e^+ & \bar d_r & d_y & \bar d_b & \vdots & d_r & d_g & d_b &e^-
\end{array} \right)$$ A feature of the $SO(10)$–theory is that the gauge group for the electroweak interactions is larger than in the $SU(5)$–theory. $SO(10)$ has the subgroup $SO(6) \times SO(4)$. Since $SO(4)$ is isomorphic into $SU(2) \times SU(2)$, one finds: $$SO(10) \rightarrow SU(4) \times SU(2)_L \times SU(2)_R \, .$$ The group $SU(4)$ must contain the color group $SU(3)^c$. The 16–representation of the fermions decomposes under $SU(4)$ into two 4–representations. These contain three quarks and one lepton, e. g. $\left( d_r, d_g, d_b \right)$ and $e^-$. One may interpret the leptons as the fourth color. But the gauge group $SU(4)$ must be broken at high energies (higher than at least 1 TeV): $$SU(4) \rightarrow SU(3) \times U(1) \, .$$ We obtain at low energies the gauge group $$SU(3)^c \times SU(2)_L \times SU(2)_R \times U(1) \, .$$ But the masses of the gauge bosons for the group $SU(2)_R$ must be much larger than the observed $W$–bosons, related to the group $SU(2)_L$.
In the $SU(5)$–theory the minimal number of fermions of the Standard Model is included. In the $SO(10)$–theory a new righthanded neutrino is added. This righthanded fermion is interpreted as a heavy Majorana particle. A mass for the lefthanded neutrino is generated by the ”see–saw“–mechanism[@Moha]. Thus in the $SO(10)$–theory the neutrinos are massive, while in the $SU(5)$–theory they must be massless. The $SO(10)$–theory is more symmetrical than the $SU(5)$–theory. It is hard to believe that Nature would stop at $SU(5)$, if Nature has chosen to unify the basic interactions.
In the $SO(10)$–theory there is one additional free parameter, related to the masses of the righthanded $W$–bosons. Since righthanded charged currents are not observed, the masses of the associated $W$–bosons must be rather high, at least 300 GeV[@Yana]. There is a new parameter $M_R$ in the $SO(10)$–theory. It can be chosen such that the coupling constant converges at very high energies, without using supersymmetry. If one chooses $M_R \sim 10^9 \ldots 10^{11}$ GeV, the convergence occurs.
The idea of Grand Unification leads to the reduction of the fundamental constants by one. The three gauge coupling constants of the Standard Model can be expressed in terms of a unified coupling constant $\alpha_u$ at the energy $\Lambda_u$, where the unification takes place. The three coupling constants $\alpha_s, \alpha_2, \alpha_1$ are replaced by $\alpha_u$ and $\Lambda_u$.
In a Grand Unified Theory the three coupling constants of the Standard Model are related to each other. If e. g. the finestructure constant shows a time variation, the other two coupling constants should also vary in time. Otherwise the unification would not be universal in time. Knowing the time variation of $\alpha $, one should be able to calculate the time variation of the other coupling constants.
We shall investigate here only the time change of the QCD coupling constant $\alpha_s $.
We use the supersymmetric $SU(5)$–theory to study the time change of the coupling constants[@Chang; @Cal]. The change of $\alpha $ is traced back to a change of the unified coupling constant at the energy of unification and to a change of the unification energy. These changes are related to each other: $$\frac{1}{\alpha} \frac{\dot{\alpha}}{\alpha} = \frac{8}{3} \cdot
\frac{1}{\alpha_s} \cdot \left( \frac{\dot{\alpha}_s}{\alpha_s} \right)
-
\frac{10}{\pi} \frac{\dot{\Lambda}_{un}}{\Lambda_{un}}.$$
We consider the following three scenarios:
1. $\Lambda_{un}$ is kept constant, $\alpha_u = \alpha_u(t)$. We obtain: $$\frac{1}{\alpha } \frac{\dot \alpha}{\alpha} = \frac{8}{3}
\frac{1}{\alpha_s} \frac{ \dot{\alpha}_s}{\alpha_s} \, .$$
Using the experimental value $\alpha_s \left( M_Z \right)
\approx 0.121$, we find for the time variation of the QCD scale[@Chang]:\
$$\begin{aligned}
\frac{\dot \Lambda}{\Lambda} & \approx & R \cdot
\frac{\dot{\alpha}}{\alpha}
\nonumber \\
\nonumber\\
R & \approx & 38 \pm 6 \, .\end{aligned}$$ The uncertainty in $R$ comes from the uncertainty in the determination of the strong interaction coupling constant $\alpha_s$. A time variation of the QCD scale $\Lambda$ implies a time change of the proton mass and of the masses of all atomic nuclei. The change of the nucleon mass during the last 10 billion years amounts to about 0.3 MeV.
In QCD the magnetic moments of the nucleon and of the atomic nuclei are inversely proportional to the QCD scale parameters $\Lambda $. We find for the nuclear magnetic moments: $$\frac{\dot{\mu}}{\mu} =
\frac{\frac{d}{dt} \left(\frac{1}{\Lambda} \right)}{\Lambda} =
- \frac{\dot{\Lambda}}{\Lambda} = - R \cdot \frac{\dot{\alpha}}{\alpha}
\, .$$ Taking the astrophysics result for $\left( \dot{\alpha} / \alpha
\right)$, we obtain: $$\frac{\dot{\Lambda}}{\Lambda} \approx 4 \cdot 10^{-14} / yr \, .$$
2. The unified coupling constant is kept invariant, but $\Lambda_{un}$ changes in time. In that case we find[@Cal]: $$\frac{\dot{\alpha}}{\alpha} \cong - \alpha \cdot \frac{10}{\pi}
\frac{\dot{\Lambda}_{un}}{\Lambda_{un}}$$ and $$\frac{\dot{\Lambda}}{\Lambda} \approx - 31 \cdot
\frac{\dot{\alpha}}{\alpha}
\, .$$ The change of the unification mass scale $\Lambda_{un}$ can be estimated, using as input the time variation of the finestructure constant $\alpha $. Thus $\Lambda_{un}$ is decreasing at the rate $$\dot{\Lambda}_{un} / \Lambda_{un} \approx - 7 \cdot 10^{-14} / yr \, .$$ The relative changes of $\Lambda $ and $\alpha $ are opposite in sign. While $\alpha $, according to ref. [@Webb], is increasing with a rate of $10^{-15} / yr$, the QCD scale $\Lambda $ and the nucleon mass are decreasing with a rate of about $3 \cdot 10^{-14} / yr$. The magnetic moments of the nucleons and of nuclei would increase: $$\frac{\dot{\mu}}{\mu} \approx 3 \cdot 10^{-14} / yr \, .$$
3. The third possibility is that both $\alpha_u$ and $\Lambda_{un}$ are time–dependent. In this case we find: $$\frac{\dot{\Lambda}}{\Lambda} \cong 46 \cdot
\frac{\dot{\alpha}}{\alpha}
+ 1,07 \cdot \frac{\dot{\Lambda}_{un}}{\Lambda_{un}} \, .$$ On the right two relative time changes appear: $\left( \dot{\alpha} / \alpha \right)$ and $\left( \dot{\Lambda}_{un} /
\Lambda_{un} \right)$. These two terms might conspire in such a way that $\left( \dot{\Lambda} / \Lambda \right)$ is smaller than about $\left(\pm 40 \cdot \dot{\alpha} / \alpha \right)$.
The question arises, whether a time change of the QCD scale parameter could be observed in the experiments. The mass of the proton and the masses of the atomic nuclei as well as their magnetic moments depend linearly on the QCD scale. If this scale changes, the mass ratio $M_{p} / m_e = \mu$ would change as well, if the electron mass is taken to be constant.
The mass ratio $\mu $ seems to show a time variation – in a linear approximation one has about $$\frac{\Delta \mu}{\mu} \approx 10^{-15} / year \, .$$ If we take the electron mass to be constant in time, this would imply that the QCD–scale $\Lambda $ changes with the rate $$\frac{\Delta \Lambda}{\Lambda} \approx 10^{-15} / year \, .$$
The connection between a time variation of the finestructure constant and of the QCD scale, discussed above, is only valid, if either the unified coupling constant or the unification scale depends on time, not both. If both the unification scale and the unified coupling constant are time dependent, we should use instead eq. (57). There might be a cancellation between the two terms. In this case the time variation of the QCD–scale would be smaller than $10^{-14} / year$. If the two terms cancel exactly, the QCD–scale would be constant, but this seems unlikely. Therefore a time variation of the QCD–scale of the order of $10^{-15} / year$ is quite possible.
Can such a small time variation of $\Lambda_c$ be observed in the experiments? In Quantum Optics one can carry out very precise experiments with lasers. In the next chapter we shall describe such an experiment at the Max–Planck–Institute of Quantum Optics in mMunich, which was designed especially to find a time variation of the QCD scale $\Lambda_c $.
Results from Quantum Optics
===========================
The hydrogen atom is a very good test object for checking fundamental theories. Its atomic properties can be calculated with very high accuracy. The level structure of the hydrogen atom can be very accurately probed, using spectroscopy methods in the visible, infrared and ultraviolett regions. Thus the hydrogen atom plays an important rôle in determining the fundamental constants like the finestructure constant.
Measurements of the Lamb shift and the 2S hyperfine structure permit very sensitive tests of quantum electrodynamics. Combining optical frequency measurements in hydrogen with results from other atoms, stringent upper limits for a time variation of the finestructure constant[@Lang] and of the QCD scale parameter can be derived.
The employment of frequency combs[@Kola] turned high–precision frequency measurements into a routine procedure. The high accuracy of the frequency comb have opened up wide perspectives for optical atomic clock applications in fundamental physics. Frequency measurements in the laboratory have become competitive recently in terms of sensitivity to a possible time variation of the fine–structure constant. Though the time interval covered by these measurements is restricted to a few years, very high accuracy compensates for this disadvantage. Their sensitivity becomes comparable with astrophysical and geological methods operating on a billion–year time scale.
Important advantages of the laboratory experiments are: The variety of different systems that may be tested, the possibility to change parameters of the experiments in order to control systematic effects, and the determination of the drift rates from the measured data. Modern precision frequency measurements deliver information about the stability of the present values of the fundamental constants, which can only be tested with laboratory measurements. But only non-laboratory methods are sensitive to processes that happened in the early universe, which can be much more severe as compared to the present time.
In the experiment of the MPQ–group in Munich[@Lang] one was able to determine the frequency of the hydrogen 1S–2S–transition to 2466061102474851(34) Hz. A comparison with the experiment performed in 1999 gives an upper limit on a time variation of the transition frequency in the time between the two measurements, 44 months apart. One finds for the difference $(-29 \pm 57) Hz$, i. e. it is consistent with zero.
The hydrogen spectrometer can be interpreted as a clock, like the cesium clock. However in the hydrogen spectrometer one uses a normal transition for the determination of the flow of time. This transition depends on the mass of the electron and on the fine structure constant. In a cesium clock the flow of time is determined by a hyperfine transition, which depends on the fine structure constant, but also on the nuclear magnetic moment.
Comparing the $1S - 2S$ hydrogen transition with the hyperfine transition of Cesium $^{133} Cs$, one can obtain information about the time variation of the ratio $\alpha /
\alpha_s$. The Cesium hyperfine transition depends on the magnetic moment of the Cesium nucleus, and the magnetic moment is proportional to ($ 1 / \Lambda_c$, ($\Lambda_c$: QCD scale parameter). If $\Lambda_c$ varies in time, the magnetic moment will also vary.
One has obtained a limit for the time variation of the magnetic moment of the Cesium nucleus[@Lang]: $$\frac{\delta \mu}{\mu} = \left( 1.5 \pm 2.0s \right) \cdot 10^{-15} /
yr \, .$$ These results are consistent with zero. The limit on the time variation of $\alpha $ is of the same order as the astrophysics result.
The result concerning the magnetic moment implies a limit on the time variation of $\Lambda_c$: $$\frac{\Delta \Lambda_c}{\Lambda_c} = \left( - 1.5 \pm 2.0 \right) \cdot
10^{-15} / yr \, .$$ This result is in disagreement with our results, based on the assumption, that either $\alpha_u$ or $\Lambda_{un}$ change in time. We obtained about $10^{-14} / yr$, which is excluded by this experiment.
The result given above is consistent with no time change for $\Lambda_c$, but it also agrees with a small time change of the order of $10^{-15}$ per year. If we assume that the electron mass does not change in time, such a change of $\Lambda_c$ would agree with the astrophysics result on the time variation of the ratio M(proton) / m(electron)[@Rein]. Theoretically we would expect such a time variation, if both $\Lambda_{un}$ and $\alpha_u$ change in time.
Conclusions and Outlook
=======================
We have summarized our present knowledge about the fundamental constants and their possible time variation. Today we do not know how these constants are generated or whether they might depend on time. There might be relations between these constants, e. g. between the flavor mixing angles and the fermion masses, or relations between the three coupling constants, implied by the idea of Grand Unification. This would reduce the number of basic constants from 28 down to a smaller number, but at least 18 fundamental constants would still exist.
A possible time variation of the fundamental constants must be rather slow, at least for those fundamental constants, which are measured very precisely, i. e. the finestructure constant, the QCD–scale $\Lambda $, and the electron mass. The constant of gravity $G$ is known with a precision of $10^{-11}$. All other fundamental constants, e. g. the masses of the other leptons or the masses of the heavy quarks, are not known with a high precision. The present limits on the time variation of the finestructure constant, the QCD scale or the electron mass are of the order of $10^{-15}$ / year. These limits should be improved by at least two orders of magnitude in the near future.
If the astrophysics experiments indicate a time variation of the order of $10^{-15}/ year$, it does not mean that experiments in quantum optics should also give such a time variation. It might be that until about 10 billion years after the Big Bang the constants did vary slowly, but after that they remained constant. No theory exists thus far for a time variation, and there is no reason to believe that a time variation should be linear, i. e. $10^{-15} / year$ throughout the history of our universe. If the fundamental constants do vary, one would expect that the variation is rather large very close to the Big Bang. In the first microseconds after the Big Bang constants like $\alpha $ or $\Lambda_c$ might have changed by a factor 2, and we would not know.
In cosmology one should consider time variations of fundamental parameters in more detail. Perhaps allowing a suitable time variation of the constants leads to a better understanding of the cosmic evolution immediately after the Big Bang. Allowing time variations might lead to better cosmological theories and to a better understanding of particle physics. Particle physics and cosmology together would give a unified view on the universe.
[99]{} A. Fritzsch and M. Gell-Mann, [*Proc. of the XV. International Conference on High Energy Physics, Chicago*]{}, (1972).\
H. Fritzsch, M. Gell–Mann and H. Leutwyler, [*Phys. Lett.*]{}[**B47**]{}, 365 (1973). S. L. Glashow, [*Adv. Nucl. Phys.*]{}[**22**]{}, 433 (1961). S. L. Glashow and J. C. Ward, [*Phys. Lett.*]{}[**13**]{}, 168 (1964).\
S. Weinberg, [*Phys. Rev. Lett.*]{} [**19**]{}, 1264 (1962).\
A. Salam, [*in: Elementary Particle Theory*]{} [**367**]{} (1969), ed. N. Svartholm, Stockholm. W. M. Yao et al. (Particle Data Group), [*J. Phys.*]{}[**G33**]{}, 1 (2006). P. Higgs, [*Phys. Rev. Lett.*]{}[**13**]{}, 508 (1964).
R. Brout and F. Englert, [*Phys. Rev. Lett.*]{}[**13**]{}, 321 (1964).
G.S. Guralnik, C. R. Hagen, and T. W. B. Kibble, [*Phys. Rev. Lett.*]{}[**13**]{}, 585 (1964). T. R. Taylor and G. Veneziano, [*Phys. Lett.*]{}[**B213**]{}, 450 (1988). E. Witten, [*Phys. Lett.*]{}[**B149**]{}, 351 (1984). A. Wyler, [*Physics Today*]{},[24 Nr.8]{}[**17**]{} (1971). R. P. Feynman, [*Princeton University Press*]{}[**129**]{} (1985). M. Gell–Mann and F. Low, [*Phys. Rev.*]{}[**25**]{}, 1300 (1954). N. S. Craigie, S. Narison and Riazuddin, [*Nucl. Phys.*]{}[**B174**]{}, 207 (1980). H. Leutwyler, [*Nucl. Phys. Troc. Suppl.*]{}[**94**]{}, 108 (2001).\
S. Narison, [*Phys. Rev.*]{}[**D74**]{}, 034013 (2006). M. A. Shifman, A. I. Vainstein, V. I. Zakharov, [*Phys. Lett.*]{}[**78B**]{}, 443 (1978). H. Fritzsch, [*Nucl. Phys.*]{}[**B155**]{}, 189 (1979). H. Fritzsch and Z. Xing, [*Phys. Lett.*]{}[**B624**]{}, 514 (2006).\
H. Fritzsch, Proc. of JCFP Peking, (2007), to appear. J. K. Webb et al., [*Phys. Rev. Lett.*]{}[**87**]{},(091301) (2001). H. Chasid, R. Srianand, P. Petitjeau and B. Aracil, [*Astron. and Astrophys.*]{}[**417**]{}, 813 (2004).\
R. Quast, D. Reimers and S.A. Levshakow, [*Astron. and Astroph.*]{}[**415**]{}, 27 (2004). H. Chaud et al., [*Astron. Astrophys.*]{}[**417**]{}, 853 (2004). P. Dirac, [*Nature*]{} [**192**]{}, 325 (1937). T. Damour and J. H. Taylor, [*ApJ*]{}[**366**]{}, 501 (1991). L. D. Landau, [*in: Niels Bohr and the Development of Physics*]{} Mc–Graw–Hill, Nes York (1955) . T. Damour and F. Dyson, [*Nucl. Phys.*]{}[**B480**]{}, 37 (1996). E. Reinhold, R. Buning, U. Hollenstein, A. Ivanchik, P. Petitjean and W. Ubachs, [*Phys. Rev. Lett.*]{}[**96**]{}, 151101 (2006). H. Georgi and S. Glashow, [*Phys. Rev. Lett.*]{}[**32**]{}, 438 (1974). U. Amaldi, W. de Boer and H. Fürstenau, [*Phys. Lett.*]{}[**B260**]{}, 447 (1991).\
J. Ellis, S. Kelly and P. V. Nanopoulos, [*Phys. Lett.*]{}[**B260**]{}, 311 (1991).\
P. Langacker and N. Polonsky, [*Phys. Rev.*]{}[**D47**]{}, 4028 (1993).\
M. Carena, S. Pokorsky and C. E.M. Wagner, [*Nucl. Phys.*]{}[**B406**]{}, 59 (1993).\
M. Shiozawa et al., [*Phys. Rev. Lett.*]{} [**81**]{}, 3319 (1998).\
Y. Hayato et al., [*Phys. Rev. Lett.*]{}[**83**]{}, 1529 (1999). H. Fritzsch and P. Minkowski, [*Annals of Physics*]{} [**93**]{}, 1293 (1975). P. Minkowski, [*Phys. Lett.*]{} [**B67**]{}, 421 (1977).\
T. Yanagida, [*Progress of Theor. Physics*]{} [**64**]{}, 1103 (1980).\
M. Gell-Mann, P. Ramond and R. Slansky, [*in: Supergravity, North Holland*]{} (1979).\
R. Mohapatra and G. Senjanovic, [*Phys. Rev. Lett.*]{} [**44**]{}, 912 (1980). S. Abachi et al., [*Phys. Rev. Lett.*]{}[**76**]{}, 327 (1996).\
D. Chang, R. Mohapatra, J. Gipson, R. E. Marshak and M. K. Parida, [*Phys. Rev.*]{} [**D31**]{}, 1718 (1985). X. Calmet and H. Fritzsch, [*Phys. Lett.*]{} [**540**]{}, 173 (2002). P. Langacker, G. Segre and M. J. Strassler, [*Phys. Lett.*]{} [**B528**]{}, 21 (2002). N. Kolachevsky, J. Alnis, A. Matveev, Th. Udem, R. Holzwarth and T. W. Hänsch: Precision Measurements in Atomic Hydrogen, in: Optical Atomic Clock, Y. Ovchinnikov. ed., to appear. T. Udem, R. Holzwarth and T. W. Haensch, [*Nature*]{} [**416**]{}, 236 (2002).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: '[**Summary:**]{} The idea is a nanoscale extension of a sporting event called a ‘color run’.[@ColorRun] The filament is wrapped with a DNA origami ‘barrel’,[@HwangCLEO15] which acts as the nanoscale equivalent of a white t-shirt. The DNA t-shirt has dangling ss-DNA ‘handles’ that bind ‘tags’ consisting of an ‘anti-handle’ oligo bound to a fluorophore. The tagging methodology exploits the recent development of DNA-based ‘metafluorophores’[@WoehrsteinSciAdv17] with digitally tunable optical properties based on a collection of fluorophores bound to a single DNA origami structure. In this idea, the filament collects different color tags at ‘color-stations’ as it travels through the network (Fig. 1). These contribute to a final DNA t-shirt color that represents of the path taken. Readout is implemented through lenseless on-chip imaging[@OzcanARBE16] via pixels in a CMOS backplane located at read-out points in the network. Color-stations are implemented via microfluidic channels on a dialysis membrane,[@ShengAnalyst14] which feed tags into the network in localised areas. A chemical countermeasure to prevent false tagging by stray tags that have drifted in from other stations by exploiting strand-displacement techniques[@ZhangNatChem11] is also proposed.'
author:
- 'A.P. Micolich'
title: 'Colorimetric path tagging of filaments using DNA-based metafluorophores'
---
[**1. Requirements Summary:**]{} Briefly summarising requirements for later reference. The goal is to encode and readout information about filament path in a network biocomputation device.[@NicolauPNAS16] Specific requirements are: A. ability to write information into filaments in regions a few micrometers in size; B. ability to readout at sites located hundreds of micrometers or more away from writing regions; C. writing at least $10$ ideally $100$s of tags; D. ability to tag and read at minimum $10$, ideally $100$s of different sites; E. information must be carried with the filaments; and F. readout must allow decoding of information from $>10^3$, ideally $10^6 - 10^9$ individual filaments. I will refer to these specifically as Req. A-F in the discussion that follows. I will focus on microtubules but see no obvious reason this couldn’t work for actin filaments.
[**2. Basic Concept:**]{} A major challenge for nanoscale tagging with significant information content is space-availability. Physical barcodes scale with the minimum resolvable bit-size and the number of bits required. For Reqs. A,C & E above, a strategy to make the bit-size tiny and/or maximise encoded information-per-bit is needed. Further, rapid read-out is needed to be practical to Reqs. D & F. This provides strong incentive for solutions giving parallel or single-shot read-out of whole-path information. This inspired an idea where whole-path information is accumulated and read-out in the optical properties, e.g., color, relative brightness, photostability, of a single microscopic ‘pixel’ (or few pixels). This colorimetric tagging would be implemented through recent advances in DNA nanotechnology, as described below. Before getting to the ‘nuts and bolts’, I will quickly use a macroscopic analogy to illustrate the basic concept.
There is a popular sporting event called the ‘Color Run’.[@ColorRun] Participants run a $5$ km course solely for fun (no winners/no prizes). Runners start wearing a white t-shirt and run through a ‘color-station’ at each km (Fig. 1a), where they are sprayed with a given paint colour. The run ends with colorful t-shirts and happy runners, but it also provides a means to determine if someone ran just the first km and took an Uber to the finish – their t-shirt would be green not rainbow colored. One could easily extend this idea to a course with multiple paths, each with a station dispensing a different color and precise amount of paint. At the end, a single read-out of one parameter, t-shirt color, tells you about the path they took. To do this with filaments, we need nanoscale equivalents for: the t-shirt, paint, an ability to control the paint supply, and a way to read the color.
[**3. Usually just a t-shirt:**]{} Microtubule-sized t-shirts don’t exist yet and microtubules don’t have arms, so my idea is to use a DNA origami ‘barrel’ structure instead (Fig. 1b).[@HwangCLEO15; @WickhamWeb; @KnudsenNatNano15] These barrels can be made small enough to fit the $25$ nm microtubule diameter,[@HwangCLEO15] carrying up to $2000$ uniquely addressable ‘handles’ spaced $9$ nm apart[@WickhamWeb] for tag-attachment. The barrels undergo two-step assembly, first forming a rectangular tile that’s closed into a cylinder via staple strands that bind opposite edges of the tile. The remaining two edges can be used to polymerize barrels, with barrel trimers of $\sim100$ nm length already demonstrated.[@KnudsenNatNano15]
The DNA handles[@WickhamWeb] on the inside surface can be used to ‘fit’ the t-shirt to the filament via an ‘anti-handle’ oligo bound to an $\alpha$-tubulin antibody or phalloidin molecule.[@AgastiChemSci17] The external handles host tags (see §4). Filaments are well-known to remain motile whilst carrying cargo, e.g., quantum dots,[@NitzscheNatNano08] and recent work shows no evidence of adversely-strong binding between DNA origami and kinesin[@WollmanNatNano14] or myosin[@IwakiNatComm16] that would shut-down motility.
[**4. Now add paint:**]{} Information encoding is inspired by the DNA metafluorophores developed by Woehrstein [*et al.*]{}[@WoehrsteinSciAdv17] These feature a DNA origami tile with three ‘stripes’ of $44$ handles. Each stripe binds anti-handles (Fig. 1d) with a different fluorophore colour (red, green or blue), giving the whole tile a highly tunable colour and intensity, hence the name ‘metafluorophore’. Ref. 3 used a $60~\times~90$ nm tile, which would translate to a t-shirt of circumference $80$ nm ($\approx 25\pi$ nm) and length $\sim 65$ nm (Fig 1b). The fluorophore colors are separated into stripes to reduce issues with self-quenching and FRET.[@WoehrsteinSciAdv17]
DNA metafluorophores have been demonstrated as single-pixel barcodes, with a maximum of $5^3~-~1~=~124$ distinct codes obtained by attributing five intensity levels to each color.[@WoehrsteinSciAdv17] Correct-read accuracies surpassing $95\%$ were obtained by requiring each color to have non-zero intensity (i.e., drop lowest intensity level from coding system), giving $4^3 = 64$ individual codes (see Table S5 of Ref. 3). In Ref. 3, all $44$ handles in a stripe have the same sequence. In my idea all handles are unique[@WickhamWeb] and color-station specific, enabling the final color to contain information about passage via $64$ separate network points.
Disentangling color to specific path would involve a filament carrying several t-shirts with differing handle combinations. A $5~\mu$m filament could carry $15$ t-shirts and still have $80\%$ free surface for motor-binding (Fig. 1c). This could be five sets of three for redundacy/error-minimisation. One set could be pre-loaded with fluorophores to act as on-board ‘standard candles’ during detection. The remaining four sets could display different handle combinations ($= 320$ bits available), cleverly chosen so that exact path information can be deduced from the specific color sequence that the set of t-shirts present. Devoting all $15$ t-shirts individually to this task gives $960$ distinct bits. Either case meets Req. C. Information density could be further increased by using the photostability of different dyes of the same color, e.g., Atto-647N & Alexa 647, as information.[@WoehrsteinSciAdv17]
[**5. Look at the pretty lights:**]{} Read-out is via t-shirt optical properties: color, relative intensity, possibly also photostability, etc. The $64$-bit set-up in §4 means $18$ fluorophores minimum ($132$ max).[@WoehrsteinSciAdv17] This is obviously possible using traditional optical microscopy. [@WoehrsteinSciAdv17] It may be a cost-effective option given super-resolution techniques can now be reasonably performed using modern cellphone CMOS imaging chips[@DiederichArxiv18] and this task shouldn’t require super-high spatial resolution. A nicer alternative would be on-chip fluorescence microscopy. [@OzcanARBE16; @HuangSSC09; @HongSSC17; @TakeharaAIPAdv17] Huang [*et al.*]{} [@HuangSSC09] demonstrated on-chip detection of targeted oligo-DNA (21-bp) binding using Qdot-655 fluorophores with detection limits down to $\sim 20$ fluorophores/$\mu$m$^2$ for a $50 \times 50~\mu$m pixel size in 2009. Better performance is expected now, with detection limits down to $\sim 1$ fluorophore/$\mu$m$^2$ predicted. [@HuangSSC09] Both are sufficient for detection in this tagging concept. Technology for microfluidics on ultra-thin glass on CMOS for fluorescence imaging also exists. [@TakeharaAIPAdv17] The underpinning CMOS imaging technology is now at single-photon detection level at room temperature, [@MaOptica17] and capable of sub-micron pixels.[@TakahashiSensors17]
A large-area CMOS array isn’t required for this idea. Instead, a custom-CMOS backplane featuring $3$-pixel (or $3n$-pixel) RGB clusters at specific read-out sites (Fig. 1f) aligned to the network structures would suffice for ‘single-shot’ t-shirt color readout. One challenge with the t-shirts is that they are wrapped around the filament and not flat like in Ref. 3. This can be overcome by noting that microtubules rotate while gliding, typically travelling $4-10~\mu$m per $2\pi$ rotation.[@NitzscheNatNano08] Local enhancement by plasmonic nanoantennas could also improve detection.[@AcunaSci12] These nanoantennas could be nanofabricated metal bow-tie structures[@LeeOptEx13] built into the nanochannel floor/walls, possibly combined with gold nanoslits to mask part of the pixel for improved resolution following Gro[ß]{} [*et al.*]{}[@GrossNatNano18]
[**6. Time for a paint shower:**]{} The last aspect is how to distribute antihandle-fluorophore molecules (hereafter ‘tags’) to specific network locations. The color-station operating concept is shown in Fig. 1f. Localised addition of tags is achieved via a structured microfluidic top-plane separated from the underlying network flow chamber by a thin polycarbonate dialysis membrane.[@ShengAnalyst14] The membrane has nanoscale pores just large enough to allow tags to pass (Fig. 1f) but small enough to prevent filament escape.[@ShengAnalyst14] The membrane also provides an impedance to reduce coupling between the two fluidic systems and prevent the network flow-cell from being flooded with tags. In other words, the membrane ensures the paint is supplied as a fine spray rather than a tsunami. The membrane properties, tag concentration and flow would need to be engineered accordingly. The microfluidic top-plane would require at least two distinct PDMS layers, which is possible using conventional casting and bonding,[@AndersonAnalChem00] or 3D-printing methods.[@GlickMN16] Hence the final device structure (Fig. 1f) would be: (from bottom) CMOS backplane,[@TakeharaAIPAdv17] thin glass floor,[@TakeharaAIPAdv17] PMMA nanochannel layer,[@NicolauPNAS16] polycarbonate membrane,[@ShengAnalyst14] and microfluidics layers.[@AndersonAnalChem00]
[**7. Blowing in the wind:**]{} An issue in this approach is tags drifting away from their station, giving false signal. My idea includes a chemical countermeasure for deactivating stray tags (Fig. 1e). The nanochannel floors are coated in either casein[@NicolauPNAS16] or biotin.[@KeyaSciRep17] Short ss-DNA oligos are bound to either anti-casein or streptavidin and primed to the nanochannel surfaces via the color-stations prior to starting computation. I will refer to these as ‘floor-strands’ (red in Fig. 1e). A series of ‘deactivation handles’ (green in Fig. 1e) are made for each color-station; these are partially complementary to the local floor strands and fully complementary to tags from adjacent/upstream color-stations. Note that these also need to be orthogonal (at least very weakly complementary) for their own color-station’s tags. The deactivation handles are flushed through prior to computation to prime the floor-strands (Fig. 1e-I) and supplied continuously at low flow during computation to replenish deactivation handles lost to the following deactivation process. The deactivation handles function by capturing stray tags (blue with red spot in Fig. 1e-II) from adjacent color-stations. In the process the deactivation handle detaches from the floor-strand by ‘strand displacement’,[@ZhangNatChem11] with the deactivated stray tag leaving the flow channel as waste in the buffer flow (Fig. 1e-III). The floor-strand is then ready to capture a new deactivation handle from its color-station, returning to the state in Fig. 1e-I to await another stray tag. Noting that the network flow-cell is operated under directional flow for ATP supply, deactivation should only be required for upstream color-stations. Misdirected handle-tag hybridization may be further minimised by careful handle sequence choice, especially between adjacent color-stations.
[**Acknowledgement:**]{} I thank Lawrence Lee for many fun discussions on DNA origami and Anders Kyrsting for introducing me to lenseless imaging.
[99]{}
‘The Color Run’, Wikipedia, Link: <http://en.wikipedia.org/wiki/The_Color_Run>.
Y.J. Hwang [*et al.*]{}, ‘Use DNA origami as a scaffold for self-assembly of optical metamolecules’, [*2015 CLEO-PR Conference Proceedings*]{}, doi: 10.1109/CLEOPR.2015.7375859 (2015).
J.B. Woehrstein [*et al.*]{}, ‘Sub-100-nm metafluorophores with digitally tunable optical properties self-assembled from DNA’, [*Sci. Adv.*]{} [**3**]{}, e1602128 (2017).
A. Ozcan & E. McLeod, ‘Lensless imaging and sensing’, [*Annu. Rev. Biomed. Eng.*]{} [**18**]{}, 77 (2016).
Y. Sheng & M.T. Bowser, ‘Isolating single stranded DNA using microfluidic dialysis device’, [*Analyst*]{} [**139**]{}, 215 (2014).
D.Y. Zhang & G. Seelig, ‘Dynamic DNA nanotechnology using strand-displacement reactions’, [*Nat. Chem.*]{} [**3**]{}, 103 (2011).
D.V. Nicolau Jr. [*et al.*]{}, ‘Parallel computation with molecular-motor-propelled agents in nanofabricated networks’, [*Proc. Natl. Acad. Sci.*]{} [**113**]{}, 2591 (2016).
H. Takehara [*et al.*]{}, ‘On-chip cell analysis platform: Implementation of contact fluorescence microscopy in microfluidic chips’, [*AIP Adv.*]{} [**7**]{}, 095213 (2017).
J.R. Anderson [*et al.*]{}, ‘Fabrication of topologically complex three-dimensional microfluidic systems in PDMS by rapid prototyping’, [*Anal. Chem.*]{} [**72**]{}, 3158 (2000).
S.F. Wickham, ‘DNA Barrels: Cylindrical NanoPegboards assembled from DNA’, <http://wordvine.sydney.edu.au/files/1699/10207/>
J.B. Knudsen [*et al.*]{}, ‘Routing of individual polymers in designed patterns’, [*Nat. Nano.*]{} [**10**]{}, 892 (2015).
S.S. Agasti [*et al.*]{}, ‘DNA-barcoded labeling probes for highly multiplexed Exchange-PAINT imaging’, [*Chem. Sci.*]{} [**8**]{}, 3080 (2017).
B. Nitzsche [*et al.*]{}, ‘Quantum-dot-assisted characterization of microtubule rotations during cargo transport’, [*Nat. Nano.*]{} [**3**]{}, 552 (2008).
A.J.M. Wollman [*et al.*]{}, ‘Transport and self-organization across different length scales powered by motor proteins and programmed by DNA’, [*Nat. Nano.*]{} [**9**]{}, 44 (2014).
M. Iwaki [*et al.*]{}, ‘A programmable DNA origami nanospring that reveals force-induced adjacent binding of myosin VI heads’, [*Nat. Commun.*]{} [**7**]{}, 13715 (2016).
B. Diederich [*et al.*]{}, ‘cellSTORM – Cost-effective super-resolution on a cellphone using dSTORM’, arXiv: 1804.06244 (2018).
T.D. Huang [*et al.*]{}, ‘A $0.18-\mu$m CMOS array sensor for integrated time-resolved fluorescence detection’, [*IEEE J. Solid-state Circuits*]{} [**44**]{}, 1644 (2009).
L. Hong [*et al.*]{}, ‘Fully integrated fluorescence biosensors on-chip employing multifunctional nanplasmonic optical structures in CMOS’, [*IEEE J. Solid-state Circuits*]{} [**52**]{}, 2388 (2017).
J. Ma [*et al.*]{}, ‘Photon-number-resolving megapixel image sensor at room temperature without avalanche gain’, [*Optica*]{} [**4**]{}, 1474 (2017).
S. Takahashi [*et al.*]{}, ‘A $45$ nm stacked CMOS image sensor process technology for submicron pixel’, [*Sensors*]{} [**17**]{}, 2816 (2017).
G.P. Acuna [*et al.*]{}, ‘Fluorescence enhancement at docking sites of DNA-directed self-assembled nanoantennas’, [*Science*]{} [**338**]{}, 506 (2012).
E.-K. Lee [*et al.*]{}, ‘Design of plasmonic nano-antenna for total internal reflection fluorescence microscopy’, [*Optics Express*]{} [**21**]{}, 023036 (2013).
H. Gro[ß]{} [*et al.*]{}, ‘Parallel mapping of optical near-field interactions by molecular motor-driven quantum dots’, [*Nat. Nano.*]{} [**13**]{}, 691 (2018).
C.C. Glick [*et al.*]{}, ‘Rapid assembly of multilayer microfluidic structures via 3D-printed transfer molding and bonding’, [*Microsys. & Nanoeng.*]{} [**2**]{}, 16063 (2016).
J.J. Keya [*et al.*]{}, ‘High-resolution imaging of a single gliding protofilament of tubulins by HS-AFM’, [*Sci. Rep.*]{} [**7**]{}, 6166 (2017).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The CMS Beam Conditions and Radiation Monitoring System, BRM, will support beam tuning, protect the CMS detector from adverse beam conditions, and measure the accumulated dose close to or inside all sub-detectors. It is composed of different sub-systems measuring either the particle flux near the beam pipe with time resolution between nano- and microseconds or the integrated dose over longer time intervals. This paper presents the Fast Beam Conditions Monitor, BCM1F, which is designed for fast flux monitoring measuring both beam halo and collision products. BCM1F is located inside the CMS pixel detector volume close to the beam-pipe. It uses sCVD diamond sensors and radiation hard front-end electronics, along with an analog optical readout of the signals. The commissioning of the system and its successful operation during the first beams of the LHC are described.'
address:
- 'Brandenburgische Technische Universität, 03046 Cottbus, Germany'
- 'CERN, 1211 Geneva 23, Switzerland'
- 'DESY, 15738 Zeuthen, Germany'
- 'Rutgers University, 08854 Piscataway, NJ, USA'
- 'Université de Genève, 1211 Geneva, Switzerland'
- 'Canterbury University, 8041 Christchurch, New Zealand'
- 'University of Wisconsin, Madison, WI 53706-1481, USA'
author:
- 'A. Bell'
- 'E. Castro'
- 'R. Hall-Wilton'
- 'W. Lange'
- 'W. Lohmann'
- 'A. Macpherson'
- 'M. Ohlerich'
- 'N. Rodriguez'
- 'V. Ryjov'
- 'R.S. Schmidt'
- 'R.L. Stone'
title: Fast Beam Conditions Monitor BCM1F for the CMS Experiment
---
LHC ,CMS ,beam conditions ,sCVD diamonds ,radiation hard sensors
Introduction {#intro}
============
The CMS experiment [@CMS] at LHC [@LHC] will be situated in an unprecedentedly high radiation field. The LHC is designed to run with 362 MJ of stored energy in one beam and with proton intensities of more than 10$^{14}$ per beam. These beams will generate a continuous flux of halo particles near the beam-pipe and when colliding also interaction secondaries, predominantly at small polar angles. The collected dose will be largest for the innermost detectors, which are therefore designed with very high radiation tolerance.
However, also short term losses of the beams may cause serious damage to detector elements, in particular to front-end electronics due to large ionisation. In addition, the innermost detectors need a sufficiently low occupancy level for successful data taking.
To monitor the particle fluxes near the beam-pipe and the radiation level in the sub-detectors, beam conditions and radiation monitors, BRM [@CMS; @BCM1], are installed in the CMS detector. Several slow systems of BRM will be used to measure the accumulated dose near the volume of all sub-detectors. These measurements are necessary to understand potential longer term damage to detector elements.
Particle flux monitors are installed close to the beam-pipe. Two such monitors measure the integral particle flux over half an orbit or over a bunch train using the signal current in polycrystalline small diamond sensors and integrate it over about 40 $\mu$s and 5 $\mu$s, respectively. These fluence measurements will assist beam tuning, indicate critical beam halo conditions for the inner detectors and initiate LHC beam aborts when conditions are such that detectors might be endangered. These systems are described in detail in Ref. [@abell]. Other LHC experiments also installed beam condition monitors using polycrystalline diamond sensors [@other_experiments].
The Fast Beam Conditions Monitor, hereafter referred to as BCM1F, will be sensitive to very fast changes of the beam conditions and provide diagnostics with a time resolution better than the time between bunch crossings, hence, for example, being able to flag problematic beam conditions resulting in bursts of beam loss over very short periods of time. Such beam losses are considered to be one of the principle damage scenarios for CMS detector components. In addition, it will store real-time data to allow post-mortem analyses in the case of beam accidents.
BCM1F System Overview {#over}
=====================
BCM1F uses single-crystal CVD[^1] diamond sensors, hereafter denoted as sCVD, for particle detection. Sensors made of sCVD are sufficiently fast to match the time resolution requirements, and small enough to be inserted into areas close to key detector components without adding substantial material or services. Four sCVD sensors, each with a volume of 5$\times$5$\times$0.5$\mm^3$, are positioned in a plane perpendicular to the beam-pipe on each side of the IP at a distance of about 1.8 m and at a radius of 4.5$\cm$ from the nominal beam position, as sketched in Figure \[fig:CMS\_detector\].
![A sketch of the CMS detector. The positions of the BCM1F in front of the pixel detector and inside the tracker are indicated by the arrows. At both sides of the IP a carbon fibre structure around the beam-pipe, as shown in the upper part, supports the sensor modules.[]{data-label="fig:CMS_detector"}](CMS_BCM_carriages.eps){width="4in"}
The position of the BCM1F is chosen to be optimal in terms of time separation between ingoing and outgoing particles from the IP. Relativistic particles need about 6$\ns$ to move between one of the detector planes and the IP. Hence, gated rate measurements of the BCM1F will allow to separate the fluxes from both beam halo of each direction and interactions products.
Each sensor is connected to a radiation hard preamplifier. Its output signal is transmitted to the counting room over an analog optical link as shown schematically in Figure \[fig:readout\_scheme\].
![ The readout scheme of BCM1F. The signal generated in the sensor is amplified and shaped in a JK16 preamplifier. The analog opto-hybrid drives a laser for analog signal transmission via a single-mode fibre. The signals are digitized and processed in the counting room. []{data-label="fig:readout_scheme"}](BCM1F_Readout_Sketch.eps){width="5.2in"}
Since neither cooling nor slow control equipment are available at the mounting positions, the modules must be operated with low power dissipation and should work over long periods without a re-adjustment of the calibration parameters.
At the back-end of the readout in the counting room signals are digitized and processed in a PC. Flash ADCs, scalers and multi-hit recording TDCs allow e.g. to monitor counts as a function of time over an orbit. Test-pulses are used to check the functionality of the system. Rates, multiplicities, timing and coincidence information are monitored and stored independently of the CMS data acquisition.
sCVD Sensors {#sCVD}
============
Outstanding properties, such as a very low leakage current with negligible temperature dependence, a fast signal response and radiation hardness, make CVD diamond sensors attractive for the locations close to the interaction region. In previous experiments polycrystalline diamond sensors have been successfully used as beam conditions monitors [@CDF; @Babar] by measuring the currents created in the sensor by the crossing particles. However, integration over a certain time limits the time resolution of such devices. In addition, due to crystal defects the charge collection efficiency of polycrystalline CVD sensors is below 50% which may result in a signal-to-noise ratio not sufficient for the detection of minimum ionising particles, MIPs.
Here single crystal CVD diamond sensors are used. They are characterised by nearly 100% charge collection efficiency and allow to count MIPs. They are operated as solid state ionisation chambers by applying high voltage to thin metal plates on both sides of a sensor to create an electric field in the bulk, as shown in Figure \[fig:readout\_scheme\]. Signals from crossing charged particles are created due to the drift of electrons and holes released in the bulk material.
The sCVD sensors are of 5$\times$5$\mm^2$ area and 500 $\mu$m thickness. They have been manufactured by Element Six [@E6] after a few years of development and research in collaboration with the CERN RD42 project [@RD42]. A first application of an sCVD diamond sensor in a collider experiment was described in Ref. [@zeus].
![ The schematic of the back-end readout. []{data-label="fig:backend"}](scheme_backend.eps){width="4in" height="6cm"}
![ The spectrum of signals from relativistic electrons of a $^{90}$Sr source taken with a fully assembled BCM1F module read out with a charge-integrating ADC. []{data-label="fig:flash_spectrum"}](characterisation_fullChain_s-w.eps){width="4in"}
![ The peak value of the signal spectrum as a function of the bias voltage. []{data-label="fig:signal_vs_voltage"}](signalYield_module10_s-w.eps){width="4in"}
![ The size of the analog pulse measured at the opto-receiver module output as a function of the amplitude of the test-pulse fed into the preamplifier. []{data-label="fig:linearity"}](linearity_largerTitle_s-w.eps){width="4in"}
![ Left side: The carbon fibre carrier structure of BCM1 with the four modules installed. Right side: Each module contains the sCVD sensor carrier (top left) and the preamplifier with the laser driver (top right), which are combined to a sandwich (middle) and then connected with cables and optical fibres and protected by an aluminum cage (bottom). []{data-label="fig:full_module"}](BCM1_subcomponents.eps){width="3.5in"}
Sensor Tests {#test}
============
Before installation the sensors were tested in the laboratory. The leakage currents and signal response to electrons from a $^{90}$Sr source were measured for all diamond sensors before assembly. The leakage currents of the sensors are in the range of a few pico-Amperes. The signal amplitude, expressed as the average collected charge, is shown as a function of the bias voltage for both polarities in Figure \[fig:cvd\_characteristics\] for one of the sensors. The signal amplitude increases for increasing bias voltage up to about 120 V, corresponding to a field strength of about 0.25 V$\mu$m$^{-1}$, and is constant thereafter. The measurements for the other sensors show a very similar behavior. In a few cases slight differences in the signal size for different bias voltage polarities are observed. These sensors are operated with the polarity of the bias voltage giving the maximum signal yield.
Two sample sensors were irradiated in a 60$\MeV$ proton beam at PSI up to a fluence of about 3$\times$10$^{14}$ protons per cm$^2$, corresponding to a fluence of 17.5$\times$10$^{14}$ MIPs per $\cm^2$ [@rad-hard_studies]. The signal amplitude obtained from electrons of a $^{90}$Sr source is shown in Figure \[fig:cvd\_characteristics\] for two sensors after irradiation. It drops to about 20% of the one measured with a non-irradiated sensor. Whereas in the non-irradiated sample the signal amplitude saturates already at about 0.25 V$\mu$m$^{-1}$, in the irradiated sample no saturation is seen up to 1 V$\mu$m$^{-1}$. Comparing these results to previous studies [@rad-hard_studies] of the performance of diamond sensors as a function of the fluence of 26$\MeV$ and 24$\GeV$ protons the hypothesis of enhanced damage at lower particle energies is supported. The fluence investigated here is approximately that expected at the location of the BCM1F detector over the baseline LHC program.
Readout Electronics {#FE}
===================
Each sensor is connected to a JK16 radiation hard amplifier ASIC [@FE1]. The chip is fabricated in a commercial 0.25 $\mu$m CMOS technology hardened by appropriate layout techniques.
Each channel comprises a fast trans-impedance preamplifier with an active feedback loop and an amplifier-integrator stage with 20$\ns$ peaking time. An excellent noise performance is achieved by a careful adjustment of the feedback current through the gate voltage of the feedback FET. For a detector capacitance of 5 pF the measured noise amounts to about 700 electrons ENC in agreement with the specifications given in Ref. [@FE1]. The measured charge gain is 20$\mV$/fC.
The analog signals are transmitted to the counting room using an analog optical chain [@optics1] developed for the CMS tracker. The preamplifier’s single-ended output is AC coupled to the custom-designed laser driver ASIC, which modulates the current of the edge-emitting laser diode. Single mode fibres from the pigtailed lasers are connected at the periphery of the tracker volume to an optical fan-in, which merges single fibres into a 12-fiber ribbon cable. In the counting room a corresponding ribbon connects directly to a 12-channel analog optical receiver card in a VME crate.
![ A snapshot of first signals in BCM1F from September 2008. The trigger was taken from the BPTX bunch pickup (yellow line). The blue and the green signals result from an analog sum of the BCM1F signals of each side. The shift of time between the two signals corresponds roughly to the time of flight of a relativistic particle between both sensor planes. []{data-label="fig:pulse_scope"}](bcm1f_scope.eps){width="4in"}
![ Distribution of the difference between arrival times of signals from sensors in different z-planes but equal azimuthal angles. []{data-label="fig:arrivel_time"}](time-lag_samePos_offsetCorrected_fit_s-w.eps){width="3.5in"}
Minor modifications on the laser driver ASIC board were done to allow mounting in two opposite orientations of the laser diode, required by the minimal bending radius of the pigtail fibres. In contrast to the tracker application, for the BCM1F the gain and the laser diode bias current cannot be programmed via the foreseen I$^2$C interface. Hence, attention was paid to choose the input polarity and the laser bias setting to preserve the dynamic range of the receiver side. In addition, the impacts of heat dissipation and the expected radiation dose on the laser diode performance degradation were taken into account. To ensure a small package size a piggy-back architecture was used for interconnecting and mounting the sensor, the amplifier and the analog optical hybrid boards on their carriage.
At the back-end side of the readout the optical signals are converted into electrical signals using an analog opto-receiver module. Its output signals are distributed by analog fan-outs to ADC inputs and to discriminators. A flash ADC performing 500 MS/s with 8 bit resolution, V1721 from CAEN, is used to digitise the signals. This module can be triggered internally or externally. It can read out in full data mode up to 45 consecutive beam orbits or a corresponding number of user definable time intervals. Data is written into a ring buffer and tagged with time stamps. It is read out via an on-board optical link and processed in a PC.
The discriminated signals are counted in all channels with a V260 scaler from CAEN and used for on-line displays of hit rates. In addition, they are digitised with multi-hit capable TDCs V767 from CAEN with 20 bit dynamic range and 0.8 ns-LSB resolution. The TDCs and the scalers are read out via a VME-bridge. They will allow orbit-by-orbit counts to be obtained as a function of time for a detailed monitoring of beam halo and interaction products.
Test-pulses are used to check the functionality of the system during operation. The amplitude of the test-pulse induces a signal similar to one MIP in the preamplifier. A schematic of the complete back-end is shown in Figure \[fig:backend\].
Performance of the System Before Installation {#performance}
=============================================
The assembled front-end modules were tested before installation using a $^{90}$Sr source. Relativistic electrons crossing the sensor trigger a scintillation counter. An example of a spectrum recorded with a charge-integrating ADC is shown in Figure \[fig:flash\_spectrum\]. The distribution of the signal charge shows the expected Landau-shape. The signal is clearly separated from the pedestal peak. A signal-to-noise ratio of about 12 is estimated. The values for the other channels are very similar.
These spectra were acquired for a range of increasing voltage applied across the sensor. Figure \[fig:signal\_vs\_voltage\] shows the most probable values of the pedestal-subtracted pulse height distributions measured as a function of the bias voltage. The maximum signal was reached, as expected, at an electric field of about 0.25 V$\mu$m$^{-1}$, corresponding to a bias voltage of 125 V for a sensor with 500 $\mu$m thickness. Above this voltage the signal is constant.
The linearity of the response of the whole readout chain was investigated using test-pulses fed in a dedicated preamplifier input. The result is shown in Figure \[fig:linearity\]. For both polarities a linear response is found up to test-pulse amplitudes corresponding to approximately 5 MIP equivalents. For test-pulse amplitudes above this value the readout becomes non-linear and approaches saturation at about 10 MIPs.
To test the proper functionality in the cooled tracker environment of CMS, the modules were operated in a climate chamber with five temperature cycles from $-$20 to $+$50 $^\circ$C. Leakage and supply currents as well as the test-pulse response were measured and found to be in the expected range. The stored results of these measurements will be compared with measurements of the same quantities taken during operation of the modules in CMS.
Installation in CMS {#installation}
===================
The components of the modules and the completed modules mounted on the carrier structure are shown in Figure \[fig:full\_module\]. The modules contain in addition to the BCM1F components also sensors from the current monitor BCM1L. Both systems are shielded with a double-cage structure. The inner cage is connected to the ground of the back-end readout. The outer cage is connected to the carbon-fibre support[^2]. The two cages are insulated from each other to mitigate frequency dependent pick-up effects on carbon fibre structures observed elsewhere [@johnson]. The eight BCM1F modules with their corresponding infrastructure were successfully installed and tested at the beginning of August 2008.
First Measurements with LHC Beams {#measurements}
==================================
When first beam circulated in the LHC at the beginning of September 2008, the BCM1F was operational and signals from beam-halo particles were recorded. One of the first beam-generated signals from BCM1F, observed on an oscilloscope, is shown in Figure \[fig:pulse\_scope\].
The readout of the BCM1F modules was triggered by a bunch pickup detector, BPTX [@bptx], indicating that a proton bunch crossed the CMS detector. A time window of about 500$\ns$ with respect to the trigger was recorded by the ADC. Figure \[fig:pulse\_height\] shows, as an example, the spectrum of signals from one of the detector modules. The signal size is obtained by integration of the signal pulse over time. The distributions for the other channels are very similar. The signal-to-noise ratios obtained from these distributions are also shown in Figure \[fig:pulse\_height\] for all channels[^3]. The values of the signal-to-noise ratio vary between 15 and 25 and are slightly better than the ones obtained in laboratory measurements before installation. Since the LHC was filled with beam in one direction only, the beam halo particles should follow this direction. To demonstrate the capabilities of BCM1F the signal arrivals times in the flash ADC are measured with respect to the BPTX trigger signal. The distribution of the difference between arrival times of signals from sensors in different z-planes but equal azimuthal positions is shown in Figure \[fig:arrivel\_time\].
From a Gaussian fit, a value of 12.4$\ns$ is obtained for the time difference, corresponding precisely to the expected time-of-flight of a relativistic particle between the two BCM1F planes on the $+$z and $-$z side. The variance of the Gaussian amounts to 1.8$\ns$, leading to an estimate of the single hit timing resolution of 1.3 ns.
Conclusions
============
The BCM1F is a fully functional sub-detector of the BRM system of CMS and will be vital for monitoring beam conditions close to the beam-pipe inside CMS. It comprises 8 modules each containing an sCVD diamond sensor, a front-end ASIC and an optical analog signal transmission to scalers, flash ADCs and TDCs. The system is operated independently from the other CMS sub-detectors.
System tests of each module in the laboratory show that performance matches requirements.
Samples of sCVD sensors were exposed to a high intensity proton beam up to fluences expected at the location of BCM1F for the nominal LHC running. The signal amplitude measured for MIPs is reduced to 20% of the original one, approaching a level critically low for MIP counting. Using the ADC the size of the MIP amplitude will be monitored as a function of the LHC operation time allowing us to replace the sensors if necessary.
BCM1F was successfully installed and was operational when LHC was filled with first beam. Data taken with beam show a slightely better signal-to-noise as reached in the laboratory tests. A measurement of the signal arrival times using a flash ADC indicates a promising single hit timing resolution of about 1.3$\ns$. This will allow the separation of incoming halo particles correlated to a certain bunch from interaction products created at the IP.
Different readout-modes and time windows for data capture are programmable. Data can be written on a local disk and published to the CMS readout system. Local data preprocessing will deliver the shift-crew a detailed picture on the beam-halo count rates as a function of the time. Data of several orbits stored in circular memory will allow diagnostics just after a beam abort.
The BCM1F detector is ready for data taking in the commissioning phase of the LHC.
Acknowledgments
===============
We thank J.P. Chatelain for his contributions to design and manufacture the mechanical structures. We would like to thank our colleagues in the BRM group for their advice and assistance. We are grateful to CMS Technical Coordination for assistance in launching and sustaining the project, particularly in its early phases. We also acknowledge the contribution of the CERN, CMS and DESY technical teams which helped bring it to a successful conclusion. We express our gratitude to PSI for allowing us to use the proton beam and in particular to K. Deiters for his collaboration there. The Rutgers University group is grateful for the support by NSF. R. Hall-Wilton is grateful for the support of the Israeli Technical Associates Program.
[00]{}
The CMS Collaboration, S. Chatrchyan et al. ,“The CMS experiment at the CERN LHC”, 2008 JINST 3S08004. L. Evans and P. Bryant (editors), “LHC Machine”, 2008 JINST 3S08001. L. Fernandez-Hernando et al., “Development of a CVD diamond Beam Condition Monitor for CMS at the Large Hadron Collider”, Nucl. Inst. and Meth. A552 (2005) 183, A. Macpherson, “Beam Condition Monitoring and radiation damage concerns of the experiment”, Proceedings LHC Project Workshop, Chamonix XV (2006) 198, D. Chong et al., “Validation of synthetic diamond for a Beam Condition Monitor for the Compact Muon Solenoid Experiment”, IEEE Trans. Nucl. Sci. 54 (2007) 182. A. J. Bell on the behalf of the BRM group, “Beam and Radiation Monitoring for CMS”, 2008 IEEE Nuclear Science Symposium Conference Record, pp 2322-2325, http://ieeexplore.ieee.org. V. Cidro et al., “The ATLAS Beam Conditions Monitor”, 2008 JINST 3P02004, The LHCb Collaboration, A. Augusto Alves Jr et al., “The LHCb Detector at the LHC”, 2008 JINST 3S08005. P. Dong et al., “Beam Condition Monitoring With Diamonds at CDF” IEEE Trans. Nucl. Sci. 55 (2008) 328. M. Brunisma et al., “CVD diamonds in the BaBar radiation monitoring system”, Nucl. Phys. B150 (2006) 164. Element Six Ltd., King’s Ride Park, Ascot, Berkshire SL5 8BP, UK. RD42 Status Report, http://cdsweb.cern.ch/record/1098155. W. Lange et al., “INVESTIGATION OF A SINGLE CRYSTAL DIAMOND SENSOR AND ITS APPLICATION IN BACKGROUND MEASUREMENTS FOR HERA”, talk given at the NoRHDia Workshop 2006, GSI Darmstadt, http://www-norhdia.gsi.de/index.php?mgid=6\#3rd.
RD-42 Collaboration, CERN, “Development of Diamond Tracking Detectors for High Luminosity Experiments at the LHC”, CERN-LHCC-2006-10 (2006), W. de Boer et al., “Radiation Hardness of Diamond and Silicon Sensors Compared”, Physica Status Solidi A204 (2007) 3004. J. Kaplan and W. Dabrowski, “Fast CMOS Binary Front End for Silicon Strip Detectors at LHC Experiments”, IEEE Trans. Nucl. Sci. 52 (2005) 2713. J. Troska et al., “Optical readout and control systems for the CMS Tracker”, IEEE Trans. Nucl. Sci. 50 (2003) 1067, G. Cervelli et al., “A radiation tolerant linear laser driver array for optical transmission in the LHC experiments”, Proc. 7th Workshop on Electronics for LHC Experiments, CERN/LHCC/2001-034 (2001) pp 155-159. M. Johnson, “Grounding and Shielding Techniques for Large Scale Experiments”, Proceedings of the 8th Workshop for Electronics for the LHC (2002), B. Quinn, “Carbon Fibre Grounding Design in the D0 Run IIb Silicon Detector Design”, Nucl. Inst. and Meth. A511 (2003) 180.
C. Ohm, “Phase and Intensity Monitoring of the Particle Beams at the ATLAS Experiment”, Linköping University, The Department of Physics, Chemistry and Biology, LITH-IFM-EX-07/1808-SE (2007), C. Ohm and T. Pauly, arxiv.org/pdf/0905.3648 (2009), T. Aumeyr, “Beam Phase and Intensity Monitoring for the Compact Muon Solenoid Experiment”, Vienna University of Technology (2008).
[^1]: Chemical Vapor Deposition
[^2]: The carbon fiber support is not connected to the CMS detector ground.
[^3]: Channel one had a faulty cable at the ADC input at the time of this measurement.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We characterise a class of SU(2) gluonic field configurations in the modified axial gauge where a zero mode component vanishes at some space point but the global Haar measure remains non-zero. The consequence of this is that gluonic wavefunctionals need not vanish at the boundary of the fundamental modular domain, which itself permits $\theta$ dependence in QCD(3+1).'
---
=6truein =21truecm =-.4truein =-.3truein
FAU-TP3-96/4\
hep-th/9604099\
April 17, 1996.\
Field Configurations and the SU(2) Haar Measure in QCD\
Harald W. Grie[ß]{}hammer [^1] and Alex C. Kalloniatis [^2]\
Institut für Theoretische Physik III\
Universität Erlangen - Nürnberg\
Staudtstra[ß]{}e 7\
D-91058 Erlangen, Germany
It has been known for some time that quantum chromodynamics in four dimensions must involve a hidden parameter known variously as the $\theta$-parameter or the QCD vacuum angle [@JaR76]. The existence of this angle, analogous to the Bloch momentum in a periodic potential, is related intimately to the presence of large (topologically non-trivial) gauge symmetries in the theory. In the Hamiltonian formalism, $\theta$ dependence creeps into the theory due to the quasiperiodicity (periodic up to an arbitrary phase) of the wavefunctional at the boundary of the so-called Fundamental Modular Domain, the region of gauge configuration space inside which unique representatives of gauge orbits have been identified. We shall further clarify this below.
The massive Schwinger model is the simplest field theory in which $\theta$ dependence plays a role in the mesonic spectrum [@CJS75]. However, QCD(1+1), the next step in non-triviality, has no $\theta$ dependence [@Wit79]. In the above terms, this can be explained as follows: the fundamental modular domain and a related region in gauge configuration space, the Gribov region, happen to be identical. At the boundary of the Gribov region, the Gribov [*horizon*]{}, wavefunctionals turn out to [*vanish*]{}. Thus wavefunctionals must vanish at the boundary of the fundamental modular domain, and there is no room left for $\theta$ to appear.
To proceed to higher dimensions now, we first recall why wavefunctionals happen to vanish at the Gribov horizon. The space of gauge configurations in non-Abelian theories is in general ‘curvilinear’, which leads to a non-trivial Jacobian, a global operator, in the course of the elimination of redundant gauge field variables. This Jacobian can either sit in the kinetic term of the gauge fields in the Hamiltonian, or be absorbed into the wavefunctional – much like the change to the radial basis in the hydrogen atom (we are evidently working in Schrödinger representation). The Gribov horizon is precisely the sub-manifold in gauge space where the Jacobian first vanishes and hence, in the ‘radial’ basis, the wavefunctional vanishes on the Gribov horizon. Now in 3+1 dimensions, one has a choice of gauges to attempt to work in. The traditional Coulomb gauge is problematic because it is impossible to perform the gauge fixing in closed form and has hindered explicit construction of the fundamental modular domain. Nonetheless, one can show [@Coul] that the fundamental modular domain is a proper subset of the Gribov region, though points on the boundary of the domain may also lie on the Gribov horizon. This shows that $\theta$ dependence is possible. In the static temporal gauge [@JLP91] and modified axial gauge [@Yab89; @LNT94], where the Jacobian is the gauge group invariant Haar measure and [*can*]{} be written down although in an UV unregulated state, one encounters a product over space points of, what we shall call, a [*local Jacobian*]{}. Now a problem ensues: if the local Jacobian vanishes for any one space point, it would seem that the entire Jacobian must also vanish. If this also happens to occur at the boundary of the fundamental modular domain, which is not unlikely as will be clarified below, then $\theta$ dependence is absent in QCD(3+1) in this gauge, a result in contradiction to conventional wisdom as well as phenomenological evidence [@Cre77]. In this letter, we show that this sequence of logic need not hold.
In order to make things more concrete we work in the Hamiltonian formulation on the three-torus of periodicity $L$ in the modified axial gauge of [@LNT94]. Our discussion can be easily translated to the static temporal gauge approach of [@JLP91]. We moreover restrict ourselves to the case of SU(2) pure gauge theory, though the generalisation of what follows is straightforward. Thus the gauge is, in the first step, $\partial_3 A^a_3 = 0, a = 1,\dots,3$. This is followed by an additional colour rotation such that the surviving zero mode of $A_3$ is diagonal in colour space.
There are now two ways to express the Haar measure in functional space: $${\cal J}[a_3] = \prod_{x_\perp}^\infty \sin^2(\pi a_3(x_\perp))
\label{jacob1}$$ or $${\cal J}[a_3] = \exp[\delta^{(2)}(0)
\int d^2x_\perp \ln \sin^2( \pi a_3(x_\perp))]
\label{jacob2}$$ where $a_3(x_\perp)$ is the dimensionless combination $gL\sqrt{A^a_3(x_\perp) A^a_3(x_\perp)} /\pi$. It is thus dependent only on time (suppressed in the Hamiltonian formalism) and the two-dimensional transverse coordinates $x_\perp$, and is a scalar in colour SU(2) space. The first expression (\[jacob1\]) allows us to clarify our earlier comments: we observe that the left hand side is a global operator while the right hand side involves the ‘local Jacobian’ $\sin^2(\pi a_3(x_\perp))$, namely the Jacobian of gauge space at each space point separately. The argument of this functional vanishes whenever dynamics lead the zero mode $a_3(x_\perp)$ itself to be an integer. However, the wavefunctional $\Psi[a_3]$ should satisfy the boundary condition ‘$\Psi[0] = \exp(- i \theta) \Psi[1]$’ (we suppress for simplicity the dependence on other variables which do not affect this line of argument). As in the case of the radial variable in the hydrogen atom, the standard kinetic energy term $-\delta^2/2\delta a_3^2$ is retrieved in the function space restricted by the condition that $\Psi = 0$ when ${\cal J} = 0$. One can now verify the following statements: (a) in the above gauge, the fundamental modular domain is the restriction of the spatially constant part of $a_3$, $\int d^2x_\perp a_3/L^2$, to the interval $[0,1]$, (b) residual gauge transformations map the boundary point $0$ into the boundary point $1$, and (c) for each configuration at the boundary of the fundamental modular domain, $a_3 \in \ZZ$, the local Jacobian must vanish for at least two space points $x_\perp$. We can now state the apparent problem precisely: if (i) $a_3 \in \ZZ$ at even one space point $x^{(0)}_\perp$, then because of the product structure in equation (\[jacob1\]) it seems to follow that (ii) ${\cal J} = 0$ globally, thus (iii) $\Psi[0] = \Psi[1] = 0$ and no further role can be played by $\theta$ in the theory. We shall now construct counterexamples to the suggestion that (ii) should follow from (i) in higher dimensions.
Observe that equations (\[jacob1\]) and (\[jacob2\]) are ill-defined expressions due to ultraviolet divergences. They actually exhibit a competition between ultralocality (seen most evidently in equation (\[jacob1\])) and non-locality. The question of choice of regulator is delicate. We shall avoid the lattice framework, since in a finite lattice the entire subtlety is thrown into the taking of the continuum limit. We also avoid dimensional regularisation it usually being only perturbatively valid. We are thus faced with the actual problem: a choice of an alternative which could respect symmetries, especially gauge symmetry. For lack of such an alternative, let us simply argue what should happen in an appropriate continuum regularisation scheme. Firstly, the form of equations (\[jacob1\]) and (\[jacob2\]) is dictated by gauge symmetry, so that in a scheme respecting this symmetry, these expressions cannot change in structure. Secondly, in a nonperturbative treatment the regulated form of the otherwise singular delta function in equation (\[jacob2\]) should bring in some new scale which should be related to $\Lambda_{\rm{QCD}}$. This is essentially the argument of Johnson et al.[@JLP91]. We can explain this somewhat further by brief recourse to dimensional regularisation: there, $\delta^{(2)}(0)$ is simply the tadpole integral in momentum space. In perturbation theory this can be set to zero because of the [*absence*]{} of any intrinsic scale in this regime. Nonperturbatively, as mentioned, this is no longer the case and the integral would be related to the appropriate scale. We assume the same holds for any other nonperturbative gauge invariant regularisation scheme. We use for the moment the second expression (\[jacob2\]), and interpret the delta function as $\delta^{(2)} (0) = 1/\kappa^2$, with $\kappa$ some fundamental length scale in the theory (similar to $a$ in [@JLP91]). This scale would be important in a full dynamical calculation, but in the following it plays no further role.
Evidently $\Psi$ is nonzero as long as ${\cal J} \neq 0$, or, using form (\[jacob2\]), when $$\int d^2 x_\perp \ln \sin^2(\pi a_3(x_\perp))\;>\;-\infty\;\;.
\label{invest}$$ Now, one notes that the Lesbegue integral in (\[invest\]) is insensitive to the exclusion of a countable number of points or lines on which the integrand diverges. However, if $a_3(x_\perp) \in \ZZ$ for $x_\perp$ lying in a two-dimensional sub-[*manifold*]{} of the torus then the integrand over these points is infinite: for such configurations the Jacobian ${\cal J}$ really is zero, namely the wavefunction vanishes. The configurations have zero probability and play no role in the theory. In particular no $\theta$ dependence can be associated with them. What is implicit in the above is that we are thinking of wavefunctions which are essentially delta functionals peaked about particular classical gauge field congurations, $\Psi[a_3] \sim \delta[ a_3 - a_3^{\rm{class}}]$. Our task is to determine examples of configurations $a_3^{\rm{class}}$ such that $\theta$ dependence can occur.
We thus look for configurations $a_3(x^{(0)}_\perp) \in \ZZ$ for only one isolated point $x^{(0)}_\perp$ on the transversal plane. The neighbourhood of $x^{(0)}_\perp$ can be parametrised by polar coordinates $(r,\varphi)$, and (\[invest\]) becomes $$\lim\limits_{\varepsilon\to 0}\int\limits_{\varepsilon} dr\int d\varphi
\;r \ln\sin^2(\pi a_3(r,\varphi)) \;>\;-\infty\;\;.
\label{limeps}$$ For a finite integral, the integrand must behave like $r^{\alpha}$ with $\alpha>-1$, $$r \ln\sin^2\pi a_3(r,\varphi) ={\cal A}(\varphi)\; r^{\alpha} +
\mbox{ ... }\;\;,\;\;\alpha > -1\;\;\mbox{ as }r\to 0\;\; .$$ Solving for the field $a_3$ yields $$\pi a_3(r, \varphi)=n\pi \pm\arcsin\exp\Big[\frac{{\cal A}(\varphi)}{2}\;
r^{\alpha-1}\Big]\;\;\mbox{ as }r\to 0\;\;.
\label{twoperdim}$$ Since $a_3(x^{(0)}_\perp)$ must be an integer, $$- 1<\alpha<1\;\;\mbox{ and }\;\;{\cal A}(\varphi)<0\;\;.$$ The same calculation gives for a line singularity whose neighbourhood is now parametrised by curvilinear coordinates, namely the distance $r$ from the line and $s$, parametrising the line itself, $$\pi a_3(r, s)=n\pi \pm\arcsin\exp\Big[\frac{{\cal A}(s)}{2}\;
r^{\alpha}\Big]\;\;,\;\;-1<\alpha<0\;,\,\;\;{\cal A}(s)<0\;\;\mbox{ as }r\to 0.
\label{oneperpdim}$$ Noting that with the integrand in (\[invest\]), also the integral cannot become positive, the extension to several singularities is easily accomplished by appropriately patching together solutions of the types (\[twoperdim\]) and (\[oneperpdim\]). Since singularities cannot cancel, the integral about each singularity has to obey (\[limeps\]). We note that the solution for point singularities approaches integer values faster than (\[oneperpdim\]), allowing for a larger range of exponents $\alpha$. Neither (\[twoperdim\]) nor (\[oneperpdim\]) is analytic at $r=0$. Finally, these expressions can be generalised to arbitrary dimensions, for example if an appropriate nonperturbative generalisation of dimensional regularisation can be validated. In $(d+1)+1$ dimensions, where $d$ is the dimensionality of the hypertorus on which $x_\perp$ lives, we take a coordinate system with $r$ the distance of a point from the hypersurface of singularities of dimension $N \leq d$, and with $(\varphi_1,\dots,\varphi_d)$ the other coordinates. Then the required field configuration is $$\begin{aligned}
\pi a_3(r, \varphi_1,\dots,\varphi_d )=n\pi \pm\arcsin\exp
\Big[\frac{{\cal A}(\varphi_1,\dots,\varphi_d)}{2}\;
r^{\alpha-(d-N)}\Big]
\;\;, \nonumber \\ \;\;-1 < \alpha < d-N
\;;\,\;\;{\cal A}(\varphi_1,\dots,\varphi_d)<0\;\;\mbox{ as }r\to 0
\;.
\label{manyperpdims}\end{aligned}$$
Now for each choice of $n$ for the configurations (\[twoperdim\]), (\[oneperpdim\]) and (\[manyperpdims\]), there exist under certain conditions [@Gri96] configurations with $n$ replaced by $n+1$ corresponding to a transformation of winding number one. Moreover, each of these classical configurations lie precisely on the boundary of the fundamental modular domain which itself corresponds to the completely space-independent part of $a_3$ taking the value zero or one. Thus, the quantum wavefunctionals corresponding to these configurations must be identified up to an arbitrary phase and this is where $\theta$ makes its appearance.
Our construction relies on the existence of an infinite number of degrees of freedom. Equation (\[jacob1\]) shows that in any [*quantum mechanical*]{} framework ‘tunneling through the Jacobian’ cannot occur. We stress moreover that (\[twoperdim\]), (\[oneperpdim\]) and (\[manyperpdims\]) represent an infinitely large [*family*]{} of gauge configurations parametrised by the functions ${\cal A}$ and exponents $\alpha$. They are not a ‘set of measure zero’. Thus indeed, with these elements alone a rich class of gauge configurations can be seen to exist. We do not claim that this is even an exhaustive classification of such configurations.
In order to proceed with dynamical calculations, the next step is to construct corresponding finite energy classical configurations of transverse gluon fields. This is delicate due to structures in the Hamiltonian of [@LNT94] involving the mode $a_3$ in the denominator and the transverse part of the Gau[ß]{} operator in the numerator, itself depending on the transverse gauge fields. Thus, finite energy contributions can only come from those configurations whose corresponding transverse gluon parts can balance the centrifugal energy of the zero mode. This work is still underway. The main goal in this direction would be to use these classical configurations finally as the basis of a WKB calculation in order to solve the Hamiltonian.
[30]{} R. Jackiw, C. Rebbi, [*Phys.Rev.Lett.*]{} [**37**]{} (1976) 172; C. Callan, R. Dashen, D.J. Gross, [*Phys.Lett.*]{} [**B63**]{} (1976) 334. S. Coleman, R. Jackiw, L. Susskind, [*Ann.Phys.(N.Y.)*]{} [**93**]{} (1975) 267. E. Witten, [*Nuov.Cim.*]{} [**51A**]{} (1979) 325. M.A. Semenov-Tyan-Shanskii, V.A. Franke, Zapiski Nauchnykh Seminarov Leningradskogo Otdeleniya Matematicheskogo Instituta im V.A. Stekhov AN SSR, 120 (1982) 159 \[Translation: (Plenum, New York, 1986) p.1999\]; G. Dell’Antonio, D. Zwanziger, [*Comm.Math.Phys.*]{} [**138**]{} (1991) 291; P. van Baal, [*Nucl.Phys.*]{} [**B369**]{} (1992) 259. K. Johnson, L. Lellouch, J. Polonyi, [*Nucl.Phys.*]{} [**B367**]{} (1991) 675. H. Yabuki, [*Phys.Lett.*]{} [**B231**]{} (1989) 271. F. Lenz, H.W.L. Naus, M. Thies, [*Ann.Phys.(N.Y.)*]{} [**233**]{} (1994) 317. R.J. Crewther, [*Phys.Lett.*]{} [**B70**]{} (1977) 349. H.W. Grie[ß]{}hammer, Ph.D. thesis, Universität Erlangen-Nürnberg, 1996.
[^1]: [email protected]
[^2]: [email protected]
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Any constructive continuous function must have a gradually varied approximation in compact space. However, the refinement of domain for $\sigma-$-net might be very small. Keeping the original discretization (square or triangulation), can we get some interesting properties related to gradual variation? In this note, we try to prove that many harmonic functions are gradually varied or near gradually varied; this means that the value of the center point differs from that of its neighbor at most by 2. It is obvious that most of the gradually varied functions are not harmonic. This note discusses some of the basic harmonic functions in relation to gradually varied functions.'
address: |
$^{1}$ Department of Computer Science & IT, University of the District of Columbia\
$^{2}$ Freddie Mac\
$^{3}$ Department of Mathematics, Rutgers University\
$^{4}$ Center for Discrete Mathematics and Theoretical Computer Science (DIMACS),\
Rutgers University\
`[email protected]`; `[email protected]`; `[email protected]`
author:
- 'Li Chen$^{1,4}$, Yong Liu$^{2,1}$, and Feng Luo$^{3,4}$'
title: A Note on Gradually Varied Functions and Harmonic Functions
---
Introduction
============
In this note, we will discuss some interesting facts about gradually varied functions (GVF) and harmonic functions. The compatibility between gradually varied functions and harmonic functions is important to the applications of gradually varied functions in real world engineering problems.
Any constructive continuous function must have a gradually varied approximation in compact space [@Che2005]. However, the refinement of domain for $\sigma-$-net might be very small. Keeping the original discretization (square or triangulation), we can obtain some interesting properties related to gradual variation. In this note, we try to prove that many harmonic functions are gradually varied or near gradually varied, meaning that the value of the center point differs from that of its neighbor by at most 2. It is obvious that most of the gradually varied functions are not harmonic. This note discusses some of the basic harmonic functions in relation to gradually varied functions.
Let $A_1, A_2, ..., A_n$ be rational numbers and $A_1< A_2< ...< A_n$. Let $D$ be a graph. $f: D\rightarrow \{A_1,...,A_n\}$ is said to be gradually varied if for any adjacent pair $p,q$ in $D$ and $f(p)=A_{i}$, then $f(q) = A_{i-1}$, $A_{i}$, or $A_{i+1}$. We usually let $A_i=i$.
Extending the concept of gradual variation to the function in continuous space: $f: D\rightarrow R$ is gradually varied if $|p-q|\le 1$ then $|f_q-f_p|\le 1$. Or
$$|f_q-f_p| \le |p-q|.$$
To some extent, gradual variation is the same as the locally Lipschitz condition. (However, $A_{i}$ may be defined differently.)
On the other hand, a harmonic function satisfies:
$$\frac {\partial ^{2} f }{\partial x ^2} + \frac {\partial ^{2} f }{\partial y ^2} =0$$
A main property of the harmonic function is that for a point $p$, $f(p)$ equals the average value of all surrounding points of $p$.
If $f$ is harmonic and $p, q$ are two points such that $f(p)<f(q)$ and $s$ is a path (curve) from $p$ to $q$, then we. We know
$$f_q-f_p = \int_{p,q} \nabla f\cdot d {\bf s}$$
If $s$ is a projection of a geodesic curve on $f$, does the gradient $\nabla f$ maintain some of its properties? For example, is it a constant or does it have any property relating to gradual variation?
What we would like to prove is that if we define
$$f_{mg} (p,q) =\max \{ |\nabla(f)|\} \mbox {on curve $s$ or entire $D$ }$$
should we have
[**Observation A:**]{} $f_{mg}(p,q)< 2\cdot |(f_q-f_p)|/ length(s)$ when $f$ is harmonic?
Therefore, our purpose is to show that many basic harmonic solutions are at least “near” GVF solutions.
Harmonic Functions with Gradual Variation
==========================================
Given the value of a set of points in domain $D$, $f: J\rightarrow R$, $J\subset D$, for 4-adjacency in 2D (grid space), using an interpolating process, we can obtain a GVF solution.[@Che2005] We also can solve a linear equation using a fast algorithm for a sparse matrix of the harmonic equation based on
$$f_{i,j} = \frac{1}{4} (f_{i-1,j}+f_{i+1,j} + f_{i,j-1}+f_{i,j+1})$$
or give an initial value for $f$ and then do an iteration. This formula gives a fast solution and also gives a definition of discrete harmonic functions [@Lov].
How we use the GVF algorithm to guarantee a near harmonic solution is a problem. We can use the divide-and-conquer method to have an $O(n log n)$ algorithm and then iterate it a few times to get a harmonic solution.
Assume $b_1$ and $b_2$ are two points in boundary $J$. $f(b_1) < f(b_2)$ and $s(b_1,b_2)$ is a path from $b_1$ to $b_2$. So
$$\frac{(f(b_2)-f(b_1))}{length(s(b_1,b_2))}$$
is the average slope of the curve. We can define $$slope (b_1,b_2) =\max \{ \frac{(f(b_2)-f(b_1))}{length(s(b_1,b_2))} | s(b_1,b_2) \mbox{ is a path}\}$$
Therefore, there is a $s(b_1,b_2)$ whose length reaches the minimum. Such a path will be a geodesic curve.
With the consideration of the maximum “slope”, the reason for $Observation A$ is $$|\nabla f|\le (\frac{\partial f}{ \partial x}^2 + \frac{\partial f}{ \partial y}^2)^{1/2} \le ? 2\cdot slope(b1,b2) \le 2.$$
In general, $$|\nabla f|\le (\frac{\partial f}{ \partial x}^k + \frac{\partial f}{ \partial y}^k)^{1/k} \le ? 2\cdot slope(b1,b2) \le 2.$$
where $k>0$. Since $slope\le 1$ based on the condition of gradual variation, we want to show that the harmonic solution is nearly gradually varied. Note that the gradual variation condition is similar to the Lipschitz condition.
There are two reasons for using “2” in the above formula as the ratio: (1) It is not possible to use “1,” and (2) anything less than 2 is almost gradual variation.
There are simple cases in discrete space that the harmonic solution reaches difference 1.5.
[*Proof*]{} Assume that we have five points in grid space in direct adjacency: $(i,j), (i-1,j), (i+1,j) , (i,j-1) , (i,j+1)$ and $f_{i-1,j}= 1$, $f_{i+1,j} = f_{i,j-1}=f_{i,j+1}=3$
We want to know what $f(i,j)$ equals. Using the GVF, we get $f(i,j)=2$ by Definition 1.1. See Fig. 3.1.
Using harmonic functions, we will have $f(i,j)=2.5$ by Definition 2.1. With the same principle, we can let $f_{i-1,j}= 3$ and $f_{i+1,j} = f_{i,j-1}=f_{i,j+1}=1$. So $f(i,j)=1.5$ for the harmonic solution and $f(i,j)=2$ still for gradually varied.\
When we use the harmonic solution to approximate gradual variation, we need to see if we can find the best value when choosing from two possible values. A simple algorithm may be needed to make this decision.
[**Observation B:**]{} There is a GVF that is almost harmonic: $|center-average Of Neighbor|<1$ or $|center-average Of Neighbor|< c$, $c$ is a constant.
The above examples show that a perfect GVF is not possible for a harmonic solution. The gradient (maximum directional derivative) less than $2\cdot f'_{m}$, $f'_{m}$ denotes the maximum average change (slope) of any path between two points on the boundary possessing the mean of gradual variation.
Every linear function is harmonic. And for quadratic functions, we have
$f(x,y)= a x^2+b y^2 + c xy $
is harmonic if and only if $a=-b$. However, the following example will not meet the case.
[**Example 1**]{} Three vertices of a triangle are $p1=(0,0)$, $p2=(9,0)$, $p3=(-8,4)$. The linear function $f(x,y)= x+3y$.
This triangle satisfies the gradually varied conditions:
$$|f(p1)-f(p2)|=9 \le |p1-p2|=9$$
$$|f(p2)-f(p3)|=|9-4| \le |p1-p2|$$
$$|f(p1)-f(p3)|=|0-4| \le \sqrt{8^2+4^2}$$
If we consider a point $p=(x,y)$ on the line $<p_2, p_3>$ when $x = 0$ and $y=36/17=2.4$, then $f(x,y) >7$. This point and $p_1$ do not maintain the condition of gradual variation. $|f(p)-f(p_1)|>7 > |p-p_1|$.\
This example seems to break the observation we have made. However, let us revisit the function $f(x,y)= x+3y$ and let $z=f(x,y)$. We have $z-x-3y=0$. We can have $y= \frac{1}{3} z- \frac{1}{3} x$ represent the triangle and associated function. In general, for a linear function in 3D
$$a x + b y + c z +d =0$$
We can always find a coefficient that has the maximum absolute value. We will have the equivalent equation that has
$$A X + B Y + D = Z$$
where $|A|$ and $|B|\le 1$. This property is often used in computer graphics.\
The Piece-wise linear function preserves the property of gradual variation.
[*Proof:*]{} We first want to discuss the case of a single triangle where any piecewise linear function is a harmonic function. In this case we can write the function like this
$$f(x,y) = a x +b y +c , \mbox{\hskip 0.1in} |a|,|b|\le 1$$
$\frac{\partial f}{\partial x} = f_x =a$ , $\frac{\partial f}{\partial y} = f_y =b$. The gradient is a constant $\sqrt{ a^2+b^2}$. There is a horizontal and vertical line that goes through boundary points. The maximum average rate of change $r$ (average slope on the path between two points on the boundary) is greater than or equal to $\max{a,b}$.
Since $|a|,|b|\le 1$ ; $r \le \sqrt {a^2+b^2}\le \sqrt{2} \max{a,b} \le \sqrt{2}$. So $r < 2$.
If this piecewise linear function is on a polygon (2D), it will still have this property.\
The problem is that in this proof, we have not used the conditions of gradual variation directly. The conditions are
$$|f(p_1)-f(p_2)|=| a (x1-x2) +b (y1-y2)| \le |p_1 -p_2| =\sqrt {(x1-x2)^2+(y1-y2)^2}$$
$$|f(p_1)-f(p_3)|= |a (x1-x3) +b (y1-y3)| \le |p_1 -p_3| = \sqrt {(x1-x3)^2+(y1-y3)^2}$$
$$|f(p_2)-f(p_3)|= |a (x2-x3) +b (y2-y3)| \le |p_3 -p_2| = \sqrt {(x2-x3)^2+(y2-y3)^2}$$
We have used the GVF general property and the triangle constraint. The next section will discuss a more general case.\
Gradually Varied Semi-Preserving
=================================
In this section, we extend the content of above sections using more rigorous mathematical definitions. Harmonic functions can be characterized by the mean value theorem. Here we are interested in harmonic functions that are gradually varied. More specifically, a function is said to be gradually varied semi-preserving if
$$\max_{D} |\nabla u| \le c \cdot \max_{p,q\in \partial D} \frac{|u(p)-u(q)|}{|p-q|}$$
where $\nabla u$ is the gradient of $u$, $D$ is a domain with the boundary $\partial D$, and $c$ is a constant.
The above formula poses a property of computational importance. We can show that linear functions and quadratic hyperbolic functions satisfy the condition of gradually varied semi-preserving.
If $u$ is linear we can assume that $u=ax+by+c$ and if $u$ is quadratic hyperbolic we can let $u =a (x^2 - y^2)$. We do not restrict the value of $a,b,c$ here.
[**Proposition 2**]{} If $u$ is linear or quadratic hyperbolic, then $$\max_{B} |\nabla u| \le \sqrt(2) \cdot \max_{p,q\in \partial B} \frac{|u(p)-u(q)|}{|p-q|}$$ where $B$ is any ball.
[*Proof*]{} Let $u$ be a linear function $u=ax+by+c$ then $$|\nabla u| = \sqrt(a^2+b^2) \le \sqrt(2) \max \{|a|,|b|\}$$
On the other hand, if we choose $p=(-r,0), q=(r,0)$ on $\partial B$,
where $r$ is the radius of the ball $B$. Then $$\frac{|u(p)-u(q)|}{|p-q|}=\frac{|-ar-ar|}{2r} = |a|$$ Choosing another pair of $p$ and $q$ on $\partial B$,
$$p=(0,r), q=(0,-r)$$
we have $$\frac{|u(p)-u(q)|}{|p-q|}= |b|$$
Combining (13), (14) and (15) we conclude (12) when $u$ is linear.
Now, consider $u$ as a quadratic hyperbolic function: $u =a (x^2 - y^2)$. Then,
$$|\nabla u| = 2 |a| \sqrt(x^2+y^2) \le 2 |a| r$$
On the other hand, if we choose $p$ and $q$ on $\partial B$,
$$p=(0,r), q=(r,0)$$
Then
$$\frac{|u(p)-u(q)|}{|p-q|}=\frac{|-ar^2-ar^2|}{\sqrt(2r^2)} = 2 |a| r$$
Combining (16) and (17) we have
$$|\nabla u| \le \sqrt(2)\frac{|u(p)-u(q)|}{|p-q|} \le \sqrt(2)\max_{x,y\in \partial B}\frac{|u(x)-u(y)|}{|x-y|}$$
and (12) follows.\
Discussion
==========
Recent studies show an increased interest in connecting discrete mathematics with continuous mathematics, especially in geometric problems. For instance, the variational principle has been used for triangulated surfaces in discrete differential geometry, see [@Luo]. This note presented an idea of combining a type of discrete function: the gradually varied function and a type of continuous function: the harmonic function in a relatively deep way in terms of continuous mathematics. The harmonic function is a weak solution to the Dirichet problem which is about how to find a surface when the boundary curve is given. The gradually varied function was proposed to solve a filling problem in computer vision. We are hesitant to use the method of the Dirichlet problem for the discrete filling problem since we do not know the exact formula (function) on the boundary, even though we know the sample points. This problem is also related to the Whitney’s problem [@Fef][@KZ]. Some ideas have been presented by the first author in the Workshop on Whitney’s problem in 2009 organized by C. Fefferman and N. Zobin at College of William and Mary (http://nxzobi.people.wm.edu/whitney/whitney.htm). During the Workshop, P. Shvartsman presented an idea of using geodesic curves in Sobolev space (See related paper [@Shv]). Our idea about using the geodesic curve presented in this note was independently obtained.
[99]{}
L. Chen, Gradually varied surfaces and Gradually varied functions, in Chinese, 1990; in English 2005 CITR-TR 156, U of Auckland.
L. Chen, The necessary and sufficient condition and the efficient algorithms for gradually varied fill, Chinese Sci. Bull. 35 (10) (1990) 870-873. L. Chen, Random gradually varied surface fitting, Chinese Sci. Bull. 37 (16) (1992) 1325-1329. L. Chen, Discrete surfaces and manifolds, Scientific and Practical Computing, Rockville, Maryland, 2004
C. Fefferman, Whitney’s extension problems and interpolation of data, Bull. Amer. Math. Soc. 46 (2009), 207-220. B. Klartag and N. Zobin, $C^1$ extensions of functions and stabilization of Glaeser refinements, Revista Math. Iberoamericana, vol. 23 no. 2, (2007), 635-669.
L. Lovasz, Discrete Analytic Functions: a survey, in: Eigenvalues of Laplacians and other geometric operators (ed. A. Grigor, S.-T. Yau), International Press, Surveys in Differential Geometry IX (2004),
F. Luo, Variational Principles on Triangulated Surfaces. http://arxiv.org/abs/0803.4232v1, 2008.
P. Shvartsman, On Sobolev extension domains in $R^n$, http://arxiv.org/abs/0904.0909, 2009.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'This paper is devoted to studying a type of contact problems modeled by hemivariational inequalities with small periodic coefficients appearing in PDEs, and the PDEs we considered are linear, second order and uniformly elliptic. Under the assumptions, it is proved that the original problem can be homogenized, and the solution weakly converges. We derive an $O(\epsilon^{1/2})$ estimation which is pivotal in building the computational framework. We also show that Robin problems— a special case of contact problems, it leads to an $O(\epsilon)$ estimation in $L^2$ norm. Our computational framework is based on finite element methods, and the numerical analysis is given, together with experiments to convince the estimation.'
author:
- 'Changqing Ye[^1], Junzhi Cui[^2]'
bibliography:
- 'references.bib'
date: '[^3]'
title: 'A First-order Two-scale Analysis for Contact Problems with Small Periodic Configurations'
---
Introduction
============
In composite material design and performance optimization, the controlling PDEs within the models frequently involve small periodic coefficients (e.g., [@Yang2017; @Dong2018; @Liu2013; @Liu2015]). For those problems, the periodic homogenization theory is the basis, and many PDE experts contributed considerable works to build this theory. For examples: qualitative results such as asymptotic expansion [@Bensoussan1978; @Oleuinik1992], H-convergence [@Tartar2010], $\Gamma$-convergence [@DalMaso1993], and two-scale convergence [@Allaire1992]; regularity results such as compactness property investigated by M. Avellaneda and Lin in [@Avellaneda1987], and recently a thorough study for Neumann boundary condition by Shen et al. [@Kenig2013; @Shen2016]. As a model problem for the multiscale phenomenon, it also attracts great attention among scientific computation community. Due to the high oscillation emerging in the solution, classical computational method, such as finite element methods (FEM) can not reveal the fine scale information. Because of strong practical background, several modern multiscale computation methods have been developed since 1990s. We can classify those as three groups in methodology: modify FEM piecewise polynomial basis to enhance the expression ability, such as MsFEM [@Hou1999; @Efendiev2000; @Chen2002] and LOD [@Malqvist2014; @Henning2014]; utilize scale separation property to decompose original solution into coarsen and fine parts, VMM [@Hughes1995; @Brezzi1997; @Larson2007]; improve the accuracy of homogenized solution by involving smallscale information, HMM [@Weinan2005; @Abdulle2012]. In those methods Dirichlet problem is chosen when conducting numerical experiments and error analysis, while Robin problem or more general contact problems are scarcely investigated.
The notion of hemivariational inequalities was first introduced by Panagiotopoulos in the early 1980s [@Panagiotopoulos1983]. Since then, hemivariational inequalities receive broad applications in nonsmooth mechanics, contact mechanics, physics, and economics [@Naniewicz1995; @Panagiotopoulos1993; @Migorski2013; @Goeleven2003]. In this paper, we focus on boundary hemivariational inequality problems, which originate from the mathematical model of elastic contact system. To solve this kind of hemivariational inequalities, a finite element method had been implemented [@Haslinger1999] while the thorough numerical analysis has not been established until recent. In [@Han2017], Han et al. derived a Céa’s inequality in an abstract framework, and figure out the influence of solution’s regularity to the numerical computation. To our knowledge, proper assumptions will balance the solvability and generality of mathematical models, and this is extremely important in nonlinear problems. Hence, we adopt the assumptions in [@Han2017] to prove our main results.
To our knowledge the study on hemivariational inequalities with the coefficients setting in small periodic configurations is few. The homogenization result could be found in [@Liu2008]. However, the result or H-convergence property does not provide a priori convergence rate which is pivotal in numerical analysis. It explains why we need to build an $O(\epsilon^{1/2})$ estimation. In the following sections, we set model problem in a scalar form merely for the simplicity of symbols, and the extension to elastic system will be straight.
The rest of paper is organized as follows. In \[sec:preliminaries\] we introduce notations, review some preliminary materials including generalized directional derivative, and state the model problem and assumptions for later proof. In \[sec:homogenization\], firstly, we prove a uniform bound for solutions which is missing in [@Liu2008], and we think it is indispensable. Then we apply div-curl lemma to prove the homogenization result, the proof will also be provided for the self-containing. We give an $O(\epsilon^{1/2})$ estimation for first order asymptotic expansion in \[sec:estimation\], and the insight most comes from [@Shen2016]. We discuss Robin problem in \[sec:robin\], and show that with the duality technique from [@Shen2016] the difference between original and homogenized solutions in $L^2$ norm is $O(\epsilon)$. A computational framework based on finite element methods will be presented in \[sec:computation\] together with its numerical analysis. Experiments are reported in \[sec:experiments\], and the results are in good agreement with predicted estimation.
Preliminaries {#sec:preliminaries}
=============
Generally, when $X$ is used, it denotes a real Banach space with its norm as ${\left\lVert \cdot \right\rVert}_X$, $X^*$ as its topological dual, ${\left\langle \cdot,\cdot \right\rangle}_{X^*\times X}$ as duality pairing. Without confusion, we omit the subscript and simply write ${\left\langle \cdot,\cdot \right\rangle}$. Weak convergence is indicated by $\rightharpoonup$. Given two normed spaces $X$ and $Y$, $\mathcal{L}(X, Y)$ is the space of all linear continuous operators from $X$ to $Y$.
Notation $d$ is always used as the space dimension. In the full text, the Einstein summation convention is adopted, means summing repeated indexes from $1$ to $d$. Without specification, $\Omega$ is a domain (open and bounded set in ${\mathbb{R}}^d$) with Lipschitz boundary $\Gamma$, and denote $\bm{n}$ as the outward unit normal to $\Gamma$. The Sobolev spaces $W^{k,p}$ and $H^{k}$ are defined as usual (see [@Brenner2008]) and we abbreviate the norm and seminorm of Sobolev space $H^k(\Omega)$ as ${\left\lVert \cdot \right\rVert}_{k,\Omega}$ and ${\left\lvert \cdot \right\rvert}_{k,\Omega}$.
To specify conditions respectively on the different parts of boundary, we rewrite $\Gamma= { \sbox{\myboxA}{$\m@th\Gamma_D$} \setbox\myboxB\null \ht\myboxB=\ht\myboxA \dp\myboxB=\dp\myboxA \wd\myboxB=0.75\wd\myboxA \sbox\myboxB{$\m@th\overline{\copy\myboxB}$} \setlength\mylenA{\the\wd\myboxA} \addtolength\mylenA{-\the\wd\myboxB} \ifdim\wd\myboxB<\wd\myboxA \rlap{\hskip 0.5\mylenA\usebox\myboxB}{\usebox\myboxA} \else
\hskip -0.5\mylenA\rlap{\usebox\myboxA}{\hskip 0.5\mylenA\usebox\myboxB} \fi} \cup { \sbox{\myboxA}{$\m@th\Gamma_N$} \setbox\myboxB\null \ht\myboxB=\ht\myboxA \dp\myboxB=\dp\myboxA \wd\myboxB=0.75\wd\myboxA \sbox\myboxB{$\m@th\overline{\copy\myboxB}$} \setlength\mylenA{\the\wd\myboxA} \addtolength\mylenA{-\the\wd\myboxB} \ifdim\wd\myboxB<\wd\myboxA \rlap{\hskip 0.5\mylenA\usebox\myboxB}{\usebox\myboxA} \else
\hskip -0.5\mylenA\rlap{\usebox\myboxA}{\hskip 0.5\mylenA\usebox\myboxB} \fi} \cup { \sbox{\myboxA}{$\m@th\Gamma_C$} \setbox\myboxB\null \ht\myboxB=\ht\myboxA \dp\myboxB=\dp\myboxA \wd\myboxB=0.75\wd\myboxA \sbox\myboxB{$\m@th\overline{\copy\myboxB}$} \setlength\mylenA{\the\wd\myboxA} \addtolength\mylenA{-\the\wd\myboxB} \ifdim\wd\myboxB<\wd\myboxA \rlap{\hskip 0.5\mylenA\usebox\myboxB}{\usebox\myboxA} \else
\hskip -0.5\mylenA\rlap{\usebox\myboxA}{\hskip 0.5\mylenA\usebox\myboxB} \fi}$, $\Gamma_D$, $\Gamma_N$ and $\Gamma_C$ are open according to the inheriting topology on $\Gamma$ and disjoint with each other, $\Gamma_C \neq \varnothing$ and $\Gamma_D \neq \varnothing$ without specification. We mainly concern functional space $V$ which its functions $u \in H^1(\Omega)$ and vanishing on $\Gamma_D$ in the sense of trace, and one can easily check that $V$ is Hilbertian, and the norm can be legally set as ${\left\lVert \cdot \right\rVert}_V={\left\lvert \cdot \right\rvert}_{1,\Omega}$ when $\Gamma_D \neq \varnothing$. Following the notations in [@Han2017], we denote $V_j=L^2(\Gamma_C)$ as the main space for hemivariational inequality and $\gamma_j \in \mathcal{L}(V,V_j)$ as trace operator from $V$ to $V_j$. We point here that the split $\Gamma= { \sbox{\myboxA}{$\m@th\Gamma_D$} \setbox\myboxB\null \ht\myboxB=\ht\myboxA \dp\myboxB=\dp\myboxA \wd\myboxB=0.75\wd\myboxA \sbox\myboxB{$\m@th\overline{\copy\myboxB}$} \setlength\mylenA{\the\wd\myboxA} \addtolength\mylenA{-\the\wd\myboxB} \ifdim\wd\myboxB<\wd\myboxA \rlap{\hskip 0.5\mylenA\usebox\myboxB}{\usebox\myboxA} \else
\hskip -0.5\mylenA\rlap{\usebox\myboxA}{\hskip 0.5\mylenA\usebox\myboxB} \fi} \cup { \sbox{\myboxA}{$\m@th\Gamma_N$} \setbox\myboxB\null \ht\myboxB=\ht\myboxA \dp\myboxB=\dp\myboxA \wd\myboxB=0.75\wd\myboxA \sbox\myboxB{$\m@th\overline{\copy\myboxB}$} \setlength\mylenA{\the\wd\myboxA} \addtolength\mylenA{-\the\wd\myboxB} \ifdim\wd\myboxB<\wd\myboxA \rlap{\hskip 0.5\mylenA\usebox\myboxB}{\usebox\myboxA} \else
\hskip -0.5\mylenA\rlap{\usebox\myboxA}{\hskip 0.5\mylenA\usebox\myboxB} \fi} \cup { \sbox{\myboxA}{$\m@th\Gamma_C$} \setbox\myboxB\null \ht\myboxB=\ht\myboxA \dp\myboxB=\dp\myboxA \wd\myboxB=0.75\wd\myboxA \sbox\myboxB{$\m@th\overline{\copy\myboxB}$} \setlength\mylenA{\the\wd\myboxA} \addtolength\mylenA{-\the\wd\myboxB} \ifdim\wd\myboxB<\wd\myboxA \rlap{\hskip 0.5\mylenA\usebox\myboxB}{\usebox\myboxA} \else
\hskip -0.5\mylenA\rlap{\usebox\myboxA}{\hskip 0.5\mylenA\usebox\myboxB} \fi}$ must be regular enough to guarantee that $\gamma_j$ is compact, and normally it is true because that fractional Sobolev space $H^{1/2}(\Gamma_C)$ is compactly embedded into $L^2(\Gamma_C)$ (refer [@Nezza2012] for more details).
To describe the periodic structure, we denote $Q=(-1/2, 1/2)^d$ as a representative cell, and call a function $f$ 1-periodic, it means: $$f(\bm{x}+\bm{z}) = f(\bm{x}) ~~~~ \forall \bm{x} \in {\mathbb{R}}^d \text{ and } \forall \bm{z} \in {\mathbb{Z}}^d ,$$ and we also use a superscript $\epsilon$ for $f(\bm{x})$ to represent scaling $f^\epsilon(\bm{x})\coloneqq f(\bm{x}/\epsilon)$ if $f$ is 1-periodic. $H^k_\sharp(Q)$ or $W^{k,p}_\sharp(Q)$ with “$\sharp$” means this functional space is the completion of smooth 1-periodic functions with respect to the $H^k(Q)$ or $W^{k,p}(Q)$ norm. We have a fundamental theorem for $f^\epsilon$:
\[thm:oscillating converge\] Let $1\leq p < \infty$, $\forall f \in L^p_\sharp(Q)$, then $f^\epsilon \rightharpoonup \mathcal{M}_Q f= 1/{\left\lvert Q \right\rvert}\int_Q f$ in $L^p(\omega)$. Here $\omega$ is an arbitrary bounded open subset in ${\mathbb{R}}^d$.
It is customary to write $C$ as a positive constant, and $C(p_1,\cdots,p_n)$ indicates that $C$ depends on $p_1,\cdots,p_n$.
Then we introduce (Clarke) generalized directional derivative and subdifferential (see [@Denkowski2003]).
Let $\varphi: X \rightarrow {\mathbb{R}}$ be a locally Lipschitz function. For $x, h \in X$, the *generalized directional derivative* of $\varphi$ at $x$ along the direction $h$, denoted by $\varphi^0(x; h)$ is defined by $$\varphi^0(x; h) \coloneqq \limsup_{y\rightarrow x, \lambda \downarrow 0} \frac{\varphi(y+\lambda h)-\varphi(y)}{\lambda} = \inf_{\epsilon, \delta>0} \sup_{\substack{{\left\lVert x-y \right\rVert}_X<\epsilon \\ 0<\lambda< \delta}} \frac{\varphi(y+\lambda h)-\varphi(y)}{\lambda} .$$ The *generalized subdifferential* of $\varphi$ at $x \in X$, is the nonempty set $\partial \varphi (x) \subset X^*$ defined by $$\partial \varphi(x) \coloneqq \{ x^* \in X^*: {\left\langle x^*,h \right\rangle} \leq \varphi^0(x; h), \forall h \in X \}.$$
From now on, the matrix function $$A^\epsilon(\bm{x})={\left[ A^\epsilon_{ij}(\bm{x})\right]}_{1\leq i, j \leq d}=A(\bm{x}/\epsilon)={\left[ A_{ij}(\frac{\bm{x}}{\epsilon})\right]}_{1\leq i, j \leq d}$$ serves as the coefficients in our PDE model. The scale parameter $\epsilon \ll 1$. Also provide $f \in V^*$, $g \in L^2(\Gamma_N)$, and $j: V_j \rightarrow {\mathbb{R}}$ is a locally Lipschitz function. Now we can formulate our contact problem, $$\label{eq:contact problem}
\left\{
\begin{aligned}
-{\mathrm{div}}(A^\epsilon(\bm{x})\nabla u_\epsilon) = f ~~~~ &\text{in} ~~ \Omega \\
u_\epsilon = 0 ~~~~ &\text{on} ~~ \Gamma_D \\
\bm{n} \cdot A^\epsilon \nabla u_\epsilon = g ~~~~ &\text{on} ~~ \Gamma_N \\
-\bm{n} \cdot A^\epsilon \nabla u_\epsilon \in \partial j(\gamma_j u_\epsilon) ~~~~ &\text{on} ~~ \Gamma_C
\end{aligned}
\right. ,$$ and its hemivariational form: $$\label{eq:contact problem hemiform}
\left\{
\begin{aligned}
&\text{Find } u_\epsilon \in V, \text{ s.t. } \forall v \in V \\
& \int_\Omega A^\epsilon \nabla u_\epsilon \cdot \nabla v + j^0(\gamma_j u_\epsilon; \gamma_j v) \geq {\left\langle f,v \right\rangle}+\int_{\Gamma_N} gv
\end{aligned}
\right. .$$ We mention that $v \mapsto \int_{\Gamma_N} gv$ is a bounded functional on $V$ since $g \in L^2(\Gamma_N)$, one can rewrite ${\left\langle \tilde{f},v \right\rangle}\coloneqq {\left\langle f,v \right\rangle}+\int_{\Gamma_N} gv$. It is clear that ${\left\lVert \tilde{f} \right\rVert}_{V^*} \leq {\left\lVert f \right\rVert}_{V^*}+c_k {\left\lVert g \right\rVert}_{0,\Gamma_N}$, where $c_k$ equals the trace operator norm from $V$ to $L^2(\Gamma_N)$.
To make this hemivariational form solvable, we need the following assumptions:
- The coefficient matrix $A(y)$ is symmetric and uniformly elliptic: $$\label{ass:A}
\begin{aligned}
& A_{ij}(\bm{y}) = A_{ji}(\bm{y}) \\
& \kappa_1 {\left\lvert \xi \right\rvert}^2 \leq A_{ij}(\bm{y})\xi_i \xi_j \leq \kappa_2 {\left\lvert \xi \right\rvert}^2 ~~~~ \text{for a.e. } \bm{y} \in {\mathbb{R}}^d \text{ and } \xi \in {\mathbb{R}}^d
\end{aligned} ~ .$$
- There exist constants $c_0,c_1,\alpha_j$, such that: $$\label{ass:B1}
{\left\lVert x^* \right\rVert}_{V_j^*} \leq c_0+c_1{\left\lVert x \right\rVert}_{V_j} \forall x \in V_j, \forall x^* \in \partial j(x) ,$$ $$\label{ass:B2}
j^0(x_1; x_2-x_1)+j^0(x_2; x_1-x_2) \leq \alpha_j {\left\lVert x_1-x_2 \right\rVert}_{V_j}^2 \forall x_1, x_2 \in V_j .$$
- Let $c_j={\left\lVert \gamma_j \right\rVert}_{V\rightarrow V_j}$ be the operator norm, there exists that, $$\label{ass:C}
\Delta \coloneqq \kappa_1 - \alpha_j c_j^2 > 0 .$$
We remark here that assumptions **B** and **C** follow [@Han2017]. Then by utilizing the framework constructed in [@Han2017] following theorem is obvious:
The solution of problem \[eq:contact problem hemiform\] exists and is unique.
The assumptions in former work [@Liu2008] are sightly different with [@Han2017]. For example, [@Liu2008] needs $j^0_N(x, r; -r) \leq d_N(1+{\left\lvert r \right\rvert})$ which is redundant in our case. Hence, we think the assumptions in [@Han2017] are more reasonable.
We close this section by illustrating a specific contact problem. Here \[fig:Rubin\] describes a domain $\Omega$ with its boundary $\Gamma$ composed by $\Gamma_D, \Gamma_N, \Gamma_{C'}, \Gamma_{C^{\prime\prime}}$. On $\Gamma_D$, we have $u=0$; On $\Gamma_N$, we have $g=\bm{n}\cdot A \nabla u$; On $\Gamma_{C'}$, a complete Robin condition is imposed, means $-\bm{n}\cdot A \nabla u = u$; while on $\Gamma_{C^{\prime\prime}}$, it is instead with a partial Robin condition $-\bm{n}\cdot A \nabla u = 0$ if $u<0$ and $-\bm{n}\cdot A \nabla u = u$ if $u\geq 0$, or write shortly as $-\bm{n}\cdot A \nabla u = u^+$.
One can check in this problem $j^0(\gamma_ju; \gamma_j v)=\int_{\Gamma_{C'}}uv+\int_{\Gamma_{C^{\prime\prime}}}u^+v $. The original locally Lipschitz function $j(u)=1/2 \int_{\Gamma_{C'}}u^2+1/2\int_{\Gamma_{C^{\prime\prime}}} {\left( u^+ \right)}^2$, and $\partial j (\gamma_ju)=u$ on $\Gamma_{C'}$ and $\partial_j(\gamma_ju)=u^+$ on $\Gamma_{C^{\prime\prime}}$. $\partial j, j^0$ coincide respectively with classical Gâteaux and Frèchet derivative of functional $j(u)$.
(0, 0) rectangle (4, 4); in [1,...,13]{} [ (0.3\*-0.05, 4) – +(135:0.3); (0, 0.3\*-0.05) – +(-0.3, 0); (4, 0.3\*-0.05) – +(0.3, 0); ]{} (-0.5, -0.5) rectangle (2, 0); (2, -0.5) rectangle (4.5, 0); at (0.4, 2) [$\Gamma_N$]{}; at (3.6, 2) [$\Gamma_N$]{}; at (2, 3.6) [$\Gamma_D$]{}; at (1, 0.4) [$\Gamma_{C'}$]{}; at (3, 0.4) [$\Gamma_{C^{\prime\prime}}$]{}; at (2, 2) [$\Omega$]{};
in [1,...,8]{} in [1,...,8]{} [ (0.5\*-0.5, 0.5\*-0.5) rectangle (0.5\*, 0.5\*); (0.5\*-0.25, 0.5\*-0.25) circle (0.15); ]{}
Homogenization {#sec:homogenization}
==============
In this section, we will give the definition of correctors and present that the problem \[eq:contact problem hemiform\] has a homogenized version, then we use div-curl lemma to prove $u_\epsilon$ can weak converge to the homogenized solution.
We denote $N_l(\bm{y})$ as correctors for $A^\epsilon$, which satisfy a group of PDEs with periodic boundary condition: $$\label{eq:correctors}
\left\{
\begin{aligned}
& -{\mathrm{div}}{\left( A(\bm{y})\nabla N_l \right)}={\mathrm{div}}(Ae_l)=\partial_i A_{il}(\bm{y}) ~~~~ \text{ in }Q\\
& N_l(\bm{y}) \in H^1_\sharp(Q) \text{ and } \int_Q N_l = 0
\end{aligned}
\right. .$$ For correctors, we have ${\left\lVert N_l \right\rVert}_{1,Q} \leq C(\kappa_1, \kappa_2)$. The homogenized coefficients are defined by $\hat{A}_{il}=\fint_Q A_{il}+A_{ij}\partial_j N_l {\mathrm{d}}\bm{y}$. The next lemma characterizes the homogenized coefficients $\hat{A}$, the proof can be found in [@Jikov1994] sect. 1.6.
Let $\kappa_1, \kappa_2$ define as previous, then $\hat{A}$ is symmetric and uniformly elliptic, means the following relation $$\begin{aligned}
& \hat{A}_{ij} = \hat{A}_{ji} \\
& \tilde{\kappa}_1 {\left\lvert \xi \right\rvert}^2 \leq \hat{A}_{ij}(\bm{y})\xi_i \xi_j \leq \tilde{\kappa}_2 {\left\lvert \xi \right\rvert}^2 ~~~~ \forall \xi \in {\mathbb{R}}^d
\end{aligned}$$ holds, where $\tilde{\kappa}_1, \tilde{\kappa}_2$ depends on $\kappa_1, \kappa_2$.
Then we have the homogenized hemivariational form: $$\label{eq:contact problem homohemiform}
\left\{
\begin{aligned}
&\text{Find } u_0 \in V, \text{ s.t. } \forall v \in V \\
& \int_\Omega \hat{A} \nabla u_0 \cdot \nabla v + j^0(\gamma_j u_0; \gamma_j v) \geq {\left\langle \tilde{f},v \right\rangle}
\end{aligned}
\right. .$$
Again, [@Han2017] tells us $u_0$ exists and is unique. To deduce $u_\epsilon \rightharpoonup u_0$, we first claim that $\{ u_\epsilon\}$ is uniformly bounded, then a subsequence of $\{ u_\epsilon\}$ will weakly converge in $V$. Finally, we prove the convergence point can only be $u_0$.
There is a constant $C$ independent with $\epsilon$, such that ${\left\lVert u_\epsilon \right\rVert}_V \leq C$, and $$C = 1/\Delta{\left( c_0c_j+{\left\lVert \tilde{f} \right\rVert}_{V^*} \right)} \leq 1/\Delta {\left( c_0c_j+{\left\lVert f \right\rVert}_{V^*}+c_k{\left\lVert g \right\rVert}_{0,\Gamma_N} \right)}.$$
Let $v=-u_\epsilon$ in \[eq:contact problem hemiform\], we have $$\begin{aligned}
\kappa_1 {\left\lVert u_\epsilon \right\rVert}_V^2 &\leq \int_{\Omega} A^\epsilon \nabla u_\epsilon\cdot \nabla u_\epsilon \leq {\left\langle \tilde{f},u_\epsilon \right\rangle}+j^0(\gamma_ju_\epsilon; -\gamma_j u_\epsilon) \\
&\leq {\left\lVert \tilde{f} \right\rVert}_{V^*} {\left\lVert u_\epsilon \right\rVert}_V+j^0(\gamma_ju_\epsilon; -\gamma_j u_\epsilon)
\end{aligned}.$$ Recall \[ass:B2\], then $$\begin{aligned}
j^0(\gamma_ju_\epsilon; -\gamma_j u_\epsilon) &\leq j^0(\gamma_ju_\epsilon; -\gamma_ju_\epsilon)+j^0(0;\gamma_ju_\epsilon)-j^0(0;\gamma_ju_\epsilon) \\ &\leq \alpha_j {\left\lVert \gamma_ju_\epsilon \right\rVert}_{V_j}^2-j^0(0;\gamma_ju_\epsilon) \\
& \leq \alpha_j c_j^2 {\left\lVert u_\epsilon \right\rVert}_V^2-j^0(0;\gamma_ju_\epsilon)
\end{aligned}.$$ By the definition of $j^0(x;h)$, we choose arbitrarily a $\xi \in \partial j(0)$ $$\begin{aligned}
-j^0(0;\gamma_ju_\epsilon) &\leq -{\left\langle \xi,\gamma_ju_\epsilon \right\rangle} \leq {\left\lVert \xi \right\rVert}_{X_j^*} {\left\lVert \gamma_ju_\epsilon \right\rVert}_{X_j}\\
& \leq c_0 c_j {\left\lVert u_\epsilon \right\rVert}_V
\end{aligned}.$$ Then we establish the desired inequality.
div-curl lemma states as follows:
Let $\bm{p}_\epsilon, \bm{q}_\epsilon \in L^2(\Omega)^d$, such that: $$\bm{p}_\epsilon \rightharpoonup \bm{p}_0 ~~~~ \bm{q}_\epsilon \rightharpoonup \bm{q}_0 ~~~~ \text{in } L^2(\Omega)^d .$$ If, in addition $\bm{q}_\epsilon = \nabla v_\epsilon$ and ${\mathrm{div}}\bm{p}_\epsilon \rightarrow f$ in $H^{-1}(\Omega)$, then $\bm{p}_\epsilon \cdot \bm{q}_\epsilon \overset{*}{\rightharpoonup} \bm{p}_0 \cdot \bm{q}_0$. Here “$\overset{*}{\rightharpoonup}$” stands for *\*-weak* convergence, it means that $\forall \phi \in C^\infty_0(\Omega), \int_\Omega \bm{p}_\epsilon \cdot \bm{q}_\epsilon \phi \rightarrow \int_{\Omega} \bm{p}_0 \cdot \bm{q}_0 \phi$.
Now we can prove the main homogenization result:
Under the assumptions \[ass:A\]-\[ass:C\], let $u_\epsilon$ and $u_0$ be the unique solution of \[eq:contact problem hemiform\] and \[eq:contact problem homohemiform\] respectively. Then $u_\epsilon \rightharpoonup u_0$ in $V$.
First, we make some justifications. Take $v \in C^\infty_0(\Omega)$ and $-v$ into \[eq:contact problem homohemiform\], it is obvious that $j^0(\gamma_ju_\epsilon; \gamma_j v)=j^0(\gamma_j u_\epsilon; -\gamma_j v)=0$. We get $\int_\Omega A^\epsilon \nabla u_\epsilon \cdot \nabla v = {\left\langle f,v \right\rangle}$, which means $-{\mathrm{div}}(A^\epsilon \nabla u_\epsilon) = f$ in the sense of weak derivative, and $f$ can naturally embed into $H^{-1}(\Omega)$.
We have already obtained that $\{ u_\epsilon\}$ is uniformly bounded, then up to a subsequence, we have $\nabla u_\epsilon \rightharpoonup \nabla u_0$, and $A^\epsilon \nabla u_\epsilon \rightharpoonup \bm{p}_0$ in $L^2(\Omega)^d$. And then we show $\bm{p}_0=\hat{A}\nabla u_0$ and $u_0$ satisfies \[eq:contact problem homohemiform\]. Combine div-curl lemma and \[thm:oscillating converge\] we get $$A^\epsilon_{ij}\partial_j u_\epsilon \partial_i (\epsilon N^\epsilon_l \lambda_l+\bm{\lambda}\cdot \bm{x}) \overset{*}{\rightharpoonup} \bm{p}_0 \cdot \bm{\lambda} .$$ On the other side, notice $A^\epsilon$ and $\hat{A}$ are both symmetric, from the definition of correctors we have $\partial_j{\left( A^\epsilon_{ij} \partial_i (\epsilon N^\epsilon_l \lambda_l + \bm{\lambda}\cdot \bm{x}) \right)}=0$, and $$A^\epsilon_{ij}\partial_j u_\epsilon \partial_i (\epsilon N^\epsilon_l \lambda_l+\bm{\lambda}\cdot \bm{x}) = A^\epsilon_{ij} \partial_j(\epsilon N^\epsilon_l \lambda_l + \bm{\lambda}\cdot \bm{x})\partial_i u_\epsilon \overset{*}{\rightharpoonup} \hat{A}\bm{\lambda} \cdot \nabla u_0 = \hat{A} \nabla u_0 \cdot \bm{\lambda} .$$ It asserts that $\bm{p}_0=\hat{A}\nabla u_0$.
Recall that $\gamma_j$ is a compact operator from $V$ to $V_j$, then up to a subsequence $\{\gamma_ju_\epsilon \}$ converges strongly in $V_j$. Because of $u_\epsilon \rightharpoonup u_0$, $\gamma_ju_\epsilon \rightarrow \gamma_ju_0$ in $L^2(\Gamma_C)$ holds. Utilize the fact that $j^0(x; h)$ is upper semicontinuous with $x$ (see [@Denkowski2003] Prop. 5.6.6), we arrive at $j^0(\gamma_ju_0; \gamma_j v) \geq \limsup_\epsilon j^0(\gamma_ju_\epsilon; \gamma_j v)$, take the sup limit in the left hand of \[eq:contact problem hemiform\], we have $\forall v \in V$: $$\int_{\Omega} \hat{A} \nabla u_0 \cdot \nabla v +j^0(\gamma_ju_0; \gamma_j v) \geq {\left\langle \tilde{f},v \right\rangle},$$ and this finishes the proof.
$\epsilon^{1/2}$ estimation {#sec:estimation}
===========================
Following \[lem:3.1\], \[lem:3.2\] and \[lem:3.4\] are quoted from [@Shen2016]. In those the standard smoothing operator ${\mathbf{S}}_\epsilon u = \phi_\epsilon \ast u$ is defined as usual (see [@Grisvard2011] sect. 7.2), while we need the convolution kernel to be contained in $B_{1/2}(\bm{0}) \subset Q$, $B_r(\bm{x})$ means an open ball centers in $\bm{x}$ with radius $r$.
\[lem:3.1\] Let $u \in H^1({\mathbb{R}}^d)$. Then $\partial_i {\mathbf{S}}_\epsilon(u)={\mathbf{S}}_\epsilon(\partial_i u)$, $${\left\lVert {\mathbf{S}}_\epsilon u \right\rVert}_{0, {\mathbb{R}}^d} \leq {\left\lVert u \right\rVert}_{0, {\mathbb{R}}^d} ,$$ and there exists a constant $C$ only depends on $d$, such that: $${\left\lVert {\mathbf{S}}_\epsilon u - u \right\rVert}_{0, {\mathbb{R}}^d} \leq \epsilon C {\left\lVert \nabla u \right\rVert}_{0, {\mathbb{R}}^d} .$$
\[lem:3.2\] Let $f \in L^2_{\text{loc}}({\mathbb{R}}^d)$ be a 1-periodic function. Then there exists a constant $C$ only depends on $d$, such that for any $u \in L^2({\mathbb{R}}^d)$, $${\left\lVert f^\epsilon {\mathbf{S}}_\epsilon u \right\rVert}_{0, {\mathbb{R}}^d} \leq C{\left\lVert f \right\rVert}_{0, Q}{\left\lVert u \right\rVert}_{0,{\mathbb{R}}^d} .$$
Denote $\tilde{\Omega}_{\epsilon}=\{\bm{x} \in {\mathbb{R}}^d: \text{dist}(\bm{x}, \partial \Omega) < \epsilon \}$ as a boundary strip to $\Omega$ with width $2\epsilon$, and $\Omega_{\epsilon}=\{\bm{x} \in \Omega: \text{dist}(\bm{x}, \partial \Omega) < \epsilon \}$. We have an estimation:
\[lem:3.4\] Let $\Omega$ be a bounded Lipschitz domain in ${\mathbb{R}}^d$ and $f \in L^2_{\text{loc}}({\mathbb{R}}^d)$ a 1-periodic function. Then there exists a constant $C$ only depends on $\Omega$, such that for any $u \in H^1({\mathbb{R}}^d)$ $$\int_{\tilde{\Omega}_{\epsilon}} {\left\lvert f^\epsilon \right\rvert}^2 {\left\lvert {\mathbf{S}}_\epsilon(u) \right\rvert}^2 \leq C \epsilon {\left\lVert f \right\rVert}_{0,Q}^2{\left\lVert u \right\rVert}^2_{1,{\mathbb{R}}^d} .$$
We can obtain a similar result comparing to \[lem:3.2\] by assuming ${\left\lVert f \right\rVert}_{L^\infty(\Omega)} < \infty$, but in here the periodic property plays key role and leads to relax on regularity assumption for $f$, which provides us more generality in the estimation.
Recall the domain we consider has Lipschitz boundary, then the extension operator $\mathbf{E}: H^1(\Omega)\mapsto H^1({\mathbb{R}}^d)$ exists and is bounded. Let $w_\epsilon = u_\epsilon - u_0-\epsilon N^\epsilon_l {\mathbf{S}}_\epsilon(\partial_l {\mkern 1.5mu\overline{\mkern-1.5muu\mkern-1.5mu}\mkern 1.5mu}_0)$, ${\mkern 1.5mu\overline{\mkern-1.5muu\mkern-1.5mu}\mkern 1.5mu}_0 = \mathbf{E}u_0 \in H^1({\mathbb{R}}^d)$ is the extension of $u_0$, our goal in this section is to prove following key lemma:
\[lem:key\] Let \[ass:A\] be satisfied, and $u_\epsilon, u_0$ the solution of \[eq:contact problem hemiform\] \[eq:contact problem homohemiform\] respectively, and suppose that $u_0 \in H^2(\Omega)$, then $\forall v \in V$: $$\int_{\Omega} A^\epsilon \nabla w_\epsilon \cdot \nabla v \leq j^0(\gamma_ju_0; \gamma_j v)+j^0(\gamma_j u_\epsilon; -\gamma_j v) + C {\left\lVert u_0 \right\rVert}_{2,\Omega}
{\left( \epsilon^{1/2}{\left\lVert \nabla v \right\rVert}_{0,\Omega_{2\epsilon}}+\epsilon{\left\lVert \nabla v \right\rVert}_{0,\Omega} \right)},$$ here constant $C$ dependents on $\Omega, \kappa_1, \kappa_2$ and $w_\epsilon = u_\epsilon - u_0-\epsilon N^\epsilon_l {\mathbf{S}}_\epsilon(\partial_l {\mkern 1.5mu\overline{\mkern-1.5muu\mkern-1.5mu}\mkern 1.5mu}_0)$.
It is harmless to assume $v \in C^\infty (\Omega) \cap V$. Since $u_0 \in H^2(\Omega)$, we have ${\left\lVert {\mkern 1.5mu\overline{\mkern-1.5muu\mkern-1.5mu}\mkern 1.5mu}_0 \right\rVert}_{2, {\mathbb{R}}^d} \leq C {\left\lVert u \right\rVert}_{2,\Omega}$. In following proof, constant $C$ only depends on $\Omega, \kappa_1, \kappa_2$. By Green formula, we have $\forall v \in V$ $$\begin{aligned}
\int_{\Omega} A^\epsilon \nabla u_\epsilon \cdot \nabla v - \int_{\Gamma_C} \bm{n}\cdot A^\epsilon\nabla u_\epsilon v &= {\left\langle f,v \right\rangle}+\int_{\Gamma_N} g v \\
&= \int_{\Omega} \hat{A} \nabla u_0 \cdot \nabla v - \int_{\Gamma_C} \bm{n} \cdot \hat{A} \nabla u_0 v
\end{aligned}.$$ The definition of $j^0$ gives: $$\begin{aligned}
\int_{\Omega} A^\epsilon \nabla u_\epsilon \cdot \nabla v &= \int_{\Omega} \hat{A} \nabla u_0 \cdot \nabla v + \int_{\Gamma_C} {\left( -\bm{n} \cdot \hat{A} \nabla u_0 \right)} v + \int_{\Gamma_C} {\left( -\bm{n}\cdot A^\epsilon\nabla u_\epsilon \right)}{\left( -v \right)} \\
& \leq \int_{\Omega} \hat{A} \nabla u_0 \cdot \nabla v+j^0(\gamma_j u_0; \gamma_j v)+ j^0(\gamma_j u_\epsilon; -\gamma_j v)
\end{aligned}.$$ Take a direct calculation, $$\begin{aligned}
&\int_\Omega A^\epsilon \nabla w_\epsilon \cdot \nabla v \\
=& \int_{\Omega} A^\epsilon \nabla u_\epsilon \cdot \nabla v-\int_{\Omega} A^\epsilon \nabla u_0 \cdot \nabla v - \int_{\Omega}A^\epsilon_{ij}{\left( \partial_j N_l \right)}^\epsilon {\mathbf{S}}_\epsilon(\partial_l {\mkern 1.5mu\overline{\mkern-1.5muu\mkern-1.5mu}\mkern 1.5mu}_0) \partial_i v -\int_{\Omega}\epsilon A^\epsilon_{ij}N^\epsilon_l {\mathbf{S}}_\epsilon(\partial^2_{lj}{\mkern 1.5mu\overline{\mkern-1.5muu\mkern-1.5mu}\mkern 1.5mu}_0)\partial_i v \\
\leq& j^0(\gamma_j u_0; \gamma_j v)+ j^0(\gamma_j u_\epsilon; -\gamma_j v) \\
&+\int_{\Omega} {\left\{ \hat{A}_{ij} \partial_j u_0-A^\epsilon_{ij}\partial_j u_0 - A^\epsilon_{ij}{\left( \partial_j N_l \right)}^\epsilon {\mathbf{S}}_\epsilon(\partial_l {\mkern 1.5mu\overline{\mkern-1.5muu\mkern-1.5mu}\mkern 1.5mu}_0)\right\}}\partial_i v\\
&-\int_{\Omega}\epsilon A^\epsilon_{ij}N^\epsilon_l {\mathbf{S}}_\epsilon(\partial^2_{lj}{\mkern 1.5mu\overline{\mkern-1.5muu\mkern-1.5mu}\mkern 1.5mu}_0)\partial_i v
\end{aligned}.$$ The techniques for last two parts of above inequality are exactly same as ([@Shen2016] Lem. 3.5). However, we will still elaborate its details for the coherence and self-containing of the proof. $$\begin{aligned}
&\int_{\Omega} {\left\{ \hat{A}_{ij} \partial_j u_0-A^\epsilon_{ij}\partial_j u_0 - A^\epsilon_{ij}{\left( \partial_j N_l \right)}^\epsilon {\mathbf{S}}_\epsilon(\partial_l {\mkern 1.5mu\overline{\mkern-1.5muu\mkern-1.5mu}\mkern 1.5mu}_0)\right\}}\partial_i v-\int_{\Omega}\epsilon A^\epsilon_{ij}N^\epsilon_l {\mathbf{S}}_\epsilon(\partial^2_{lj}{\mkern 1.5mu\overline{\mkern-1.5muu\mkern-1.5mu}\mkern 1.5mu}_0)\partial_i v \\
\coloneqq & \int_\Omega B^\epsilon_{il} {\mathbf{S}}_\epsilon(\partial_l {\mkern 1.5mu\overline{\mkern-1.5muu\mkern-1.5mu}\mkern 1.5mu}_0) \partial_i v+\int_\Omega T_\epsilon \cdot \nabla v \\
\coloneqq & J_1+J_2
\end{aligned}.$$ Here $B^\epsilon={\left[ B_{il}(\bm{x}/\epsilon)\right]}_{1\leq i,l \leq n} \coloneqq \hat{A}-A^\epsilon-{\left( A_{ij}\partial_j N_l \right)}^\epsilon$, and $$T_\epsilon \coloneqq {\left( \hat{A}\nabla u_0 - \hat{A}{\mathbf{S}}_\epsilon(\nabla {\mkern 1.5mu\overline{\mkern-1.5muu\mkern-1.5mu}\mkern 1.5mu}_0) \right)}+{\left( A^\epsilon\nabla u_0-A^\epsilon {\mathbf{S}}_\epsilon(\nabla {\mkern 1.5mu\overline{\mkern-1.5muu\mkern-1.5mu}\mkern 1.5mu}_0) \right)}-\epsilon A^\epsilon_{ij} N^\epsilon_l {\mathbf{S}}_\epsilon(\partial^2_{lj}{\mkern 1.5mu\overline{\mkern-1.5muu\mkern-1.5mu}\mkern 1.5mu}_0),$$ \[lem:3.1\] together with \[lem:3.2\] implies ${\left\lVert T_\epsilon \right\rVert}_{0,\Omega} \leq \epsilon {\left\lVert u_0 \right\rVert}_{2,\Omega}$, then: $${\left\lvert J_2 \right\rvert} \leq C \epsilon {\left\lVert u_0 \right\rVert}_{2,\Omega} {\left\lVert \nabla v \right\rVert}_{0,\Omega} .$$ Because $\partial_i B_{il}=\bm{0}$ and $\fint_Q B_{il} = \bm{0}$, we can construct $E_{ijl}(\bm{y}) \in H^1_\sharp(Q)$, such that $B_{il}=\partial_j E_{ijl}$, $E_{ijl}=-E_{jil}$, and ${\left\lVert E_{ijl} \right\rVert}_{1,Q} \leq C {\left\lVert B_{il} \right\rVert}_{0,Q} \leq C(\kappa_1, \kappa_2)$, see ([@Jikov1994] p. 6-7). Take $\theta_\epsilon \in C^\infty({\mathbb{R}}^d)$, let $\theta_\epsilon(\bm{x}) \equiv 1$ when $\text{dist}(\bm{x}, \partial \Omega) \leq \epsilon$; $\theta_\epsilon(\bm{x}) \equiv 0$ when $\text{dist}(\bm{x}, \partial \Omega) \geq 2\epsilon$; $0 \leq \theta_\epsilon(\bm{x}) \leq 1$ when $\epsilon \leq \text{dist}(\bm{x}, \partial \Omega) \leq 2\epsilon$. We have ${\left\lVert \nabla \theta_\epsilon \right\rVert}_{L^\infty({\mathbb{R}}^d)} \leq C(\Omega)/\epsilon$. Then split $J_1$ as $$J_1 = \int_\Omega (\partial_j E_{ijl})^\epsilon \theta_\epsilon {\mathbf{S}}_\epsilon(\partial_l {\mkern 1.5mu\overline{\mkern-1.5muu\mkern-1.5mu}\mkern 1.5mu}_0) \partial_i v +\int_\Omega (\partial_j E_{ijl})^\epsilon (1-\theta_\epsilon) {\mathbf{S}}_\epsilon(\partial_l {\mkern 1.5mu\overline{\mkern-1.5muu\mkern-1.5mu}\mkern 1.5mu}_0) \partial_i v \coloneqq J_{11}+J_{12} .$$ Use \[lem:3.4\], we have: $$\begin{aligned}
{\left\lvert J_{11} \right\rvert} \leq &{\left\lVert \nabla v \right\rVert}_{0,\Omega_{2\epsilon}} {\left[ \sum_i\int_{\Omega_{2\epsilon}} {\left\lvert (\partial_j E_{ijl})^\epsilon \right\rvert}^2 {\left\lvert {\mathbf{S}}_\epsilon(\partial_l {\mkern 1.5mu\overline{\mkern-1.5muu\mkern-1.5mu}\mkern 1.5mu}_0) \right\rvert}^2 \theta_\epsilon^2\right]}^{1/2} \\
\leq & {\left\lVert \nabla v \right\rVert}_{0,\Omega_{2\epsilon}} {\left[ \sum_i\int_{\tilde{\Omega}_{2\epsilon}} {\left\lvert (\partial_j E_{ijl})^\epsilon \right\rvert}^2 {\left\lvert {\mathbf{S}}_\epsilon(\partial_l {\mkern 1.5mu\overline{\mkern-1.5muu\mkern-1.5mu}\mkern 1.5mu}_0) \right\rvert}^2\right]}^{1/2} \\
\leq & C \epsilon^{1/2} {\left\lVert u_0 \right\rVert}_{2,\Omega}{\left\lVert \nabla v \right\rVert}_{0,\Omega_{2\epsilon}}
\end{aligned}.$$ Due to $1-\theta_\epsilon \equiv 0$ on $\partial \Omega$, $(1-\theta_\epsilon) {\mathbf{S}}_\epsilon(\partial_l {\mkern 1.5mu\overline{\mkern-1.5muu\mkern-1.5mu}\mkern 1.5mu}_0) \partial_i v \in C^\infty_0(\Omega)$, and $(\partial_j E_{ijl})^\epsilon=\epsilon \partial_j E_{ijl}^\epsilon$. According to integration by parts: $$\begin{aligned}
J_{12}=&\int_{\Omega} \epsilon \partial_j E_{ijl}^\epsilon (1-\theta_\epsilon) {\mathbf{S}}_\epsilon(\partial_l {\mkern 1.5mu\overline{\mkern-1.5muu\mkern-1.5mu}\mkern 1.5mu}_0) \partial_i v
= -\int_{\Omega} \epsilon E_{ijl}^\epsilon \partial_j{\left[ (1-\theta_\epsilon){\mathbf{S}}_\epsilon(\partial_l {\mkern 1.5mu\overline{\mkern-1.5muu\mkern-1.5mu}\mkern 1.5mu}_0)\partial_i v\right]} \\
= & \int_{\Omega} E_{ijl}^\epsilon \epsilon \partial_j \theta_\epsilon {\mathbf{S}}_\epsilon(\partial_l {\mkern 1.5mu\overline{\mkern-1.5muu\mkern-1.5mu}\mkern 1.5mu}_0) \partial_i v \\
& - \int_{\Omega} \epsilon E_{ijl}^\epsilon (1-\theta_\epsilon) {\mathbf{S}}_\epsilon(\partial^2_{jl}{\mkern 1.5mu\overline{\mkern-1.5muu\mkern-1.5mu}\mkern 1.5mu}_0) \partial_i v \\
& - \int_{\Omega} \epsilon E_{ijl}^\epsilon (1-\theta_\epsilon) {\mathbf{S}}_\epsilon(\partial_l {\mkern 1.5mu\overline{\mkern-1.5muu\mkern-1.5mu}\mkern 1.5mu}_0) \partial^2_{ij} v
\end{aligned},$$ $E_{ijl}=-E_{jil}$ implies $\int_{\Omega} \epsilon E_{ijl}^\epsilon (1-\theta_\epsilon) {\mathbf{S}}_\epsilon(\partial_l {\mkern 1.5mu\overline{\mkern-1.5muu\mkern-1.5mu}\mkern 1.5mu}_0) \partial^2_{ij} v =0$, and $$\begin{aligned}
\int_{\Omega} {\left\lvert E_{ijl}^\epsilon \epsilon \partial_j \theta_\epsilon {\mathbf{S}}_\epsilon(\partial_l {\mkern 1.5mu\overline{\mkern-1.5muu\mkern-1.5mu}\mkern 1.5mu}_0) \partial_i v \right\rvert} \leq& C {\left\lVert \nabla v \right\rVert}_{0,\Omega_{2\epsilon}} {\left[ \sum_{i,j}\int_{\tilde{\Omega}_{2\epsilon}} {\left\lvert E^\epsilon_{ijl} \right\rvert}^2 {\left\lvert {\mathbf{S}}_\epsilon(\partial_l {\mkern 1.5mu\overline{\mkern-1.5muu\mkern-1.5mu}\mkern 1.5mu}_0) \right\rvert}^2\right]}^{1/2} \\\leq& C \epsilon^{1/2} {\left\lVert \nabla v \right\rVert}_{0,\Omega_{2\epsilon}} {\left\lVert u_0 \right\rVert}_{2,\Omega}
\end{aligned},$$ $$\begin{aligned}
\int_{\Omega} {\left\lvert \epsilon E_{ijl}^\epsilon (1-\theta_\epsilon){\mathbf{S}}_\epsilon(\partial_{jl}{\mkern 1.5mu\overline{\mkern-1.5muu\mkern-1.5mu}\mkern 1.5mu}_0)\partial_i v \right\rvert} \leq& \epsilon {\left\lVert \nabla v \right\rVert}_{0,\Omega} {\left[ \sum_i \int_{\Omega} {\left\lvert E_{ijl}^\epsilon \right\rvert}^2 {\left\lvert {\mathbf{S}}_\epsilon(\partial_{jl}{\mkern 1.5mu\overline{\mkern-1.5muu\mkern-1.5mu}\mkern 1.5mu}_0) \right\rvert}^2\right]}^{1/2} \\ \leq& C \epsilon {\left\lVert \nabla v \right\rVert}_{0,\Omega} {\left\lVert u_0 \right\rVert}_{2,\Omega}
\end{aligned}.$$ Finally, we derive ${\left\lvert J_1 \right\rvert}+{\left\lvert J_2 \right\rvert}\leq C {\left\lVert u_0 \right\rVert}_{2,\Omega}
{\left( \epsilon^{1/2}{\left\lVert \nabla v \right\rVert}_{0,\Omega_{2\epsilon}}+\epsilon{\left\lVert \nabla v \right\rVert}_{0,\Omega} \right)}$.
A corollary of this lemma is following theorem:
Let the assumptions \[ass:A\]-\[ass:C\] be satisfied, and $u_0 \in H^2(\Omega)$. Then $${\left\lVert \nabla w_\epsilon \right\rVert}_{0,\Omega} \leq C\epsilon^{1/2}{\left\lVert u_0 \right\rVert}_{2,\Omega},$$ here constant $C$ depends on $\Omega, \kappa_1, \kappa_2, \Delta$.
Take $\theta_\epsilon$ defined as previous, and let $\tilde{u}_\epsilon=u_0+\epsilon N_l^\epsilon {\mathbf{S}}_\epsilon(\partial_l {\mkern 1.5mu\overline{\mkern-1.5muu\mkern-1.5mu}\mkern 1.5mu}_0)(1-\theta_\epsilon)$. Then $w_\epsilon = u_\epsilon-\tilde{u}_\epsilon-\epsilon N_l^\epsilon {\mathbf{S}}_\epsilon(\partial_l {\mkern 1.5mu\overline{\mkern-1.5muu\mkern-1.5mu}\mkern 1.5mu}_0)\theta_\epsilon \coloneqq u_\epsilon-\tilde{u}_\epsilon-r_\epsilon$. It is obvious to see $\tilde{u}_\epsilon \in V$ and $\gamma_j u_0 = \gamma_j \tilde{u_\epsilon}$. We use \[lem:3.1\]-\[lem:3.4\] to handle $r_\epsilon = \epsilon N^\epsilon_l {\mathbf{S}}_\epsilon(\partial_l {\mkern 1.5mu\overline{\mkern-1.5muu\mkern-1.5mu}\mkern 1.5mu}_0) \theta_\epsilon$: $$\begin{aligned}
{\left\lVert \partial_i r_\epsilon \right\rVert}_{0,\Omega} \leq& {\left\lVert (\partial_i N_l)^\epsilon {\mathbf{S}}_\epsilon(\partial_l {\mkern 1.5mu\overline{\mkern-1.5muu\mkern-1.5mu}\mkern 1.5mu}_0)\theta_\epsilon \right\rVert}_{0,\Omega} + {\left\lVert N_l^\epsilon \epsilon \partial_i \theta_\epsilon {\mathbf{S}}_\epsilon(\partial_l {\mkern 1.5mu\overline{\mkern-1.5muu\mkern-1.5mu}\mkern 1.5mu}_0) \right\rVert}_{0,\Omega}\\
&+{\left\lVert \epsilon N_l \theta_\epsilon {\mathbf{S}}_\epsilon(\partial_{il}^2 {\mkern 1.5mu\overline{\mkern-1.5muu\mkern-1.5mu}\mkern 1.5mu}_0) \right\rVert}_{0,\Omega} \\
\leq& C(\Omega, \kappa_1, \kappa_2){\left( \epsilon^{1/2}{\left\lVert u_0 \right\rVert}_{2,\Omega}+\epsilon {\left\lVert u_0 \right\rVert}_{2,\Omega} \right)} \leq C(\Omega, \kappa_1, \kappa_2) \epsilon^{1/2} {\left\lVert u_0 \right\rVert}_{2,\Omega}
\end{aligned}.$$ Substitute $v=u_\epsilon-\tilde{u}_\epsilon \in V$ into \[lem:key\], and recall $\gamma_j u_0 = \gamma_j \tilde{u_\epsilon}$, we have: $$\begin{aligned}
&\int_{\Omega} A^\epsilon \nabla(u_\epsilon - \tilde{u}_\epsilon) \cdot \nabla (u_\epsilon -\tilde{u}_\epsilon) \\
=& \int_\Omega A^\epsilon \nabla w_\epsilon \cdot \nabla (u_\epsilon -\tilde{u}_\epsilon) + \int_{\Omega} A^\epsilon \nabla r_\epsilon \cdot \nabla (u_\epsilon -\tilde{u}_\epsilon) \\
\leq& j^0(\gamma_j u_0; \gamma_j (u_\epsilon-\tilde{u}_\epsilon))+j^0(\gamma_j u_\epsilon; \gamma_j(\tilde{u}_\epsilon-u_\epsilon))\\
&+C\epsilon^{1/2}{\left\lVert u_0 \right\rVert}_{2,\Omega} {\left\lVert \nabla(u_\epsilon -\tilde{u}_\epsilon) \right\rVert}_{0, \Omega}+\kappa_2 {\left\lVert \nabla r_\epsilon \right\rVert}_{0,\Omega} {\left\lVert \nabla (u_\epsilon-\tilde{u}_\epsilon) \right\rVert}_{0,\Omega} \\
\leq& j^0(\gamma_j \tilde{u}_\epsilon; \gamma_j (u_\epsilon-\tilde{u}_\epsilon))+j^0(\gamma_j u_\epsilon; \gamma_j(\tilde{u}_\epsilon-u_\epsilon))+C\epsilon^{1/2}{\left\lVert u_0 \right\rVert}_{2,\Omega} {\left\lVert \nabla(u_\epsilon -\tilde{u}_\epsilon) \right\rVert}_{0,\Omega}\\
\leq& \alpha_j c_j^2 {\left\lVert u_\epsilon-\tilde{u}_\epsilon \right\rVert}_V^2 + C\epsilon^{1/2} {\left\lVert u_0 \right\rVert}_{2,\Omega} {\left\lVert u_\epsilon-\tilde{u}_\epsilon \right\rVert}_V
\end{aligned}.$$ By a direct calculation, we have ${\left\lVert u_\epsilon-\tilde{u}_{\epsilon} \right\rVert}_V \leq 1/\Delta C\epsilon^{1/2} {\left\lVert u_0 \right\rVert}_{2,\Omega}$. Then ${\left\lVert \nabla w_\epsilon \right\rVert}_{0,\Omega} \leq C\epsilon^{1/2} {\left\lVert u_0 \right\rVert}_{2,\Omega}$, $C$ depends on $\Omega, \kappa_1, \kappa_2, \Delta$, since ${\left\lVert \nabla w_\epsilon \right\rVert}_{0,\Omega} \leq {\left\lVert u_\epsilon-\tilde{u}_\epsilon \right\rVert}_V+ {\left\lVert \nabla r_\epsilon \right\rVert}_{0,\Omega}$.
We have following corollary to quantify the $L^2$ convergence rate:
Let the assumptions \[ass:A\]-\[ass:C\] be satisfied, and $u_0 \in H^2(\Omega)$. Then: $${\left\lVert u_\epsilon - u_0 \right\rVert}_{0,\Omega} \leq C\epsilon^{1/2} {\left\lVert u_0 \right\rVert}_{2,\Omega}.$$ Here the constant $C$ depends on $\Omega,\kappa_1,\kappa_2,\Delta$
Since $\Gamma_D \neq \varnothing$, by Poincaré inequality, ${\left\lVert u_\epsilon-\tilde{u}_\epsilon \right\rVert}_{0,\Omega} \leq C(\Omega) {\left\lVert u_\epsilon-\tilde{u}_\epsilon \right\rVert}_V \leq C\epsilon^{1/2} {\left\lVert u_0 \right\rVert}_{2,\Omega}$. It is easy to show ${\left\lVert \tilde{u}_\epsilon - u_0 \right\rVert}_{0,\Omega}\leq C(\Omega) \epsilon {\left\lVert u_0 \right\rVert}_{1,\Omega}$. Then the triangle relation tells us ${\left\lVert u_\epsilon - u_0 \right\rVert}_{0,\Omega} \leq C\epsilon^{1/2}{\left\lVert u_0 \right\rVert}_{2,\Omega}$.
This $L^2$ convergence rate presented here is not optimal while We conjecture that best results is ${\left\lVert u_\epsilon-u_0 \right\rVert}_{0,\Omega} \leq C\epsilon{\left\lVert u_0 \right\rVert}_{2,\Omega}$ (see [@Shen2016]), and \[sec:experiments\] supports this claim. However, gradient information is far more valuable in application, this is why we mainly consider norm ${\left\lvert \cdot \right\rvert}_{1,\Omega}$ or ${\left\lVert \nabla \cdot \right\rVert}_{0,\Omega}$.
A Special Case—Robin Problem {#sec:robin}
============================
The assumptions \[ass:B1\]-\[ass:C\] are relatively general to cover considerable situations. For example, let us consider a simplified version of \[eq:contact problem hemiform\]:
\[pp:simplified contact problem\] Let $\mathrm{B}(x)$ be a function on ${\mathbb{R}}$ satisfying uniform Lipschitz condition, that is ${\left\lvert \mathrm{B}(x)-\mathrm{B}(y) \right\rvert} \leq L_\mathrm{B} {\left\lvert x-y \right\rvert} \forall x, y \in {\mathbb{R}}$, and the other notations are defined as previous. Then the following hemivariational inequality (variational equality) is solvable: $$\label{eq:simplified contact problem}
\left\{
\begin{aligned}
&\text{Find } u \in V, \text{ s.t. } \forall v \in V \\
&\int_\Omega A \nabla u_\epsilon \cdot \nabla v + \int_{\Gamma_C} \mathrm{B}{\left( u \right)} v = {\left\langle f,v \right\rangle}+\int_{\Gamma_N} gv
\end{aligned} ~~,
\right.$$ if $\kappa_1-L_\mathrm{B} c_j^2 > 0$.
We can reprove this proposition by directly utilizing strongly monotone operator theory (see [@Zeidler1995] sect. 2.14). Moreover, in a special case— Robin problems, we can obtain $L^2(\Omega)$ estimation by utilizing the bilinearity of its variational form.
Let $A^\epsilon$ define as previous and satisfy the assumption \[ass:A\], for simplicity, we state Robin problem on the whole boundary $\Gamma$: $$\left\{
\begin{aligned}
-{\mathrm{div}}(A^\epsilon \nabla u_\epsilon) = f ~~~~ &\text{ in } L^2(\Omega) \\
\bm{n}\cdot A^\epsilon \nabla u_\epsilon + \alpha(\bm{x})u_\epsilon = g ~~~~ &\text{ in } L^2(\Gamma)
\end{aligned} .
\right.$$ And the corresponding variational form is: $$\label{eq:robin variation}
\left\{
\begin{aligned}
&\text{Find } u_\epsilon \in H^1(\Omega), \text{ s.t. } \forall v \in H^1(\Omega) \\
&\int_{\Omega} A^\epsilon \nabla u_\epsilon \cdot \nabla v + \int_{\Gamma} \alpha u_\epsilon v = \int_{\Omega} fv + \int_{\Gamma} gv
\end{aligned} .
\right.$$ We need assumption $$0 < \alpha_1 \leq \alpha(\bm{x}) \leq \alpha_2 < \infty ~~ \forall \bm{x} \in \Gamma$$ to prove this bilinear form is coercive, then Lax-Milgram theorem asserts the solvability of \[eq:robin variation\]. We have following lemma, the proof is postponed in appendix.
\[lem:coercive for Robin\] Let $\Omega$ be a Lipschitz domain and $\Gamma$ its boundary. Then there exists a constant $C(\Omega)$, such that $\forall \psi \in H^1(\Omega)$: $$C(\Omega) {\left\lVert \psi \right\rVert}_{1,\Omega}^2 \leq \int_{\Omega} {\left\lvert \nabla \psi \right\rvert}^2 + \int_{\Gamma} \psi^2 .$$
Similarly, we can prove
The solutions $u_\epsilon$ of Robin problems \[eq:robin variation\] converge weakly to $u^0$, while $u^0$ is the solution of homogenized Robin problem: $$\left\{
\begin{aligned}
-{\mathrm{div}}(\hat{A}\nabla u_0) = f ~~~~ &\text{ in } L^2(\Omega) \\
\bm{n}\cdot \hat{A}\nabla u_0+\alpha(\bm{x})u_0=g ~~~~ &\text{ in } L^2(\Gamma)
\end{aligned} .
\right.$$ And the associated variation form is: $$\label{eq:homogenized robin problem variation form}
\left\{
\begin{aligned}
&\text{Find} ~~ u_0 \in H^1(\Omega) \text{ s.t. } \forall v \in H^1(\Omega)\\
&\int_{\Omega}\hat{A} \nabla u_0 \cdot \nabla v + \int_{\Gamma} \alpha u_0 v = \int_{\Omega}fv + \int_{\Gamma} gv
\end{aligned} .
\right.$$
Estimation \[lem:boundary estimation\] is cited from Thm 1.5.1.10 in [@Grisvard2011], we use this lemma to prove $O(\epsilon^{1/2})$ convergence rate in $H^1(\Omega)$ norm.
\[lem:boundary estimation\] Let $\Omega$ be a Lipschitz domain in ${\mathbb{R}}^d$, then for $u \in H^1(\Omega)$, $$\int_{\Gamma} u^2 \leq C(\Omega) {\left( t\int_{\Omega} {\left\lvert \nabla u \right\rvert}^2 + t^{-1} \int_{\Omega} u^2 \right)} .$$ Here $t$ can choose arbitrarily in $(0, 1)$.
Let $w_\epsilon$ be defined as previous, we have a parallel version of \[lem:key\]:
\[lem:Robin problem estimation 1/2\] Assume $u_0 \in H^2(\Omega)$, then $\forall v \in H^1(\Omega)$ $$\int_\Omega A^\epsilon \nabla w_\epsilon \cdot \nabla v + \int_{\Gamma} \alpha w_\epsilon v \leq C(\Omega, \kappa_1, \kappa_2, \alpha_1, \alpha_2) {\left\lVert u_0 \right\rVert}_{2,\Omega} {\left( \epsilon^{1/2}{\left\lVert \nabla v \right\rVert}_{0,\Omega_{2\epsilon}}+\epsilon{\left\lVert \nabla v \right\rVert}_{0,\Omega}+\epsilon {\left\lVert v \right\rVert}_{0, \Gamma} \right)} .$$
Compare with the proof of \[lem:key\], we are left to show: $$J_3 \coloneqq \int_{\Gamma} \epsilon \alpha N^\epsilon_l {\mathbf{S}}_\epsilon(\partial_l {\mkern 1.5mu\overline{\mkern-1.5muu\mkern-1.5mu}\mkern 1.5mu}_0) v \leq C \epsilon {\left\lVert u_0 \right\rVert}_{2,\Omega} {\left\lVert v \right\rVert}_{0,\Gamma} .$$ By calculation: $${\left\lvert J_3 \right\rvert} \leq C\int_{\Gamma} {\left\lvert \epsilon N^\epsilon_l {\mathbf{S}}_\epsilon(\partial_l {\mkern 1.5mu\overline{\mkern-1.5muu\mkern-1.5mu}\mkern 1.5mu}_0)\theta_\epsilon v \right\rvert} \leq C{\left\lVert \epsilon N^\epsilon_l {\mathbf{S}}_\epsilon(\partial_l {\mkern 1.5mu\overline{\mkern-1.5muu\mkern-1.5mu}\mkern 1.5mu}_0) \right\rVert}_{0,\Gamma} {\left\lVert v \right\rVert}_{0,\Gamma} .$$ Then $$\begin{aligned}
& \int_{\Gamma} {\left\lvert \epsilon N^\epsilon_l {\mathbf{S}}_\epsilon(\partial_l {\mkern 1.5mu\overline{\mkern-1.5muu\mkern-1.5mu}\mkern 1.5mu}_0)\theta_\epsilon \right\rvert}^2\\
\underset{\text{use } \eqref{lem:boundary estimation}}{\leq}& Ct \sum_i \int_\Omega {\left\lvert (\partial_i N_l)^\epsilon {\mathbf{S}}_\epsilon(\partial_l {\mkern 1.5mu\overline{\mkern-1.5muu\mkern-1.5mu}\mkern 1.5mu}_0)\theta_\epsilon \right\rvert}^2+{\left\lvert N^\epsilon_l {\mathbf{S}}_\epsilon(\partial_l {\mkern 1.5mu\overline{\mkern-1.5muu\mkern-1.5mu}\mkern 1.5mu}_0)\epsilon \partial_i \theta_\epsilon \right\rvert}^2+\epsilon^2 {\left\lvert N^\epsilon_l {\mathbf{S}}_\epsilon(\partial^2_{il}{\mkern 1.5mu\overline{\mkern-1.5muu\mkern-1.5mu}\mkern 1.5mu}_0)\theta_\epsilon \right\rvert}^2 \\
&+ t^{-1}\epsilon^2 \int_{\Omega} {\left\lvert N^\epsilon_l {\mathbf{S}}_\epsilon(\partial_l {\mkern 1.5mu\overline{\mkern-1.5muu\mkern-1.5mu}\mkern 1.5mu}_0)\theta_\epsilon \right\rvert}^2\\
\underset{\text{take $t=\epsilon$}}{\leq} &C\epsilon \sum_i \int_{\tilde{\Omega}_{2\epsilon}}{\left\lvert (\partial_i N_l)^\epsilon {\mathbf{S}}_\epsilon(\partial_l {\mkern 1.5mu\overline{\mkern-1.5muu\mkern-1.5mu}\mkern 1.5mu}_0) \right\rvert}^2+{\left\lvert N^\epsilon_l {\mathbf{S}}_\epsilon(\partial_l {\mkern 1.5mu\overline{\mkern-1.5muu\mkern-1.5mu}\mkern 1.5mu}_0) \right\rvert}^2+\epsilon^2\int_{\Omega}{\left\lvert N^\epsilon_l {\mathbf{S}}_\epsilon(\partial^2_{il}{\mkern 1.5mu\overline{\mkern-1.5muu\mkern-1.5mu}\mkern 1.5mu}_0) \right\rvert}^2 \\
&+ \epsilon \int_{\tilde{\Omega}_{2\epsilon}} {\left\lvert N^\epsilon_l {\mathbf{S}}_\epsilon(\partial_l {\mkern 1.5mu\overline{\mkern-1.5muu\mkern-1.5mu}\mkern 1.5mu}_0) \right\rvert}^2 \\
\underset{\text{use}\eqref{lem:3.4}\eqref{lem:3.2}}{\leq} &C\epsilon^2 {\left\lVert u \right\rVert}_{2,\Omega}^2+\epsilon^3 {\left\lVert u_0 \right\rVert}_{2,\Omega}^2 \leq C\epsilon^2 {\left\lVert u_0 \right\rVert}^2_{2,\Omega}
\end{aligned} .$$
Then, we have:
\[thm:Robin weakly conv\] Let $w_\epsilon = u_\epsilon - u_0-\epsilon N^\epsilon_l {\mathbf{S}}_\epsilon(\partial_l {\mkern 1.5mu\overline{\mkern-1.5muu\mkern-1.5mu}\mkern 1.5mu}_0)$, $u_\epsilon$ and $u_0$ be the solution of \[eq:robin variation\] and \[eq:homogenized robin problem variation form\] respectively. Assume $u_0(\Omega) \in H^2(\Omega)$, then $$\int_{\Omega} {\left\lvert \nabla w_\epsilon \right\rvert}^2 + \int_{\Gamma} \alpha w_\epsilon^2 \leq C(\Omega, \kappa_1, \kappa_2, \alpha_1, \alpha_2) \epsilon {\left\lVert u_0 \right\rVert}_{2,\Omega}^2 .$$
As a corollary, we have $$\begin{aligned}
{\left\lVert \nabla w_\epsilon \right\rVert}_{0,\Omega} &\leq C \epsilon^{1/2} {\left\lVert u_0 \right\rVert}_{2,\Omega} , \\
{\left\lVert u_\epsilon - u_0 \right\rVert}_{0,\Omega} &\leq C \epsilon^{1/2} {\left\lVert u_0 \right\rVert}_{2,\Omega} .
\end{aligned}$$
Next, we will show that ${\left\lVert u_\epsilon - u_0 \right\rVert}_{0,\Omega}$ can reach $O(\epsilon)$. We need a regularity result for Robin boundary problem:
Suppose that $\Omega$ has $C^{1,1}$ boundary. In addition to uniformly ellipticity, coefficients $A(x)={\left[ A_{ij}\right]}_{1\leq i,j \leq d}$ are in $C^{0,1}(\Omega)$, and $\alpha(\bm{x})$ is $C^{0,1}(\Gamma)$ (in the sense of local coordinate). Then $\forall f \in L^2(\Omega)$, $u$ is the solution of Robin problem: $$\left\{
\begin{aligned}
-{\mathrm{div}}(A \nabla u) = f \\
\bm{n}\cdot A \nabla u + \alpha(\bm{x})u = 0
\end{aligned} .
\right.$$ Then $u \in H^2(\Omega)$ with estimation ${\left\lVert u \right\rVert}_{2,\Omega} \leq C {\left\lVert f \right\rVert}_{0,\Omega}$, here $C$ depends on $\Omega, {\left\lVert A_{ij} \right\rVert}_{C^{0,1}(\Omega)}, {\left\lVert \alpha \right\rVert}_{C^{0,1}(\Gamma)}$ and $\kappa_1, \kappa_2$.
A proof is provided in the appendix. The $L^2$ estimation states as following:
Suppose that $\Omega$ has $C^{1,1}$ boundary. In addition to the hypotheses in \[thm:Robin weakly conv\], $\alpha(\bm{x})$ is uniformly Lipschitz continuous on $\Gamma$. Then we have: $${\left\lVert u_\epsilon - u_0 \right\rVert}_{0,\Omega} \leq C \epsilon {\left\lVert u_0 \right\rVert}_{2,\Omega}.$$ Where $C=C(\Omega, \kappa_1, \kappa_2, {\left\lVert \alpha \right\rVert}_{C^{0,1}(\Gamma)})$.
It is sufficient to show ${\left\lVert w_\epsilon \right\rVert}_{0,\Omega} \leq C\epsilon {\left\lVert u_0 \right\rVert}_{2,\Omega}$, because of the fact that ${\left\lVert \epsilon N_l {\mathbf{S}}_\epsilon (\partial_l {\mkern 1.5mu\overline{\mkern-1.5muu\mkern-1.5mu}\mkern 1.5mu}_0) \right\rVert}_{0,\Omega} \leq C\epsilon {\left\lVert u_0 \right\rVert}_{2,\Omega}$.
We will take duality technique from [@Shen2016] for the rest proof. First $\forall G \in L^2(\Omega)$, we have $\rho \in H^2(\Omega)$ which satisfies homogenized Robin problem: $$\left\{
\begin{aligned}
-{\mathrm{div}}(\hat{A} \nabla \rho_0) = G \\
\bm{n}\cdot \hat{A} \nabla \rho_0 + \alpha(\bm{x})\rho_0 = 0
\end{aligned} .
\right.$$ We also let $\rho_\epsilon$ be the solution of original Robin problem: $$\left\{
\begin{aligned}
-{\mathrm{div}}(A^\epsilon \nabla \rho_\epsilon) = G \\
\bm{n}\cdot A^\epsilon \nabla \rho_\epsilon + \alpha(\bm{x})\rho_\epsilon = 0
\end{aligned} .
\right.$$ According to the regularity result, we have ${\left\lVert \rho \right\rVert}_{2,\Omega} \leq C {\left\lVert G \right\rVert}_{0,\Omega}$. Take $w_\epsilon$ as test function into previous equation, $$\int_{\Omega} G w_\epsilon = \int_{\Omega} A^\epsilon \nabla \rho_\epsilon \cdot \nabla w_\epsilon + \int_\Gamma \alpha \rho_\epsilon w_\epsilon .$$ Split $\rho_\epsilon$ into three parts $\rho_\epsilon = \rho + \epsilon N_l {\mathbf{S}}_\epsilon(\partial_l {\mkern 1.5mu\overline{\mkern-1.5mu\rho\mkern-1.5mu}\mkern 1.5mu}) + {\left( \rho_\epsilon - \rho - \epsilon N_l {\mathbf{S}}_\epsilon(\partial_l {\mkern 1.5mu\overline{\mkern-1.5mu\rho\mkern-1.5mu}\mkern 1.5mu}) \right)} \eqqcolon \rho + \psi_\epsilon + \eta_\epsilon$. From \[thm:Robin weakly conv\], we will have ${\left\lVert \nabla \eta_\epsilon \right\rVert}_{0,\Omega} \leq \epsilon^{1/2} {\left\lVert \rho \right\rVert}_{2,\Omega}$ and ${\left\lVert \eta_\epsilon \right\rVert}_{0,\Gamma} \leq \epsilon^{1/2} {\left\lVert \rho \right\rVert}_{2,\Omega}$. Then, $$\begin{aligned}
\int_{\Omega} G w_\epsilon =& \int_\Omega A^\epsilon \nabla \rho \cdot \nabla w_\epsilon + \int_\Gamma \alpha \rho w_\epsilon \\
&+ \int_\Omega A^\epsilon \nabla \psi_\epsilon \cdot \nabla w_\epsilon + \int_\Gamma \alpha \psi_\epsilon w_\epsilon \\
&+ \int_\Omega A^\epsilon \nabla \eta_\epsilon \cdot \nabla w_\epsilon + \int_\Gamma \alpha \eta_\epsilon w_\epsilon \\
\coloneqq & J_1+J_2+J_3 .
\end{aligned}$$ For $J_3$, we have ${\left\lvert J_3 \right\rvert} \leq C\epsilon {\left\lVert \rho \right\rVert}_{2,\Omega}{\left\lVert u_0 \right\rVert}_{2,\Omega}$. Use \[lem:Robin problem estimation 1/2\], we obtain: $$\begin{aligned}
{\left\lvert J_1 \right\rvert} &\leq C{\left\lVert u_0 \right\rVert}_{2,\Omega} {\left( \epsilon^{1/2}{\left\lVert \nabla \rho \right\rVert}_{0, \Omega_{2\epsilon}} + \epsilon {\left\lVert \nabla \rho \right\rVert}_{0,\Omega} + \epsilon {\left\lVert \rho \right\rVert}_{0,\Gamma} \right)} \\
&\leq C{\left\lVert u_0 \right\rVert}_{2,\Omega} {\left( \epsilon{\left\lVert \rho \right\rVert}_{2, \Omega} + \epsilon {\left\lVert \nabla \rho \right\rVert}_{0,\Omega} + \epsilon {\left\lVert \rho \right\rVert}_{0,\Gamma} \right)} \leq C\epsilon {\left\lVert \rho \right\rVert}_{2,\Omega} {\left\lVert u_0 \right\rVert}_{2,\Omega}
\end{aligned}.$$ Similarly, ${\left\lvert J_2 \right\rvert} \leq C\epsilon {\left\lVert \rho \right\rVert}_{2,\Omega} {\left\lVert u_0 \right\rVert}_{2,\Omega}$. Together, we have $$\int_\Omega G w_\epsilon \leq C \epsilon {\left\lVert \rho \right\rVert}_{2,\Omega} {\left\lVert u_0 \right\rVert}_{2,\Omega} \leq C \epsilon {\left\lVert G \right\rVert}_{0,\Omega} {\left\lVert u_0 \right\rVert}_{2,\Omega}$$.
Computational Method {#sec:computation}
====================
After the completing of $O(\epsilon^{1/2})$ estimation, \[eq:contact problem hemiform\] will be computable because $u_0+\epsilon N^\epsilon_l{\mathbf{S}}_\epsilon(\partial_l {\mkern 1.5mu\overline{\mkern-1.5muu\mkern-1.5mu}\mkern 1.5mu}_0)$ can approximate well to original high oscillating $u_\epsilon$. However, obtain ${\mkern 1.5mu\overline{\mkern-1.5muu\mkern-1.5mu}\mkern 1.5mu}_0$ and perform smoothing action ${\mathbf{S}}_\epsilon$ is impractical in real computation. Instead, we should calculate $\partial_i u_0 + (\partial_i N_l)^\epsilon \partial_l u_0$ as an approximation for $\partial_i u_\epsilon$. Here is a lemma for the error analysis.
\[lem:gradient error analysis\] Let $u_\epsilon$ and $u_0$ be the solution of \[eq:contact problem hemiform\] and \[eq:contact problem homohemiform\] respectively, and assumptions \[ass:A\]-\[ass:C\] be satisfied, and assume $u_0 \in H^2(\Omega), N_l \in W^{1,\infty}_\sharp(Q)$. Then: $$\sum_i \int_{\Omega}{\left\lvert \partial_i u_\epsilon-\partial_i u_0 -(\partial_i N_l)^\epsilon \partial_l u_0 \right\rvert}^2 \leq \epsilon C(\Omega, \kappa_1, \kappa_2, \Delta, {\left\lVert N_l \right\rVert}_{W^{1,\infty}(Q)}) {\left\lVert u_0 \right\rVert}_{2,\Omega}^2 .$$
Directly calculate the error $\partial_i w_\epsilon = \partial_i u_\epsilon - \partial_i u_0 - (\partial_i N_l)^\epsilon {\mathbf{S}}_\epsilon(\partial_l {\mkern 1.5mu\overline{\mkern-1.5muu\mkern-1.5mu}\mkern 1.5mu}_0) - \epsilon N_l^\epsilon {\mathbf{S}}_\epsilon(\partial^2_{il}{\mkern 1.5mu\overline{\mkern-1.5muu\mkern-1.5mu}\mkern 1.5mu}_0)$, \[lem:3.2\] tells us ${\left\lVert N_l^\epsilon {\mathbf{S}}_\epsilon(\partial^2_{il}{\mkern 1.5mu\overline{\mkern-1.5muu\mkern-1.5mu}\mkern 1.5mu}_0) \right\rVert}_{0,\Omega} \leq C {\left\lVert u_0 \right\rVert}_{2, \Omega}$. By Hölder inequality, we have $$\begin{aligned}
\int_{\Omega} {\left\lvert (\partial_i N_l)^\epsilon \right\rvert}^2 {\left\lvert {\mathbf{S}}_\epsilon (\partial_l {\mkern 1.5mu\overline{\mkern-1.5muu\mkern-1.5mu}\mkern 1.5mu}_0)-\partial_l u_0 \right\rvert}^2 \leq& C \sum_l \int_{{\mathbb{R}}^d} {\left\lvert {\mathbf{S}}_\epsilon (\partial_l {\mkern 1.5mu\overline{\mkern-1.5muu\mkern-1.5mu}\mkern 1.5mu}_0)-\partial_l {\mkern 1.5mu\overline{\mkern-1.5muu\mkern-1.5mu}\mkern 1.5mu}_0 \right\rvert}^2 \leq C\epsilon^2 \int_{{\mathbb{R}}^d} {\left\lvert \nabla^2 {\mkern 1.5mu\overline{\mkern-1.5muu\mkern-1.5mu}\mkern 1.5mu}_0 \right\rvert}^2 \\
\leq& C \epsilon^2 {\left\lVert u_0 \right\rVert}_{2,\Omega}^2
\end{aligned}.$$ Then the conclusion holds because ${\left\lVert \nabla w_\epsilon \right\rVert}_{0,\Omega}$ dominates the error: $$\begin{aligned}
&\sum_i \int_{\Omega} {\left\lvert \partial_i u_\epsilon-\partial_i u_0 -(\partial_i N_l)^\epsilon \partial_l u_0 \right\rvert}^2 \\
\leq & \int_{\Omega} {\left\lvert \nabla w_\epsilon \right\rvert}^2\\
&+\sum_i\epsilon^2 \int_{\Omega} {\left\lvert N_l^\epsilon {\mathbf{S}}_\epsilon(\partial^2_{il}{\mkern 1.5mu\overline{\mkern-1.5muu\mkern-1.5mu}\mkern 1.5mu}_0) \right\rvert}^2 + \sum_i \int_{\Omega} {\left\lvert (\partial_i N_l)^\epsilon \right\rvert}^2 {\left\lvert {\mathbf{S}}_\epsilon (\partial_l {\mkern 1.5mu\overline{\mkern-1.5muu\mkern-1.5mu}\mkern 1.5mu}_0)-\partial_l u_0 \right\rvert}^2
\end{aligned}.$$
It seems that we can not weaken the regularity assumption for $N_l(\bm{y})$ because we can not prove a strengthened version of \[lem:3.2\], that is:
> Let $f \in L_{\text{loc}}^2({\mathbb{R}}^d)$ be a 1-periodic function, Then for any $u \in H^1({\mathbb{R}}^d)$, $${\left\lVert f^\epsilon {\left( {\mathbf{S}}_\epsilon u - u \right)} \right\rVert}_{0,{\mathbb{R}}^d} \leq C \epsilon {\left\lVert f \right\rVert}_{0, Q} {\left\lVert \nabla u \right\rVert}_{0,{\mathbb{R}}^d}.$$
We also mention that when the coefficients $A(\bm{y})$ is piecewise smooth, which is a suitable assumption in application, and the $W^{1,\infty}$ proposition can be verified by the works in [@Li2003].
We can implement finite element method (FEM) to obtain the numerical solution of $u_0$. Let $V_h$ be the finite element space (see [@Brenner2008]), then $$\label{eq:contact problem homofem}
\left\{
\begin{aligned}
&\text{Find } u_{0,h} \in V_h, \text{ s.t. } \forall v_h \in V \\
& \int_\Omega \hat{A} \nabla u_{0,h} \cdot \nabla v_h + j^0(\gamma_j u_{0,h}; \gamma_j v_h) \geq {\left\langle \tilde{f},v_h \right\rangle}
\end{aligned} .
\right.$$ The existence and uniqueness of this problem were also shown by utilizing the framework in [@Han2017]. Then the computational method is direct:
Solve the equations on correctors \[eq:correctors\] and obtain numerical solution $N_l^\star(\bm{y})$. Calculate numerical homogenized coefficients $\hat{A}^\star$. In this step, the cost of computation is independent with original problem, thus we can implement high accuracy method.
Choose grid size $h$ to mesh the domain $\Omega$. Solve \[eq:contact problem homofem\] and obtain $u_{0,h}$.
Construct numerical gradient value $(\partial_i u_\epsilon)^\star$ by $$(\partial_i u_\epsilon)^\star(\bm{x}) \coloneqq \partial_i u_{0,h}(\bm{x}) + \partial_i N_l^\star(\bm{x}/\epsilon) \partial_l u_{0,h}(\bm{x}) .$$
Rewrite the error analysis \[lem:gradient error analysis\] in FEM framework, we derive following numerical error expression: $${\left\lVert \nabla u_\epsilon-(\nabla u_\epsilon)^\star \right\rVert}_{0,\Omega} \leq C{\left( \epsilon^{1/2} {\left\lVert u_0 \right\rVert}_{2,\Omega}+{\left\lvert u_0-u_{0,h} \right\rvert}_{1,\Omega} \right)},$$ here $(\nabla u_\epsilon)^\star={\left[ (\partial_i u_\epsilon)^\star\right]}_{1\leq i \leq d}={\left[ \partial_i u_{0,h}(\bm{x}) + \partial_i N_l^\star(\bm{x}/\epsilon) \partial_l u_{0,h}(\bm{x})\right]}_{1\leq i \leq d}$. The estimation for ${\left\lvert u_0-u_{0,h} \right\rvert}_{1,\Omega}$ merely involves the theory of FEM. Fortunately, a Céa’s inequality has been proved in [@Han2017] sect. 4.2, and combine their works, we have following theorem:
Assume $u_0(\bm{x}) \in H^2(\Omega)$ and $N_l(\bm{y}) \in W^{1,\infty}_\sharp(Q)$. Neglect the error brought by calculating numerical correctors $N^\star_l(\bm{y})$ and homogenized coefficients $\hat{A}^\star$. Use Lagrange FEM to Solve \[eq:contact problem homofem\] with $h$ grid size. Recall $\gamma_ju_0 \in H^{3/2}(\Gamma_C)$ and ${\left\lVert \gamma_ju_0 \right\rVert}_{3/2,\Gamma_C} \leq C {\left\lVert u_0 \right\rVert}_{2,\Omega}$. Then error between numerical gradient $(\nabla u_\epsilon)^\star$ and $\nabla u_\epsilon$ satisfies the relation: $${\left\lVert \nabla u_\epsilon-(\nabla u_\epsilon)^\star \right\rVert}_{0,\Omega} \leq C{\left[ (\epsilon^{1/2}+h) {\left\lVert u_0 \right\rVert}_{2,\Omega}+h^{3/4} \sqrt{{\left\lVert u_0 \right\rVert}_{2,\Omega}}\right]} .$$ If $\gamma_ju_0 \in H^2(\Gamma_C)$, there exists optimal numerical error order: $${\left\lVert \nabla u_\epsilon-(\nabla u_\epsilon)^\star \right\rVert}_{0,\Omega} \leq C (h+\epsilon^{1/2}) {\left( {\left\lVert u_0 \right\rVert}_{2,\Omega}+ \sqrt{{\left\lVert \gamma_ju_0 \right\rVert}_{2, \Gamma_C}} \right)},$$ here the constant $C$ depends on $\Omega, \kappa_1, \kappa_2, \Delta, {\left\lVert N_l \right\rVert}_{W^{1,\infty}(Q)}$ and quality of the mesh (see [@Brenner2008] for detail description).
For Robin problem, the conclusion is more elegant:
Assume $u_0(\bm{x}) \in H^2(\Omega)$ and $N_l(\bm{y}) \in W^{1,\infty}_\sharp(Q)$. Neglect the error brought by calculating numerical correctors $N^\star_l(\bm{y})$ and homogenized coefficients $\hat{A}^\star$. Use Lagrange FEM to Solve \[eq:homogenized robin problem variation form\] with $h$ grid size. The numerical error is: $${\left\lVert \nabla u_\epsilon-(\nabla u_\epsilon)^\star \right\rVert}_{0,\Omega} \leq C (h+\epsilon^{1/2}) {\left\lVert u_0 \right\rVert}_{2,\Omega},$$ here the constant $C$ depends on $\Omega, \kappa_1, \kappa_2, \alpha_1, \alpha_2, {\left\lVert N_l \right\rVert}_{W^{1,\infty}(Q)}$ and quality of the mesh.
Numerical Experiments {#sec:experiments}
=====================
Generally, when to solve contact problems described by hemivariational inequalities (for examples, \[eq:contact problem homohemiform\] and \[eq:contact problem hemiform\]), we need to convert the original problems to an optimization problem and use optimizaiton algorithm. Inspired by [@Barboteu2013], we discover an iterative method to solve simplified problem \[eq:simplified contact problem\], and it performs effectively in our numerical experiment.
Mesh the domain $\Omega$ and construct FEM space $V_h$. Then set tolerance $\mathit{tol}$ and initial solutions $u^{(0)}_h$. Solve a variational problem: \[al:1\] $$\left\{
\begin{aligned}
&\text{Find } u^{(n+1)}_h \in V, \text{ s.t. } \forall v_h \in V_h \\
&\int_\Omega A \nabla u^{(n+1)}_h \cdot \nabla v_h = {\left\langle f,v_h \right\rangle}+\int_{\Gamma_N} gv_h - \int_{\Gamma_C} \mathrm{B}{\left( u^{(n)}_h \right)} v_h
\end{aligned}
\right. ,$$ which is equivalent to solve a linear system: $$A_h u^{(n+1)} = b_h + r_h{\left( u^{(n)} \right)} .$$ Loop until [${\left\lVert u^{(n+1)}_h - u^{(n)} \right\rVert} < \mathit{tol} {\left\lVert u^{(n)} \right\rVert}$]{}.
Actually, One can prove if \[pp:simplified contact problem\] holds, the algorithm above will converge linearly to the real solution. The technique used here is to elucidate nonlinear map $u^{(n)} \mapsto u^{(n+1)}$ is contractive, and we omit the details here.
Now, we can set up our experiment problem:
Take $\Omega$ as square $(0,1)\times(0,1)$, and partition $\Omega$ into $N\times N$ whole cells. Hence, in this case $\epsilon = 1/N$. We set $A^\epsilon(\bm{x})=\kappa^\epsilon(\bm{x}) I$, $I$ is identity matrix, and $\kappa^\epsilon(\bm{x})$ can merely take two values respectively in different subdomains on each cell. Here a figure to illustrate those relations:
(-2.0, -1) – (-1, -1); (-2.0, -1) – (-2.0, 0); at (-0.7, -1) [$x$]{}; at (-2.0, 0.3) [$y$]{}; (0, 0) rectangle (4, 4); (0, 0) – (0, -0.3) (2, 0) – (2, -0.3) (4, 0) – (4, -0.3); (0, -0.15) – (0.5, -0.15); (2, -0.15) – (2.5, -0.15); (1.5, -0.15) – (2.0, -0.15); (3.5, -0.15) – (4.0, -0.15); (0, 0) – (-0.3, 0) (0, 4) – (-0.3, 4); (-0.15, 4) – (-0.15, 2.5); (-0.15, 1.5) – (-0.15, 0); (4, 0) – (4.3, 0) (4, 4) – (4.3, 4); (4.15, 4) – (4.15, 2.5); (4.15, 1.5) – (4.15, 0); (0, 4) – (0, 4.3) (4, 4) – (4, 4.3); (0, 4.15) – (1.5, 4.15); (2.5, 4.15) – (4, 4.15); at (1.0, -0.25) [$\Gamma_{C'}$]{}; at (3.0, -0.25) [$\Gamma_{C''}$]{}; at (-0.25, 2) [$\Gamma_N$]{}; at (4.3, 2) [$\Gamma_N$]{}; at (2.0, 4.25) [$\Gamma_D$]{};
in [1,...,4]{} in [1,...,4]{} [ (-1, -1) rectangle (, ); (-0.75, -0.75) rectangle (-0.25, -0.25); ]{} iin [1,...,3]{} [ (0, i) – (4, i); (i, 0) – (i, 4); ]{}
(7,0) rectangle (11, 4); (8,1) rectangle (10, 3); (7, -0.3) – (7, 4.3) (6.7, 4) – (11.3, 4) (11, 4.3) – (11, -0.3) (11.3, 0.0) – (6.7, 0); (7, 2) – (7.25, 2); (7.75, 2) – (8, 2); at (7.5, 2) [$\rho$]{}; (9, 0) – (9, 0.25); (9, 1) – (9, 0.75); at (9, 0.5) [$\rho$]{}; at (9, 2) [$\kappa(\bm{y}) \equiv \kappa_2$]{}; at (9, 3.5) [$\kappa(\bm{y}) \equiv \kappa_1$]{};
(4.3, 2.6) – (6.6, 3.6); (4.3, 1.4) – (6.6, 0.4); at (5.5, 2.0) [$\bm{y}=\bm{x}/\epsilon$]{};
at (2.0, 2.0) [$\Omega$]{};
Boundary $\Gamma$ is divided into four parts $\Gamma_D, \Gamma_N, \Gamma_{C'}, \Gamma_{C''}$, and the boundary conditions are expressed in following variational problem: $$\label{eq:numerical experiment}
\left\{
\begin{aligned}
&\text{Find } u^\epsilon \in V= \{ v \in H^1(\Omega): v \equiv 0 \text{ on } \Gamma_D\}, \text{ s.t. } \forall v \in V \\
& \int_\Omega \kappa^\epsilon(\bm{x}) \nabla u^\epsilon \cdot \nabla v + \alpha \int_{\Gamma_{C'}} u^\epsilon v + \alpha \int_{\Gamma_{C''}} {\left( u^\epsilon \right)}^+v = g \int_{\Gamma_N} v + f \int_{\Omega} v
\end{aligned}
\right. .$$ Here for simplicity, we set $\alpha, g, f$ as constants. To guarantee that this problem is solvable, we give following proposition:
If $\kappa_1 > {\left\lvert \alpha \right\rvert}$, then the solution of \[eq:numerical experiment\] exists.
According to \[ass:C\], we are left to show $c_j \leq 1$. Use the fact that $u(x, 1)\equiv 0$, we have $$\int_0^1 {\left\lvert u(x, 0) \right\rvert}^2 {\mathrm{d}}x = \int_0^1 {\left\lvert \int_{0}^{1} \partial_y u(x,y) {\mathrm{d}}y \right\rvert}^2 {\mathrm{d}}x \leq \int_{0}^{1} \int_{0}^{1} {\left\lvert \partial_y u(x, y) \right\rvert}^2 {\mathrm{d}}x {\mathrm{d}}y ~ ,$$ and this gives $c_j \leq 1$.
We slice each cell equally into $M \times M$ elements, therefore, we actually solve the original problem \[eq:numerical experiment\] and its homogenized version on a $NM\times NM$ grid. Our numerical experiment focus on verifying the homogenization error, we use following notation to measure the errors: $$\begin{aligned}
\mathbf{ERR_0} &\coloneqq {\left\lVert u_\epsilon - u_0 \right\rVert}_{0, \Omega}/ {\left\lVert u_0 \right\rVert}_{0,\Omega} \\
\mathbf{ERR_1} &\coloneqq {\left[ \sum_i{\left\lVert \partial_i u_\epsilon - \partial_i u_0 -(\partial_i N_l)^\epsilon \partial_l u_0 \right\rVert}_{0, \Omega}^2\right]}^{1/2} / {\left\lvert u_0 \right\rvert}_{1, \Omega} \\
\mathbf{ERR_2} &\coloneqq {\left\lvert u_\epsilon - u_0 \right\rvert}_{1,\Omega}/ {\left\lvert u_0 \right\rvert}_{1,\Omega}
\end{aligned} ~~ .$$ We set $\alpha=0.5, f=1.0, g=1.0, \kappa_1=1.0, \kappa_2 = 2.0$ and list the results in \[tab:1\].
$\mathbf{ERR_0}$ $\mathbf{ERR_1}$ $\mathbf{ERR_2}$
------------------ ------------------ ------------------ ------------------
$N=16, M=128$ $0.00328$ $0.00737$ $0.21898$
$N=32, M=64$ $0.00164$ $0.00480$ $0.21899$
$N=64, M=32$ $0.00082$ $0.00331$ $0.21886$
$N=128, M=16$ $0.00049$ $0.00233$ $0.21843$
Convergence rate $0.92$ $0.55$ -
$N=32, M=128$ $0.00164$ $0.00480$ $0.21904$
: The results of numerical experiments[]{data-label="tab:1"}
From this table, the numerical convergence rate is actually close to its theoretical value $1.0$ and $0.5$, some differences may credit to that grid get coarser as heterogeneity or $1/\epsilon$ increase. In the last row of the table, we show the case $N=32, M=128$ as a compare to $N=32, M=64$, and the difference is few. This means that the $\mathbf{ERR_0}, \mathbf{ERR_1}$ and $\mathbf{ERR_2}$ we compute are accurate enough when $M$ is not small. Because of the limitation on computation resource, further and refined experiments such as $N=64, M=128$ or $N=128, M=128$ do not get conducted. We also notice that $\mathbf{ERR_2}$ do not decrease, This observation convinces the necessity of using First-order asymptotic solution.
Conclusions
===========
To model real scientific or engineering problems, Only studying the Dirichlet or Neumann boundary conditions is not completely adequate, and the situation has been encountered commonly in contact problems. Hence, the study on more suitable boundary condition is needed. A hemivariational inequalities framework for contact problems has been developed and also proved to be effective. Many physical and mechanical phenomena occur in highly heterogeneous media, and the simplified occasion is setting the coefficients of governing PDEs to have small periodicity. Contact problems in small periodicity setting have two major difficulties: one comes from nonlinearity in hemivariational inequalities, and the other originates from high oscillation due to multiscale property.
In this paper, several relatively reasonable assumptions are postulated to make the problems well posed, and a homogenization theorem is obtained by div-curl lemma. The key part is to derive $O(\epsilon^{1/2})$ estimation, and this result quantifies the convergence rate for first order expansion. Then, a computational method is proposed, and its numerical accuracy is also analyzed in FEM framework. We examine the special case— Rubin problem and find out that an optimal $L^2$ estimation is obtainable.
It should be emphasized that, direct computational methods will cost enormous resources because of nonlinearity and high heterogeneity in this problem. It leads to the development of specialized computational methods. A thorough comparison of these two approaches and nontrivial numerical experiments will be more persuasive, and it will be provided in the future work.
Let $\Omega$ be a Lipschitz domain and $\Gamma$ its boundary. Then there exists a constant $C(\Omega)$, such that $\forall \psi \in H^1(\Omega)$: $$C(\Omega) {\left\lVert \psi \right\rVert}_{1,\Omega}^2 \leq \int_{\Omega} {\left\lvert \nabla \psi \right\rvert}^2 + \int_{\Gamma} \psi^2 .$$
If not, we have a sequence of $\{ \psi_n\}$ with ${\left\lVert \psi_n \right\rVert}_{1,\Omega}=1$ and $\int_{\Omega} {\left\lvert \nabla \psi_n \right\rvert}^2 + \int_{\Gamma} \psi_n^2 \leq 1/n$. Up to a subsequence, we will have: $$\left\{
\begin{aligned}
& \psi_n \rightarrow \psi_0 ~~~~ \text{ in } ~~ L^2(\Omega) \\
& \nabla \psi_n \rightharpoonup \nabla \psi_0 ~~~~ \text{ in } ~~ L^2(\Omega)^d
\end{aligned} ~~ .
\right.$$ Since $\int_{\Omega} {\left\lvert \nabla \psi_n \right\rvert}^2 \leq 1/n$, we obtain $\nabla \psi_n \rightarrow \bm{0}$ in $L^2(\Omega)^d$. We now have $\nabla \psi_0 \equiv \bm{0}$, and $\psi_0 \equiv C$. By $\int_\Omega \psi_0^2=\lim\limits_{n}\int_{\Omega} \psi_n^2 = 1-\lim\limits_{n}\int_{\Omega} {\left\lvert \nabla \psi_n \right\rvert}^2=1$, we know $C \neq 0$. Due to trace theorem, ${\left\langle u,\phi \right\rangle}=\int_{\Gamma}u\phi$ is a bounded functional on $H^1(\Omega)$ for any $u \in H^1(\Omega)$. Use weak convergence we have $\int_{\Gamma} \psi_0^2 = \lim\limits_{n} \int_{\Gamma} \psi_0 \psi_n \leq {\left\lVert \psi_0 \right\rVert}_{0,\Gamma} \liminf_n {\left\lVert \psi_n \right\rVert}_{0,\Gamma}$, and $$0 = \lim\limits_{n} {\left\lvert \nabla \psi_n \right\rvert}^2 + \int_{\Gamma} \psi_n^2 \geq \liminf_n \int_{\Gamma} \psi_n^2 \geq \int_{\Gamma} \psi_0^2 = \int_{\Gamma} C^2 ,$$ and this contradicts $C\neq 0$.
Suppose that $\Omega$ has $C^{1,1}$ boundary. In addition to uniformly ellipticity, coefficients $A(x)={\left[ A_{ij}\right]}_{1\leq i,j \leq d}$ are in $C^{0,1}(\Omega)$, and $\alpha(\bm{x})$ is $C^{0,1}(\Gamma)$ (in the sense of local coordinate). Then $\forall f \in L^2(\Omega)$, $u$ is the solution of Robin problem: $$\left\{
\begin{aligned}
-{\mathrm{div}}(A \nabla u) = f \\
\bm{n}\cdot A \nabla u + \alpha(\bm{x})u = 0
\end{aligned} .
\right.$$ Then $u \in H^2(\Omega)$ with estimation ${\left\lVert u \right\rVert}_{2,\Omega} \leq C {\left\lVert f \right\rVert}_{0,\Omega}$, here $C$ depends on $\Omega, {\left\lVert A_{ij} \right\rVert}_{C^{0,1}(\Omega)}, {\left\lVert \alpha \right\rVert}_{C^{0,1}(\Gamma)}$ and $\kappa_1, \kappa_2$.
By flatting boundary technique, we only need to consider a half sphere $B_1^+$ in ${\mathbb{R}}^d_+$ as domain, and plane $T=\{\bm{x}\in {\mathbb{R}}^d \colon x_d=0\}$ as boundary where Robin condition imposed. The original problem can locally transfer to following form: $$\int_{B_1^+} A \nabla u \cdot \nabla v + \int_T \alpha u v = \int_{B_1^+} f v ~~~~ \forall v \in C^\infty(B_1^+) \text{ s.t. } \mathrm{dist}{\left( {\mathrm{supp}}(v), \partial B_1^+ \cap {\mathbb{R}}^d_+ \right)} > 0 .$$ Take differential quotient $\Delta^h=\Delta^h_k$, and by a similar process we will get a result as in [@Gilbarg1983] sect. 8.3, here $k\neq d$: $$\begin{aligned}
&\int_{B_1^+} A(\bm{x}+h\bm{e}_k) \nabla(\Delta ^h u) \cdot \nabla v + \int_T \alpha(\bm{x}+h\bm{e}_k) (\Delta ^h u) v \\
\leq & {\left( {\left\lVert f \right\rVert}_{0, B_1^+}+{\left\lVert A \right\rVert}_{C^{0,1}(B_1^+)}{\left\lVert \nabla u \right\rVert}_{0, B_1^+} \right)} {\left\lVert \nabla v \right\rVert}_{0, B_1^+} + {\left\lVert \alpha \right\rVert}_{C^{0,1}(T)}{\left\lVert u \right\rVert}_{0, T}{\left\lVert v \right\rVert}_{0, T}
\end{aligned}.$$ Take $\eta$ as cut-off function which $\eta \equiv 1$ in $B_{1/2}^+$ and $\eta \equiv 0$ in $B_{1}^+\setminus B_{3/4}^+$, then substitute $\eta^2 \Delta ^h u$ for $v$ into above inequality. We have: $$\int_{B_{1}^+} {\left\lvert \eta \nabla \Delta ^h u \right\rvert}^2 + \int_T {\left\lvert \eta \Delta ^h u \right\rvert}^2 \leq C {\left( {\left\lVert u \right\rVert}_{1, B_{1}^+}+{\left\lVert f \right\rVert}_{0, B_{1}^+} \right)}.$$ This implies ${\left\lVert \partial_{ij}u \right\rVert}_{0, B_{1/2}^+} \leq C{\left( {\left\lVert u \right\rVert}_{1, B_{1}^+}+{\left\lVert f \right\rVert}_{0, B_{1}^+} \right)}$ for $i \neq d$ or $j \neq d$. Recall $$f = -a_{dd} \partial_{dd} u - \sum_{i \neq d \text{ or } j \neq d} a_{ij} \partial_{ij} u - \sum_{ij} \partial_i a_{ij} \partial_j u$$ It then follows ${\left\lVert \partial_{dd}u \right\rVert}_{0, B_{1/2}^+} \leq C{\left( {\left\lVert u \right\rVert}_{1, B_{1}^+}+{\left\lVert f \right\rVert}_{0, B_{1}^+} \right)}$. Finally, we have ${\left\lVert u \right\rVert}_{2, B_{1/2}^+} \leq C {\left\lVert f \right\rVert}_{0, B_{1}^+}$.
[^1]: LSEC, ICMSEC, Academy of Mathematics and Systems Science, Chinese Academy of Sciences, Beijing 100190, China, and School of Mathematical Sciences, University of Chinese Academy of Sciences, Beijing 100049, China
[^2]: LSEC, ICMSEC, Academy of Mathematics and Systems Science, Chinese Academy of Sciences, Beijing 100190, China
[^3]: Version 0.6
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'RE J1034+396, a narrow-line Seyfert-1 active galactic nucleus (AGN), is the first example of AGNs that exhibited a nearly coherent quasi-periodic oscillation (QPO) for the data collected by XMM-Newton in 2007. The spectral behaviors and timing properties of the QPO have been studied since its discovery. We present an analysis of the QPO in RE J1034+396 based on the Hilbert-Huang transform (HHT). Comparing other time-frequency analysis methods, the Hilbert spectrum reveals the variation of the QPO period in great detail. Furthermore, the empirical mode decomposition provides band-pass filtered data that can be used in the O – C and correlation analysis. We suggest that it is better to divide the evolution of the QPO in this observation into three epochs according to their different periodicities. In addition to the periodicities, the correlations between the QPO periods and corresponding mean count rates are also different in these three epochs. Further examining the phase lags in these epochs, we found no significant phase lags between the soft and hard X-ray bands, which is also confirmed in the QPO phase-resolved spectral analysis. Finally, we discuss the indications of current models including a spotted accretion disk, diskoseismology, and oscillation of shock, according to the observed time-frequency and spectral behaviors.'
author:
- 'Chin-Ping Hu, Yi Chou, Ting-Chang Yang, Yi-Hao Su'
bibliography:
- 'reference.bib'
title: 'Tracking the Evolution of QPO in RE J1034+396 Using the Hilbert-Huang Transform'
---
Introduction
============
Quasi-periodic oscillations (QPOs), which contain useful information about the inner accretion disk, are commonly observed in black hole X-ray binaries. Active galactic nuclei (AGNs), as upscaling of black hole X-ray binaries, are expected to have QPO phenomenon. However, AGNs were never significantly found to exhibit QPOs until the first detection in a narrow-line Seyfert 1 (NLS1) galaxy RE J1034+396 made by XMM-Newton observation [@Gierlinski2008]. A QPO with period of 3733 s was found to be coherent with the data segment after $\sim25$ ks of this XMM-Newton observation at a $5.6 \sigma$ confidence level [@Gierlinski2008]. From the observation on bulge stellar velocity dispersion, @Bian2010 estimated that the mass of the supermassive black hole (SMBH) is $(1-4) \times 10^6 M_{\odot}$. This estimation indicated that this QPO corresponds to a high-frequency QPO in the black hole X-ray binary [@Gierlinski2008; @Middleton2009; @Bian2010].
RE J1034+396, as well as other NLS1s, are also known for their strong soft X-ray excess and highly variability [@Puchnarewicz2001]. The spectral behavior may also relate to the presence of QPO. Using the rms spectrum, @Middleton2009 concluded that the energy spectrum of RE J1034+396 could be decomposed by a low-temperature Comptonization of disk emission and a hard power law tail. The QPO is dominated by the variability of the hard power law tail, while the soft component remains constant. Furthermore, @Middleton2011 folded the light curve of different energy bands with the QPO period and found a significant soft lag, which can be interpreted by the reprocessing. @Maitra2010 divided the X-ray photons into high and low phases and found that the low phase spectrum exhibits an absorption edge at $\sim 0.86$ keV, which corresponds to a warm absorber lying at $\sim 9 r_g$ for a $4\times 10^6 M_{\odot}$ black hole.
QPOs in X-ray binaries result in broad peaks in their power spectra. This broadening is probably caused by the dramatic variation in the QPO frequency or a modulation with a relatively stable period but fragmented in time. Fourier analysis is insufficient to study the time variability of frequencies because the frequencies defined in the Fourier spectra are assumed to be constant over the entire observation. Based on the timing analysis technique, adding a moving window is a straightforward solution to investigate the variation of the frequency. Several similar analysis methods were introduced, such as the dynamic power spectrum [@Clarkson2003a], and spectrogram [@Oppenheim1989]. An improved time-frequency analysis method, i.e., the Morlet wavelet analysis, was also invented and widely applied on astronomical time series, especially quasi-periodic modulations in various time scales. For example, @Lachowicz2010 studied the 4 Hz low-frequency QPO in XTE J1550–564 using the wavelet analysis and Matching Pursuit algorithm. They concluded that the QPO is composed of multiple independent oscillations with the same frequencies, but the oscillation was present intermittently. For the case of RE J1034+396, @Czerny2010 found a drift in the QPO central frequency based on the wavelet analysis. A possible $\sim24$ ks time scale of a QPO period variation was also marginally detected but could not be firmly concluded as a periodicity because the data time span was limited. In addition, they also found a positive correlation between the QPO frequencies, which were determined by the frequencies at the peak of wavelet spectrum, and the X-ray fluxes.
Another possible method to investigate the variation of QPO period is the Hilbert-Huang Transform (HHT) proposed by @Huang1998. The HHT has been successfully applied on the superorbital modulation of SMC X-1 [@Hu2011], 11-years sunspot variability [@Barnhart2011], and the search for gravitational waves [@Camp2007]. In the HHT, the instantaneous frequency, which is different from that in the Fourier analysis, is defined as the time derivative of phase function. Thus, the Hilbert spectrum can provide us detailed information in both the time and frequency domains. However, the time interval between the samplings must be much shorter than the variability time scale. The fast oscillations in X-ray binaries, such as the kilohertz QPOs and burst QPOs, cannot be analyzed by the HHT with current observatories. Fortunately, the X-ray intensity of RE J1034+396 is high enough to provide us with a sufficiently sampled X-ray light curve that can be analyzed using the HHT. In addition, HHT can provide us the phase function of the QPO. Thus, phase-resolved spectral analysis, as well as the phase lags between folded light curves of different energy bands are also applicable.
We present our analysis on the evolution of the QPO period for RE J1034+396 as well as the QPO phase-resolved spectral variation. In Section \[obs\], we briefly introduce the observation made by XMM-Newton and the data selection criteria. The time-frequency analysis, including the HHT analysis, O – C result, issues with phase lags, and relationship between the QPO period and X-ray flux, are presented in Section \[analysis\]. The spectral model as well as the phase-resolved spectral analysis are described in Section \[spectral\]. We further discuss different possible scenarios, including the spotted accretion disk model, diskoseismology, and shock oscillation model in Section \[discussion\]. Finally, we summarize our work in Section \[summary\].
Observation {#obs}
===========
The QPO in RE J1034+396 was detected by a $\sim93$ ks XMM-Newton observation on May 31, 2007 (OBSID: 0506440101). Both the MOS and PN detector were operated in full-frame mode. Following the selection criteria of @Gierlinski2008, we extracted the photon events within 45” radius around the source, and then excluded the final $\sim7$ ks data because they were highly affected by the background flare. All the photon arrival times were corrected to the barycenter of the solar system. We further combined the photons collected by MOS1, MOS2, and PN detectors with the energy range between 0.3 and 10 keV to perform a time-frequency analysis.
Time-Frequency Analysis {#analysis}
=======================
In order to have sufficient samplings within one cycle, the X-ray photons were evenly binned into a 64-s resolution light curve shown as the upper panel of \[LC\_LS\]. As described in @Gierlinski2008, the QPO modulation is easily seen. A flux-drop event in which the count rate drops from $\sim6.5$ counts/s to $\sim5$ counts/s occurs between $t\approx40-55$ ks. This event was noticed by @Middleton2009 and believed to be an occultation. In addition, a similar event is also observed in $\sim10-20$ ks although it is less pronounced.
To check the consistency with the Fourier analysis performed by @Gierlinski2008, we applied the Lomb-Scargle periodogram [@Scargle1982; @Press1989] on the entire light curve and segment 2 defined by @Gierlinski2008, i.e., after 25 ks. The power spectra are shown in the lower panel of Figure \[LC\_LS\]. The spectrum of the segment 2 light curve exhibits a strong peak at $P=3733.6\pm3.9$ s, where the error comes from a $10^4$ times Monte Carlo simulation. This period is consistent with the one (3733 s) reported by @Gierlinski2008. On the other hand, the power spectrum of the entire data set shows two significant peaks located at $P = 3794 \pm 19$ s and $P=4095 \pm 4$ s. This suggests that the QPO period is probably not stable during the observation. The presence of these two peaks motivated us to further study its time-frequency properties using the HHT.
HHT Analysis {#HHT_analysis}
------------
To obtain the variation of QPO frequency, we applied a modified version of the empirical mode decomposition (EMD), i.e., the ensemble EMD (EEMD) proposed by @Wu2009, on the combined PN+MOS light curve. The EMD is an iterative “sifting” process for extracting oscillation modes by subtracting the local means from the original data [@Huang1998]. The decomposed components are intrinsic mode functions (IMFs) that satisfy the following two conditions: (1) the number of extrema and the number of zero crossings must be identical or differ by one; and (2) the local mean of the data is zero [@Huang1998; @Hu2011]. The instantaneous frequency obtained by the Hilbert transform is meaningful only if it is calculated from an IMF. EEMD overcomes the mode-mixing problem, i.e., a modulation with the same time scale distributed across different IMFs, by taking the emsemble mean of the IMFs from combinations of the original data and added white noises [@Wu2009]. Because the ensemble means of IMFs may not satisfied the condition of IMF, a post-processing EMD [@Wu2009] was applied on the decomposed components to guarantee that the final result satisfied the IMF criteria.
The light curve, $x(t)$, was decomposed into 10 IMFs, denoted as $c_1$ to $c_{10}$, and a residual, $r_{10}$. The completeness of the IMFs are guaranteed so that the light curve can be represented as:
$$x(t)=\sum_{j=1}^{10}c_j+r_{10}$$
The QPO signal lies within $c_5$ can be examined by its variability time-scale as well as its timing behavior such as its Lomb-Scargle periodogram and dynamic power spectrum. The orthogonality between neighboring components $c_f$ and $c_g$ can be expressed using the orthogonality index $OI_{fg}$ [@Huang1998]:
$$OI_{fg}=\sum_{t}\frac{c_fc_g}{c_f^2+c_g^2}$$
The orthogonality index between $c_4$ and $c_5$ in this decomposition is $OI_{45}=0.033$, whereas the orthogonality index between $c_5$ and $c_6$ is $OI_{56}=0.007$. This means that the QPO signal is mostly concentrated within the IMF $c_5$, and the power barely leaks out to neighboring IMFs. Figure \[EEMD\] shows the original light curve, EEMD high-pass filtered light curve, QPO component, and long-term modulation. The EEMD high-pass filtered light curve is a summation from $c_1$ to $c_4$, which represents the high-frequency noise in the data. The long-term modulation light curve, which represents the variability with time scale longer than the QPO, is a summation from $c_6$ to the residual that contains no extrema.
We then applied the normalized Hilbert transform [@Huang2009] to the QPO component to obtain the instantaneous frequency and amplitude. The instantaneous amplitude $a_j(t)$ can be defined by the upper envelope of the absolute value of an IMF $c_j(t)$. The Hilbert transform of a normalized IMF $X_j(t)=c_j(t)/a_j(t)$ can be represented as $$Y(t)=\frac{1}{\pi}P\int_{-\infty}^{\infty}\frac{X(t')}{t-t'}dt'$$ where P indicates the Cauchy principal value. Therefore, an analytical signal $Z_j(t)$ as well as the angular phase function $\theta(t)$ can be defined as $$Z_j(t)=X_j(t)+iY_j(t)=e^{i\theta_j(t)}$$ Then, the instantaneous angular frequency $\omega_j(t)$ is defined as the time derivative of $\theta_j(t)$. As a result, the data can be expressed as $$x(t)=\sum_{j=1}^{10}a_j(t)\exp\left( i\int\omega_j(t)dt\right) +r_{10}$$
The variation of frequency and amplitude was plotted on a three-dimensional map, as shown in Figure \[HHT\_DPS\] to demonstrate the drift of the QPO frequency and the variation of the QPO amplitude. We also produced the dynamic Lomb-Scargle periodogram for comparison. The window size of the dynamic power spectrum was set to 10 ks and the moving step was 100 s. Instead of the original light curve, we applied the dynamic Lomb-Scargle algorithm on the detrended light curve, which was obtained by subtracting the long-term modulation light curve (Figure \[EEMD\](d)) from the original light curve. Because the window size is approximately the same as the time scale of flux-drop event $\sim40-55$ ks mentioned in @Middleton2009, the power spectra are dominated by the long-term trend, and the QPO signals are strongly depressed when the window moves around here. The dynamic power spectrum is shown as the contour stacked on top of the Hilbert spectrum in Figure \[HHT\_DPS\].
From Figure \[HHT\_DPS\], we found that the Hilbert spectrum agrees with the dynamic power spectrum. Both of them can describe the drift of the QPO frequency for time scales longer than $\sim 10$ ks. The flux-drop event, together with a frequency increase in $\sim40-55$ ks, can be observed in this time-frequency map. Moreover, the Hilbert spectrum shows more details in both the time and frequency domains than that in the dynamic power spectrum. The HHT provides us not only the instantaneous frequency and amplitude, but a well-defined phase function of QPO. We obtained a folded light curve by folding all the data according to the HHT phase. Figure \[fold\_lc\] shows the light curves folded by the HHT phase, as well as ones folded by the cycle length and by a fixed period of 3733 s. The cycle length is the time difference between the neighboring fiducial points, which is defined by the local minima of the EEMD low-pass filtered light curve obtained from a summation of QPO signal and the long-term modulation light curve. This is similar to the definition of local minima in @Trowbridge2007, but we used the EEMD as a local filter instead of a Gaussian filter, and we used the local minima without fitting them with any functions. Because the QPO period is not stationary, the light curve folded with a fixed period of 3733 s would be expected to have distortion in its modulation shape. From Figure \[fold\_lc\], the folded light curves produced by the HHT phase and cycle length are similar, whereas the one folded by a fix period is significantly different. We believe that the light curve folded by the HHT phase as well as that folded by the cycle length is a better description of the QPO profile than that folded by the fixed period.
O – C Result {#o-c}
------------
To study the evolution of the QPO period in detail, we applied the O – C analysis on the QPO cycle length. The variation of QPO period in RE J1034+396 is so dramatic that it can be detected in both the dynamic power spectrum and Hilbert spectrum. However, it is still worth tracing the evolution of the QPO period using O – C as long as the modulation is clearly seen without large observation gaps, which may cause the cycle count ambiguity. The fiducial point of each QPO cycle is defined as the one for the cycle length described in Section \[HHT\_analysis\]. This low-passed filtered light curve provided us a model independent way to define the fiducial point, i.e., the minimum, in each cycle for the O – C analysis. We further applied a $10^4$ times Monte Carlo simulation to determine the error of the occurance time of fiducial points. Then, we applied a linear ephemeris with the location of the first minimum as the phase zero epoch, and the computed period was 3733 s. Figure \[o\_c\] (a) shows the phase evolution of QPO and Figure \[o\_c\] (b) shows the corresponding light curve, EEMD low-pass filtered light curve, and local minima. From the O – C results, we found that it is better to divide the evolution of the QPO in three epochs rather than two segments proposed by @Gierlinski2008. Epoch A is from the beginning of the observation to $\sim 42$ ks, and the corresponding period obtained by fitting the phase evolution with a straight line is $P_A=4082.7\pm24.6$ s. Epoch B starts at $\sim 42$ ks and ends at $\sim 55$ ks, which corresponds to the flux-drop event. The period of epoch B obtained by O – C is $P_B=3094.4\pm40.3$ s, which much shorter than that of epoch A. This decrease in period is also observed in the time-frequency map (see Figure \[HHT\_DPS\]). Because the flux-drop occurs with an increase in frequency, it strongly implies that this event is caused by the internal change in the QPO status rather than an external occultation as suggested by @Middleton2009. Epoch C is $t\gtrsim55$ ks, and the period is derived as $P_C=3784.1\pm43.1$ s, which is also significantly shorter than that of epoch A. The overall $\chi_{\nu}^2$ is 2.24, which implies a moderate fitting. Because the cycle length varies dramatically even between neighboring cycles, the large $\chi_{\nu}^2$ can be interpreted as resulting from the instability of the QPO period. We also fitted the segment 2 data with a straight line and found a period of $P(t>25ks)=3722.4\pm27.1$ s, which is consistent with the 3733 s determined by @Gierlinski2008. However, the large $\chi_{\nu}^2$ value of 15.9 indicates an unacceptable fitting.
The periodicities obtained by O – C can be further examined by the Lomb-Scargle periodogram. The power spectra of the three epochs are shown in Figures \[o\_c\] (c)–(e). The power spectrum of epoch A shows a strong peak at $P_A=4086.0\pm10.6$ s, whereas the one for epoch C also shows a strong peak at $P_C=3795.8\pm16.2$ s. The periods determined for epoch A and C are consistent with the result obtained by O – C. These periods are also consistent with the peaks in the power spectra of the entire data set (Figure \[LC\_LS\]). Furthermore, the periods of epoch A and C significantly deviate from each other at the $\sim5\sigma$ and $\sim10\sigma$ levels for the results obtained from the O – C analysis and Lomb-Scargle periodograms, respectively. The periodogram of epoch B exhibits a weak broad peak at $P_B=3282.2\pm51.4$ s. This period marginally agrees with that determined by O – C at only the $2\sigma$ level. This is acceptable because epoch B only consists of four cycles, and the QPO period changes between neighboring cycles.
Phase Lag {#lag}
---------
Time lags either between different X-ray bands or between X-ray, UV, and optical wavelengths, are intriguing issues of AGNs. Furthermore, time lags in the different time scales indicate different physical mechanisms. @Middleton2011 found a soft lag in the folded QPO light curve. However, the QPO period was unstable, which implies that the phase lags are probably an artificial effect due to folding the non-stationary light curve with a fixed period, as we have shown in Section \[HHT\_analysis\]. We investigated this issue with the help of the HHT and O – C results.
We first re-examined the results obtained in @Middleton2011. The light curve was divided into four energy bands: band 1 (0.2 – 0.3 keV), band 2 (0.3 – 0.5 keV), band 3 (0.5 – 1.0 keV), and band 4 (1.0 – 10.0 keV). We then folded the light curves of these four energy bands using a period of 3733 s, and calculated the phase lags of bands 2 – 4 with respect to band 1 using the cross-correlation method. The folded light curves with the largest phase lags, i.e., bands 1 and 4, are shown in Figure \[phase\_lag\_3733\_4083\](a), where the soft lag can be easily seen. Figure \[phase\_lag\_3733\_4083\](b) shows the relationship between the phase lags and energies, which is similar to that obtained by @Middleton2011. However, from both the Lomb-Scargle periodogram and time-frequency analysis in the previous sections, we noticed that a single period of 3733 s is insufficient to describe the QPO behavior. For example, we also folded the light curve using a 4083 s period, which is another significant peak in the Lomb-Scargle periodogram in the entire data set. The folded light curves of bands 1 and 4 are shown as Figure \[phase\_lag\_3733\_4083\](c), and the corresponding phase lags are shown as Figure \[phase\_lag\_3733\_4083\](d). No significant soft lag can be observed.
Because the QPO period experienced three evolutionary epochs, we may obtain the phase lags between different energy bands of individual epochs. We first folded the light curves of four bands of epoch A using $P_A=4083$ s, and then observed the phase lags between the folded light curves. No significant phase lags were detected between bands 1 and 2 or bands 1 and 3 (Figure \[phase\_lag\_3784C\_4083A\_HHT\](a)). The cross correlation analysis shows that the folded light curve in band 4 slightly lags band 1, but the significance is only at $\sim1\sigma$ level. We processed similar analysis for epoch C by folding the light curve with period of $P_C=3784$ s. Figure \[phase\_lag\_3784C\_4083A\_HHT\](b) shows that the result is similar to that of epoch A, but band 4 appears to slightly lead band 1, although the significance is also at $\sim1\sigma$ level. Finally, we folded the entire data set using the phase defined by the HHT to check the phase lags between different bands. However, no significant phase lags were detected (Figure \[phase\_lag\_3784C\_4083A\_HHT\](c)).
Correlation between QPO Period and Flux
---------------------------------------
We then examined the relationship between the flux and QPO period for the entire data and 3 epochs, as defined in Section \[o-c\]. To obtain an independent measurement of the period and to avoid the influence of intra-wave frequency modulation, i.e., the frequency variation within one cycle, we used the cycle length obtained in Section \[o-c\] as the period of each QPO cycle. On the other hand, we determined the flux of each QPO cycle by averaging the corresponding count rate of a long-term modulation light curve, which is shown in Figure \[EEMD\](d), to decrease the influence of the noise and QPO modulation. Twenty QPO cycles were used in the correlation analysis. The relationship between the mean flux and QPO period is shown in Figure \[corr\_mp\_rate\]. The linear correlation coefficient of the entire data set is 0.67 with a null hypothesis probability of $3.4\times10^{-4}$. This indicates a strong correlation between the mean flux and QPO period. For the individual epochs, epoch B contains only 3 cycles, so we ignore it and focus on the other two epochs. The linear correlation coefficient of epoch A is 0.11 with a null hypothesis probability of 0.68, which indicates no significant correlation. In contrast, the linear correlation coefficient of epoch C is 0.85 with a null hypothesis probability of $7.4\times10^{-3}$, which is much stronger than that of epoch A.
We further examined the slope between the QPO period and count rate to compare our results with that from @Czerny2010. We made a logarithmic plot of the relationship between the mean count rate and QPO cycle length, and then fitted the data points of the entire data set with a straight line. The slope is $0.30\pm0.08$, which is marginally consistent with the results of @Czerny2010 ($0.42\pm0.05$). We calculated the slopes for epoch A and C to be $0.09\pm0.26$ and $0.43\pm0.12$, respectively. The interpretation of the slopes as well as the dramatic change in the QPO period is discussed with the current QPO models in Section \[discussion\].
Spectral Analysis {#spectral}
=================
From the fractional rms amplitude, @Middleton2009 concluded that the spectral behavior of RE J1034+396 can be described as a constant low-temperature Comptonized component plus a highly variable power law component. @Maitra2010 divided the QPO into high- and low-intensity phases, and then fitted the spectra using a blackbody component plus a broken power law component. They found an absorption feature by assuming that the spectral shape did not vary during the QPO cycle. To investigate the variation of spectral behavior during the QPO cycle, we performed a phase-resolved spectral analysis.
Phase-resolved spectral analysis was achieved using the PN data because the PN collected more photons than the MOS detectors. The following selection criteria were used to achieve high quality spectra: pattern $\leq$ 4 and FLAG $=$ 0. The energy range was set between 0.2 to 10 keV. We first attempted to find the best spectral model to describe both the high-intensity and low-intensity phases. Then, the X-ray photons were divided into 16 phase bins to perform a phase-resolved spectral analysis with the best model.
Fits to High- and Low-Intensity Phases {#high_low}
--------------------------------------
To determine the best spectral model, we divided the QPO phase into a high-intensity phase (0.2 – 0.55) and low-intensity phase (0.65–1.0) according to the folded light curve using the HHT phase. All the X-ray photons in the transition were discarded. Because the image of RE J1034+396 was highly concentrated and had a heavy pile-up in the XMM-Newton PN and MOS chips [@Middleton2009; @Maitra2010], excising the X-ray photons within the central region is needed. For the PN data of the entire data set, the $0.5-2$ keV observed-to-model fraction of single events after excising the inner $32"$ is $0.983\pm0.017$, which is acceptable at 1$\sigma$ level [@Maitra2010]. We also used the SAS task [epatplot]{} to examine this fraction of both the high- and low- intensity phases. The observed-to-model fraction of single events is $0.986\pm0.014$ after excising the inner $15"$ region for the low-intensity state, whereas the fraction is $0.981\pm0.021$ after excising the inner $26"$ region for the high-intensity state. To reduce the statistical error, we grouped the X-ray photons using [grppha]{} with a minimum of 100 photons in each energy bin. The [XSPEC v12.8.0]{} of the [HEASOFT]{} package was used to perform the spectral fitting. The galactic hydrogen column density was fixed at $1.31\times10^{20}$ cm$^{-2}$; however, an additional absorption to represent the absorption around the AGN was also set by adding a [zphabs]{} with a fixed redshift of 0.042. We then used the spectral model proposed by @Middleton2009. A low-temperature Comptonized disk plus a hard power law tail (hereafter known as [compTT+pl]{}) was used to fit both the high- and low-intensity phase data. Furthermore, we also added an additional blackbody component to the [compTT+pl]{} as the seed photons, hereafter known as [bb+compTT+pl]{}. Table \[fitting\_diff\_model\] lists the best-fit parameters of these two models for both the high- and low-intensity phases.
Phase
------------------------ --------------- ------------------ --------------- ------------------
Model [compTT+pl]{} [bb+compTT+pl]{} [compTT+pl]{} [bb+compTT+pl]{}
$kT_{BB} (keV)$ $-$ $0.012\pm0.007$ $-$ $0.013\pm0.005$
$\Gamma$ $1.71\pm1.1$ $2.60\pm0.30$ $1.91\pm0.52$ $2.22\pm0.25$
$T_0 (keV)$ $0.02\pm0.08$ $=kT_{BB}$ $0.02\pm0.06$ $=kT_{BB}$
$kT_{comp} (keV)$ $0.42\pm0.59$ $0.16\pm0.04$ $0.41\pm0.28$ $0.24\pm0.05$
$\tau$ $7.1\pm6.3$ $13.8\pm3.3$ $7.05\pm3.45$ $10.6\pm1.6$
$\chi_{\nu}^{2}$ (dof) 1.107(83) 1.124(82) 1.127(116) 0.982(115)
The fitted results reveal a blackbody emission of $kT\sim10-\sim15$ eV, which probably represents the tail of the thermal emission from the gas in the inner accretion disk. Treating this blackbody emission as the seed photons for the [bb+compTT+pl]{} model, we obtained a good fit for both the high- and low-intensity phases with $\chi_{\nu}^2\approx1.1$ and $1.0$, respectively. The [compTT+pl]{} model, by comparison, cannot constrain a meaningful soft photon temperature in both the high- and low-intensity phases. Thus, we chose the [bb+compTT+pl]{} model to fit the phase-resolved spectra.
Phase-Resolved Analysis
-----------------------
We further divided the X-ray photons into 16 phase bins according to the HHT phase. All the X-ray photons collected in each phase bin were then fitted by the best-fit model in Section \[high\_low\]. To increase the number of energy bins but keep sufficient statistics for each of the bins, especially the hard power law tail with energy higher than 2 keV, we grouped the X-ray photons with a minimum of 25 photons in each energy bin. All the spectral parameters were first set to be free, and we found that they all have no significant variation during the QPO cycle except for the normalizations. Then, we froze the spectral parameters to the values obtained in Section \[high\_low\], except for the normalizations, to estimate the flux variation of the spectral components during one QPO cycle. Because of the lack of sufficient hard X-ray photons with energies higher than 6 keV in individual phase bins, we calculated the flux between 0.3 and 6.0 keV. The fluxes of black body component cannot be significantly detected because the normalization term of the black body component cannot be well constrained.
The left panel of Figure \[spectral\_paremeter\] shows the variation of those three spectral components. The power law tail is more variable than the Comptonized component. We obtained the flux variation of these two components shown in the right panel of Figure \[spectral\_paremeter\]. We found that the fractional rms amplitude of the power law component is 20.9%, which is approximately three times larger than that of the Comptonized component (6.6 %). Furthermore, the phase lag between the Comptonized disk component and the power law tail is $-0.017\pm0.030$ where the error is estimated using a $10^4$ times Monte Carlo simulation, which indicates no significant phase lag. The lack of significant phase lag implies that if the Comptonized component is the reprocessed X-ray from the power law component, the light travel scale should be negligible
Discussion
==========
The QPO detected in RE J1034+396 is believed to be a high-frequency QPO because of its short periodicity. This type of QPO in X-ray binary systems appears in the very high state, while the accretion disk nearly reaches the innermost region of the central black hole. A variety of mechanisms capable of interpreting the QPO phenomenon were proposed. We discussed three of them according to our discoveries, including the evolution of characteristic periods for each epoch, correlation between the QPO period and mean flux, and variation of spectral components.
The Spotted Disk Model
----------------------
The most straightforward model to understand the QPO is that it is caused by the Keplerian motion of a temporary hot spot on the accretion disk [@Pechacek2006; @Bachetti2010]. In this scenario, the QPO frequency is related to the radius of Keplerian orbit. From the HHT and O – C analysis, we found that the QPO period changes dramatically even between neighboring cycles. This indicates that the hot spot is not strictly anchored on the accretion disk. Instead, the hot spot is wobbling around a characteristic radius. Furthermore, the evolution of the QPO period can be described by three epochs in this observation. This evolution can be interpreted as the change of characteristic radius caused by the instability of the accretion disk.
From the analysis of the relationship between modulation period and the flux, we observed a positive correlation between the QPO period and the flux. This relationship, which is basically agree with that observed in @Czerny2010, does not agree with the prediction of this model. However, some indications of this model are still worth to be addressed. @Czerny2010 compared the fractional rms amplitude of RE J1034+396 and that of the light curve calculated by a spotted accretion disk [@Pechacek2006]. They concluded that the inclination angle should be 2 – 3 degrees if the hot spot is close to the central black hole. However, not only the flare but also the disk, corona, and probably other materials near the black hole contribute to the X-ray emissions. For example, we obtained three components in the spectral fitting. Those X-ray sources would contribute non-modulated or lower-amplitude X-ray emissions such that the fractional amplitude as well as the inclination angle would be under estimated.
Besides the fractional amplitude, another possible indicator that can be used to estimate the inclination angle is the modulation shape. From the simulated light curve in @Czerny2010 and @Pechacek2006, the QPO modulation shape became highly non-sinusoidal for higher inclination angles. A non-sinusoidal light curve can be expressed as harmonics in the Fourier analysis; however, we could not observe any significant harmonics in the power spectrum. There are two possible explanations for this. The first is that the inclination angle of RE J1034+396 is low, and the QPO profile is close to sinusoidal such that the harmonics are not significantly detected. Another possibility is that the QPO period varies so dramatically that the harmonics are more strongly depressed than the fundamental. Fortunately, the HHT provides us an indicator to probe the non-sinusoidal level of the data. Unlike harmonics in a Fourier analysis, the non-sinusoidal light curve would produce [*intra-wave*]{} frequency modulations, i.e., the frequency variation within one cycle [@Huang1998] in the instantaneous frequency. The frequency derived from the cycle length in the O – C analysis can be treated as the [*inter-wave*]{} frequency modulation, i.e., the frequency variation between cycles. Thus, the intra-wave frequency modulation can be estimated by taking the difference between the instantaneous frequency and inter-wave frequency modulation. We then created simulated light curves according to equations 4 and 10 in @Pechacek2006 for different black hole masses and inclination angles. The orbital radius of the hot spot is calculated using Kepler’s law for orbital periods of 3784 and 4083 s. The amplitude of the intra-wave frequency modulation of a simulated light curve can also be calculated using the Hilbert transform on the simulated light curve. For a mass range $1-4\times10^6$ M$_{\odot}$, we can estimate the inclination angle by matching the intra-wave frequency modulation amplitude of the simulated light curves with the observed one. Figure \[inclination\_mass\] shows the relationship between the estimated inclination angle and black hole mass. We found that the inclination angle was between $\sim 27$ and $\sim 57$ degrees. This estimation of inclination angle is probably an over-estimate because the noise in the light curve would also contribute to the intra-wave modulation.
Diskoseismology
---------------
Relativistic diskoseismology, i.e., the oscillations that are trapped in the inner accretion disk, may also be responsible for the presence of QPOs [@Perez1997; @Silbergleit2001; @Wagoner2001; @Ortega2002]. Three oscillation modes, i.e., g-, c-, and p-modes, are investigated. The fundamental frequency of these three modes mainly depend on the mass of the black hole, i.e., $f\propto1/M_{BH}$, where $f$ is the QPO frequency and $M_{BH}$ is the mass of black hole. From @Wagoner2001, the g-mode is dominated by the gravitational-centrifugal force, in which the oscillation frequency anti-correlates with the flux of accretion disk. The c-mode is caused by the non-radial corrugation on the inner accretion disk, while the oscillation frequency is highly dependent on the spin of black hole. The p-mode is dominated by the restoring-force due to the pressure gradient, in which the oscillation frequency shows a positive correlation with the flux of disk. Only the p-mode shows a positive correlation between the flux and QPO frequency, but it is opposite of our results. Thus, the QPO in RE J1034+396 is not likely caused by the p-mode oscillation. For a black hole with mass of 10 M$_{\odot}$, the g-mode oscillation is between $\sim 70$ to $\sim 110$ Hz for different angular momentums and luminosities [see Figure 1 of @Wagoner2001]. If we enlarge the black hole mass to that in RE J1034+396, the g-mode oscillation period is between $\sim900$ and $\sim1400$ s for a $1\times10^6$ M$_{\odot}$ black hole, or between $\sim3600$ and $\sim5700$ s for a $4\times10^6$ M$_{\odot}$ black hole. Thus, the preferred black hole mass is close to the upper limit of the estimation made by @Bian2010 if we assume that the QPO is caused by the g-mode oscillation. However, it is difficult for the slope between the g-mode oscillation period and luminosity to exceed $\sim0.1$ unless the angular momentum of the black hole is extremely high [see equation 3 in @Wagoner2001].Thus, the g-mode oscillation cannot provide the observed slope ($\sim0.3$), although this value is less reliable relative to the correlation coefficient because the observed X-ray flux may also contain the emissions from other components. On the other hand, the c-mode oscillation can produce a larger slope between the flux and QPO frequency in the range of our observed value, but the angular momentum of the black hole should be close to zero [see equation 6 in @Wagoner2001]. However, the fundamental c-mode frequency of such a slow-rotating black hole is very low. For example, a 10 M$_{\odot}$ black hole with angular momentum of $a=0.05$ has a fundamental c-mode oscillation frequency of $f\lesssim1$ Hz. This oscillation corresponds to a period longer than $10^5$ s for a $10^6$ M$_{\odot}$ black hole. Thus, if the c-mode oscillation is responsible for the observed QPO, it should be high-order harmonics.
We also studied the relationship between the QPO frequency and flux in detail by dividing the evolution of the QPO into three epochs. A change in correlation between epochs was observed: a significant anti-correlation between the frequency and flux was observed in epoch C, but no significant correlation was found in epoch A. This is possibly caused by a change in oscillation mode. A correlation in epoch A was not significantly detected; however, we found that the first three cycles appear to behave similar to the flux drop event in epoch B. If we ignore these three cycle, the remaining cycles in epoch A show a strong anti-correlation between the cycle length and flux with a linear correlation coefficient of $-0.61$. Thus, the QPO in epoch A is probably caused by the p-mode oscillation. After epoch B, i.e., the flux-drop event, the g-mode oscillation dominated the observed QPO phenomenon such that we observed a significant direct correlation between the cycle length and flux.
All the discussions above are based on the assumption that the luminosity is less than the Eddington luminosity. This is a necessary condition for the thin-disk assumption. However, the luminosity of RE J1034+396 is probably a super-Eddington [@Gierlinski2008; @Middleton2009]. Thus, we could not exclude all the possible oscillation modes in a super-Eddington accretion disk.
Oscillation of Shock
--------------------
@Das2011 proposed a novel model of QPO to investigate the observed slope between the QPO period and X-ray flux in RE J1034+396. In their model, QPO is caused by the oscillations of the shock formed inside the hot accretion flow. The variation in the QPO period is interpreted as the drifting of shock location. This model successfully interpreted the correlation between the QPO period and X-ray flux in @Czerny2010 with slope of $0.92\pm0.03$. However, we have demonstrated that the slope changes as QPO evolves in this observation. In fact, even the relationship likely changes. For example, the QPO period and flux exhibit no significant correlation in epoch A, or even anti-correlation when we omit the first three cycles that show another possible flux-drop event. Thus, the entire observation cannot be interpreted by this single scenario.
Summary
=======
This research successfully applied the novel time-frequency analysis technique, i.e., HHT, on the QPO of RE J1034+396. The Hilbert spectrum enabled us to trace the variation of QPO period even between neighboring cycles. The EEMD filtered light curve allowed us to directly determine the fiducial points for the O – C analysis. From the O – C analysis, we found that the variation in the QPO period can be described as three evolutionary epochs. For a spotted accretion disk model and the shock oscillation model, the change of characteristic periods indicates a dramatic change in the flare or shock location, respectively. For the diskoseismology scenario, the change in QPO characteristic period as well as the change in flux-period correlation between epochs can be interpreted as the change in oscillation modes.
We further examined the phase lag phenomenon between soft and hard X-ray bands. No phase lags were determined when folding the light curve in different epochs with their corresponding periodicities, or folding the entire data set by the HHT phase. The phase lag determined by folding the entire data set with a fixed 3733 s period is probably an artificial effect that resulted from folding the variable periodicity with a fixed period. The lack of phase lags indicates that the emission regions for both spectral components are close to each other.
The time scale of QPO in AGN is much longer than that in the X-ray binary system. Such long time scales make it easier to study the variability in the QPO period in detail. We believe that future X-ray observations can discover more QPO samples in AGNs. With the help of the HHT and dynamic power spectrum, it is possible to study the evolution of QPO periodicities systematically.
We thank Prof. Christopher Reynolds and Prof. Chris Done for useful discussions in the QPO mechanisms. We thank the anonymous referee for useful suggestions and comments. The HHT codes were provided by the Research Center for Adaptive Data Analysis in National Central University of Taiwan. This work is based on observations obtained with XMM-Newton, an ESA science mission with instruments and contributions directly funded by ESA Member States and the USA (NASA). This research was supported by the NSC 102-2112-M-008-020-MY3 grant from the Ministry of Science and Technology of Taiwan.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
We prove several results asserting that the action of a Banach–Lie group on Hilbert spaces of holomorphic sections of a holomorphic Hilbert space bundle over a complex Banach manifold is multiplicity free. These results require the existence of compatible antiholomorphic bundle maps and certain multiplicity freeness assumptions for stabilizer groups. For the group action on the base, the notion of an $(S,\sigma)$-weakly visible action (generalizing T. Koboyashi’s visible actions) provides an effective way to express the assumptions in an economical fashion. In particular, we derive a version for group actions on homogeneous bundles for larger groups. We illustrate these general results by several examples related to operator groups and von Neumann algebras.
**Keywords.** unitary representation, infinite dimensional Lie group, holomorphic Hilbert bundle, multiplicity-free representation, reproducing kernel, visible action
address:
- 'IAM-CONICET. Saavedra 15, Piso 3, (1083) Buenos Aires, Argentina'
- 'Department Mathematik, University Erlangen-Nuremberg, Cauerstrasse 11, 91058 Erlangen, Germany'
author:
- Martín Miglioli
- 'Karl-Hermann Neeb'
title: Multiplicity freeness of unitary representations in sections of holomorphic Hilbert bundles
---
Introduction
============
A unitary representation $(\pi,\cH)$ of a group $G$ on a complex Hilbert space $\cH$ is called *multiplicity-free* if its commutant, the von Neumann algebra $\pi(G)'$ of continuous $G$-intertwining operators, is commutative. Multiplicity-free representations are special in the sense that one may expect to find natural decompositions into irreducible ones based on direct integrals over the spectrum of the commutant. We refer to [@kob2] and the reference therein for a survey of multiplicity free theorems and its applications in the context of finite dimensional Lie groups.
The main results of this paper consist in propagation theorems for the multiplicity-free property (MFP) from the representation of a stabilizer group in a fiber Hilbert space to Hilbert spaces of holomorphic sections of holomorphic Hilbert bundles. Our results extend those of T. Kobayashi concerning finite-dimensional bundles [@kob] to Hilbert bundles over Banach manifolds. More specifically, the fibers are complex Hilbert spaces and the groups act as fiberwise isometric holomorphic bundle automorphisms. We apply these propagation theorems to branching problems of representations of infinite dimensional groups constructed by holomorphic induction. Here an essential part is that the isotropy representations are not finite dimensional, hence in general not direct sums of irreducible representations. As we shall see, this difficulty can be overcome by working systematically with the commutant as a von Neumann algebra.
A variant of the propagation theorem is formalized as in [@kob] in terms of so-called visible actions. The $G$-action on $M$ is called $(S,\sigma^M)$-weakly visible if $S\subseteq M$ is a subset for which the closure of $G.S$ has interior points and $\sigma^M$ is an antiholomorphic diffeomorphism of $M$ preserving all $G$-orbits through $S$ and leaving $S$ invariant. If $\sigma^M$ lifts to an anti-holomorphic bundle endomorphism $\sigma^{\V}$ which is compatible with the $G$-action with respect to an automorphism $\sigma^G$ of $G$ satisfying $\sigma^\bV(g.v) = \sigma^G(g).\sigma^\bV$, then one can formulate a variant of the propagation theorem (Theorem \[multfreevis\]) asserting the multiplicity-freeness of the $G$-representation on $\cH \subseteq \Gamma(\bV)$ if, for every $s \in S$, the antiunitary operator $\sigma^\bV_s$ commutes on $\bV_s$ with the hermitian part of the commutant of the $G_s$-action. A third form of the propagation theorem is obtained in the setting where the bundle $\V=G\times_{\rho,H} V$ is associated to a homogeneous $H$-principal bundle $G\to M=G/H$ by a norm continuous unitary representation $(\rho,V)$ of $H$. We plan to use this formulation for concrete branching problems in the representation theory of Banach–Lie groups.
In the infinite dimensional context there are no general results on the existence of solutions of ${\overline}{\partial}$-equations that can be used to verify integrability of complex structures on Banach manifolds and in particular on vector bundles. Here [@neeb13a] provides effective methods to treat unitary representations of Banach–Lie groups in spaces of holomorphic sections of homogeneous Hilbert bundles. For a real homogeneous Banach vector bundle $\V=G\times_H V$ over $G/H$ associated to a norm continuous representation of $H$, compatible complex structures are obtained by extensions $\beta:\q\to \gl (V)$ of the differential $\operatorname{d}\rho\colon\h\to \gl(V)$ to a representation of the complex subalgebra $\q\subseteq \g_{\C}$ specifying the complex structure on $M$ by $T_{1H}M\simeq\g_{\C}/\q$.
Another particularity of the infinite dimensional context is that in general self-adjoint operators are not conjugate to operators in a given maximal abelian subalgebra. This is well known for the algebra of bounded operators acting on a Hilbert space, see [@bw] for the case of Hilbert-Schmidt operators and the example after [@ak06]\*[Thm. 5.4]{} for operators in a finite von Neumann factor. In the fundamental examples the group action is derived from the two-sided action of the unitary group $\operatorname{U}(\cM)$ on the algebra $\cM$ endowed with a scalar product derived from its trace, and the slice $S$ is derived from the hermitian operators in a maximal abelian subalgebra. That the closure of $G.S$ has interior points comes from an approximate Cartan decomposition which follows from diagonalizabity on a dense subset. Also, since the isotropy representations in the infinite dimensional context may not be discretely decomposable as in the finite dimensional case treated in [@kob] we do not impose this condition in the propagation theorems.
The structure of this paper is as follows. Section \[prel\] contains some preliminary results on equivariant holomorphic Hilbert bundles and representations in Hilbert spaces of holomorphic sections. In Section \[compgeomultfree\] we start with our First Propagation Theorem \[multfree1\], which asserts the multiplicity freeness of a unitary representation of a group $G$ in a Hilbert space of holomorphic sections, provided there exists an antiholomorphic bundle map satisfying certain compatibility conditions formulated in terms of stabilizer representations in points $m$ belonging to a subset $D{\subseteq}M$.
For the sake of easier application of this result in the infinite dimensional context, we slightly extend T. Kobayashi’s notion of a visible action ([@kob2]). We call the action of the group $G$ on a complex manifold $M$ by holomorphic maps $(S,\sigma^M)$-weakly visible if $S {\subseteq}G$ is a subset for which the closure of $G.S$ has interior points and $\sigma^M {\colon}M \to M$ is an antiholomorphic diffeomorphism fixing $S$ pointwise and preserving the $G$-orbits through $S$. For a visible action one requires in addition that $G.S$ is an open subset of $M$, but this is not satisfied in many interesting infinite dimensional situations, where the weak visibility can be verified. This leads us to our Second Propagation Theorem \[multfreevis\], where the assumptions are formulated in terms of a weakly visible action.
We then turn to the special case where the bundle $\bV$ is a homogeneous bundle over a homogeneous space $G/H$ of a Banach–Lie group $G$. In Section \[assocbundles\] we discuss $G$-invariant complex structures on such bundles and antiholomorphic isomorphisms. This is used in Section \[propagassocbund\] to obtain a propagation theorem for the multiplicity freeness of the representation of a subgroup $K {\subseteq}G$ on Hilbert spaces of holomorphic sections of $\bV$ (Theorem \[multfreegroup\]).
In Section \[exampleswv\] we eventually discuss various concrete situations in the Banach context, where the results of this paper apply naturally. In particular we exhibit several kinds of weakly visible actions on infinite dimensional spaces and state some corresponding propagation theorems. A thorough investigation of these particular representations and concrete branching results are the topic of ongoing research.
Preliminaries {#prel}
=============
Equivariant holomorphic Hilbert bundles
---------------------------------------
Let $q:\V=\coprod_{m\in M}\V_m\to M$ be a holomorphic vector bundle over a connected complex Banach manifold $M$ whose fibers are complex Hilbert spaces. We write $\Gamma (\V)$ for the space of holomorphic sections of $\V\to M$. We further assume that $G$ is a group (at this point no topology on $G$ is assumed) which acts on $\V$ by isometric holomorphic bundle automorphisms $(\gamma_g)_{g \in G}$. We denote the action of $G$ on the base space simply by $m\mapsto g.m$ for $g\in G$. In particular, we obtain for each $m \in M$ a unitary representation $$\rho_m:G_m\to \operatorname{U}(\V_m)$$ of the isotropy subgroup $G_m:=\{g\in G:g.m=m\}$ on the fiber $\V_m$. Finally, the action of $G$ on the bundle $\V\to M$ gives rise to a representation $\delta$ of $G$ on $\Gamma (\V)$ by $$\begin{aligned}
(\delta_gs)(m):=\gamma_{g}(s(g^{-1}.m)) \quad\mbox{for}\quad g\in G\quad\mbox{and}\quad s\in\Gamma (\V).
\label{actionbundle}\end{aligned}$$
Reproducing kernels for Hilbert bundles {#repkern}
---------------------------------------
Let $q:\V\to M$ be a holomorphic Hilbert bundle on the complex manifold $M$. A Hilbert subspace $\H\subseteq \Gamma (\V)$ is said to have *continuous point evaluations* if all the evaluation maps $$\operatorname{ev}_m:\H\to\V_m, \, s\mapsto s(m)$$ are continuous and the function $m\mapsto \|\operatorname{ev}_m\|_{B(\H,\V_m)}$ is locally bounded. Then $$Q(m,n):=(\operatorname{ev}_m)(\operatorname{ev}_n)^*\in B(\V_n,\V_m),$$ defines a holomorphic section of the operator bundle $$B(\V):=\coprod_{(m,n)\in M\times M^{\rm op}}B(\V_n,\V_m)\to M\times M^{\rm op},$$ where $ M^{\rm op}$ is the complex manifold $M$ endowed with the opposite complex structure. This section is the *reproducing kernel* of $\H$.
[ We call a Hilbert subspace $\H\subseteq\Gamma (\V)$ with continuous point evaluations *$G$-invariant*, if $\H$ is invariant under the action defined by (\[actionbundle\]) and the so obtained representation $\pi$ of $G$ on $\H$ is unitary. In this case we say that $(\pi,\H)$ is *realized* in $\Gamma (\V)$.]{}
\[lemeqkern\]
1. For $j =1,2$, let $Q_j$ be the reproducing kernels of the Hilbert spaces $\H_j\subseteq\Gamma(\V)$ with inner products $\langle\cdot, \cdot \rangle_{\H_j}$. If $Q_1=Q_2$, then the subspaces $\H_1$ and $\H_2$ coincide and the inner products $\langle\cdot, \cdot\rangle_{\H_1}$ and $\langle\cdot, \cdot\rangle_{\H_2}$ are the same.
2. If $M$ is connected and $Q_1(m,m)=Q_2(m,m)$ for all $m$ in a subset $D$ which is dense in an open subset of $M$, then $Q_1=Q_2$.
We can represent holomorphic sections of the bundle $q:\V\to M$ by holomorphic functions on the total space of the dual bundle $\V^*$ which are linear on each fiber via the $G$-equivariant embedding $$\Psi:\Gamma(\V)\to\O(\V^*),\quad \Psi(s)(\alpha_m)=\alpha_m(s(m))\quad\mbox{for}\; s\in \Gamma(\V),\; \alpha_m\in \V_m^*,$$ where $\O(\V^*)$ is the space of holomorphic functions on $\V^*$, see [@neeb13a]\*[Remark 2.2]{}. For a reproducing kernel Hilbert space $\H\subseteq \Gamma(\V)$ with reproducing kernel $Q$ we obtain a reproducing kernel Hilbert space of holomorphic functions $\Psi(\H)\subseteq \O(\V^*)$ with reproducing kernel $$K(\alpha_m,\beta_n)=\operatorname{ev}_{\alpha_m}\operatorname{ev}_{\beta_n}^*\in B(\C,\C)\simeq \C\quad\mbox{for}\quad\alpha_m\in\V_m^*,\beta_n\in\V_n^*.$$ Since $$\operatorname{ev}_{\alpha_m} \circ \Psi = \alpha_m \circ \operatorname{ev}_m
\quad\mbox{for}\quad \alpha_m\in \V_m^*,$$ we obtain for $f:=\Psi(s)\in \Psi(\H)$, $s\in \H$, the relation $$\operatorname{ev}_{\alpha_m}(f)=\alpha_m(\operatorname{ev}_m(\Psi^{-1}(f))),$$ so that $$\begin{aligned}
K(\alpha_m,\beta_n)&=\operatorname{ev}_{\alpha_m}\operatorname{ev}_{\beta_n}^*
=(\alpha_m\circ\operatorname{ev}_m\circ\Psi^{-1})(\beta_n\circ\operatorname{ev}_n\circ\Psi^{-1})^*\\
&=\alpha_m\operatorname{ev}_m\operatorname{ev}_n^*\beta_n^*=\alpha_mQ(m,n)\beta_n^*.\end{aligned}$$ Therefore, if $\alpha_m=\langle\cdot,v_m\rangle_{\V_m}$ for $v_m\in\V_m$ and $\beta_n=\langle\cdot,w_n\rangle_{\V_n}$ for $w_n\in\V_n$, then $$K(\alpha_m,\beta_n)=\langle Q(m,n)w_n,v_m\rangle_{\V_m}.$$ If $M$ is connected, then the total space $\bV^*$ is also connected and if $D$ is dense in an open subset of $M$, then $\coprod_{m\in D}\V^*_m$ is dense in a open subset of $\V^*$. The first assertion now follows from general facts about reproducing kernel spaces [@neeb00]\*[Lemma I.1.5]{} and the second by the the discussion in [@neeb00]\*[Lemma A.III.8]{}.
\[equivkernel\] If $(\pi,\H)$ is realized in $\Gamma(\V)$, then the kernel $Q$ of $\H$ satisfies $$Q(g.m,g.n)=(\gamma_g|_{\V_m})Q(m,n)(\gamma_g|_{\V_n})^{-1}
\quad \mbox{ for } \quad m,n\in M, g\in G.$$ In particular the hermitian operators $Q(m,m)$ commute with $\rho_m(G_m))$ for every $m\in M$.
Since $$(\pi(g)^{-1}s)(m)=\gamma_{g^{-1}}(s(g.m))\quad\mbox{for}\quad s\in \H,g\in G,m\in M$$ we have $$\operatorname{ev}_{g.m}=\gamma_g|_{\V_m}\circ \operatorname{ev}_m\circ\pi(g)^{-1}.$$ Therefore $$\begin{aligned}
Q(g.m,g.n)&=\operatorname{ev}_{g.m}\operatorname{ev}_{g.n}^*
=((\gamma_g|_{\V_m}) \operatorname{ev}_m\pi(g)^{-1})(\pi(g)\operatorname{ev}_n^*(\gamma_g|_{\V_n})^{-1})\\
&=(\gamma_g|_{\V_m})Q(m,n)(\gamma_g|_{\V_n})^{-1}.
\qedhere\end{aligned}$$
Complex geometry and multiplicity-free property {#compgeomultfree}
===============================================
In this section we prove the propagation of the multiplicity-freeness from the isotropy representations to the representation on Hilbert spaces of holomorphic sections of equivariant holomorphic vector bundles. This result is proved through the construction of an anti-unitary operator $J$ on the representation space which implements a conjugation in the commutant of the representation, from which the multiplicity-free property of the representation follows. Then we introduce the concept of $(S,\sigma^M)$-weakly visible action to prove a second version of the Propagation Theorem where the conditions are imposed on a slice of the group action on a dense in an open subset of the base space.
Propagation of the multiplicity free property from fibers to sections
---------------------------------------------------------------------
The following lemma captures the key idea that is mostly used to show that a commutant is commutative. We refer to [@FT99] for one of the first systematic applications of this idea.
\[conjcommute\] Let $\M\subseteq B(\H)$ be a von Neumann algebra. The following conditions are equivalent:
- $\M$ is commutative.
- There is an anti-unitary $J$ on $\H$ such that $JAJ^{-1}=A^*$ for $A\in \M$.
- There is an anti-unitary $J$ on $\H$ which commutes with the self-adjoint part $\M_h$ of $\M$, i.e., $J\in(\M_h)'^{\R}$, where $(\cdot)'^{\R}$ denotes the real linear commutant.
- There is an anti-unitary $J$ which commutes with the positive invertible operators in $\M$.
\(a) $\Rightarrow$ (b): We decompose $\H$ into cyclic representations of $\M$. Hence $\H=\bigoplus_{i\in I}\H_i$ where $\H_i\cong
L^2(X_i,\mu_i)$ for compact spaces $X_i$ and regular probability measures $\mu_i$ on $X_i$ and $\M|_{\H_i}= L^{\infty}(X_i,\mu_i)$ acts as multiplication operators on $\H_i=L^2(X_i,\mu_i)$, see [@fa]\*[Thm. 11.32]{}. We define $J(\oplus_{i\in I}f_i)=\oplus_{i\in I}\overline{f_i}$ and the assertion follows since $\M_h|_{\H_i}$ are real valued functions on $X_i$.
\(b) $\Leftrightarrow$ (c): Since the map $\M \to \M, A\mapsto JAJ^{-1}$ is anti-linear, it coincides with the antilinear map $A \mapsto A^*$ if and only if it does on the subspace $\M_h$ of hermitian elements.
\(b) $\Rightarrow$ (a): For $A,B\in \M$ we have $$AB=J^{-1}(AB)^*J=J^{-1}B^*JJ^{-1}A^*J=BA,$$ so that $\M$ is commutative.
\(c) $\Rightarrow$ (d): This is trivial.
\(d) $\Rightarrow$ (c): Assume that $A\in \M_h$ and choose $c\in\R$ such that $B=A+c\operatorname{id}$ is positive and invertible. Since $B=J^{-1}BJ$ we obtain $$\begin{aligned}
A&=B-c\operatorname{id}=J^{-1}BJ-c\operatorname{id}=J^{-1}(B-c\operatorname{id})J=J^{-1}AJ.
\qedhere\end{aligned}$$
<span style="font-variant:small-caps;">(First Propagation Theorem)</span>\[multfree1\] Let a group $G$ act by automorphisms on the holomorphic Hilbert bundle $q:\V\to M$. Assume that there exists an anti-holomorphic bundle endomorphism $(\sigma^{\V},\sigma^M)$ and a $G$-invariant subset $D\subset M$ with $\overline{D}^\mathrm{o}\neq \emptyset$, such that for any $m\in D$ there is $g\in G$ such that $g.m=\sigma^M(m)$, and $(\gamma_g|_{\V_m})^{-1}\sigma^{\V}_m$ commutes with the hermitian part $\rho(G_m)'_h$ of $\rho(G_m)'$. Then any unitary representation $(\pi,\H)$ of $G$ realized in $\Gamma(\V)$ is multiplicity-free.
Note that the group $G$ is not assumed to be a Banach–Lie group. Also note that the existence of an anti-unitary operator $(\gamma_g|_{\V_m})^{-1}\sigma^{\V}_m$ commuting with $\rho_m(G_m)'_h$ implies by Lemma \[conjcommute\] that the representation $\rho_m:G_m\to\operatorname{U}(\V_m)$ is multiplicity-free.
**First step.** We define a conjugate linear map $$\begin{aligned}
J:\Gamma( \V)\to \Gamma(\V),\quad s\mapsto (\sigma^{\V})^{-1}\circ s\circ \sigma^M.\label{conj}\end{aligned}$$ We will prove that $J:\H\to\H$ is a anti-unitary operator for any unitary representation $(\pi,\H)$ realized in $\Gamma (\V)$.
Consider the Hilbert space $\widetilde{\H}:=J(\H)\subseteq \Gamma(\V)$ equipped with the inner product $$\langle Js_1,Js_2\rangle_{\widetilde{\H}}:=\langle s_2,s_1\rangle_{\H} \quad\mbox{for}\quad s_1,s_2\in\H,$$ so that $J:\H\to \widetilde{\H}$ is an anti-unitary operator. Denote by $\widetilde{\operatorname{ev}}_m:\widetilde{\H}\to \V_m$ the evaluations of $\widetilde{\H}$. Since $$\widetilde{\operatorname{ev}}_m(Js)=(\sigma^{\V}_m)^{-1}(s(\sigma^M(m)))\quad\mbox{for}\quad s\in \H, m\in M,$$ we get $$\widetilde{\operatorname{ev}}_m=(\sigma^{\V}_m)^{-1}\circ\operatorname{ev}_{\sigma^M(m)}\circ J^{-1},$$ so that $$\begin{aligned}
Q_{\widetilde{\H}}(m,n)&=\widetilde{\operatorname{ev}}_m\widetilde{\operatorname{ev}}_n^*=((\sigma^{\V}_m)^{-1}\operatorname{ev}_{\sigma^M(m)} J^{-1})(J\operatorname{ev}_{\sigma^M(n)}^*\sigma^{\V}_n)\\
&=(\sigma^{\V}_m)^{-1}Q_{\H}(\sigma^M(m),\sigma^M(n))\sigma^{\V}_n.\end{aligned}$$ We fix $m\in D$. By assumption there is $g\in G$ such that $\sigma^M(m)=g.m$ and $(\gamma_g|_{\V_m})^{-1}\sigma^{\V}_m$ commutes with the hermitian part $\rho(G^1_m)'_h$ of $\rho(G^1_m)'$. Since $Q(m,m)\in \rho_m(G_m)'_h$ by Lemma \[equivkernel\], $$\begin{aligned}
Q_{\widetilde{\H}}(m,m)
&=(\sigma^{\V}_m)^{-1}Q_{\H}(\sigma^M(m),\sigma^M(m))\sigma^{\V}_m\\
&=(\sigma^{\V}_m)^{-1}Q_{\H}(g.m,g.m)\sigma^{\V}_m\\
&=(\sigma^{\V}_m)^{-1}(\gamma_g|_{\V_m})Q_{\H}(m,m)(\gamma_g|_{\V_m})^{-1}\sigma^{\V}_m=Q_{\H}(m,m),\end{aligned}$$ so that $Q_{\widetilde{\H}}=Q_{\H}$ on $\operatorname{diag}(D)\subseteq M\times M$. By Lemma \[lemeqkern\], the Hilbert space $\widetilde{\H}$ coincides with $\H$ and $$\label{eq:j-rel}
\langle Js_1,Js_2\rangle_{\H}=\langle Js_1,Js_2\rangle_{\widetilde{\H}}=\langle s_2,s_1\rangle_{\H} \quad\mbox{for}\quad s_1,s_2\in \H.$$
**Second step**. Assume that $A\in \pi(G)'$ is positive and invertible. We define a compatible inner product on $\H$ by $$\langle s_1,s_2\rangle_{\H_A}:=\langle As_1,s_2\rangle_{\H}.$$ The space $\H$ with the new inner product will be denoted by $\H_A$. For $s_1,s_2\in \H$ and $g\in G$ we have $$\begin{aligned}
\langle \pi(g)s_1,\pi(g)s_2\rangle_{\H_A}&=\langle A\pi(g)s_1,\pi(g)s_2\rangle_{\H}
=\langle \pi(g)As_1,\pi(g)s_2\rangle_{\H}=\langle As_1,s_2\rangle_{\H}=\langle s_1,s_2\rangle_{\H_A}.\end{aligned}$$ Therefore $\pi$ also defines a unitary representation on $\H_A$ and we can apply to $\H_A$ to obtain $$\begin{aligned}
\langle As_1,s_2\rangle_{\H}&=\langle s_1,s_2\rangle_{\H_A}=\langle Js_2,Js_1\rangle_{\H_A}=\langle AJs_2,Js_1\rangle_{\H}=\langle Js_2,AJs_1\rangle_{\H}\\
&=\langle Js_2,JJ^{-1}AJs_1\rangle_{\H}=\langle J^{-1}AJs_1,s_2\rangle_{\H}.\end{aligned}$$ Hence $A=J^{-1}AJ$, i.e., $J$ commutes with $A$, and by Lemma \[conjcommute\] the von Neumann algebra $\pi(G)'$ is commutative.
Discretely decomposable representation on the fiber
---------------------------------------------------
The following lemma shows how the commuting condition in the First Propagation Theorem can be expressed in the classical context where the isotropy representation decomposes discretely.
\[conjcommutedisc\] Let $\M\subseteq B(\H)$ be a commutative von Neumann algebra, so that $\H=\oplus_{i\in I}\H_i$ and the action of $\M'$ on $\H_i$ is irreducible for each $i\in I$. Then the following conditions on an anti-unitary $J$ acting on $\H$ are equivalent:
- $J$ commutes with $\M_h$.
- $J(\H_i)\subseteq\H_i$ for $i\in I$.
As $\M_h=\bigoplus^{\infty}_{i\in I}\R \operatorname{id}_{\H_i}$ by Schur’s Lemma, $J\in(\M_h)'^{\R}$ is equivalent to $J(\H_i)\subseteq \H_i$ for $i\in I$.
\[doublecom1\] If, for $m\in M$, the isotropy representation on the fiber $\V_m$ restricted to a subgroup $G^1_m\subseteq G_m$ is multiplicity free with irreducible decomposition $\V_m=\bigoplus_{i\in I} \V_m^{(i)}$ and there exists a $g\in G$ such that $(\gamma_g|_{\V_m})^{-1}\sigma^{\V}_m(\V_m^{(i)})= \V_m^{(i)}$ for all $i\in I$, then the anti-unitary operator $(\gamma_g|_{\V_m})^{-1}\sigma^{\V}_m$ commutes with $\rho_m(G_m)'_h$.
We apply Lemma \[conjcommutedisc\] with $\M=\rho_m(G^1_m)'$, $J=(\gamma_g|_{\V_m})^{-1}\sigma^{\V}_m$ and $\H_i=\V_m^{(i)}$ and conclude that the anti-unitary operator $(\gamma_g|_{\V_m})^{-1}\sigma^{\V}_m$ commutes with $\rho_m(G^1_m)'_h$, and hence also with $\rho_m(G_m)'_h$.
Using Proposition \[doublecom1\], we obtain a special case of Theorem \[multfree1\] which is an infinite dimensional version of [@kob]\*[Theorem 2.2]{}.
\[multfree2\] Let a group $G$ act by automorphisms on the holomorphic Hilbert bundle $q:\V\to M$. Assume that there exists an anti-holomorphic bundle endomorphism $(\sigma^{\V},\sigma^M)$ and a $G$-invariant subset $D$ which is dense in an open subset of $M$ such that for any $m\in M$:
- The isotropy representation on the fiber $\V_m$ restricted to a subgroup $G^1_m\subseteq G_m$ is multiplicity free with irreducible decomposition $\V_m=\bigoplus_{i\in I} \V_m^{(i)}$.
- There exists $g\in G$ such that $(\gamma_g|_{\V_m})^{-1}\sigma^{\V}_m(\V_m^{(i)})= \V_m^{(i)}$ for all $i\in I$.
Then any unitary representation $(\pi,\H)$ of $G$ realized in $\Gamma(\V)$ is multiplicity-free.
Weakly visible actions on complex manifolds
-------------------------------------------
\[defweakvis\]
If a group $G$ acts by holomorphic maps on a connected complex manifold $M$ we say that the action is *$(S,\sigma^M)$-weakly visible* if $S$ is a subset $M$ and $\sigma^M$ is an antiholomorphic diffeomorphism of $M$ satisfying
- \[visible1\] $D=G.S$ is dense in an open subset of $M$.
- \[visible2\] $\sigma^M|_S=\operatorname{id}$.
- \[visible3\] $\sigma^M$ preserves every $G$-orbit in $D$.
\[section\] [Since $S$ meets every $G$-orbit in $D$, there exists a subset $S_0 \subset S$ which meets every $G$-orbit in a single point. Then $S_0$ also satisfies (WV1-3).]{}
[Condition (WV3) in Definition \[defweakvis\] follows from conditions (WV1/2) if there is an automorphism $\sigma^G$ of $G$ such that $$\begin{aligned}
\sigma^G (g).\sigma^M(m)=\sigma^M(g.m) \quad\mbox{for}\quad g\in G,m\in D. \end{aligned}$$ ]{}
[A bundle endomorphism $(\sigma^{\V},\sigma^M)$ is said to be *compatible with the $G$-action and the automorphism $\sigma^G$ of $G$* if $\sigma^{\V}$ is an intertwining map between the action $\gamma$ and the action $\gamma\circ\sigma^G$, i.e., on $\bV$ we have $$\begin{aligned}
\gamma_{\sigma^G(g)}=\sigma^{\V}\circ\gamma_g\circ(\sigma^{\V})^{-1}
\quad \mbox{ for } \quad g \in G. \label{resp}\end{aligned}$$ This implies in particular $$\begin{aligned}
\sigma^G(g).\sigma^M(m)=\sigma^M(g.m) \quad\mbox{for}\quad g\in G,m\in M.
\label{compat}\end{aligned}$$ ]{}
If the anti-holomorphic bundle endomorphism $(\sigma^{\V},\sigma^M)$ is compatible with the $G$-action and with an automorphism $\sigma^G$ of $G$, then the representation $\delta$ of $G$ on $\Gamma(\V)$ satisfies $$\delta_{\sigma^G(g)}=J^{-1}\delta_gJ,\quad\mbox{for}\quad g\in G,$$ where $J$ is the anti-linear operator defined in .
This is immediate from , and .
Conditions on a slice $S$
-------------------------
By using the concept of a weakly-visible action we give a second form of the propagation theorem where the conditions are imposed on the slice $S$ instead on all of $D$.
<span style="font-variant:small-caps;">(Second Propagation Theorem).</span>\[multfreevis\] Let $q:\V\to M$ be a $G$-equivariant holomorphic vector bundle with an anti-holomorphic vector bundle endomorphim $(\sigma^{\V},\sigma^M)$ which is compatible with the $G$-action and the automorphism $\sigma^G$ of $G$. Assume also that
- \[cond21\] The action on $M$ is $(S,\sigma^M)$-weakly visible.
- \[cond22\] For any $s\in S$ the anti-unitary operator $\sigma^{\V}_s$ commutes with $\rho_s(G_s)'_h$.
Then any unitary representation $(\pi,\H)$ of $G$ realized in $\Gamma (\V)$ is multiplicity-free.
We set $D:=G.S$ which is dense in an open subset of $M$ since the action on the base space is $(S,\sigma^M)$-weakly visible. By Remark \[section\] we may assume that each $G$-orbit intersects $S$ only once. Then for $m\in D$ we get $m=g.s$ for some $g\in G$ and a unique $s\in S$.
If we define $c_g:G_s\xrightarrow{\sim} G_m$, $\ell\mapsto g\ell g^{-1}$, then $\gamma_g|_{\V_s}:\V_s\to \V_m$ satisfies $$\xy
\xymatrix{
\V_s \ar[d]_{\rho_s(\ell)} \ar[r]^{\gamma_g} & \V_m \ar[d]^{\rho_m(c_g(\ell))}
\\
\V_s \ar[r]^{\gamma_g} & \V_m
}\endxy$$ for $\ell\in G_s$. Hence, since $G_m=c_g(G_s)$ we get $$\begin{aligned}
((\rho_m(G_m))'_h)'^{\R}=(\gamma_g|_{\V_s})((\rho_s(G_s))'_h)'^{\R}(\gamma_g|_{\V_s})^{-1}.
\label{isomcom}\end{aligned}$$ For $g':=\sigma^G(g)g^{-1}\in G$ we have $$\begin{aligned}
\sigma^M(m)&=\sigma^M(g.s)=\sigma^G(g).\sigma^M(s)&\mbox{by \eqref{compat}}\\
&=\sigma^G(g).s&\mbox{by (WV2) in Def. \ref{defweakvis}}\\
&=\sigma^G(g)g^{-1}g.s=g'.m.\end{aligned}$$ Using we obtain $$\begin{aligned}
(\gamma_{g'}|_{\V_m})^{-1}\sigma_m^{\V}=(\gamma_g|_{\V_s})(\gamma_{\sigma^G(g^{-1})}|_{\V_{\sigma^M(m)}})\sigma_m^{\V}
=(\gamma_g|_{\V_s})\sigma_s^{\V}(\gamma_g|_{\V_s})^{-1}.\end{aligned}$$ Since $\sigma^{\V}_s$ commutes with $\rho_s(G_s)'_h$ we conclude with that $(\gamma_{g'}|_{\V_m})^{-1}\sigma_m^{\V}$ commutes with $\rho_s(G_m)'_h$. Hence all the assumptions of Theorem \[multfree1\] hold and the conclusion follows.
Choosing $g=e$ for $m\in S$ in Proposition \[doublecom1\] we can replace condition (F) in Theorem \[multfreevis\] by the weaker condition:
- For $s\in S$, the isotropy representation on the fiber $\V_s$ restricted to a subgroup $G^1_s\subseteq G_s$ is multiplicity-free with irreducible decomposition $\V_s=\bigoplus_{i\in I} \V_s^{(i)}$ and $\sigma^{\V}_s(\V_s^{(i)})=\V_s^{(i)}$ for any $i\in I$.
We thus obtain a special case of Theorem \[multfreevis\]
Associated bundles {#assocbundles}
==================
In this section we recall from [@bel] and [@neeb13a] how to define complex structures on associated Banach bundles. Based on these complex structures, we prove that certain vector bundle endomorphisms are antiholomorphic.
Let $G$ be a Banach–Lie group with Lie algebra $\g$ and $H\subseteq G$ be a *split Lie subgroup*, i.e., the Lie algebra $\h$ of $H$ has a closed complement in $\g$. Hence the homogeneous space $M:=G/H$ has a smooth manifold structure such that the projection $q_M:G\to G/H, g \mapsto gH$ is a submersion and defines a smooth $H$-principal bundle.
Let $q:\V=G\times_H V\to M$ be a homogeneous vector bundle defined by the norm continuous representation $\rho:H\to \operatorname{GL}(V)$ on a Banach space $V$. We associate to each section $s:M\to \V$ the function ${\widetilde}{s}:G\to V$ specified by $s(gH)=[g,{\widetilde}{s}(g)]$ for $g\in G$. Then a function $f:G\to V$ is of the form ${\widetilde}{s}$ for a section $s$ of $\V$ if and only if $$\begin{aligned}
\label{funsec}
f(gh)=\rho(h)^{-1}f(g)\quad\mbox{for}\quad g\in G,h\in H. \end{aligned}$$ We write $C^{\infty}(G,V)_\rho$ for the set of smooth functions $f:G\to V$ satisfying .
For every $g\in G$, we then have isomorphisms $$\iota_g:V\to \V_{gH}=[g,V],\ v\mapsto [g,v]$$ and the group $G$ acts in a canonical way on $\V\to M$ by bundle automorphism $\gamma_g([g',v])=[gg',v]$ and $g.g'H=gg'H$ for $g,g'\in G$ and $v\in V$.
Complex structures on associated bundles {#compstructassoc}
----------------------------------------
We assume that the coset space $M:=G/H$ carries the structure of a complex manifold such that $G$ acts on $M$ by biholomorphic maps. Let $m_0:=q_M(e)\in M$ be the canonical base point and $\q\subseteq \g_{\C}$ be the kernel of the complex linear extension of the map $\g\to T_{m_0}(G/H)$, so that $\q$ is a complex linear subalgebra of $\g_{\C}$ invariant under $\operatorname{Ad}_H$. We call $\q$ the *subalgebra defining the complex structure* on $M=G/H$ because specifying $\q$ means to identify $T_{m_0}(G/H)\simeq \g/\h$ with the complex Banach space $\g_{\C}/\q$, and thus specifying the complex structure on $M$. The subalgebra $\q$ satisfies $\q+{\overline}{\q}=\g_{\C}$, $\q\cap{\overline}{\q}=\h_{\C}$ and $\operatorname{Ad}_H(\q)=\q$. See [@bel] for further information on complex structures on homogeneous Banach manifolds.
We want to define complex structures on vector bundles $\V=G\times_H V$ over Banach homogeneous spaces $M=G/H$ associated to a norm continuous representation $\rho:H\to \operatorname{GL}(V)$ of the isotropy group on a complex Banach space $V$. Suppose that $H\subseteq G$ is a split Lie subgroup and, as above, $\q\subseteq \g_{\C}$ is a closed complex $\operatorname{Ad}(H)$-invariant subalgebra containing $\h_{\C}$ and specifying the complex structure on $G/H$. If $\rho:G\to\operatorname{GL}(V)$ is a norm continuous representation on the Banach space $V$, then a morphism $\beta:\q\to\gl(V)$ of complex Banach–Lie algebras is said to be an *extension* of $\operatorname{d}\rho$ if $$\begin{aligned}
\operatorname{d}\rho=\beta|_{\h}\quad\mbox{and}\quad\beta(\operatorname{Ad}(h)x)=\rho(h)\beta(x)\rho(h)^{-1} \quad\mbox{for}\quad h\in H ,x\in \q .
\label{defext}\end{aligned}$$
We associate to each $x\in \g$ the left invariant differential operator $L_x$ on $C^{\infty}(G,V)$ given by $$(L_xf)(g)=\frac{d}{dt}\biggl|_{t=0}f(g\exp(tx))$$ and by complex linear extension we define the operators $$\begin{aligned}
L_{x+iy}:=L_x+iL_y\quad\mbox{ for }\quad x,y\in \g.
\label{defdifop}\end{aligned}$$ For any extension $\beta$ of $\rho$, we write $C^{\infty}(G,V)_{\rho,\beta}$ for the subspace of those elements of $C^{\infty}(G,V)_\rho$ satisfying $$\begin{aligned}
L_wf=-\beta(w)f\quad \mbox{ for }\quad w\in\q.
\label{defholsec}\end{aligned}$$
We recall the following theorem from [@neeb13a]\*[Thm. 1.6]{}.
Let $V$ be a complex Banach space and $\rho:H\to \operatorname{GL}(V)$ be a norm continuous representation. Then, for any extension $\beta:\q\to\gl(V)$ of $\rho$, the associated bundle $\V=G\times_H V$ carries a unique structure of a holomorphic vector bundle over $M= G/H$, which is determined by the characterization of holomorphic sections $s:M\to \V$ as those for which ${\widetilde}{s}\in C^{\infty}(G,V)_{\rho,\beta}$. Any such holomorphic bundle structure is $G$-invariant in the sense that $G$ acts on $\V$ by holomorphic vector bundle automorphisms. Conversely, every $G$-invariant holomorphic vector bundle structure on $\V$ is obtained from this construction.
The endomorphism bundle {#end}
-----------------------
Let $\V=G\times_H V\to M$ be a $G$-homogeneous Hilbert bundle as in Theorem \[compstructassoc\], where the representation $\rho$ is unitary. Then the complex manifold $M\times M^{\rm op}$ is a complex homogeneous space $(G\times G)/(H\times H)$, where the complex structure is defined by the closed subalgebra $\q\oplus {\overline}{\q}$ of $\g_\C\oplus\g_\C$. On the Banach space $B(V)$ we consider the norm continuous representation $${\widetilde}{\rho}:H\times H\to \operatorname{GL}(B(V))\quad \mbox{ given by } \quad
{\widetilde}{\rho}(h_1,h_2)A=\rho(h_1)A\rho(h_2)^* = \rho(h_1)A\rho(h_2)^{-1}$$ and the corresponding extension ${\widetilde}{\beta}:\q\oplus{\overline}{\q}\to\gl(B(V))$ by $${\widetilde}{\beta}(x_1,x_2)A=\beta(x_1)A+A\beta(\overline{x_2})^*.$$ We write $\L:=(G\times G)\times_{(H\times H)}B(V)$ for the corresponding holomorphic Banach bundle over $M\times M^{\rm op}$. For every pair $(g_1,g_2)\in G\times G$ we have an isomorphism $$\gamma_{(g_1,g_2)} :B(V)\to B(\V_{q_M(g_2)},\V_{q_M(g_1)}),\quad \gamma_{(g_1,g_2)}(A)[g_2,v]= [g_1,Av].$$ This defines a map $$\gamma :G\times G\times B(V)\to B(\V)=\coprod_{m,n\in M} B(\V_m,\V_n),\quad \gamma (g_1,g_2,A)=\gamma_{(g_1,g_2)}(A).$$ For $h_1, h_2\in H$, we have $$\begin{aligned}
\gamma(g_1h_1,g_2h_2,{\widetilde}{\rho}(h_1,h_2)^{-1}A)[g_2,v]&=\gamma(g_1h_1,g_2h_2,{\widetilde}{\rho}(h_1,h_2)^{-1}A)[g_2h_2,\rho(h_2)^{-1}v]
\\
&=[g_1h_1,\rho(h_1)Av]=[g_1,Av]=\gamma(g_1,g_2,A)[g_2,v]\end{aligned}$$ so that $\gamma$ factors through a bijection ${\overline}{\gamma}: \L\to B(\V)$. This provides a description of the bundle $\L$ as the endomorphism bundle of the Hilbert bundle $\V$, see [@br07].
Holomorphic morphism of equivariant principal bundles
-----------------------------------------------------
Let $q_i:\V_i=G_i\times_{H_i} V_i\to M_i=G_i/H_i$ for $i=1,2$ be homogeneous vector bundles defined by norm continuous unitary representations $\rho_i:H_i\to \operatorname{U}(V_i)$ on Hilbert spaces $V_i$.
Let $\lambda:G_1\to G_2$ be a homomorphism of Banach–Lie groups satisfying $\lambda(H_1)\subseteq H_2$, so that there is an induced map $\lambda^M:M_1\to M_2$ defined by $gH_1\mapsto \lambda(g)H_2$. Let $\psi:V_1\to V_2$ be an operator such that $$\begin{aligned}
\rho_2(\lambda(h))\circ\psi=\psi\circ\rho_1 (h) \quad\mbox{for}\quad h\in H_1.\label{intertwin}\end{aligned}$$ We say that $(\lambda,\psi)$ is a morphism between the representations $\rho_1$ and $\rho_2$. These conditions imply that the map $$\lambda^{\V}:\V_1\to \V_2,\quad [g,v]\mapsto [\lambda (g),\psi (v)]$$ is well defined. It is a lift of the map $\lambda^M$ and it is complex linear on each fiber. If we differentiate we obtain $$\begin{aligned}
\operatorname{d}\rho_2(\operatorname{d}\lambda(w))\circ\psi=\psi\circ\operatorname{d}\rho_1 (w) \quad\mbox{for}\quad w\in \h_1.\end{aligned}$$ If the $M_i=G_i/H_i$ carry complex structures defined by subalgebras $\q_i$, then $\lambda^M$ is holomorphic if and only if the complex linear extension $\operatorname{d}\lambda_{\C}:\g_{1,\C}\to\g_{2,\C}$ satisfies $\operatorname{d}\lambda_{\C}(\q_1)\subseteq \q_2$. Assume further that $q_i:\V_i\to M_i$ have complex bundle structures defined by extensions $\beta_i:\q_i\to \gl(V_i)$ of $\operatorname{d}\rho_i$ for $i=1,2$. If the intertwining operator $\psi$ also satisfies $$\begin{aligned}
\beta_2(\operatorname{d}\lambda(w))\circ\psi=\psi\circ\beta_1 (w) \quad\mbox{for}\quad w\in \q_1,\label{intertwindif}\end{aligned}$$ then we say that $(\lambda,\psi)$ is a *morphism* between the representation $(\rho_1,\beta_1)$ of $(H,\fq)$ on $V_1$ and the representation $(\rho_2,\beta_2)$ on $V_2$.
\[holoend\] Assume that $(\lambda,\psi)$ is an intertwiner for the $H$-representations $(\rho_1, V_1)$ and $(\rho_2, V_2)$. Then $(\lambda,\psi)$ is a morphism between the representation $(\rho_1,\beta_1)$ of $V_1$ and the representation $(\rho_2,\beta_2)$ of $V_2$ if and only if $\lambda^M:M_1\to M_2$, $gH_1\mapsto \lambda(g)H_2$ and $\lambda^{\V}:\V_1\to \V_2$ are holomorphic.
The condition $\operatorname{d}\lambda_{\C}(\q_1)\subseteq \q_2$ holds if and only if the induced map $\lambda^M:M_1\to M_2$, $gH_1\mapsto \lambda(g)H_2$ is holomorphic. Assume that this is the case. We represent $\lambda^{\V}$ in local trivializations $$\alpha_i:U_i\times V_i\to \V_i|_{U_i},\quad (gH_i,v)\mapsto
[g,F_i(g)v]\quad\mbox{for}\quad i=1,2,$$ where $U_i\subseteq M_i$ are open subsets and $$F_i:q_M^{-1}(U_i)\to \operatorname{GL}(V_i)$$ satisfies $$\begin{aligned}
&F_i(gh)=\rho_i(h)^{-1}F_i(g) \quad\mbox{for}\quad g\in q_M^{-1}(U_i),h\in H_i\nonumber\\
&L_wF_i=-\beta_i (w)F_i \quad\mbox{for}\quad w\in \q_i\label{difext1}\end{aligned}$$ (see the proof of [@neeb13a]\*[Thm 1.6]{}). Observe that $$\xy
\xymatrix{
(gH_1,v) \ar@{|->}[d] \ar@{|->}[r] & [g,F_1(g)v]\ar@{|->}[d]^{\lambda^{\V}}
\\
(\lambda(g)H_2, F_2(\lambda (g))^{-1}\psi F_1(g)v) & [\lambda(g),\psi(F_1(g)v)]\ar@{|->}[l]
}\endxy$$ so that in these trivializations the map is given by $$(gH_1,v)\mapsto (\lambda(g)H_2, F_2(\lambda (g))^{-1}\psi F_1(g)v)
\quad\mbox{for}\quad gH_1\in U_1\cap (\lambda^M)^{-1}(U_2),v\in V.$$ Since the map $F_2:q_M^{-1}(U_2)\to \operatorname{GL}(V_2)$ satisfies we have $$\begin{aligned}
\label{adifext2}
L_w(F_2\circ\lambda)=L_{\operatorname{d}\lambda(w)}(F_2)\circ\lambda=-\beta_2 (\operatorname{d}\lambda(w))(F_2\circ\lambda).\end{aligned}$$ The map $gH_1\mapsto \lambda(g)H_2$ is holomorphic and the map $F_2(\lambda (g))^{-1}\psi F_1(g) {\colon}V_1\to V_2$ is complex linear. We have to prove that the map $gH_1\mapsto F_2(\lambda (g))^{-1}\psi F_1(g)\in B(V_1,V_2)$ for $gH_1\in U_1\cap (\lambda^M)^{-1}(U_2)$ is holomorphic, i.e., that the map $g\mapsto F_2(\lambda (g))^{-1}\psi F_1(g)$ is annihilated by the differential operators $L_w$ for $w\in\q$, if and only if holds. Observe that $$\begin{aligned}
L_w((F_2\circ\lambda)^{-1} (\psi F_1))=&L_w((F_2\circ\lambda)^{-1})(\psi F_1)+(F_2\circ\lambda)^{-1}L_w(\psi F_1)\\
=&-(F_2\circ\lambda)^{-1}L_w(F_2\circ\lambda)(F_2\circ\lambda)^{-1}(\psi F_1)+(F_2\circ\lambda)^{-1}L_w(\psi F_1)\\
=&-(F_2\circ\lambda)^{-1}(-\beta_2(\operatorname{d}\lambda(w)))(F_2\circ\lambda)(F_2\circ\lambda)^{-1}(\psi F_1)\\
&+(F_2\circ\lambda)^{-1}(\psi L_w F_1)\qquad\qquad \qquad\qquad\qquad \qquad
\mbox{ by \eqref{adifext2}}\\
=&-(F_2\circ\lambda)^{-1}(-\beta_2(\operatorname{d}\lambda(w))(\psi F_1))\\
&+(F_2\circ\lambda)^{-1}\psi (-\beta_1(w) F_1)
\qquad \qquad\qquad\qquad \qquad \mbox{ by \eqref{difext1}}\\
=&(F_2\circ\lambda)^{-1}(\beta_2(\operatorname{d}\lambda(w))\psi -\psi (\beta_1(w)))F_1.\end{aligned}$$ Therefore $L_w((F_2\circ\lambda)^{-1} (\psi F_1))=0$ if and only if $\beta_2(\operatorname{d}\lambda(w))\psi -\psi (\beta_1(w))=0$ for ${w\in \q_1}$.
[It is easy to check that the correspondence between morphisms of representations and holomorphic bundle morphisms is functorial. In particular one has for a smooth Banach $G$-module $W$ the following linear bijection, which is a variant of Frobenius reciprocity: $$\label{eq:frob}
\Gamma {\colon}\operatorname{Hom}_G(W,C^{\infty}_{\rho,\beta}(G,V))\to \operatorname{Hom}_{\rho,\beta}(W,V),\quad
\Phi\mapsto {\rm ev}_e \circ \Phi.$$ Here the intertwining operators are assumed to be continuous and the space $C^{\infty}_{\rho,\beta}(G,V)$ carries the locally convex topology of pointwise convergence. Then the inverse of $\Gamma$ is simply given by $\Gamma^{-1}(\varphi)(w)(g)=\varphi(\pi(g^{-1})w)$. ]{}
Opposite complex structure on associated bundles {#oppositebundle}
------------------------------------------------
Given an associated vector bundle $q:\V=G\times_{H, \rho}V\to M=G/H$ endowed with a complex structure defined in Subsection \[compstructassoc\] by a subalgebra $\q\subseteq \g_{\C}$ and an extension $\beta$ of $\operatorname{d}\rho$, we can define a bundle $q:\V^{\rm op}\to M^{\rm op}$ with the same underlying real spaces and opposite complex structures on the total and base spaces.
\[propoppositebundle\] If ${\overline}{\rho}:H\to\operatorname{U}(V^{\rm op})$ is the complex conjugate representation of $\rho$ then the map ${\overline}{\beta}:{\overline}{\q}\to \gl(V^{\rm op})$ given by ${\overline}{\beta}(w):=\overline{\beta(\overline{w})}$ for $w\in{\overline}{\q}$ is an extension of of $\operatorname{d}{\overline}{\rho}$. The bundle $q:\V^{\rm op}=G\times_{H,{\overline}{\rho}}V^{\rm op}\to M^{\rm op}=G/H$ with the complex structure defined by the subalgebra ${\overline}{\q}\subseteq \g_{\C}$ and the extension ${\overline}{\beta}$ of $\operatorname{d}{\overline}{\rho}$ has the same holomorphic sections as $q:\V=G\times_{H,\rho}V\to M=G/H$, so it carries the opposite complex structure on total and base space, respectively.
The algebra ${\overline}{\q}$ defines the opposite complex structure on $M=G/H$. Since ${\overline}{\beta}$ is a complex linear Lie algebra homomorphism and since $$\operatorname{d}{\overline}{\rho}(w)=\overline{\operatorname{d}\rho(w)}=\overline{\beta(w)}=\overline{\beta(\overline{w})}={\overline}{\beta}(w)\quad\mbox{for}\quad w\in \h$$ we get $\operatorname{d}{\overline}{\rho}={\overline}{\beta}|_{\h}$. Also, $$\begin{aligned}
{\overline}{\beta}(\operatorname{Ad}(h)w)&=\overline{\beta(\overline{\operatorname{Ad}(h)w})}={\overline}{\beta}(\operatorname{Ad}(h)\overline{w})={\overline}{\rho}(h)\overline{\beta(\overline{w})}{\overline}{\rho}(h)^{-1}={\overline}{\rho}(h){\overline}{\beta}(w){\overline}{\rho}(h)^{-1}\end{aligned}$$ for $h\in H$ and $w\in {\overline}{\q}$, so that ${\overline}{\beta}:{\overline}{\q}\to \gl(V^{\rm op})$ is an extension of $\operatorname{d}{\overline}{\rho}$.
For a map $f:G\to V$ we write ${\overline}{f}:G\to V^{\rm op}$ for the same map to $V$ with the opposite complex structure. Then $${\overline}{f}(gh)={\overline}{\rho}(h)^{-1}{\overline}{f}(g)\quad\mbox{for}\quad g\in G,h\in H,$$ and $$L_w{\overline}{f}=\overline{L_{\overline{w}}f}
=-\overline{\beta(\overline{w})f}=-\overline{\beta(\overline{w})}{\overline}{f}
=-{\overline}{\beta}(w){\overline}{f}\quad \mbox{ for }\quad w\in{\overline}{\q}.$$ Therefore $C^{\infty}(G,V)_{\rho,\beta}= C^{\infty}(G,V^{\rm op})_{{\overline}{\rho},{\overline}{\beta}}$, and the bundles $q:\V\to M$ and $q:\V^{\rm op}\to M^{\rm op}$ have the same holomorphic sections. As both bundles have the same sections and $ M^{\rm op}$ and $V^{\rm op}$ carry the complex structure opposite to the one $M$ and $V$, respectively, the bundle $\V^{\rm op}$ carries the complex structure opposite to $\V$.
Anti-holomorphic automorphisms of equivariant principal bundles
---------------------------------------------------------------
Assume that there is an automorphism $\sigma$ of $G$ which stabilizes $H$ and such that the complex linear extension $\operatorname{d}\sigma_{\C}:\g_{\C}\to\g_{\C}$ of $\operatorname{d}\sigma$ satisfies $\operatorname{d}\sigma_{\C}(\q)={\overline}{\q}$, i.e., the induced map $\sigma^M$ on $M=G/H$ defined by $gH\mapsto \sigma(g)H$ is anti-holomorphic. The condition ${\overline}{\rho}\circ\sigma\simeq\rho$, where ${\overline}{\rho}:H\to \operatorname{U}(V^{\rm op})$ is the complex conjugate representation, holds if and only if there is an anti-unitary operator $\psi:V\to V$ satisfying $$\begin{aligned}
\rho(\sigma(h))\circ\psi=\psi\circ\rho (h) \quad\mbox{for}\quad h\in H.\label{dualrep}\end{aligned}$$ This condition implies that the endomorphism of $\V$ given by $$\sigma^{\V}:\V\to \V,\quad [g,v]\mapsto [\sigma (g),\psi (v)]$$ is well-defined. It is a lift of the anti-holomorphic map $\sigma^M:M\to M$, $gH\mapsto \sigma(g)H$ and it is anti-unitary on each fiber.
If the antiunitary map $\psi {\colon}V \to V$ also satisfies $$\begin{aligned}
\beta(\overline{\operatorname{d}\sigma(w)})\circ\psi=\psi\circ\beta (w)\label{dualrepdif}
\quad \mbox{ for } \quad w\in \q, \end{aligned}$$ then $(\sigma,\psi)$ is a morphism between $(\rho,\beta)$ and $({\overline}{\rho},{\overline}{\beta})$. If we consider the endomorphism $(\sigma^{\V},\sigma^M)$ of $\V$ as a morphism $$\xy
\xymatrix{
\V \ar[d]_{q} \ar[r]^{\sigma^{\V}} & \V^{\rm op} \ar[d]^{q}
\\
M \ar[r]^{\sigma^M} & M^{\rm op},
}\endxy$$ then Propositions \[holoend\] and \[propoppositebundle\] lead to:
\[aholoend\] If $(\sigma,\psi)$ intertwines the representations $(\rho,\beta)$ and $({\overline}{\rho},{\overline}{\beta})$, then ${\sigma^M:M\to M}$, $gH\mapsto \sigma(g)H$ and $\sigma^{\V}:\V\to \V$, $[g,v]\mapsto [\sigma (g),\psi (v)]$ are anti-holomorphic.
A Propagation Theorem for associated bundles {#propagassocbund}
============================================
Based on the constructions of the previous section we give a formulation of the propagation theorem for infinite-dimensional associated holomorphic vector bundles, which is best suited for specific examples.
Let $K$ be a subgroup of the connected Banach–Lie group $G$. For $g\in G$, the isotropy group of $m=gH$ for the action of $K$ on $M=G/H$ is $K_m=K\cap G_m=K\cap gHg^{-1}$. We define a homomorphism $$c_{g^{-1}}:K_{gH}\to H,\quad k\mapsto g^{-1}kg$$ with range $K_{(g)}:=c_{g^{-1}}(K_{gH})=g^{-1}Kg\cap H\subseteq H$. If $\sigma$ is an automorphism of $G$ such that $\sigma(K)=K$ and $\sigma(H)=H$, for $g\in G^{\sigma}$ we get $\sigma(K_{(g)})=K_{(g)}$.
<span style="font-variant:small-caps;">(Third Propagation Theorem).</span>\[multfreegroup\] Let $\V=G\times_{H,\rho} V\to M$ be a $G$-equivariant holomorphic vector bundle with a complex subalgebra $\q$ and an extension $\beta:\q\to \gl(V)$ of $\operatorname{d}\rho$ defining a complex structure on it. Let $K\subseteq G$ be a subgroup and $\sigma$ be an automorphism of the Banach–Lie group $G$ stabilizing $K$ and $H$ such that $(\sigma,\psi)$ is a morphism between the representations $(\rho,\beta)$ and $({\overline}{\rho},{\overline}{\beta})$. Suppose that there is a subset $B$ of $G^{\sigma}$ such that:
- \[cond31\] The closure of the subset $KBH{\subseteq}G$ has interior points.
- \[cond32\] For every $b\in B$ the anti-unitary operator $\psi$ commutes with $\rho(K_{(b)})'_h$.
Then any unitary representation $(\pi,\H)$ of $K$ realized in $\Gamma (\V)$ is multiplicity free.
Since $(\sigma,\psi)$ is a morphism between the representations $(\rho,\beta)$ and $({\overline}{\rho},{\overline}{\beta})$, by Proposition \[aholoend\] the bundle endomorphisms $(\sigma^{\V},\sigma^M)$ given by $$\sigma^{\V}([g,v])=[\sigma^G(g),\psi(v)]\quad\mbox{and}\quad \sigma^M(gH)
=\sigma(g)H \quad\mbox{for}\quad g\in G, v\in V$$ is antiholomorphic.
The antiholomorphic vector bundle endomorphim $(\sigma^{\V},\sigma^M)$ is compatible with the $K$-action and with the automorphism $\sigma^K:=\sigma|_K$ of $K$ since for $k\in K$, $g\in G$ and $v\in V$ $$\sigma^M(k.gH)=\sigma^M(kg).H=\sigma^K(k).\sigma(g).H=\sigma^K(k).\sigma^M(gH),$$ and $$\begin{aligned}
(\gamma_{\sigma^K(k)}\circ\sigma^{\V})([g,v])&=\gamma_{\sigma^K(k)}([\sigma(g),\psi(v)])=[\sigma^K(k)\sigma(g),\psi(v)]\\
&=([\sigma(kg),\psi(v)])=\sigma^{\V}([kg,v])=(\sigma^{\V}\circ\gamma_k)([g,v]).\end{aligned}$$
Since $KBH$ is dense in an open subset of $G$, the $K$-invariant subset $D:=KBH/H
\subseteq G/H$ is dense in an open subset of $M=G/H$. We set $$S:=BH/H\subseteq D,$$ so that $K.S=D$. Then $\sigma^M|_S=\operatorname{id}$ because $B\subseteq G^{\sigma^G}$. For $s\in S$ we have $$\sigma^M(K.s)=\sigma^K(K).\sigma^M(s)=K.s.$$ Therefore the $K$-action on the base space is $(S,\sigma^M)$-weakly visible and condition (B) in Theorem \[multfreevis\] is satisfied.
Let $m=gH$ for $g\in G$. Via the bijection $\iota_g:V\xrightarrow{\sim}\V_{gH}$, $v\mapsto [g,v]$ and the group homomorphism $c_{g^{-1}}:K_{gH}\to K_{(g)}$ the isotropy representation $\rho_m:K_m\to \operatorname{U}(\V_m)$ factors through the representation $\rho|_{K_{(g)}}:K_{(g)}\to \operatorname{U}(V)$, namely $$\xy
\xymatrix{
V \ar[d]_{\rho(c_{g^{-1}}(k))} \ar[r]^{\sim}_{\iota_g} & \V_m \ar[d]^{\rho_m(k)}
\\
V \ar[r]^{\sim}_{\iota_g} & \V_m,
}\endxy$$ and thus $$\iota_g\rho(K_{(g)})\iota_g^{-1}=\rho_s(K_m).$$
Let $s\in S$ and $b\in B$ with $s=bH$. Observe that $$\begin{aligned}
\iota_{\sigma (g)}\circ \psi=\sigma^{\V}_{gH}\circ\iota_g\quad\mbox{for}\quad g\in G,\end{aligned}$$ and $\sigma(b)=b$, so that $\sigma_s^{\V}=\iota_b\psi\iota_b^{-1}$. Since $\psi$ commutes with $\rho(K_{(b)})'_h$ we conclude that $\sigma_s^{\V}=\iota_b\psi\iota_b^{-1}$ commutes with $(\iota_b\rho(K_{(b)})\iota_b^{-1})'_h=\rho_s(K_s)'_h$ so that condition (F) in Theorem \[multfreevis\] holds as well.
\[samesubgroup\]
Note that $$N:=Z_{K\cap H}(B)=\{k\in K\cap H:kb=bk \quad\mbox{for}\quad b\in B\}\subseteq b^{-1}Kb\cap H=K_{(b)} \quad \mbox{ for } \quad b\in B.$$ Since the commutant of $\rho(N)'_h$ is contained in the commutant of $\rho(K_{(b)})'_h$, we obtain a corollary of Theorem \[multfreegroup\] by replacing condition (F) by:
- The anti-unitary operator $\psi$ commutes with $\rho(N)'_h$ for $N=Z_{K\cap H}(B)$.
[ In the previous formulation of the theorem, if the subset $B$ is bigger, condition (B) becomes weaker and condition (F’) stronger. ]{}
\[multfreedual\] If ${\overline}{\rho}\circ\sigma\simeq\rho$ as representations of $H$, with the isomorphism given by an anti-unitary $\psi$, the restriction $\rho|_{K^1}$ to a subgroup $K^1\subseteq H$ is multiplicity free with irreducible decomposition $\rho|_{K^1}\simeq\bigoplus_{i\in I}\nu^{(i)}$ on $V=\bigoplus_{i\in I} V^{(i)}$ and ${\overline}{\nu}^{(i)}\circ\sigma\simeq\nu^{(i)}$ as representations of $K^1$, then $\psi$ commutes with $\rho(K^1)'_h$.
Since ${\overline}{\nu}^{(i)}\circ\sigma\simeq\nu^{(i)}$ as representations of $K^1$, there are anti-unitary $K^1$-intertwining operators $\psi_i:V^{(i)}\to V^{(i)}$ for $i\in I$. Then the unitary operator $\Psi := \psi^{-1}\circ \bigoplus_{i\in I}\psi_i$ commutes with the representation $\rho|_{K^1}\simeq\bigoplus_{j\in I} \nu^{(j)}$ which is multiplicity free. Therefore $\Psi(V^{j})= V^{j}$ for every $j\in I$, which implies that $\psi(V^{j})= V^{j}$ for $j\in I$. By Lemma \[conjcommutedisc\], this means that the anti-unitary operator $\psi$ commutes with $\rho(K^1)'_h$.
\[cond2’\]
With Proposition \[multfreedual\], we obtain a special case of Theorem \[multfreegroup\] where the representations on the fibers are discretely decomposable by changing condition (F) to:
- For every $b\in B$ the restriction of $\rho$ to a subgroup $K^1_{(b)}\subseteq K_{(b)}$ is multiplicity free with irreducible decomposition $\rho|_{K^1_{(b)}}\simeq\bigoplus_{i\in I}\nu^{(i)}_b$ on $V=\bigoplus_{i\in I} V^{(i)}_b$ and ${\overline}{\nu}^{(i)}_b\circ\sigma\simeq\nu^{(i)}_b$ as representations of $K^1_{(b)}$.
By Proposition \[multfreedual\] and Remark \[samesubgroup\] we obtain the following reformulation of Theorem \[multfreegroup\].
\[multfreegroup2\] Let $\V=G\times_{H,\rho} V\to M$ be a $G$-equivariant holomorphic vector bundle with a complex subalgebra $\q$ and an extension $\beta:\q\to \gl(V)$ of $\operatorname{d}\rho$ defining a complex structure on it. Let $K\subseteq G$ be a subgroup and $\sigma$ be an automorphism of the Banach–Lie group $G$ stabilizing $K$ and $H$ such that $(\sigma,\psi)$ is a morphism between the representations $(\rho,\beta)$ and $({\overline}{\rho},{\overline}{\beta})$. Suppose also that there is a subset $B$ of $G^{\sigma}$ such that:
- \[cond41\] The subset $KBH$ of $G$ satisfies $\overline{KBH}^\mathrm{o}\neq \emptyset$.
- \[cond42\] For $N=Z_{K\cap H}(B)$ the restriction $\rho|_N$ is multiplicity free with irreducible decomposition $\rho|_N\simeq\bigoplus_{i\in I} \nu^{(i)}$ on $V=\bigoplus_{i\in I} V^{(i)}$ and ${\overline}{\nu}^{(i)}\circ\sigma\simeq\nu^{(i)}$ for $i\in I$ as representations of $N$.
Then any unitary representation $(\pi,H)$ of $K$ is realized in $\Gamma (\V)$ is multiplicity free.
Examples of weakly visible actions and propagation theorems {#exampleswv}
===========================================================
In this final section we discuss various concrete situations in the context of operator algebras and Hilbert spaces which fit into the setting developped above. In particular we exhibit several kinds of weakly visible actions on infinite dimensional spaces and state some corresponding propagation theorems.
Propagation theorem for linear base spaces {#proplinear}
------------------------------------------
Let $\H$ be a complex Hilbert space and let $\Fo(\H)\subseteq \O(\H)$ be the Fock space on $\H$ with reproducing kernel $K(x,y)=e^{\langle x,y\rangle}$ for $x,y\in \H$. Let $(e_j)_{j\in J}$ be an orthonormal basis of $\H$. Then the polynomial functions $p_m:\H\to\C$ $$p_m(z)=\frac{1}{\sqrt{m!}}\prod_{j\in J}\langle z,e_j\rangle^{m_j}
\quad\mbox{for}\quad m\in \N_{0}^{(J)}$$ form an orthonormal basis of $\Fo(\H)$, where $\N_{0}^{(J)}$ is the set of finitely supported tuples indexed by $J$ and $m!=\prod_{j\in J}m_j!$. The Hilbert subspace of $\Fo(\H)$ of homogeneous functions of degree $n\in\N_0$ has reproducing kernel $K_n(x,y)=\frac{1}{n!}\langle x,y\rangle$ for $x,y\in \H$.
Let $\delta:H\to\operatorname{U}(\H)$ be a norm-continuous representation of a Banach–Lie group $H$ and let $G:=\H\rtimes_{\delta}H$, so that $M:=G/H\simeq \H$. Let $\rho:H\to \operatorname{U}(V)$ be another norm-continuous representation, so we can define an associated (trivial) vector bundle $$q:\V=G\times_{\rho,H}V\to G/H\simeq \H.$$ Its space of holomorphic sections $\Gamma(\V)$ can be identified with the space of holomorphic functions $\O(\H,V)$, where a section of the bundle $$s:(z,1)(\{0\}\times H)\mapsto [(z,1),f(z,1)]$$ is defined by a holomorphic function $f':\H\to V$, $f'(z)=f(z,1)$ for $z\in \H$.
With the representation $\delta:H\to\operatorname{U}(\H)$ we define a representation $\delta':H\to\operatorname{U}(\Fo(\H))$, $(\delta'_hg)(z)=g(\delta_{h^{-1}}z)$ for $h\in H$, $g\in \Fo(\H)$ and $z\in \H$. We can realize the representation $\delta'\otimes\rho$ of $H$ in the space of holomorphic sections of the bundle via the embedding $$\begin{aligned}
&\Fo(\H)\otimes V\to \O(\H,V)\simeq \Gamma(\V)\\
&f\otimes v\mapsto (z\mapsto f(z)v).\end{aligned}$$ For $z\in \H$ we have $\operatorname{ev}_z:\Fo(\H)\otimes V\to V$, $f\otimes v\mapsto f(z)v$ and $(\operatorname{ev}_z)^*:V\to \Fo(\H)\otimes V$, $(\operatorname{ev}_z)^*(w)=e^{\langle \cdot,z\rangle}w$ so that the reproducing kernel is given by $$Q(y,z):=(\operatorname{ev}_y)(\operatorname{ev}_z)^*=(\operatorname{ev}_y)e^{\langle \cdot,z\rangle}\operatorname{id}_V
=e^{\langle y,z\rangle}\operatorname{id}_V=e^{\langle y,z\rangle}\operatorname{id}_V.$$
\[multfreegrouplinear\] Let $\sigma$ be an involution on $H$, and let $\sigma^{\H}$ be an anti-unitary operator on $\H$ such that $$\delta_{\sigma(h)}=\sigma^{\H}\circ\delta_{h} \circ (\sigma^{\H})^{-1},$$ and $\psi$ be an anti-unitary operator on $V$ such that $$\rho_{\sigma(h)}=\psi\circ\rho_{h} \circ \psi^{-1}\quad \mbox{ for } \quad h \in H.$$ Suppose further that there is a subset $S$ in the fixed point set of $\sigma^{\H}$ such that:
- \[cond51\] The closure of $\delta(H)S$ in $\cH$ has interior points.
- \[cond52\] The operator $\psi$ commutes with the hermitian operators in the commutant $\rho(Z_H(S))'$, where $Z_H(S) := \{ h \in H : (\forall s \in S)\ \delta_h(s)=s\}$ is the pointwise stabilizer of $S$ in $H$.
Then the representation $\delta'\otimes\rho$ of $H$ on $\Fo(\H)\otimes V$ is multiplicity free.
For the subset $B=S\times\{1\}\subseteq\H\times\{1\}\subseteq G$, the relations $(b,1)(0,h)=(b,h)$ and $(0,h)(b,1)=(\delta_hb,h)$ for $h\in H$ and $b\in B$ lead to $$Z_H(B)=\{h\in H:\delta_h(b)=b\quad\mbox{for}\quad b\in B\}.$$
If $B$ is in the fixed point set of $\sigma^{\H}$, $HBH$ is dense in an open subset of $G$ (which is equivalent to ${\overline}{H.B}^\circ \not={\emptyset}$ in $\cH$), and $\psi$ commutes with the hermitian operators in $\rho(Z_H(B))'$, then Theorem \[multfreegroup\] with condition (F’) in Remark \[samesubgroup\] implies that the representation $\delta'\otimes\rho$ of $H$ on $\Fo(\H)\otimes V$ is multiplicity free.
\[cormultfreegrouplinear\] Let $\sigma$ be an involution on $H$, $\sigma^{\H}$ an anti-unitary operator on $\H$ such that $$\delta_{\sigma(h)}\circ \sigma^{\H}=\sigma^{\H}\circ\delta_{h}.$$ Suppose further that $S {\subseteq}\cH^{\sigma^\cH}$ is such that the closure of $\delta(H)S$ in $\cH$ has interior points. Then the representation $\delta'$ of $H$ on $\Fo(\H)$ is multiplicity free.
Hilbert-Schmidt operators
-------------------------
In Section 5.6 of [@kob2] finite dimensional examples of visible actions on linear spaces are presented. We extend some of these results to the context of Hilbert–Schmidt operators. We denote by $B_2(\H_1,\H_2)$ the Hilbert–Schmidt operators from $\H_1$ to $\H_2$ and by $$\operatorname{U}_2(\H):=\operatorname{U}(\H)\cap({\mathbf 1}+B_2(\H))$$ the unitary Hilbert–Schmidt perturbations of the identity. The first result is about torus actions and the remaining results involve an approximate Cartan decomposition in this context.
The two sided action of the group $\operatorname{U}_2(\H)\times \operatorname{U}_2(\H)$ on $\operatorname{GL}_2(\H)$ or $B_2(\H)$ is weakly visible. We take an orthonormal basis $(e_n)_{n\in \N}$ of $\H$ and define a conjugation on $\H$ by $J(\sum_n\alpha_ne_n)=\sum_n\overline{\alpha_n}e_n$ and an automorphism of $\operatorname{GL}_2(\H)$ by $\sigma(g)=J gJ$. If we define $S$ as the subset of positive diagonal operators in $\operatorname{GL}_2(\H)$, the Cartan decomposition and a finite-dimensional approximation argument imply that $D:=(\operatorname{U}_2(\H)\times \operatorname{U}_2(\H)).S=\operatorname{U}_2(\H)S \operatorname{U}_2(\H)$ is dense in $\operatorname{GL}_2(\H)$ and also in $B_2(\H)$. Furthermore the $\operatorname{U}_2(\H)\times \operatorname{U}_2(\H)$-orbits are $\sigma$-invariant since $\sigma(\operatorname{U}_2(\H))=\operatorname{U}_2(\H)$, and $\sigma(s)=s$ for $s\in S$. Therefore the action is $(S,\sigma)$-weakly visible.
\[torushilbert\]
A particular case is the multiplication action of the abelian Banach–Lie group $$\ell^2(\N,\R)/\Z^{(\N)} \cong
\exp(i\ell^2(\N,\R))\subseteq \operatorname{U}_2(\ell^2(\N,\C))
= \operatorname{U}(\ell^2(\N,\C)) \cap ({\mathbf 1}+ B_2(\ell^2(\N,\C))$$ on $\ell^2(\N,\C)$. It is $S$-weakly-weakly visible if we take $S=\ell^2(\N,\R)$ and $\sigma$ is conjugation on $\ell^2(\N,\C)$. Here the finitely supported sequences $f:\N\to \C$ are contained in $D:=\exp(i\ell^2(\N,\R)).\ell^2(\N,\R)$, so that this subset is dense in $\ell^2(\N,\C)$. The sequence $f(n)=\frac{i}{2^n}$ is contained in $\ell^2(\N,\C)$ but not in $D$ because $(-i,-i,-i,\ldots)$ is not contained in $\exp(i\ell^2(\N,\R))$.
Assume $H:=\exp(i\ell^2(\N,\R))$ and $\ell^2(\N,\C)$ are endowed with the canonical involutions. Corollary \[cormultfreegrouplinear\] implies that the representation of $H$ on $\Fo(\ell^2(\N,\C)){\subseteq}\cO(\ell^2(\N,\C))$ is multiplicity free. This is the fact that in the Taylor expansion of a function $f\in\Fo(\ell^2(\N,\C))$ each monomial appears exactly once. In the finite-dimensional context this is the most basic example of a multiplicity-free representation (see the introduction of [@kob2]).
[ Let $\H=L^2(X,\mu)$ with a $\sigma$-finite measure space $(X,\mu)$. Then the multiplication algebra $\A=L^{\infty}(X,\mu)$ is maximal abelian in $B(L^2(X,\mu))$. If $\sigma(f) = \overline{f}$ is complex conjugation on $L^2(X,\mu)$ and $S=L^2(X,\mu;\R)$ denotes the subspace of real-valued functions, then the action of the unitary group $\operatorname{U}_{\A}$ of $\cA$ on $L^2(X,\mu)$ is $(S,\sigma)$-visible. If $\mu$ is infinite, then the action of the subgroup $\exp(i L^2(X,\mu;\R))\subseteq \operatorname{U}_{\A}$ is $(S,\sigma)$-weakly visible. ]{}
\[linearmfhs\]
Let $\H_1$ be a closed subspace of the Hilbert space $\H_2$ and take an orthonormal basis $(e_i)_{i\in I_1}$ of $\H_1$ and an orthonormal basis $(e_i)_{i\in I_2}$ of $\H_2$ such that $I_1\subseteq I_2$. Consider the action $\delta$ of $H:=\operatorname{U}_2(\H_1)\times \operatorname{U}_2(\H_2)$ on $M=B_2(\H_1,\H_2)$ given by $$\label{eq:doubleact}
\delta(u_1,u_2)(x)=u_2xu_1^{-1}.$$
Consider the subset $$S:=\{A\in B_2(\H_1,\H_2):(\forall j\in I)\ A(e_j)\in\R e_j\}$$ and the conjugations of $\H_1$ and $\H_2$ given by $J_2(\sum_{j\in I_2}\alpha_j e_j)=\sum_{j\in I_2}\overline{\alpha_j}e'_j$ and $J_1=J_2|_{\H_1}$ respectively. Define a conjugation on $B_2(H_1,\H_2)$ by $\sigma^M(x)=J_2 xJ_1$. It is easy to verify that the action is $(S,\sigma^M)$-weakly visible and compatible with the automorphism of $\operatorname{U}_2(\H_1)\times \operatorname{U}_2(\H_2)$ given by $\sigma(u_1,u_2)=(J_1u_1J_1,J_2u_2J_2)$. Corollary \[cormultfreegrouplinear\] implies that the representation of $H$ on $\Fo(B_2(\H_1,\H_2)){\subseteq}\cO(B_2(\H_1,\H_2))$ is multiplicity free.
\[kacexample\]
Let $\H_1$ be a closed subspace of the Hilbert space $\H_2$ and take an orthonormal basis $(e_i)_{i\in I_1}$ of $\H_1$ and an orthonormal basis $(e_i)_{i\in I_2}$ of $\H_2$ such that $I_1\subseteq I_2$. Consider the action $\delta$ of $\operatorname{U}_2(\H_1)\times \operatorname{U}_2(\H_2)$ on $M := B_2(\H_1,\H_2\oplus\C)\simeq B_2(\H_1,\H_2)\oplus~\H_1$ given by $$\delta(u_1,u_2)(x,\xi)=(u_2xu_1^{-1},u_1\xi),$$ the subset $$S':=S\oplus \H_1^{J_1}, \quad \mbox{ where } \quad
\H_1^{J_1} :=\{\xi\in\H_1:J_1\xi=\xi\}$$ with $S$ as in Example \[linearmfhs\], and the conjugation on $B_2(\H_1,\H_2)\oplus\H_1$ given by $\sigma(x,\xi)=(J_2xJ_1,J_1\xi)$. The action is $(S',\sigma)$-weakly visible since the action of the subgroup $$\begin{aligned}
\{(u,(u,\operatorname{id}_{\H_1^{\perp}}))\in\operatorname{U}_2(\H_1)\times \operatorname{U}_2(\H_2):&
(\forall j \in I_1)\ u(e_j)\in \T e_j\}\end{aligned}$$ fixes $S$ and rotates the vectors in a dense subset of $\H_1$ into $\H_1^{J_1}$.
Corollary \[cormultfreegrouplinear\] implies that the action of $H$ on $$\Fo(B_2(\H_1,\H_2)\oplus\H_1)\simeq \Fo(B_2(\H_1,\H_2))\otimes\Fo(\H_1) {\subseteq}\cO(B_2(\H_1,\H_2)\oplus\H_1)$$ is multiplicity free.
[ Example \[linearmfhs\] can be interpreted as the isotropy representation of the group $\operatorname{U}_2(\H_1)\times \operatorname{U}_2(\H_2)$ on the tangent space at the base point of the restricted Graßmannian $$G_{\rm res}(\cH_1\oplus \cH_2) \cong \operatorname{U}_2(\H)/(\operatorname{U}_2(\H_1)\times \operatorname{U}_2(\H_2)).$$ ]{}
\[linearmfwedge\] [ Assume the context and notation of Example \[linearmfhs\], and for simplicity we set $\H_1 = \cH_2 = \cH$. The action $\delta$ of $H:=\operatorname{U}_2(\H)\times \operatorname{U}_2(\H)$ on $\cK := B_2(\cH)$ given by $$\delta(u_1,u_2)(x)=u_2xu_1^{-1}$$ is a unitary representation $\delta:H\to \operatorname{U}(\cK)$. As uniformly bounded isotropy representations, we can take for example the representations $$\rho(u_1, u_2) :=
\rho_1(u_1) \oplus \rho_2(u_2),\quad \mbox{ where } \quad
\rho_j:= \bigoplus_{k=1}^{m_j}\Lambda^{n_{j,k}}.$$ Here, for $n\in\N$ the representation $\Lambda^{n}:\operatorname{U}_2(\H)\to \operatorname{U}(\Lambda^{n}\H)$ is defined by $$\Lambda^{n}(u)(v_1\wedge\ldots\wedge v_n)=uv_1\wedge\ldots\wedge uv_n.$$ Consider the group $$N=Z_H(B)=\{(u,u)\in \operatorname{U}_2(\H_1)\times
\operatorname{U}_2(\H_2):(\forall j \in I)\ u(e_j)\in \T e_j\},$$ where $$B=\{x\in B_2(\H_1,\H_2):x(e_i)\in\R e'_{i}\quad\mbox{for}\quad j\in J\}.$$ Since $$\Lambda^{n}(\operatorname{diag}(t_j)_{j\in J})(e_{j_1}\wedge\ldots\wedge e_{j_n})
=t_{j_1}\ldots t_{j_n}(e_{j_1}\wedge\ldots\wedge e_{j_n}),$$ if the $n_{j,k}$ for $j=1,2$ and $k=1,\ldots,m_j$ are all distinct, $\rho|_N$ is multiplicity free. We can take the canonical anti-unitary operators given by complex conjugation on $$\Lambda^{n}\big(\ell^2(J)\big)\subseteq \bigotimes^{n}\ell^2(J)=\ell^2(J^n).$$ Theorem \[multfreegrouplinear\] implies that the representation $\delta'\otimes\rho$ of $H=\operatorname{U}_2(\H)\times \operatorname{U}_2(\H)$ is multiplicity-free. ]{}
\[linearmftrace\] [For a Hilbert space $\cH$, let $\operatorname{U}_1(\H) :=\operatorname{U}(\H)\cap(1+B_1(\H))$ denote the group of unitary operators which are trace class perturbations of the identity. In the previous example we can take instead of $\operatorname{U}_2(\H)\times \operatorname{U}_2(\H)$ the group $\operatorname{U}_1(\H)\times \operatorname{U}_1(\H)$ to construct isotropy representations from the operator determinant. The isotropy representation is $\rho(u_1, u_2) = \rho_1(u_1) \oplus \rho_2(u_2)$, where $$\rho_j = \bigoplus_{k=1}^{m_j}(\operatorname{det}_{\operatorname{U}_1(\H)})^{n_{j,k}}.$$ If the $n_{j,k}$ for $j=1,2$ and $k=1,\ldots,m_j$ are all distinct, then $\rho|_N$ is multiplicity free, where $$N=Z_H(B)=\{(u,u)\in \operatorname{U}_1(\H_1)\times
\operatorname{U}_1(\H_2): (\forall j \in I)\ u(e_j)\in \T e_j\}.$$ ]{}
Finite von Neumann algebras
---------------------------
Some basic examples of type II$_1$ factors have representations such that the Hilbert space where the factor $\cM$ acts can be endowed with an anti-unitary operator $J$ such that $$\label{eq:j-cond}
J\cM J = \cM \quad \mbox{ and } \quad Jx J = x\quad\mbox{ for }x\in\cA_h$$ for a maximal abelian $*$-subalgebra $\cA$ of $\cM$ (a [*masa*]{} for short), see Examples \[exvncartype\] and \[exgroupmeasurespace\]. We are going to use Theorem \[multfreegrouplinear\] to construct examples of representations of the product $H := \operatorname{U}_{\cM}\times\operatorname{U}_{\cM}$ of the unitary group of a finite factor $\cM$, where the base is the GNS construction of $\cM$, the slice $S$ consists of the hermitian operators in a masa, and the conjugations are constructed from the conjugation $J$.
We denote by $L^2(\M)$ the GNS Hilbert space, which is the completion of $\M$, endowed with the inner product $\langle x,y\rangle=\tau(xy^*)$ for $x,y\in \cM$, where $\tau$ is the trace of the algebra $\M$. Let the unitary representation $\delta:H\to \operatorname{U}(\H)$ of $H=U_{\M}\times U_{\M}$ on $\H=L^2(\M)$ given by $$\delta(u_1,u_2)(x)=u_1 xu_2^{-1}
\quad \mbox{ for } \quad u_1,u_2\in U_{\M}, x\in \M\subseteq L^2(\M).$$ Let $J$ be as in . If we define $\sigma(x)=J xJ$ for $x\in \M$, then $\sigma$ extends to an anti-unitary involution $\sigma^{L^2(\M)}$ on $L^2(\M)$ compatible with the involutions on $U_{\M}$ and $H=U_{\M}\times U_{\M}$ given by $\sigma(u)=J uJ$ for $u\in U_{\M}$, and $\sigma(h)=(J,J) h(J,J)$ for $h\in H=U_{\M}\times U_{\M}$, respectively. In fact, for $u_1,u_2\in U_{\M}$ and $x\in \M\subseteq L^2(\M)$, we have $$\begin{aligned}
(\sigma(u_1),\sigma(u_2)).\sigma^{L^2(\M)}(x)&=\sigma(u_1)\sigma^{L^2(\M)}(x)\sigma(u_2^{-1})=(J u_1J)J xJ(J u_2^{-1}J)\\
&=J u_1xu_2^{-1}J=\sigma^{L^2(\M)}(u_1xu_2^{-1})
=\sigma^{L^2(\M)}((u_1,u_2).x).\end{aligned}$$
Note that $\sigma^{L^2(\M)}$ fixes $S:=\A_h$ pointwise. To prove that the action of $\operatorname{U}_{\M}\times \operatorname{U}_{\M}$ on $L^2(\M)$ is $(S,\sigma^{L^2(\M)})$-weakly visible it remains to show that $D:=U_{\M}\A_h U_{\M}$ is dense in $L^2(\M)$. This is a consequence of the following proposition.
Let $\M$ be type II$_1$-factor with trace $\tau$, let $\A\subseteq\M$ be a masa, and denote the cone of positive invertible elements in $\A$ by $\A^+$. Then $D:=U_{\M}\A^+ U_{\M}$ is dense in $L^2(\M)$
For a self-adjoint operator $x\in \M$ there exists a unique Borel probability measure $m_x$ on $\R$ such that $$\int_{-\infty}^{+\infty}\lambda^ndm_x(\lambda)=\tau(x^n)\quad\mbox{for}\quad n\in \{0\}\cup\N.$$ If, conversely, $m$ is a compactly supported probability measure on $\R$, then there is a self-adjoint operator $x\in \A$ with spectral measure $m_x = m$, see [@ak06]\*[Prop. 5.2]{}. The norm closure of the unitary orbit $\O(x)=\{uxu^{-1}:u\in \operatorname{U}_{\M}\}$ of a self-adjoint $x\in \M$ consists of the selfadjoint operators in $\M$ with the same spectral measure as $x$ ([@ak06]\*[Thm. 5.4]{}), so the spectral measure is a complete invariant of approximate unitary equivalence.
Then the polar decomposition implies that $D:=U_{\M}\A^+ U_{\M}$ is dense in the unit group $\M^\times$ of $\M$. In [@ch70]\*[Thm. 5]{} it is shown that a von Neumann algebra $\cM$ has dense unit group in the norm topology if and only if $\M$ is of finite type. Therefore, if $\M$ is a finite type factor, then $D$ is norm dense in $\M$. Since on $\M\subseteq L^2(\M)$, the Hilbert space norm is dominated by the uniform norm, $D:=U_{\M}\A^+ U_{\M}$ is dense in $L^2(\M)$ with its Hilbert space topology.
[ As the example after [@ak06]\*[Thm. 5.4]{} shows, there exist selfadjoint operators whith the same spectral measure which are not unitarily equivalent. ]{}
Set $S:=\A_h$, so that $$N=Z_H(S\times\{1\})=\{(t,t)\in U_{\M}\times U_{\M}:t\in U_{\A}\}.$$ If there is a unitary representation $\rho:U_{\M}\times U_{\M}\to U(V)$ on a Hilbert space $V$, and there is an antiunitary involution $\psi$ on $V$ such that
- $\rho_{\sigma(h)}\circ \psi=\psi\circ\rho_{h} \quad \mbox{ for } \quad h \in H$.
- $\psi$ commutes with the hermitian operators in $\rho(\{(t,t)\in U_{\M}\times U_{\M}:t\in U_{\A}\})'$
then Theorem \[multfreegrouplinear\] implies that the representation $\delta'\otimes\rho$ of $H=U_{\M}\times U_{\M}$ on $\Fo(L^2(\M))\otimes V$ is multiplicity free.
We now turn to examples of factors $\cM {\subseteq}B(\cH)$ for which there exists an anti-unitary operator $J$ such that $J\cM J = \cM$ and $Jx J = x$ for selfadjoints $x$ in some masa $\cA {\subseteq}\cM$.
[ There are two types of masas $\cA$ in a II$_1$ factor $\cM$ specified in terms of the algebra generated by the normalizer of the masa. The normalizer of the masa is defined by $$N(\cA)=\{u\in U_{\cM}:u\cA u^*=\cA\}.$$ The masa $\cA$ is called [*regular*]{} or [*Cartan*]{} if $N(\cA)''=\cM$, and it is called singular if $N(\cA)''=\cA$, see the first paragraph of [@ss08]\*[3.2]{}. ]{}
\[exvncartype\] [ (von Neumann algebras of CAR type). Let $M_2(\C)$ be the $2\times 2$ complex matrices with diagonal masa $D$. For each $k\in \N$ let $M_k$ be a copy of $M_2(\C)$ with a copy of $D$ as diagonal masa $D_k$. The algebra $\cM_n=\otimes_{k=1}^{n}M_k\simeq M_{2^n}(\C)$ has a masa $\cA_n=\otimes_{k=1}^{n}D_k$. We have embeddings $\cM_n\hookrightarrow\cM_{n+1}$, $x\mapsto x\otimes \operatorname{id}_{M_2(\C)}$ which preserve the normalized traces. Therefore $\bigcup_{n=1}^{\infty}\otimes_{k=1}^nM_k$ is a $*$-algebra with trace $$\tau(\otimes_{k=1}^{\infty}x_k)=\prod_{k=1}^{\infty}\operatorname{tr}(x_k),$$ where all but finitely many $x_i$ are equal to the identity of $M_2(\C)$. The weak closure $\cM$ of the GNS representation of $\bigcup_{n=1}^{\infty}\otimes_{k=1}^nM_k$ is a copy of the hyperfinite II$_1$-factor and the weak closure $\cA$ of $\bigcup_{n=1}^{\infty}\otimes_{k=1}^nD_k$ in this representation is a Cartan masa, see the first part of Subsection 3.4 in [@ss08]. The canonical complex conjugation on $\bigcup_{n=1}^{\infty}\otimes_{k=1}^nM_k$ yields a complex conjugation $J$ on the GNS space such that $J\M J=\M$, $Jx J=x$ for $x\in\cA_h$ and $\tau(JxJ)=\overline{\tau(x)}$. ]{}
[ In [@st80] it is shown that there is, up to conjugacy, a unique real von Neumann algebra $\cR$ in the hyperfinite factor $\cM$ of type II$_1$, i.e. $\cR$ is a $*$-algebra over the reals such that $\cR+\sqrt{-1}\cR=\cM$ and $\cR\cap\sqrt{-1}\cR=\{0\}$. It follows that any two involutive antilinear automorphisms of this factor are conjugate under $\operatorname{Aut}(\cR)$. This contrasts with the situation in $B(\H)$, where there are two distinct conjugacy classes of such automorphisms, induced by conjugation with antiunitary operators $J$ satisfying $J^2 = \pm {\mathbf 1}$. ]{}
\[exgroupmeasurespace\]
(Group-measure space construction). See Subsection 8.6 in [@kr97] for detailed information on this construction. Let $\Gamma\curvearrowright (X,\mu)$ be a probability measure preserving action of a countable discrete group $\Gamma$. We consider the unitary [*Koopman representation*]{} $$\alpha:\Gamma\to\operatorname{U}(L^2(X,\mu)), \qquad (\alpha_sf)(x)=f(s^{-1}.x)
\quad \mbox{ for } \quad s\in\Gamma, f\in L^2(X,\mu), x\in X.$$ Let $\lambda:\Gamma\to\operatorname{U}(\ell^2(\Gamma))$ be the left regular representation. We regard $L^{\infty}(X)\simeq L^{\infty}(X)\otimes 1\subseteq B(L^2(X)\otimes \ell^2(\Gamma))$. For $s\in \Gamma$ we consider the unitaries $u_s=\alpha_s\otimes \lambda_s\in\operatorname{U}(L^2(X)\otimes \ell^2(\Gamma))$. The crossed product von Neumann algebra is defined as $$L^{\infty}(X)\rtimes\Gamma :=\Big\{\sum_{s\in \Gamma}a_su_s:
a_s\in L^{\infty}(X)\Big\}''\subseteq B(L^2(X)\otimes \ell^2(\Gamma)).$$ The trace is given by the extension of $$\tau\bigg(\sum_{s\in\Gamma}a_su_s\bigg)=\int_Xa_ed\mu.$$ The action is called [*free*]{} if $\A:=L^{\infty}(X)$ is maximal abelian in $L^{\infty}(X)\rtimes\Gamma$, and in this case $L^{\infty}(X)$ is a Cartan subalgebra. The von Neumann algebra $L^{\infty}(X)\rtimes\Gamma$ is a II$_1$ factor if and only if the action $\Gamma\curvearrowright X$ is ergodic, i.e., $L^{\infty}(X,\mu)^{\Gamma}=\C 1$.
Let $J$ be the complex conjugation on $L^2(X)\otimes \ell^2(\Gamma)\simeq L^2(X\times \Gamma)$. Observe that $J a J=\overline{a}$ for $a\in L^{\infty}(X)$ and $J u_s J=u_s$ for $s\in \Gamma$. Therefore $J(L^{\infty}(X)\rtimes\Gamma)J=L^{\infty}(X)\rtimes\Gamma$ and $J x J=x$ for $x\in\cA_h$. Note that $\tau(J x J)=\overline{\tau(x)}$ for $x\in L^{\infty}(X)\rtimes\Gamma$.
Symmetric spaces
----------------
In Subsections 5.3 and 5.4 of [@kob2] finite dimensional examples of visible actions on symmetric spaces are presented. The action of $H$ on $G/H$ for a vast class of symmetric spaces is studied in [@kob08]. We begin by presenting weakly visible actions on Graßmannians and symmetric domains modeled on Banach spaces. We first present a weak visibility result for the group of bounded invertible operators acting on a Hilbert space to illustrate the approximate Cartan decomposition which is involved in the subsequent arguments.
\[cartanunif\] [ Let $\H$ be a Hilbert space with orthonormal basis $(e_j)_{j\in I}$ and define a complex conjugation on $\H$ by $J(\sum_j\alpha_j e_j )=\sum_j \overline{\alpha_j}e_j$ and an automorphism on $\operatorname{GL}(\H)$ by $\sigma(g)=J gJ$. If we define $S$ as the set of positive diagonal operators in $\operatorname{GL}(\H)$ then $D:=(\operatorname{U}(\H)\times \operatorname{U}(\H)).S=\operatorname{U}(\H)S \operatorname{U}(\H)$ is dense in $\operatorname{GL}(\H)$: For $g\in\operatorname{GL}(\H)$ we have the polar decomposition $g=up$, with $u\in \operatorname{U}(\H)$ and an invertible $p>0$. Given $\epsilon>0$ we can use the measurable functional calculus of $p$ to find a positive invertible $q$ with finite spectrum such that $\|q-p\|<\epsilon$. Since $q$ has finite spectrum, there is a unitary $v$ such that $s:=vqv^{-1}\in S$. Hence $$\|g-(uv^{-1})sv\|=\|up-uq\|<\epsilon.$$ Furthermore the $\operatorname{U}(\H)\times \operatorname{U}(\H)$-orbits are preserved under $\sigma$ since $\sigma(\operatorname{U}(\H))=\operatorname{U}(\H)$, and $\sigma(s)=s$ for $s\in S$, so the action is $(S,\sigma)$-weakly visible. ]{}
\[pcss\]
Let $\cH_1$ and $\cH_2$ be complex Hilbert spaces. We consider the identical representation of $\operatorname{U}(\K)$ on the complex Hilbert space $\K=\H_1\oplus\H_2$. Then the subgroup $Q:=\{g\in\operatorname{GL}(\K):g\H_1=\H_1\}$ is a complex Lie subgroup of $\operatorname{GL}(\K)$ and the Graßmannian $\operatorname{Gr}_{\H_1}(\K):=\operatorname{GL}(\K)\H_1\simeq\operatorname{GL}(\K)/Q$ carries the structure of a complex homogeneous space on which the unitary group $G=\operatorname{U}(\K)$ acts transitively
[Here we use that, for $\cE := g\cH_1$, the orthogonal space $\cE^\bot$ is the image of $\cH_2$ under $(g^{-1})^*$. Hence there exists a unitary isomorphism $\cE \oplus \cE^\bot \to \cH_1 \oplus \cH_2$ mapping $\cE$ to $\cH_1$.]{}
and which is isomorphic to $G/H$ for $H:=\operatorname{U}(\K)_{\H_1}\simeq\operatorname{U}(\H_1)\times\operatorname{U}(\H_2)$. Writing elements of $B(\H)$ as $(2\times 2)$-matrices according to the decomposition $\K=\H_1\oplus \H_2$, we have $$\q=
\Big\{ {\begin{pmatrix} a & b \\ c & d \end{pmatrix}} {\colon}a\in B(\H_1),b\in B(\H_2,\H_1),d\in B(\H_2)\Big\}.$$ and $\q\simeq\p^+\rtimes\h_{\C}$ and $\gl(\K)=\q\oplus\p^-$ holds for $$\p^-=
\Big\{ {\begin{pmatrix} 0 & 0 \\ c & 0 \end{pmatrix}} {\colon}c\in B(\H_1,\H_2) \Big\}\quad \mbox{ and }\quad
\p^+=\overline{\p^-}.$$
Assume that $\H_1 := \H \oplus \{0\}$ and $\H_2 := \{0\} \oplus \H$ for a Hilbert space $\H$. We fix an orthonormal basis $(e_i)_{i \in I}$ in $\H$. We can define a conjugation $J$ on $\H$ by $J(\sum_{j\in I}\alpha_j e_j)=\sum_{j\in I}\overline{\alpha_j}e_j$ and defined $J_{\K}(v,w) := (Jv,Jw)$ on $\K$. Let $\sigma(u)=J_{\K} uJ_{\K}$ be the corresponding involution on $\operatorname{U}(\K)$ and $\sigma^M(u\H_1)=J u\H_1=\sigma(u)\H_1$ be the corresponding involution on the Graßmannian $\operatorname{Gr}_{\H_1}(\K)$.
Consider the embedding $$\tau:\B(\H)\to \g,\quad A\mapsto
{\begin{pmatrix} 0 & A \\ -A^* & 0 \end{pmatrix}},$$ the diagonal real subalgebra of $\B(\H)$ given by $$\cA_{\R}=\{A\in\B(\H): (\forall j \in I)\
A(e_j)\in\R e_j\},$$ and let $B:=\exp(\tau(A_{\R}))$, where $\exp:\g\to G$ is the exponential map of $\operatorname{U}(\K)$. Observe that $B\subseteq G^{\sigma}$ and that $$\begin{aligned}
Z_{H}(B)=\{(u,u)\in\operatorname{U}(\H)^2 {\colon}(\forall j \in I)\
u(e_j) \in \T e_j\}.\end{aligned}$$ To prove that the action of $K=H=\operatorname{U}(\H)^2$ on $\operatorname{Gr}_{\H_1}(\K)$ is $(B,\sigma^M)$-weakly-visible it remains to show that $$\overline{\operatorname{U}(\H)^2\exp(\tau(\cA_{\R}))\operatorname{U}(\H)^2}^\mathrm{o}\neq \emptyset.$$ Observe that $$HBH=\operatorname{U}(\H)^2\exp(\tau(\operatorname{U}(\H)\cA_{\R}\operatorname{U}(\H)))\operatorname{U}(\H)^2.$$ For ${\varepsilon}>0$, the argument used to prove the approximate Cartan decomposition of Example \[cartanunif\] leads to $$\overline{\operatorname{U}(\H)(\cA_{\R}\cap B_{{\varepsilon}/2}({\varepsilon}{\mathbf 1}))\operatorname{U}(\H)}
=B_{\leq{\varepsilon}/2}({\varepsilon}{\mathbf 1}).$$ We choose ${\varepsilon}>0$ small enough to ensure that $\tau(B_{{\varepsilon}/2}({\varepsilon}{\mathbf 1}))$ is contained in a neighborhood of $0$ on which $\exp$ is a local diffeomorphism, and the result follows.
\[ncss\]
Let $\K=\H\oplus\H$, $\H_1 := \H \oplus \{0\}$, and $\H_2 := \{0\} \oplus \H$ for a Hilbert space $\H$. We fix an orthonormal basis $(e_j)_{j \in I}$ in $\H$. We can define a conjugation $J$ on $\H$ by $J(\sum_{j\in I}\alpha_j e_j)=\sum_{j\in I}\overline{\alpha_j}e_j$ and define $J_{\K}(v,w) := (Jv,Jw)$ on $\K$.
We endow the Hilbert space $\K$ with the indefinite hermitian form given by $$h((v_1,v_2),(w_1,w_2))=\langle v_1,w_1 \rangle
-\langle v_2, w_2\rangle .$$ We can write $\D:=\{z\in B(\cH):\|z\|<1\}$ as $G/H$, where $G$ is the pseudo-unitary group $$G=\operatorname{U}(\H_1,\H_2)=\{g\in \operatorname{GL}(\K):h(g.v,g.v)=h(v,v)\quad\mbox{for all}\quad
v\in\K\},$$ and $H=\operatorname{U}(\H_1)\times\operatorname{U}(\H_2)\cong \operatorname{U}(\cH)^2$ is the subgroup of diagonal matrices in $G$. In fact, $G$ acts transitively on $M$ by fractional linear transformations $\bigl(\begin{smallmatrix}
a&b \\ c&d
\end{smallmatrix} \bigr).z=(az+b)(cz+d)^{-1}$, where the $(2\times 2)$-block matrix is written according to the decomposition of $\K$. The stabilizer of $0\in \D$ is the group $H$.
A conjugation on $Z:=B_2(\H,\H)$ is given by $\sigma^Z(x)=J xJ$. Let $$\tau:\B(\H)\to \g,\quad A\mapsto
{\begin{pmatrix} 0 & A \\ A^* & 0 \end{pmatrix}}.$$ Consider the diagonal real subalgebra of $\B(\H)$ given by $$\cA_{\R}=\{A\in\B(\H): (\forall j \in I)\ A(e_j)\in \R e_j\},$$ and let $B:=\exp(\tau(\cA_{\R}))$, where $\exp(X) = \sum_{n = 0}^\infty \frac{X^n}{n!}$ is the operator exponential. Observe that $B\subseteq G^{\sigma}$ and that $$\begin{aligned}
Z_{H}(B)=\{(u_1,u_2)\in\operatorname{U}(\H)\times \operatorname{U}(\H):&
(\forall j \in I)\ u_1(e_j)\in \T e_j, u_2(e_j)\in \T e_j\}. \end{aligned}$$ To prove that the action of $K=H=\operatorname{U}(\H_1)\times \operatorname{U}(\H_2)$ on $D$ is $(B,\sigma^M)$-weakly-visible it remains to show that $$\overline{\operatorname{U}(\H)^2.\exp(\tau(\cA_{\R})).\operatorname{U}(\H)^2}^\mathrm{o}\neq \emptyset.$$ This follows from an argument as in the last part of Example \[ncss\]. The action is also compatible with the automorphism of $\operatorname{U}(\H_1,\H_2)$ given by $\sigma(u)=J_{\K}uJ_{\K}$.
\[pcssr\]
Let $\K$ be a complex Hilbert space which is a direct sum $\K=\H_1\oplus\H_2$. $$B_{\operatorname{res}}(\K):=
\Big\{{\begin{pmatrix} a & b \\ c & d \end{pmatrix}} \in B(\K):c\in B_2(\H_1,\H_2),b\in B_2(\H_2,\H_1) \Big\}$$ is a complex Banach-$*$-algebra. Its unit group is $$\operatorname{GL}_{\operatorname{res}}(\K)=\operatorname{GL}(\K)\cap B_{\operatorname{res}}(\K).$$ The restricted unitary group is $$\operatorname{U}_{\operatorname{res}}(\K)=\operatorname{U}(\K)\cap \operatorname{GL}_{\operatorname{res}}(\K).$$ The homogeneous space $\operatorname{Gr}_{\operatorname{res}}:=\operatorname{U}_{\operatorname{res}}(\K)/(\operatorname{U}(\H_1)\times\operatorname{U}(\H_2))$ is the restricted Graßmannian. There is an isomorphism $$\operatorname{U}_2(\K)/(\operatorname{U}_2(\H_1)\times\operatorname{U}_2(\H_2))\simeq \operatorname{U}_{\operatorname{res}}(\K)_0/(\operatorname{U}(\H_1)\times\operatorname{U}(\H_2)),$$ where $\operatorname{U}_{\operatorname{res}}(\K)_0$ is the connected component of $\operatorname{U}_{\operatorname{res}}(\K)$ given by operator $(2\times 2)$-block matrices in $\operatorname{U}_{\operatorname{res}}(\K)$ with diagonal operators with Fredholm index equal to zero.
As in the previous example we assume that $\H_1 := \H \oplus \{0\}$ and $\H_2 := \{0\} \oplus \H$ for a Hilbert space $\H$. Let $$\tau: B_2(\H)\to \g,\quad A\mapsto
{\begin{pmatrix} 0 & A \\ -A^* & 0 \end{pmatrix}}.$$ Consider the diagonal real subalgebra of $B_2(\H_1)$ given by $$\cA_{\R}=\{f\in B_2(\H): f(e_i)\in\R e_i\quad\mbox{for}\quad i\in I\},$$ and let $B:=\exp(\tau(\cA_{\R}))$, where $\exp:\g\to G$ is the exponential map of $\operatorname{U}_2(\H)$. Observe that $B\subseteq G^{\sigma}$ and that $$\begin{aligned}
Z_{H}(B)=\{(u,u)\in\operatorname{U}_2(\H)^2:& u(e_i)\in \T
e_i \quad \mbox{for}\quad i\in I\}.\end{aligned}$$ To prove that the action of $K=H=\operatorname{U}_2(\H_1)\times \operatorname{U}_2(\H_2)$ on the restricted Gassmannian is $(B,\sigma^M)$-weakly-visible it remains to show that $$\overline{\operatorname{U}_2(\H)^2.\exp(\tau(\cA_{\R})).\operatorname{U}_2(\H)^2}^\mathrm{o}\neq \emptyset.$$ Observe that $$HBH=\operatorname{U}_2(\H)^2\exp(\tau(\operatorname{U}_2(\H)\cA_{\R}\operatorname{U}_2(\H)^2))
\operatorname{U}_2(\H)^2.$$ A finite dimensional approximation argument leads to $$\overline{\operatorname{U}_2(\H)\cA_{\R}\operatorname{U}_2(\H)}=B_2(\H),$$ and the result follows.
\[ncssr\]
The restricted pseudo-unitary group is $$G=\operatorname{U}_{\operatorname{res}}(\H_1,\H_2)=\operatorname{U}(\H_1,\H_2)\cap\operatorname{GL}_{\operatorname{res}}(\H).$$ It acts transitively on $$\D=\{z\in B_2(\H_2,\H_1):\|z\|<1\}$$ by fractional linear transformations. The stabilizer of $0$ in $G$ is $H=\operatorname{U}(\H_1)\times\operatorname{U}(\H_2)$.
We can adapt the previous example to this restricted group context as we did with both examples of positively curved symmetric spaces.
Approximate triunity
--------------------
In [@kob04] three multiplicity-free results are shown to stem from a single geometry. The basic result is the following, which we state in a form suited to our infinite dimensional context.
\[approxtriun\] If $G$ is a topological group with subgroups $K$, $B$ and $H$, then the following conditions are equivalent:
1. $KBH$ is dense in $G$.
2. $HBK$ is dense in $G$.
3. $\operatorname{diag}(G)(B\times B)(K\times H)$ is dense in $G\times G$.
The equivalence of (1) and (2) is trivial. If (3) holds, then the image $KBH$ of $\operatorname{diag}(G)(B\times B)(K\times H)$ under the continuous map $\tau {\colon}G \times G \to G,
\tau(g,h) := g^{-1}h$ is dense in $G$, which is (1).
Now we assume that (1) and (2) hold. From $(hbk,e)=(h,h)(b,e)(k,h^{-1})$ we conclude that $$\overline{HBK}\times\{e\}\subseteq \overline{\operatorname{diag}(G)(B\times B)(K\times H)}.$$ From a similar equation we conclude that $$\{e\} \times\overline{KBH}\subseteq\overline{\operatorname{diag}(G)(B\times B)(K\times H)},$$ so that (1) and (2) imply (3).
The next theorem was proved in [@kob04]\*[Thm. 3.1]{}:
\[thm:6.24\] Let $n_1+n_2+n_3=p+q=n$, and consider the naturally embedded groups $K:=\operatorname{U}(n_1)\times\operatorname{U}(n_2)\times\operatorname{U}(n_3)$ and $H:=\operatorname{U}(p)\times\operatorname{U}(q)$ in $G:=\operatorname{U}(n)$. We define an automorphism $\sigma$ of $G$ by $\sigma(g)=\overline{g}$ and let $B:=G^{\sigma}=\operatorname{O}(n)$. Then $G=KBH$ is equivalent to $\min(p,q)\leq 2$ or $\min(n_1,n_2,n_3)\leq 1$.
Note that $BH/H\simeq \operatorname{Gr}_p(\R^n)$ is the real Graßmannian of $p$ dimensional subspaces in $\R^n$ and that $BK/K\simeq \B_{n_1,n_1+n_2}(\R^n)$ is the real flag manifold of pairs of subspaces $(F_1,F_2)$ of dimension $n_1$ and $n_1+n_2$ respectively such that $F_1\subseteq F_2$. Using a finite dimensional approximation argument we can prove a version of Theorem \[thm:6.24\] in the case of groups of operators which are Hilbert-Schmidt plus identity.
\[triun\] Let $I_1\cup I_2\cup I_3=J_1\cup J_2=I$ be partitions of a countable infinite index set $I$, let $(e_i)_{i\in I}$ be an orthonormal basis of a Hilbert space $\H$ and denote $\H_J:=\operatorname{span}\{e_j\}_{j\in J}$ for $J\subseteq I$. Consider the naturally embedded groups $K:=\operatorname{U}_2(\H_{I_1})\times\operatorname{U}_2(\H_{I_2})\times\operatorname{U}_2(\H_{I_3})$ and $H:=\operatorname{U}_2(\H_{J_1})\times\operatorname{U}_2(\H_{J_2})$ in $G:=\operatorname{U}_2(\H)$. We define an automorphism $\sigma$ of $G$ by $\sigma(u)=\overline{u}=JuJ$, where $J$ is the canonical complex conjugation on $\H=\ell^2(I)$, and let $B:=G^{\sigma}\cong\operatorname{O}_2(\H)$. Then $KBH$ is dense in $G$ if and only if $\min(|J_1|,|J_2|)\leq 2$ or $\min(|I_1|,|I_2|,|I_3|)\leq 1$, where $|I|$ denotes the cardinality of a set $I$.
[ For a Hilbert space $\H$, we write $\P(\H)$ for its projective space, i.e., the set of all $1$-dimensional subspaces of $\H$. The standard action of $\exp(i\ell^2(\N,\R))$ on $\P(\ell^2(\N,\C))$ is weakly visible if we take $S=\P(\ell^2(\N,\R))$ and the canonical complex conjugation on $\ell^2(\N,\C)$. This follows from the density of $KBH$ in $G$, whith $G=\operatorname{U}_2(\ell^2(\N,\C))$, $H=\operatorname{U}(1)\times\operatorname{U}_2(\ell^2(\N_{>1},\C))$, $K=\exp(i\ell^2(\N,\R))
\subseteq \operatorname{U}_2(\ell^2(\N,\C))$ and $B=G^{\sigma}$. Here $\sigma(u)=\overline{u}=JuJ$, where $J$ is the canonical complex conjugation on $\ell^2(\N,\C)$. ]{}
Lemma \[approxtriun\] can be used to prove multiplicity-free branching rules for representations realized on spaces of holomorphic sections of line bundles over flag manifolds as in Section VII of [@neeb04] or [@neeb12]. Lemma \[approxtriun\](3) can be used to prove the multiplicity-free property of tensor product representations by taking the diagonal action of a group $G$ on a product of line bundles over spaces $G/H$ and $G/K$ as in Example 2.4 of [@kob04].
Acknowledgments {#acknowledgments .unnumbered}
===============
K.-H. Neeb acknowledges support by DFG-grant NE 413/9-1. M. Miglioli acknowledges support by a CONICET postdoctoral fellowship and a DAAD grant for short term visit, and is grateful for the excellent working conditions provided by the FAU Erlangen-Nürnberg.
[XXXXx]{}
W. Arveson, R. Kadison, [*Diagonals of self-adjoint operators.*]{} in “Operator Theory, Operator Algebras, and Applications,” 247–263, Contemp. Math., 414, Amer. Math. Soc., Providence, RI, 2006.
D. Beltiţă, [*Integrability of analytic almost complex structures on Banach manifolds.*]{} Ann. Global Anal. Geom. 28 (2005), no. 1, 59–73.
D. Beltiţă, S. Patnaik, G. Weiss, [*Cartan subalgebras of operator ideals.*]{} Indiana Univ. Math. J. 65 (2016), no. 1, 1–37.
D. Beltiţă, T. S. Ratiu, [*Geometric representation theory for unitary groups of operator algebras.*]{} Adv. Math. 208 (2007), no. 1, 299–317.
R. V. Kadison, J. R. Ringrose, “Fundamentals of the theory of operator algebras. Vol. II. Advanced theory. Corrected reprint of the 1986 original.” Graduate Studies in Mathematics, 16. American Mathematical Society, Providence, RI, 1997.
H. Choda, [*An extremal property of the polar decomposition in von Neumann algebras.*]{} Proc. Japan Acad. 46 (1970), 341-344.
J. Faraut, J., and E. G. F. Thomas, [*Invariant Hilbert Spaces of Holomorphic Functions*]{}, J. Lie Theory [**9**]{} (1999), 383–402
D. Farenick, [*Fundamentals of Functional Analysis.*]{} Universitext. Springer, 2016.
T. Kobayashi, [*Geometry of multiplicity-free representations of ${\rm GL}(n)$, visible actions on flag varieties, and triunity.*]{} Acta Appl. Math. 81 (2004), no. 1-3, 129–146.
T. Kobayashi, [*Multiplicity-free representations and visible actions on complex manifolds.*]{} Publ. Res. Inst. Math. Sci. 41 (2005), no. 3, 497–549.
T. Kobayashi, [*Multiplicity-free theorems of the restrictions of unitary highest weight modules with respect to reductive symmetric pairs.*]{} Representation theory and automorphic forms, 45–109, Progr. Math., 255, Birkhäuser Boston, Boston, MA, 2008.
T. Kobayashi, [*Propagation of multiplicity-freeness property for holomorphic vector bundles.*]{} in “Lie Groups: Structure, Actions, and Representations”, 113–140, Progr. Math. 306, Birkhäuser, New York, 2013.
C. Müller, K.-H. Neeb, H. Seppänen, [ *Borel–Weil Theory for Root Graded Banach–Lie Groups*]{}, International Mathematics Research Notices 5 (2010), 783–823
K.-H. Neeb, B. Oersted, [*Unitary highest weight representations in Hilbert spaces of holomorphic functions on infinite-dimensional domains.*]{} J. Funct. Anal. 156 (1998), no. 1, 263–300.
K.-H. Neeb, “Holomorphy and Convexity in Lie Theory,” de Gruyter Expositions in Mathematics. 28. Walter de Gruyter & Co., Berlin, 2000.
K.-H. Neeb, [*Infinite-dimensional groups and their representations.*]{} in “Lie Theory,” 213–328, Progr. Math., 228, Birkhäuser Boston, Boston, MA, 2004.
K.-H. Neeb, [*Semibounded representations of Hermitian Lie groups*]{} Trav. Math., 21, 2–109; Fac. Sci. Technol. Commun. Univ. Luxemb., Luxembourg, 2012.
K.-H. Neeb, [*Holomorphic realization of unitary representations of Banach–Lie groups.*]{} in “Lie Groups: Structure, Actions, and Representations,” 185–223, Progr. Math., 306, Birkhäuser, New York, 2013.
K.-H. Neeb, [*Unitary representations of unitary groups,*]{} in “Lie Theory Workshops,” Eds. G. Mason, I. Penkov, J. Wolf, Developments in Math. 37, Springer, 2014, 197–243
A. M. Sinclair, R. R. Smith, “Finite von Neumann algebras and masas.” London Mathematical Society Lecture Note Series 351, Cambridge University Press, Cambridge, 2008.
E. Størmer, [*Real structure in the hyperfinite factor,*]{} Duke Math. J. 47 (1980), no. 1, 14–153.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We study the effects of generalised surface disorder on the monomer-monomer model of heterogeneous catalysis, where disorder is implemented by allowing different adsorption rates for each lattice site. By mapping the system in the reaction-controlled limit onto a kinetic Ising model, we derive the rate equations for the one and two-spin correlation functions. There is good agreement between these equations and numerical simulations. We then study the inclusion of desorption of monomers from the substrate, first by both species and then by just one, and find exact time-dependent solutions for the one-spin correlation functions.'
author:
- 'D.A.Head and G.J.Rodgers'
date: |
Department of Physics, Brunel University, Uxbridge,\
Middlesex. UB8 3PH, UK
title: Kinetics of Catalysis with Surface Disorder
---
[PACS numbers: 05.70.Ln, 82.65.Jv, 82.20.Mj]{}
e-mail: [email protected], [email protected]
[**I. INTRODUCTION**]{}
Diffusionless surface-reaction models were first introduced by Ziff, Gulari and Barshad [@ziff2], who investigated a monomer-dimer reaction corresponding to the chemical reaction $2CO+O_{2}\rightarrow 2CO_{2}$ on a catalytic surface. A well-studied variant [@ziff3; @meakin; @evans] employs the simpler monomer-monomer reaction, described by
$$\begin{aligned}
A_{gas}+S & \stackrel{\rm k_{A}}{\longrightarrow} & A_{surface} \nonumber\\
B_{gas}+S & \stackrel{\rm k_{B}}{\longrightarrow} & B_{surface} \nonumber\\
A_{surface}+B_{surface} & \stackrel{\rm k_{R}}{\longrightarrow} & AB_{gas}+2S,\end{aligned}$$
[where S denotes an empty site. This process exhibits a [*kinetic phase*]{} when there are equal propensities of A and B species, in which the long-time kinetics become dominated by domain coarsening. Mean-field analysis [@ben-A], in which every site is taken to be connected to every other site in a ‘complete graph’, demonstrated that finite lattices will always saturate - that is, the lattice will either become full of A’s, or full of B’s, and the process will stop. Krapivsky [@krapivsky] recently solved the model exactly in the reaction-controlled limit $k_{R}\rightarrow\infty$ by mapping the system onto the standard Ising model. ]{}
Many enhancements to these models have been studied with a view to more closely modelling actual chemical processes, including nearest neighbour excluded adsorption [@zhuo] and surface diffusion [@evans; @vlachos]. However, only recently have the effects of surface disorder been touched upon by Frachebourg [@redner]. They chose to model a disordered surface by taking a lattice of two different types of site, one which favours adsorption by the A-species, and one which favours adsorption by the B’s. They showed numerically that such disorder allows for a reactive equilibrium in two dimensions.
In this paper, we extend the analytical method used in [@krapivsky] to a general form of surface disorder, based on [@redner] but allowing for a range of different types of site in the lattice. Furthermore, we also investigate separately the effects of desorption in the system. All the results presented are for the physically relevant case of two dimensions.
This paper is organised as follows. In section II we define the model and derive the general rate equations for the n-spin correlation functions. In section III, these equations are applied to a model similar to that in [@redner] and their solutions are compared to numerical simulations. In sections IV and V we include the effects of desorption, first by both species and then by just one, and derive exact solutions. The conclusions are summarised in section VI.
[**II. RATE EQUATIONS**]{}
We consider the surface reaction $A+B\rightarrow2S$ on a periodic $L\times L$ square lattice, ignoring the effects of diffusion and desorption. For simplicity, we take the reaction-controlled limit, where the adsorption of A and B species is taken to be infinitely fast so that the substrate is always full. The algorithm employed here is to select a nearest-neighbour (NN) pair at random, check for an AB-reaction, and, if so, remove the particles and immediately refill both sites.
With the usual homogeneous model, the probability of filling a site with and A or B is independant of the site chosen - in this model, however, that probability is allowed to vary. Specifically, we introduce the [*site inhomogeneity*]{} matrix $P_{ij}$, $0\leq P_{ij}\leq 1\; \forall i,j$, such that the probability of filling the site $(i,j)$ with an A is given by $P_{ij}$ (or, equivalently, a probability $1-P_{ij}$ of filling the site with a B).
Since in the reaction-controlled limit each site $(i,j)$ has only two possible states, we can map this model onto an Ising model with mixed Glauber-Kawasaki dynamics [@krapivsky], identifying A’s with $S_{ij}=+1$ and B’s with $S_{ij}= -1$. The master equation for $P(S,t)$, the probability distribution for the system to be in the state $S=\{S_{ij}\}$ at time t, is
$$\begin{aligned}
\frac{d}{dt}P(S,t)&=&\sum_{i,j}[\: U_{ij}(F_{ij}S)P(F_{ij}S,t)-U_{ij}(S)P(S,t)
\: ]\nonumber \\
&+&\sum_{i,j}[\: V_{ij}(F_{ij}F_{i+1j}S)P(F_{ij}F_{i+1j}S,t)-V_{ij}(S)P(S,t)
\: ]\nonumber \\
&+&\sum_{i,j}[\: W_{ij}(F_{ij}F_{ij+1}S)P(F_{ij}F_{ij+1}S,t)-W_{ij}(S)P(S,t)
\:].
\label{k:master}\end{aligned}$$
The flip operator $F_{ij}$ acts on the system state vector $S$ by flipping the sign of the $S_{ij}$ component, leaving the remaining components unchanged. $U_{ij}$ corresponds to Glauber spin-flip dynamics [@glauber], whereas $V_{ij}$ and $W_{ij}$ correspond to Kawasaki exchange dynamics. Equation (\[k:master\]) is identical the the homogeneous case, except that now the full expressions for $U_{ij}$,$V_{ij}$ and $W_{ij}$ are given by
$$\begin{aligned}
\label{uij}
4\tau_{1}U_{ij} & = &
(1-S_{ij}S_{i+1j})\{1-d_{ij}+S_{ij}(1-a^{+}_{ij})\} \nonumber\\
&&+\; (1-S_{ij}S_{i-1j})\{1-d_{i-1j}+S_{ij}(1-a^{+}_{i-1j})\}
\nonumber\\
&&+\;(1-S_{ij}S_{ij+1})\{1-e_{ij}+S_{ij}(1-b^{+}_{ij})\}\nonumber\\
&&+\;(1-S_{ij}S_{ij-1})\{1-e_{ij-1}+S_{ij}(1-b^{+}_{ij-1})\}, \\
\label{vij}
4\tau_{2}V_{ij}&=&(1-S_{ij}S_{i+1j})\left\{d_{ij}+a^{-}_{ij}
S_{ij}\right\}, \\
\label{wij}
4\tau_{2}W_{ij}&=&(1-S_{ij}S_{ij+1})\left\{
e_{ij}+b^{-}_{ij}S_{ij}\right\},\end{aligned}$$
where the constant coefficients $a_{ij}^{\pm}$,$b_{ij}^{\pm}$, $d_{ij}$ and $e_{ij}$ are related to the inhomogeneity matrix $P_{ij}$,
$$\begin{aligned}
a^{\pm}_{ij} & = & P_{i+1j}\pm P_{ij}, \nonumber \\
b^{\pm}_{ij} & = & P_{ij+1}\pm P_{ij}, \nonumber \\
d_{ij} & = & P_{ij}+P_{i+1j}-2P_{ij}P_{i+1j}, \nonumber \\
e_{ij} & = & P_{ij}+P_{ij+1}-2P_{ij}P_{ij+1}.\end{aligned}$$
We proceed by deriving the rate equations for the one and two-spin correlation functions, where the general n-spin function is given by
$$\langle S_{i_{1}j_{1}}\ldots S_{i_{n}j_{n}}\rangle
=\sum_{S}S_{i_{1}j_{1}}\ldots S_{i_{n}j_{n}}P(S,t).
\label{spincor}$$
Using this and (\[k:master\]), some lengthy but straightforward calculations result in the following hierarchy of differential equations, using the renormalised time scale $\tau$ defined by $\tau^{-1} = \tau_{1}^{-1}+\tau_{2}^{-1}$, and setting $\tau_{1}=\tau_{2}$,
$$\begin{aligned}
\label{1pt}
4\tau \frac{d}{dt}\langle S_{ij}\rangle&=&\Delta_{ij}\langle S_{ij}\rangle
+(1-2P_{ij})\langle S_{ij}\{\Delta_{ij}S_{ij}\}\rangle, \\
4\tau \frac{d}{dt}\langle S_{ij}S_{kl}\rangle& = & (\Delta_{ij}+\Delta_{kl})
\langle S_{ij}S_{kl}\rangle \nonumber\\
&&+(1-2P_{ij})\langle S_{ij}S_{kl}\{\Delta_{ij}S_{ij}\}\rangle\nonumber\\
&&+(1-2P_{kl})\langle S_{ij}S_{kl}\{\Delta_{kl}S_{kl}\}\rangle .\end{aligned}$$
…for $|i-k|+|j-l|>1$
Here, $\Delta_{ij}\langle S_{ij}
\rangle =-4\langle S_{ij}\rangle + \langle S_{i+1j}\rangle +
\langle S_{i-1j}\rangle +\langle S_{ij+1}\rangle +\langle S_{ij-1}
\rangle $ is the discrete Laplacian. For $|i-k|+|j-l|=1$, i.e. for nearest neighbour 2-point correlations, the rate equation has a more complex form. For example,
$$\begin{aligned}
4\tau \frac{d}{dt}\langle S_{ij}S_{i+1j}\rangle&=&
(2d_{ij}-8)\langle S_{ij}S_{i+1j}\rangle \nonumber\\
&&+\langle S_{i-1j}S_{i+1j}\rangle+\langle S_{ij}S_{i+2j}\rangle
+\langle S_{ij}S_{i+1j+1}\rangle \nonumber\\
&&+\langle S_{ij}S_{i+1j-1}\rangle
+\langle S_{ij-1}S_{i+1j}\rangle+\langle S_{ij+1}S_{i+1j}\rangle \nonumber\\
&&+(1-2P_{ij})\left\{\langle S_{i-1j}S_{ij}S_{i+1j}\rangle
-\frac{3}{2}\langle S_{i+1j}\rangle\right\} \nonumber\\
&&+(1-2P_{i+1j})\left\{\langle S_{ij}S_{i+1j}S_{i+2j}\rangle
- \frac{3}{2}\langle S_{ij}\rangle\right\} \nonumber\\
&&+2(1-2d_{ij}).
\label{2pt:nn}\end{aligned}$$
In the homogeneous limit $P_{ij}\rightarrow\frac{1}{2}$, the results in [@krapivsky] are recovered.
[**III. TWO-SITE DISORDER**]{}
We now turn to the case where $P_{ij}$ can take just two different values, $p$ or $q=1-p$, with an equal number of $p$-sites and $q$-sites. This corresponds to the model given in [@redner] with equal fluxes of A and B species, $\epsilon =|p-\frac{1}{2}|$ and $c_{-}=c_{+}=\frac{1}{2}$, using the notation given there.
Since $p+q=1$, the global dynamics of the system must be unchanged under the transformation $(p,q)\rightarrow(1-p,1-q)=(q,p)$. This symmetry means that the system cannot favour one state over the other, and so the average of $\langle S_{ij}
\rangle$ taken over the entire $L\times L$ lattice, $\frac{1}{L^{2}}\sum_{i,j}\langle S_{ij}\rangle$, will always tend to zero in the $L\rightarrow\infty$ limit. An important consequence of this is that if a finite system always saturates, then it does so with equal probability of saturating either to every site being +1, or every site being -1, and so $\langle S_{ij}
\rangle|_{t=\infty}=0\;\forall i,j$, regardless of whatever $P_{ij}$ may be. If a reactive steady-state occurs - that is, if the average saturation time diverges at least as fast as $e^{L^{2}}$ [@ben-A] - then it should be expected that $\langle S_{ij}\rangle$ may be non-zero for $t\rightarrow\infty$ (if $p\neq\frac{1}{2}$). It is the purpose of this section to apply the rate equations derived in section II to predict the equilibrium value of $\langle S_{ij}\rangle$ on $p$-sites in any such non-trivial steady state.
Although the concentrations of $p$-sites and $q$-sites are equal, different arrangements of the sites can dramatically alter the long-time dynamics of the system. For instance, choosing to split the lattice into two alternating $c(2\times 2)$ sublattices, with one sublattice full of $p$-sites and the other full of $q$-sites, results in a system with no non-trivial steady states for $p\neq 0$ or $1$. Since saturation always occurs, $\langle S_{ij}\rangle_{t=\infty}=0$ on either type of site.
A more informative model can be constructed by randomly arranging the $p$ and $q$ sites. This allows for regions of $p$-sites, which will all tend to be fixed into the same state, and regions of $q$-sites, which will all tend to be fixed into the other state, to [*‘pin’*]{} the dynamics into a reactive equilibrium. Although exact analysis of this model is obviously impossible, a useful approximation can be made by assuming that every site is surrounded by exactly 2 $p$-sites and 2 $q$-sites. It is then possible to write down (\[1pt\]) and (\[2pt:nn\]) for the two sorts of site, $\langle S_{ij}\rangle_{p}$ and $\langle S_{ij}
\rangle_{q}$, and the various two-point functions.
To obtain a closed set of equations, further approximations must be made to reduce the three-point functions in (\[2pt:nn\]) to one and two point functions. The obvious choice is
$$\langle S_{ij}S_{kl}S_{mn}\rangle\approx \langle S_{ij}S_{kl}\rangle
\langle S_{kl}S_{mn}\rangle,$$
[but this has the unwanted side-effect that $\langle S_{ij}\rangle_{p}
+\langle S_{ij}\rangle_{q}$ $\neq 0$, something which cannot be true since $p+q=1$. To restore the required symmetry we must also include an alternative three-point approximation,]{}
$$\langle S_{ij}S_{kl}S_{mn}\rangle\approx \langle S_{ij}\rangle
\langle S_{kl}S_{mn}\rangle.$$
For greater clarity, we denote the one-spin correlation function$\langle S_{ij}\rangle_{p}=-\langle S_{ij}\rangle_{q}$ by $y_{p}$, the two-spin correlation function between two NN $p$-sites (or, equivalently, two NN $q$-sites) by $z_{pp}$, and use $z_{pq}$ for the two-point function between nearest neighbour $p$ and $q$ sites. Setting $\tau =1$, we can now obtain a closed set of equations,
$$\begin{aligned}
\label{dyp}
2\frac{d}{dt}y_{p}&=&-2y_{p}+(1-2p)(z_{pq}+z_{pp}-2), \\
4\frac{d}{dt}z_{pp}&=&(4pq-8)z_{pp}-3(1-2p)y_{p}+
\nonumber\\
&&\{(1-2p)y_{p}+3z_{pp}\}(z_{pp}+z_{pq})+(2-4pq),\\
4\frac{d}{dt}z_{pq}&=&4pq-(4pq+6)z_{pq}+3(1-2p)y_{p}+3z_{pq}(z_{pp}
+z_{pq}).
\label{dzpq}\end{aligned}$$
The most constructive way to test the validity of this analysis is to compare the value of $y_{p}$ at equilibrium, as predicted by iterating eqns.(\[dyp\]–\[dzpq\]), to numerical simulations. In these simulations, the sites are initially randomly filled with $+1$’s or $-1$’s, so the corresponding initial conditions for the iteration procedure are $y_{p}|_{t=0}=z_{pp}|_{t=0}=z_{pq}|_{t=0}=0$. The results are compared in fig.1., where the simulation results compare favourably with the approximate analysis, the agreement improving for larger values of $p$.
Note that even when $p=1$, $y_{p}$ does [*not*]{} tend to +1, either in the theory or in the numerical work. This is because it is possible to have a jammed state where, for instance, a $q$-site surrounded by 4 $p$-sites may initially start at +1 but be unable to change, since if all 4 NN $p$-sites get fixed into a +1 state before they have reacted with the central $q$-site, then the $q$-site will never be able to react and so it will stay as +1 for all time, despite the fact that it has $P_{ij}=0$.
[**IV. INHOMOGENEOUS DESORPTION**]{}
We now turn to an enhanced model studied by Fichthorn, Gulari and Ziff [@ziff], which introduces noise into the system in the form of the desorption of A and B species from the substrate. They demonstrated numerically, later confirmed by mean-field analysis [@considine], that even a small desorption rate induces steady-state reactivity onto finite lattices. In our version of the model, sites vacated by desorption are refilled by an A or a B as defined by the inhomogeneity matrix, which we now call $Q_{ij}$. $Q_{ij}$ differs from $P_{ij}$ in that now it [*only*]{} applies to sites refilled after desorption - sites vacated after an reaction have an equal chance of being refilled either by an A or by a B. Thus, the reaction kinetics alone are the same as the usual homogeneous model, and the $U_{ij}$,$V_{ij}$ and $W_{ij}$ operators without the desorption take their simpler form found by setting $P_{ij}=\frac{1}{2}$ in (\[uij\]–\[wij\]). Explicitly,
$$\begin{aligned}
\label{homo-start}
8\tau_{1}U_{ij}(S)&=&4-S_{ij}(S_{i+1j}+S_{i-1j}+S_{ij+1}+S_{ij-1}),\\
8\tau_{2}V_{ij}(S)&=&1-S_{ij}S_{i+1j},\\
8\tau_{2}W_{ij}(S)&=&1-S_{ij}S_{ij+1}.
\label{homo-end}\end{aligned}$$
To include inhomogeneous desorption within this formulation, we replace $U_{ij}$ with $U^{d}_{ij}$,
$$U^{d}_{ij}=U_{ij}+\frac{1}{2\tau_{3}}\{1+S_{ij}(1-2Q_{ij})\},
\label{uij_d}$$
[where, as in [@krapivsky], we introduce a renormalised time scale $\tau$ and the spin-flip parameter $\gamma$, defined by ]{}
$$\begin{aligned}
\frac{1}{\tau}&=&\frac{1}{\tau_{1}}+\frac{1}{\tau_{2}}+\frac{1}{\tau_{3}}, \\
\gamma&=&1-\tau/\tau_{3}.\end{aligned}$$
The one point spin-correlation rate equation can now be recalculated using (\[k:master\]) and (\[homo-start\]–\[uij\_d\]),
$$\label{desorb}
4\tau\frac{d}{dt}\langle S_{ij}\rangle = \gamma\Delta_{ij}
\langle S_{ij}\rangle-4(1-\gamma)\{\langle S_{ij}\rangle
+(1-2Q_{ij})\}.$$
This can be solved by using a generating function, $G(X,Y,t)$, defined in terms of the time-dependant one-spin correlation function $\langle S_{ij}\rangle$,
$$G(X,Y,t)=\sum_{i=-\infty}^{\infty}\sum_{j=-\infty}^{\infty}
X^{i}Y^{j}\langle S_{ij}\rangle.
\label{gen}$$
Combining this with (\[desorb\]) gives rise to a differential equation for $G$,
$$\frac{\partial G}{\partial t} = \frac{G}{\tau}\left\{
\frac{\gamma}{4}\left(X+\frac{1}{X}+Y+\frac{1}{Y}\right) -1\right\}
-\frac{1}{\tau_{3}}\sum_{i,j=-\infty}^{\infty}X^{i}Y^{j}(1-2Q_{ij}).
\label{dgdt}$$
Noting that, except for G(X,Y,t), the right hand side of (\[dgdt\]) is independent of time, it is not difficult to derive an explicitly time-dependent expression for $\langle S_{ij}\rangle$ in terms of its initial state, $\sigma_{ij}=\langle S_{ij}\rangle|_{t=0}$,
$$\begin{aligned}
\langle S_{ij}\rangle &=& e^{-t/\tau}\sum_{k,l=-\infty}^{\infty}
\sigma_{kl}I_{i-k}\left(\frac{\gamma t}{2\tau}\right)
I_{j-l}\left(\frac{\gamma t}{2\tau}\right)
\nonumber\\ &&
-\frac{1}{\tau_{3}}\sum_{k,l=-\infty}^{\infty}(1-2Q_{kl})
\int_{0}^{t}e^{-t'/\tau}
I_{i-k}\left(\frac{\gamma t'}{2\tau}\right)
I_{j-l}\left(\frac{\gamma t'}{2\tau}\right) dt',
\label{desij}\end{aligned}$$
[where $I_{i}(t)$ is the $i^{th}$ order modified Bessel function. In the special case $\Delta_{ij}Q_{ij}=0$ it is possible to to rewrite the second term on the right hand side of (\[desij\]) as]{}
$$-\frac{1}{4\tau_{3}}\sum_{k,l=-\infty}^{\infty}(1-2Q_{kl})\left\{
f_{i-k+1\:j-l}+f_{i-k-1\:j-l}+f_{i-k\:j-l+1}+f_{i-k\:j-l-1} \right\},
\label{delf}$$
[where for clarity we have introduced]{}
$$f_{ij}(t)=\int_{0}^{t}e^{-t'/\tau}
I_{i}\left(\frac{\gamma t'}{2\tau}\right)
I_{j}\left(\frac{\gamma t'}{2\tau}\right)
dt',$$
[which obeys the identity]{}
$$\begin{aligned}
f_{i+1j}+f_{i-1j}+f_{ij+1}+f_{ij-1}&=&\frac{4}{\gamma}f_{ij}
-\frac{4\tau}{\gamma}\delta_{i0}\delta_{j0}
\nonumber\\
&&+\frac{4\tau}{\gamma}e^{-t/\tau}
I_{i}\left(\frac{\gamma t}{2\tau}\right)
I_{j}\left(\frac{\gamma t}{2\tau}\right),
\label{feqn}\end{aligned}$$
[with $\delta_{ij}$ the usual Krönecker Delta. Substituting (\[feqn\]) into (\[desij\]) and (\[delf\]) results in an exact expression, ]{}
$$\langle S_{ij}\rangle=2Q_{ij}-1\: +e^{-t/\tau}\sum_{k,l=-\infty}^{\infty}
(1-2Q_{kl}+\sigma_{kl})
I_{i-k}\left(\frac{\gamma t}{2\tau}\right)
I_{j-l}\left(\frac{\gamma t}{2\tau}\right).$$
So when $\Delta_{ij}Q_{ij}=0$, $\langle S_{ij}\rangle\rightarrow 2Q_{ij}-1$ exponentially as $t\rightarrow\infty$, again in agreement with the homogeneous result of [@ziff]. With desorption, jamming is no longer possible and so now $\langle S_{ij}\rangle\rightarrow 1$ when $Q_{ij}=1$. Although this final solution is exact, it is hard to see what physical applications a mixed homo/inhomogeneous model such as this one may have.
[**V. INHOMOGENEOUS ONE-SPECIES DESORPTION**]{}
Whilst investigating the monomer-dimer model, Ziff, Gulari and Barshad [@ziff2] briefly discussed the additional feature of allowing just the monomers to desorb. Physically, this corresponds to the reaction where only the $CO$ can desorb from the substrate, which is a good approximation for this reaction at the usual operating temperatures.
To apply a similar principle to our monomer-monomer model, we extend the analysis in section IV to allow for the desorption of A-species only, with the inhomogeneity matrix $Q_{ij}$ only applying to sites vacated after desorption. Thus, the flip-exchange operators are unchanged from (\[homo-start\]–\[homo-end\]), but now we replace $U_{ij}$ with
$$U^{d}_{ij}=U_{ij}+\frac{1-Q_{ij}}{2\tau_{3}}\left(1+S_{ij}\right).$$
Furthermore, $Q_{ij}$ is also taken to be a constant matrix, $Q_{ij}=q\;\forall i,j$. The rate equation for the one-spin correlation function (\[desorb\]) is now
$$\label{A-desorb}
4\tau\frac{d}{dt}\langle S_{ij}\rangle = \Delta_{ij}
\langle S_{ij}\rangle-\gamma (1-q)(1+\langle S_{ij}\rangle).$$
The definitions of $\tau$ and $\gamma$ have now altered from the previous case,
$$\begin{aligned}
\frac{1}{\tau}&=&\frac{1}{\tau_{1}}+\frac{1}{\tau_{2}}, \\
\gamma &=& \frac{4\tau}{\tau_{3}}.\end{aligned}$$
Applying the same generating function (\[gen\]) results in a new partial differential equation for $G(X,Y,t)$ acting on an $L\times L$ lattice,
$$\frac{\partial G}{\partial t}=\frac{G}{\tau}\left\{
\frac{1}{4}\left(X+\frac{1}{X}+Y+\frac{1}{Y}\right)-1\right\}
-\frac{1-q}{\tau_{3}}(G+L^{2}).$$
Continuing as before, an explicit time-dependent expression for $\langle S_{ij}\rangle$ is reached,
$$\begin{aligned}
\langle S_{ij}\rangle&=&e^{-t\left(\frac{1}{\tau}+\frac{1-q}{\tau_{3}}\right)}
\sum_{k,l=-L/2}^{L/2}\sigma_{kl}
I_{i-k}\left(\frac{t}{2\tau}\right)
I_{j-l}\left(\frac{t}{2\tau}\right)
\nonumber\\ &&
-\frac{(1-q)L^{2}}{\tau_{3}}\int_{0}^{t}
e^{-t'\left(\frac{1}{\tau}+\frac{1-q}{\tau_{3}}\right)}
I_{i}\left(\frac{t'}{2\tau}\right)
I_{j}\left(\frac{t'}{2\tau}\right) dt'.\end{aligned}$$
[**VI. CONCLUSIONS AND DISCUSSION**]{}
We have introduced a methodology for dealing with the effects of generalised surface disorder on the monomer-monomer reaction process , by mapping the system in the reaction-controlled limit onto an Ising Model. The two-dimensional rate equations were derived, including the very concise one-spin correlation equation (\[1pt\]), and used to study the special case of two-site disorder. Here, it was found that the global system dynamics are sensitive to the choice of layout of the two different types of site. Catalysts consisting of two different molecules arranged in a regular manner, such as on two alternating $c(2\times2)$ sublattices, allow for no reactive equilibrium and will always saturate on finite lattices. Choosing to randomly arrange the sites, however, allowing compact clusters of the same site, was shown to produce a reactive steady-state. Analysis based on the rate equations was used to predict the concentration of A’s and B’s on the different types of site, showing reasonable agreement between theory and simulation despite the rather crude approximations involved in the analysis.
The model was then extended to include desorption from the substrate, either by one or both species, and was solved exactly in both cases.
Extending this work to dimensions other than $d=2$ is straightforward once the mapping onto the Ising model has been achieved. Indeed, the rate equations for $d=1$ can been immediately seen from those given here (\[1pt\]–\[2pt:nn\]). We have focused on $d=2$ since the most useful physical application is of surface catalysis.
It should be noted that the definition of inhomogeneity we chose to employ here is only one of many ways of modelling surface disorder. For instance, requiring that each site be ‘hit’ a different number of times before adsorbing a particle, or assigning a quenched random ‘energy’ to each site and always adsorbing the particles onto the vacant site with the lowest energy, are just two alternative possibilities. We intend to study some of these in future work.
[99]{}
R.Ziff, E.Gulari and Y.Barshad, Phys.Rev.Lett [**56**]{} 2553(1986).
R.Ziff and K.Fichthorn, Phys.Rev.B [**34**]{} 2038(1986).
P.Meakin and D.Scalapino, J.Chem.Phys. [**87**]{} 731(1987).
J.Evans and T.Ray, Phys.Rev.E [**47**]{} 1018(1993).
D.ben-Avraham, S.Redner, D.B.Considine and P.Meakin, J.Phys.A [**23**]{} L613(1990).
P.L.Krapivsky, Phys.Rev.A [**45**]{} 1067(1992).
J.Zhuo, S.Redner and H.Park, J.Phys A [**26**]{} 4197(1993).
D.Vlachos, L.Schmidt and R.Aris, J.Chem.Phys [**93**]{} 8306(1990).
L.Frachebourg, P.L.Krapivsky and S.Redner, Phys.Rev.Lett [**75**]{} 2891(1995).
G.Forgacs, D.Mukamel and R.A.Pelcovits, Phys.Rev.B [**30**]{} 205(1984).
K.Fichthorn, E.Gulari and R.Ziff, Phys.Rev.Lett [**63**]{} 1527(1989)
D.Considine, S.Redner and H.Takayasu, Phys.Rev.Lett [**63**]{} 2857(1989).
Figure 1. Plot of $p$ vs. $y_{p}|_{t=\infty}$. The line gives the values predicted by the rate equations. Numerical simulation results are plotted as crosses. The simulations were performed on a $200\times200$ lattice, and averaged over 100 runs.
=
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'In the present paper we investigate distributional properties of sparse sequences modulo almost all prime numbers. We obtain new results for a wide class of sparse sequences which in particular find applications on additive problems and the discrete Littlewood problem related to lower bound estimates of the $L_1$-norm of trigonometric sums.'
author:
- |
[Victor C. García]{}\
\
\
\
title: On the distribution of sparse sequences in prime fields and applications
---
#### 2000 Mathematics Subject Classification: {#mathematics-subject-classification .unnumbered}
11B39, 11B50, 11L07.
Introduction
============
Throughout the paper $\{x_n\}$ is an increasing sequence of positive integers. The study of distributional properties of the sequence $$x_n \pmod p; \quad n=1,2,\ldots \,,$$ and additive problems connected with such sequences are classical questions in number theory with a variety of results in the literature. When $\{x_n\}$ grows rapidly the problem becomes harder for individual moduli, but it is possible to obtain strong results modulo most of the primes $p.$ We mention the work of Banks, Conflitti, Friedlander and Shparlinski [@BanksConflittiFriedShpar], where a series of results on distribution of Mersenne numbers $M_q = 2^q-1$ in residue classes have been obtained. This question has also been considered by Bourgain in [@Bourgain1]. General results on distribution of sequences of type $2^{x_n},$ (and generally of the form $\lambda^{x_n}$), modulo most of the primes have been obtained by Garaev and Shparlinski [@GaraevShparlinski], and by Garaev [@Garaev1]. For instance, Garaev [@Garaev1] has obtained a non-trivial bound for the exponential sum $$\max_{(a,p)=1}\left|\sum_{n\le T}e^{2 \pi i \frac{a}{p}{\lambda}^{x_n}}\right|,$$ for $\pi(N)(1+o(1))$ primes $p\le N$ and $T=N(\log N)^{2+\varepsilon},$ where $\{x_n\}$ is any strictly increasing sequence of positive integers satisfying $x_n\le n^{15/14 + o(1)}.$ Banks, Garaev, Luca and Shparlinski [@BanksGaraevLucaShparlinski] obtained uniform distributional properties of the sequences $$f_g(n)=\frac{g^{n-1}-1}{n}, \qquad h_g(n)=\frac{g^{n-1}-1}{P(n)},$$ where $g$ and $n$ are positive integers, $n$ is composite and $P(n)$ is the largest prime factor of $n.$
Now consider a simpler sequence $$2^n \pmod p ; \qquad n=1, 2,\ldots \, .$$ From a result of Erdös and Murty [@ErdosMurty] it is well-known that $2$ has the multiplicative order $t_p\ge N^{1/2 + o(1)}$ for $\pi(N)(1+o(1))$ primes $p\le N$. Combining this with a result of Glibichuk [@Glibichuk] it follows that for almost all primes $p$ every residue class modulo $p$ can be represented in the form $$2^{n_1} + \ldots +2^{n_8} \pmod p,$$ for certain positive integers $n_1,\ldots, n_8.$ García, Luca and Mejía [@GarciaLucaMejia] have applied similar arguments to obtain analogous results for the sequence of Fibonacci numbers $$F_n \pmod p; \qquad n=1,2,\ldots \,,$$ where $$F_{n+2}=F_{n+1}+F_n, \qquad n\ge 1,$$ with $F_1=F_2=1.$ They proved that for almost all primes $p,$ every residue class modulo $p$ is a sum of 32 Fibonacci numbers.
In the present paper using a different approach we obtain new results on additive properties for general sparse sequences for almost all the prime moduli. In particular we prove that for $\pi(N)(1+o(1))$ primes $p \le N$ every residue class is a sum of 16 Fibonacci numbers $F_n,$ with $n \le N^{1/2 + o(1)},$ improving upon the mentioned result of García, Luca and Mejía. Moreover, we establish that for any $\varepsilon >0$ there is an integer $s\le 100/\varepsilon$ such that for $\pi(N)(1+o(1))$ primes, $p\le N,$ every residue class can be written as $$F_1 +\ldots +F_{s}\pmod p,$$ with $1\le n\le N^{\varepsilon}.$ We note that the value $s$ has the optimal order $s={{\mathcal O}}(1/\varepsilon).$
From the work of Karatsuba [@Karatsuba] it is known the connection between additive problems and the Littlewood problem on lower bound estimates for the $L_1$-norm of exponential sums. Namely, for any coefficents $\gamma_n,$ $|\gamma_n|=1,$ and any strictly increasing sequence of integers $\{f(n)\},$ Karatsuba established $$\label{ineq:KaratsubaL1}
\int \limits_0^1\left| \sum_{n=1}^{N}\gamma_n e^{2\pi i \alpha f(n)}\right| {{\,\mathrm d}}\alpha \ge
\left(N^3/J\right)^{1/2},$$ where $J$ denotes the number of solutions of the diophantine equation $$f(n)+f(m)=f(k)+f(\ell); \qquad 1 \le n,m,k,\ell \le N.$$ Solving the Littlewood conjecture, Konyagin [@Konyagin], and McGehee, Pigno and Smith [@McGePigSmith] proved that $$\label{eq:LittlewoodConj}
\int \limits_{0}^{1}\left| \sum_{n=1}^{N}e^{2\pi i \alpha f(n)}\right| {{\,\mathrm d}}\alpha \gg \log N,$$ where $f(n)$ is an integer valued function. This bound reflects the best possible lower bound in general settings, as it shown by the example $f(n)=n$. However, due to the connection with certain additive problems, for a very wide class of integer valued sequences $f(n),$ estimate has been improved, see, for example, Garaev [@Garaev], Karatsuba [@Karatsuba] and Konyagin [@Konyagin1].
Green and Konyagin [@GreenKony] established a variant of the Littlewood problem in prime fields ${{\mathbb F}}_p.$ One of their results claims that if ${{\mathcal A}}$ is a subset of ${{\mathbb F}}_p,$ with $|{{\mathcal A}}|=(p-1)/2,$ then $$\frac{1}{p}\sum_{x=0}^{p-1}\left|\sum_{a \in {{\mathcal A}}}e^{2 \pi i x \frac{x}{p}a}\right|
\gg (\log p)^{1/3 - \varepsilon}.$$
One can use a version of Karatsuba’s inequality to get a variety of result for specific sequences. For instance, we employ a recent result of Bourgain and Garaev [@BourgainGaraev] on additive energy of the set $g^x \pmod p; \, 1\le x\le N,$ and show that for any $N < p^{1/2}$ we have the bound $$\frac{1}{p}\sum_{x=0}^{p-1}\left|\sum_{n \le N}e^{2 \pi i \frac{x}{p}g^n}\right| \gg N^{1/48 +o(1)}.$$ For the sequence $\{F_n\}$ of Fibonacci numbers we shall prove the following result: given any positive real $\gamma < 1/3$ there are positive constants $c_1=c_1(\gamma), c_2=c_2(\gamma)$ such that for $\pi(N)(1+o(1))$ primes $p\le N$ the following estimate holds $$c_1 N^{\gamma/2}\le \frac{1}{p}\sum_{x=0}^{p-1}\left|\sum_{n\le N^{\gamma}}
e^{2 \pi i \frac{x}{p}F_n}\right|
\le c_2 N^{\gamma/2}.$$
[**Acknowledgement.**]{} The author is grateful to M. Z. Garaev for sharing his thoughts which has led to the present improvement of the result of [@GarciaLucaMejia].
Formulation of results
======================
Throughout the paper $N$ and $M$ always denote positive large parameters. Let ${{\mathcal X}}$ be any subset of $\{1,\ldots, 10^M\}.$ The first result of our present paper relies on ideas of arithmetic combinatorics.
\[thm:average\] Let ${{\mathcal J}}(N)$ be the number of solutions of $$\label{eq:sim}
x\equiv y \pmod p; \qquad x,y \in {{\mathcal X}}, \quad p\le N.$$ The following asymptotic formula holds $$\label{asymp:average}
{{\mathcal J}}(N)=\pi(N)|{{\mathcal X}}| + {{\mathcal O}}\left(\frac{|{{\mathcal X}}|^2M}{\log M}\right).$$
Using Theorem \[thm:average\] we can get the following result on the value set of any sequence modulo most of the primes $p$.
\[thm:main\] For $\pi(N)(1+{{\mathcal O}}(1/\Delta))$ prime numbers $p\le N,$ we have the following asymptotic formula $$\label{eq:Size}
\#\{x \pmod p \;:\;x \in {{\mathcal X}}\;\}= |{{\mathcal X}}| +
{{\mathcal O}}\left(\frac{|{{\mathcal X}}|}{1 + \frac{\pi(N)\log M}{M|X|\Delta}}\right),$$ where $\Delta=\Delta(N)$ is any function with $\Delta\to \infty.$
In particular we have the following corollary which can be applied for a large class of sparse sequences.
\[coro:asympSparseSet\] If $M |{{\mathcal X}}|\Delta^2 \le \pi(N)\log M ,$ then $$\label{eq:SizeSparse}
\#\{x \pmod p \;:\;x \in {{\mathcal X}}\;\}= |{{\mathcal X}}|\left( 1 +
{{\mathcal O}}\left({\Delta}^{-1}\right)\right ).$$
Additive properties of Fibonacci numbers
----------------------------------------
Theorem \[thm:main\] finds application on additive problems for well known very fast increasing sequences. For example the following theorems on additive properities of the Fibonacci sequence $\{F_n\}.$
\[thm:Waring\_0\] For primes $p\le N,$ every integer $\lambda$ can be written as $$F_{n_1}+\ldots +F_{n_{16}}\equiv \lambda \pmod p,$$ where $1 \le n_1, \ldots, n_{16} \le N^{1/2+o(1)}.$
This improves the result of García, Luca and Mejía [@GarciaLucaMejia] on the representation of any residue class $\lambda$ in the form $$F_{n_1}+\ldots +F_{n_{32}}\equiv \lambda \pmod p,$$ for certain integers $n_1, \ldots, n_{32},$ for almost all primes $p.$ Moreover, combining Theorem \[thm:main\] with exponential sum techniques we obtain a more general result.
\[thm:Waring\_1\] Let $0 < \varepsilon <1/2.$ There is an integer $s < 100/\varepsilon$ such that for $\pi(N)(1 + o(1))$ primes $p \le N,$ every integer $\lambda$ can be written as $$F_{n_1}+\ldots +F_{n_s}\equiv \lambda \pmod p,$$ where $n_i \le N^{\varepsilon},$ $i=1,\ldots, s.$
Observe that the number of terms on the sumatory has the expected order, apart from the value 100. Indeed we obtain $s=4([8/\varepsilon]-1).$ However, we do not consider a reduction to be essential in this paper.
Application to the discrete Littlewood
---------------------------------------
The following theorem presents an application of a result of Bourgain and Garaev [@BourgainGaraev Theorem 1.4].
\[thm:L1PrimitiveRoot\] Let $g$ be a primitive root modulo $p.$ If $N < p^{1/2}$ then $$\frac{1}{p}\sum_{x=0}^{p-1}\left|\sum_{n\le N}e^{2 \pi i \frac{x}{p}g^n}\right|\gg N^{1/48 + o(1)}.$$
Regarding the Fibonacci sequence, we prove the following theorem.
\[thm:L1Fibonacci\] Let $0<\gamma <1/3.$ There are two positive absolute constants $c_1=c_1(\gamma), c_2=c_2(\gamma)$ such that for $\pi(N)(1 + o(1))$ primes, $p \le N,$ we have $$c_1 N^{\gamma/2}\le \frac{1}{p}\sum_{x=0}^{p-1}\left|\sum_{n\le N^{\gamma}}
e^{2 \pi i \frac{x}{p}F_n}\right|
\le c_2 N^{\gamma/2}.$$
Notation and lemmas
===================
For given subsets ${{\mathcal A}}$ and ${{\mathcal B}}$ of the residue field ${{\mathbb F}}_p$ and any integer $k\ge 2,$ as usual, we denote $$\begin{aligned}
{{\mathcal A}}+{{\mathcal B}}&= \{a + b \;:\;a\in {{\mathcal A}},\; b \in {{\mathcal B}}\},\\
{{\mathcal A}}\cdot {{\mathcal B}}&= \{a b \;:\;a\in {{\mathcal A}},\; b \in {{\mathcal B}}\},\\
k {{\mathcal A}}&= \{a_1+\ldots+a_k\;:\; a_1,\ldots, a_k \in {{\mathcal A}}\}.
\end{aligned}$$ For any finite subset of integers ${{\mathcal X}}$ we denote $${{\mathcal X}}\pmod p = \{x \pmod p \;:\; x \in {{\mathcal X}}\}.$$ The next lemma is a result of Glibichuk [@Glibichuk].
\[lemma:Glibichuk\] Let ${{\mathcal A}}, {{\mathcal B}}$ be subsets of ${{\mathbb F}}_p$ such that $|{{\mathcal A}}||{{\mathcal B}}|> 2p.$ Then $$8 {{\mathcal A}}\cdot {{\mathcal B}}= {{\mathbb F}}_p. \qed$$
Given a fixed prime number $p,$ we denote by $t_p$ the [*multiplicative order*]{} of 2 modulo $p.$ That is $$t_p = \min \{\ell \;:\;2^{\ell}\equiv 1 \pmod p\}.$$ As we mentioned in the introduction, the result of Erdös–Murty [@ErdosMurty] establish that for $\pi(N)(1+o(1))$ primes, $p\le N,$ we have $t_p > N^{1/2}e^{(\log N)^{\rho_0}},$ with some sufficiently small $\rho_0 >0.$ We present an analogous result for the [*order of appearance,*]{} defined by $$z(k)=\min\{\ell \;:\; F_{\ell}\equiv 0 \pmod k\},$$ where $k$ is a fixed integer $k\ge 2$ and $F_n$ denotes the $n$th term of the sequence of Fibonacci numbers.
\[lemma:OrderApp\] For almost all primes $p \le N,$ we have $$z(p)\ge N^{1/2}e^{(\log N)^{\rho}},$$ with some appropriate $\rho >0.$
We require the following lemma which follows from exponential sums estimates, see for example the proof of [@GaraevKueh Theorem1.1] or [@Sarkozy].
\[lemma:ternary\] Let $X, Y$ and $Z$ be subsets of $\{0,1,\ldots, p-1\}.$ Denote by $T$ the number of solutions of the congruence $$\label{eq:trigsum}
x y + z_1 + z_2 \equiv \lambda \pmod p,$$ where $$x \in X, \quad y \in Y, \quad z_1,z_2 \in Z.$$ Then, the asymptotic formula $$T = \frac{|X||Y||Z|^2}{p} + \theta\sqrt{p|X||Y|}|Z|, \quad |\theta|\le 1,$$ holds uniformly over $\lambda.$ In particular Eq. has solution if $|X||Y||Z|^{2}> p^{3}.$
We shall use some results concernig the values of Fibonacci sequence. $$\begin{aligned}
F_{u+v} &=\frac{1}{2}(F_uL_{v} + L_u F_{v}), \label{eq:FibIdentity1}\\
F_{u-v}&=\frac{(-1)^{v}}{2}(F_uL_{v} - L_u F_{v}),\label{eq:FibIdentity2}
\end{aligned}$$ where $\{L_m\}$ is the Lucas sequence given by $$L_{m+2}=L_{m+1}+L_m, \quad L_1=1, \,L_2=3.$$
The following lemma is due to Bourgain and Garaev [@BourgainGaraev]
\[lemma:EnergyPrimRoots\] Let $g$ be a fixed primitive root modulo $p.$ Let $1 < M < p^{1/2}$ and denote by $T$ be the number of solutions of the congruence $$g^x +g^y \equiv g^z + g^w \pmod p; \qquad 1 \le x,y,z,w \le M.$$ Then $$T < M^{3 - 1/24 + o(1)}.$$
Proof of Theorems
=================
Proof of Theorem \[thm:average\]
--------------------------------
If $x=y$ then Eq. has $\pi(N)|{{\mathcal X}}|$ solutions. Therefore $$\label{eq:J}
{{\mathcal J}}(N)=\pi(N)|{{\mathcal X}}| + {{\mathcal J}}',$$ where ${{\mathcal J}}'$ denotes the number of solutions of subject to $x\neq y.$ Given $x,y$ in ${{\mathcal X}}$ with $x\neq y,$ the equation $$pk = x-y, \qquad p \le N,$$ has at most $\omega(|x-y|)$ solutions, where $\omega(n)$ denotes the number of prime divisors of $n.$ If , using the well-known estimate $\omega(n)\ll (\log n)/(\log \log n),$ we obtain that has at most ${{\mathcal O}}(|{{\mathcal X}}|^2M/\log M)$ solutions. Otherwise, if $0< |x-y|<4,$ then has no more than ${{\mathcal O}}(|{{\mathcal X}}|)$ solutions. Thus $${{\mathcal J}}' \ll |{{\mathcal X}}|^2\frac{M}{\log M}.$$ Inserting this upper bound for ${{\mathcal J}}'$ in , Theorem \[thm:average\] follows. $\qed$
Proof of Theorem \[thm:main\]
-----------------------------
Before the proof, we shall introduce the following lemma
\[lemma:asympJ\_p\] Let $J_p$ be the number of solutions of the congruence $$\label{eq:J_p}
x \equiv y \pmod p; \quad x,y \in {{\mathcal X}}.$$ For $\pi(n)=(1+{{\mathcal O}}(1/\Delta))$ primes $p\le N$ we have $$\label{asymp:J_p}
J_p = |{{\mathcal X}}|+{{\mathcal O}}\left(\frac{|{{\mathcal X}}|^2M}{\pi(N)\log M}\Delta\right).$$
Note that $J_p \ge |{{\mathcal X}}|,$ because the case $x=y$ satisfies . It is clear that $${{\mathcal J}}(N)= \sum_{p \le N} J_p.$$ Denote by ${{\mathcal P}}$ the set of prime numbers $p\le N$ such that $$J_p - |{{\mathcal X}}| > \frac{|{{\mathcal X}}|^2 M}{\pi(N)\log M}\Delta.$$ If $p$ runs through the set ${{\mathcal P}},$ recalling that $J_p - |{{\mathcal X}}|\ge 0,$ we get $$|{{\mathcal P}}| \frac{|{{\mathcal X}}|^2 M}{\pi(N)\log M}\Delta \le \sum_{p \in {{\mathcal P}}}(J_p -|{{\mathcal X}}|)\le
\sum_{p \le N}(J_p -|{{\mathcal X}}|) = {{\mathcal J}}(N) - \pi(N)|{{\mathcal X}}|.$$ Thus, applying Theorem \[thm:average\], we derive that $$|{{\mathcal P}}|\ll \frac{\pi(N)}{\Delta}.$$ Therefore, if ${{\mathcal Q}}$ denotes the number of primes $p\le N$ such that $$J_p - |{{\mathcal X}}| \le \frac{|{{\mathcal X}}|^2 M}{\pi(N)\log M}\Delta,$$ then $$|{{\mathcal Q}}|= \pi(N)-|{{\mathcal P}}|= \pi(N)(1+{{\mathcal O}}(\Delta^{-1})).$$
Theorem \[thm:main\] follows from the relation $$\#\{x \pmod p \;:\; x \in {{\mathcal X}}\} \ge \frac{|{{\mathcal X}}|^2}{J_p}. \qed$$
Proof of Theorem \[thm:Waring\_0\]
----------------------------------
Lemma \[lemma:OrderApp\] allow us to establish the order of the value set of the Fibonacci sequence for most primes $$\#\{F_n \pmod p \;:\; n \le \delta N^{1/2 }\} \asymp \delta N^{1/2 },$$ where $\delta=\delta(N)=e^{(\log N)^{\rho}}$ and $\rho>0 $ is the refered constant in Lemma \[lemma:OrderApp\]. In order to establish the last relation, it is sufficent to prove that for $${{\mathcal F}}=\{F_{2n} \;:\; \delta N^{1/2} /10< n \le \delta N^{1/2} /5\},$$ we have $$\label{eq:ValueSetFib}
|{{\mathcal F}}\pmod p| = |{{\mathcal F}}| = \frac{\delta N^{1/2}}{10} + {{\mathcal O}}(1).$$ Let $n,n'$ be positive integers such that $$\label{eq:I}
F_{2n}\equiv F_{2n'}\pmod p; \quad \delta N^{1/2} /10<n,n' \le \delta N^{1/2} /5.$$ Without loss of generality we can assume that $n\ge n'.$ Substituting $u= n +n'$ and $v=n-n'$ in and , we can obtain [ $$F_{2n} - F_{2 n'} =
\frac{1}{2}\left((1-(-1)^{n-n'})F_{n+n'}L_{n-n'}+ (1+(-1)^{n-n'})L_{n+n'}F_{n-n'}\right).$$ ]{} Suppose that $n- n' \equiv 0 \pmod 2,$ then from Eq. follows $$p | L_{n+n'}F_{n-n'}.$$ If $n \neq n',$ then $0 < n-n'< N^{1/2}\delta \le z(p),$ which implies $(p,F_{n-n'})=1.$ Thus $$p| L_{n+n'}, \quad\textrm{in particular }\,\; p|F_{n +n'}L_{n+n'},$$ where $F_{n +n'}L_{n+n'}=F_{2(n+n')}.$ Hence $p|F_{2(n +n')},$ with $2(n+ n') < z(p).$ This contradicts the choice of $z(p).$ Therefore in the case $n - n' \equiv 0 \pmod 2$ Eq. has only trivial solutions $n=n'.$ Similarly, it is possible to verify that has not solutions if $n - n' \equiv 1 \pmod 2.$
Now, consider the subset of Lucas sequence $${{\mathcal L}}= \{L_{2m} \;:\; 1\le m\le N^{1/2}/\sqrt{\delta}\}.$$ Taking in Theorem \[thm:main\]; $M=N^{1/2}/\sqrt{\delta}$ and $\Delta={\delta}^{1/4}$ we obtain $$\label{eq:ValueSetLucas}
| {{\mathcal L}}\pmod p| =\frac{N^{1/2}}{\sqrt{\delta}}(1+{{\mathcal O}}({\delta}^{-1/4})).$$ Observe that equalities and are valid respectively for most primes. Thus, for $\pi(N)(1 + o(1))$ primes $p \le N$ we have $$|{{\mathcal F}}\pmod p||{{\mathcal L}}\pmod p|\gg \sqrt{ \delta} N \ge 2p.$$ Applying Lemma \[lemma:Glibichuk\], we obtain that for almost all primes $p$ every integer $\lambda$ can be written as $$F_{2n_1}L_{2m_1} +\ldots+F_{2n_8}L_{2m_8}\equiv \lambda \pmod p,$$ where $$N^{1/2}\delta /10< n_i \le N^{1/2}\delta /5, \quad 1 \le m_i \le N^{1/2}/\sqrt{\delta}, \quad 1 \le i \le 8.$$ Using the identity $$F_uL_{v}= F_{u+v} + (-1)^{v}F_{u-v},$$ for every $1\le i\le 8$ we get $$F_{2n_i}L_{2m_i}= F_{2(n_i+m_i)} + F_{2(n_i-m_i)}.$$ Thus, Theorem \[thm:Waring\_0\] follows. $\qed$
Proof of Theorem \[thm:Waring\_1\]
----------------------------------
Let $k$ be the minimal integer such that ${1}/{(k+2)} < {\varepsilon}/{8}.$ Define the sets $$\begin{aligned}
X &=\{F_{2n_1-1}+ \ldots + F_{2n_k-1}\;:\;1\le n_1, \ldots, n_k \le N^{\frac{1}{k+2}}\}, \\
Y &= \{L_m \;:\;\frac{1}{2}N^{\frac{7}{k+2}} < m \le N^{\frac{7}{k+2}}\}, \\
Z &=\{F_{2\ell_1}+ \ldots + F_{2\ell_k}\;:\; 1 \le \ell_1, \ldots, \ell_k \le N^{\frac{1}{k+2}} \}.
\end{aligned}$$ Observe that $|Y| \gg N^{\frac{7}{k+2}}$ and exist a positive constant $c=c(k) < 1$ such that $$|X|, |Z| \ge cN^{\frac{k}{k+2}}.$$ In order to estimate the value set of $X \pmod p$ note that if $x\in X,$ then $x \le 10^{\log k N^{1/(k+2)}}.$ Thus, applying Corollary \[coro:asympSparseSet\] with $M=\log k N^{1/(k+2)},$ ${{\mathcal X}}=Z $ and $\Delta = (\log N)^A,$ (for any integer $A>0$), we have that for most of primes $p\le N$ $$|X \pmod p | =|X|(1 + o(1)).$$ Analogously, we can obtain $$|Y \pmod p|=|Y|(1 + o(1)), \qquad |Z \pmod p|=|Z|(1 + o(1)),$$ for almost all primes respectively. Therefore, there is a constant $c_1=c_1(k),$ $0 < c_1 <1,$ such that for $\pi(N)(1+o(1))$ primes $p\le N$ we have $$|X \pmod p||Y \pmod p||Z \pmod p|^2 \ge c_1 N^{3 + \frac{1}{k+2}}> p^{3 + \frac{1}{k+2}}.$$ Applying Lemma \[lemma:ternary\] it follows that for almost all primes every integer $\lambda$ can be represented as $$\label{eq:TernaryFibonacci}
\sum_{i=1}^{k}L_mF_{2n_i-1}+ \sum_{j=1}^{k}(F_{2\ell_j}+F_{2\ell'_j}) \equiv \lambda \pmod p,$$ where $$\frac{1}{2}N^{\frac{7}{k+2}} < m \le N^{\frac{7}{k+2}}, \quad
1\le n_i\le N^{\frac{1}{k+2}}, \quad
1 \le \ell_j,\ell'_j \le N^{\frac{1}{k+2}}, \quad (1 \le i,j \le k).$$ We recall the identity $$L_u F_v = F_{u+v}+(-1)^{v+1}F_{u-v}.$$ Thus, for every $1 \le i \le k$ in we get $$L_mF_{2n_i-1}=F_{m + 2n_i - 1}+ F_{m - 2n_i +1}.$$ taking $s=4k$ (that is, $s = 4([ 8/\varepsilon]-1)$), we conclude that for almost all primes every residue class $\lambda$ has a representation in the form $$F_{n_1}+\ldots +F_{n_s}\equiv \lambda \pmod p,$$ for some integers $$1\le n_1, \ldots, n_s \le N^{\varepsilon}. \qed$$
Proof of Theorem \[thm:L1PrimitiveRoot\]
----------------------------------------
Note that the congruence $$g^x \equiv g^y \pmod p; \qquad1 \le x,y \le N,$$ has exactly $N$ solutions. Therefore, $$N = \frac{1}{p}\sum_{x=0}^{p-1}\left|\sum_{n\le N}e^{2 \pi i \frac{x}{p}g^n}\right|^2 =
\frac{1}{p}\sum_{x=0}^{p-1}\left|\sum_{n\le N}e^{2 \pi i \frac{x}{p}g^n}\right|^{2/3}
\left|\sum_{n\le N}e^{2 \pi i \frac{x}{p}g^n}\right|^{4/3},$$ using Hölder’s inequality we obtain $$\begin{aligned}
N &\le &\frac{1}{p}\left(\sum_{x=0}^{p-1}\left|\sum_{n\le N}e^{2 \pi i \frac{x}{p}g^n}\right|\right)^{2/3}
\left(\sum_{x=0}^{p-1}\left|\sum_{n\le N}e^{2 \pi i \frac{x}{p}g^n}\right|^4\right)^{1/3}\\
& \le &{T^{1/3}}\left(\frac{1}{p}\sum_{x=0}^{p-1}\left|\sum_{n\le N}e^{2 \pi i \frac{x}{p}g^n}\right|\right)^{2/3},
\end{aligned}$$ where, $T$ denotes the number of solutions of the congruence $$g^x + g^y \equiv g^z + g^w \pmod p; \qquad 1\le x,y,z,w \le N.$$ Thus, from Lemma \[lemma:EnergyPrimRoots\] we know that $T < N^{3 - 1/24 + o(1)}.$ Therefore $$\frac{1}{p}\sum_{x=0}^{p-1}\left|\sum_{n\le N} e^{2 \pi i }\right| > N^{1/48 - o(1)}. \qed$$
Proof of Theorem \[thm:L1Fibonacci\]
------------------------------------
Observe that the congruence $$F_n\equiv F_{n'}\pmod p; \qquad 1 \le n,n' \le N^{\gamma},$$ has at least $N^{\gamma}$ solutions. Therefore $$N^{\gamma}\le \frac{1}{p}\sum_{x=0}^{p-1}
\left|\sum_{n\le N^{\gamma}}e^{2 \pi i \frac{x}{p}F_n}\right|^2.$$ From Hölder’s inequality, as in the proof of Theorem \[thm:L1PrimitiveRoot\], it is possible to obtain $$\label{ineq:L1Fibonacci}
N^{\gamma} \le T_p^{1/3}\left(\frac{1}{p}\sum_{x=0}^{p-1}
\left|\sum_{n\le N^{\gamma}}e^{2 \pi i \frac{x}{p}F_n}\right|\right)^{2/3},$$ where $T_p$ denotes the number of solutions of the congruence $$F_{n_1}+F_{n_2}\equiv F_{m_1}+F_{m_2}\pmod p; \quad 1 \le n_1,n_2,m_1,m_2 \le N^{\gamma}.$$ Let $${{\mathcal X}}= \{F_{n_1}+F_{n_2} \;:\; 1 \le n_1,n_2 \le N^{\gamma}\}.$$ Then $|{{\mathcal X}}|\asymp N^{2\gamma}.$ Applying Lemma \[lemma:asympJ\_p\] with $M=N^{\gamma}$ and $\Delta=N^{(1-3\gamma)/2}$ we get, for $\pi(N)(1+o(1))$ primes $p\le N,$ the estimation $$T_p \le N^{2\gamma}\left(1 + N^{-(1-3\gamma)/2}\right).$$ Combining this estimation with relation we conclude that there is a positive constant $c_1(\gamma)$ such that $$\frac{1}{p}\sum_{x=0}^{p-1}
\left|\sum_{n\le N^{\gamma}}e^{2 \pi i \frac{x}{p}F_n}\right| \ge c_1(\gamma)N^{\gamma/2}.$$ Finally, to obtain an upper bound of the same order, using the Cauchy-Schwartz inequality we have $$\label{ineq:UpperBoundFibonacci}
\left( \frac{1}{p}\sum_{x=0}^{p-1}
\left|\sum_{n\le N^{\gamma}}e^{2 \pi i \frac{x}{p}F_n}\right|\right)^2 \le
\frac{1}{p}\sum_{x=0}^{p-1}
\left|\sum_{n\le N^{\gamma}}e^{2 \pi i \frac{x}{p}F_n}\right|^2,$$ where the right term is, indeed, the number of solutions of the congruence $$F_n\equiv F_m \pmod p; \qquad 1 \le n,m \le N^{\gamma}.$$ Applying again Lemma \[lemma:asympJ\_p\] with $M=N^{\gamma}$ and $\Delta=N^{(1-2\gamma)/2},$ we obtain, for $\pi(N)(1+o(1))$ primes $p\le N,$ the estimation $$\frac{1}{p}\sum_{x=0}^{p-1}
\left|\sum_{n\le N^{\gamma}}e^{2 \pi i \frac{x}{p}F_n}\right|^2 \le N^{\gamma}\left(1 + N^{-(1-2\gamma)/2}\right)
\le c_2(\gamma) N^{\gamma},$$ for some positive constant $c_2(\gamma).$ Putting together with and taking square root we conclude the proof. $\qed$
[999]{}
W. D. Banks, A. Conflitti, J. B. Friedlander and I. E. Shparlinski, ‘Exponential sums over Mersenne numbers,’ [*Compos. Math.,*]{} [**140**]{} (1), 15–30 (2004).
W. D. Banks, M. Z. Garaev, F. Luca and I. E. Shparlinski, ‘Uniform distribution of fractional parts related to pseudprimes,’ [*Canad. J. Math.,*]{} [**61**]{} (3), 481–502 (2009).
J. Bourgain, ‘Estimates on exponential sums related to the Diffie–Hellman distributions,’ [*Geom. Funct. Anal.,*]{} [**15**]{} (1), 1–34 (2005).
J. Bourgain, M. Z. Garaev, ‘On a variant of sum-product estimates and explicit exponential sum bounds in prime fields,’ [*Math. Proc. Cambridge Philos. Soc.,*]{} [**146**]{} (1), 1–21 (2009).
P. Erdős and M. R. Murty, ‘On the order of $a\pmod p$’, in [*Number theory (Ottawa, ON, 1996)*]{}, 87–97, CRM Proc. Lecture Notes [**19**]{}, Amer. Math. Soc., Providence, RI, 1999.
M. Z. Garaev, ‘Upper bounds for the number of solutions of a diophantine equation,’ [*Trans. Amer. Math. Soc.,*]{} [**357**]{}, 2527–2534 (2005).
M. Z. Garaev, ‘The large sieve inequality for the exponential sequence $\lambda^{[O(n^{15/14+o(1)})]}$ modulo primes’ [*Canad. J. Math.,*]{} [**61**]{} (2), 336–350 (2009).
M. Z. Garaev and Ka–Lam Kueh, ‘Distribution of special sequences modulo a large prime’ [*Int. J. Math. Math. Sci.,*]{} [**50**]{}, 3189-3194 (2003).
M. Z. Garaev and I. E. Shparlinski, ‘The large sieve inequality with exponential functions and the distribution of Mersenne numbers modulo primes,’ [*Int. Math. Res. Notices,*]{} (39), 2391–2408 (2005).
V. C. Garcia, F. Luca and V. J. Mejia, ‘On sums of Fibonacci numbers modulo $p$,’ [*Bull. Aust. Math. Soc.*]{}, (to appear).
A. A. Glibichuk, ‘Combinatorial properties of sets of residues modulo a prime an the Erdős–Graham problem’, [*Mat. Zametki* ]{} [**79**]{}:3, 384–395 (2006); English transl., [*Math. Notes*]{} [**79**]{}:3–4, 356–365 (2006).
B. Green and S. V. Konyagin, ‘On the Littlewood problem modulo a prime,’ Canad. J. Math. [**61**]{} (1), 141–164 (2009).
A. A. Karatsuba, ‘An estimate of the $L_1$-norm of an expontential sum,’ Math. Notes, [**64**]{}, 401–404 (1998).
S. V. Konyagin, ‘On the problem of Littlewood,’ Izv. Acad. Nauk SSSR Ser. Mat. \[Math. USSR-Izv.\], [**45** ]{}(2), 243–265 (1981).
S. V. Konyagin, ‘An estimate of the $L_1$-norm of an exponential sum,’ The Theory of Approximations of Functions and Operators. Abstracts of Papers of the International Conference Dedicated to StechkinÕs 80th Anniversary \[in Russian\]. Ekaterinburg (2000), pp. 88-89.
O. C. McGehee, L. Pigno and B. Smith, ‘Hardy’s inequality and the $L_1$ norm of exponential sums,’ Ann. of Math. (2), [**113**]{}(3), 613–618 (1981).
A. Sárközy, ‘On sums and products of residues modulo $p$,’ [*Acta Arith.,*]{} [**118**]{} (4), 403–409 (2005).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We probe the rheology of weakly vibrated granular flows as function of flow rate, vibration strength and pressure by performing experiments in a vertically vibrated split-bottom shear cell. For slow flows, we establish the existence of a novel vibration dominated granular flow regime, where the driving stresses smoothly vanish as the driving rate is diminished. We distinguish three qualitatively different vibration dominated rheologies, most strikingly a regime where the shear stresses no longer are proportional to the pressure.'
author:
- 'Geert H. Wortel'
- 'Joshua A. Dijksman'
- Martin van Hecke
title: Rheology of Weakly Vibrated Granular Media
---
Granular media are collections of macroscopic, athermal grains which interact through dissipative, frictional contact forces. In the presence of gravity and in the absence of additional external forces, they jam in metastable configurations; however, external forcing can easily lead to yielding and flow [@1995_book_duran; @1996_revmodphys_jaeger; @2004_epje_gdrmidi; @2008_annurevfluid_forterre; @2010_prl_nichol; @2007_pre_sanchez; @2006_pre_rubin; @1989_prl_jaeger; @2009_jrheol_marchal; @2009_epl_janda]. The best known scenario that leads to granular flow is by exerting shear stresses that exceed the yield stress, as when tilting a quiescent layer of sand sufficiently far away from the horizontal [@1995_book_duran; @1996_revmodphys_jaeger]. To understand such dense granular flows, it is becoming increasingly clear that both stress and mechanical agitations play a crucial role. Indeed, a given stress can give rise to a wide range of flow rates depending on the magnitude of these agitations [@2011_prl_reddy; @2010_prl_nichol; @2013_elie_arxiv; @andreotti_arxiv]. Moreover, agitations make granular media lose their rigidity, although in the absence of shear stresses this does not need to cause flow [@2010_prl_nichol; @2011_prl_reddy; @2012_pre_nichol; @2013_elie_arxiv]. We note here that the idea that both the stress and the amount of agitations determine the flow rate lies at the basis of numerous models for slowly flowing disordered materials [@1998_pre_falk; @1997_prl_sollich; @2012_pre_krimer; @2008_nature_goyon].
![Sketch of the vibrated split-bottom setup in which the rotation of a disk of radius $r_s$ is used to probe the rheology of agitated granular media. The crucial experimental parameters are the filling height, $H$, vibration amplitude, $\Gamma$, the torque, $T$, and the rotation rate, $\Omega$.[]{data-label="fig:intro"}](Figure_1.eps){width="5cm"}
In a granular context, such agitations may be provided by external vibrations. In a classic experiment, the slope of a granular pile was found to relax under vertical vibrations [@1989_prl_jaeger], and similarly, horizontal vibrations have been used to induce flow on inclined planes [@2007_pre_sanchez; @2006_pre_rubin]. Piezo transducers inside the medium have been used to inject tiny rearrangements or force fluctuations, either locally [@2009_epl_janda; @2009_epje_caballero] or along a complete boundary [@2005_jphyscm_caballero]. Shear induced agitations similarly have induced microscopic rearrangements [@2012_epl_coulais].
Naturally, flow itself also induces mechanical agitations [@2002_prl_longhi]. For example, for shear banded flows [@schall], the observation of particle rearrangements and fluidization far away from the flowing region suggest that the effect of agitations can be carried far through the material [@2010_prl_nichol; @2008_jstatmech_crassous; @2011_prl_reddy; @2012_prl_amon]. Agitations thus form a crucial ingredient for non-local extensions of models for the rheology of granular flows [@2008_annurevfluid_forterre; @2008_pre_jagla; @2007_pre_torok; @2006_pre_depken], and may explain, among other things, the large extension of shear bands in split-bottom granular flows [@2012_prl_kamrin; @PNAS_kamrin].
Recently, we have explored how weak vibrations influence the rheology of dry granular media [@2011_prl_dijksman] by performing experiments in a vibrated split-bottom cell, as shown in Fig. \[fig:intro\]. We found that weak external vibrations suppress the yield stress of the material and strongly influence the rheology of slow granular flows.
In this paper, we reveal the intriguing rheology of weakly vibrated granular media in much more detail. We find that we can distinguish a variety of qualitatively different flow regimes. First, for large flow rates, inertial effects dominate, and the effect of vibrations is small. Second, for slower flow rates, we cross over to a regime similar to the well-known quasi static flows that have been studied at length in the absence of vibrations [@1980_powtech_nedderman; @2004_prl_fenistein; @2006_prl_fenistein; @2010_sm_dijksman; @2011_prl_dijksman]. Third, for slower flows, we enter a regime where the vibrations lead to completely new rheological behavior.
The focus of this paper is on these [*vibration dominated flows*]{}. By probing the equilibration times of the stresses and the variation of the steady state stresses with filling height, we find evidence for three qualitatively different regimes. For slow enough flows, vibration effects increasingly dominate the physics, leading to compaction of the material for weak vibrations, and to fluidization of the material for vibrational accelerations approaching gravity. Most strikingly, in the latter regime, we see a breakdown of the proportionality of shear stresses and pressure, a highly unusual phenomenon in granular flows.
We also study how the rheology of weakly vibrated granular media behaves in stress controlled flow experiments. We find that equilibration times can be dramatically longer than in rate controlled experiments, akin to what has been observed in several soft materials that exhibit a so-called viscosity bifurcation [@coussot_pre_2002; @bonn_epl_2009], and leading to Andrade creep like phenomena [@andrade_1914], providing an interesting analogy with thermal, disordered flows [@2004_jphyscondmat_petekidis; @ballauff_prl_2012].
The outline of this paper is as follows. In Section \[sec:setup\] we describe the details of our experimental setup and the measurement protocols used in the current and previous study [@2011_prl_dijksman]. In Section \[sec:pheno\] we describe the main phenomenology of a complete set of experiments probing $T(\Gamma,\Omega,H)$. In Section \[sec:unger\] we introduce the canonical perspectives on granular rheology, including a model for the stresses in split-bottom flows by Unger [*et al.*]{} [@2004_prl_unger]. In Section \[sec:Ggt0\] we use this model to extract effective friction coefficients from our data, as well as exploring the quality of the fit between the data and this model. In Section \[sec:fluid\] we provide strong evidence for the existence of a pressure independent flow regime though measurements of the flow structure. Additionally, we compare steady state results obtained at constant $\Omega$ driving with a constant stress driving mode in Section \[sec:constantT\].
Setup and Protocol {#sec:setup}
==================
In this section we briefly introduce the main parts of our experimental setup (for details, see Appendix \[app:setup\]), discuss our measurement protocols, and show how we ensure that we measure steady values for the rheology.
Setup
-----
We probe the rheology of weakly vibrated granular flows in a modified split-bottom cell, as shown in Fig. 1. The acrylic container has an inner radius of 7 cm. We mount a hollow smooth cylinder of 6 mm height and 4 cm radius on the bottom of the container. The rotating disk (radius $r_s$ of 4 cm and thickness 5 mm) that drives the granular flow is mounted just above the cylinder. The gap between the container and the disk is about 0.3 mm so no particles can get underneath the disk. To ensure a no-slip boundary condition, the top surface of the disk is made rough by gluing glass particles with diameter of 2 mm to it.
The container is filled with particles (black soda-lime glass beads, Sigmund Lindner 4504-007-L), a polydisperse mixture with a diameter between 1 and 1.3 mm, and a bulk density $\rho$ of 1.7$\times 10^3$ kg/m$^3$, up to a filling height $H$. To ensure good reproducibility, we use to total mass of the particles to control $H$.
All experiments are carried out under ambient temperature, pressure and relative humidity. We have verified that our experiments are insensitive to relative humidities ranging from 6 and 55%. After several months of use, the black coating of the particles visibly deteriorates, and the rheological behavior becomes more sensitive to humidity. We therefore renew our particles on a trimonthly basis, and found that our experiments reproduce well over the course of several years.
In the absence of vibrations, the phenomenology of the flow is determined by the dimensionless filling height $h\equiv H/r_s$ [@2010_sm_dijksman]. In our experiments, we stay in the low filling height regime ($h < 0.6$), where the shear bands are mainly vertical, and all grains above the disk co-rotate along with it (trumpet-flow). We have found that in all but one flow regimes, the flow profiles observed at the free surface are insensitive to the magnitude of vibrations. For the exception – which is for slow flow and strong vibrations – see Section \[sec:fluid\].
To drive the rotation of the bottom disk and shear the granular media we use a rheometer (Anton Paar DSR 301), which can be used both in stress control (imposing a torque $T$ and measuring the resulting rotation rate $\Omega$) or in rate control (impose $\Omega$, measure $T$). We shake the system with a sinusoidal oscillation $Asin(2\pi f t)$, with a fixed $f$ of 63 Hz, using an electromagnetic shaker (VTS systems VG100). The amount of vibrations is characterized by the dimensionless parameter $\Gamma$=$A(2\pi f)^2/g$, where g is the gravitational acceleration. We image the surface of the system using a mirror and a Foculus FO114B camera, allowing us to extract the surface flow by particle image velocimetry.
![(Color online) The torque as function of deflection angle $\theta$ for $\Omega$=10$^{-4}$ rps and $\Gamma=0.7$ but different waiting time $t_w$ between the preshear and the actual measurement. []{data-label="fig:waiting"}](Figure_2.eps){width="\columnwidth"}
Protocol
--------
Our experiments focus on the rheological curves which relate the driving torque $T$ and the driving rate $\Omega$. In section \[sec:constantT\] we describe some experiments performed at constant torque, but our main focus is on experiments where we fix the driving rate in the range from 1 to 10$^{-4}$ rps, and probe the torque. We perform these experiments for a range of vibration amplitudes $\Gamma = 0, 0.2, 0.5, 0.7, 0.83, 0.95$ and $1$, and moreover use seven different filling heights ($h = 0.19, 0.25, 0.31, 0.38, 0.44, 0.50$ and $0.56$). Varying $h$ allows us to probe the role of the confining pressure for the rheology.
Each experiments starts with switching on the vibrations, after which we allow the shaker feedback loop 30 s to settle to the required value of $\Gamma$. We proceed by applying an amount of pre-shear to the granular material, in order to obtain similar starting conditions for each experiment. Unless noted otherwise, the protocol consists of the following steps: *(i)* 2 s of 1 rps rotation clockwise; *(ii)* 4 s of 1 rps rotation counter-clockwise; *(iii)* 2 s of 1 rps rotation clockwise. *(iv)* 5 s without imposed stress or shear. *(v)* start of actual measurement. The rotation in the experiments is in the clockwise direction to minimize anisotropy effects [@wortel_aniso].
Vibrations lead to compaction of granular media, although this process is very slow for $\Gamma < 1$ [@2005_jphyscm_caballero], whereas flow typically leads to dilatation [@1885_philmag_reynolds]. Additionally, anisotropy in the fabric of the granular media needs a finite amount of strain to build up, but may be relaxed by vibrations [@wortel_aniso; @2013_elie_arxiv]. For both density and anisotropy, vibrations and flow are in competition, and as we are interested in steady state flow properties, we need to ask: what is the minimum time or strain necessary to get into a steady state flow regime?
We have probed the relaxation of our flows to a steady state by starting the flow from a denser or less densely packed state as follows: Before each experiment, we perform pre-shear as described above. After pre-shear, during stage *(iv)*, we stop the shear and vibrate the material for a waiting period $t_w$, during which the granular packing density increases by compaction, and then start the actual measurements. By varying the $t_w$, we thus manipulate the packing fraction at the start of the flow. By measuring the torque as function of time we capture the evolution of the torque to its steady state value. As we expect this equilibration to be slowest for small $\Omega$ we perform this test at the smallest $\Omega$ (10$^{-4}$ rps) that we explore in our experiments.
![(Color online) Flow curves for different averaging times $t_a$. Below $\Omega \approx 0.5\times10^{-3}$, the torque increases with the waiting time. The measurements are for $\Gamma=0.6$ and $h$=0.56.[]{data-label="fig:aging"}](Figure_3.eps){width="\columnwidth"}
The results of this test are shown in Fig. \[fig:waiting\], where we plot $T$ as function of the total angle of rotation of the bottom disk $\theta$. This figure shows that for small $t_w$, $T$ grows monotonically before reaching steady state, whereas for large $t_w$, the torque peaks at values larger than the steady state value. This is consistent with a simple picture where the longer the waiting time, the denser the grains are packed at the start of the experiment, and the larger the torque needed to cause flow.
For all waiting times, the torque reaches its steady state value for $\theta < 0.1$ rad, corresponding to a measurement time of 150 s at 10$^{-4}$ rps. We claim that this represents the longest equilibration time necessary to reach a steady state flow situation, as all our experiments are carried out for $\Omega \geq 10^{-4}$ rps. Moreover, in many experiments our data is acquired in a so called strain rate sweep, where the rotation rate is varied by a small amount so that equilibration will be faster. In all cases, an equilibration strain or time of $\theta > 0.1$ rad, or 150 s, will be sufficient to obtain steady state flow curves. We choose 180 s for all the experiments described in this paper.
To independently verify that equilibration times of 180 s are long enough, we perform a strain rate sweep at fixed $\Gamma=0.6$. We sweep the flow rate from fast to slow rates, and then compare flow curves obtained for different times $t_a$ per step, as shown in Fig. \[fig:aging\]. While for small values of $\Omega$ and $t_a$, the torque shows a variation with $t_a$, for all the flow rates probed here we note that the data for $t_a=180$ s and $1800$ s are indistinguishable, showing that for $t_a \geq 180$ s the obtained values of $T$ are steady state values. As a final additional test, we have also inspected $T(t)$ to confirm we reach steady state (Appendix \[proto\]).
Phenomenology {#sec:pheno}
=============
{width="17cm"}
![(Color online) The boundaries between the regions where the flow curves have a positive and a negative slope for $h$= 0.19 ($\diamond$), $h$= 0.38 ($\ast$) and $h$= 0.56 (+). For $\Gamma=0$ and $\Omega<0.3$ rps, the flow curve always has a negative slope. The region extends to $\Gamma>0$, and it extends to higher $\Gamma$ for lower $h$.[]{data-label="fig:slope"}](Figure_5.eps){width="\columnwidth"}
We now turn our attention to the rheological curves $T(\Omega)$. As shown in the $T(\Omega)$ curves in Fig. \[fig:overview\_g\], the flows in our system exhibit a rich rheology. There are two simple trends we see illustrated in these curves: increasing the filling height always increases $T$, whereas increasing the vibration strength always decreases $T$. The role of the flow rate is not as simple, with the torque often being a non monotonic function of the disk rate $\Omega$ — moreover, the details of the rheological curves depend on both the vibration strength $\Gamma$ and filling height $h$. We note here that the sign of $\partial T/ \partial\Omega$ has a crucial rheological implication: flows for which $\partial T/ \partial\Omega > 0$ can also be accessed in experiments where the torque is fixed, whereas flows for which $\partial T/ \partial\Omega < 0$ are [*unstable*]{} in torque controlled experiments. As we discussed in [@2011_prl_dijksman], this range of unstable flows leads to hysteretic switching between two different flow regimes when the torque is varied, and is deeply connected to the yielding behavior of granular media observed for $\Gamma=0$.
Here we focus on rate controlled experiments, and as a first step in characterizing these curves, we plot the boundaries between the regions where $\partial T/ \partial\Omega$ is positive and negative for three values of $h$ in Fig. \[fig:slope\]. Roughly speaking, we can distinguish three regimes.
[*Inertial flows —*]{} For $\Omega \gtrsim 0.3$ rps, $\partial T/ \partial\Omega > 0$; the flow curves show an increasing $T$ for increasing $\Omega$. This increase corresponds to the onset of the inertial regime [@pouliquen_I]. To estimate the inertial number $I=\dot{\gamma}d/ \sqrt{P/\rho}$ at $\Omega=0.3$ rps, we have to choose a characteristic pressure and strain rate scale, as both $\dot{\gamma}$ and $P$ vary throughout the system. Taking $P$ as the hydrostatic pressure at $0.5H$, and $\dot{\gamma}$ corresponding to a shear band of 3 particles wide, we get $I=0.09$ for $h=0.38$ and $\Omega=0.3$ rps. Considering that the inertial regime typically starts at $I=0.1$ [@2006_nature_jop], there is good agreement between the onset of increasing $T(\Omega)$ and the onset of the inertial regime. In the remainder of the paper we will focus on slower flows.
[*Unstable flows —*]{} For intermediate flow rates, $T(\Omega)$ has a negative slope for small $\Gamma$ — for $\Gamma \rightarrow 0$, this regime extends to arbitrarily small flow rates, although there the flow curves become essentially flat. Despite the unstable character of the global rheology, and in contrast to unstable flows in e.g. micelles [@schall], we do not see any changes in the shear bands as we move in and out of this unstable regime. As the variation of the stress with flow rate is not very large, this regime can also be referred to as [*quasistatic*]{}.
[*Vibration dominated flows —*]{} Both the unstable/quasistatic and inertial regime have been studied in great detail already [@1980_powtech_nedderman; @2006_pre_depken; @2007_epl_depken; @2004_epje_gdrmidi; @2006_nature_jop; @2008_annurevfluid_forterre], as they also arise in the absence of vibrations. Hence, in the remainder of the paper we will focus on the new vibration dominated regime of slow, stable flows that arises for $\Gamma >0$ and $\Omega < 10^{-2}$ rps.
One striking qualitative feature of this regime we already want to point out is the pronounced ”kink” in the flow curves that can be seen for $10^{-2}$ rps$<\Omega<10^{-3}$ rps in Fig. \[fig:overview\_g\]. In Fig. \[fig:overview\_g\]b we explicitly mark such a kink. The kinks coincide with the flow rates where $t_a$ needs to be sufficiently large for $T$ to equilibrate (see Fig. \[fig:aging\]). We suggest that at sufficiently low $\Omega$, compaction effects become significant, leading to an increase of $T$ with time, and a ”kink” in the flow curves.
{width="17cm"}
Vibration Dominated Flows {#granorl}
=========================
We will now turn our attention to the increase of $T$ with $h$, which allows us to probe the underlying mechanisms that govern the rheology of vibration dominated flows. The canonical starting point of descriptions of [*non-vibrated*]{} slow granular flows is that the shear stresses $\tau$ are proportional to the pressure $P$ [@1980_powtech_nedderman; @2006_pre_depken; @2007_epl_depken], and the ratio of $\tau$ and $P$ is an effective friction coefficient, $\mu$. For inertial flows, a description where $\mu$ becomes rate dependent (through the inertial number) has been shown to capture much of the phenomenology [@2004_epje_gdrmidi; @2006_nature_jop; @2008_annurevfluid_forterre], and for slow, non-vibrated flows, this Mohr-Coulomb picture combined with a non-local rheology captures the essentials of steady, slow granular flow [@2012_prl_kamrin; @PNAS_kamrin].
By varying the filling height $h$, we can modify the pressure $P$ and probe its role for the rheology in the different regimes. Clear predictions for $T(h)$ exist from a well-studied rheological model for the driving torques in a split-bottom geometry [@2004_prl_unger]. In addition, this model provides clues to the flow’s special structure and how it depends on friction and other factors. In this section, we describe how our experiments allow us to build on these basic ingredients to identify two qualitatively different regimes in vibration dominated flows. We find a frictional regime in which $P \sim \tau$, yet with $\mu(\Omega)$ a rate dependent friction for $\Omega \gtrsim$ 10$^{-3}$ rps or $\Gamma \lesssim 0.8$. For even slower, more strongly vibrated flows, both the rheology and the location of the shear band presents strong evidence for a regime where $T$ becomes *independent* of $P$.
![(Color online) $\mu(\Omega)$ as obtained from the fit with the frictional model. The inset shows one example of the fit for $\Omega=8.5\cdot\times10^{-4}$ rps. The upper curve (+) shows the raw data $T$, the bottom curve ($\diamond$) is the raw data minus the correction term, $\tilde{T}$ – which goes through the origin. The fit matches the data very well resulting in a $\chi^2$ of 2.0$\times 10^{-3}$ (for the upper curve).[]{data-label="fig:g0metinset"}](Figure_7.eps){width="\columnwidth"}
Torque Minimization Model {#sec:unger}
-------------------------
To interpret the observed filling height dependence of the shear stresses, we start from a simple frictional model due to Unger [*et al.*]{}, which was developed to describe the 3D shape of the shear zones in the split-bottom geometry, but which also makes a precise prediction for the driving torque as function of filling height for purely frictional flows [@2004_prl_unger; @2006_prl_fenistein; @2010_sm_dijksman]. This model is based on the following three ingredients. First, think of the shear zones as localized along a narrow sheet $r(z)$ (corresponding to the center of the shear zones [@2003_nature_fenistein; @2004_prl_fenistein; @2006_prl_fenistein; @2010_sm_dijksman]). Second, assume that the stress tensor is colinear with the strain rate tensor [@2006_pre_depken] and proportional to the hydrostatic pressure. Third, assume that the sheet shape $r(z)$ minimizes the driving torque [^1]: $$\tilde{T}[r(z)] = 2 \pi g \rho \mu \int^{H}_{0}(H-z)~r^2\sqrt{1+(dr/dz)^2}~dz ~, \label{ttildeH}$$ where $g$ denotes the gravitational acceleration, and $\rho$ the bulk density (1.7$\times$10$^3$ kg/m$^{3}$) of the granular material. Minimizing $\tilde{T}$ for a given $h$ determines the shear sheet $r(z)$, from which the torque can be determined as function of $h$. As expected, we can write this torque as $\tilde{T}(h)=\mu \tilde{T}_f(h)$, where $\tilde{T}_f$ is a universal function of $h$. Note that for shallow filling heights, the torque is approximately proportional to the product of pressure and the extension of the shear band, so that $\tilde{T}_f(h)$ is quadratic in $h$ for $h \ll 1$.
In contrast to the original split-bottom cell for which Eq. (\[ttildeH\]) was developed, in our system the driving disk is slightly elevated with respect to the bottom. This is done in order to avoid observing spurious torque fluctuations that we associate with the diverging strain rate in the original split-bottom setup. The elevated disk leads to a $\mu$ depended addition in the experimental torque signal $T$, due to slip between the side of the disk and the stationary particles next to it. We have found that this drag term can be estimated as: $$T_{drag}(H)=2\pi r_s^2\mu \rho g \int_0^s (H+z)dz~, \label{drageq}$$ where $s$ is the disk thickness (5 mm), and $\mu$ is the effective friction coefficient for sliding of the disk past the particles, for which we use the same effective friction coefficient as for the granular flow. The contribution of $T_{drag}$ to the torque varies with $H$ and is proportional to $\mu$, so that we can write $T_{drag} = \mu T_d$, where $T_d$ can be deduced from Eq. (\[drageq\]).
We conclude that the measured torque $T$ is composed of two contributions: $$T=\tilde{T}(H)+T_{drag}= \mu \left[ \tilde{T}_f(H) + T_d\right]=\mu T_f~. \label{ungereq}$$
In conclusion, we can extract $T(h)$ from our flow curves, and check whether the flow appears frictional, and if so, determine $\mu$ [@2010_pre_dijksman] and $\tilde{T}(h)$.
### $\Gamma$=0
In Fig. \[fig:overview\_h\]a we show flow curves for $\Gamma=0$ and a range of $h$. Clearly, the torque only weakly varies with $\Omega$, and we expect the stresses to be frictional. For each fixed $\Omega$, we extract $T(h)$ from our data and fit it to $\mu T_f$ (Eq. (\[ungereq\])), as shown in the inset of Fig. \[fig:g0metinset\]. We find that this fit is excellent, which implies that the stresses are frictional, and which allows us to extract $\mu(\Omega)$. As shown in Fig. \[fig:g0metinset\], $\mu(\Omega)$ is almost flat, and has the same shape as the flow curves. We stress here that $\mu(\Omega)$ together with the frictional model predicts the stresses for *all* values of $h$, thus representing all the flow curves taken at different $h$. We note that our values for $\mu$ are comparable to those found previously in a standard split-bottom cell using the same particles [@2010_sm_dijksman].
Frictional Model for $\Gamma>0$ {#sec:Ggt0}
-------------------------------
In Fig. \[fig:overview\_h\]b and Fig. \[fig:overview\_h\]c we show examples of flow curves for a range of $h$ and $\Gamma>0$. We will now use $T(h,\Gamma>0)$ to test if the basic assumptions for the Unger model break down in the vibration dominated regime. We will find two flow regimes with the distinguishing features $T \sim \mu(\Omega) P$ and $T \nsim \mu P$. We describe here how we can distinguish these regimes in the rheological data.
![(Color online) (a) The effective friction coefficient $\mu(\Gamma,\Omega)$ as found by fitting the data with Eq. \[ungereq\]. (b) The $\chi^2$ of the fits, as defined in Eq. (\[chi2\]). []{data-label="fig:3x3"}](Figure_8.eps){width="\columnwidth"}
From Fig. \[fig:overview\_h\]b-c we see that that the flow curves for $\Gamma>0$ all show significant rate dependence. Even so, we attempt to fit Unger’s model to the rheological data. We thus fit $T(h,\Omega)$ to try to obtain a $\mu(\Omega)$. If this rate dependence were captured by an effective friction coefficient that only depends on $\Omega$, with $T(h,\Omega) = \mu(\Omega) T_f(h)$, the rate dependence would lead to a good fit of our data to the frictional model. To quantify the deviations between the data and fits to the frictional model, we calculate the best estimate of $\mu$ and the corresponding $\chi^2_f$ as follows. For each fixed $\Omega$ and $\Gamma$, we have measured the torque for seven values of $h$, and then determine: $$\label{chi2}
\chi^2 := \langle (\mu T_f(h) - T(h))^2 \rangle / \sigma^2_{T(h)}~,$$
We apply this procedure for each value of $\Omega$ and $\Gamma$, and show the result for $\mu$ and $\chi^2$ of these calculations in Fig. \[fig:3x3\]. For $\Omega>0.3$ rps - the inertial regime - the fit works very well and results in a weakly rate dependent $\mu$, just as for $\Gamma=0$. In addition, we find a large region for $\Gamma \leq 0.83$ and $\Omega<0.5\times 10^{-3}$ where the fit also works well, but this time with a more strongly rate dependent effective fiction $\mu(\Omega)$. This tells us that even in this rate dependent, vibration dominated regime, a frictional prediction is perfectly capable of describing the flow.
We do however observe two distinct regimes where $\chi^2$ is large, indicating a poor fit. First, there is a significant peak in $\chi^2$ around $\Omega=10^{-2}$ rps for $\Gamma\geq0.7$. Second, for $\Gamma\geq0.95$ and $\Omega<10^{-3}$ rps, $\chi^2$ also is substantial. The underlying physics in these two regimes is different. As we will show in the next section, the first peak is associated with a broad crossover regime between rate dependent and rate independent flows – a rather trivial consequence of the flow profiles in the split-bottom geometry. The second peak we associate with a flow regime in which the rheology becomes pressure independent, as shown in in section \[sec:fluid\].
![(Color online) (a) $T(h)$ for $\Gamma$=0.95 and a range in $\Omega$ at the point where the rate dependence starts. It can be seen that the curve drop for high $h$, resulting in an s-shaped $T(h)$ curve rather than an upwards curved $T_f$ one. The black line is the fit with the frictional model to the top curve. (b) A theoretical prediction of $\dot{\gamma}/\Omega$ -which decreases with $z$- in the split-bottom cell [@2010_sm_dijksman; @wandesman_epl_2012]. Dark color presents high $\dot{\gamma}/\Omega$, white is low $\dot{\gamma}/\Omega$.[]{data-label="fig:ratedep"}](Figure_9.eps){width="\columnwidth"}
### Onset of Rate Dependence
The peak in $\chi^2$ around $\Omega=10^{-2}$ rps is consistent with the onset of rate dependence below $\Omega =10^{-2}$ rps as per the following reasoning. First, both our raw data for $T$ as well as the best fits for $\mu$ show that rate dependence sets in rather abruptly for $\Omega<0.1$ rps, and that rate dependence is strongest for large $\Gamma$, consistent with the location and strength of the peak in $\chi^2$. Crucially, this onset of rate dependence sets in at different flow rates for different heights (see Fig. \[fig:overview\_h\]), so that at a given $\Omega$, the data for $T(h)$ mixes rate independent and rate dependent flows.
In Fig. \[fig:ratedep\]a we show examples of $T$ as function of height, that illustrate that when $\Omega$ enters this rate dependent regime, $T(h)$ strongly deviates from the quadratic form predicted by Eq. (\[ungereq\]). To interpret this deviation, it is important to realize that at a given $\Omega$, the local strain rate ${\dot{\gamma}}$ spans a wide range of values and has a strong $z$ dependence [@2004_prl_unger; @2010_sm_dijksman; @wandesman_epl_2012] — see Fig. \[fig:ratedep\]b. Hence, as the torque $T$ is an integral over the local stress in different layers in the material, $T(\Omega)$ mixes different local rheologies. More precisely: under the assumption that rate dependence sets in below a given ${\dot{\gamma}}$, there is a range of values of $\Omega$ for which the lower part of the system (where strain rates are largest) is still rate independent, whereas the top part of the system (where strain rates are smallest) are already rate dependent. This is consistent with the ”drop” in the $T(h)$ curves at large $h$ shown in Fig. \[fig:ratedep\]a — the deviations from the Unger model emerge first for large $h$, for which the range of strain rates is biggest and regime mixing is thus most pronounced. Our data also shows that once $\Omega$ is sufficiently low, so that all of the material is in a rate dependent state, $T(h,\Omega)$ is close to $\mu T_f(h)$ so that $\chi^2$ drops to low values again, and $\mu$ can be replaced with a rate dependent $\mu(\Omega)$. The range of $\Omega$ over which this crossover exists broadens with $\Gamma$, since the rate dependence becomes stronger with $\Gamma$.
In conclusion, the lowering of the friction coefficient $\mu$ and the peak in $\chi^2$ around $\Omega=0.01$ rps are caused by the onset of rate dependence which occurs at different $\Omega$ for different vertical locations in the flow cell. For $\Gamma \lesssim 0.8$, we also observe that once all the material is in the slow, rate dependent regime, the fit to the frictional model achieves a low $\chi^2$ again, so that $T(h,\Omega) \approx \mu(\Omega) T_f(h)$.
![(Color online) $T(h)$ curves for $\Gamma=1$ and $\Omega<10^{-3}$ rps. Towards lower $\Omega$, the curves lose their curvature and become straight lines. []{data-label="fig:toline"}](Figure_10.eps){width="\columnwidth"}
Fluidized Region {#sec:fluid}
----------------
The growth of $\chi^2$ for large $\Gamma$ and low $\Omega$ signals a breakdown of the frictional picture, where shear stresses are proportional to the pressure, as we will describe in this subsection. To gain deeper insight in the flow phenomenology in this regime, we plot $T(h)$ for $\Gamma=1$ and a range in $\Omega$ in Fig. \[fig:toline\]. We see that for all filling heights the stresses drop with $\Omega$, and at low $\Omega$, $T(h)$ becomes approximately [*linear*]{}. The standard Unger model predicts a quadratic dependence of $T(h)$ on $h$, as mentioned above. A linear dependence would suggest a pressure independent rheology, for which the increase of $T$ with $h$ is only due to increasing surface area on which the shear stress acts.
Note that the large values of $\chi^2$ here cannot be due to the existence of a crossover regime, as presented above for $\Omega \approx 10^{-2}$ rps. Evidence for this comes from Fig. \[fig:overview\_h\]c, which shows that both rate dependence of $T(\Omega)$ is small, and that there is no strong difference in the rate dependence for different values of the height in this regime. First of all, that means that there is little mixing of different rheologies in the global torque signal; second, the rate dependence is weak, so even if there were some mixing, it would not produce a strong $h$ dependence.
It it perhaps not surprising that new phenomena occur around the special value $\Gamma=1$. For $\Gamma \approx 1$, the grains lose contact during part of the vibration cycle — the precise value of $\Gamma$ where this happens depends on details [@durian_prl_1997; @2005_pre_mobius]. As a result, the confining pressure becomes zero during part of the cycle, and as most slip can be expected to occur when the normal grain forces are absent, the flows may become pressure independent, as in a viscous liquid.
![(Color online) (a) $\omega(r)$ for $\Omega=10^{-1}$ rps, $\Gamma=0.2$ ($\times$), $\Omega=10^{-3}$ rps, $\Gamma=1$ (+), $\Omega=10^{-4}$ rps, $\Gamma=1$ ($\diamond$). In red, we add the fit with $\omega(r)=1/2-1/2$ erf$[(r-r_c)/W]$. (b) The center of the shear band at the surface $r_c$ as a function of $\Gamma$ for h=0.47 and $\Omega$ as indicated. For low $\Omega$, $r_c$ clearly decreases.[]{data-label="fig:flowprofile"}](Figure_11.eps){width="\columnwidth"}
### Rate Dependent Flow Structure
Additional evidence for the loss of pressure dependence for high $\Gamma$ and low $\Omega$ comes from measurements of the flow structure. From finite element calculations on the flow structure of a viscous liquid in the split-bottom geometry, it is known that the shear band is much closer to the center of the cell than for frictional flow [@2010_pre_dijksman]. As such, a pressure independent rheology for the granular flows in this regime can be expected to be accompanied by similar changes in the flow structure.
To test this, we have measured the velocity profiles $\omega(r)$ at the surface of our system for a range in $\Gamma$ and $\Omega$ using particle image velocimetry [@2004_prl_fenistein; @2006_prl_fenistein; @2010_sm_dijksman]. In Fig. \[fig:flowprofile\]a we show examples of $\omega(r)$, showing a broadening and shift of the shear zones when $\Omega$ enters the pressure independent regime. We fit the velocity profiles with $\omega(r)=1/2-1/2$ erf$\left[ (r-r_c)/W \right]$, where $r_c$ is the center of the shear band at the free surface [@2004_prl_fenistein]. In Fig. \[fig:flowprofile\]b, we plot $r_c$ for $\Omega$ ranging from $10^{-1}$ to $10^{-4}$ rps and a range in $\Gamma$. Clearly, the location of the shear band is mostly independent of $\Omega$ and $\Gamma$, including in most of the rate dependent regime. However, in the regime where we observed the pressure independent rheology, we observe significant deviations in the flow profiles. The deviations show a trend towards a shear band moving inwards — consistent with the idea of a pressure independent regime.
Moreover, we can modify the Unger model to test which rheological scenario is most compatible with the observed shift in the shear band. Throughout, we assume that the torque minimization principle is robust. The frictional torque model assumes $\sigma(z) \sim 1-z/H$, in which the shear stress, being proportional to the hydrostatic pressure, goes to zero at the surface. We can replace this model with $\sigma(z) \sim 1-(1-\alpha_1)z/H$, in which the shear stress reaches a final value when approaching the free surface – see Fig. \[fig:ungershift\]a. The extreme case $\alpha_1 = 1$ represents a pressure independent rheology. We compute the location of the shear band at the free surface as a function of model parameter $\alpha_1$. The results are shown in Fig. \[fig:ungershift\]b. We find that for larger $\alpha_1$, the location of the shear band at the free surface moves inwards. Thus, the closer the model resembles a Newtonian rheology, the more the shear band moves towards the center. This can be understood intuitively as follows: the penalty for having a shear band at large radius at the surface is zero in the pressure dependent model, because the shear stress goes to zero at the free surface. Once a finite amount of shear stress is present in the shear band at the surface, torque minimization will move the shear band inwards precisely as we observe in the experiments at $\Omega < 10^{-3}$ rps, $\Gamma > 0.9$.
Conversely, for a frictional, rate dependent rheology, the shear stress closer to the surface is *lower* than that of a simple frictional model. We model this with a $\sigma(z)$ that can be captured with $\sigma=(1-z/H)+\alpha_2\sin(2 \pi z/(2H))$, as shown in Fig. \[fig:ungershift\]c. The torque penalty for having a shear band at finite $r$ is thus reduced, and the model predicts indeed an *increase* of the shear band radius at the free surface (Fig. \[fig:ungershift\]d), contrary to what we observe. We thus conclude that our observation of the inward displacement of the shear band location at $\Omega < 10^{-3}$ rps, $\Gamma > 0.9$ is consistent with the granular flow obtaining a rheology which becomes pressure independent.
![The center of the shear band at the surface, $r_c$, can be found using the method by Unger for given $z$ dependent stress $\sigma(z)$. In (a), we plot $\sigma(z)$ for case $\sigma(z) \sim 1-(1-\alpha_1)z/H$ where $\sigma$ is still finite at $z=H$, in contrast to the frictional description where $\sigma(H)=0$. The resulting $r_c$ is shown in (b), we recover in inwards moving shear band as we observe in experiments. In (c) we plot $\sigma(z)$ for $\sigma=(1-z/H)+\alpha_2\sin(2 \pi z/(2H))$, where $P$ and $\mu$ respectively vanish and decrease towards the surface, corresponding the a strain rate dependent frictional picture, as we show in (d), this predict an outwards moving shear band, contrary to what we observe.[]{data-label="fig:ungershift"}](Figure_12.eps){width="\columnwidth"}
Constant Torque Driving {#sec:constantT}
=======================
In this section we discuss stress controlled experiments. These experiments consist of the following protocol: after pre-shearing the sample, a fixed value of the driving torque $T$ is set, and the ensuing rotation angle $\theta$ of the disk is monitored. We show several examples of the time evolution of $\theta$ and its derivative $\Omega$ for a range of values of the torque $T$ and vibration strength $\Gamma$ and fixed $h = 0.33$ in Fig. \[fig:Tmerge\]. Data for other values of $T$, $\Gamma$ and $h$ look similar; the small initial oscillatory motion, visible for small $\Gamma$, is due to transient flexure oscillations.
As shown in Fig. \[fig:Tmerge\], in most cases the disk rotation rate reaches a steady state value after some equilibration time. For small $T$, this equilibration time becomes extremely long, much longer than in the rate controlled experiments shown above. This is typical for complex fluids that exhibit a so-called viscosity bifurcation [@coussot_pre_2002; @bonn_epl_2009], and we suggest that weakly vibrated granular media fall into this category. We also note that for small $T$, the angle as function of time (or alternatively, $\Omega(t)$) exhibits power law behavior reminiscent of Andrade creep [@andrade_1914], again a typical feature of weakly driven disordered media.
For the data sets where $\Omega$ reaches a steady state, we find that $\Omega(T)$ and $T(\Omega)$ are consistent — in other words, the steady state flow properties are the same as measured through the rheology under constant rate or constant stress driving. Consistent with this picture, we observe that the data in Fig. \[fig:Tmerge\] illustrates that the steady state values of $\Omega$ exhibit a jump between inertial rates ($\Omega>0.5$ rps) and significantly slower flow rate ($\Omega < 10^{-2}$ rps), consistent with the hysteretic transition between inertial and vibration dominated flows observed in [@2011_prl_dijksman].
![(Color online) The phenomenology for constant torque experiments at $H/R_s = 0.33$. In (a), we show the $\theta$ as a function of time, in (b), we show the derivative of $\theta$: $\Omega$. It can be seen that the curves for low $\Gamma$ and $\tau$ show very long transients. The black line has a slope -1. The curves are down up for $\Gamma=$0.2,0.3,0.4,0.4,0.4,0.5,0.6,0.7,0.75,0.8,0.8,0.8,0.9,0.9,1 and $T$ \[mNm\]=3.36, 4.62, 5.74, 5.88, 7.14, 8.54, 9.94, 10.64, 11.3, 10.64, 11.3, 12.04,13.02,13.44, as also indicated in the inset of (b).[]{data-label="fig:Tmerge"}](Figure_13.eps){width="\columnwidth"}
Conclusion {#sec:conc}
==========
To summarize, we probe the rheology of weakly vibrated granular media and find several different flow regimes. First, for $\Omega>0.3$ rps, our data shows the well known *inertial flow* regime, consistent with a rough estimate of the inertial number. In the absence of vibrations, lower flow rates lead to an essentially rate independent, *quasistatic*, regime, where the variation of the torque is small, and where $T(h)$ is well fitted using Eq. (\[ungereq\]), implying that the shear stresses are proportional to the pressure here. For $\Gamma=0$, this regime covers all $\Omega < 0.1$ rps, whereas the range of flow rates where this rate independent regime resides shrinks in the presence of vibrations, and almost vanishes for $\Gamma=1$. For $\Omega$ below the rate independent regime and $\Gamma>0$, we have described three *vibration dominated* regimes. For two of these regimes, our data shows that the shear stresses are still proportional to the normal stresses, but now via rate dependent $\mu(\Omega)$. For the slowest of these two regimes, we see a slow densification, leading to a kink in the flow curves. Finally, for $\Gamma$ close to one, the vibrations affect the rheology of the granular medium so significantly, that the shear stresses are no longer proportional to the normal stresses, signifying a complete departure of the frictional nature that is a hallmark of all other types of slow granular flows.
Acknowledgements {#acknowledgements .unnumbered}
================
We appreciate helpful discussions with O. Dauchot and H. Jaeger, J. Mesman for outstanding support in constructing the setup, and T. Tampung for support in developing the electronics. We thank L. van Dellen for his contribution to the experiments and data analysis. This work was supported by the Dutch physics foundation FOM.
Setup {#app:setup}
=====
Our setup allows to do sensitive rheology on vibrated materials and its main parts are illustrated in Fig. \[fig:setup\]. The granular split-bottom flow cell forms the heart of setup, and consists of a cylindrical flow cell, the bottom of which is formed by a rotating disk that drives the flow. The disk is attached to an axle which is mounted in a cylindrical air bearing ($\varnothing$ 1/4", New Way): rotational motion is friction-free which allows precise rheological measurements.
In ordinary split-bottom cells, the strain rate diverges at the split. We have found that the torques occasionally show strong fluctuations in this case, presumable due to individual particle being trapped just above the split — moreover, these fluctuations depend on the precise roughness near the split, thus leading to a dependence of the average $T$ on such experimental details. To avoid this, we make sure that the strain rate field is smooth at the grain level and have elevated the driving disk by 6 mm (Fig. \[fig:diskfric\]). The side of the disk is smooth, and particles immediately next to the disk hardly move, creating a static bottom layer flush with the disk. Hence, the boundary conditions are essentially the same as for the ordinary split-bottom disk, and the elevation does not affect the overall flow field for $h\equiv H/r_s$ larger than about 0.1. The elevation does ensue that the torques are insensitive to experimental details and do not show the aforementioned spurious fluctuations.
We induce vibrations in the split-bottom cell with an electromagnetic shaker (VTS systems VG100), driven by a function generator (Thurlby Thandar TG1010) via a commercial audio amplifier (Crown CE1000). The shaker is coupled to the flow cell via a long rod, and the cell is mounted on a square air bearing (4“x4”, New Way), which ensures that the slider and flow cell move virtually friction free in the vertical direction, with the amplitude of the residual horizontal vibrations in the submicron range due to the stiffness of the air bearing. The total combined weight of the slider and the split-bottom cell is approximately 12 kg and too large to be carried by the shaker alone. Therefore we support their weight with four suspension springs with a stiffness of 2 kN/m, connected to the aluminum frame shown in Fig. \[fig:setup\]a,b. The vertical acceleration of the split-bottom cell is measured by a combination of two accelerometers (Dytran E3120AK and a modified ADXL322EB [^2]). To limit the mechanical coupling between the table and the shaker, the shaker is placed on a stack of concrete tiles with rubber mats sandwiched between them. We do not observe any appreciable heaping [@1995_prl_pak].
The vibrations in the flow cell should not induce relative vertical motion between the disk that drives the granular flow and the container. To ensure that the disk only induces rotational shear deformations and no additional vibrations, we attach a flange to the disks axle. The surface of the flange is polished, and clamped between four flat air bearings ($\varnothing$ 25 mm, New Way) – see Fig. \[fig:cusairbearing\]. By mounting the flat air-bearing assembly on the container, we ensure that disk and container move together during the vibration cycle: the stiffness of one flat bearing, for a typical gap between the bearing and the flange of 5 $\mu$m, is 18 N/$\mu$m. Since the peak acceleration of the system is never larger than $\sim 1.5$ $g$, and the weight of the flange/axle/disk construction is less than a kilogram, the maximum residual motion between the disk and the container is smaller than a micrometer, which is the typical length scale of asperities on glass beads [@2005_pre_schroter].
A rheometer (Anton Paar DSR 301), coupled to the flow cell via a system of flexures and airbearings, allows to probe and control the granular flow. As the rheometer cannot be vibrated, and its rotational axis is housed in a very stiff air-bearing, we need to decouple the vibrations from the rheometer, and also compensate for inevitable misalignments of rheometer and flow cell. We therefore couple the two axes with a flexure, as shown in Fig. \[fig:setup\]c. This flexure has a torsional spring constant of 4 Nm/rad and compressional spring constant of 5 $\times 10^2$ N/m. The flexure can be compressed easily and therefore accommodate the vertical motion of the disk axle. The copper blades used in the flexure are flexible enough to compensate for any small misalignment between the axis of the disk and the rheometer, without exerting significant forces on the bearings.
Our DSR 301 rheometer is a stress driven rheometer, so its native mode is to set a torque to control rotation. However, it is also possible to use the rheometer in a so-called constant shear rate mode. In this mode, the rheometer applies a feedback routine that dynamically adjusts its torque in order to achieve a constant rotation rate. This process has one control parameter: the rate at which the torque is increased or decreased. This timescale is set by the desired driving rate, and by a user-defined time constant, and in principle could affect all constant shear rate experiments. In these experiments we set the feedback constant such that we observe smooth rotation, which is achieved for a feedback parameter, the so called CSR-value, of 25%. If the torque adaptation rate is too high and becomes of the order of the eigenfrequency of the rheometer axle, the feedback routine will be affected by the rotational eigenmodes of the mass-spring system of the rheometer axle.
![Schematic drawing of the custom-built air-bearing assembly that fixes the disk axle in the vibrating reference frame of the container. The flange is held in place by four flat bearings, the axle is fixed by a cylindrical bearing.\[fig:cusairbearing\]](Figure_16.eps){width="\columnwidth"}
Alignment
---------
To compensate for inevitable misalignments between the shaker axis and vertical air-bearing axis, we couple them via a 40 cm long and flexible aluminum rod. The aluminum rod is flexible enough to compensate for a horizontal alignment mismatch of a few millimeter, but stiff enough to ensure a mechanical coupling for vertical motion (see Fig. \[fig:setup\]a).
Misalignment between the disk and rheometer axle causes rotationally periodic variations in the force that the rheometer exerts on the disk and grains. This periodic modulation of the torque can be reduced by improving the alignment of the flexure-coupled axles. We manually align the rheometer with the disk axis by setting a constant rotation rate to the disk, with no particles in the container. While the disk rotates, we monitor the required torque to run the disk at the set speed. This torque has a periodic modulation; alignment of the two axis with the set screws on the air bearing reduces the amplitude of this periodic modulation on the torque.
A typical alignment run is shown in Fig. \[fig:residue\]: the modulations can be minimized to an amplitude of less than 20 $\mu$Nm, which is $<$ 0.1% of the largest torque need to drive our granular flows. Nota that a small torque offset also remains since the flat air bearings that confine the vertical motion of the flange cannot be positioned perfectly parallel to the flange. This induces an asymmetry in the air flow between the bearings and the flange. This exerts a torque of the order of 20 $\mu$Nm on the flange; a negligible amount in comparison to the typical driving torques employed in our experiments.
![A typical torque signal under constant rotation rate during the alignment of the air bearing. One oscillation period corresponds to a full rotation of the disk[]{data-label="fig:residue"}](Figure_17.eps){width="6cm"}
Vibration Control
-----------------
![Steady state rotation rate dependence on the vibration frequency, measured with $\Gamma = 0.5$, $\tau = 0.8$. []{data-label="fig:vibrheo:mechresp"}](Figure_18.eps){width="6cm"}
To choose an appropriate external vibration frequency, we measure how the steady state rotation rate $\Omega$ under a constant applied torque $T=25$ mNm and fixed vibration amplitude $\Gamma = 0.5$ depends on the vibration frequency $f$. We carry out the measurements with a filling height of $h$ = 0.55. The result is shown in Fig. \[fig:vibrheo:mechresp\]. We see that there is a resonance above 105 Hz, and for vibration frequencies between 50 and 75 Hz, the rotation rate depends only marginally on the vibration frequency. We choose the middle of this regime, $f$ = 63 Hz to carry out all experiments discussed in this paper. The vertical vibration amplitude at a peak acceleration of 1 $g$ at this frequency is 64 $\mu$m.
It is critical that the vibration amplitude is constant during the duration of the experiments, which can last for hours. The shaker generates considerable heat, which affects its operation; with constant driving current to the shaker coils, the vibration amplitude slowly drifts over $\sim 0.1 g$ per hour.
To keep the vibration amplitude constant during the experiments, we employ a LabVIEW driven feedback scheme that continuously adjusts the driving current for the shaker coil. A function generator, under computer control, sends a sine wave to the amplifier that drives our shaker. Accelerometers detect the actual shaking amplitude. We find that the harmonic distortion of the sinusoidal shaker motion is about 1%, so we use a lock-in amplifier (SRS 830) to achieve a high accuracy measurement of the vibration amplitude. The computer then adjust the output of the function generator, and with this feedback scheme we can maintain a stability of $\Gamma$ to within $<10^{-3}$ $g$.
Steady state $T(t)$ {#proto}
-------------------
![Rotation rate (a,b) and driving torque (c,d) shown as a function of time for an experiment in which the set speed is reduced logarithmically. (a,c) are measured for $\Gamma = 0.2$, (b,d) are measured for $\Gamma = 0.8$. Both experiments are carried out at $h = 0.33$.[]{data-label="fig:transient"}](Figure_19.eps){width="\columnwidth"}
We verify that a steady state flow rate can be obtained by the rheometer and that we observe no long term transients in the rate control mode. To do this, we measure the instantaneous driving torque and speed as a function of time for different rotation rates $\Omega$ and vibration amplitudes $\Gamma$. We apply pre-shear before the start of the experiment, and start a logarithmic ramp down in $\Omega$, from $5\times 10^{-1}$ to $5\times 10^{-5}$ rps with 2 steps per decade. We measure for all $\Omega$ over a minimum strain of 0.5 rotation, or a total shear time of 10 seconds, whichever is longer. We sample the rotation rate and torque with a rate of 10 Hz. The results for $h$ = 0.33 are shown in Fig. \[fig:transient\], both the instantaneous speed of the disk and the driving torque as a function of time. Note the logarithmic scale on the time axis. We see that for $\Gamma = 0.2$, the set speeds are achieved very accurately and no large fluctuations or transients are observable. The fluctuations in the torque signal are however appreciable. Transients are short, even for the slowest run — for more on transients, see the next section. For $\Gamma = 0.8$, the fluctuations in both speed and torque are strongly reduced and we can clearly see that no long time transients are present in these experiments. We conclude that the mean driving torque $T$ and mean rotation rate $\Omega$ are well defined quantities in steady state shear experiments.
[100]{} natexlab\#1[\#1]{}bibnamefont \#1[\#1]{}bibfnamefont \#1[\#1]{}citenamefont \#1[\#1]{}url \#1[`#1`]{}urlprefix\[2\][\#2]{} \[2\]\[\][[\#2](#2)]{}
, ** ().
, , , ****, ().
, ****, ().
, ****, ().
, , , , , ****, ().
, , , , , , ****, ().
, , , ****, ().
, , , ****, ().
, , , ****, ().
, , , , , , ****, ().
, , , ****, ().
, ****, ().
, ().
, , , , , ().
, ****, ().
, , , , ****, ().
, , , ****, ().
, , , , , ****, ().
, ****, ().
, , , , , ****, ().
, , , ****, ().
, , , ****, ().
, ****, ().
, , , , p. ().
, , , , , ****, ().
, ****, ().
, , , , ****, ().
, , , ****, ().
, ****, ().
, ****, ().
, , , , , ****, ().
, ****, ().
, , , ****, ().
, , , ****, ().
, ****, ().
, , , , **** ().
, , , **** ().
, **** ().
, , , ****, ().
, , , **** ().
, , , , ****, ().
, , (),
, ().
, , , , , ().
, , , ****, ().
, , , , , ****, ().
, ****, ().
, , , , , , , ****, ().
, , , ().
, **** ().
, , , , , , ****, ().
, , , ****, ().
, , , ****, ().
[^1]: Note that whereas we characterize the filling height with the dimensionless $h$, we write the frictional model in terms of $H$, the common notation.
[^2]: Its factory default bandwidth of 50 Hz is increased to 2 kHz.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The speed of light is a complicated synthesizer quantity with distinctive origins which lead to coincident values in the standard theory. Due to the fact that different aspects of speed of light do not coincide in the for different Palatini modified gravity theories, when deviating from general relativity, one should consider which aspect of speed of light should be taken into account meticulously and unambiguously. The aim of this study is mainly investigating the modification of the *SN Ia* (Supernovae Type Ia) luminosity distance, for two and extended theories in the in Palatini formalism considering different aspects of the speed of light. Besides the itself should be determined in the Palatini formalism as a frame in which the Einstein Equivalence Principle is valid. SN Ia luminosity distance should be modified considering the variation of the space-time causal structure constant for two well-known extended models in Palatini formalism.'
author:
- Azam Izadi $^1$
- Shadi Sajedi Shacker $^2$
title: Studying the Supernova luminosity distance in Palatini formalism considering the role of causal structure constant
---
Introduction
============
Nowadays the major challenge to the cosmological models is the late-time cosmic accelerating expansion of the universe. According to the cosmic observational evidences based on Type Ia supernovae (SN Ia) [@Riess; @Perlmutter], and standard rulers[@Shinji; @palatini_gravity_theory], our universe is experiencing an accelerating expansion phase. Two general classes of models have been put forward in order to explain this expansion behavior. In the first class, it is attributed to the new gravitating component for the energy content of the universe, dark energy, with repulsive gravitational properties due to its negative pressure. The second class of models looks for an accelerating expansion via modification of general relativity theory on cosmological scales or interpreting cosmological observations in another theoretical frame [@uzan]. A variety of approaches with richer space–time concepts are proposed to modify Einstein’s theory of gravity. These can be put into different categories such as Scalar–Tensor theories, Tensor–Vector–Scalar theories (TeVeS), higher dimensions theories, Kaluza–Klein theory, and theories with a modified Lagrangian ($f(\az, \az_{\mu\nu}, \cdots)$) in three metric, Palatini, and metric–affine formulations. Among all these various options, the advantage of $f(\az)$ and $f(\az^{\mu\nu} \az_{\mu\nu})$ extended gravity theories is that no extra degree of freedom is suggested and the scalar curvature with well-understood physical origin causes the accelerating expansion.\
As it mentioned there are several formalisms in dealing with General Relativity such as *metric formalism* and *Palatini formalism*. In metric formalism Lagrangian is modified for extended theories and the connection is the Christoffel symbol, but Birkhoff’s theorem does not hold. In contrast, the metric and affine connection are regarded as two independent variables in Palatini formalism, the Birkhoff’s theorem holds there and the energy–momentum tensor is taken independent of the connection [@palatini_approach; @palatini_approach_fR; @fRgravity; @palatini_gravity_theory]. Therefore, conservation of the energy–momentum tensor is still valid.\
It is worth noting that according to Ostrogradski theorem[@Ostro], Lagrangians with higher derivatives may suffer from instability problems. Based on Ostrogradski theorem, Hamiltonians containing more than first time derivative would have linear instabilities, on condition that the higher derivatives cannot be eliminated by partial integration. The result of Ostrogradski theorem is not relevant here since for the $f(\az)$ and $f(\az^{\mu\nu}\az_{\mu\nu})$ theories in Palatini formalism, the Lagrangian depends just on the first derivative of the fields.\
The fact is that Palatini formalism gives a more generic geometric description of the space time. In addition, there are some relations between some extended models such as Ricci scalar and Ricci squared theories in the Palatini version and non-perturbative approaches to quantum gravity, Loop Quantum Gravity, which motivate researchers to study Palatini extended theories. Due to all these motivations, special attention has been paid to the Palatini formalism and its applications (e.g.[@tavakol_tsujikawa; @Olmo_stellar; @Binary]) in recent years. In this paper we will consider two extended gravity theories, non-linear Ricci scalar and Ricci squared theories, in the framework of Palatini in which $\az_{\mu\nu}$ represents the Ricci tensor, which is made by a connection independent of the metric. The metric formalism can be obtained from Palatini one, assuming the simple form $f(\az_{\mu\nu})=\az_{\mu\nu}$ [@fRgravity].
As a matter of fact the cornerstone of general relativity theory is the *equivalence principle*. The equivalence principle is true where one can find a local frame in which the gravitational effects are vanishing, which is called the *local inertial frame*. Despite the fact that in the standard theory of gravity, a is considered a frame in which the metric is locally Minkowskian and therefore the Christoffel symbol is zero, this is a questionable case in the modified gravity theories in the Palatini formalism [@Izadi_1; @OlmoPRL; @open_problems_gr_phys]. Since in such theories the connection is different from the Christoffel symbols, two different local frames can be defined in principle. One in which the connection is locally zero and the other in which the Christoffel symbols are locally vanishing. Therefore the main concern in the Palatini formalism is where the equivalence principle is true. These two frames are conformally related for Palatini $f(\az)$ gravity and thus the causal structure of both frames is the same, but this does not mean that they are completely equivalent from the physical point of view. For Palatini gravity, these two frames are not conformally related and thus their causal structure is different.\
The point is that the action of a theory should be defined on some space–time with predefined properties. In the Palatini formalism the space–time structure is defined via both the metric (as the device for defining length) and the independent affine connection (as the device for parallel transporting). This means that in any experiment that the concept of parallelism is important, the effect of the independent connection can be seen and should be evaluated. This controversial subject was meticulously reconsidered in [@Izadi_1]. In the foregoing paper the in Palatini formalism, in which the equivalence principle holds, was chosen wherever affine connection is locally zero. In [@Izadi_1] it is confessed that all the gravitational effects are removed from the equations in this introduced and then two samples of extended gravity theories were studied there. Since the Christoffel symbols are not zero, matter fields and the test particle trajectory are not the special relativistic ones, namely the particle’s trajectory is not a straight line. For a general case, one can identify the extra terms appeared in the particle’s trajectory with the force due to the variation of the speed of light [@Izadi_1].\
On the other hand the speed of light, $c$, is a complicated synthesizer quantity which has different distinctive origins in Physics, which results in coincident values in standard theory. As a matter of fact $c$ is not the speed of ”light” in all these theories. Indeed, there is no persuasive reason to consider all different aspects of light speed such as (the space-time causal structure constant), $c_{GW}$ (the gravitational wave velocity), $c_{EM}$ (the electromagnetic wave velocity) and $c_{E}$ (the space-time–matter coupling constant appearing at the right hand side of Einstein’s equations) the same in all formalisms [@Ellis] inasmuch as this fundamental constant enters many physical laws with different origins which are a prior unrelated. In any Deviation from GR or using a different formalism, one should consider precisely at first which aspect of speed of light is involved in the theoretical formulation and secondly whether all these facets still coincide and thirdly whether in an extended theory in different formalism, these facets remain constant or not. Furthermore the right technique of measuring spatial distances should be used [@Izadi_1; @Izadi_2].
On the whole, considering the advantages of Palatini extended theories and the fact that distinguishing different aspects of light speed may affect the analysis of cosmological data such as SN Ia, brings the idea of interpreting these data in a true .
In what follows, the local inertial frame in Palatini formalism will be briefly introduced according to the result of [@Izadi_1]. We will see that for the foregoing extended models, working in palatini formalism leads to not coinciding different aspects of speed of light in the .\
The precise formulas of redshift, Hubble parameter and SN Ia luminosity distance will be rewritten by considering the right aspect of speed of light, (coming from metric). Then we will show how the modifications, due to studying the extended Palatini theories in the right local inertial frame, affect the SN Ia luminosity distance in different and models in section \[sec:redshift\]. In section \[sec:Friedmannology\], the equations related to the SN Ia luminosity distance will be modified due to the variable feature of in the in Palatini formalism. In section \[sec:proposed\_model\], an appropriate model with a free parameter is proposed for modifying SN Ia relations to fit with observational data and its properties are studied as well.
Local Inertial Frame in Palatini extended theories
==================================================
Two well-known extended gravity models, $f(\az)$ and $f(\az^{\mu\nu} \az_{\mu\nu})$ are under study in the present paper.
Palatini $f(\az)$ and $f(\az^{\mu\nu} \az_{\mu\nu})$ models
-----------------------------------------------------------
In Palatini approach the Lagrangian density of the non-linear Ricci scalar gravity model is chosen to be an arbitrary function of the scalar curvature, $f(\az)$: $$\mathcal{L} = \frac{1}{2\kappa} f(\az[g,\Gamma]) + \mathcal{L}_m
\label{action}$$ in which $\mathcal{L}_m$ is the matter Lagrangian density and $\kappa=8\pi G/c_0^4$. It should be noted that the matter action does not depend on the connection in the Palatini formalism. The following equations of motion will be obtained by considering the metric and the connection as two independent variables: $$f'(\az)\az_{\mu\nu}-\frac{1}{2}f(\az)g_{\mu\nu}=\kappa T_{\mu\nu}
\label{eq1}$$ and $$\label{eq01}
D_{\alpha}\left (\sqrt{-g}f'g^{\mu\nu}\right )=0$$ where $D_{\mu} X^{\nu}= \partial_{\mu}X^{\nu} + \Gamma^{\nu}_{\alpha\mu}X^{\alpha}$. The above equation shows that by a conformal transformation, $h_{\mu\nu}=f'g_{\mu\nu}$, the connection is compatible with the metric $h_{\mu\nu}$ which is given by: $$\label{ggama}
\Gamma^{\alpha}_{\mu\nu}= {\alpha\brace \mu\nu}+ \gamma^{\alpha}_{\mu\nu}={\alpha\brace \mu\nu}+\frac{1}{2f'}\left [
2\delta^{\alpha}_{(\mu}\partial_{\nu)}f'-g_{\mu\nu}g^{\alpha\beta}.
\partial_{\beta} f'\right ]$$
In Palatini $f (\az_{\mu\nu}\az^{\mu\nu})$ gravity, after ADM decomposition, Lagrangian density and the energy–momentum tensor are defined as below [@Izadi_1; @Mota].
$$\mathcal{L} = \frac{1}{2\kappa}(\az + f(\az_{\mu\nu}\az^{\mu\nu})) + \mathcal{L}_m$$
$$\label{Tmunu}
T_{\mu\nu}=\rho u_\mu u_\nu + 2q _{(\mu} u_{\nu)} - p h_{\mu\nu} + \pi_{\mu\nu},$$
in which $u^\mu = \frac{dx^\mu}{d\tau} $ is the 4-velocity normalized as $ u^\mu u_\mu=1$, $ h_{\mu\nu} = g_{\mu\nu} - u_\mu u_\nu$ which determines the orthogonal metric properties of observers moving with 4−velocity $ u^\mu$, $ \pi_{\mu\nu}= h^\alpha_\mu h^\beta_\nu T_{\alpha \beta}$ is the projected symmetric trace free anisotropic pressure, $ \rho = T_{\mu\nu} u^\mu u^\nu$ is the relativistic energy density relative to $ u^\mu$, $ q_\mu = h^\alpha_\mu u^\beta T_{\beta\alpha}$ is the relativistic momentum density, and $ p = \frac{-1}{3} h^{\mu\nu}T_{\mu\nu}$ is the isotropic pressure. In a similar way $\az_{\mu\nu}$ can be written as $$\label{Rmunu}
\az_{\mu\nu} = \Delta u_\mu u_\nu + \Xi h_{\mu\nu} + 2 u_{(\mu}\gamma_{\nu)} + \varSigma_{\mu\nu}.$$ Putting and in the modified Einstein equation, four equations are obtained for all the unknown coefficients.
The relation between the affine connection and the Christoffel symbol of physical metric is: $$\Gamma^{\alpha}_{\mu\nu}={\alpha\brace \mu\nu}+ \gamma^{\alpha}_{\mu\nu}
\label{18}$$ where some of the above coefficients appears in $\gamma^{\alpha}_{\mu\nu}$ relation [@Izadi_1; @Mota]. As it mentioned in the introduction section, there are different aspects of speed of light with different origins in physics. They will be considered in the next subsection. But it should be noted that in the [@Izadi_1] for this modified theory $$\label{cST}
c_{ST} = \frac{c_0}{\sqrt{\omega}}.$$ In order to find in Palatini f($\az_{\mu\nu}\az^{\mu\nu}$) gravity one needs to know $\omega$ [@Izadi_1; @Mota]:
$$\omega=\frac{1+2F\Xi}{1+2F\Delta}.$$
F is the derivative of $f(\az^{\mu\nu}\az_{\mu\nu})$ with respect to $\az^{\mu\nu}\az_{\mu\nu}$ and $\Xi$ and $\Delta$ can be obtained from below equations:
$$\label{delta}\Delta+2F\Delta^2 -\frac{1}{2}(\Delta+3\Xi+f)=\kappa\rho c_E^2$$
$$\label{xi}\Xi+2F\Xi^2 -\frac{1}{2}(\Delta+3\Xi+f)=-\kappa p.$$
Here we consider $c_E=c_0$ so $\kappa=\frac{8 \pi G}{c_0^4}$. Putting the right values for $\rho $ and $p$, $\omega$ can be obtained in different cosmological eras. If we take a simple model $f(\az^{\mu\nu} \az_{\mu\nu}) = a_1 \az^{\mu\nu} \az_{\mu\nu}$,
$$\label{omega}
\frac{1}{\sqrt{\omega}} =
\begin{cases}
\sqrt{\frac{1+3\sqrt{3+16 \kappa a_1 \rho_r^0 c_0^2 (1+z)^4}}{5-\sqrt{3+16 \kappa a_1 \rho_r^0 c_0^2 (1+z)^4}} }& \text{radiation dominated era} \\
\sqrt{\frac{1+\kappa a_1 \rho_m^0 c_0^2 (1+z)^3}{1-\kappa a_1 \rho_m^0 c_0^2 (1+z)^3}} & \text{matter dominated era.}
\end{cases}$$
And for the present de Sitter era $\omega\rightarrow 1$.
In addition, as the variation of the speed of light comes from the variation of energy content of the universe (energy momentum tensor), for local tests which can be interpreted by Schwarzschild solution ($T_{\mu\nu}=0$), we have no variation. This means that our results are compatible with local varying speed of light tests such as the shift of the Mercury perihelion.
Local inertail frame
--------------------
As we emphasized before the main concern in the Palatini formalism is where the equivalence principle is true or in other words in which local frame there is no gravity. In a general case the space-–time structure is defined by both the metric (as the device for defining length) and the independent affine connection (as the device for parallel transporting) in the Palatini formalism. This clearly means that in any experiment that the concept of parallelism is important, the effects of the independent connection should be examined. In principle one is able to define two local frames in Palatini approach. In the first one, metric is locally Minkowskian, but the independent connection is not zero’ namely $g_{\mu\nu}=\eta_{\mu\nu}$, ${\alpha\brace \mu\nu}=0$ and $\Gamma^{\alpha}_{\mu\nu}=\gamma^{\alpha}_{\mu\nu}$. There is also a second local frame in which the connection is zero and thus: $g_{\mu\nu}\neq \eta_{\mu\nu}$, $\Gamma^{\alpha}_{\mu\nu}=0$ and ${\alpha\brace \mu\nu}=-\gamma^{\alpha}_{\mu\nu}$.\
In order to understand the physical significance of the above mentioned frames, let’s study the trajectory of a test particle in these two local frames. Hence we consider a dust with $T^{\mu\nu}=\rho u^{\mu} u^{\nu}$, where $\rho$ is the particle density and $u^{\mu}$ is the $4$–velocity. The energy–momentum conservation relation leads to the test particle trajectory. In the first local frame, the particle trajectory is a straight line $$\frac{d^2x^{\mu}}{ds^2}=0$$ while in the second frame one has $$\label{trajectory}
\frac{d^2 x^\mu}{ds^2}-\gamma^\mu_{\alpha\beta} \frac{dx^\alpha}{ds} \frac{dx^\beta}{ds}=0,$$ where the $\gamma^\mu_{\alpha\beta}$ is the difference of the affine connection and Christoffel symbol for both models. From the above equation we can see that the particle’s trajectory is not a special relativistic one, straight line, in the second local frame.\
Now the main question is that which one is the local inertial frame?\
It has to be noted that although in the first frame the particle moves on a straight line but this does not mean that equivalence principle is satisfied since gravity is present via the non–vanishing connection in other physical relations. So one expects to have changes in the geometrical concepts like geodesic deviation and Raychaudhuri’s equation representing how a flux of geodesics expands. In this theory the existence of two different connection fields has new consequences. The geodesics are determined by the Christoffel symbols (this choice is motivated by energy–momentum conservation) but the equation that governs the evolution of the deviation vector involves the affine connection (motivated by the fact that the covariant derivative or parallel propagation along any arbitrary curve is defined by the affine connection). In the case that the Christoffel symbols are not zero, matter fields and the test particle trajectory are not the special relativistic ones. For a general case, one can identify the extra terms appeared in the particle’s trajectory with the force due to the variation of the velocity of clock synchronization [@Izadi_1]. To see this, consider the case of Palatini $f(\az)$ gravity, in the second local frame (which is called the inertial frame). The observer at point 2 has to adjust its clock ahead with the amount $\sqrt{f'}d\ell/c_0$ (with $d\ell^2=f'|d\vec{x}|^2$). This shows that one should use the velocity $c_C=c_0/\sqrt{f'}$ for synchronization and as a result the test particle motion is given by: $$\frac{d^2x^\mu}{d\tau^2}+\frac{dx^\mu}{d\tau}\frac{1}{c_C}\frac{dc_C}{d\tau}-\eta^{\mu\nu}c_0^2\frac{\partial_\nu c_C}{c_C}=0$$ It is clear now that the extra terms are forces exerted on the particle because of the variation of the clock synchronization velocity. A similar discussion can be done for non-linear Ricci squared theories [@Izadi_1]. In standard theory the clock synchronization velocity coincides with the speed of light if we use the electromagnetic waves to do that.
In addition to the above discussions, we should notice the crucial point that the speed of light plays different key roles in physics and different aspects of the light speed may appear in various physical theories. It is a complicated synthesizer which has different distinctive origins, which result in coincident values in standard theory. As it was shown in [@Izadi_1], different aspects of light speed in $f(\az)$ and f($\az_{\mu\nu}\az^{\mu\nu}$) gravity became different from the amount $c_0$ ($\sim3\times10^8 m/{s^2}$), we normally consider for light speed in MKS units, in the . So whenever light speed appears in an equation, not only one should consider which kind of $c$ it is, but also the kind of extended gravity model must be known. For instance, is not the same in Palatini $f(\az)$ and f($\az_{\mu\nu}\az^{\mu\nu}$) gravity in the . A summary of these attempts is shown in table \[speedt\] [@Izadi_1]. According to the table \[speedt\] if $\omega$ is known then can be calculated in Palatini $f(\az_{\mu\nu}\az^{\mu\nu})$ gravity. Also, it is important to imply that according to its behavior in these modified theories, has been larger in the past than its present time value. This would be a convincing motivator to investigate how a cosmological equation would be modified considering has not always been $c_0$.
------------------- --------------------------- ------------------------------------------------
Palatini $f(\az)$ gravity Palatini f($\az_{\mu\nu}\az^{\mu\nu}$) gravity
\[0.5ex\] $c_{E}$ $c_{0}$ $c_{0}$
$c_{GW}$ $c_{0}$ $c_{0}$
$c_{EM}$ varying varying
$c_{0}$ varying$(\frac{1}{\sqrt{\omega}} c_{0})$
\[1ex\]
------------------- --------------------------- ------------------------------------------------
: \[speedt\]Comparison of different speeds of light in the Palatini $f(\az)$ and f($\az_{\mu\nu}\az^{\mu\nu}$) theories.
At first this controversial subject was considered in [@OlmoPRL]. Olmo considered quantum applications of Palatini formalism for theories in Schwarzschild solution and shows that the equivalence principle is violated for microscopic applications [@OlmoPRL; @Olmo_Hydrogen] in the local frame in which the metric is locally Minkowskian. He called this frame inertial. He considered the microscopic applications. However for macroscopic applications following the steps in [@OlmoPRL], one can show as the metric outside the mass is Schwarzschild, the geodesic equations would be the usual ones in Schwarzschild case.\
In macroscopic applications we meticulously reconsidered in [@Izadi_1] this important issue. In the foregoing paper the in Palatini formalism, in which the equivalence principle holds, was chosen wherever affine connection is zero in such a way that all the gravitational effects are vanishing from the equations and then two samples of extended gravity theories were studied. Of course if one considers a Schwarzschild metric for the case, this would turn into a straight line. The main goal of this work is to examine the behavior of fundamental constants in the in Palatini formalism for two well-known extended gravity theories, $f(\az)$ and $f(\az^{\mu\nu})$ models, and considering the right aspect of speed of light in the relations which should predict the observational data.
\[sec:redshift\] Changes in redshift
====================================
Redshift is a cosmological parameter which most of the cosmological equations are expressed in term of. As the goal of this study is to rewrite SN Ia equations and interprete them in in Palatini formalism considering different facets of the speed of light, here we tend to rewrite redshift and study the result. For incoming null ray [@Coles],
$$\label{fr}\int_{t}^{t_0}{\frac{cdt}{a(t)}}=-\int_{r}^{0} \frac{dr}{\sqrt{1-kr^2}}=f(r).$$
For a small time duration one can get
$$\int_{t+\delta t}^{t_0+\delta t_0}{c\frac{dt}{a(t)}}= \int_{t}^{t_0}{c\frac{dt}{a(t)}} \rightarrow c\frac{\delta t_0}{a_0(t)}=c\frac{\delta t}{a(t)},$$
and the conventional redshift is defined as
$$\label{z} \frac{\nu_0}{\nu}=\frac{a(t)}{a_0(t)}=\frac{1}{1+z},$$
where $ a_0 (t)$ is the present scale factor. On the other hand, it should be noticed that in above equations light speed is considered constant c. But as the definition of redshift comes from line element $ds^2$ and according to [@Ellis]
$$ds^2 = g_{\mu\nu} dx^{\mu}dx^{\nu},$$
where $g_{00} = -c_{ST}^2$. Equation must be rewritten as
$$\label{newfr}\int_{t}^{t_0}{\frac{c_{ST} dt}{a(t)}}=-\int_{r}^{0} \frac{dr}{\sqrt{1-kr^2}}=f(r).$$
Due to the fact that in this equation speed of light is coming from space-time description of the line element of Special Relativity [@Ellis], it is . Therefore if one distinguishes between different kinds of light speed, redshift would take the form
$$\label{zprime}
1+z^\prime \equiv \frac{\nu}{\nu_0}=\frac{c_{ST}}{c_0}(1+z).$$
It is obvious from the above equation that if in a formalism $c_{ST}=c_0$ then $z=z^{\prime}$. Otherwise, the redshift must be modified in each cosmological observation for the model which has been used. Specifically according to table \[speedt\], in both radiation and matter eras the above relation takes this form $(1+z^\prime)>(1+z)$ for $f(\az_{\mu\nu}\az^{\mu\nu})$ model.\
In addition, if one wants to replace the frequency with its corresponding wavelength, $c_{EM}$ would be needed as it is the velocity of any electromagnetic wave [@Ellis]. In other words,
$$\lambda=\frac{c_{EM}}{\nu} \nonumber$$
$$\frac{\lambda_0}{\lambda}=\frac{c_{ST}}{c_{EM}} (1+z),$$
in which $ c_{EM}$ is considered to approach $c_0$ in the present time.
Due to the modification of redshift, Hubble parameter will change as well. $H'$ has simple relation with $H$: $$\label{Hprime}
H'=H(1+\frac{1+z}{a_0} \frac{d}{dz}(Ln (c_{ST})).$$ As a result redshift and Hubble parameter should be modified in extended gravity theories in Palatini formalism, when one takes into account different facets of the speed of light.
\[sec:Friedmannology\]Extended Friedmannology
=============================================
Type Ia Supernovae luminosity distance
--------------------------------------
### Standard theory
Before we proceed further, let us review some general points. In standard cosmology, Luminosity distance $(d_L)$ is defined as [@review_on_DE]:
$$d_L = f(\chi)\sqrt{\frac{L_s}{L_o}},$$
in which $\sqrt{\frac{L_s}{L_o}} = (\frac{\Delta \nu_1}{\Delta \nu_0}) = (1+z)$ and $f(\chi) = \frac{c}{H_0\sqrt{\Omega^0_k}} sinh (\frac{\sqrt{\Omega^0_k}}{c} \int \frac{c dz}{E(z)})$. Therefore, $$\label{dl}d_L=\frac{c}{H_0\sqrt{\Omega_k^0}}sinh(\sqrt{\Omega_k^0}\int_{0}^{z}{\frac{dz}{E(z)}})(1+z),$$ where density parameter is defined as$\Omega_k^0=\frac{-kc^2}{(a_0H_0)^2}$ and $k$ shows the curvature of the universe (open ($k=-1$), flat ($k=0$) or closed ($k=+1$)). Also considering a universe without dark energy, the Hubble parameter can be shown by
$$H^2=H_0^2 (\Omega_m^0(1+z)^3 + \Omega_r^0(1+z)^4 + \Omega_k^0 (1+z)^2).$$
Here $H_0$ is the Hubble constant at the present time, $\Omega_m^0$, $\Omega_r^0$, $\Omega_k^0$ are matter, radiation and curvature density parameters at the present time respectively.
Without considering dark energy, expanding the integral in results in $$d_L=\frac{c}{H_0}[z+\frac{1}{4}(1+\Omega_k^0)z^2+...].$$ Willing to consider the presence of dark energy, expanding the integral in results in [@Shinji] $$\label{usual_dL} d_L=\frac{c}{H_0}[z+\frac{1}{4}(1-3\omega_D\Omega_D^0+\Omega_k^0)z^2+...].$$
In order to explain the discrepancy of SN Ia luminosity distance, between theory and observations, this equation has to have a term like $-3\omega_D\Omega_D^0$. This extra term comes from observations and it can be eliminated or less than this amount if a suitable theory is being proposed.\
Furthermore from observational point of view, is considered as [@Coles]
$$\label{dL_observation}
d_L = 10^{1+\mu/5} pc,$$
in which $\mu= m-M $ is the distance modulus, m is the apparent magnitude and M is the absolute magnitude. Also, as all equations are considered to be in m.k.s units here, $1 pc \simeq 3.08 \times 10^{16} m$.
### The effect of considering Palatini formalism on Supernovae type Ia data
Here we are willing to put the right velocities according to their origins in the corresponding notions. Rewriting the analogous relations, the first step is to specify which speeds are involved. For Palatini formalism, $c$ gives its place to in some equations; because when it refers to calculating lengths, one has to use the concept of space-time. In fact, if one is looking for an operational technique for measuring distances, using radar demands the modification in the line element, since the space-time velocity is varying in the in Palatini formalism [@Izadi_2]. What will be appeared in the line element is , therefore the mentioned equations will get the forms $\sqrt{\frac{L_s}{L_o}} = (\frac{\Delta \nu_1}{\Delta \nu_0}) = (1+z')$ and $ f(\chi) = \frac{c_{ST}}{H'_0\sqrt{\Omega^0_k}} sinh (\frac{\sqrt{\Omega^0_k}}{c_{ST}} \int \frac{c_{ST} dz'}{E'(z')}) $.\
As a result,
$$\label{newdl} d_L=\frac{c_{ST}}{H_0^\prime\sqrt{\Omega_k^0}}sinh[\frac{\sqrt{\Omega_k^0}}{c_{ST}}\int_{0}^{z'}{c_{ST}\frac{d {z'}}{E'({z'})}}](1+z^\prime),$$
where $ E'(z')=\frac{H'}{H'_0}$ and $H'$ is introduced in . Willing to consider low redshift region and using $sinh x$ expansion, one gets to
$$d_L=(1+z^\prime)\int_{0}^{z'}{c_{ST}\frac{d {z'}}{H'}}.$$
Palatini Lg gravity
-------------------
As it is shown in table \[speedt\], in the is the same as $c_0$ in Palatini $f(\az)$ gravity. Hence rewriting equations in this model results in no change.
Palatini Lg gravity
-------------------
According to and also table \[speedt\], considering $f(\az_{\mu\nu}\az^{\mu\nu})$ model in Palatini formalism redshift would be:
$$\begin{aligned}
1+z^\prime &=& \frac{c_{ST}}{c_0}(1+z) \nonumber\\
&=&\frac{1}{\sqrt{\omega}}(1+z). \end{aligned}$$
According to it is obvious that is greater in the past. Consequently, $$\label{zrelation}
\frac{1+z^\prime}{1+z}=\frac{1}{\sqrt{\omega}}=\begin{cases}
\sqrt{\frac{1+3\sqrt{3+16 \kappa a_1 \rho_r^0 c_0^2 (1+z)^4}}{5-\sqrt{3+16 \kappa a_1 \rho_r^0 c_0^2 (1+z)^4}} } & \text{radiation dominated era} \\
\sqrt{\frac{1+\kappa a_1 \rho_m^0 c_0^2 (1+z)^3}{1-\kappa a_1 \rho_m^0 c_0^2 (1+z)^3}} & \text{matter dominated era.}
\end{cases}$$ This means that in Palatini f($\az_{\mu\nu}\az^{\mu\nu}$) gravity, interpreting redshift of a signal in leads in a greater value. ($z^\prime>z$)
Moreover, according to equation , the luminosity distance would be rewritten in the form of:
$$\label{palatinidl}d_L=\frac{c_0}{\sqrt{\omega}H'_0\sqrt{\Omega_k^0}}sinh[\sqrt{\omega}\sqrt{\Omega_k^0}\int_{0}^{z'}{\frac{1}{\sqrt{\omega}}\frac{dz'}{E'(z')}}](1+z').$$
It can be concluded here that rewriting the luminosity distance in this way, leads in some changes. The exact difference can be specified by putting the right numerical values and a suitable model (see section \[sec:proposed\_model\]). These results are indicative of larger luminosity distance for Palatini compared to the Einstein de Sitter model and they emphasize on the power of Palatini formalism. The price of this achievement is varying some aspects of speed of light in the of extended gravity theories in Palatini formalism.
\[sec:proposed\_model\] Proposed model to have the most consistent modified luminosity distance for SN Ia
=========================================================================================================
Here the main goal is to fit a model in a way that fits the most with the concordance model ($\Omega^0_{\Lambda}=0.6825$ and $\Omega^0_{m}=0.3175$ [@Planck]) In what follows, a suitable model is being proposed in order to check the effect of varying on SN Ia Luminosity distance.
Considering flat universe in low redshift region, equation becomes:
$$\label{modeldl}
d_L=(1+z')\int_{0}^{z'}c_{ST}\frac{dz'}{H'},$$
in which , $z'$ and $H'$ are defined in , and .
If we choose a linear form of $f(\az_{\mu\nu}\az^{\mu\nu})=a_1 \az_{\mu\nu}\az^{\mu\nu}$ as [@Izadi_1], $a_1$ whose dimension is length squared, can be used as the model’s degree of freedom. Choosing $$\label{a1_model}
a_1 = \Lambda^\varepsilon l_p^{2(1+\varepsilon)},$$ $\varepsilon$ can be set by the observational data. In this equation, $\Lambda$ is the cosmological constant and $l_p$ is the Planck length.
Studying more than 650 SN Ia (data from [@website:union] ) in $0<z<1$ region, gives a more or less similar value for $\varepsilon$, ($\varepsilon\sim-0.985$) in order to have a modified Luminosity distance which is more compatible with the observations compared to the Einstein de-Sitter model. In fact, taking model in Palatini formalism leads to larger luminosity distance with respect to metric formalism, and gets closer to the observational data. In figure \[fig\_dL\] is plotted for different values of $\varepsilon$ and is compared to models both with and without dark energy.
![\[fig\_dL\] The luminosity distance vs. redshift for i) observations, using union 2 data for the distance modulus ($\mu$) [@website:union]. ii) Einstein de- Sitter model, no dark energy is considered. iii) concordance model, (considering dark energy $\Omega^0_{\Lambda}=0.6825$ [@Planck]) iv) Modified gravity in Palatini formalism considering the proposed model in equation and also distinguishing from $c_0$ regarding the results in [@Izadi_1]. As it is clear from the figure, $-0.987<\varepsilon<-0.98$ has the best fitting with the concordance model. Compared to Einstein de-Sitter model, the modified gravity model in Palatini is more consistent with the concordance model. In all models, the universe is considered to be flat. ](fig1.jpg)
It is obvious from figure \[fig\_dL\] that the line with $\varepsilon\sim-0.985$ is the closest to the concordance model (in which $\Omega^0_{\Lambda}=0.6825$ and $\Omega^0_{m}=0.3175$ [@Planck]). This leads to less need of dark energy for explaining the inconsistency between observations and theory for . Of course by testing this model on more cosmological equations, $\varepsilon$ can be fitted more precisely (future work).\
Using the different values for $\varepsilon$, and $z'$ are plotted in figures (\[cST\_for\_985\]) and (\[zprime\_for\_985\]). Obviously for different values of $\varepsilon$, one gets different plots.
![\[cST\_for\_985\]Space time speed of light vs. redshift. in Palatini formalism for model, increases with the redshift and for the present time it approaches to $c_0$ as for today’s deSitter universe in $z=0$, $c_{ST}=c_0$. ](fig2.jpg)
![\[zprime\_for\_985\]Modified redshift vs. redshift. Considering modified gravity in Palatini formalism makes a shift in the value of the redshift. Considering the right value of $\varepsilon$ regarding SN Ia luminosity distance, this graph is plotted for $\varepsilon=-0.985$. This modification affects so many observational data. Of course this value of $z$ is suitable for $0<z<1$ region and for higher redshift regions, more tests should be done (future work).](fig3.jpg)
Conclusion
==========
The speed of light plays a synthesizer role in physics. Various aspects of speed of light appear in different physical theories with different origins [@Ellis]. In standard theory the value of all these aspects such as , $c_{EM}$, $c_E$ and $c_{GW}$ coincide; i.e. $c_{ST} = c_E = c_{GW} = c_{EM} = c_0$. However considering modified gravity in Palatini formalism, all these aspects may not coincide at all. In fact in the in which equivalence principle is satisfied, several forms of light speed such as , $c_{EM}$, $c_E$ and $c_{GW}$ become different from each other considering modified gravity in Palatini formalism [@Izadi_1]. Therefore unlike the standard case, one cannot use $c_0$ as the speed of light in all theories. So in the determination of distance which needs both the measurement of time and a signal going from one point to another, it is important to distinguish between and $c_{EM}$. All in all, for some modified gravity models in Palatini formalism, may be varying in the and one should consider the effects of this change on the analysis of cosmological data such as the ones referring to dark energy.
For the redshift, rewriting equations leads to a new redshift which is indeed more than the usual one. This may considerably affect analysing observational data in which redshift plays an important role. Hence using modifications above, redshift will have a larger value being interpreted in our choice of .
Finally, the SN Ia luminosity distance has been changed in Palatini formalism. The modification is shown in equation . Also a consistent model has been proposed. Within this model, there is a degree of freedom and choosing the right value for it, can help in fitting the modified luminosity distance to cosmological observations. It is shown in figure (\[fig\_dL\]) that if one chooses the right value for $\varepsilon$, the luminosity distance for the proposed model would fit better to the concordance model in $z \sim 1$ and there would be less need to dark energy. One should also notice that for the local tests ($z=0$) using the Schwarzschild metric instead of FRW, gets the value $c_0$. In this model, the SN Ia modifications has been studied for the $0<z<1$ range. Studying these effects in larger redshifts needs more details and considerations. These considerations and also more tests such as the shift parameter of CMB and Baryonic Acoustic Oscillations are under study so that $\varepsilon$ can be fitted with the observations the most.\
The modifications are listed in table (\[alldata\]).
Metric formalism Palatini gravity (model)
---------- --------------------------------------------------------------------- ----------------------------------------------------------------------------------------------
redshift $ z $ $1+z^\prime=\frac{c_{ST}}{c_0}(1+z)$
$H$ $ H_0 (1+z)^{3/2} $ $H(1+\frac{d lnc_{ST}}{dz}(1+z))$
$d_L$ $ \frac{c}{H_0\sqrt{\Omega_k^0}} $ $\frac{c_{ST}}{H'_0\sqrt{\Omega_k^0}}$
$\times sinh[\sqrt{\Omega_k^0}\int_{0}^{z'}{\frac{dz}{E(z)}}](1+z)$ $\times sinh[\frac{\sqrt{\Omega_k^0}}{c_{ST}}\int_{0}^{z'}{c_{ST}\frac{dz'}{E'(z')}}](1+z')$
: \[alldata\] Several observational data in two different formalisms: metric and Palatini. The standard Hubble constant $H$ is considered for Einstein de-Sitter model.
**Acknowledgments**: The authors are grateful to Robi Banerjee and Ali Shojai for their outstanding comments and also helpful discussions. This work is supported by K. N. Toosi University of Technology.
[100]{}
S. Perlmutter, G. Aldering, G. Goldhaber, R. A. Knop, P. Nugent, P. G. Castro, S. Deustua, S. Fabbro, A. Goobar, D. E. Groom, I. M. Hook, A. G. Kim, M. Y. Kim, J. C. Lee, N. J. Nunes, R. Pain, C. R. Pennypacker, R. Quimby, C. Lidman, R. S. Ellis, M. Irwin, R. G. McMahon, P. Ruiz-Lapuente, N. Walton, B. Schaefer, B. J. Boyle, A. V. Filippenko, T. Matheson, A. S. Fruchter, N. Panagia, H. J. M. Newberg, W. J. Couch, and T. S. C. Project. Measurements of Ω and Λ from 42 High-Redshift Supernovae. Astrophys.J., 517:565–586, June 1999.
A. G. Riess, A. V. Filippenko, P. Challis, A. Clocchiatti, A. Diercks, P. M. Garnavich, R. L. Gilliland, C. J. Hogan, S. Jha, R. P. Kirshner, B. Leibundgut, M. M. Phillips, D. Reiss, B. P. Schmidt, R. A. Schommer, R. C. Smith, J. Spyromilio, C. Stubbs, N. B. Suntzeff, and J. Tonry. Observational Evidence from Supernovae for an Accelerating Universe and a Cosmological Constant. Astron.J., 116:1009–1038, September 1998.
L. Amendola and S. Tsujikawa. Dark Energy: Theory and Observations. 2010.
M. Bastero-Gil, M. Borunda, and B. Janssen. The Palatini formalism for higher-curvature gravity theories. In K. E. Kunze, M. Mars, and M. A. V ́azquez-Mozo, editors, American Institute of Physics Conference Series, volume 1122 of American Institute of Physics Conference Series, pages 189–192, May 2009.
J.-P. Uzan. Variation of the Constants of Nature in the Early and Late Universe. In C. J. A. P. Martins, P. P. Avelino, M. S. Costa, K. Mack, M. F. Mota, and M. Parry, editors, Phi in the Sky: The Quest for Cosmological Scalar Fields, volume 736 of American Institute of Physics Conference Series, pages 3–20, November 2004.
G. J. Olmo. Palatini Approach Beyond Einstein’s Gravity. ArXiv e-prints, December 2011.
G. J. Olmo. Palatini Approach to Modified Gravity:. f(R) Theories and Beyond. International Journal of Modern Physics D, 20:413–462, 2011.
T. P. Sotiriou and V. Faraoni. f(R) theories of gravity. Reviews of Modern Physics, 82:451–497, January 2010.
M. Ostrogradski, Mem. Ac. St. Petersbourg VI 4 385, 1850
S. Fay, R. Tavakol, and S. Tsujikawa. f(R) gravity theories in Palatini formalism: Cosmological dynamics and observational constraints. Phys. Rev. D, 75(6):063509, March 2007.
G. J. Olmo, H. Sanchis-Alepuz, and S. Tripathi. Stellar structure equations in extended Palatini gravity. Phys. Rev. D, 86(10):104039, November 2012.
K. Enqvist, H. J. Nyrhinen, and T. Koivisto. Binary systems in Palatini f(R) gravity. Phys. Rev. D, 88(10):104008, November 2013.
A. Izadi and A. Shojai. The speed of light in extended gravity theories. Classical and Quantum Gravity, 26(19):195006, October 2009.
G. J. Olmo. Violation of the Equivalence Principle in Modified Theories of Gravity. Physical Review Letters, 98(6):061101, February 2007.
S. Capozziello and G. Lambiase. Open problems in gravitational physics. ArXiv e-prints, September 2014.
G. F. R. Ellis and J.-P. Uzan. c is the speed of light, isn’t it? American Journal of Physics, 73:240–247, March 2005.
A. Izadi and A. Shojai. Measurement of the space-time interval in modified gravity theories in Palatini formalism. General Relativity and Gravitation, 45:229–241, January 2013.
B. Li, J. D. Barrow, and D. F. Mota. Cosmology of Ricci-tensor-squared gravity in the Palatini variational approach. Phys. Rev. D, 76(10):104047, November 2007.
G. J. Olmo. Hydrogen atom in Palatini theories of gravity. Phys. Rev. D, 77(8):084021, April 2008.
P. Coles and F. Lucchin. Cosmology: The Origin and Evolution of Cosmic Structure, Second Edition. July 2002.
S. Tsujikawa. Dark energy: investigation and modeling. ArXiv e-prints, April 2010.
Planck Collaboration, P. A. R. Ade, N. Aghanim, C. Armitage-Caplan, M. Arnaud, M. Ashdown, F. Atrio-Barandela, J. Aumont, C. Baccigalupi, A. J. Banday, and et al. Planck 2013 results. XVI. Cosmological parameters. Astron.Astrophys, 571:A16, November 2014.
Supernova cosmology project.
| {
"pile_set_name": "ArXiv"
} |
---
address:
- '$^a$Max-Planck-Institut für Festkörperforschung, Heisenbergstr.1, D-70569 Stuttgart, FRG.'
- '$^b$Geophysical Laboratory, Carnegie Institution of Washington, 5251 Broad Branch Rd., NW, Washington, DC 20015. '
author:
- 'A.I.Liechtenstein$^a$, I.I. Mazin$^{a,b}$, and O.K. Andersen$^a$'
title: '$s$-wave superconductivity from antiferromagnetic spin-fluctuation model for bilayer materials'
---
The currently most exciting discussion about high-$T_c$ superconductivity deals with the symmetry of the pairing state[@Nature]. Intimately related to this, is the question of whether the superconductivity is due to antiferromagnetic spin fluctuations (see e.g. Monthoux and Pines (MP) Ref. [@MP], and also Refs. [@scal; @bickers]), to electron-phonon (EP) interaction enhanced by inter-layer pair tunnelling[@Science], or to neither of two. In this discussion, it is indirectly assumed that the antiferromagnetic spin-fluctuation (AFSF) mechanism necessarily leads to d-wave pairing, and that the AFSF and EP mechanisms cannot coexist.
In this Letter we point out that, whereas the AFSF mechanism leads to $d$-pairing for one layer, it may lead to (two-dimensional) $s$-symmetry for a bilayer. The condition for that is existence of strong antiferromagnetic correlations between the two layers in a bilayer, as found experimentally in YBa$_2$Cu$_3$O$_7$ [@tranq; @mook; @rossat]. We find that, for a given coupling strength, $T_c(s)$ is about twice as high as $T_c(d)$ thus making it easier to achieve the observed values of $T_c\sim 100$K. Essential for the positive influence of layer-doubling is the [*single*]{}-particle tunneling which splits the one-electron plane-bands into even and odd with respect to the mirror-plane between the layers. In this aspect our mechanism is very different from the interlayer [*pair*]{}-tunnelling (IPT) mechanism discussed by Anderson et al.[@Science]. Nevertheless, similar to the IPT model, any attractive interaction between electrons in the same band, such as the one mediated by even (Raman-active) phonons, enhances $T_c$. This is opposite to the previously considered single-layer AFSF models in which such interactions are mutually destructive.
Support for such an enhancement mechanism may be found in the experimental fact (e.g. Refs.[@T*mwave; @T*tunn]) that some members of the cuprate family (Nd$_{2-x}$Ce$_x$CuO$_4$, HgBa$_2$CuO$_4$) behave as conventional $s$-wave EP superconductors. MP AFSF theory, on the other hand, would have to imply principally different mechanisms for this compound and for those with high $T_c$’s. Another experimental fact which suggests a constructive interplay between phonon- and non-phonon mechanisms is that in YBa$_2$Cu$_3$O$_7,$ the isotope effect increases smoothly when the superconductivity is suppressed [@Frank]. Finally, the most impressive argument is that in all high-$T_c$ materials $T_c$ is [*anticorrelated*]{} with the in-plane antiferromagnetic correlation length $\xi $. In particular, in YBa$_2$Cu$_3$O$_{7.0}$, $\xi $ is about one lattice parameter, which would make the single-layer AFSF superconductivity virtually inoperative. To the contrary, as we shall argue below, the proposed bilayer model is barely sensitive to the in-plane AF correlation length $\xi $ at all.
In the following we shall assume a conventional picture in the sense that the one-electron tunneling between the planes is allowed both in the normal and in the superconducting states. This is in contradiction with the IPT scenario[@Science], but in agreement with some photoemission experiments[@photo]. In this case, the single-particle eigenstates for a bilayer are the even $|+,{\bf k}\rangle $ and odd $|-,{\bf k}\rangle $ combinations of the individual plane states and ${\bf k}$ is the 2D Bloch-vector. The properties of the even and odd bands are discussed in detail in Ref.[@TB8], but, for the purpose of comparison with the MP model, we use the same band model as they did. We neglect completely the $k_z$ dispersion due to small intercell $c$-hopping, which can lead to interesting effects (see, e.g., Ref.[@klem]) but which are however beyound the scope of this Letter. Accordingly, in the following the term “bands” always means “two-dimensional bands”. As regards the interplane hopping inside the unit cell, $t_{\perp
}\left( {\bf k}\right)$, we assume that it is sufficiently large to set even and odd symmetry of the two-dimensional bands, but we neglect, for simplicity, in the following numerical calculations the even-odd splitting $
\epsilon _{-}\left( {\bf k}\right) -\epsilon_{+}\left( {\bf k}\right) =
2t_{\perp}({\bf k})$.
The generalization of the MP AFSF model to two-bands is straightforward; one has only to take into account that the effective vertex for scattering of an electron from band $i$ to band $j$ by a spin-fluctuation depends on $%
i,j$, while the spectrum of the fluctuations $\chi $ is the same as in MP. Then, Eqs. (6-8) of MP become$$\begin{aligned}
\Sigma _{ij}\left( {\bf k},i\omega _n\right)&=&T\sum_{{\bf q}m}
\sum_{kl}V_{ik,lj} \left(
{\bf k-q},i\omega _n-i\omega _m\right) \nonumber \\
&\times &G_{kl}\left( {\bf q},i\omega _m\right)
\nonumber \\
G_{ij}^{-1}\left( {\bf k},i\omega _n\right)& =&\left[ i\omega
_n-\epsilon \left( {\bf k}\right) +\mu \right] \delta _{ij}-\Sigma
_{ij}\left( {\bf k},i\omega _n\right)\nonumber\\
\Phi _{ij}\left( {\bf k},i\omega _n\right)& =&-T\sum_{{\bf q}m}\sum_{klst}V
_{ik,tj}\left( {\bf k-q},i\omega _n-i\omega _m\right)
\label{MP}\\
\times &G&_{kl}\left( -{\bf q},-i\omega _m\right)\Phi
_{ls}\left( {\bf q},i\omega _m\right)G_{st}\left( {\bf q},i\omega
_m\right) \nonumber\end{aligned}$$ where $\Sigma $ and $\Phi $ are respectively the normal and anomalous self-energies, $G$ is the single-particle Green function, $\epsilon $ the bare electron energy, and $\sum_{{\bf q}m}$ denotes the [*average*]{} over the Brillouin zone plus the sum over the Matsubara frequencies. $V$ is AFSF pairing interaction, determined by the exchange interaction of electrons with the AFSF’s, $Vij,kl=\int {d{\bf R}d{\bf R}^{\prime }} \sum_{\alpha
\beta \gamma \delta } \langle i\alpha |J({\bf r-R})\sigma _{\alpha \beta
}|j\beta \rangle \\ \times \tilde \chi ({\bf R-R}^{\prime })\langle k\gamma
|J({\bf r-R})\sigma _{\gamma \delta }|l\delta \rangle $ where $J$ is exchange interaction and $\tilde \chi =\langle {\bf S(R)S(R}^{\prime
})\rangle $ is spin-spin correlation function. For a bilayer, one can let [**R**]{} be a two-dimensional vector and introduce $\tilde \chi =\chi {\bf %
(R-R}^{\prime })I_{uv}$, where $u,v=1,2$ label layers, and $I$ accounts for interplane correlations, if any. Then the function $\chi $ is the same as in MP.
The key to the our bilayer AFSF model is the experimental fact that the spin fluctuations in the bilayer of YBa$_2$Cu$_3$O$_{7-x}$ are always antiferromagnetically correlated between the planes [@tranq; @mooK; @rossat]. Even fully oxygenated samples, where the in-plane correlation length is already of the order of the lattice parameter, show nearly perfect interlayer correlation[@rossat]. The exchange potential set up by such a spin-fluctuation is therefore [*odd*]{} with respect to the mid-layer mirror plane and, correspondingly, couples exclusively [*even and odd*]{} electron states (but neither odd to odd, nor even to even). In other words, $I_{u\neq v}=-I_{uu}=-1.$ In this case, after summation over $u,v$ in the expression for $V_{ij,kl}$ and defining the appropriate coupling constant $g$, Eqs. \[MP\] become:$$\begin{aligned}
\Sigma _{-}\left( {\bf k},i\omega _n\right)&=&Tg^2\sum_{{\bf q}m}\chi \left(
{\bf k-q},i\omega _n-i\omega _m\right)
G_{+}\left( {\bf q},i\omega _m\right) \nonumber
\\
G_{+}^{-1}\left( {\bf k},i\omega _n\right)&=& i\omega
_n-\epsilon \left( {\bf k}\right) +\mu -\Sigma
_{+}\left( {\bf k},i\omega _n\right) \nonumber \\
\Phi _{+}\left( {\bf k},i\omega _n\right)&=&-Tg^2\sum_{{\bf q}m}\chi
\left( {\bf k-q},i\omega _n-i\omega _m\right) \label{MP1}\\
&\times &G_{-}\left( -{\bf q},-i\omega _m\right) \Phi
_{-}\left( {\bf q},i\omega _m\right) G_{-}\left( {\bf q},i\omega
_m\right) \nonumber \end{aligned}$$ and the same with $+$ and $-$ subscripts interchanged, and $g$ and $\chi $ are the same as in MP.
For reasons of symmetry, the solution of these equations must have the form $%
G_{+}=G_{-}$, $\Phi _{+}=\pm \Phi _{-}$. For the upper choice of the sign, the Eqs. (\[MP1\]) reduce precisely to the original MP pairing state. For the lower choice, the Eqs. (\[MP1\]) again reduce to the one-plane case, [*but now the interaction in the equation for* ]{}$\Phi $[* is effectively attractive*]{}. In other words, now the order parameter has the opposite sign in the two bands, and therefore the last Eq. in (\[MP1\]) can be rewritten in terms of $|\Phi |$, and with plus instead of minus on the right-hand side.
The concept of a superconducting state where two distinctive bands had the order parameters of the opposite signs was first discussed in 1973 in connection with semimetals[@AS]. More recently, in a two-layer Hubbard model, such a solution was found by Bulut et al[@Scalap-PRB] (which they labeled as “$d_z$” state) and in the conventional superconductivity theory[@we], where it appears in case of strongly anisotropic electron-phonon and/or Coulomb interaction, or because of a strong interband scattering by magnetic impurities. In all cases, order parameter has $s$-symmetry inside each band and changes sign between the bands.
From Eqs. (\[MP1\]) it is quite plausible that such an instability is stronger than the $d_{x^2-y^2}$ one, and will occur at a higher $T_c$. Below we shall prove this numerically, but before going to numerical results, it is instructive to get a conceptual understanding about these two different solutions. The physical reason for having $d$-symmetry in the one-plane case, is that the AFSF interaction makes pairing energetically favorable only when it couples parts of the Fermi surface which have opposite signs of the order parameter[@?]. In Y123 the AFSF interaction is peaked at [**Q**]{}=($\pi /a,\pi /a)$. The shape of the Fermi surface is such that the condition is satisfied only for $d_{x^2-y^2}$ symmetry. On the other hand, the small-[**q**]{} interaction couples parts of the Fermi surface where the order parameter has the same sign. This makes pairing unfavorable. Since $%
\chi ({\bf q}\approx {\bf Q})\gg \chi ({\bf q}\approx 0)$, nevertheless, more is lost by making an $s$-state than by making a $d$-state (which has been found numerically by MP), because the latter loss is the difference between the small-$q$ loss and the large-$q$ gain, while in an $s$-state one loses over the whole Fermi surface[@spin-bag].
Now, coming to the bilayer case, we observe that there is no conflict between the small and the large $q$’s any more. The AFSF interaction spans two different sheets of the Fermi surface, which always have order parameters of the opposite signs. Thus the AFSF interaction is as attractive for $s$-pairing in a bilayer as it is repulsive in a single plane, and consequently more attractive than $d$-pairing in a single plane. Of course, the resulting $s$-state is likely to be highly anisotropic, to take better advantage of the large $\chi ({\bf q})$ at ${\bf q}\approx {\bf Q}$. This is similar to Anderson’s model[@Science]. To demonstrate this effect numerically, we have solved Eqs.\[MP1\] with the parameters from MP paper, and using the same numerical technique. As expected, the maximal eigenvalue of the last Eq.\[MP1\] is larger than that for the MP $d$-pairing (about 1.5 compared to 1). Fig.\[tc\] shows the plot of $T_c$ as a function of the interaction constant $g$ for both cases. To test the numerics, we have also solved the original MP equations and obtained the similar results as MP.
From Fig.\[tc\], one immediately observes that the value (0.69 eV) of the coupling constant $g$ which yields $T_c\approx 90$K for two planes and $s$-symmetry, is much smaller than the corresponding value (1.24 eV) for one plane and $d$-symmetry. Actually, $T_c(s)\sim 2T_c(d)$ for $g$ up to about 1 eV. At stronger couplings, $T_c(d)$ saturates faster than $T_c(s)$ due to stronger effect of mass renormalization. Similarly, as we shall see below, the ratio of the maximal gap to $T_c$ tends to be larger for the one-plane model, for the same $T_c.$
One can also obtain the self-consistent solution for $\Phi $ at $T\ll T_c$. To do that, one has to include higher-order terms (see, e.g., Ref. [@scal]). In this case one of the Green functions in the Eq.\[MP1\] for $%
\Phi $ should be replaced by:$$\begin{aligned}
\tilde G_-^{-1}({\bf k},i\omega _n)&=&i\omega _n-(\epsilon _{{\bf k}}-\mu
)-\Sigma _-({\bf k},i\omega _n),
\label{MP2} \\
G_-^{-1}({\bf k},i\omega _n)&=&\tilde G_-^{-1}({\bf k},i\omega _n) \nonumber\\
&-&\Phi _-({\bf k},i\omega _n)\tilde G_-(-{\bf k},-i\omega _n)
\Phi _-(-{\bf q},-i\omega_n)]. \nonumber\end{aligned}$$ This new set of equations can be solved iteratively, starting with $G=\tilde
G$ (which is correct to first order in $\Phi )$. The actual solution for $%
T=T_c/2$, shown in Fig.\[gaps\], was achieved by making two iterations of Eqs.\[MP2\]. The frequency-dependent superconducting gap is related to $%
\Phi $ as$$\Delta ({\bf p},i\omega _n)=\frac{\Phi ({\bf p},i\omega _n)}{Z({\bf p}%
,i\omega _n)}=\frac{\Phi ({\bf p},i\omega _n)}{1-Im\Sigma ({\bf p},i\omega
_n)/\omega _n}.$$ From Fig.\[gaps\] we observe that the [*absolute value* ]{}of $\Delta $ behaves similarly in both cases, having a minimum along (11) directions and a maximum along (10) directions. Furthermore, $|\Delta |$’s in both cases differ by less than 10% on two-thirds of the whole Fermi surface, thus making it extremely difficult to distinguish between the two in an experiment which does not probe the relative phases of $\Delta .$ In other words, the order parameter, formally having $s-$symmetry, is still strongly anisotropic, but nodeless.
Now we shall briefly discuss some experimental consequences of the bilayer AFSF superconductivity model. It turns out that many difficulties associated with the original AFSF superconductivity model disappear in the present version.
\(1) In-plane vs. perpendicular-to-the-planes Josephson tunneling. Recent searches for the $d$-pairing in YBa$_2$Cu$_3$O$_7$ (Refs.[@illi; @dynes] and others) still do not give a definite answer. Experiments probing the angular dependence of the order parameter, as well the existence of the so-called “paramagnetic Meissner effect”[@par-M], indicate the existence of order parameters of opposite signs; Many experiments were interpreted in terms of $d_{x^2-y^2}$, but such an interpretations can be questioned because of presence of the chain band[@we]. On the other hand, the existence of the finite tunneling current perpendicular to the planes[@dynes] is incompatible with $d$-pairing, but compatible with our model once the simplifying assumption $\epsilon _{-}\left( {\bf k}\right)
=\epsilon _{+}\left( {\bf k}\right) $ is removed[@intra].
\(2) The original MP model is very sensitive to the in-plane correlation length $\xi $: According to Ref.[@MP], $T_c$ drops from 90K to 35K when $%
\xi $ is reduced from 2.3$a$ to $a.$ Experimentally, in fully oxygenated samples (O$_{6.9-7.0}$) $\xi $ is small and it still decreases closer to the O$_7$-composition where it becomes less than $a$. $T_c$ is, however, not sensitive to the oxygen content in this regime. In our model and contrary to MP, $T_c$ is hardly sensitive to the sharpness of $\chi (q)$, which predominantly influences the gap anisotropy.
\(3) One of the arguments in favor of the AFSF mechanism has been the strong $%
T_c$ suppression upon Zn doping. However, it has remained unclear why chain disorder hardly affects $T_c.$ (One could argue that the chains are completely decoupled from the planes, but this is inconsistent with the strong effect on the Ba $A_{1g}$ phonon mode of the onset of superconductivity). This finds natural explanation in our AFSF model: The chain impurity potential is even with respect to the mid-plane reflection and does therefore not produce scattering between the even and odd bands.
\(4) The MP model has difficulty in explaining the continuous change of the isotope effect with oxygen content, as well as in reconciling the apparent phonon $s-$wave superconductivity in Nd-cuprate with the assumed AFSF superconductivity in YBa$_2$Cu$_3$O$_7.$ The basics of this conflict is as follows: The AFSF interaction is pairing when a pair changes the sign of its order parameter upon scattering; this is the case MP model for ${\bf q}%
\approx (1,1)\pi /a$. It is [*de*]{}pairing if there is no sign change. The opposite is true for the electron-phonon interaction. Obviously, the only way to reconcile the MP model with the known facts about the role of phonons in superconductivity is to assume that the electron-phonon interaction, contrary to the AFSF interaction, is strong for ${\bf q}\rightarrow 0$ and weak for large $q$’s; this is opposite to common wisdom. In our model the corresponding assumption is much less painful: One has to assume that the [*even*]{} phonons, like for instance $A_{1g}$ Raman-active phonons, interact with electrons stronger than the [*odd*]{} ones. This seems quite plausible and some indirect arguments can be given in support of this assumption (e.g., the even phonons strongly influence the extended van Hove singularities[@TB8].
In conclusion, we have shown that from the observed strong antiferromagnetic correlations between the Cu-O planes in bilayer materials the strongest superconducting instability due to antiferromagnetic spin fluctuations appears in the anisotropic $s$-channel, but so that the order parameters in the bonding (symmetric) and antibonding (antisymmetric) bands have opposite signs. This helps to reconcile the magnetic-induced superconductivity model with many experiments which previously seemed to contradict the magnetic scenario. In particular, the interrelation between the doping dependencies of the magnetic and the superconducting properties can be much easier understood.
We want to thank O. Gunnarsson for many helpful discussions, especially regarding the relation of this model to the two-band Hubbard model.
P.W. Anderson, Physics Today, [**3**]{}, 11(1994); D. Pines, [*ibid.,* ]{} p. 13; D. Scalapino, [*ibid.,*]{} p120.
P. Monthoux and D. Pines, Phys. Rev. B [**49**]{}, 4261 (1994).
P. Monthoux and D.J. Scalapino, Phys. Rev. Lett. [**72**]{}, 1874 (1994).
C.H. Pao and N.E. Bickers, Phys. Rev. Lett. [**72**]{}, 1870 (1994).
S. Chakravarty, A. Sudbø, P.W.Anderson, and S. Strong, Science, [**261**]{}, 337 (1993)
J.M. Tranquada [*et al*]{}, Phys. Rev. B [**46**]{}, 5561 (1992).
H.A. Mook, Phys. Rev. Lett. [**70**]{}, 3490 (1993).
J.Rossat-Mignod, [*et al*]{}, Physica [**B199&200**]{}, 281 (1994)
S. Anlage [*et al*]{}, Phys. Rev. B [**50**]{}, 523 (1994).
J. Chen [*et al*]{}, Phys. Rev. B [**49**]{}, 3683 (1994).
J.P. Franck, [*et al*]{}, Phys. Rev. B [**44**]{}, 5318 (1991).
R.Liu et al, Phys. Rev., [**B46,**]{} 11056 (1992).
R.A. Klemm and S.H. Liu, Phys. Rev. B. [**44**]{}, 7526 (1991).
O.K.Andersen [*et al*]{}, Phys. Rev. B [**49**]{}, 4145 (1994).
H. Suhl, B.T. Matthias, and L.R. Walker, Phys. Rev. Lett. [**3**]{}, 552 (1959).
A.G. Aronov and E.B. Sonin, JETP Lett. [**36,** ]{} 556 (1973).
N.Bulut, D.J. Scalapino, and R.T. Scalettar, Phys. Rev. B [**45**]{}, 5577 (1992).
A. Golubov and I.I. Mazin, Physica C, to be published.
P. Monthoux, A.V. Balatsky, and D. Pines, Phys. Rev. B [**46**]{}, 14803 (1992).
In the spin-bag model of J.R Schrieffer et. al., Phys. Rev. Lett. [**60,** ]{}944 (1988), the attractive interaction is due to spin fluctuations and the symmetry of the pairing state is $s$. Our two-plane model bears some resemblance to the spin-bag model because, due to the assumed 2$\times $1 spin-density-wave state, the spin-bag model includes two bands with the associated interference of the wave functions. The spin-bag model makes the potential attractive for all q because of the mixing of the spin states in each band; this is in contrast to our interband sign reversal of the order parameter.
D.A. Wollman [*et al*]{}, Phys. Rev. Lett. [**71**]{}, 2134 (1993).
A.G. Sun [*et al*]{}, Phys. Rev. Lett. [**72**]{}, 2267 (1994).
D. Khomskii, J. Low Temp. [**95**]{}, 205 (1994).
This assumption makes Eq. fully symmetric with respect to band indices; correspondingly, two components of a superconducting pair reside on two different planes. In general, intraband pairing leads to both intra- and interplane anomalous averages (cf., e.g., M.U. Ubbens and P.A. Lee, Phys. Rev. B, [**50**]{}, 438, 1994, or E. Dagotto et al [*ibid*]{}, [**45**]{}, 5744, 1992).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We give detailed insight into photon counting OTDR ($\nu-$OTDR) operation, ranging from Geiger mode operation of avalanche photodiodes (APD), analysis of different APD bias schemes, to the discussion of OTDR perspectives. Our results demonstrate that an InGaAs/InP APD based $\nu-$OTDR has the potential of outperforming the dynamic range of a conventional state-of-the-art OTDR by 10 dB as well as the 2-point resolution by a factor of 20. Considering the trace acquisition speed of $\nu-$OTDRs, we find that a combination of rapid gating for high photon flux and free running mode for low photon flux is the most efficient solution. Concerning dead zones, our results are less promising. Without additional measures, e.g. an optical shutter, the photon counting approach is not competitive.'
author:
- 'Patrick Eraerds, Matthieu Legré, Jun Zhang, Hugo Zbinden, Nicolas Gisin [^1][^2] [^3]'
title: |
Photon Counting OTDR :\
Advantages and Limitations
---
[Shell : Photon Counting OTDR :\
Advantages and Limitations]{}
$\copyright$ 20xx IEEE. Personal use of this material is permitted. However, permission to reprint/republish this material for advertising or promotional purposes or for creating new collective works for resale or redistribution to servers or lists, or to reuse any copyrighted component of this work in other works must be obtained from the IEEE.\
Distributed detection, fiber metrology, optical time-domain reflectometry, photon counting
Introduction
============
Time Domain Reflectometry [@Barnoski] is a well known technique for fiber link characterization. Most of today’s commercially available optical time domain reflectometers (OTDRs) are based on linear photon detectors, such as p-i-n or avalanche photodiodes (APDs). Although single photon detection features unmatched sensitivity, OTDRs based on this technique ($\nu-$OTDR) [@Healey] have reached the market only in niches [@ComNUOTDR].\
Several single photon detection techniques are possible [@HamamatsuPMT]-[@Friedrich], but only few of them are suitable for in-field measurements. Geiger-mode operated InGaAs/InP APDs (for telecom wavelengths) [@Ribordy][@Stucki][@Rarity] are the most promising candidates, due to their robustness and manageable cooling.\
In this paper we discuss the advantages and limitations of these devices, when used in an $\nu-$OTDR. We concentrate in particular on the dynamic range, 2-point resolution, measurement time and dead zone. All $\nu-$OTDR measurements are supplemented by measurements using a conventional state-of-the-art OTDR (Exfo, FTB7600). This makes it easier to evaluate the $\nu-$OTDR performance. Our discussion also contains the possible yield of newly emerged gating techniques like *rapid gating* [@Shields][@Inoue][@Jun2].\
We note that some years ago $\nu$-OTDRs based on silicon APDs, suitable for C-band operation, were demonstrated [@Diamanti][@Legre]. Although silicon APDs show superior behavior, concerning afterpulsing and timing jitter, the upconversion of telecom photons to the visible regime demands more expensive optics and more sophisticated alignment. Therefore we believe that they are less suitable when robustness is required.\
Paper organization : In Sect.II we provide information about Geiger-mode operation of InGaAs/InP APDs and discuss its major impairment, the afterpulsing effect. Sect.III focusses on $\nu-$OTDR operation and performance (dynamic range, 2-point resolution, measurement time, dead zone) and compares it with the performance of a conventional state-of-the-art long haul OTDR (Exfo FTB7600). Sect.IV considers time efficient bias schemes (*rapid gating, free running*) and finally we summarize our results in Sect.V.
Geiger-mode APD
===============
Basic operation {#Basic operation}
---------------
In Geiger-mode, the APD is biased beyond its breakdown voltage, typically by a few percent. This provides a sufficiently large gain (order of $10^6$) to detect a single incident photon (with detection efficiency $\eta$). In contrast to a linear mode APD, the output signal is no longer proportional to the number of primary charges. Whenever an avalanche occurs and the current reaches a certain discrimination level, a detection is counted, independent of how many primary charges caused or were created during the avalanche.\
To reset the APD for the next detection, the avalanche needs to be quenched. This is typically done by lowering the bias voltage, either actively or passively [@Cova].\
An APD based on InGaAs/InP is particularly well suited for use with the principal telecom wavelength bands. Although the dark count[^4] rate is higher than in silicon based APDs, high sensitivity can be regained by cooling, typically around $-50^\circ$C (see also Sect.\[DetSens\]).\
There are different ways of applying the overbias ($V_{bias}>V_{bd}$ (breakdown voltage)) to the diode. The most common ones are the *gated mode* and the *free running mode* [@Thew]. In *gated mode* the overbias is applied only during a short time $\Delta t_{gate}$ (called gate), in a repetitive manner with frequency $f_{gate}$ (respecting $f_{gate}<\frac{1}{\Delta t_{gate}})$. Typically $\Delta t_{gate}\in [2 \mbox{ns}, 20\mu\mbox{s}]$ and $f_{gate}\in [100\mbox{Hz},10\mbox{MHz}]$. In *free running mode*, the overbias is applied until a photon or noise initiates an avalanche.\
While the *gated mode* achieves high signal to noise ratios when a synchronized signal is being detected, the *free running mode* is most suited when the photon arrival time is not known (e.g. in OTDR).\
More recent developments, summarized by the name *rapid gating* [@Shields][@Inoue][@Jun2], apply very short gates ($\approx$ 200 ps) in order to severely limit avalanche evolution and reduce afterpulsing (see Sect.\[Afterpulsing\]). The technical challenge consists in discriminating the rather small avalanche signal from the capacitive response to overbias of the diode itself. In “classical gating”, described in the previous paragraph, one usually waits until the avalanche signal is easy to discriminate. Typical gating frequencies in *rapid gating* are of the order of 1 GHz.\
In Sect.IV we will discuss pros and cons of these different approaches, in particular concerning their applicability for $\nu-$OTDRs.\
Detection sensitivity {#DetSens}
---------------------
A figure of merit for the sensitivity of a detector is its noise equivalent power ($NEP$). For example, the bandwidth normalized $NEP_{norm}$ of a linear photo detector is given by [@Hamamatsu][@Perkin]\
$$NEP_{norm} = \frac{\Delta I_{noise}}{S\cdot G} \quad [\frac{\mbox{W}}{\sqrt{\mbox{Hz}}}]$$\
where $\Delta I_{noise}$ \[A/$\sqrt{\mbox{Hz}}$\] is the standard deviation of the total noise current (thermal-, dark-, signal shot- and in case of gain also gain noise), normalized with respect to the bandwidth of the detector, $S$ \[A/W\] is the detector photosensitivity and $G$ is the gain of the diode (p-i-n diode : $G=1$, linear APD (typically) : $G=10-100$).\
APDs are superior to p-i-n diodes in the circuit noise limited regime[^5] [@Teich], but lose their advantage when the gain noise becomes important, i.e. at stronger signal powers. The minimal detectable power $NEP_{norm,0}~$\[W/$\sqrt{\mbox{Hz}}$\] is obtained by setting the signal power and thus the signal shot-noise equal to zero. $NEP_{norm,0}$ can usually be found in the data sheet of the diode, typically $10^{-15}- 10^{-13} ~$\[W/$\sqrt{\mbox{Hz}}$\] for InGaAs APDs at $25^\circ$C.\
A similar expression can be derived for Geiger-mode APDs (see App.\[APPC\], Eq.\[NEPnorm\]):\
$$NEP_{norm}=\frac{h\nu}{\eta}\cdot\sqrt{2\cdot\hat{p}_{noise}}\quad\quad [\frac{\mbox{W}}{\sqrt{\mbox{Hz}}}]
\label{NEPGeiger}$$\
where $\eta$ is the detection efficiency and $\hat{p}_{noise}$ is the noise detection probability per gate (including signal and dark count shot noise), normalized with respect to the gate width $\Delta t_{gate}$ in seconds. Again, setting the input optical power equal to zero, we infer the minimal detectable power (App.\[APPC\], Eq.\[NEPnorm0\]) $$NEP_{norm,0}=\frac{h\nu}{\eta}\cdot\sqrt{2\cdot\hat{p}_{dc}}\quad\quad [\frac{\mbox{W}}{\sqrt{\mbox{Hz}}}]
\label{NEP0Geiger}$$\
where $\hat{p}_{dc}$ denotes the dark count probability per gate, normalized with respect to the gate width $\Delta t_{gate}$ in seconds. Inserting the parameters of the Geiger mode APD used in our experiments ($\hat{p}_{dc}=2000~\mbox{s}^{-1}$, $\eta=10$%, $T=-50^\circ$C, Sect.\[DynamicR\]), we estimate $NEP_{norm,0}\approx 10^{-16}$ \[W/$\sqrt{\mbox{Hz}}$\].\
In Fig.\[NEPTEMP\] we see the evolution of $NEP_{norm,0}$ as function of temperature. We observe that when approaching ambient temperatures, we almost reach the regime of the best linear mode diodes.
![Bandwidth normalized noise equivalent power ($NEP_{norm,0}$, see Eq.\[NEP0Geiger\]) as function of Geiger-mode APD temperature. The detection efficiency $\eta$ is kept constant at 10%. At ambient temperatures the noise equivalent power is increased by almost a factor of 10 with respect to the usual operating temperature of $-50^\circ$C.[]{data-label="NEPTEMP"}](Eraer1.eps){width="8cm"}
Conversely, one might be tempted to cool linear diodes to $-50^\circ$C to reach the $NEP$ of Geiger mode APDs. Even if this might be in general achievable, one should not forget, that the output signal still needs to be amplified. Even at ambient temperatures the small pulse amplifier noise usually constitutes the dominating noise source leading to much higher effective NEPs.\
By analyzing the performance of a conventional OTDR in Sect.III, we will gain more insight into the sensitivity limits of linear mode APD detection systems.
Afterpulsing {#Afterpulsing}
------------
One of the major impairments of InGaAs/InP APDs is the afterpulsing effect. Imperfections and impurities in the semiconductor material are responsible for intermediate energy levels (also called trap levels), located between the valence band and the conduction band. During an avalanche, these levels get overpopulated with respect to the thermal equilibrium population. If the APD gets reactivated right after the quenching of an avalanche, the probability of thermal excitation or tunneling of one of these charges into the conduction band and the subsequent initiation of an afterpulse avalanche, is high. Although fundamentally the improvement of semiconductor purity and thus the reduction of the number of trap levels is preferable, different mitigation measures can be carried out :\
*a) dead time* : A purely passive measure is the introduction of a dead time. The trap population decreases exponentially with time, due to thermal diffusion. Finally the thermal equilibrium configuration is restored. The impact of afterpulsing can therefore be mitigated by maintaining the bias voltage below breakdown, i.e. the application of a dead time $\tau$, after a detection takes place. Dead times severely limit the maximum achievable detection rate.\
*b) heating* : An increased temperature accelerates the diffusion of trapped charges. However, at the same time charge excitation from the valence into the conduction band increases, leading to globally increased noise, which eventually reduces the detector sensitivity. Thus one cannot achieve low afterpulsing and high sensitivity at the same time. It is necessary to find a trade-off depending on the particular application.\
*c) quenching technique* : As soon as the avalanche has gained enough strength such that the current pulse can be detected, it needs to be quenched. The quenching speed is crucial to limiting the number of secondary charges which can populate trap levels. Here, fully integrated active quenching circuits yield much better results than non-integrated ones [@Jun]. Another approach is *rapid gating* (Sect.\[Basic operation\]). Avalanche evolution is terminated by short gate durations (200 ps) and the number of secondary charges is kept low.\
![Afterpulse probability as function of dead time $\tau$. The detection efficiency $\eta$ is equal to 10% and the temperature T=$-50^\circ C$. An active quenching application specific integrated circuit (ASIC) [@Jun] was used.[]{data-label="APCurves"}](Eraer2.eps){width="\linewidth"}
\
In Fig.\[APCurves\] we plot an example of afterpulse probability as function of dead time $\tau$, using a fully integrated ASIC based active quenching circuit [@Jun]. Whenever a detection takes place, we activate a second gate of width $\Delta t_{gate}= 10~ns$ with a temporal delay of $\tau$. In this second gate, no incident photons are present. If there is a detection it is either a dark count or an afterpulse. Since for large $\tau$ only the actual dark counts remain, we can subtract it from the total count rate and obtain the pure afterpulsing probability ($\rightarrow$ Fig.\[APCurves\]). During larger gates, the afterpulse probability sums up and afterpulsing increases. One can easily calculate the afterpulse probability of a gate of width $\Delta t_{gate}$ ($\leq 10 \mu s$) by $$p_{AP,\Delta t_{gate}}(\tau)=1-(1-p_{AP,10 ns}(\tau))^m \\
\label{InferingAP}$$ where $p_{AP,10 ns}(\tau)$ is the afterpulse probability in a gate of 10 ns width and $m=\frac{\Delta t_{gate}}{10ns}$.\
We note that if no afterpulse occurs in the first activated gate after a detection, it can also happen in any succeeding gate. However, the probability decreases due to trap charge diffusion. To get the total afterpulse probability, or rather the signal to afterpulse ratio, one needs to account for this summing effect as well. The lower the signal detection rate, the more summing-up takes place. Thus a higher signal detection rate improves the signal to afterpulse ratio.\
In Fig.\[APtrace\] we illustrate the impact afterpulsing can have in a $\nu-$OTDR measurement. Firstly and most importantly we must consider dead zones (see Sect.\[DEADZ\] for definition, not to be confused with dead time). Whenever an important loss (at 25 km) or a reflection (at 36 km) occurs, it is followed by a tail which prevents the detection of the Rayleigh backscatter directly behind it. Secondly, the backscatter trace is shifted to higher values, since more detections than in the pure signal case occur (pile-up effect). Thirdly, the slope of the trace is flatter than it should be.
![Illustration of the afterpulsing effect on $\nu-$OTDR trace. Most severe are the dead zones after large loss events (at 25 km) and reflections (at 35 km). More subtle is the change of the slope of the trace which is usually smaller than what is measured when afterpulsing can be neglected.[]{data-label="APtrace"}](Eraer3.eps){width="\linewidth"}
How much afterpulsing can be tolerated, generally depends on the particular measurement. For instance in a coarse measurement on a long span of fiber, where only peak positions or large loss events are of interest, one can tolerate a fairly high afterpulsing contribution. On the other hand, in the case of short links, where high precision for fiber attenuation measurement and dead zone minimization is desired, afterpulsing must be kept to a few percent or even lower, depending on required precision of the measurement.\
Numerical afterpulse correction methods were also analyzed [@Wegmuller], but it was found that for high precision measurements the algorithm lacks robustness due to possible variations in the afterpulse probability. It should therefore be used only when the requirements on precision are not stringent.\
Photon counting vs. conventional OTDR
=====================================
Although this section is mainly concerned with the $\nu$-OTDR technique, we also perform measurements using a state-of-the-art conventional[^6] OTDR (FTB-7600, EXFO), a product especially designed for long-haul applications (up to 50 dB dynamic range). This makes it easier for us to highlight the advantages and drawbacks of the photon counting approach. The experimental setup and a detailed explanation is given in Fig.\[setup\].
![Basic $\nu$-OTDR setup. The laser (we use the laser of the FTB-7600, $P_{peak}=400$ mW) emits pulses with a frequency $f_{pulse}$ adapted to the length of the fiber under test $L_{fiber}$ ($\rightarrow f_{pulse}= \frac{c}{2\cdot L_{fiber}}$). The signal is split at a 99/1-coupler. The 99% part is launched into the fiber under test (FUT) via a circulator. Backscattered light from the fiber exits the lower port of the circulator and illuminates the InGaAs/InP-APD. The 1% part is used to measure the time of departure $t_0$ of the laser pulse (for synchronization reasons) using a conventional photodiode (Newport, 1GHz). The output signal is sent to a delay generator. A delayed signal at $t_0+t_{delay}$ is sent to the APD to apply a detection gate of length $\Delta t_{gate}$ and the backscattered intensity corresponding to the applied delay is measured. The APD reverse bias is equal 48.7 V, yielding a detection efficiency $\eta = 10$% at $-57^\circ$C (minimal value). We measure a dark count probability per gate equal to $\hat{p}_{dc}= 2000~\mbox{s}^{-1}$ (normalized with respect to the gate width, Eq.\[NEP0Geiger\]). This corresponds to a dark count probability of $2\cdot10^{-5} $ for a gate width of 10 ns and $2\cdot10^{-2} $ for a 10 $\mu$s gate.[]{data-label="setup"}](Eraer4.eps){width="\linewidth"}
\
For a fixed delay, a number of laser pulses $N_{pulse}$ (repetition frequency $f_{pulse}$) are sent and $N_{gate} (=N_{pulse}$) gates are activated. A counter records the number of detections. The incident signal power can be inferred from the ratio of detections to activated gates (for details, see App.\[APPA\]).\
To get information on the backscatter of the entire fiber, the delay needs to be scanned, repeating the procedure explained before for each single delay position. The sampling resolution, i.e. the delay step $\Delta t_{delay} ~(t_{delay}=i\cdot\Delta t_{delay}, i=1,2,3...)$, needs to be adapted to the requirements of the particular measurement (e.g. zooming or coarse full trace measurement). The detection bandwidth is given by $B=\frac{1}{2\Delta t_{gate}}$.\
We note that this is only the most basic version of a photon counting OTDR. One of the advantages of this system is that due to the low gating frequency, we can totally exclude afterpulse effects (dead time 2 ms). Therefore we can determine the unadulterated dynamic range and 2-point resolution. Nevertheless, data acquisition is very time consuming. For example the measurement of the entire 200 km fiber, discussed in the next section (Fig.\[Comparison1mus\]), took about 6 hours. In Sect.IV we will see, how it can be performed more efficiently.
Dynamic range {#DynamicR}
-------------
To measure the dynamic range of both devices for different laser pulse widths, we take a 200 km fiber, composed of a 50 km spool and an installed fiber link of 150 km (Swisscom, Geneva-Neuchatel), which itself consists of several fibers. The length of the fiber allows a maximal laser pulse repetition rate of $f_{laser}=\frac{c}{2\cdot L_{fiber}}=500$ Hz, where $c$ is the speed of light in standard optical fiber.\
We start measuring the trace with the FTB-7600 for 3 minutes[^7] with a laser pulse width of 1 $\mu$s. The device acquires $180~s\cdot500~\mbox{Hz}=9\cdot10^4$ different traces. The final output trace is the numerical average of these single traces (Fig.\[Comparison1mus\], light grey curve). For a fair comparison the detection bandwidth should be equal for the two devices. For the conventional OTDR it is automatically chosen by the device and not available to us. We infer its value by looking at the noise period at the end of the measurement range. For a pulse width of $1\mu s$ we obtain 4 MHz. Under these conditions the dynamic range is found to be 34.5 dB.\
We then perform the $\nu$-OTDR measurement, ensuring that we do not saturate the detector with the backscatter from the beginning of the fiber. Therefore we insert an additional attenuator in front of the APD to reach the unsaturated regime. We adjust the attenuation to yield a detection rate of about 90% of the gate rate for the first delay position. At each delay position we count the number of detections within 3 minutes, which yields the same statistics per sampling point as in the previous case ($180~s\cdot500~\mbox{Hz}=9\cdot10^4$ samplings). We choose $\Delta t_{delay}$ equal to 3 $\mu$s (=300 m sampling point separation in fiber) and $\Delta t_{gate}=\Delta t_{pulse}$. With increasing delay the backscatter power and thus the detection rate decreases. When we start to approach the noise level of the detector, we pause the measurement and remove a part of the attenuation (to regain 90% detection rate), reduce the delay for a few kilometers (to get an overlap with the previous part) and resume the measurement. In this way we obtain several single traces of adjacent parts of the fiber. In the following we will refer to this as partial trace measurement. By means of the overlaps, the entire trace can be reconstructed. Each partial trace measurement contributes approximately 20 dB to the overall $\nu$-OTDR dynamic range. For example, to cover 50 dB of fiber loss, we need to perform three partial trace measurements[^8].\
![OTDR traces of 200 km fiber link using a laser pulse width of $1\mu s$. The light gray curve represents the result of the conventional OTDR (Exfo FTB-7600) after 3 minutes of measurement in standard configuration. In this configuration it uses a detection band width of 4 MHz at the end of the measurement range. The $\nu-$OTDR result is represented by the blue curve. The measurement bandwidth is 500 kHz using the same number of samplings for each point as the conventional device. The bandwidth adapted results for different pulse widths can be found in Fig.\[DynRanges\].[]{data-label="Comparison1mus"}](Eraer5.eps){width="\linewidth"}
The result of the $\nu$-OTDR measurements is also shown in Fig.\[Comparison1mus\] (blue curve). It is important to note that we adapt the gate width to the laser pulse width to obtain the minimal $NEP_0$ (App.\[APPC\], Eq.\[NEP0\]) without affecting the 2-point resolution (limited by the laser pulse width). This means that in the case of Fig.\[Comparison1mus\], the detection bandwidth of the $\nu-$OTDR is $B =\frac{1}{2\cdot\Delta t_{gate}}= 500$ kHz[^9]. To be able to compare the measured results in a representative manner, we average the conventional OTDR trace in order to obtain the same bandwidth as was used in the $\nu-$OTDR measurement. We gain 2.5 dB, yielding a corrected dynamic range of 37 dB. The $\nu-$OTDR advantage is found to be roughly 9 dB in this case.\
We repeat the measurement for different pulse widths keeping all other parameters unchanged. The final results are shown in Fig.\[DynRanges\]. The detection bandwidths were adapted as before. It holds that $B=\frac{1}{2\Delta t_{pulse}}$ for both devices.\
For pulse widths between 30 ns and 1 $\mu$s the dynamic range difference is about 9-10 dB. This is a direct consequence of the smaller $NEP_{norm,0}$ (see Sect.\[DetSens\]) of the Geiger-mode APD, since bandwidth and integration time per sampling point were equally chosen. This means that the $NEP_{norm,0}$ of the $\nu-$OTDR is roughly a factor 63-100 smaller[^10] than the $NEP_{norm,0}$ of the conventional OTDR (noise dominated by the small pulse amplifier).
![Dynamic ranges of FTB-7600 and $\nu-$OTDR for different laser pulse widths. The length of the used fiber was 200 km. The detection bandwidths $B$ were adapted in each case, it holds that $B=\frac{1}{2\Delta t_{pulse}}$.[]{data-label="DynRanges"}](Eraer6.eps){width="8cm"}
However, going to larger laser pulses, we observe increased noise for the $\nu$-OTDR and the advantage gets smaller. We suppose that this happens due to the increased backscatter power from the beginning of the fiber. Although the diode is not active, the charge persistence effect (also sometimes called charge subsistence) can have a non negligible impact on the noise counts in a subsequently activated gate (for more details see also Sect.\[DEADZ\]).\
In summary, by adapting sampling statistics and detection bandwidth of both devices we find an advantage of about 9-10 dB in dynamic range for the $\nu$-OTDR. By increasing the laser pulse width we observe increased detector noise and the effective advantage gets smaller. We uncouple the question of measurement time since it is highly dependent on the gating technique used in the $\nu-$OTDR. This discussion is postponed to section \[Timeeff\].
2-point resolution
------------------
When considering the 2-point resolution[^11], one can divide OTDR operation into two regimes a) the receiver limited and b) the laser peak power limited regime. In case a) the ultimate timing resolution is either given by the amplifier bandwidth or the detector jitter (using fine laser pulses), whereas in case b) the limited laser peak power makes it necessary to use larger pulse widths (larger than the limit given in case a)) in order to reach high dynamic ranges. In the receiver limited regime, the advantages of photon counting were already discussed in [@Wegmuller], yielding a maximal 2-point resolution of 10 cm for the $\nu$-OTDR and 1 m for the conventional device. In long haul OTDR applications, we operate in the laser peak power limited regime.\
It is easy to see that in this regime the sensitivity advantage ($NEP_{norm,0}$) of photon counting translates directly into an advantage in 2-point resolution.\
Example : we consider a reflective event at the end of the dynamic range (with a certain pulse width and measurement time) of the FTB-7600, see Fig.\[LimitTrace\].\
The $\nu-$OTDR can achieve the same dynamic range with a much smaller pulse width, see Fig.\[DynRanges\]. According to App.\[APPE\], an advantage of 10 dB in dynamic range[^12] corresponds to an advantage in 2-point resolution by a factor of 20. In Fig.\[Zoom\] we see the results of three $\nu-$OTDR measurements focusing on the reflective event at about 102.8 km. With each reduction in pulse width more and more structure is revealed and we actually find 3 reflections. The ratio of the peak widths agrees well with the calculated one.
![OTDR trace measured using the FTB-7600 with a laser pulse width of 300 ns, acquisition time = 3min. The achieved dynamic range is about 30 dB. At the transition to the noise level (at about 102.8 km), we can see a reflection peak (maybe two, see inset). The FTB-7600 cannot achieve a better resolution of the peak without decreasing the laser pulse width. As the dynamic range would drop (keeping the acquisition time unchanged) and the peak would not be seen anymore.[]{data-label="LimitTrace"}](Eraer7.eps){width="\linewidth"}
\
![Step by step reduction of laser pulse width in $\nu-$OTDR measurement, when zooming on the reflective event seen by the FTB-7600 (Fig.\[LimitTrace\]). The lower graph is again a zoom on the two reflective events which were revealed by the second graph (peak 2 and 3). Due to its larger sensitivity the $\nu-$OTDR can afford smaller laser pulses at distances where the FTB-7600 reaches its limits.[]{data-label="Zoom"}](Eraer8.eps){width="\linewidth"}
In summary, when the OTDRs are operated in the laser peak power limited regime, the sensitivity advantage of the $\nu-$OTDR translates directly into an advantage in 2-point resolution. Its amount is described by Eq.\[2pointAdv\] in App.\[APPE\].
Measurement time {#meastime}
----------------
We start discussing the measurement time by taking a look at the time necessary to obtain a sufficient signal to noise ratio (SNR) for a specific delay position in the fiber, from which we receive a backscatter power $P_{opt}$ (see App.\[APPF\]): $$t=\frac{1}{f_{pulse}}\cdot\left(\frac{SNR\cdot NEP_{norm}\cdot\sqrt{B}}{P_{opt}}\right)^2
\label{tmeas}$$ where $NEP_{norm}$ (see Eq.\[NEPnorm\]) is the noise equivalent power normalized with respect to detector bandwidth in \[W/$\sqrt{\mbox{Hz}}$\], $B$ the detector bandwidth in \[Hz\] and $f_{pulse}$ the laser pulse repetition rate in \[Hz\]. This formula applies to both the Geiger and linear mode operation as long as linearity between input and output signal is guaranteed[^13].\
While the Geiger mode exhibits linearity only in a relatively restricted domain of $P_{opt}$, the linear mode is able to cover several orders of magnitude of input power. Therefore Eq.\[tmeas\] is applicable for a much wider range of optical powers and shows the significant advantage of the linear mode when larger powers need to be measured ($\propto P_{opt}^{-2}$).\
More interesting from the $\nu-$OTDR perspective is the case when Eq.\[tmeas\] applies as well for Geiger-mode, i.e. for sufficiently small powers (order -100 dBm). We can then easily calculate the ratio of measurement times (assuming $f_{pulse},B$ and $P_{opt}$ to be equal): $$\frac{t^{(conv)}}{t^{(pc)}}=\left(\frac{NEP_{norm}^{(conv)}}{NEP_{norm}^{(pc)}}\right)^2
\label{timeratio}$$ where the superscript $pc$ signifies photon counting (i.e. Geiger mode), and $conv$ represents the conventional detection mode (i.e. linear mode).\
$NEP_{norm}$ can be split into a signal initiated noise contribution $NEP_{norm,sig}$ (e.g. due to signal shot noise) and a contribution from signal independent sources (dark current, dark counts) represented here by $NEP_{norm,0}$ : $$NEP_{norm}^2=NEP_{norm,sig}^2+NEP_{norm,0}^2$$ For sufficiently small input power, $NEP_{norm}$ becomes $NEP_{norm,0}$. In Sect.\[DynamicR\] we estimated the ratio of the $NEP_{norm,0}$ of conventional and $\nu-$OTDR to lie within 63 and 100. Thus the ratio given in Eq.\[timeratio\] approaches a value between 4000 and 10000. This means that we continuously pass from a huge linear mode advantage ($P_{opt}$ large) to a huge Geiger mode advantage ($P_{opt}$ small).\
We stress at this point that this result applies for the measurement of a specific delay position. OTDR measurements consist of scanning a range of different powers, i.e. the exponentially decreasing backscatter power. In case of the conventional OTDR the achievement of a sufficient SNR for a certain delay position, e.g. far down the fiber, means that all delays closer to the beginning of the fiber have been scanned at least with the same SNR. Thus we already obtain the full trace up to that point. How large the actually scanned interval is in the case of photon counting depends on the gating technique. For example, using the basic approach, explained earlier (Fig.\[setup\]), one would scan exactly one position. In Sect.IV we discuss how the NEP advantage of photon counting can be used more efficiently.\
Dead zone {#DEADZ}
---------
Dead zones are parts of a fiber link where the OTDR trace does not display the actual Rayleigh backscatter, but a signal induced by another source. One example that we have already encountered is the afterpulsing effect. If not accounted for it leads to tails after large loss events or reflections (see Fig.\[APtrace\]). Unfortunately this is not the only origin of dead zones. If we mitigate the afterpulsing effect by using an appropriate dead time, another effect, very similar to afterpulsing gets dominant : charge persistence. Even though the APD is not biased beyond breakdown between adjacent gates, there is still a bias which can weakly multiply primary charges created by photons incident during that time. These weak avalanches might however lead to increased trap population and increased noise avalanche probability in the next gate ($V_{bias}>V_{bdown}$). This effect, although less severe than afterpulsing, becomes visible under the same circumstances, namely after large loss events and reflections.\
![Behavior of conventional and photon counting OTDR when subjected to significant change in backscatter power level (here : 17 dB). We introduced a reflection of -45dB in front of the loss, simulating for example a bad connector. The black line represents a fit of the backscatter behind 8 km and can be used as a reference to assess the magnitude of the dead zone.[]{data-label="deadzone"}](Eraer9.eps){width="9cm"}
To estimate the impact of this effect on the OTDR output, we perform a measurement on a fiber link containing a large loss event (17 dB), with a reflection (-45 dB) just before it. This link simulates a typical situation encountered in passive optical networks, where a splitter of high multiplicity induces a significant loss. Weak reflections right in front can be induced by bad connectors.\
We perform a measurement of this particular fiber link situation, using our $\nu-$OTDR in basic mode, which ensures that no afterpulsing effect is present and the charge persistence effect becomes visible. Our results, including the measurement using the conventional OTDR, are shown in Fig.\[deadzone\]. We observe the emergence of a tail approximately 10 dB below the loss edge, which decays by approximately 3.5 dB/km. The lower Rayleigh backscatter level gets visible again after about 2 km ($20\mu s$).\
The result obtained with the conventional OTDR is much better. The emergence of the tail starts on a significantly lower level. Full sensitivity is regained after 1 km. The results found for the $\nu-$OTDR, confirm the observations made in [@Wegmuller] and [@Scholder]. One possibility to mitigate dead zones, induced by charge persistence, is the use of an optical shutter as performed in [@Scholder]. If the initial backscatter is blocked by the shutter and the gate gets activated, as well as the shutter deactivated right after the loss event, much better results can be obtained.\
In summary, concerning dead zones, the conventional OTDR shows superior behavior, when no additional measures are taken in the case of the $\nu-$OTDR, e.g. using an optical shutter.
Time efficient bias schemes {#Timeeff}
===========================
The way we implemented the $\nu$-OTDR in Sect.III is one of the simplest and trace acquisition is time consuming. It is well suited to study general characteristics, but not for other applications. Its apparent drawback is the wasting of backscattered signal, due to the fact that $f_{gate}=f_{pulse}$, i.e. only one gate per laser pulse is activated.\
A more efficient approach is the *train of gates* scheme [@Wegmuller]. Unlike to the basic mode, the gating frequency is higher than the laser pulse repetition rate and more than one gate gets activated per laser pulse, see Fig.\[GSchemes\]. In the ideal case we could choose a gating frequency $f_{gate}$ in the way that the designated sampling resolution is obtained. Unfortunately we have to account for the afterpulsing effect and thus need to apply a dead time $\tau$ (whose length depends on the tolerable afterpulsing)[^14]. In App.\[APPG\] we discuss the impact of dead time on the detection statistics in detail. To follow the general discussion here it can be skipped though.\
A measure for the acquisition speed is the achievable detection rate $f_{det}$. We state a linearized formula for $f_{det}$, which illustrates the most important relations very well : $$f_{det}=\frac{1}{\frac{1}{\eta\cdot\mu\cdot\Gamma}+\tau}
\label{fdet}$$ where $\eta$ is the detection efficiency$, \mu$ is the incident photon flux \[photons/sec\], $\Gamma=f_{gate}\cdot\Delta t_{gate}$ is the detection duty cycle and $\tau$ is the dead time.\
*a) high flux :* If the photon flux is large, the detection rate is limited by the dead time, i.e. $f_{det,max}=\frac{1}{\tau}$. In order to increase the detection rate, the dead time needs to be decreased. In Sect.\[Afterpulsing\] we have already started to discuss the possibilities of afterpulse mitigation. The quenching technique, that yields to date the lowest afterpulsing, is *rapid gating* [@Shields][@Inoue][@Jun2]. Dead times of the order of 10 ns can be considered realistic. This is approximately a factor 1000 better than the best actively quenched circuits can deliver. To maintain a reasonable duty cycle, gating frequencies of the order of 1 GHz are used ($\Delta t_{pulse}\approx 200$ ps).\
![Different bias schemes for backscatter measurement : basic, train of gates and free running mode. $\tau$ represents the dead time.[]{data-label="GSchemes"}](Eraer10.eps){width="\linewidth"}
*b) low flux :* If the flux is small, the dead time is insignificant and $f_{det}=\eta\cdot\mu\cdot\Gamma$. Here the most important parameter is the duty cycle $\Gamma$ which should be preferably high. Increasing the duty cycle will finally lead to a situation that is called *free running mode* [@Thew], Fig.\[GSchemes\]. The overbias is applied until a signal photon or a noise effect initiates an avalanche. The photon flux $\mu$ below which the *free running mode* yields higher detection rates is roughly $\frac{1}{\eta\cdot\tau}$, see also App.\[APPG\]. We see that in this case improved afterpulsing and thus smaller dead times, would extend the application range to higher photon fluxes. To date the best low afterpulsing solution for the *free running mode* is the earlier discussed integrated active quenching approach [@Jun]. Fig.\[Regions\] illustrates our discussion.\
At this point we want to demonstrate, that using the *free running mode* in its application regime (low flux), the acquisition speed of the conventional OTDR can be considerably outperformed.\
Example : We want to scan a 10 km interval of a 200 km fiber ($\Rightarrow f_{pulse}=500$ Hz). We assume $\tau = 10\mu s$, $\eta=0.1$ $\Rightarrow$ the maximal flux is $\mu=10^6$ photons/s, which corresponds to a power of $P_{start}=-99$ dBm[^15]. At the end of the 10 km interval the power drops to about $P_{stop}=-103$ dBm (assuming regular fiber behavior, attenuation =0.2 dB/km). From $P_{stop}$ one can infer $\hat{p}_{sig}$ (Eq.\[phat\]) and using $\hat{p}_{dc}$= 2000 $\mbox{s}^{-1}$ we obtain $NEP_{norm}^{(pc)}= 3.6\cdot10^{-16} ~\mbox{W}/\sqrt{\mbox{Hz}}$ (see Eq.\[NEPnorm\]), which enters into Eq.\[tmeas\] for measurement time calculation. The bandwidth $B$ of the measurement is managed by appropriate averaging of adjacent points after the full data is acquired. Here we suppose a bandwidth $B$ of 10 MHz (averaging on 50 ns intervals) and a signal to noise ratio (SNR) of 4. All together we compute a measurement time of approximately 20 s.\
![According to Eq.\[fdet\] we can determine the flux regimes where each of the techniques at disposal deliver their best performance. In the low flux regime $\mu<\frac{1}{\eta\cdot\tau}$ the detection duty cycle is most important, therefore the *free running mode* is the optimal solution. As soon as the dead time becomes the limiting factor for the detection rate, *rapid gating* can take advantage of its significantly lower afterpulsing, resulting in much smaller required dead times.[]{data-label="Regions"}](Eraer11.eps){width="\linewidth"}
To obtain the time of the conventional OTDR we calculate the $NEP_{norm}^{(conv)}$ and the ratio of Eq.\[timeratio\]. In the beginning of Sec.\[meastime\], we stated that the $NEP_{norm,0}^{(conv)}$ was about 63-100 times larger than the $NEP_{norm,0}^{(pc)}=10^{-16}~\mbox{W}/\sqrt{\mbox{Hz}}$. The signal power is low enough to neglect the signal induced noise contribution and we obtain simply $NEP_{norm}^{(conv)}\approx 63\cdot NEP_{norm,0}^{(pc)}=6.3\cdot10^{-15}~\mbox{W}/\sqrt{\mbox{Hz}}$ $$\Rightarrow \frac{t^{(conv)}}{t^{(pc)}}=\left(\frac{6.3\cdot10^{-15}}{3.6\cdot10^{-16}}\right)^2\approx 300$$ yielding a total measurement time for the conventional device of about 1.5 h, assuming the same detection bandwidth $B$.\
We stress that during this time, the conventional device obtains information not only on our 10 km measurement range, but also about the entire range before. The $\nu-$OTDR only scans the designated 10 km, but finishes doing this in around 20 s.
Conclusion
==========
The huge advantage of Geiger-mode APDs with respect to its linear mode counterparts is the small core noise equivalent power $NEP_{norm,0}$. Our comparison of a state-of-the-art conventional OTDR (based on linear mode APD) and a photon counting OTDR (based on Geiger-mode APD) reveals a difference of roughly two orders of magnitude. We demonstrate that this resource has the potential to improve the dynamic range by 10 dB as well as the 2-point resolution by a factor of 20 (in the laser peak power limited regime). The important question is, how efficient can it be used in OTDR applications (concerning the measurement time)? To sample the backscatter of a fiber we have different possibilities at hand, i.e. *train of gates* (with classical gating), *free running mode* or *rapid gating*. For sufficiently low backscatter power (order of -100 dBm; long fibers) we show that the *free running mode* is capable of efficiently using the NEP advantage. For example, measuring a 10 km interval far down the fiber, yielded a measurement time of around 20 s, while the conventional device needs to integrate for about 1.5 h to average out the noise sufficiently.\
At higher backscatter power, i.e. closer to the beginning of the fiber, we show that *rapid gating* can largely profit from its reduced afterpulsing, which makes dead times of the order of 10 ns realistic.\
We see a combination of *rapid gating* for the beginning of the fiber and *free running mode* for its end part as the currently best $\nu-$OTDR solution. Alternatively one can also imagine a hybrid of conventional OTDR for high flux and photon counting for low backscatter power, including high resolution scans with fine laser pulses on short distances.\
Concerning dead zones, the conventional OTDR is clearly superior. It is more robust to sudden changes in backscatter power, while the $\nu-$OTDR suffers from the charge persistence effect. This effect can for example be mitigated by using an additional optical shutter.
Power measurement with gated APD {#APPA}
================================
We consider coherent light, with a mean photon number per second $\mu$, incident on the diode (detection efficiency $\eta$). We apply gates of duration $\Delta t_{gate}$, which means that on average $\mu\cdot\Delta t_{gate}$ photons hit the diode during a gate. Due to the limited detection efficiency not every photon leads to a detection. If our detector would be photon number resolving, the average number of signal detections per gate would be given by $\eta\cdot\mu\cdot\Delta t_{gate}$. Since our APD does not have this ability, all cases where more than one signal detection would occur, results in only one detection output. According to Poissonian statistics, the probability of having no signal detection is given by $e^{-\eta\cdot\mu\cdot\Delta t_{gate}}$, hence the probability of having an APD signal detection output is given by : $$p_{sig,gate}=1-e^{-\eta\cdot\mu\cdot\Delta t_{gate}}.
\label{detProb}$$ $\mu$ can be expressed by the incident optical power $P_{opt}$ as $$\mu=\frac{P_{opt}}{h\nu}
\label{muPopt}$$ Solving for $P_{opt}$ yields $$P_{opt}=\frac{-h\nu}{\eta\cdot\Delta t_{gate}}\mbox{ln}(1-p_{sig,gate})
\label{Popt}$$ Measuring the ratio of the number of detections $N_{det}$ (including signal detections $N_{sig}$ and dark counts $N_{dc}$ ) and the number of activated gates ($N_{gate}$), yields the signal detection probability per gate $$p_{sig,gate}=\frac{N_{det}-N_{dc}}{N_{gate}}$$ and finally the incident optical power $P_{opt}$.\
If the signal is weak, then it is apparent that first of all the signal needs to be separated from noise by applying a sufficiently large number of gates $N_{gate}$, see also App.\[APPC\]. Once this is achieved, one has to consider the precision or statistical error of the result, which also is a function of $N_{gate}$. Two examples : a) $N_{gates}=10,~\Delta t_{gate}=100ns,~p_{dc,gate}=2\cdot10^{-4},~p_{sig,gate}=0.2$, the probability of a dark count to appear is negligible and we obtain $2\pm \sqrt{2} $ signal counts= total counts, from which we can calculate an optical input power lying in \[0.7 pW, 5.3 pW\]. b) $N_{gates}=10000$ and the other parameters like before, we obtain a number of total counts of $2002\pm \sqrt{2002} $ from which $2\pm \sqrt{2} $ are dark counts and $2000\pm \sqrt{2000} $ are signal counts. From the signal counts we infer an optical input power lying within \[2.8 pW, 2.9 pW\]. The statistical error of the second measurement is much smaller.\
Noise equivalent power of Geiger-mode APD {#APPC}
=========================================
In the treatment of the APD noise we mainly consider two contributions, i.e. the shot noise of a) the signal counts and b) the dark counts (assuming that afterpulse contributions can be neglected, for example by choosing a sufficiently large dead time).\
Let $P_{opt}$ be the incident optical power on the diode. With the energy per photon $h\nu$, the detection efficiency $\eta$ and the gate width $\Delta t_{gate}$, we infer the signal detection probability per gate (linearized version of Eq.\[detProb\], for sufficiently small $P_{opt}$) : $$p_{sig,gate}=\eta\cdot\frac{P_{opt}\cdot\Delta t_{gate}}{h\nu}
\label{Psig}$$ After applying $N_{gate}$ gates of the same width the mean number of signal detections $N_{sig}$ is $$N_{sig}=p_{sig,gate}\cdot N_{gate}
\label{Nsig}$$ Assuming Poissonian statistics we calculate the fluctuation $$\Delta N_{sig} = \sqrt{p_{sig,gate}\cdot N_{gate}}
\label{aux1}$$ The same derivation holds for the dark counts : we introduce a dark count probability per gate $p_{dc,gate}$ (which will be measured directly) leading to $$\Delta N_{dc} = \sqrt{p_{dc,gate}\cdot N_{gate}}
\label{aux2}$$ thus the total noise fluctuation : $$\Delta N_{tot}=\sqrt{\Delta N_{sig}^2+\Delta N_{dc}^2}
\label{totalNoise}$$ The noise equivalent power ($NEP$) is inferred by calculating the optical power necessary to produce $\Delta N_{tot}$ counts when applying $N_{gate}$ gates. In order to achieve this we replace $P_{opt}$ by $NEP$ in Eq.\[Psig\] and multiply by $N_{gate}$: $$\Delta N_{tot}=\eta\cdot\frac{NEP\cdot\Delta t_{gate}}{h\nu}\cdot N_{gate}$$ Using Eq.\[aux1\]-\[totalNoise\] and solving for $NEP$ yields $$NEP=\frac{h\nu}{\eta}\cdot\sqrt{\frac{p_{sig,gate}+p_{dc,gate}}{N_{gate} \Delta t_{gate}^2 }}\quad\quad [\mbox{W}]
\label{NEPAnnexe}$$ The minimal detectable power $NEP_0$ is obtained by setting the signal shot noise contribution equal to zero : $$NEP_0=\frac{h\nu}{\eta}\cdot\sqrt{\frac{p_{dc,gate}}{N_{gate} \Delta t_{gate}^2 }}\quad\quad [\mbox{W}]
\label{NEP0}$$ The existence of $N_{gate}$ in these equations represents the iteration of a measurement and is a function of time ($N_{gate}=f_{pulse}\cdot t)$). The elemental measurement time is represented by $\Delta t_{gate}$, the duration of a single gate, which can be interpreted as the detection bandwidth via $B:=\frac{1}{2\cdot\Delta t_{gate}}$. These formulas are practical when a $NEP$ for a particular measurement needs to be calculated (see also App.\[APPD\]). In order to obtain a formula which makes it easy to compare different detectors, we normalize with respect to $N_{gate}$ (which represents nothing else than the measurement time) and bandwidth $B$ :\
\
$$NEP_{norm}=\frac{h\nu}{\eta}\cdot\sqrt{2\cdot(\hat{p}_{sig}+\hat{p}_{dc})}\quad\quad [\frac{\mbox{W}}{\sqrt{\mbox{Hz}}}]
\label{NEPnorm}$$ and\
$$NEP_{norm,0}=\frac{h\nu}{\eta}\cdot\sqrt{2\cdot\hat{p}_{dc}}\quad\quad [\frac{\mbox{W}}{\sqrt{\mbox{Hz}}}]
\label{NEPnorm0}$$ where $$\hat{p}_{dc}:=\frac{p_{dc,gate}}{\Delta t_{gate}}\quad,\quad
\hat{p}_{sig}:=\frac{p_{sig,gate}}{\Delta t_{gate}}
\label{phat}$$ are the signal and dark count probability per gate, normalized with respect to the gate width in seconds [^16].
Dynamic Range of Optical Time Domain Reflectometer {#APPD}
==================================================
The strongest backscatter signal is observed right after the emission at time $t_0=\frac{\Delta l_{p}}{c}$, coming from the fiber locations within the interval $[0;\frac{\Delta l_{p}}{2}]$, where $c$ is the speed of light in the fiber and $\Delta l_{p}$ is the width of the laser pulse. The corresponding backscatter power is given by [@Derickson] $$P_{BS,0}= S\cdot P_{0,eff} \cdot e^{-2\alpha_s L}(1-e^{-\alpha_s\Delta l_{p}})$$ where $S$ is the fibers caption ratio, $P_{0,eff}$ is the effective laser peak power corrected for internal component (e.g. circulator) and connector loss and $\alpha_s$ the scattering coefficient. If we assume that $\alpha_s \Delta l_{p} << 1$, which is true in standard fiber ($\alpha_s\approx0.04 \mbox{km}^{-1}$) and $\Delta l_{p} < 2$ km ($\Delta t_{p}<10\mu s$) we can expand the exponential and get $$P_{BS,0}\approx\ S\cdot P_{0,eff}\cdot\alpha_s\cdot\Delta l_{p}
\label{RB0}$$ The dynamic range is then given by the ratio of $P_{BS,0}$ and the minimal detectable power $NEP_0~[W]$ (Eq.\[NEP0\]): $$\mbox{dynR} =5 ~\mbox{log}(\frac{P_{BS,0}}{NEP_0})
\label{EquDynR}$$ where the factor 5 accounts for the roundtrip in the fiber. Finally we obtain : $$\mbox{dynR} \approx\ 5 ~\mbox{log}(\frac{S\cdot P_{0,eff}\cdot\alpha_s\cdot\Delta l_{p}}{NEP_0})
\label{EquDynRApprox}$$ We note that $NEP_0$ like used here, includes the measurement time and decreases $\propto\sqrt{t}$ (see also Eq.\[NEP0\] in the case of the $\nu-$OTDR, where the measurement time $t$ is represented by the number of applied gates, $N_{gate}=f_{pulse}\cdot t$).\
The operational definition of the dynamic range of an OTDR, given for example in [@Derickson], contains a measurement time of 3 minutes. It is apparent that an extended measurement time enhances the $NEP_0$ and thus the dynamic range. In general, if the measurement time is increased by a factor $d$, the standard deviation of the noise is lowered by a factor $\sqrt{d}$ and thus the $NEP_0$ by the same amount.
2-point resolution advantage of $\nu$-OTDR {#APPE}
==========================================
We assume an $x$ dB $\nu$-OTDR advantage in dynamic range (with respect to conventional OTDR, using the same laser pulse width $\Delta l_{p}$). Now we look for a factor $\alpha$ such that $\alpha\cdot\Delta l_{p}$ yields a reduction of the $\nu$-OTDR dynamic range by $x$ dB. Using Eq.\[EquDynRApprox\],\[NEP0\] and $\Delta t_{gate}=\frac{\Delta l_{p}}{c}$ (adapting laser pulse width and gate width) we have to fulfill $$5\mbox{log}((\alpha\cdot\Delta l_{p})^{\frac{3}{2}})=5\mbox{log}((\Delta l_{p})^{\frac{3}{2}})-x$$ yielding $$\alpha=10^{-\frac{2}{15}x}
\label{2pointAdv}$$ For example : $x=10$ dB $\rightarrow \alpha=0.046$. Thus the $\nu$-OTDR can achieve the same dynamic range with a 20 times smaller pulse width.
Signal to noise ratio as function of measurement time {#APPF}
=====================================================
We derive a formula for the SNR ratio as a function of time from the photon counting perspective. However, the final result, after linearization will not contain any photon counting specific quantity and is therefore generally applicable.\
We define the signal to noise ratio $(SNR)$ as the ratio of signal counts $N_{sig}$ to the total fluctuation of the counts $\Delta N_{tot}$ including fluctuation of signal and dark counts (like defined in App.\[APPC\]). $$SNR=\frac{N_{sig}}{\Delta N_{tot}}=\frac{p_{sig,gate}\cdot N_{gate}}{\sqrt{(p_{sig,gate}+p_{dc,gate})\cdot N_{gate}}}$$ using Eq.\[Nsig\] and \[totalNoise\]. Now we introduce the bandwidth normalized NEP (Eq.\[NEPnorm\]) and use $N_{gate}=f_{pulse}\cdot t$ where $t$ is the measurement time $$\Rightarrow SNR=\frac{\sqrt{2}\cdot h \nu}{\eta}\cdot\frac{p_{sig,gate}\cdot\sqrt{f_{pulse}\cdot t}}{NEP_{norm}\cdot\sqrt{\Delta t_{gate}}}$$ then replacing $p_{sig,gate}$ using Eq.\[detProb\] and \[muPopt\]: $$SNR=\frac{\sqrt{2}\cdot h \nu}{\eta}\cdot\frac{(1-e^{-\frac{\eta}{h\nu}\cdot P_{opt}\cdot\Delta t_{gate}})\cdot\sqrt{f_{pulse}\cdot t}}{NEP_{norm}\cdot\sqrt{\Delta t_{gate}}}$$ If the optical input power is sufficiently small, the signal detection probability increases linearly with the optical power and we can expand the exponential to obtain $$SNR=\frac{P_{opt}\cdot\sqrt{f_{pulse}\cdot t}\cdot\sqrt{2\cdot\Delta t_{gate}}}{NEP_{norm}}$$ $$=\frac{P_{opt}\cdot\sqrt{f_{pulse}\cdot t}}{NEP_{norm}\cdot\sqrt{B}}\quad\quad\quad$$ where we used $B=\frac{1}{2\cdot\Delta t_{gate}}$, the detection bandwidth.\
This final formula is independent of any photon counting quantities and does also apply for the general case, including linear APD detection. In the linear regime there is even no such severe restrictions as in photon counting mode since much higher $P_{opt}$ can be processed.\
On the other hand, if measurement time needs to be calculated as a function of $SNR$, we obtain straight forward $$t=\frac{1}{f_{pulse}}\cdot\left(\frac{SNR\cdot NEP_{norm}\cdot\sqrt{B}}{P_{opt}}\right)^2
\label{tmeasSNR}$$
Train of gates discussion {#APPG}
=========================
The application of a dead time can have considerable impact on the gate activation statistics. Fig.\[activationminimum\] shows what happens if the product $f_{gate}\cdot\tau$ is chosen too large.\
![Dead time application (here : $\tau=10 \mu s$ corresponding to 1 km in terms of sampling distance) has considerable impact on the gate activation statistics. If the gating frequency $f_{gate}$ is chosen too high, then the activation minimum approaches zero, no signal can be acquired there.[]{data-label="activationminimum"}](Eraer12.eps){width="\linewidth"}
If we assume, that the probability of detecting a signal photon in the first gate of the train of gates is $p_{sig,gate1}$, then gate 2 gets activated with probability $(1-p_{sig,gate1})$ (otherwise it falls into the dead time of gate 1). The probability that gate i gets activated ($i\leq f_{gate}\cdot\tau$) is then given by $$p_{act,gate~i}=(1-p_{sig,gate1})^{(i-1)}
\label{pact}$$ where we assume, that the signal detection probability is almost constant at the beginning of the fiber. This expression approaches 0 when i is large. In the worst case, “activation holes” appear in a repetitive manner (see Fig.\[activationminimum\], in the case where the minima of the periodic structure at the beginning touch zero probability) and therefore detections around these locations are impossible or only possible with very low statistics.\
To avoid this, we can define a criteria, which ensures that each gate has a sufficient number of activations. The first minimum plays the role of a bottleneck. When it is above some threshold (to be defined), all the other minima are as well[^17]. The minimum depends on $p_{sig,gate1}$ and $f_{gate}\cdot\tau$.\
Using Eq.\[pact\] we can calculate, the maximally suitable gating frequency $f_{gate,max}$, depending on a designated threshold, see Fig.\[fgatemax\].\
Example : assuming $\tau=1~ \mu s,~ p_{sig,gate1}=0.25$ and an activation minimum of 0.4. Then we infer $f_{gate,max}\cdot\tau = 4$ (see arrows in Fig.\[fgatemax\]) and therefore $f_{gate,max}=\frac{4}{\tau}=4$ MHz.\
If the maximally suitable gating frequency $f_{gate,max}$ is not large enough to obtain the designated sampling resolution, it is necessary to shift the start delay of the train of gates. For instance, if we want a sampling resolution of 5 m, but $f_{gate,max}=4$ MHz, yielding only 25 m, we need to delay the train (with respect to the laser pulse departure) four times by 50 ns. This of course increases the total measurement time by a factor 5.\
We note that $f_{gate,max}$, is in principle bounded by $1/\Delta t_{gate}$. If the suggested $f_{gate,max}$, according to Fig.\[fgatemax\], is larger than $1/\Delta t_{gate}$, we are naturally led to the *free running mode*, where the diode stays active until a detection is obtained [@Thew], see Fig.\[GSchemes\].\
![The maximally allowed gating frequency $f_{gate,max}$, in order to avoid activation holes, is a function of the photon detection probability in the first gate $p_{ph,gate1}$ and the designated dead time, which itself is set by the designated afterpulsing probability. Depending on the choice of the activation minimum value one can infer the product $f_{gate,max}\cdot\tau$, which also signifies the number of non activated gates after a detection. An example is given in the text.[]{data-label="fgatemax"}](Eraer13.eps){width="\linewidth"}
This happens if $$\mu<\frac{b}{\eta\cdot\tau}
\label{fluxequ}$$ or equivalently $$P_{opt}<h\nu\cdot\frac{b}{\eta\cdot\tau}$$ where $\mu$ is the photon flux (number of photons per second, cw) and $P_{opt}$ the corresponding power, $\eta$ the detector efficiency, $\tau$ the detector dead time and $b$ a constant depending on the activation minimum criteria, explicitly : for an activation minimum of $0.2 (0.4,0.6,0.8)$, one obtains $b = 1.61 (0.92,0.51,0.23)$. Due to its superior duty cycle, the *free running mode* is the ideal low power solution.\
Acknowledgment {#acknowledgment .unnumbered}
==============
The authors would like to thank EXFO for providing the FTB-7600 OTDR module and support, in particular S. Perron, J. Gagnon and G. Schinn. We also like to thank J.D. Gautier for providing technical support and Bruno Sanguinetti for thorough cross-reading. We thank Swisscom for providing access to the fiber link between Geneva and Neuchatel.
[99]{} “Fiber waveguides: a novel technique for investigating attenuation characteristics”, M. K. Barnoski and S. M. Jensen, Appl. Opt., Vol. 15, Issue 9, pp. 2112-2115, (1976)
“Optical time domain reflectometry by photon counting”, P.Healey, P. Hensel, Electronics Letters Vol.16 Issue 16, p.631-633 (1980)
Luciol instruments (www.sunrisetelecom.ch), Scientex (www.scientex.co.jp/english/index.html)
“Photomultiplier Tubes: Basics and Applications (3rd Edition)”, Hamamatsu Photonics, Hamamatsu City, Japan, (2006)
“Performance of InGaAs/InP avalanche photodiodes as gated-mode photon counters”, G.Ribordy, J.D. Gautier, H.Zbinden, N. Gisin, Applied Optics 37 (12), pp.2272-2277, (1998).
“Picosecond superconducting single-photon optical detector”, G. N. Gol’tsman, O. Okunev, G. Chulkova, A. Lipatov, A. Semenov, K. Smirnov, B. Voronov, A. Dzardanov, C. Williams, R.Sobolewski, Appl. Phys. Lett. 79, 705 (2001)
“Detection of single infrared, optical, and ultraviolet photons using superconducting transition edge sensors”, B. Cabrera, R. M. Clarke, P. Colling, A. J. Miller, S. Nam, R. W. Romani, Appl. Phys. Lett., Volume 73, Iss. 6, (1998)
“A single-photon detector in the far-infrared range”, S. Komiyama, O. Astafiev, V. Antonov, T. Kutsuwa, H. Hirai, Nature, Vol. 403, pp. 405-407, (2000)
“Superconducting Tunnel Junction Photon Detectors: Theory and Applications”, S. Friedrich, J.Low Temp. Phys. 151, pp. 277-286, (2008)
“Photon counting for quantum key distribution with Peltier cooled InGaAs/InP APDs” D.Stucki, G.Ribordy, A. Stefanov, H. Zbinden, J.G. Rarity, T. Wall, Journal or Modern optics Vol.48 No. 13, 1967-1982, (2001)
“Single-photon counting for the 1300-1600-nm range by use of a Peltier-cooled and passively quenched InGaAs avalanche photodiodes”, J.G.Rarity, T.E.Wall, K.D.Ridley, P.C.M. Owens, P.R.Tapster, Appl.Opt. Vol.39, No.36 (2000)
“High speed single photon detection in the near infrared”, Z.L.Yuan, B.E.Kardynal, A.W. Sharpe, A.J. Shields, Appl.Phys.Lett. 91, 041114 (2007)
“800 MHz Single-photon detection at 1550-nm using an InGaAs/InP avalanche photodiode operated with a sine wave gating”, N. Namekata, S. Sasamori, S. Inoue, Opt. Express, Vol. 14, No. 21 (2006)
“Practical fast gate rate InGaAs/InP single-photon avalanche photodiodes”, J. Zhang, R. Thew, C. Barreiro, H. Zbinden, Appl. Phys. Lett. 95, 091103 (2009) “1.5 $\mu$m photon counting optical time-domain reflectometry with a single-photon detector based on upconversion in a periodically poled lithium niobate waveguide”, E. Diamanti, C.Langrock, M.M. Fejer, Y. Yamamoto, Opt.Lett. Vol. 31 No. 6 (2006)
“High resolution optical time domain reflectometer based on 1.55 $\mu$m up-conversion photon-counting module”, M. Legré, R.Thew, H.Zbinden, N. Gisin, Opt. Express, Vol. 15 Issue 13, pp.8237-8242 (2007)
“Avalanche photodiodes and quenching circuits for single-photon detection”, S. Cova, M. Ghioni, A. Lacaita, C. Samori, F. Zappa, Appl. Opt., Vol. 35, Iss. 12, pp. 1956-1976, (1996)
“Free-running InGaAs/InP Avalanche Photodiode with Active Quenching for Single Photon Counting at Telecom Wavelengths”, R.T. Thew, D. Stucki, J.D. Gautier, A. Rochas, H.Zbinden, Appl. Phys. Lett. 91, 201114 (2007)
“Comprehensive Characterization of InGaAs/InP Avalanche Photodiodes at 1550 nm with an Active Quenching ASIC”, J. Zhang, R. Thew, J.D. Gautier, N. Gisin, H. Zbinden, IEEE J. Quantum Electron., vol. 45, no. 7, pp. 792-799, (2009)
“Photon-Counting OTDR for Local Birefringence and Fault Analysis in the Metro Environment” M.Wegmuller, F. Scholder, N. Gisin, Journal of Lightwave Technology, Vol. 22, No.2 (2004)
“Photodiode Technical Guide” Hamamatsu, http://sales.hamamatsu.com/assets/html/ssd/si-photodiode/index.htm
“Avalanche Photodiode : A User Guide”, Application Note Perkin Elmer
“Fundamentals of Photonics” B.E.A. Saleh, M.C.Teich, John Wiley & Sons second edition Chapter 17 “Semiconductor Photon Detectors”
Fiber Optic Test and Measurement, Dennis Derickson, Prentice Hall (1998)
“Long-distance OTDR using photon counting and large detection gates at telecom wavelength”, F.Scholder, J.D. Gautier, M. Wegmuller, N.Gisin, Optics Communications 213, 57–61,(2002)
[^1]: P. Eraerds, J. Zhang, H. Zbinden, N. Gisin are with the Group of Applied Physics, University of Geneva, 1211 Geneva 4 Switzerland, e-mail: [email protected]
[^2]: M. Legré is with idQuantique SA, 1227 Carouge/Geneva, Switzerland
[^3]: Manuscript received April x, 200y; revised January x, 200y. Financial supports from the Swiss Federal Department for Education and Science (OFES), in the framework of the European COST299 project, from the Swiss NCCR “Quantum photonics” are acknowledged.
[^4]: A detection which was not initiated by a signal photon but thermal excitation or tunneling.
[^5]: Circuit noise results for example from thermal motion of charges in resistors or charge fluctuation in transistors in the receiver amplifier.
[^6]: Based on linear-mode APD detection.
[^7]: We choose 3 min because this is the time specified in the definition of OTDR dynamic range for conventional OTDRs [@Derickson].
[^8]: The first measurement covers 0-20 dB, the second 15-35 dB and the third 30-50 dB respecting the necessary overlap between different partial trace measurements.
[^9]: The conventional device uses a higher bandwidth since higher sampling resolution is useful when the position of an event needs to be determined with higher precision.
[^10]: Respecting the functional dependence between NEP and dynamic range given in Eq.\[EquDynR\], using $NEP_0\propto NEP_{norm,0}$
[^11]: By 2-point resolution we mean the minimal distance, necessary between two reflective events, in order to be able to recognize them as distinct peaks on the OTDR trace output (e.g. dip between peaks at least 1 dB lower than the peaks themselves)
[^12]: Bandwidth normalized.
[^13]: Linearity in Geiger mode applies if the signal detection probability per gate $p_{sig,gate}$ depends linearly on the input power $P_{opt}$, see App.\[APPA\] Eq.\[detProb\] & \[muPopt\]. For sufficiently small $P_{opt}$ it holds that : $p_{sig,gate}=\eta\cdot\frac{P_{opt}\cdot\Delta t_{gate}}{h\nu}$
[^14]: Like depicted in Fig.\[GSchemes\], we can number each gate in the train of gates, $1...n$. In the ideal case (no dead time) each of these gates gets activated for each laser pulse sent into the fiber. In reality, when a dead time needs to be applied, it is not certain that gate i gets activated. It is possible that it falls into the dead time application range (gate 4-7 in Fig.\[GSchemes\]).
[^15]: It depends on the laser power, pulse width and the fiber link quality, to which distance this corresponds. Choosing $P_{peak}=400$ mW and assuming that this is also the effective power that reaches the fiber (no internal loss), $\Delta t_{pulse}= 100$ ns and a regular fiber behavior (loss = 0.2 dB/km), then -99 dBm correspond to backscatter power coming from a distance of 158 km.
[^16]: The relation between signal count/dark count probability per gate and gate width is almost linear over a large range of practical gate widths (typically ranging from nanoseconds to microseconds).Knowing $\hat{p}_{sig}$ and $\hat{p}_{dc}$ makes it easy to calculate the signal count and dark count probability of a particular gate of widths $\Delta t_{gate}$, just by multiplying it by $\Delta t_{gate}$.
[^17]: The reason for this is the fiber loss, which decreases the backscattered signal and therefore yields less detections and dead time applications. The dead time effect gets less severe. When the backscatter power is very low there is almost no difference to the ideal case
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We study a one-dimensional (iso)spin $1/2$ Bose gas with repulsive $\delta$-function interaction by the Bethe Ansatz method and discuss the excitations above the polarized ground state. In addition to phonons the system features spin waves with a quadratic dispersion. We compute analytically and numerically the effective mass of the spin wave and show that the spin transport is greatly suppressed in the strong coupling regime, where the isospin-density (or “spin-charge”) separation is maximal. Using a hydrodynamic approach, we study spin excitations in a harmonically trapped system and discuss prospects for future studies of two-component ultracold atomic gases.'
author:
- 'J.N. Fuchs'
- 'D.M. Gangardt'
- 'T. Keilmann'
- 'G.V. Shlyapnikov'
title: 'Spin waves in a one-dimensional spinor Bose gas'
---
Recent experiments have shown the possibility of studying ultra-cold atomic gases confined in very elongated traps [@1d_exp; @PhilEss; @Bloch; @Weiss]. In such geometries, the gas behaves kinematically as if it were truly one-dimensional (1D). Many theoretical studies [@Girardeau; @LL; @Lieb; @Olshanii; @PSW; @DLO] have predicted and discussed interesting effects in 1D Bose gases, such as the occurrence of fermionization in the strong coupling Tonks-Girardeau (TG) regime, where elementary excitations are expected to be similar to those of a non-interacting 1D Fermi gas [@Girardeau]. Manifestations of strong interactions have been found in the experiments [@PhilEss], and recently the TG regime has been achieved for bosons in an optical lattice [@Bloch] and in the gas phase [@Weiss].
Present facilities allow one to create spinor Bose gases which has been demonstrated in experimental studies of two-component Bose-Einstein condensates [@Cornell]. These systems are produced by simultaneously trapping atoms in two internal states, which can be referred to as (iso)spin 1/2 states. Relative spatial oscillations of the two components can be viewed as spin waves [@spin_dyn] (see [@Nikuni2003] for review). A variety of interesting spin-related effects such as phase separation [@phasesep], exotic ground states [@exotic], and counter intuitive spin dynamics [@spin_dyn] due to the exchange mean field, have been studied both theoretically and experimentally. However, most of these studies are restricted to the weakly interacting Gross-Pitaevskii (GP) regime. There is a fundamental question to what extent these effects survive in the strongly correlated regime characteristic of one spatial dimension. The purpose of the present Letter is to study spin excitations of an interacting 1D spinor Bose gas. This is done using an exact solution by the Bethe Ansatz.
We start with a spinor (two component) gas of $N$ bosons with mass $m$ at zero temperature, interacting with each other via a repulsive short-range potential in a narrow three-dimensional waveguide. In general, the interaction depends on the internal (spin) states of the colliding particles. Here we consider the case of a spin-independent interaction characterized by a single 3D scattering length $a>0$. This is a reasonable approximation for the commonly used internal levels of $^{87}$Rb (see e.g. [@Cornell]). The waveguide has length $L$ and we assume periodic boundary conditions for simplicity. The transverse confinement is due to a harmonic trapping potential of frequency $\omega_0$. When the chemical potential of the gas is much smaller than $\hbar\omega_0$, the transverse motion is frozen to zero point oscillations with amplitude $l_0= \sqrt{\hbar/m\omega_0}$. In such a quasi-1D geometry, the interaction between atoms is characterized by an effective one-dimensional delta-potential $g\delta(x)$. For $a\ll l_0$, the coupling constant $g$ is related to the 3D scattering length as $g=2\hbar^2 a/ml_0^2>0$ [@Olshanii]. The behavior of the system depends crucially on the dimensionless parameter $\gamma = mg/\hbar^2
n$, where $n= N/L$ is the 1D density. For $\gamma\ll 1$ one has the weak coupling GP regime, whereas for $\gamma\gg 1$ the gas enters the strongly interacting TG regime.
Under the above conditions, the system is governed by the following spin-independent 1D total Hamiltonian: $$\label{eq:ham}
H=-\frac{\hbar^2}{2m}\sum_{i=1}^{N}\frac{\partial^2}{\partial
x_i^2}+g\sum_{i<j}\delta(x_i-x_j).$$ This Hamiltonian was introduced by Lieb and Liniger [@LL] for describing spinless bosons, and their solution by the Bethe Ansatz (BA) has been generalized to bosons or fermions in two internal states by M. Gaudin and C.N. Yang [@Gaudin; @Yang]. In the case of a two-component Bose gas (spin 1/2 bosons), due to the $SU(2)$ symmetry of the Hamiltonian the eigenstates are classified according to their total (iso)spin $S$ ranging from 0 to $N/2$. In this case, which was recently considered by Li, Gu, Yang and Eckern [@LGYE], the ground state is fully polarized ($S=N/2$) and has $2S+1$-fold degeneracy, in agreement with a general theorem [@EL; @YL]. At a fixed $S=N/2$, the system is described by the Lieb-Liniger (LL) model [@LL], for which elementary excitations have been studied in Ref. [@Lieb] for any value of the interaction constant. Spin excitations above the ground state are independent of the ground-state spin projection $M_S$ and represent transverse spin waves. For $M_S=0$ they correspond to relative oscillations of the two gas components.
We first give a brief summary of the BA diagonalization [@Gaudin; @Yang; @LGYE] of the Hamiltonian (\[eq:ham\]). An eigenstate with total spin $S=N/2-K$ ($0\leq K\leq N/2$) is characterized by two sets of quantum numbers: $N$ density quantum numbers $I_j$ with $j=1,..,N$ and $K$ spin quantum numbers $J_{\mu}$ with $\mu=1,..,K$. If $N-K$ is odd (resp. even), $I_j$ and $J_\mu$ are integers (resp. half-integers). These quantum numbers define $N$ quasi-momenta $k_j$ and $K$ spin rapidities $\lambda_{\mu}$, which satisfy the following set of BA equations (we set $\hbar=2m=1$): $$\begin{aligned}
\!\!\frac{L k_j}{2}\!\!&=&\!\!\pi I_j\!-\!\sum_{l=1}^{N}\!\arctan
\!\!\left(\!\frac{k_j\!-\!k_l}{g/2}\!\!\right) \!+\!
\sum_{\nu=1}^{K}\!\arctan\!\!\left(\!\frac{k_j\!-\!\lambda_\nu}{g/4}\!\!\right),
\label{BAE1}\\
\!\!\pi J_\mu\!\! &=& \!\! \sum_{l=1}^{N}\!\arctan
\!\left(\!\frac{\lambda_\mu\!-\!k_l}{g/4}\!\!\right)
-\sum_{\nu=1}^K\!\arctan\!
\left(\!\frac{\lambda_\mu\!-\!\lambda_\nu}{g/2}\!\!\right).
\label{BAE2}\end{aligned}$$ The energy of the corresponding state is $E=\sum_{j} k_j^2$, and its momentum is given by: $$p=\sum_{j=1}^{N} k_j=\frac{2\pi}{L}\Big(\sum_{j=1}^N I_j
-\sum_{\mu=1}^K J_{\mu}\Big) .
\label{mom}$$ As we are also interested in finite size effects, we do not take the thermodynamic limit at this point.
The ground state corresponds to the quantum numbers $\{I_j^{0}\}=\{-(N-1)/2,..,(N-1)/2\}$ and $K=0$, which shows that the BA equations reduce to those of LL [@LL]. The wave function is given by the orbital wave function of the LL ground state multiplied by a fully polarized spin wave function. All ground state orbital properties (energy, chemical potential, correlation functions, etc.) are therefore identical to those of the LL model. Elementary excitations in the density sector correspond to modifying the density quantum numbers $I_j$ while leaving the total spin unchanged, i.e., $K=0$. At low energy, the density excitations are phonons propagating with the Bogoliubov sound velocity $v_s=\sqrt{2gn}$ in the GP limit and with the Fermi velocity $v_s=2\pi n$ in the TG regime.
We now focus on the spin sector. Elementary spin excitations correspond to reversing one spin ($K=1$), and the total spin changes from $N/2$ to $N/2-1$. Thus, we have a single spin rapidity $\lambda$ and the corresponding quantum number $J$. In general, this procedure creates a density excitation and a spin wave (isospinon) [@LGYE]. Here, we choose specific quantum numbers $I_j, J$ in order to excite the isospinon alone [@footnote]. Accordingly, the momentum $p$ of the excitation is $$p=\frac{2\pi}{L}\left(\frac{N}{2}-J\right),
\label{pJ}$$ which follows from the definition (\[mom\]).
In the long wavelength limit, where $|p|\ll n$, due to the $SU(2)$ symmetry one expects [@Halperin] a quadratic dispersion for the spin-wave excitations above the ferromagnetic ground state: $$\varepsilon_p \equiv E(p)-E_0\simeq p^2/2m^{*}
\label{defiem},$$ where $E(p)$ is the energy of the system in the presence of a spin wave with momentum $p$, $E_0$ is the ground state energy and $m^{*}$ is an effective mass (or inverse spin stiffness). This quadratic behavior is due to a vanishing inverse spin susceptibility, which is a consequence of the $SU(2)$ symmetry [@Halperin]. A variational calculation in the spirit of Feynman’s single mode approximation [@YL], shows that $\varepsilon_p \leq p^2/2m$ implying that $m^{*}\geq m$. Below we show that strong interactions greatly enhance the effective mass.
In the strong coupling limit it is possible to solve the BA equations (\[BAE1\]) and (\[BAE2\]) perturbatively in $1/\gamma$ [@LL]. We solve these equations both for the ground state $\{I_j^{0}\}$ and the excited state $\{I_j;\, J\}$. We anticipate that in the limit of strong interactions, for small momenta ($|p|/n\ll 1$) and a large number of particles ($N\gg 1$), the dimensionless spin rapidity is $\tilde{\lambda}\equiv 2\lambda/ g \gg 1$ and the dimensionless quasi-momenta are $|k_j|/g\ll 1$. This allows us to expand Eqs. (\[BAE1\]) and (\[BAE2\]) to first order in $1/\gamma$ and $1/N$. The ground state quasi-momenta are then given by: $$\begin{aligned}
k_j^0 L=2\pi I_j^{0}\left(1-2/\gamma\right).
\label{k0}\end{aligned}$$ Here we used the relation $\sum_l \arctan(2(k_j^{0}-k_l^{0})/g)\simeq
2Nk_j^0/g-2(\sum_l k_l^0)/g=2Nk_j^0/g$, which is a consequence of the vanishing ground state momentum. Similarly, the excited state quasi-momenta obey the equations: $$\begin{aligned}
k_j L = \left(1-\frac{2}{\gamma}\right)2\pi I_j +\frac{2pL}{N\gamma}
-\pi+\frac{1}{\tilde{\lambda}}\left(1+\frac{k_jL}{\gamma N
\tilde{\lambda}} \right),
\label{k}\end{aligned}$$ where $p$ is given by Eq. (\[mom\]). Neglecting quasi-momenta $k_l$ in the argument of arctangent in the BA equation (\[BAE2\]), we obtain the excited state spin rapidity: $$2\pi J=2N\arctan(2\tilde{\lambda})\simeq \pi N -N/\tilde{\lambda}
\label{J}$$ Equations (\[pJ\]) and (\[J\]) then give: $$\tilde{\lambda}=N/pL,
\label{lambda}$$ which justifies that $\tilde{\lambda} \gg 1$ for $|p|/n \ll 1$. Combining this result with Eq. (\[k\]) shows that $|k_j|/g \ll 1$, as anticipated. Let us now define the shift of the quasi-momenta $\Delta k_j \equiv k_j-k_j^{0}$. Taking the difference between equations (\[k\]) and (\[k0\]), we find: $$\Delta k_j=\frac{1}{L \tilde{\lambda}}+\frac{k_j^{0}}{\gamma N
\tilde{\lambda}^2} +\frac{2p}{\gamma N}-\frac{2\pi}{L \gamma}
\label{Dkj}$$ where we used that $I_j-I_j^{0}=1/2$. We can now compute the energy of the spin wave, as defined in Eq. (\[defiem\]): $$\varepsilon_p= \sum_{j=1}^{N} \left[ 2k_j^{0} \Delta k_j + (\Delta
k_j)^2 \right].$$ Using Eq. (\[Dkj\]) for $\Delta k_j$ and Eq. (\[lambda\]) for $\tilde{\lambda}$ gives $\varepsilon_p=p^2\left(1/N+2\pi^2/3\gamma \right)$. Note that the last two terms in the right hand side of Eq. (\[Dkj\]) give no contribution, as the ground state momentum is zero. According to the definition (\[defiem\]), the inverse effective mass is therefore: $$\frac{m}{m^{*}}=\frac{1}{N}+\frac{2\pi^2}{3\gamma},
\label{strongres}$$ where we restored the units. Remarkably, the effective mass reaches the total mass $Nm$ for $\gamma\to \infty$: the bosons are impenetrable and therefore a down spin boson can move on a ring only if all other bosons move as well.
In the opposite limit of weak interactions it is possible to compute the effective mass from the Bogoliubov approach [@LP]. The validity of this procedure when considering a 1D Bose gas, i.e. in the absence of a true Bose-Einstein condensate, is justified in [@Popov]. The Hamiltonian of the system can be written as $H_0+H_{int}$, where $H_0$ is the Hamiltonian of free Bogoliubov quasiparticles and free spin waves: $$H_0=\sum_{p}\epsilon_p \alpha_{p}^{\dagger}\alpha_{p}
+\sum_{p}e_p \beta_{p}^{\dagger}\beta_{p},$$ with $\alpha_p,\beta_p$ being the Bogoliubov quasiparticle and the spin wave field operators, $\epsilon_p = \sqrt{e_p(e_p+2gn)}$ the Bogoliubov spectrum, and $e_p = p^2/2m$ the spectrum of free spin waves [@footnote2]. The Hamiltonian $H_{int}$ describes the interaction between Bogoliubov quasiparticles and spin waves and provides corrections to the dispersion relations $\epsilon_p$ and $e_p$. The most important part of $H_{int}$ reads: $$H_{int}=g\sqrt{\frac{n}{L}}\sum_{k,q\neq 0}
\left(u_q\alpha_{q}^{\dagger}-v_q\alpha_{-q}\right)
\beta_{k-q}^{\dagger}\beta_{k}+\text{h.c.},$$ where $u_q$ and $v_q$ are the $u,v$ Bogoliubov coefficients satisfying the relations $u_q+v_q=\sqrt{\epsilon_q/e_q}$ and $u_q-v_q=\sqrt{e_q/\epsilon_q}$ [@LP]. Neglected terms contribute only to higher orders in the coupling constant. To second order in perturbation theory, in the thermodynamic limit the presence of a spin wave changes the energy of the system by: $$\label{prec}
\Delta E(p)=e_p +\frac{g^2n}{2\pi\hbar}\int dq\;
\frac{e_q}{\epsilon_q}\; \frac{1}{e_p-[\epsilon_q+e_{p+q}]}.$$ In order to calculate a correction to the effective mass of the spin wave, we expand Eq. (\[prec\]) in the limit of $p\rightarrow 0$. Terms which do not depend on $p$ modify the ground state energy, linear terms vanish, and quadratic terms modify the spin wave spectrum as follows: $$\varepsilon_p=e_p \left(1-\frac{4g^2 n}{\pi\hbar}\int_{0}^{\infty}
dq\; \frac{e_q}{\epsilon_q}\; \frac{e_q}{[\epsilon_q+e_q]^3} \right),$$ where the main contribution to the integral comes from momenta $q\sim
\sqrt{mgn}$. Using the definition (\[defiem\]), we then obtain the inverse effective mass: $$\frac{m}{m^{*}}=1-\frac{2\sqrt{\gamma}}{\pi}\int_0^\infty
dx\,\frac{(\sqrt{1+x^2}-x)^3}{\sqrt{1+x^2}}
=1-\frac{2\sqrt{\gamma}}{3\pi},
\label{weakres}$$ which clearly shows non-analytical corrections to the bare mass due to correlations between particles. This result can also be obtained directly from the BA equations.
![Inverse effective mass $m/m^*$ as a function of the dimensionless coupling constant $\gamma$ (logarithmic scale). The stars ($*$) show numerical results for $N=111$ particles, the solid curve represents the behavior in the strong coupling limit (Eq. (\[strongres\])), and the dashed curve the behavior for a weak coupling (Eq. (\[weakres\])). []{data-label="effmass111"}](effmass111.eps){width="8cm"}
For intermediate couplings, we obtained the effective mass by numerically solving the BA equations (\[BAE1\]) and (\[BAE2\]). Our results are displayed in Fig.\[effmass111\]. Note that when solving the BA equations, one should take care of choosing $N^{-2}\ll \gamma \ll N^2$. Indeed, if $\gamma < N^{-2}$, the potential energy per particle in the weak coupling limit is lower than the zero point kinetic energy $\hbar^2/m L^2$ and the gas is therefore non-interacting (effectively $\gamma=0$). In the strong coupling limit and for the same reason, if $\gamma>N^2$, the system behaves as a TG gas (effectively $\gamma=\infty$).
We now turn to harmonically trapped bosons in the TG regime and rely on spin hydrodynamics introduced for uniform systems [@Halperin]. As the ground state is fully polarized we assume the equilibrium (longitudinal) spin density $\vec{S} (x)=n(x)\hat{e}_3$ and study small transverse spin density fluctuations $\delta\vec{
S}(x,t)=\delta S_1\hat{e}_1+\delta S_2\hat{e}_2 $, where $\hat{e}_1,\hat{e}_2,\hat{e}_3$ form an orthonormal basis in the spin space. For a large $N$, the equilibrium density profile $n(x)$ in a harmonic trapping potential $V(x)=m\omega^2x^2/2$ is given by the Thomas-Fermi expression $$\label{eq:dens_prof}
n(x)=n_0\sqrt{1-(x/R)^2}.$$ Here $n_0=n(0)$ is the density in the center of the trap and $R=\sqrt{2\hbar N/m\omega}$ is the Thomas-Fermi radius. For a strong but finite coupling Eq. (\[eq:dens\_prof\]) represents the leading term, with corrections proportional to inverse powers of $\gamma_0=mg/\hbar^2
n_0$. The spin density fluctuations $\delta\vec{ S}$ obey the following linearized Landau-Lifshitz equations [@Halperin]: $$\delta \dot{S}_{1,2}=\mp\frac{\hbar}{2}\partial_x\frac{
n(x)}{m^*(x)} \partial_x \frac{\delta S_{2,1}}{n(x)}
.
\label{eq:hydro_spin}$$ In the TG regime the effective mass entering the equation of motion (\[eq:hydro\_spin\]) depends on the density profile $n(x)$ as $$\label{eq:eff_mass_trap}
m^* (x)/m\approx 3\gamma (x)/2\pi^2=3 m g /2\pi^2\hbar^2 n(x).$$ Using the density profile (\[eq:dens\_prof\]) and introducing a complex function $$\label{eq:psi}
n(x) \Phi(x,t)=\delta S_1 (x,t)+i \delta S_2 (x,t),$$ one obtains from Eqs. (\[eq:hydro\_spin\]): $$\label{eq:motion_psi}
i\dot{\Phi}=\Omega\Phi =-\frac{\pi^2}{6}\frac{\omega}{\gamma_0 N}
\frac{1}{\sqrt{1-X^2}}\partial_X\left(1-X^2\right) \partial_X \Phi,$$ where $X=x/R$ is the dimensionless coordinate, and we assumed the stationary time dependence $\Phi (X,t)=
e^{-i\Omega t}\Phi(X)$. Equation (\[eq:motion\_psi\]) shows that the typical frequency scale of the isospin excitations is given by $\omega/\gamma_0 N$, which is smaller than the scale $\omega$ of acoustic frequencies by a large factor $\gamma_0 N$. The exact solution to this equation was obtained numerically using the shooting method, and the spectrum shows only a small difference from the semi-classical result $$\label{eq:semiclass}
\Omega_j = \frac{A\omega}{\gamma_0 N}\left(j+\frac{1}{2}\right)^2,
\qquad j=0,1,2,\ldots,$$ where the numerical factor is $A=\pi^5/48 \Gamma^4(3/4)\approx 2.83$. For $\omega \sim 100$ Hz, $\gamma_0 \sim 10$ and $N \sim 100$ as in the experiment [@Weiss], the lowest eigenfrequencies $\Omega_j$ are two or three orders of magnitude smaller than acoustic frequencies and are $\sim 0.1$ Hz.
In conclusion, we have found extremely slow (iso)spin dynamics in the strong coupling TG regime, originating from a very large effective mass of spin waves. In an experiment with ultra-cold bosons, this should show up as a spectacular isospin-density separation: an initial wave packet splits into a fast acoustic wave traveling at the Fermi velocity and an extremely slow spin wave [@Recati]. One can even think of “freezing" the spin transport, which in experiments with two-component 1D Bose gases will correspond to freezing relative oscillations of the two components.
We thank F. Laloë and W. Zwerger for useful discussions. This work was supported by the Ministère de la Recherche (grant ACI Nanoscience 201), and by the Nederlandse Stichting voor Fundamenteel Onderzoek der Materie (FOM). The studies were performed in part at Laboratoire Kastler Brossel, ENS Paris. LPTMS is UMR 8626 of CNRS and Université Paris XI. LPS is UMR 8502 of CNRS and Université Paris XI.
[99]{}
F. Schreck *et al.*, Phys. Rev. Lett. **87**, 080403 (2001); A. Görlitz *et al.*, *ibid.* **87**, 130402 (2001); M. Greiner *et al.*, *ibid.* **87**, 160405 (2001). H. Moritz et al., Phys. Rev. Lett. **91**, 250402 (2003); B. Laburthe Tolra et al., *ibid.* **92**, 190401 (2004).
B. Paredes et al., Nature [**429**]{}, 277 (2004).
T. Kinoshita, T. Wenger and D.S. Weiss, Science [**305**]{}, 1125 (2004).
M. Girardeau, J. Math. Phys. **1**, 516 (1960).
E.H. Lieb and W. Liniger, Phys. Rev. **130**, 1605 (1963).
E.H. Lieb, Phys. Rev. **130**, 1616 (1963).
M. Olshanii, Phys. Rev. Lett. **81**, 938 (1998).
D.S. Petrov, G.V. Shlyapnikov and J.T.M. Walraven, Phys. Rev. Lett. **85**, 3745 (2000).
V. Dunjko, V. Lorent and M. Olshanii, Phys. Rev. Lett. **86**, 5413 (2001).
D.S. Hall *et al.*, Phys. Rev. Lett. **81**, 1539 (1998); D.S. Hall *et al.*, *ibid.* **81**, 1543 (1998).
H.J. Lewandowski *et al.*, Phys. Rev. Lett. **88**, 070403 (2002); M.Ö. Oktel and L.S. Levitov, *ibid.* **88**, 230403 (2002); J.N. Fuchs, D.M. Gangardt and F. Laloë, *ibid.* **88**, 230404 (2002); J.E. Williams, T. Nikuni and C.W. Clark, *ibid.* **88**, 230405 (2002); J.M. McGuirk *et al.*, *ibid.* **89**, 090402 (2002).
T. Nikuni and J.E. Williams, J. Low Temp. Phys. **133**, 323 (2003).
W.B. Colson and A. Fetter, J. Low Temp. Phys. **33**, 231 (1978); T.-L. Ho and V.B. Shenoy, Phys. Rev. Lett. **77**, 3276 (1996); B.D. Esry *et al.*, *ibid.* **78**, 3594 (1997); C.K. Law *et al.*, *ibid.* **79**, 3105 (1997); H. Pu and N.P. Bigelow, *ibid.* **80**, 1130 (1998).
T.-L. Ho and S.K. Yip, Phys. Rev. Lett. **84**, 4031 (2000); A.B. Kuklov and B.V. Svistunov, *ibid.* **89**, 170403 (2002).
M. Gaudin, Phys. Lett. A **24**, 55 (1967); *La Fonction d’onde de Bethe*, (Masson, Paris, 1983).
C.N. Yang, Phys. Rev. Lett. **19**, 1312 (1967).
Y.Q. Li *et al.*, Europhys. Lett. **61**, 368-374 (2003).
E. Eisenberg and E.H. Lieb, Phys. Rev. Lett. **89**, 220403 (2002).
K. Yang and Y.Q. Li, Int. J. Mod. Phys. B **17**, 1027 (2003).
In order to excite an isospinon alone we choose density quantum numbers $I_j=j-N/2$ and the spin quantum number $J=1-N/2+l$, where $l$ varies between $1$ and $N-2$ [@LGYE].
B.I. Halperin and P.C. Hohenberg, Phys. Rev. **188**, 898 (1969); B.I. Halperin, Phys. Rev. B [**11**]{}, 178 (1975).
See, e.g., E.M. Lifshitz and L.P. Pitaevskii, *Statistical Physics, Part 2* (Butterworth-Heinemann, 1980).
V.N. Popov, *Functional Integrals in Quantum Field Theory and Statistical Physics* (Reidel, Dordrecht, 1983).
The mechanism responsible for spin waves in the GP regime is the so-called “quantum torque” (see e.g. [@Nikuni2003]).
We find that in the 1D spinor Bose gas the “spin-charge separation” is significantly stronger than in the recent proposal for spin $1/2$ atomic *fermions*, see A. Recati *et al.*, Phys. Rev. Lett. **90**, 020401 (2003).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We present a constructive solution to the $N$-representability problem—a full characterization of the conditions for constraining the two-electron reduced density matrix (2-RDM) to represent an $N$-electron density matrix. Previously known conditions, while rigorous, were incomplete. Here we derive a hierarchy of constraints built upon (i) the bipolar theorem and (ii) tensor decompositions of model Hamiltonians. Existing conditions $D$, $Q$, $G$, $T1$, and $T2$, known classical conditions, and new conditions appear naturally. Subsets of the conditions are amenable to polynomial-time computations of strongly correlated systems.'
author:
- 'David A. Mazziotti'
date: 'Submitted November 8, 2011; Published [*Phys. Rev. Lett*]{} [**108**]{}, 263002 (2012)'
title: 'Structure of Fermionic Density Matrices: Complete $N$-representability Conditions'
---
The wavefunction of a many-electron quantum system contains significantly more information than necessary for the calculation of energies and properties. In 1955 Mayer proposed in [*Physical Review*]{} computing the ground-state energy variationally as a functional of the two-electron reduced density matrix (2-RDM) which, unlike the wavefunction, scales polynomially with the number $N$ of electrons [@RDM; @CY00; @M55]. However, the 2-electron density matrix must be constrained to represent a many-electron (or $N$-electron) density matrix (or wavefunction); otherwise, the minimized energy is unphysically below the ground-state energy for $N>2$. Coleman called these constraints [*$N$-representability conditions*]{} [@C63], and the search for them became known as the $N$-representability problem [@GP64; @H78; @E78; @E79; @P78; @R07]. In 1995 the National Research Council ranked the $N$-representability problem as one of the top unsolved theoretical problems in chemical physics [@NRC]. While progress was limited for many years, recent advances in theory and optimization [@EJ00; @N01; @M04; @P04; @C06; @E07; @A09; @S10; @M11] have enabled the application of the variational 2-RDM method to studying strong correlation in quantum phase transitions [@GM06], quantum dots [@RM09], polyaromatic hydrocarbons [@GM08], firefly bioluminescence [@GM10], and metal-to-insulator transitions [@SGM10].
Despite the recent computational results with 2-RDM methods, a complete set of $N$-representability conditions on the 2-RDM—not dependent upon higher-order RDMs—has remained unknown. While formal solutions of the $N$-representability problem were developed in the 1960s [@GP64; @K67], practically they required the $N$-electron density matrix [@RDM; @CY00]. In this Letter we present a constructive solution of the $N$-representability problem that generates a complete set of $N$-representability conditions on the 2-RDM. The approach is applicable to generating the $N$-representability conditions on the $p$-RDM for any $p \le N$. The conditions arise naturally as a hierarchy of constraints on the 2-RDM, which we label the $(2,q)$-positivity conditions, where the $(2,2)$- and $(2,3)$-positivity conditions include the already known $D$, $Q$, $G$, $T1$, and $T2$ conditions [@C63; @GP64; @E78; @P04]. The second number in $(2,q)$ corresponds to the higher $q$-RDM which serves as the starting point for the derivation of the condition.
A key advance in extending the $(2,q)$-positivity conditions for $q>3$ is the use of tensor decompositions in the model Hamiltonians that expose the boundary of the $N$-representable 2-RDM set. The decompositions allow the terms in the model Hamiltonians to have no more than two-body interactions through the cancelation of all higher 3-to-$q$-body terms. A second important element is the recognition that when $q=r$ where $r$ is the rank of the one-electron basis set the positivity conditions are complete. The hierarchy of conditions can be thought of as a collection of model Hamiltonians [@P78]. For example, the ‘basic’ (2,2)-positivity conditions are both necessary and sufficient constraints for computing the ground-state energies of pairing model Hamiltonians [@CY00; @M04], often employed in describing long-range order and superconductivity.
Consider a quantum system composed of $N$ fermions. A matrix is a fermionic [*density matrix*]{} if and only if it is: ([*i*]{}) Hermitian, ([*ii*]{}) normalized (fixed trace), ([*iii*]{}) antisymmetric in the exchange of particles, and ([*iv*]{}) positive semidefinite. A matrix is [*positive semidefinite*]{} if and only if its eigenvalues are nonnegative. The $p$-particle reduced density matrix ($p$-RDM) can be obtained from the $N$-particle density matrix by integrating over all but the first $p$ particles $$\label{eq:Dp} {}^{p} D = {N \choose p} \int{ {}^{N} D \, d(p+1)
\dots dN } .$$ The set of ${}^{N} D$ is a convex set which we denote as $P^{N}$ while the set ${}^{p} D$ is a convex set which we denote as $P^{p}_{N}$, the set of $N$-representable $p$-particle density matrices. A set is [*convex*]{} if and only if the convex combination of any two members of the set is also contained in the set $$w \, {}^{N} D_{1} + (1-w) \, {}^{N} D_{2} \in P^{N},$$ where $0 \le w \le 1$. The integration in Eq. (\[eq:Dp\]) defines a linear mapping from $P^{N}$ to $P^{p}_{N}$, which preserves its convexity.
The energy of a quantum system in a stationary state can be computed from the Hamiltonian traced against the state’s density matrix. For a system of $N$ fermions we have $$\label{eq:EN} E = {\rm Tr}({\hat H} \, {}^{N} D) .$$ If the Hamiltonian is a $p$-body operator, meaning that it has at most $p$-particle interactions, then the energy can be written as a functional of only the $p$-RDM $$\label{eq:Ep} E = {\rm Tr}({\hat H} \, {}^{p} D) .$$ For a system of $N$ electrons the Hamiltonian generally has at most pairwise interactions, and hence, the energy can be expressed as a linear functional of the 2-RDM. Except when $N=2$, however, minimizing the energy as a functional of a two-electron density matrix ${}^{2} D \in P^{2}$ yields an energy that is much too low. To obtain the correct ground-state energy, we must constrain the two-electron density matrix to be $N$-representable, that is $^{2} D
\in P^{2}_{N}$.
Based on the equivalence of the energy expectation values in Eqs. (\[eq:EN\]) and (\[eq:Ep\]), we can use the set $P^{p}_{N}$ of $N$-representable $p$-particle density matrices to define a set ${P^{p}_{N}}^{*}$ of $p$-particle (Hamiltonian) operators $^{p}
{\hat O}$ that are positive semidefinite in their trace with any $N$-particle density matrix $$\label{eq:Oset} {P^{p}_{N}}^{*} = \{ ^{p} {\hat O} | {\rm Tr}(^{p}
{\hat O} \, ^{p} D) \ge 0~{\rm for~all}~^{p} D \in {P^{p}_{N}} \}.$$ The set ${P^{p}_{N}}^{*}$ is said to be the [*polar*]{} (or dual) of the set $P^{p}_{N}$. Importantly, by the [*bipolar theorem*]{} [@K67; @R71], the set ${P^{p}_{N}}^{*}$ also fully defines its polar set $P^{p}_{N}$ as follows $$\label{eq:Dset} P^{p}_{N} = \{ ^{p} D | {\rm Tr}(^{p} {\hat O} \,
^{p} D) \ge 0~{\rm for~all}~^{p} {\hat O} \in {{P^{p}_{N}}^{*}} \}.$$ By Eq. (\[eq:Dset\]) we have a complete characterization of the $N$-representable $p$-RDMs from a knowledge of all operators $^{p}
{\hat O} \in {P^{p}_{N}}^{*}$ [@K67]. This analysis shows formally that there exists a solution to the $N$-representability problem [@GP64; @K67], but it does not provide a mechanism for characterizing the set ${P^{p}_{N}}^{*}$.
![The convex set ${P^{2}_{N}}^{*}$ of 2-body operators that are positive semidefinite in their trace with any $N$-particle density matrix is contained within the convex set ${P^{3}_{N}}^{*}$ of analogous 3-body operators, which in turn is contained within the set ${P^{r}_{N}}^{*}$. Hence, the extreme points of ${P^{2}_{N}}^{*}$ can be characterized completely by the convex combination of the extreme points of ${P^{r}_{N}}^{*}$, which are given by Eq. (\[eq:rpos\]).[]{data-label="f:n2"}](polar_set_v3.eps)
To characterize ${P^{p}_{N}}^{*}$, we assume that the $N$-fermion quantum system has $r$ orbitals and hence, $r-N$ holes. A convex set can be defined by the enumeration of its [*extreme elements*]{}, that is the elements (or members) that cannot be expressed by a convex combination of other elements [@CY00; @R71]. The definition of ${P^{p}_{N}}^{*}$ in Eq. (\[eq:Oset\]) for $p \le N$ can be extended in second quantization to include $p > N$ $$\label{eq:Oset2} {P^{p}_{N}}^{*} = \{ ^{p} {\hat O} | {\rm Tr}(^{p}
{\hat O} \, ^{N} D) \ge 0~{\rm for~all}~^{N} D \}$$ with the $^{p} {\hat O}$ being polynomials in creation and annihilation operators of degree $2p$. Because in second quantization the value of $N$ is defined in the density matrices $^{N} D$ rather than in the operators $^{p} {\hat O}$ [@S89], the set ${P^{p}_{N}}^{*}$ provides complete $N$-representability conditions on the $p$-RDM for any $N$ between 2 and $r$. The extreme operators in the set ${P^{r}_{N}}^{*}$ can be written as Hermitian squares of operators [@H02] $$\label{eq:rpos} {^{r} {\hat O}_{i}} = {^{r} {\hat C}_{i}} \, {^{r}
{\hat C}_{i}^{\dagger}},$$ where the ${}^{r} {\hat C}_{i}$ are polynomials in the creation and annihilation operators of degree less than or equal to $r$ (i.e., Eqs. (\[eq:T21\]) and (\[eq:T22\])). Because any operator ${}^{p} {\hat C}$ with $p>r$ reduces to a polynomial of degree $r$ in its operation on any ${}^{N} D$, the sets ${P^{p}_{N}}^{*}$ with $p>r$ do not contain additional information about the positivity of ${}^{N} D$. To establish this reduction, we rearrange terms in ${}^{p} {\hat C}$ of degree greater than $r$ into a normal order with either more than $N$ annihilation operators to the right of the creation operators or more than $r-N$ creation operators to the right of the annihilation operators; in either situation, the terms of degree greater than $r$ vanish in their operation upon any $^{N}
D$.
The operators ${}^{p} {\hat O}$ that constrain the $p$-RDM to be $N$-representable in Eq. (\[eq:Dset\]) are also necessary to constrain the $q$-RDM to be $N$-representable where $q>p$; formally, each ${}^{p} {\hat O} \in {P^{p}_{N}}^{*}$ can be lifted by inserting the number operator to the $(q-p)$ power to form a ${}^{q}
{\hat O} \in {P^{q}_{N}}^{*}$ [@M04]. Therefore, as illustrated in Fig. 1, we have the following set relations $${P^{2}_{N}}^{*} \subseteq {P^{3}_{N}}^{*} \subseteq {P^{p}_{N}}^{*}
... \subseteq {P^{r}_{N}}^{*} .$$ Consequently, extreme operators $^{r} {\hat O}_{i}$ of ${P^{r}_{N}}^{*}$ can be combined convexly to produce all $p$-body operators $^{p} {\hat O} \in {P^{p}_{N}}^{*}$, and hence, the extreme points of ${P^{p}_{N}}^{*}$ can be characterized completely by the convex combination of the extreme points of ${P^{r}_{N}}^{*}$. More generally, convex combinations of extreme $^{q} {\hat O}_{i} \in {P^{q}_{N}}^{*}$ generate all $p$-body operators $^{p} {\hat O} \in {P^{p}_{N}}^{*}$ for $p < q$. Depending upon the order of the creation and annihilation operators in $^{r}
{\hat O}_{i}$, the normal-ordered terms will have either positive or negative coefficients. Convex combinations of the $^{r} {\hat
O}_{i}$ can be chosen to cancel the coefficients of all terms of degree greater than $p$. Extreme elements are generated from the minimum number of convex combinations to effect the cancelation. This characterization of the set ${P^{p}_{N}}^{*}$ provides a [*constructive solution*]{} of the $N$-representability problem for the $p$-RDM.
The constructive solution—convex combinations of the operators in Eq. (\[eq:rpos\])—generates the existing $N$-representability conditions as well as new conditions. The [*(1,1)-positivity conditions*]{} [@C63] are derivable from the subset of $^{r} {\hat
C}_{i}$ operators in Eq. (\[eq:rpos\]) of degree 1 $$\begin{aligned}
{\hat C}_{D} & = & \sum_{j}{ b_{j} {\hat a}^{\dagger}_{j} } \\
{\hat C}_{Q} & = & \sum_{j}{ b_{j} {\hat a}_{j} } .\end{aligned}$$ Keeping the trace of the corresponding one-body operators $^{1}
{\hat O}_{D}$ and $^{1} {\hat O}_{Q}$ against the 1-RDM nonnegative for all values of $b_{j}$ yields the conditions, $^{1} D \succeq 0$ and $^{1} Q \succeq 0$, where ${}^{1} D$ and $^{1} Q$ are matrix representations of the 1-particle and the 1-hole RDMs and the symbol $\succeq$ indicates that the matrix is constrained to be positive semidefinite.
Similarly, the [*(2,2)-positivity conditions*]{} [@GP64] follow from considering the $^{r} {\hat C}_{i}$ operators of degree 2 in Eq. (\[eq:rpos\]) $$\begin{aligned}
{\hat C}_{D} & = & \sum_{jk}{ b_{jk} {\hat a}^{\dagger}_{j} {\hat a}^{\dagger}_{k} } \\
{\hat C}_{Q} & = & \sum_{jk}{ b_{jk} {\hat a}_{j} {\hat a}_{k} } \\
{\hat C}_{G} & = & \sum_{jk}{ b_{jk} {\hat a}^{\dagger}_{j} {\hat
a}_{k} } .\end{aligned}$$ Restricting the trace of the corresponding two-body operators $^{2}
{\hat O}_{D}$, $^{2} {\hat O}_{Q}$, and $^{2} {\hat O}_{G}$ against the 2-RDM to be nonnegative for all values of $b_{jk}$ defines the conditions, $^{2} D \succeq 0$, $^{2} Q \succeq 0$, and $^{2} G
\succeq 0$, which constrain the probabilities for finding two particles, two holes, and a particle-hole pair to be nonnegative, respectively.
-------------------------------------------------------------------
(2,4)-Positivity Conditions
-------------------------------------------------------------------
${\rm Tr}( (3 {\hat C}_{\rm xxxx} {\hat C}_{\rm xxxx}^{\dagger} +
{\hat C}_{\rm xxxo} {\hat C}_{\rm xxxo}^{\dagger} + {\hat C}_{\rm
xxox} {\hat C}_{\rm xxox}^{\dagger} + {\hat C}_{\rm xoxx} {\hat
C}_{\rm xoxx}^{\dagger} + {\hat C}_{\rm oxxx} {\hat C}_{\rm
oxxx}^{\dagger} + {\hat C}_{\rm oooo} {\hat
C}_{\rm oooo}^{\dagger}) \, {}^{2} D) \ge 0$
${\rm Tr}( (3 {\hat C}_{\rm xxxo} {\hat C}_{\rm xxxo}^{\dagger} +
{\hat C}_{\rm xxxx} {\hat C}_{\rm xxxx}^{\dagger} + {\hat C}_{\rm
xxoo} {\hat C}_{\rm xxoo}^{\dagger} + {\hat C}_{\rm xoxo} {\hat
C}_{\rm xoxo}^{\dagger} + {\hat C}_{\rm oxxo} {\hat C}_{\rm
oxxo}^{\dagger} + {\hat C}_{\rm ooox} {\hat
C}_{\rm ooox}^{\dagger}) \, {}^{2} D) \ge 0$
${\rm Tr}( (3 {\hat C}_{\rm xxox} {\hat C}_{\rm xxox}^{\dagger} +
{\hat C}_{\rm xxoo} {\hat C}_{\rm xxoo}^{\dagger} + {\hat C}_{\rm
xxxx} {\hat C}_{\rm xxxx}^{\dagger} + {\hat C}_{\rm xoox} {\hat
C}_{\rm xoox}^{\dagger} + {\hat C}_{\rm oxox} {\hat C}_{\rm
oxox}^{\dagger} + {\hat C}_{\rm ooxo} {\hat
C}_{\rm ooxo}^{\dagger}) \, {}^{2} D) \ge 0$
${\rm Tr}( (3 {\hat C}_{\rm xoxx} {\hat C}_{\rm xoxx}^{\dagger} +
{\hat C}_{\rm xoxo} {\hat C}_{\rm xoxo}^{\dagger} + {\hat C}_{\rm
xoox} {\hat C}_{\rm xoox}^{\dagger} + {\hat C}_{\rm xxxx} {\hat
C}_{\rm xxxx}^{\dagger} + {\hat C}_{\rm ooxx} {\hat C}_{\rm
ooxx}^{\dagger} + {\hat C}_{\rm oxoo} {\hat
C}_{\rm oxoo}^{\dagger}) \, {}^{2} D) \ge 0$
${\rm Tr}( (3 {\hat C}_{\rm oxxx} {\hat C}_{\rm oxxx}^{\dagger} +
{\hat C}_{\rm oxxo} {\hat C}_{\rm oxxo}^{\dagger} + {\hat C}_{\rm
oxox} {\hat C}_{\rm oxox}^{\dagger} + {\hat C}_{\rm ooxx} {\hat
C}_{\rm ooxx}^{\dagger} + {\hat C}_{\rm xxxx} {\hat C}_{\rm
xxxx}^{\dagger} + {\hat C}_{\rm xooo} {\hat
C}_{\rm xooo}^{\dagger}) \, {}^{2} D) \ge 0$
${\rm Tr}( (3 {\hat C}_{\rm xxoo} {\hat C}_{\rm xxoo}^{\dagger} +
{\hat C}_{\rm xxox} {\hat C}_{\rm xxox}^{\dagger} + {\hat C}_{\rm
xxxo} {\hat C}_{\rm xxxo}^{\dagger} + {\hat C}_{\rm xooo} {\hat
C}_{\rm xooo}^{\dagger} + {\hat C}_{\rm oxoo} {\hat C}_{\rm
oxoo}^{\dagger} + {\hat C}_{\rm ooxx} {\hat
C}_{\rm ooxx}^{\dagger}) \, {}^{2} D) \ge 0$
${\rm Tr}( (3 {\hat C}_{\rm xoox} {\hat C}_{\rm xoox}^{\dagger} +
{\hat C}_{\rm xooo} {\hat C}_{\rm xooo}^{\dagger} + {\hat C}_{\rm
xoxx} {\hat C}_{\rm xoxx}^{\dagger} + {\hat C}_{\rm xxox} {\hat
C}_{\rm xxox}^{\dagger} + {\hat C}_{\rm ooox} {\hat C}_{\rm
ooox}^{\dagger} + {\hat C}_{\rm oxxo} {\hat
C}_{\rm oxxo}^{\dagger}) \, {}^{2} D) \ge 0$
${\rm Tr}( (3 {\hat C}_{\rm xoxo} {\hat C}_{\rm xoxo}^{\dagger} +
{\hat C}_{\rm xoxx} {\hat C}_{\rm xoxx}^{\dagger} + {\hat C}_{\rm
xooo} {\hat C}_{\rm xooo}^{\dagger} + {\hat C}_{\rm xxxo} {\hat
C}_{\rm xxxo}^{\dagger} + {\hat C}_{\rm ooxo} {\hat C}_{\rm
ooxo}^{\dagger} + {\hat C}_{\rm oxox} {\hat
C}_{\rm oxox}^{\dagger}) \, {}^{2} D) \ge 0$
-------------------------------------------------------------------
In general, the $(q,q)$-positivity conditions [@EJ00; @M04] follow from restricting all $q$-body operators ${}^{q} {\hat O}$ in Eq. (\[eq:rpos\]) to be nonnegative in their trace against the $q$-RDM [@M04]. While the $(q,q)$-positive operators are not two-body operators for $q>2$, convex combinations of them generate two-body operators $^{2} {\hat O} \in {P^{2}_{N}}^{*}$ that enforce the $N$-representability of the 2-RDM. We refer to necessary $N$-representability conditions arising from convex combinations of $(q,q)$-positivity conditions as $(2,q)$-positivity conditions.
The simplest such constraints, the [*(2,3)-positivity conditions*]{}, arise from keeping convex combinations of 3-body operators in Eq. (\[eq:rpos\]) nonnegative; for example, $$\begin{aligned}
^{2} {\hat O}_{T1} & = & \frac{1}{2} ( {\hat C}_{T1,1} \, {\hat
C}_{T1,1}^{\dagger} + {\hat C}_{T1,2} {\hat
C}_{T1,2}^{\dagger} ) \\
^{2} {\hat O}_{T2} & = & \frac{1}{2} ( {\hat C}_{T2,1} \, {\hat
C}_{T2,1}^{\dagger} + {\hat C}_{T2,2} {\hat C}_{T2,2}^{\dagger} )\end{aligned}$$ where $$\begin{aligned}
{\hat C}_{T1,1} & = & \sum_{jkl}{ b_{jkl} {\hat a}^{\dagger}_{j}
{\hat a}^{\dagger}_{k} {\hat a}^{\dagger}_{l} } \\
{\hat C}_{T1,2} & = & \sum_{jkl}{ b^{*}_{jkl} {\hat a}_{j} {\hat a}_{k} {\hat a}_{l}} \\
{\hat C}_{T2,1} & = & \sum_{jkl}{ b_{jkl} {\hat a}^{\dagger}_{j}
{\hat a}^{\dagger}_{k} {\hat a}_{l} } + \sum_{j}{ b_{j} {\hat a}^{\dagger}_{j} } \label{eq:T21} \\
{\hat C}_{T2,2} & = & \sum_{jkl}{ b^{*}_{jkl} {\hat a}_{j}
{\hat a}_{k} {\hat a}^{\dagger}_{l} } + \sum_{j}{d_{j} {\hat a}_{j}}
\label{eq:T22} .\end{aligned}$$ These conditions, known as the $T1$ and generalized $T2$ conditions were developed by Erdahl [@E78] and implemented by Zhao [*at al.*]{} [@P04] and Mazziotti [@M04]. In general, they significantly improve the accuracy of the 2-positivity conditions.
Although the constructive proof given above indicates that a complete set of $N$-representability conditions can be generated from convex combinations of extreme elements of ${P^{r}_{N}}^{*}$, additional conditions have not been discovered beyond the (2,2)- and (2,3)-positivity conditions. For example, what about (2,4)-positivity conditions—that is, $N$-representability constraints on the 2-RDM arising from convex combinations of 4-body operators in Eq. (\[eq:rpos\])? First, we derive a class of (3,4)-positivity conditions on the 3-RDM.
Consider the nonnegativity of the following operator ${\hat O}$ formed by the convex combination of a pair of 4-body operators from Eq. (\[eq:rpos\]) $$\label{eq:O} {\hat O} = \frac{1}{2} ( {\hat C}_{\rm xxxx} \, {\hat
C}_{\rm xxxx}^{\dagger} + {\hat C}_{\rm xooo} {\hat
C}_{\rm xooo}^{\dagger} )\\$$ where the symbols ${\rm x}$ and ${\rm o}$ represent creation and annihilation operators, respectively, in the ${\hat C}$ operators defined as follows $$\begin{aligned}
{\hat C}_{\rm xxxx} & = & \sum_{jklm}{ b_{jklm} {\hat
a}^{\dagger}_{j}
{\hat a}^{\dagger}_{k} {\hat a}^{\dagger}_{l} {\hat a}^{\dagger}_{m} } \\
{\hat C}_{\rm xooo} & = & \sum_{jklm}{ d_{jklm} {\hat
a}^{\dagger}_{j} {\hat a}_{k} {\hat a}_{l} {\hat a}_{m} } .\end{aligned}$$ Importantly, the expectation value of ${\hat O}$ with $d_{jklm} =
b_{jklm}$ requires the 4-RDM because the cumulant part ${}^{4}
\Delta$ of the 4-RDM [@M98; @RDM] does not vanish $$\sum_{jklmpqst}{ b_{jklm} b^{*}_{pqst} \, (^{4} \Delta^{jklm}_{pqst}
- ^{4} \Delta^{jqst}_{pklm} ) } \neq 0 .$$ To obtain additional $N$-representability conditions requires that the dependence of the ${\hat C}$ operators on the expansion coefficients be [*generalized from linear to nonlinear*]{}. Specifically, to obtain 3-RDM conditions beyond the (3,3)-positivity constraints, we must factor the 4-particle expansion coefficients $b_{jklm}$ and $d_{jklm}$ into products of 3- and 1-particle coefficients $b_{j} b_{klm}$ and $b_{j} b^{*}_{klm}$ which cause the cumulant part of the 4-RDM in $\langle \Psi | {\hat O} | \Psi
\rangle $ to vanish $$\sum_{jklmpqst}{ b_{j} b_{klm} b^{*}_{p} b^{*}_{qst} \, (^{4}
\Delta^{jklm}_{pqst} - ^{4} \Delta^{jklm}_{pqst} ) } = 0 .$$ The (3,4)-positivity condition, represented by Eq. (\[eq:O\]) and the tensor decomposition of the expansion coefficients, is part of a class of (3,4)-conditions that arises from all distinct combinations of two 4-particle metric matrices that differ from each other in the replacement of [*three*]{} second-quantized operators by their adjoints.
A class of [*(2,4)-positivity conditions*]{}, shown in Table \[t:24\], can be derived from convex combinations of the above (3,4)-positivity conditions that cancel the 3-particle operators, that is the products of six second-quantized operators. To effect the cancelation, the nonlinearity of the expansion coefficients of ${\hat C}$ must be increased from $b_{j} b_{klm}$ to $b_{j} c_{k} d_{l} e_{m}$. Specifically, the ${\hat C}$ operators in Table \[t:24\] are defined as $${\hat C}_{\rm uvwz} = \sum_{jklm}{ b^{\rm u}_{j} c^{\rm v}_{k}
d^{\rm w}_{l} e^{\rm z}_{m} {\hat a}^{\rm u}_{j} {\hat a}^{\rm
v}_{k} {\hat a}^{\rm w}_{l} {\hat a}^{\rm z}_{m} } ,$$ where ${\hat a}^{u}_{j}$ and $b^{\rm u}_{j}$ are ${\hat
a}^{\dagger}_{j}$ and $b^{*}_{j}$ if ${\rm u}={\rm x}$ and ${\hat
a}_{j}$ and $b_{j}$ if ${\rm u}={\rm o}$. Each of the eight (2,4)-positivity conditions in Table \[t:24\] generates an additional condition by switching all ${\rm x}$’s and ${\rm o}$’s in accordance with [*particle-hole duality*]{}, the symmetry between particles and holes. The (2,4)-conditions become the diagonal $N$-representability conditions [@E78; @Cuts; @D02; @KM08] when $b$, $c$, $d$, and $e$ are restricted to be unit vectors; they are more general than the unitarily invariant diagonal conditions because these four vectors are not required to be orthogonal. These (2,4)-positivity conditions are only representative of the process by which complete conditions can be constructed from the solution of the $N$-representability problem presented in this Letter. Additional (2,4)-conditions in this class can be generated from reordering creation and annihilation operators in the conditions of Table \[t:24\], and other extreme (2,4)-conditions can be constructed from lifting the (2,3)-conditions. A comprehensive list of (2,4)-positivity conditions as well as (2,3)-, (2,5)-, and (2,6)-positivity conditions, which are consistent with the constructive solution, will be presented elsewhere [@M12]. The (2,5)- and (2,6)-conditions include extensions of three and eighteen classes of known diagonal conditions, respectively.
The set ${P^{2}_{N}}^{*}$ of $N$-representability conditions on the 2-RDM contains the set ${C^{2}_{N}}^{*}$ of [*classical $N$-representability conditions*]{} [@E78; @Cuts; @D02; @KM08], which ensure that the two-electron reduced density function (2-RDF), the diagonal (classical) part of the 2-RDM, can be represented by the integration of a $N$-particle density function. In different fields the set $C^{2}_{N}$ of $N$-representable 2-RDF has been given different names: cut polytope [@Cuts] in combinatorial optimization and the correlation (or Boole) polytope [@Cuts; @P89] in the study of 0-1 programming or Bell’s inequalities. The set $C^{2}_{N}$, previously characterized, has important applications in global optimization including the search for the global energy minima of molecular clusters [@KM08], the study of classical fluids [@Fluid], the max-cut problem in circuit design and spin glasses [@Cuts], lattice holes in the geometry of numbers, pair density (2-RDF) functional theory [@D02], and the investigation of generalized Bell’s inequalities [@P89]. The characterization of the set $P^{2}_{N}$ of $N$-representable 2-RDMs represents a significant generalization of the solution of the classical $N$-representability problem (the Boole 0-1 programming problem). In addition to its potentially significant applications to the study of correlation in many-fermion quantum systems, knowledge of the set $P^{2}_{N}$ may have important applications to “quantum” analogues of problems in circuit design and the geometry of numbers.
The complete set of $N$-representability conditions firmly solidifies 2-RDM theory as a fundamental theory of many-body quantum mechanics with two-particle interactions. Rigorous lower bounds to the ground-state energy of strongly correlated quantum systems can be computed and improved in polynomial time from subsets of the complete $N$-representability conditions [@M11] (Minimizing the energy with a fully $N$-representable 2-RDM is a non-deterministic polynomial (NP) complete problem because $C^{2}_{N} \subset
P^{2}_{N}$ with optimization in $C^{2}_{N}$ known to be NP-complete [@Cuts]). The present result raises challenges and opportunities for future research that include (i) implementing the higher $N$-representability conditions which are not in the form of traditional semidefinite programming [@M04; @P04; @M11], and (ii) determining which of the new conditions are most appropriate for different problems in many-particle chemistry and physics. Beyond their potential computational applications, the complete $N$-representability conditions for fermionic density matrices provide new fundamental insight into many-electron quantum mechanics including the identification and measurement of correlation and entanglement.
The author thanks D. Herschbach, H. Rabitz, and A. Mazziotti for encouragement, and the NSF, ARO, Microsoft Corporation, Dreyfus Foundation, and David-Lucile Packard Foundation for support.
[99]{}
, edited by D. A. Mazziotti, Advances in Chemical Physics Vol. 134 (Wiley, New York, 2007).
A. J. Coleman and V. I. Yukalov, [*Reduced Density Matrices: Coulson’s Challenge*]{} (Springer, New York 2000).
J. E. Mayer, Phys. Rev. [**100**]{}, 1579 (1955).
A. J. Coleman, Rev. Mod. Phys. [**35**]{} 668 (1963).
C. Garrod and J. Percus, J. Math. Phys. [**5**]{}, 1756 (1964).
J. E. Harriman, Phys. Rev. A [**17**]{}, 1257 (1978).
R. M. Erdahl, Int. J. Quantum Chem. [**13**]{}, 697 (1978).
R. M. Erdahl, Rep. Math. Phys. [**15**]{}, 147 (1979).
J. K. Percus, Int. J. Quantum Chem. [**13**]{}, 89 (1978).
M. Rosina, Adv. Chem. Phys. [**134**]{}, 11 (2007).
F. H. Stillinger et al., [*Mathematical Challenges from Theoretical/Computational Chemistry*]{} (National Academic Press, Washington, D.C., 1995).
R. M. Erdahl and B. Jin in [*Many-electron Densities and Density Matrices*]{}, edited by J. Cioslowski, (Kluwer, Boston, 2000).
M. Nakata, H. Nakatsuji, M. Ehara, M. Fukuda, K. Nakata, and K. Fujisawa, J. Chem. Phys. **114**, 8282 (2001).
D. A. Mazziotti, Phys. Rev. A **65**, 062511 (2002); Phys. Rev. Lett. [**93**]{}, 213001 (2004); Phys. Rev. A **74**, 032501 (2006).
Z. Zhao, B. J. Braams, H. Fukuda, M. L. Overton, and J. K. Percus, J. Chem. Phys. [**120**]{}, 2095 (2004); M. Fukuda, B. J. Braams, M. Nakata, M. L. Overton, J. K. Percus, M. Yamashita, and Z. Zhao, Math. Program. Ser. B **109**, 553 (2007).
E. Cancés, G. Stoltz, and M. Lewin, J. Chem. Phys. **125**, 064101 (2006).
R. M. Erdahl, Adv. Chem. Phys. [**134**]{}, 61 (2007).
B. Verstichel, H. van Aggelen, D. Van Neck, P. W. Ayers, and P. Bultinck, Phys. Rev. A [**80**]{}, 032508 (2009).
N. Shenvi and A. F. Izmaylov, Phys. Rev. Lett. [**105**]{}, 213003 (2010).
D. A. Mazziotti, Phys. Rev. Lett. [**106**]{}, 083001 (2011).
G. Gidofalvi and D. A. Mazziotti, Phys. Rev. A [**74**]{}, 012501 (2006).
A. E. Rothman and D. A. Mazziotti, Phys. Rev. A [**78**]{}, 032510 (2008).
G. Gidofalvi and D. A. Mazziotti, J. Chem. Phys. [**129**]{}, 134108 (2008).
L. Greenman and D. A. Mazziotti, J. Chem. Phys. [**133**]{}, 164110 (2010).
A. Sinitskiy, L. Greenman, and D. A. Mazziotti, J. Chem. Phys. [**133**]{}, 014104 (2010).
H. Kummer, J. Math. Phys. [**8**]{}, 2063 (1967).
R. T. Rockafellar, [*Convex Analysis*]{} (Princeton University Press, Princeton, 1970).
P. R. Surj[á]{}n, [*Second Quantized Approach to Quantum Chemistry: An Elementary Introduction*]{} (Springer-Verlag, New York, 1989).
J. W. Helton, Ann. of Math. [**156**]{}, 675 (2002); J. W. Helton and S. McCullough, Trans. Amer. Math. Soc. [**356**]{}, 3721 (2004).
D. A. Mazziotti, Chem. Phys. Lett. [**289**]{}, 419 (1998).
D. A. Mazziotti, Phys. Rev. A [**85**]{}, 062507 (2012).
M. M. Deza and M. Laurent, [*Geometry of Cuts and Metrics*]{} (Springer, New York, 1997).
P. W. Ayers and E. R. Davidson, Adv. Chem. Phys. [**134**]{}, 443 (2007).
E. Kamarchik and D. A. Mazziotti, Phys. Rev. Lett. [**99**]{}, 243002 (2007).
J. Crawford, S. Torquato, and F. H. Stillinger, J. Chem. Phys. [**119**]{}, 7065 (2003).
I. Pitowsky, Math. Program. [**50**]{}, 395 (1991).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Recent experimental results have pushed the limits of magnetization dynamics to pico- and femtosecond timescales. This ultra-fast spin dynamics occurs in extreme conditions of strong and rapidly varying fields and high temperatures. This situation requires new description of magnetization dynamics, even on a phenomenological level of the atomistic Landau-Lifshitz-Gilbert equation, taking into account that the correlation time for electron system could be of the order of the inverse characteristic spin frequency. For this case we introduce the thermodynamically correct phenomenological approach for spin dynamics based on the Landau-Lifshitz-Miyasaki-Seki equation. The influence of the noise correlation time on longitudinal and transverse magnetization relaxation is investigated. We also demonstrate the effect of the noise correlation time on demagnetisation rate of different materials during laser-induced dynamics.'
author:
- 'U.Atxitia$^1$, O. Chubykalo-Fesenko$^1$, R.W.Chantrell$^2$, U.Nowak$^3$, and A Rebei$^4$'
title: 'Ultra-fast spin dynamics: the effect of colored noise.'
---
One of the fundamental questions of modern solid state physics is how rapidly the magnetization can respond to an external excitation. The recent development of time-resolved pump-probe experimental techniques using X-ray spectroscopy based on synchrotron radiation [@Siegmann] and the Stanford linear accelerator (SLAC) [@Tudosa; @Stamm] has allowed the investigation of magnetization dynamics on the pico-second time scale. The use of the powerful femto-second lasers [@beaurepairePRL96; @koopmansPRL00; @Gerrits] has pushed this limit down to the femto-second time-scale. The physical processes underlying the response of the magnetization on this ultra-short time-scale are complicated and far from being understood, but clearly involve the excitation and consequent non-equilibrium interaction of electron, phonon and spin sub-systems. Spin dynamic processes on this time-scale occur under extreme conditions remarkably different from those typical for dynamics at longer time-scales. Firstly, the three subsystems (electron, phonon and spin) are not in equilibrium with each other. Secondly, spin dynamic processes occur under very strong fields with different sources. In particular, in the experiments using the SLAC the magnitude of the external field can be as large as 20T. The strongly non-homogeneous magnetization processes are driven by the exchange field, having a magnitude $\gtrsim 100$T. Finally, in the laser-induced magnetization dynamics the effective temperature is increased up to and often above the Curie temperature [@koopmansPRL00; @beaurepairePRL96].
Atomistic spin models have proved to be a powerful approach to model ultra-fast magnetization dynamics [@kazantsevaEPL08; @kazantsevaPRB08]. For example, in the case of laser-induced magnetization changes, spin models provide important physical insight into the spin-reordering process, establishing the linear character of the demagnetization during the sub-picosecond regime and predicting the origin of different recovery rates in the pico-second regime. The basis of these models is the stochastic Landau-Lifshitz (LL) equation for each localized magnetic moment ${{\ensuremath{\mathbf{s}}}}_i$: $$\label{LLG}
{{\ensuremath{\mathbf{\dot{s}}}}}_i=\gamma[{{\ensuremath{\mathbf{s}}}}_i\times{{\ensuremath{\mathbf{H}}}}_i]-\gamma\alpha[{{\ensuremath{\mathbf{s}}}}_i\times[{{\ensuremath{\mathbf{s}}}}_i\times{{\ensuremath{\mathbf{H}}}}_i]]$$ Here ${{\ensuremath{\mathbf{H}}}}_i$ is the local effective field which includes Zeeman, exchange, anisotropy and magnetostatic contributions, augmented with a stochastic field ${{\ensuremath{\mathbf{\xi}}}}_i(t)$ with the following properties for both components and different spin sites: $$\label{whitenoise}
\left\langle \xi_i(t) \right\rangle=0, \ \ \ \left\langle\xi_i(t)\xi_j(t')\right\rangle=
\frac{2\alpha k_B T}{\gamma \mu_{s}}\delta(t-t')\delta_{ij}.$$ Here $T$ is the temperature, $\gamma$ is the gyro-magnetic ratio, $\mu_{s}$ is the magnetic moment and $\alpha$ is the parameter describing the coupling to the bath system. The basis of this equation is the separation of timescales, assuming that the bath (phonon or electron system) is much faster than the spin system. In this case, the bath degrees of freedom can be averaged out and replaced by a stochastic field with white noise correlation functions. The coefficient in front of the delta function in Eq.(\[whitenoise\]) is determined by the fluctuation-dissipation theorem. The assumption of white noise is therefore invalid for magnetization dynamics occurring on a timescale comparable to the the relaxation time of the electron system. The typical correlation time for the electron system in metals is $\lesssim 10 fs$ [@Siegmann]. Such magnetization dynamics time scale is now commonly achieved by applying femtosecond laser pulses. A further limitation of this approach comes from the fact that characteristic frequencies of the magnetization process are now also of the order of the timescale of the noise (electron) variable. Therefore, for modelling the ultra-fast magnetization experiments the approach (\[LLG\]) could break down.
The aim of the present Letter is to present a classical formalism beyond the white noise approximation to form a strong physical basis for models of ultra-fast magnetization dynamics in extreme conditions. First, we introduce the formalism and show that our approach is consistent with the equilibrium Boltzmann distribution and coincides with the previous atomistic approach for small correlation times for the bath variable. As main implications of the correlated noise approach, we discuss the influence of noise correlations on the most relevant characteristics of magnetization dynamics: the longitudinal and transverse relaxation times. Finally, we model the laser-induced demagnetisation rate for materials with different noise correlation times.
The standard generalization of the white noise to include correlations is the Ornstein-Uhlenbeck stochastic process [@Sancho]. However, when we implemented this process within the LL dynamics, we have seen that in agreement with the general theory [@Hasegawa], the Boltzmann distribution at equilibrium is not recovered for magnetization dynamics with correlated noise in this case. The deviations invariably correspond to precessional frequencies of the order of the inverse correlation time. The colored noise approach based on the Ornstein-Uhlenbeck process may provide a reasonable description in some situations when the system goes to a stationary condition not necessary coinciding with the equilibrium one (an example of this could be the spin-torque pumping into a magnetic system). However, in experiments such as those corresponding to laser-pulsed induced dynamics, a stochastic approach giving the correct equilibrium magnetization long after the laser pulse is gone, is necessary.
A suitable approach has been found in the work of Miyazaki and Seki [@MSeki] who generalized the Langevin equation for one spin to a non-Markovian case. The approach has been introduced for one spin at high temperatures, neglecting the interactions with other spins and assuming that their role is to provide the bath environment. In the present paper we generalize this approach to a many spin case, similar to the standard way of Eq.(\[LLG\]) where the applied field is substituted by the local field. We assume that the bath variable is due to external sources such as electrons. The other assumption made in this approach is that the spin is connected locally to the bath. Consequently, the set of equations for magnetization dynamics (in the following called Landau-Lifshitz-Miyazaki-Seki (LLMS)) reads:
$$\begin{aligned}
\label{MS}
\mathbf{\dot{s}}_i&=&\gamma [\mathbf{s}_i \times (\mathbf{H}_i + \boldsymbol{\eta}_i)], \nonumber \\
\boldsymbol{\dot{\eta}}_i&=&-{1 \over \tau_c}(\boldsymbol{\eta}_i - \chi \mathbf{s}_i)+ \mathbf{R}_i\end{aligned}$$
with the fluctuation-dissipation theorem for the bath variable: $\left\langle\mathbf{R}_i(t)\right\rangle=0$ ; $\left\langle\mathbf{R}_i(t)\mathbf{R}_j(t')\right\rangle=(2\chi k_B T /
\tau_c)\delta_{ij}\delta(t-t')$. The parameter $\chi$ describes the coupling of the bath variable to the spin. The precession term in the first equation of the set (\[MS\]) has the same form as in the Eq.(\[LLG\]). However, the damping is now described by the second equation in this set where also the bath variable adjusts to the direction of the spin due to the interaction with it. In the limit $\tau_c
\rightarrow 0$ the stochastic LL equation (\[LLG\]) is recovered [@MSeki]. This also provides a relation between the damping and the coupling constants as $\alpha=\gamma \chi \tau_c $, giving a more precise physical sense to the LL damping constant at atomistic level.
For integration of Eqs.(\[MS\]) the Heun integration scheme was modified for this special case. First of all we investigated the equilibrium properties for an ensemble of non-interacting spins. In all cases of large fields, temperatures and correlation times, the correct Boltzmann distribution is obtained at equilibrium (see inset in Fig.\[f:LLMS\]).
![Equilibrium magnetization as a function of temperature for a system of $\mathcal{N}=32^3$ interacting spins, integrating the LLMS equation with different correlation times and integrating the LL equation. The inset shows distribution functions for non-interacting spin system modeled within the LLMS approach for different values of the reduced field $\zeta=\mu_s H/k_B T$ ($\gamma H \tau_c=1.76$) and correlation time $\tau_c=10$fs. The solid line in the inset represents the Boltzmann distribution. \[f:LLMS\]](figure2bis.eps)
Now we turn to the multi-spin system. First of all, we prove that the stochastic Eqs.(\[MS\]) for a multi-spin system are consistent with the standard equilibrium properties. For this purpose we use the formalism of the Onsager kinetic coefficient method applied in Ref.[@Brownian] for the LL multispin system (\[LLG\]). The system (\[MS\]) is linearized near equilibrium and represented in a general form of the Langevin equation: $$\begin{aligned}
\label{genLan}
{dx_i \over dt}&=&-\sum_j \gamma_{ij}X_j + r_i\\ \nonumber
\left\langle r_i(t)\right\rangle &=&0;\;\;\left\langle r_i(0)r_j(t)\right\rangle =\mu_{ij}\delta(t)\end{aligned}$$ Here $x_i$ stands for small deviations of the stochastic variables $\mathbf{s}_i$ or $\eta_i$ from their equilibrium values, $X_i$ represent their thermodynamically conjugate variables and $\mu_{ij}=\gamma_{ij}+\gamma_{ji}$. For the spin variable we have: $X_j=-(\mu_s / k_B T) H_j$ , where $H_j$ is the internal field corresponding to a particular lattice site and spin component. Unlike Eq.(\[LLG\]), the first equation in Eqs.(\[MS\]) contains only a precessional term and, therefore, the corresponding kinetic coefficients are antisymmetric in spin components, giving for this equation $\mu_{ij}=0$. Taking into account the generalization of the internal energy to include the bath variable as $F(\{\mathbf{s}_i\},\{\mathbf{\eta}_i\})
=F_0(\{\mathbf{s}_i\})+\sum_{i}[\eta_i^2/(2
\chi)-\mathbf{\eta}_i\mathbf{s}_i]$, where $F_0(\{\mathbf{s}_i\})$ is the internal energy without the bath variable, the conjugate variable to the bath one is $\mathbf{X}_j=(\mathbf{\eta}_j-\chi \mathbf{s}_j)/(k_B T \chi)$. Therefore, the corresponding matrix of the kinetic coefficients is diagonal and for the second equation we obtain $\mu_{ij}= (2k_BT
\chi /T_c) \delta_{ij}$. Consequently, we have proven that, under the supposition of local coupling of the spin to the bath variables, the set of multi-spin equations (\[MS\]) is consistent with the equilibrium properties.
In our simulations for the multispin system we use a Heisenberg Hamiltonian on a cubic system of $32^3$ magnetic moments with nearest-neighbor interactions only, with $\mu_s=1.45 \mu_B$ and the Curie temperature $T_c=700$K ($k_B
T_c \approx 1.44$J). The coupling parameter $\chi$ was chosen to give the LL damping parameter $\alpha=0.01$. In Fig.\[f:LLMS\] we present calculations of the equilibrium magnetization as a function of temperature for spin systems with different values of the correlation times. Independence of the equilibrium properties on the correlation time, and the agreement with calculations using the LL equation with uncorrelated noise demonstrates our generalization of the LLMS equation to multi-spin systems. Consequently, the LLMS equation provides a basis for the phenomenological description of magnetization dynamics in extreme situations of high temperatures, and large and rapidly varying external fields. The advantage of the approach is also that the fluctuation-dissipation theorem is not applied directly to the spin variable. Therefore, the bath variable (for example electrons) and the spin system need not be in equilibrium with each other.
Next we discuss the most important implications of this new approach to the ultra-fast dynamics. It is known that during the excitation with spatially inhomogeneous fields in the Terahertz range [@Siegmann; @Stamm] and also during laser-induced magnetization dynamics [@beaurepairePRL96; @koopmansPRL00], strong local disordering of the spin system occurs. The dynamics in these cases is governed by field or temperature excited high-frequency spin waves which are responsible for the effective damping. The important dynamical feature then is the rate of magnetization recovery. During these processes two types of relaxation could be distinguished. The first one known as longitudinal relaxation is responsible for linear magnetization recovery, i.e. the magnetization magnitude. During the laser-induced demagnetization, the longitudinal relaxation is responsible for the femtosecond demagnetization. The second one is the transverse relaxation when the magnetization vector relaxes to the direction parallel to the internal field via magnetization precession. The longitudinal relaxation time increases with the temperature while the transverse relaxation time has minimum at $T_c$ [@chubykaloPRB06].
To simulate the longitudinal relaxation, we start with the initial condition $\{s_i^z\}=1.0$ and observe the system to relax at given temperature $T$. The obtained relaxation curves are then fitted to exponential decay to extract the longitudinal relaxation rate. The longitudinal relaxation time, normalized to the uncorrelated case, is presented in Fig.\[f:Long\] as a function of the noise correlation time. The longitudinal time calculated by means of the LL approach (\[LLG\]) is of the order of $10$fs ($\tau^{0}_{||}= 28$fs at $T=300$K). For correlation time $\tau_c \lesssim 1$fs the uncorrelated approach gives the same results as the LLMS one. However, one can see that $\tau_c \simeq 10-100$fs gives a dramatic increment of the longitudinal relaxation time. The effect is less pronounced at higher temperature since in this case the temperature contributes to the loss of correlations.
![Longitudinal relaxation time (normalized to the uncorrelated case) as a function of the correlation time for various temperatures calculated within LLMS approach. The inset shows longitudinal relaxation for various correlation times and $T=600$K. \[f:Long\]](figure3.eps)
Next, we investigate the transverse relaxation in Fig. \[f:Trans\]. The transverse relaxation time is defined by the magnetisation precession and normally is much slower than the longitudinal one. For one spin $\tau_{\perp}=\tau_{\perp}^0 [1+(\omega_H \tau_c)^2]$ [@MSeki], where $\omega_H$ is the field-dependent precessional frequency. Consequently, the influence of the correlation time on the transverse relaxation may be expected only for strong applied field for which $\omega_{H}\sim \tau_c^{-1}$ and, thus, could be relevant for SLAC experiments. To show the influence of the noise correlation on the transverse relaxation, we model the precessional dynamics at strong applied field $H=24.85$T ($0.05J/\mu_s$). In this case, the spin system was first equilibrated at given temperature and applied field. After that the whole system was rotated to an angle $30^o$ and the relaxation to the direction parallel to the applied field was observed. For this particular strong applied field, the correlation times $\tau_c \gtrsim 100fs$ are necessary in order to see their influence on the transverse relaxation.
![Transverse relaxation time (normalized to the uncorrelated case) as a function of the correlation time for various temperatures calculated within LLMS approach. The insert shows transverse relaxation for two temperatures $T=5$K and $T=600$K and $\tau_c=10$fs. \[f:Trans\]](figure4.eps)
![Laser-induced demagnetisation as a function of time for various noise correlation times $\tau_c$ modelled within LLMS and LL approaches. \[f:pump\_probe\]](figure5.eps)
To demonstrate the immediate physical consequence of the correlated noise for a more concrete experimental situation, we present in Fig.\[f:pump\_probe\] the modelling results for the laser-induced demagnetisation for various noise correlation times and constant damping parameter $\alpha=0.05$. We suppose that the magnetization dynamics is produced by a Gaussian laser pulse with 50 fs duration and the fluency $33.7 mJ/cm^2$. Similar to Ref.[@kazantsevaEPL08], we assume that the photon energy is transfered to the electrons and lattice but the magnetization is directly coupled to the electron temperature $T_e$. The latter is calculated within the two-temperature model [@...] with the electron and lattice specific heat constants $C_e= 700 J/m^3 K^2 \times T_e^{max},\; C_l=3\times 10^6 J/m^3K $ and the coupling constant $G_{el}=8\times10^{17} W/m^3 K $. Our results clearly show strong impact of the noise correlation time on the degree of demagnetisation during the laser-induced process. Namely, the materials with small $\tau_c$ are demagnetised easier. This could be true, for example, for $d$-electrons in metals with large scattering rate or for $f$-electrons in rare-earths which have strongly relativistic nature.
In conclusion, the standard phenomenological approach to model spin dynamics has been generalized to the non-Markovian case. This approach is necessary in the extreme situations of large characteristic magnon frequencies occurring during ultra-fast magnetization processes. The advantages of the new approach are the following: (i) the memory (correlation) effects arising from the fact that the bath variable responds to the spin direction are taken into account. This corresponds to the situation when the bath variable is not in equilibrium with the spin system. (ii) the fluctuation-dissipation theorem is not applied to the spin systems as in the standard LL approach (iii) In equilibrium the Boltzmann distribution is recovered. The price for this new approach is the use of two phenomenological constants: the phenomenological damping parameter $\alpha$ for the LL approach is substituted by two phenomenological parameters in the LLMS approach: the correlation time $\tau_c$ and the coupling constant $\chi$. Several processes may be important in determining these constants, as for example, the spin-orbit coupling, momentum relaxation, scattering rate and de-phasing time of conduction electrons. As in the LL approach, these parameters will be material-specific and their physical origins should be clarified on the basis of first-principle approaches. We have shown that the ultra-fast magnetization dynamics is strongly influenced by these parameters which stresses the necessity of first-principle models, capable to clarify their physical origins.
The authors acknowledge the financial support from Seagate Technology, USA and from European COST-P19 Action.
[99]{}
J. Stöhr and H. C. Siegmann, “*Magnetism*” (Springer-Verlag, Berlin, 2006) and references therein.
I.Tudosa, C.Stamm, A.B.Kashubam F.King, H.C.Siegmann, J.J. Stöhr, G.Ju, B.Lu and D.Weller, Nature [**428**]{}, 831 (2004)
C. Stamm, I. Tudosa, H. C. Siegmann, J. Stöhr, A. Yu. Dobin, G. Woltersdorf, B. Heinrich and A. Vaterlaus, Phys. Rev. Lett. **94** 197603 (2005).
E. Beaurepaire, J.-C. Merle, A. Daunois, and J. Y. Bigot, Phys. Rev. Lett. **76**, 4250 (1996).
B. Koopmans, M. van Kampen, J. T. Kohlhepp, and W. J. M. de Jonge, Phys. Rev. Lett. [**85**]{}, 844 (2000).
Th. Gerrits, H.A.M. van der Berg, J.Hohlfeld, L. Bür and Th.Rasing, Nature [**418**]{} 509 (2002)
N. Kazantseva, U. Nowak, R. W. Chantrell, J. Hohlfeld, and A. Rebei, Europhys. Lett. [**81**]{}, 27004 (2008).
N.Kazantseva, D.Hinzke, U.Nowak, R.W.Chantrell, U.Atxitia and O.Chubykalo-Fesenko, Phys Rev B [**77**]{}, 184428 (2008)
J. M. Sancho and M. San Miguel, Phys. Rev. A **26**, 1589 (1982).
H. Hasegawa, Phys.A **387** 2697 (2008).
K. Miyazaki and K. Seki, J. Chem. Phys. **108**, 7052 (1998)
O. A. Chubykalo, R. Smirnov-Rueda, J. M. González, M. A. Wongsam, R. W. Chantrell and U. Nowak, J. Magn. Magn. Mater. **266**, 28 (2003).
O. Chubykalo-Fesenko, U. Nowak, R. W. Chantrell, and D. Garanin, Phys. Rev. B [**74**]{}, 094436 (2006).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Effect of carbon content on magnetostructural transformation in antiperovskites of the type Mn$_3$GaC$_x$ ($x$ = 0.8, 1.0 and 1.05) has been investigated. It is found that, increase in carbon content changes the ground state from ferromagnetic metallic ($x$ = 0.8) to antiferromagnetic semiconducting ($x$ = 1.05) type. This has been attributed to localization of itinerant Mn $3d$ electrons due to increased Mn$3d$ - C$2p$ hybridization. Such a hybridization strengthens Mn-C-Mn antiferromagnetic interactions over Mn-Mn ferromagnetic interactions. Further, magnetic field can be used as a tool to modulate the relative strengths of these ferromagnetic and antiferromagnetic interactions thereby affecting the nature and strength of magnetocaloric properties.'
address:
- 'Department of Physics, Goa University, Goa 403206, India.'
- 'Tata Institute of Fundamental Research, Dr. Homi Bhabha Road, Colaba, Mumbai 400005, India.'
author:
- 'E. T. Dias'
- 'K. R. Priolkar'
- 'A. K. Nigam'
title: 'Effect of carbon content on magnetostructural properties of Mn$_3$GaC'
---
,
,
Antiperovskites, magnetostructural transformation, Mn$_3$GaC
Introduction
============
The antiperovskite Mn$_{3}$GaC has a cubic structure [@e-dias1; @e-dias10] and belongs to a class of materials that are gaining interest due to their magnetocaloric properties [@e-dias5; @e-dias6; @e-dias7; @e-dias8; @e-dias11]. Stoichiometric composition of Mn$_3$GaC undergoes a series of magnetic transformations as a function of temperature. Firstly, it undergoes a second order transformation from a paramagnetic (PM) state to a ferromagnetic (FM) state at $T_{C}\sim 249K$ followed by another second order transformation to a canted ferromagnetic state (CFM) at $T\sim 164K$. Finally the material undergoes a volume discontinuous first order magnetic transformation to an an antiferromagnetic (AFM) state at $T\sim 160K$ [@e-dias3; @e-dias4]. Concentration of carbon plays a vital role in inducing the first order magnetic transformation. Literature suggests that samples deficient in carbon, Mn$_3$GaC$_{0.8}$, exhibits only a second order PM to FM transition at $T_C\sim 318K$ [@e-dias5] while the compound with carbon concentration equal to 1.0, undergoes additional magnetic transitions leading to an antiferromagnetic ground state. The first-order magnetic transition from a ferromagnetic to antiferromagnetic state in Mn$_3$GaC produces a large change in entropy in presence of magnetic field giving rise to a large magnetocaloric effect (MCE), a property useful in magnetic refrigeration [@e-dias6; @e-dias7].
Although the role of carbon content in inducing the magnetic phase transformation has been contemplated in literature [@e-dias1; @e-dias12], the exact nature of the driving force in inducing the magnetostructural transformation is not yet understood. Neutron diffraction studies have shown that increase in C concentration results in a lattice expansion, increasing Mn-Mn next nearest neighbor distance. This is implied to cause an increase in Mn $3d$ - C $2p$ hybridization which strengthens Mn-C-Mn antiferromagnetic interactions over Mn-Mn ferromagnetic interactions causing the sample to transform via a discontinuous volume expansion [@e-dias1; @e-dias7; @e-dias13; @e-dias14; @e-dias16]. However, X-ray absorption near edge structure (XANES) studies at Mn L edges did not show any change in near edge features as a function of C content that would suggest an increase in Mn $3d$ - C $2p$ hybridization [@e-dias5]. It must be mentioned here that these studies were carried out on compositions wherein carbon content varied from $0.8 \le x \le 1.0$. In this paper for the first time we examine Mn$_3$GaC$_x$ with $x$ = 1.05 (C-excess) along with $x$ = 0.8 (C-deficient) and $x$ = 1.0 (C-stoichiometric) compounds. From structural, transport and magnetic properties of these three compositions, it is shown that increase in carbon concentration leads to an increase in cell volume, a change in character of resistivity from metallic in $x$ = 0.8 to semiconducting in $x$ = 1.05 and dominance of antiferromagnetic interactions. These characteristics point towards localization of conduction electrons and highlight the importance of Mn-C-Mn interactions in inducing the first order magnetic transformation from ferromagnetic to antiferromagnetic state. Furthermore, it is also shown that magnetic field can be effectively used as a tool to modulate the nature of magnetic interactions and therefore the magnitude and nature of magnetocaloric properties.
Experimental
============
The polycrystalline samples of Mn$_3$GaC$_x$ were prepared by solid state reaction method. First the constituent elements, Mn, Ga and C were thoroughly mixed in molar ratio of 3:1:1, pressed into pellets, sealed in evacuated quartz tubes and heated for five days at 1073K. The sample was then quenched to room temperature, pulverised, pelletized and annealed again for five more days at 1073K [@e-dias7]. A comparison of results obtained from X-ray diffraction studies, density measurements and magnetization measurements (described later) with those reported in literature [@e-dias4] showed that the resulting compound was sub stoichiometric having a composition Mn$_3$GaC$_{0.8}$. After preserving a third of this powder, the other two parts were respectively mixed with additional carbon equal to 0.20 and 0.25 times the original carbon content and were further annealed at 1073K for five days in evacuated and sealed quartz tubes to obtain Mn$_3$GaC and Mn$_3$GaC$_{1.05}$. X-ray diffraction (XRD) patterns, to identify the crystal structure and phase purity of the powdered samples were recorded at room temperature in the range of $ 20^\circ \leq 2\theta \leq 80^\circ$ using Cu K$_\alpha $ radiation and at Indian beamline, BL18B, Photon Factory, Japan at photon energy of 13 keV. In order to study the effect of carbon content on the transport properties, resistivity measurements (10K - 330K) on all the samples were performed using conventional D.C. four probe technique in a Closed Cycle Refrigerator (CCR). To investigate the role of carbon on magnetic properties of these samples, magnetization (M) measurements were carried out as a function of temperature (T) and magnetic field (H) in the temperature range of 5K to 300K and in fields up to 7T using a Quantum Design SQUID magnetometer and a Quantum Design Vibrating sample magnetometer up to fields of 14T.
Results and Discussion
======================
Figure \[fig:mt-rt\] depicts temperature dependence of magnetization and resistivity of Mn$_3$GaC$_{x}$ ($x$ = 0.80, 1.0 and 1.05). Temperature dependent magnetization curves M(T) for the three compositions was measured in an applied field of 0.1T during zero-field cooling (ZFC) and field cooled warming (FCW) cycles (Figure \[fig:mt-rt\](a)). The C-deficient compound shows a PM to FM transition with $T_{C}$ just above 300K, whereas the C-stoichiometric compound shows the same PM to FM transition at a lower temperature ($ T_{C} = 242K $) followed by a FM to AFM first-order magnetic transition at $T_N = 178K$. Interestingly, in case of C-excess sample, a clear PM to FM transition is absent. It appears as if the sample transforms to an AFM ground state via a first-order transition at $T_N = 195K$ directly from a PM state. Such a behaviour has been observed in related antiperovskites Mn$_3$Cu$_{0.5}$Ge$_{0.5}$N and Mn$_3$GaC$_{0.85}$N$_{0.15}$ [@nitride; @nitride1]. However, ferromagnetic correlations are present especially below 240K which manifest in terms of a relatively sharper rise in magnetization as can be seen in M versus T plot of Mn$_3$GaC$_{1.05}$ presented in Figure \[fig:mt-rt\](a). It may be noted that this temperature at which ferromagnetic correlations appear is lower than the T$_C$ of C-stoichiometric sample. Thus it can be inferred that with increase in C content, the strength of FM interactions decreases while the strength of AFM interactions increases.
![**(a)** Temperature dependence of the magnetization M(T) during ZFC and FCW cycles in 0.1T for Mn$ _{3} $GaC$ _{x} $, $x$ = 0.8, 1.0 and 1.05. **(b)** Variation of electrical resistivity as a function of temperature for Mn$ _{3} $GaC$ _{x} $, $x$ = 0.8, 1.0 and 1.05.[]{data-label="fig:mt-rt"}](mt-rt.eps){width="\columnwidth"}
Figure \[fig:mt-rt\](b) compares the electrical resistivity behavior of the three compounds as a function of temperature. It can be seen that the magnitude of resistivity increases with increasing C content. The C-deficient sample has the lowest value of resistance and exhibits a metallic behavior with a distinct slope change around T$_C \sim$ 300K depicting PM to FM transition. Likewise, the resistivity of C-stoichiometric sample also has an overall positive temperature coefficient but with a magnitude which is about an order of magnitude higher than the C-deficient sample. This resistivity behavior is quite similar to that reported for Mn$_3$GaC composition in literature [@e-dias2]. Two features that stand out are a subtle change in slope around $ T_{C}$ ($\sim$ 240K) followed by a sharp increase in resisitance and a hysteresis between warming and cooling cycles near the first order transformation temperature. The higher resistance at lower temperature hints at localization of conduction electrons in the high volume cubic phase to which the compound transforms at T$_N$. On the other hand C-excess compound not only has the highest magnitude of resistivity but also displays a completely opposite variation of resistivity with temperature. Here the resistivity increases with decrease in temperature which is akin to semiconductor like behavior. A weak discontinuity is also visible around 190K corresponding to the first order transformation to antiferromagnetic state. The gradual increase in resistivity as a function of carbon content and the change in temperature gradient of resistivity from positive to negative with increase in carbon content is a signature of the localization of $3d$ conduction electrons in the carbon valence band.
In order to understand the variations in magnetic and transport properties exhibited by these compounds room temperature XRD patterns of Mn$ _{3} $GaC$ _{x} $ ($x$ = 0.80, 1.0 and 1.05) were refined using the LeBail method in FullProf Suite. It can be seen in Figure \[fig:xrd\] that all the three compositions crystallize largely in a simple cubic Mn$ _{3} $GaC phase with space group *Pm3m* [@e-dias1] along with a tiny impurity ($<$ 1%) of MnO. Such an insignificant amount of impurity does not to affect the magnetic and transport properties as can be seen in Figure \[fig:mt-rt\] and which are in good agreement with those reported in literature [@e-dias5; @e-dias2]. Although very little changes in the intensities of Bragg peaks were observed for compounds with varying carbon concentration, the unit cell volume and the density calculated from XRD exhibit an increasing trend with increase in carbon content. While the density increases from 6.95 g/cm$^3$ to 6.97 g/cm$^3$ with increase in carbon content, the variation of unit cell volume is depicted in Figure \[fig:lattice\]. Increase in cell volume will directly increase the Mn-Mn bond distance and cause the ferromagnetic interactions to weaken[@e-dias17]. Ferromagnetic interactions in these compounds is believed to be mediated by itinerant electrons [@e-dias2]. With increase in carbon content, the conduction electrons seem to get localized resulting in weakening of ferromagnetism and a concomittant increase in resistivity. Indeed such a trend is readily seen from the resistivity and magnetization measurements. Furthermore, a satisfactory LeBail fit for Mn$_3$GaC$_{1.05}$ could be obtained only by taking into account an additional antiperovskite cubic phase with a higher cell volume. The refined lattice parameter obtained for this phase were approximately equal to that of the transformed antiferromagnetic phase of Mn$_3$GaC [@e-dias1]. This high volume antiperovskite phase is antiferromagnetic in nature and its presence could be the reason for the absence of a clear ferromagnetic ordering in Mn$3$GaC$_{1.05}$.
![Room temperature XRD patterns of Mn$ _{3} $GaC$ _{x} $, $x$ = 0.8, 1.0 and 1.05 refined using LeBail method. Insets show diffraction pattern for $ 30^\circ \leq 2\theta \leq 50^\circ$. Astericks indicate the MnO phase and the additional high volume phase in the C-excess sample is marked with an arrow.[]{data-label="fig:xrd"}](xrd.eps){width="\columnwidth"}
Apart from the decrease in T$_C$, it is also observed that antiferromagnetic ordering not only appears but also increases in strength with increasing carbon content. The C-deficient compound can be conceived as consisting of randomly distributed vacancies in the carbon sublattice and relatively fuller Mn and Ga sublattices. Therefore the ferromagnetic Mn-Mn interactions dominate and there is no transformation to the antiferromagnetic state. While in case of the C-stoichiometric sample where all the individual sublattices are likely have near optimal occupation, a strong competition between ferromagnetic Mn-Mn and antiferromagnetic Mn-C-Mn interactions exists. In the C-excess sample, presence of additional carbon further strengthens antiferromagnetic interactions leading to an increase in T$_N$. It is reported that due to increase in Mn-Mn bond distance the hybridization between Mn $3d$ and C $2p$ bands increases inducing antiferromagnetic Mn-C-Mn superexchange interactions which compete with ferromagnetic interactions. This competition drives the compound to undergo a first order magnetic transformation from ferromagnetic cubic to an antiferromagnetic cubic phase of higher volume. Furthermore, the localization of itinerant Mn $3d$ conduction electrons due to Mn $3d$ - C $2p$ hybridization is clearly brought out from the resistivity measurements. Here, not only does the magnitude of resistivity increases with increasing carbon content, its character itself changes from metallic to one displaying semiconductor like behavior.
![Variation of unit cell volume with carbon concentration for Mn$ _{3} $GaC$ _{x} $, $x$ = 0.8, 1.0 and 1.05.[]{data-label="fig:lattice"}](cell.eps){width="\columnwidth"}
In order to develop a deeper understanding of the competing ferromagnetic and antiferromagnetic interactions, magnetization loops as a function of field M(H) were recorded at 5K in $H = \pm$7T range and up to 14T in the temperature range of 140K to 280K. The M(H) loops at 5K are shown in Figure \[fig:mh\](a). The carbon deficient compound shows a typical soft ferromagnetic behavior with a saturation moment of 3.6 $\mu _{B} $/f.u. and a coercivity of about 91Oe, confirming a FM phase at 5K. Interestingly, the stoichiometric and the carbon rich compound show a [*S*]{} shaped loop with no saturation up to 7T and with the virgin curve lying outside the envelope (Figure 4(b)). This behavior suggests presence of competing FM and AFM interactions with contribution of FM growing at the expense of AFM interactions in presence of magnetic field. It can be clearly seen that in the case of C-excess sample, the virgin magnetization curve is completely outside the main loop as compared to that in C-stoichiometric sample indicating stronger antiferromagnetic interactions in Mn$_3$GaC$_{1.05}$. Surprisingly however, the C-excess sample displays higher values of magnetization than C-stoichiometric sample. Plots of magnetization as a function of temperature indicates AFM ground states in both C-stoichiometric and C-excess samples at 5K. This is also corroborated by extrapolating M(H) curve to meet the [*y-axis*]{}. In both the samples the M $\approx$ 0 at H = 0 indicating abscence of spontaneous magnetization. Therefore the presence of FM interactions in these compounds is rather unexpected and appears to have been induced by external magnetic field. This field induced ferromagnetism possibly explains higher magnetic moment values at 5K in C-excess sample as compared to that of C-stoichiometric sample.
![Hysteresis loops recorded at 5K in the fields between $ \pm $±7T for Mn$ _{3} $GaC$ _{x} $, $x$ = 0.8, 1.0 and 1.05) []{data-label="fig:mh"}](mh.eps){width="\columnwidth"}
![Isothermal magnetization curves for Mn$ _{3} $GaC$ _{x} $, $x$ = 1.0 and 1.05 between 155K and 190K recorded at an interval of 5K in the range 0T $\le H \le$ 14T.[]{data-label="fig:mh-1st-quad"}](mh-1st-quad.eps){width="\columnwidth"}
![Change in entropy ($\Delta S$) calculated from magnetisation isotherms recorded in the temperature range 140K-270K for fields varying from 0.1T to 14T for Mn$ _{3} $GaC$ _{x} $, $x$ = 1.0 and 1.05. []{data-label="fig:mce"}](mce.eps){width="\columnwidth"}
The presence of field induced ferromagnetism is further confirmed from isothermal magnetization studies conducted in the temperature interval from 140K to 270K. The isothermal magnetization curves in the limited temperature range of 155K to 190K and the corresponding entropy plots are presented in Figure \[fig:mh-1st-quad\] and Figure \[fig:mce\] respectively. A comparison of isothermal magnetization curves recorded at 190K (ferromagnetic phase) and 155K (antiferromagnetic phase) clearly indicates the field induced effect. In the C-stoichiometric compound, up to H = 5T, the magnetization at 155K is very small, but shows a sharp increase and approaches the ferromagnetic value (M at 190K) for H $>$ 6T. Such a field induced metamagnetic behavior is also seen in the C-excess compound but the transition takes place at much higher fields and over a broad field range. This indicates that even in antiferromagnetic state, ferromagnetism can be induced by external magnetic field. The effect of such a metamagnetic transition can also been seen on the magnetocaloric properties of these compounds. In stoichiometric Mn$_3$GaC an isothermal magnetic entropy change, $\Delta S$ = 16 J/kg-K can be obtained near the first order transition temperature which agrees well the the reported value [@e-dias6; @e-dias7; @e-dias15]. In addition a negative peak in $ \Delta S$ is observed near $T_{C}$ ($\sim$ 240K). This change in entropy reportedly translates to an adiabatic temperature difference of about $\Delta T \approx 4K$ [@e-dias8]. Further the $\Delta S$ exhibits very little change as a function of magnetic field (H) for H $\ge$ 2T and remains absolutely constant above H = 6T. This behavior of $\Delta S$ can be directly linked to the metamagnetic transition observed in magnetization plots.
The carbon excess sample also shows a similar positive maximum at its first order transition temperature but with a slightly smaller value of 13.6 J/Kg-K at 14T. There is no negative peak in $\Delta S$ confirming absence of a transition to ferromagnetic state. Furthermore, $\Delta S$ exhibits clear field dependency with its value increasing continuously as well as moves to a lower temperature with increase in $H$. An estimate of $\Delta T$ made using the Clausius Clayperon equation, in low values of magnetic field (H $\sim$ 2T) indicates it to be nearly the same as in case of C-stoichiometric sample. However, the induced ferromagnetism in C-stoichiometric sample causes it to decrease very sharply for higher magnetic field values. This decrease is much slower in the sample which has carbon content in excess.
Conclusions
===========
In summary, the carbon content critically controls the magnetostructural properties of Mn$_3$GaC type antiperovskites. It is seen that an increase in C content strengthens the antiferromagnetic interactions while weakening the feromagnetic interactions in the compounds. Incorporation of carbon into the lattice increases the unit cell volume which increases the Mn-Mn bond distance and thus weakening the ferromagnetic interactions. The change in character of resistivity from metallic to semiconducting type indicates localization of itinerant Mn $3d$ conduction electrons. This leads to strengthening of Mn-C-Mn antiferromagnetic superexchange interactions. Furthermore, even in the antiferromagnetic state, magnetic field can be used to tune the magnetic interactions. In Mn$_3$GaC, a applied magnetic field of 6T converts antiferromagnetic state to ferromagnetic one. Such a conversion cannot be fully achieved in Mn$_3$GaC$_{1.05}$ even up to a field of 14T. This field induced ferromagnetism affects the magnetocaloric properties of the two compounds.
Acknowledgments {#acknowledgments .unnumbered}
===============
Authors thank Board of Research in Nuclear Sciences (BRNS) for the financial support under the project 2011/37P/06. M/s Devendra D. Buddhikot and Ganesh Jangam are acknowledged for the experimental assistance. Thanks are also due to Department of Science and Technology, India for the financial support and Saha Institute of Nuclear Physics, India for facilitating the experiments at the Indian Beamline, Photon Factory, KEK, Japan.
[100]{} D. Fruchart and E. Bertaut, *J. Phys. Soc. Jpn.* **44** (1978) 781 L. Howe and H. Myers, *Phil. Mag.* **2** (1957) 554 L. Lewis, D. Yoder, A. Moodenbaugh, D. Fischer and M. Yu, *J. Phys.: Condens. Matter* **18** (2006) 1677 T. Tohei, H. Wada and T. Kanomata, *J. Appl. Phys.* **94** (2003) 1800 M. Yu, L. Lewis and A. Moodenbaugh, *J. Appl. Phys.***93** (2003) 10128 $\ddot{\rm O}$. $\c{\rm C}$akir and M. Acet, *Appl. Phys. Lett.* **100** (2012) 202404 T. Tohei, H. Wada, and T. Kanomata, *J. Magn. Magn. Mater.* **E585** (2004) 272 K. Kamishima, M. Bartashevich, T. Goto, M. Kikuchi and T. Kanomata, *J. Phys. Soc. Jpn.* **67** (1998) 1748 K. Kamishima, T. Goto, T. Sasaki, T. Kanomata, and T. Inami, *J. Phys. Soc. Jpn.* **71** (2002) 922 D. Fruchart, E. Bertaut, B. Le Clerc, K. Le Dang, P. Veillet, G. Lorthioir, E. Fruchart, and R. Fruchart *J. Solid State Chem.* **8** (1973) 182 B. Wang, P. Tong, Y. Sun, W. Tang, L. Li, X. Zhu, Z. Yang and W. Song, *Physica B* **405** (2010) 2427 B. Wang, C. Li, J. Lin, L. Li, P. Tong, X. Zhu, Z. Yang, W. Song, J. Dai and Y. Sun *J. Magn. Magn. Mater.* **323** (2011) 2017 T. Kanomata, M. Kikuchi, T. Kaneko, K. Kamishima, M. Bartashevich, H. Katori, and T. Goto, *Solid State Commun.* **101** (1997) 811 Z. H. Sun, X. Y. Song, F. X. Lin, L. X. Sun, X. K. Yuan and X. M. Liu, *J. Phys. D: Appl. Phys.* **42** (2009) 122004 $\ddot {\rm O}$. Cakır and M. Acet, *J. Magn. Magn. Mater.* **344** (2013) 207 T. Harada, T. Makabe, T. Kanamota and T. Kaneko, *J. Magn, Magn. Mater.* **104-107** (1992) 1955 J. Bouchaud, *Ann. Chim.* **3** (1968) 81 M. Yu, L. Lewis and A. Moodenbaugh, *J. Magn. Magn. Mater.* **299** (2006) 317
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The existence of one-phonon and possible two-phonon transverse wobbling bands is proposed for the first time in two even-even nuclei, $^{134}$Ce and $^{136}$Nd. The predominant $E2$ character of the $\Delta I = 1$ transitions connecting the one-phonon wobbling band in $^{134}$Ce to the two-quasiparticle yrast band supports the wobbling interpretation. The extracted wobbling frequencies decrease with increasing spin, indicating the transverse character of the wobbling motion, with the angular momenta of the two quasiparticles aligned perpendicular to the axis of collective rotation. A candidate for two-phonon wobbling motion is also proposed in $^{136}$Nd. The wobbling frequencies calculated in the harmonic frozen approximation are in good agreement with the experimental ones for both the$^{134}$Ce and $^{136}$Nd nuclei.'
address: 'Centre de Sciences Nucléaires et Sciences de la Matière, CNRS/IN2P3, Université Paris-Saclay, Bât. 104-108, 91405 Orsay, France'
author:
- 'C. M. Petrache'
- 'S. Guo'
bibliography:
- 'mybibfile.bib'
title: 'Transverse wobbling motion in $^{134}$Ce and $^{136}$Nd'
---
`Nuclear structure of ^{134}Ce and ^{136}Nd`,model calculations,collective excitations ,wobbling motion
Introduction
============
The recent studies of the Lanthanide nuclei with $A \approx 130$ were mainly driven by the search for chiral bands in odd-odd and odd-even nuclei [@fra]. Candidates for chiral partners were proposed in nuclei from $^{126}$Cs to $^{138}$Eu. Very recently, the transverse wobbling mode has been discovered in the odd-even $^{135}$Pr nucleus [@fra-trans; @135pr], in which case the unpaired $\pi h_{11/2}$ proton lies at the bottom of the $\pi h_{11/2}$ sub-shell, having therefore particle character. As a consequence, in order to maximize the overlap of the density distribution with the triaxial core, its angular momentum is along the short axis of the triaxial core, being therefore perpendicular to the angular momentum of the core which is built along the intermediate axis if one assumes hydrodynamic moments of inertia. The wobbling excitations can also occur in odd-even nuclei with an unpaired $\nu h_{11/2}$ neutron lying at the top of the $\pi h_{11/2}$ sub-shell, having therefore hole character. In this case the angular momentum of the unpaired neutron is aligned along the long axis, being again perpendicular to the collective angular momentum of the triaxial core. Such excitations have been recently identified in $^{133}$Ce [@133ce-wobb].
In the Lu-Hf region, wobbling bands were only observed in odd-even triaxial superdeformed nuclei [@163lu]. Even if wobbling bands are predicted since long in even-even triaxial nuclei, no clear experimental evidence has been reported until now in configurations without unpaired nucleons. However, the band built on the $\nu h_{11/2}^{-2}$ two-quasihole neutron configuration of the normal-deformed $^{138}$Nd was recently interpreted as an wobbling collective excitation [@138nd]. Even a two-phonon wobbling excitation was suggested in the same $^{138}$Nd nucleus, but the limited number of observed states and lack of angular distribution data did not allow to draw a definite conclusion.\
The recent evidence for wobbling motion in normal-deformed triaxial nuclei of the $A\sim130$ mass region showed that the transverse coupling is realized over a frequency range much shorter than for the triaxial superdeformed bands observed in the Lu region, but sufficient to allow experimental observation [@133ce-wobb]. It is therefore natural to imagine a transverse geometry with two $h_{11/2}$ particles coupled to a triaxial core. The resulting bands would be built on the $I^{\pi}=10^+$ maximum aligned states of even-even nuclei from the $A\sim 130$ mass region. The angular momentum of two maximally-aligned $h_{11/2}$ quasiparticles in an even-even nucleus, $I=10$ $\hbar$, being larger than that of one $h_{11/2}$ quasiparticle in an odd-even nucleus, $I=11/2$ $\hbar$, would assure the observation of the wobbling motion over a larger frequency interval before the decoupling from the core under the effect of the Coriolis force. Rotational bands built on the $\pi h_{11/2}^2$ and $\nu h_{11/2}^{-2}$, $I=10^+$ states are known in several nuclei of the $A\sim130$ mass region [@138nd; @134ce; @136nd-plb; @136nd-sun; @136nd-mergel; @136nd-vre]. Calculations performed using different models converge to the conclusion that these 2-qp bands are based on triaxial shapes. In addition to the bands with even spins built on the lowest $10^+$ states, also bands with odd spins which decay to the bands built on the $10^+$ states were identified in the three nuclei $^{134}$Ce, $^{136}$Nd and $^{138}$Nd [@138nd; @134ce; @136nd-plb]. These odd-spin bands were tentatively assigned as two-quasiparticle (2-qp) triaxial configurations in $^{134}$Ce and $^{136}$Nd, while in $^{138}$Nd the odd-spin band which decays to the band built on the $I^{\pi}=10^+$ isomer was interpreted as wobbling collective excitation. Excepting $^{134}$Ce, for which angular distribution results were recently published [@134ce], no detailed experimental information on the angular distribution of the connecting transitions between the 2-qp odd-spin bands of $^{136}$Nd and $^{138}$Nd were reported.
The present paper proposes a new interpretation of the 2-qp odd-spin bands of $^{134}$Ce and $^{136}$Nd in terms of transverse wobbling motion. Simple calculations using the recently introduced Harmonic Frozen Approximation (HFA) [@fra-trans] are used to interpret the observed wobbling frequencies. The adopted deformation of the various configurations are calculated with the Cranked Nilsson-Strutinsky (CNS) model.
Experimental information
========================
The wobbling interpretation is based on the behavior of the wobbling frequency, on the recently published mixing ratios of the $\Delta I=1$ connecting transitions between the bands 7 and 9 of $^{134}$Ce [@134ce], and on the similarities between the observed 2-qp bands of $^{134}$Ce and $^{136}$Nd. The partial level schemes showing the discussed bands of $^{134}$Ce and $^{136}$Nd are given in Figs. \[level-scheme-1\] and \[level-scheme-2\].
The bands 7 and 10 of $^{134}$Ce have been recently reported in Ref. [@134ce]. The absolute values of the mixing ratios extracted from the angular distribution of the 664- and 741-keV, $\Delta I=1$ transitions are significantly larger than one, 2.13(48) and 1.73(52) respectively, clearly indicating their predominant $E2$ character, with 82(5)% and 75(6)% $E2$ component, respectively. We identified two new weak transitions of 691 and 795 keV populating and de-exciting the $(13^+)$ state of band 10, respectively.
The bands 5 and 6 of $^{136}$Nd have been reported in Ref. [@136nd-plb; @136nd-sun], while the bands 3 and 4 have been reported in Refs. [@136nd-sun; @136nd-mergel]. Some of the transitions reported in the previous papers are drawn differently in Fig. \[level-scheme-2\], to reveal the band pattern. We identified two new weak transitions of 401 and 438 keV de-exciting the $14^+$ and $16^+$ states of band 6, respectively. The experimental mixing ratios of the decaying transitions of bands 4, 5 and 6 could not be extracted due to limited statistics.
-1.0 cm -. cm -.0 cm
.0 cm
-.0 cm -. cm -.0 cm
.0 cm
Discussion
==========
The wobbling energies (also called wobbling frequencies) of the 1-phonon ($n_{w}=1$) band defined as the energy difference with respect to the interpolated energies of the $n_{w}=0$ band $$E_{wobb}(I)=E(I,n_w=1)-
[E(I+1,n_w=0)+E(I-1,n_w=0)]/2$$ are shown in Fig. \[e-wobb-exp\]. One can observe the decrease of the wobbling frequencies with increasing spin, which is characteristic for the transverse wobbling mode proposed theoretically in Ref. [@fra-trans] and observed experimentally in $^{135}$Pr [@135pr]. This behavior, together with the predominant $E2$ character of the $\Delta I=1$ decaying transitions of band 7 in $^{134}$Ce strongly support the transverse wobbling interpretation. No detailed angular distribution and polarization information is available for the other bands of $^{134}$Ce and $^{136}$Nd discussed in the present paper, and therefore the electromagnetic character of the transitions is uncertain. The wobbling interpretation of these bands has to wait a definite confirmation from future experiments dedicated to the measurement of the electromagnetic character of the $\Delta I=1$ connecting transitions.
The extracted wobbling frequencies for the bands reflect their different configurations. The two bands of $^{134}$Ce based on the 3207-keV and 3719-keV, $I=10^+$ states are assigned to the same $\nu h_{11/2}^{-2}$ configuration, but corresponding to coexisting shapes with different deformations [@134ce-zemel], or to low-$K$ configurations built on the ground-state band and the $\gamma$-band [@sheikh]. The yrast $\nu h_{11/2}^{-2}$ configuration calculated by the CNS model can be naturally assigned to band 10 based on the yrast $I=10^+$ state. Its deformation changes from $(\varepsilon_2 \sim 0.14, \gamma \sim -35^{\circ})$ at spin $I=12$, to $(\varepsilon_2 \sim 0.12, \gamma \sim -49^{\circ})$ at spin $I=22$ [@134ce]. The deformation of the non-yrast $\nu h_{11/2}^{-2}$ configuration that can be assigned to band 7 built on the 3719-keV, $I=10^+$ state cannot be calculated within the CNS model. Based on the similarity with bands observed in many neighboring nuclei, in particular with band 2 of the presently studied $^{136}$Nd nucleus, one can assume that the deformation of band 7 is similar to that of the $\pi h_{11/2}^{2}$ configuration assigned to band 2 in $^{136}$Nd, which changes from $(\varepsilon_2 \sim 0.21, \gamma \sim 25^{\circ})$ at spin $I=12$, to $(\varepsilon_2 \sim 0.19, \gamma \sim 25^{\circ})$ at spin $I=22$. One can therefore make a step further in the interpretation of the bands based on the two lowest $I=10^+$ states in $^{134}$Ce, combining the two previously interpretations in terms of coexisting shapes with different deformations [@134ce-zemel] and bands built on the ground-state band and the $\gamma$-band [@sheikh]: the band based on the yrast $I=10^+$ state is based on a configuration with lower deformation and the unpaired neutrons aligned along the long axis ($\gamma < 0^{\circ}$), while the band based on the non-yrast $I=10^+$ state is based on a configuration with higher deformation and the unpaired neutrons aligned along the short axis ($\gamma > 0^{\circ}$). The configurations assigned to the bands 7 and 10 of $^{134}$Ce involve therefore two neutron holes in orbitals from the top of the $h_{11/2}$ sub-shell, which, in order to maximize the overlap of the density distribution with the triaxial core, are aligned parallel to either the short or the long axes, respectively. As the hydrodynamical moments of inertia (MoI) are largest for rotation along the intermediate axis, the angular momenta of the two quasiholes are in both cases perpendicular to the angular momentum of the core, corresponding to transverse wobbling. As this is the first time that transverse wobbling bands are proposed in even-even nuclei, we will call them [*2-qp transverse wobbling bands*]{}, to be distinguished from the [*1-qp transverse wobbling bands*]{} observed in odd-even nuclei [@135pr; @133ce-wobb]. However, one note that the wobbling mode built on 2-qp configurations has been recently proposed for a similar $\nu h_{11/2}^{-2}$ configuration in $^{138}$Nd, but without invoking the transverse coupling geometry [@138nd].
Based on the similarity between the bands 10 of $^{134}$Ce and 4 of $^{136}$Nd, one can also interpret band 4 of $^{136}$Nd as a [*2-qp transverse wobbling band*]{} built on the $\nu h_{11/2}^{-2}$ configuration assigned to band 3 towards which it decays [@136nd-paul]. As one can see in Fig. \[e-wobb-exp\], the wobbling phonon energies of bands 10 of $^{134}$Ce and 4 of $^{136}$Nd are smaller than those of band 7 of $^{134}$Ce, and 5 and 6 of $^{136}$Nd, and decrease more sharply with spin, like in the case of the 1-qp wobbling bands in $^{133}$Ce in which one $h_{11/2}$ neutron hole is coupled in a transverse geometry with the triaxial core [@133ce-wobb]. The decrease of the wobbling frequency in band 10 of $^{134}$Ce is observed for the two lowest spins, being followed by an increase which signals the decoupling from the transverse geometry.
Band 5 of $^{136}$Nd decays to band 2 which has a $\pi h_{11/2}^2$ configuration [@136nd-paul]. The wobbling frequency is larger and decreases more slowly than in band 4. The calculated deformation within the CNS formalism changes from $(\varepsilon_2 \sim 0.21, \gamma \sim 25^{\circ})$ at spin $I=12$, to $(\varepsilon_2 \sim 0.19, \gamma \sim 25^{\circ})$ at spin $I=22$, which is therefore larger than that of the $\nu h_{11/2}^{-2}$ configuration assigned to band 10 of $^{134}$Ce and band 4 of $^{136}$Nd.
Band 6 of $^{136}$Nd decays to both band 2 and band 5. Its wobbling frequency is higher than that of band 5 and has the same gradual decrease as that observed in band 5. It is a good candidate for a two-phonon 2-qp wobbling band.
To further investigate if the 2-qp transverse wobbling interpretation of the bands observed in $^{134}$Ce and $^{136}$Nd is realistic, we calculated the wobbling frequencies using the Harmonic Frozen Approximation, in which the angular momenta of the unpaired nucleons are frozen perpendicular to the collective angular momentum of the core [@fra-trans]. The wobbling frequency depends on the three MoI’s and on the spins of the unpaired quasiparticles. It is reasonable to assume that the two unpaired $h_{11/2}$ quasiparticles are coupled to maximum angular momentum $I=10$. For the MoI’s of the $\nu h_{11/2}^{-2}$ configuration we can adopt values similar to those calculated using the RPA model applied to 2-qp configurations employed for $^{138}$Nd [@138nd]: $\mathcal{J}_1=16$, $\mathcal{J}_2=32$ and $\mathcal{J}_3=8$ (in $\hbar^2$/MeV). Note that $\mathcal{J}_1$, $\mathcal{J}_2$ and $\mathcal{J}_3$ correspond to rotations around the long, intermediate and short axes, or alternatively around the 1-axis, 2-axis and 3-axis, respectively. For the MoI’s of the $\pi h_{11/2}^{2}$ configuration we adopt larger values, which correspond to the overall larger deformation calculated within the CNS model : $\mathcal{J}_1=12$, $\mathcal{J}_2=48$ and $\mathcal{J}_3=24$ (in $\hbar^2$/MeV). In this case $\mathcal{J}_1 < \mathcal{J}_3$, which is opposite to the relation $\mathcal{J}_1 > \mathcal{J}_3$ valid for the $\nu h_{11/2}^{-2}$ configuration. These moments of inertia have to be considered as effective moments of inertia, which are larger than the collective ones by the contribution of the unpaired quasiparticles, which are two neutron holes for the 1-axis and two proton particles for the 3-axis. The ratio of the MoI’s for the axes without aligned particles is obtained from the hydrodynamical model with maximal triaxiality $\gamma=30^{\circ}$: $\mathcal{J}_2/\mathcal{J}_3=32/8=4$ for the $\nu h_{11/2}^{-2}$ configuration or $\mathcal{J}_2/\mathcal{J}_1=48/12=4$ for the $\pi h_{11/2}^{2}$ configuration. The contribution of the unpaired quasiparticles to the effective MoI is taken as large as the hydrodynamical MoI corresponding to the axis with aligned quasiparticles: 8 $\hbar^2$/MeV for the $\nu h_{11/2}^{-2}$ configuration and 12 $\hbar^2$/MeV for the $\pi h_{11/2}^{2}$ configuration. As one can see in Fig. \[e-wobb\], an overall good agreement with the experimental values is obtained. Of course, the present schematic calculations have to be confirmed in the future by more realistic and detailed calculations like RPA built on 2-qp configurations.
-.0 cm -. cm -.0 cm
.0 cm
-.0 cm -. cm -.0 cm
.0 cm
Summary
=======
One-phonon and possible two-phonon transverse wobbling bands are proposed for the first time in the even-even nuclei $^{134}$Ce and $^{136}$Nd. The experimental wobbling frequencies decrease with increasing spin, indicating the transverse character of the wobbling motion, with the angular momenta of the two quasiparticles perpendicular to the collective angular momentum of the triaxial core. A candidate for two-phonon wobbling motion is also proposed in $^{136}$Nd. The calculated wobbling frequencies in the harmonic frozen approximation are in good agreement with the experimental values for both $^{134}$Ce and $^{136}$Nd nuclei. To give a solid support to the present wobbling interpretation, future experiments to determine the electromagnetic character of the $\Delta I=1$ connecting transitions and the reduced transitions probabilities are necessary. Detailed calculations like RPA built on 2-qp configurations are also needed to confirm the present schematic calculations.
References {#references .unnumbered}
==========
[99]{} S. Frauendorf and J. Meng, A [**617**]{}, 131 (1997).
S. Frauendorf and F. Dönau, C [**89**]{}, 14322 (2014).
J. T. Matta et al., , 082501 (2015).
C. M. Petrache et al., to be published.
S. W. Ødegård et al., , 5866 (2001).
C. M. Petrache et al., C [**86**]{}, 044321 (2012).
C. M. Petrache et al., to be published.
C. M. Petrache et al., B [**373**]{}, 275 (1996).
C. M. Petrache et al., Phys. Rev. C [**53**]{}, R2581 (1996).
E. Mergel et al., A [**15**]{}, 417 (2002).
D. Vretenar et al., C [**57**]{}, 675 (1998).
A. Zemel et al., Nucl. Phys. A [**383**]{}, 165 (1982).
J.A. Sheikh et al., Nucl. Phys. A [**824**]{}, 58 (2009).
E. Paul et al., C [**36**]{}, 153 (1997).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We analyze [[*[WMAP ]{}*]{}]{}3 year data using the one-point distribution functions to probe the non-Gaussianity in the Cosmic Microwave Background (CMB) Anisotropy data. Computer simulations are performed to determine the uncertainties of the results. We report the non-Gaussianity parameter ${f_{\mathrm{NL}}}$ is constrained to $26<{f_{\mathrm{NL}}}<82$ for Q-band, $12<{f_{\mathrm{NL}}}<67$ for V-band, $7<{f_{\mathrm{NL}}}<64$ for W-band and $23<{f_{\mathrm{NL}}}<75$ for Q+V+W combined data at 95% confidence level (CL).'
author:
- 'E. Jeong$^1$ and G. F. Smoot$^{1,2}$'
bibliography:
- 'ms.bib'
title: 'Probing Non-Gaussianity In The Cosmic Microwave Background Anisotropies: One Point Distribution Function'
---
introduction
============
Non-Gaussianity is one of the most important tests of models of the inflation. Among the various theoretical models on the inflation, slow-roll inflation is currently most lively being studied. There are various predictions on the magnitude of non-Gaussianity based on the simple model of slow-roll inflation and its extensions, ranging from undetectably tiny values to large enough values to be detectable with currently available data [@barnabycline; @battefeldeasther; @calcagni; @creminelli; @bartolo.et.al]. On the other hand, observational works have claimed both detection and non-detection of non-Gaussianity (for reviews on recent works, see [@komatsu.et.al; @Spergel.et.al; @troia.et.al; @creminelli.et.al; @gaztanagawagg]). Among the popular techniques for detecting non-Gaussianity are one-point distribution function fitting, bispectrum, trispectrum and Minkowski functionals. Here, we investigate the one-point distribution functions to probe primordial non-Gaussianity in the CMB anisotropy data. An observed CMB anisotropy at a direction ($\delta T_{obs}$) can be regarded as the superposition of three parts: physical fluctuation of cosmic origin ($\delta T_p$), instrumental noise ($\delta T_n$), and foreground emissions ($T_{fg}$). Since the foreground templates are separately prepared, we start with foreground-removed data of which the CMB anisotropy can be decomposed into two uncorrelated components, $$\label{eq1}
\delta T=\delta T_{obs}-T_{fg}=\delta T_{p}+\delta T_n.$$ The primary source for the cosmic fluctuation of CMB at the large scale is attributed to the Sachs-Wolfe effect which is again triggered by the primordial curvature perturbation. The curvature perturbation $\Phi$ by primordial seed during the inflation is transferred to CMB anisotropy with the relation $$\label{eq2}
\frac{\delta T_p{\left( {{\bf{x}}}\right)}}{T_0}=\eta_{t}\Phi{\left( {{\bf{x}}}\right)}$$ where $T_0=2.725$ K, the thermodynamic temperature of the CMB today, and $\eta_{t}$ is the radiation transfer function. For the super-horizon scale, we take $\eta_{t}=-1/3$ from the Sachs-Wolfe effects. At the first-order of perturbation, we may replace $\Phi =\Phi_g$, where $\Phi_g$ is an auxiliary random Gaussian field with its mean ${\langle \Phi_g\rangle}=0$ and its variance denoted by ${\langle \Phi_g^2\rangle}$. When the second-order perturbation is considered, it is conventional to prescribe the nonlinear coupling of the curvature perturbation as [@komatsuspergel] $$\label{eq3}
\Phi ({{\bf{x}}})\simeq\Phi_g({{\bf{x}}})+{f_{\mathrm{NL}}}{\left( \Phi_g^2({{\bf{x}}})-{\langle \Phi_g^2\rangle}\right)}$$ where ${f_{\mathrm{NL}}}$ is the non-Gaussianity parameter. The second term in is responsible for the non-Gaussianity of the primordial fluctuation. Then, the probability distribution function of the non-Gaussian field $\Phi$ can be derived as $$\begin{aligned}
f_{\Phi}(\Phi)&=&\int f_G(\Phi_g)\delta_D
{\left[ \Phi-\Phi_g-{f_{\mathrm{NL}}}{\left( \Phi_g^2-{\langle \Phi_g^2\rangle}\right)}\right]}d\Phi_g\nonumber\\
&=&\frac{1}{\sqrt{2\pi{\langle \Phi_g^2\rangle}{f_{\mathrm{NL}}}^2{\left( \Phi_+-\Phi_-\right)}^2}}\nonumber\\
& &\times{\left[ \exp{\left( -\frac{\Phi_+^2}{2{\langle \Phi_g^2\rangle}}\right)}
+\exp{\left( -\frac{\Phi_-^2}{2{\langle \Phi_g^2\rangle}}\right)}\right]}\label{eq4}\end{aligned}$$ where $\Phi_{\pm}$ are defined by $$\label{eq5}
\Phi_{\pm}=\frac{1}{2{f_{\mathrm{NL}}}}
{\left[ -1\pm\sqrt{1+4{f_{\mathrm{NL}}}\Phi+4{f_{\mathrm{NL}}}^2{\langle \Phi_g^2\rangle}}\right]}$$ and $\Phi$ has to be limited by the reality of $\Phi_{\pm}$ as $$\label{eq6}
{f_{\mathrm{NL}}}\Phi > -\frac{1}{4}-{f_{\mathrm{NL}}}^2{\langle \Phi_g^2\rangle}.$$ ${\langle \Phi_g^2\rangle}$ can be expressed in terms of $\eta_t$, $T_0$ and $\sigma_{CMB}$, $$\label{eq7}
{\langle \Phi_g^2\rangle}=\frac{1}{4{f_{\mathrm{NL}}}^2}{\left[ -1+\sqrt{1+8
{\left( \frac{{f_{\mathrm{NL}}}\sigma_{CMB}}{\eta_t T_0}\right)}^2}\right]}.$$ For a pixelized CMB anisotropy data set, the probability distribution function for Gaussian instrumental noise becomes $$\label{eq8}
f_N(\delta T_n)=\frac{1}{N_{pix}}\sum_{i=1}^{N_{pix}}
\frac{1}{\sqrt{2\pi\sigma_0^2/n_i}}
\exp{\left[ -\frac{\delta T_n^2}{2\sigma_0^2/n_i}\right]}$$ where $n_i$ is the effective number of measurements at the $i_{th}$ pixel and $\sigma_0$ represents the dispersion of the instrumental noise per observation ($\sigma_0$=2.1898, 3.1249, 6.5112 mK for Q, V, W-band, respectively [@Limon.et.al]). Now, it is straightforward to express the probability density function for $\delta T$ in an integral form, $$\begin{aligned}
f(\delta T)&=&\int f_{\delta T_p}(\delta T_p)f_N(\delta T_n)\nonumber\\
& &\times\delta_D{\left( \delta T-\delta T_p-\delta T_n\right)}
d\delta T_p d\delta T_n\nonumber\\
&=&\int f_{\Phi}(\Phi )f_N(\delta T_n)\nonumber\\
& &\times\delta_D{\left( \delta T-\eta_t T_0\Phi-\delta T_n\right)}d\Phi d\delta T_n.
\label{eq9}\end{aligned}$$ The probability density function derived in explicitly contains the non-Gaussianity parameter ${f_{\mathrm{NL}}}$, and it can serve as the prediction of one-point distribution function with a given ${f_{\mathrm{NL}}}$ for a (ideally) foreground-removed CMB anisotropy data set to estimate the magnitude of deviation from Gaussian distribution in a quantitative manner.
Application to [[*[WMAP ]{}*]{}]{}data
======================================
We use the three channels of [[*[WMAP ]{}*]{}]{}3 year CMB anisotropy data sets (Q (33GHz)-, V (61GHz)-, W (94GHz)-band) which contain dominant signal over contaminations to investigate the non-Gaussianity of the CMB anisotropy data. To remove the foreground emissions, the Maximum Entropy Method (MEM) maps of the synchrotron, free-free and thermal dust are used. The Kp0-mask is applied to the sky maps to remove the intense Galactic emissions and scattered bright point sources, which leaves 76.5% of the sky. We also prepare a combined map (Q+V+W) by taking a weighted sum for a pixel temperature, $$\label{eq10}
\delta T{\left( {{\bf{x}}}\right)}=\frac{\sum_i\delta T_i{\left( {{\bf{x}}}\right)}n_i{\left( {{\bf{x}}}\right)}/\sigma_{0i}^2}
{\sum_in_i{\left( {{\bf{x}}}\right)}/\sigma_{0i}^2},\quad i={\mathrm{Q,\: V,\: W}}$$ where $n_i{\left( {{\bf{x}}}\right)}$ is the effective number of measurements at the pixelized position ${{\bf{x}}}$ and $\sigma_{0i}$ is the dispersion of the instrumental noise of the $i_{\mathrm{th}}$ channel. We can trace the effective variance of the instrumental noise as a result of weighted sum defined in as $$\label{eq11}
\sigma^2{\left( {{\bf{x}}}\right)}={\left[ \sum_in_i{\left( {{\bf{x}}}\right)}/\sigma_{0i}^2\right]}^{-1}
,\quad i={\mathrm{Q,\: V,\: W}}.$$ The sky map data are degraded from $N_{\mathrm{side}}=512$ ($6.87^{\prime}$ pixel) to $N_{\mathrm{side}}=128$ ($27.48^{\prime}$ pixel) where the number of pixels in a full sky map is given by $12\times N_{\mathrm{side}}^2$. The purpose of demotion of the resolution is to suppress the small scale fluctuation which is dominated by the instrumental noise. We perform the $\chi^2$-test for the goodness of fit for the probability density function given in as a prediction to the observed probability density function which is directly calculated from the [[*[WMAP ]{}*]{}]{}data. Figure \[fig1\] and Table \[t1\] show the results of $\chi^2$ fitting of [[*[WMAP ]{}*]{}]{}data sets with varying ${f_{\mathrm{NL}}}$ as a free parameter. All data sets are best fitted at positive ${f_{\mathrm{NL}}}$ (dubbed ${f_{\mathrm{NL}}}^{{\left( \mathrm{opt}\right)}}$) which are consistent with one another as well as the results with previous work [@Spergel.et.al] within the statistical errors.
{width="6.2cm"} {width="6.2cm"} {width="6.2cm"} {width="6.2cm"}
Map
---------------------------------------------------- ---- ----- ----- ----- ----- ----- ----- -----
${f_{\mathrm{NL}}}^{{\left( \mathrm{opt}\right)}}$
68% -9 86 -6 101
95% 34 71 -37 113 -20 89 -29 123
99% 3 102 -49 126 -37 106 -40 134
DOF
: \[t1\]The results of $\chi^2$-tests for Goodness-of-fit for [[*[WMAP ]{}*]{}]{} 3 year data. The percentages on the first column represent the tail probabilities at which the statistic $\chi^2$ would be smaller than the observed. DOF on the bottom row stands for [*[Degrees of freedom]{}*]{}.
Simulation And Statistical Uncertainties
========================================
As is shown in Figure \[fig1\], [[*[WMAP ]{}*]{}]{}data fit well with finite range of the non-Gaussianity parameter ${f_{\mathrm{NL}}}$. We pick ${f_{\mathrm{NL}}}^{{\left( \mathrm{opt}\right)}}$ as the representative magnitude of non-Gaussianity for a data set, and carry out the Monte-Carlo simulations to test the pertinence of ${f_{\mathrm{NL}}}^{{\left( \mathrm{opt}\right)}}$ as a proper measure of non-Gaussianity for a data set in a quantitative manner. A simulated data set is prepared as follows: First, a Gaussian field $\Phi_g$ with its variance equal to is generated and we use it to generate $\Phi$-field of which the deviation from Gaussianity is denoted by ${f_{\mathrm{NL}}}$. Second, we prepare a noise map in which each pixel contains a random value picked from a normal distribution with variance $\sigma_0^2/n_i$ as defined in and add this to $\Phi$-field. A simulated map generated in this way has the same noise structure and the dispersion of physical fluctuation ($\sigma_{CMB}$) as a real data set. Since a data set contains nontrivial instrumental noise and the number of pixels is finite, the returned magnitude of non-Gaussianity (${f_{\mathrm{NL}}}^{{\left( \mathrm{opt}\right)}}$) would have some uncertainty. For a given value of ${f_{\mathrm{NL}}}$, we repeat the simulation and find that the algorithm returned an unbiased, normal distribution of ${f_{\mathrm{NL}}}^{{\left( \mathrm{opt}\right)}}$ which is centered at the input value of ${f_{\mathrm{NL}}}$. Thus, from the results of the simulations, we are able to set the bounds on ${f_{\mathrm{NL}}}$ for the real data. The simulation results and error bands are plotted in Figure \[fig2\] and the deduced uncertainties for ${f_{\mathrm{NL}}}$ are summarized in Table \[t2\]. It is very remarkable that this analysis strongly disfavors the null hypothesis (${f_{\mathrm{NL}}}=0$) and all the data sets show consistent results within the statistical errors. The results are also consistent with the works by the [[*[WMAP ]{}*]{}]{}team ($-54<{f_{\mathrm{NL}}}<114$ at 95% CL from bispectrum [@Spergel.et.al]) but with much tighter limits and more importantly, it excludes ${f_{\mathrm{NL}}}=0$ at 95% CL.
{width="6.2cm"} {width="6.2cm"} {width="6.2cm"} {width="6.2cm"}
Map
-------------------- ---- ---- ---- ---- ---- ---- ---- ----
$N_{\mathrm{sim}}$
68% 40 68 26 53 21 50 36 62
95% 26 82 12 67 7 64 23 75
99% 12 96 -1 80 -7 78 9 88
: \[t2\]Summary of results from Simulations with [[*[WMAP ]{}*]{}]{}3 year data profiles and the bounds at three confidence levels from the simulations.
Conclusion
==========
We developed an algorithm that uses the one-point distribution function to investigate the non-Gaussianity of CMB anisotropy data, and applied it to [[*[WMAP ]{}*]{}]{}3 year data. We found that the null result (${f_{\mathrm{NL}}}$=0) is manifestly excluded at 95% CL. The estimated magnitude of non-Gaussianity parameter is $23<{f_{\mathrm{NL}}}<75$ at 95% CL and $9<{f_{\mathrm{NL}}}<88$ at 99% CL for the (Q+V+W)-combined map. Since the quadratic term in takes a generic form of Taylor series for a perturbative expansion, it is a good possibility that the observed non-Gaussianity in this work is a combined effects of various physical processes, while the primordial seeds are very likely to be the leading one. There are two premises we have taken in developing the algorithm, which, provided they are not precise enough, could cause non-Gaussianity of not cosmic but systematic origin: (1) the probability distribution function of the instrumental noise for each pixel is centered at zero, and (2) the foreground emissions are removed efficiently enough in the foreground-removed maps. The first condition can be broken when the thermal and radiation environments of the [[*[WMAP ]{}*]{}]{}satellite in its orbit are taken into account, while the [[*[WMAP ]{}*]{}]{}team assessed they are insufficient to influence the science data [@Limon.et.al]. So, we tested the effects of the alternative noise distributions with a random mean in each of the Gaussian distribution in and the algorithm was not misled to show non-Gaussianity within the statistical error. It is difficult to directly estimate how much residual foreground emissions after foreground subtraction would affect the one-point distribution function. We solely rely on the quality of foreground templates and it is remarkably successful, showing that the observed total Galactic emission matches the model to less than 1% [@bennett.et.al2; @Hinshaw.et.al]. We also analyzed simulated maps which are (Gaussian map + foreground templates), and all the templates for Q, V and W-channel showed negative values of the non-Gaussianity parameter with ${\left| {f_{\mathrm{NL}}}\right|}\sim\mathcal{O}{\left( 10^1\right)}$ at the resolution $N_{\mathrm{side}}=512$.
We would like to thank Dr. Sara Ricciardi and Dr. Oliver Zahn for valuable discussion and comments on the foreground emission and other topics. Computer simulation and data analysis with [[*[WMAP ]{}*]{}]{}data set were done using the HEALPix[@Gorski.et.al]. This work was supported by LBNL and the Department of Physics at University of California, Berkeley.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We analyze representative ill-posed scenarios of tomographic PIV with a focus on conditions for unique volume reconstruction. Based on sparse random seedings of a region of interest with small particles, the corresponding systems of linear projection equations are probabilistically analyzed in order to determine (i) the ability of unique reconstruction in terms of the imaging geometry and the critical sparsity parameter, and (ii) sharpness of the transition to non-unique reconstruction with ghost particles when choosing the sparsity parameter improperly. The sparsity parameter directly relates to the seeding density used for PIV in experimental fluids dynamics that is chosen empirically to date. Our results provide a basic mathematical characterization of the PIV volume reconstruction problem that is an essential prerequisite for any algorithm used to actually compute the reconstruction. Moreover, we connect the sparse volume function reconstruction problem from few tomographic projections to major developments in compressed sensing.'
address:
- 'Image and Pattern Analysis Group, University of Heidelberg, Speyerer Str. 6, 69115 Heidelberg, Germany'
- 'Institute of Aerodynamics and Flow Technology, German Aerospace Center, Bunsenstr. 10, 37073 Göttingen, Germany'
author:
- 'Stefania Petra, Christoph Schnörr, Andreas Schröder'
title: '[C]{}ritical [P]{}arameter [V]{}alues and [R]{}econstruction [P]{}roperties of [D]{}iscrete [T]{}omography: [A]{}pplication to [E]{}xperimental [F]{}luid [D]{}ynamics'
---
Introduction
============
Motivated by an application from fluid dynamics [@Els-06], we investigate conditions for an highly underdetermined nonnegative system of linear equations to have a unique nonnegative solution, provided it is sparse. The sought solution is a sparse 3D image of particles immersed in a fluid known only from its projection. This projection represents the simultaneous 2D images captured by few camera sensors from different viewing directions, see Fig. \[fig:TomoPIV\]. The reconstruction of the 3D image from the 2D images employs a standard algebraic image reconstruction model, which assumes that the image consists of an array of unknowns (cells, voxels), and sets up algebraic equations for the unknowns in terms of measured projection data. The latter are the pixel entries in the recorded 2D images that represent the integration of the original 3D light intensity distribution along the pixels line-of-sight. The number of cameras is limited to 3 to 6 cameras, typically 4. As a consequence, the reconstruction problem becomes severely ill-posed.
Thus, we consider a huge and severely underdetermined linear system $$\label{eq:Ax=b}
A x = b,\qquad A \in {\mathbb{R}}^{m \times n},\qquad
m \ll n,$$ with the following properties: a *very sparse* nonnegative measurement matrix $A$ with *constant small support* of length $\ell$ of all column vectors, $$\label{eq:Axb-properties}
A \geq 0,\; x \geq 0,\qquad \operatorname{supp}({{A}_{\bullet,j}}) = \ell \ll m,\qquad \forall j = 1,\dotsc,n.$$ and a nonnegative $k$-sparse solution vector $x$. While $\ell$ equals the number of cameras, $k$ is equal related to the particle density (equal, in the present work, proportional, in practice). We also consider the discretization (or resolution) parameter $d$, and relate it to the number of discretization cells and number of measurements: $$\begin{aligned}
\label{eq:m_n_2D}
m&=\ell\cdot O(d), \quad n=O(d^2), \quad {\rm \ in \ 2D},\\
\label{eq:m_n_3D}
m&=\ell\cdot O(d^2), \quad n=O(d^3), \quad {\rm in \ 3D}\ .\end{aligned}$$
We will answer the following question: what is the maximal number of particles, depending on the image resolution parameter $d$, that can be localized uniquely? Formally, we want to relate the *exact* recovery of $x$ from it’s noiseless measurements $b$ to the sparsity $k$ and to the dimensions of $m,n$ of the projection matrix $A$. Moreover, we investigate critical values of the sparsity parameter $k$ such that most $k$-sparse nonnegative solutions are the unique nonnegative solutions of with high probability.
![ Typical camera arrangements: in circular configuration (right) or all in line (left). []{data-label="fig:TomoPIV"}](Gray_setupTomoPIV "fig:"){height="0.2\textheight"} ![ Typical camera arrangements: in circular configuration (right) or all in line (left). []{data-label="fig:TomoPIV"}](Gray_FreeJetLaVision.png "fig:"){height="0.2\textheight"}
Related Work
------------
Research on compressed sensing [@CompressedSensing-06; @Candes-CompressiveSampling-06] focuses on properties of underdetermined linear systems that guarantee exact recovery of sparse or compressible signals $x$ from measurements $b$. Donoho and Tanner [@DonohoT10; @DonTan10_Precise] have computed sharp reconstruction thresholds for random measurement matrices, such that for given a signal length $n$ and numbers of measurements $m$, the maximal sparsity value $k$ which guarantees perfect reconstruction can be determined explicitly. The authors derived their results by connecting it to problem from geometric probability that $n$ points in general position in ${{\mathbb R^m}}$ can be linearly separated [@Wendel-62]. This holds with probability $\Pr(n,m)=1$ for $n/m \leq 1$, and with $\Pr(n,m) \to 1$ if $m \to \infty$ and $1 \leq n/m < 2$, where $$\label{eq:WendelPR}
\Pr(n,m) = \frac{1}{2^{n-1}} \sum_{i=0}^{m-1}
\binom{n-1}{i}\ .$$ The authors show in [@DonohoT10 Thm. 1.10] that the probability of uniqueness of a $k$-sparse nonnegative vector equals $\Pr(n-m,n-k)$, provided $A$ satisfies certain conditions which do not hold in our considered application. Likewise, by exploiting again Wendel’s theorem, Mangasarian and Recht showed [@Man09ProbInteger] that a binary solution is most likely unique if $m/n>1/2$, provided that $A$ comes from a centrosymmetric distribution. Unfortunately, the underlying distribution $A$ lacks symmetry with respect to the origin. However, we recently showed [@Petra-Schnoerr-12a] for a three camera scenario that there are thresholds on sparsity (i.e. density of the particles), below which exact recovery will succeed and above which it fails with high probability. This explicit thresholds depend on the number of measurements (recording pixel in the camera arrays). The current work investigates further geometries and focus on an *average case analysis* of conditions under which *uniqueness* of $x$ can be expected with *high probability*. A corresponding tail bound implies a weak threshold effect and criterion for adequately choosing the value of the sparsity parameter $k$.
Notation
--------
$|X|$ denotes the cardinality of a finite set $X$ and $[n] = \{1,2,\dotsc,n\}$ for $n \in {\mathbb{N}}$. We will denote by $\|x\|_{0} = |\{i \colon x_{i} \neq 0 \}|$ and ${\mathbb{R}}_{k}^{n} = \{ x \in {\mathbb{R}}^{n} \colon \|x\|_{0} \leq k \}$ the set of $k$-sparse vectors. The corresponding sets of non-negative vectors are denoted by ${\mathbb{R}}_{+}^{n}$ and ${\mathbb{R}}_{k,+}^{n}$, respectively. The support of a vector $x\in {\mathbb{R}}^{n}$, ${\mathrm{supp}}(x) \subseteq [n]$, is the set of indices of non-vanishing components of $x$. For a finite set $S$, the set ${\mathcal{N}}(S)$ denotes the union of all neighbors of elements of $S$ where the corresponding relation (graph) will be clear from the context.
${{A}_{\bullet,i}}$ denotes the $i$-th column vector of a matrix $A$. For given index sets $I, J$, matrix $A_{I J}$ denotes the submatrix of $A$ with rows and columns indexed by $I$ and $J$, respectively. $I^{c},
J^{c}$ denote the respective complement sets. Similarly, $b_{I}$ denotes a subvector of $b$.
${\mathbb{E}}[\cdot]$ denotes the expectation operation applied to a random variable and $\Pr(A)$ the probability to observe an event $A$.
Graph Related Properties of Tomographic Projection Matrices {#sec:expander}
===========================================================
Recent trends in compressed sensing [@RIP-P-SMM-08; @XuHassibi_Expander] tend to replace random dense matrices by adjacency matrices of ”high quality” expander graphs. Explicit constructions of such expanders exist, but are quite involved. However, random $m\times n$ binary matrices with nonreplicative columns that have $\lfloor \ell n\rfloor$ entries equal to $1$, perform numerically extremely well, even if $\ell$ is small, as shown in [@RIP-P-SMM-08]. In [@HassibiIEEE; @Petra2009] it is shown that perturbing the elements of adjacency matrices of expander graphs with low expansion, can also improve performance.
Preliminaries
-------------
For simplicity, we will restrict on situations were the intersection lengths of projection rays corresponding to each camera with each discretization cell are all equal. Thus, we can make the assumption that the entries of $A$ are binary. It will be useful to denote the set of cells by $C = [n]$ and the set of rays by $R = [m]$. The incidence relation between cells and rays is then given by $$\label{eq:def-A}
(A)_{ij}=
\begin{cases} 1,&
\quad \text{if $j$-th ray intersects $i$-th cell},\\
0, & \quad \text{otherwise}, \end{cases}$$ for all $i\in[m]$, $j\in[n]$. Thus, cells and rays correspond to columns and rows of $A$.
This gives the equivalent representation in terms of a *bipartite graph* $G = (C,R;E)$ with left and right vertices $C$ and $R$, and edges $(c,r) \in E$ iff $(A)_{rc} = 1$. $G$ has *constant left-degree* $\ell$ equal to the number of projecting directions.
For any non-negative measurement matrix $A$ and the corresponding graph, the set $${{\mathcal N}}(S) = \{i\in[m] \colon A_{ij}>0,\, j\in S\}$$ contains all neighbors of $S$. The same notation applies to neighbors of subsets $S \subset [m]$ of right nodes. Further, we will call any non-negative matrix *adjacency matrix*, based on the incidence relation of its non-zero entries.
If $A$ is the non-negative adjacency matrix of a bipartite graph with constant left degree $\ell$, the **perturbed matrix** $\tilde A$ is computed by uniformly perturbing the non-zero entries $A_{ij} > 0$ to obtain $\tilde A_{ij} \in [A_{ij}-{\varepsilon},A_{ij}+{\varepsilon}]$, and by normalizing subsequently all column vectors of $\tilde A$. In practice, such perturbation can be implemented by discretizing the image by radial basis functions of unequal size or and choose their locations on an irregular grid.
The following class of graphs plays a key role in the present context and in the field of compressed sensing in general.
\[def:Expander\] A $(\nu,\delta)$-unbalanced expander is a bipartite simple graph $G
= (L,R;E)$ with constant left-degree $\ell$ such that for any $X \subset L$ with $|X| \leq \nu$, the set of neighbors ${{\mathcal N}}(X) \subset R$ of $X$ has at least size $|{{\mathcal N}}(X)| \geq \delta \ell |X|$.
Recovery of a $k$-sparse nonnegative solution via an $(\nu,\delta)$-unbalanced expander was derived in [@WangIEEE]. It employs the smallest expansion constant $\delta$ with respect to other similar results in the literature.
\[thm:wang\] Let $A$ be the adjacency matrix of a $(\nu,\delta)$-unbalanced expander and $1 \geq \delta>\frac{\sqrt{5}-1}{2}$. Then for any $k$-sparse vector $x^*$ with $k\le \frac{\nu}{(1+\delta)}$, the solution set $\{x \colon Ax=Ax^*,x \ge 0\}$ is a singleton.
Now let $A$ denote the tomographic projection matrix, and consider a subset $X \subset C$ of $|X|=k$ columns and a corresponding $k$-sparse vector $x$. Then $b = A x$ has support ${\mathcal{N}}(x)$, and we may remove the subset of ${\mathcal{N}}(X)^{c} = ({\mathcal{N}}(X))^{c}$ rows from the linear system $A x = b$ corresponding to $b_{r}=0,\, \forall r \in R$. Moreover, based on the observation ${\mathcal{N}}(X)$, we know that $$\label{eq:reduced-dimensions}
X \subseteq {\mathcal{N}}({\mathcal{N}}(X))
\qquad\text{and}\qquad
{\mathcal{N}}({\mathcal{N}}(X)^{c}) \cap X = \emptyset.$$
We continue by formalizing the system reduction just described.
The *reduced system* corresponding to a given non-negative vector $b$, $$\label{eq:red-system}
A_{red} x = b_{red},\qquad A_{red} \in
{\mathbb{R}}_{+}^{m_{red} \times n_{red}},$$ results from $A, b$ by choosing the subsets of rows and columns $$\label{eq:def-RbCb}
R_{b} := \operatorname{supp}(b),\qquad
C_{b} := {\mathcal{N}}(R_{b}) \setminus {\mathcal{N}}(R_{b}^{c})$$ with $$\label{def:mn-red}
m_{red} := |R_{b}|,\qquad
n_{red} := |C_{b}|.$$
Note that for a vector $x$ and the bipartite graph induced by the measurement matrix $A$, we have the correspondence (cf. ) $$X = \operatorname{supp}(x),\qquad
R_{b} = {\mathcal{N}}(X),\qquad
C_{b} = {\mathcal{N}}({\mathcal{N}}(X)) \setminus {\mathcal{N}}({\mathcal{N}}(X)^{c}).$$ We further define $$\label{def:feasSet}
{{\mathcal S}}^+:=\{x \colon Ax=b, x\ge 0\}$$ and $$\label{def:redfeasSet}
{{\mathcal S}}_{red}^+:=\{x \colon A_{R_b C_b}x=b_{R_b}, x\ge 0\}\ .$$ The following proposition asserts that solving the reduced system will always recover the support of the solution to the original system $A x = b$
[@Petra-Schnoerr-12a Prop. 5.1]\[prop:redfeasSet\] Let $A\in {\mathbb{R}}^{m\times n}$ and $b\in{\mathbb{R}}^m$ have nonnegative entries only, and let ${{\mathcal S}}^+$ and ${{\mathcal S}}_{red}^+$ be defined by and , respectively. Then $$\label{eq:feasSet}
{{\mathcal S}}^+=\{x\in {\mathbb{R}}^n \colon x_{(C_b)^c}=0\;\text{ and }\; x_{C_b}\in
{{\mathcal S}}_{red}^+\}.$$
Consequently, we can restrict the linear system $A x = b$ to the subset of columns ${\mathcal{N}}({\mathcal{N}}(X)) \setminus {\mathcal{N}}({\mathcal{N}}(X)^{c}) \subset C$, and only consider properties of this reduced systems.
Guaranteed Uniqueness {#sec:uniqueness}
---------------------
Uniqueness of $x\in{{\mathbb R^n}}_{\delta k,+}$ is guaranteed if all $k$ or less-sparse supported on ${\rm supp}(x)$ induce overdetermined reduced systems with $m_{red}/n_{red}> \delta \ell \ge \frac{\sqrt{5}-1}{2}\ell$.
[@Petra-Schnoerr-12a Th. 3.4]\[thm:Wang\] Let $A$ be the adjacency matrix of a bipartite graph such that for all random subsets $X \subset C$ of $|X| \leq k$ left nodes, the set of neighbors ${{\mathcal N}}(X)$ of $X$ satisfies $$\label{eq:condition-Wang}
|{{\mathcal N}}(X)| \geq \delta
\ell |{{\mathcal N}}({{\mathcal N}}(X))\setminus {{\mathcal N}}({{\mathcal N}}(X)^c)|
\qquad\text{with}\qquad
\delta>\frac{\sqrt{5}-1}{2}.$$ Then, for any $\delta k$-sparse nonnegative vector $x^*$, the solution set $\{x \colon Ax=Ax^*,x \ge 0\}$ is a singleton.
For perturbed matrices uniqueness is guaranteed for square reduced systems, and thus less high sparsity values.
[@Petra-Schnoerr-12a Th. 3.4]\[thm:SPCS1\] Let $A$ be the adjacency matrix of a bipartite graph such that for all subsets $X \subset C$ of $|X| \leq k$ left nodes, the set of neighbors ${{\mathcal N}}(X)$ of $X$ satisfies $$\label{eq:Hassibi-condition}
|{{\mathcal N}}(X)| \geq \delta
\ell |{{\mathcal N}}({{\mathcal N}}(X))\setminus {{\mathcal N}}({{\mathcal N}}(X)^c)|
\qquad\text{with}\qquad
\delta > \frac{1}{\ell}.$$ Then, for any $k$-sparse vector $x^*$, there exists a perturbation $\tilde A$ of $A$ such that the solution set $\{x \colon \tilde Ax=\tilde Ax^*,x \ge 0\}$ is a singleton.
Recovery via perturbed underdetermined reduced systems is possible and our numerical results from Section \[sec:num\] suggest the following.
\[conj:SPCS2\] Let $A$ be the adjacency matrix of a bipartite graph such that for all subsets $X \subset C$ of $|X| \leq k$ left nodes, the set of neighbors ${{\mathcal N}}(X)$ of $X$ satisfies $$\label{eq:inv-condition}
|{{\mathcal N}}(X)| \geq \frac{1+\delta}{\ell}
|{{\mathcal N}}({{\mathcal N}}(X))\setminus {{\mathcal N}}({{\mathcal N}}(X)^c)|
\qquad\text{with}\qquad
\delta >\frac{\sqrt{5}-1}{2}.$$ Then, for any $\frac{k}{\ell}$-sparse vector $x^*$, there exists a perturbation $\tilde A$ of $A$ such that the solution set $\{x \colon \tilde Ax=\tilde Ax^*,x \ge 0\}$ is a singleton.
The consequences of Propositions \[thm:Wang\], \[thm:SPCS1\] and Conjecture \[conj:SPCS2\] are investigated in the following sections \[sec:hexagon\_red\_dim\] and \[sec:cube\_red\_dim\] by working out critical values of the sparsity parameter $k$ for which the respective conditions are satisfied with high probability.
3 Cameras - Left Degree equals 3 {#sec:Hex3C}
================================
In this section, we analyze the imaging set-up depicted in Figure \[fig:3\_cam\], left panel, which also represents typical 3D scenarios encountered in practice with a coarse resolution only along the third coordinate, as shown by Figure \[fig:3\_cam\], center panel.
![Sketch of a 3-cameras setup in 2D. Left: The hexagonal area discretized in $3\frac{d^2+1}{4}$ equally sized cells projected on 3 1D cameras. The resulting projection matrix $A\in\{0, 1\}^{m\times n}$ is underdetermined, with $m=3d$ and $n=3\frac{d^2+1}{4}$, where $\frac{d-1}{2}+1$ is the number of cells on each hexagon edge. Middle: This geometry can be easily extended to 3D by enhancing both cameras and volume by one dimension, thus representing scenarios of practical relevance when cameras are aligned on a line. Right: When considering a square area along with three projection directions (two orthogonal, one diagonal) one obtains a projection matrix with analogous reconstruction properties. The projection matrix *equals* up to scaling the previous projection matrix corresponding to the hexagonal area if we remove the $2\cdot \frac{d^2-1}{8}$ cells in the marked corners along with incident rays.[]{data-label="fig:3_cam"}](Gray_Setup-2D-Hex.png "fig:"){width="35.00000%"} ![Sketch of a 3-cameras setup in 2D. Left: The hexagonal area discretized in $3\frac{d^2+1}{4}$ equally sized cells projected on 3 1D cameras. The resulting projection matrix $A\in\{0, 1\}^{m\times n}$ is underdetermined, with $m=3d$ and $n=3\frac{d^2+1}{4}$, where $\frac{d-1}{2}+1$ is the number of cells on each hexagon edge. Middle: This geometry can be easily extended to 3D by enhancing both cameras and volume by one dimension, thus representing scenarios of practical relevance when cameras are aligned on a line. Right: When considering a square area along with three projection directions (two orthogonal, one diagonal) one obtains a projection matrix with analogous reconstruction properties. The projection matrix *equals* up to scaling the previous projection matrix corresponding to the hexagonal area if we remove the $2\cdot \frac{d^2-1}{8}$ cells in the marked corners along with incident rays.[]{data-label="fig:3_cam"}](Gray_HexagonVolumeSetup.png "fig:"){width="30.00000%"} ![Sketch of a 3-cameras setup in 2D. Left: The hexagonal area discretized in $3\frac{d^2+1}{4}$ equally sized cells projected on 3 1D cameras. The resulting projection matrix $A\in\{0, 1\}^{m\times n}$ is underdetermined, with $m=3d$ and $n=3\frac{d^2+1}{4}$, where $\frac{d-1}{2}+1$ is the number of cells on each hexagon edge. Middle: This geometry can be easily extended to 3D by enhancing both cameras and volume by one dimension, thus representing scenarios of practical relevance when cameras are aligned on a line. Right: When considering a square area along with three projection directions (two orthogonal, one diagonal) one obtains a projection matrix with analogous reconstruction properties. The projection matrix *equals* up to scaling the previous projection matrix corresponding to the hexagonal area if we remove the $2\cdot \frac{d^2-1}{8}$ cells in the marked corners along with incident rays.[]{data-label="fig:3_cam"}](Gray_Hex_Square.png "fig:"){width="30.00000%"}
Imaging Geometry {#sec:hexagon_geom}
----------------
Cell centers $x_{c}$ of hexagonal cells $c \in C$ that partition a region of interest, are given by lattice points corresponding to integer linear combinations of two vectors $d^{i},\,i=1,2$, $$\begin{gathered}
x_{c} = i_{1} d^{1} + i_{2} d^{2},\qquad
d^{1} = \frac{1}{2} {\begin{pmatrix}}\sqrt{3} \\ 1 {\end{pmatrix}},\quad
d^{2} = {\begin{pmatrix}}0 \\ 1 {\end{pmatrix}},\quad
(i_{1},i_{2}) \in {\mathcal{I}},\end{gathered}$$ for the index set $${\mathcal{I}} = \big\{(i,j) \colon
-(d-1)/2 \leq i,j \leq (d-1)/2,\; |i+j| \leq (d-1)/2
\big\},$$ with problem size $d \in {\mathbb{N}}$ that we assume (in this section) to be an odd number for simplicity. The number $|R|$ of projections $r \in R = R_{1} \cup R_{2} \cup R_{3}$, where $R_{i},\,i=1,2,3$, corresponds to the rays of direction $i$, is $$\label{eq:def-NR}
|R| = 3 |R_{i}| = 3 d.$$ The number of cells incident with projection rays ranges over the interval $$\{(d+1)/2, (d+1)/2+1, \dotsc, d\}$$ from the periphery towards the center. Thus, indexing with $r$ each projection ray along any particular direction $R_{i},\,i=1,2,3$, from one side of the hexagon across the center towards the opposite side, the numbers of cells incident with ray $r$ is $$\label{eq:def-Nc}
|r| \in \{(d+1)/2,\dotsc, d,\dotsc, (d+1)/2\},\qquad
r \in R_{i},\; i=1,2,3.$$ The total number of cells is $$\label{eq:def-NC}
|C| = \sum_{r \in R_{i}} |r|
= 2 \sum_{j=(d+1)/2}^{d-1} j + d
= \frac{1}{4}(3 d^{2}+1),\quad i \in \{1,2,3\},$$ and $$\label{eq:sum-NR-NC}
\sum_{r \in R} |r| = 3 |C|.$$ Accordingly, the system of equations representing the imaging geometry depicted by Figure \[fig:3\_cam\], left panel, has dimensions $$\label{eq:def-Ab}
A x = b,\qquad\qquad A \in \{0,1\}^{|R| \times |C|},\qquad
b \in {\mathbb{R}}^{|R|}.$$ Note that $|R| \ll |C|$. For further reference, we define the quantities
\[eq:def-pqr\] $$\begin{aligned}
q_{r} &= \frac{|r|}{|C|}, &
p_{r} &= 1-q_{r}, \\
{\underline{q}}_{d} &= \min_{r \in R} q_{r}, &
{\overline{q}}_{d} &= \max_{r \in R} q_{r}, \\
{\underline{p}}_{d} &= 1-{\overline{q}}_{d}, &
{\overline{p}}_{d} &= 1-{\underline{q}}_{d},\end{aligned}$$
and list some further relations and approximations for large $d$,
\[eq:basic-approximations\] $$\begin{aligned}
{\underline{q}}_{d} &= \frac{2(d+1)}{3 d^{2}+1}
\approx \frac{2}{3 d}, &
{\overline{q}}_{d} &= \frac{1}{3} \frac{|R|}{|C|}
\approx \frac{4}{3 d}, \\
|R| {\underline{q}}_{d} &\approx 2, &
|R| {\overline{q}}_{d} &\approx 4.\end{aligned}$$
Dimensions of Reduced Systems {#sec:hexagon_red_dim}
-----------------------------
We estimate the *expected* dimensions of the reduced system based on uniformly selecting $k$ cells at random locations.
To each projection ray $r \in R$, we associate a binary random variable $X_{r}$ taking the value $X_{r}=1$ if *not any* of the $k$ cells is incident with ray $r$, and $X_{r}=0$ otherwise. We call the event $X_{r}=1$ *zero-measurement*.
We are interested in the random variable $$\label{eq:def-X}
X = \sum_{r \in R} X_{r}$$ that determines the number of projection rays not incident with any of the $k$ cells, that is the number of zero measurements. We set $$\label{eq:def-NR0}
N_{R}^{0} := {\mathbb{E}}[X],\qquad
N_{R} := |R| - N_{R}^{0}.$$ Hence, $N_{R}$ is the *expected* size of the support $m_{red} = |\operatorname{supp}(b)|$ of the measurement vector $b$.
\[rem:Xr-dependence\] Note that random variables $X_{r}$ are *not* independent because different projection rays may intersect. This dependency does not affect the expected value of $X$, but it does affect the deviation of observed values of $X$ from its expected value – cf. Section \[sec:hexagon-tailbound\].
\[rem:ass-replace\] We do *not* assume in the derivation below that $k$ *different* cells are selected. In fact, a single cell may be occupied by more than a single particle in practice, because real particles are very small relative to the discretization cells $c$. The imaging optics enlarges the appearance of particles, and the action of physical projection rays is adequately represented by linear superposition.
\[def:k\] We refer to the number $k$ introduced above as sparsity parameter. Thus, highly sparse scenarios correspond to low values $k$.
\[lem:NR0\] The expected number $N_{R}^{0}$ of zero measurements is $$\label{eq:NR0}
N_{R}^{0} = N_{R}^{0}(k) = {\mathbb{E}}[X] = \sum_{r \in R} p_{r}^{k}.$$
For $k=1$, $X_{r}$ has a Bernoulli distribution with $$\label{eq:def-pr}
{\mathbb{E}}[X_{r}] = \Pr[X_{r}=1] = 1 - \frac{|r|}{|C|}
= 1-q_{r} = p_{r}.$$ For $k$ independent trials, we have (cf. Remark \[rem:ass-replace\]) $$\label{eq:exp-Xr}
{\mathbb{E}}[X_{r}] = \Pr[X_{r}=1] = p_{r}^{k}.$$ By the linearity of expectations and , we obtain , $$\label{eq:exp-p0}
N_{R}^{0} = {\mathbb{E}}[X]
= \sum_{r \in R} {\mathbb{E}}[X_{r}] =
\sum_{r \in R} p_{r}^{k}.$$
We discuss few specific scenarios depending on the sparsity parameter $k$.
No particles
: For $k=0$, we obviously have $$N_{R}^{0} = \sum_{r \in R} 1 = |R|.$$
High sparsity
: By , we have ${\underline{q}}_{d} \leq q_{r} \leq {\overline{q}}_{d}$, hence $q_{r} = {\mathcal{O}}(d^{-1})$. Thus, for large problem sizes $d$ and small values of $k$, $$N_{R}^{0} \approx \sum_{r \in R} \Big(
\binom{k}{0} 1^{k} q_{r}^{0} -
\binom{k}{1} 1^{k-1} q_{r}^{1} \Big)
= \sum_{r \in R} (1-k q_{r}).$$ By , $\sum_{r \in R} q_{r} = 3$, hence $$\label{eq:NR0-linear-lower-bound}
N_{R}^{0} \approx |R| - 3 k.$$ This approximation says that for sufficiently small values of $k$ each randomly selected cell can be expected to create 3 independent measurements, which just reflects the fact that each cell is met by three projection rays.
Less high sparsity
: For increasing values of $k$ higher-order terms cannot longer be ignored, due to the increasing number of projection rays meeting *several* particles. Taking the second-order term into account, we obtain in an analogous way $$\label{eq:NR0-quadratic-upper-bound}
\begin{aligned}
N_{R}^{0} &\approx
\sum_{r \in R} (1-k q_{r}+\frac{k (k-1)}{2} q_{r}^{2}) \\
&\leq \sum_{r \in R} (1-k q_{r}+\frac{k (k-1)}{2} q_{r} {\overline{q}}_{d})
= |R| - 3 k + \frac{3}{2} k (k-1) {\overline{q}}_{d},
\end{aligned}$$ which is a fairly tight upper bound for values of $k$ and $N$ that are relevant to applications.
We consider next the *expected* number of cells $n_{red} = |C_{b}|$ of cells supporting the set $R_{b}$ according to . We denote this expected number by $$N_{C} := {\mathbb{E}}[|C_{b}|],\qquad N_{C}^{0} := |C|-N_{C},$$ and by $N_{C}^{0}$ the expected size of the complement.
Let $R = R_{1} \cup R_{2} \cup R_{3}$ denote the partition of all projection rays by the three directions. For each cell $c$, there are three unique rays $r_{i}(c) \in R_{i},\, i=1,2,3$, incident with $c$. Furthermore, for $i \neq j$ and some ray $r_{i} \in R_{i}$, let $R_{j}(r_{i})$ denote the set of rays that intersect with $r_{i}$. As before, $|r|$ denotes the number of cells covered by projection ray $r \in R$.
\[prop:NC0\] For a given sparsity parameter $k$, the expected number of cells that can be recognized as empty based on the observations of random variables $\{X_{r}\}_{r \in R}$, is
\[eq:NC\] $$\begin{aligned}
N_{C}^{0} &= N_{C}^{0}(k)
= 3 N_{C}^{1} - 3 N_{C}^{2} + N_{C}^{3}, \label{eq:NC0} \\
N_{C}^{1} &= \sum_{r \in R_{i}} |r| \left(1 -
\frac{|r|}{|C|} \right)^{k},\qquad
\text{for any} \quad i \in \{1,2,3\},
\label{eq:NC0-N1} \\
N_{C}^{2} &= \sum_{r_{i} \in R_{i}} \sum_{r_{j} \in R_{j}(r_{i})}
\left( 1 - \frac{|r_{i}| + |r_{j}|-1}{|C|} \right)^{k},
\; \text{for any}\; i,j \in \{1,2,3\},\, i \neq j,
\label{eq:NC0-N2} \\ \label{eq:NC0-N3}
N_{C}^{3} &= \sum_{c \in C} \left( 1 - \frac{\sum_{i=1}^{3}
|r_{i}(c)|-2}{|C|} \right)^{k}.\end{aligned}$$
Each cell intersects with three projection rays $r_{i}(c),\,i=1,2,3$. Hence, given the rays corresponding to zero measurements, each cell that can be recognized as empty if either one, two or three rays from the set $\{r_{i}(c)\}_{i=1,2,3}$ belong to this set.
We therefore determine separately the expected number of removable cells (i) due to individual rays corresponding to zero measurements, (ii) due to all pairs of rays that intersect and correspond to zero measurements, and (iii) due to all triples of rays that intersect and correspond to zero measurements. The estimate then results from the inclusion-exclusion principle that combines these numbers so as to avoid overcounting, to obtain the desired estimate corresponding to the union of these events.
Consider each projection ray $r \in R_{i}$ for any fixed direction $i=1,2,3$. Because these rays do not intersect, the expected number of cells that can be removed based on the observation $\{X_{r}\}_{r \in R}$, is $$N_{C}^{1} = {\mathbb{E}}\Big[\sum_{r \in R_{i}} X_{r} |r| \Big]
= \sum_{r \in R_{i}} p_{r}^{k} |r|,$$ by the linearity of expectations and . Due to the symmetry of the setup, this number is the same for each direction $i=1,2,3$. Hence we multiply $N_{C}^{1}$ by $3$ in .
Consider next pairs of directions $i,j \in \{1,2,3\},\, i \neq j$. For $i$ fixed, the expected number of empty cells based on a zero measurement corresponding to some ray $r_{i} \in R_{i}$ *and all* rays $r_{j} \in R_{j}(r_{i})$ intersecting with $r_{i}$, is $$N_{C}^{2}
= {\mathbb{E}}\Big[ \sum_{r_{i} \in R_{i}} \sum_{r_{j} \in R_{j}(r_{i})}
X_{r_{i}} X_{r_{j}} \Big].$$ The linearity of expectations and ${\mathbb{E}}[X_{r_{i}} X_{r_{j}}] = \Pr\big[(X_{r_{i}}=1) \wedge (X_{r_{j}}=1)\big]$ gives . Due to symmetry, we have to multiply $N_{C}^{2}$ by $3$ in .
Finally, the expected number of empty cells that correspond to observed zero measurements along *all three* projection directions, is $$N_{C}^{3} = {\mathbb{E}}\Big[\sum_{c \in C}
\prod_{i=1}^{3} X_{r_{i}(c)} \Big],$$ which equals .
An immediate consequence of Lemma \[lem:NR0\] and Prop. \[prop:NC0\] is
\[cor:Ared\] For a given value of the sparsity parameter $k$, the expected dimensions of the reduced system are $$\label{eq:Ared}
m_{red}=N_{R}-N_{R}^{0},\quad
n_{red}=N_{C}-N_{C}^{0},$$ with $N_{R}^{0}, N_{C}^{0}$ given by and .
A Tail Bound {#sec:hexagon-tailbound}
------------
We are interested in how sharply the random number $X$ of zero measurements peaks around its expected value $N_{R}^{0} = {\mathbb{E}}[X]$ given by .
Because the random variables $X_{r},\,r \in R$, are *not* independent due to the intersection of projection rays, we apply the following classical inequality for bounding the deviation of a random variable from its expected value based on martingales, that is on sequences of random variables $(X_{i})$ defined on a finite probability space $(\Omega, {\mathcal{F}}, \mu)$ satisfying $$\label{eq:condition-martingale}
{\mathbb{E}}[X_{i+1}|{\mathcal{F}}_{i}] = X_{i},\qquad
\text{for all}\quad i \geq 1,$$ where ${\mathcal{F}}_{i}$ denotes an increasing sequence of $\sigma$-fields in ${\mathcal{F}}$ with $X_{i}$ being ${\mathcal{F}}_{i}$-measurable.
\[thm:Azuma\] Let $(X_{i})_{i=0,1,2,\dotsc}$ be a sequence of random variables such that for each $i$, $$\label{eq:def-ci}
|X_{i}-X_{i-1}| \leq c_{i}.$$ Then, for all $j \geq 0$ and any $\delta > 0$, $$\Pr\big(|X_{j}-X_{0}| \geq \delta\big) \leq
2 \exp\Big( -\frac{\delta^{2}}{2 \sum_{i=1}^{j} c_{i}^{2}}\Big).$$
Let ${\mathcal{F}}_{i} \subset 2^{R},\, i=0,1,2,\dotsc$, denote the $\sigma$-field generated by the collection of subsets of $R$ that correspond to all possible events after having observed $i$ randomly selected cells. We set ${\mathcal{F}}_{0} = \{\emptyset,R\}$. Because observing cell $i+1$ just further partitions the current state based on the previously observed $i$ cells by possibly removing some ray (or rays) from the set of zero measurements, we have a nested sequence (filtration) ${\mathcal{F}}_{0} \subseteq {\mathcal{F}}_{1} \subseteq \dotsb \subseteq {\mathcal{F}}_{k}$ of the set $2^{R}$ of all subsets of $R$.
Based on this, for a fixed value of the sparsity parameter $k$, we define the sequence of random variables $$Y_{i} = {\mathbb{E}}[X|{\mathcal{F}}_{i}],\quad i=0,1,\dotsc,k,$$ where $Y_{i},\,i=0,1,\dotsc,k-1$, are the random variable specifying the expected number of zero measurements after having observed $k$ randomly selected cells, conditioned on the subset of events ${\mathcal{F}}_{i}$ determined by the observation of $i$ randomly selected cells. Consequently, $Y_{0}={\mathbb{E}}[X]=N_{R}^{0}$ due to the absence of any information, and $Y_{k} = X$ is just the observed number of zero measurements. The sequence $(Y_{i})_{i=0,\dotsc,k}$ is a martingale by construction satisfying ${\mathbb{E}}[Y_{i+1}|{\mathcal{F}}_{i}]=Y_{i}$, that is condition .
\[prop:NR0-deviation\] Let $N_{R}^{0}={\mathbb{E}}[X]$ be the expected number of zero measurements for a given sparsity parameter $k$, given by . Then, for any $\delta > 0$, $$\label{eq:NR0-deviation}
\Pr\big[|X-N_{R}^{0}| \geq \delta\big] \;\leq\;
2 \exp\bigg(
-\frac{1-{\overline{p}}_{d}^{2}}{18 (1-{\overline{p}}_{d}^{2k})}
\;\delta^{2}
\bigg).$$
Let $R^{0}_{i-1} \subset R$ denote the subset of rays with zero measurements after the random selection of $i-1 < k$ cells. For the remaining $k-(i-1)$ trials, the probability that not any cell incident with some ray $r \in R^{0}_{i-1}$ will be selected, is $$p_{r}^{k-(i-1)} = {\mathbb{E}}[X_{r}|{\mathcal{F}}_{i-1}],$$ with $p_{r}$ given by . Consequently, by the linearity of expectations, the expectation $Y_{i-1}$ of zero measurements, given the number $|R^{0}_{i-1}|$ of zero measurements after the selection of $i-1$ cells, is $$Y_{i-1} = {\mathbb{E}}[X|{\mathcal{F}}_{i-1}]
= \sum_{r \in R^{0}_{i-1}} p_{r}^{k-(i-1)}.$$ Now suppose we observe the random selection of the $i$-th cell. We distinguish two possible cases.
1. Cell $i$ is not incident with any ray $r \in R^{0}_{i-1}$. Then the number of zero measurements remains the same, and $$Y_{i} = \sum_{r \in R^{0}_{i-1}} p_{r}^{k-i}.$$ Furthermore, $$\label{eq:Azuma-estimate-2}
\begin{aligned}
Y_{i}-Y_{i-1} &= \sum_{r \in R^{0}_{i-1}} \big(
p_{r}^{k-i} - p_{r}^{k-(i-1)} \big)
= \sum_{r \in R^{0}_{i-1}} p_{r}^{k-i} (1-p_{r}) \\
&\leq {\overline{p}}_{d}^{k-i} \sum_{r \in R} q_{r}
= 3 {\overline{p}}_{d}^{k-i}.
\end{aligned}$$
2. Cell $i$ is incident with one, two or three rays contained in $R^{0}_{i-1}$. Let $R^{0}_{i}$ denote the set $R^{0}_{i-1}$ after removing these rays. Then $$Y_{i} = \sum_{r \in R^{0}_{i}} p_{r}^{k-i}.$$ Furthermore, since $R^{0}_{i} \subset R^{0}_{i-1}$ and $|R^{0}_{i-1} \setminus R^{0}_{i}| \leq 3$, $$\begin{aligned}
Y_{i-1} - Y_{i} &=
\sum_{r \in R^{0}_{i-1} \setminus R^{0}_{i}} p_{r}^{k-(i-1)}
- \sum_{r \in R^{0}_{i}} \big(
p_{r}^{k-i} - p_{r}^{k-(i-1)} \big) \\
&\leq 3 {\overline{p}}_{d}^{k-i+1}
- \sum_{r \in R^{0}_{i}}
{\overline{p}}_{d}^{k-i} {\underline{q}}_{d}.\end{aligned}$$ Further upper bounding by dropping the second sum shows that the resulting first term is still smaller than the bound .
As a result, we consider the larger bound of these two cases and compute $$\sum_{i=1}^{k} \big(3 {\overline{p}}_{d}^{k-i}\big)^{2}
= 9 \frac{1-{\overline{p}}_{d}^{2k}}{1-{\overline{p}}_{d}^{2}}.$$ Applying Theorem \[thm:Azuma\] completes the proof.
\[rem:choice-of-k\] Expanding the r.h.s. of around $0$ in terms of the variable $d^{-1}$ shows $$\Pr\big[|X-N_{R}^{0}| \geq \delta\big] \;\leq\;
2 \exp\Big(-\frac{\delta^{2}}{18 k} \Big)
\qquad\text{for}\qquad
d \to \infty.$$ This indicates appropriate choices $k = k(d)$ for large but finite problem sizes $d$ occurring in applications, so as to bound the deviation of $N_{R}^{0}$ from its expected value. As a result, for such choices of $k$, our analysis based on expected values of the key system parameters will hold in applications with high probability.
Critical Sparsity Values and Recovery
-------------------------------------
We derived the expected number $N_R(k)$ of nonzero measurements $m_{red}$ induced by random $k$-sparse vectors $x\in{\mathbb{R}}^n_{k,+}$ and the corresponding expected number $N_R(k)$ of non redundant cells $n_{red}$. The tail bound, Prop. \[prop:NR0-deviation\], guarantees that the dimensions of reduced systems concentrate around the derived expected values, explaining the threshold effects of unique recovery from sparse tomographic measurements.
We now introduce some further notations and discuss the implication of Section \[sec:uniqueness\] on exact recovery of $x\in{{\mathbb R^n}}_{k,+}$. Let $N_R(k)$ and $N_C(k)$ be the expected dimensions of the reduced system induced by a random $k$-sparse nonnegative vector as detailed in Corollary \[cor:Ared\]. Let $\delta=\frac{\sqrt{5}-1}{2}$ and denote by $k_\delta, k_{crit}$ and $k_{1/\delta}$ the on $d$ dependent sparsity values which solve the equations $$\begin{aligned}
N_R(k_\delta) & =\delta \ell N_C(k_\delta), \label{eq:kdelta-criterion}\\
N_R(k_{crit}) & = N_C(k_{crit}), \label{eq:kcrit-criterion}\\
N_R(k_{opt}) & =\frac{1}{2} N_C(k_{opt}), \label{eq:kopt-criterion}\\
N_R(k_{1/\delta})& = \frac{1+\delta}{\ell}N_C(k_{1/\delta}). \label{eq:kinv-criterion}\end{aligned}$$
In what follows, the phrase *with high probability* refers to values of the sparsity parameter $k$ for which random supports $|\operatorname{supp}(b)|$ concentrate around the crucial expected value $N_{R}$ according to Prop. \[prop:NR0-deviation\], thus yielding a desired threshold effect.
\[prop:appl-Wang\] The system $A x = b$, with measurement matrix $A$, admits unique recovery of $k$-sparse non-negative vectors $x$ with high probability, if $$\label{tilde-kdelta-criterion}
k \leq \frac{N_{C}(k_{\delta})}{1+\delta}=:\tilde k_\delta\ .$$
For perturbed systems we have.
\[prop:appl-Hassibi\] The system $\tilde A x = b$, with perturbed measurement matrix $\tilde A$, admits unique recovery of $k$-sparse non-negative vectors $x$ with high probability, if $k$ satisfies condition $k \leq k_{crit}$.
In case Conjecture \[conj:SPCS2\] holds, uniqueness of $x\in{{\mathbb R^n}}_{k,+}$ would be guaranteed if $k\le k_{1/\delta}$. Finally, we comment on the maximal sparsity threshold $k_{opt}$, in case reduced systems would follow a symmetric distribution with respect to the origin and columns would be in general position.
![ Imaging setup with 4 cameras corresponding to the image planes, shown as two pairs in the left and center panel, respectively. Right panel: Cell centers projected onto the first image plane are shown as dots for the case $d=5$. The cube $\Omega = [0,d]^{3}$ is discretized into $d^{3}$ cells and projected along $4 \cdot d (2 d-1)$ rays. []{data-label="fig:4C3D"}](4C3D-a.pdf "fig:"){width="30.00000%"} ![ Imaging setup with 4 cameras corresponding to the image planes, shown as two pairs in the left and center panel, respectively. Right panel: Cell centers projected onto the first image plane are shown as dots for the case $d=5$. The cube $\Omega = [0,d]^{3}$ is discretized into $d^{3}$ cells and projected along $4 \cdot d (2 d-1)$ rays. []{data-label="fig:4C3D"}](4C3D-b.pdf "fig:"){width="30.00000%"} ![ Imaging setup with 4 cameras corresponding to the image planes, shown as two pairs in the left and center panel, respectively. Right panel: Cell centers projected onto the first image plane are shown as dots for the case $d=5$. The cube $\Omega = [0,d]^{3}$ is discretized into $d^{3}$ cells and projected along $4 \cdot d (2 d-1)$ rays. []{data-label="fig:4C3D"}](4C3D-c.pdf "fig:"){width="25.00000%"}
4 Cameras - Left Degree equals 4 {#sec:4C3D}
================================
We consider the imaging set-up depicted by Figure \[fig:4C3D\] and conduct a probabilistic analysis of its recovery properties, analogous to \[sec:hexagon\_red\_dim\]. This scenario is straightforward to realize and should also be particularly relevant to practical applications.
Imaging Geometry {#imaging-geometry}
----------------
Each coordinate of the unit cube $\Omega = [0,d]^{3}$ is discretized into the intervals $\{0,1,2,\dotsc,d\}$, resulting in $d^{3}$ voxels with coordinates $$\label{eq:def-C-cells}
C = \big\{c = (i,j,l)-\frac{1}{2}(1,1,1) \colon i,j,l \in [d]\big\}.$$ There are 4 sets of parallel projection rays corresponding to the normals of the image planes depicted in Fig. \[fig:4C3D\], $$\label{eq:4-normals}
n^{1} = \frac{1}{\sqrt{2}} (-1,0,1),\quad
n^{2} = \frac{1}{\sqrt{2}} (1,0,1),\quad
n^{3} = \frac{1}{\sqrt{2}} (0,-1,1),\quad
n^{4} = \frac{1}{\sqrt{2}} (0,1,1).$$ We denote the set of projection rays and its partition corresponding to the 4 directions $$\label{eq:R-1-4}
R = \cup_{l=1}^{4} R_{l}.$$ Each set $R_{i}$ contains $(2 d-1) \cdot d$ projection rays whose measurements yields a projection image with $(2 d-1) \times d$ pixels. We index and denote the pixels by $(s,t)$, and the projection rays through these pixels by $$r^{i}_{s,t} \in R_{i},\quad i\in\{1,2,3,4\}.$$ For each cell $c \in C$ indexed by $i,j,l \in [d]$ according to , we represent the corresponding pixels after a suitable transformation by
\[eq:voxel-projections\] $$\begin{aligned}
(s_{1},t_{1}) &= (i+l-1-d,j), \qquad&\qquad
s_{1} &\in [1-d,d-1],\; t_{1} \in [d], \\
(s_{2},t_{2}) &= (i-l,j), \qquad&\qquad
s_{2} &\in [1-d,d-1],\; t_{2} \in [d], \\
(s_{3},t_{3}) &= (i,j+l-1-d), \qquad&\qquad
s_{3} &\in [d],\; t_{3} \in [1-d,d-1], \\
(s_{4},t_{4}) &= (i,j-l), \qquad&\qquad
s_{4} &\in [d],\; t_{4} \in [1-d,d-1].
\end{aligned}$$
The cardinalities of the projection rays, i.e. the number of cells covered by each projection ray, are
\[eq:r1234-cells\] $$\begin{aligned}
a \in \{1,2\}\colon\qquad
|r^{a}_{s,t}| &= d-|s|,
\qquad s \in [1-d,d-1],\quad
t \in [d],
\label{eq:r12-cells} \\ \label{eq:r34-cells}
b \in \{3,4\}\colon\qquad
|r^{b}_{s,t}| &= d-|t|,
\qquad s \in [d],\quad
t \in [1-d,d-1].\end{aligned}$$
We observe the symmetries $$\label{eq:r1234-symmetry}
|r^{a}_{-s,t}| = |r^{a}_{s,t}|,\qquad
|r^{a}_{t,s}| = |r^{b}_{s,t}|$$ and define $$\label{eq:def-rs}
|r_{s}| := |r^{a}_{s,1}|$$ because $|r^{a}_{s,t}|$ does not vary with $t$. Summing up the cells covered by all rays along the first direction, for example, we obtain by , and , $$\begin{aligned}
\sum_{r^{1} \in R_{1}} |r^{1}|
&= \sum_{t \in [d]} \sum_{s=1-d}^{d-1} |r^{1}_{s,t}| =
d \sum_{s=1-d}^{d-1}|r_{s}| =
d \big(d + 2 \sum_{s=1}^{d-1} (d-s)\big) = d^{3} = |C|.\end{aligned}$$ We set $$\label{eq:def-Rk1k2}
\begin{aligned}
R(k_{1},k_{2}) &:=
\Big(1-\frac{|r_{k_{1}}| + |r_{k_{2}}|-1}{d^{3}}\Big)^{k}, \\
R(k_{1},k_{2},k_{3}) &:=
\Big(1-\frac{|r_{k_{1}}| + |r_{k_{2}}| + |r_{k_{3}}|
-2}{d^{3}}\Big)^{k}, \\
R(k_{1},k_{2},k_{3},k_{4}) &:=
\Big(1-\frac{|r_{k_{1}}| + |r_{k_{2}}| + |r_{k_{3}}| + |r_{k_{4}}|
-3}{d^{3}}\Big)^{k}.
\end{aligned}$$ We conduct next for this setup the analysis analogous to Section \[sec:hexagon\_red\_dim\], in order to compute the expected size of the reduced system for random $k$-sparse vectors $x$.
Dimensions of Reduced Systems {#sec:cube_red_dim}
-----------------------------
We first compute the expected number of measurements $m_{red}$ as a function of the sparsity parameter $k$.
\[lem:NR-cube\] The expected number $m_{red} = N_{R}$ of non-zero measurements is
$$\begin{aligned}
N_{R} &= N_{R}(k) = {\mathbb{E}}[|\operatorname{supp}(b)|] = |R|-N_{R}^{0}
= 4 d (2 d-1) - N_{R}^{0}, \\
N_{R}^{0} &= 4 d \bigg( \Big(1-\frac{1}{d^{2}}\Big)^{k} + 2
\sum_{s=1}^{d-1} \Big(1-\frac{s}{d^{3}}\Big)^{k} \bigg).\end{aligned}$$
Taking into account symmetry, we have $$\begin{aligned}
N_{R}^{0} &= {\mathbb{E}}\Big[\sum_{r \in R} X_{r}\Big]
= \sum_{r \in R} p_{r}^{k}
= 4 \sum_{r^{1} \in R_{1}} \Big(1-\frac{|r^{1}|}{|C|}\Big)^{k}.\end{aligned}$$ Applying yields the assertion.
\[prop:NC-cube\] The expected size $n_{red} = N_{C}$ of subset of cells that support random subsets $R_{b} \subset {\mathbb{R}}$ of observed non-zero measurements, is $$N_{C} = N_{C}(k) =
d^{3} - N_{C}^{1} + N_{C}^{2} - N_{C}^{3} + N_{C}^{4}$$ where $$\begin{aligned}
N_{C}^{1} &= 4 d \bigg( d \Big(1-\frac{1}{d^{2}}\Big)^{k} +
2 \sum_{s=1}^{d-1} s \Big(1-\frac{s}{d^{3}}\Big)^{k} \bigg), \\
N_{C}^{2} &= 2 d \sum_{i,l \in [d]} R(l+i-1-d,i-l)
+ 4 \sum_{i,j,l\in [d]} R(l-i,l-j), \\
N_{C}^{3} &= 2 \sum_{i,j,l\in [d]}\big(
R(l+i-1-d,l-i,l-j) + R(l-i,l-j,l+j-1-d) \big), \\
N_{C}^{4} &= \sum_{i,j,l \in [n]} R(l+i-1-d,l-i,l+j-1-d,l-j),
\end{aligned}$$ end the functions $R$ are given by .
We consider for each cell $c \in C$ the quadruple of projection rays $(r^{1}_{C},r^{2}_{C},r^{3}_{C},r^{4}_{C})$ meeting in this cell, and the corresponding partition of projection rays. Cell $c$ is contained in the set $C_{b}$ supporting $R_{b}$ if not any ray of the corresponding quadruple returns a zero measurement. Thus, $$\label{eq:NC-all-proof}
\begin{aligned}
N_{C} &= {\mathbb{E}}\Big[\sum_{c \in C}
(1-X_{r^{1}_{c}}) (1-X_{r^{2}_{c}})
(1-X_{r^{3}_{c}}) (1-X_{r^{4}_{c}})
\Big] \\
&= \sum_{c \in C} \Big(1 - \sum_{i=1}^{4} {\mathbb{E}}[X_{r^{i}_{c}}]
+ \sum_{1 \leq i \leq j \leq 4} {\mathbb{E}}[X_{r^{i}_{c}} X_{r^{j}_{c}}]
- \sum_{1 \leq i \leq j \leq l \leq 4}
{\mathbb{E}}[X_{r^{i}_{c}} X_{r^{j}_{c}} X_{r^{l}_{c}}]
+ {\mathbb{E}}[X_{r^{1}_{c}} X_{r^{2}_{c}} X_{r^{3}_{c}} X_{r^{4}_{c}}]
\Big) \\
&= \sum_{c \in C} \bigg(1
- \sum_{i=1}^{4} \Big(1-\frac{|r^{i}_{c}|}{d^{3}}\Big)^{k}
+ \sum_{1 \leq i \leq j \leq 4}
\Big(1-\frac{|r^{i}_{c} \cup r^{j}_{c}|}{d^{3}}\Big)^{k} \\
&\qquad
- \sum_{1 \leq i \leq j \leq l \leq 4}
\Big(1-\frac{|r^{i}_{c} \cup r^{j}_{c} \cup r^{l}_{c}|}{d^{3}} \Big)^{k}
+ \Big(1-\frac{|\cup_{i=1}^{4} r^{i}_{c}|}{d^{3}} \Big)^{k}
\bigg)
\end{aligned}$$ We consider each term in turn.
(i) As for the first term, we obviously have $|C|=d^{3}$.
(ii) Concerning the second term, taking symmetry into account we compute, $$\begin{aligned}
\sum_{c \in C} \sum_{i=1}^{4}
\Big(1-\frac{|r^{i}_{c}|}{d^{3}}\Big)^{k}
&= 4 \sum_{c \in C} \Big(1-\frac{|r^{1}_{c}|}{d^{3}}\Big)^{k}
= 4 \sum_{r^{1} \in R_{1}} \sum_{c \in r^{1}}
\Big(1-\frac{|r^{1}_{c}|}{d^{3}}\Big)^{k}.\end{aligned}$$ Since $r^{1}_{c} = r^{1}$ for all $c \in r^{1}$, we obtain using , $$\begin{aligned}
\sum_{c \in C} \sum_{i=1}^{4} {\mathbb{E}}[X_{r^{i}_{c}}]
&= 4 \sum_{r^{1} \in R_{1}}
|r^{1}| \Big(1-\frac{|r^{1}|}{d^{3}}\Big)^{k}
= 4 d \sum_{s_{1} = 1-d}^{d-1}
|r^{1}_{s_{1},t_{1}}|
\Big(1-\frac{|r^{1}_{s_{1},t_{1}}|}{d^{3}}\Big)^{k} \\
&= 4 d \bigg( d \Big(1-\frac{1}{d^{2}}\Big)^{k} +
2 \sum_{s=1}^{d-1} s \Big(1-\frac{s}{d^{3}}\Big)^{k} \bigg).\end{aligned}$$
(iii) Concerning the third term of , we consider first the contribution of the pair of directions $(i,j) = (1,2)$. Replacing cell-indices of projection rays by pixel indices according to , we have using and , $$\label{eq:NC-r12-term}
\begin{aligned}
\sum_{c \in C}
\Big(1-\frac{|r^{1}_{c} \cup r^{2}_{c}|}{d^{3}}\Big)^{k}
&= \sum_{i,j,l \in [d]}
\Big(1-\frac{|r^{1}_{(i+l-1-d,j)}| + |r^{2}_{(i-l,j)}|
- 1}{d^{3}}\Big)^{k} \\
&= d \sum_{i,l \in [d]}
\Big(1-\frac{|r_{i+l-1-d}| + |r_{i-l}|
- 1}{d^{3}}\Big)^{k} \\
&= d \sum_{i,l \in [d]} R(i+l-1-d,i-l),
\end{aligned}$$ where the factor $d$ appears because the summand does not depend on $j$ by , and $R$ is defined by . For the pair of directions $(3,4)$, we get $$\label{eq:NC-r34-term}
\begin{aligned}
\sum_{c \in C}
\Big(1-\frac{|r^{3}_{c} \cup r^{4}_{c}|}{d^{3}}\Big)^{k}
&= \sum_{i,j,l \in [d]}
\Big(1-\frac{|r^{3}_{(i,j+l-1-d)}| + |r^{4}_{(i,j-l)}|
- 1}{d^{3}}\Big)^{k},
\end{aligned}$$ which equals due to the symmetry .
Next, we consider the pair of directions $(1,3)$. Taking into account the symmetry and using , we obtain $$\begin{aligned}
&\sum_{c \in C}
\Big(1-\frac{|r^{1}_{c} \cup r^{3}_{c}|}{d^{3}}\Big)^{k}
= \sum_{i,j,l \in [d]}
\Big(1-\frac{|r^{1}_{(i+l-1-d,j)}| + |r^{3}_{(i,j+l-1-d)}|
- 1}{d^{3}}\Big)^{k} \\
&\quad= \sum_{i,j,l \in [d]}
\Big(1-\frac{|r_{i+l-1-d}| + |r_{j+l-1-d}|
- 1}{d^{3}}\Big)^{k}
= \sum_{i,j,l \in [d]}
\Big(1-\frac{|r_{l-i}| + |r_{l-j}| - 1}{d^{3}}\Big)^{k}
\end{aligned}$$ In the same way it can be shown that the remaining pairs of directions $(1,4), (2,3), (2,4)$ each contributes the last expression.
(iv) Concerning the fourth term of (4.10), we get for the triple of directions $(1,2,3)$ the contribution $$\begin{aligned}
\sum_{c \in C} \Big(1-\frac{|r^{1}_{c} \cup r^{2}_{c} \cup r^{3}_{c}|}{d^{3}} \Big)^{k}
&= \sum_{i,j,l \in [n]}
\Big(1-\frac{|r_{i+l-1-d}| + |r_{i-l}|
+ |r_{j+l-1-d}| - 2}{d^{3}}\Big)^{k} \\
&= \sum_{i,j,l \in [n]}
R(i+l-1-d,l-i,j+l-1-d)\big),
\end{aligned}$$ and likewise for the remaining triples $$\begin{aligned}
(1,2,4)\colon\qquad &
\sum_{i,j,l \in [n]} R(i+l-1-d,l-i,l-j), \\
(1,3,4)\colon\qquad &
\sum_{i,j,l \in [n]} R(i+l-1-d,j+l-1-d,l-j), \\
(2,3,4)\colon\qquad &
\sum_{i,j,l \in [n]} R(l-i,j+l-1-d,l-j).
\end{aligned}$$ Evidently, the first and last pair of expressions are equal, respectively.
(v) Finally, the expression for the last term of is immediate.
We conclude this section by stressing that critical sparsity values $k_\delta$ , $k_{crit}$ , $k_{opt}$ , $k_{1/\delta})$ , can be worked out based on the just derived values $N_R$ and $N_C$. A tail bound may be derived analogously to Prop. 3.5. We omit this redundant detail due to space constraints.
-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
![ Success and failure empirical phase transitions for the 2D, 3-cameras case, hexagonal area, from Section \[sec:Hex3C\], Fig. \[fig:3\_cam\], left. Reduced unperturbed (top left) and perturbed (top right) matrices are overdetermined and of full rank with hight probability if the corresponding sparsity level is below $k_{\delta}$ (unperturbed case) or $k_{crit}$ (perturbed case). The $\triangleleft$-marked curve depicts $\tilde k_{\delta}/d^2$ , and the $\square$-marked curve, $k_{crit}/d^2$ from . Probability of uniqueness in $[0,1]^n$ of a $k=\rho d^2$ sparse binary vector for unperturbed (bottom left) and perturbed matrices (bottom right). This probability is high below $\tilde k_{\delta}/d^2$ and decreases slowly, for the unperturbed case (bottom left). In the perturbed case the empirical probability of uniqueness exhibits a sharp transition accurately described by the $\triangleright$-marked curve $k_{1/\delta}/d^2$ from , which for the 3 camera case lies under the $\circ$-marked curve $k_{opt}$ from . []{data-label="fig:Prob_Hex_3C"}](Gray_Final_RankA0Hex3C.png "fig:"){width="45.00000%"} ![ Success and failure empirical phase transitions for the 2D, 3-cameras case, hexagonal area, from Section \[sec:Hex3C\], Fig. \[fig:3\_cam\], left. Reduced unperturbed (top left) and perturbed (top right) matrices are overdetermined and of full rank with hight probability if the corresponding sparsity level is below $k_{\delta}$ (unperturbed case) or $k_{crit}$ (perturbed case). The $\triangleleft$-marked curve depicts $\tilde k_{\delta}/d^2$ , and the $\square$-marked curve, $k_{crit}/d^2$ from . Probability of uniqueness in $[0,1]^n$ of a $k=\rho d^2$ sparse binary vector for unperturbed (bottom left) and perturbed matrices (bottom right). This probability is high below $\tilde k_{\delta}/d^2$ and decreases slowly, for the unperturbed case (bottom left). In the perturbed case the empirical probability of uniqueness exhibits a sharp transition accurately described by the $\triangleright$-marked curve $k_{1/\delta}/d^2$ from , which for the 3 camera case lies under the $\circ$-marked curve $k_{opt}$ from . []{data-label="fig:Prob_Hex_3C"}](Gray_Final_RankA2Hex3C.png "fig:"){width="45.00000%"}
![ Success and failure empirical phase transitions for the 2D, 3-cameras case, hexagonal area, from Section \[sec:Hex3C\], Fig. \[fig:3\_cam\], left. Reduced unperturbed (top left) and perturbed (top right) matrices are overdetermined and of full rank with hight probability if the corresponding sparsity level is below $k_{\delta}$ (unperturbed case) or $k_{crit}$ (perturbed case). The $\triangleleft$-marked curve depicts $\tilde k_{\delta}/d^2$ , and the $\square$-marked curve, $k_{crit}/d^2$ from . Probability of uniqueness in $[0,1]^n$ of a $k=\rho d^2$ sparse binary vector for unperturbed (bottom left) and perturbed matrices (bottom right). This probability is high below $\tilde k_{\delta}/d^2$ and decreases slowly, for the unperturbed case (bottom left). In the perturbed case the empirical probability of uniqueness exhibits a sharp transition accurately described by the $\triangleright$-marked curve $k_{1/\delta}/d^2$ from , which for the 3 camera case lies under the $\circ$-marked curve $k_{opt}$ from . []{data-label="fig:Prob_Hex_3C"}](Gray_Final_ProbLPbinA0Hex3C.png "fig:"){width="45.00000%"} ![ Success and failure empirical phase transitions for the 2D, 3-cameras case, hexagonal area, from Section \[sec:Hex3C\], Fig. \[fig:3\_cam\], left. Reduced unperturbed (top left) and perturbed (top right) matrices are overdetermined and of full rank with hight probability if the corresponding sparsity level is below $k_{\delta}$ (unperturbed case) or $k_{crit}$ (perturbed case). The $\triangleleft$-marked curve depicts $\tilde k_{\delta}/d^2$ , and the $\square$-marked curve, $k_{crit}/d^2$ from . Probability of uniqueness in $[0,1]^n$ of a $k=\rho d^2$ sparse binary vector for unperturbed (bottom left) and perturbed matrices (bottom right). This probability is high below $\tilde k_{\delta}/d^2$ and decreases slowly, for the unperturbed case (bottom left). In the perturbed case the empirical probability of uniqueness exhibits a sharp transition accurately described by the $\triangleright$-marked curve $k_{1/\delta}/d^2$ from , which for the 3 camera case lies under the $\circ$-marked curve $k_{opt}$ from . []{data-label="fig:Prob_Hex_3C"}](Gray_Final_ProbLPbinA2Hex3C.png "fig:"){width="45.00000%"}
-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
![Left: The analytically derived curves from Section \[sec:Hex3C\] correctly follow the contour lines of the average fraction of reduced systems determined empirically as a function of $d$ and relative sparsity $k/d$. From bottom to top: $\tilde k_{\delta}/d$ ($\triangleleft$-marked curve), $k_{crit}/d$ (${\square}$-marked curve), $k_{opt}/d$ ($\circ$-marked curve) and $k_{1/\delta}/d$ ($\triangleleft$-marked curve). Right: These curves are plotted again (light gray) for a wider range and compared to the analogous curves for the geometry in Fig. \[fig:3\_cam\], right, square area, 2 orthogonal, one diagonal projecting direction.[]{data-label="fig:HexvsSquare"}](Gray_Final_FracHex3C.png "fig:"){width="45.00000%"} ![Left: The analytically derived curves from Section \[sec:Hex3C\] correctly follow the contour lines of the average fraction of reduced systems determined empirically as a function of $d$ and relative sparsity $k/d$. From bottom to top: $\tilde k_{\delta}/d$ ($\triangleleft$-marked curve), $k_{crit}/d$ (${\square}$-marked curve), $k_{opt}/d$ ($\circ$-marked curve) and $k_{1/\delta}/d$ ($\triangleleft$-marked curve). Right: These curves are plotted again (light gray) for a wider range and compared to the analogous curves for the geometry in Fig. \[fig:3\_cam\], right, square area, 2 orthogonal, one diagonal projecting direction.[]{data-label="fig:HexvsSquare"}](Gray_Final_curvesHex_vs_Square_3C.png "fig:"){width="45.00000%"}
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Numerical Experiments and Discussion {#sec:num}
====================================
In this section we relate the previously derived bounds on the required sparsity that guarantee unique nonnegative or binary $k$-sparse solutions to numerical experiments. In analogy to [@DonTan05] we assess the so called *phase transition* $\rho$ as a function of $d$, which is reciprocally proportional to the undersampling ratio $\frac{m}{n}\in(0,1)$. We vary $d$, build a specific matrix projection matrix $A$ along with its perturbed version $\tilde A$ and consider the sparsity as a fraction of $d$ in 2D or $d^2$ in 3D, respectively, thus $k=\rho d^{D-1}$ , with $\rho\in (0,4)$ and $D\in\{2,3\}$. This phase transition $\rho(d)$ indicates the necessary relative sparsity to recover a $k$-sparse solution with overwhelming probability. More precisely, if $\|x\|_0\le\rho(d) \cdot d^{D-1}$, then with overwhelming probability a random $k$-sparse nonnegative (or binary) vector $x^*$ is the unique solution in ${{\mathcal F}}_+:=\{x \colon Ax= Ax^*, x\ge 0\}$ or ${{\mathcal F}}_{ 0,1}:=\{x \colon Ax= Ax^*, x\in[0,1]^n\}$, respectively. Uniqueness can be ”verified” by minimizing and maximizing the same objective $f^\top x$ over ${{\mathcal F}}_+$ or ${{\mathcal F}}_{ 0,1}$, respectively. If the minimizers coincide for several random vectors $f$ we claim uniqueness. The resulting linear programs we solved by a standard LP solver [^1]. As shown e.g. in Fig. \[fig:sliceA3D\] and confirmed by all our numerical experiments the threshold for a unique nonnegative solution and a unique $0/1$-bounded solution are quite close, especially for high values of $d$.
We note that $\tilde A$ has the same sparsity structure as $A$, but random entries drawn from the standard uniform distribution on the open interval $(0.9,1.1)$.
Then for $\rho\in[0, 1]$ a $\rho d^{D-1}$-sparse nonnegative or binary vector was generated to compute the right hand side measurement vector and for each $(d,\rho)$-point 50 random problem instances were generated. A threshold-effect is clearly visible in all figures exhibiting parameter regions where the probability of exact reconstruction is close to one and it is much stronger for the perturbed systems. The results are in excellent agreement with the derived analytical thresholds. We refer to the figure captions for detailed explanations and stress that a threshold-effect is clearly visible in all figures exhibiting parameter regions where the probability of exact reconstruction is close to one.
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ -----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
![Empirical relative critical curves in for 3 to 8 cameras in 2D. Top left: $\tilde k_{\delta}/d$ ($\triangleright$-marked curve). Top right: $k_{crit}/d$ (${\square}$-marked curve). Bottom left: $k_{opt}/d$ ($\circ$-marked curve). Bottom right: $k_{1/\delta}/d$ ($\triangleleft$-marked curve). Perfect recovery is possible for sparsity levels below $k_{1/\delta}$ for perturbed systems.[]{data-label="fig:k_MoreCams"}](Gray_AllCams_tilde_k_delta_Square.png "fig:"){width="45.00000%"} ![Empirical relative critical curves in for 3 to 8 cameras in 2D. Top left: $\tilde k_{\delta}/d$ ($\triangleright$-marked curve). Top right: $k_{crit}/d$ (${\square}$-marked curve). Bottom left: $k_{opt}/d$ ($\circ$-marked curve). Bottom right: $k_{1/\delta}/d$ ($\triangleleft$-marked curve). Perfect recovery is possible for sparsity levels below $k_{1/\delta}$ for perturbed systems.[]{data-label="fig:k_MoreCams"}](Gray_AllCams_k_crit_Square.png "fig:"){width="45.00000%"}
![Empirical relative critical curves in for 3 to 8 cameras in 2D. Top left: $\tilde k_{\delta}/d$ ($\triangleright$-marked curve). Top right: $k_{crit}/d$ (${\square}$-marked curve). Bottom left: $k_{opt}/d$ ($\circ$-marked curve). Bottom right: $k_{1/\delta}/d$ ($\triangleleft$-marked curve). Perfect recovery is possible for sparsity levels below $k_{1/\delta}$ for perturbed systems.[]{data-label="fig:k_MoreCams"}](Gray_AllCams_k_opt_Square.png "fig:"){width="45.00000%"} ![Empirical relative critical curves in for 3 to 8 cameras in 2D. Top left: $\tilde k_{\delta}/d$ ($\triangleright$-marked curve). Top right: $k_{crit}/d$ (${\square}$-marked curve). Bottom left: $k_{opt}/d$ ($\circ$-marked curve). Bottom right: $k_{1/\delta}/d$ ($\triangleleft$-marked curve). Perfect recovery is possible for sparsity levels below $k_{1/\delta}$ for perturbed systems.[]{data-label="fig:k_MoreCams"}](Gray_AllCams_k_inv_Square.png "fig:"){width="45.00000%"}
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ -----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
![Left: $k_{opt}/d$ ($\circ$-marked curve) along with $k_{1/\delta}/d$ ($\triangleleft$-marked curve). For 3 cameras $k_{1/\delta}/d$ lies below $k_{opt}/d$, but starting with 5 cameras $k_{1/\delta}$ significantly outperforms $k_{opt}$. This shows the fundamental difference between the considered 0/1-matrices and random matrices underlying a symmetrical distribution with respect to the origin. For random matrices recovery of $k$-sparse positive or binary vectors with sparsity levels beyond $k_{opt}$ would be *impossible*. Right: For $d=200$ the 6 curves depict the average ratio of $m_{red}(k)/n_{red}(k)$ as a function of sparsity $k$ for 3 to 8 cameras from bottom to top.[]{data-label="fig:AvrFracMoreCams"}](Gray_AllCams_k_opt_k_inv_Square.png "fig:"){width="45.00000%"} ![Left: $k_{opt}/d$ ($\circ$-marked curve) along with $k_{1/\delta}/d$ ($\triangleleft$-marked curve). For 3 cameras $k_{1/\delta}/d$ lies below $k_{opt}/d$, but starting with 5 cameras $k_{1/\delta}$ significantly outperforms $k_{opt}$. This shows the fundamental difference between the considered 0/1-matrices and random matrices underlying a symmetrical distribution with respect to the origin. For random matrices recovery of $k$-sparse positive or binary vectors with sparsity levels beyond $k_{opt}$ would be *impossible*. Right: For $d=200$ the 6 curves depict the average ratio of $m_{red}(k)/n_{red}(k)$ as a function of sparsity $k$ for 3 to 8 cameras from bottom to top.[]{data-label="fig:AvrFracMoreCams"}](Gray_AllCams_Frac_Square.png "fig:"){width="45.00000%"}
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
![Success and failure empirical phase transitions for the 2D, 6 cameras case. Probability of uniqueness in $[0,1]^n$ of a $k=\rho d^2$ sparse binary vector for unperturbed (left) and perturbed matrices (right), along with $\tilde k_{\delta}/d^2$ ($\triangleright$-marked curve), $k_{crit}/d^2$ (${\square}$-marked curve), $k_{opt}/d^2$ ($\circ$-marked curve), and $k_{1/\delta}/d^2$ ($\triangleleft$-marked curve) from bottom to top. The dashed line depicts $k/d^2$, with $k$ solving $m_{red}(k)=\frac{2}{\ell} n_{red}(k)$, which accurately follows the border of the highly success area *for all* considered number of cameras, $3, 4 \dots 8$. Recovery is possible *beyond* $k_{opt}$, accurately described by $k_{1/\delta}$. In the 6 cameras case there is no evident performance boost for perturbed systems, since the columns of reduced systems are most likely to be in general position for both perturbed and unperturbed systems. However, in the perturbed case, recovery is more stable.[]{data-label="fig:Square_6C"}](Gray_Square_6C_Unpert_probLPbinA0.png "fig:"){width="45.00000%"} ![Success and failure empirical phase transitions for the 2D, 6 cameras case. Probability of uniqueness in $[0,1]^n$ of a $k=\rho d^2$ sparse binary vector for unperturbed (left) and perturbed matrices (right), along with $\tilde k_{\delta}/d^2$ ($\triangleright$-marked curve), $k_{crit}/d^2$ (${\square}$-marked curve), $k_{opt}/d^2$ ($\circ$-marked curve), and $k_{1/\delta}/d^2$ ($\triangleleft$-marked curve) from bottom to top. The dashed line depicts $k/d^2$, with $k$ solving $m_{red}(k)=\frac{2}{\ell} n_{red}(k)$, which accurately follows the border of the highly success area *for all* considered number of cameras, $3, 4 \dots 8$. Recovery is possible *beyond* $k_{opt}$, accurately described by $k_{1/\delta}$. In the 6 cameras case there is no evident performance boost for perturbed systems, since the columns of reduced systems are most likely to be in general position for both perturbed and unperturbed systems. However, in the perturbed case, recovery is more stable.[]{data-label="fig:Square_6C"}](Gray_Square_6C_Pert_probLPbinA2.png "fig:"){width="45.00000%"}
---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
2D: Hexagonal Volume and 3 Cameras
----------------------------------
In this 2D, 3-cameras case, we generated $A$ according to the geometry described in Section \[sec:Hex3C\], Fig. \[fig:3\_cam\], left. We considered $d\in\{51,71,91,\dots ,251\}$ and varied the sparsity $k=\rho d$, by varying $\rho$ in $(0,0.5)$ with constant stepsize $0.01$. The obtained empirical phase transitions are depicted in \[fig:Prob\_Hex\_3C\]. Fig. \[fig:AvrFracMoreCams\] additionally shows how the analytically determined critical curves compare to the critical curves obtained empirically for the geometry in Fig. \[fig:3\_cam\], right. The $(4d-1)\times d^2$ projection matrix corresponding to a square area and three projecting directions has similar reconstruction properties and the critical curves slightly change by factor $\frac{10}{9}$.
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
![ Recovery via the unperturbed matrix from Section \[sec:4C3D\], Fig. \[fig:4C3D\] (dark gray $\square$-marked curves), for $d=30$ (left) and $d=40$ (right) versus the perturbed counterpart (light gray $\circ$-marked curves). The dashed lines depict the empirical probability (500 trials) that reduced systems are overdetermined and of full rank. The solid lines show the probability that a $k$ sparse nonnegative vector is unique. The dash-dot curves shows the probability that a $k$ sparse binary solution is the unique solution of in $[0,1]^n$. Additional information like binarity gives only a slight performance boost, as $d$ increases. The curve $\tilde k_{\delta}$ correctly predicts that 500 ($d=30$) and 787 ($d=40$) particles are reconstructed with high probability via the unperturbed systems. Up to the sparsity level 1136 ($d=30$) and 1714 ($d=40$) perturbed systems are overdetermined and of full rank according to $k_{crit}$ . Moreover, perturbed systems have a unique $k$ sparse solution if $k$ lies in between $k_{opt}$ - 2028 ($d=30$) and 2856 ($d=40$) and $k_{1/\delta}$ - 2136 ($d=30$) and 3128 ($d=40$).[]{data-label="fig:sliceA3D"}](Gray_All_markerA3D4K_d30_with_legend.png "fig:"){width="40.00000%"} ![ Recovery via the unperturbed matrix from Section \[sec:4C3D\], Fig. \[fig:4C3D\] (dark gray $\square$-marked curves), for $d=30$ (left) and $d=40$ (right) versus the perturbed counterpart (light gray $\circ$-marked curves). The dashed lines depict the empirical probability (500 trials) that reduced systems are overdetermined and of full rank. The solid lines show the probability that a $k$ sparse nonnegative vector is unique. The dash-dot curves shows the probability that a $k$ sparse binary solution is the unique solution of in $[0,1]^n$. Additional information like binarity gives only a slight performance boost, as $d$ increases. The curve $\tilde k_{\delta}$ correctly predicts that 500 ($d=30$) and 787 ($d=40$) particles are reconstructed with high probability via the unperturbed systems. Up to the sparsity level 1136 ($d=30$) and 1714 ($d=40$) perturbed systems are overdetermined and of full rank according to $k_{crit}$ . Moreover, perturbed systems have a unique $k$ sparse solution if $k$ lies in between $k_{opt}$ - 2028 ($d=30$) and 2856 ($d=40$) and $k_{1/\delta}$ - 2136 ($d=30$) and 3128 ($d=40$).[]{data-label="fig:sliceA3D"}](Gray_All_markerA3D4K_d40_with_legend.png "fig:"){width="40.00000%"}
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
![ Success and failure empirical phase transitions for the 3D, 4 cameras case, from Section \[sec:4C3D\], Fig. \[fig:4C3D\]. Probability of uniqueness in $[0,1]^n$ of a $k=\rho d^2$ sparse binary vector for unperturbed (top left) and perturbed matrices (top right). The $\triangleright$-marked curve depicts $\tilde k_{\delta}/d^2$ , the $\square$-marked curve $k_{crit}/d^2$ , the $\circ$-marked curve $k_{opt}$ and the $\triangleleft$-marked curve $k_{1/\delta}/d^2$ . In case of the perturbed matrix $\tilde A$ exact recovery is possible *beyond* $k_{opt}/d^2$. Moreover $k_{1/\delta}/d^2$ follows most accurately the empirical phase transition for perturbed systems for high values of $d$. The empirical probability that the reduced unperturbed matrices are overdetermined and of full rank (bottom figure), exhibits a threshold in between the estimated relative critical sparsity level $k_{\delta}$ and $k_{crit}$. This explains the performance boost for perturbed systems.[]{data-label="fig:A4C3D"}](Gray_Final_ProbLPbinA0Cube4C.png "fig:"){width="45.00000%"} ![ Success and failure empirical phase transitions for the 3D, 4 cameras case, from Section \[sec:4C3D\], Fig. \[fig:4C3D\]. Probability of uniqueness in $[0,1]^n$ of a $k=\rho d^2$ sparse binary vector for unperturbed (top left) and perturbed matrices (top right). The $\triangleright$-marked curve depicts $\tilde k_{\delta}/d^2$ , the $\square$-marked curve $k_{crit}/d^2$ , the $\circ$-marked curve $k_{opt}$ and the $\triangleleft$-marked curve $k_{1/\delta}/d^2$ . In case of the perturbed matrix $\tilde A$ exact recovery is possible *beyond* $k_{opt}/d^2$. Moreover $k_{1/\delta}/d^2$ follows most accurately the empirical phase transition for perturbed systems for high values of $d$. The empirical probability that the reduced unperturbed matrices are overdetermined and of full rank (bottom figure), exhibits a threshold in between the estimated relative critical sparsity level $k_{\delta}$ and $k_{crit}$. This explains the performance boost for perturbed systems.[]{data-label="fig:A4C3D"}](Gray_Final_ProbLPbinA2Cube4C.png "fig:"){width="45.00000%"}\
![ Success and failure empirical phase transitions for the 3D, 4 cameras case, from Section \[sec:4C3D\], Fig. \[fig:4C3D\]. Probability of uniqueness in $[0,1]^n$ of a $k=\rho d^2$ sparse binary vector for unperturbed (top left) and perturbed matrices (top right). The $\triangleright$-marked curve depicts $\tilde k_{\delta}/d^2$ , the $\square$-marked curve $k_{crit}/d^2$ , the $\circ$-marked curve $k_{opt}$ and the $\triangleleft$-marked curve $k_{1/\delta}/d^2$ . In case of the perturbed matrix $\tilde A$ exact recovery is possible *beyond* $k_{opt}/d^2$. Moreover $k_{1/\delta}/d^2$ follows most accurately the empirical phase transition for perturbed systems for high values of $d$. The empirical probability that the reduced unperturbed matrices are overdetermined and of full rank (bottom figure), exhibits a threshold in between the estimated relative critical sparsity level $k_{\delta}$ and $k_{crit}$. This explains the performance boost for perturbed systems.[]{data-label="fig:A4C3D"}](Gray_Final_ProbRankA0Cube4C.png "fig:"){width="45.00000%"}
2D: Square Volume and 3 to 8 Cameras
------------------------------------
In our theoretical analysis in the previous sections we derived the expected number of nonzero rows $N_R(k)$ induced by the $k$-sparse vector along with the number $N_C(k)$ of ”active” cells which cannot be empty. This can be done also empirically, compare Fig. \[fig:HexvsSquare\], left. We have done this in 2D up to 8 projecting directions and obtained *empirically* the critical curves $k_\delta$, $k_{crit}$, $k_{opt}$ and $k_{1/\delta}$. To generate the curves we varied $k/d\in(0,4)$ by a constant stepsize $0.01$ and $d\in\{50,100, \cdots, 1050\}$ and generated for each point $(k/d,d)$ 500 problem instances. Further we determined the contour lines of $N_R(k/d,d)/N_C(k/d,d)$ corresponding to the levels $\{\delta\ell,1,0.5,\frac{1+\delta}{\ell}\}$. The relative sparsity curves $k_\delta/d$, $k_{crit}/d$, $k_{opt}/d$ and $k_{1/\delta}/d$ are plotted in Fig. \[fig:k\_MoreCams\] and accurately follow the empirical recovery thresholds, as shown e.g. in Fig. \[fig:Square\_6C\] for 6 cameras. Fig. \[fig:AvrFracMoreCams\] shows $N_R(k,200)/N_C(k,200)$ for varying number of cameras. The projection angles we chosen such that the intersection with all cells is constant, yielding binary projection matrices after scaling. Each camera resolution differs with different angle. We summarize the used parameters in Table \[tab:1\].
\# $m$ $n$ projection angles
------------ --------------------------------------- ------- ---------------------------------------------------------------------
3rd camera $4d-1$ $d^2$ $0^\circ, 90^\circ, 45^\circ$
4th camera $6d-2$ $d^2$ $0^\circ, 90^\circ, \mp 45^\circ$
5th camera $7d + \lfloor \frac{d}{2}\rfloor-2$ $d^2$ $0^\circ, 90^\circ, \mp 45^\circ, \arctan(2)$
6th camera $8d + 2\lfloor \frac{d}{2}\rfloor-2$ $d^2$ $0^\circ, 90^\circ, \mp 45^\circ, \mp \arctan(2)$
7th camera $9d + 3\lfloor \frac{d}{2}\rfloor-2$ $d^2$ $0^\circ, 90^\circ, \mp 45^\circ, \mp \arctan(2), \arctan(0.5)$
8th camera $10d + 4\lfloor \frac{d}{2}\rfloor-2$ $d^2$ $0^\circ, 90^\circ, \mp 45^\circ, \mp \arctan(2), \mp \arctan(0.5)$
: Dimensions of full projection matrices.[]{data-label="tab:1"}
3D: 4 Cameras
-------------
In 3D we consider the matrix from Section \[sec:4C3D\], Fig. \[fig:4C3D\], vary $d\in\{25,26,\dots,55\}$ and $k=\rho d^2$ by varying $\rho\in(0,4)$ with stepsize $0.01$. For larger values of $d$ the empirical thresholds follow accurately the estimated curves. Note that for $d=55$, $A$ is a $48180\times 166375$ matrix.
Conclusion
==========
The new measurement paradigm of compressed sensing seeks to capture the ”essential” aspects of a high-dimensional but sparse object using as few measurements as possible by randomization. Provided that the measurements satisfy certain properties nonnegative sparse signals can be reconstructed exactly from a surprisingly small number of samples. Moreover, there exist precise thresholds on sparsity such that for any nonnegative solution that is sparser than the threshold is also the unique nonnegative solution of the underlying linear system. Tomographic projection matrices do not satisfy the conditions which allow applying these results on the image reconstruction problem even if the sought solution is very sparse. However, we analytically showed in the present work that there are thresholds on sparsity depending on the numbers of measurements, below which uniqueness is guaranteed and recovery will succeed and above which it fails with high probability. When recovery succeeds it yields perfect reconstructions, without any ghost-particles.
[10]{}
K. Azuma. eighted sums of certain dependent random variables. , 19(3):357–367, 1967.
R. Berinde and P. Indyk. Sparse recovery using sparse random matrices, 2008. MIT-CSAIL Technical Report.
E. Cand[è]{}s. ompressive sampling. In [*Int. Congress of Math.*]{}, volume 3, Madrid, Spain, 2006.
A. DasGupta. . Springer, 2008.
D. Donoho. ompressed [S]{}ensing. , 52:1289–1306, 2006.
D.L. Donoho and J. Tanner. Sparse nonnegative solution of underdetermined linear equations by linear programming. , 102(27):9446–9451, 2005.
D.L. Donoho and J. Tanner. Counting the faces of randomly-projected hypercubes and orthants, with applications. , 43(3):522–541, 2010.
D.L. Donoho and J. Tanner. Precise undersampling theorems. , 98(6):913–924, 2010.
G. Elsinga, F. Scarano, B. Wieneke, and B. van Oudheusden. Tomographic particle image velocimetry. , 41:933–947, 2007.
M.A. Khajehnejad, A.G. Dimakis, W. Xu, and B. Hassibi. Sparse recovery of positive signals with minimal expansion. , 59:196–208, 2011.
O.L. Mangasarian and B. Recht. Probability of unique integer solution to a system of linear equations. , 214(1):27–30, 2011.
S. Petra and C. Schnörr. Tomo[PIV]{} meets compressed sensing. , 20(1-2):49–76, 2009.
S. Petra and C Schnörr. verage [C]{}ase [R]{}ecovery [A]{}nalysis of [T]{}omographic [C]{}ompressive [S]{}ensing. arXiv:1208.5894v2 \[math.NA\], August 30 2012.
M. Wang, W. Xu, and A. Tang. A unique “nonnegative” solution to an underdetermined system: From vectors to matrices. , 59(3):1007–1016, 2011.
J.G. Wendel. roblem in [G]{}eometric [P]{}robability. , 11:109–111, 1962.
W. Xu and B. Hassibi. . In [*Information Theory Workshop, 2007. ITW ’07. IEEE*]{}, pages 414–419, 2007.
[^1]: MOSEK, <http://www.mosek.com/>
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We present velocities of galactic outflows in seven star-forming galaxies at $z=5\nn6$ with stellar masses of $M_* \sim10^{10.1}$ . Although it is challenging to observationally determine the outflow velocities, we overcome this by using ALMA \[\]158 $\mu$m emission lines for systemic velocities and deep Keck spectra with metal absorption lines for velocity profiles available to date. We construct a composite Keck spectrum of the galaxies at $z=5\nn6$ with the \[\]-systemic velocities, and fit outflow-line profiles to the $\lambda1260$, $\lambda1335$, and $\lambda\lambda1394,1403$ absorption lines in the composite spectrum. We measure the maximum (90%) and central outflow velocities to be $\vmax=700^{+180}_{-110}$ and $\vout= 400^{+100}_{-150}$ on average, respectively, showing no significant differences between the outflow velocities derived with the low to high-ionization absorption lines. For $M_* \sim10^{10.1}$ , we find that the value of our $z=5\nn6$ galaxies is 3 times higher than those of $z\sim0$ galaxies and comparable to $z\sim2$ galaxies. Estimating the halo circular velocity from the stellar masses and the abundance matching results, we investigate a – relation. Interestingly, for galaxies with $M_*=10^{10.0\nn10.8}$ shows a clear positive correlation with and/or the galaxy star formation rate over $z=0\nn6$ with a small scatter of $\simeq \pm 0.1$ dex, which is in good agreement with theoretical predictions. This positive correlation suggests that the outflow velocity is physically related to the halo circular velocity, and that the redshift evolution of at fixed $M_*$ is explained by the increase in toward high redshift.'
bibliography:
- '\$HOME/Documents/set\_TeX/reference.bib'
title: 'Fast Outflows Identified in Early Star-Forming Galaxies at $z=5\nn6$'
---
INTRODUCTION {#sec:intro}
============
The energy and momentum inputs from supernovae (SNe) and active galactic nuclei (AGNs) accelerate the inter-stellar medium (ISM) outwards, and launch galactic-scale outflows. The outflows are composed of the various ISM phases from cold molecular to hot gas [e.g., @Veilleux:2005]. AGN-driven outflows are thought to play an important role to quench the star-forming activity in massive galaxies, and recent observational work reports the evidence that the AGN feedback may also operate in low-mass galaxies [@Penny.S:2018a; @Manzano-King.C:2019a]. On the other hand, SN-driven outflows are thought to affect galaxies primarily in low-mass regime. The mass, momentum, energy, and metal budgets of the outflows leaked from the star-forming galaxies are theoretically important for regulating the star-forming activity in low-mass galaxies, creating the mass-metalicity relation of galaxies, and polluting the circum-galactic medium and intergalactic medium (IGM) [for a review, see @Somerville:2015; @Dayal.P:2018a]. Thus, the outflows in star-forming galaxies driven by SNe have a large impact on the galaxy and IGM evolution. Many theoretical studies contribute to revealing the details of the outflow properties using cosmological (zoom-in) simulations [@Oppenheimer.B:2010a; @Muratov:2015; @Christensen:2016a; @Ceverino.D:2016b; @Mitchell.P:2018a; @Nelson.D:2019a].
In the rest-frame far-ultraviolet (FUV; 1000–2000Å) to optical bands, metal absorption and emission lines are useful to trace the kinematics of the cold and warm outflowing gas. The outflow velocity along the line of sight is estimated with the “down-the-barrel” technique, which measures blueshifts of the absorption lines in the galaxy spectra [e.g., @Heckman:2000; @Martin:2005; @Rupke:2005a; @Rupke:2005b; @Steidel:2010; @Martin:2012; @Zhu.G:2015a; @Heckman:2015; @Chisholm:2015a; @Chisholm:2016b; @Chisholm.J:2017a; @Roberts-Borsani.G:2019a; @Concas.A:2019a], while the outflowing gas far from the galaxy is detected with the absorption lines in the background-quasar spectra [e.g., @Bouche:2012; @Kacprzak.G:2015a; @Muzahid:2015b; @Schroetter:2015; @Schroetter:2016a]. Broad components in emission lines also provide a signature of outflows [@Cicone.C:2016a; @Finley.H:2017a; @Concas.A:2017a; @Freeman.K:2017a], which have also recently been observed with integral-field-units spectroscopy [@Davies.R:2019a; @Forster-Schreiber.N:2019a; @Swinbank.M:2019a].
The outflows are ubiquitously observed in the star-forming galaxies at $z<1.5$ [@Weiner:2009; @Chen:2010; @Rubin:2014]. Their outflow velocities are probed to have a positive correlation with the star formation rate (), the stellar mass ($M_*$), and the surface density ($\Sigma_\SFR$) [e.g., @Rubin:2014; @Heckman:2016; @Chisholm:2016a]. @Sugahara.Y:2017a use archival spectra to show that maximum outflow velocity increases from $z\sim0$ to 2 in star-forming galaxies that are in a similar $M_*$ and range.
The “down-the-barrel” technique is also appropriate for outflow studies at $z > 2$. Unlike the emission from the outflows whose detection becomes difficult toward high redshift, the absorption can be detected with a bright background continuum source. @Shapley:2003 construct composites of almost 1000 Lyman-break galaxy (LBG) spectra at $z\sim3$ to discuss the relation between the FUV spectral features and the outflow properties. Recently, @Du.X:2018a report no evolution of central outflow velocities at $z\sim2\nn4$ using composites of the rest-frame FUV spectra presented in @Steidel:2003 [@Steidel:2004], @Reddy.N:2008a, and @Jones:2012a. Although the profile provides us the information on the neutral-gas kinematics around Lyman alpha emitters at high redshift [e.g., @Erb:2014a; @Shibuya:2014; @Hashimoto.T:2015a; @Trainor:2015a; @Karman.W:2017a], even at $z\sim6$ [@Ajiki.M:2002a], it is difficult to directly estimate the outflow properties only from the profile due to its strong resonance scattering.
One of the keys to estimate outflow properties is to determine the systemic redshifts of the galaxies. At the low redshift, the systemic redshifts are measured by nebular emission lines (e.g., , \[\], and \[\]), but observations of the emission lines become expensive at high redshift. Some outflow studies at $z>1.5$ conduct additional near-infrared (IR) observations [@Steidel:2010; @Shibuya:2014], while others determine the redshifts from emission or interstellar absorption, which includes the uncertainties based on the outflows [@Shapley:2003; @Du.X:2018a]. Moreover, precise measurements of the systemic redshifts are challenging at $z>5$, where the strong optical emission lines fall into the mid-IR bands. Although there are several nebular emission lines in the rest-frame FUV band such as \]$\lambda\lambda1660,1666$ and \]$\lambda\lambda1906,1908$, these lines are weak to be detected in typical star-forming galaxies at high redshift. This problem makes it difficult to extend the outflow studies to $z > 5$.
A solution in this paper is observations with the Atacama Large Millimeter/submillimeter Array (ALMA). Recent ALMA observations detect \[\] $158$ and \[\] $88$ emission lines in high-$z$ galaxies [e.g., @Capak:2015; @Inoue.A:2016a; @Hashimoto.T:2019a], which enables us to measure the systemic redshifts of the galaxies. Combining the redshift determined from the ALMA observations with deep observed-frame optical spectra, we can address the outflow properties at $z>5$. As a case study, @Pavesi.R:2016a discuss the rest-frame FUV absorption lines in HZ10, a IR-luminous LBG at $z\simeq 5.6$, and find the blueshifts with respect to the \[\] emission line.
This paper presents estimates of outflow velocities in star-forming galaxies at $z=5\nn6$ and discuss the redshift evolution of the outflows from $z\sim0$ to $6$. Section \[sec:sample\] describes the sample of galaxies at $z=5\nn6$. Section \[sec:analysis\] explains the analysis of the absorption lines in the observed-frame optical spectra. We obtain a composite spectrum of the galaxies to measure an outflow velocity. Section \[sec:results\] shows the results on the outflow velocity and its redshift evolution. Section \[sec:discussion\] discusses relations between the outflow and galaxy properties. Section \[sec:conclusion\] summarizes our conclusion. The $\Lambda$CDM cosmology is used throughout this paper: $\Omega_\mathrm{M} = 0.27$, $\Omega_\mathrm{\Lambda} = 0.73$, $h = H_0/(100\ \mathrm{km\ s^{-1}\ Mpc^{-1}}) = 0.70$, $n_s = 0.95$, and $\sigma_8 = 0.82$. All transitions are referred to by their wavelengths in vacuum.
Sample and Data Reduction {#sec:sample}
=========================
Our sample consists of seven galaxies at $z=5\nn6$ whose spectra are taken in the optical and millimeter wavelengths. We use the galaxies presented in @Capak:2015, who observe nine LBGs and one low-luminosity quasar at $z\sim5$–$6$ in the Cosmic Evolution Survey [COSMOS; @Scoville.N:2007a] field. @Capak:2015 obtain the rest-frame FUV spectra of the galaxies with the DEep Imaging Multi-Object Spectrograph [DEIMOS; @Faber:2003] at the Keck II telescope. The spectroscopic configuration is the 830 grating with the OG550 filter, which gives the wavelength coverage of 6000–9500 Å and the spectral resolution of $R\sim2500\nn3500$. The total integration time is $\sim$3.5 hr for each object.
We download raw DEIMOS data of the galaxies from the Keck Observatory Archive[^1] (KOA). The raw data are reduced with the [IDL]{} package, the DEIMOS [spec2d]{} pipeline, developed by the Deep Extragalactic Evolutionary Probe 2 (DEEP2) Redshift Survey team [@Cooper:2012; @Newman:2013]. From the reduced two-dimensional multi-object-slit data, the pipeline extracts the one-dimensional spectra of the science targets. Finally, we obtain the rest-frame FUV spectra of seven out of the nine LBGs in @Capak:2015, other than two objects (HZ3 and HZ9) whose spectra we could not identify from the archive.
The ALMA follow-up observations are conducted in a project of \#2012.1.00523.S (PI: P. Capak). The Band 7 observations have detected the \[\] $158$ emission lines in all of the nine LBGs. Previous studies report possible present and past outflow signatures in the \[\] emission lines of these galaxies [@Gallerani.S:2018a; @Fujimoto.S:2019a]. In this study, we use the systemic redshifts measured from the \[\] emission lines by @Capak:2015. In contrast to \[\] $88$ or optical nebular emission lines that come from only regions, the \[\] emission arises from regions and photo-dissociated regions, which may results in some uncertainties in the measured redshifts. However, by detecting both \[\] and \[\] emission lines in objects at $z > 6$ with ALMA, recent studies reveal that redshifts determined by \[\] and \[\] are consistent within the errors [@Marrone.D:2018a; @Decarli.R:2017a; @Walter.F:2018a] or show offsets less than $50$ at most [@Hashimoto.T:2019a]. There is good evidence therefore that measurements of the \[\] emission lines provide reliable systemic redshifts. The median redshift error is $\sim 2\times10^{-4}$, corresponding to $\sim 10$ . The systemic redshifts of our galaxies are listed in Table \[tb:1\].
We use and $M_*$ derived by @Capak:2015. The is estimated from the sum of the rest-frame UV and IR luminosity. The stellar mass $M_*$ is estimated from the spectral energy distribution fitting to the optical to IR photometry taken from the COSMOS photometric redshift catalog [@Ilbert.O:2013a] and the Spitzer-Large Area Survey with Hyper-Suprime-Cam [SPLASH; @Steinhardt.C:2014a]. The halo circular velocity is estimated from $M_*$. First, we convert $M_*$ into the halo mass $M_\mathrm{h}$ with the stellar-to-halo mass ratio (SHMR) given by @Behroozi:2013, who derive the SHMR at $z=0\nn8$ with the abundance matching method, although it should be noted that observational constraints are less complete and potentially less robust at $z > 4$ than at $z \lesssim 2$. Then, are calculated by equations in @Mo:2002 expressed as $$\begin{aligned}
\label{eq:0a}
\vcir &=& \left( \frac{GM_\mathrm{h}}{r_\mathrm{h}} \right)^{1/2}, \\
\label{eq:0b}
r_\mathrm{h} &=& \left( \frac{GM_\mathrm{h}}{100 \Omega_\mathrm{M} H_0^2} \right)^{1/3} (1+z)^{-1} , \end{aligned}$$ where $G$ is the gravitational constant and $r_\mathrm{h}$ the halo radius.
Analysis and Measurements {#sec:analysis}
=========================
Because the outflowing gas gives rise to the blueshifted metal absorption lines due to the Doppler shift, the blueshift reflects the line-of-sight velocity of the outflowing gas. The absorption-line analysis requires high signal-to-noise ratios (S/N) of the continuum spectra. Our rest-frame FUV spectra have the average S/N of $\simeq 0.47$ , which is not enough for the absorption-line analysis. Therefore, we obtain a high-S/N composite spectrum by stacking all of the spectra with an inverse-variance weighted mean. The top panel of Figure \[fig:spec\] shows the composite spectrum and its error spectrum. The continuum S/N of the composite spectrum is $1.4$ around the $\lambda1260$ absorption line. The physical parameters of the composite spectrum are the truncated mean discarding the maximum and minimum values. We note that HZ10 has much higher than other galaxies. By constructing another composite spectrum without HZ10, we check whether this high- galaxy affects our results to confirm that our conclusion does not change.
In the wavelength range from 1150 to 1450 Å in the rest frame, we use the absorption lines of $\lambda1260$, $\lambda1335$, and $\lambda\lambda1394,1403$ for the analysis, without the $\lambda1304$ line that has a nearby strong $\lambda1302$ absorption line. We hereafter refer to $\lambda1260$ as .
We measure outflow velocities by fitting a line profile to the absorption lines. As the line profile, we adopt a physical profile based on the assumption of the curve of growth [@Rupke:2005a]. This line profile $I(\lambda)$, as a function of the wavelength $\lambda$, is expressed by $$\begin{aligned}
\label{eq:1}
I(\lambda)/I_0 &=& 1-C_f+C_f\exp(-\tau(\lambda)), \\
\label{eq:2}
\tau(\lambda) &=& \tau_0\exp(-(v(\lambda)-v_0)^2/b^2),\end{aligned}$$ where $I_0$ is the continuum level, $C_f$ the covering fraction, $\tau(\lambda)$ the optical depth, $\tau_0$ the optical depth at the line center, $v(\lambda)$ the velocity measured from the rest wavelength, $v_0$ the velocity at the line center, and $b$ the Doppler width. The line profile is convolved with a Gaussian profile representing the spectral resolution. The free parameters are five: $I_0$, $v_0$, $C_f$, $\tau_0$, and $b_D$. Because the composite spectrum has large noises, we treat $I_0$ as a free parameter instead of normalizing the spectrum by a stellar continuum. We fit the line profile to , , and , using an [IDL]{} procedure [MPFIT]{}, which performs non-linear least-squares fitting in a robust manner [@Markwardt:2009]. The bottom panel of Figure \[fig:spec\] shows the best-fit model of the , , and absorption lines with the red lines.
The best-fit $v_0$ values, listed in Table \[tb:2\], are all significantly negative, implying that the absorption lines are blueshifted by the outflowing gas. The errors of $v_0$ are evaluated by the parametric bootstrap method. We obtain a $v_0$ distribution by fitting the line profile to 1000 resampled fluxes based on the spectral noise and use the $\pm 34$th $v_0$ values for its error. These velocities are consistent with the values estimated in the literature that analyzes the data of the same galaxies [@Pavesi.R:2016a; @Gallerani.S:2018a]. HZ10 has the , $\lambda1304$/, and absorption lines blueshifted by $100\pm180$ with respect to the \[\] emission line [@Pavesi.R:2016a]. The composite emission of the \[\] line in HZ1–9, without HZ5, is reported to have the broad wings that are likely generated by the outflows with the velocity of $\sigma=220\nn500$ [@Gallerani.S:2018a].
We define the maximum outflow velocity $\vmax$, the highest velocity of the outflowing gas, as $$\label{eq:3}
\vmax = - v_0 + b
\sqrt{-\ln\left(\frac{1}{\tau_0}\ln\frac{1}{0.9}\right)},$$ which represents the velocity where the best-fit model has a 90% flux from the continuum to the bottome of the line.[^2]. The errors of are evaluated by the parametric bootstrap method that is the same as used to evaluate the $v_0$ error.
The derived maximum outflow velocities for , , and are $\vmaxSiII = 690^{+260}_{-120}$ , $\vmaxCII = 720^{+140}_{-460}$ , and $\vmaxSiIV = 610^{+240}_{-96}$ , respectively. Low-ionized elements (and ) have ionization potentials lower than that of hydrogen (13.6 eV), while high-ionized elements () have a much higher ionization potential. Although the low- and high-ionized elements trace the different state of the ISM, $\vmaxSiII$ and $\vmaxCII$ are consistent with $\vmaxSiIV$ within the 1$\sigma$ errors. This consistency agrees with previous work on outflows at $z\sim0$ [@Chisholm:2016a].
and have similar ionization potentials and oscillator strengths, and exhibit similar maximum outflow velocities. To obtain a typical value of the $z=5\nn6$ galaxies, we additionally measure the maximum outflow velocity by a simultaneous fitting to and , adopting as a free parameter instead of $v_0$. Both lines are assumed to have the same $C_f$. The measured value is $\vmax = 700^{+180}_{-110}$ . This value is consistent with $\vmaxSiII$ and $\vmaxCII$, but its error is smaller than those of and . Table \[tb:2\] lists the measurements of and $v_0$ for each absorption lines. These values are consistent with the results of the \[\] emission analysis by @Gallerani.S:2018a. They stack the ALMA data of HZ1–9, without HZ5, to show the highest velocity of $\sim500\nn700$ at which the \[\] flux excess can be observed, although this broad flux excess may include emission from satellites around the central galaxies.
Galaxies at $z\sim0\nn2$
------------------------
To investigate the redshift evolution of the outflow velocity, we mainly compare the measurements at $z=5\nn6$ with the results in @Sugahara.Y:2017a, who study outflows at $z\sim0\nn2$. Here, we briefly describe the sample and analysis in @Sugahara.Y:2017a. They obtain outflow velocities using optical spectra of the Sloan Digital Sky Survey Data Release 7 [SDSS DR7; @Abazajian:2009] at $z\sim0$, the DEEP2 DR4 survey [@Newman:2013] at $z\sim1$, and the Keck/LRIS sample in @Erb:2006c at $z\sim2$. The galaxies at $z\sim0$ are the highly star-forming galaxies in which the absorption lines can be analyzed, while the galaxies at $z\sim1$ and $2$ are the star-forming main-sequence galaxies. The stellar masses of the galaxies are listed in Table \[tb:3\], which are similar to the galaxies at $z=5\nn6$ in this work.
@Sugahara.Y:2017a basically perform the same analysis as this study. The main difference in the method is to use a two-component absorption-line profile that consists of a systemic component fixed at the systemic velocity and a blueshifted component produced by outflowing gas. This difference of the fitting profile is negligible for the spectra at $z\sim0$ and $1$, where the absorption lines have small systemic components. The spectra at $z\sim2$, however, have large systemic components. We re-analyze the absorption lines of the normalized composite spectrum at $z\sim2$ to measure the maximum outflow velocity with the one-component absorption-line profile described in Section \[sec:analysis\]. The new maximum outflow velocity becomes lower than the previous value, but the conclusions in @Sugahara.Y:2017a are not affected by this re-analysis.
Although the available absorption lines differ depending on redshifts, @Sugahara.Y:2017a carefully compare the outflow velocities at $z\sim0\nn2$ measured from different lines and show an increase in the outflow velocities from $z\sim0$ to $2$. Table \[tb:3\] lists the values measured from the , , and absorption lines for the $z\sim0$, $1$, and $2$ galaxy spectra, respectively.
Results {#sec:results}
=======
Maximum Outflow Velocity vs. Galaxy Properties {#sec:results1}
----------------------------------------------
Figure \[fig:res1\] shows the maximum outflow velocity as a function of . The , , , and values are plotted with the open orange square, diamond, triangle, and filled red square, respectively. @Sugahara.Y:2017a illustrate that the outflow velocity increases from $z\sim0$ (blue) to $2$ (green) in star-forming galaxies with similar $M_*$ and . We find that the value at $z=5\nn6$ is $\sim0.2$ dex higher than the relation at $z\sim0$ and comparable to the value at $z\sim2$, although the values at $z\sim0$ are not as high as those at $z=5\nn6$. This means that the outflow velocity shows a strong increase from $z\sim0$ to $2$ and a slight or no increase from $z\sim2$ to $6$ in galaxies with similar $M_*$ and .
Figure \[fig:res2\] illustrates as a function of that are calculated from $M_*$ in Section \[sec:sample\]. Because the galaxies at $z=0\nn6$ have similar $M_*$ and $M_\mathrm{h}$ values (Table \[tb:3\]), the data points are located in different ranges depending on the redshifts. In the figure, tightly correlates with at $z\sim0$. A correlation with a similar slope at $z\sim0$ is also seen in the cyan diamonds at $z\sim1$. Although only one measurement is available at $z\sim2$ and $z=5\nn6$, respectively, the two data points at $z\sim2\nn6$ appear to follow the relation at $z\sim0\nn1$. Therefore, Figure \[fig:res2\] suggests a single relation between and that holds over $z\sim0\nn6$. The dotted line indicates a relation at $z=0$ obtained from observations by the Cosmic Origin Spectrograph mounted on the Hubble Space Telescope [@Heckman:2016], which has a similar slope to our measurements. The offset between our data points and the dotted line may arise from the fact that our data points represent the average properties of galaxies at each redshift while their extreme-starburst galaxies have much higher SFR than our galaxies.
Redshift evolution of outflow velocities {#sec:results2}
----------------------------------------
We illustrate the redshift evolution of in star-forming galaxies with $M_*\sim 10^{10.1}$ in the top panel of Figure \[fig:evolution\]. As shown in Section \[sec:results1\], $\vmax$ strongly increases from $z\sim0$ to $2$ and slightly from $z\sim2$ to $6$. Although many studies measure outflow velocities at fixed redshift, a few studies investigate the redshift evolution of the velocities in wide redshift ranges. @Jones.T:2013b present the maximum outflow velocity of gravitationally lensed sources at $z\sim2\nn4$ (cross). The stellar masses of these sources are not estimated and the outflow velocity of them is measured in a different manner from ours. However, the sources show similar outflow velocities to our values at $z\sim2$ and $z=5\nn6$, except for a data point of $\vmax \simeq 300$ . @Jones.T:2013b suggest a decrease in at high redshift that is not statistically significant. We find no decrease at $z=5\nn6$.
@Sugahara.Y:2017a and @Du.X:2018a discuss the redshift evolution of the central outflow velocity () measured with a two-component profile. While the maximum outflow velocity shows the highest velocity that the outflowing gas reaches, the central outflow velocity represents the bulk motion of the outflowing gas. Because the absorption lines include absorption components from outflows and ISM at the systemic velocity, should be estimated by a two-component fitting. Although a two-component fitting produces larger errors of the best-fit values than a one-component fitting, we fit a two-component Gaussian profile to the absorption line in the composite spectrum at $z=5\nn6$, in order to compare the values with previous studies at $z \lesssim 4$.
The two-component Gaussian profile consists of the systemic and outflow components; is defined as the central velocity of the outflow component. This analysis is identical to that used in @Du.X:2018a. Before the fitting, the composite spectrum is smoothed by a Gaussian kernel so that the spectral resolution become similar to the composite spectrum at $z\gtrsim2$ in @Sugahara.Y:2017a and @Du.X:2018a. We also analyze the composite spectrum at $z\sim2$ presented in @Sugahara.Y:2017a.
The measured velocities are $\vout=400^{+100}_{-150}$ at $z=5\nn6$ and $\vout=352^{+26}_{-27}$ at $z\sim2$. In the bottom panel of Figure \[fig:evolution\], we plot the measured values, showing that the redshift evolution has similar features to the evolution: a strong increase from $z\sim0$ to $2$ and no increase from $z\sim2$ to $6$ within the errors. The latter is consistent with a result of @Du.X:2018a. The and values at $z\sim0$, $1$, $2$, and $5\nn6$ are listed in Table \[tb:4\].
The open diamonds indicate at $z\sim2$, $3$, and $4$ given by @Du.X:2018a. The value at $z=5\nn6$ is comparable to those at $z\sim3$ and $4$ within the marginally large error bars. However, the value at $z\sim2$ denoted by the green diamond is not consistent with the one denoted by the open diamond. In addition, the error bars of the open diamonds are generally larger than those of the filled symbols, in spite of the fact that @Du.X:2018a stacked a larger number of galaxy spectra than this study and @Sugahara.Y:2017a. These results may be attributed to the uncertainty of the systemic redshifts in @Du.X:2018a, who determine the systemic redshifts from the emission or interstellar absorption lines. When individual spectra are stacked using the systemic redshifts, the uncertainties of the systemic redshifts broaden absorption lines in the composite spectrum. It is possible that this broadened absorption line produces low values and large errors of the best-fit parameters measured with the two-component fitting, which are sensitive to absorption-line profiles. We note that the median stellar masses of the galaxies in @Du.X:2018a are $\log (M_*/M_\odot) = 10.00$, $9.87$, and $9.72$ at $z\sim2$, $3$, and $4$, respectively, which are less than $M_*$ of our galaxies. It is also possible that this small $M_*$ (i.e., small ) may lead to the low value at $z\sim2$.
Discussion {#sec:discussion}
==========
Comparisons with theoretical models {#sec:physical-background}
-----------------------------------
Recent numerical and zoom-in simulations can be used to predict outflow velocities. These simulations compute energy input to the ISM surrounding SNe and investigate the statistics of galaxy and outflow properties [e.g., @Muratov:2015; @Christensen:2016a; @Mitchell.P:2018a; @Nelson.D:2019a]. Here we compare our results with simulation work that studies the redshift evolution of the outflow velocities. In the figures we convert $M_*$ () into ($M_*$) in the simulation work by using Equations (\[eq:0a\])–(\[eq:0b\]) and the SHMR in @Behroozi:2013.
@Nelson.D:2019a analyze $\sim$20,000 galaxies in the IllustrisTNG simulation to provide statistical relations between outflow and galaxy properties, including SN and AGN feedback. Figure \[fig:res2\] shows outflow velocities, $v_\mathrm{out,90,r=10\ kpc}$, defined as the 90th percentile of the flux-weighted velocity distribution at a radius of 10 kpc, at $z=0.2$, $1$, $2$, and $6$ in stellar mass ranges similar to observational data points at the redshifts. This theoretical prediction by @Nelson.D:2019a agrees well with our observational measurements, although the trend at $z\sim0$ is different.
@Muratov:2015 calculate the flux-weighted velocity of the outflowing gas at one quarter of the halo virial radius with the Feedback in Realistic Environments (FIRE) simulations, which computes the thermal and momentum input to the ISM considering the stellar and SN feedback. The outflow velocity in the FIRE simulations tightly correlates with the halo circular velocity and the correlation does not exhibit the significant evolution over $z\sim0.5\nn4$. Figure \[fig:res2\] illustrates that the correlation in the FIRE simulations at $z=0.5\nn4$ is in good agreement with the tight linear relation which we present in Section \[sec:results1\], although the outflow velocity at $z\sim0$ is $\sim0.1$ dex lower than the theoretical prediction.
These agreements with theoretical work support our result that correlates with in a given stellar mass range. However, we note two factors that are important when one compares observations and theories: gas phases and galactocentric radii of outflows. As described in Section \[sec:analysis\] our observational technique traces low-ionized elements in warm gas ($\lesssim 10^4\ \mathrm{K}$). On the other hand, @Muratov:2015 and @Nelson.D:2019a compute the –($M_*$) relation from outflowing gas with all temperatures. It is noteworthy that a non-negligible fraction of outflowing gas in numerical simulations would be in a hot, diffuse phase which are not observable with optical absorption lines [e.g., @Mitchell.P:2018a; @Nelson.D:2019a]. Moreover, in some numerical simulations, outflows in a hot phase tend to exhibit faster velocities than those in cold phases [e.g., @Gallerani.S:2018a; @Mitchell.P:2018a]. In addition to gas phases, galactocentric distances of outflowing gas are different between observations and simulations. Observations with the “down-the-barrel” technique integrate outflowing-gas absorption along the line of sight and cannot distinguish absorption components at different radii, while most simulations compute outflow velocities at fixed radii. Even among the simulations, @Nelson.D:2019a and @Muratov:2015 compute velocities at different radii, $10$ and $0.25$ halo virial radius, respectively, despite a radial dependence of outflow velocities [@Nelson.D:2019a]. Considering these all factors, it is difficult to interpret the similar – relation in this work, @Nelson.D:2019a, and @Muratov:2015. Nevertheless, the agreement perhaps suggests that multi-phase outflows are accelerated following a common – relation, irrespective of gas phases.
In Figure \[fig:evolution\] the black solid and dot-dashed lines indicate the redshift evolution of predicted outflow velocities at $M_* = 10^{10.1}$ based on the results given by the FIRE [@Muratov:2015] and the IllustrisTNG [@Nelson.D:2019a] simulations, respectively. The evolution based on the simulations is in good agreement with the and values in this study and @Sugahara.Y:2017a, and also with those in @Du.X:2018a and @Jones.T:2013b, except for one at $z\sim4$.
These simulations support a monotonic increase in from $z=0$ to $6$ at a fixed stellar mass of $M_*\sim10^{10.1}$ . This increase will be explained by a monotonic increase in . While $M_\mathrm{h}$ does not significantly change around $M_*\sim10^{10.1}$ at $z\sim0\nn6$ [@Behroozi:2013], $r_\mathrm{h}$ is proportional to $(1+z)^{-1}$ at a fixed $M_h$ (Equation \[eq:0b\]). Hence, Equation (\[eq:0a\]) gives the redshift dependence of the halo circular velocity as $\vcir \propto (1+z)^{0.5}$. Given that has the linear correlation with as shown in Figure \[fig:res2\], the redshift evolution in (Figure \[fig:evolution\]) is explained as reflecting the redshift dependence of . The power-law index of $0.5$ reproduces the strong increase in from $z\sim0$ to $2$ and the slight increase from $z\sim2$ to $6$.
Outflow-velocity correlation with {#sec:relation-between-sfr}
----------------------------------
The outflow maximum velocity tightly correlates with the halo circular velocity, but it also has a strong correlation with . To study the fundamental parameter that determines the outflow velocity over all redshifts, it is worth discussing correlations of with galaxy properties over the wide redshift range. If there exists the fundamental parameter, it should exhibit a single scaling relation with that holds at fixed redshifts and throughout all redshifts.
Figure \[fig:vm-galprop\] plots as a function of , $M_*$, , $\SFR/M_*$, and $\Sigma_\SFR$. Because galaxy sizes are unavailable at high-$z$ due to the spacial resolutions, $\Sigma_{\SFR}$ is estimated on the assumption that galaxy sizes are proportional to redshifts by $(1+z)^{-1}$ [@Shibuya:2015]. First, we calculate the Spearman’s rank correlations, $r$, between and the galaxy properties over all redshifts. While $M_*$ has no correlation with , the other galaxy properties exhibit strong correlations of $r=0.81$ (), $0.78$ (), $0.90$ ($\SFR/M_*$), and $0.89$ ($\Sigma_\SFR$) with the $> 3\sigma$ significance levels. Next, we perform a linear fitting to the data points at all redshifts and only at $z\sim0$. The best-fit results are illustrated in Figure \[fig:vm-galprop\]. The best-fit slopes at all redshifts (black line) are positive for , , $\SFR/M_*$, and $\Sigma_\SFR$. For and , the data points show relatively small scatter within $\sim0.1$ dex with respect to the best-fit relation at all redshifts. For $\SFR/M_*$ and $\Sigma_\SFR$, however, the best-fit relations at $z\sim0$ (blue line) have large offsets from the data points at $z\sim1\nn6$, and the slopes of the best-fit relations at $z\sim0$ and at all redshifts are very different from each other. These correlation and linear-fitting tests demonstrate that $M_*$, $\SFR/M_*$ and $\Sigma_\SFR$ show scaling relations at $z\sim0$, but those scaling relations cannot explain the outflow velocity throughout all redshifts. Therefore, and are likely to have the tightest single relations with from $z\sim0$ to $6$.
The strong -correlations with and imply a strong correlation between and . To understand the – relation independent of redshifts, it is helpful to see the distribution of the star-forming main-sequence galaxies on a – plane. The top left panel of Figure \[fig:sfr-vcir\] illustrates the main sequences at $z\sim0.5$, $1$, $2$, and $6$ that are presented by @Speagle:2014a. The galaxies in this work and @Sugahara.Y:2017a have similar stellar masses in the range of $10.0 < \log(M_*/\mathrm{M_\odot}) < 11.0$. We note that the main sequence at $z\sim0.5$ (blue line) agrees with galaxies at $z\sim0.1$ (blue circles) because @Sugahara.Y:2017a construct the $z\sim0$ sample with the highly star-forming galaxies to analyze the absorption line. By converting $M_*$ into with the method in Section \[sec:sample\], we plot the main sequences on a – plane in the panel (a) of Figure \[fig:sfr-vcir\]. They show similar positive relations at all redshifts, leading to a positive correlation of the main-sequence galaxies, irrespective of redshifts. The data points indeed exhibit a strong positive correlation ($r=0.99$) at the $5.8\sigma$ significance level. This result naturally explains that has a correlation with and simultaneously. In other words, constraining the fundamental parameter requires more measurements in a wide range of the stellar masses, , and redshifts.
In the panels (b) and (c) of Figure \[fig:sfr-vcir\], we plot the main sequences on $\SFR/M_*$– and $\Sigma_\SFR$– planes, respectively, which are useful to understand Figure \[fig:vm-galprop\]. Contrary to those on the – plane, the main sequences have offsets in the positive direction from low to high redshifts. This demonstrates that the apparent positive correlation of the data points on the $\SFR/M_*$– and $\Sigma_\SFR$– planes are simply because the galaxies have the similar stellar masses. Given a redshift-independent correlation between and as discussed above, Figure \[fig:sfr-vcir\] illustrates scaling relations between and galaxy properties at each redshift, which clearly reproduces the distribution of the data points in Figure \[fig:vm-galprop\]. These simple models suggest that –$\SFR/M_*$ and –$\Sigma_\SFR$ relations of the main-sequence galaxies depend on redshifts, namely, that $\SFR/M_*$ and $\Sigma_\SFR$ are unlikely to be the fundamental parameter.
The parameters which most strongly correlate with are and , suggesting that the fundamental parameter is or . This result agrees with previous observational studies that present positive correlations of with $M_*$ [@Martin:2005; @Rubin:2014; @Erb:2012] or [@Kornei:2012; @Heckman:2015; @Heckman:2016]. In many cases, the outflow properties are assumed to be connected with star-forming activities in galaxies. However, affects through the halo accretion rate [e.g., @Harikane.Y:2018a; @Tacchella.S:2018a] and this process contributes to form the – correlation in Figure \[fig:sfr-vcir\]. Thus, because represents two key parameters for the outflow velocity, the gravitational potential and the star-forming activity, it is important to consider the possibility that is the fundamental parameter to determine the outflow velocity.
Lyman-continuum leakage {#sec:lyman-cont-leak}
-----------------------
The redshift $z=5\nn6$ is near the end of the cosmic reionization, when the neutral IGM has been ionized. Plausible ionizing sources are young, low-mass galaxies [e.g., @Robertson:2015; @Ishigaki.M:2018b], but their contribution is still a matter of debate. A key physical parameter is the escape fraction of the Lyman-continuum (LyC) photons from galaxies (). Because it is possible that the outflows help increase by creating holes in the neutral ISM from which the LyC photons can escape, the high-$z$ galaxies with outflows in this work are appropriate for the reionization study. However, direct measurements of are challenging for galaxies at $z=5\nn6$ because the LyC photons almost disappear by ionizing the neutral IGM. In this section, we discuss the value of our galaxies at $z=5\nn6$ with two indirect methods regarding the absorption lines.
In the first method, we calculate the covering fraction from the metal absorption lines. In cases of the optically-thick outflowing gas, absorption lines are saturated and the line depth gives the covering fractions. Assuming that the low-ionized elements are associated with the neutral-hydrogen gas, @Jones.T:2013b evaluate the maximum covering fraction () of the low-ionized elements from the low-ionized absorption lines as an upper limit of . Because our composite spectrum has the low continuum S/N, we define as $\Cfmax = 1-F_\mathrm{SiII}$, where $F_\mathrm{SiII}$ is the median flux density of the line from $-350$ to $-100$ in the normalized spectrum. Its error is calculated with the parametric bootstrap method based on the spectral noise. The measured value is $\Cfmax = 0.8\pm0.2$. We note that this value is likely smaller than the value evaluated by the method in @Jones.T:2013b because our value is calculated in the wide velocity range of 250 . We additionally measure the equivalent width () of the composite spectrum to be $\EWLya=6.05\pm0.45$ Å, using the emission strength from the stellar continuum at 1216–1221 Å.
Figure \[fig:fabs\] illustrates as a function of . Our measurement at $z=5\nn6$ (red square) is consistent with previous results [@Jones.T:2013b; @Leethochawalit.N:2016a] and on the linear relation at $z\sim2\nn4$ presented by @Jones.T:2013b [dashed line]. This is the first observational result showing that the linear relation between and holds even at $z>5$, provided that the relation is independent of the stellar mass. Using the value corresponding to $\EWLya=6.05$ Å on the relation, we obtain an upper limit of to be $\simeq 0.2$. This secure upper limit is too weak to constrain models where bright galaxies contribute to the cosmic reionization [e.g., $\sim10\%$; @Sharma.M:2017a]. However, @Jones.T:2013b emphasize that the property derived by this method is an upper limit. Following an equation derived by @Chisholm.J:2018b, who propose indirect estimations of using local LyC leaking galaxies, we obtain $\fescc \lesssim 0.5-0.6\Cfmax = 0.02$. Hence, the intrinsic is likely much lower than the upper-limit value.
In the second method, we consider the shape of the absorption-line profile using the outflow velocities. @Chisholm.J:2017a calculate the ratio of the maximum outflow velocity to the central outflow velocity ($v_{90}/v_\mathrm{cen}$) of galaxies at $z=0$. They find that the LyC leaking galaxies exhibit smaller ratios, $v_{90}/v_\mathrm{cen}\lesssim 5$, than galaxies without LyC leakage, although there are several galaxies with $v_{90}/v_\mathrm{cen} < 5$ but $\fescc = 0$. Here we use $|\vmax/v_0|$ for an alternative to $v_{90}/v_\mathrm{cen}$ used in @Chisholm.J:2017a. The ratio for the galaxies at $z=5\nn6$ is obtained to be $|\vmax/v_0| = 2.0\pm0.2$. This result suggests that the galaxies at $z=5\nn6$ are the LyC leaking galaxies, in contrast to the result of the first method. Further studies on both the LyC photons and the absorption-line properties will provide key quantities to address the challenge of estimating for galaxies at the epoch of reionization.
Conclusion {#sec:conclusion}
==========
We study the outflow velocities of star-forming galaxies at $z=5\nn6$ and discuss the redshift evolution of the outflow velocities from $z\sim0$ to $6$ by analyzing rest-frame FUV spectra of seven LBGs at $z=5\nn6$ taken by DEIMOS available to date. We construct a high-S/N composite FUV spectrum based on the systemic redshifts determined by ALMA \[\] 158 $\mu$m observations [@Capak:2015] to fit a line profile to the $\lambda1260$, $\lambda1335$, and $\lambda\lambda1394,1403$ absorption lines. One of the best-fit parameters $v_0$, the central velocity of the line profile, is significantly negative, which implies that the absorption lines are blueshifted by the outflows.
The maximum outflow velocity is measured from the best-fit parameters. The values for the low-ionized lines (and ) are comparable to the one for the high-ionized line (), within the moderately large errors. By a simultaneous fit to the and lines, we obtain $\vmax = 700^{+180}_{-110}$ , which is higher than those at $z\sim0$ and comparable to the one at $z\sim2$ presented by @Sugahara.Y:2017a. This result represents the redshift evolution of that strongly increases from $z\sim0$ to $2$ and weakly increases from $z\sim2$ to $6$, at the fixed stellar mass of $\log (M_*/\mathrm{M_\sun}) \sim10.1$. We additionally measure the central outflow velocity () by fitting a two-component Gaussian profile to the line, and confirm that the redshift evolution of is similar to the evolution.
Over $z\sim0\nn6$, $\log \vmax$ is linearly correlated with the halo circular velocity ($\log \vcir$) that are estimated from the stellar mass. This linear correlation can explain the increasing features of the evolution because is proportional to $(1+z)^{0.5}$ for the galaxies with $\log (M_*/\mathrm{M_\sun}) \sim10.1$, at which the halo mass is almost constant over $z\sim0\nn6$ [@Behroozi:2013]. In addition, the correlation between and is in good agreement with a relation predicted by the IllustrisTNG [@Nelson.D:2019a] and the FIRE [@Muratov:2015] simulations. Although there are differences of gas phases and galactocentric radii between the simulation and observational work, this good agreement perhaps suggest that the multi-phase outflows are driven by a common – relation.
The outflow maximum velocity strongly correlates with , , $\SFR/M_*$, and $\Sigma_\SFR$ over $z=0\nn6$. Moreover, on the – and – planes, the linear scaling relations at $z=0$ explain the whole distribution from $z=0$ to $6$. Given that the – relation holds at any redshifts, the models of the star-forming main sequences reproduce the relation between and galaxy properties at $z=0\nn6$. For these reasons, or are likely to be the fundamental parameter to determine with a single relation throughout all redshifts. Considering that has an impact on through the halo accretion rate, it is possible that is the fundamental parameter.
Absorption-line profiles are also used for indirect estimations of the escape fraction of the LyC photons (). We find that the maximum covering fraction of the line and the equivalent width of the composite spectrum at $z=5\nn6$ are consistent with a relation at $z\sim2\nn4$. The intrinsic would be much lower than the secure upper limit $\fescc < 0.2$, while the ratio $|\vmax/v_0|$ is comparable to the values of the local LyC leaking galaxies.
We thank Xinnan Du, Kate Rubin, John Chisholm, Léo Michel-Dansac, Timothy M. Heckman, and Hidenobu Yajima for very useful discussion and comments. We wish to thank the referee for constructive and valuable suggestions for improvement. We acknowledge Peter Capak, the PI of the data in this work. The data presented herein were obtained at the W. M. Keck Observatory, which is operated as a scientific partnership among the California Institute of Technology, the University of California and the National Aeronautics and Space Administration. The Observatory was made possible by the generous financial support of the W. M. Keck Foundation. This research has made use of the Keck Observatory Archive (KOA), which is operated by the W. M. Keck Observatory and the NASA Exoplanet Science Institute (NExScI), under contract with the National Aeronautics and Space Administration. The authors wish to recognize and acknowledge the very significant cultural role and reverence that the summit of Maunakea has always had within the indigenous Hawaiian community. We are most fortunate to have the opportunity to conduct observations from this mountain. This work is supported by World Premier International Research Center Initiative (WPI Initiative), MEXT, Japan, and KAKENHI (15H02064, 17H01110, and 17H01114) Grant-in-Aid for Scientific Research (A) through Japan Society for the Promotion of Science. Y.S. acknowledges support from the JSPS through the JSPS Research Fellowship for Young Scientists.
[^1]: KOA website: <http://www2.keck.hawaii.edu/koa/public/koa.php.>
[^2]: Throughout this paper, is not the maximum circular velocity in the rotation curve of a galaxy or halo, which is often used in theoretical papers.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We study theoretically the effects of confinement on active polar gels such as the actin network of eukaryotic cells. Using generalized hydrodynamics equations derived for active gels, we predict, in the case of quasi one-dimensional geometry, a spontaneous flow transition from a homogeneously polarized immobile state for small thicknesses, to a perturbed flowing state for larger thicknesses. The transition is not driven by an external field but by the activity of the system. We suggest several possible experimental realizations.'
author:
- 'R.Voituriez$^{1}$, J.F. Joanny$^1$ and J. Prost$^{1,2}$'
title: Spontaneous flow transition in active polar gels
---
Introduction and statement of the problem
=========================================
Active materials are a challenging class of systems driven out of equilibrium by an internal or an external energy source. Many examples of active systems are provided by the biological world such as self–propelled particle assemblies in bacterial colonies, or the membrane or the cytoskeleton of eukaryotic cells [@albe02]. The cell cytoskeleton is a complex network of long filamentary proteins (mostly F-actin, microtubules and intermediate filaments) interacting with a variety of smaller proteins [@howa01] which can, among other things, crosslink or cap the filaments. A well studied class of proteins interacting with actin and microtubules are motor proteins, myosin, kinesin or dyneins. These proteins use the chemical energy of Adenosinetriphosphate (ATP) hydrolysis to “walk” along the filaments, and exert stresses that deform the filament network [@taki91; @nede97]. The active properties of the cytoskeleton play a crucial role in most for cell functions such as intracellular transport, motility and cell division.
Many efforts towards understanding the mechanical properties of the cytoskeleton have focused on the description of its passive visco-elastic properties which are well understood in terms of a gel built by cross-linked semi-flexible polymers [@head03; @wilh03]. More recently, Kruse et al. [@krus04; @krus05] have proposed a generalized hydrodynamic theory based on conservation laws and symmetry considerations, to describe macroscopically [*active*]{} polar gels. A typical example is given by the network of actin cytoskeletal filaments in the presence of myosin II motor proteins which generate active processes by hydrolyzing ATP. Since cytoskeletal filaments are structurally polar (with a + and - end), each filament locally defines a unit vector. The filamental structure gives rise on large scales to a macroscopic polarity if the filaments are on average aligned.
Experiments, numerical simulations and analytical descriptions, have shown that the cell cytoskeleton has a rich and complex mechanical behavior [@taki91; @nede97; @krus00; @krus01; @lee01; @kim03; @live03; @seki91]. In particular, self-organized patterns, including asters, vortices, and rotating spirals have been observed as a function of motor and ATP concentrations in a two-dimensional geometry [@nede97], and have been recently reproduced theoretically [@krus04; @menon].
It is well known that boundary effects can play a very important role in the formation of self-organized patterns. Here, we use the generalized hydrodynamic description of Ref. [@krus04; @krus05] to study analytically the effect of confinement of active polar gels confined between two parallel surfaces for various kinds of boundary conditions (free, no–slip, mixed and active). We predict a “Frederiks–like” flow transition from a homogeneously polarized immobile state for small thicknesses, to a inhomogeneous flowing state for larger thicknesses. This transition is reminiscent of the classical Frederiks transition in thin nematic liquid-crystal films since the system switches from a homogeneous to a non homogeneous polarization state. However there are two important differences, the transition in active polar gels does not require any external field and the non homogeneous active state is mobile. The critical length at which the transition occurs is monitored by the active stress proportional to the ATP/ADP chemical potential difference $\Delta \mu$. An experimental study of this spontaneousflow transition could give access to the active parameters of the actin-myosin cytoskeleton.
We now expose briefly the model in 2 dimensions, following Ref.[@krus04; @krus05]. The network of actin filaments has a macroscopic polarity described by a unit vector polarization field ${\bf p}=(\cos \theta,\sin\theta)$ (see Fig.\[figure:fig1\]). The associated polarization free energy is given by the standard expression for a polar liquid crystal [@dege93] : $$F=\int dxdy\left[ \frac{K_1}{2}(\nabla\cdot {\bf p} )^2 +\frac{K_3}{2}(\nabla\times{\bf p})^2 -
\frac{1}{2}h_\parallel {\bf p}^2 \right]$$ where $K_1=K$ and $K_3$ are the splay and bend elastic moduli. Note that there is no twist free energy in 2 dimensions, and that we do not write here the spontaneous splay term $k\nabla\cdot {\bf p}$ allowed by polar symmetry. This term is equivalent to a surface free energy of the form $k({\bf p}\cdot{\bf n})$ (${\bf n}$ being the unit vector normal to the surface). In the strong anchoring limit, this term is small compared to the anchoring energy that could be written as $k'({\bf p}\cdot{\bf n})^2$ with a large energy $k'\to\infty$. In this strong anchoring limit, the elastic free energy is the same as that of a classical nematic liquid crystal and specific polarity effects only show up in the dynamical equations \[uab\], \[eq:dpdt\] below. In the case of weak anchoring, the spontaneous splay could become dominant; as it does not allow for a state of uniform polarization, it would exclude the transition described at length in this paper. The Lagrange multiplier $h_{||}$ is introduced in order to satisfy the constraint ${\bf p}^2=1$. The molecular field, conjugate to the polarization is given by $h_\alpha=-\delta F/\delta p_\alpha$; in the following, we use its parallel and perpendicular coordinates $(h_\parallel,h_\perp)$ in the local frame linked to the polarization ${\bf p}$.
The gel motion is described by the velocity field ${\bf v}$ or the strain rate tensor $u_{\alpha\beta}=(\partial_{\alpha}v_\beta+\partial_{\beta}v_\alpha)/2$. The vorticity tensor $\omega_{\alpha\beta}=(\partial_{\alpha}v_\beta-\partial_{\beta}v_\alpha)/2$ is its antisymmetric counterpart. We assume here that the gel is incompressible. The gel is driven out of equilibrium by a constant chemical potential difference $\Delta \mu$ between ATP and its hydrolysis products. We consider here a visco-elastic gel which has a liquid behavior at long time scales. In a steady state, the gel behaves as a Newtonian liquid and its elasticity is irrelevant. The case of active nematic elastomers would lead to a different behaviour not described by our theory, as the classical Frederiks transition in these materials is very different from that of usual nematic liquids [@warner]. The linear generalized hydrodynamics equations for an incompressible active polar gel, at long time scales, read [@krus04; @krus05]: $$\begin{aligned}
2\eta u_{\alpha\beta} & = & \sigma_{\alpha\beta}
+ \zeta\Delta\mu p_\alpha p_\beta -\frac{\nu}{2}(p_\alpha h_\beta+p_\beta h_\alpha)
+\frac{1}{2}(p_\alpha h_\beta-p_\beta h_\alpha ) +{\bar \zeta}\Delta\mu\delta_{\alpha\beta}
\label{uab}\\
\frac{D p_\alpha}{D t} &=& \frac{1}{\gamma} h_\alpha
+ \lambda p_\alpha \Delta\mu - \nu u_{\alpha\beta}p_\beta \label{eq:dpdt}\end{aligned}$$ where we have used the corotational time derivative of the vector $p_\alpha$, $\frac{D}{Dt}p_\alpha=\frac{\partial p_\alpha}{\partial t}+(v_\gamma\partial_\gamma)p_\alpha
+\omega_{\alpha\beta} p_\beta$. The full derivation of these constitutive equations is given in [@krus05] and will not be discussed here. The rotational viscosity $\gamma$ and the coupling constant between flow and polarization $ \nu$ are standard liquid crystal parameters [@dege93]. The active contributions to the mechanical stress and to the rate of variation of the polarization are proportional to $\Delta \mu$ and are characterized by the coefficients $\zeta, {\bar \zeta}$ and $\lambda$. We limit here the theory to linear order in the activity and we neglect the geometric non-linearities introduced in [@krus04; @krus05]. The diagonal active term proportional to $\Delta\mu\delta_{\alpha,\beta}$ contribute to the absolute value of the pressure in an incompressible gel.
This set of constitutive equations is completed by the force balance equation: $\partial_\alpha (\sigma_{\alpha\beta}-\Pi\delta_{\alpha\beta})=0 $; locally, there are two forces acting on the gel, the deviatory stress tensor $\sigma_{\alpha\beta}$ and the pressure $\Pi$ that insures the incompressibility of the gel.
Spontaneous flow transition
===========================
We now look for the stationnary states of an active film in a confined geometry: we assume that the system is translationally invariant along the $y$ direction and that the gel is confined between $x=0$ and $x=L$. We consider several hydrodynamic boundary conditions. As discussed above, in the strong anchoring limit the polarization orientation is strictly defined on the confining surfaces. For simplicity, we only consider here the planar anchoring where the polarization at the confining surfaces is parallel to the $y$ direction, $\theta=\frac{\pi}2$. Similar results are obtained for an homeotropic anchoring, $\theta =0$.
The symmetry of the problem implies that $v_x={\rm const}=0$, and that the transverse component of the stress tensor $\sigma_{xy}$ is constant (the force balance imposes that $\partial_x \sigma_{xy}=0$). The perpendicular component of the molecular field is $h_\perp=K\partial_x^2\theta$. This relation is exact if $K_1=K_3=K$, and only valid for small values of $\theta -\pi/2$ if $K_1=K \neq K_3$. In this slab geometry, the constitutive equations for the gel motion are rewritten as $$\begin{aligned}
2\eta u= \frac{\sigma_{yx}+\sigma_{xy}}{2}+\frac{\zeta\Delta\mu}{2}\sin2\theta-
\frac{\nu}{2}(h_\parallel\sin2\theta+h_\perp\cos2\theta);\qquad
\frac{\sigma_{xy}-\sigma_{yx}}{2}+\frac{h_\perp}{2} =0
\label{u}\end{aligned}$$ where for simplicity we have denoted $u\equiv u_{xy}$. In turn, the polarization equation \[eq:dpdt\] gives $$\begin{aligned}
u\nu\sin2\theta =\frac{h_\parallel}{\gamma}+\lambda\Delta\mu;\qquad
u(\nu\cos2\theta-1) = \frac{h_\perp}{\gamma}
\label{p}\end{aligned}$$
Hydrodynamic free boundary conditions
=====================================
We first consider a free standing film where the gel slides freely on the confining surfaces. The transverse stress $\sigma_{xy}(x)$ vanishes since $\sigma_{xy}(x=0)=\sigma_{xy}(x=L)=0$. With this boundary condition, equations \[u\],\[p\] can be recast into a differential equation for the angle $\theta$ giving the polarization orientation: $$\partial_x^2\theta=\frac{{\tilde \zeta}\Delta\mu\sin2\theta(\nu\cos2\theta-1)}
{K[4\frac{\eta}{\gamma}+\nu^2-2\nu\cos2\theta+1]}=\Phi_f(\theta)$$ with ${\tilde \zeta}=\zeta +\nu\gamma\lambda$. Here we assume ${\tilde \zeta}\Delta\mu<0$, which has been shown to be the case generating self–motion of point–like defect in Ref. [@krus04]. A full understanding of the microscopic origin of the penomenological couplings $\zeta, \lambda$ is still missing. However, it is known experimentally [@thou] that for the actin-myosin gels forming the cell cytoskeleton, the stress is contractile corresponding to $\zeta\Delta\mu<0$ (a contractile effect requires a positive stress $\sigma_{\alpha\beta}$, all other terms being zero in Eq.\[uab\]). The contractility has also been reproduced by microscopic models [@live03]. Eq. \[eq:dpdt\] suggests that $\lambda\Delta\mu$ must be positive in order to account for the self–alignement effects such as the zipping effects observed in [@uhde04]. The effective potential $V_f$ associated to the force $\Phi_f(\theta)$, $V_f=-\int\Phi_f(\theta)d\theta$, can be expanded around the value imposed by the anchoring on the confining surfaces, $\theta=\pi/2+\epsilon$: $$\label{Vdev}
V_f(\theta)=V_f(\pi/2)-\frac{{\tilde \zeta}\Delta\mu(\nu+1)}
{K(4\eta/\gamma+(\nu+1)^2)}\epsilon^2+o(\epsilon^2)=
V_f(\pi/2)+\frac{\epsilon^2}{\ell^2}+o(\epsilon^2)$$ Following the standard argument for nematic liquid crystals [@dege93], if the length $\ell$ defined by Eq.\[Vdev\] is real (i.e. if the effective active stress is actually contractile ${\tilde \zeta}<0$ ), the system exhibits a continuous spontaneous flow transition at constant activity for a critical size $L_c=\ell\pi/\sqrt{2}$, or equivalently at constant thickness for a critical activity $$\Delta\mu_c=\frac{\pi^2K(4\eta/\gamma+(\nu+1)^2)}{-2L^2{\tilde \zeta}(\nu+1)}$$ For $L<L_c$ (or $\Delta\mu<\Delta\mu_c$), the system is dominated by anchoring effects and the orientation of the polarization is constant $\theta(x)=\pi/2$, whereas for $L>L_c$ ( $\Delta\mu>\Delta\mu_c$) the polarization is tilted in the bulk down to a minimum angle $\theta_{min}=\theta(L/2)<\pi/2$. We recall that close to threshold, $L\sim L_C$, the tilt amplitude is $\epsilon(x)=\epsilon_m\sin (x\pi/L)$ with $\epsilon_m\propto\sqrt{L-L_c}$ or $\epsilon_m\propto\sqrt{\Delta\mu-\Delta\mu_c}$.
The flow field can be calculated at linear order in $\epsilon_m$ from the velocity gradient $u=({\tilde \zeta}\Delta\mu\sin2\theta)/(4\eta+\gamma(\nu^2-2\nu\cos2\theta+1))$. We choose here the integration constant (the velocity at the midplane) in such a way that the total net flow vanishes (see Fig\[fig3\]a):
$$v_y=\frac{4L{\tilde \zeta}\Delta\mu\epsilon_m\cos(x\pi/L)}{\pi(4\eta+\gamma(\nu+1)^2)}.$$
In the case of an infinite system along the $x$ direction, the polarization reaches the minimum tilt angle defined by $\cos 2\theta_{min} =1/\nu$ when $\vert \nu \vert >1$. If $\vert \nu \vert <1$ the angle decreases to zero. Note that when $\nu =0$ the problem can be mapped exactly onto the original Frederiks problem for any angle value and $K_1 \neq K_3$. As opposed to a liquid crystal in an externally imposed shear flow, there is no flow tumbling here because the transverse stress $\sigma_{xy}$ vanishes. One can check directly by plotting the effective potential $V_f(\theta)$ that the state $\theta=0$ is stable.
Hydrodynamic no-slip boundary conditions
========================================
For solid confining surfaces when the gel does not slide on the walls, $v_y(x=0)=v_y(x=L)=0$. The equation for the polarization angle $\theta$ is in this case: $$\label{ns}
\partial_x^2\theta=\frac{({\tilde \zeta}\Delta\mu\sin2\theta+2\sigma_{xy})
(\nu\cos2\theta-1)}{K[4\eta/\gamma+\nu^2-2\nu\cos2\theta+1]}=\Phi_{ns}(\theta)$$
At linear order in the polarization tilt, the polarization equation reads $$\partial_x^2\epsilon=\frac{2(\nu+1)}{K[4\eta/\gamma+(\nu+1)^2]}
({\tilde \zeta}\Delta\mu\epsilon -\sigma_{xy})$$ and the polarization angle is given by $$\epsilon=\frac{\sigma_{xy}}{{\tilde \zeta}\Delta\mu}\left(1-\cos(\pi x/L_c)-\tan(\frac{\pi L}{2L_c})\sin(\pi x/L_c)\right)
\label{tilt}$$
If $\sigma_{xy}$ were different from $0$, there would be no spontaneous flow transition. However, we can show that in the low activity regime $\sigma_{xy}=0$. Indeed when the activity $\Delta\mu$ is small, the tilt angle $\epsilon$ is small. One can then directly check using equation \[tilt\] that the velocity gradient does not change sign for any constant value of $\sigma_{xy}\neq 0$. This is incompatible with the no-slip boundary conditions and necessarily $\sigma_{xy}= 0$. In turn the polarization remains uniform. For larger $\Delta \mu$ a spontaneous flow transition is reached.
In the vicinity of the spontaneous flow transition, the maximum tilt angle is small and one can again linearize the polarization equation. The polarization tilt $\epsilon$ is then given by equation \[tilt\]. The integration of the velocity gradient with the boundary condition $v_y(x=0)=0$ leads to
$$v_y=\frac{4\sigma_{xy}L_c}{\pi(4\eta+\gamma(\nu+1)^2)}\left(\sin(\pi x/L_c)-\tan(\frac{\pi L}{2L_c})(\cos(\pi x/L_c)-1)\right)$$
The boundary condition $v_y(L)=0$ can be satisfied only if $L=2L_c$. The spontaneous flow transition occurs therefore at $L=2L_c$, or equivalently at $\Delta\mu=4\Delta\mu_c$ (see Fig\[fig3\]b).
Above threshold, the tilt cannot be considered as small and a full non-linear analysis is required. As in the previous case, close to the transition, the maximum tilt angle and the transverse stress $\sigma_{xy}$ scale like $\sigma_{xy} \propto \epsilon_m\propto\sqrt{L-L_c}$.
Hydrodynamics mixed boundary conditions
=======================================
The easiest experimental conditions are that of a film on a solid plane with a free surface ($v_y(x=0)=0, \sigma_{xy}(x=L)=0$). As the stress vanishes everywhere, the polarization field is the same as for a free standing film with free boundary conditions. The transition threshold is given by Eq. (8). In this case a finite gel flux $Q$ is generated above threshold $$Q = \int_0^L v_y dx=-\frac{4L{\tilde \zeta}\Delta\mu \quad \epsilon_m}{\pi[4\eta+\gamma(\nu+1)^2]}.$$ with $\epsilon_{m}=-\sqrt{2\pi(L-L_c)/L_c}$. The velocity profile is sketched on figure \[fig4\]. This provides a striking example of self-generated motion that could be observed in an annular geometry. The Fredericks transition could further play a role in lamellipodium motility [@albe02].
For completeness, we also study the case where the slip is different on the two surfaces: $v_y(x=0)=\mu \sigma_{xy}; \quad v_y(x=L)=-\mu' \sigma_{xy}$. This could perhaps be realized experimentally by inserting a liquid film between the gel and a solid confining surface and working with homeotropic boundary conditions ($\theta=0$). There is, in this case, a spontaneous flow transition at a thickness $L$ such that $L_c<L<2L_c$ given by $$\tan (\frac {\pi L}{2L_c})=-\frac{\pi(\mu+\mu')(4\eta+\gamma(\nu+1)^2)}{8L_c}$$ If the boundary conditions are asymmetric ($\mu\neq \mu'$), the transition leads to the apparition of a finite gel flux in the direction $y$ parallel to the confining surfaces.
Active boundary conditions
==========================
Active boundary conditions can be generated by confining the gel between two planar surfaces coated with molecular motors. The motors impose a finite gel velocity on the surfaces. If the confining surfaces are identical, the velocities on the two surfaces are equal; this is equivalent to no-slip boundary conditions with a constant drift at the motors velocity. If the two velocities imposed by the motors are different, there is always a finite shear stress $\sigma_{xy}$ and a finite tilt of the polarization: there is no continuous spontaneous flow transition in this case.
Concluding remarks
==================
In this letter we predict a spontaneous Frederiks–like transition in active polar materials confined in one dimension: for small thicknesses or small activity, boundary effects are prevailing and the gel remains in an unperturbed, static, homogeneous state. Above a critical thickness or a critical activity, a polarization tilt appears and the system flows. The transition depends strongly on the very nature of the boundary conditions imposed by the confinement: homeotropic or planar anchoring are necessary, and the threshold value depends on the nature of the hydrodynamic boundary conditions. Interestingly, for asymmetric active boundary conditions, there is no continuous transition.
A detailed experimental study of the transition, using for instance actin-myosin gels in micro-channels is very promising. A measurement of the transition threshold will give direct access to the effective active stress $\tilde \zeta \Delta\mu$. The two-fold aspect of the transition, involving both dynamical and polarization properties should allow for various methods of visualization coming either from liquid crystal physics or from micro-fluidics. We are not aware at the moment of any quantitative measurement of these active stresses that would be essential to give a proper mesoscopic description of active actin gels. More generally, any geometry where anchoring imposes a non uniform polarization field in an active gel would lead to a hydrodynamic flow driven by the activity because of the couplings between polarization, activity and flow. A careful choice of the geometry and of the anchoring conditions could then allow for a determination of the various active coefficients of the gel ($\zeta$, $\lambda$) as well as that of the more standard liquid crystalline parameters ($\nu$,$\gamma$).
Our results are remarkable in the sense that they predict a spontaneous flow transition in the absence of any external field. This opens the way to a quantitative characterization of this new class of materials. We think that the understanding of this physics will prove to be useful in the study of the cell cytoskeleton. In particular, lamellipodia share several features with a polar active gel slab.
We are grateful to S.Fraden (Boston) for important discussions on possible experimental realizations of the spontaneous flow transition in active polar gels.
[0]{} B. Alberts [*et al.*]{}, [*Molecular Biology of the Cell*]{} 4th ed. (Garland, New York, 2002).
J. Howard, [*Mechanics of Motor Proteins and the Cytoskeleton*]{} (Sinauer Associates, Inc., Sunderland, 2001).
F.J. Nédélec [*et al.*]{}, [*Nature*]{} [**389**]{}, 305 (1997), F.J. Nédélec, T. Surrey, and A.C. Maggs, [*Phys. Rev. Lett.*]{} [**86**]{}, 3192 (2001), T. Surrey [*et al.*]{}, [*Science*]{} [**292**]{}, 1167 (2001).
K. Kruse and F. Jülicher, [*Phys. Rev. Lett.*]{} [**85**]{}, 1778 (2000).
K. Kruse, S. Camalet, and F. Jülicher, [*Phys. Rev. Lett.*]{} [**87**]{}, 138101 (2001). H.Y. Lee and M. Kardar, [*Phys. Rev. E*]{} [**64**]{}, 056113 (2001).
J. Kim [*et al.*]{}, [*J. Korean Phys. Soc.*]{} [**42**]{}, 162 (2003).
S. Sankararaman, G.I. Menon and P.B. Sunil Kumar, [*Phys. Rev. E*]{} [**70**]{}, 031905 (2004), M.C. Aronson [*Unpublished*]{}
T.B. Liverpool and M.C. Marchetti, [*Phys. Rev. Lett.*]{} [**90**]{}, 138102 (2003).
A. Simha and S. Ramaswamy, [*Phys. Rev. Lett.*]{} [**89**]{}, 058101 (2002).
E.M. Terentjev, M. Warner, R.B. Meyer and J. Yamamoto, Phys. Rev.E [**60**]{} (1999)
K.Sekimoto J:Phys.II (France) [**1**]{}, 19 (1991).
O. Thoumine, A. Ott, [*J. Cell. Sc.*]{} [**110**]{} (1997) J. Uhde, M. Keller and E. Sackmann, [*Phys. Rev. Lett.*]{} [**93**]{} (2004) K. Kruse, J.F. Joanny, F. Jülicher, J. Prost and K. Sekimoto, Phys. Rev. Lett [**92**]{}, 078101 (2004).
K. Kruse, J.F. Joanny, F. Jülicher, J. Prost and K. Sekimoto, Eur. Phys. J. E [**16**]{}, 5 (2005).
D.A. Head, A.J. Levine, F.C. MacKintosh, [*Phys. Rev. Lett.*]{} [**91**]{}, 108102 (2003).
J. Wilhelm and E, Frey, [*Phys. Rev. Lett.*]{} [**91**]{}, 108103 (2003).
K.Takiguchi J.Biochem. [**109**]{}, 520 (1991).
P.G. De Gennes and J. Prost, [*The Physics of Liquid Crystals*]{} (Clarendon Press, Oxford 1993).
| {
"pile_set_name": "ArXiv"
} |
---
author:
- |
Tao Li\
Department of Computer Science\
Purdue University\
[[email protected]]{}
- |
Lei Lin\
Goergen Institute for Data Science\
University of Rochester\
[[email protected]]{}
bibliography:
- 'db.bib'
title: 'AnonymousNet: Natural Face De-Identification with Measurable Privacy'
---
| {
"pile_set_name": "ArXiv"
} |
---
author:
- Yahel DavidDotan Di CastroZohar Karnin
bibliography:
- 'bibliography.bib'
title: 'One-Shot Session Recommendation Systems with Combinatorial Items'
---
Introduction
============
Related Work
============
Problem Formulation
===================
Approximation Of the Value Function {#sec:theory}
===================================
Experiments [^1] {#add_exp}
================
Discussion and Conclusions {#Discussion:Exp}
==========================
[^1]: Additional experiments are provided in Section \[sup:add:exp\] of the supplementary material.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
We study the zero-temperature properties of the Kondo lattice model within the dynamical mean-field theory. As impurity solver we use the numerical renormalization group. We present results for the paramagnetic case showing the anticipated heavy Fermion physics, including direct evidence for the appearance of a large Fermi surface for antiferromagnetic exchange interaction. Allowing for the formation of a Néel state, we observe at finite doping an antiferromagnetic metal below a critical exchange interaction, which shows a crossover from a local-moment antiferromagnet with a small Fermi surface for weak exchange coupling to a heavy-fermion antiferromagnet with a large Fermi surface for increasing exchange.
Including lattice degrees of freedom via an additional Holstein term we observe a significant suppression of the Kondo effect, leading to strongly reduced low-energy scale. For too large electron-phonon coupling we find a complete collaps of the heavy Fermi liquid and the formation of polarons.
address:
- 'Institute for Theoretical Physics, University of Göttingen, Friedrich-Hund-Platz 1, 37077 Göttingen, Germany'
- 'J. Stefan Institute, Jamova 39, SI-1000 Ljubljana, Slovenia'
- 'University of Kyoto, Department of Physics, Graduate School of Science, Kitashirakawa-Oiwakecho, 606-8502 Kyoto, Japan'
author:
- 'O Bodensiek$^1$, R Žitko$^2$, R Peters$^3$ and T Pruschke$^1$'
bibliography:
- 'references.bib'
title: 'Low-energy properties of the Kondo lattice model'
---
Introduction
============
Heavy Fermion systems based on 4f or 5f intermetallics are paradigms for electronic correlations in solid state physics. The low-temperature physics of these compounds is strongly influenced by the local moment on the f shell, subject to an antiferromagnetic exchange to the conduction electrons. The resulting physical properties are in many cases again Fermi liquid like, however with extremely enhanced Landau parameters, in particular an effective mass up to three orders of magnitude larger than the one found in conventional metals[@stewart:1984; @GreweSteglich:1991; @stewart:2001]. This large effective mass is the reason why these systems are referred to as *Heavy-Fermion materials* (HF). Moreover, in addition to these extreme Fermi liquid properties, the HF also show various phase transitions and the thermodynamics show that these transitions actually occur within the heavy Fermi liquid [@stewart:1984]. Finally, the appearance of superconductivity in a system with initially well-defined magnetic moments is a rather unconventional feature, and close investigation revealed early on that the nature of the ordered state may be rather unconventional [@GreweSteglich:1991]. This observation has been substantiated by the development over the past 15 years which showed that a larger number of these HF systems exhibit rather peculiar quantum-phase transitions, partially identified as the driving force behind the superconducting transitions [@stewart:2001; @loehneysen:2007].
Other materials which show a coupling between itinerant quasi-particles and localized spins are certain transition metal oxides [@imada:98], magnetic semiconductors or semi-metals in the series of the rare earth monopnictides and monochalcogenides [@ovchinnikov:91; @sharma:06], and diluted magnetic semiconductors such as Ga$_{1-x}$Mn$_x$As [@ohno:98; @dms_review]. Here, the coupling between local spin and conduction electrons usually mediated through Hund’s exchange and thus typically is ferromagnetic.
A theoretical description of HF compounds is conventionally based on the Kondo-lattice model (KLM) $$\label{eq:KLM}
H_{\rm KLM}=\sum_{\vec{k},\sigma}\epsilon_{\vec{k}}\hat{c}_{\vec{k},\sigma}^\dagger \hat{c}_{\vec{k},\sigma}-J\sum_i\vec{s}_i\cdot\vec{S}_i\;\;.$$ The operators $\hat{c}_{\vec{k},\sigma}^{(\dagger)}$ denote annihilation (creation) operators of itinerant quasi-particles with dispersion $\epsilon_{\vec{k}}$, $\vec{s}_i$ is the operator for the conduction states’ spin density at lattice site $\vec{R}_i$ and $\vec{S}_i$ describes a spin of magnitude $S$ localized at site $\vec{R}_i$. The interaction between the spin of the conduction states and the localized spin is modeled as conventional isotropic exchange interaction $J$.
The dilute version of the KLM [(\[eq:KLM\])]{}, the so-called single-impurity Kondo model (SIKM) where there exists only one additional spin at site $\vec{R}_i=0$, is well understood and shows for antiferromagnetic coupling $J<0$ the Kondo effect [@hewson:book], which precisely leads to the phenomena observed in the Fermi liquid phase of HF systems, viz a strongly enhanced mass. There are nowadays several computational tools to treat the SIKM, for instance continuous-time Monte-Carlo [@Otsuki:2007] or Wilson’s numerical renormalization group [@bullareview].
In theoretical treatments one usually ignores the lattice degrees of freedom. On the other hand, all the above mentioned materials have a rather strong electron-phonon coupling [@allen:1982; @imada:98] and one can expect that the charge physics driven by phonons somehow competes with the spin physics due to the exchange interaction with the localized spin. Moreover, without exchange coupling, phonons will lead to conventional $s$-wave superconductivity. Thus the investigation of the interplay between a coupling to a spin and the lattice degrees of freedom is highly interesting.
This paper is intended to give a summary of the physical properties of the KLM as seen by the dymanical mean-field theory. This necessesarily excludes nonlocal phenomena like unconventional superconductivity, but allows for the study of antiferromagnetism and whether it is always accompanied by a breakdown of the large Fermi surface. Inclusion of phonons eventually leads to superconductivity [@Bodensiek:2009], which however is of the standard local $s$-wave type. A detailed account of an investigation of the interplay of phonon-mediated superconductivity and HF physics will be presented elsewhere.
The paper is organized as follows. In the next section we discuss the model and the approximation used to solve it. The case without phonons, i.e. the conventional heavy-fermion physics both in the paramagnetic and the antiferromagnetically ordered state is the subject of section \[sec:parm\_metal\]. The effect of phonons on the low-energy properties will be discussed in section \[sec:phonons\]. A short summary and outlook will close the paper.
Model and Method
================
The KLM Hmiltonian [(\[eq:KLM\])]{} will be again the basic model. Except for one dimension, no analytical solution exists, and even conventional numerical tools such as Quantum Monte-Carlo (QMC) become rather cumbersome due to a severe sign problem away from particle-hole symmetry. Thus, a reliable approximate method is needed. If one is not interested in the properties too close to a phase transition or in the rather complicated, non-local ordering phenomena, a suitable tool is the dynamical mean-field theory (DMFT) [@georges:96]. Here, the lattice is mapped onto an effective single-impurity problem, which can be then solved using standard techniques. Here, we use the numerical renormalization group (NRG) approach [@wilson:75; @bullareview]. One of its apparent advantages is the possibility to access small energy scales without problem and cover the whole range from $T=0$ to finite temperatures of the order of the bare energy scales. Furthermore, it also allows to include phonons to a certain extent, namely an Einstein mode coupled through a Holstein term to the charge degrees of freedom. As is suggestive from effects like Kondo volume collapse [@allen:1982; @imada:98], such a term could be rather important. We will thus work with a Hamiltonian (for a detailed introduction and further references see [@assaad:2009]) $$\label{eq:KLM_Holstein}
H=H_{\rm KLM}
+
\omega_0\sum\limits_{i}b^\dagger_i b^{\phantom{\dagger}}_i
+g\sum\limits_{i\sigma}\left(
c_{i\sigma}^\dagger c_{i\sigma}^{\phantom{\dagger}}-1\right)\left(b^\dagger_i
+b^{\phantom{\dagger}}_i\right)$$ where $\omega_0$ is the frequency of an appropriate optical mode and $g$ a measure of the electron-phonon coupling. How such an additional local coupling can be treated within NRG, is described in detail in [@bullareview]. To obtain reasonably accurate spectra also for higher energies, we use the broadening strategy introduced by [@Freyn:2009].
As is well-known, one major effect of the Holstein model is to introduce an effective attractive interaction to the electronic subsystem: If the phonon frequency $\omega_0$ and the electron-phonon coupling $g$ become large, keeping $g^2/\omega_0$ constant, the phonons can be integrated out, yielding an attractive local Coulomb interaction $U_{\rm eff}=-2g^2/\omega_0$. Without explicit exchange interaction $J$, one will then obtain a Hubbard model with attractive $U$, which shows charge-density and superconducting ordering phenomena [@Freericks:1993; @Bauer:2009]. From the point of view of Kondo physics, the negative $U$ will lead to a Kondo-like behavior in the charge sector, strongly competing with the spin Kondo effect introduced by $J$. We thus can expect interesting physics to occur when both couplings are present.
KLM without phonons\[sec:parm\_metal\]
======================================
Let us begin with a comprehensive summary of the basics of heavy-fermion physics. We will use a $2D$ square lattice with nearest-neighbor hopping for the conduction states. Note that the DMFT is rather insensitive to the dimensionality, and we chose the $2D$ lattice to facilitate visualization of the results. Calculations were done with an NRG discretization parameter $\Lambda=2$, between 1000 and 5000 states were kept per NRG step and, where applicable, 50 bosons kept initially. These values were systematically changed for selected calculations to ensure that the results are independent of these numerical parameters.
Paramagnetic metal
------------------
Let us start with a comparison of the properties at finite $J$, but with $g=0$. We can distinguish two cases, namely an antiferromagnetic exchange $J<0$ and a ferromagnetic $J>0$. The resulting density of states (DOS) for $T=0$
![DOS at $T=0$ for the KLM [(\[eq:KLM\])]{} with $|J|=0.25W$, where $W$ denotes the bandwidth of the conduction electrons. The filling of the band is $n_c=0.8$. The dotted curve displays the DOS at $J=0$ as reference.[]{data-label="fig:DOS_no_phonons"}](figure1){width="80.00000%"}
and $|J|/W=0.25$ at a filling $n_c=0.8$ of the conduction band is shown in Fig. \[fig:DOS\_no\_phonons\]. The quantity $W$ denotes the bandwidth of the conduction band and will serve as energy scale hereafter. There are notable differences between the two cases $J<0$ and $J>0$. The DOS for $J>0$ looks very much like the DOS of the bare conduction band (dotted curve in Fig. \[fig:DOS\_no\_phonons\]), although it is somewhat broadened.
![Spectral functions (left) and momentum distribution (right) for $|J|=0.25W$, other parameters as in Fig. \[fig:DOS\_no\_phonons\]. The thick line in the lower left corner of the contour plot for $J<0$ denotes the position of the jump in the momentum distribution.[]{data-label="fig:Aofkw_no_phonons"}](figure2a){width="\textwidth"}
![Spectral functions (left) and momentum distribution (right) for $|J|=0.25W$, other parameters as in Fig. \[fig:DOS\_no\_phonons\]. The thick line in the lower left corner of the contour plot for $J<0$ denotes the position of the jump in the momentum distribution.[]{data-label="fig:Aofkw_no_phonons"}](figure2b){width="125.00000%"}
\
$J>0$
\
![Spectral functions (left) and momentum distribution (right) for $|J|=0.25W$, other parameters as in Fig. \[fig:DOS\_no\_phonons\]. The thick line in the lower left corner of the contour plot for $J<0$ denotes the position of the jump in the momentum distribution.[]{data-label="fig:Aofkw_no_phonons"}](figure2c){width="\textwidth"}
![Spectral functions (left) and momentum distribution (right) for $|J|=0.25W$, other parameters as in Fig. \[fig:DOS\_no\_phonons\]. The thick line in the lower left corner of the contour plot for $J<0$ denotes the position of the jump in the momentum distribution.[]{data-label="fig:Aofkw_no_phonons"}](figure2d){width="125.00000%"}
\
$J<0$
\
For $J<0$, however, the DOS is strongly modified [@otsuki:2009], showing a pseudo gap close to the Fermi energy $\omega=0$. This latter feature is a fingerprint of heavy Fermion physics, resulting from a picture of hybridized bands [@GreweSteglich:1991]. This interpretation becomes even more apparent when one looks at the spectral function along the standard $\vec{k}$ directions in the first Brillouin zone of the $2D$ square lattice in Fig. \[fig:Aofkw\_no\_phonons\], left part. Dark color means low intensity, bright color high. Included as full black line is the bare bandstructure of the $2D$ nearest-neighbor tight-binding band. For $J>0$ we basically see the band structure of the $2D$ nearest neighbor tight-binding band. There is a moderate broadening, which actually is to be expected, because for $J>0$ the local spin effectively acts as a potential scatterer for the band states [@hewson:book]. For such a situation the DMFT is equivalent to a CPA calculation, which yields a constant broadening. For $J<0$ (right panel), on the other hand, there is a flattening of the band structure close to the Fermi energy and a structure similar to a hybridization gap opens. The flat portion of the lower band corresponds to a large effective mass of the quasi-particles. Note that we have a rather sharp structure at the Fermi energy, i.e. one can indeed talk about quasi-particles here.
Another remarkable difference between the cases $J>0$ and $J<0$ is observed when one looks at the momentum distribution function $n(\vec{k})$ displayed in the right part of Fig. \[fig:Aofkw\_no\_phonons\]. As already noted for the spectral functions, the result for $J>0$ resembles the Fermi function, slightly smeared out by incoherent scattering from the spins. In any case, the Fermi surface is located at the $\vec{k}_F$ of a non-interacting $2D$ tight-binding band with a filling $n_c=0.8$. On the other hand, the momentum distribution for $J<0$ does not show any distinct features at this particular value of $\vec{k}$. Instead, one notes a small jump in $n(\vec{k})$ at a vector outside the square marking the Fermi surface of a half-filled system. A closer inspection shows that this $\vec{k}$ vector corresponds to a Fermi surface of a system with $n_c=1.8$, i.e. the system shows a “large” Fermi surface with the spin degrees of freedom contributing to the quasi-particles now. Note that the height of the jump in $n(\vec{k})$ is directly related to the inverse effective mass of the quasi-particles.
The results for $J<0$ in Fig. \[fig:Aofkw\_no\_phonons\] represent the essence of HF physics, namely the generation of heavy quasi-particles, represented by flat bands with a structure known from hybridized bands and a large Fermi surface.
Antiferromagnetic ordering
--------------------------
As is well known, the Kondo lattice model shows a variety of magnetically ordered states [@lacroix:1979; @fazekas:1991; @kienert:2006; @henning:2009]. Within DMFT, an overview was presented in [@Peters:2007; @chattopadhyay:2001]. The major findings at $T=0$ are: At half filling (Kondo insulator regime) one finds a critical $J_c<0$, with no magnetic phases present for $J<J_c$ and antiferromagnetism for $J>J_c$. The staggered magnetization of local spins and band electrons are opposite for $J<0$, leading to an effectively reduced total moment especially close to the “quantum critical” point. Note that this effect is enhanced when one includes interactions in the band electron system (see e.g. [@Peters:2007]).
Away from half filling there appear, at strong enough doping, ferromagnetic phases in addition [@santos:2002; @Peters:2007]. For $J<J_c$, no further magnetic phase was observed in the vicinity of half filling. For $J>J_c$, one can however again stabilize antiferromagnetic phases. For example, for a filling $n_c=0.9$ of the conduction band, such an ordered state is found for $J\lesssim-W/4$. To obtain a reasonable convergence and a stable solution, the DMFT calulations must be done with Broyden mixing [@Zitko:2009Broyden]. Quite interestingly, we are not able to find stable and reasonably converged solutions for $-0.125\lesssim J/W\lesssim-0.1$. The results for
\
![Total DOS for the $A$ sublattice at $T=0$ for $J/W=-0.05$ (full line) and $J/W=-0.15$ (dashed line). The table to the left shows the staggered magnetizations.[]{data-label="fig:DOS_AFM"}](figure3 "fig:"){width="99.00000%"}
\
[@l@[ ]{}l@[ ]{}l@[ ]{}l]{} $J/W$ & $m_s^I$ & $m_s^c$ & $m_s^t$\
$-0.05$ & $0.485$ & $0.057$ & $0.425$\
$-0.15$ & $0.2$ & $0.08$ & $0.12$\
$J/W=-0.05$ and $J/W=-0.15$ are shown in Fig. \[fig:DOS\_AFM\] together with a table of the values for the staggered magetization of the local spin, $m_s^I$, the conduction band, $m_s^c$, and their sum, $m_s^t$. A rather interesting question in this connection is whether the ordered state corresponds to a “local moment” regime, i.e. where the local spins are effectively decoupled from the band states and one has a small Fermi surface, or if it is a “heavy Fermion” magnet with a large Fermi surface. The most interesting case of course is when both appear as function of $J$ and there might be a phase transition associated with the change in Fermi surface topology. Obviously, inspection of the DOS in Fig. \[fig:DOS\_AFM\] alone is not sufficient to identify a possible transition.
Such information can however be obtained by inspecting the spectral function. The results are shown in Fig. \[fig:Aofkw\_AFM\_small\_J\], where the $\vec{k}$-vectors are now restricted to the first magnetic Brillouin zone (see for example [@pruschke:03]).
$J/W=-0.05$
![Spectral functions for $\vec{k}$ from the magnetic Brillouin zone for $n_c=0.9$. Left: Full energy range. Right: Magnified view around the chemical potential.[]{data-label="fig:Aofkw_AFM_small_J"}](figure4a){width="\textwidth"}
![Spectral functions for $\vec{k}$ from the magnetic Brillouin zone for $n_c=0.9$. Left: Full energy range. Right: Magnified view around the chemical potential.[]{data-label="fig:Aofkw_AFM_small_J"}](figure4b){width="\textwidth"}
\
$J/W=-0.15$
![Spectral functions for $\vec{k}$ from the magnetic Brillouin zone for $n_c=0.9$. Left: Full energy range. Right: Magnified view around the chemical potential.[]{data-label="fig:Aofkw_AFM_small_J"}](figure4c){width="\textwidth"}
![Spectral functions for $\vec{k}$ from the magnetic Brillouin zone for $n_c=0.9$. Left: Full energy range. Right: Magnified view around the chemical potential.[]{data-label="fig:Aofkw_AFM_small_J"}](figure4d){width="\textwidth"}
Included in the figure is as continuous black line the dispersion calculated with the Hartree term of the self-energy alone [@Negele]. This curve thus represents the proper approximation for the situation of ordered local moments polarizing the Fermi sea. The left part shows the overall structures and the right part a magnified view of the region around the Fermi energy.
Quite apparently, the concept of a small Fermi surface, i.e. ordered local spins polarizing the band states, describes the situation very well for $J/W=-0.05$. The situation changes completely for $J/W=-0.15$. First, the staggered magnetization has dropped to $m_s\approx0.12$ now, with $m_s^I\approx0.2$ and $m_s^c\approx0.08$ (see Tab. \[fig:DOS\_AFM\]). Thus, while the polarization of the band states has not changed significantly, the one for the local spins has dropped by more than 50%. Further increasing $|J|$ does not lead to any new features, the polarizations of both local moments and band electrons smoothly drop to zero for $J/W\approx-W/4$. Second, the spectral function has become much more HF like with a flat band around $\Gamma$ (c.f. Fig. \[fig:Aofkw\_no\_phonons\]). Therefore, the scenario of a decoupled Fermi sea and local moments ordering does surely not apply any more, in fact the whole system looks much more band like, with a large Fermi surface.[^1] This scenario is also supported by the fact, that for increasing coupling $|J|$ the system more strongly shows tendencies to rather form spin density waves than Néel order. These observations strongly hint towards a phase transition from a local moment like phase at small $|J|$ to a phase where magnetic ordering appears in the heavy quasiparticles at larger $|J|$. However, a decisive answer what type of phase transition this is at present cannot be given.
Effect of phonons\[sec:phonons\]
================================
![DOS at $T=0$ for the KLM [(\[eq:KLM\])]{} with $J=-0.5W$ and $g=0$, $g=0.2W$ and $g=0.4W$. The filling of the band is $n_c=0.8$. The inset shows the dependence of the effective mass on $g$.[]{data-label="fig:DOS_with_phonons"}](figure5){width="80.00000%"}
An extended account of the effect of Einstein phonons on HF physics is given in [@assaad:2009], where we concentrated on the periodic Anderson model. Here, we show further results for the KLM, which supplement the findings reported there, in particular the behavior as the Kondo effect collapses due to polaron formation. We restrict the discussion here to $J<0$ and fix the phonon frequency to $\omega_0=0.5W$. Calculations were done for $J=-0.5W$ at a filling $n_c=0.8$. The values for $\omega_0$ and $J$ are apparently rather large. However, smaller $J$ and $\omega_0$ do not change the qualitative picture, but make it much harder to visualize the structures.
The results are summarized in Fig. \[fig:DOS\_with\_phonons\] for $g=0$, $g=0.2W$ and $g=0.4W$. The first thing to note is that the phonons lead to a reduction of the width of the pseudo gap close to the Fermi energy and also a reduction of the overall bandwidth. In addition there occur new structures at higher energy with increasing electron-phonon coupling, which are related to the formation of polarons. The reduction of the overall bandwidth is expected and can be interpreted as an increase of the effective mass of the bare conduction states due to the coupling to the phonons. The reduction of the width of the pseudo gap, on the other hand, signals a likewise reduction of the low-energy scale generated by the Kondo effect. Both effects lead to an effective mass as function of $g$ as depicted in the inset to Fig. \[fig:DOS\_with\_phonons\]. Note that $m^\ast$ initially depends only weakly on $g$. However it diverges very strongly as $g\to\omega_0$.
$g=0.2W$\
![Spectral functions for $J=-0.5W$, other parameters as in Fig. \[fig:DOS\_with\_phonons\]. High intensity is represented by red, low by violet.[]{data-label="fig:Aofkw_with_phonons"}](figure6a "fig:"){width="\textwidth"}
$g=0.4W$\
![Spectral functions for $J=-0.5W$, other parameters as in Fig. \[fig:DOS\_with\_phonons\]. High intensity is represented by red, low by violet.[]{data-label="fig:Aofkw_with_phonons"}](figure6b "fig:"){width="\textwidth"}
These feature becomes more apparent by inspecting the spectral functions in Fig. \[fig:Aofkw\_with\_phonons\]. Both effects, the overall reduction of the bare bandwidth and the increased HF mass, are clearly visible. Moreover, with increasing $g$ one finds roughly $\vec{k}$-independent structures representing the polaronic modes. Further increasing $g$, we observe a rather sharp crossover
![DOS (left panel) respectively spectral function (right panel) for $g=0.5W=\omega_0$. Other parameters as in Fig. \[fig:DOS\_with\_phonons\].[]{data-label="fig:Phonons_large_g"}](figure7a){width="\textwidth"}
![DOS (left panel) respectively spectral function (right panel) for $g=0.5W=\omega_0$. Other parameters as in Fig. \[fig:DOS\_with\_phonons\].[]{data-label="fig:Phonons_large_g"}](figure7b){width="\textwidth"}
around $g\approx \omega_0$ to a completely incoherent behavior. The results of a calculation for $g=\omega_0$ are shown in Fig. \[fig:Phonons\_large\_g\]. Note that there is no Kondo feature left either in the DOS or in the spectral function, and all structures are rather broad. We also would like to mention that for $g\gtrsim \omega_0$ it becomes increasingly hard to stabilize a given occupation $n\ne0,1,2$ of the conduction band. This indicates that the system is close to a Peierls instability, i.e. the formation of a charge density wave together with a lattice distortion.
Summary
=======
We have presented a summary of properties of the Kondo lattice model within dynamical mean-field theory at $T=0$ using the numerical renormalization group as impurity solver. We have extended the Kondo lattice model by including an Einstein mode coupled to the electrons via a Holstein term. The importance of such modes for HF materials can be deduced from strong effects such as the Kondo volume collapse observed in Ce.
Without the phonons, we find for antiferromagnetic coupling the expected heavy-fermion behavior, with hybridized bands appearing in the spectrum and a large Fermi surface. For ferromagnetic coupling on the other hand, the bare band structure is only weakly modified due to incoherent scattering from the local degrees of freedom.
Allowing for a magnetically ordered state, we are able to stabilize a Néel order for $J>J_c$, where $J_c<0$ depends on the filling of the conduction band. For the case $n_c=0.9$ discussed here, we find $J_c/W\approx-0.25$. Quite interestingly, there appears to be a qualitative difference between the ground-state for$|J|\to0$ and $|J|\to|J_c|$. In the former case, the results can well be interpreted with a system consisting of local moments which order antiferromagnetically via an RKKY-like exchange. In this case, we have a small Fermi surface and the band states are weakly polarized by the presence of the local moments. For the latter case we find a rather different behavior. First, although the polarization of the local moments is considerably smaller than for small $J$, the band polarization has actually increased pointing towards a much stronger entanglement between local spins and conduction electrons. Second, the spectral functions again form flat bands around points of the magnetic Brillouin zone which map to the heavy fermion bands without Néel order. A more detailed investigation of the crossover respectively transitions could not yet be accomplished due to convergence problems of the DMFT in the interesting region.
Adding phonons, we find a general narrowing of the bare band, which also leads to a reduction of the Kondo scale. Eventually, when the coupling becomes of the order of the phonon frequency, the electrons tend to localize and form polarons with the phonons. At that point, the effective mass diverges and the electronic spectrum becomes incoherent. As a side observation we note that in this region one sees a tendency of the system to form a charge density wave.
The results for $T=0$ presented here strongly motivate further investigations, in particular at finite $T$, searching for the critical temperature and in particular the interplay between HF and local moment physics with respect to magnetism, charge odering and superconductivity.
We want to thank helpful discussions with Andreas Honecker, Akihisa Koga, Achim Rosch, Fakher Assaad and Dieter Vollhardt. RP wants to thank the Japan Society for the Promotion of Science (JSPS) together with the Alexander von Humboldt-Foundation for a postdoctoral fellowship. Computer support was provided by the Gesellschaft für wissenschaftliche Datenverarbeitung in Göttingen and the Norddeutsche Verbund für Hoch- und Höchstleistungsrechnen.
[^1]: Note that “large” in the magnetic Brillouin zone means centered around $\Gamma$, as the $M$ point has been mapped back to the zone center.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Great advances have been made in the study of ultra-high energy cosmic rays (UHECR) in the past two decades. These include the discovery of the spectral cut-off near $5\times10^{19}$eV and complex structure at lower energies, as well as increasingly precise information about the composition of cosmic rays as a function of energy. Important improvements in techniques, including extensive surface detector arrays and high resolution air fluorescence detectors, have been instrumental in facilitating this progress. We discuss the status of the field, including the open questions about the nature of spectral structure, systematic issues related to our understanding of composition, and emerging evidence for anisotropy at the highest energies. We review prospects for upgraded and future observatories including Telescope Array, Pierre Auger and JEM-EUSO and other space-based proposals, and discuss promising new technologies based on radio emission from extensive air showers produced by UHECR.'
author:
- 'B.R. Dawson'
- 'M. Fukushima'
- 'P. Sokolsky'
title: 'Past, Present and Future of UHECR Observations'
---
Introduction
============
Cosmic rays were discovered a little over 100 years ago [@DiscoveryCosmicRays; @KampertWatson2012]. They were called “cosmic” to distinguish them from the then equally mysterious “X rays” emanating from laboratory instruments and particular minerals. The increase in intensity of this cosmic radiation with altitude made it clear that the sources were extraterrestrial. Over the next few decades, the grand questions were developed about their origin, the extent of the energy spectrum, and their composition. It took close to one hundred years to find the end of the remarkable, approximately power law energy spectrum, at energies near 50 Joules per nucleus.
We now know a great deal more, and at lower energies in significant detail (as in the isotopic composition and gamma/electron/positron fluxes). But at the “frontier” energies of $>10^{18}$eV, where the flux is most likely extra-galactic, the experimental tools, while greatly improved, are still too imprecise. While our ability to measure energy with reasonable certitude has improved dramatically, we attempt to measure the cosmic ray composition with what is effectively a blunt instrument. The measurement of arrival directions has also improved markedly, but there, Nature has been unkind and only hints of cosmic ray anisotropy and sources have appeared.
One hundred years ago, the argument was whether the radiation came from the Earth or Space, whether it was composed of charged or neutral particles, and what could be inferred about its energy by measuring “penetrating power”. Now the arguments relate to the nature of observed structures in the energy spectrum - the knees, ankles and final cut-off, whether the composition is protonic, a mixture of p, He and CNO group nuclei, or significantly heavy up to Fe, and how this interplays with the spectral structure. Hints of departure from arrival direction isotropy come and go and we fervently hold on to the most recent observation hoping that this time the significance will strengthen with additional data and a source, or sources, will finally be found. But except for the spectrum and its structures, much of what we argue about is ephemeral and can easily change with modification of hadronic models or a decrease in statistical significance of a source. It is a hard fact that we still do not know with any real certainty the origin of cosmic rays above a few tens of GeV in energy. What is it then that we do know, how well, and what are the implications? This is what will be discussed at length in this volume. Our knowledge of our deficiencies also leads to new ideas for better detectors and new programmatic approaches to fill in needed extrapolations from accelerator data and ancillary measurements. This too will be explored in subsequent pages.
In the present paper we address broadly the experimental status above $10^{18}$eV and briefly describe the current status and the evolution of ultra-high energy (UHE) detection techniques. The pioneering Volcano Ranch, Haverah Park, SUGAR, Yakutsk and Akeno arrays [@NaganoWatson2000; @Linsley1963; @Allan1962; @McCusker1963; @Bell1974; @Afanasiev1993; @Nagano1984] led to the major leaps forward represented by the AGASA, Fly’s Eye, HiRes, Auger and TA experiments [@Nagano1984; @Bergeson1977; @Baltrusaitis1985; @Thomson2004; @Abraham2004; @Sokolsky2007]. The early detectors led us to more precise formulations of the questions we now ask. The decades-long development of these second and third generation detectors culminated in the reliable results with well-understood energy and geometrical resolution that have given us new hope that many of the puzzles presented by UHE cosmic rays may soon be answered. The current generation of detectors has brought definitive confirmation on hints of an ankle structure above $10^{17}$eV [@Abbasi2005] and settled once and for all the reality of a cut-off at energies between 4 and 6 $\times$ $10^{19}$eV [@Abbasi2008; @Abraham2008]. The “holy grail” of UHE cosmic ray physics has been found. It is striking that the suppression is in the same range of energy as the prediction of the Greisen-Zatsepin-Kuzmin (GZK) effect [@greisen; @zatsepin_kuzmin] (depending somewhat on the assumed distribution of sources, composition and the injection spectrum).
Improvements in the determination of extensive air shower (EAS) profiles and the depth of shower maximum, [$X_{\rm max}$]{}, brought about by either using stereo air fluorescence measurements (HiRes and TA) [@Bird1993; @Abbasi2010; @Abbasi2009] or by hybrid surface detector and air fluorescence measurements (Auger and TA) [@AbuZayyad2000b; @AbuZayyad2001; @TA_Composition_Hybrid; @Abraham2010; @Auger_longXmax] have reduced the reconstruction uncertainties in [$X_{\rm max}$]{} to near 10[g/cm$^2$]{} with systematic uncertainties approaching this number. We have reached the point where the experimental measurements are becoming more precise than the theoretical underpinnings of the shower simulations used to extract composition information. While data from the LHC in the forward region at an equivalent energy of $10^{17}$eV are very helpful in tuning the various hadronic models [@Engel2011; @dEnterria2011], there are significant issues in extrapolating to p-nucleus and nucleus - nucleus interactions at much higher energies ($10^{18}$ - $10^{20}$eV) [@dEnterria2011]. Currently the combined systematic uncertainties for data and simulations make it difficult to reliably find the mix of protons, He, CNO and Fe that would match the observed [$X_{\rm max}$]{} distributions. What can be said, and this is a great accomplishment, is that there is very little iron nucleus component [@AugerMassMixtures; @TA_Composition_Hybrid]. Why this should be the case, given the relative stability of the iron nucleus as it travels through the relic photon fields, and its relatively high acceleration efficiency, is a puzzle which we are just now beginning to confront.
Achievements in the Era of Very Large Observatories
===================================================
The previous generation of detectors
------------------------------------
At the beginning of the 21st century, three experiments were studying the highest energy cosmic rays: the Yakutsk array, the Akeno Giant Air Shower Array (AGASA), and the High Resolution Fly’s Eye (HiRes).
The Yakutsk array in Russia had operated in various forms since 1967, and had reached a maximum collecting area of 17km$^2$ around 1990. Subsequently, it was reconfigured to study lower energy cosmic rays, and today it has an area of 8km$^2$. While its focus has changed, analyses are still done on the data from high energy showers already collected e.g. [@Ivanov2015].
AGASA, located 100km west of Tokyo at an average altitude of 667m, operated from 1990 to 2004 as a 100km$^2$ array consisting of over one hundred scintillator detectors inter-connected by optical fibers for timing measurements and data collection [@agasa_nim]. It pioneered many of the techniques employed today in more modern observatories, and produced important results on the UHECR energy spectrum, anisotropy and mass composition [@Shinozaki2004]. Of particular historical interest was the observation by AGASA of the continuation of the cosmic ray spectrum beyond $10^{20}$eV, with no sign of a flux suppression [@Takeda2003].
HiRes was the successor to the first successful air fluorescence detector, the Fly’s Eye [@Baltrusaitis1985; @YoshidaDai1998] which operated from 1981 to 1993 at the Dugway Proving Grounds in Utah, USA. The Fly’s Eye achieved a time-averaged aperture of about 100km$^2$sr at the highest energies, taking into account that it only operated on clear, moonless nights. HiRes was an advancement in resolution and sensitivity, achieved by increasing the telescope effective mirror areas to 3.8m$^2$, and reducing the camera pixel angular diameters to $1^\circ$ [@AbuZayyad2000a]. Two sites, 12.8km apart, were instrumented, allowing for stereo observations of approximately $30-40$% of air showers that triggered either detector near $10^{19}$eV. The collecting area of HiRes was close to an order of magnitude larger than that of the Fly’s Eye. The first HiRes site at Five Mile Hill began full operation in 1997, followed by the Camel’s Back Mountain site in 1999. HiRes ceased operations in 2006. A summary of the important physics results from HiRes is given in [@Sokolsky2007]. This includes the first unambiguous detection of a flux suppression at the highest energies, using monocular data from HiRes I and published in 2008 [@Abbasi2008].
At the beginning of the 1990s discussions began about the next step in UHE cosmic ray observations, where it was recognised that apertures even larger than that of HiRes would be necessary to answer some of the long-standing questions in cosmic ray astrophysics. The experimental challenge was enhanced by the apparent disagreement in the energy spectra presented by AGASA and HiRes in the first few years of the new century. This led to the next generation of experiments adopting hybrid designs, with combinations of surface arrays and air fluorescence detectors.
![The Telescope Array layout, showing the locations of the 507 surface detectors. The three FD stations at Middle Drum, Black Rock Mesa and Long Ridge are indicated, each having an azimuthal field of view of about $110^\circ$. The TALE detector is situated at the MD site [@TA_Spectrum_ICRC15].[]{data-label="fig:TA"}](figures/Layout_of_Pierre_Auger_Observatory.png){width="0.95\linewidth"}
![The Telescope Array layout, showing the locations of the 507 surface detectors. The three FD stations at Middle Drum, Black Rock Mesa and Long Ridge are indicated, each having an azimuthal field of view of about $110^\circ$. The TALE detector is situated at the MD site [@TA_Spectrum_ICRC15].[]{data-label="fig:TA"}](figures/TA_layout_ICRC15.pdf){width="0.95\linewidth"}
Currently operating Observatories
---------------------------------
### The Pierre Auger Observatory {#Sect2:auger}
The Pierre Auger Observatory had its beginnings in 1991 when James Cronin and Alan Watson began discussions with a number of experimenters in the field. Its design evolved through an initial meeting in Paris in 1992, a two week design workshop in Adelaide in 1993, and a six-month design study hosted by Fermilab in 1995. Initial ideas were based on a 5000km$^2$ surface array without fluorescence telescopes, but the advantages of the hybrid approach soon became apparent [@Watson2014; @SommersHybrid; @DawsonHybrid]. The Auger Observatory is near Malargüe, Argentina, and its construction began in 2001 with an engineering array. The Observatory was completed in 2008, though official data taking began in 2004 during construction.
Auger’s surface detector (SD) consists of 1660 water-Cherenkov detectors (WCDs) arranged on a 1.5km triangular grid covering 3000km$^2$ [@AugerNIM], see Figure \[fig:Auger\]. The WCDs are 10m$^2$ in area and 1.2m deep, and build on the experience gained from the Haverah Park detector in the UK (see [@KampertWatson2012]). Such detectors have the advantage of having a broad zenith angle sensitivity, and are deep enough to produce signal from the numerous photons in the extensive air shower. The SD is fully efficient for cosmic ray energies greater than $
3\times 10^{18}$eV and zenith angles less than $60^\circ$. A small 23.5km$^2$ area of the array hosts a denser 750m spacing of WCDs, which is fully efficient for $E > 3 \times 10^{17}$eV and zenith angles less than $55^\circ$.
The main fluorescence detector (FD) consists of 4 sites on the perimeter of the surface array. Each site hosts six telescopes, each with a field of view of $30^\circ$ in azimuth and an elevation range from $1.5^\circ$ to $30^\circ$. Each telescope is of the Schmidt design and consists of a 13m$^2$ segmented spherical mirror with a 2.2m diameter entrance aperture (including a ring of corrector lenses) and a camera composed of 440 photomultiplier pixels, each viewing a $1.5^\circ$ diameter area of sky [@AugerNIM]. The entrance aperture also contains a UV transmitting filter to match the air fluorescence spectrum spanning approximately 300–400nm. The telescope design benefits from experience with the Fly’s Eye and HiRes experiments, with the primary difference being the Schmidt optics design, allowing for a wide field of view with minimal coma aberration. Apart from the 24 telescopes of the main FD described here, an additional three telescopes make up the HEAT (High Elevation Auger Telescopes) system. These view the elevation range between $30^\circ$ and $58^\circ$ to study lower energy air showers (currently down to $10^{17}$eV) that, due to their lower brightness, are observed closer to the FD site [@AugerNIM].
The combination of the surface and fluorescence detectors to make a “hybrid” observatory has been exploited in much of Auger’s scientific output. The SD has many strengths, including robust WCDs that operate with a 100% duty cycle. It also measures the lateral characteristics of the air shower, albeit at one altitude, which are being used for several studies including mass composition. The FD, while having the disadvantage of a 15% duty cycle [@AugerNIM], measures fluorescence light which is produced in direct proportion to the energy deposited by the air shower. Thus, the fluorescence technique measures air shower energies calorimetrically, and it is the FD measurements of energy that calibrate the SD energy scale, as described below. The FD views the developing shower and has access to the depth of shower maximum, $X_{\rm max}$, used in mass composition studies. Finally, the FD reconstruction of the direction and position of the air shower axis is greatly assisted by the SD measurement of the shower arrival time at ground level [@DawsonHybrid]. This hybrid reconstruction produces an FD arrival direction resolution of about $0.5^\circ$ [@Bonifazi2009], which helps achieve typical resolutions in energy and $X_{\rm max}$ of $\sim10$% and 20g/cm$^2$, respectively, at $10^{19}$eV [@Dawson2007].
Descriptions of the calibration procedures for both the SD and FD are given in [@AugerNIM]. In both cases the calibration is “end-to-end”, either using unaccompanied muons (in the case of the SD) or a large “drum” to illuminate an FD aperture, so as to calibrate the full detector and data acquisition chain in one step. The atmosphere is also carefully monitored. The density of the lower atmosphere has a well known effect on the lateral distribution of the air shower at ground level, and these weather corrections are applied to the SD energy measurements for certain studies, such as large scale anisotropy measurements [@Auger_weather2016]. Finally, the light attenuation characteristics of the atmosphere are measured on an hourly basis during FD operations to account for varying molecular and aerosol scattering, and to monitor cloud cover [@AugerAtmosphere2010].
The Auger Collaboration had always planned to build a northern array in order to achieve full sky coverage. The site was chosen to be in south-eastern Colorado, USA [@AugerNorth]. Currently there is a strong focus on an upgrade of the southern site [@AugerPrime] (discussed in Section 4), and exploration of the northern sky is being undertaken by the Telescope Array collaboration.
### The Telescope Array
The Telescope Array (TA) project was originally proposed around 1997 by members of the AGASA and HiRes experiments as a large fluorescence telescope complex with an effective aperture (after accounting for a 10% duty factor) of 5,000km$^2$ sr for 10$^{20}$ eV cosmic ray particles [@ta_durban_1; @ta_durban_2; @ta_dr_2000]. An apparent discrepancy in the measurements of the UHECR flux above the GZK energy by AGASA and HiRes (see [@Takeda2003; @Abbasi2008]), however, encouraged the members of TA to make a critical examination of the experimental methods used. This led to the present form of TA, started in 2003, as two complete and co-sited SD and FD detectors, each observing the same UHECR events and allowing for a critical comparison of the measured shower parameters [@ta_tsukuba]. The TA experiment occupies a large area near the town of Delta, 200km south-west of Salt Lake City, Utah, USA, and is now operated by a collaboration of 120 members from five countries: Japan, USA, Korea, Russia and Belgium. The experiment is conducting a high-statistics exploration of the northern sky as a hybrid array of surface and fluorescence detectors.
The TA surface detector comprises 507 detectors on a 1.2km square grid covering an area of 700km$^2$ (Figure \[fig:TA\]). Each detector has an area of 3m$^2$ and consists of two layers of 1.2cm plastic scintillator separated by a 1mm thickness of stainless steel [@TA_SD]. The scintillators are equally sensitive to all minimum ionising charged particles ($e^\pm$, $\mu^\pm$), with the SD energy determination being dominated by the EM (electromagnetic) component. This is seen as an advantage as it reduces uncertainties in the energy scale due to mass composition or hadronic physics. The array reaches full efficiency above $10^{19}$eV for zenith angles less than $45^\circ$, providing an aperture of 1100km$^2$sr.
Three fluorescence detector sites sit near the boundary of the array. One of the sites, at Middle Drum, uses 14 refurbished telescopes originally part of the HiRes detector. They are arranged in two “rings” that together view an elevation range from $3^\circ$ to $31^\circ$, and an azimuth range of $112^\circ$. Each telescope consists of a 2m diameter spherical mirror and a camera of 256 hexagonal pixels, with pixels viewing approximately a $1^\circ$ diameter section of the sky [@AbuZayyad2000a].
The other two FD sites at Black Rock Mesa (BR) and Long Ridge (LR) each contains 12 newly fabricated telescopes. Each telescope consists of a 3.3m diameter segmented spherical mirror focusing light onto a 256 pixel camera. The pixels also have a field of view of $1^\circ$ diameter and each site covers a field of view of $3^\circ - 33^\circ$ in elevation and $108^\circ$ in azimuth using a two ring structure [@TA_newFD]. The electronics in the new FD stations digitise the pixel signals at 10MHz with 14 bits of precision [@TA_FD_electronics].
The TA detectors began full operation in March 2008. Various data sets are being collected, and the consistency between different data sets and different analysis procedures has been carefully examined. Surface detector energy spectrum studies use contained events within the array with zenith angles $< 45^\circ$, while anisotropy studies use looser cuts and zenith angles $< 55^\circ$ [@Tinyakov2014]. For fluorescence analysis, some studies are done with mono observations (requiring observations from one FD site), stereo observations (two FD sites), and hybrid observations. In the latter case, FD observations are coupled with SD measurements of the shower at ground level, much in the style of Auger analysis, except that timing information from more than one SD station is used, and the shower core location derived from SD data alone is used to constrain the hybrid core [@TA_Composition_Hybrid].
The atmospheric transparency of the TA site is monitored by a suite of instruments, including a Central Laser Facility, IR cloud cameras and a LIDAR station [@tomida_2011; @Tomida2013]. In analyses performed to this point, the aerosol content of the atmosphere has been assumed to be constant with time. Atmospheric transparency data has been used to determine an average value of the vertical aerosol optical depth of 0.04 (see e.g. [@TA_Composition_Hybrid]), and the effect on systematic uncertainties of fluctuations about the mean has been studied.
The TA project has in recent years extended its reach towards lower energy cosmic rays with the TALE (TA Low Energy) extension. With an additional 10 telescopes in an extra two “rings” at the Middle Drum site, the field of view there is now $3 - 59^\circ$ in elevation and approximately $120^\circ$ in azimuth. By using Cherenkov-rich events, the energy threshold of TALE is below $10^{16}$eV, complementary to the standard fluorescence observations [@Zundel_ICRC15]. The surface detector array is being increased in density in front of this FD site to assist with lower energy hybrid observations.
Advances in techniques {#Sect2:advances}
----------------------
The advances in our understanding of UHECR that we will discuss in the following sections owe much to the large collecting areas now instrumented by the Auger and TA collaborations. Additionally, the number of scientists now studying UHECR is much larger than in early generations of experiments, meaning that more manpower is available for the maintenance of the experiments, for calibration of the detectors and the atmosphere, and for devising new and creative analysis techniques and cross-checks. Thus the increase in sensitivity of the modern observatories is due to more than an increase in the collecting area alone. We give some examples here of recently exploited advances in detectors, tools and techniques.
A stable surface detector is necessary for optimal energy resolution, and for searching for weak broad-scale anisotropies. For example, temperature-dependent particle density measurements can introduce diurnal variations into shower rates above some energy threshold, which may be wrongly interpreted as sidereal harmonics. Both Auger and TA avoid this by monitoring their SD detector performance on short time scales. This is done by collecting histograms of the fundamental unit used to calibrate SD signals - in the case of TA scintillator detectors, a histogram of integrated charges from $\sim$0.4 million penetrating particles is collected every 10 minutes for each detector [@TA_SD], while for Auger WCDs, histograms of signals from through-going muons are collected every minute, from which the signal due to a “vertical-equivalent muon” (VEM) can be derived [@AugerNIM]. In addition to this basic calibration, Auger also makes a correction (in broad-scale anisotropy studies) for the effect of diurnal atmospheric variations on air-shower development. These weather effects make small but significant corrections to the SD energy estimator $S(1000)$ (the detector signal 1000m from the shower core) using the local air density and pressure [@Auger_weather2016].
In fluorescence light detection, both collaborations have benefitted from experience with the HiRes and Fly’s Eye experiments. The electronics in the new FDs in both TA [@TA_newFD] and Auger [@AugerNIM] include sophisticated triggering circuitry, and digitisation of each pixel signal is performed at 10MHz. Improved telescope design is a feature in both experiments, with Auger using a Schmidt optics design which gives a coma-free optical spot over a $30^\circ$ field of view [@AugerNIM]. At TA, signals from UV-bright stars are used to verify ray-tracing estimations of optical aberrations, and to check telescope alignment [@TA_newFD]. Calibrations of the FDs feature end-to-end procedures. At Auger, a “drum” calibrating system is moved from telescope to telescope periodically. It uniformly illuminates the aperture of a telescope with an absolutely calibrated light source at a number of wavelengths. Between these absolute drum calibrations, the system calibration is monitored with light sources illuminating the mirror and camera [@AugerNIM]. At TA, a small number of absolutely-calibrated PMTs are in each camera [@kawana_crays], and their gains are monitored with a radioactive source-scintillator YAP unit [@shin_yap]. The other PMTs in the camera are cross-calibrated and monitored every 30 minutes during observation time using a diffuse Xenon light source installed in front of each camera [@TA_FDcalib]. The Telescope Array is going one step further, experimenting at the Black Rock FD station with an electron light source (ELS), which shoots a vertical beam of 40MeV electrons 100m from a FD telescope, creating artificial air fluorescence in-situ [@shibata_els]. The ELS aims to make an end-to-end calibration of the FD - from energy deposits in the air to the detection of fluorescence light by the telescope [@TA_ELS_ICRC15].
Since both experiments rely on fluorescence measurements to calibrate the SD energy scales, much effort has recently gone into laboratory measurements of the fluorescence efficiency, that fraction of the shower’s ionisation energy deposit going into light production. After the pioneering work of Bunner, Kakimoto et al. and Nagano et al. [@Bunner; @Kakimoto; @Nagano], new measurements include those of the AIRFLY [@AIRFLY] and FLASH [@FLASH] experiments. Results include precise measurements of the fluorescence efficiency and spectrum, and the pressure, temperature and humidity dependence of the light [@AIRFLY]. Auger, which uses the AIRFLY results, has been able to reduce the systematic uncertainty in shower energy associated with the fluorescence yield from 14% to 3.6% [@Auger_EScale_ICRC13]. TA has used the Kakimoto et al. [@Kakimoto] fluorescence yield model, also used by HiRes, for the sake of consistency and continuity. It is now re-examining its energy scale using more contemporary measurements.
Air shower simulations are used in a variety of applications in both experiments, including the extraction of mass composition estimates from air shower development measurements. The CORSIKA three dimensional shower simulation program [@CORSIKA] is still the most widely used, with other code such as the longitudinal profile simulator CONEX [@CONEX] being used in certain applications. The continuing challenge is to improve the implementation of high energy hadronic interactions in the code. In the last decade important new constraints on these interactions have come from measurements at the Large Hadron Collider, spawning new hadronic models for CORSIKA and CONEX, including EPOS-LHC [@EPOS], QGSJetII-04 [@QGSJet] and Sibyll 2.3 [@Sibyll]. Information is also moving in the other direction, with cosmic ray experiments like Auger and TA testing certain aspects of these predictions (see Section \[Sect2:interactions\]).
The energy spectrum of UHECR {#Sect2:energy}
----------------------------
The surface detectors of Auger and TA are the “work-horses” of the respective experiments, since they are operational 24 hours per day and contribute the majority of exposure to studies such as the energy spectrum. However, both experiments use their fluorescence detectors to determine the energy scale of their surface detectors, avoiding much of the uncertainty associated with the alternative, air shower simulations. Systematic uncertainties in the energy scale have been derived taking account of uncertainties in detector calibration and stability, atmospheric transmission, fluorescence yield, and reconstruction. For Auger this amounts to a total systematic uncertainty of 14% [@Auger_EScale_ICRC13], and for TA 21% [@TA_hybrid_spectrum2015]. Both quoted uncertainties are independent of energy.
The most recent energy spectra were presented in 2015 by Auger [@Auger_Spectrum_ICRC15] and TA [@TA_Spectrum_ICRC15]. The Telescope Array presented a spectrum over 4.5 decades of energy from below $10^{16}$eV, combining results from the SD, the two Japanese FD sites (monocular reconstruction) and TALE (Figure \[fig:TASpectrum\]). At $10^{20}$eV the exposure is approximately 6200km$^2$sryr for the SD and 650km$^2$sryr for the FDs. A remarkable set of features is present in the combined spectrum, represented by a series of five power-law segments. The flux suppression at the highest energies, identified by TA as consistent with the GZK mechanism [@greisen; @zatsepin_kuzmin], is present above $10^{19.80\pm0.05}$eV ($6.3\times 10^{19}$eV), and the spectral ankle appears at $10^{18.72\pm0.02}$eV ($5.2\times
10^{18}$eV), again consistent with proton interactions with the cosmic microwave background (CMB) [@SpectralDip]. At lower energies, TALE detects two other features, a second knee at $10^{17.30\pm0.05}$eV ($2.0\times 10^{17}$eV), and a low energy ankle at $10^{16.34\pm0.04}$eV ($2.2\times 10^{16}$eV). The systematic uncertainties of the TALE measurements are currently being evaluated.
![The Auger combined energy spectrum with data from the 750m spaced SD, FD (hybrid) and the 1500m SD array. The energy systematic uncertainty is 14%. Event numbers are shown, and a spectrum model is fitted [@Auger_Spectrum_ICRC15].[]{data-label="fig:AugerSpectrum"}](figures/TASpectrum2015.pdf){width="1.0\linewidth"}
![The Auger combined energy spectrum with data from the 750m spaced SD, FD (hybrid) and the 1500m SD array. The energy systematic uncertainty is 14%. Event numbers are shown, and a spectrum model is fitted [@Auger_Spectrum_ICRC15].[]{data-label="fig:AugerSpectrum"}](figures/AugerSpectrum2015.pdf){width="1.0\linewidth"}
The latest Auger spectrum [@Auger_Spectrum_ICRC15] (Figure \[fig:AugerSpectrum\]) is a combined measurement from the 1500m spaced SD (zenith angle $\theta < 80^\circ$), the FDs operating in hybrid mode, and the 750m spaced SD ($\theta <
55^\circ$). Two separate analyses were performed for the 1500m SD array, for “vertical” ($\theta < 60^\circ$) and for “inclined” ($60^\circ < \theta < 80^\circ$) showers, with the latter events being muon dominated. The integrated exposures for the 1500m vertical SD, inclined SD and 750m SD are 42,500km$^2$sryr, 10,900km$^2$sryr and 150km$^2$sryr, respectively. The energy-dependent exposure for the hybrid spectrum is 1500km$^2$sryr at $10^{19}$eV. The combined spectrum extends from $10^{17.5}$eV to the highest energies, and shows the ankle feature at $(4.82 \pm 0.07 \pm 0.8) \times 10^{18}$eV (statistical and systematic uncertainties are quoted). The flux suppression is characterised by a smooth function with $E_s = (4.21 \pm 0.17 \pm
0.76)\times 10^{19}$eV, with $E_s$ representing the energy at which the flux falls to one-half of the value of the power-law extrapolation.
When comparing the Auger and TA spectra, the following points have been made by a joint Auger/TA energy spectrum working group [@UHECR2016_spectrum]:
- In the overlapping region of energy, the spectral slopes are consistent within uncertainties, and the energy of the “ankle” is consistent given the statistical and systematic uncertainties. A flux difference at energies from $10^{17.5} - 10^{19.3}$eV of $\sim
20\%$ could be the result of a shift in the energy scale within systematic uncertainties.
- On the other hand, the energy of the flux suppression at the highest energies, characterised by $E_{1/2}$ (a measure of the suppression energy favoured by TA [@Ehalf]) is inconsistent. The TA measurement is $(6.0 \pm 0.7 {\rm (stat)}) \times 10^{19}$eV, compared with the Auger measurement of $(2.47 \pm 0.01 {\rm
(stat)} ^{+ 0.82}_{- 0.34} {\rm (syst)}) \times 10^{19}$eV.
- The agreement in the position of the ankle and the disagreement in the suppression energy might be explained by an energy-dependent systematic uncertainty in energy, or a real difference in the physics of cosmic rays in the northern and southern hemispheres. At the current time, no source of the former has been identified and, as mentioned above, both experiments quote an energy independent systematic. As an example, differences in the correction for [*invisible energy*]{} used by Auger and TA in the FD analyses (i.e. that energy carried by high energy muons and neutrinos that does not result in proportionate fluorescence light), and differences in fluorescence yield models, produce only a small shift in the energy scale of $5-10$% which is essentially energy independent.
- The possibility that the UHECR sky is different in the northern and southern hemispheres has been studied by determining the energy spectrum as a function of declination. The Auger SD “vertical” ($\theta < 60^\circ$) spectrum covers a declination range from $-90^\circ$ to $+25^\circ$, 71% of the total sky. Four energy spectra have been derived for independent declination bands [@Auger_Spectrum_ICRC15], which are then compared with the total Auger spectrum. A small, and statistically insignificant declination dependence in the flux is observed ($< 5$% below the suppression energy $E_s$ and $<13$% above) within the declinations studied. The conclusion is that the Auger/TA spectrum difference in the suppression region cannot be explained in terms of a declination dependence, unless there is a significant change in the spectrum north of $25^\circ$. The TA collaboration see a hint of such an effect when considering an SD energy spectrum extended to include zenith angles $< 55^\circ$, covering a range of declinations from $-16^\circ$ to $+90^\circ$. The position of the suppression energy $E_{1/2}$ is approximately $3\sigma$ higher for declinations north of $26^\circ$ compared with those south of $26^\circ$ [@Verzi_PTEP]. The question of a declination dependent spectrum is connected to the observations of anisotropies of the flux discussed below in Section \[Sect2:anisotropy\].
Currently the source of the disagreement in the energy spectrum at the highest energies is an open question. Any declination-dependence of the spectrum appears weak, but more studies are on-going. In parallel, the Auger and TA groups are working together to understand differences in the analysis procedures, and how these might lead to an experimental explanation for the spectral differences.
The methods used by the collaborations for measuring the energy spectrum have many things in common, most importantly in the use of fluorescence measurements to set the energy scale. But there are significant differences in other areas, either necessitated by detector differences (scintillators vs. water-Cherenkov detectors), or because of the philosophy of the collaborations. A case in point is how each experiment accounts for the zenith angle-dependent attenuation of showers for SD measurements. The Auger collaboration has a philosophy of avoiding the use of air shower simulations, wherever possible, in deriving energy estimates. To account for shower attenuation, Auger uses the method of constant intensity cuts (CIC), a data-driven procedure which uses the fact that the intensity of cosmic rays above a certain energy threshold should be independent of the zenith angle of the showers [@AugerNIM]. The analysis converts Auger’s SD energy estimator $S(1000)$ (the WCD signal 1000m from the shower core) to $S_{38}$, the value of $S(1000)$ the shower would have possessed if it had arrived at the median zenith angle of $38^\circ$. While, in principle, the conversion from $S(1000)$ to $S_{38}$ could be energy-dependent, no dependence has been detected. In contrast, the TA analysis uses simulations of proton showers to account for shower attenuation in their analysis. Both methods are valid, but both have possible weaknesses, which can be explored in future studies under the joint working group structure created by the two collaborations [@UHECR2016_spectrum].
The astrophysical interpretation of the energy spectrum is, of course, coupled to other measurements made by the collaborations, in particular the mass composition. The TA collaboration finds that features of its spectrum can be satisfactorily explained by models of production and propagation of a pure protonic cosmic ray flux [@TA_ICRC15_spectrum_interpretation]. Here, the ankle is interpreted as a “dip” caused by pair production on the CMB and IR photons, and the suppression is due to the classic Greisen-Zatsepin-Kuzmin photopion production on the same photon fields. Propagation simulations for both a uniform distribution of proton sources, and a distribution which follows the local large scale structure of the Universe, are compared with the measured spectrum to fit the power-law index $\gamma$ of the spectrum at the sources, a parameter $m$ related to the source evolution with redshift, and a logarithmic shift of the experimental energy scale $\Delta \log E$. Good fits were obtained for both source distributions. For the uniform distribution, the $\chi^2/d.o.f.$ was 12.4/17 with $\gamma =
2.21^{+0.10}_{-0.15}$, $m = 6.7^{+1.7}_{-1.4}$, and $\Delta
\log E =0.03\pm0.03$ [@TA_ICRC15_spectrum_interpretation]. The assumption of the pure proton flux is consistent with TA’s measurements of mass composition (see Section \[Sect2:mass\]). The best value of energy shift ($\sim 3$%) is well within the systematic energy uncertainty of TA. The parameters $\gamma$ and $m$ apply only for sources with $z < 0.7$, since the contribution of protons arriving from sources beyond that redshift is negligible for $E >
10^{18.2}$eV.
Such a mass composition is not favoured by Auger’s measurements. In this case it is necessary to use source production and propagation modelling to fit both the energy spectrum, and the mass composition measured at Earth. This has been done by several authors, including the Auger collaboration [@Auger_ICRC15_spectrum_interpretation]. The input to the simulation is a population of uniformly distributed sources accelerating protons, and nuclei of He, N and Fe, to a maximum rigidity with a power-law spectrum of index $\gamma$. The standard interactions of protons and nuclei with background photons are taken into account. The result of this simple model is a rather hard input spectrum ($\gamma \sim 1$) with a rather low maximum rigidity of the source accelerators. While the authors point out the naivety of the model, the results are in real contrast to the protonic model favoured by the TA collaboration, and thus stress the importance of the mass composition assumptions when interpreting the energy spectrum.
Arrival direction studies {#Sect2:anisotropy}
-------------------------
As described in Section \[Sect2:advances\], improvements in detector size, design and operations have led to major advances in sensitivity for anisotropy studies. In parallel, new Faraday rotation studies have improved our understanding of cosmic magnetic fields, particularly those within our Galaxy and its halo [@Pshirkov; @Jansson]. While the new observatories have not uncovered the strong anisotropies that had been predicted by some, a number of interesting results have ruled out several scenarios for UHECR sources and propagation.
### Broad-scale anisotropy searches
Broad-scale anisotropies are often searched for using a harmonic analysis in right ascension (RA), though increasingly more sophisticated multipole analyses are undertaken. Results are challenging to interpret, as they depend not only on distribution of sources, but also on the distribution (including turbulence) of the galactic and extragalactic magnetic fields, and the magnetic rigidity of the cosmic rays.
One case study is the analysis by the Auger collaboration using both the 750m and 1500m SD arrays, covering energies from around $2\times 10^{16}$eV to the highest energies [@SamaraiICRC15]. For most of the reported energy range, the amplitude of the first harmonic in RA is not significant, but the phase of the harmonic shows an interesting energy dependence, changing from roughly the Galactic Center direction at low energies to a direction almost $180^\circ$ away at the highest energies. Linsley pointed out many years ago that the phase information may have some validity even for anisotropy amplitudes that are not significant (see [@Linsley]). Two energy bins have amplitudes approaching acceptable significance - the bin from 1 to $2\times 10^{18}$eV and, especially, the bin for energies above $8\times 10^{18}$eV (Figure \[fig:AugerAnisotropy\]). The latter amplitude in RA of 4.4% has a chance probability of $6.4\times10^{-5}$ [@AugerAnisot_Broad2015]. Expressed as a dipole amplitude and direction, the excess is $7.3\pm1.5$% (approaching $5\sigma$) in a direction ${\rm (RA,
dec)}=(95^\circ\pm13^\circ , -39^\circ\pm13^\circ)$.
![The TA “hotspot” in 2014 in equatorial coordinates. Nineteen events are observed above $5.7\times10^{19}$eV within a 20$^\circ$ radius area of sky when 4.49 are expected, giving a post-trial significance of 3.4$\sigma$ [@TA_Hotspot2014]. (Colour scale represents $\sigma$).[]{data-label="fig:TAHotspot"}](figures/Auger_anisotropy.pdf){width="1.0\linewidth"}
![The TA “hotspot” in 2014 in equatorial coordinates. Nineteen events are observed above $5.7\times10^{19}$eV within a 20$^\circ$ radius area of sky when 4.49 are expected, giving a post-trial significance of 3.4$\sigma$ [@TA_Hotspot2014]. (Colour scale represents $\sigma$).[]{data-label="fig:TAHotspot"}](figures/TA_hotspot.pdf){width="1.0\linewidth"}
This apparent transition of the phase of the anisotropy from the galactic center direction to the opposite direction coincides in energy with the ankle of the spectrum, an energy range often seen as the transition between galactic and extragalactic sources (e.g. [@Aloisio2012]). It is also an energy where both Auger and TA find that protons seem to dominate the flux (see Section \[Sect2:mass\]). The [*dominance*]{} of protons of *galactic* origin around $10^{18}$eV is excluded by the low limits on the amplitude of the anisotropy as measured by both Auger and TA [@AugerAnisot_Broad2012; @TA_Anisot2017] with TA concluding that less than 1.3% (95% CL) of cosmic rays with energies between $10^{18}$ and $3\times10^{18}$eV are galactic protons (given certain assumptions about the galactic and halo magnetic fields, and assuming an isotropic extragalactic flux). If, in this region, an extragalactic flux is taking over from a galactic origin, the low level of anisotropy could be explained by the flux being the sum of two fluxes with first-harmonic phases almost $180^\circ$ apart.
The Auger and TA collaborations have combined data to examine broad-scale anisotropies above $10^{19}$eV with a full-sky coverage [@AugerTA_anis1; @AugerTA_anis2]. Only in such a full sky analysis can a true dipole moment be measured unambiguously, and higher moments searched for confidently. With the current statistics, a dipole moment of amplitude $6.5\pm1.9$% is seen with a chance probability of 0.5% and a direction consistent with the Auger-only result above $8\times10^{18}$eV described above. Future joint analyses are awaited with interest.
### Small and medium-scale anisotropy searches
At the highest energies, source distances are likely to be closer than 100Mpc because of energy loss interactions of cosmic rays (of all masses) on various photon fields (e.g. [@Allard]). Then, if magnetic deflections are not too extreme, the arrival direction distribution will mirror the distribution of sources in the local Universe. Both collaborations have searched for event clustering, and for cross-correlations with various astronomical catalogs over a range of angular scales and above a number of energy thresholds.
The most recent Auger data-set (including inclined events out to a zenith angle of $80^\circ$) has been used for searches with energy thresholds between $4\times10^{19}$eV and $8\times10^{19}$eV [@AugerAnisot_small2015]. Self clustering, and clustering around the galactic plane, the galactic center and the super-galactic plane have been tested. In addition, cross correlation analyses have been performed with catalogs of extragalactic objects. No significant anisotropies were found. Of the studies done, the two with the smallest post-trial probabilities (both 1.4% as it happens) were a correlation of cosmic ray arrival directions ($E > 5.8\times
10^{19}$eV) with directions of active galaxies in the Swift-BAT X-ray catalog closer than 130Mpc and with luminosities greater than $10^{44}$erg/s, using a $18^\circ$ search radius; and a clustering of cosmic rays above the same energy threshold within a $15^\circ$ radius of our closest active galaxy, Centaurus A.
The TA collaboration have done similar searches [@TA_AnisotSmall1; @TA_AnisotSmall2] with similar null results. However, an excess on a medium angular scale of $20^\circ$ radius has been detected above $5.7\times10^{19}$eV in the direction ${\rm
(RA,dec)}=(146.7^\circ , 43.2^\circ)$. With five years of TA data, the “hotspot” contained 19 events when the background expectation was 4.49 [@TA_Hotspot2014]. After accounting for trials, including the choice of angular scale, the significance of the excess is 3.4$\sigma$ (Figure \[fig:TAHotspot\]). There is no obvious source or galaxy cluster in this direction, though the excess may be associated with large scale structure, its center being $19^\circ$ away from the supergalactic plane. An update of the result with an additional two years of exposure showed a total of 24 events when the background expectation was 6.88 [@TA_Tinyakov_Kyoto]. This represents a 3.4$\sigma$ post-trial significance, no change since the original result. The future evolution of this analysis will be followed with interest.
The lack of strong statistical evidence for small-scale anisotropies at the highest energies starts to put constraints on the source characteristics, but those constraints are tightly coupled to the mass (charge) of the particles and the magnetic fields. If the UHECR were proton dominated, and if extragalactic magnetic fields are generally at the nano-gauss scale, we would need to conclude that there was a high source density within the 100Mpc horizon (e.g. [@SourceDensity]). On the other hand, if most of the UHECR have medium to high charge, the lack of strong anisotropy could be blamed on magnetic deflections. This emphasises the importance of the next topic in our discussion.
Interpretations of mass composition from air shower measurements {#Sect2:mass}
----------------------------------------------------------------
Unfortunately our access to information on the mass of UHECR is rather indirect, through observations of the extensive air showers they initiate. We must rely on models of hadronic interactions at extreme energies to interpret these observations in terms of the mass of the cosmic ray. As described in Section \[Sect2:advances\], hadronic models have improved in recent years with the availability of measurements from the LHC at centre of mass energies of up to 13TeV. However, given that this corresponds to a fixed target energy of around $10^{17}$eV, laboratory measurements still fall short of the energies involved in UHECR interactions.
With this large caveat in mind, we can attempt to transform measurements of air shower development into estimates of primary mass.
### Fluorescence detector measurements of shower depth of maximum
For many years the depth of shower maximum [$X_{\rm max}$]{} has been the prime measurement for this purpose, first in Cherenkov light experiments around the knee of the energy spectrum in the 1970s, and more recently in fluorescence detector measurements at the highest energies. [$X_{\rm max}$]{} is the slant depth in the atmosphere (in [g/cm$^2$]{}) at which the air shower reaches its maximum size (number of particles) or, near equivalently, at which the shower reaches the maximum of its energy deposit, $dE/dX$. From simple arguments it can be shown that the depth of maximum increases with the logarithm of the primary energy for a fixed primary mass, and with the logarithm of the primary mass number, $A$, at fixed energy (e.g. [@Matthews]).
Unprecedented resolution in [$X_{\rm max}$]{} is now possible with the fluorescence technique, particularly due to reliable reconstruction of the shower axis with the hybrid or stereo techniques, and partly due to finer pixelisation and digitisation in FD cameras. Statistical resolution can be better than 20[g/cm$^2$]{} above $10^{19}$eV [@Auger_longXmax; @Hanlon_UHECR2016], with measurement systematics below 10[g/cm$^2$]{} for Auger [@Auger_longXmax] and somewhat higher for TA [@TA_Composition_Hybrid]. Now it is possible to confidently quote not only mean values of [$X_{\rm max}$]{} as a function of energy, but also the width (RMS or $\sigma$) of the distribution in some energy range.
In the interpretation of [$X_{\rm max}$]{} measurements, one needs to be aware of any biases imposed by the detection or reconstruction processes. A simple example of detection bias would be a bias against the detection of showers with very deep [$X_{\rm max}$]{} (say 900[g/cm$^2$]{}), since vertical showers of this type would have their maxima very close to, or below, ground level. The Auger and TA collaborations have approached the detection bias issue in quite different ways, both valid. The Auger approach [@Auger_longXmax] is to apply strict cuts on the axis geometry of air showers to avoid bias in the detection of both shallow and deep showers. Despite the cost of lower statistics, this allows evaluation of the energy dependence of the “true” (free of detector bias) [$X_{\rm max}$]{} distributions, which can then be compared with theoretical predictions for various mass groups. When computing the RMS of the [$X_{\rm max}$]{} distributions, the experimental resolution is subtracted in quadrature, and care is taken (with more than one method) to account for possible undersampling of the tails of the distributions [@Auger_longXmax].
The alternate philosophy, practised by TA [@TA_Xmax2015] and inherited from the Fly’s Eye and HiRes approaches, is to only apply cuts based on data quality, not potential bias. The theoretical expectation for a particular mass group is then derived using simulations of the detection and reconstruction processes, so that any biases and resolution effects are also present in the expectation. This procedure maximises the event statistics for analysis. However, the results are not [*easily*]{} comparable with measurements from other detectors.
![Auger results on the mean [$X_{\rm max}$]{} (left) and its RMS (right), compared with expectations for protons and iron using the EPOS-LHC, QGSJetII-04 and Sibyll 2.1 hadronic models. Statistical and systematic error bars are indicated [@AugerMass_ICRC15]. []{data-label="fig:AugerMass"}](figures/Auger_MassICRC15.pdf){width="0.95\linewidth"}
The Auger collaboration has presented results on both the mean and RMS of [$X_{\rm max}$]{} ($\langle X_{\rm max}\rangle$ and $\sigma(X_{\rm max})$) from $10^{17}$ to $10^{19.6}$eV [@AugerMass_ICRC15], as shown in Figure \[fig:AugerMass\]. The reduction in the lower energy limit below the previous value of $10^{17.8}$eV [@Auger_longXmax] is due to the inclusion of data from the HEAT FD enhancement (see Section \[Sect2:auger\]). The number of events in the latest analysis is 23872, including 7142 events above $10^{18.2}$eV. The results can be summarised as follows,
- The rate of change of $\langle X_{\rm max}\rangle$ per decade of energy (known as the elongation rate) is not consistent at any energy with that expected of an unchanging mass composition, namely about 60[g/cm$^2$]{} per decade. Below $10^{18.3}$eV the elongation rate is 85[g/cm$^2$]{} per decade, while above that energy it becomes much flatter at approximately 26[g/cm$^2$]{} per decade. This is interpreted as the average mass of cosmic rays decreasing with energy up to the break-point, and then increasing again up to the highest energies. (The lower energy elongation rate is compatible with the measurement by the HiRes/MIA experiments in the same energy range [@AbuZayyad2000b]).
- The behaviour of $\sigma(X_{\rm max})$ is broadly consistent with the behaviour of the mean value. Up to $10^{18.3}$eV the spread of [$X_{\rm max}$]{} is roughly constant (plausible even if the mean mass is decreasing, since a significant proton component appears to remain throughout this energy range), after which the spread appears to decrease with energy.
- Using two post-LHC hadronic models, EPOS-LHC and QGSJetII-04, the experimental data are expressed in terms of $\langle \ln A
\rangle$ and $\sigma^2(\ln A)$, where $A$ is the mass number of a cosmic ray nucleus [@Auger_longXmax]. With both models the mean value of $A$ is similar at the lowest energies and at the highest energies explored, while reducing to a minimum at around $10^{18.3}$eV. The EPOS-LHC model interprets the data with slightly heavier mean $A$ at all energies, compared with QGSJetII-04. At the higher energies $\sigma^2(\ln A)$ approaches zero for the EPOS-LHC model (implying a single type of nucleus) and becomes unphysically negative for the QGSJetII-04 model.
- [$X_{\rm max}$]{} distributions for energy bins from $10^{17.8}$eV to the highest energies have been fitted with model expectations for mixtures of protons with nuclei of helium, nitrogen and iron [@AugerMassMixtures]. With the current models, a simple mixture of protons and iron is not a good fit at any energy, but acceptable fits are obtained when intermediate masses are introduced. For all models there is a significant reduction in the proton fraction with increasing energy above $10^{18.3}$eV, and no model requires any significant fraction of iron at any energy. However, the intermediate masses concluded to be present at any energy have a strong model dependence.
Despite the interpretational problems associated with hadronic physics models, the Auger results show a clear structure in the evolution with energy of both $\langle X_{\rm max}\rangle$ and $\sigma(X_{\rm max})$. The data do not appear consistent with a mass composition unchanging with energy.
![Telescope Array observed mean [$X_{\rm max}$]{} results from seven years of hybrid data from the BR/LR fluorescence detectors (preliminary data). Data are compared with the expectations for protons and iron from the QGSJetII-03 and QGSJetII-04 hadronic models. A systematic uncertainty of 20[g/cm$^2$]{} is indicated by the shaded region [@Hanlon_UHECR2016].[]{data-label="fig:TAMass"}](figures/TA_BLLRMass.pdf){width="0.48\linewidth"}
In the past three years the TA collaboration have discussed results of three [$X_{\rm max}$]{} analyses, all using the previously discussed philosophy of maximising statistics by applying only data quality cuts. Detection biases are accounted for by comparing real data with simulations having the same biases. Those analyses are a hybrid study of data from the Middle Drum (MD) FD detector using five years [@TA_Composition_Hybrid] and seven years [@Hanlon_UHECR2016] of exposure; a study of data from all three FDs using “stereo” geometrical reconstruction [@TA_stereoXmaxICRC15]; and a recent study of hybrid-reconstructed showers viewed by the Black Rock Mesa and Long Ridge (BR/LR) fluorescence detectors over seven years [@Hanlon_UHECR2016]. We summarise the conclusions of those studies here,
- The MD hybrid study published in early 2015 [@TA_Composition_Hybrid] detailed the analysis of showers with energies above $10^{18.2}$eV viewed by the refurbished HiRes FD detector over five years. Using improved profile reconstruction cuts (based on a pattern recognition approach), [$X_{\rm max}$]{} resolution better than 25[g/cm$^2$]{} was achieved, with a systematic uncertainty in [$X_{\rm max}$]{} of better than 18[g/cm$^2$]{}. Data were compared with expectations of the QGSJetII-03 hadronic model, both in terms of the mean [$X_{\rm max}$]{} as a function of energy, and by comparing the shapes of the [$X_{\rm max}$]{} distributions in a number of energy bins. The overall conclusion was that, taking into account systematic uncertainties, the mean behaviour and the distributions are consistent with the expectations for a light, mainly protonic composition. An additional two years of hybrid MD data were included in an analysis presented in 2016 [@Hanlon_UHECR2016] with no change in conclusions.
- An alternative to “hybrid” geometrical reconstruction of FD events using information from the SD is to use the stereo technique combining views of the shower from at least two FD sites. An [$X_{\rm max}$]{} analysis of stereo data from all three FD sites over seven years was published in 2015 [@TA_stereoXmaxICRC15], with an energy threshold of $10^{18.4}$eV. The [$X_{\rm max}$]{} resolution and systematic uncertainty were similar to the MD hybrid analysis above. Comparisons were made with 5 hadronic interaction models, including two making use of recent LHC input. The trend is for more recent hadronic models to predict deeper developing air showers. Based on the behaviour of the mean [$X_{\rm max}$]{} as a function of energy, and on the shape of the [$X_{\rm max}$]{} distribution for all energies, the authors conclude that no iron is required at any energy, and that the data are consistent with protons from the early QGSJet-01c model. While pure protons from post-LHC models are disfavoured, a light composition remains consistent with the data within the systematic uncertainties.
- Finally, a new analysis of seven years of hybrid data from the BR/LR FD stations has been presented at the UHECR 2016 conference [@Hanlon_UHECR2016]. This data set is the largest with 2597 events above $10^{18.2}$eV, compared with 1346 events from the stereo analysis ($E > 10^{18.4}$eV) and 623 from the MD hybrid analysis. With the aid of the FADC digitisation of the signals in these FD sites, the [$X_{\rm max}$]{} resolution is improved to better than 20[g/cm$^2$]{}, though the systematic uncertainty is now conservatively quoted as 20.3[g/cm$^2$]{}. The data are compared with expectations from the QGSJetII-03 and QGSJetII-04 models, see Figure \[fig:TAMass\]. One conclusion is that, within the systematic uncertainty, the mean [$X_{\rm max}$]{} versus energy is consistent with that expected for a “light” composition. In addition, the [*shapes*]{} of the [$X_{\rm max}$]{} distributions in five energy bins are consistent with the protonic expectations, and inconsistent with those of iron.
A joint group of collaborators from both Auger and TA have been working to understand the differences in [$X_{\rm max}$]{} results from HiRes, TA and Auger [@XmaxWG2012; @XmaxWG2014; @XmaxWG2016]. A particular question is, are the differences related to experimental factors, or due to the interpretation via hadronic models? As we have discussed, the comparisons are complicated by the different philosophies of the experiments, with Auger applying cuts designed to remove detection bias. In their latest report [@XmaxWG2016], the group has asked the following question: are the measurements of [$X_{\rm max}$]{} made by Auger (both the mean values and the distributions) consistent with the measurements of TA? This is a question quite separate from any particular hadronic model and mass interpretation, though such models must be used to “translate” Auger measurements into TA expectations. The Auger fractions of protons and nuclei of He, N and Fe in energy bins above $10^{18.2}$eV were taken from [@AugerMassMixtures] under the assumption of the QGSJetII-04 model. Those mixtures were then processed through the TA detector simulation and reconstruction to give the [$X_{\rm max}$]{} distributions expected in each energy bin at TA for the Auger “mix”, taking into account any detection bias. In particular, the comparison was done for the seven-year, higher-statistics BR/LR hybrid data set described above. The conclusions are that the Auger mix produces a mean [$X_{\rm max}$]{} as a function of energy that is consistent with the TA measurements within the current systematic uncertainty of 20[g/cm$^2$]{}; and that there is also qualitative agreement between the shapes of the Auger mix distributions of [$X_{\rm max}$]{} and TA distributions in several energy bins below $10^{19}$eV where TA has sufficient statistics. Above $10^{19}$eV the TA data still suffer from insufficient statistics to come to more definite conclusions about the distribution widths. This important study removes much of the doubt about the consistency of Auger and TA results, and shows the importance of continuing dialog between the two experiments.
### Other mass-related measurements
Surface detector arrays are sensitive to variations in shower development (and hence mass) through measurements of parameters such as the pulse rise-time in a detector, the radius of curvature of the shower front and the lateral distribution function (e.g. see historical examples in [@NaganoWatson2000]). Those arrays with particular sensitivity to muons can also attempt to tackle the mass issue through measuring the muon content of air showers, as will be discussed in Section \[Sect2:interactions\].
Recently the Auger Observatory, in particular, has explored air shower development with several SD methods, and we briefly mention two here. While the resolution in inferred mass-related parameters such as [$X_{\rm max}$]{} is typically poorer than the equivalent FD measurement, the SD has the advantage of a 100% duty cycle.
The rise-time of a signal in a water-Cherenkov detector, defined as the time taken for the signal to increase from 10% to 50% of the total integrated value, is related to the core-distance of the WCD and the zenith angle of the air shower. It also displays azimuthal asymmetry with respect to the azimuth of the shower axis, which can be exploited to study shower development [@AugerRiseTime2016]. The conclusion of the study is that above $10^{18.5}$eV there appears to be an increase in the mean mass of cosmic rays, but that the detail of the mass increase depends on the hadronic model assumed, and the core radius range used in the analysis. The latter dependence implies a deficiency in both of the (post-LHC) hadronic models used.
Similar interpretational issues occur with Auger’s measurements of the muon production depth, MPD, using inclined energetic showers above $10^{19.3}$eV [@MPD1; @MPD2]. In such showers, the electromagnetic component of the shower is essentially absent at ground level, and the digitised signals from the WCDs can be analysed to give the longitudinal profile of the production depths of muons, and the depth of the maximum of that profile, $X^\mu_{\rm max}$. While not the same as the depth of maximum of the overall shower [$X_{\rm max}$]{} (dominated by the electromagnetic component), $X^\mu_{\rm
max}$ also has sensitivity to mass. The measurements show that the mean $X^\mu_{\rm max}$ is effectively flat with energy above $10^{19.3}$eV, implying a mass increasing with energy. The mean mass implied by the QGSJetII-04 model is heavy, but that implied by the EPOS-LHC model is unphysically heavier than iron. Again, this is an indication that the current hadronic models are not describing the measurements well.
Finally, there is one Auger measurement, this time at energies just above the spectral ankle ($10^{18.5}-10^{19}$eV), where it is claimed that its main conclusion is insensitive to details of hadronic models [@AugerMass_correlation]. Here, hybrid data are used to produce a scatter plot of [$X_{\rm max}$]{} vs $S(1000)$, and a correlation coefficient is determined. (The energy and zenith angle dependence of the variables is removed before plotting). The value of the correlation coefficient is found to be inconsistent with any [ *pure*]{} composition of [*any*]{} mass, with the conclusion the same for all three post-LHC hadronic models tested. This result disfavours, for example, a pure protonic cosmic ray flux around the spectral ankle, that proposed by the so-called “dip” model of this feature [@SpectralDip].
Photons and neutrinos {#Sect2:photonsNeutrinos}
---------------------
Apart from attempting to characterise the nuclei within the cosmic ray flux, there is great interest in searching for photon and neutrino candidates within the events detected by the experiments. Photons and neutrinos will be produced at some level in the sources due to interactions of hadronic cosmic rays with ambient gas and photon fields. They will also be produced through photo-pion production when the highest energy protons interact with photons of the cosmic microwave background (often called “GZK” or cosmogenic photons and neutrinos) [@CMB1; @CMB2], and several exotic models of “top-down” cosmic ray production (e.g. from super-heavy dark matter) predict significant photon fluxes (e.g. [@SuperHeavyDM]). Thus measurements, or limits, on the flux of UHE photons and neutrinos are important.
![The Auger single-flavour limits to the UHE neutrino flux (90% CL) in half-decade bins, with the equivalent limits from IceCube and ANITA. For references to these other measurements, and the cosmogenic models shown, see [@Auger_Neutrino2015].[]{data-label="fig:Neutrinos"}](figures/Photons.pdf){width="1.\linewidth"}
![The Auger single-flavour limits to the UHE neutrino flux (90% CL) in half-decade bins, with the equivalent limits from IceCube and ANITA. For references to these other measurements, and the cosmogenic models shown, see [@Auger_Neutrino2015].[]{data-label="fig:Neutrinos"}](figures/Neutrinos.pdf){width="1.\linewidth"}
### Recent photon limits
Both TA and Auger have produced updates to their photon limits in the past two years. For TA, the discrimination between hadronic and photon initiated air showers is done using a multivariate analysis of TA SD data using machine learning techniques [@TAPhoton_UHECR2012; @TAPhoton_ICRC15]. Among the variables tested are a shower front curvature parameter, and the signals in the top and bottom layers of the SD scintillators, the latter seeking to exploit the deficit of muons in photon initiated showers. No photon candidates were observed with $\theta < 60^\circ$, and 95% CL upper limits on the integral flux were derived: 0.032, 0.0047, 0.0021 and 0.0011 km$^{-2}$sr$^{-1}$yr$^{-1}$ above 3, 10, 30 and 100$\times 10^{18}$eV respectively.
The Auger collaboration searches for photons using two techniques. At lower energies hybrid data are used, and a multivariate analysis of variables including [$X_{\rm max}$]{} and a measure of the SD lateral distribution function (LDF) is the basis of the photon discrimination [@AugerPhoton2016]. Using 9 years of hybrid data, three photon candidates have been identified near $10^{18}$eV, a number consistent with the expected mis-classification of hadronic showers. Thus, upper limits on the integral photon flux are calculated for five lower energy thresholds of 1,2,3,5 and 10$\times10^{18}$eV, namely 0.027, 0.009, 0.008, 0.008, and 0.007 km$^{-2}$sr$^{-1}$yr$^{-1}$ (95% CL). This puts the photon fraction of the flux at less than 0.1% in the first bin and less than 2.7% in the last.
In the decade of energy above $10^{19}$eV, SD data alone were used in the Auger study [@AugerPhoton_ICRC15]. There, discriminating shower parameters are related to the signal LDF, and the rise-times of the WCD signals - photon showers have steeper LDFs and longer rise-times than hadronic initiated showers. With 8.5 years of data, five photon candidates are observed (consistent with expectations for hadronic mis-classification), and upper limits on the photon integral flux are $(1.9, 1.0, 0.49) \times
10^{-3}$km$^{-2}$sr$^{-1}$yr$^{-1}$ (95% CL) above thresholds of 1, 2 and $4\times 10^{19}$eV.
These limits are summarised in Figure \[fig:Photons\]. Note that while the Auger results are stronger because of the larger exposure, the TA experiment explores a different hemisphere, relevant in the case of point sources. The figure shows expectations for models of top-down production of UHECR, now disfavoured at almost all energies, as are two models of cosmogenic photons which assume a pure proton UHECR flux. The experimental limits are encroaching on the cosmogenic model with optimistic selections of the source spectral index and maximum energy. The other model expectation, assuming a proton source spectral index of $\gamma=2$ and a maximum energy of $10^{21}$eV, is 4 times lower than the integral limit at $10^{19}$eV. Sensitivity to this model may be reached by the current experiments in the next decade.
### Recent neutrino limits
The current competitive limits on UHE neutrinos come from the Pierre Auger Observatory and the IceCube experiment. The Telescope Array has not yet published results of their searches. IceCube and Auger have similar sensitivities at the highest energies.
The basis of Auger’s neutrino search is to identify “young” showers at large zenith angles (or indeed, upward-going) in the SD dataset. A young shower at ground level is one with both electromagnetic and muonic components intact. The electromagnetic component of a large zenith angle shower initiated by a hadron will be absorbed by the atmosphere before hitting the ground, so “normal” inclined events are characterised by SD station signals with fast rise-times and short durations. Auger’s latest limits have combined results from three searches to give its most sensitive single-flavour limits to date [@Auger_Neutrino2015]. The searches include one for earth-skimming showers (sensitive to $\nu_\tau$), and two searches in two zenith angle bands ($\theta \in (60^\circ,75^\circ)$ and $\theta
\in (75^\circ,90^\circ)$) sensitive to all three flavours. No candidates were identified, and the limits are shown in Figure \[fig:Neutrinos\]. These limits are now having some astrophysical significance, with some models of neutrino production in sources, and exotic production mechanisms, being ruled out. In particular, cosmogenic neutrino production models that assume pure proton fluxes at high-redshift sources and strong source evolution (like FR-II galaxies) are highly disfavoured by the Auger analysis [@Auger_Neutrino2015]. Similarly, the recent IceCube analysis excludes with 90% CL proton sources evolving strongly with the evolution parameter $m > 5$ and with redshifts $z$ up to 1.4 [@Aartsen:2016ngq].
In other neutrino-related studies, the Auger, TA and IceCube experiments have reported a negative finding on a search for coincident arrival directions of IceCube neutrinos and Auger and TA UHECR [@AugerTAIceCube]. The Auger Observatory has also searched for neutrino events associated with the first two LIGO gravitational wave observations [@Auger_NeutrinoGW].
Interaction cross-sections and tests of hadronic physics {#Sect2:interactions}
--------------------------------------------------------
As we have seen, modern cosmic ray observatories rely on models of hadronic interactions to interpret shower development measurements in terms of the primary cosmic ray mass. Thankfully, the near-calorimetric fluorescence technique has meant that energy assignments have very little dependence on these models.
Despite the very indirect nature of our observations of cosmic ray interactions, modern observatories can contribute knowledge to the nature of hadronic physics at energies well above those probed by the LHC. Two example areas are measurements of interaction cross-sections, and the identification of model deficiencies in predicting ground signals.
### The Proton-Air Inelastic Cross Section at Ultra-high Energies
The first measurements of $\sigma_{\rm p-air}^{inel}$ using cosmic rays at extreme energies were made by the Akeno [@Akeno_CrossSection] and Fly’s Eye [@FlysEye_CrossSection; @GaisserCrossSection] experiments, followed later by HiRes [@Belov2006]. While Akeno showed that this measurement was possible using a surface array (characterising shower development using electromagnetic and muon content at ground level), the Fly’s Eye and subsequent experiments have used FD observations of [$X_{\rm max}$]{}. The exponential tail of a histogram of [$X_{\rm max}$]{} measurements is fitted with a function $\exp(-X_{\rm max}/\Lambda)$ to yield the scale of the exponential $\Lambda$. Provided the showers contributing to the tail are initiated by protons, $\Lambda$ can be converted to $\sigma_{\rm p-air}^{inel}$ with a relatively small sensitivity to hadronic interaction models. For comparisons to accelerator data, the inelastic proton-air cross-section may be converted to the inelastic and total proton-proton cross-sections using Glauber theory (see e.g. [@Auger_CrossSection1]).
The Auger collaboration measurements were published in 2012 [@Auger_CrossSection1] and updated with increased statistics in 2015 [@Auger_CrossSection2]. The energy range of interest is around $10^{18}$eV where the mass composition appears proton-rich. In the latest analysis, two energy bins are used, $10^{17.8}-10^{18}$eV and $10^{18}-10^{18.5}$eV, corresponding to centre-of-mass energies of 39TeV and 56TeV, respectively. Only the deepest 20% of the showers are used in the analysis to minimise contamination from primaries other than protons. Nevertheless, an important systematic uncertainty is related to the possible contamination by helium nuclei. Conservatively a 25% contamination of helium is assumed. Results for $\sigma_{\rm p-air}^{inel}$ are $[457.5 \pm 17.8{\rm (stat)} ^{+19}_{-25}{\rm (syst)}]$mb at 39TeV and $[485.8 \pm 15.8{\rm (stat)} ^{+19}_{-25}{\rm (syst)}]$mb at 56TeV. Of the total systematic uncertainty, $\pm 10$mb is attributed to hadronic model sensitivity at both energies.
The recent Telescope Array analysis is of showers observed by the Middle Drum FD detector in hybrid mode [@TA_CrossSection]. Air showers over the energy range from $10^{18.3}-10^{19.3}$eV are used, corresponding to an average centre-of-mass energy of 95TeV. Showers in the tail of the [$X_{\rm max}$]{} distribution beyond 790[g/cm$^2$]{} are assumed to be protons. Assuming a 25% contamination of helium in the tail, the $\sigma_{\rm p-air}^{inel}$ is determined to be $[567.0 \pm
70.5{\rm (stat)} ^{+29}_{-25}{\rm (syst)}]$mb.
The results from both experiments have been converted to proton-proton cross-sections to rule out the more extreme extrapolations of accelerator data [@Auger_CrossSection1; @TA_CrossSection].
### Characterisation of Deficiencies in Hadronic Models {#Sect2:deficiencies}
Both the Auger and Telescope Array experiments have detected likely deficiencies in the hadronic interaction models employed in air shower simulations. That there are deficiencies is unsurprising, given that these models are extrapolations of direct accelerator measurements, but it is encouraging that some of the more recent models, based on LHC data, are less discrepant with respect to the cosmic ray measurements (see below).
An example from the TA experiment relates to the SD energy estimator $S(800)$. It has been related to primary energy using simulations of proton showers (a preference based on HiRes and TA interpretations of mass composition) and the QGSJetII-03 hadronic model [@TA_2013Spectrum]. The SD energy ($E_{SD}$) at a given zenith angle is determined as the (simulation) energy that reproduces the measured $S(800)$ at the same zenith angle. For hybrid events, the ratio $E_{SD}/E_{FD}$ is found to be 1.27, where $E_{FD}$, the FD energy, is obtained calorimetrically and is essentially free of hadronic physics uncertainties. The ratio has no significant dependence on energy or zenith angle for $E > 10^{18.5}$eV and $\theta < 45^\circ$. The lateral distribution and other experimentally measured variables are well reproduced by the energy-rescaled shower simulation (proton, QGSJetII-03). The required rescaling points to a deficiency in the simulations that predicts fewer charged particles (electrons and/or muons) hitting the SD, although a further quantitative analysis studying the dependence on hadronic model, assumed mass composition and zenith angle is necessary. The uncertainty of the ratio $E_{SD}/E_{FD}$ is currently dominated by the FD energy uncertainty, which is 21% for TA, and its improvement will help to pin down the nature and the level of the deficiency.
The Auger collaboration has studied the muon content of inclined (zenith angle $62^\circ-80^\circ$) air showers above $4\times
10^{18}$eV [@Auger_Muon2015]. At these angles the electromagnetic component of the showers is absorbed by the atmosphere, and the WCD signal is essentially due to muons. Using hybrid showers, the energy is known from the FD (to within its systematic uncertainty of 14%) and the muon content measured by the SD can be compared with expectations from simulations (using two pre-LHC and two post-LHC hadronic models) of proton, iron and mixed compositions. The mixed composition is that implied (for each hadronic model) from Auger FD measurements at these energies. A relative integrated muon number $R_\mu$, designed to remove the energy and zenith-angle dependence of the measurement, is used to compare real measurements with simulations. It is found that the simulations underestimate the muon measurements by $(30^{+17}_{-20}{\rm
(sys.)})$% to $(80^{+17}_{-20}{\rm (sys.)})$% for the assumed mixed composition at $10^{19}$eV, over the range of models tested. The models with smallest discrepancy are the post-LHC QGSJetII-04 and EPOS-LHC models. The quoted systematic uncertainties arise primarily from the experimental measurement, a significant part of which is due to the 14% systematic uncertainty in FD energy. In this study, the energy systematic is necessarily transferred to a systematic in muon number.
This mixing of systematics is largely avoided in another Auger study, this time with more vertical ($0^\circ - 60^\circ$) hybrid showers with energies between $6\times10^{18}$ and $1.6\times10^{19}$eV [@Auger_Hadron2016]. Simulated showers were generated using the EPOS-LHC and QGSJetII-04 hadronic models for pure protons, and for mass mixtures consistent with the Auger measurement for each model. Then for every one of the 411 real showers, the simulation library for a given model and mass option was searched for the best match to the real longitudinal profile as measured by the FD. The lateral distribution function of that simulated shower was then compared with that measured by the SD. On average, the simulations underestimated the signal $S(1000)$ for both models and both compositions, and the deficit was [*not*]{} constant with zenith angle. It is this zenith angle dependence that reduces the degeneracy between a systematic shift in energy or muon content, since the FD is sensitive mainly to the electromagnetic component, and the SD is sensitive to both EM and muonic components, the mixture of which changes with zenith angle. The analysis results in rescale factors $R_{\rm had}$ for the muon content and $R_E$ for the energy scale, for each model/composition combination. The data and the simulation can be brought into agreement with the application of $R_{\rm had}$ and/or $R_E$ to the simulation. The results are that, for the mass mixture, the energy rescale factor was consistent with unity for both hadronic models: $R_E = 1.00 \pm 0.10$ for EPOS-LHC and $R_E = 1.00 \pm 0.14$ for QGSJetII-04, where the error is the statistical and systematic uncertainties added in quadrature. However, the magnitude of the rescale factors necessary for the simulated muon numbers to match the experiment were $R_{\rm had} = 1.33 \pm 0.16$ for EPOS-LHC and $R_{\rm
had} = 1.61 \pm 0.21$ for QGSJetII-04, both improvements in the significance of the model discrepancies compared with the inclined air shower study discussed above. Obviously, the hadronic rescale factors required for a pure-proton composition were even larger.
Recent preliminary results from the TA collaboration indicate that the muon content of EAS at distances between 2 and 4km from the core of the shower is substantially larger by factors of two to three (at the $3\,\sigma$ level) than predictions from any of the current hadronic models for both proton and iron primaries. The improved sensitivity to muons is provided by making very selective cuts that maximise the absorption of electrons by the atmosphere. Results indicate that the discrepancy increases with core distance, which may imply problems with our understanding of the early part of EAS development, since the muons at large core distances would originate there [@TAMuonExcess].
These examples show that there is sensitivity for testing hadronic models with the current observatories, taking advantage of hybrid measurements of air showers with surface and fluorescence detectors. We can expect even better sensitivity in the future using surface detectors that can separately measure muon and electromagnetic shower components, for example with the upgraded Auger Observatory [@AugerPrime].
Challenges
==========
Composition
-----------
### Using the [$X_{\rm max}$]{} measurement {#Sect3:using_xmax}
Linsley [@Linsley1977] first proposed a simple way to look for changes in the cosmic ray composition as a function of energy. This involves the so-called “elongation rate” or mean [$X_{\rm max}$]{} as a function of energy. In a simple superposition model, a pure single component composition will have $\langle$[$X_{\rm max}$]{}$\rangle$ depend logarithmically on $E$ with a constant change per decade (the elongation rate). A change in the composition would create an energy dependent change in this rate. A change from a light to heavier composition would produce a decrease in the rate of change of mean [$X_{\rm max}$]{} with energy (a flatter or even negative elongation rate) for example. As long as there are no rapid changes in hadronic interaction physics, this is true in a model independent way. However, the elongation rate does not tell us what the composition actually is. For this, hadronic models must be used to simulate air showers, the response of the fluorescence detector must be folded in or dealt with using cuts, and the resultant absolute position of $\langle$[$X_{\rm max}$]{}$\rangle$ compared with data at a number of energies.
There are a number of problems with this approach. Firstly, the absolute predicted value of $\langle$[$X_{\rm max}$]{}$\rangle$ is hadronic model dependent, with variations of 10-20[g/cm$^2$]{} between extreme models at any given energy. Then, the actual [$X_{\rm max}$]{} distribution is asymmetric and if there is a significant protonic component it will have a long tail extending to deep [$X_{\rm max}$]{}. Heavier nuclei will have less pronounced tails. The mean value, $\langle$[$X_{\rm max}$]{}$\rangle$, is sensitive to these tails which can be affected by detector systematics. This is one possible source of bias that can produce a systematic difference between simulations and data. Studies indicate that all the various effects can produce a net residual systematic in the mean [$X_{\rm max}$]{} as large as 10-20[g/cm$^2$]{} for the TA experiment [@TA_Composition_Hybrid], and up to 10[g/cm$^2$]{} for the Auger experiment [@Auger_longXmax]. Unless care is taken, undersampling due to low statistics may also shift $\langle$[$X_{\rm max}$]{}$\rangle$. The second moment of the distribution, the RMS, is sensitive to both the tails and the width of the distribution and hence carries additional information. The RMS is less hadronic model dependent since the distribution width mostly depends on superposition. A change of RMS from 60[g/cm$^2$]{} (characteristic of proton showers for essentially all hadronic models) to 30[g/cm$^2$]{} as a function of energy is observed by the Auger collaboration [@Auger_longXmax] in the energy range above 10$^{18}$eV and it can be considered evidence for a change in composition. The smaller RMS can only be produced by heavier nuclei such as CNO or Fe, again in an essentially model independent way. However, the RMS measurement suffers some of the same systematic problems as the elongation rate. Undersampling of the tail of a distribution either due to low statistics or detector bias can mimic a composition change. The Auger collaboration has been able to address this in some detail because of its high statistics [@Auger_longXmax]. The TA measurements at the highest energies still suffer from insufficient statistics to address this issue completely.
A puzzling issue that has emerged from this approach is that it is difficult to reconcile the $\langle$[$X_{\rm max}$]{}$\rangle$ with the RMS distributions. In a simple two component p/Fe model for example, the RMS at the highest energies agrees well with a nearly pure Fe composition but the $\langle$[$X_{\rm max}$]{}$\rangle$ requires a much lighter mix. Reproducing this is a struggle even with a four component composition.
For all these reasons, comparison of the full [$X_{\rm max}$]{} distribution between data and Monte Carlo (MC) simulations seems the best approach. In principle it provides the maximum information. However, a straightforward statistical comparison of data and simulations is made impractical because of the presence of significant systematic uncertainties both in the data (overall [$X_{\rm max}$]{} position) and the hadronic model.
There are two approaches to deal with the problem of determining composition. The TA and HiRes collaborations apply loose cuts to data (sufficient to ensure good resolution) and carefully simulate p, He, N and Fe air showers based on a variety of hadronic models. Whatever distortions in the [$X_{\rm max}$]{} distribution are generated by the detection method and reconstruction should then be evident in the reconstructed simulated data. The Auger collaboration instead applies much tighter fiducial volume cuts which minimise any detector and reconstruction bias. The resulting data can then be directly compared to the “thrown” simulations. Direct comparison of the data from these two approaches can be problematic since the detector distortions will be different, though the biases in the most recent TA hybrid analysis are much smaller than for previous results.
Recently, the Auger and TA groups have developed a method to improve comparison of [$X_{\rm max}$]{} distributions [@XmaxWG2016]. The Auger group fits their cut and unbiased data to a simulated composition mixture as a function of energy. The resultant composition fractions are then used by the TA group to generate “thrown” simulations. These are then processed through the TA reconstruction process and compared to the data. Preliminary results show agreement within the systematic uncertainty for the overall elongation rate. This approach is, in principle, independent of the hadronic model used, since this is only used as a method to port one data set into another experiment’s acceptance.
But how does one deal with systematic uncertainties in these comparisons? What is needed is a comparison method that allows for a sliding [$X_{\rm max}$]{} scale (to take account of overall systematics) while preserving the shape of the distribution. The TA group has proposed such a method [@Hanlon_UHECR2016] which first removes the energy dependence of the distributions and allows an [$X_{\rm max}$]{} shift for the data which is determined by the best overall [$X_{\rm max}$]{} profile fit. One can then compare the distribution shape goodness of fit to the required sliding [$X_{\rm max}$]{} scale factor to see how well any given composition assumption does when compared to the data. For example, if the scale factor shift required is well beyond the estimated systematic uncertainties and there is a poor profile shape fit, then that hypothesis can be discarded.
Another approach, used by Auger [@AugerMassMixtures], is to directly compare a multi-component mix of simulated showers with the data. This is done for a variety of hadronic models. This approach uses p, He, N and Fe as markers for the actual cosmic ray composition. The QGSJetII-04, Sibyll 2.1, and EPOS-LHC hadronic models are used. The individual components have separate weights that vary as a function of energy and the experimental systematics are folded in to the degree that they are known. The results change as one shifts hadronic model assumptions; while an overall trend of moving from p to He in the energy range from $10^{18.3}-10^{18.8}$eV is shared by all 3 models, only EPOS-LHC gives a nearly constant and large fraction of N ( 40%) across the full energy span, while the other two models are consistent with almost zero N fraction. Above $10^{18.8}$eV, dominant components are He and N, but their proportions are very much model dependent; EPOS-LHC favors N while the other two models strongly support He. These differences make an astrophysical interpretation challenging as they are most likely due to model inadequacies. While the details are not clear, the required proton fraction decreases above 3$\times$10$^{18}$eV and there is no requirement for any significant iron fraction.
The lack of consistency is not surprising given that even for a single hadronic model, HiRes publications [@Abbasi2005b] have noted that introducing more than two components into a fit to an [$X_{\rm max}$]{} distribution does not lead to an easily interpretable result as various combinations can give equally good fits. In the case of Auger, the best fits are produced with more than two components, but the uniqueness of the interpretation remains problematic.
Any particular approach to reconstruction shower profiles has hidden systematics which are intrinsic to the chosen approach and the particular software implementation. This systematic uncertainty is separate from detector or atmospheric systematics. Two different, error-free, reconstruction programs that use different approaches (different binning, least-square fitting routines, tabular vs functional corrections etc.) will produce slightly different results. The TA group has explored this “intrinsic” systematic by comparing completely independent and otherwise well-vetted hybrid reconstruction programs as well as by comparing results from stereo data. They find that it is very difficult to make the [$X_{\rm max}$]{} distributions (for the same data or simulations) agree to better than 5-10[g/cm$^2$]{} [@HanlonPC; @Auger_longXmax]. This seems to be an irreducible systematic uncertainty.
A particular complication in the study of cosmic ray composition is the fact that any nucleus heavier than a proton will eventually fragment to a lighter nucleus as it travels from its source to the Earth. This fragmentation is due to the interaction of the nucleus with both the relic 2.7K black body photons and the IR radiation fields produced by stellar radiation [@Puget1976; @Stecker1999; @Aab2016; @Allard]. As a result, even a pure single nucleus composition heavier than a proton at the source should appear as a mixed composition. A pure proton primary composition will arrive intact, but observation of a proton component cannot rule out that part of this component is due to heavier nuclei. On the other hand, observation of an iron component uniquely indicates the existence of a primary iron at the source, since stellar nucleo-synthesis does not provide any significant concentration of nuclei above iron. Propagation models show that He, CNO and Fe have different spallation probabilities as a function of energy [@Taylor2006; @Hooper2008]. This is particularly evident above 3$\times$10$^{19}$eV where the He mean free path is on the order of 10 Mpc, compared to $\sim$ 100 Mpc for N [@Allard]. A He dominated flux above 3$\times$10$^{19}$eV only makes sense if sources are very close. The lack of anisotropy makes any such assertion implausible. Indeed propagation calculations indicate that integrated over the large scale structure, the mean $A$ of nuclei originating as He is essentially one [@Hooper2008]. If the observed He is the result of fragmentation of heavier nuclei, then a proper proportion of these heavy nuclei, whose mean free paths are much longer, must also be seen in the composition distribution. Incorporating the observed cosmic ray nuclear abundances found with a particular hadronic model directly with propagation effect weights in a more direct fashion than is currently done could be very helpful. In some cases, this may rule out an otherwise well fitting model as leading to astrophysically implausible scenarios.
### Implications of the lack of iron in UHECR {#Sect3:lack_of_iron}
While the systematic uncertainties in [$X_{\rm max}$]{} determination and comparison to simulated compositions are still too large to make strong statements about the relative abundance of elements in the cosmic ray flux at Earth or at their origin, we have learned that there is very little iron in the flux above 10$^{18}$eV. Given existing systematic uncertainties, it is safe to say that any direct or secondary heavy nuclei from Fe to Si are absent from the spectrum. We know this absence with better precision than we know what elements are present in the flux. What does this imply about the sources and acceleration mechanisms of UHECR? UHE primary iron can easily reach the Earth from as far away as 100 Mpc and its spallated byproducts down to Si from much further distances. Are magnetic field effects strong enough to substantially increase the effective path length? Is there a deficit of iron in the cosmic material feeding the accelerator? There is astronomical evidence that iron attaches itself to dust particles and hence appears to be somewhat depleted in its free form, for example [@DeCia2016; @Krogager2016; @Gail2015]. However, why the iron-rich dust particles cannot be swept from an accretion disk into the accelerator beam, decomposed to their atomic constituents and provide the original iron abundance is not clear. With a charge 26 times that of a proton, iron will be accelerated quite efficiently at the highest energies. If iron is indeed accelerated at the source then its absence must require photon fluxes at the source that essentially eliminate it from the cosmic ray flux. The absence of heavy elements in the cosmic ray spectrum may thus be an important constraint and clue to cosmic ray origins.
### FD/SD energy mismatch, muon excess {#Sect3:missmatch}
Much of the progress in establishing the structures in the cosmic ray spectrum come from the reliable energy scale provided by air fluorescence. With a $\sim$15% FD energy resolution and similar systematics, calibrating the SD energy scale to simultaneously observed FD events has made the SD spectrum energy largely hadronic model independent [@Song2000]. For the TA case, the energy scale adjustment that needs to be made between the FD and SD is on the order of 25-30% if one uses QGSJetII-03 proton simulations for the SD energy. The measured particle densities produce a lateral distribution which, were it analyzed based on hadronic model simulations of air showers, would generate too high an energy by this amount. In other words, there are too many charged particles at ground level for a shower with an energy as determined by the FDs. Because Auger water Cherenkov detectors are quite sensitive to muons, this mismatch has been attributed to an excess of muons in the data compared with expectation. In the case of TA’s plastic scintillation detectors, electrons and muons have similar detector response and the mismatch can only be partly attributed to a muon excess. Studies are proceeding to investigate whether scintillation detectors at large distances from the core which should have mainly muon initiated signals are consistent with the Auger results [@TAMuonExcess]. In any case, it is clear that the hadronic models that are used to simulate showers are not adequate. Until this issue is resolved it is difficult to use muon density to measure cosmic ray composition precisely, though trends can certainly be established (see below).
### Other techniques
#### a. Radio
The fluorescence technique revolutionised the study of UHECR physics because it made possible a largely calorimetric determination of the energy scale and a relatively direct measure of composition using [$X_{\rm max}$]{} of the showers. It requires clear moonless nights which restricts its on-time to 10-15% of the SD operation times. Recently a great deal of work has been done in investigating the possibility of using radio emission from EAS in much the same way as one now uses fluorescence [@huege-kyoto]. Radio can, in principle, determine the shower energy and [$X_{\rm max}$]{} and would have the advantage of $\sim$100% on-time. A number of radio arrays have now been operating either stand-alone or in conjunction with surface and air fluorescence detectors [@lofar_xmax; @aera_instrument; @aera_observation]. Because there are several mechanisms in the air shower development that can generate radio waves, the detailed simulation has taken some time to develop but now seems sufficiently advanced. Meaningful comparisons with real data have been done and good agreement is now evident between simulations and radio and SD measurements [@radio-1]. These studies have been largely limited to energies less than 10$^{18}$eV however, and in this energy regime it appears that an array of radio antennas with spacings not dissimilar to SD spacings are required for good energy and [$X_{\rm max}$]{} resolution. If similar spacings is required for $>$ 10$^{18}$eV energies, the costs associated with instrumenting $>$ 1000 km$^2$ arrays become significant. Until more complete optimisation and cost/benefit analyses for the UHECR regime are done it is not clear that this technique will supplant fluorescence and particle SD arrays. In any case, significant physics from the low energy arrays is required before this new technique can be considered fully vetted.
#### b. [$X_{\rm max}$]{} - SD signal correlations
Recently an approach to studying composition has been proposed using the correlation between [$X_{\rm max}$]{} and the SD signal [@AugerMass_correlation]. This is based on the very old idea that if iron and proton showers have different [$X_{\rm max}$]{} distributions and different $N_\mu$ distributions then superposing their [$X_{\rm max}$]{}-$N_\mu$ scatter plots will lead to a negative correlation even though the pure distributions have a positive correlation. Since this is generally true of any hadronic model, the claim of this approach is that this is more model independent than either a pure [$X_{\rm max}$]{} or pure $N_\mu$ analysis. However, since neither Auger or TA actually measure $N_\mu$ (except at large zenith angles in the case of Auger) the searched for correlation is with $S(1000)$ or $S(800)$. The recent work on this by Auger [@AugerMass_correlation] shows that in the $10^{18.5}-10^{19.0}$eV energy region the correlation is inconsistent with a pure composition. A preliminary study by the TA collaboration on the other hand shows no inconsistency with the assumption of a protonic composition [@LindquistPC]. However, in the case of TA, the method is not nearly as sensitive as the [$X_{\rm max}$]{} method. This may be because the muon content in TA is not as large a component of the SD signal as for Auger. Since the method is “de facto” dependent on detector muon sensitivity it must also be to some extent model dependent, although the Auger study checks this with two independent models. Studying the applicability of this method in the lower energy region ($10^{18}-10^{18.5}$eV) would be of interest since there the composition is likely to be more pure, given both the Auger and TA [$X_{\rm max}$]{} data.
Energy {#Sect3:energy}
------
One of the most significant results coming from Auger and TA is the overall agreement in the shape of the UHECR spectrum. At first glance, both spectra show a clear ankle structure and a cutoff, although the precise energies for these structures differ. However, a shift of either experiment’s energy scale by 10-15% brings the ankle structure into excellent agreement [@Verzi_PTEP]. Since such a shift is within the systematic uncertainties of either experiment, it would seem that there are no significant north - south differences here. A closer look at the ratio of the two spectra shows, however, that the location and shape of the cutoff seems different at the $\sim$3$\sigma$ level [@VerziICRC15; @TA_Spectrum_ICRC15]. The Auger-TA combined working groups have looked at this difference and, so far, have found no reason to believe it is a result of systematic uncertainties in energy. If this is truly a difference in the flux of northern and southern sources at the highest energies, there should be an overall declination dependence. Preliminary evidence from TA indicates that the TA spectrum becomes much more like the Auger spectrum near the cutoff if a declination cut of $<25^\circ$ is made [@IvanovPC]. The difference (a higher energy cutoff for TA) must then come from higher declinations which also contain the “hot spot” that may be a signature for a relatively nearby source. While this is suggestive, much more work needs to be done to demonstrate that this cannot be a systematic effect either in energy or aperture estimation.
### Energy Scale Shift Systematics
As indicated above, the ankle structure which is seen with high statistics in both TA and Auger data can be used to estimate the difference in energy scale of the two experiments. While the result is within systematic uncertainty estimates, it is important to understand the nature of the energy shift as well as possible. Given the current precise nature of shower reconstruction, differences in energy can most likely be attributed to systematic uncertainties in optical properties (mirror reflectivity, light collection efficiency etc.), phototube gain calibration, atmospheric transmission and air fluorescence efficiency. All but the last are by their nature detector dependent and we must rely on the diligence of the experimenters in estimating how well they know these parameters.
The air fluorescence efficiency is in principle a common factor, though it depends on humidity and temperature corrections which may be somewhat different in the two locations. For reasons of keeping a historically consistent energy scale, the HiRes and TA groups have used the original Kakimoto et al. overall yield measurement [@Kakimoto] while a subset of Auger collaborators has launched a series of special experiments to measure the fluorescence yield more precisely with AIRFLY [@AIRFLY0; @AIRFLY]. The HiRes group also performed a series of measurements using an electron beam at SLAC (FLASH [@FLASH2; @FLASH3; @FLASH4]) but only the relative spectral line strength measurement has so far been incorporated in the TA analysis. The TA experiment also includes a 40 MeV electron linac whose vertical beam is seen in the field of view of one of the fluorescence detectors. Work on understanding the results of this in-situ measurement is proceeding [@TA_ELS_ICRC15]. All contemporary measurements of the absolute value of air fluorescence rely on a fixed energy electron or proton beam which deposits energy in a small pressure controlled chamber. Significant corrections for deposited energy escaping the chamber (in the form of delta and gamma rays) must be made. The MACFLY [@MACFLY2007] and thick target FLASH experiments generated a shower in an air-equivalent material and observed the fluorescence as a function of absorber. Neither of these experiments was able to produce an absolute value for the fluorescence efficiency with sufficiently small uncertainties to compete with the thin target experiments, though they did show that the relative longitudinal development of showers is well tracked by the resultant air fluorescence. Recently a new experiment at SLAC called sFLASH [@SokolskyPC] is attempting a $<$ 10% total systematic uncertainty measurement of air fluorescence from a $\sim$10 GeV electron shower developing at sea level and observed near shower maximum. If successful, this will be a valuable cross check on the thin target results. What is lacking is a common air fluorescence result that is used for both TA and Auger analysis. It is to be hoped that such a convergence can occur in the near future.
Anisotropy {#Sect3:anisotropy}
----------
Large scale anisotropy can be searched for using a multipole expansion. This is however tricky to do without bias unless one has full coverage over the celestial sphere. It is thus very advantageous to combine TA and Auger arrival direction data. There are several challenges to using this data set however. Because of the energy scale difference one cannot simply apply the same energy cut for both data samples. There are also potential systematic differences in determining the detector apertures of the two detectors. The Auger/TA anisotropy working group has developed an approach that uses the overlapping declination band for the two detectors [@AugerTA_anis1]. The fluxes in this band are normalised and this normalisation is carried over to the total data set. The assumption here is that the spectrum has no significant declination dependence in the overlap band. The resultant distributions have yet to show any statistically significant dipole or quadrupole moments, though Auger itself observes a significant dipole enhancement [@AugerAnisot_Broad2015]. A better understanding of the energy scale shift between the two experiments, and strategies to deal with it, could provide a simpler method of combining data without additional assumptions.
TA hot spot {#Sect3:hotspot}
-----------
With the fading of the Auger association of UHECR with AGN [@AugerAGN2007; @AugerAnisot_small2015], the community’s hope for finding clear associations of cosmic ray arrival directions with astrophysical sources has received a lift with the possible observation (at the 3.4$\sigma$ level) of a concentration of cosmic rays with energies above 5.7$\times$10$^{19}$eV in the northern sky by the TA experiment [@TA_Hotspot2014]. This “hot spot” of $20^\circ$ radius is observed near Ursa Major, about 10 degrees off the supergalactic plane. If this intermediate-scale anisotropy is confirmed with more statistics its location raises interesting questions, since none of the previously assumed cosmic ray sources (e.g. the Virgo cluster) are in the immediate vicinity. If the sources are actually in the adjacent portion of the supergalactic plane, then there must be a magnetic field effect shifting the flux to the observed location. A suggestion has been proposed that there is a magnetic flux tube produced by a filament of galaxies connecting the hot spot to sources such as M87 [@Ryu2016]. Another possibility is M82 which is sufficiently close to account for the hot spot using currently estimated magnetic fields. Tidal disruption events creating one or more flashes of extremely high energy protons or nuclei have been proposed for the acceleration mechanism [@Pfeffer; @He]. If this hot spot strengthens in significance it will pose a challenge to our understanding of sources and magnetic field configurations.
If medium or small-scale anisotropy is finally observed the next major challenge is to correlate our composition related information with arrival direction information. Hybrid FD plus SD data would be the most convincing, but requires the most running time for any given source. If the muon content of showers can be better understood and correlated with composition, this could give the most sensitive composition dependent anisotropy measurement. It is unfortunate that the currently most likely source (TA hot spot) and the major Auger upgrade of their SDs to better detect muons correspond to disconnected parts of the sky. If the hot spot is confirmed, and the muon content becomes better understood, coming to grips with this issue will be one of the major challenges for this community.
Future Observations
===================
Extension and Upgrade of Ground Observatories
---------------------------------------------
Above the “knee” at around $10^{16}$eV, the cosmic ray energy spectrum and [$X_{\rm max}$]{} measurement demonstrate rich features, and around $10^{19}$eV and above, various anisotropies seem to show up in the energy spectrum and flux. Where statistics are adequate, no obvious inconsistency is found in the [$X_{\rm max}$]{} measurements in the northern and southern hemispheres above $10^{18}$eV, but their interpretation allows a range of composition mixes and energy dependencies due to statistical and systematic limitations. It is important for Auger and TA to cover the entire sky and the whole energy region together, in order to bring these indications to a consistent set of observational facts. It will become the basis of locating the galactic to extra-galactic transition energy of cosmic rays sources, and of building a viable astrophysical model to explain the production and propagation of UHECR. Continuing to challenge this physics, TA and Auger are both planning to start the operation of extended and upgraded detectors around 2018–19.
In the northern hemisphere, TA$\times$4, the extension of TA [@sagawa_icrc2015; @kido_uhecr2016] is in preparation. It will extend the aperture of the SD by a factor of four by 2018. Leaving a part of the SD intact with 1.2 km spacing, an extended part will have a 2.08 km spacing, together covering a 3,000 km$^2$ ground area. Adding two more FD stations, the hybrid coverage will be tripled. The trigger efficiency of the extended SD will be larger than 95 % for E $>~10^{19.8}$eV. Resolutions will be slightly compromised to become $\sim$25 % for energy and 2.2$^\circ$ for the arrival direction. In three years of running over 2018-2021, the number of SD events above 57EeV (=$10^{19.76}$eV) will be quadrupled to become 300, of which $\sim$80 would be in the hotspot region, assuming the flux of [@TA_Hotspot2014]. The measurement range of $\langle$[$X_{\rm max}$]{}$\rangle$ using hybrid events will be extended to $\sim 10^{19.6}$eV from the present $10^{19.4}$eV [@Hanlon_UHECR2016]. The SD design of the TA$\times$4 was re-optimised to use a much shorter length of wavelength shifting fibers (1/3 of the length in the TA/SD) while keeping the same number of photo-electrons collected by the PMT for a minimum ionising particle. The quantum efficiency and the linear range of the PMT is nearly doubled.
In the southern hemisphere, the Auger Observatory plans to upgrade the detector to AugerPrime by 2018 [@AugerPrime; @engel_uhecr2016]. All the $\sim$1600 stations will be equipped with a 3.8 m$^2$ plastic scintillator on top and the waveform sampling electronics will become three times faster (to 120 MHz). An integrated analysis of water-Cherenkov and scintillator signals will enable an isolation of muonic and electromagnetic (EM) energy deposits, and enable the counting of the number of muons hitting the SD. New methods are being developed to estimate [$X_{\rm max}$]{} from SD measurements alone, taking advantage of so-called shower universality [@Universality]. The search for small-scale anisotropy and source correlation is expected to improve significantly by selecting SD events with high likelihood of being protons or light nuclei. The muon identification is double-checked for a portion of SD events using an array of scintillators buried 2.3 m underground. The duty cycle of FD operation is expected to become 1.5 times larger by tolerating data collection with a higher night sky background. The mixed composition result [@AugerMassMixtures] will be further checked with measurements from the FD together with the enhanced SD with its own measurements of [$X_{\rm max}$]{} and muon content. AugerPrime and TA$\times$4 together will have all-sky coverage with a total of 6,000 km$^2$ of surface area; one at 39$^\circ$ North and the other at 35$^\circ$ South. The overlapping region at low declinations ($-16^\circ$ to $+45^\circ$) will be important in understanding the relative exposures and to examine systematics of the detectors and data analyses.
Air shower detectors operating in the last decade have reported a series of Earth-science related findings; TA’s SD recorded bursts of particle showers associated with lightning [@okuda_icrc2015; @belz_uhecr2016], the development of distant atmospheric “elves” was recorded by Auger [@tonachini_icrc2013; @colalillo_uhecr2016], and the LOPES radio signal from air showers was modulated by thunder-clouds [@buitink_thundercloud] etc.. UHECR observatories may become an interesting research tool for Earth and atmospheric sciences in the next decade.
Development of Radio Detection
------------------------------
Understanding the mechanism of air showers generating radio signals in the sub-100 MHz range has advanced greatly in the last decade (see [@huege_pr; @huege_uhecr2016] for reviews). Newly developed simulation codes tell us that the radio signal comes from two types of time-varying, fast-moving effective charges generated in the air shower; one is the lateral movement of shower e$^\pm$ under the geomagnetic field and the other is the longitudinal movement of net charge in the shower front (the Askaryan effect). Both signals scale with the square of the electromagnetic energy ($\propto E^2$). The signal is sharply forward peaked in the direction of air shower development and stands well above the galactic radio noise for energies exceeding 10$^{16}$eV. The radio telescope LOFAR, operating in cosmic ray detection mode, realised a very fine radio sampling of air showers, and succeeded in observing air showers of energy 10$^{17}$ - 10$^{17.5}$eV with a typical [$X_{\rm max}$]{} reconstruction uncertainty of 17g/cm$^2$ [@lofar_xmax].
The AERA radio array with varied antenna spacing has been deployed at the Auger site for testing the detection of the highest energy showers [@aera_instrument; @aera_observation]. The results demonstrate that a dense deployment of antennae is required for the effective detection of UHECR that have a footprint of several 100 meters in diameter. Even though the elements of the RD (Radio Detector) may be relatively simple and inexpensive, the total cost of deploying and operating a large area detector would become prohibitive. One practical application for UHECR is for the measurement of very inclined air showers with $\theta > 70^\circ$, which has an extended oval footprint larger than 10 km$^2$. Such RDs deployed together with the SD may also be used for the calibration of SD energy, making use of the fact that the radio signal originates predominantly from the EM component of the shower, and that it has a negligible attenuation in the atmosphere. Note that this is the kinematic region where precise measurements by standard SD techniques, using a water-Cherenkov station or a scintillator, have large uncertainties, and redundant information is useful. The radio detector is also expected to have a high duty factor of $\sim$95%.
Another direction of progress foreseen for using radio signals from EAS is the detection of high energy neutrino-induced showers in the Antarctic ice via the Askaryan effect. Pioneering work searching for such short, GHz polarised radio signals from the horizon in Antarctica began with the ANITA balloon experiment in 2006 [@anita_1]. Its 4th flight was launched in December 2016. The ARA and ARIANNA experiments were recently proposed and extensive RD is underway to detect the Askaryan signal from cosmogenic neutrinos near or on the surface of the Antarctic ice [@barwick_ana_arianna].
Searches for GHz “Molecular Bremsstrahlung” radio emission from particle showers in the atmosphere [@mol-brems_gorham] have so far not been successful [@mol-brems_crome; @mol-brems_els]. Also, a limit has been set by the TARA experiment at the TA site for detecting the modulation of 54.1MHz carrier radio waves by the ionised column generated by a UHECR shower in the atmosphere [@radio-5].
Observations from Space
-----------------------
The EUSO international collaboration was formed in 2000 to install a wide field of view (FoV) telescope at the International Space Station (ISS) to look down on the Earth’s atmosphere and search for air fluorescence flashes from UHECR [@euso_proposal]. The JEM-EUSO detector employs a Fresnel lens telescope with a diameter of 2.4 m and a 60$^\circ$ FoV, covering a ground area of 200km radius from an altitude of 400km [@jemeuso_acceptance]. The effective ground coverage with the expected duty factor of 20% is 28,000km$^2$, or approximately five times that of AugerPrime and TA$\times$4 combined. A tilted mode of observation would increase the acceptance by a factor of three or more at the cost of reduced resolution and higher detection threshold. The ISS inclination angle of 51.6$^\circ$ allows a uniform survey of the Earth’s atmosphere in the northern and southern hemispheres with nearly the same acceptance and event geometry. The observations from high altitude and the limited optical entrance pupil of the Fresnel lens will however limit the JEM-EUSO detection threshold to be $\sim 10^{19.5}$eV, and the [$X_{\rm max}$]{} resolution is foreseen to be larger than 60g/cm$^2$ in the nadir mode [@jemeuso_resolution], making the differentiation of nuclear composition difficult.
The mission schedule of JEM-EUSO is yet to be determined, but an extensive series of tests of the prototype instruments is being performed [@casolino_uhecr2016]. Major efforts include the balloon borne EUSO-SPB test (2017), and deployments of mini-EUSO (2017) and K-EUSO (2020) at the ISS. The K-EUSO experiment will have a segmented Fresnel mirror 3.4 m in diameter, and its effective coverage of the ground will be 6,200 km$^2$ above 10$^{19.5}$eV, about equivalent to AugerPrime and TA$\times$4 combined. An exploring Russian satellite experiment, TUS, with similar optics was launched in 2016 and is being commissioned [@tus_uhecr2016]. The uniform all-sky coverage of K-EUSO will be very important in understanding the nature of the north-south anisotropy, or the inconsistency of the flux, being seen by TA and Auger.
A Future Ground Observatory
---------------------------
Given that observations by the extended ground detectors will proceed well into the next decade, and that exploratory space projects will start giving a large acceptance coverage of the entire sky, what are the new and/or remaining challenges for future ground observatories (FGO) of ultra-high energy cosmic rays? In this section, we take “Auger$\times$10” as a hypothetical example of an FGO, and discuss how the FGO might look, and how research might proceed with the FGO.
[**FGO:**]{} We assume “Auger$\times$10” is a symmetric set of northern and southern observatories, each with 30,000 km$^2$ ground area, covered in whole by hybrid arrays of FDs and SDs. An array of radio detectors (RDs) may be overlaid on the SD to enhance the energy and possibly the [$X_{\rm max}$]{} determination for inclined events, improving the quality of all sky coverage. We assume the SD is equipped with a particle identification function for a fraction of shower particles, and that this is to be used for the likelihood tagging of the primary composition. Approximately 10 % of events are SD-FD hybrid, which offers a direct means of composition determination via [$X_{\rm max}$]{}.
[**Composition at the cutoff:**]{} Such an FGO will collect approximately 10,000 SD events above $E_{1/2}$ (10$^{19.8}$eV for TA) or above $E_{s}$ (10$^{19.6}$eV for Auger) in 10 years of operation, of which about 1,000 events will be SD-FD hybrid. Protons and iron are the natural nuclear species to compose a cutoff structure, due to their expected abundance at the acceleration site and their comparative stability in the subsequent propagation in the nearby ($\sim$100Mpc) universe. Indeed propagation calculations indicate that cosmic rays above energies of $10^{19.5}$eV will have a simplified, approximately bimodal arrival composition, even if they are produced in equal proportions from protons to iron at the source. Intermediate mass nuclei will appear mostly as proton and He spallation by-products. Thanks to the high statistics, the improved [$X_{\rm max}$]{} resolution and the additional N$_\mu$ information of the FGO hybrid events, we expect that contributions of protons and iron will clearly stand out in a [$X_{\rm max}$]{} $-$ $N_\mu$ scatter plot. Protons or iron at the cutoff will be a straightforward confirmation of the existence of the corresponding astrophysical mechanism that creates the strong suppression, either the GZK or the acceleration limit scenarios.
If protons and iron were both identified in the hybrid sample it would allow the measured estimators of composition, [$X_{\rm max}$]{} and N$_\mu$, and their predictions by the simulation code, to be “calibrated” by the observation. Even when contributions of He, CNO and heavier nuclei are significant (and the isolation of protons and iron is not obvious), we still expect proton and iron contribution, because the existence of He results in (spallation) protons, and the existence of CNO calls for the parent Fe of the spallation (see Sections \[Sect3:using\_xmax\] and \[Sect3:lack\_of\_iron\] for a discussion). The statistics of the FGO hybrid sample, 1,000 events above $E_{1/2}$ or $E_{s}$, would allow a reasonable “calibration” or cross-check to be performed for compositions in the range of protons (Z=1) to iron (Z=26).
[**Composition dependent anisotropy and energy spectrum:**]{} The SD and RD events of the FGO are tagged with a likely primary mass derived from the [$X_{\rm max}$]{} and N$_\mu$ analysis, both of which are being calibrated using hybrid events. The statistics of these events, 10,000 or more in total above the flux suppression, is enough to allow the flux, energy spectrum and composition of UHECR to be separately determined in $\sim$100 different sections of the sky. Their correlations, such as the “proton/iron sky above a certain energy” and the “energy spectrum of proton/He/CNO/iron in particular sections of the sky", can be plotted from a single unified event sample. This will be very effective in establishing astrophysical models to explain the observed features of UHECR. Searches for auto-correlation and association with astrophysical sources, as well as the multi-messenger analysis, will be effectively made using the tag of primary composition.
As a result, we can continue investigating the nature of galactic and extra-galactic magnetic fields, background photons in the universe, cosmological development of UHECR sources, special relativity with exceptionally high Lorentz factors, and other subjects in astro-particle physics.
[**Cosmogenic $\nu$s and $\gamma$s:**]{} The search for UHE neutrinos and gamma rays by the FGO will be limited only by statistics, using the primary composition tagging of the FGO/SD. The sensitivity to cosmological neutrinos and gamma rays will allow us to enter the region of possible detection, or of placing significant limits on standard predictions (see Section \[Sect2:photonsNeutrinos\]).
[**UHE interactions:**]{} Our understanding of ultra-high energy air showers is incomplete; the data from present detectors indicate that the number of shower particles in the off-core region is larger than what the simulation program predicts, or that the simulated air shower is “slimmer” than the real one (see Section \[Sect2:deficiencies\]). Using a large collecting area and high sensitivity for penetrating particles, this is most clearly demonstrated for muons detected in the Auger water-Cherenkov stations. The difference between the data and the simulation remains after the ambiguities from energy determination and primary composition are removed, and updated hadronic interaction models with the LHC data are used [@Auger_Hadron2016]. Using composition-tagged SD events of the FGO, the lateral distribution of muons and electrons in the off-core region, and its relation to the primary energy and composition will be studied. The measurements are to be compared with a variety of model predictions with the highest energy LHC data, including taking into account nucleus-nucleus collisions.
In the case where a certain region of the sky is identified as protonic without significant contributions of heavier components, the ankle and the cutoff features in this region could possibly be attributed to the pair-production on the CMB and the GZK effect, and the corresponding energies can be used for calibrating the energy scale of the incoming protons. While it is possible that there may still be a contribution from galactic protons, this can be checked by examining the ratio of the ankle to GZK energies and the Berezinsky modification factor [@Aloisio2007] as a function of light/heavy anisotropy. In any case, the possible anisotropies of these ratios would be of great importance in constraining cosmic ray origin and propagation models. Given that these features are able to be associated with pair production and the GZK cut-off, then the difference between the expected primary energy and the measured calorimetric energy by the FD or RD is to be accounted for by the “invisible energy” carried underground by the very high energy muons and neutrinos in the shower core region. In this way, we expect ultra-high energy air showers will remain as a source of observational information for the study of the nature of hadronic and nuclear interactions beyond collider energies.
[**FD:**]{} Measuring the energy and composition of UHECR will remain a basic mission of the FGO. The FGO/FD does this by covering the entire acceptance, but with a limited duty cycle of $\sim$10 %. The FGO/RD would cover a limited acceptance at large $\theta$ but with a duty cycle higher than 90%, and its eventual contribution may become significant. When working in hybrid mode with the SD, some of the FD information is redundant. This leads to the FAST concept of deploying an array of compact, wide-angle and essentially single pixel FD telescopes that record the time development of air fluorescence in multiple stations [@fast_design]. The reconstruction of the shower core location and arrival direction may be achieved mainly by using the SD information, and the time variation of the FAST signal is then converted into the longitudinal development of the shower. The FAST detector would work exclusively for supplying the calorimetric energy and [$X_{\rm max}$]{} information of the event. Optimisation studies have been performed with the goal of obtaining good resolution with limited photon statistics. Controlling the effect of background photons on-line for the trigger and data acquisition remains a technical challenge. A small FD telescope with a similar concept has been tested by the CRAFFT team [@crafft_test]. The current design of FAST assumes FD stations with 360$^\circ$ azimuthal coverage on a rectangular grid of 20km separation. A total of 75$\times$2 FD stations will be necessary to cover the entire FGO acceptance in the northern and southern hemispheres.
[**SD:**]{} The FGO/SD is expected to have a good particle identification capability for shower particles. The isolation of muons will be of particular importance, and various types of detectors have been tested during the design study of AugerPrime using the existing water-Cherenkov station as a bulk muon counter and absorber of the EM component. Here we remind the reader of another example, the “lead burger”: a sandwich of segmented scintillator and lead absorber, tested in the AGASA array as a candidate for the original Auger SD detector [@hashimoto_icrc1995; @lead_burger]. Advances in photo-detectors and electronics may now allow a significantly finer detector segmentation and fast waveform sampling, strengthening the multi-hit capability of the lead burger [@mu_ray; @nonaka_uhecr2013; @peters_uhecr2016]. dE/dX measurement and coarse tracking of individual particles may also be incorporated. Besides muon identification, detection of spallation neutrons may be possible for tagging the nuclear composition and identifying primary gamma rays. TA and Auger are ideal testing grounds for developing the FGO/SD and for optimising its performance. Taking the 2.1km grid spacing of TA$\times$4 as an example, 6,900$\times$2 SD units will be necessary to fill the FGO acceptance.
[**Electronics and Network:**]{} The FGO electronics may follow the base design of Auger and TA with FADC sampling of wave-forms, multi-level digital triggers and wireless communication networks. For the FGO/SD, the number of readout channels must be significantly increased for the segmented detector and for the integration of FGO/RD, and faster digitisation is required for better timing measurement. The biggest challenge would be that all these performance upgrades need to be realised with a limited power budget due to local electricity generation and storage. Taking advantage of the low duty cycle of SD digitisation electronics, clever methods could be invented to save electric power consumption.
Reliability and fault-tolerance are required for the stable operation of many FGO/SDs distributed over a large ground area. With a steady increase of locally available computing and data storage capacities, real-time requirements for trigger generation and data acquisition may be loosened, and greater autonomy may be allowed for the operation of individual SDs. This in turn reduces the network load of communications between SDs, and increases the overall system reliability. It may also allow for the whole communication system of the FGO to reside on a standard communication network. This will be advantageous in terms of construction cost, long-term operation and for taking advantage of progress in communication technologies and updates in the DAQ system. Implementing a prompt coincidence trigger formed over several clustered SDs, assuming it is required, may be a technical challenge in building such an autonomous DAQ system.
For the FGO/FD array, the load on digitisation electronics will decrease compared with the current many-channel system, but triggering on a limited number of sensor pixels will be itself a challenging task. A clever time coincidence method between the neighboring SDs and FDs, or the introduction of an external trigger via the network, may be a resolution to this problem. Reliable remote operation of the telescopes, monitoring and calibration devices via the wireless network is the key to the success of achieving good operational performance of the FGO.
[**Collaboration:**]{} The construction and operation of the FGO will be a challenge for technology, management and resources. It may be accomplished only through a collaboration of people with zeal, having excellent expertise and experience. Some of the features of the FGO detector, electronics and communication system were already discussed in the proposals for northern Auger and AugerPrime [@AugerNorth; @AugerPrime]. The physics issues, and an experiment of similar sensitivity to the FGO known as “TA2”, have been discussed in physics community meetings in Japan.
Conclusion
==========
The era of the very large observatories has produced many new and important insights into the properties of the ultra-high energy cosmic ray flux. This has been achieved with well designed detectors and very large collecting areas, with important design input from previous generations of experiments. An important feature of both Auger and TA is the hybrid nature of the observations. Apart from providing calorimetric energy measurements, a hybrid observatory offers a multitude of cross-checks which have improved the measurements and reduced the systematic uncertainties.
The UHECR energy spectrum is now measured with high statistics, resulting from a combined Auger/TA exposure of over 60,000km$^2$sryr at the highest energies. The spectrum now reveals several features including an unambiguous suppression beyond $4\times10^{19}$eV. A dipole anisotropy has been observed for the first time at ultra-high energies, and there is great interest in the possible northern hemisphere “hot spot” over a $20^\circ$ radius area of sky which will be monitored by TA$\times$4 for an increase in significance in coming years.
Small-scale anisotropies and associations of cosmic ray arrival directions with astronomical catalogs have not been convincingly observed. It is probably fair to say that this lack of success was not expected. This may be related to an apparent increase in the cosmic ray mass and charge above the ankle of the energy spectrum. While not currently embraced by the entire community, a heavier flux would help to explain the lack of small-scale anisotropy. On the other hand, if the northern hemisphere “hot spot” persists, its appearance may be related to mass composition differences between the north and the south. The difference of the flux suppression energies between Auger and TA may also suggest that the astrophysics is not identical in both hemispheres. In that context, a composition dependent anisotropy study will be of great interest. The promising aspect is that our measurements of air shower development, whether they be from FDs or SDs, are continuing to improve, with the likely future outcome being the ability to (at least) identify the lighter fraction of the flux with a surface detector, 24 hours per day, as being realised by AugerPrime.
Photon and neutrino limits set by the experiments have ruled out certain exotic production mechanisms for the highest energy cosmic rays, including the decay of super-heavy dark matter. They are now probing cosmogenic photon and neutrino models that may provide information on the fraction of protons present at the highest energies. Finally, measurements at the observatories are constraining some aspects of hadronic and nuclear interaction models at very high energy, a remarkable feat given the indirect view of the interactions afforded by characteristics of the air shower.
As well as these achievements, we have also discussed the present challenges for the field. For example, despite much progress in measurements at accelerators, and despite constraints from ultra-energetic cosmic rays, we do not know the systematic uncertainty to attach to a simulation prediction for the mean [$X_{\rm max}$]{} of a proton (or any) shower, reducing the power of our mass measurements and their influence on the open astrophysics questions. It also means that we are currently limited in being able to confidently select showers initiated by low charge primaries in our attempts to improve the sensitivity to anisotropies, especially since we do not yet have a mass estimate for every event we observe. Finally, we still lack sufficient collecting area to answer some of the big questions, given that we appear to be faced with a cosmic ray sky remarkably free of strong anisotropies at the highest energies.
For the future, we must build on this impressive progress with new ideas and techniques that will lead to new observatories that are sufficiently large (and cost-effective) to answer the remaining questions. While further increases in collecting area are of prime importance, it seems crucial that future ground observatories be endowed with at least some mass-composition sensitivity for all collected events.
Acknowledgment {#acknowledgment .unnumbered}
==============
We thank all of our colleagues in the Pierre Auger and Telescope Array experiments for their inspiring collaboration. While we have benefitted greatly from discussions with many of them over the years, the responsibility for any errors in this article is entirely ours.
[100]{}
J. F. [Ormes]{}, editor, , volume 1516 of [*American Institute of Physics Conference Series*]{} (2013).
K.-H. [Kampert]{} and A. A. [Watson]{}, Eur. Phys. J. H, [**37**]{}, 359–412 (2012), [[arXiv:1207.4827]{}]{}.
M. [Nagano]{} and A. A. [Watson]{}, Rev. Mod. Phys., [**72**]{}, 689–732 (2000).
J. [Linsley]{}, Proc. 8th Int. Cosmic Ray Conf., Jaipur, [**4**]{}, 77 (1963).
H. R. [Allan]{}, R. F. W. [Beamish]{}, W. M. [Glencross]{}, D. M. [Thomson]{}, and R. D. [Wills]{}, Proc. Phys. Soc. London, [**79**]{}, 1170–1182 (1962).
C. B. A. McCusker and M. M. Winn, Il Nuovo Cimento, [**28**]{}(1), 175–178 (1963).
C. J. Bell, A. D. Bray, S. A. David, et al., J. Phys. A, [**7**]{}(8), 990 (1974).
B.N. Afanasiev et al., , In M. [Nagano]{}, editor, [*Tokyo Workshop on Techniques for the Study of EHE Cosmic Rays*]{}, page 35. University of Tokyo (1993).
M. [Nagano]{}, T. [Hara]{}, Y. [Hatano]{}, et al., J. Phys. G: Nucl. Phys., [**10**]{}, 1295–1310 (1984).
H. E. [Bergeson]{}, G. L. [Cassiday]{}, T.-W. [Chiu]{}, et al., Phys. Rev. Lett., [**39**]{}, 847–849 (1977).
R. M. [Baltrusaitis]{}, R. [Cady]{}, G. L. [Cassiday]{}, et al., Nucl. Instrum. Methods Phys. Res., Sect. A, [**240**]{}, 410–428 (1985).
G. [Thomson]{}, Nucl. Phys. B Proc. Suppl., [**136**]{}, 28–33 (2004).
J. [Abraham]{}, M. [Aglietta]{}, I. C. [Aguirre]{}, et al., Nucl. Instrum. Methods Phys. Res., Sect. A, [**523**]{}, 50–95 (2004).
P. [Sokolsky]{} and G. B. [Thomson]{}, J. Phys. G: Nucl. Phys., [**34**]{}, R401–R429 (2007), [[arXiv:0706.1248]{}]{}.
R. U. Abbasi et al., Phys. Lett. B, [**619**]{}, 271–280 (2005), [[arXiv:astro-ph/0501317]{}]{}.
R. U. [Abbasi]{}, T. [Abu-Zayyad]{}, M. [Allen]{}, et al., Phys. Rev. Lett., [ **100**]{}(10), 101101 (2008), [[astro-ph/0703099]{}]{}.
J. [Abraham]{}, P. [Abreu]{}, M. [Aglietta]{}, et al., Phys. Rev. Lett., [ **101**]{}(6), 061101 (2008), [[arXiv:0806.4302]{}]{}.
K. Greisen, Phys. Rev. Lett., [**16**]{}, 748 (1966).
G.T. Zatsepin and V.A. Kuz’min, JETP Lett., [**4**]{}, 78–80 (1966).
D. J. [Bird]{}, S. C. [Corbat[ó]{}]{}, H. Y. [Dai]{}, et al., Phys. Rev. Lett., [ **71**]{}, 3401–3404 (1993).
R. U. [Abbasi]{}, T. [Abu-Zayyad]{}, M. [Al-Seady]{}, et al., Phys. Rev. Lett., [ **104**]{}(16), 161101 (2010), [[arXiv:0910.4184]{}]{}.
R. U. [Abbasi]{}, T. [Abu-Zayyad]{}, M. [Al-Seady]{}, et al., Astropart. Phys., [ **32**]{}, 53–60 (2009), [[arXiv:0904.4500]{}]{}.
T. [Abu-Zayyad]{}, K. [Belov]{}, D. J. [Bird]{}, et al., Phys. Rev. Lett., [**84**]{}, 4276–4279 (2000), [[astro-ph/9911144]{}]{}.
T. [Abu-Zayyad]{}, K. [Belov]{}, D. J. [Bird]{}, et al., Astropart. Phys., [**16**]{}, 1–11 (2001), [[astro-ph/0008206]{}]{}.
R. U. [Abbasi]{}, M. [Abe]{}, T. [Abu-Zayyad]{}, et al., Astropart. Phys., [**64**]{}, 49–62 (2015), [[arXiv:1408.1726]{}]{}.
J. [Abraham]{}, P. [Abreu]{}, M. [Aglietta]{}, et al., Phys. Rev. Lett., [ **104**]{}(9), 091101 (2010), [[arXiv:1002.0699]{}]{}.
A. Aab, P. Abreu, M. Aglietta, et al., Phys. Rev. D, [**90**]{}, 122005 (2014).
R. [Engel]{}, D. [Heck]{}, and T. [Pierog]{}, Annu. Rev. Nucl. Part. Science, [ **61**]{}, 467–489 (2011).
D. [d’Enterria]{}, R. [Engel]{}, T. [Pierog]{}, S. [Ostapchenko]{}, and K. [Werner]{}, Astropart. Phys., [**35**]{}, 98–113 (2011), [[arXiv:1101.5596]{}]{}.
A. Aab, P. Abreu, M. Aglietta, et al., Phys. Rev. D, [**90**]{}, 122006 (2014).
A. A. [Ivanov]{}, Astrophys. J., [**804**]{}, 122 (2015), [[arXiv:1502.07496]{}]{}.
N. [Chiba]{} et al., Nucl. Instrum. Methods in Phys. Res., Sect. A, [**311**]{}, 338–349 (1992).
K. [Shinozaki]{}, M. [Teshima]{}, and [AGASA Collaboration]{}, Nucl. Phys. B Proc. Suppl., [**136**]{}, 18–27 (2004).
M. [Takeda]{}, N. [Sakaki]{}, K. [Honda]{}, et al., Astropart. Phys., [**19**]{}, 447–462 (2003), [[astro-ph/0209422]{}]{}.
S. [Yoshida]{} and H. [Dai]{}, J. Phys. G: Nucl. Phys., [**24**]{}, 905–938 (1998), [[astro-ph/9802294]{}]{}.
T. [Abu-Zayyad]{}, M. [Al-Seady]{}, K. [Belov]{}, et al., Nucl. Instrum. Methods Phys. Res., Sect. A, [**450**]{}, 253–269 (2000).
A. Aab et al., Nucl. Instrum. Methods Phys. Res., Sect. A, [**798**]{}, 172–213 (2015).
D. Ivanov et al., Proc. 34th Int. Cosmic Ray Conf., The Hague, Proc. of Science, [**ICRC2015**]{}, 349 (2016).
A. A. [Watson]{}, Astropart. Phys., [**53**]{}, 107–114 (2014).
P. [Sommers]{}, Astropart. Phys., [**3**]{}, 349–360 (1995).
B. R. [Dawson]{}, H. Y. [Dai]{}, P. [Sommers]{}, and S. [Yoshida]{}, Astropart. Phys., [**5**]{}, 239–247 (1996).
C. Bonifazi for the [Pierre Auger Collaboration]{}, Nucl. Phys. B Proc. Suppl., [**190**]{}, 20–25 (2009), [[arXiv:0901.3138]{}]{}.
B. R. Dawson for the Pierre Auger Collaboration, Proc. 30th Int. Cosmic Ray Conf., Merida, Mexico, [**4**]{}, 425–428 (2007), [[arXiv:0706.1105]{}]{}.
A. Aab, P. Abreu, M. Aglietta, et al., J. Instrum., [**12**]{}(02), P02006 (2017).
J. [Abraham]{}, P. [Abreu]{}, M. [Aglietta]{}, et al., Astropart. Phys., [**33**]{}, 108–129 (2010), [[arXiv:1002.0366]{}]{}.
J. [Bl[ü]{}mer]{} and [Pierre Auger Collaboration]{}, New J. Phys., [**12**]{}(3), 035001 (2010).
A. [Aab]{}, P. [Abreu]{}, M. [Aglietta]{}, et al., ArXiv e-prints (2016), [[arXiv:1604.03637]{}]{}.
S. Aiso et al., The [Telescope Array Project]{}, In [*Proc. of the 25th Int. Cosmic Ray Conf., Durban*]{} (1997).
The Telescope Array Collaboration, as an extremely high energy cosmic ray detector, In [*Proc. of the 25th Int. Cosmic Ray Conf., Durban*]{} (1997).
The TA Collaboration, The [Telescope Array Project Design Report]{} (July 19th, 2000).
Y. Arai et al., The [Telescope Array]{} experiment: An overview and physics aims, In [*Proc. of the 28th Int. Cosmic Ray Conf., Tsukuba*]{}, pages 1025–1028 (2003).
T. [Abu-Zayyad]{}, R. [Aida]{}, M. [Allen]{}, et al., Nucl. Instrum. Methods Phys. Res., Sect. A, [**689**]{}, 87–97 (2012), [[arXiv:1201.4964]{}]{}.
H. [Tokuno]{}, Y. [Tameda]{}, M. [Takeda]{}, et al., Nucl. Instrum. Methods Phys. Res., Sect. A, [**676**]{}, 54–65 (2012), [[arXiv:1201.0002]{}]{}.
Y. [Tameda]{}, A. [Taketa]{}, J. D. [Smith]{}, et al., Nucl. Instrum. Methods Phys. Res., Sect. A, [**609**]{}, 227–234 (2009).
P. [Tinyakov]{}, Nucl. Instrum. Methods Phys. Res., Sect. A, [**742**]{}, 29–34 (2014).
T. Tomida et al., Nucl. Instrum. Methods Phys. Res., Sect. A, [**654**]{}, 653–660 (2011).
T. [Tomida]{}, M. [Chikawa]{}, M. [Fukushima]{}, et al., , In [*European Physical Journal Web of Conferences*]{}, volume 53 of [*European Physical Journal Web of Conferences*]{}, page 10003 (2013).
Z. [Zundel]{} et al., Proc. 34th Int. Cosmic Ray Conf., The Hague, Proc. of Science, [**ICRC2015**]{}, 445 (2016).
S. Kawana et al., Nucl. Instrum. Methods Phys. Res., Sect. A, [**681**]{}, 68–77 (2012).
B.K. Shin et al., Nucl. Instrum. Methods Phys. Res., Sect. A, [**768**]{}, 96–103 (2014).
H. [Tokuno]{}, Y. [Murano]{}, S. [Kawana]{}, et al., Nucl. Instrum. Methods Phys. Res., Sect. A, [**601**]{}, 364–371 (2009).
T. Shibata et al., Nucl. Instrum. Methods Phys. Res., Sect. A, [**597**]{}, 61–66 (2001).
B. Shin et al., Proc. 34th Int. Cosmic Ray Conf., The Hague, Proc. of Science, [**ICRC2015**]{}, 640 (2016).
A.N. Bunner, , PhD thesis, Cornell University (1967).
F. [Kakimoto]{}, E. C. [Loh]{}, M. [Nagano]{}, et al., Nucl. Instrum. Methods Phys. Res., Sect. A, [**372**]{}, 527–533 (1996).
M. [Nagano]{}, K. [Kobayakawa]{}, N. [Sakaki]{}, and K. [Ando]{}, Astropart. Phys., [**22**]{}, 235–248 (2004), [[astro-ph/0406474]{}]{}.
M. [Ave]{}, M. [Bohacova]{}, E. [Curry]{}, et al., Astropart. Phys., [**42**]{}, 90–102 (2013), [[arXiv:1210.6734]{}]{}.
C. C. [Chen]{}, C. W. [Chen]{}, P. [Chen]{}, et al., Astropart. Phys., [**25**]{}, 129–139 (2006).
V. Verzi for the Pierre Auger Collaboration, [Proc. 33rd Int. Cosmic Ray Conf., Rio de Janeiro, <http://www.cbpf.br/~icrc2013/papers/icrc2013-0928.pdf>]{} (2013).
D. [Heck]{}, J. [Knapp]{}, J. N. [Capdevielle]{}, G. [Schatz]{}, and T. [Thouw]{}, , (Forschungszentrum Karlsruhe Report FZKA 6019, 1998).
T. Bergmann, R. Engel, D. Heck, et al., Astropart. Phys., [**26**]{}, 420–432 (2007), [[arXiv:0606564]{}]{}.
T. Pierog, Iu. Karpenko, J. M. Katzy, E. Yatsenko, and K. Werner, Phys. Rev. C, [**92**]{}(3), 034906 (2015), [[arXiv:1306.0121]{}]{}.
S. Ostapchenko, EPJ Web Conf., [**52**]{}, 02001 (2013).
F. Riehn, R. Engel, A. Fedynitch, T. K. Gaisser, and T. Stanev, Proc. 34th Int. Cosmic Ray Conf., The Hague, Proc. of Science, [**ICRC2015**]{}, 558 (2016).
T. [Abu-Zayyad]{}, R. [Aida]{}, M. [Allen]{}, et al., Astropart. Phys., [**61**]{}, 93–101 (2015).
I. Valino for the Pierre Auger Collaboration, Proc. 34th Int. Cosmic Ray Conf., The Hague, Proc. of Science, [**ICRC2015**]{}, 271 (2016).
V. Berezinsky, A.Z. Gazizov, and S.I. Grigorieva, Phys. Lett. B, [ **612**]{}(3–4), 147 – 153 (2005).
V. Verzi et al., , In [*[Proceedings, UHECR 2016, Kyoto, Japan, 2016]{}*]{} (in preparation 2016).
V. [Berezinsky]{}, A. [Gazizov]{}, and S. [Grigorieva]{}, Phys. Rev. D, [**74**]{}(4), 043005 (2006), [[hep-ph/0204357]{}]{}.
V. Verzi, D. Ivanov, and Y. Tsunesada, Progr. Theor. Exp. Physics, [**[this issue]{}**]{} (2017).
E. Kido, O. Kalashev, et al., Proc. 34th Int. Cosmic Ray Conf., The Hague, Proc. of Science, [**ICRC2015**]{}, 258 (2016).
A. di Matteo for the Pierre Auger Collaboration, Proc. 34th Int. Cosmic Ray Conf., The Hague, Proc. of Science, [**ICRC2015**]{}, 249 (2016).
M. S. Pshirkov, P. G. Tinyakov, P. P. Kronberg, and K. J. Newton-McGee, Astrophys. J., [**738**]{}(2), 192 (2011).
R. Jansson and G. R. Farrar, Astrophys. J., [**757**]{}(1), 14 (2012).
I. Al Samarai for the Pierre Auger Collaboration, Proc. 34th Int. Cosmic Ray Conf., The Hague, Proc. of Science, [**ICRC2015**]{}, 372 (2016).
D. M. [Edge]{}, A. M. T. [Pollock]{}, R. J. O. [Reid]{}, A. A. [Watson]{}, and J. G. [Wilson]{}, J. Phys. G: Nucl. Phys., [**4**]{}, 133–157 (1978).
A. [Aab]{}, P. [Abreu]{}, M. [Aglietta]{}, et al., Astrophys. J., [**802**]{}, 111 (2015), [[arXiv:1411.6953]{}]{}.
R. U. [Abbasi]{}, M. [Abe]{}, T. [Abu-Zayyad]{}, et al., Astrophys. J., [**790**]{}, L21 (2014), [[arXiv:1404.5890]{}]{}.
R. Aloisio, V. Berezinsky, and A. Gazizov, Astropart. Phys., [**39–40**]{}, 129 – 143, Cosmic Rays Topical Issue (2012).
P. [Abreu]{}, M. [Aglietta]{}, M. [Ahlers]{}, et al., Astrophys. J. (Suppl.), [ **203**]{}, 34 (2012), [[arXiv:1210.3736]{}]{}.
R. U. [Abbasi]{}, M. [Abe]{}, T. [Abu-Zayyad]{}, et al., Astropart. Phys., [**86**]{}, 21–26 (2017), [[arXiv:1608.06306]{}]{}.
A. [Aab]{}, P. [Abreu]{}, M. [Aglietta]{}, et al., Astrophys. J., [**794**]{}, 172 (2014), [[arXiv:1409.3128]{}]{}.
O. Deligny for the Pierre Auger and Telescope Array Collaborations, Proc. 34th Int. Cosmic Ray Conf., The Hague, Proc. of Science, [**ICRC2015**]{}, 395 (2016).
D. [Allard]{}, Astropart. Phys., [**39**]{}, 33–43 (2012), [[arXiv:1111.3290]{}]{}.
A. [Aab]{}, P. [Abreu]{}, M. [Aglietta]{}, et al., Astrophys. J., [**804**]{}, 15 (2015), [[arXiv:1411.6111]{}]{}.
T. [Abu-Zayyad]{}, R. [Aida]{}, M. [Allen]{}, et al., Astrophys. J., [**757**]{}, 26 (2012), [[arXiv:1205.5984]{}]{}.
T. [Abu-Zayyad]{}, R. [Aida]{}, M. [Allen]{}, et al., Astrophys. J., [**777**]{}, 88 (2013), [[arXiv:1306.5808]{}]{}.
Tinyakov P. and The Telescope Array Collaboration, , In [*[Proceedings, UHECR 2016, Kyoto, Japan, 2016]{}*]{} (in preparation 2016).
P. Abreu et al., J. Cosmol. Astropart. Phys., [**2013**]{}(05), 009 (2013).
J. [Matthews]{}, Astropart. Phys., [**22**]{}, 387–397 (2005).
W. Hanlon and The Telescope Array Collaboration, , In [*[Proceedings, UHECR 2016, Kyoto, Japan, 2016]{}*]{} (in preparation 2016).
R.U. Abbasi, M. Abe, T. Abu-Zayyad, et al., Astropart. Phys., [**64**]{}, 49 – 62 (2015).
A. Porcelli for the Pierre Auger Collaboration, Proc. 34th Int. Cosmic Ray Conf., The Hague, Proc. of Science, [**ICRC2015**]{}, 420 (2016).
T. Stroman, Y. Tameda, et al., Proc. 34th Int. Cosmic Ray Conf., The Hague, Proc. of Science, [**ICRC2015**]{}, 361 (2016).
E. Barcikowski, J. Bellido, J. Belz, et al., EPJ Web Conf., [**53**]{}, 01006 (2013), [[arXiv:1306.4430]{}]{}.
R. Abbasi, J. Bellido, J. Belz, et al., Report of the working group on the composition of ultra high energy cosmic rays, In [*Proceedings of International Symposium for Ultra-High Energy Cosmic Rays (UHECR2014)*]{}. The Physical Society of Japan (2016), [[http://journals.jps.jp/doi/pdf/10.7566/JPSCP.9.010016]{}]{}.
J. Bellido, J. Belz, S. Blaess, et al., , In [*[Proceedings, UHECR 2016, Kyoto, Japan, 2016]{}*]{} (in preparation 2016).
A. [Aab]{}, P. [Abreu]{}, M. [Aglietta]{}, et al., Phys. Rev. D, [**93**]{}(7), 072006 (2016).
A. [Aab]{}, P. [Abreu]{}, M. [Aglietta]{}, et al., Phys. Rev. D, [**90**]{}(1), 012012 (2014), [[arXiv:1407.5919]{}]{}.
L. Collica for the Pierre Auger Collaboration, Eur. Phys. J. Plus, [**131**]{}, 301 (2016), [[arXiv:1609.02498]{}]{}.
A. [Aab]{}, P. [Abreu]{}, M. [Aglietta]{}, et al., Phys. Lett. B, [**762**]{}, 288–295 (2016), [[arXiv:1609.08567]{}]{}.
G. B. [Gelmini]{}, O. E. [Kalashev]{}, and D. V. [Semikoz]{}, J. Exp. Theor. Phys., [**106**]{}, 1061–1082 (2008), [[astro-ph/0506128]{}]{}.
B. [Sarkar]{}, K.-H. [Kampert]{}, et al., Proc. 32nd Int. Cosmic Ray Conf., Beijing,, [**2**]{}, 198 (2011).
J. [Ellis]{}, V. E. [Mayes]{}, and D. V. [Nanopoulos]{}, Phys. Rev. D, [**74**]{}(11), 115003 (2006), [[astro-ph/0512303]{}]{}.
A. [Aab]{}, P. [Abreu]{}, M. [Aglietta]{}, et al., Submitted to J. Cosmol. Astropart. Phys. (2016), [[arXiv:1612.01517]{}]{}.
G. Rubtsov et al., Proc. 34th Int. Cosmic Ray Conf., The Hague, Proc. of Science, [**ICRC2015**]{}, 331 (2016).
A. [Aab]{}, P. [Abreu]{}, M. [Aglietta]{}, et al., Phys. Rev. D, [**91**]{}(9), 092008 (2015), [[arXiv:1504.05397]{}]{}.
G. I. [Rubtsov]{}, M. [Fukushima]{}, D. [Ivanov]{}, et al., , In [*European Physical Journal Web of Conferences*]{}, volume 53 of [*European Physical Journal Web of Conferences*]{}, page 05001 (2013).
C. Bleve for the Pierre Auger Collaboration, Proc. 34th Int. Cosmic Ray Conf., The Hague, Proc. of Science, [**ICRC2015**]{}, 1103 (2016).
M. G. Aartsen et al., Phys. Rev. Lett., [**117**]{}(24), 241101 (2016), [[arXiv:1607.05886]{}]{}.
, [Pierre Auger Collaboration]{}, and [Telescope Array Collaboration]{}, J. Cosmol. Astropart. Phys., [**1**]{}, 037 (2016), [[arXiv:1511.09408]{}]{}.
A. [Aab]{}, P. [Abreu]{}, M. [Aglietta]{}, et al., Phys. Rev. D. (in press) (2016), [[arXiv:1608.07378]{}]{}.
M. [Honda]{}, M. [Nagano]{}, S. [Tonwar]{}, et al., Phys. Rev. Lett., [**70**]{}, 525–528 (1993).
R. M. Baltrusaitis, G. L. Cassiday, J. W. Elbert, et al., Phys. Rev. Lett., [**52**]{}, 1380–1383 (1984).
R. W. Ellsworth, T. K. Gaisser, Todor Stanev, and G. B. Yodh, Phys. Rev. D, [**26**]{}, 336–339 (1982).
K. Belov et al., Nuclear Physics B - Proceedings Supplements, [**151**]{}(1), 197 – 204 (2006).
P. [Abreu]{}, M. [Aglietta]{}, E. J. [Ahn]{}, et al., Phys. Rev. Lett., [**109**]{}(6), 062002 (2012), [[arXiv:1208.1520]{}]{}.
R. Ulrich for the Pierre Auger Collaboration, Proc. 34th Int. Cosmic Ray Conf., The Hague, Proc. of Science, [**ICRC2015**]{}, 401 (2016).
R. U. [Abbasi]{}, M. [Abe]{}, T. [Abu-Zayyad]{}, et al., Phys. Rev. D, [**92**]{}(3), 032007 (2015), [[arXiv:1505.01860]{}]{}.
T. [Abu-Zayyad]{}, R. [Aida]{}, M. [Allen]{}, et al., Ap. J. Lett., [**768**]{}, L1 (2013), [[arXiv:1205.5067]{}]{}.
A. [Aab]{} et al., Phys. Rev. D, [**91**]{}(5), 059901 (2015).
A. [Aab]{}, P. [Abreu]{}, M. [Aglietta]{}, et al., Phys. Rev. Lett., [**117**]{}(19), 192001 (2016).
R. [Takeishi]{}, , PhD thesis, University of Tokyo (2017).
J. [Linsley]{}, International Cosmic Ray Conference, [**12**]{}, 89–96 (1977).
R. U. Abbasi, T. Abu-Zayyad, G. Archbold, et al., Astrophys. J., [**622**]{}(2), 910 (2005).
W Hanlon, personal communication (2016).
J. L. [Puget]{}, F. W. [Stecker]{}, and J. H. [Bredekamp]{}, Ap. J., [**205**]{}, 638–654 (1976).
F. W. [Stecker]{} and M. H. [Salamon]{}, Ap. J., [**512**]{}, 521–526 (1999), [[astro-ph/9808110]{}]{}.
A. [Aab]{}, P. [Abreu]{}, M. [Aglietta]{}, et al., Submitted to J. Cosmol. Astropart. Phys. (2016), [[arXiv:1612.07155]{}]{}.
A. M. [Taylor]{}, , PhD thesis, University of Oxford (2007), https://ora.ox.ac.uk/objects/uuid:63572ebe-fb32-41b6-8b91-a7294db135a6.
D. [Hooper]{}, S. [Sarkar]{}, and A. M. [Taylor]{}, Phys. Rev. D, [**77**]{}(10), 103007 (2008), [[arXiv:0802.1538]{}]{}.
A. [De Cia]{}, C. [Ledoux]{}, L. [Mattsson]{}, et al., A & A, [**596**]{}, A97 (2016), [[arXiv:1608.08621]{}]{}.
J.-K. [Krogager]{}, J. P. U. [Fynbo]{}, P. [Noterdaeme]{}, et al., MNRAS, [**455**]{}, 2698–2711 (2016), [[arXiv:1510.04695]{}]{}.
H. [Gail]{}, Condensation of dust in astrophysical environments, In S. [Schlemmer]{}, T. [Giesen]{}, C. [Jager]{}, and H. [Mutschke]{}, editors, [*Laboratory Astrochemistry: From Molecules to Nanoparticles to Grains*]{}. Wiley –VCH (2015).
C. [Song]{}, Z. [Cao]{}, B. R. [Dawson]{}, et al., Astropart. Phys., [**14**]{}, 7–13 (2000), [[astro-ph/9910195]{}]{}.
T. Huege, Radio detection of cosmic rays - achievements and future potential, In [*[Proceedings, UHECR 2016, Kyoto, Japan, 2016]{}*]{} (in preparation 2016).
S. Buitink et al., Phys. Rev. D, [**90**]{}, 082003 (2014), [[arXiv:1408.7001]{}]{}.
P. Abreu et al., J. Instrum., [**7**]{}, P10011 (2012).
A. Aab et al., Phys. Rev. D, [**93**]{}, 122005 (2016).
T. Huege, Phys. Rep., [**620**]{}, 1–52 (2016).
J.P. Lindquist, personal communication (2016).
V. Verzi, Proc. 34th Int. Cosmic Ray Conf., The Hague, Proc. of Science, [ **ICRC2015**]{}, 015 (2016).
D. Ivanov, personal communication (2016).
M. Ave, M. Bohacova, B. Buonomo, et al., Nucl. Instrum. Methods Phys. Res., Sect. A, [**597**]{}(1), 41–54, Proceedings of the 5th Fluorescence Workshop (2008).
R. [Abbasi]{}, T. [Abu-Zayyad]{}, K. [Belov]{}, et al., Nucl. Instrum. Methods Phys. Res., Sect. A, [**597**]{}, 37–40 (2008).
R. [Abbasi]{}, T. [Abu-Zayyad]{}, K. [Belov]{}, et al., Astropart. Phys., [**29**]{}, 77–86 (2008), [[arXiv:0708.3116]{}]{}.
R. [Abbasi]{}, T. [Abu-Zayyad]{}, K. [Belov]{}, et al., Nucl. Instrum. Methods Phys. Res., Sect. A, [**597**]{}, 32–36 (2008).
P. [Colin]{}, A. [Chukanov]{}, V. [Grebenyuk]{}, et al., Astropart. Phys., [**27**]{}, 317–325 (2007), [[astro-ph/0612110]{}]{}.
P. Sokolsky, personal communication (2016).
J. [Abraham]{}, P. [Abreu]{}, M. [Aglietta]{}, et al., Science, [**318**]{}, 938 (2007), [[arXiv:0711.2256]{}]{}.
D. Ryu, A close correlation of [TA]{} hotspot events with filaments of galaxies connected to the [Virgo Cluster]{}, In [*[Proceedings, UHECR 2016, Kyoto, Japan, 2016]{}*]{} (in preparation 2016).
D. N. [Pfeffer]{}, E. D. [Kovetz]{}, and M. [Kamionkowski]{}, MNRAS, [**466**]{}, 2922–2926 (2017), [[arXiv:1512.04959]{}]{}.
H.-N. [He]{}, A. [Kusenko]{}, S. [Nagataki]{}, et al., Phys. Rev. D, [**93**]{}(4), 043011 (2016), [[arXiv:1411.5273]{}]{}.
H. Sagawa and [The Telescope Array Collaboration]{}, Proc. 34th Int. Cosmic Ray Conf., The Hague, Proc. of Science, [**ICRC2015**]{}, 657 (2016).
E. Kido, , In [*[Proceedings, UHECR 2016, Kyoto, Japan, 2016]{}*]{} (in preparation 2016).
R. Engel for the Pierre Auger Collaboration, , In [*[Proceedings, UHECR 2016, Kyoto, Japan, 2016]{}*]{} (in preparation 2016).
P. [Lipari]{}, Phys. Rev. D, [**79**]{}(6), 063001 (2009), [[arXiv:0809.0190]{}]{}.
T. Okuda et al., Proc. 34th Int. Cosmic Ray Conf., The Hague, Proc. of Science, [**ICRC2015**]{}, 298 (2016).
J. Belz, , In [*[Proceedings, UHECR 2016, Kyoto, Japan, 2016]{}*]{} (in preparation 2016).
A. Tonachini for the Pierre Auger Collaboration, [Proc. 33rd Int. Cosmic Ray Conf., Rio de Janeiro, <http://www.cbpf.br/~icrc2013/papers/icrc2013-0676.pdf>]{} (2013).
R. Colalillo for the Pierre Auger Collaboration, , In [*[Proceedings, UHECR 2016, Kyoto, Japan, 2016]{}*]{} (in preparation 2016).
S. Buitink et al., A & A, [**467**]{}, 385–394 (2007).
T. Huege, Phys. Rep., [**620**]{}, 1–52 (2016).
T. [Huege]{}, detection of cosmic rays - achievements and future potential, In [*[Proceedings, UHECR 2016, Kyoto, Japan, 2016]{}*]{} (in preparation 2016).
P.W. Gorham et al., Phys. Rev. Lett., [**103**]{}, 051103 (2009).
S. Barwick, , In [*[Proceedings, UHECR 2016, Kyoto, Japan, 2016]{}*]{} (in preparation 2016).
P.W. Gorham et al., Phys. Rev. D, [**78**]{}, 032007 (2008).
R. Smida et al., Phys. Rev. Lett., [**113**]{}, 221101 (2014).
I.S. Ohta et al., Nucl. Instrum. Methods Phys. Res., Sect. A, [**810**]{}, 44–50 (2016).
R. U. [Abbasi]{} et al., Astropart. Phys., [**87**]{}, 1–17 (2017), [[arXiv:1603.05217]{}]{}.
L. Scarsi et al., , In [*Proposal for the ESA F2/F3 Missions*]{} (2000).
J.H. Adams et al., Astropart. Phys., [**44**]{}, 76–90 (2013).
J.H. Adams et al., Exp. Astron., [**40**]{}, 183–214 (2015).
M. Casolino, The [JEM-EUSO]{} program to study [UHECR]{} from space, In [*[Proceedings, UHECR 2016, Kyoto, Japan, 2016]{}*]{} (in preparation 2016).
M.Yu. Zotov, The first results of the [TUS]{} orbital detector of ultra high energy cosmic rays and transient luminous events in the atmosphere, In [*[Proceedings, UHECR 2016, Kyoto, Japan, 2016]{}*]{} (in preparation 2016).
R. [Aloisio]{}, V. [Berezinsky]{}, P. [Blasi]{}, et al., Astropart. Phys., [**27**]{}, 76–91 (2007), [[astro-ph/0608219]{}]{}.
T. Fujii et al., Astropart. Phys., [**74**]{}, 64–72 (2016).
Y. Tameda et al., Development of the cosmic ray air fluorescence fresnel lens telescope for a next generation [UHECR]{} observatory, In [*[Proceedings, UHECR 2016, Kyoto, Japan, 2016]{}*]{} (in preparation 2016).
K. [Hashimoto]{}, K. [Honda]{}, N. [Kawasumi]{}, et al., , In [*[Proc. 24th Int. Cosmic Ray Conf. Rome]{}*]{} (1995).
K. Honda et al., Phys. Rev. D, [**56**]{}, 3833 (1997).
A. Anastasio et al., Nucl. Instrum. Methods Phys. Res., Sect. A, [**718**]{}, 134–137 (2013).
T. Nonaka et al. for the Telescope Array Collaboration, , In [*[Proceedings of UHECR 2014, Springdale, USA, October 12-15, 2014, [JPS Conf. Proc. [**9**]{}, 010013]{}]{}*]{} (2016).
C. Peters, Prospects of silicon photomultipliers for ground-based cosmic ray experiments, In [*[Proceedings, UHECR 2016, Kyoto, Japan, 2016]{}*]{} (in preparation 2016).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We study static, spherically symmetric vacuum solutions to Quadratic Gravity, extending considerably our previous Rapid Communication \[Phys. Rev. D 98, 021502(R) (2018)\] on this topic. Using a conformal-to-Kundt metric ansatz, we arrive at a much simpler form of the field equations in comparison with their expression in the standard spherically symmetric coordinates. We present details of the derivation of this compact form of two ordinary differential field equations for two metric functions. Next, we apply analytical methods and express their solutions as infinite power series expansions. We systematically derive all possible cases admitted by such an ansatz, arriving at six main classes of solutions, and provide recurrent formulas for all the series coefficients. These results allow us to identify the classes containing the Schwarzschild black hole as a special case. It turns out that one class contains only the Schwarzschild black hole, three classes admit the Schwarzschild solution as a special subcase, and two classes are not compatible with the Schwarzschild solution at all since they have strictly nonzero Bach tensor. In our analysis, we naturally focus on the classes containing the Schwarzschild spacetime, in particular on a new family of the Schwarzschild–Bach black holes which possesses one additional non-Schwarzschild parameter corresponding to the value of the Bach tensor invariant on the horizon. We study its geometrical and physical properties, such as basic thermodynamical quantities and tidal effects on free test particles induced by the presence of the Bach tensor. We also compare our results with previous findings in the literature obtained using the standard spherically symmetric coordinates.'
author:
- |
J. Podolský$^\star$, R. Švarc$^\star$, V. Pravda$^\diamond$, A. Pravdov' a$^\diamond$\
\
\
[$^\star$ Institute of Theoretical Physics, Faculty of Mathematics and Physics,]{}\
[Charles University, V Holešovičkách 2, 180 00 Prague 8, Czech Republic.]{}\
[$^\diamond$ Institute of Mathematics, Academy of Sciences of the Czech Republic]{},\
[Žitn' a 25, 115 67 Prague 1, Czech Republic.]{}\
[E-mail: `[email protected], [email protected], `]{}\
[`[email protected], [email protected]`]{}
title: Black holes and other exact spherical solutions in Quadratic Gravity
---
PACS numbers: 04.20.Jb, 04.50.–h, 04.70.Bw, 04.70.Dy, 11.25.–w
Keywords: black holes, exact solutions, Quadratic Gravity, Einstein–Weyl gravity, Schwarzschild metric, Bach tensor, Robinson–Trautman spacetimes, Kundt spacetimes
Introduction {#intro}
============
Soon after Albert Einstein formulated his General Relativity in November 1915 and David Hilbert found an elegant procedure how to derive Einstein’s field equations from the variational principle, various attempts started to extend and generalize this gravity theory. One possible road, suggested by Theodor Kaluza exactly a century ago in 1919, was to consider higher dimensions in an attempt to unify the field theories of gravitation and electromagnetism. In the same year, another road was proposed by Hermann Weyl. In this case, the idea was to derive alternative field equations of a metric theory of gravity by starting with a different action. Instead of using the Einstein–Hilbert Lagrangian of General Relativity, which is simply the Ricci curvature scalar $R$ (a double contraction of a single Riemann tensor), Weyl proposed a Lagrangian containing *contractions of a product of two curvature tensors*. Such a Lagrangian is thus not linear in curvature — it is quadratic so that this theory can be naturally called “quadratic gravity”. Einstein was well aware of these attempts to formulate such alternative theories of gravity, and for some time he also worked on them. Interestingly, expressions for the quadratic gravity theory can be found even in his last writing pad (at the bottom of its last but one page) which he used in spring 1955.
Although it turned out rather quickly that these original classical theories extending General Relativity led to specific conceptual, mathematical and physical problems, the nice ideas have been so appealing that — the whole century after their conception — they are still very actively investigated. Both the higher dimensions of the Kaluza–Klein theory and Weyl’s higher-order curvature terms in an effective action are now incorporated into the foundations of string theory. Quadratic Gravity (QG) also plays an important role in contemporary studies of relativistic quantum field theories.
Quadratic Gravity is a very natural and quite “conservative” extension of the Einstein theory, the most precise gravity theory today. Quadratic terms in the QG Lagrangian can be understood as corrections to General Relativity, which may play a crucial role at extremely high energies. In the search for a consistent quantum gravity theory, which could be applicable near the Big Bang or near spacetime singularities inside black holes, it is important to understand the role of these higher-order curvature corrections.
Interestingly, it was suggested by Weinberg and Deser, and then proved by Stelle [@Stelle:77] already in the 1970s that adding the terms quadratic in the curvature to the Einstein–Hilbert action renders gravity renormalizable, see the very recent review [@Salvio]. This property is also preserved in the general coupling with a generic quantum field theory. However, due to the presence of higher derivatives, “massive ghosts” also appear (the corresponding classical Hamiltonian is unbounded from below). Nevertheless, there is a possibility that these ghosts could be benign [@Smilga]. For all these reasons, this QG theory has attracted considerable attention in recent years.
In our work, we are interested in *classical solutions to QG in four dimensions*. It can be easily shown that all Einstein spacetimes obey the vacuum field equations of this theory. However, QG also admits additional vacuum solutions with nontrivial Ricci tensor. In this paper, we focus on such *static, spherically symmetric vacuum solutions* without a cosmological constant. They were first studied in the seminal work [@Stelle:1978], in which three families of such spacetimes were identified by using a power expansion of the metric functions around the origin. The failure of the Birkhoff theorem in [Quadratic]{} Gravity has also been pointed out therein. Spherically symmetric solutions were further studied in [@Holdom:2002], where also numbers of free parameters for some of the above-mentioned classes were determined. Recently it has been pointed out in [@LuPerkinsPopeStelle:2015; @LuPerkinsPopeStelle:2015b; @PerkinsPhD] that, apart from the Schwarzschild black hole and other spherical solutions, QG admits a *non-Schwarzschild* spherically symmetric and static black holes.
The field equations of a generic Quadratic Gravity theory form a highly complicated system of fourth-order nonlinear PDEs. Only a few nontrivial exact solutions are thus known so far, and various approximative and numerical methods have had to be used in their studies. Specifically, in the new class of black holes presented in [@LuPerkinsPopeStelle:2015], the two unknown metric functions of the standard form of spherically symmetric metric were given in terms of two complicated coupled ODEs which were (apart from the first few orders in the power expansion) solved and analyzed numerically. Interestingly, all QG corrections to the four-dimensional vacuum Einstein equations for constant Ricci scalar are nicely combined into a conformally well-behaved Bach tensor. Together with a conformal-to-Kundt metric ansatz [@PravdaPravdovaPodolskySvarc:2017], this leads to a considerably simpler autonomous system of the field equations. We employed this approach in our recent letters [@PodolskySvarcPravdaPravdova:2018] and [@SvarcPodolskyPravdaPravdova:2018] for vanishing and nonvanishing cosmological constant, respectively. In [@PodolskySvarcPravdaPravdova:2018] we were thus able to present an explicit form of the corresponding nontrivial black-hole spacetimes — the so-called *Schwarzschild–Bach black holes* with two parameters, a position of the horizon and an additional Bach parameter. By setting this additional Bach parameter to zero, the Schwarzschild metric of General Relativity is directly recovered. In the present considerably longer paper, we are now giving the details of the derivation summarized in [@PodolskySvarcPravdaPravdova:2018], and also survey and analysis of other classes of spherically symmetric solutions to Quadratic Gravity.
Our paper is organized as follows. In Sec. \[QGandEWtheory\] we recall the Quadratic Gravity and the Einstein–Weyl theory, and we put the corresponding field equations into a convenient form in which the Ricci tensor is proportional to the Bach tensor. In Sec. \[BHmetricsec\] we introduce a suitable spherically symmetric metric ansatz in the conformal-to-Kundt form, and we give relations to the standard metric form. In Sec. \[derivingFE\] we overview the derivation of the field equations with various technical details and thorough discussion being postponed to Appendices A–C. In Sec. \[invariants\] expressions for curvature invariants are derived. In Sec. \[integration\] expansions in powers of ${\Delta \equiv r-r_0}$ around a fixed point $r_0$, and for $r \rightarrow \infty$ are introduced. In Sec. \[expansiont\_0\] the leading orders in ${\Delta }$ of the field equations are solved and four main classes of solutions are obtained. For these solutions, in Sec. \[description\] all coefficients of the metric functions in the power expansions in $\Delta$ are given in the form of recurrent formulas, convenient gauge choices are found, and various aspects of the solutions are discussed. Sections \[expansiont\_INF\] and \[description\_INF\] focus on the same topics as Secs. \[expansiont\_0\] and \[description\], respectively, but this time for expansions $r \rightarrow \infty$. In Sec. \[summary\] the relation of the solutions obtained in Secs. \[expansiont\_0\]–\[description\_INF\] (including their special subcases) to the solutions given in the literature is discussed, and summarized in Table \[tab:3\]. Mathematical and physical aspects (specific tidal effects and thermodynamical quantities) of the Schwarzschild–Bach solutions are discussed in Sections \[discussion-and-figures\] and \[physics\], respectively. Finally, concluding remarks are given in Sec. \[conclusions\].
Quadratic Gravity and the Einstein–Weyl theory {#QGandEWtheory}
==============================================
Quadratic Gravity (QG) is a natural generalization of Einstein’s theory that includes higher derivatives of the metric. Its action in four dimensions contains additional quadratic terms, namely square of the Ricci scalar $R$ and a contraction of the Weyl tensor $C_{abcd}$ with itself [@Weyl1919; @Bach1921]. In the absence of matter, the most general QG action generalizing the Einstein–Hilbert action reads[@PravdaPravdovaPodolskySvarc:2017][^1] S = [[[d]{}]{}]{}\^4 x ( (R-2) +R\^2 - C\_[abcd]{} C\^[abcd]{} ), \[actionQG\] where ${\gamma=1/G}$ ($G$ is the Newtonian constant), $\Lambda$ is the cosmological constant, and $\alpha$, $\beta$ are additional QG theory parameters. The Einstein–Weyl theory is contained as a special case by setting ${\beta=0}$.
*Vacuum field equations* corresponding to the action (\[actionQG\]) are $$\begin{aligned}
&\gamma \left(R_{ab} - {\pul} R\, g_{ab}+\Lambda\,g_{ab}\right)-4 \alpha\,B_{ab} \nonumber \\
&\quad +2\beta\left(R_{ab}-\tfrac{1}{4}R\, g_{ab}+ g_{ab}\, \Box - \nabla_b \nabla_a\right) R = 0 \,, \label{GenQGFieldEq}\end{aligned}$$ where $B_{ab}$ is the *Bach tensor* defined as B\_[ab]{} ( \^c \^d + R\^[cd]{} )C\_[acbd]{} . \[defBach\] It is traceless, symmetric, and conserved: $$g^{ab}B_{ab}=0 \,, \qquad B_{ab}=B_{ba} \,, \qquad \nabla^b B_{ab}=0 \,,
\label{Bachproperties}$$ and also conformally well-behaved (see expression (\[OmBach\]) below).
Now, *assuming* ${R=\hbox{const.}}$, the last two terms in (\[GenQGFieldEq\]) containing covariant derivatives of $R$ vanish. Using (\[Bachproperties\]), the trace of the field equations thus immediately implies R=4. \[R=4Lambda\] By substituting this relation into the field equations (\[GenQGFieldEq\]), they simplify considerably to R\_[ab]{}-g\_[ab]{}=4k B\_[ab]{}, k . \[fieldeqsgen\]
In this paper, *we restrict ourselves to* investigation of solutions with *vanishing cosmological constant* $\Lambda$ (see [@SvarcPodolskyPravdaPravdova:2018] for the study of a more general case ${\Lambda\ne0}$). In view of (\[R=4Lambda\]), this implies vanishing Ricci scalar, R=0, \[R=0\] and the field equations (\[fieldeqsgen\]) further reduce to a simpler form R\_[ab]{}=4k B\_[ab]{}, \[fieldeqsEWmod\] where the constant $k$ is now a shorthand for the combination of the theory parameters ${ k \equiv \alpha/\gamma= G\alpha}$. For ${k=0}$ we recover vacuum Einstein’s equations of General Relativity. Interestingly, all solutions of (\[fieldeqsEWmod\]) in *Einstein–Weyl gravity* (${\beta=0}$) with ${R=0}$ *are also solutions to general Quadratic Gravity* (${\beta\ne0}$) since for ${\Lambda=0}$ the QG parameter $\beta$ does not contribute to the constant $k$ defined by (\[fieldeqsgen\]).
Black hole metrics {#BHmetricsec}
==================
For studying static, nonrotating black holes, it is a common approach to employ the canonical form of a general spherically symmetric metric $${{\rm{d}}}s^2 = -h(\bar r)\,{{\rm{d}}}t^2+\frac{{{\rm{d}}}\bar r^2}{f(\bar r)}+\bar r^2({{\rm{d}}}\theta^2+\sin^2\theta\,{{\rm{d}}}\phi^2) \,.
\label{Einstein-WeylBH}$$ In particular, for the famous *Schwarzschild solution* of Einstein’s General Relativity [@Schwarzschild:1916] (and also of QG), the two metric functions *are the same* and take the well-known form $$f(\bar{r}) = h(\bar{r})=1-\frac{2m}{\bar{r}} \,.
\label{SchwarzschildBH}$$ The metric (\[Einstein-WeylBH\]) was also used in the seminal papers [@LuPerkinsPopeStelle:2015; @LuPerkinsPopeStelle:2015b] to investigate generic spherical black holes in Quadratic Gravity, in which it was surprisingly shown, mostly by numerical methods, that such a class contains further black-hole solutions *distinct* from the Schwarzschild solution (\[SchwarzschildBH\]). It turned out that while the Schwarzschild black hole has ${f=h}$, this non-Schwarzschild black hole is characterized by ${f\not=h}$. However, due to the complexity of the QG field equations (\[GenQGFieldEq\]) for the classical metric form (\[Einstein-WeylBH\]), it has not been possible to find an explicit analytic form of the metric functions ${f(\bar{r}), h(\bar{r})}$.
A new convenient metric form of the black hole geometry {#BH metric}
-------------------------------------------------------
As demonstrated in our previous works [@PodolskySvarcPravdaPravdova:2018; @SvarcPodolskyPravdaPravdova:2018], it is much more convenient to employ an *alternative metric form* of the spacetimes represented by (\[Einstein-WeylBH\]). This is obtained by performing the transformation $$\bar{r} = \Omega(r)\,, \qquad t = u - \int\! \frac{{{\rm{d}}}r}{\H(r)} \,, \label{to static}$$ resulting in [[[d]{}]{}]{}s\^2 = \^2(r) . \[BHmetric\] The two new metric functions $\Omega(r)$ and $\H(r)$ are related to $f(\bar r)$ and $h(\bar r)$ via simple relations h = -\^2 , f = -()\^2 , \[rcehf\] where prime denotes the derivative with respect to $r$. Of course, the argument $r$ of both functions $\Omega$ and $\H$ must be expressed in terms of $\bar{r}$ using the inverse of the relation ${\bar{r} = \Omega(r)}$.
The metric admits a *gauge freedom* given by a constant rescaling and a shift of $r$, r r+, u \^[-1]{}u . \[scalingfreedom\]
More importantly, this new black hole metric is *conformal* to a much simpler Kundt-type metric, [[[d]{}]{}]{}s\^2 =\^2(r)[[[d]{}]{}]{}s\^2\_. \[confrelation\] Indeed, ${{{\rm{d}}}s^2_{\hbox{\tiny Kundt}}}$ belongs to the famous class of *Kundt geometries*, which are nonexpanding, shear-free and twist-free, see [@Stephanietal:2003; @GriffithsPodolsky:2009]. In fact, it is a subclass of Kundt spacetimes which is the *direct-product of two 2-spaces*, and is of Weyl algebraic type D and Ricci type II [@GriffithsPodolsky:2009; @PravdaPravdovaPodolskySvarc:2017]. The first part of [[[d]{}]{}]{}s\^2\_=[[[d]{}]{}]{}\^2+\^2[[[d]{}]{}]{}\^2 -2[[[d]{}]{}]{}u[[[d]{}]{}]{}r+[H]{}(r)[[[d]{}]{}]{}u\^2 \[Kundt seed\] spanned by ${\theta, \phi}$ is a round 2-sphere of Gaussian curvature ${K=1}$, while the second part spanned by ${u, r}$ is a 2-dim Lorentzian spacetime. With the usual stereographic representation of a 2-sphere given by ${x+\hbox{i}\, y = 2\tan(\theta/2)\exp(\hbox{i}\phi)}$, this *Kundt seed* metric can be rewritten as [[[d]{}]{}]{}s\^2\_= -2[[[d]{}]{}]{}u[[[d]{}]{}]{}r+[H]{}(r)[[[d]{}]{}]{}u\^2 . \[Kundt seed xy\]
The black hole horizon {#BH horizon}
----------------------
In the usual metric form (\[Einstein-WeylBH\]), the Schwarzschild horizon is defined by the zeros of the same two metric functions $h{({\bar r})=f({\bar r})}$. Due to (\[SchwarzschildBH\]), it is located at ${{\bar r}_h=2m}$, where $m$ denotes the total mass of the black hole.
In a general case, such a horizon can be defined as the *Killing horizon* associated with the vector field ${\partial_t}$. Its norm is determined by the metric function $-h({\bar r})$. In the regions where ${h({\bar r})>0}$, the spacetime is static and $t$ is the corresponding temporal coordinate. The Killing horizon is generated by the *null vector field* ${\partial_t}$, and it is thus located at a specific radius ${\bar r}_h$ satisfying $$h \big|_{{\bar r}={\bar r}_h}=0\,. \label{standardhorizon}$$
In terms of the new metric form , we may similarly employ the vector field ${\partial_u}$ which coincides with ${\partial_t}$ everywhere. Its norm is given by $\Omega^2\, \H$. Since the conformal factor $\Omega$ is nonvanishing throughout the spacetime, the Killing horizon is uniquely located at a specific radius $r_h$ satisfying the condition $$\H \big|_{r=r_h}=0\,. \label{horizon}$$ Interestingly, via the relations this automatically implies ${h({\bar r_h})=0=f({\bar r_h})}$.
It is also important to recall that there is a *time-scaling freedom* of the metric tt/, \[scaling-t\] where ${\sigma \ne 0}$ is any constant, which implies ${h\to h\,\sigma^2}$. This freedom can be used to adjust an appropriate value of $h$ at a chosen radius ${\bar r}$. Or, in an asymptotically flat spacetime such as (\[SchwarzschildBH\]) it could be used to achieve ${h \to 1}$ as ${{\bar r}\to \infty}$, thus enabling us to determine the mass of a black hole.
The Kundt seed of the Schwarzschild solution {#Kundt seed of the Schwarzschild}
--------------------------------------------
It is also important to explicitly identify the Kundt seed geometry (\[Kundt seed\]) which, via the conformal relation (\[confrelation\]), generates the well-known vacuum *Schwarzschild solution*. This is simply given by $$\bar{r}=\Omega(r)=-\frac{1}{r}\,,\qquad
\H(r) = -r^2-2m\, r^3 \,.
\label{Schw}$$ Indeed, the first relation implies ${r=-1/\bar{r}}$, so that ${\H(\bar{r}) = - (1-2m/\bar{r})/\bar{r}^2}$. Using (\[rcehf\]), we easily obtain (\[SchwarzschildBH\]). It should be emphasized that the standard physical range ${\bar{r}>0}$ corresponds to ${r<0}$. Also, the auxiliary Kundt coordinate $r$ *increases from negative values to* $0$, as $\bar{r}$ increases to $\infty$.
Notice that ${\cal H}$ given by (\[Schw\]) is simply a *cubic* in the coordinate $r$ of the Kundt geometry. For ${m=0}$, the Kundt seed with ${{\cal H} = -\,r^2}$ is the Bertotti–Robinson spacetime with the geometry ${S^2\times AdS_2}$ (see chapter 7 of [@GriffithsPodolsky:2009]), and the corresponding conformally related metric (\[confrelation\]) is just the flat space. It should also be emphasized that, while the Schwarzschild and Minkowski spacetimes are (the simplest) vacuum solutions in Einstein’s theory, their Kundt seeds (\[Schw\]) *are not vacuum solutions* in Einstein’s theory since their Ricci tensor is nonvanishing. In fact, the Bertotti–Robinson geometry is an electrovacuum space of Einstein’s theory.
Since conformal transformations preserve the Weyl tensor, both ${{\rm{d}}}s^2$ and ${{\rm{d}}}s^2_{\hbox{\tiny Kundt}} $ are of the *same algebraic type*. Indeed, in the null frame ${{\mbox{\boldmath$k$}}= \mathbf{\partial}_r}$, ${{\mbox{\boldmath$l$}}= {\textstyle\frac{1}{2}}{\cal H}\,\mathbf{\partial}_r+\mathbf{\partial}_u}$, ${{\mbox{\boldmath$m$}}_i = \big(1+\ctvrt(x^2+y^2)\big)\mathbf{\partial}_i}$, the only Newman–Penrose Weyl scalar for (\[Kundt seed xy\]) is ${\Psi_2=-\frac{1}{12}({\cal H}''+2)}$, and both ${\mbox{\boldmath$k$}}$ and ${\mbox{\boldmath$l$}}$ are double principal null directions. For the specific function (\[Schw\]), ${\Psi_2=m\,r}$. The Kundt seed geometry for the Schwarzschild solution is thus of algebraic type D. It is conformally flat if, and only if, ${m=0}$, in which case it is the Bertotti–Robinson spacetime.
The Robinson–Trautman form of the black hole metrics {#RT}
----------------------------------------------------
Recently, we have proven in [@PravdaPravdovaPodolskySvarc:2017] that *any metric conformal to a Kundt geometry must belong to the class of expanding Robinson–Trautman geometries* (or it remains in the Kundt class). Indeed, performing a simple transformation ${r(\tilde r)}$ of (\[confrelation\]), (\[Kundt seed xy\]), such that $$r = \int\!\!\frac{{{\rm{d}}}\tilde r}{\Omega^2(\tilde r)}\, , \qquad
{{H}}\equiv \Omega^{2}\, \H \,,
\label{guu_RT}$$ we obtain $${{\rm{d}}}s^2_{\hbox{\tiny RT}} = \Omega^2(\tilde r)\,\frac{{{\rm{d}}}x^2 + {{\rm{d}}}y^2}{\big(1+\ctvrt(x^2+y^2)\big)^2}
-2\,{{\rm{d}}}u\,{{\rm{d}}}\tilde r+{{H}}(\tilde r)\,{{\rm{d}}}u^2 \,. \label{confRT}$$ This has the canonical form of the Robinson–Trautman class [@Stephanietal:2003; @GriffithsPodolsky:2009] with the identification \_[,r]{} = ,[[H]{}]{}= - h. The Schwarzschild black hole is recovered for ${\Omega(\tilde r)=\tilde r}$ that is ${\Omega_{,\tilde r}=1}$, equivalent to ${f(\bar{r}) = h(\bar{r})}$. Other distinct non-Schwarzschild black hole solutions are identified by ${f(\bar{r}) \ne h(\bar{r})}$. The Killing horizon is obviously given by ${{{H}}(\tilde r_h)=0}$, corresponding to ${\H(r_h)=0=h(\bar r_h)}$ and ${f(\bar r_h)=0}$.
The field equations {#derivingFE}
===================
The conformal approach to describing and studying black holes and other spherical solutions in Einstein–Weyl gravity and fully general Quadratic Gravity, based on the new form of the metric , is very convenient. Due to (\[confrelation\]), it enables to evaluate easily the Ricci and Bach tensors, entering the field equations (\[fieldeqsEWmod\]), from the Ricci and Bach tensors of the much simpler Kundt seed metric ${{{\rm{d}}}s^2_{\hbox{\tiny Kundt}}}$. In particular, to derive the explicit form of the field equations, it is possible to proceed as follows:
1. Calculate all components of the Ricci and Bach tensors $R_{ab}^{{\hbox{\tiny Kundt}}}$ and $B_{ab}^{{\hbox{\tiny Kundt}}}$ for the Kundt seed metric $g_{ab}^{{\hbox{\tiny Kundt}}}$. Since such a metric (\[Kundt seed xy\]) is simple, containing only one general metric function of one variable $\H(r)$, its key curvature tensors are also simple. Their explicit form is presented in Appendix A.
2. Use the well-known geometric relations for the Ricci and Bach tensors of conformally related metrics $g_{ab}^{{\hbox{\tiny Kundt}}}$ and ${g_{ab}=\Omega^2 \,g_{ab}^{{\hbox{\tiny Kundt}}}}$. Thus it is straightforward to evaluate the curvature tensors $R_{ab}$ and $B_{ab}$ for spherically symmetric geometries, starting from their forms of the Kundt seed calculated in the first step. In particular, since the Bach tensor trivially rescales under the conformal transformation as ${B_{ab} = \Omega^{-2}\,B_{ab}^{{\hbox{\tiny Kundt}}}}$, it remains simple. These calculations are performed in Appendix B.
3. These explicit components of the Ricci and Bach tensors are substituted into the field equations of Quadratic Gravity, which we already reduced to the expression ${R_{ab}=4k\, B_{ab}}$, see . This immediately leads to a very simple and compact form of these field equations. Moreover, using the Bianchi identities, it can be shown that the whole system reduces just to two equations , for the metric functions $\Omega(r)$ and $\H(r)$, see Appendix C.
By this procedure, we thus arrive at a remarkably simple form of the field equations (\[fieldeqsEWmod\]) for spherically symmetric vacuum spacetimes in Einstein–Weyl gravity and general Quadratic Gravity with ${R=0}$, namely *two ordinary differential equations* for the *two metric functions* $\Omega(r)$ and ${\cal H}(r)$: $$\begin{aligned}
\Omega\Omega''-2{\Omega'}^2 = &\ \tfrac{1}{3}k\, \B_1 \H^{-1} \,, \label{Eq1}\\
\Omega\Omega'{\cal H}'+3\Omega'^2{\cal H}+\Omega^2
= &\ \tfrac{1}{3}k \,\B_2 \,. \label{Eq2}\end{aligned}$$ The functions $\B_1(r)$ and $\B_2(r)$ denote *two independent components of the Bach tensor*, && \_1 ””, \[B1\]\
&& \_2 ’[H]{}”’-\^2 +2. \[B2\]
Recall also the relation (\[R=0\]), that is ${R=0}$, which is a trace of the field equations . This relation takes the explicit form $${\cal H}\Omega''+{\cal H}'\Omega'+{\textstyle \frac{1}{6}} ({\cal H}''+2)\Omega = 0 \,,
\label{trace}$$ see (\[barR\]). Indeed, it immediately follows from , : just subtract from the derivative of the second equation the first equation multiplied by $\H'$ (and divide the result by $6\Omega'$).
It is a great advantage of our conformal approach with the convenient form of the new metric (\[BHmetric\]) that the field equations (\[Eq1\]), (\[Eq2\]) are *considerably simpler* than the previously used field equations for the standard metric . Moreover, they form an *autonomous system*, which means that the differential equations *do not explicitly depend on the radial variable $r$*. This will be essential for solving such a system, finding their analytic solution in the generic form , or , in subsequent Section \[integration\].
Fundamental scalar invariants and geometric classification {#invariants}
==========================================================
For a geometrical and physical interpretation of spacetimes that are solutions to the field equations (\[Eq1\]), (\[Eq2\]), it will be crucial to investigate the behaviour of scalar curvature invariants constructed from the Ricci, Bach, and Weyl tensors themselves. A direct calculation yields $$\begin{aligned}
R_{ab}\, R^{ab} &= 16k^2\, B_{ab} B^{ab} \,, \label{invR}\\
B_{ab}\, B^{ab} &= \tfrac{1}{72}\,\Omega^{-8}\,\big[(\B_1)^2 + 2(\B_1+\B_2)^2\big] \,,\label{invB}\\
C_{abcd}\, C^{abcd} &= \tfrac{1}{3}\,\Omega^{-4}\,\big({\cal H}'' +2\big)^2 \,. \label{invC}\end{aligned}$$ To derive these expressions, we have used the field equations, the quantities (\[RT\_R rr\])–(\[RT\_R xx\]), (\[Bach rr\])–(\[Bach xx\]), (\[WeyliK\])–(\[WeylfK\]), and relations (\[contraEinstein-WeylBHC\]), (\[confrel\]), (\[OmBach\]) together with ${C_{abcd}\,C^{abcd}=\Omega^{-4}\, C_{abcd}^{{\hbox{\tiny Kundt}}}\, C^{abcd}_{{\hbox{\tiny Kundt}}}}$ which follows from the invariance of the Weyl tensor under conformal transformations.
It is interesting to observe from (\[invB\]) and (\[Bach rr\])–(\[Bach xx\]) with (\[OmBach\]) that B\_[ab]{}=0B\_[ab]{}B\^[ab]{} =0. \[Bach=0iffINV=0\] Moreover, C\_[abcd]{}C\^[abcd]{}=0B\_[ab]{} =0, \[Weylinv=0thenBach=0\] because the relation ${{\cal H}'' +2=0}$ substituted into gives ${B_{ab}\,B^{ab} =0}$, i.e., ${B_{ab} =0}$ due to .
Notice also that the *first Bach component* ${\B_1=\H \H''''}$ *always vanishes on the horizon* where ${\H=0}$, see the condition .
In view of the key invariant , there are *two geometrically distinct classes of solutions* to (\[Eq1\]), (\[Eq2\]), depending on the Bach tensor ${B_{ab}}$. The first simple case corresponds to ${B_{ab}=0}$, while the much more involved second case, not allowed in General Relativity, arises when ${B_{ab}\ne0}$. This invariant classification has geometrical and physical consequences. In particular, the distinction of spacetimes with ${B_{ab}=0}$ and with ${B_{ab}\ne0}$ can be detected by measuring geodesic deviation of test particles, see Section \[geodeviation\] below.
${B_{ab}=0}$: Uniqueness of Schwarzschild {#integration:Schw}
-----------------------------------------
First, let us assume the metrics (\[BHmetric\]) such that ${B_{ab}=0}$ everywhere. In view of and , this condition requires ${\B_1=0=\B_2}$, that is ””=0,’[H]{}”’-[[H]{}”]{}\^2 +2 =0. \[Bab=0-RHS\] Therefore, all left-hand sides and right-hand sides of equations (\[Eq1\]) and (\[Eq2\]) *vanish separately*, i.e., ”=2[’]{}\^2,’[H]{}’+3’\^2[H]{}+\^2 =0. \[Bab=0-LHS\] The first equations of (\[Bab=0-RHS\]) and (\[Bab=0-LHS\]) imply that ${\cal H}$ must be *at most cubic*, and $\Omega^{-1}$ must be *at most linear* in $r$. Using a coordinate freedom of the metric (\[BHmetric\]), without loss of generality we obtain ${\Omega=-1/r}$. The remaining equations (\[Bab=0-RHS\]), (\[Bab=0-LHS\]) then admit a unique solution $$\Omega(r)=-\frac{1}{r}\,,\qquad
{\cal H}(r) = -r^2-2m\, r^3 \,.
\label{IntegrSchwAdS}$$ Not surprisingly, this is exactly the Schwarzschild solution of General Relativity, see equation (\[Schw\]). Thus we have verified that the *Schwarzschild black hole spacetime is the only possible solution with vanishing Bach tensor*. Its corresponding scalar invariants (\[invR\])–(\[invC\]) are R\_[ab]{} R\^[ab]{} = 0 = B\_[ab]{} B\^[ab]{},C\_[abcd]{} C\^[abcd]{} = 48m\^2r\^6 . \[SchwarzInvariants\] Clearly, for ${m\not=0}$ there is a curvature singularity at ${r\to\infty}$ corresponding to ${\bar{r}=\Omega(r)=0}$.[^2]
${B_{ab}\ne0}$: New types of solutions to QG {#integration:nonSchw}
--------------------------------------------
Many other spherically symmetric vacuum solutions to Quadratic Gravity and Einstein–Weyl gravity exist when the Bach tensor is nontrivial. They are *much more involved, and do not exist in General Relativity*. Indeed, the field equations (\[fieldeqsEWmod\]) imply ${R_{ab}=4k\, B_{ab}\ne0}$, which is in contradiction with vacuum Einstein’s equations ${R_{ab}=0}$.
In the rest of this paper, we now concentrate on these new spherical spacetimes in QG, in particular on black holes generalizing the Schwarzschild solution. First, we integrate the field equations (\[Eq1\]), (\[Eq2\]) for the metric functions $\Omega(r)$ and ${\cal H}(r)$. Actually, we demonstrate that there are several classes of such solutions with ${B_{ab}\ne0}$. After their explicit identification and description, we will analyze their geometrical and physical properties.
Solving the field equations {#integration}
===========================
For nontrivial Bach tensor (${\B_1, \B_2 \ne0}$), the right-hand sides of the field equations (\[Eq1\]), (\[Eq2\]) are nonzero so that the nonlinear system of two ordinary differential equations for $\Omega(r)$, ${\cal H}(r)$ is coupled in a complicated way. Finding explicitly its general solution seems to be hopeless. However, *it is possible to write the admitted solutions analytically, in terms of (infinite) mathematical series expressed in powers of the radial coordinate $r$*.
In fact, there are *two natural possibilities*. The first is the expansion in powers of the parameter ${\Delta \equiv r-r_0}$ which expresses the solution around any finite value $r_0$ (including ${r_0=0}$). The second possibility is the expansion in powers of $r^{-1}$ which is applicable for large values of $r$. Let us now investigate both these cases.
Expansion in powers of ${\Delta \equiv r-r_0}$ {#expansio_DElta}
----------------------------------------------
It is a great advantage that , is an *autonomous system*. Thus we can find the metric functions in the form of an *expansion in powers of $r$ around any fixed value* ${r_0}$, $$\begin{aligned}
\Omega(r) {\!\!\!& = &\!\!\!}\Delta^n \sum_{i=0}^\infty a_i \,\Delta^{i}\,, \label{rozvojomeg0}\\
\H(r) {\!\!\!& = &\!\!\!}\Delta^p \,\sum_{i=0}^\infty c_i \,\Delta^{i}\,, \label{rozvojcalH0}\end{aligned}$$ where r-r\_0, \[DElta\] and $r_0$ is *any real constant*.[^3] In particular, in some cases this allows us to find solutions close to any black hole horizon $r_h$ by choosing ${r_0=r_h}$.
It is assumed that ${i=0, 1, 2, \ldots}$ are integers, so that the metric functions are expanded in integer steps of ${\Delta=r-r_0}$. On the other hand, the *dominant real powers* $n$ and $p$ in the expansions (\[rozvojomeg0\]) and (\[rozvojcalH0\]) *need not be* positive integers. We only assume that ${a_0\not=0}$ and ${c_0\not=0}$, so that the coefficients $n$ and $p$ are uniquely defined as the leading powers.
By inserting – into the field equations , , we prove in Section \[expansiont\_0\] that *only 4 classes of solutions of this form are allowed*, namely =\[-1,2\],=\[0,1\],=\[0,0\],=\[1,0\]. \[4classes\] In subsequent Section \[description\], it will turn out that the only possible solution in the class ${[n,p]=[-1,2]}$ is the Schwarzschild black hole for which the Bach tensor vanishes. Explicit Schwarzschild–Bach black holes with ${B_{ab}\ne0}$ are contained in the classes ${[0,1]}$ and ${[0,0]}$. The fourth class ${[n,p]=[1,0]}$ represents singular solutions without horizon, and it is equivalent to the class ${(s,t)=(2,2)}$ identified previously in [@Stelle:1978; @LuPerkinsPopeStelle:2015b; @PerkinsPhD].
Expansion in powers of $r^{-1}$ {#expansion_INF}
-------------------------------
Analogously, we may study and classify all possible solutions to the QG field equations for an asymptotic expansion as ${r\rightarrow \infty}$. Instead of , with , for very large $r$ we can assume that the metric functions $\Omega(r)$, $\mathcal{H}(r)$ are expanded in *negative powers* of $r$ as $$\begin{aligned}
\Omega(r) {\!\!\!& = &\!\!\!}r^N \sum_{i=0}^\infty A_i \,r^{-i}\,, \label{rozvojomegINF}\\
\mathcal{H}(r) {\!\!\!& = &\!\!\!}r^P \,\sum_{i=0}^\infty C_i \,r^{-i}\,. \label{rozvojcalHINF}\end{aligned}$$
Inserting the series (\[rozvojomegINF\]), (\[rozvojcalHINF\]) into the field equations , , it can be shown that *only 2 classes of such solutions are allowed*, namely =\[-1,3\]\^,=\[-1,2\]\^, \[2classes\] see Section \[expansiont\_INF\]. In subsequent Section \[description\_INF\], it will be shown that the class ${[N,P]=[-1,3]^\infty}$ represents the Schwarzschild–Bach black holes, whereas the class ${[N,P]=[-1,2]^\infty}$ is a specific Bachian generalization of a flat space which does not correspond to a black hole.
Discussion of solutions using the expansion in powers of $\Delta$ {#expansiont_0}
=================================================================
By inserting the series (\[rozvojomeg0\]), (\[rozvojcalH0\]) into the first field equation (\[Eq1\]), the following key relation is obtained $$\begin{aligned}
&\sum_{l=2n-2}^{\infty}\Delta^{l}\sum^{l-2n+2}_{i=0}a_i\, a_{l-i-2n+2}\,(l-i-n+2)(l-3i-3n+1) \nonumber \\
& \hspace{35.0mm}=\tfrac{1}{3}k \sum^{\infty}_{l=p-4}\Delta^{l}\,c_{l-p+4}\,(l+4)(l+3)(l+2)(l+1) \,.
\label{KeyEq1}\end{aligned}$$ The second field equation (\[Eq2\]) puts further constraints on the admitted solutions, namely $$\begin{aligned}
&\sum_{l=2n+p-2}^{\infty}\Delta^{l}\sum^{l-2n-p+2}_{j=0}\sum^{j}_{i=0}a_i\,a_{j-i}\,c_{l-j-2n-p+2}\,(j-i+n)(l-j+3i+n+2)
+\sum_{l=2n}^{\infty}\Delta^{l}\sum^{l-2n}_{i=0}a_i\,a_{l-i-2n}
\nonumber \\
& = \tfrac{1}{3}k \bigg[2+\sum^{\infty}_{l=2p-4}\Delta^{l}\sum^{l-2p+4}_{i=0}c_{i}\,c_{l-i-2p+4}\,(i+p)(l-i-p+4)(l-i-p+3)(l-\tfrac{3}{2}i-\tfrac{3}{2}p+\tfrac{5}{2})\bigg]\,.
\label{KeyEq2}\end{aligned}$$ A considerably simpler is the additional (necessary but not sufficient) condition following from the trace equation (\[trace\]) which reads $$\begin{aligned}
&\sum_{l=n+p-2}^{\infty}\Delta^{l}\sum^{l-n-p+2}_{i=0}c_i\,a_{l-i-n-p+2}\,\big[(l-i-p+2)(l+1)+\tfrac{1}{6}(i+p)(i+p-1)\big] =-\tfrac{1}{3}\sum^{\infty}_{l=n}\Delta^{l}\,a_{l-n}
\,.
\label{KeyEq3}\end{aligned}$$
Now we analyze the consequences of the equations (\[KeyEq1\])–(\[KeyEq3\]).
First, by comparing the corresponding coefficients of the same powers of $\Delta^l$ on both sides of the key relation (\[KeyEq1\]), we can express the coefficients $c_j$ in terms of (products of) $a_j$. Moreover, the *terms with the lowest order* put further restrictions. In particular, comparing the lowest orders on both sides (that is ${l=2n-2}$ and ${l=p-4}$) it is obvious that *we have to discuss three distinct cases*, namely:
- **Case I**: ${\ \ 2n-2<p-4}$, i.e., ${\ p>2n+2}$,
- **Case II**: ${\ 2n-2>p-4}$, i.e., ${\ p<2n+2}$,
- **Case III**: ${2n-2=p-4}$, i.e., ${\ p=2n+2}$.
Now let us systematically derive all possible solutions in these three distinct cases.
**Case I**
----------
In this case, ${2n-2<p-4}$, so that the *lowest* order in the key equation (\[KeyEq1\]) is on the *left hand* side, namely $\Delta^l$ with ${l=2n-2}$, and this yields the condition $$n(n+1)=0 \,.
\label{KeyEq1CaseI}$$ There are thus only two possible cases, namely ${n=0}$ and ${n=-1}$. Next, it is convenient to apply the equation (\[KeyEq3\]) whose lowest orders on its both sides are $$\big[6n(n+p-1)+p(p-1)\big]c_0\,\Delta^{n+p-2}+\cdots=-2\,\Delta^{n}+\cdots \,.
\label{KeyEq3CaseI}$$ For ${n=0}$, these powers are ${\Delta^{p-2}}$ and ${\Delta^{0}}$, respectively, but ${p-2>2n=0}$ by the definition of Case I. The lowest order ${0=-2\Delta^{0}}$ thus leads to a contradiction. Only the possibility ${n=-1}$ remains, for which reduces to $$(p-3)(p-4)c_0\,\Delta^{p-3}+\cdots =-2\,\Delta^{-1}+\cdots
\,.
\label{KeyEq3CaseIn=-1}$$ Since ${c_0\ne0}$, the only possibility is ${p=2}$, in which case ${c_0=-1}$.
**To summarize**: The only possible class of solutions in Case I is given by $$[n,p]=[-1,2]\qquad \hbox{with}\quad c_0=-1\,.
\label{CaseI_summary}$$
**Case II**
-----------
In this case, ${2n-2>p-4}$, so that the *lowest* order in the key equation (\[KeyEq1\]) is on the *right hand* side, namely $\Delta^l$ with ${l=p-4}$, and this gives the condition $$p(p-1)(p-2)(p-3)=0 \,.
\label{KeyEq1CaseII}$$ Thus there are four possible cases, namely ${p=0}$, ${p=1}$, ${p=2}$, and ${p=3}$. Equation has the lowest orders on both sides the same as given by equation , that is $$\begin{aligned}
\hbox{for}\quad p=0:\qquad &
\big[6n(n-1)\big]c_0\,\Delta^{n-2}+\cdots=-2\,\Delta^{n}+\cdots&&
\hbox{necessarily}\quad n=0, 1\,,\\
\hbox{for}\quad p=1:\qquad &
\big[6n^2\big]c_0\,\Delta^{n-1}+\cdots=-2\,\Delta^{n}+\cdots&&
\hbox{necessarily}\quad n=0\,,\\
\hbox{for}\quad p=2:\qquad &
\big[6n(n+1)+2\big]c_0\,\Delta^{n}+\cdots=-2\,\Delta^{n}+\cdots&&
(3n^2+3n+1)c_0=-1\,,\label{contrp=2c0}
\\
\hbox{for}\quad p=3:\qquad &
\big[6n(n+2)+6\big]c_0\,\Delta^{n+1}+\cdots=-2\,\Delta^{n}+\cdots&&
\hbox{not compatible}\,.\end{aligned}$$ Moreover, the lowest orders of all the terms in the field equation for the case ${p=2}$, implying ${n>0}$, are 3a\_0\^2\[n(3n+2)c\_0+1\]\^[2n]{} +2k(c\_0\^2 -1) + =0 , \[eq2rozvoj0omeg\] which requires ${c_0=\pm 1}$, but the constraint ${3n^2+3n+1=\pm 1}$ cannot be satisfied for ${n>0}$.
**To summarize**: The only possible three classes of solutions in Case II are given by \[n,p\]=\[0,1\], \[n,p\]=\[0,0\], \[n,p\]=\[1,0\]. \[CaseII\_summary\]
**Case III**
------------
Now ${2n-2=p-4}$, that is ${n=-1+p/2}$ equivalent to ${p=2n+2}$. In such a case, the *lowest* order in the key equation (\[KeyEq1\]) is *on both sides*, namely $\Delta^l$ with ${l=p-4}$. This implies the condition p(p-2)=0. \[KeyEq1CaseIII\] There are three subcases to be considered, namely ${p=0}$, ${p=2}$, and ${3a_0^2=-4kc_0 (p-1)(p-3)}$ with ${p\not= 0,1,2,3}$. This corresponds to ${n=-1}$, ${n=0}$, and ${3a_0^2=-4kc_0(4n^2-1)}$ with ${n\not= -1,-1/2,0,1/2}$, respectively. The leading orders of the trace equation on both sides are 2(11n\^2+6n+1) c\_0\^[3n]{} +[& = &]{}-2 \^n+. \[eqtr00omegIII\] Consequently, we obtain $$\begin{aligned}
&\hbox{for}\quad n=-1\Leftrightarrow p=0:\quad &
12c_0\,\Delta^{-3}+\cdots=-2\,\Delta^{-1}+\cdots& \qquad\hbox{not compatible}\,,\label{contrp=2c0IIIa}\\
&\hbox{for}\quad n=0\Leftrightarrow p=2:\quad & 2c_0+\cdots=-2+\cdots&\qquad c_0=-1\,,\label{contrp=2c0IIIb}\\
&\hbox{for}\quad 3a_0^2=4kc_0(1-4n^2): & (11n^2+6n+1)c_0+\cdots=0&\qquad
\hbox{not compatible}\,. \label{contrp=2c0IIIc}\end{aligned}$$ The incompatibility in the cases and are due to the fact that ${c_0\ne 0}$ and ${11n^2+6n+1}$ is always positive. In the case , we employ the field equation which for ${n=0, p=2}$ gives the condition ${3a_0^2+2k(c_0^2-1) = 0}$. Since ${c_0=-1}$ implies ${a_0=0}$, we again end up in a contradiction.
**To summarize**: There are no possible solutions in Case III.
Description and study of all possible solutions in powers of $\Delta$ {#description}
=====================================================================
Let us analyze all spherically symmetric solutions contained in the possible four classes and contained in Case I and Case II, respectively.
Uniqueness of the Schwarzschild black hole in the class ${[n,p]=[-1,2]}$ {#Schw_[n,p]=[-1,2]}
------------------------------------------------------------------------
Starting with the only admitted class ${[n,p]=[-1,2]}$ in the Case I, see , now we prove that *the only solution in this class is the Schwarzschild solution* with vanishing Bach tensor. Such a solution can be easily identified within the complete form (\[rozvojomeg0\])–(\[DElta\]), with ${r_0=0}$, using the expression (\[IntegrSchwAdS\]) as && a\_0=-1,a\_i=0 i1,\
&& c\_0=-1,c\_1=-2m,c\_i=0 i2, where $m$ is a free parameter.
Let us prove the uniqueness. The full key equation for ${n=-1}$ ${p=2}$ reads $$\begin{aligned}
2a_1a_0\,\Delta^{-3}
+ 6a_2a_0\,\Delta^{-2}
+12a_3a_0\,\Delta^{-1}
&
+\sum_{l=0}^{\infty}\Delta^{l}\sum^{l+4}_{i=0}a_i\, a_{l+4-i}\,(l+3-i)(l+4-3i)
\nonumber \\
&
=\tfrac{1}{3}k \sum^{\infty}_{l=0}\Delta^{l}\,c_{l+2}\,(l+4)(l+3)(l+2)(l+1) \,,
\label{KeyEq1[n,p]=[-12]}\end{aligned}$$ which necessarily implies a\_1=0,a\_2=0,a\_3=0, \[Schwinitcond1a\] and \^[l+4]{}\_[i=0]{}a\_i a\_[l+4-i]{}(l+3-i)(l+4-3i) =k c\_[l+2]{}(l+4)(l+3)(l+2)(l+1) l0, \[Schwinitcond1b\] that is (l+4)(l+5)a\_0a\_[l+4]{} =k c\_[l+2]{}(l+4)(l+3)(l+2)(l+1) -\^[l+3]{}\_[i=1]{}a\_i a\_[l+4-i]{}(l+3-i)(l+4-3i) l0. \[Schwinitcond1c\] The second field equation , using (\[Schwinitcond1a\]), takes the explicit form $$\begin{aligned}
-c_2a_0^2\,\Delta^{0}&+ \sum_{l=1}^{\infty}\Delta^{l}\sum^{l+2}_{j=0}\sum^{j}_{i=0}a_i\,a_{j-i}\,c_{l-j+2}\,(j-i-1)(l-j+3i+1)
+\sum_{l=1}^{\infty}\Delta^{l}\sum^{l+2}_{i=0}a_i\,a_{l-i+2}\nonumber \\
&
= \tfrac{1}{3}k \,\sum^{\infty}_{l=1}\Delta^{l}\sum^{l}_{i=0}c_{i}\,c_{l-i}\,(i+2)(l-i+2)(l-i+1)
(l-\tfrac{3}{2}i-\tfrac{1}{2})\,,
\label{KeyEq2[n,p]=[-12]}\end{aligned}$$ which implies c\_2=0. \[Schwinitcond2\] However, instead of solving (\[KeyEq2\[n,p\]=\[-12\]\]) for a general $l$, it is convenient to employ the “trace equation” \^[l+1]{}\_[i=0]{}c\_ia\_[l+1-i]{}=-a\_[l+1]{} l2. \[KeyEq3\[n,p\]=\[-12\]\] This can be rewritten as (l-1)la\_0c\_[l+1]{}=6l(l+1)a\_[l+1]{}-\^[l]{}\_[i=1]{}c\_ia\_[l+1-i]{} l2, \[KeyEq3\[n,p\]=\[-12\]b\] i.e., by relabeling the index ${l \to l+2}$, as (l+1)(l+2)a\_0c\_[l+3]{}=6(l+2)(l+3)a\_[l+3]{}-\^[l+2]{}\_[i=1]{}c\_ia\_[l+3-i]{} l0. \[KeyEq3\[n,p\]=\[-12\]c\]
Now, we employ the mathematical induction. Let us assume that for some ${l\ge 0}$ && a\_i=0 i=1,…,l+3,\
&& c\_i=0 i=2,…,l+2. For ${l=0}$ this is true due to , . Then the field equation reduces to (l+4)(l+5)a\_0a\_[l+4]{}=0,\[caseIbeq1\] while equation gives (l+1)(l+2)a\_0c\_[l+3]{}=0. This obviously implies ${a_{l+4}=0}$ and ${c_{l+3}=0}$, completing the induction step.
Therefore, *all* coefficients $a_i$ for $i\geq 1$ and *all* $c_i$ for $i\geq 2$ vanish, which means that the only possible solution in Case I is =,=-\^2 +c\_[1]{}\^[3]{}. \[Schwarzschild\[-1,2\]\] With the coordinate freedom , enabling us to set ${a_0=-1}$ and ${\Delta=r}$, this is exactly the explicit Schwarzschild solution .
**To conclude**: The class of solutions ${[n,p]=[-1,2]}$ represents spherically symmetric Schwarzschild solution , and it is the only solution in this class.
Schwarzschild–Bach black holes in the class ${[n,p]=[0,1]}$: near the horizon {#SchwaBach_[n,p]=[0,1]}
-----------------------------------------------------------------------------
Now we will prove that this second class represents spherically symmetric *non-Schwarzschild solutions to QG* that describe *black holes with nonvanishing Bach tensor*. Thus it is natural to call this family *Schwarzschild–Bach black holes*. The first three terms in the expansion of the full solution take the explicit form $$\begin{aligned}
\Omega(r) {\!\!\!& = &\!\!\!}-\frac{1}{r} + \frac{b}{r_h^2}(r-r_h)
-\frac{b}{r_h^3}\Big(2 +\frac{1}{8k r_h^2}+b\Big)(r-r_h)^2+\ldots\,, \label{IIbOmegaFULL}\\
\mathcal{H}(r) {\!\!\!& = &\!\!\!}(r-r_h)\bigg[ \frac{r^2}{r_h}
+ 3b\,(r-r_h)+\frac{b}{r_h}\Big(4-\frac{1}{2k r_h^2} + 3b \Big)(r-r_h)^2 + \ldots \bigg]\,,
\label{IIbH0FULL}\end{aligned}$$ where $r_h$ localizes the *black hole horizon* since ${\H(r_h)=0}$. In fact, for the *whole* class ${[n,p]=[0,1]}$, the metric function $\H$ given by , takes the generic form ${\H(r) = (r-r_0)\,\big(c_0+c_1(r-r_0)+\ldots\big)}$, which means that ${r=r_0}$ is the root of $\H$, and thus the horizon. Therefore, we can identify the constant $r_0$ (around which the solution is expanded) with the location of geometrical/physical horizon, r\_0r\_h . \[r0=rh\]
When the additional new *“Bach parameter”* $b$ in , is set to zero, the Bach tensor vanishes, and this solution reduces to the Schwarzschild spacetime (\[IntegrSchwAdS\]) with ${r_h=-1/(2m)}$.
Let us systematically derive the complete analytic form of these Schwarzschild–Bach black holes, leading to , . The equation for ${[n,p]=[0,1]}$ gives \^[l+1]{}\_[i=0]{}a\_i a\_[l+2-i]{}(l+2-i)(l+1-3i) =k c\_[l+3]{}(l+4)(l+3)(l+2)(l+1) , \[KeyEq1\[n,p\]=\[01\]\] where ${l\ge 0}$. Relabeling ${l \to l-1 }$, we thus obtain c\_[l+2]{}=\^[l]{}\_[i=0]{}a\_i a\_[l+1-i]{}(l+1-i)(l-3i) l1, \[nonSchwinitcondc\] which enables us to express all coefficients $c_{l+2}$ in terms of ${a_0,\ldots, a_{l+1}}$, starting from $c_3$. In the lowest nontrivial order ${l=0}$, the “trace equation” implies a\_1=-(1+c\_1), \[nonSchwinitcond3\] while for higher orders ${l=1, 2, \ldots}$, yields a\_[l+1]{}= l1, \[nonSchwinitconda\] which expresses all $a_{l+1}$ in terms of ${a_0,\ldots,
a_l}$ and ${c_1,\ldots, c_{l+1}}$. Finally, in the lowest nontrivial order ${l=0}$, the field equation gives the constraint ${6kc_0c_2=3a_0(a_0+a_1c_0)+2k(c_1^2-1)}$. Using , this becomes c\_2=. \[nonSchwinitcond2\]
There are thus *three free initial parameters*, namely $a_0$, $c_0$, and $c_1$ (apart from ${r_0=r_h}$). Using , , we obtain ${a_1, c_2}$, and then $a_{l+1}$, $c_{l+2}$ for all ${l = 1, 2, \ldots }$ by the alternate application of the *recurrent relations* , . This gives the complete analytic solution.
Now, the scalar invariants (\[invB\]), (\[invC\]) evaluated at ${r=r_h\equiv r_0}$ take the form B\_[ab]{}B\^[ab]{}(r\_h) = ( )\^2 ,C\_[abcd]{} C\^[abcd]{}(r\_h) = (1 + c\_1)\^2 .\[BInv2\] *The Bach tensor is in general nonvanishing*. In fact, for a physical interpretation of this family of solutions, it is convenient to introduce a new parameter $b$ proportional to ${1-c_1^2+3c_0c_2}$. Setting ${b=0}$ then gives the necessary condition for the Bach tensor to vanish. In view of , such *Bach parameter* $b$ can be defined simply as b (c\_1-2), \[b\_definice\] so that the Bach scalar invariant at the black hole horizon $r_h$ becomes B\_[ab]{}B\^[ab]{}(r\_h) = . \[BachInvariant\] Using $b$ as the dimensionless key parameter in the expansion , , the recurrent relations , readily yield an explicit solution of the field equations in the form $$\begin{aligned}
& a_1 = -\frac{a_0}{c_0} \Big( 1 + b \Big) \,,\label{IIb_expansiona}\\
& a_2 = +\frac{a_0}{c_0^2}\Big( 1 + \big(2+\tfrac{a_0^2}{8k}\big)b+b^2 \Big) \,,\nonumber\\
& a_3 = -\frac{a_0}{c_0^3}\Big( 1 + \tfrac{1}{9}\big(25+\tfrac{29a_0^2}{8k}+\tfrac{a_0^4}{16k^2}\big)b
+\tfrac{1}{9}\big(23+\tfrac{35a_0^2}{8k}\big)b^2+\tfrac{7}{9}b^3 \Big) \,, \ldots\,,
\nonumber\end{aligned}$$ and $$\begin{aligned}
& c_1 = 2 + 3 b \,,\label{IIb_expansionc}\\
& c_2 = \frac{1}{c_0}\Big( 1 + \big(4-\tfrac{a_0^2}{2k}\big)b + 3 b^2 \Big) \,,\nonumber\\
& c_3 = \frac{a_0^4}{32k^2c_0^2}\, b\,,\nonumber\\
& c_4 = \frac{a_0^2}{30kc_0^3}\, b\,\Big(\big(1-\tfrac{5a_0^2}{4k}-\tfrac{a_0^4}{32k^2}\big)
+\big(2-\tfrac{13a_0^2}{8k}\big)b+ b^2 \Big)\,, \ldots\,,
\nonumber\end{aligned}$$ and so on, where $a_0$, $c_0$, and $b$ are three free parameters.
### Identification of the Schwarzschild black hole
Now, it is possible to *identify the Schwarzschild black hole*. This is defined geometrically by the property that its Bach tensor vanishes. In view of , it requires to set the key parameter $b$ to zero. Interestingly, with ${b=0}$, the expansion coefficients , simplify enormously to && a\_i=a\_0(-)\^i i 0 ,\
&& c\_1=2, c\_2=, c\_i=0 i 3. The first sequence clearly corresponds to a *geometrical series*, while the second series is *truncated to a polynomial of the 3rd order*. The metric functions thus take the explicit closed form $$\begin{aligned}
\Omega(r) {\!\!\!& = &\!\!\!}a_0 \,\sum_{i=0}^\infty \,\Big(\!\!-\frac{\Delta}{c_0}\Big)^i
=\frac{a_0\,c_0}{c_0+\Delta}=\frac{a_0\,c_0}{r-r_h+c_0}\,, \label{IIbOmega}\\
\mathcal{H}(r) {\!\!\!& = &\!\!\!}c_0(r-r_h)+2(r-r_h)^2+c_0^{-1}(r-r_h)^3 \,. \label{IIbH0}\end{aligned}$$ Using the gauge freedom (a constant rescaling and shift of the coordinate $r$), *we are free to chose* a\_0=-,c\_0=r\_h,\[IIb\_a0\] so that the metric functions become =(r) = -, (r) = -r\^2+ = (r-r\_h) . \[IIbH0Schw\] Clearly, there is a *black hole horizon located at* ${r_h}$. This is the *Schwarzschild horizon* given by the usual condition ${\,h=1-2m/\bar{r}=0\,}$. In terms of ${r=-1/{\bar r}}$, it is equivalent to ${r_h=-1/(2m)}$. Thus for the case ${b=0}$, we have fully recovered the standard form of the Schwarzschild solution, since the metric functions are exactly the same as .
### More general Schwarzschild–Bach black holes
When ${b\not=0}$, the corresponding solution given by , , that is , , can be naturally interpreted as *generalized black holes with a nontrivial Bach tensor* whose invariant value ${B_{ab}\,B^{ab}}$ at the horizon is proportional to $b^2$, according to . Moreover, as ${b \to 0}$ we explicitly obtain the Schwarzschild black hole . Using the summation of the “background” terms independent of $b$ as in , and the same gauge fixing , it is possible to write this solution explicitly as , . Recall that $r_h$ *still gives the exact value of the horizon* even if $b$ is now nonzero, see the text below equation .
To express a *general solution in this class completely*, it is convenient to introduce coefficients $\alpha_i, \gamma_i$ as *those parts of* $a_i, c_i$, respectively, which *do not involve* the ${b=0}$ Schwarzschild “background”, i.e., using the following definitions: a\_i [& &]{}a\_i(b=0)-,a\_i(b=0) ,\[def\_alphai\]\
c\_1 [& &]{}2 + 3b\_1 ,c\_2 +3b ,c\_i 3b i 3. \[def\_gammai\] With the natural gauge choice , the complete solution then takes the explicit form (r) [& = &]{}- -\_[i=1]{}\^\_i(1-)\^i , \[Omega\_\[0,1\]\]\
(r) [& = &]{}(r-r\_h), \[H\_\[0,1\]\] with the initial coefficients $$\alpha_1=1 \,, \qquad \gamma_1=1\,, \qquad \gamma_2 = \frac{1}{3}\Big(4-\frac{1}{2kr_h^2}+3b\Big) \,,
\label{alphasgammaIIbinitial}$$ and all other coefficients $\alpha_l, \gamma_l$ for any ${l \ge 1}$ given by the recurrent relations (defining ${\alpha_0=0}$) $$\begin{aligned}
\alpha_{l+1}= &\, \frac{1}{(l+1)^2}\Big[\alpha_l\big(2l^2+2l+1\big)-\alpha_{l-1}l^2
-3\sum_{i=1}^{l+1}(-1)^i\,\gamma_i\,(1+b\,\alpha_{l+1-i})\big[(l+1)(l+1-i)+\tfrac{1}{6}i(i+1)\big]\Big],
\nonumber\\
\gamma_{l+2}= &\, \frac{(-1)^{l+1}}{kr_h^2\,(l+3)(l+2)(l+1)l}\,\sum_{i=0}^{l}
\big(\alpha_i+\alpha_{l+1-i}(1+b\,\alpha_i) \big)(l+1-i)(l-3i) \,,
\label{alphasIIbgeneral}\end{aligned}$$ which follow from and for $a_{l+1}$ and $c_{l+2}$, respectively. The first terms generated by these relations are $$\begin{aligned}
& \alpha_2 = 2+\frac{1}{8kr_h^2}+b \,, \nonumber\\
& \alpha_3 = \frac{1}{9}\Big(25+\frac{29}{8kr_h^2}+\frac{1}{16k^2r_h^4}\Big)
+\frac{1}{9}\Big(23+\frac{35}{8kr_h^2}\Big)\,b+\frac{7}{9}\,b^2 \,, \ldots\,,
\label{alphasIIb0}\\
& \gamma_3 = \frac{1}{96k^2r_h^4} \,, \nonumber\\
& \gamma_4 = \frac{1}{18kr_h^2}\Big(\frac{1}{5}-\frac{1}{4kr_h^2}-\frac{1}{160k^2r_h^4}\Big)
+\frac{3}{720kr_h^2}\Big(16-\frac{13}{kr_h^2}\Big)\,b+\frac{1}{90kr_h^2}\,b^2 \,, \ldots\,,
\label{gammasIIb0}\end{aligned}$$ yielding , .
This family of spherically symmetric black-hole spacetimes , in Einstein–Weyl/Quadratic Gravity depends on *two parameters* with a *clear geometrical and physical interpretation*, namely:
- The parameter $r_h$ identifies the *horizon position*. Indeed, ${r=r_h}$ is the root of the metric function $\H(r)$ given by .
- The dimensionless *Bach parameter* $b$ *distinguishes* the Schwarzschild solution (${b=0}$) from the more general Schwarzschild–Bach black hole spacetime with nonzero Bach tensor (${b\ne0}$).
In fact, we have chosen the parameter $b$ in such a way that it *determines the value of the Bach tensor , on the horizon* $r_h$, namely \_1(r\_h) = 0, \_2(r\_h) = -b . \[bonhorizon\] Thus on the horizon, the invariants and of the Bach and Weyl tensors take the values B\_[ab]{}B\^[ab]{}(r\_h) = b\^2 ,C\_[abcd]{} C\^[abcd]{}(r\_h) = 12r\_h\^4(1+b)\^2 , \[BCInvariants\_\[0,1\]\] respectively.
**To conclude**: The class of solutions ${[n,p]=[0,1]}$ represents spherically symmetric Schwarzschild–Bach black holes (abbreviated as Schwa–Bach), expressed in terms of the series , *around the horizon* $r_h$, i.e., for the special choice ${r_0=r_h}$. These Schwa–Bach black holes include and generalize the well-known Schwarzschild black hole.
Restricting to Einstein’s theory, corresponding to ${k=0}$, requires ${a_0+a_1c_0=0}$, see the constraint above equation . Substituting this into , we obtain ${c_1=2}$, and thus ${b=0}$. This again confirms that the only possible spherical vacuum solution in General Relativity is the Schwarzschild solution.
Let us finally remark that the explicit recurrent relations can be rewritten in a slightly more compact form if we relabel the index $l$ to ${j \equiv l+1}$, so that the relations for any ${j \ge 2}$ become $$\begin{aligned}
\alpha_{j}= &\, \frac{1}{j^2}\Big[\alpha_{j-1}\big(2j^2-2j+1\big)-\alpha_{j-2}(j-1)^2
-3\sum_{i=1}^{j}(-1)^i\,\gamma_i\,(1+b\,\alpha_{j-i})\big[j(j-i)+\tfrac{1}{6}i(i+1)\big]\Big]\,,
\nonumber\\
\gamma_{j+1}= &\, \frac{(-1)^{j}}{kr_h^2\,(j+2)(j+1)j(j-1)}\,\sum_{i=0}^{j-1}
\big(\alpha_i+\alpha_{j-i}(1+b\,\alpha_i) \big)(j-i)(j-1-3i) \,.
\label{alphasgammasgeneral_[0,1]}\end{aligned}$$
Schwarzschild–Bach black holes in the class ${[n,p]=[0,0]}$: near a generic point {#SchwaBach_[n,p]=[0,0]}
---------------------------------------------------------------------------------
This more general class of possible spherically symmetric vacuum solutions to QG (see ) *may, as a special case, also represent the family of Schwarzschild–Bach black holes* with nonvanishing Bach tensor. In contrast to the previous case ${[n,p]=[0,1]}$, the expansion is now considered around *an arbitrary fixed value* $r_0$ which is *distinct from the position of the black hole horizon* $r_h$, r\_0 r\_h. \[r0\_NOT=\_rh\] Indeed, for ${[n,p]=[0,0]}$ the metric function $\H$ given by , is ${\H(r) = c_0+c_1(r-r_0)+\ldots}$, where ${c_0\ne0}$, so that the value ${r=r_0}$ is *not* the root of $\H$ and thus cannot be the horizon.
In such a case, the first few terms in the expansion of the full solution take the explicit form $$\begin{aligned}
\Omega(r) {\!\!\!& = &\!\!\!}-\frac{1}{r} + b_1\,\frac{r_h}{2r_0^3}\,\frac{(r-r_0)^2}{r_h-r_0} +\ldots\,,
\label{[0,0]_OmegaFULL}\\
\mathcal{H}(r) {\!\!\!& = &\!\!\!}\big(r-r_h\big)\frac{r^2}{r_h} + (b_1-b_2)\,r_0(r-r_0) -3b_2\,(r-r_0)^2
\nonumber\\
&& +\frac{(b_2-b_1)\big(1+\gamma+\frac{1}{2kr_0^2}\big)-2(2+3\gamma)b_2+3b_2^2}{(1+3\gamma+b_1-b_2)\,r_0} \,(r-r_0)^3 \ldots \,,
\label{[0,0]_HFULL}\end{aligned}$$ where $b_1$ and $b_2$ are *two independent Bach parameters* proportional to values of the two components of the Bach tensor at $r_0$. By setting ${b_1=0=b_2}$, the Schwarzschild solution (which has vanishing Bach tensor) is immediately obtained.
Let us derive this analytic form of the Schwa–Bach black holes. For ${[n,p]=[0,0]}$ the complete solution to , of the form – is given by the Taylor expansions (r) = a\_0 + \_[i=1]{}\^a\_i (r-r\_0)\^[i]{},(r) = c\_0 + \_[i=1]{}\^c\_i (r-r\_0)\^[i]{}. \[rozvoj\[0,0\]\] The key equation for ${n=0=p}$, after relabeling ${l \to l-1}$, gives c\_[l+3]{}=\^[l]{}\_[i=0]{} a\_i a\_[l+1-i]{}(l+1-i)(l-3i) l1. \[\[0,0\]initcondc\] Equation , relabeling ${l \to l-1}$, implies a\_[l+1]{}= l1. \[\[0,0\]initconda\] Finally, the field equation in the lowest nontrivial order ${l=0}$ gives one additional constraint c\_3=. \[\[0,0\]initcond2\]
Thus there are *five free initial parameters*, namely ${a_0,\, a_1,\, c_0,\, c_1,\, c_2}$ (in addition to $r_0$). All the remaining coefficients $a_{l+1}$, $c_{l+3}$ in are then obtained by applying the recurrent relations , , respectively, starting as && a\_2 = -, …\
&& c\_4 = -, …. \[\[0,0\]initcond3\]
Now we show that *three of the five initial parameters* (namely ${a_0, a_1, c_0}$) *can be conveniently fixed using the gauge freedom* in such a way that the Schwarzschild solution and flat Minkowski background are uniquely identified and directly seen.
### Identification of the Schwarzschild black hole
Specific geometry can be identified by the scalar invariants , with , . In particular, the Bach invariant evaluated at ${r=r_0}$ is $$\begin{aligned}
& B_{ab}\, B^{ab} (r_0) = \frac{1}{72\,a_0^8}\,\big[ (\B_1)^2 + 2(\B_1+\B_2)^2\big]\,,\nonumber\\
& \hbox{where}\quad \B_1 (r_0)= 24 c_0c_4 \,,\qquad \B_2 (r_0)= 2(3 c_1 c_3 - c_2^2 + 1)\,.
\label{[0,0]_BachInvariant}\end{aligned}$$ *Vanishing of the Bach tensor* (${B_{ab}=0 \Leftrightarrow \B_1=0=\B_2}$), which uniquely identifies the Schwarzschild solution, thus requires ${c_4=0}$ and ${3 c_1 c_3 - c_2^2 + 1 =0}$. In combination with , , this implies two necessary conditions c\_1=-(1+3c\_0),c\_2=2+3c\_0 \[\[0,0\]BachzeroA\] that only depend on the fraction ${a_1/a_0}$ and $c_0$. Interestingly, for such a choice of parameters the recurrent relations , give a very simple complete solution a\_i=a\_0()\^i i 0 ,c\_3=-(1+c\_0), c\_i=0 i 4. The first sequence clearly yields a geometrical series, while the second series is truncated to the 3rd-order polynomial. Thus the metric functions take the closed form $$\begin{aligned}
\Omega(r) {\!\!\!& = &\!\!\!}a_0 \,\sum_{i=0}^\infty \,\Big(\frac{a_1}{a_0}\Delta\Big)^i
=\frac{a_0^2}{a_0-a_1\Delta}=\frac{a_0^2}{(a_0+a_1r_0)-a_1r}\,, \label{[0,0]Omega}\\
\mathcal{H}(r) {\!\!\!& = &\!\!\!}c_0+c_1(r-r_0)+c_2(r-r_0)^2+c_3(r-r_0)^3 \,. \label{[0,0]H0}\end{aligned}$$ Using the gauge freedom , the most convenient choice a\_0=-,a\_1= \[\[0,0\]\_a0a1\] can always be made, so that the metric functions reduce to =(r) = -, (r) = (r-r\_0)+r\^3 . \[\[0,0\]\_Schw\] Notice that this function $\mathcal{H}$ can be rewritten as (r) = -r\^2+ = (r-r\_h) , r\_h . \[\[0,0\]\_SchwSHIFT\] This is exactly the standard form of the Schwarzschild solution, with the *black hole horizon located at* $r_h$ (clearly the root of $\H$). Thus the constant $c_0$ is uniquely determined in terms of the physical/geometrical parameter $r_h$ (the horizon) and an arbitrary parameter $r_0$ (entering the expansion variable ${\Delta=r-r_0}$) as c\_0 r\_0\^2,-1,r\_0 r\_h . \[\[0,0\]\_DEF c0\] Thus we have proven that all solutions in the class ${[n,p]=[0,0]}$ *with vanishing Bach tensor are equivalent to the Schwarzschild black hole solution*, as also identified in the classes ${[n,p]=[0,1]}$ and ${[n,p]=[-1,2]}$, see expressions and , respectively. The *main difference* is that in the class ${[n,p]=[0,1]}$, it is possible (and, in fact, necessary) to choose the expansion parameter $r_0$ equal to the horizon $r_h$, see , naturally allowing to expand the solution around the black hole horizon, while in the present case of the class ${[n,p]=[0,0]}$, *such a choice is forbidden* ($r_0$ is *not* the root of $\mathcal{H}$). Indeed, for the choice ${r_0=r_h}$, the expression would lead to ${c_0=0}$ which is not allowed. Otherwise the constant $r_0$, determining the initial position around which the solution is expanded, can be chosen *arbitrarily*.
These conclusions are consistent with the behavior of the Weyl curvature invariant at $r_0$, C\_[abcd]{} C\^[abcd]{}(r\_0) = 12 = 48 m\^2r\_0\^6 , where we have used the conditions and the Schwarzschild horizon position ${r_h=-1/(2m)}$. This invariant value at the horizon agrees with . For ${m=0}$, flat Minkowski background is obtained, corresponding to ${c_2=-1}$, that is ${c_0=-r_0^2}$, in which case ${\mathcal{H}(r) = -r^2}$, and there is no horizon.
### More general black hole solutions with nontrivial Bach tensor
Returning to the generic case – in the class ${[n,p]=[0,0]}$ with nonvanishing Bach tensor, it is now necessary to *introduce two distinct Bach parameters $b_1$ and $b_2$*, corresponding to the *two components $\B_1(r_0)$ and $\B_2(r_0)$* of the Bach tensor and , respectively, evaluated at $r_0$. They enter via the coefficients $c_4$ and $c_3$, which are expressed in terms of the two remaining initial parameters $c_1$ and $c_2$ using and . For ${B_{ab}=0}$, they take the form , i.e., with the gauge and fixing , ${c_1=(1+3\gamma)\,r_0}$ and ${c_2=2+3\gamma}$. It turns out to be useful to define two dimensionless Bach parameters $b_1$ and $b_2$ via the relations c\_1 = (1+3+b\_1-b\_2)r\_0 ,c\_2 = 2+3-3b\_2 , \[Bach\_c1c2\] that is b\_1 (-1-6-c\_2+3c\_1/r\_0) , b\_2 (2+3-c\_2) . \[Bach\_b1b2\] Then $b_1$ and $b_2$ are *directly proportional* to the two Bach tensor components $\B_1(r_0)$ and $\B_2(r_0)$, b\_1=kr\_0\^2\_1(r\_0) ,b\_2=kr\_0\^2(\_1(r\_0)+\_2(r\_0)), \[\[0,0\]\_BachInvariantb1b2B1B2\] and the Bach invariant at $r_0$ is simply expressed as B\_[ab]{} B\^[ab]{} (r\_0)= ( b\_1\^2 + 2 b\_2\^2) . \[\[0,0\]\_BachInvariantb1b2\]
With the parametrization by $b_1$, $b_2$ introduced in , assuming again the natural gauge and fixing , the coefficients $a_i$, $c_i$ of the explicit solution – are then given as $$\begin{aligned}
&
a_0 = -\frac{1}{r_0}\,,\qquad
a_1 =\frac{1}{r_0^2}\,,\qquad
a_2 = -\frac{1}{r_0^3}-\frac{b_1}{2\gamma\, r_0^3}\,, \ldots\,,\label{[0,0]_expansionaINFa} \\
& c_0 = \gamma\,r_0^2 \,,\qquad
c_1 = (1+3\gamma)\,r_0+(b_1-b_2)\,r_0 \,,\qquad
c_2 = 2+3\gamma-3b_2 \,,\nonumber\\
&
c_3 = \frac{(1+\gamma)(1+3\gamma)-2(2+3\gamma)b_2+3b_2^2+(b_2-b_1)/(2kr_0^2)}{(1+3\gamma+b_1-b_2)\,r_0} \,,\qquad
c_4 = \frac{b_1}{8k\gamma r_0^4}\,, \ldots\,.
\label{[0,0]_expansioncINFc}\end{aligned}$$ For ${b_1=0=b_2}$, we immediately recover the Schwarzschild solution , that is . In a generic case, the complete solution can be understood as the Schwarzschild black hole “background” modified by a nonzero Bach tensor, encoded in the terms that are proportional to (powers of) the dimensionless Bach parameters $b_1$ and $b_2$. The expansion of this full solution takes the explicit form , .
### Identification of the Schwa–Bach black hole solutions ${[0,1]}$ in the class ${[0,0]}$
Now a natural question arises about the explicit relation between the form , and the form , of the family of Schwarzschild–Bach black holes. The problem is that we cannot simply express the single Bach parameter $b$ in terms of the two parameters ${b_1, b_2}$. The reason is that $b$ determines the value of the Bach tensor *at the horizon* $r_h$, namely \_1(r\_h) = 0, \_2(r\_h) = -b , \[bonhorizonN\] while $b_1$ and $b_2$ determine its two independent values *at any given* $r_0$ \_1(r\_0) = b\_1,\_2(r\_0) = (b\_2 - b\_1), \[\[0,0\]\_BachInvariantb1b2B1B2N\] see and , respectively. Since the functions ${\B_1(r),\, \B_2(r)}$ are complicated, the relations between the constants $b$ and ${b_1,\, b_2}$ are obscured.
However, this problem can be circumvent by the following procedure. In order to explicitly identify the Schwa–Bach black hole solution , , expressed around the horizon $r_h$ in the class ${[0,1]}$, within the generic class ${[0,0]}$ given by , we just have to determine its five free parameters ${a_0,\, a_1,\, c_0,\, c_1,\, c_2}$ properly. Instead of considering , , we can simply *evaluate the functions* , (and their derivatives) at ${r=r_0}$, and then compare them with the Taylor expansions (and their derivatives) evaluated at ${r=r_0}$, obtaining[^4] $$\begin{aligned}
a_0 & = -\frac{1}{r_0}-\frac{b}{r_h}\sum_{i=1}^\infty \alpha_i\,\Big(1-\frac{r_0}{r_h}\Big)^i \,, \label{[0,0]a0}\\
a_1 & = \frac{1}{r_0^2}+\frac{b}{r_h^2}\sum_{i=1}^\infty i\,\alpha_i\,\Big(1-\frac{r_0}{r_h}\Big)^{i-1} \,, \\
c_0 & = (r_0-r_h)\bigg[\frac{r_0^2}{r_h}+3b\,r_h\sum_{i=1}^\infty \gamma_i\,\Big(\frac{r_0}{r_h}-1\Big)^i\, \bigg] \,, \\
c_1 & = (3r_0-2r_h)\frac{r_0}{r_h}+3b\,r_h\sum_{i=1}^\infty (i+1)\, \gamma_i\,\Big(\frac{r_0}{r_h}-1\Big)^i \,, \\
c_2 & = (3r_0-r_h)\frac{1}{r_h}+\frac{3}{2}b\,\sum_{i=1}^\infty i(i+1)\,\gamma_i\,\Big(\frac{r_0}{r_h}-1\Big)^{i-1} \label{[0,0]c2}\,.\end{aligned}$$ Then using the recurrent relations –, we are able to express the Schwarzschild–Bach black holes using the complete expansion around *any value* $r_0$ and just a *single Bach parameter* $b$ which determines the value of the Bach tensor at the horizon $r_h$.
When ${b=0}$, the coefficients $a_i$ form a geometrical series, and the metric functions simplify to , which is again the Schwarzschild solution . Both the classes ${[0,0]}$ and ${[0,1]}$ with ${B_{ab}=0}$ thus reduce to the Schwarzschild black hole. The difference is that in the class ${[0,1]}$ the radial distance parameter $r_0$ is *equal* to $r_h$, while ${r_0 \ne r_h}$ can be chosen *arbitrarily* in the class ${[0,0]}$.
### Formal limit ${r_0 \to r_h}$
Let us consider a “consistency check” between the two series expressing the Schwa–Bach black hole solution, namely , in the class ${[0,1]}$ and , in the class ${[0,0]}$.
To this end, let us denote temporarily the coefficients in the class ${[0,0]}$ by ${\hat c}_i$ and ${\hat a}_i$. The limit ${r_0 \to r_h}$ in – can be trivially performed, just by setting ${r_0=r_h}$, leading to the relations \_0 [& = &]{}- a\_0,\_1 = (1+b) a\_1,\
[c]{}\_0 [& = &]{}0,\_1 = r\_h c\_0,\_2 = 2+3b c\_1, where equation and the first relations in , have also been employed. By comparing and it is also seen that ${\hat c}_{j+1}$ satisfies the same recurrent relation as $c_j$, so that the functions $\H$ agree. Moreover, from the relation it follows that the condition ${\hat c}_0=0$ requires 0=a\_[l-1]{} +l\^2 [c]{}\_1[a]{}\_l +\^[l+1]{}\_[i=2]{}c\_ia\_[l+1-i]{}. This implies \_l =- , which (with the identification ${\hat c_{i+1}=c_i}$) is equivalent to the recurrent expression for $a_{l+1}$, so that the functions $\Omega$ also agree. In other words, in the limit ${r_0\to r_h}$ we obtain \_0 0, \_[j+1]{} c\_j, \_j a\_j j 0, demonstrating the consistency of the two expressions for the Schwa–Bach black holes in these two classes of solutions.
Bachian singularity in the class ${[n,p]=[1,0]}$ {#SchwaBach_[n,p]=[1,0]}
------------------------------------------------
This last possible class of spherically symmetric vacuum solutions represents spacetimes which are *not* black holes with horizon localized at $r_0$. Instead, it seems to be a specific family containing a naked singularity with ${B_{ab} \ne 0}$.
The key equation for ${[n,p]=[1,0]}$, relabeling ${l \to l-3}$, gives c\_[l+1]{}=\^[l-3]{}\_[i=0]{} a\_i a\_[l-3-i]{}(l-2-i)(l-5-3i) l3, \[(2,2)initcondc\] expressing $c_{l+1}$, starting from ${c_4}$. Equation in the lowest order ${l=0}$ implies a\_1=-, \[(2,2)initcond3\] and in higher orders a\_[l+1]{}= l1. \[(2,2)initconda\] Finally, the field equation in the lowest nontrivial order ${l=0}$ gives the condition c\_3=. \[(2,2)initcond2\] All coefficients $a_{l+1}$, $c_{l+1}$ are obtained by applying the recurrent relations , . This yields an explicit solution (r) = (r-r\_0),(r) = c\_0 + \_[i=1]{}\^c\_i (r-r\_0)\^[i]{}, \[rozvoj\[1,0\]\] where a\_2 [& = &]{}-,\
a\_3 [& = &]{}-, …,\
c\_4 [& = &]{}-, c\_5 = , …, and ${a_0,\, c_0,\, c_1,\, c_2}$ are *four initial parameters* (apart from $r_0$), but not all of them are independent. Due to the gauge freedom , we can set, for example, ${a_0=1}$ and also ${r_0=0}$.
To determine the main geometric properties we employ the scalar invariants , , which read B\_[ab]{} B\^[ab]{}(r) = +…,C\_[abcd]{} C\^[abcd]{}(r) = +….\[(2,2)BInv2\] *The Bach tensor $B_{ab}$ is thus nonvanishing near* $r_0$. And since ${R_{ab}=4k\, B_{ab} \ne 0}$, this class of solutions *does not contain Ricci-flat subcases*. The Bach invariant always diverges at ${r=r_0}$, and there is also a *Weyl curvature singularity* at ${r=r_0}$ (maybe unless ${c_2=-1}$).
Moreover, for the expressions – in the limit ${r\rightarrow r_0}$ behave as && [|r]{}=(r) \~a\_0(r-r\_0) 0 , \[(2,2)horizon\]\
&& h \~-c\_0[|r]{}\^2 0, f \~-a\_0\^2c\_0 ([|r]{})\^[-2]{} . \[(2,2)horizonC\] It shows a very specific and unusual behavior of the metric functions $f$ and $h$ close to the curvature singularity at ${{\bar r}=0 }$, in terms of the physical radial coordinate $\bar r$.
This class ${[n,p]=[1,0]}$ of solutions corresponds to the family which has been identified in [@Stelle:1978; @LuPerkinsPopeStelle:2015b; @HoldomRen:2017] as ${(s,t)=(2,2)}$, and nicknamed *(2,2)-family* in [@PerkinsPhD], see Section \[summary\] for more details.
Discussion of solutions using the expansion in powers of $r^{-1}$ {#expansiont_INF}
=================================================================
By inserting the series (\[rozvojomegINF\]), (\[rozvojcalHINF\]), that is $$\Omega(r) = r^N \sum_{i=0}^\infty A_i \,r^{-i}\,, \qquad
\mathcal{H}(r) = r^P \,\sum_{i=0}^\infty C_i \,r^{-i}\,,\label{rozvojomagAcalHINF}$$ into the key field equation (\[Eq1\]), we obtain the relation $$\begin{aligned}
&\sum_{l=-2N+2}^{\infty}r^{-l}\sum^{l+2N-2}_{i=0}A_i\,A_{l-i+2N-2}\,(l-i+N-2)(l-3i+3N-1) \nonumber \\
& \hspace{45.0mm}=\tfrac{1}{3}k\sum^{\infty}_{l=-P+4}r^{-l}\,C_{l+P-4}\,(l-4)(l-3)(l-2)(l-1) \,.
\label{KeyEq1INF}\end{aligned}$$ The second field equation (\[Eq2\]) puts further constraints, namely $$\begin{aligned}
&\sum_{l=-2N-P+2}^{\infty}r^{-l}\sum^{l+2N+P-2}_{j=0}\sum^{j}_{i=0}A_i\,A_{j-i}\,C_{l-j+2N+P-2}
\,(j-i-N)(l-j+3i-N-2) \nonumber \\
& \hspace{10.0mm} +\sum_{l=-2N}^{\infty}r^{-l}\sum^{l+2N}_{i=0}A_i\,A_{l-i+2N}
\nonumber \\
& = \tfrac{1}{3}k \bigg[2+\sum^{\infty}_{l=-2P+4}r^{-l}\sum^{l+2P-4}_{i=0}C_{i}\,C_{l-i+2P-4}\,(i-P)(l-i+P-4)(l-i+P-3)
(l-\tfrac{3}{2}i+\tfrac{3}{2}P-\tfrac{5}{2})\bigg]\,.
\label{KeyEq2INF}\end{aligned}$$ The supplementry condition following from the “trace equation” (\[trace\]) reads $$\begin{aligned}
&\sum_{l=-N-P+2}^{\infty}r^{-l}\sum^{l+N+P-2}_{i=0}C_i\,A_{l-i+N+P-2}\,\big[(l-i+P-2)(l-1)+\tfrac{1}{6}(i-P)(i-P+1)\big] \nonumber \\
& \hspace{80mm}=-\tfrac{1}{3}\sum^{\infty}_{l=-N}r^{-l}\,A_{l+N}
\,.
\label{KeyEq3INF}\end{aligned}$$
By comparing the corresponding coefficients of the same powers of $r^{-l}$ on both sides of the relation (\[KeyEq1INF\]), we can express the coefficients $C_j$ in terms of $A_j$s. Moreover, the *terms with the lowest order* imply that we have to discuss *three distinct cases*, namely:
- **Case I**$^\infty$: ${\ \ -2N+2<-P+4}$, i.e., ${\ P<2N+2}$,
- **Case II**$^\infty$: ${\ -2N+2>-P+4}$, i.e., ${\ P>2N+2}$,
- **Case III**$^\infty$: ${-2N+2=-P+4}$, i.e., ${\ P=2N+2}$.
Let us derive all possible solutions in these cases.
**Case I**$^\infty$
-------------------
In the case, ${-2N+2<-P+4}$, the *highest* order in the key equation (\[KeyEq1INF\]) is on the *left hand* side, namely $r^{-l}$ with ${-l=2N-2}$, which yields the condition $$N(N+1)=0 \,.
\label{KeyEq1CaseIINF}$$ The only two admitted cases are ${N=0}$ and ${N=-1}$. The highest orders on both sides of equation (\[KeyEq3INF\]) are $$\big[6N(N+P-1)+P(P-1)\big]C_0\,r^{N+P-2}+\cdots=-2\,r^N+\cdots \,.
\label{KeyEq3CaseIINF}$$ For ${N=0}$, these powers are ${r^{P-2}}$ and ${r^0}$, respectively, but ${P-2<2N=0}$ by the definition of Case I$^\infty$. The highest order ${0=-2r^0}$ thus leads to a contradiction. Similarly, for the second possibility ${N=-1}$, the powers are ${r^{P-3}}$ and ${r^{-1}}$, respectively, but ${P-3<2N-1=-3<-1}$. The highest order is thus ${0=-2r^{-1}}$, which is again a contradiction.
**To summarize**: There are no possible solutions in Case I$^\infty$.
**Case II**$^\infty$
--------------------
In this case, ${-2N+2>-P+4}$, so that the *highest* order in the key equation (\[KeyEq1INF\]) is on the *right hand* side, namely $r^{-l}$ with ${l=-P+4}$, which gives the condition $$P(P-1)(P-2)(P-3)=0 \,.
\label{KeyEq1CaseIIINF}$$ Thus there are four possible cases, namely ${P=0}$, ${P=1}$, ${P=2}$, and ${P=3}$. Equation has the highest orders on both sides as given by equation , that is $$\begin{aligned}
\hbox{for}\quad P=0:\qquad &
\big[6N(N-1)\big]C_0\,r^{N-2}+\cdots=-2\,r^N+\cdots&&
\hbox{not compatible}\,,\\
\hbox{for}\quad P=1:\qquad &
\big[6N^2\big]C_0\,r^{N-1}+\cdots=-2\,r^N+\cdots&&
\hbox{not compatible}\,,\\
\hbox{for}\quad P=2:\qquad &
\big[6N(N+1)+2\big]C_0\,r^N+\cdots=-2\,r^N+\cdots&&
(3N^2+3N+1)C_0=-1\,,\label{contrp=2c0INF}
\\
\hbox{for}\quad P=3:\qquad &
\big[6N(N+2)+6\big]C_0\,r^{N+1}+\cdots=-2\,r^N+\cdots&&
\hbox{necessarily}\quad N=-1\,.\end{aligned}$$ The highest orders of all terms in equation for the case ${P=2}$, implying ${N<0}$, are 3A\_0\^2\[N(3N+2)C\_0+1\]r\^[2N]{} +2k(C\_0\^2 -1) + =0 , \[eq2rozvoj0omegINF\] which requires ${(3N^2+2N)C_0=-1}$. Together with constraint this implies ${N=-1}$, ${C_0=-1}$.
**To summarize**: The only possible two classes of solutions in Case II$^\infty$ are given by \[N,P\]=\[-1,3\]\^, \[N,P\]=\[-1,2\]\^. \[CaseII\_summaryINF\]
**Case III**$^\infty$
---------------------
Now, ${-2N+2=-P+4}$, that is ${N=-1+P/2}$ and ${P=2N+2}$. In such a case, the *highest* order in the key equation (\[KeyEq1INF\]) is *on both sides*, namely $r^{-l}$ with ${l=2-2N}$. This implies the condition P(P-2)=0. \[KeyEq1CaseIIIINF\] There are three subcases to be considered, namely ${P=0}$, ${P=2}$, and ${3A_0^2=-4kC_0 (P-1)(P-3)}$ with ${P\not= 0,1,2,3}$. This corresponds to ${N=-1}$, ${N=0}$, and also ${3A_0^2=-4kC_0(4N^2-1)}$ with $N\not= -1, -1/2, 0, 1/2$, respectively. The leading orders of the trace equation on both sides are 2(11N\^2+6N+1) C\_0r\^[3N]{} +[& = &]{}-2 r\^N+. \[eqtr00omegIIIINF\] Consequently, we obtain $$\begin{aligned}
&\hbox{for}\quad N=-1\,,\ P=0:\quad &
12C_0\,r^{-3}+\cdots=-2\,r^{-1}+\cdots& \qquad\hbox{not compatible}\,,\label{contrp=2c0IIIaINF}\\
&\hbox{for}\quad N=0\,,\ P=2:\quad & 2C_0+\cdots=-2+\cdots&\qquad C_0=-1\,,\label{contrp=2c0IIIbINF}\\
&\hbox{for}\quad 3A_0^2=4kC_0(1-4N^2): & (11N^2+6N+1)C_0+\cdots=0&\qquad
\hbox{not compatible}\,. \label{contrp=2c0IIIcINF}\end{aligned}$$ The incompatibility in the case is due to the fact that ${11N^2+6N+1}$ is always positive. In the case , we employ the field equation , which for ${N=0, P=2}$ requires ${3A_0^2+2k(C_0^2-1) = 0}$. Since ${C_0=-1}$ would imply ${A_0=0}$, we also end up in a contradiction.
**To summarize**: There are no possible solutions in Case III$^\infty$.
Description and study of all possible solutions in powers of $r^{-1}$ {#description_INF}
=====================================================================
Now we derive and investigate spherically symmetric solutions in the domain as ${r \to \infty}$ by completely solving the equations (\[KeyEq1INF\]), (\[KeyEq2INF\]), and their consequence (\[KeyEq3INF\]). As it has been proven in previous Section \[expansiont\_INF\], there are only two distinct cases to be discussed.\
Schwarzschild–Bach black holes in the class ${[-1,3]^\infty}$: near the singularity {#Schw_[N,P]=[-1,3]}
-----------------------------------------------------------------------------------
In the class given by ${N=-1}$, ${P=3}$ in the expansion , in *negative* powers of $r$, the only possible black hole solutions are $$\begin{aligned}
\Omega(r) {\!\!\!& = &\!\!\!}-\frac{1}{r}
+ \frac{B}{r}\,\bigg(\frac{2}{9}\,\frac{r_h^3}{r^3}
+\frac{1}{6}\,\frac{r_h^4}{r^4}
+\frac{2}{15}\,\frac{r_h^5}{r^5} +\ldots\bigg)\,, \label{IIdOmegaFULL}\\
\mathcal{H}(r) {\!\!\!& = &\!\!\!}(r-r_h)\frac{r^2}{r_h}
+ B\,\bigg( r_h^2-\frac{1}{90k}\,\frac{r_h^3 }{r^3}
-\frac{1}{140k}\,\frac{r_h^4}{r^4}-\frac{1}{210k}\,\frac{r_h^5}{r^5}+ \ldots \bigg) \,. \label{IIdH0FULL}\end{aligned}$$ These solutions represent the *class of Schwarzschild–Bach black holes* in Quadratic Gravity/the Einstein–Weyl theory. By setting ${B=0}$, the *Schwarzschild solution* is again obtained, with the *horizon located at* $r_h$.
In the limit ${r\to\infty}$, the relation implies ${{\bar r}=\Omega(r)\sim -1/r \to 0 }$. In such a limit, *the curvature singularity at ${\bar r = 0}$ is approached*, where ${\H \to \infty}$. Moreover, from the relations it follows that ${h({\bar r})\sim 1/( r_h \, \bar r) \to \infty}$ and ${f({\bar r})\sim h({\bar r})}$. Thus both metric functions of *diverge exactly in the same way as for the Schwarzschild solution, independently of the Bach parameter* $B$.
Let us derive this class of solutions. The key equation , relabeling ${l \to l+2 }$, implies C\_[l+1]{}=\^[l-2]{}\_[i=0]{} A\_iA\_[l-2-i]{}(l-1-i)(l-2-3i) l3, \[nonSchwinitcondcINF\] which gives all $C_{l+1}$ in terms of ${A_0,\ldots, A_{l-2}}$, starting form ${C_4=0}$. The trace equation yields A\_[l]{}= l1, \[nonSchwinitcondaINF\] which expresses all $A_{l}$ in terms of ${A_0,\ldots, A_{l-1}}$ and ${C_1,\ldots, C_{l}}$. Finally, the second field equation in the lowest nontrivial order ${l=0}$ gives the additional constraint C\_2=. \[nonSchwinitcond2INF\] Therefore, in this case there are *four free parameters*, namely ${A_0, C_0, C_1, C_3}$. Using we obtain ${C_2}$, and then $A_l$, $C_{l+3}$ for all ${l\ge 1}$ by the application of the recurrent relations , .
### Identification of the Schwarzschild black hole
The scalar invariants , for , now take the form B\_[ab]{} B\^[ab]{}(r ) = (45 C\_6)\^2 , C\_[abcd]{} C\^[abcd]{}(r ) \~12 r\^6 . \[BInv2INF\] Since ${A_0, C_0}$ are nonzero by definition, the necessary condition for the *Bach tensor to vanish* (which geometrically identifies the classical Schwarzschild solution) is C\_6 = 0. \[C\_6=0\] Interestingly, for such a setting, the expansion coefficients simplify enormously to && A\_i=A\_0(-)\^i i 0 ,\
&& C\_2=, C\_3=, C\_i=0 i 4. The first sequence is a geometrical series, while the second series is truncated to the 3rd-order polynomial. Thus the metric functions can be written in the closed form $$\begin{aligned}
\Omega(r) {\!\!\!& = &\!\!\!}\frac{A_0}{r} \,\sum_{i=0}^\infty \,\Big(\!\!-\frac{C_1+1}{3C_0\,r}\Big)^i
=\frac{A_0}{r+(C_1+1)/(3C_0)}\,, \label{IIdOmega}\\
\mathcal{H}(r) {\!\!\!& = &\!\!\!}C_0\,r^3+C_1\,r^2+\frac{C_1^2-1}{3C_0}\, r+\frac{(C_1+1)^2(C_1-2)}{27 C_0^2 } \,. \label{IIdH0}\end{aligned}$$ In view of , we are free to chose the gauge A\_0=-1,C\_1=-1, \[IId\_a0\] so that the metric functions become =(r) = -, (r) = -r\^2+C\_0r\^3 . \[IIdH0Schw\] This is exactly the *Schwarzschild black hole metric* in the form and . It also identifies the physical meaning of the coefficient $C_0$ as $$C_0 = \frac{1}{r_h} \,,
\label{IIdH0SchwC0}$$ where $r_h$ determines the *horizon position*, the root of $\mathcal{H}$ given by . Of course, the Schwarzschild horizon is given by ${r_h=-1/(2m)}$, i.e., ${C_0=-2m}$. All free parameters of such solution are thus fixed and fully determined.
### More general Schwarzschild–Bach black holes
For the physical interpretation of the more general solutions in this family, it is convenient to introduce the *Bach parameter $B$ proportional to ${C_6}$* entering , which for the gauge choice reads ${ C_6 = -C_3 / (90k C_0)}$. We also naturally require $B$ to be a *dimensionless* parameter, so that the best choice seems to be B C\_0\^2C\_3 = . \[b\_definiceINF\] With such $B$ as the key parameter in the expansions , and the same natural gauge , the recurrent relations , yield an explicit solution of the field equations in a simple form $$\begin{aligned}
& A_0 = -1\,,\qquad
A_1 = 0\,,\qquad
A_2 = 0\,,\nonumber\\
& A_3 = \frac{2}{9}\,r_h^3\,B \,, \qquad
A_4 = \frac{1}{6}\,r_h^4\,B \,, \qquad
A_5 = \frac{2}{15}\,r_h^5\,B \,, \nonumber\\
& A_6 = \frac{1}{9}\,r_h^6\,\Big( 1-\frac{7}{360kr_h^2}-\frac{10}{9}\,B \Big)\,B\,, \ldots\,,\label{IId_expansionaINFa} \\
& C_0 = r_h^{-1} \,,\qquad
C_1 = -1 \,,\qquad
C_2 = 0 \,,\nonumber\\
& C_3 = r_h^2\,B \,,\qquad
C_4 = 0 \,,\qquad
C_5 = 0 \,,\nonumber\\
& C_6 = -\frac{1}{90k}\,r_h^3\,B \,,\qquad
C_7 = -\frac{1}{140k}\,r_h^4\,B \,,\qquad
C_8 = -\frac{1}{210k}\,r_h^5\,B \,, \ldots\,.
\label{IId_expansioncINFc}\end{aligned}$$ This gives the explicit expansion , .
The corresponding scalar invariants at ${\bar r = 0}$ are B\_[ab]{} B\^[ab]{}(r ) = B\^2 ,C\_[abcd]{} C\^[abcd]{} (r ) \~r\^6 , \[BInv2INFfinal\] which can be compared with the invariants evaluated at the horizon $\bar r_h$ B\_[ab]{}B\^[ab]{}(r\_h) = b\^2 ,C\_[abcd]{} C\^[abcd]{}(r\_h) = 12r\_h\^4(1+b)\^2 , \[BInv2final\] obtained previously for the class ${[n,p]=[0,1]}$ of the Schwarzschild–Bach black holes. There is a striking similarity between the two expressions for ${B_{ab}\,B^{ab}}$, and thus we could be inclined to directly identify the Bach parameter $B$ with the parameter $b$. However, is should again be emphasized that $B$ determines the value of the Bach invariant *at the Weyl curvature singularity* ${\bar r=0}$, while $b$ determines its value *at the horizon* $\bar r_h$. And these values are, in general, distinct.
Bachian vacuum in the class ${[N,P]=[-1,2]^\infty}$ {#Schw_[N,P]=[-1,2]}
---------------------------------------------------
Finally, it remains to analyze the second possibility in the Case II$^\infty$. For ${N=-1}$, ${P=2}$ the key equation , relabeling ${l \to l+2}$, gives C\_l=\^[l-2]{}\_[i=0]{} A\_i A\_[l-2-i]{}(l-1-i)(l-2-3i) l3. \[\[-1,2\]initcondc\] Equation in its lowest orders ${l=1, 2 }$ puts the constraints $$A_1=\tfrac{1}{2}A_0C_1 \,, \qquad C_0=-1\,,$$ and for higher $l$ implies A\_[l-1]{}= \^[l-1]{}\_[i=1]{} C\_iA\_[l-1-i]{} l3. \[\[-1,2\]initconda\] The equation gives no additional constraint. There are thus *three free parameters*, namely ${A_0, C_1, C_2}$, and all other coefficients are determined by the relations , , starting as $$\begin{aligned}
& A_2 = \frac{A_0}{3}(C_1^2 + C_2)\,,\qquad
A_3 = \frac{ A_0 }{4}C_1 (C_1^2 + 2C_2)\,,\nonumber\\
& A_4 = \frac{A_0 }{5}\Big(C_1^4 + 3 C_1^2 C_2 + C_2^2 +\frac{A_0^2}{192 k}(C_1^2 + 4 C_2)\Big)\,, \ldots\,,
\label{[-1,2]_expansionaINFa} \\
& C_3 = 0 \,,\qquad
C_4 = \frac{A_0^2 }{240 k}(C_1^2 + 4C_2) \,,\qquad
C_5 = \frac{A_0^2}{240 k} C_1 (C_1^2 + 4C_2) \,,\nonumber\\
& C_6 = \frac{A_0^2 }{67200 k^2}\Big(3A_0^2+ 4k(59 C_1^2 + 26 C_2)\Big)(C_1^2 + 4C_2) \,, \ldots\,.
\label{[-1,2]_expansioncINFc}\end{aligned}$$
### Identification of flat Minkowski space
Now, for very large $r$ the scalar invariants , behave as B\_[ab]{} B\^[ab]{}(r) = C\_4\^2 ,C\_[abcd]{} C\^[abcd]{}(r) \~C\_4\^2 . \[\[-1,2\]\_Inv\_INF\] Interestingly, they *remain finite*, so that for ${r\to\infty}$ there is *no physical singularity*. Moreover, for ${C_4 \ne0}$ they are nonzero. In fact, the necessary condition for both the Bach and Weyl tensor invariants to vanish is ${C_4=0}$, that is ${C_1^2 + 4C_2 =0}$. For such a choice, we obtain the relation ${C_2=-\frac{1}{4} C_1^2}$, and then all the coefficients , simplify enormously to ${A_i=A_0\,(\frac{1}{2}C_1)^i}$ for all $i$, and ${C_i=0}$ for all ${i\ge3}$. The metric functions thus reduce to (r) = \_[i=0]{}\^()\^i =, (r) = -(r-C\_1)\^2 . \[\[-1,2\]\_Omega\_H0\] Using the gauge freedom we can always set A\_0=-1,C\_1=0, \[\[-1,2\]\_a0\] and the functions take the trivial form =(r) = -, (r) = -r\^2 . \[\[-1,2\]\_solution\] In view of , , we conclude that the case ${C_4=0}$ gives the Schwarzschild metric with trivial value ${C_0=-2m=0}$ which is *just flat Minkowski space without any horizon* (formally ${r_h=\infty}$). Of course, for flat space, both the Bach and the Weyl tensor vanish everywhere.
### Bachian vacuum
Now, the complete class of solutions ${[N,P]=[-1,2]^\infty}$ can be naturally analyzed if we introduce the *Bach parameter $B_v$ proportional to ${C_4}$* because, due to , such solutions admit general Bach and Weyl tensors. With the same gauge , we observe from that ${ C_4 = C_2/(60 k)}$, so that it is more convenient to choose the equivalent parameter $C_2$, instead. The simplest choice is B\_v C\_2 . \[beta\_definiceINF\] With the only remaining parameter $B_v$ (in this case it is not dimensionless), the coefficients , simplify to $$\begin{aligned}
& A_0 = -1\,,\quad
A_1 = 0\,,\quad
A_2 = -\frac{1}{3}\,B_v\,,\quad
A_3 = 0\,, \nonumber\\
& A_4 = -\frac{1}{5}\,\Big(\frac{1}{48k}+B_v \Big)\,B_v \,, \quad
A_5 = 0 \,, \ldots \label{[-1,2]_expansionaINFaa} \\
& C_0 = -1 \,,\quad
C_1 = 0 \,,\quad
C_2 = B_v \,,\quad
C_3 = 0 \,,\nonumber\\
& C_4 = \frac{1}{60k}\,B_v \,,\quad
C_5 = 0 \,,\quad
C_6 = \frac{1}{700k}\,\Big(\frac{1}{8k}+\frac{13}{3}B_v \Big)\,B_v \,, \ldots
\label{[-1,2]_expansioncINFcc}\end{aligned}$$ yielding an explicit solution $$\begin{aligned}
\Omega(r) {\!\!\!& = &\!\!\!}-\frac{1}{r}
- B_v\,\bigg(\frac{1}{3\,r^3}
+\frac{1}{5\,r^5}\,\Big(\frac{1}{48k}+B_v \Big) +\ldots\bigg)\,, \label{[-1,2]_OmegaFULL}\\
\mathcal{H}(r) {\!\!\!& = &\!\!\!}-r^2
+ B_v\,\bigg( 1+\frac{1}{60k\,r^2}
+\frac{1}{700k\,r^4}\,\Big(\frac{1}{8k}+\frac{13}{3}B_v \Big)
+ \ldots \bigg) \,. \label{[-1,2]_H0FULL}\end{aligned}$$ The corresponding scalar invariants now read B\_[ab]{} B\^[ab]{} (r)= B\_v\^2 , C\_[abcd]{} C\^[abcd]{} (r) \~ 0 . \[\[-1,2\]\_Inv\_INF\_beta\] Therefore, we may conclude that this class of metrics ${[N,P]=[-1,2]^\infty}$ can be understood as a *one-parameter Bachian generalization of flat space* (that is the limit of black hole solutions without mass and horizon) *with a nonzero Bach tensor whose magnitude is determined by the parameter $B_v$*, i.e., the “massless limit” of the previous class ${[N,P]=[-1,3]^\infty}$.
Interestingly, in the limit ${r\to\infty}$, the expressions , now imply && [|r]{}=(r) \~-1/r 0 , \[\[-1,2\]A\]\
&& h \~1, f \~1. \[\[-1,2\]C\] Both the metric functions $h$ and $f$ thus remain nonzero and finite, i.e., in this limit we are not approaching a horizon nor a singularity. In fact, for ${\bar r \to 0}$ the metric *becomes conformally flat*. Interestingly, the Bach invariants and are very similar.
Consistency check of the limit $[-1,3]^\infty\,\rightarrow\, [-1,2]^\infty$
---------------------------------------------------------------------------
Let us consider a “consistency check” between the class of solutions $[-1,3]^\infty$, described by –, and the class $[-1,2]^\infty$, described by –, where the coefficients will now be denoted by hats.
The transition from $[-1,3]^\infty$ to $[-1,2]^\infty$ requires C\_0 0,C\_i C\_[i-1]{}, i1,A\_i A\_i, i0. The relation for ${l=1}$, that is $3C_0A_1=-A_0(1+C_1)$, in this limit leads to C\_1 -1, C\_0=-1 , while the relations for $C_{l+1}$ and for $\hat C_l$ remain the same. Moreover, the relation for $A_l$, -l\^2C\_0A\_[l]{}=A\_[l-1]{} +\^[l]{}\_[i=1]{} C\_iA\_[l-i]{} l1, for $C_0=0$ leads to A\_[l-1]{} = \^[l-1]{}\_[i=1]{} C\_iA\_[l-1-i]{} l2, which is exactly and thus concludes the consistency check.
Note that from the free parameters of the family $[-1,3]^\infty$, two parameters become determined, namely ${C_0\rightarrow0}$, ${C_1\rightarrow\hat C_0=-1}$, and one parameter ${C_2\rightarrow\hat C_1}$ becomes undetermined since $3C_0C_2=C_1^2-1\rightarrow 0$. Therefore, four free parameters ${A_0,C_0,C_1,C_3}$ of the $[-1,3]^\infty$ family reduce to three free parameters ${\hat A_0,\hat C_1, \hat C_2}$ of the $[-1,2]^\infty$ family.
Summary and relations to previous results {#summary}
=========================================
In this section, let us summarize all the distinct and explicit families of spherically symmetric vacuum spacetimes in QG, expressed both in powers of ${\Delta \equiv r-r_0}$ and $r^{-1}$. Moreover, we identify these families with solutions previously discussed in the literature.
In particular, in [@Stelle:1978; @LuPerkinsPopeStelle:2015; @LuPerkinsPopeStelle:2015b], various classes of static spherically symmetric solutions to higher-derivative gravity equations were identified and denoted by the symbol $(s,t)$, using the *standard spherically symmetric form* . Such a classification was based on the powers $s$ and $t$ of the *leading terms* of a Laurent expansion of the two metric functions, namely[^5] f\^[-1]{}([|r]{}) [& = &]{}A([|r]{}) \~[|r]{}\^[s]{}, \[rcef2\]\
h([|r]{}) [& = &]{}B([|r]{}) \~[|r]{}\^[t]{} ,\[rceh2\] in the domain ${\bar r \to 0}$. It was shown in [@Stelle:1978; @LuPerkinsPopeStelle:2015b] that there are *three main solution families* corresponding to the following choices of ${(s,t)}$: (s,t) [& = &]{}(0,0)\_0, \[rodina1\]\
(s,t) [& = &]{}(1,-1)\_0, \[rodina2\]\
(s,t) [& = &]{}(2,2)\_0, \[rodina3\] where the subscript “$\,_0$” indicates the expansion around the origin ${\bar r =0}$.
In addition, the following *three families* ${(w,t)}$ were identified in [@LuPerkinsPopeStelle:2015b; @PerkinsPhD] using a series expansion around a *finite* point ${{\bar r}\to {\bar r}_0 \ne0}$: (w,t)=(1,1)\_[[|r]{}\_0]{},\
(w,t)=(0,0)\_[[|r]{}\_0]{},\
(w,t)=(1,0)\_[[|r]{}\_0]{}, where w=-s , that is ${f \sim {\bar r}^{\,w}}$ and ${h \sim {\bar r}^{\,t}}$. The subscript “$\,_{{\bar r}_0}$” indicates the expansion around ${\bar r_0}$.
In fact, *we have recovered all these families* of solutions in the present paper, and *we have also identified some additional families*.
To find the specific mutual relations, first let us note that from the relation between the spherically symmetric radial coordinate $\bar r$ and the Kundt coordinate $r$, that is ${\bar r=\Omega(r)}$, it follows using and that
- ${\bar r \rightarrow 0\,\,\,}$ for ${r\rightarrow r_0}$, ${n>0}$, and also for ${r\rightarrow \infty}$, ${N<0}$,
- ${\bar r \rightarrow \bar r_0\,}$ for ${r\rightarrow r_0}$, ${n=0}$, and also for $r\rightarrow\ \infty$, ${N=0}$,
- ${\bar r \rightarrow \infty}$ for ${r\rightarrow r_0}$, ${n<0}$, and also for $r\rightarrow\ \infty$, ${N>0}$.
Now let us find a relation between the powers $(s,t)$ introduced by and , respectively, and the coefficients ${[n, p]}$ employed in this paper. They are the analogous leading powers of the two metric functions $\Omega$ and $\H$, respectively. For ${n\not=0}$, such a relation is found using the expressions with ${\bar{r} = \Omega(r)}$ and , for ${r \to r_0}$. It turns out that s = , t = 2+. \[st-np\] Analogously, using , , we obtain the relations s = , t = 2+ \[st-NP\] for the asymptotic expansion of the metric functions as ${r\rightarrow \infty}$. Thus, for ${n\not=0}$ and ${N\not=0}$, it immediately follows that
- the family ${(s,t)=(0,0)_0}$ corresponds to ${[N,P]=[-1,2]^\infty}$,
- the family ${(s,t)=(0,0)^ \infty}$ corresponds to ${[n, p]=[-1,2]}$,
- the family ${(s,t)=(1,-1)_0}$ corresponds to ${[N,P]=[-1,3]^\infty}$,
- the family ${(s,t)=(2,2)_0}$ corresponds to ${[n,p]=[1,0]}$,
where the superscript “$\,^\infty$” in ${(0,0)^ \infty}$ indicates the expansion as ${{\bar r}\to\infty}$.
The two admitted cases with ${n=0}$ have to be analyzed separately (there are no cases with ${N=0}$). In the generic case when ${a_1\not=0}$, using , , , we obtain that w=p, t=p. Therefore, for ${n=0}$ and ${a_1\not=0}$ we conclude that
- the family ${(w,t)=(0,0)_{\bar r_0}}$ corresponds to ${[n, p]=[0,0]}$,
- the family ${(w,t)=(1,1)_{\bar r_0}}$ corresponds to ${[n,p]=[0,1]}$,
*completing the identification of all our main six classes of solutions*. Note that for ${n=0}$, ${a_1\not=0}$ the relation between $\Delta$ and $\bar \Delta$ is ${\bar\Delta \equiv \bar r -\bar r_0 \sim a_1\Delta}$. Therefore, a series expansion with *integer steps in* $\Delta$ corresponds to a series expansion with *integer steps in* $\bar\Delta$ in the physical radial coordinate $\bar r$.
All four possible generic families compatible with the field equations as ${r \to r_0}$ and the series expansion – are summarized in Table \[tbl:01\], while the two cases compatible with the field equations as ${r \to \infty}$ and , , are summarized in Table \[tbl:02\]. We also indicate their physical interpretation and the corresponding Section, in which these solutions are described and studied.
-------------------- --------------------- -------------------------------------------------------- -------------------------------- -- -- --
Class $[n,p]$ Family $(s,t)$ Interpretation Section
\[0.5mm\] $[-1,2]$ $(0,0)^\infty$ Schwarzschild black hole \[Schw\_\[n,p\]=\[-1,2\]\]
\[1mm\] $[0,1]$ $(-1,1)_{\bar r_0}$ Schwarzschild–Bach black holes (near the horizon) \[SchwaBach\_\[n,p\]=\[0,1\]\]
\[1mm\] $[0,0]$ $(0,0)_{\bar r_0}$ generic solution, including the Schwa–Bach black holes \[SchwaBach\_\[n,p\]=\[0,0\]\]
\[1mm\] $[1,0]$ $(2,2)_0$ Bachian singularity (near the singularity) \[SchwaBach\_\[n,p\]=\[1,0\]\]
\[1mm\]
-------------------- --------------------- -------------------------------------------------------- -------------------------------- -- -- --
: All possible generic types of solutions to Quadratic Gravity and the Einstein–Weyl theory that can be written as the power series – expanded around any constant value $r_0$. []{data-label="tbl:01"}
\
--------------------------- ---------------- ------------------------------------------------------- ---------------------------- -- -- --
Class $[N,P]^\infty$ Family $(s,t)$ Interpretation Section
\[0.5mm\] $[-1,3]^\infty$ $(1,-1)_0$ Schwarzschild–Bach black holes (near the singularity) \[Schw\_\[N,P\]=\[-1,3\]\]
\[1mm\] $[-1,2]^\infty$ $(0,0)_0$ Bachian vacuum (near the origin) \[Schw\_\[N,P\]=\[-1,2\]\]
\[1mm\]
--------------------------- ---------------- ------------------------------------------------------- ---------------------------- -- -- --
: All possible generic types of solutions to Quadratic Gravity and the Einstein–Weyl theory that can be written as the power series , expanded as ${r\to\infty}$. []{data-label="tbl:02"}
\
Special subclasses with $n=0$ {#relation_to_previous_results 2}
-----------------------------
In addition to the above six main classes of solutions, in the case given by ${n=0}$ we have identified some other special subclasses, including a new one. These are *not* given as integer steps in $\bar r$ or ${\bar\Delta}$, so that these are additional classes from the point of view of expansions in powers of ${\bar r -\bar r_0}$ in the physical radial coordinate. In our Kundt coordinate $r$, they just naturally appear as special cases of the solutions with ${n=0}$, namely when ${a_1=0\not=a_2}$ and ${a_1=0=a_2}$.
When ${a_1=0\not=a_2}$, the relation is ${\bar\Delta\sim a_2\,\Delta^2}$, and thus a series expansion with integer steps in $\Delta$ leads to *(half integer) steps* $\bar \Delta^{1/2}$. Using , in such a case we obtain w=+1,t=. For ${a_1=0}$ and ${a_2\not=0}$, we thus conclude that
- the family $(w,t)=(\tfrac{3}{2},\tfrac{1}{2})_{\bar r_0,1/2}$ corresponds to ${[n, p]=[0,1]_{a_1=0}}$,
- the family ${(w,t)=(1,0)_{\bar r_0,1/2}}$ corresponds to ${[n,p]=[0,0]_{a_1=0}}$.
Analogously, when ${a_1=0=a_2}$ and ${a_3\not=0}$, the relation is ${{\bar \Delta}\sim a_3\,\Delta^3}$, and thus integer steps in $\Delta$ corresponds to steps in $\bar \Delta^{1/3}$. The relations are now w=,t=. Thus for ${a_1=0=a_2}$ and ${a_3\not=0}$, we conclude that
- the family $(w,t)=(\tfrac{4}{3},0)_{\bar r_0,1/3}$ corresponds to ${[n, p]=[0,0]_{a_1=0=a_2}}$.
Concerning the geometrical and physical interpretation of these special solutions, it can be generally said that the classes with ${n=0}$ contain (among other solutions) black holes and wormholes. In particular, the class ${[n=0,p=1]}$ represents a *black hole spacetime* since it admits a Killing horizon at ${r_h=r_0}$, see . As pointed out in [@LuPerkinsPopeStelle:2015b], a *wormhole spacetime* is characterized by admitting a finite value of ${{\bar r}_0}$ where ${f=0}$ while ${h\not=0}$. Therefore, for a series expansion around this point, necessarily ${n=0=p}$ (since $\H \not= 0$), and ${a_1=0}$ (since $\Omega'=0$). Thus wormholes may appear only in the class ${[0,0]_{a_1=0}}$.
The family of solutions ${(\tfrac{3}{2},\tfrac{1}{2})_{\bar r_0,1/2}}$ was identified in [@LuPerkinsPopeStelle:2015b] and interpreted in [@PerkinsPhD] as an “unusual” type of a horizon. However, it was stated therein that it is a solution to QG only for ${\beta\not =0}$, which implies ${R\not=0}$. Thus it seems that this class does not coincide with our class ${[0,1]_{a_1=0}}$ since, for all our classes, ${R=0}$ by assumption.
Our family ${[0,0]_{a_1=0}}$ corresponds to the family ${(1,0)_{\bar r_0,1/2}}$ of [@LuPerkinsPopeStelle:2015b; @PerkinsPhD], while our family $[0,0]_{a_1=0=c_1=c_3}$, where *only even powers* in $\Delta$ are considered (indicated by the subscript “$\,_E$”), corresponds to the family $(1,0)_{\bar r_0,E}$ of [@LuPerkinsPopeStelle:2015b; @PerkinsPhD]. Both these families describe wormholes with two different (half-integer wormhole) and two same patches (integer wormhole), respectively, see [@PerkinsPhD]. Note that the Bach invariant for wormholes in the ${[0,0]_{a_1=0}}$ class is always nonvanishing.
To our knowledge, the specific family ${[0,0]_{a_1=0=a_2}}$ *has not yet been considered*, and it corresponds to a new family ${(\tfrac{4}{3},0)_{\bar r_0,1/3}}$ in the notation of [@LuPerkinsPopeStelle:2015b].
It also seems that the *generic solution* $[0,0]$, with the highest number of free parameters, can be connected to all other solutions, and it represents an expansion around a generic point in these spacetimes.
In Table \[tab:3\], we summarize all the classes and subclasses found and identified both in the physical and Kundt coordinates, grouped according to the regions in which the expansions are taken in the usual radial coordinate $\bar r$.
Family $[n,p]$ or $[N,P]^\infty\!\!$ Parameters Free param. Interpretation
-------------------------------- ------------------------------- ----------------------------- ------------------ ------------------------------
$(s,t)$
$(2,2)_0$ $[1,0]$ $a_0,c_0,c_1,c_2,r_0$ $5\rightarrow 3$ Bachian singularity (nS)
$(2,2)_{0,E}$ $[1,0]_{c_1=0=c_3}$ $a_0,c_0,r_0$ $3\rightarrow 1$ Bachian singularity (nS)
$(1,-1)_0$ $[-1,3]^\infty$ $A_0,C_0,C_1,C_3$ $4\rightarrow 2$ Schwa–Bach black holes (S)
$(0,0)_0$ $[-1,2]^\infty$ $A_0,C_1,C_2$ $3\rightarrow 1$ Bachian vacuum (nS)
$(w,t)$
$(1,1)_{\bar r_0}$ $[0,1]$ $a_0,c_0,c_1,r_0=r_h$ $4\rightarrow 2$ Schwa–Bach black holes (S)
$(3/2,1/2)_{\bar r_0,1/2}\!\!$ $[0,1]_{a_1=0}$ $a_0,c_0,r_0$ $3\rightarrow 1$ “unusual” horizon (nS)
$(0,0)_{\bar r_0}$ $[0,0]$ $a_0,a_1,c_0,c_1,c_2,r_0\!$ $6\rightarrow 4$ generic solution (S)
$(1,0)_{\bar r_0,1/2}$ $[0,0]_{a_1=0}$ $a_0,c_0,c_1,c_2,r_0$ $5\rightarrow 3$ half-integer wormhole (nS)
$(1,0)_{\bar r_0,E}$ $[0,0]_{a_1=0=c_1=c_3}$ $a_0,c_0,r_0$ $3\rightarrow 1$ symmetric wormhole (nS)
$(4/3,0)_{\bar r_0,1/3}$ $[0,0]_{a_1=0=a_2}$ $a_0,c_0,c_1,r_0$ $4\rightarrow 2$ not known (nS) — new
$(s,t)$
$(0,0)^\infty$ $[-1,2]$ $a_0,c_1,r_0$ $3\rightarrow 1$ Schwarzschild black hole (S)
: All solutions, sorted according to the physical regions in which the expansions are taken. The subscripts “$\,_0$”, “$\,_{{\bar r}_0}$” and the superscript “$\,^\infty$” denote solutions $(s,t)$ or $(w,t)$ near ${\bar r=0}$, ${\bar r={\bar r_0}}$, and ${\bar r\rightarrow\infty}$, respectively. The subscript “$\,_E$” indicates that only even powers are present in the expansion, while “$\,_{1/2}$” and “$\,_{1/3}$”indicate that fractional powers are present. Specific number of free parameters is given before and after removing two parameters by the gauge freedom in the Kundt coordinates. In physical coordinates, only one parameter can be removed by rescaling . The symbols “(S)” or “(nS)” indicate that a class of solutions contains or does *not* contain the Schwarzschild black hole, respectively. []{data-label="tab:3"}
\
Discussion and analysis of the Schwarzschild–Bach black holes {#discussion-and-figures}
=============================================================
In this section, we discuss the behavior of the series expressing the Schwarzschild–Bach black hole solutions , . For our analysis, we choose the same values of the parameters as in our previous paper [@PodolskySvarcPravdaPravdova:2018], namely ${r_h=-1}$, ${k = 0.5}$, ${b=0.3633018769168}$. Such a very special value of $b$ is “close” to the asymptotically flat case.[^6]
The key observation for estimating the radius of convergence can be made from Figure \[Figasymp\]. Interestingly, the ratios of subsequent terms ${\frac{\alpha_n}{\alpha_{n-1}}}$ and ${-\frac{\gamma_n}{\gamma_{n-1}}}$ given by the recurrent relations are *approaching a constant asymptotically*. This suggests that both series given by $\alpha_n$ and $\gamma_n$ behave as *geometric series for large* $n$, with the ratio $q$ being apparently *equal for both the series*. Therefore, the series for $\Omega$ and $\cal H$, given by , , should be convergent for ${-1-\frac{1}{q}<r<-1+\frac{1}{q}}$, where ${q \approx 1.494}$, that is in the interval ${r\in(-1.67,-0.33)}$.
![The Schwarzschild–Bach solution ${[0,1]}$ given by , . The ratios $\frac{\alpha_n}{\alpha_{n-1}}$ (blue) and ${-\frac{\gamma_n}{\gamma_{n-1}}}$ (red) for the first 3000 coefficients $\alpha_i$ and $\gamma_i$ given by the recurrent formula are ploted. []{data-label="Figasymp"}](fig1.pdf){height="62mm"}
Figure \[fig:OmegaH\] illustrates the convergence of the metric functions $\Omega(r)$ and $\H(r)$ in the Kundt coordinate $r$. In the domain of convergence, denoted by vertical dashed lines, the solution fully agrees with the numerical solution of the field equations.
For comparison, Figure \[fig:fh\] illustrates the convergence of the corresponding metric functions $f(\bar r)$ and $h(\bar r)$ in the standard spherically symmetric coordinates. The solution quickly converges, and approaches a numerical solution even at a large distance from the horizon located at ${{\bar r}_h =1}$.
From the value of ${\Omega(r)\equiv \bar r}$ at the *lower* boundary of the domain of convergence shown in Figure \[fig:OmegaH\], we can easily read off its value ${\bar r \approx 0.53}$ in the usual radial coordinate. In contrast, the value of the coordinate $\bar r$ given by $\Omega(r)$ at the *upper* boundary remains unclear since it depends on the precise value of the series at the upper boundary of the domain of convergence. In fact, we cannot even say with certainty that the radius of convergence in the standard spherical coordinate $\bar r $ is finite — it may well extend up to ${\bar r \to \infty}$.
Finally, it is illustrative to show explicitly that, in contrast to the Schwarzschild solution, the metric functions $f(\bar r)$ and $h(\bar r)$ for the Schwarzschild–Bach black holes are *not equal*. This is clearly seen from their plots in Figure \[fig:fhnearhorizon\].
![The metric functions $\Omega(r)$ (left) and $\H(r)$ (right) for the Schwarzschild–Bach solution $[0,1]$. The first 20 (red), 50 (orange), 100 (green), and 500 (blue) terms of the series , for $\Omega$ and $\H$ are also compared with a numerical solution (black). Boundaries of the domain of convergence are denoted by vertical dashed lines. Within this radius of convergence, all these functions overlap with the numerical solution, except the lowest shown 20th order of $\Omega$ near the top right corner on the left graph. []{data-label="fig:OmegaH"}](fig2.pdf){height="50mm"}
![The metric functions $f(\bar r)$ (left) and $h(\bar r)$ (right) for the Schwarzschild–Bach solution \[0,1\] in the standard coordinates. The first 20 (red), 50 (orange), 100 (green), and 300 (blue) terms of the series are plotted. A numerical solution (black) overlaps with the blue curve, even far above the horizon located at ${{\bar r}_h=1}$ (here up to ${\bar r = 20\, \bar r_h}$).[]{data-label="fig:fh"}](fig3.pdf){height="48mm"}
![The metric functions $f(\bar r)$ (blue) and $h(\bar r)$ (red) in the near-horizon region for the Schwarzschild–Bach solution \[0,1\]. These two functions are clearly distinct. They both vanish at the horizon, located here at ${{\bar r}_h=1}$[]{data-label="fig:fhnearhorizon"}](fig4.pdf){height="50mm"}
There are *three classes* of solutions *containing the Schwarzschild black hole as a special case*, namely the ${[0,0]}$ class with *four* free parameters and the classes ${[0,1]}$ and ${[-1,3]^{\infty}}$, both with *two* free parameters, see Table \[tab:3\]. (The class ${[-1,2]}$ contains *only* the Schwarzschild solution.) The solution ${[0,0]}$ describes a generic point of a static, spherically symmetric spacetime in QG, including also black-hole and wormhole solutions. A natural question is whether the solutions ${[0,1]}$ and ${[-1,3]^{\infty}}$ describe *the same black hole at two different regions* (near the horizon and near the singularity, respectively). We have not arrived at a definite answer yet. Nevertheless the Bach invariant for the class ${[-1,3]^{\infty}}$ approaches a *finite constant* as ${|r| \rightarrow \infty}$ corresponding to ${\bar r \to 0}$, see expression , while analytical and numerical results describing the behavior of the Bach invariant of the ${[0,1]}$ class of solutions as the value of $r$ decreases below the horizon seems to suggest that in this case the Bach invariant is *unbounded*, see Figure \[fig:bach\]. If this is indeed the case, then the classes ${[0,1]}$ and ${[-1,3]^{\infty}}$ must describe *distinct* generalizations of the Schwarzschild black hole admitting a nontrivial Bach tensor.
![The Bach invariant inside the horizon of the Schwarzschild–Bach black holes ${[0,1]}$ calculated from first 20 (red), 50 (green), and 300 (blue) terms, compared with the numerical solution (black). The lower boundary of the domain of convergence is indicated by the vertical dashed line. The horizon is located at ${r_h=-1}$. The insert in the upper right corner shows the numerical value to much lower value of the coordinate $r$, indicating a possible divergence as ${r \rightarrow -\infty}$, that is as ${\bar r \to 0}$.[]{data-label="fig:bach"}](fig5.pdf){height="55mm"}
Main physical properties of the Schwarzschild–Bach black holes {#physics}
==============================================================
Specific observable effects on test particles caused by the Bach tensor {#geodeviation}
-----------------------------------------------------------------------
In this section we demonstrate that the two parts $\B_1$, $\B_2$ of the Bach tensor , , entering the invariant , that distinguish the Schwa–Bach and the Schwarzschild black holes, can be explicitly observed via a *specific influence on particles*. It is well known that a *relative motion* of freely falling test particles (observers) directly encodes specific components of the spacetime curvature, such as the tidal deformation in the vicinity of a black hole, or a transverse effect of gravitational waves measurable by a laser interferometer detector. This is described by the *equation of geodesic deviation*, see [@BicakPodolsky:1999; @PodolskySvarc:2012] for a recent review with historical remarks and description of the formalism that we are going to employ here.
### Interpreting solutions to Quadratic Gravity using geodesic deviation
To obtain physically measurable information about the relative motion, we have to choose an *orthonormal frame* ${\{{\mbox{\boldmath$e$}}_{(0)}, {\mbox{\boldmath$e$}}_{(1)}, {\mbox{\boldmath$e$}}_{(2)}, {\mbox{\boldmath$e$}}_{(3)}}\}$ such that ${{\mbox{\boldmath$e$}}_{(a)}\cdot{\mbox{\boldmath$e$}}_{(b)}=\eta_{ab}}$, where the time-like vector ${{\mbox{\boldmath$e$}}_{(0)}={\mbox{\boldmath$u$}}}$ is observer’s $4$-velocity. Projecting the equation of geodesic deviation onto this frame we obtain Z\^[([i]{})]{}= R\^[([i]{})]{}\_[(0)(0)([j]{})]{}Z\^[([j]{})]{} ,[i]{},[j]{}=1,2,3, \[InvGeoDev\] where $$\label{PhysAccel}
\ddot Z^{(\rm{i})} \equiv e^{(\rm{i})}_a\,\frac{{{\rm D}}^2 Z^a}{{{\rm{d}}}\, \tau^2}
=e^{(\rm{i})}_a\,{Z^a}_{;cd}\, u^c u^d \,, \qquad \hbox{and} \qquad
{R_{(\rm{i})(0)(0)(\rm{j})}\equiv R_{abcd} \,e^a_{(\rm{i})}u^b u^c e^d_{(\rm{j})}} \,.$$ Spacetime curvature, characterized by the Riemann tensor, can then be decomposed into the traceless Weyl tensor, the Ricci tensor, and the scalar curvature $R$. Its projection (\[PhysAccel\]) gives $$\begin{aligned}
\label{DecompFrame}
R_{(\rm{i})(0)(0)(\rm{j})}=C_{(\rm{i})(0)(0)(\rm{j})}+\tfrac{1}{2}\big(R_{(\rm{i})(\rm{j})}
-\delta_{\rm{i}\rm{j}}\,R_{(0)(0)}\big)-\tfrac{1}{6}\,R\,\delta_{\rm{i}\rm{j}} \,.\end{aligned}$$ Moreover, the vacuum field equations of Quadratic Gravity (including the Einstein–Weyl theory), ${R_{ab}=4k\, B_{ab}}$ implying ${R=0}$, can be employed. Substituting these relations into (\[DecompFrame\]), we finally obtain the *invariant form of the equation of geodesic deviation* (\[InvGeoDev\]) as $$\begin{aligned}
\ddot{Z}^{(\rm{i})}= C_{(\rm{i})(0)(0)(\rm{j})}\,Z^{(\rm{j})}
+2k\big(B_{(\rm{i})(\rm{j})}\,Z^{(\rm{j})}-B_{(0)(0)}\,Z^{(\rm{i})}\big)\,. \label{InvGeoDevExpl}\end{aligned}$$ Of course, ${C_{(\rm{i})(0)(0)(\rm{j})}=C^{(\rm{i})}_{\quad(0)(0)(\rm{j})}}$ and ${B_{(\rm{i})(\rm{j})}=B^{(\rm{i})}_{\quad(\rm{j})}}$ since the spatial part of the frame is Cartesian. The Weyl tensor projections ${C_{(\rm{i})(0)(0)(\rm{j})}}$ can be further decomposed and expressed in terms of the Newman–Penrose scalars $\Psi_A$ with respect to the (real) *null frame* ${\{{\mbox{\boldmath$k$}}, {\mbox{\boldmath$l$}}, {\mbox{\boldmath$m$}}_{i} \}}$ which is defined by $$\begin{aligned}
{\mbox{\boldmath$k$}}={{\textstyle\frac1{\sqrt{2}}}}({\mbox{\boldmath$u$}}+{\mbox{\boldmath$e$}}_{(1)})\,, \qquad {\mbox{\boldmath$l$}}={{\textstyle\frac1{\sqrt{2}}}}({\mbox{\boldmath$u$}}-{\mbox{\boldmath$e$}}_{(1)})\,,
\qquad {\mbox{\boldmath$m$}}_{i}={\mbox{\boldmath$e$}}_{(i)} \quad \hbox{for} \quad i=2,\,3 \,. \label{NullFrame}\end{aligned}$$ Thus, ${\mbox{\boldmath$k$}}$ and ${\mbox{\boldmath$l$}}$ are future oriented null vectors, and ${\mbox{\boldmath$m$}}_{i}$ are two spatial vectors orthogonal to them, normalized as ${{\mbox{\boldmath$k$}}\cdot{\mbox{\boldmath$l$}}=-1}$ and ${{\mbox{\boldmath$m$}}_{i}\cdot{\mbox{\boldmath$m$}}_{j}=\delta_{ij}}$. Such a generic decomposition was found in [@BicakPodolsky:1999; @PodolskySvarc:2012].
Using these results, we obtain the corresponding *general form of the equation of geodesic deviation* (\[InvGeoDevExpl\]) *in Quadratic Gravity/the Einstein–Weyl theory*: $$\begin{aligned}
\ddot{Z}^{(1)} = & \quad \Psi_{2S}\,Z^{(1)}+ \tfrac{1}{\sqrt{2}}(\Psi_{1T^j}-\Psi_{3T^j})\,Z^{(j)} \nonumber\\
& \qquad +2k\,\big[(B_{(1)(1)}-B_{(0)(0)})\,Z^{(1)}+B_{(1)(j)} \,Z^{(j)}\,\big]\,,\label{InvGeoDevFinal1}\\
\ddot{Z}^{(i)} = & - \tfrac{1}{2}\Psi_{2S}\,Z^{(i)} + \tfrac{1}{\sqrt{2}}(\Psi_{1T^i}-\Psi_{3T^i})\,Z^{(1)} -\tfrac{1}{2}(\Psi_{0^{ij}}+\Psi_{4^{ij}})\,Z^{(j)} \nonumber\\
& \qquad +2k\,\big[\,B_{(i)(1)} \,Z^{(1)}+B_{(i)(j)} \,Z^{(j)}-B_{(0)(0)}\,Z^{(i)}\,\big]\,, \label{InvGeoDevFinal2}\end{aligned}$$ where we have used the relation ${\Psi_{2T^{(ij)}}=\tfrac{1}{2}\Psi_{2S}\,\delta_{ij}}$ valid in ${D=4}$, see [@PodolskySvarc:2012]. This system of equations admits a *clear physical interpretation*: The Newtonian component $\Psi_{2S}$ of the gravitational field causes classical tidal deformations, $\Psi_{3T^i}, \Psi_{1T^i}$ are responsible for longitudinal motions, while $\Psi_{4^{ij}}, \Psi_{0^{ij}}$ represent the transverse effects of gravitational waves (propagating in the directions ${\mbox{\boldmath$e$}}_{(1)}, -{\mbox{\boldmath$e$}}_{(1)}$, respectively). The additional specific effects caused by the nonvanishing Bach tensor are encoded in the frame components $B_{(a)(b)}$.
### Geodesic deviation in the Schwarzschild–Bach black hole spacetimes
Let us concentrate on the spherically symmetric black hole metric in the form , or with . In particular, we introduce the “interpretation” orthonormal frame associated with a *radially falling observer*, i.e., assuming ${\dot{x}=0=\dot{y}}$. Such a frame reads $$\begin{aligned}
& {\mbox{\boldmath$e$}}_{(0)}\equiv {\mbox{\boldmath$u$}}= \dot{r}\,\partial_r +\dot{u}\,\partial_u \,, \nonumber \\
& {\mbox{\boldmath$e$}}_{(1)}= \tfrac{1}{2}\big[( {\Omega^2\dot{u}} )^{-1}-{\cal H}\dot{u}\big]\partial_r -\dot{u}\,\partial_u \,, \nonumber \\
& {\mbox{\boldmath$e$}}_{(i)}= \Omega^{-1}\big[1+\tfrac{1}{4}(x^2+y^2)\big]\partial_i \,, \label{OrtFrame}\end{aligned}$$ where the normalisation of observer’s four-velocity ${{\mbox{\boldmath$u$}}\cdot{\mbox{\boldmath$u$}}=-1}$ implies ${\dot{r}=\tfrac{1}{2}\big[({\Omega^2\dot{u}})^{-1}+{\cal H}\dot{u}\big]}$. Using (\[NullFrame\]), the associated null interpretation frame thus takes the form $${\mbox{\boldmath$k$}}= \frac{1}{\sqrt{2}\,\dot{u}\,\Omega^2}\,\partial_r \,, \qquad
{\mbox{\boldmath$l$}}= \frac{\dot{u}\,{\cal H}}{\sqrt{2}}\,\mathbf{\partial}_r+\sqrt{2}\dot{u}\,\partial_u\,, \qquad
{\mbox{\boldmath$m$}}_i = \Omega^{-1}\big[1+\tfrac{1}{4}(x^2+y^2)\big]\partial_i \,. \label{NullIntFrame}$$ A direct calculation shows that *the only nonvanishing Weyl tensor component* with respect to is $$\Psi_{2S} \equiv C_{abcd}\; k^a\, l^b\, l^c\, k^d
=\tfrac{1}{6}\,\Omega^{-2}({\cal H}''+2)\,. \label{Psi2Int}$$ This is consistent with the fact that the spherically symmetric black hole metric is of algebraic type D. The explicit Bach tensor projections with respect to the orthonormal frame (\[OrtFrame\]) are $$\begin{aligned}
B_{(0)(0)}&= \frac{1}{24\,\Omega^6\dot{u}^2}\Big[-(1-\Omega^2{\cal H}\dot{u}^2)^2\,{\cal H}''''+2\Omega^2\dot{u}^2({\cal H}'{\cal H}'''-\tfrac{1}{2}{{\cal H}''}^2+2)\Big]\,, \\
B_{(1)(1)}&= \frac{1}{24\,\Omega^6\dot{u}^2}\Big[-(1+\Omega^2{\cal H}\dot{u}^2)^2\,{\cal H}''''-2\Omega^2\dot{u}^2({\cal H}'{\cal H}'''-\tfrac{1}{2}{{\cal H}''}^2+2)\Big]\,, \\
B_{(0)(1)}&= -\frac{1}{24\,\Omega^6\dot{u}^2}\,(1-\Omega^4{\cal H}^2\dot{u}^4)\,{\cal H}'''' \,,
\qquad B_{(0)(i)}= 0\,, \\
B_{(i)(j)}&= \frac{\delta_{ij}}{12\,\Omega^4}({\cal H}{\cal H}''''+{\cal H}'{\cal H}'''-\tfrac{1}{2}{{\cal H}''}^2+2)\,,
\qquad B_{(1)(i)}= 0 \,.\end{aligned}$$ Therefore, the equation of geodesics deviation , explicitly becomes $$\begin{aligned}
\ddot{Z}^{(1)} = & \hspace{6mm} \frac{1}{6}\, \Omega^{-2}\big({\cal H}''+2\big)\,Z^{(1)}\,
-\,\frac{1}{3}\,k\,\Omega^{-4}\big({\cal H}{\cal H}''''+{\cal H}'{\cal H}'''-\tfrac{1}{2}{{\cal H}''}^2+2\big)Z^{(1)} \,, \label{InvGeoDevBH1}\\
\ddot{Z}^{(i)} = & - \frac{1}{12}\,\Omega^{-2}\big({\cal H}''+2\big)\,Z^{(i)}
+\frac{1}{12}\,k\,\Omega^{-4}\big((\Omega^2\H\dot{u}^2)^{-1}+\Omega^2{\cal H}\dot{u}^2\big){\cal H}{\cal H}''''\,Z^{(i)} \,. \label{InvGeoDevBHi}\end{aligned}$$ We conclude that there is a classical *tidal deformation* caused by the *Weyl curvature* proportional to ${\Omega^{-2}({\cal H}''+2)}$, i.e., the square root of the invariant . Moreover, in Quadratic Gravity (with ${k\not=0}$) there are *two additional effects* caused by the presence of a nonvanishing *Bach tensor*. The first can be observed in the longitudinal component of the acceleration , while the second can be observed in the transverse components . Interestingly, up to a constant they are exactly the square roots of the two parts of the invariant , that is the amplitudes $\B_1$, $\B_2$ given by , .
The influence of these two distinct components $\B_1$ and $\B_2$ of the Bach tensor $B_{ab}$ on test particles is even more explicitly seen in the geodesic deviation of *initially static test particles with* ${\dot{r}=0}$. The 4-velocity normalization then implies ${\Omega^2\H\,\dot{u}^2=-1}$, which simplifies , to $$\begin{aligned}
\ddot{Z}^{(1)} = & \hspace{6mm} \frac{1}{6}\, \Omega^{-2}\big({\cal H}''+2\big)\,Z^{(1)}
-\frac{1}{3}\,k\,\Omega^{-4}\big(\B_1+\B_2\big)Z^{(1)} \,, \label{InvGeoDevBH1r0}\\
\ddot{Z}^{(i)} = & - \frac{1}{12}\,\Omega^{-2}\big({\cal H}''+2\big)\,Z^{(i)}
-\frac{1}{6}\,k\,\Omega^{-4}\,\B_1\,Z^{(i)} \,. \label{InvGeoDevBHir0}\end{aligned}$$ From these expressions, it immediately follows that the first component $\B_1$ of the Bach tensor is directly observed in the *transverse* components of the acceleration along ${{\mbox{\boldmath$e$}}_{(2)}, {\mbox{\boldmath$e$}}_{(3)}}$, that is ${\partial_x, \partial_y}$ (equivalent to ${\partial_\theta, \partial_\phi}$), while the second component $\B_2$ only occurs in the *radial* component along ${\mbox{\boldmath$e$}}_{(1)}= -\dot{u}\,(\partial_u+\H\,\partial_r)= -\H\,\Omega'\,\dot{u}\,\partial_{\bar r}$, proportional to $\partial_{\bar r}$.
Interestingly, *on the horizon there is only the radial effect* given by ${\B_2(r_h)}$ since ${\B_1(r_h)=0}$ due to and , see also .
It can also be proven by direct calculation that the specific character of $\B_1, \B_2$ *cannot mimic* the Newtonian tidal effect in the Schwarzschild solution, i.e., cannot be “incorporated” into the first terms ${\Omega^{-2}\big({\cal H}''+2\big)}$ in , . Therefore, by measuring the free fall of a set of test particles, it is *possible to distinguish* the pure Schwarzschild black hole from the Schwarzschild–Bach black hole geometry which has nonvanishing Bach tensor ${B_{ab}\ne 0}$.
Thermodynamic properties: horizon area, temperature, entropy
------------------------------------------------------------
It is also important to determine main geometrical and thermodynamic properties of the family of Schwarzschild–Bach black holes. The *horizon* in these spherically symmetric spacetimes is generated by the rescaled null Killing vector ${\xi\equiv\sigma\partial_u=\sigma\partial_t}$, considering the time-scaling freedom represented by a parameter $\sigma$. Thus it appears at *zero of the metric function* ${\H(r)}$, where the norm of $\xi$ vanishes, see . In the explicit form , this is clearly located at ${r=r_h}$ since ${\H(r_h)=0}$. By simply integrating the angular coordinates of the metric , we immediately obtain the *horizon area* as = 4\^2(r\_h)= = 4[|r]{}\_h\^2 . \[horizon\_area\]
The only nonzero derivatives of $\xi$ are ${\xi_{u;r}=-\xi_{r;u}=\frac{1}{2}\sigma(\Omega^2\H)'}$, and thus ${\xi^{\,r;u}=-\xi^{\,u;r}=\Omega^{-4}\xi_{u;r}}$. From the definition [@Wald:1984] of *surface gravity* ${\kappa^2\equiv-\frac{1}{2}\,\xi_{\mu;\nu}\,\xi^{\,\mu;\nu}}$, we obtain ${\kappa=-\frac{1}{2}\sigma(\H'+2\H\,\Omega'/\Omega)}$. On the horizon, where ${\H=0}$, using this simplifies to /= -’(r\_h) = - = . \[surface\_gravity\] It is *the same expression as for the Schwarzschild solution* (in which case ${\kappa=1/4m}$). The standard expression for *temperature* of the black hole horizon ${T\equiv\kappa/(2\pi)}$, which is valid even in higher-derivate gravity theories [@FanLu:2015], thus yields T/= - = , \[temperature\] *independent of the Bach parameter $b$*.
However, in higher-derivative theories it is *not* possible to use the usual formula ${S=\frac{1}{4}{\cal A}}$ to determine the *black hole horizon entropy*. Instead, it is necessary to apply the generalized formula derived by Wald [@Wald:1993; @IyerWald:1994], namely S=, \[WaldS\] where the Noether charge 2-form $\mathbf{Q}$ on the horizon is [& = &]{}\_Q\^[[[d]{}]{}]{}x\^[[[d]{}]{}]{}x\^ ,\
Q\^ [& = &]{}2X\^\_[;]{} +4[X\^]{}\_[;]{}\_X\^ , \[NoetherCharge\] in which ${\cal L}$ is the Lagrangian of the theory. In the case of Quadratic Gravity , it can be shown that X\^ [& = &]{}. \[Xabcd\] Subsequent lengthy calculation for the metric with ${\Lambda=0}$ then leads to (r\_h) = -\^2’ |\_[r=r\_h]{} [[[d]{}]{}]{}[[[d]{}]{}]{}. \[NoetherCharge2\] Evaluating the integral , and using , , , we finally obtain S = [A]{}(1-4kr\_h\^2b) = [A]{}(1-4k ) . \[entropySB\] This explicit formula for the Schwarzschild–Bach black hole entropy agrees with the numerical results presented in [@LuPerkinsPopeStelle:2015], with the identification ${k=\alpha}$ and ${b=\delta^*}$. In fact, it gives a geometrical interpretation of the “non-Schwarzschild parameter” $\delta^*$ as the dimensionless Bach parameter $b$ that determines the *value of the Bach tensor on the horizon* $r_h$, see relations . Of course, for the Schwarzschild black hole (${b=0}$) or in Einstein’s General Relativity (${k=0}$) we recover the standard expression ${S=\frac{1}{4G}\,{\cal A}}$. Notice also from that for a given ${b \ne 0}$, the *deviation* from this standard Schwarzschild entropy *is larger when the Schwarzschild–Bach black holes are smaller* because they have smaller $\bar r_h$.
Conclusions
===========
The class of spherically symmetric black holes in Quadratic Gravity and the Einstein–Weyl theory was studied in many previous works, in particular [@Stelle:1978; @Holdom:2002; @LuPerkinsPopeStelle:2015; @LuPerkinsPopeStelle:2015b; @PerkinsPhD], often by numerical methods applied to complicated field equations corresponding to the standard form of the spherical metric (\[Einstein-WeylBH\]). In [@PodolskySvarcPravdaPravdova:2018; @SvarcPodolskyPravdaPravdova:2018], using a convenient form of the line element (\[BHmetric\]) conformal to a simple Kundt seed, we obtained a surprisingly simple form of the field equations (\[Eq1\]), (\[Eq2\]). This enabled us to find an *explicit form* of their exact solutions. Moreover, we identified the *Bach tensor* as the key ingredient which makes the Schwarzschild solution geometrically distinct from the other branch of “non-Schwarzschild” ones. This is a direct consequence of the extension of Einstein’s theory to include higher derivative corrections.
The present paper contains a thorough analysis of all such solutions and their derivation, including the details which had to be omitted in our brief letter [@PodolskySvarcPravdaPravdova:2018].
We have started with the conformal-to-Kundt metric ansatz (\[BHmetric\]). Together with the Bianchi identities, this leads to a compact form of the Quadratic Gravity field equations (\[fieldeqsEWmod\]), assuming ${R=0}$, namely the autonomous system of *two ordinary differential equations* (\[Eq1\]) and (\[Eq2\]) for two metric functions ${\Omega(r)}$ and ${{\cal H}}(r)$. They have been solved in terms of *power series* representing these metric functions, expanded around any *fixed point* $r_0$ (\[rozvojomeg0\]), (\[rozvojcalH0\]), or using the *asymptotic expansion* (\[rozvojomegINF\]), (\[rozvojcalHINF\]), respectively. The field equations have become the algebraic constraints (\[KeyEq1\]), (\[KeyEq2\]) in the fixed point case (near ${r_0}$), and (\[KeyEq1INF\]), (\[KeyEq2INF\]) in the asymptotic region (as ${r \to \infty}$). Their dominant orders restrict the admitted solutions to (\[4classes\]) and (\[2classes\]), respectively. The detailed discussion of all the possible six main classes, together with a suitable fixing of the gauge freedom, can be found in subsequent Sections \[description\] and \[description\_INF\]. The classes are summarized in Tables \[tbl:01\] and \[tbl:02\] in Section \[summary\].
The most prominent case corresponds to the spherically symmetric *black hole spacetimes with* (in general) *nonvanishing Bach tensor*. This solution has been expanded around the event horizon, see Subsection \[SchwaBach\_\[n,p\]=\[0,1\]\]. The metric functions ${\Omega(r)}$ and ${{\cal H}}(r)$ are given by the series (\[Omega\_\[0,1\]\]), (\[H\_\[0,1\]\]) with the initial coefficients specified by (\[alphasgammaIIbinitial\]), and all other coefficients determined by the recurrent relations (\[alphasIIbgeneral\]). Thus we have obtained the two-parametric family of black holes characterized by the radial position ${r_h}$ of the horizon and by the additional parameter $b$. The new Bach parameter distinguishes this more general *Schwarzschild–Bach* solutions (${b\neq0}$) from the classical Schwarzschild spacetime with vanishing Bach tensor (${b=0}$). The main mathematical properties of the Schwarzschild–Bach metric functions are presented and visualized in Section \[discussion-and-figures\]. Subsequent Section \[physics\] contains the physical and geometrical analysis. We have discussed specific behavior of freely falling test observers, described by the equation of geodesic deviation, and demonstrated that their *relative motion* encodes the presence of the Bach tensor. The physical investigation is completed by a fully explicit evaluation of the *thermodynamic quantities*. In particular, the expression for entropy (\[entropySB\]) exhibits the key role of the Bach parameter ${b}$.
Finally, for convenience, in Section \[summary\] we have also *summarized all the admitted classes* of solutions, including their physical interpretation, the number of free parameters and, most importantly, relations to previous works. See, in particular, Table \[tab:3\].
We hope that our approach to spherically symmetric vacuum solutions to Quadratic Gravity and the Einstein–Weyl theory may elucidate some of their properties that are not easily accessible by numerical simulations. Of course, we are aware of many remaining open questions. For example, complete analytic identification of the same physical solution in distinct classes and their mutual relations are still missing. It is also of physical interest to understand the effect of nontrivial Bach tensor in the Schwarzschild–Bach spacetimes on perihelion shift and light bending, studied thoroughly during the last century in Einstein’s theory using the Schwarzschild solution.
Acknowledgements {#acknowledgements .unnumbered}
================
This work has been supported by the Czech Science Foundation Grants No. GAČR 17-01625S (JP, R[Š]{}) and 19-09659S (VP, AP), and the Research Plan RVO: 67985840 (VP, AP).
The Ricci and Bach tensors for the Kundt seed {#derivingRBseed}
=============================================
We start with the seed Kundt metric (\[Kundt seed xy\]). Its nontrivial metric components $g_{ab}$ are $$\label{Einstein-WeylBHC}
g_{xx}^{{\hbox{\tiny Kundt}}}= g_{yy}^{{\hbox{\tiny Kundt}}}= \textstyle{\left(1+\frac{1}{4}(x^2+y^2)\right)^{-2}}\,,\qquad
g_{ru}^{{\hbox{\tiny Kundt}}}= -1 \,,\qquad
g_{uu}^{{\hbox{\tiny Kundt}}}= {\cal H}\,,$$ so that the contravariant components $g^{ab}$ read $$\label{contraEinstein-WeylBHC}
g^{xx}_{{\hbox{\tiny Kundt}}}= g^{yy}_{{\hbox{\tiny Kundt}}}= \textstyle{\left(1+\frac{1}{4}(x^2+y^2)\right)^2}\,,\qquad\quad
g^{ru}_{{\hbox{\tiny Kundt}}}= -1 \,,\qquad
g^{rr}_{{\hbox{\tiny Kundt}}}= -{\cal H} \,.$$ Recall that the spatial 2-metric ${g_{ij}}$ is a round sphere of unit radius, with the Gaussian curvature ${K=1}$ and thus its Ricci scalar is ${{\cal R}=2K=2}$. The nontrivial Christoffel symbols for this metric are \^r\_[ru]{} = [-’]{} , \^r\_[uu]{} = [’]{} , \^u\_[uu]{} = [’]{} , \^k\_[ij]{} = [\^[S]{}\^k\_[ij]{}]{} , \[ChristoffelEnd\] where ${\,^{S}\Gamma^k_{ij}\equiv\frac{1}{2}g^{kl}(2g_{l(i,j)}-g_{ij,l})}$ are the symbols with respect to the spatial metric $g_{ij}$ of the 2-sphere. The only nontrivial Riemann curvature tensor components are R\_[ruru]{}\^= [-”]{} , R\_[kilj]{}\^= g\_[kl]{}g\_[ij]{}-g\_[kj]{}g\_[il]{} , and the only nontrivial Ricci tensor components of (\[Einstein-WeylBHC\]) are $$\begin{aligned}
R_{ru}^{{\hbox{\tiny Kundt}}}{\!\!\!& = &\!\!\!}-\pul\,{\cal H}'' \,, \label{Ricci ru5} \\
R_{uu}^{{\hbox{\tiny Kundt}}}{\!\!\!& = &\!\!\!}-{\cal H}\,R_{ru}^{{\hbox{\tiny Kundt}}}\,,\label{Ricci uu5}\\
R_{xx}^{{\hbox{\tiny Kundt}}}= R_{yy}^{{\hbox{\tiny Kundt}}}{\!\!\!& = &\!\!\!}g_{xx} \,, \label{Ricci ij5}\end{aligned}$$ while the Ricci scalar reads R\^=[H]{}” + 2 , so that the only nontrivial Weyl tensor components are $$\begin{aligned}
C_{ruru}^{{\hbox{\tiny Kundt}}}{\!\!\!& = &\!\!\!}{\textstyle -\frac{1}{6} R} \,, \label{WeyliK}\\
C_{riuj}^{{\hbox{\tiny Kundt}}}{\!\!\!& = &\!\!\!}{\textstyle \frac{1}{12} R \,g_{ij}} \,, \\
C_{kilj}^{{\hbox{\tiny Kundt}}}{\!\!\!& = &\!\!\!}{\textstyle \frac{1}{6}R\,(g_{kl}g_{ij}-g_{kj}g_{il}) } \,, \\
C_{uiuj}^{{\hbox{\tiny Kundt}}}{\!\!\!& = &\!\!\!}-{\cal H}\, C_{riuj} \,. \label{WeylfK}\end{aligned}$$ The nonzero components of the Bach tensor are $$\begin{aligned}
B_{rr}^{{\hbox{\tiny Kundt}}}{\!\!\!& = &\!\!\!}{\textstyle -\frac{1}{6}\,{\cal H}'''' } \,, \label{Bach rr}\\
B_{ru}^{{\hbox{\tiny Kundt}}}{\!\!\!& = &\!\!\!}{\textstyle \frac{1}{12}\,
\big(2\,{\cal H}{\cal H}''''+{\cal H}'{\cal H}'''-{\textstyle\frac{1}{2}}{{\cal H}''}^2 +2\big) } \,, \label{Bach ru}\\
B_{uu}^{{\hbox{\tiny Kundt}}}{\!\!\!& = &\!\!\!}-{\cal H}\,B_{ru}^{{\hbox{\tiny Kundt}}}\,, \label{Bach uu}\\
B_{xx}^{{\hbox{\tiny Kundt}}}= B_{yy}^{{\hbox{\tiny Kundt}}}{\!\!\!& = &\!\!\!}{\textstyle \frac{1}{12}}\,g_{xx}\,
\big({\cal H}{\cal H}''''+{\cal H}'{\cal H}'''-{\textstyle\frac{1}{2}}{{\cal H}''}^2 +2\big) \,,\label{Bach xx}\end{aligned}$$ involving up to the 4th derivative of the metric function $\H(r)$.
The Ricci and Bach tensors for the conformal metric {#App_derivingFE}
===================================================
Taking the class of Kundt geometries (\[Kundt seed xy\]) as a seed, we can generate the metric of spherically symmetric geometries by the conformal transformation (\[confrelation\]), that is [[[d]{}]{}]{}s\^2 = \^2(r). \[BHmetric-xy\]
Now, it is well-known [@Wald:1984] that under a conformal transformation of the seed metric g\_[ab]{}=\^2 g\_[ab]{}\^, \[confrel\] the Ricci scalar and the Ricci and Bach tensors transform as $$\begin{aligned}
R {\!\!\!& = &\!\!\!}\Omega^{-2}R^{{\hbox{\tiny Kundt}}}-6\Omega^{-3} \square\Omega\,, \label{OmRiccscalar}\\
R_{ab} {\!\!\!& = &\!\!\!}R_{ab}^{{\hbox{\tiny Kundt}}}- 2\Omega^{-1}\nabla_a\nabla_b\Omega - \Omega^{-1} g_{ab}^{{\hbox{\tiny Kundt}}}\square\Omega+ \Omega^{-2}(4\Omega_{,a}\Omega_{,b}-g_{ab}^{{\hbox{\tiny Kundt}}}g^{cd}_{{\hbox{\tiny Kundt}}}\Omega_{,c}\Omega_{,d})\,, \label{OmRicc}\\
B_{ab} {\!\!\!& = &\!\!\!}\Omega^{-2}B_{ab}^{{\hbox{\tiny Kundt}}}\,. \label{OmBach}\end{aligned}$$
For the Kundt seed metric $g_{ab}^{{\hbox{\tiny Kundt}}}$ (\[Einstein-WeylBHC\]), its Ricci and Bach tensors $R_{ab}^{{\hbox{\tiny Kundt}}}$ and $B_{ab}^{{\hbox{\tiny Kundt}}}$ are given by (\[Ricci ru5\])–(\[Ricci ij5\]) and (\[Bach rr\])–(\[Bach xx\]), respectively. The nontrivial derivatives (with respect to the Kundt seed) of the conformal factor $\Omega(r)$ are, in view of (\[ChristoffelEnd\]), $$\begin{aligned}
&& \Omega_{,r} \equiv \Omega'\,, \nonumber\\
&& \nabla_r\nabla_r \Omega= \Omega''\,,\quad
\nabla_r\nabla_u \Omega= {\textstyle\frac{1}{2}}{\cal H}'\Omega' = \nabla_u\nabla_r \Omega\,,\quad
\nabla_u\nabla_u \Omega= {-\textstyle\frac{1}{2}}{\cal H}{\cal H}'\Omega' \,,\\
&& \square\Omega = -({\cal H}\Omega''+{\cal H}'\Omega')\,. \nonumber\end{aligned}$$ Employing (\[OmRicc\]), the nonvanishing Ricci tensor components of the metric (\[BHmetric-xy\]) are thus $$\begin{aligned}
R_{rr} {\!\!\!& = &\!\!\!}-2\Omega^{-2}\big(\Omega\Omega''-2{\Omega'}^2\big) \,, \label{RT_R rr}\\
R_{ru} {\!\!\!& = &\!\!\!}-\pul \Omega^{-2}\big(\Omega^2 {\cal H}\big)'' \,, \label{RT_R ru}\\
R_{uu} {\!\!\!& = &\!\!\!}-{\cal H}\, {R}_{ru} \,, \label{RT_R uu}\\
R_{xx} = {R}_{yy} {\!\!\!& = &\!\!\!}\Omega^{-2}g_{xx}\,
\big[ \big({\cal H}\Omega\Omega'\big)'+\Omega^2 \big] \,,\label{RT_R xx}\end{aligned}$$ and using (\[OmRiccscalar\]) we obtain R = 6\^[-3]{} . \[barR\] The nonvanishing Bach tensor components $B_{ab}$ are obtained by a trivial rescaling (\[OmBach\]) of (\[Bach rr\])–(\[Bach xx\]).
Derivation and simplification of the field equations {#analysingFE}
====================================================
The vacuum field equations in the Einstein–Weyl theory and also general Quadratic Gravity for the metric $g_{ab}$ are (\[fieldeqsEWmod\]), that is $$R_{ab} = 4k\, B_{ab} \,.
\label{EWfield equations}$$ Using the expressions (\[RT\_R rr\])–(\[RT\_R xx\]) and (\[OmBach\]) with (\[Bach rr\])–(\[Bach xx\]), these *field equations explicitly read* $$\begin{aligned}
\Omega\Omega''-2{\Omega'}^2 & = \tfrac{1}{3}k\, {\cal H}'''' \,, \label{Neq_rr} \\
\big(\Omega^2 {\cal H}\big)'' & = -\tfrac{2}{3}k \big(2\,{\cal H}{\cal H}''''+{\cal H}'{\cal H}'''-{\textstyle\frac{1}{2}}{{\cal H}''}^2 +2\big) \,, \label{Neq_ru} \\
\big({\cal H}\Omega\Omega'\big)'+\Omega^2 & = \tfrac{1}{3}k \,\big({\cal H}{\cal H}''''+{\cal H}'{\cal H}'''-{\textstyle\frac{1}{2}}{{\cal H}''}^2 +2 \big) \,. \label{Neq_xx}\end{aligned}$$ The equations (\[Neq\_rr\]), (\[Neq\_ru\]), (\[Neq\_xx\]) represent the nontrivial components $rr$, $ru$, $xx$ (identical to $yy$), respectively. The $uu$ component of the field equations is just the ${(-{\cal H})}$-multiple of (\[Neq\_ru\]).
Moreover, recall that the trace of the field equations (\[EWfield equations\]) is ${ {R}=0}$, cf. (\[R=0\]). Using (\[barR\]) we obtain the explicit condition $$\T\equiv{\cal H}\Omega''+{\cal H}'\Omega'+{\textstyle \frac{1}{6}} ({\cal H}''+2)\Omega = 0 \,.
\label{traceC}$$ It can be checked that this is a direct *consequence* of equations (\[Neq\_rr\])–(\[Neq\_xx\]). Notice that it is a *linear differential equation for the function* ${\cal H}(r)$, and also linear differential equation for $\Omega(r)$.
We have thus obtained *three nontrivial field equations* (\[Neq\_rr\])–(\[Neq\_xx\]) for *two unknown functions* $\Omega(r)$ and ${\cal H}(r)$, and also their consequence (\[traceC\]). Therefore, this coupled system seems to be overdetermined. However, now we prove that the key metric functions $\Omega(r)$ and ${\cal H}(r)$ are, in fact, *solutions of just two coupled equations*.
To this end, let us introduce the *auxiliary symmetric tensor* $J_{ab}$ defined as J\_[ab]{}R\_[ab]{}-Rg\_[ab]{} - 4kB\_[ab]{} . \[defJab\] Using $J_{ab}$, the vacuum field equations (\[GenQGFieldEq\]) of Quadratic Gravity (assuming a constant $R$ and ${\Lambda=0}$) or Einstein–Weyl gravity (with ${\beta=0=\Lambda}$) are simply J\_[ab]{}=0 . \[Jab=0\]
Now, by employing the contracted *Bianchi identities* ${\nabla^b R_{ab}=\frac{1}{2} R_{,a}}$ and the conservation property of the Bach tensor ${\nabla^b B_{ab}=0}$, see (\[Bachproperties\]), we obtain \^b J\_[ab]{}0 . \[BianchiIdEW\] Interestingly, this is actually a *geometrical identity* which is valid without employing any field equations, namely (\[Jab=0\]), or (\[EWfield equations\]) in particular.
An explicit evaluation of the identity (\[BianchiIdEW\]) for the metric $ {g}_{ab}$ of the form (\[BHmetric-xy\]) leads to the following equations, which are *always satisfied*: $$\begin{aligned}
& \nabla^b J_{rb} = - \Omega^{-3}\Omega'\big(J_{ij}\,{g}^{ij}
+{\cal H} J_{rr}\big)-\Omega^{-2}\big({\cal H} J_{rr,r}+ J_{ru,r}+\tfrac{3}{2}{\cal H}' J_{rr}\big)
\hspace{-25mm}&\equiv 0 \,, \label{BI_r} \\
& \nabla^b J_{ub} = -2\Omega^{-3}\Omega'\big(J_{uu}+{\cal H} J_{ru}\big)-\Omega^{-2}\big(J_{uu}+
{\cal H} J_{ru}\big)_{,r} &\equiv 0 \,, \label{BI_u} \\
& \nabla^b J_{ib} = \Omega^{-2} J_{ik||l}\,{g}^{kl} &\equiv 0 \,. \label{BI_i}\end{aligned}$$ Here the spatial covariant derivative $_{||}$ is calculated with respect to the spatial part $g_{ij}$ of the Kundt seed metric (\[Einstein-WeylBHC\]). Moreover, a direct calculation of $J_{ab}$ defined by (\[defJab\]) gives $$J_{uu}=-{\cal H}\, J_{ru}\,, \qquad J_{xx}=\mathcal{J}(r)\, g_{xx}
= J_{yy} \,, \label{EqDep}$$ where the function $\mathcal{J}(r)$ is defined as $$\mathcal{J} \equiv \Omega^{-2}\,\big[\big({\cal H}\Omega\Omega'\big)'+\Omega^2
-3\T\Omega
-\tfrac{1}{3}k \,\big({\cal H}{\cal H}''''+{\cal H}'{\cal H}'''-{\textstyle\frac{1}{2}}{{\cal H}''}^2 +2 \big)\big] \,, \label{calJ}$$ and $$\begin{aligned}
J_{rr} = 2\Omega^{-2}&\,\big[-\Omega\Omega''+2{\Omega'}^2+\tfrac{1}{3}k\,{\cal H}''''\big] \,, \label{bJrr}\\
J_{ru} = \Omega^{-2}&\,\big[-\tfrac{1}{2}\big(\Omega^2 {\cal H}\big)''
+3\T\Omega
-\tfrac{1}{3}k \big(2\,{\cal H}{\cal H}''''+{\cal H}'{\cal H}'''-{\textstyle\frac{1}{2}}{{\cal H}''}^2 +2\big)\big] \label{bJru}\,.\end{aligned}$$
By substituting the relations (\[EqDep\]) into (\[BI\_u\]) and (\[BI\_i\]), it can be seen immediately that these two conditions are automatically satisfied. Interestingly, the remaining Bianchi identity (\[BI\_r\]) gives a *nontrivial* result. If the metric functions $\Omega(r)$ and ${\cal H}(r)$ satisfy the two field equations ${J_{rr}=0}$ and ${J_{ru}=0}$ then necessarily ${J_{ij}\,g^{ij}\equiv0}$, that is ${J_{xx}\,g^{xx}+J_{yy}\,g^{yy}=2\mathcal{J}(r)=0}$ and thus ${J_{xx}=0=J_{yy}}$.
Therefore, we conclude that *all field equations* for the metric (\[BHmetric-xy\]) *reduce just to two key equations*, namely ${J_{rr}=0}$ and ${J_{ru}=0}$. Since ${g^{ab}J_{ab}=0}$, it also implies ${R=0}$ and thus ${\T=0}$, cf. . This coupled system of two equations completely determines all possible exact vacuum solutions of the type (\[BHmetric-xy\]) in Einstein–Weyl gravity, and since ${ R=0}$, also in a general Quadratic Gravity. The key point is that, due to the Bianchi identities, the two key equations *imply* the nontrivial field equations ${J_{xx}=0= J_{yy}}$ since necessarily ${\mathcal{J}=0}$, that is using (\[calJ\]) $$\big({\cal H}\Omega\Omega'\big)'+\Omega^2-3\T\Omega
= \tfrac{1}{3}k \big({\cal H}{\cal H}''''+{\cal H}'{\cal H}'''-{\textstyle\frac{1}{2}}{{\cal H}''}^2 +2 \big) \,. \label{J=0}$$ The equation ${J_{rr}=0}$ is exactly equation , and equation is simply ${J_{ru}=0}$ with ${\T=0}$. Finally, substituting ${\T=0}$ into , we immediately obtain . This completes the proof of the equivalence.
To integrate the field equations, it is necessary to solve the equation . Simultaneously, we must solve the equation (\^2 [H]{})” -6= -k (2[H]{}[H]{}””+[H]{}’[H]{}”’-[[H]{}”]{}\^2 +2) \[bJru0\]. Remarkably, this equation *can further be simplified* by expressing the term ${{\cal H}''''}$ from . We thus finally obtain *two very simple field equations* $$\begin{aligned}
\Omega\Omega''-2{\Omega'}^2 = &\ \tfrac{1}{3}k\,{\cal H}'''' \,, \label{Eq1C}\\
\Omega\Omega'{\cal H}'+3\Omega'^2{\cal H}+\Omega^2
= &\ \tfrac{1}{3}k \big({\cal H}'{\cal H}'''-{\textstyle\frac{1}{2}}{{\cal H}''}^2 +2 \big)\,, \label{Eq2C}\end{aligned}$$ for the two metric functions $\Omega(r)$ and ${\cal H}(r)$. Alternatively, instead of solving the single equation (\[Eq2C\]), it is also possible to solve *any two of the three equations* (\[Neq\_ru\]), (\[Neq\_xx\]), (\[traceC\]).
[10]{}
Stelle K S 1977 Renormalization of higher derivative quantum gravity [*Phys. Rev. D*]{} [**16**]{} 953
Salvio A 2018 Quadratic gravity [*Front. Phys.*]{} [**6**]{} 77
Smilga A V 2014 Supersymmetric field theory with benign ghosts [*J. Phys. A*]{} [**47**]{} 052001
Stelle K S 1978 Classical gravity with higher derivatives [*Gen. Relativ. Gravit.*]{} [**9**]{} 353
Holdom B 2002 On the fate of singularities and horizons in higher derivative gravity [*Phys. Rev. D*]{} [**66**]{} 084010
Lü H, Perkins A, Pope C N, and Stelle K S 2015 Black holes in higher derivative gravity [*Phys. Rev. Lett.*]{} [**114**]{} 171601
Lü H, Perkins A, Pope C N, and Stelle K S 2015 Spherically symmetric solutions in higher derivative gravity [*Phys. Rev. D*]{} [**92**]{} 124019
Perkins A 2016 [*Static spherically symmetric solutions in higher derivative gravity*]{} (Ph.D. thesis, Imperial College London)
Pravda V, Pravdová A, Podolský J, and Švarc R 2017 Exact solutions to quadratic gravity [*Phys. Rev. D*]{} [**95**]{} 084025
Podolský J, Švarc R, Pravda V, and Pravdová A 2018 Explicit black hole solutions in higher-derivative gravity [*Phys. Rev. D*]{} [**98**]{} 021502(R)
Švarc R, Podolský J, Pravda V, and Pravdová A 2018 Exact black holes in quadratic gravity with any cosmological constant [*Phys. Rev. Lett*]{} [**121**]{} 231104
Weyl H 1919 Eine neue Erweiterung der Relativitätstheorie [*Ann. der Physik*]{} [**59**]{} 101
Bach R 1921 Zur Weylschen Relativitätstheorie und der Weylschen Erweiterung des Krümmungstensorbegriffs [*Math. Zeitschrift*]{} [**9**]{} 110
Schwarzschild K 1916 Über das Gravitationsfeld eines Massenpunktes nach der Einsteinschen Theorie [*Sitz. Preuss. Akad. Wiss. Berlin*]{} [**7**]{} 189
Stephani H, Kramer D, MacCallum M, Hoenselaers C, and Herlt E 2003 [*Exact Solutions of Einstein’s Field Equations*]{} (Cambridge: Cambridge University Press)
Griffiths J and Podolský J 2009 [*Exact Space-Times in Einstein’s General Relativity*]{} (Cambridge: Cambridge University Press)
Holdom B and Ren J 2017 Not quite a black hole, [*Phys. Rev. D*]{} [**95**]{} 084034
Bičák J and Podolský J 1999 Gravitational waves in vacuum spacetimes with cosmological constant. II. Deviation of geodesics and interpretation of nontwisting type N solutions [*J. Math. Phys. (N.Y.)*]{} [**40**]{} 4506
Podolský J and Švarc R 2012 Interpreting spacetimes of any dimension using geodesic deviation [*Phys. Rev. D*]{} [**85**]{} 044057
Wald R M 1984 [*General Relativity*]{} (Chicago: University of Chicago Press)
Zhong-Ying Fan and Lü H 2015 Thermodynamical first laws of black holes in quadratically-extended gravities [*Phys. Rev. D*]{} [**91**]{} 064009
Wald R M 1993 Black hole entropy is the Noether charge [*Phys. Rev. D*]{} [**48**]{} R3427(R)
Iyer V and Wald R M 1994 Some properties of Noether charge and a proposal for dynamical black hole entropy [*Phys. Rev. D*]{} [**50**]{} 846
[^1]: In four dimensions, the Gauss–Bonnet term ${R_{abcd}R^{abcd}-4R_{ab}R^{ab}+R^2}$ does not contribute to the field equations.
[^2]: For brevity, in this paper the symbol ${r\to\infty}$ means ${|r|\to\infty}$, unless the sign of $r$ is explicitly specified.
[^3]: There may also exist other solutions such that their expansion contains logarithmic or exponential terms in $r$.
[^4]: Of course, provided $r_0$ is within the convergence radius od , .
[^5]: To make the identification, we have relabeled the arguments of the metric functions $A(r), B(r)$ of [@LuPerkinsPopeStelle:2015b] to $\bar r$.
[^6]: We obtained this value from the Mathematica code kindly provided by H. Lü, cf. also [@PerkinsPhD] for a very close value of $b$.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We study the $\Gamma$-convergence of a family of non-local, non-convex functionals in $L^p(I)$ for $p \ge 1$, where $I$ is an open interval. We show that the limit is a multiple of the $W^{1, p}(I)$ semi-norm to the power $p$ when $p>1$ (resp. the $BV(I)$ semi-norm when $p=1$). In dimension one, this extends earlier results which required a monotonicity condition.'
address:
- |
Department of Mathematics, Rutgers University, Hill Center, Busch Campus, 110 Frelinghuysen Road, Piscataway, NJ 08854, USA,\
Departments of Mathematics and Computer Science, Technion, Israel Institute of Technology, 32.000 Haifa, Israel,\
Laboratoire Jacques-Louis Lions, Sorbonne Universités, UPMC Université Paris-6, 4 place Jussieu, 75005 Paris, France,\
[email protected]
- 'Department of Mathematics, EPFL SB CAMA, Station 8 CH-1015 Lausanne, Switzerland, [email protected]'
author:
- HAÏM BREZIS
- 'HOAI-MINH NGUYEN'
title: '$\Gamma$-convergence of non-local, non-convex functionals in one dimension'
---
[**Key words**]{}: non-local, non-convex, pointwise convergence, $\Gamma$-convergence, Sobolev norms.
Introduction and statement of the main results
==============================================
Assume that $\varphi:[0, +\infty) \to [0, + \infty)$ is defined at [*every*]{} point of $[0, + \infty)$, $\varphi$ is continuous on $[0, +\infty)$ except at a finite number of points in $(0, +\infty)$ where it admits a limit from the left and from the right, and $\varphi(0) = 0$. Let $I$ denote an open interval of ${\mathbb{R}}$. Fix $p \ge 1$. Given a measurable function $u$ on $I$, and a parameter $\delta > 0$, we define, as in [@BrNg-Nonlocal1], the following non-local functionals $$\label{def-Lambda}
\Lambda (u, I): = \int_I \int_I \frac{\varphi(|u(x) - u(y)|) }{|x - y|^{p + 1}} \, dx \,
dy \le + \infty$$ and $$\Lambda_{\delta}(u, I): = \delta^p \Lambda (u/\delta, I).$$ Throughout the paper, we make the following three assumptions on $\varphi$: $$\label{cond-varphi-0}
\varphi(t) \le \alpha t^{p+1} \mbox{ on } [0,1] \mbox{ for some positive constant } \alpha,$$ $$\label{cond-varphi-1}
\varphi(t) \le \beta \mbox{ on } [0, + \infty) \mbox{ for some positive constant } \beta,$$ and $$\label{cond-varphi-3}
\int_0^\infty \varphi(t) t^{-(p+1)} \,dt=1/2.$$
Our main result is the following
\[thm-gamma\] Let $p \ge 1$ and let $\varphi$ satisfy -. Then, as $\delta \to 0$, $$\Lambda_{\delta}(\cdot, I) \; \Gamma\mbox{-converges in $L^p(I)$ to } \Lambda_0(\cdot, I),$$ where $$\label{def-Lambda0}
\Lambda_0(u, I) = {\kappa}\int_{I} |u' |^p \, dx \mbox{ in } L^p(I),$$ for some constant ${\kappa}$, depending on $\varphi$ but independent of $I$, such that $$\label{constant-gamma}
0 \le {\kappa}\le 1.$$
Some comments on Theorem \[thm-gamma\] are in order.
$\bullet$ On the precise definition of $\Lambda_0$. If ${\kappa}= 0$, by convention, $\Lambda_0(u, I) = 0$ for all $u \in L^p(I)$. In other words, the conclusion of asserts that [*either*]{} $\Lambda_\delta(\cdot, I)$ $\Gamma$-converges to 0 in $L^p(I)$ [*or*]{} there exists a constant $0< {\kappa}\le 1$ such that $\Lambda_\delta(\cdot, I)$ $\Gamma$-converges to $\Lambda_0(\cdot, I)$ defined by with the usual convention: $\Lambda_0(u, I) = + \infty$ if $u \not \in BV(I)$ for $p=1$, or if $u \not \in W^{1, p}(I)$ for $p>1$. The first part of the alternative, i.e., ${\kappa}= 0$, occurs e.g. when $\varphi$ has a compact support in $[0, + \infty)$ (see [@BrNg-Nonlocal1 Remark 3]; only the case $p=1$ was considered in [@BrNg-Nonlocal1], however, the same conclusion holds for $p>1$ with the same proof). The second part of the alternative, i.e., ${\kappa}>0$, happens e.g. when $\varphi$ is non-decreasing (see [@BrNg-Nonlocal1; @BrNg-Nonlocalp] with roots in [@BourNg]); more generally, ${\kappa}> 0$ when $\liminf_{t \to + \infty} \varphi(t) > 0$. It would be very interesting to find a natural weaker sufficient condition on $\varphi$ at infinity such that ${\kappa}>0$.
$\bullet$ On the condition . This is just a normalization condition. Without this assumption, the conclusion of Theorem \[thm-gamma\] holds with replaced by $$0 \le {\kappa}\le 2 \int_0^{\infty} \varphi(t) t^{-(p+1)} \, dt.$$ This suggests that assumptions - might be substituted by the weaker condition $$\int_0^{\infty} \varphi(t) t^{-(p+1)} \, dt < + \infty.$$
It is worth noting that the following pointwise convergence property holds for $\Lambda_\delta$:
\[pro-pointwise\] Let $p \ge 1$ and let $\varphi$ satisfy -. Then,
1. for $p>1$ and for $u \in W^{1, p}(I)$,
or
1. for $p=1$ and for $u \in C^1(\bar I)$ if $I$ is bounded (resp. $u \in C^1_c(\bar I)$ if $I$ is unbounded),
we have $$\label{pro-pointwise-s}
\lim_{\delta \to 0} \Lambda_{\delta} (u, I) = \int_{I} |u' |^p \, dx.$$
The conclusion of under the assumption $i)$ follows from [@BrNg-Nonlocalp Theorem 1] (the only remaining case to be considered is the case $I = (0, + \infty)$ which can be deduced from the cases $I$ bounded and $I = {\mathbb{R}}$ by standard arguments). The proof of under the assumption $ii)$ appeared in [@BrNg-Nonlocal1 proof of Proposition 1] under the additional assumption $$\label{cond-varphi-mono}
\mbox{$\varphi$ is non-decreasing};$$ however, this assumption can be easily removed from the proof. The conclusion of contrasts with the conclusion of Theorem \[thm-gamma\] since it may happen, for some functions $\varphi$ $\big(\mbox{e.g. } \varphi : = \frac{p}{2} \mathds{1}_{(1, + \infty)}\big)$, that ${\kappa}$ is [*strictly*]{} less than 1 (see [@NgGammaCRAS]); an explicit value of ${\kappa}$ for this $\varphi$ is given in [@AGMP]. As established in [@AGP], it may happen that ${\kappa}(\varphi) = 1$ for some $\varphi$.
This work is a follow-up of our previous papers [@BrNg-Nonlocal1; @BrNg-Nonlocalp] where we investigated a similar problem in any dimension $d \ge 1$. More precisely, $I$ is replaced by a domain $\Omega \subset {\mathbb{R}}^d$ and the RHS in is replaced by $$\int_\Omega \int_\Omega \frac{\varphi(|u(x) - u(y)|) }{|x - y|^{p + d}} \, dx \,
dy.$$ Assuming , , and the [*additional condition*]{} , we established in [@BrNg-Nonlocal1; @BrNg-Nonlocalp] the $\Gamma$-convergence of $\Lambda_\delta$ to a multiple of $\int_{\Omega} |\nabla u|^p \, dx$. In these works, the monotonicity assumption played a crucial role at almost every level of the proofs. The proof of has its roots in [@BourNg; @NgGamma; @BrNg-Nonlocal1; @BrNg-Nonlocalp]. However, many new ideas are required to overcome the lack of assumption . We do not know whether can be removed when $d > 1$.
Proof of the main result {#sect-proof}
========================
We first recall the meaning of $\Gamma$-convergence. One says that $\Lambda_\delta(\cdot, I)$ ${\displaystyle}\mathop{\to}^{\Gamma} \Lambda_0(\cdot, I)$ in $L^p(I)$ for $p \ge 1$ as $\delta \to 0$ if the following two properties hold
1. For each $g \in L^p(I)$ and for [*every*]{} family $(g_\delta) \subset L^p(I)$ such that $g_\delta$ converges to $g$ in $L^p(I)$ as $\delta \to 0$, one has $$\liminf_{\delta \to 0} \Lambda_\delta(g_\delta, I) \ge \Lambda_0(g, I).$$
2. For each $g \in L^p(I)$, there [*exists*]{} a family $(g_\delta) \subset L^p(I)$ such that $g_\delta$ converges to $g$ in $L^p(I)$ as $\delta \to 0$, and $$\limsup_{\delta \to 0} \Lambda_\delta(g_\delta, I) \le \Lambda_0(g, I).$$
In this section, we establish properties (G1) and (G2) with $\Lambda_0$ defined by and ${\kappa}$ defined by $$\label{def-k}
{\kappa}:= \inf \liminf_{\delta \to 0} \Lambda_{\delta} (v_\delta, (0, 1)),$$ where the infimum is taken over all families $(v_\delta) \subset L^p(0, 1)$ such that $v_\delta \to U$ in $L^p(0, 1)$ as $\delta \to 0$, where $$U(x) := x \mbox{ for } x \in (0, 1).$$
Choosing $I = (0, 1)$ and $u = U$ in , we see that the constant ${\kappa}$ given by satisfies $0 \le {\kappa}\le 1$.
\[rem-k\] As a direct consequence of the definition of ${\kappa}$ in , the following property holds $$\liminf_{k \to + \infty} \Lambda_{\delta_k}(g_k, (0, 1)) \ge {\kappa},$$ for every $(\delta_k) \subset {\mathbb{R}}_+$ and $(g_k) \subset L^p(0, 1)$ such that $\delta_k \to 0$ and $g_k \to U$ in $L^p(0, 1)$ as $k \to + \infty$.
We will only consider the case $I = {\mathbb{R}}$. The other cases can be handled as in [@BrNg-Nonlocal1] and are left to the reader. The rest of this paper is organized as follows. is devoted to the proof of Property (G2). The proofs of Property (G1) for $p=1$ and $p>1$ are given in , respectively.
For $p \ge 1$ and $\delta > 0$, we will denote $$\varphi_\delta(t) := \delta^p \varphi(t/\delta) \mbox{ for } t \ge 0.$$
Proof of Property (G2) {#sect-G2}
----------------------
The proof of Property (G2) is based on the following three lemmas which are valid for ${\kappa}$ defined by , possibly equal to $0$. We begin with
\[lem1\] Let $p \ge 1$ and let $\varphi$ satisfy -. There exists a family $(v_\delta) \subset L^p(0, 1)$ converging to $U$ in $L^p(0, 1)$, as $\delta \to 0$, such that $$\label{lem1-statement}
\lim_{\delta \to 0} \Lambda_{\delta} (v_\delta, (0, 1)) = {\kappa}.$$
From the definition of ${\kappa}$ in , there exist a sequence $(\delta_k) \subset {\mathbb{R}}_+$ converging to 0 and a sequence $(u_{k}) \subset L^p(0, 1)$ converging to $U$ in $L^p(0, 1)$ such that $$\label{lem1-start}
\lim_{k \to + \infty} \Lambda_{\delta_k} (u_k, (0, 1)) = {\kappa}.$$ Let $(c_k)$ be a sequence of positive numbers converging to $0$ such that, for large $k$, $$\label{lem1-p1}
c_k \ge \delta_k^{1/2},$$ $$\label{lem1-p2}
\int_{0}^1 |u_k - U|^p \, dx \le c_k^{p+1},$$ $$\label{lem1-p3}
\Lambda_{\delta_{k}}(u_k, (0, 1)) \le {\kappa}+ c_{k},$$ $$\label{lem1-p4}
\Lambda_{\delta_{k}}(u_k, (c_k, 1 - c_k)) \ge {\kappa}(1- 2 c_k) - c_k.$$ Such a sequence $(c_k)$ exists; indeed, from the definition of ${\kappa}$, by a change of variables, we have $$\liminf_{k \to + \infty} \Lambda_{\delta_k}(u_k, (c, 1 - c)) \ge {\kappa}(1- 2 c),$$ for every $c \in (0, 1/2)$. Hereafter, we only consider large $k$ so that - hold.
In what follows in this proof, $C$ denotes positive constants which depend only on $\alpha$, $\beta$, and $p$ and can vary from one place to another. From and and the fact that ${\kappa}\le 1$, we obtain $$\label{est-mean}
\int_{c_k/2}^{c_k} \int_0^1 \frac{\varphi_{\delta_k}(|u_k(x) - u_k(y)|)}{|x- y|^{p+1}} \, dx \, dy \le C c_k.$$ By and , there exists $$\label{lem1-x1k}
x_{1, k} \in (c_k/2, c_k)$$ such that $$\label{lem1-o1}
|u_k(x_{1, k}) - x_{1, k}| \le C c_k$$ and $$\label{lem1-o2}
\int_0^1 \frac{\varphi_{\delta_k}(|u_k(x_{1,k}) - u_k(y)|)}{|x_{1, k}- y|^{p+1}} dy \le C.$$ Similarly, there exists $$\label{lem1-x2k}
x_{2, k} \in (1 - c_k, 1- c_{k}/2)$$ such that $$\label{lem1-o3}
|u_k(x_{2, k}) - x_{2, k}| \le C c_k$$ and $$\label{lem1-o4}
\int_0^1 \frac{\varphi_{\delta_k}(|u_k(x_{2,k}) - u_k(y)|)}{|x_{2, k}- y|^{p+1}} dy \le C.$$
We now modify $u_k$ to obtain a new sequence $(\hat u_k)$ such that $\hat u_k \to U$ in $L^p(0, 1)$, is preserved for $\hat u_k$, i.e., $$\label{lem1-start-1}
\lim_{k \to + \infty} \Lambda_{\delta_k} (\hat u_k, (0, 1)) = {\kappa},$$ and [*in addition*]{} $$\hat u_k = U \mbox{ in suitable neighborhoods of 0 and 1}.$$ Define $\hat u_k: (0, 1 ) \to {\mathbb{R}}$ as follows $$\hat u_k (x) : = \left\{\begin{array}{cl}
x& \mbox{ if } 0 < x < \frac{x_{1, k}}{3}, \\[6pt]
u_k(x_{1, k}) & \mbox{ if } \frac{2x_{1, k}}{3} < x < x_{1, k}, \\[6pt]
u_k(x) & \mbox{ if } x_{1, k} \le x \le x_{2, k}, \\[6pt]
u_k(x_{2, k}) & \mbox{ if } x_{2, k} < x < \frac{1+2x_{2, k}}{3}, \\[6pt]
x & \mbox{ if } \frac{2+x_{2, k}}{3} < x < 1,
\end{array}\right.$$ and $\hat u_k$ is chosen in $[\frac{x_{1,k}}{3}, \frac{2 x_{1, k}}{3}] \cup [ \frac{1+2x_{2, k}}{3}, \frac{2+x_{2, k}}{3}]$ in such a way that it is affine there and $\hat u_k$ is continuous at the end points.
We claim that $$\label{lem1-e}
\Lambda_{\delta_k} (\hat u_k, (0, 1)) \le {\kappa}+ C c_k.$$ For this purpose, we estimate $\Lambda_{\delta_k} (\hat u_k, (0, 1))$ writing $$\begin{aligned}
\label{lem1-decomp}
\Lambda_{\delta_k} & (\hat u_k, (0, 1)) \\[6pt]
\le & \Lambda_{\delta_k}(\hat u_k, (0, x_{1, k})) + 2
\int_{2x_{1,k}/3}^{x_{1, k}} \, dx \int_{x_{1, k}}^{x_{2, k}} \frac{\varphi_{\delta_k}(|\hat u_k(x) - \hat u_k(y)|)}{|x - y|^{p+1}} dy \nonumber \\[6pt]
& + \Lambda_{\delta_k}(\hat u_k, (x_{2, k}, 1))
+ 2 \int_{x_{2,k}}^{ (1 + 2x_{2, k})/3} \int_{x_{1, k}}^{x_{2, k}} \frac{\varphi_{\delta_k}(|\hat u_k(x) - \hat u_k(y)|)}{|x - y|^{p+1}} dy \, dx \nonumber \\[6pt]
& + \Lambda_{\delta_k}(\hat u_k, (x_{1, k}, x_{2, k})) + \mathop{\int_0^1 \int_0^1}_{|x - y| > \min\{x_{1, k}/3, (1- x_{2, k})/3 \}} \frac{\varphi_{\delta_k}(|\hat u_k(x) - \hat u_k(y)|)}{|x - y|^{p+1}} dy \, dx. \nonumber \\[6pt]
& : = I + II + III + IV + V + VI. \nonumber\end{aligned}$$ We begin with $I$. We have, by and $$\begin{aligned}
I = & \Lambda_{\delta_k}(\hat u_k, (0, x_{1, k})) = \int_0^{x_{1, k}} \int_0^{x_{1, k}} \frac{\varphi_{\delta_k}(|\hat u_k(x) - \hat u_k(y)|) }{|x - y|^{p + 1}} \, dx \, dy \\[6pt]
\le & C \mathop{\int_0^{x_{1, k}} \int_0^{x_{1, k}}}_{|x - y | \le \delta_k} \frac{ \delta_k^{p} |\hat u_k(x) - \hat u_k(y)|^{p+1}}{\delta_k^{p+1}} \frac{1}{|x-y|^{p+1}} \, dx \, dy \\[6pt]
& + C \mathop{\int_0^{x_{1, k}} \int_0^{x_{1, k}}}_{|x - y | > \delta_k} \frac{ \delta_k^p }{|x - y|^{p + 1}} \, dx \, dy. \end{aligned}$$ Since $|\hat u_k(x) - \hat u_k(y)| \le C |x - y|$ for $x,y \in (0, x_{1, k})$, we obtain $$I
\le C \mathop{\int_0^{x_{1, k}} \int_0^{x_{1, k}}}_{|x - y | \le \delta_k} \delta_k^{-1} \, dx \, dy + C \mathop{\int_0^{x_{1, k}} \int_0^{x_{1, k}}}_{|x - y | > \delta_k} \frac{ \delta_k^p }{|x - y|^{p + 1}} \, dx \, dy.$$ It follows from straightforward integral estimates that $$\label{lem1-e1}
I \le C x_{1, k} \mathop{\le}^{\eqref{lem1-x1k}} C c_k.$$
We next consider $II$. It is clear from the definition of $\hat u_k$ that $$\begin{gathered}
II = 2 \int_{2x_{1,k}/3}^{x_{1, k}} \, dx \int_{x_{1, k}}^{x_{2, k}} \frac{\varphi_{\delta_k}(|\hat u_k(x) - \hat u_k(y)|)}{|x - y|^{p+1}} dy \\[6pt]
\le \frac{2 x_{1, k}}{3}\int_{x_{1, k}}^{x_{2, k}} \frac{\varphi_{\delta_k}(|u_k(x_{1, k}) - u_k(y)|)}{|x_{1, k}- y|^{p+1}} dy, \end{gathered}$$ which implies, by and , $$\label{lem1-e2}
II \le C c_k.$$
Similarly, using and , one has $$\label{lem1-e3}
III = \Lambda_{\delta_k}(\hat u_k, (x_{1, k}, x_{2, k})) \le C c_k$$ and $$\label{lem1-e4}
IV = 2 \int_{x_{2,k}}^{ (1 + 2x_{2, k})/3} \int_{x_{1, k}}^{x_{2, k}} \frac{\varphi_{\delta_k}(|\hat u_k(x) - \hat u_k(y)|)}{|x - y|^{p+1}} dy \, dx \le C c_k.$$
It is clear from that $$\label{lem1-e4-1}
V = \Lambda_{\delta_k}(\hat u_k, (x_{1, k}, x_{2, k})) \le \Lambda_{\delta_k} (u_k, (0, 1)) \le {\kappa}+ c_k.$$
We now consider $VI$. We have, for every $c > 0$, $$\mathop{\int_0^1 \int_0^1}_{|x - y| > c} \frac{\varphi_{\delta_k}(|\hat u_k(x) - \hat u_k(y)|)}{|x - y|^{p+1}} dy \, dx \mathop{\le}^{\eqref{cond-varphi-1}} C \delta_k^p/ c^p,$$ which yields $$\begin{gathered}
VI = \mathop{\int_0^1 \int_0^1}_{|x - y| > \min\{x_{1, k}/3, (1- x_{2, k})/3 \}} \frac{\varphi_{\delta_k}(|\hat u_k(x) - \hat u_k(y)|)}{|x - y|^{p+1}} dy \, dx \\[6pt]
\le \frac{C\delta_k^p}{\min\{x_{1, k}/3, (1- x_{2, k})/3 \}^p}. \end{gathered}$$ From and , we derive that $$\label{lem1-e5}
VI \le C \delta_k^p/c_k^p \mathop{ \le }^{\eqref{lem1-p1}} C c_k.$$ Combining - yields $$\Lambda_{\delta_k} (\hat u_k, (0, 1)) \le {\kappa}+ C c_k.$$ The proof of Claim is complete. In view of the definition of ${\kappa}$, we obtain from .
As a consequence of , we have $$\label{lem1-e-***}
\limsup_{k \to + \infty} \Lambda_{\delta_k} (\hat u_k, (0, 1)) \le \limsup_{k \to + \infty} \Lambda_{\delta_k}(u_k, (0, 1)).$$
From $(\hat u_k)$, we now construct [*a family*]{} $(v_\delta) \subset L^p(0, 1)$ such that $v_\delta \to U$ in $L^p(0, 1)$ and $$\lim_{\delta \to 0} \Lambda_{\delta} (v_\delta, (0, 1)) = {\kappa}.$$
Let $(\tau_k) \subset (0,1)$ be a decreasing sequence converging to 0 such that $$\label{tau}
\tau_k \le \delta_k c_k.$$ For each $\delta \in (0, 1)$ small, let $k = k(\delta)$ be such that $\tau_{k(\delta)+1} < \delta \le \tau_{k(\delta)}$. Define $ \hat m = \hat m (\delta)=\delta_{k(\delta)}/\delta$ and set $m = m (\delta)= [\hat m (\delta)]$, the largest integer less than or equal to $\hat m (\delta)$. Then, by , $$\hat m (\delta) \ge \delta_{k(\delta)}/ \tau_{k(\delta)} \ge 1/ c_{k(\delta)}.$$ This implies $$\label{lem1-m}
\hat m (\delta)/ m (\delta) \to 1 \mbox{ as } \delta \to 0.$$ In what follows, for notational ease, we delete the dependence on $\delta$ in $k(\delta)$, $m (\delta)$ and $\hat m (\delta)$.
Consider $v_\delta: (0,1) \to
{\mathbb{R}}$ defined as follows $$v_\delta(x) : = \frac{1}{\hat m} \hat v_\delta (m x),$$ where $\hat v_\delta: (0, m) \to {\mathbb{R}}$ is given by, for $\tau_{k+1} < \delta \le \tau_{k}$, $$\label{lem1-def-hatv}
\hat v_\delta (x) : = [x] + \hat u_k (x - [x]).$$ Then $$\label{lem1-est1*}
\Lambda_\delta(v_\delta, (0, 1)) = \frac{m^{p-1}}{\hat m^p} \Lambda_{\delta_k} (\hat v_\delta, (0, m)).$$ We have $$\begin{gathered}
\label{lem1-decomposition}
\Lambda_{\delta_k} (\hat v_\delta, (0, m)) = \sum_{i=0}^{m-1}
\Lambda_{\delta_k} (\hat v_\delta, (i, i+1))
\\[6pt]+ \sum_{i=0}^{m-1} \int_{i}^{i+1} \, dx \mathop{\int}_{(0, m) \setminus (i, i+1)} \frac{\varphi_{\delta_k}(|\hat v_\delta(x) - \hat v_\delta(y)|)}{|x-y|^{p+1}} \, dy. \end{gathered}$$
It is clear from the definition of $v_\delta$ that, for $0 \le i \le m-1 $, $$\label{lem1-est1*-1}
\Lambda_{\delta_k} (\hat v_\delta, (i, i+1)) = \Lambda_{\delta_k} (\hat u_k, (0, 1))$$ and, that, with $s_k = \min\{x_{1,k}, 1- x_{2, k} \}/3$, $$\begin{gathered}
\label{lem1-est1*-2-1}
\int_{i}^{i+1} \, dx \mathop{\int}_{(0, m) \setminus (i, i+1)} \frac{\varphi_{\delta_k}(|\hat v_\delta(x) - \hat v_\delta(y)|)}{|x-y|^{p+1}} \, dy \\[6pt]
= \int_{i}^{i+1} \, dx \mathop{\mathop{\int}_{(0, m) \setminus (i, i+1)}}_{|y-x| \le s_k} \frac{\varphi_{\delta_k}(|\hat v_\delta(x) - \hat v_\delta(y)|)}{|x-y|^{p+1}} \, dy \\[6pt]
+ \int_{i}^{i+1} \, dx \mathop{\mathop{\int}_{(0, m) \setminus (i, i+1)}}_{|y-x| \ge s_k} \frac{\varphi_{\delta_k}(|\hat v_\delta(x) - \hat v_\delta(y)|)}{|x-y|^{p+1}} \, dy. \end{gathered}$$ Note that, by and , $$s_k \sim c_k.$$ By the same method used to establish , we have $$\label{lem1-est1*-2-2}
\int_{i}^{i+1} \, dx \mathop{\mathop{\int}_{(0, m) \setminus (i, i+1)}}_{|y-x| \ge s_k} \frac{\varphi_{\delta_k}(|\hat v_\delta(x) - \hat v_\delta(y)|)}{|x-y|^{p+1}} \, dy \le C c_k.$$ Note, from the definition of $\hat v_\delta$, that $|\hat v_\delta(x) - \hat v_{\delta} (y)| = |x -y|$ for $x \in (i, i+1)$ and $y \in (0, m) \setminus (i, i+1)$ with $|y-x| \le s_k$. By the same method used to establish , we have $$\label{lem1-est1*-2-3}
\int_{i}^{i+1} \, dx \mathop{\mathop{\int}_{(0, m) \setminus (i, i+1)}}_{|y-x| \le s_k} \frac{\varphi_{\delta_k}(|\hat v_\delta(x) - \hat v_\delta(y)|)}{|x-y|^{p+1}} \, dy \le C c_k.$$ We derive from - that $$\label{lem1-est1*-2}
\int_{i}^{i+1} \, dx \mathop{\int}_{(0, m) \setminus (i, i+1)} \frac{\varphi_{\delta_k}(|\hat v_\delta(x) - \hat v_\delta(y)|)}{|x-y|^{p+1}} \, dy \le C c_k.$$ Combining - and using and , we have $$\label{lem1-est2}
\Lambda_{\delta_k} (\hat v_\delta, (0, m)) \le m ({\kappa}+ C c_k).$$ We deduce from , , and that $$\label{lem1-cl1}
\limsup_{\delta \to 0} \Lambda_{\delta} (v_\delta, (0, 1)) \le {\kappa}.$$
It is clear from that $$\lim_{\delta \to 0} \int_{0}^1 |v_\delta - U|^p \, dx = \lim_{\delta \to 0} \frac{1}{m^p} \int_{0}^1 |\hat v_\delta (mx) - mx|^p \, dx,$$ which yields, by a change of variables, $$\lim_{\delta \to 0} \int_{0}^1 |v_\delta - U|^p \, dx = \lim_{\delta \to 0} \frac{1}{m^{p+1}} \int_{0}^m |\hat v_\delta (x) - x|^p \, dx.$$ Since $$\int_{0}^m |\hat v_\delta (x) - x|^p \, dx = \sum_{j=0}^{m-1} \int_{j}^{j+1} |\hat v_\delta (x) - x|^p \, dx,$$ it follows from and that $$\label{lem1-cl2}
\lim_{\delta \to 0} \int_{0}^1 |v_\delta - U|^p \, dx = 0.$$
Combining and , and using the definition of ${\kappa}$, we obtain . The proof is complete.
We next establish
\[lem2\] Let $p \ge 1$ and let $\varphi$ satisfy -. Let $a< b$ and let $u$ be an affine function on $(a, b)$. There exists a family $(u_\delta) \subset L^p(a, b)$ such that $u_\delta \to u$ in $L^p(a, b)$, as $\delta \to 0$, $$\limsup_{\delta \to 0} \Lambda_{\delta} (u_\delta, (a, b)) \le {\kappa}\int_a^b |u'|^p \, dx,$$ and, for small $\delta$, $$u_{\delta} = u \mbox{ on } (a, a + \delta^{1/2}/6) \cup (b - \delta^{1/2}/ 6, b).$$
By Lemma \[lem1\], after a change of variables, there exists a family $(v_\delta) \subset L^p(a,b)$ converging to $u$ in $L^p(a, b)$ as $\delta \to 0$, and $$\label{lem2-p1}
\lim_{\delta \to 0} \Lambda_{\delta} (v_\delta, (a, b)) = {\kappa}\int_a^b |u'|^p \, dx.$$ As in the proof of Lemma \[lem1\], there exist $(c_\delta) $, $(x_{1, \delta})$, $(x_{2, \delta})$ such that $$\lim_{\delta \to 0} c_\delta = 0, \quad c_\delta \ge \delta^{1/2},$$ $$\Lambda_{\delta}(v_\delta, (a, b)) \le {\kappa}\int_a^b |u'|^p \, dx + c_{\delta},$$ $$x_{1, \delta} \in (a + c_\delta/ 2, a + c_\delta ), \quad x_{2, \delta} \in (b - c_\delta, b - c_\delta/ 2),$$ $$|v_{\delta}(x_{1, \delta}) - u(x_{1 \delta})| \le c_{\delta}, \quad |v_{\delta}(x_{2, \delta}) - u(x_{2, \delta})| \le c_{\delta},$$ $$\int_a^b \frac{\varphi_{\delta}(|v_\delta(x_{1,\delta}) - v_\delta(y)|)}{|x_{1, \delta} - y|^{p+1}} dy \le C, \quad \int_a^b \frac{\varphi_{\delta}(|v_\delta(x_{2,\delta}) - v_\delta(y)|)}{|x_{2, \delta} - y|^{p+1}} dy \le C,$$ for small $\delta$ and for some positive constant $C$ independent of $\delta$. Here we used the fact, for $c \in (0, (b-a)/2)$, $$\liminf_{\delta \to 0} \Lambda_{\delta} (v_\delta, (a+c, b-c)) \ge {\kappa}\int_{a+c}^{b-c} |u'|^p \, dx.$$
Define $\hat v_\delta: (a, b) \to {\mathbb{R}}$ as follows $$\hat v_\delta (x) : = \left\{\begin{array}{cl}
u(x)& \mbox{ if } a < x < \frac{2 a + x_{1, \delta}}{3}, \\[6pt]
v_\delta(x_{1, \delta}) & \mbox{ if } \frac{a + 2x_{1, \delta}}{3} < x < x_{1, \delta}, \\[6pt]
v_\delta(x) & \mbox{ if } x_{1, \delta} \le x \le x_{2, \delta}, \\[6pt]
v_\delta(x_{2, \delta}) & \mbox{ if } x_{2, \delta} < x < \frac{b+2x_{2, \delta}}{3}, \\[6pt]
u(x) & \mbox{ if } \frac{2b+x_{2, \delta}}{3} < x < b,
\end{array}\right.$$ and $\hat v_\delta$ is chosen in $[\frac{2 a + x_{1,\delta}}{3}, \frac{a + 2 x_{1, \delta}}{3}] \cup [ \frac{b+2x_{2, \delta}}{3}, \frac{2 b+x_{2, \delta}}{3}]$ in such a way that it is affine there and $\hat u_\delta$ is continuous at the end points. It is clear that $\hat v_\delta \to u$ in $L^p(a, b)$.
As in the proof of , we have $$\limsup_{\delta \to 0} \Lambda_{\delta}(\hat v_\delta, (a, b)) \le \limsup_{\delta \to 0} \Lambda_{\delta}(v_\delta, (a, b)).$$ By , the conclusion now holds for $(u_\delta)$ with $u_\delta : = \hat v_\delta$.
Using Lemma \[lem2\], we can establish the following key ingredient in the proof of (G2).
\[lemaffine1\] Let $p \ge 1$ and let $\varphi$ satisfy -. Let $u$ be a continuous piecewise linear function defined on ${\mathbb{R}}$ with compact support. There exists a family $(u_\delta) \subset L^p({\mathbb{R}})$ such that $u_\delta \to u$ in $L^p({\mathbb{R}})$, as $\delta \to 0$, and $$\limsup_{\delta \to 0} \Lambda_\delta(u_\delta, {\mathbb{R}}) \le {\kappa}\int_{{\mathbb{R}}}
|u'|^p \, dx.$$
Since $u$ is a continuous piecewise linear function defined on ${\mathbb{R}}$ with compact support, there exist $a_1 < a_2 < \dots <
a_m$ such that $u$ is affine on $(a_i, a_{i+1})$, $1 \le i < m-1$, $u(x) = 0 $ if $x< a_1$ or $x > a_m$, and $u$ is continuous at $a_i$ for $1 \le i \le m$. In what follows, we denote $a_0 = - \infty$ and $a_{m+1} = + \infty$.
For $1 \le i \le m-1$, by Lemma \[lem2\], there exist a family $(v_{i, \delta}) \subset L^p(a_i, a_{i+1})$ such that $$\label{lem3-vi-1-1}
v_{i, \delta} \to u \mbox{ in } L^{p}(a_{i}, a_{i+1}) \mbox{ as } \delta \to 0,$$ $$\label{lem3-vi-A-1}
\limsup_{\delta \to 0} \Lambda_{\delta} ( v_{i, \delta}, (a_i, a_{i+1}) ) \le {\kappa}\int_{a_i}^{a_{i+1}} |u'|^p \, dx,$$ and, for small $\delta$, $$\label{lem3-vi-2}
v_{i, \delta} = u \mbox{ on } (a_i, a_i + \delta^{1/2}/6) \cup (a_{i+1} - \delta^{1/2}/6, a_{i+1}).$$ Set $$\label{lem3-vi-1-2}
v_{0, \delta} = 0 \mbox{ in } (a_0, a_1) \quad { and } \quad v_{m, \delta} = 0 \mbox{ in } (a_m, a_{m+1}).$$ Then $$\label{lem3-vi-A-2}
\Lambda_{\delta} ( v_{0, \delta}, (a_0, a_{1}) ) = \Lambda_{\delta} ( v_{m, \delta}, (a_m, a_{m+1}) ) = 0.$$ Define $u_\delta : {\mathbb{R}}\to {\mathbb{R}}$ as follows $$\label{lem3-def-u}
u_\delta (x) = v_{i, \delta} (x) \mbox{ for } x \in (a_i, a_{i+1}) \mbox{ and } 0 \le i \le m.$$ As in , we have $$\label{lem3-vi-***}
\Lambda_{\delta}(u_\delta, {\mathbb{R}}) \le \sum_{i=1}^{m-1} \Lambda_{\delta} (u_\delta, (a_i, a_{i+1})) + C m \delta^{1/2},$$ for some positive constant $C$ independent of $\delta$ (but $C$ depends on the slope of $u$ on each interval $(a_i, a_{i+1})$ for small $\delta$). It follows from , , and that $$\label{lem3-est-u-1}
\limsup_{\delta \to 0} \Lambda_{\delta}(u_\delta, {\mathbb{R}}) \le {\kappa}\int_{{\mathbb{R}}} |u'|^p \, dx.$$ From and , we have $$\label{lem3-est-u-2}
u_\delta \to u \mbox{ in } L^p({\mathbb{R}}) \mbox{ as } \delta \to 0.$$ The conclusion now follows from and .
We are ready to complete the
We distinguish two cases.
[**Case 1**]{}: ${\kappa}> 0$. In this case, for any function $g \in W^{1, p}({\mathbb{R}})$ with $p>1$ (resp. $g \in BV({\mathbb{R}})$ with $p=1$), we will construct a family $(g_\delta) \subset L^p({\mathbb{R}})$ such that $g_\delta \to g$ in $L^p({\mathbb{R}})$, as $\delta \to 0$, and $$\limsup_{\delta \to 0} \Lambda_{\delta} (g_\delta, {\mathbb{R}}) \le {\kappa}\int_{{\mathbb{R}}} |g'|^p \, dx.$$
[**Case 2**]{}: ${\kappa}= 0$. In this case, for any function $g \in L^p({\mathbb{R}})$ with $p \ge 1$, we will construct a family $(g_\delta) \subset L^p({\mathbb{R}})$ such that $g_\delta \to g$ in $L^p({\mathbb{R}})$, as $\delta \to 0$, and $$\lim_{\delta \to 0} \Lambda_{\delta} (g_\delta, {\mathbb{R}}) = 0.$$
[**Proof in Case 1**]{}: Let $(g_n) \subset L^p({\mathbb{R}})$ be a sequence of continuous piecewise linear functions with compact support such that $g_n \to g$ in $L^p({\mathbb{R}})$, as $n \to + \infty$, and $$\lim_{n \to + \infty} \int_{{\mathbb{R}}} |g_n \sp{\prime} |^p \, dx = \int_{{\mathbb{R}}} |g'|^p \, dx.$$ For each $n \in {\mathbb{N}}$, by Lemma \[lemaffine1\], there exists a family $(g_{n, \delta}) \subset L^p({\mathbb{R}})$ such that $g_{n, \delta} \to g_n$ in $L^p({\mathbb{R}})$, as $\delta \to 0$, and $$\limsup_{\delta \to 0 } \Lambda_{\delta}(g_{n, \delta}, {\mathbb{R}}) \le {\kappa}\int_{{\mathbb{R}}} |g_n \sp{\prime}|^p \, dx.$$ The conclusion now follows from a standard selection process.
[**Proof in Case 2**]{}: Let $(g_n) \subset L^p({\mathbb{R}})$ be a sequence of continuous piecewise linear functions with compact support such that $g_n \to g$ in $L^p({\mathbb{R}})$ as $n \to \infty$. For each $n \in {\mathbb{N}}$, by Lemma \[lemaffine1\], there exists a family $(g_{n, \delta}) \subset L^p({\mathbb{R}})$ such that $g_{n, \delta} \to g_n$ in $L^p({\mathbb{R}})$, as $\delta \to 0$, and $$\lim_{\delta \to 0 } \Lambda_{\delta}(g_{n, \delta}, {\mathbb{R}}) = 0.$$ The conclusion now follows from a standard selection process.
Proof of Property (G1)
----------------------
This section containing two subsections is devoted to the proof of Property (G1). In the first subsection, we consider the case $p=1$. The case $p>1$ is studied in the second subsection.
### Proof of Property (G1) for $p=1$ {#sect-G1-1}
In this section, we consider $p=1$ and assume ${\kappa}> 0$ since there is nothing to prove otherwise. Define $$\label{defb1}
\gamma := \inf \liminf_{\delta \to 0} \Lambda_{\delta} (v_\delta, (0, 1)),$$ where the infimum is taken over all families $(v_\delta) \subset L^1(0, 1)$ such that $v_\delta \to H_{1/2}$ in $L^1(0, 1)$ as $\delta \to 0$. Here and in what follows $H_c(x) := H(x- c)$ for any $c \in {\mathbb{R}}$, where $H$ is the function defined on ${\mathbb{R}}$ by $$H(x) : = \left\{\begin{array}{ll} 0 & \mbox{if } x < 0, \\[6pt]
1 & \mbox{otherwise}.
\end{array}\right.$$
There are two key ingredients.
\[lem-G1\] Let $p = 1$ and let $\varphi$ satisfy -. We have $$\gamma = {\kappa},$$ where ${\kappa}$ is the constant defined in .
\[lem-C4\] Let $p = 1$ and let $\varphi$ satisfy -. Let $u \in L^1( a, b)$ and let $a < t_1 < t_{2} < b$ be two Lebesgue points of $u$. Let $(u_\delta) \subset L^1(t_1, t_2)$ such that $u_\delta \to u$ in $L^{1}(t_1, t_2)$. We have $$\label{lem-C4-state}
\liminf_{\delta \to 0} \Lambda_{\delta}(u_{\delta}, (t_1, t_2))
\ge \gamma |u(t_2) - u(t_1)|.$$
Assuming , we give the
Since $\gamma = {\kappa}$ by , Property (G1) is now a direct consequence of and the fact that for $u \in L^1({\mathbb{R}})$, then $$\label{G1-pro}
\int_{{\mathbb{R}}} |u'| \, dx = \sup \left\{ \sum_{j=1}^m |u(t_{j+1}) - u(t_j)|\right\},$$ where the supremum is taken over all finite sets $\big\{t_j; 1 \le j \le m+1 \big\}$ such that $t_1 < \cdots < t_{m+1}$ and each $t_j$ is a Lebesgue point of $u$, see, e.g., [@EGMeasure Theorem 1 on page 217]. Indeed, we have $$\liminf_{\delta \to 0} \Lambda_{\delta} (u_\delta, {\mathbb{R}}) \ge \sum_{j=1}^{m} \liminf_{\delta \to 0} \Lambda_{\delta} (u_\delta, (t_j, t_{j+1})) \ge \sum_{j=1}^{m} \gamma |u(t_{j+1}) - u(t_j)| \mbox{ by Lemma~\ref{lem-C4}},$$ which implies, by , $$\liminf_{\delta \to 0} \Lambda_{\delta} (u_\delta, {\mathbb{R}}) \ge \gamma \int_{{\mathbb{R}}} |u'| \, dx.$$
The proof of relies on the two lemmas below. The first one is
\[lem-C1\] Let $p = 1$ and let $\varphi$ satisfy -. There exist a sequence $(h_k) \subset L^1(0, 1)$ and a sequence $(\delta_k)\subset {\mathbb{R}}_+$ converging to 0 such that $$\lim_{k \to + \infty} h_k = H_{1/2} \mbox{ in } L^1(0, 1),$$ $$h_k(x) = 0 \mbox{ for } x < 1/16, \quad h_k(x) = 1 \mbox{ for } x > 1-1/16,$$ and $$\limsup_{k \to + \infty} \Lambda_{\delta_k} (h_k, (0, 1))
\le \gamma.$$
Let $(\delta_k) \subset {\mathbb{R}}_+$ and $(g_k) \subset L^1(0, 1)$ be such that $$\lim_{k \to + \infty} \delta_k =0, \quad \lim_{k \to + \infty} g_k = H_{1/2} \mbox{ in } L^1(0, 1),$$ and $$\lim_{k \to + \infty} \Lambda_{\delta_k} (g_k, (0, 1)) = {\gamma}.$$ Let $(c_k) \subset {\mathbb{R}}_+$ be such that, for large $k$, $$\label{lem-C1-ob-1}
\lim_{k \to + \infty} c_k = 0, \quad c_k \ge \delta_k^{1/2},$$ $$\label{lem-C1-ob-2}
\Lambda_{\delta_k} (g_k, (0, 1)) \le \gamma + c_k,$$ $$\label{lem-C1-ob-3}
\int_{0}^1 |g_k - H_{1/2}| \, dx \le c_k^{2},$$
In what follows in this proof, $C$ denotes a positive constant depending only on $\alpha$ and $\beta$. From -, we derive that, for some $\tau_k \in (1/8, 1/5)$ and with $\hat \tau_k = \tau_k + 1/ 2$, $$\label{lem-C1-ob-4}
\int_{\tau_k}^{\tau_k + c_k} |g_k - H_{1/2}| \, dx + \int_{\hat \tau_k}^{\hat \tau_k + c_k} |g_k - H_{1/2}| \, dx \le C c_k$$ and $$\begin{gathered}
\label{lem-C1-ob-5}
\mathop{\iint}_{(\tau_k, \tau_k + c_k) \times (0, 1) } \frac{\varphi_{\delta_k} (|g_k(x) - g_k(y)|)}{|x-y|^{2}} \, dx \, dy \\[6pt]
+ \mathop{\iint}_{(\hat \tau_{k}, \hat \tau_k + c_k) \times (0, 1) } \frac{\varphi_{\delta_k} (|g_k(x) - g_k(y)|)}{|x-y|^{2}} \, dx \, dy \le C c_k. \end{gathered}$$ It follows from and that, for some $b_k \in [1/8, 1/4]$, with ${\hat b}_k = b_k + 1/2$, $$\label{lem-C1-p1}
|g_k(b_k)| + |g_k({\hat b}_k) - 1| \le C c_k$$ and $$\label{lem-C1-p2}
\int_0^1 \frac{\varphi_{\delta_k}(|g_k(b_k) - g_k(y)|)}{|b_k-y|^{2}} \, dy + \int_0^1 \frac{\varphi_{\delta_k}(|g_k({\hat b}_k) - g_k(y)|)}{|{\hat b}_k-y|^{2}} \, dy\le C.$$
Define $h_k: (0, 1) \to {\mathbb{R}}$ as follows $$h_k (x) = \left\{\begin{array}{cl}
0 & \mbox{ for } 0 < x < b_k - 2 c_k, \\[6pt]
g_k(b_k) & \mbox{ for } b_k - c_k < x < b_k, \\[6pt]
g_k(x) & \mbox{ for } b_k \le x \le {\hat b}_k, \\[6pt]
g_k({\hat b}_k) & \mbox{ for } {\hat b}_k < x < {\hat b}_k + c_k, \\[6pt]
1 & \mbox{ for } {\hat b}_k + 2 c_k < x < 1,
\end{array}\right.$$ and $h_k$ is chosen in $[b_k - 2c_k, b_k - c_k] \cup [{\hat b}_k + c_k, {\hat b}_k + 2 c_k]$ in such a way that it is affine there and $h_k$ is continuous at the end points. As in the proof of in Lemma \[lem1\], one can check that $$\label{lem-C1-p6}
\Lambda_{\delta_k}(h_k, (0, 1)) \le \gamma + C c_k.$$ Therefore, the conclusion holds for $h_k$.
The second lemma used in the proof of is
\[lem-C2\] Let $p = 1$ and let $\varphi$ satisfy -. There exist a sequence $(u_k) \subset L^1(0, 1)$ and a sequence $(\mu_k) \subset {\mathbb{R}}_+$ such that $$\lim_{k \to + \infty} \mu_k = 0, \quad \lim_{k \to + \infty} u_k = U \mbox{ in } L^1(0, 1),$$ and $$\limsup_{k \to + \infty} \Lambda_{\mu_k} (u_k, (0, 1))
\le \gamma.$$
Let $(\delta_k)$ and $(h_k) \subset L^1(0, 1)$ be the sequences satisfying the conclusion of Lemma \[lem-C1\]. Given $n \in {\mathbb{N}}$, set $I_j= (j/n, (j+1)/ n) $ for $0 \le j \le n-1$, and define $$\label{def-fkn}
f_{k, n} (x) = \frac{1}{n} h_k \Big( n(x - j/n) \Big) + \frac{j}{n} \mbox{ for } x \in I_j.$$ By a change of variables, we obtain $$\label{lem-C2-p1}
\int_{0}^1 |f_{k, n} (x) - x| \, dx = \frac{1}{n} \int_{0}^1 |h_k (x) - x| \, dx.$$
We next estimate $$\Lambda_{\delta_k/n} (f_{k, n}, (0, 1)).$$ It is clear that $$\begin{gathered}
\label{lem-C2-p2}
\Lambda_{\delta_k/n} (f_{k, n}, (0, 1))
\le \sum_{j=0}^{n-1} \Lambda_{\delta_k/n} (f_{k, n}, I_j) \\[6pt]
+ \sum_{j=0}^{n-1} \mathop{\iint}_{I_j \times ( (0, 1) \setminus I_j)}
\frac{ \varphi_{\delta_k/n} (|f_{k, n}(x) - f_{k,n}(y)|)}{|x-y|^{2}} \, dx \, dy.\end{gathered}$$ We have, by a change of variables, $$\label{lem-C2-p3}
\Lambda_{\delta_k/n} (f_{k, n}, I_j) = \frac{1}{n}\Lambda_{\delta_k} (h_k, (0, 1)).$$ It is clear that $$\begin{gathered}
\mathop{\iint}_{I_j \times ((0, 1) \setminus I_j)}
\frac{ \varphi_{\delta_k/n} (|f_{k, n}(x) - f_{k,n}(y)|)}{|x-y|^{2}} \, dx \, dy \\[6pt]
= \mathop{ \mathop{\iint}_{I_j \times ((0, 1) \setminus I_j)}}_{|x - y| < 1/(16 n)}
\frac{ \varphi_{\delta_k/n} (|f_{k, n}(x) - f_{k,n}(y)|)}{|x-y|^{2}} \, dx \, dy \\[6pt] + \mathop{ \mathop{\iint}_{I_j \times ((0, 1) \setminus I_j)}}_{|x -y| > 1/(16n)}
\frac{ \varphi_{\delta_k/n} (|f_{k, n}(x) - f_{k,n}(y)|)}{|x-y|^{2}} \, dx \, dy. \end{gathered}$$ Since the first term on the RHS of the above identity is 0, by straightforward integral estimates, we obtain $$\label{lem-C2-p4}
\mathop{\iint}_{I_j \times ((0, 1) \setminus I_j)}
\frac{ \varphi_{\delta_k/n} (|f_{k, n}(x) - f_{k,n}(y)|)}{|x-y|^{2}} \, dx \, dy \le C \frac{\delta_k}{n} \ln n.$$ Set $$\mbox{$n_k= [\ln \delta_k^{-1}]$ (the integer part of $\ln \delta_k^{-1}$)} \quad \mbox{ and } \quad \mu_k = \delta_k/ n_k,$$ so that $n_k \to + \infty$ and $\mu_k \to 0$ as $k \to + \infty$. Combining , , and yields, with $u_k = f_{k, n_k}$, $$\label{lem-C2-p5}
\Lambda_{\mu_k} (u_k, (0, 1)) \le \Lambda_{\delta_k} (h_k, (0, 1)) + C \delta_k \ln n_k.$$ It follows that $$\label{lem-C2-p6}
\limsup_{k \to + \infty} \Lambda_{\mu_k} (u_k, (0, 1)) \le \gamma$$ since $\limsup_{k \to + \infty} \Lambda_{\delta_k}(h_k, (0, 1)) \le \gamma$ by Lemma \[lem-C1\]. Since $n_k \to + \infty$, we have, by , $$\label{lem-C2-p7}
\lim_{k \to + \infty} \int_{0}^1 |u_{k} - U| \, dx = 0.$$ The conclusion follows from and .
We are ready to give the
By Property (G2) applied with $g = H_{1/2}$ and $I =(0, 1)$, there exists a family $(g_\delta) \subset L^1(0, 1)$ such that $g_\delta \to H_{1/2}$ in $L^1(0, 1)$ and $$\limsup_{\delta \to 0} \Lambda_\delta(g_\delta, (0, 1)) \le {\kappa}.$$ This implies, by the definition of $\gamma$ in , that $\gamma \le {\kappa}$. By and Remark \[rem-k\], one obtains ${\kappa}\le \gamma$. The conclusion follows.
We now give the
We begin with the following
For any ${\varepsilon}> 0$, there exist two positive numbers $\hat \delta_1$, $\hat \delta_2$ such that for any $a_1, b_1, c \in {\mathbb{R}}$ with $a_1<b_1$, and for any $u \in
L^1(a_1, b_1)$ satisfying $$\|u - c H_{a_1 + \frac{1}{2} (b_1-a_1)}\|_{L^{1}(a_1, b_1)} \le |c| (b_1 - a_1) \hat \delta_{1},$$ one has $$\Lambda_{\delta}(u, (a_1, b_1)) \ge |c| (\gamma - {\varepsilon}) \mbox{ for all } \delta \in (0, |c| \hat \delta_2).$$ To establish the claim, we first consider the case $(a_1, \, b_1) = (0, 1)$ and $c = 1$. The existence of $\hat \delta_{1}$ and $\hat \delta_{2}$ in this case is a direct consequence of the definition of $\gamma$ by a contradiction argument. The general case follows from this case by a change of variables.
We now prove . Without loss of generality, one may assume that $t_1= 0$, $t_2 = 1$, $u(t_1) = 0$, and $u(t_2) = 1$. It suffices to prove that $$\label{lem-C4-main}
\liminf_{k \to + \infty} \Lambda_{\delta_k}(u_{k}, (0, 1)) \ge \gamma,$$ for every $(\delta_k) \subset {\mathbb{R}}_+$ converging to 0, and for every $(u_k) \subset L^1(0, 1)$ such that $u_k \to u$ in $L^1(0, 1)$ and $\sup_{k} \Lambda_{\delta_k}(u_{k}, (0, 1)) < + \infty$.
Set $$T = \sup_{k} \Lambda_{\delta_k}(u_{k}, (0, 1)) < + \infty.$$ Fix ${\varepsilon}>0$ (arbitrary). Let $\hat \delta_{1} $ be the constant in the Claim corresponding to ${\varepsilon}$. Without loss of generality, one may assume that $\hat \delta_1 < 1$. Let $c$ be a small positive number such that $$\int_0^{c} |u(x) | \, dx + \int_{1-c}^{1} |u(x) - 1| \, dx \le c \hat \delta_1^2/64.$$ Since $0$ and $1$ are Lebesgue points, such a $c$ exists. Since $u_k \to u$ in $L^1(0, 1)$, it follows that, for large $k$, $$\int_0^{c} |u_k(x)| \, dx + \int_{1-c}^{1} |u_k(x) - 1| \, dx \le c \hat \delta_1^2/32.$$ This implies, for large $k$, $$| A_k | \ge c/2,$$ where $A_k = \{ x \in (0, c); |u_k(x)| \le \delta_1^2/16 \} $. There exists $x_{1, k} \in A_k$ such that $$\label{lem-C4-p1}
\int_{0}^{1} \frac{\varphi_{\delta_k} ( |u_{k} (x_{1, k}) - u_{k} (y) |)}{|x_{1, k} - y|^{2}} \, dy
\le \frac{1}{|A_k|} \int_{A_k} \, dx \int_{0}^{1} \frac{\varphi_{\delta_k}(|u_{k}(x) - u_{k}(y)| )}{|x - y|^{2}} \, dy \le 2T /c.$$ Similarly, there exists $x_{2, k} \in \{ x \in (1-c, 1); |u_k(x)-1| \le \delta_1^2/16 \}$ such that $$\label{lem-C4-p2}
\int_{0}^{1} \frac{\varphi_{\delta_k} ( |u_{k} (x_{2, k}) - u_{k} (y) |)}{|x_{2, k} - y|^{2}} \, dy
\le 2T/c.$$ It is then clear that $$\label{lem-C4-p3}
|u_{k}(x_{1, k})| + |u_{k}(x_{2, k}) - 1 | \le \hat \delta_{1}^{2}/8,$$ for large $k$.
For each (fixed) $n>0$ (large), define $v_{k} : (-n, n) \to {\mathbb{R}}$ as follows $$v_{k} (x) = \left\{\begin{array}{cl} u_{k} (x_{1, k}) & \mbox{ if } -n < x < x_{1, k}, \\[6pt]
u_{k}(x) & \mbox{ if } x_{1, k} \le x \le x_{2, k}, \\[6pt]
u_{k}(x_{2, k}) & \mbox{ if } x_{2, k} < x < n.
\end{array} \right.$$ We have, since $v_k$ is constant on $(-n, x_{1, k})$ and on $(x_{2, k}, n)$, $$\begin{aligned}
\label{lem-C4-d1}
\Lambda_{\delta_k}(v_{k}, (-n, n) ) \le & \Lambda_{\delta_k}(v_{k}, (x_{1, k}, x_{2, k}) ) \nonumber \\[6pt]
& + 2 \int_{x_{1, k} - \delta_k}^{x_{1, k}} \int_{x_{1, k}}^{x_{2, k}} \frac{\varphi_{\delta_k}(|v_{k}(x) - v_{k}(y)| )}{|x - y|^{2}} \, dy \, dx\nonumber \\[6pt]
& + 2 \int_{x_{1, k}}^{x_{2, k}} \int_{x_{2, k}}^{x_{2, k} + \delta_k} \frac{\varphi_{\delta_k}(|v_{k}(x) - v_{k}(y)| )}{|x - y|^{2}} \, dy \, dx \nonumber \\[6pt]
& + 2 \int_{-n}^{x_{1, k} - \delta_k} \int_{x_{1, k}}^{x_{2, k}} \frac{\varphi_{\delta_k}(|v_{k}(x) - v_{k}(y)| )}{|x - y|^{2}} \, dy \, dx\nonumber \\[6pt]
& + 2 \int_{x_{1, k}}^{x_{2, k}} \int_{x_{2, k} + \delta_k}^{n} \frac{\varphi_{\delta_k}(|v_{k}(x) - v_{k}(y)| )}{|x - y|^{2}} \, dy \, dx \nonumber \\[6pt]
& + \mathop{\int_{-n}^{n} \int_{-n}^{n}}_{|x - y|> x_{2, k} - x_{1, k}} \frac{\varphi_{\delta_k}(|v_{k}(x) - v_{k}(y)| )}{|x - y|^{2}} \, dy \, dx. \end{aligned}$$ By straightforward integral estimates, we have $$\begin{gathered}
\int_{x_{1, k} - \delta_k}^{x_{1, k}} \int_{x_{1, k}}^{x_{2, k}} \frac{\varphi_{\delta_k}(|v_{k}(x) - v_{k}(y)| )}{|x - y|^{2}} \, dy \, dx \\[6pt] + \int_{x_{1, k}}^{x_{2, k}} \int_{x_{2, k}}^{x_{2, k} + \delta_k} \frac{\varphi_{\delta_k}(|v_{k}(x) - v_{k}(y)| )}{|x - y|^{2}} \, dy \, dx
\mathop{\le}^{\eqref{lem-C4-p1}-\eqref{lem-C4-p2}} C \delta_k, \end{gathered}$$ $$\begin{gathered}
\int_{-n}^{x_{1, k} - \delta_k} \int_{x_{1, k}}^{x_{2, k}} \frac{\varphi_{\delta_k}(|v_{k}(x) - v_{k}(y)| )}{|x - y|^{2}} \, dy \, dx \\[6pt]
+ \int_{x_{1, k}}^{x_{2, k}} \int_{x_{2, k} + \delta_k}^{n} \frac{\varphi_{\delta_k}(|v_{k}(x) - v_{k}(y)| )}{|x - y|^{2}} \, dy \, dx \mathop{\le}^{\eqref{cond-varphi-1}} C \delta_k \ln n + \delta_k |\ln \delta_k|, \end{gathered}$$ $$\mathop{\int_{-n}^{n} \int_{-n}^{n}}_{|x - y|> x_{2, k} - x_{1, k}} \frac{\varphi_{\delta_k}(|v_{k}(x) - v_{k}(y)| )}{|x - y|^{2}} \, dy \, dx \mathop{\le}^{\eqref{cond-varphi-1}} C \delta_k \ln n,$$ for some positive constant $C$ independent of $k$ and $n$. It follows from that $$\Lambda_{\delta_k}(v_{k}, (-n, n)) \le \Lambda_{\delta_k}(u_{k}, (0, 1) ) + C \delta_k \ln n + C \delta_k |\ln \delta_k|.$$ This implies, for every $n$, $$\label{lem-C4-e2}
\liminf_{k \to + \infty} \Lambda_{\delta_k}(u_{k}, (0, 1) ) \ge \liminf_{k \to + \infty} \Lambda_{\delta_k}(v_{k}, (-n, n)).$$
We have, by , $$\label{lem-C4-e3}
\|v_{k}(x) - H_{0} (x) \|_{L^{1}(-n, n)} \le \hat \delta_{1}^{2} n / 2 + C,$$ since $(\| u_k\|_{L^1(0,1)})$ is bounded. We now fix $n \ge 2 C / \hat \delta_1$ so that the RHS of is less than $n \hat \delta_{1}$ since $\hat \delta_1 < 1$. Applying the Claim with $c=1$ and $(a_1,b_1) = (-n, n)$, we have, for large $k$, $$\label{lem-C4-e4}
\Lambda_{\delta_k}(v_{k}, (-n, n)) \ge \gamma -{\varepsilon}.$$ Combining and yields $$\liminf_{k \to + \infty} \Lambda_{\delta_k}(u_{k}, (0, 1) ) \ge \gamma -{\varepsilon}.$$ Since ${\varepsilon}>0$ is arbitrary, follows.
The proof of Property (G1) for $p=1$ is complete.
### Proof of Property (G1) for $p>1$ {#sect-G1-p}
Throughout this section, we assume that ${\kappa}> 0$ since there is nothing to prove otherwise. The first key ingredient of the proof is
\[lem1-G1\] Let $p > 1$ and let $\varphi$ satisfy -. Let $ a < b$, and $u \in L^p(a, b)$ and let $t_1, t_2 \in (a, b)$ be two Lebesgue points of $u$. Then, for some positive constant $\sigma$ depending only on $\alpha$, $\beta$, and $p$, $$\label{lem1-G1-es}
\liminf_{\delta \to 0} \Lambda_{\delta} (u_\delta, (t_1, t_2)) \ge \sigma {\kappa}(t_2 - t_1)^{1-p} |u(t_2) - u(t_1)|^p,$$ for any family $(u_\delta) \subset L^p(t_1, t_2)$ such that $u_\delta \to u$ in $L^p(t_1, t_2)$, as $\delta \to 0$.
Without loss of generality, one may assume that $t_1 = 0$, $t_2 = 1$, $u(t_1) = 0$, and $u(t_2) = 1$. Let $(\delta_k)$ and $(u_k)$ be arbitrary such that $\delta_k \to 0$, $u_k \to u$ in $L^p(0, 1)$, and $$\lim_{k \to + \infty} \Lambda_{\delta_k} (u_k, (0, 1)) \mbox{ exists and is finite}.$$ Denote $\tau$ the limit of $\Lambda_{\delta_k} (u_k, (0, 1))$. In order to establish , it suffices to prove $$\label{lem1-claim}
{\kappa}\le C \tau.$$ Here and in what follows, $C$ denotes a positive constant depending only on $\alpha$, $\beta$, and $p$.
Let $(c_k) \subset {\mathbb{R}}_+$ be such that $$\label{lem1-G1-p1}
\lim_{k \to + \infty} c_k =0, \quad c_k \ge \delta_k^{1/2},$$ $$\label{lem1-G1-p1-1}
\Lambda_{\delta_k} (u_k, (0, 1)) \le \tau + c_k,$$ $$\label{lem1-G1-p2}
\int_{0}^1 \frac{\varphi_{\delta_k}(|u_k(c_k) - u_k(y)|)}{|c_k - y|^{p+1}} \, dy \le C c_k^{-1}(\tau + c_k),$$ $$\label{lem1-G1-p3}
\int_{0}^1 \frac{\varphi_{\delta_k}(|u_k(1-c_k) - u_k(y)|)}{|1-c_k - y|^{p+1}} \, dy \le C c_k^{-1}(\tau + c_k).$$ $$\label{lem1-G1-p4}
|u_k(c_k)| + |u_k(1- c_k) - 1| \to 0,$$ for large $k$.
For simplicity of presentation, we will assume that $u_k(c_k) = 0$ and $u_k(1 - c_k) = 1$. Define $\hat u_k: (0, 1 ) \to {\mathbb{R}}$ as follows $$\hat u_k (x) = \left\{\begin{array}{cl}
0& \mbox{ if } 0 < x < c_k, \\[6pt]
u_k(x) & \mbox{ if } c_k \le x \le 1 - c_k, \\[6pt]
1 & \mbox{ if } 1 - c_k < x < 1.
\end{array}\right.$$ For $n \in {\mathbb{N}}$, set $$f_{k, n}(x) = \hat u_k(x - [x]) + [x] \mbox{ for } x \in (0, n)$$ and $$g_{k, n} (x) = \frac{1}{n} f_{k, n} (nx) \mbox{ for } x \in (0, 1).$$
We have, by a change of variables, $$\Lambda_{\delta_k/n} (g_{k, n}, (0, 1)) = \frac{1}{n} \Lambda_{\delta_k} (f_{k, n}, (0, n)).$$ Using , one can check, by straightforward integral estimates, that $$\Lambda_{\delta_k} (f_{k, n}, (0, n)) \le C n (\tau + c_k) + C n \delta_k^p/ c_k^p.$$ This implies, by and , $$\label{lem1-p5}
\Lambda_{\delta_k/n} (g_{k, n}, (0, 1)) \le C \tau + C c_k.$$
On the other hand, we have $$\label{lem1-p6}
\int_{0}^1 |g_{k, n}(x) - x|^p \, dx = \frac{1}{n^p} \int_0^n |f_{k, n} (nx) - n x|^p \, dx =
\frac{1}{n^{p-1}} \int_0^1 |\hat u_{k}(x) - x|^p \, dx.$$ Taking $n = n_k = [\ln \delta_k^{-1}]$, we derive from that $$\lim_{k \to + \infty} \int_{0}^1 |g_{k, n_k}(x) - U|^p \, dx = 0,$$ since $p>1$ and $(\|\hat u_k \|_{L^p(0, 1)})$ is bounded. By noting that $\delta_k/ n_k \to 0$ as $k \to + \infty$, we derive from the definition of ${\kappa}$ that $$\label{lem1-p7}
{\kappa}\le \liminf_{k \to + \infty} \Lambda_{\delta_k/n_k} (g_{k, n_k}, (0, 1)).$$ Combining and yields $${\kappa}\le C \tau,$$ which is .
From Lemma \[lem1-G1\], we now derive
\[lem1-G1-1\] Let $p > 1$ and let $\varphi$ satisfy -. Let $u \in L^p({\mathbb{R}})$ and assume that, for some $(u_\delta) \subset L^p({\mathbb{R}})$ converging to $u$ in $L^p({\mathbb{R}})$, $$\liminf_{\delta \to 0} \Lambda_{\delta}(u_\delta, {\mathbb{R}}) < + \infty.$$ Then $u \in W^{1, p}({\mathbb{R}})$.
As a consequence of , one has, for every $ - \infty < a< b< + \infty$, $$\label{lem1-G1-1-p1}
\liminf_{\delta \to 0} \Lambda_{\delta} (u_\delta, (a, b)) \ge \sigma {\kappa}(b - a)^{1-p} |\mathop{\mathrm{ess \; sup} }_{x \in (a, b )}
u - \mathop{\mathrm{ess \; inf} }_{x \in (a, b)} u|^p,$$ for some constant $\sigma > 0$, independent of $a$ and $b$. Set, for $h \in (0, 1)$, $$\tau_{h}(u)(x) = \frac{1}{h} \Big(u(x+h) - u (x) \Big) \quad \mbox{ for } x \in {\mathbb{R}}.$$ For each $m \ge 2$ and $h \in (0, 1)$, fix $K > 0$ such that $Kh \ge m$. Then $$\label{lem1-G1-1-p2}
\int_{-m}^m |\tau_h(u)|^p \, dx \le \sum_{k = -K}^{K}\int_{k
h}^{(k+1)h } |\tau_h(u)|^p \, dx.$$ Since, for every $a \in {\mathbb{R}}$, $$\int_a^{a + h } |\tau_{h}(u)|^p \, dx \le \int_a^{a + h
}\frac{1}{h^p} |\mathop{\mathrm{ess \; sup} }_{t \in (a, a + 2h )}
u - \mathop{\mathrm{ess \; inf} }_{t \in (a, a + 2h ) } u|^p \,
dx,$$ it follows from that $$\int_a^{a + h } |\tau_{h}(u)|^p \, dx \le \frac{2^{p-1}}{\sigma {\kappa}} \liminf_{\delta \to 0} \Lambda_{\delta} (u_\delta, (a, a+2h)).$$ We derive from that $$\int_{-m}^m |\tau_h(u)|^p \, dx \le \frac{2^{p}}{\sigma {\kappa}} \liminf_{\delta \to 0} \Lambda_{\delta} (u_\delta, {\mathbb{R}}).$$ Since $m \ge 2$ is arbitrary, we obtain, for all $h \in (0, 1)$, $$\label{lem1-G1-1-p3}
\int_{{\mathbb{R}}} |\tau_h(u)|^p \, dx \le \frac{2^{p}}{\sigma {\kappa}} \liminf_{\delta \to 0} \Lambda_{\delta} (u_\delta, {\mathbb{R}}).$$ It follows that $u
\in W^{1,p}({\mathbb{R}})$ (see e.g. [@BrAnalyse1 Chapter 8]).
The second key ingredient in the proof of Property (G1) is the following useful property of functions in $W^{1, p}({\mathbb{R}})$.
\[lem-lem\] Let $p > 1$ and $u \in W^{1, p}({\mathbb{R}})$ (so that $u$ admits a continuous representative still denoted by $u$). Given ${\varepsilon}_1 > 0$, there exist a subset $B$ of Lebesgue points of $u'$ and $\ell \ge 1$ such that $$\label{G1-p-pro-Am}
\int_{{\mathbb{R}}\setminus B} |u'|^p \, dx \le {\varepsilon}_1 \int_{{\mathbb{R}}} |u'|^p \, dx,$$ and, for every open interval $I'$ with $|I'| \le 1/\ell$ and $I' \cap B \neq \emptyset$, and for every $x \in I' \cap B$, $$\label{lem-lem-1}
\frac{1}{|I'|^p }\fint_{I'} |u(y) - u(x) - u'(x) (y - x)|^p \, d y \le {\varepsilon}_1$$ and $$\label{lem-lem-2}
|u'(x)|^p \ge (1 - {\varepsilon}_1) \fint_{I'} |u'(y)|^p \, dy.$$
We first recall the following property of $W^{1, p}({\mathbb{R}})$ functions (see e.g., [@Ziemer Theorem 3.4.2]): Let $f \in W^{1,p} ({\mathbb{R}})$. Then, for a.e. $x \in {\mathbb{R}}$, $$\label{G1-p-measure}
\lim_{r \to 0} \frac{1}{r^p} \fint_{x-r}^{x+r} \big|f(y) - f(x) - f'(x)(y - x) \big|^p \, dy = 0.$$
Given $n \in {\mathbb{N}}$, define, for a.e. $x \in {\mathbb{R}}$, $$\label{G1-p-def-rho}
\rho_n(x) = \sup \left\{ \frac{1}{r^{p}}
\fint_{x-r}^{x+r} \big|u(y) - u(x) - u'(x) (y-x) \big|^p \, dy ; \; r \in (0, 1/n) \right\}$$ and $$\label{G1-p-def-tau}
\tau_n(x) = \sup \left\{ \fint_{x-r}^{x+r}| u'(y) - u'(x)|^p \, dy; r \in (0, 1/n) \right\}.$$ Note that, by , $\rho_n (x) \to 0$ for a.e. $x \in {\mathbb{R}}$ as $n \to + \infty$. We also have, $\tau_n (x)\to 0$ for a.e. $x \in {\mathbb{R}}$ as $n \to + \infty$ (and in fact at every Lebesgue points of $u'$). For $m \ge 1$, set $$D_m = \big\{x \in (-m, m); \mbox{$x$ is a Lebesgue point of $u'$ and } |u'(x)| \ge 1/ m \big\}.$$ Then there exists $m \ge 1$ such that $$\label{G1-p-pro-Am-1}
\int_{{\mathbb{R}}\setminus D_m} |u'|^p \, dx \le \frac{{\varepsilon}_1}{2} \int_{{\mathbb{R}}} |u'|^p \, dx.$$ Fix such an $m$. By Egorov’s theorem, there exist a subset $B$ of $ D_m$ such that $(\rho_n)$ and $(\tau_n)$ converge to $0$ uniformly on $B$, and $$\label{G1-p-pro-Am-2}
\int_{D_m \setminus B} |u'|^p \, dx \le \frac{{\varepsilon}_1}{2} \int_{{\mathbb{R}}} |u'|^p \, dx.$$ Combining and yields .
We have, for every non-empty, open interval $I'$ and $x \in {\mathbb{R}}$ (in particular for $x \in I' \cap B$), $$\label{lem-lem-*}
\left(\fint_{I'} |u'(y)|^p \, dy \right)^{1/p} \le \left(\fint_{I'} |u'(y) - u'(x) |^p \, dy \right)^{1/p} + |u'(x)|,$$ Since $(\rho_n)$ and $(\tau_n)$ converge to $0$ uniformly on $B$ and $|u'(x)| \ge 1/ m$ for $x \in B$, it follows from that there exists an $\ell \ge 1$ such that and holds. The proof is complete.
We are ready to give the
We begin with
[*Claim 2:*]{} For ${\varepsilon}> 0$, there exist two positive constants $\hat \delta_1, \hat \delta_2$ such that for every $c, d \in {\mathbb{R}}$, for every open bounded interval $I'$ of ${\mathbb{R}}$, and for every $f \in L^p(I')$ satisfying $$\fint_{I'} |f(y) - (c y + d)|^p \, dy < \hat \delta_1 |c|^p |I'|^p,$$ one has $$\Lambda_{\delta} (f, I') \ge ({\kappa}- {\varepsilon})|c|^p |I'| \mbox{ for all } \delta \in (0, \hat \delta_2 |c| |I'|).$$ This claim is a consequence of the definition of ${\kappa}$ and its proof is omitted (it is similar to the one of Claim 1 in the proof of ).
In order to establish Property (G1), it suffices to prove that $$\liminf_{k \to + \infty} \Lambda_{\delta_k}(g_k, {\mathbb{R}}) \ge {\kappa}\int_{{\mathbb{R}}} |g'|^p \,dx$$ for every $(\delta_k) \subset {\mathbb{R}}_+$ and $(g_k) \subset L^p({\mathbb{R}})$ such that $\delta_k \to 0$ and $g_k \to g$ in $L^p({\mathbb{R}})$.
Without loss of generality, one may assume that $\liminf_{k \to + \infty} \Lambda_{\delta_k}(g_k, {\mathbb{R}}) < + \infty$. It follows from Lemma \[lem1-G1-1\] that $g \in W^{1, p}({\mathbb{R}})$. Fix ${\varepsilon}> 0$ (arbitrary) and let $\hat \delta_1$ be the positive constant corresponding to ${\varepsilon}$ in Claim 2. Set $$A_m = \{x \in {\mathbb{R}}; \; \mbox{$x$ is a Lebesgue points of $g'$ and } |g'(x)| \le 1/m \} \mbox{ for } m \ge 1.$$ Since $$\lim_{m \to + \infty} \int_{A_m} |g'|^p \, dx = 0,$$ there exists $m \ge 1$ such that $$\label{G1-p-pro-Am-11}
\int_{A_m} |g'|^p \, dx \le \frac{{\varepsilon}}{2} \int_{{\mathbb{R}}} |g'|^p \, dx.$$ Fix such an $m$. By Lemma \[lem-lem\] applied to $u = g$ and ${\varepsilon}_1 = \min\{ {\varepsilon}/ 2, \hat \delta_1/ (2m)^p \}$, there exist a subset $B$ of Lebesgue points of $g'$ and a positive integer $\ell$ such that $$\label{estBm-11}
\int_{{\mathbb{R}}\setminus B} | g'|^p \, dx \le \frac{{\varepsilon}}{2} \int_{{\mathbb{R}}}
|g'|^p \, dx,$$ and for every open interval $I'$ with $|I'| \le 1 / \ell$ and $I' \cap B \neq \emptyset$, and, for every $x \in I' \cap B$, $$\label{G1-p-rho-p1}
\frac{1}{|I'|^p}
\fint_{I'} \big|g(y) - g(x) - g'(x)(y-x) \big|^p \, dy \le \hat \delta_1/ (2m)^p$$ and $$\label{G1-p-tau-p2}
|g'(x)|^p |I'| \ge (1 - {\varepsilon}) \int_{I'} |g' |^p \, dy.$$ Fix such an $\ell$. Set $$B_m = (B \setminus A_m)$$ and denote $${\bf \Omega}_\ell = \Big\{(i/\ell, (i+1) /\ell); i \in {\mathbb{Z}}\Big\} \quad \mbox{ and } \quad { \bf
J}_\ell = \Big\{ J \in {\bf \Omega}_\ell ; \; J
\cap B_m \neq \emptyset \Big\}.$$ Since ${\mathbb{R}}\setminus (B \setminus A_m) \subset ({\mathbb{R}}\setminus B) \cup A_m$, it follows from and that $$\label{estBm}
\int_{{\mathbb{R}}\setminus B_m} | g'|^p \, dx = \int_{{\mathbb{R}}\setminus (B \setminus A_m)} | g'|^p \, dx \le {\varepsilon}\int_{{\mathbb{R}}}
|g'|^p \, dx.$$
Take $J \in {\bf J}_\ell$ and $x \in J \cap B_m$. Since $g_k \to g$ in $L^p(J)$, we derive from (applied with $I' = J$ which is admissible since $B_m \subset B$) that $$\lim_{k \to + \infty} \frac{1}{|J|^p} \fint_{J} \big|g_k(y) - g(x) - g'(x)(y-x) \big|^p \, dy \le \hat \delta_1/ (2m)^p.$$ Applying Claim 2 with $I' = J$, $f = g_k$ for large $k$, $c = g'(x)$, and $d = g(x)$, we have $$\liminf_{k \to + \infty} \Lambda_{\delta_k}(g_k, J) \ge
({\kappa}- {\varepsilon}) |g'(x)|^p |J|,$$ which implies, by , $$\label{estI'}
\liminf_{k \to + \infty}\Lambda_{\delta_k}(g_k, J) \ge
({\kappa}- {\varepsilon}) (1- {\varepsilon})\int_{J} |g'|^p \, dy.$$ Since $$\begin{aligned}
\liminf_{k \to + \infty} \Lambda_{\delta_k}(g_k, {\mathbb{R}}) \ge \sum_{J \in
{\bf J}_\ell }\liminf_{k \to + \infty} \Lambda_{\delta_k}(g_k, J),\end{aligned}$$ it follows from that $$\begin{gathered}
\liminf_{k \to + \infty} \Lambda_{\delta_k}(g_k, {\mathbb{R}})
\ge ({\kappa}- {\varepsilon}) (1- {\varepsilon}) \sum_{J \in {\bf J}_\ell} \int_{J} |g'|^p \, dx \\[6pt]
\ge ({\kappa}- {\varepsilon}) (1- {\varepsilon})\int_{B_m} |g'|^p \, dx \mathop{\ge}^{\eqref{estBm}}
({\kappa}- {\varepsilon}) (1- {\varepsilon})^2 \int_{{\mathbb{R}}} |g'|^p \, dx; \end{gathered}$$ here in the second inequality, we have used the fact $B_m$ is contained in $\bigcup_{J \in {\bf J}_\ell} J$ up to a null set. Since ${\varepsilon}>0$ is arbitrary, one has $$\begin{aligned}
\liminf_{k \to + \infty}\Lambda_{\delta_k}(g_k, {\mathbb{R}}) \ge
{\kappa}\int_{{\mathbb{R}}} |g'|^p \, dx.\end{aligned}$$ The proof is complete.
[**Acknowledgments.**]{} This work was completed during a visit of H.-M. Nguyen at Rutgers University. He thanks H. Brezis for the invitation and the Department of Mathematics for its hospitality.
[10]{}
C. Antonucci, M. Gobbino, M. Migliorini, and N. Picenni, *[Optimal constants for a non-local approximation of Sobolev norms and total variation]{}*, to appear.
C. Antonucci, M. Gobbino, and N. Picenni, *On the gap between gamma-limit and pointwise limit for a non-local approximation of the total variation*, to appear.
J. Bourgain and H.-M. Nguyen, *[A new characterization of Sobolev spaces]{}*, C. R. Acad. Sci. Paris, Ser. [**I**]{} **343** (2006), 75–80.
H. Brezis, *[Functional Analysis, Sobolev Spaces and Partial Differential Equations]{}*, Springer, 2010.
H. Brezis and H.-M. Nguyen, *[Non-local functionals related to the total variation and applications in Image Processing]{}*, Ann. PDE **4** (2018), 77 pp.
[to3em]{}, *[Non-local, non-convex functionals converging to Sobolev norms]{}*, Nonlinear Anal., to appear.
L. C. Evans and R. F. Gariepy, *[Measure theory and fine properties of functions]{}*, Studies in Advanced Mathematics, CRC Press, Boca Raton, FL, 1992.
H.-M. Nguyen, *[$\Gamma$-convergence and Sobolev norms]{}*, C. R. Acad. Sci. Paris **345** (2007), 679–684.
[to3em]{}, *[$\Gamma$-convergence, Sobolev norms, and BV functions]{}*, Duke Math. J. **157** (2011), 495–533.
W. Ziemer, *[Weakly differentiable functions. Sobolev spaces and functions of bounded variation]{}*, Graduate Texts in Mathematics, vol. 120, Springer-Verlag, New York, 1989.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We are interested in the Logarithmic Sobolev Inequality for the infinite volume Gibbs measure with no quadratic interactions. We consider unbounded spin systems on the one dimensional Lattice with interactions that go beyond the usual strict convexity and without uniform bound on the second derivative. We assume that the one dimensional single-site measure with boundaries satisfies the Log-Sobolev inequality uniformly on the boundary conditions and we determine conditions under which the Log-Sobolev Inequality can be extended to the infinite volume Gibbs measure.'
author:
- 'Ioannis Papageorgiou [^1]'
---
**Keywords**: Logarithmic Sobolev inequality, Gibbs measure, Infinite dimensions, Spin systems.
**Mathematics Subject Classification (2000)**: 60E15, 26D10
Introduction
============
We are interested in the $q$ Logarithmic Sobolev Inequality (LSq) for measures related to systems of unbounded spins on the one dimensional Lattice with nearest neighbour interactions that are not strictly convex. Suppose that the Log-Sobolev Inequality is true for the single site measure with a constant uniformly bound on the boundary conditions. The aim of this paper is to present a criterion under which the inequality can be extended to the infinite volume Gibbs measure. More specifically, we extend the already know results for interactions $V$ that satisfy $\left\Vert \nabla_i \nabla_j V(x_i,x_j) \right\Vert_{\infty}<\infty$ to the more general case of interactions with $\left\Vert \nabla_i \nabla_j V(x_i,x_j) \right\Vert_{\infty}=\infty$.
Regarding the Log-Sobolev Inequality for the local specification $\{\mathbb{E}^{\Lambda,\omega}\}_{\Lambda\subset\subset \mathbb{Z}^d,\omega \in
\Omega}$ on a d-dimensional Lattice, criterions and examples of measures $\mathbb{E}^{\Lambda,\omega}$ that satisfy the Log-Sobolev -with a constant uniformly on the set $\Lambda$ and the boundary conditions $\omega-$ are investigated in \[Z2\], \[B-E\], \[B-L\], \[Y\] and \[B-H\]. For
$\left\Vert \nabla_i \nabla_j V(x_i,x_j) \right\Vert_{\infty}<\infty$ the Log-Sobolev is proved when the phase $\phi$ is strictly convex and convex at infinity. Furthermore, in \[G-R\] the Spectral Gap Inequality is proved to be true for phases beyond the convexity at infinity, while in \[M-M\] and \[B-J-S\] the Decay of Correlation is studied.
For the measure $\mathbb{E}^{\{i\},\omega}$ on the real line, necessary and sufficient conditions are presented in \[B-G\], \[B-Z\] and \[R-Z\], so that the Log-Sobolev Inequality is satisfied uniformly on the boundary conditions $\omega$.
The problem of the Log-Sobolev inequality for the Infinite dimensional Gibbs measure on the Lattice is examined in \[G-Z\], \[Z1\] and \[Z2\]. The first two study the LS for measures on a d-dimensional Lattice for bounded spin systems, while the third one looks at continuous spins systems on the one dimensional Lattice.
In \[M\] and \[O-R\], criterions are presented in order to pass from the Log-Sobolev Inequality for the single-site measure $\mathbb{E}^{\{i\},\omega}$ to the LS2 for the Gibbs measure $\nu_{N}$ on a finite N-dimensional product space. Furthermore, using these criterions one can conclude the Log-Sobolev Inequality for the family $\{\nu_N,N\in\mathbb{N}\}$ with a constant uniformly on $N$. Concerning the same problem for the LSq ($q\in (1,2]$) inequality in the case of Heisenberg groups with quadratic interactions in \[I-P\] a similar criterion is presented for the Gibbs measure based on the methods developed in \[Z1\] and \[Z2\].
All the pre mentioned developments refer to measures with interactions $V$ that satisfy $\left\Vert \nabla_i \nabla_j V(x_i,x_j) \right\Vert_{\infty}<\infty$. The question that arises is whether similar assertions can be verified for the infinite dimensional Gibbs measure in the case where $\left\Vert \nabla_i \nabla_j V(x_i,x_j) \right\Vert_{\infty}= \infty$ and in this paper we present a strategy to solve this problem.
Consider the one dimensional measure $$\mathbb{E}^{\{i\},\omega}(dx_{i})=\frac{e^{-
\phi(x_i)-\sum_{j\sim
i}J_{ij}V(x_i
,\omega_j)}dX_i} {Z^{\{i\},\omega}} \text{\; with \;} \left\Vert\partial_x \partial_y V(x,y) \right\Vert_{\infty}= \infty$$ Assume that $\mathbb{E}^{\{i\},\omega}$ satisfies the (LS) inequality with a constant uniformly on $\omega
$. Our aim is to set conditions, so that the infinite volume Gibbs measure $\nu$ for the local specification $\{\mathbb{E}^{\Lambda,\omega}\}_{\Lambda\subset\subset \mathbb{Z},\omega \in
\Omega}$ satisfies the LS inequality. We will focus on measures on the the one dimensional Lattice, but our result can also be easily extended on trees.
Our general setting is as follows:
*The Lattice.* When we refer to the Lattice we mean the 1-dimensional Lattice $\mathbb{Z}$.
*The Configuration space.* We consider continuous unbounded random variables in $\mathbb{R}$, representing spins. Our configuration space is $\Omega=\mathbb{R}^{\mathbb{Z}}$. For any $\omega\in \Omega $ and $\Lambda \subset \mathbb{Z}$ we denote $$\omega=(\omega_i)_{i\in \mathbb{Z}}, \omega_{\Lambda}=(\omega_i)_{i\in\Lambda},\omega_{\Lambda^c}=(\omega_i)_{i\in\Lambda^c} \text{\; and \;}\omega=\omega_{\Lambda}\circ\omega_{\Lambda^c}$$ where $\omega_i \in \mathbb{R}$. When $\Lambda=\{i\}$ we will write $\omega_i=\omega_{\{i\}}$. Furthermore, we will write $i\sim j$ when the nodes $i$ and $j$ are nearest neighbours, that means, they are connected with a vertex, while we will denote the set of the neighbours of $k$ as $\{\sim k\}=\{r:r \sim k\}$.
*The functions of the configuration.* We consider integrable functions $f$ that depend on a finite set of variables $\{x_i\}, i\in{\Sigma_f}$ for a finite subset $\Sigma_f\subset\subset \mathbb{Z}$. The symbol $\subset\subset$ is used to denote a finite subset.
*The Measure on $\mathbb{Z}$.* For any subset $\Lambda\subset\subset
\mathbb{Z}$ we define the probability measure $$\mathbb{E}^{\Lambda,\omega}(dx_\Lambda) =
\frac{e^{-H^{\Lambda,\omega}}dx_\Lambda} {Z^{\Lambda,\omega}}$$ where
- $x_{\Lambda}=(x_i)_{i\in\Lambda}$ and $dx_\Lambda=\prod_{i\in\Lambda}dx_i$
- $Z^{\Lambda,\omega}=\int e^{-H^{\Lambda,\omega}}dx_\Lambda$
- $H^{\Lambda,\omega}=\sum_{i\in\Lambda}
\phi (x_i)+\sum_{i\in\Lambda,j\sim
i}J_{ij}V(x_i
,z_j)$
and
- $z_j=x_{\Lambda}\circ\omega_{\Lambda^c}=\begin{cases}x_j & ,i\in\Lambda \\
\omega_j & ,i\notin\Lambda \\
\end{cases}$
We call $\phi$ the phase and $V$ the potential of the interaction. For convenience we will frequently omit the boundary symbol from the measure and will write $\mathbb{E}^{\Lambda}\equiv\mathbb{E}^{\Lambda,\omega}$.
*The Infinite Volume Gibbs Measure.* The Gibbs measure $\nu$ for the local specification $\{\mathbb{E}^{\Lambda,\omega}\}_{\Lambda\subset \mathbb{Z},\omega \in \Omega}$ is defined as the probability measure which solves the Dobrushin-Lanford-Ruelle (DLR) equation $$\nu \mathbb{E}^{\Lambda,\star}=\nu$$ for finite sets $\Lambda\subset \mathbb{Z}$ (see \[P\]). For conditions on the existence and uniqueness of the Gibbs measure see e.g. \[B-HK\] and \[D\]. In this paper we consider local specifications for which the Gibbs measure exists and it is unique. It should be noted that $\{\mathbb{E}^{\Lambda,\omega}\}_{\Lambda\subset\subset \mathbb{Z},\omega \in
\Omega}$ always satisfies the DLR equation, in the sense that $$\mathbb{E}^{\Lambda,\omega}\mathbb{E}^{M,\ast}=\mathbb{E}^{\Lambda,\omega}$$ for every $M\subset\Lambda$. \[P\].
*The gradient $\nabla$ for continuous spins systems.* For any subset $\Lambda\subset \mathbb{Z}$ we define the gradient $$\left\vert \nabla_{\Lambda} f
\right\vert^q=\sum_{i\in\Lambda}\left\vert \nabla_{i} f
\right\vert^q\text{,\; where \;}\nabla_{i}=\frac{\partial}{\partial x_i}$$ When $\Lambda=\mathbb{Z}$ we will simply write $\nabla=\nabla_\mathbb{Z}$. We denote $$\mathbb{E}^{\Lambda,\omega}f=\int f d\mathbb{E}^{\Lambda,\omega}(x_\Lambda)$$ We can define the following inequalities
*The $q$ Log-Sobolev Inequality (LS$_q$).* We say that the measure $\mathbb{E}^{\Lambda,\omega}$ satisfies the q Log-Sobolev Inequality for $q\in (1,2]$, if there exists a constant $C_{LS}$ such that for any function $f$, the following holds $$\mathbb{E}^{\Lambda,\omega}\left\vert f\right\vert^qlog\frac{\left\vert f\right\vert^q}{\mathbb{E}^{\Lambda,\omega}\left\vert f\right\vert^q}\leq C_{LS}
\mathbb{E}^{\Lambda,\omega}\left\vert \nabla_{\Lambda} f
\right\vert^q$$ with a constant $ C_{LS}\in(0,\infty)$ uniformly on the set $\Lambda$ and the boundary conditions $\omega$.
*The $q$ Spectral Gap Inequality.* We say that the measure $\mathbb{E}^{\Lambda,\omega}$ satisfies the q Spectral Gap Inequality for $q\in (1,2]$, if there exists a constant $C_{SG}$ such that for any function $f$, the following holds $$\mathbb{E}^{\Lambda,\omega}\left\vert f-\mathbb{E}^{\Lambda,\omega}f \right\vert^q\leq C_{SG}
\mathbb{E}^{\Lambda,\omega}\left\vert \nabla_{\Lambda} f
\right\vert^q$$ with a constant $C_{SG}\in(0,\infty)$ uniformly on the set $\Lambda$ and the boundary conditions $\omega$.
\[rem1.1\] We will frequently use the following two well known properties about the Log-Sobolev and the Spectral Gap Inequality. If the probability measure $\mu$ satisfies the Log-Sobolev Inequality with constant $c$ then it also satisfies the Spectral Gap Inequality with a constant $\hat c= \frac{4 c}{\log 2}$. More detailed, in the case where $q=2$ the optimal constant is less or equal to $\frac{c}{2}<\hat c$, while in the case $1<q<2$ it is less or equal to $\frac{4c}{\log 2}$. The constant $\hat c$ does not depend on the value of the parameter $q\in(1,2]$.
Furthermore, if for a family $I$ of sets $\Lambda_i \subset \mathbb{Z}$, $dist(\Lambda_i,\Lambda_j)>1 \ , i\neq j$ the measures $\mathbb{E}^{\Lambda_{i},\omega},
i\in I$ satisfy the Log-Sobolev Inequality with constants $c_i,i\in I$, then the probability measure $\mathbb{E}^{\{\cup_{i\in I}\Lambda_i\},\omega}$ also satisfies the (LS) Inequality with constant $c=max_{i\in I} c_i$. The last result is also true for the Spectral Gap Inequality. The proofs of these two properties can be found in \[G\] and \[G-Z\] for $q=2$ and in \[B-Z\] for $1<q<2$.
The Main Result
===============
We want to extend the Log-Sobolev Inequality from the single-site measure $\mathbb{E}^{\{i\},\omega}$ to the Gibbs measure for the local specification $\{\mathbb{E}^{\Lambda,\omega}\}_{\Lambda\subset\subset \mathbb{Z},\omega \in
\Omega}$ on the entire one dimensional Lattice.
**Hypothesis** We consider four main hypothesis:
**(H0)**: The one dimensional measures $\mathbb{E}^{i,\omega}$ satisfies the Log-Sobolev-q Inequality with a constant $c$ uniformly with respect to the boundary conditions $\omega$.
**(H1)**: The restriction $\nu_{\Lambda(k)}$ of the Gibbs measure $\nu$ to the $\sigma-$algebra $\Sigma_{\Lambda(k)}$, $$\Lambda(k)=\{k-2,k-1,k,k+1,k+2\}$$ satisfies the Log-Sobolev-q Inequality with a constant $C\in (0,\infty)$.
**(H2)**: For some $\epsilon>0$ and $K>0$ $$\nu_{\Lambda(i)} e^{2^{q+2}\epsilon
V(x_r,x_s)}\leq e^K\text{\; and \;} \nu_{\Lambda(i)} e^{2^{q+2}\epsilon \left\vert\nabla_{r}V(x_{r},x_{s})\right\vert^q}\leq
e^K$$ for $r,s\in\{i-2,i-1,i,i+1,i+2\}$
**(H3)**: The coefficients $J_{i,j}$ are such that $\left\vert J_{i,j}\right\vert\in[0,J]$ for some $J<1$ sufficiently
small.
\[rem2.1n\] From Hypothesis $(H2)$ and Jensen’s inequality it follows that $$\nu e^{\epsilon(\left\vert F(r)\right\vert+\mathbb{E}^{S(r),\omega}\left\vert F(r)\right\vert)^q}\leq
e^K,\text{\; for \;}r=i-2,i-1,i,i+1,i+2$$where the functions $F(r)$ are defined by $$F(r)=\begin{cases}\nabla_{r}V(x_{i-1},x_{i})+\nabla_{r}V(x_{i+1},x_{i})& \text{\; for \;}r=i-1,i,i+1 \\
\nabla_{r}V(x_s,x_r)\mathcal{I}_{s\sim r:s\in\{i-3,i+3\}} & \text{\; for \;}r=i-2,i+2 \\
\end{cases}$$ and the sets $S(r)$ by $$S(r)=\begin{cases}\{\sim i\} & \text{\; for \;}r=i-1,i,i+1 \\
\{i+3,i+4,...\} & \text{\; for \;}r=i+2 \text{\;and\;}s=i+3\\
\{...,i-4,i-3\} & \text{\; for \;}r=i-2 \text{\;and\;}s=i-3 \\
\end{cases}$$ These bounds will be frequently used through out the paper.
\[newremark1\] Throughout this paper we will consider differentiable functions that satisfy $$\nu\left\vert f\right\vert^q <\infty \text{\; and \;} \nu \left\vert \nabla f\right\vert^q <\infty$$
The main theorem follows.
\[thm2.1\] If hypothesis (H0)-(H3) are satisfied, then the infinite dimensional Gibbs measure $\nu$ for the local specification $\{\mathbb{E}^{\Lambda,\omega}\}_{\Lambda\subset\subset \mathbb{Z},\omega \in
\Omega}$ satisfies the $q$ Log-Sobolev inequality $$\nu \left\vert f\right\vert^q log\frac{\left\vert f\right\vert^q}{\nu \left\vert f\right\vert^q}\leq \mathfrak{C} \ \nu \left\vert \nabla f
\right\vert^q$$ for some positive constant $\mathfrak{C}$.
For the proof of the theorem it is sufficient to consider $f\geq 0$. This is an assumption that we will make through all the proofs presented in this paper. We want to extend the Log-Sobolev Inequality from the single-site measure $\mathbb{E}^{\{i\},\omega}$ to the Gibbs measure for the local specification $\{\mathbb{E}^{\Lambda,\omega}\}_{\Lambda\subset\subset \mathbb{Z},\omega \in
\Omega}$ on the entire one dimensional lattice. To do so, we will follow the iterative method developed by Zegarlinski in \[Z1\] and \[Z2\]. Define the following sets $$\Gamma_0= \text{\;even integers, \;} \Gamma_1=\mathbb{Z}\smallsetminus\Gamma_0$$ One can notice that $\{dist(i,j)>1, \ \forall i,j \in\Gamma_k,k=0,1\}$, $\Gamma_0\cap\Gamma_1=\emptyset$ and $\mathbb{Z}=\Gamma_0\cup\Gamma_1$. For convenience we will write $\mathbb{E}^{\Gamma_i}=\mathbb{E}^{\Gamma_i,\omega}$ for $i=0,2$.We will denote $$\mathcal{P}=\mathbb{E}^{\Gamma_1}\mathbb{E}^{\Gamma_{0}}$$ In order to prove the Log-Sobolev Inequality for the measure $\nu$, we will express the entropy with respect to the measure $\nu$ as the sum of the entropies of the measures $\mathbb{E}^{\Gamma_0}$ and $\mathbb{E}^{\Gamma_1}$ which are easier to handle. We can write $$\begin{aligned}
\nonumber \nu (f^q log\frac{f^q}{\nu f^q})=&\nu\mathbb{E}^{\Gamma_0} (f^q log\frac{f^q}{\mathbb{E}^{\Gamma_0} f^q})+\nu\mathbb{E}^{\Gamma_{1}} (\mathbb{E}^{\Gamma_0}f^q log\frac{\mathbb{E}^{\Gamma_0}f^q}{\mathbb{E}^{\Gamma_{1}}\mathbb{E}^{\Gamma_0} f^q})\\ & \label{2.1}+
\nu (\mathbb{E}^{\Gamma_{1}}\mathbb{E}^{\Gamma_{0}}f^q log\mathbb{E}^{\Gamma_{1}}\mathbb{E}^{\Gamma_{0}}f^q)-\nu
(f^q log \nu f^{q})\end{aligned}$$ According to hypothesis (H0), the Log-Sobolev Inequality is satisfied for the single-state measures $\mathbb{E}^{\{j\}}$ and the sets $\Gamma_0$ and $\Gamma_1$ are unions of one dimensional sets of distance greater than the length of the interaction one. Thus, as we mentioned in Remark \[rem1.1\] in the introduction, the (LS) holds for the product measures $\mathbb{E}^{\Gamma_0}$ and $\mathbb{E}^{\Gamma_1}$ with the same constant c. If we use the LS for $\mathbb{E}^{\Gamma_i},i=0,1$ we get $$\begin{aligned}
\nonumber ~\eqref{2.1}\leq & c\nu(\mathbb{E}^{\Gamma_0}\left\vert \nabla_{\Gamma_0} f
\right\vert^q)+c\nu \mathbb{E}^{\Gamma_1}\left\vert \nabla_{\Gamma_1}(\mathbb{E}^{\Gamma_0} f^q)^{\frac{1}{q}}
\right\vert^q\\ &\label{2.2}+\nu (\mathbb{E}^{\Gamma_{1}}\mathbb{E}^{\Gamma_{0}}f^q log\mathbb{E}^{\Gamma_{1}}\mathbb{E}^{\Gamma_{0}}f^q)-\nu (f^q log \nu f^q)\end{aligned}$$ For the third term of we can write $$\begin{aligned}
\nonumber\nu (\mathcal{P}f^q log \mathcal{P} f^q)=&\nu \mathbb{E}^{\Gamma_{0}}(\mathcal{P}f^q log \frac{\mathcal{P} f^q}{\mathbb{E}^{\Gamma_{0}}\mathcal{P} f^q})+\nu \mathbb{E}^{\Gamma_{1}}(\mathbb{E}^{\Gamma_{0}}\mathcal{P}f^q log \frac{\mathbb{E}^{\Gamma_{0}}\mathcal{P} f^q}{\mathbb{E}^{\Gamma_{1}}\mathbb{E}^{\Gamma_{0}}\mathcal{P} f^q})\\ & \nonumber
+\nu( \mathbb{E}^{\Gamma_{1}}\mathbb{E}^{\Gamma_{0}}\mathcal{P}f^q log \mathbb{E}^{\Gamma_{1}}\mathbb{E}^{\Gamma_{0}}\mathcal{P}f^q )\end{aligned}$$ If we use again the Log-Sobolev Inequality for the measures $\mathbb{E}^{\Gamma_{i}},i=0,1$ we get $$\label{2.3}\nu (\mathcal{P}f^qlog \mathcal{P} f^q)\leq c \nu\left\vert \nabla_{\Gamma_0}(\mathcal{P}f^q)^\frac{1}{q}
\right\vert^q+c \nu\left\vert \nabla_{\Gamma_1}(\mathbb{E}^{\Gamma_{0}} \mathcal{P}f^q)^\frac{1}{q}
\right\vert^q+\nu (\mathcal{P}^2f^q log \mathcal{P}^2f^q)$$ If we work similarly for the last term $\nu (\mathcal{P}^2f^q log \mathcal{P}^2f^q)$ of and inductively for any term $\nu (\mathcal{P}^kf^q log \mathcal{P}^kf^q)$, then after $n$ steps and will give $$\begin{aligned}
\nonumber\nu (f^q log\frac{f^q}{\nu f^q})\leq &\nu (\mathcal{P}^n f^q \log\mathcal{P}^n f^q)-\nu (f^qlog \nu f^q)+c\nu \left\vert \nabla_{\Gamma_0}f\right\vert^q\\ &
\label{2.4}+c \sum_{k=1}^{n-1} \nu \left\vert \nabla_{\Gamma_0}( \mathcal{P}^kf^q)^\frac{1}{q}
\right\vert^q+c \sum_{k=0}^{n-1} \nu\left\vert \nabla_{\Gamma_1}(\mathbb{E}^{\Gamma_{0}} \mathcal{P}^kf^q)^\frac{1}{q}
\right\vert^q \end{aligned}$$ In order to calculate the fourth and fifth term on the right-hand side of we will use the following proposition
\[prp2.2\]Suppose that hypothesis (H0)-(H3) are satisfied. Then the following bound holds $$\label{2.5}\nu\left\vert \nabla_{\Gamma_i}(\mathbb{E}^{\Gamma_{j}} \vert f\vert^q)^\frac{1}{q}
\right\vert^q\leq C_1\nu\left\vert \nabla_{\Gamma_i}f\right\vert^q+C_2\nu\left\vert \nabla_{\Gamma_j}f\right\vert^q$$ for $\{i,j\}=\{0,1\}$ and constants $C_1\in (0,\infty)$ and $0<C_2<1$.
The proof of Proposition \[prp2.2\] will be the subject of Section 4. If we apply inductively relationship k times to the fourth and the fifth term of we obtain $$\label{2.6} \nu\left\vert \nabla_{\Gamma_0}(\mathcal{P}^kf^q)^\frac{1}{q}
\right\vert^q \leq C_2^{2k-1}C_1\nu\left\vert \nabla_{\Gamma_1} f
\right\vert^q+C_2^{2k}\nu\left\vert \nabla_{\Gamma_0} f
\right\vert^q$$ and $$\label{2.7}\nu\left\vert \nabla_{\Gamma_1}(\mathbb{E}^{\Gamma_{0}}\mathcal{P}^kf^q)^\frac{1}{q}
\right\vert^q \leq C_2^{2k}C_1\nu\left\vert \nabla_{\Gamma_1} f
\right\vert^q+C_2^{2k+1}\nu\left\vert \nabla_{\Gamma_0} f
\right\vert^q$$ If we plug and in we get $$\begin{aligned}
\nonumber\nu (f^qlog\frac{f^q}{\nu f^q})\leq&\nu (\mathcal{P}^n f^q log \mathcal{P}^n f^q)-\nu (f^qlog \nu f^q)\\ \nonumber
&+c(\sum_{k=0}^{n-1}C_2^{2k-1})C_1\nu\left\vert \nabla_{\Gamma_1} f
\right\vert^q+c(\sum_{k=0}^{n-1}C_2^{2k})\nu\left\vert \nabla_{\Gamma_0} f
\right\vert^q\\ &
\label{2.8}+c(\sum_{k=0}^{n-1}C_2^{2k})C_1\nu\left\vert \nabla_{\Gamma_1} f
\right\vert^q+c(\sum_{k=0}^{n-1}C_2^{2k+1})\nu\left\vert \nabla_{\Gamma_0} f
\right\vert^q\end{aligned}$$ If we take the limit of $n$ to infinity in the first two terms on the right hand side cancel with each other, as explained on the proposition bellow.
\[prp2.3\] Under hypothesis (H0)-(H3), $\mathcal{P}^nf$ converges $\nu$-almost everywhere to $\nu f$.
The proof of this proposition will be presented in Section 3. So, taking the limit of $n$ to infinity in leads to $$\nu (\vert f\vert^qlog\frac{\vert f\vert^q}{\nu \vert f\vert^q})\leq cA\left(\frac{C_1}{C_2}+C_2+C_1\right)\nu\left\vert \nabla_{\Gamma_1} f
\right\vert^q+cA\nu\left\vert \nabla_{\Gamma_0} f
\right\vert^q$$ where $A=lim_{n\rightarrow\infty}\sum_{k=0}^{n-1}C_2^{2k}<\infty$ for $C_2<1$, and the theorem follows for a constant $C=max\{cA\left(\frac{C_1}{C_2}+C_2+C_1\right),cA\}$
Proof of Proposition \[prp2.3\].
=================================
Before proving Proposition \[prp2.3\] we will present three useful lemmata. These lemmata will also be used in the next section 4 where Proposition \[prp2.2\] is proved.
In the case of quadratic interactions $V(x,y)=(x-y)^2$ one can calculate $$\mathbb{E}^{i,\omega}\left( f^2(\nabla_j V(x_i-x_j)-\mathbb{E}^{i,\omega}\nabla_j V(x_i-x_j))^2 \right)$$ (see \[B-H\] and \[H\]) with the use of the Deuschel-Stroock relative entropy inequality (see \[D-S\]) and the Herbst argument (see \[L\] and \[H\]). Herbst’s arguement states that if a probability measure $\mu$ satisfies the LS2 inequality and a function $F$ is Lipschitz continues with $\Vert F \Vert_{Lips}\leq 1$ and such that $\mu (F)=0$, then for some small $\epsilon$ we have $$\mu e^{\epsilon F^2}<\infty$$ For $\mu=\mathbb{E}^{i,\omega}$ and $F=\frac{\nabla_j V(x_i-x_j)-\mathbb{E}^{i,\omega}\nabla_j V(x_i-x_j)}{2}$ we then obtain $$\mathbb{E}^{i,\omega} e^{\frac{\epsilon}{4} (\nabla_j V(x_i-x_j)-\mathbb{E}^{i,\omega}\nabla_j V(x_i-x_j))^2}<\infty$$uniformly on the boundary conditions $\omega$, because of hypothesis $(H0)$. In the more general case however of non quadratic interactions that we examine in this work, the Herbst argument cannot be applied. In this and next sections we show how one can bound exponential quantities like the last one with the use of the projection of the infinite dimensional Gibbs measure and hypothesis (H1) and (H2).
For every probability measure $\mu$, we define the correlation function $$\mu(f;g)\equiv\mu(fg)-\mu(f)\mu(g)$$If for the set $M(k)=\mathbb{Z}\smallsetminus\Lambda(k)$ and $h_k:=f-\mathbb{E}^{\{ \sim k\}}f$ we define$$Q(u,k)\equiv\nu_{\Lambda(u)}\left\vert \nabla_{\Lambda(u)}\left( \mathbb{E}^{M(u)}
\vert h_k\vert ^q \right)^{\frac{1}{q}}\right\vert^q$$ then the following lemma presents an estimate for the correlation function, in terms of $Q(k,k)$.
\[lem3.1\] For any functions $u$ localised in $\Lambda(k)$ for which $\nu_{\Lambda(k)} e^{2^{q}\epsilon \vert u\vert^q}<\infty$ the following inequalities are satisfied
\(a) under hypothesis (H1)$$\begin{aligned}
\nu \left\vert\mathbb{E}^{k-1}\mathbb{E}^{k+1}(f;u)\right\vert^q\leq&\frac{C}{\epsilon }Q(k,k)+ \frac{1}{\epsilon }\left(log\nu_{\Lambda(k)} e^{\epsilon \vert u-\mathbb{E}^{k-1}\mathbb{E}^{k+1}u\vert^q}\right)\nu\left\vert f-\mathbb{E}^{k-1}\mathbb{E}^{k+1}f\right \vert^{q}\end{aligned}$$
\(b) under hypothesis (H0) and (H1)$$\nu \left\vert\mathbb{E}^{k-1}\mathbb{E}^{k+1}(f;u)\right\vert^q\leq\frac{C}{\epsilon }Q(k,k)+\frac{\hat c}{\epsilon }\left(log\nu_{\Lambda(k)} e^{\epsilon \vert u-\mathbb{E}^{k-1}\mathbb{E}^{k+1}u\vert^q}\right)\sum_{i=k-1,k+1}\nu\left\vert \nabla_ i f\right\vert^q$$ where $\hat c =\frac{4 c}{\log 2}$.
From the definition of the correlation function we can write $$\begin{aligned}
\nonumber\nu\left\vert\mathbb{E}^{k-1}\mathbb{E}^{k+1}
(f;u)\right\vert^q= & \nu\left\vert\mathbb{E}^{k-1}\mathbb{E}^{k+1}((f-\mathbb{E}^{k-1}\mathbb{E}^{k+1}f)(u-\mathbb{E}^{k-1}\mathbb{E}^{k+1}u))\right\vert^q
\\ \leq & \nu\mathbb{E}^{k-1}\mathbb{E}^{k+1}\left(\vert f-\mathbb{E}^{k-1}\mathbb{E}^{k+1}
f\vert^q\vert u-\mathbb{E}^{k-1}\mathbb{E}^{k+1}u\vert^q\right) \nonumber\\ = & \label{3.1}\nu\left(\vert f-\mathbb{E}^{k-1}\mathbb{E}^{k+1}
f\vert^q\vert u-\mathbb{E}^{k-1}\mathbb{E}^{k+1}u\vert^q\right)\end{aligned}$$ where above we first used the Jensen’s Inequality and then the fact that the Gibbs measure $\nu$ satisfies the DLR equation. Because the function $u$ is localised in $\Lambda(k)$ and the measure $\mathbb{E}^{\{k-1,k+1\},\omega}=\mathbb{E}^{k-1}\mathbb{E}^{k+1}$ has boundary in $\{k-2,k,k+2\}\subset\Lambda(k)$, we have that $u-\mathbb{E}^{k-1}\mathbb{E}^{k+1}u$ is also localised in $\Lambda(k)$ and so for $M(k)$ being the complementary of $\Lambda(k)$ we can write $$\begin{aligned}
\nonumber\nu(\vert f-\mathbb{E}^{k-1}\mathbb{E}^{k+1}
f\vert^{q}&\vert u-\mathbb{E}^{k-1}\mathbb{E}^{k+1}u\vert^q)=&\\ &\label{3.2}\nu_{\Lambda(k)}\left(\left(\mathbb{E}^{M(k)}\vert f-\mathbb{E}^{k-1}\mathbb{E}^{k+1}f\vert^{q}\right)\vert u-\mathbb{E}^{k-1}\mathbb{E}^{k+1}u\vert^q\right)\end{aligned}$$ On the right hand side of we can use the following entropic inequality (see \[D-S\]) $$\label{3.3}\forall t>0, \ \mu(uy)\leq\frac{1}{t}log\left(\mu(e^{tu})\right)+\frac{1}{t}\mu(y\log y)$$ for any probability measure $\mu$ and $y\geq 0$, $\mu y=1$. Then from and we will obtain $$\begin{aligned}
\nonumber \nu&\left\vert\mathbb{E}^{k-1}\mathbb{E}^{k+1}
(f;u)\right\vert^q\leq& \\ & \frac{1}{\epsilon }\nu_{\Lambda(k)}\mathbb{E}^{M(k)}
\vert f-\mathbb{E}^{k-1}\mathbb{E}^{k+1}
f\vert^{q}\log\frac{\mathbb{E}^{M(k)}
\vert f-\mathbb{E}^{k-1}\mathbb{E}^{k+1}
f\vert^{q}}{\nu_{\Lambda(k)}\mathbb{E}^{M(k)}
\vert f-\mathbb{E}^{k-1}\mathbb{E}^{k+1}
f\vert^{q}}\nonumber \\ &\label{3.4}+\frac{1}{\epsilon}\left(log\nu_{\Lambda(k)}e^{\epsilon
\vert u-\mathbb{E}^{k-1}\mathbb{E}^{k+1}
u\vert^q}\right)\nu_{\Lambda(k)}\mathbb{E}^{M(k)}\vert f-\mathbb{E}^{k-1}\mathbb{E}^{k+1}f\vert^{q}\end{aligned}$$ The first term on the right hand side of can be bounded from hypothesis (H1) by the Log-Sobolev inequality for $\nu_{\Lambda(k)}$ $$\begin{aligned}
\nonumber \nu_{\Lambda(k)}\mathbb{E}^{M(k)}
\vert f-\mathbb{E}^{k-1}\mathbb{E}^{k+1}
f\vert^{q}&\log\frac{\mathbb{E}^{M(k)}
\vert f-\mathbb{E}^{k-1}\mathbb{E}^{k+1}
f\vert^{q}}{\nu_{\Lambda(k)}\mathbb{E}^{M(k)}
\vert f-\mathbb{E}^{k-1}\mathbb{E}^{k+1}
f\vert^{q}}\\ \leq & \label{3.5}C\nu_{\Lambda(k)}\left\vert \nabla_{\Lambda(k)}(\mathbb{E}^{M(k)}
\vert f-\mathbb{E}^{k-1}\mathbb{E}^{k+1}
f\vert^{q})^{\frac{1}{q}}\right \vert ^q=CQ(k,k)\end{aligned}$$ Using and we get $$\begin{aligned}
\label{3.5+1equation}
\nu \left\vert\mathbb{E}^{k-1}\mathbb{E}^{k+1}(f;u)\right\vert^q\leq\frac{C}{\epsilon }Q(k,k)+\frac{1}{\epsilon }\left(log\nu e^{\epsilon \vert u-\mathbb{E}^{k-1}\mathbb{E}^{k+1}u\vert^q}\right)\nu\left\vert f-\mathbb{E}^{k-1}\mathbb{E}^{k+1}f\right \vert^{q}\end{aligned}$$which proves (a). If we assume hypothesis (H0), then we can bound the second term on the right hand side of (\[3.5+1equation\]) from the $SG_q$ for the measures $\mathbb{E}^{k-1},\mathbb{E}^{k+1}$ from hypothesis (H0) and the product property for the $SG_q$ (Remark \[rem1.1\]), to obtain $$\begin{aligned}
\nu\vert f-\mathbb{E}^{k-1}\mathbb{E}^{k+1}f\vert^{q}=&\nu\mathbb{E}^{k-1}\mathbb{E}^{k+1}\vert f-\mathbb{E}^{k-1}\mathbb{E}^{k+1}f\vert^{q}
\leq &\label{3.6}\hat c\sum_{i=k-1,k+1}\nu\left\vert \nabla_ i f\right\vert^q\end{aligned}$$ where $\hat c =\frac{4 c}{\log 2}$. Using (\[3.5+1equation\]) and we finally get (b)$$\nu \left\vert\mathbb{E}^{k-1}\mathbb{E}^{k+1}(f;u)\right\vert^q\leq\frac{C}{\epsilon }Q(k,k)+\frac{\hat c}{\epsilon }\left(log\nu e^{\epsilon \vert u-\mathbb{E}^{k-1}\mathbb{E}^{k+1}u\vert^q}\right)\sum_{i=k-1,k+1}\nu\left\vert \nabla_ i f\right\vert^q$$
The following lemma gives an explicit bound for the quantity $Q(k,k)$.
\[lem3.2\]Suppose that hypothesis (H0)-(H3) are satisfied. Then $$\begin{aligned}
Q(k,k)\leq&D\sum_{r=k-2}^{k+2}\nu\left\vert \nabla_{r}f
\right\vert^q\\ & +D\sum_{ n=0 }^{\infty} J^{(n+1)(q-1)}\sum_{r=0}^3\left(\nu\left\vert \nabla_{k+3+4n+r} f
\right\vert^q+\nu\left\vert \nabla_{k-3-4n-r} f
\right\vert^q\right)\end{aligned}$$for some positive constant $D$.
The proof of this lemma will be the subject of Section 5.
\[lem3.3\]Suppose that hypothesis (H0)-(H3) are satisfied. Then for $\{i,j\}=\{0,1\}$ $$\nu\left\vert \nabla_{\Gamma_i}(\mathbb{E}^{\Gamma_{j}}f)
\right\vert^q \leq D_1\mathcal{\nu}\left\vert \nabla_{\Gamma_i} f
\right\vert^q+D_2\nu\left\vert \nabla_{\Gamma_j} f
\right\vert^q$$ holds for constants $D_1\in (0,\infty)$ and $0<D_2<1$.
Assume $i=1,j=0$. We have $$\label{3.7}\nu\left\vert \nabla_{\Gamma_1}(\mathbb{E}^{\Gamma_{0}}f)
\right\vert^q=\sum_{i\in \Gamma_1} \nu\left\vert \nabla_{i}(\mathbb{E}^{\Gamma_{0}}f)
\right\vert^q\leq\sum_{i\in \Gamma_1} \nu\left\vert \nabla_{i}(\mathbb{E}^{i-1}\mathbb{E}^{i+1}f)
\right\vert^q$$ If we denote $\rho_i= \frac{e^{-H(x_{i-1})}e^{-H(x_{i+1})}}{\int e^{-H(x_{i-1})}dx_i\int e^{-H(x_{i+1})}dx_i}$ the density of the measure $\mathbb{E}^{i-1}\mathbb{E}^{i+1}$ we can then write $$\begin{aligned}
\nonumber\nu&\left\vert\nabla_{i}(\mathbb{E}^{i-1}\mathbb{E}^{i+1}f)\right\vert^{q}=
\nu\left\vert\nabla_{i}(\int \int \rho_{i} f dx_{i-1}dx_{i+1})\right\vert^{q}\leq
&\\ & \label{3.8}2^{q-1}\nu\left\vert\int \int (\nabla_{i}f) \rho_{i} dx_{i-1}dx_{i+1}\right\vert^{q}+\ 2^{q-1}\nu\left\vert\int \int f(\nabla_{i}\rho_{i} )dx_{i-1}dx_{i+1}\right\vert^{q}\leq
\\ & \label{3.9}c_{1}\nu\left\vert\mathbb{E}^{i-1}\mathbb{E}^{i+1}(\nabla_{i}f)\right\vert^{q}+\ c_{1}J^q\nu\left\vert\mathbb{E}^{i-1}\mathbb{E}^{i+1}(f; \nabla_{i}V(x_{i-1},x_{i})+\nabla_{i}V(x_{i+1},x_{i}))\right\vert^{q}\end{aligned}$$ where in we used hypothesis (H3) to bound the coefficients $J_{i,j}$ and we have denoted $c_1=2^{4q}$. If we apply the Hölder Inequality to the first term of and Lemma \[lem3.1\] (b) to the second term, we obtain $$\label{3.10}\nu\left\vert\nabla_{i}(\mathbb{E}^{i-1}\mathbb{E}^{i+1}f)\right\vert^q
\leq c_{1}\nu\left\vert\nabla_{i}f\right\vert^q+\frac{J^{q}c_1C}{\epsilon }Q(i,i)+\frac{J^{q}\hat cc_{1}K}{\epsilon }\sum_{k=i-1,i+1}\nu\left\vert \nabla_ k f
\right\vert^q$$ where the constant $K$ as in hypothesis $(H2)$. From and we have $$\begin{aligned}
\nu \left\vert \nabla_{\Gamma_1}(\mathbb{E}^{\Gamma_0}f)
\right\vert^q\leq c_1\nu\left\vert\nabla_{ \Gamma_1}f\right\vert^q+\frac{J^q c_1C}{\epsilon}\sum_{i\in \Gamma_1}Q(i,i)+
\frac{J^q\hat cc_1K}{\epsilon }\sum_{i\in \Gamma_1}\sum_{k=i-1,i+1}\nu\left\vert \nabla_ k
f
\right\vert^q\end{aligned}$$ If we use Lemma \[lem3.2\] to replace $Q(k,k)$ in the above expression we get $$\begin{aligned}
\nonumber\nu \left\vert \nabla_{\Gamma_1}(\mathbb{E}^{\Gamma_0}f)\right\vert^q&\leq c_1\nu\left\vert\nabla_{ \Gamma_1}f\right\vert^q+\frac{J^q\hat c2c_1K}{\epsilon}\nu\left\vert\nabla_{ \Gamma_0}f\right\vert^q+\frac{J^q c_1DC}{\epsilon }\sum_{i\in \Gamma_1}\sum_{r=i-2}^{i+2}\nu\left\vert \nabla_r f\right\vert^q+ \\ & \frac{J^q c_1DC}{\epsilon }\sum_{i\in \Gamma_1}\sum_{n=0 }^{\infty} J^{(n+1)(q-1)}\sum_{r=0}^3\left(\nu\left\vert \nabla_{i+3+4n+r} f
\right\vert^q+\nu\left\vert \nabla_{i-3-4n-r} f
\right\vert^q\right)\end{aligned}$$ for constant $D>0$ as in Lemma \[lem3.2\]. For coefficients $J_{i,j}$ sufficiently small such that $J<1$ in (H3) we finally obtain
$\nu \left\vert \nabla_{\Gamma_1}(\mathbb{E}^{\Gamma_0}f)
\right\vert^q\leq J^{q}\left(\frac{2\hat cc_{1}K}{\epsilon}+\frac{2c_{1}CD}{\epsilon }+2D\frac{c_{1}C}{\epsilon }\frac{J^{(q-1)}}{1-J^{(q-1)}}\right)\nu\left\vert \nabla_ { \Gamma_0}
f
\right\vert^q$
$ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ +\left( c_1q+\frac{J^{q}c_{1}C}{\epsilon }3D+D\frac{2J^{q}c_{1}C}{\epsilon }\frac{J^{(q-1)}}{1-J^{(q-1)}}\right)\nu\left\vert\nabla_{ \Gamma_1}f\right\vert^q$
and the lemma follows for $J$ sufficiently small such that $$D_2=J^{q}\left(\frac{\hat cc_{1}K}{\epsilon}2+\frac{2c_1CD}{\epsilon }+D\frac{2c_1C}{\epsilon }\frac{J^{(q-1)}}{1-J^{(q-1)}}\right)<1$$
Now we can prove Proposition \[prp2.3\].
***Proof of Proposition \[prp2.3\]**.* Following \[G-Z\] we will show that in $L_1(\nu)$ we have $lim_{n\rightarrow \infty}\mathcal{P}^n=\nu$. For $i\not=j$ we have that $$\begin{aligned}
\nonumber\nu\vert\mathbb{E}^{\Gamma_{j}} f- \mathbb{E}^{\Gamma_{i}}\mathbb{E}^{\Gamma_{j}} f\vert^q&=\nu\mathbb{E}^{\Gamma_{i}}\vert\mathbb{E}^{\Gamma_{j}} f- \mathbb{E}^{\Gamma_{i}}\mathbb{E}^{\Gamma_{j}} f\vert^q\\ &
\label{3.11}\leq \hat c\nu\left\vert \nabla_{\Gamma_i}(\mathbb{E}^{\Gamma_{j}} f
)\right\vert^q\end{aligned}$$ The last inequality due to the fact that both the measures $\mathbb{E}^{\Gamma_{0}}$ and $\mathbb{E}^{\Gamma_{1}}$ satisfy the Log-Sobolev Inequality and the Spectral Gap inequality with constants independently of the boundary conditions. If we use Lemma \[lem3.3\] we get $$\nu\vert\mathbb{E}^{\Gamma_{j}} f- \mathbb{E}^{\Gamma_{i}}\mathbb{E}^{\Gamma_{j}} f\vert^q\leq \hat cD_1\nu \vert \nabla_{\Gamma_i}f\vert^q+\hat cD_2\nu \vert \nabla_{\Gamma_j}f\vert^q$$ From the last inequality we obtain that for any $n\in \mathbb{N}$, $$\begin{aligned}
\nu\vert \mathcal{P}^{n}f- \mathbb{E}^{\Gamma_0}\mathcal{P}^{n} f\vert^q &\leq \hat cD_1\nu \vert \nabla_{\Gamma_0}(\mathbb{E}^{\Gamma_0}\mathcal{P}^{n-1} f)\vert^q+\hat cD_2\nu \vert \nabla_{\Gamma_1}(\mathbb{E}^{\Gamma_0}\mathcal{P}^{n-1} f)\vert^q\\&=\hat cD_2\nu \vert \nabla_{\Gamma_1}(\mathbb{E}^{\Gamma_0}\mathcal{P}^{n-1} f)\vert^q\end{aligned}$$ If we use Lemma \[lem3.3\] to bound the last expression we have the following $$\label{BorelCant1} \nu\vert \mathcal{P}^{n}f- \mathbb{E}^{\Gamma_0}\mathcal{P}^{n} f\vert^q \leq \hat cD_2^{n}\left(D_1\nu\left\vert \nabla_{\Gamma_1} f
\right\vert^q+D_2\nu\left\vert \nabla_{\Gamma_0} f
\right\vert^q\right)$$ Similarly we obtain $$\label{BorelCant1+2} \nu\vert\mathbb{E}^{\Gamma_0} \mathcal{P}^{n}f- \mathcal{P}^{n+1} f\vert^q\leq \hat cD_2^{n}\left(D_1\nu\left\vert \nabla_{\Gamma_1} f
\right\vert^q+D_2\nu\left\vert \nabla_{\Gamma_0} f
\right\vert^q\right)$$
Consider the sequence $\{\mathcal{Q}^n\}_{n\in \mathbb{N}}$ defined as $$\mathcal{Q}^{n}f=\begin{cases}\mathcal{P}^{\frac{n}{2}}f & \text{ if \ } n \text{ \ even} \\
\mathbb{E}^{\Gamma_0}\mathcal{P}^{\frac{n-1}{2}}f & \text{ if \ } n \text{ \ odd} \\
\end{cases}$$ for every $n\in\mathbb{N}$. Hence, if we define the sets $$A_n=\{\vert \mathcal{Q}^nf- \mathcal{Q}^{n+1} f\vert \geq (\frac{1}{2})^n\}$$ we obtain $$\nu (A_n)=\nu\left( \{\vert \mathcal{Q}^nf- \mathcal{Q}^{n+1} f\vert \geq (\frac{1}{2})^n\} \right)\leq 2^{qn}\nu\vert \mathcal{Q}^nf- \mathcal{Q}^{n+1} f\vert^q$$by Chebyshev inequality. If we use (\[BorelCant1\]) and (\[BorelCant1+2\]) to bound the last we have $$\nu (A_n)\leq (2^{q}D_2^\frac{1}{2} )^{n}\hat c\left(D_1\nu\left\vert \nabla_{\Gamma_1} f
\right\vert^q+D_2\nu\left\vert \nabla_{\Gamma_0} f
\right\vert^q\right)$$ We can choose $J$ sufficiently small such that $2^{q}D_2^\frac{1}{2}<\frac{1}{2}$ in which case we get that $$\sum_{n=0}^\infty\nu (A_n)\leq\left(\sum_{n=0}^\infty(\frac{1}{2})^{n} \right)\hat c\left(D_1\nu\left\vert \nabla_{\Gamma_1} f
\right\vert^q+D_2\nu\left\vert \nabla_{\Gamma_0} f
\right\vert^q\right)<\infty$$ From the Borel-Cantelli lemma, only finite number of the sets $A_n$ can occur, which implies that the sequence $$\{\mathcal{Q}^{n} f\}_{n \in \mathbb{N}}$$ is a Cauchy sequence and that it converges $\nu-$almost surely. Say $$\mathcal{Q}^nf\rightarrow \theta (f) \ \ \ \ \ \nu-\text{a.e.}$$ We will first show that $\theta (f)$ is a constant, i.e. it does not depend on variables on $\Gamma_0$ or $\Gamma_1$. To show that, first notice that $\mathcal{Q}^n(f)$ is a function on $\Gamma_1$ and $\Gamma_0$ when $n$ is odd and even respectively, which implies that the limits $$\theta_{o}(f)=\lim_{n \text{\ odd}, n\rightarrow \infty}\mathcal{Q}^nf \text{ \ and \ }\theta_e(f)=\lim_{n \text{\ even}, n\rightarrow \infty}\mathcal{Q}^nf$$ do not depend on variables on $\Gamma_0$ and $\Gamma_1$ respectively. Since both the subsequences $\{\mathcal{Q}^nf\}_{n\text{\ even}}$ and $\{\mathcal{Q}^nf\}_{n\text{\ odd}}$ converge to $\theta (f)$ $\nu-$a.e. we have that $$\theta_{o}(f)=\theta(f)=\theta_{e}(f)$$ which implies that $\theta (f)$ is a constant. From that we obtain that $$\label{BorelCant6} \nu \left( \theta (f) \right)=\theta(f)$$ Since the sequence $\{ \mathcal{Q}^n f\}_{n \in \mathbb{N}}$ converges $\nu-$almost, the same holds for the sequence $\{ \mathcal{Q}^n f-\nu \mathcal{Q}^n f\}_{n \in \mathbb{N}}$. We have $$\lim_{n\rightarrow \infty}(\mathcal{Q}^n f-\nu \mathcal{Q}^n f)=\theta (f)-\nu\left( \theta(f) \right) =\theta (f)-\theta (f)=0$$ where above we used (\[BorelCant6\]). On the other side, we also have $$\label{BorelCant7}\lim_{n\rightarrow \infty}( \mathcal{Q}^n f-\nu \mathcal{Q}^n f)=\lim_{n\rightarrow \infty}(\mathcal{Q}^n f-\nu f)=\theta (f)-\nu (f)$$ From (\[BorelCant6\]) and (\[BorelCant7\]) we get that $$\theta (f)=\nu (f)$$ We finally get $$\lim_{n\rightarrow \infty}\mathcal{P}^n f=\lim_{n \text{\ even}, n\rightarrow \infty}\mathcal{Q}^nf=\nu f, \ \ \nu \ \ a.e.$$
Proof of Proposition \[prp2.2\]
===============================
Before we prove Proposition \[prp2.2\] we present some useful lemmata. First we define $$\label{defineW_k}W_k=\nabla_kV(x_{k},x_{k-1})+\nabla_kV(x_{k},x_{k+1}) \text{\; and \;}
U_k=\left\vert W_{k}\right\vert^q+\mathbb{E}^{\{\sim k\}}\left\vert W_{k}\right\vert^q$$ where $\{\sim k \}\equiv\{j:j\sim k\}=\{k-1,k+1\}$.
\[lem4.1\] The following inequality holds $$\begin{aligned}
\mathbb{E}^{\{\sim k\}}( f^q; W_{k})
\leq c_{0}\left(\mathbb{E}^{\{\sim k\}}\vert f\vert^q\right)^{\frac{1}{p}}\left( \mathbb{E}^{\{\sim k \}}(\vert f-\mathbb{E}^{\{\sim k\}}f\vert^qU_k) \right)^\frac{1}{q}\end{aligned}$$ for some constant $c_0$ uniformly on the boundary conditions and $\frac{1}{q}+\frac{1}{p}=1$.
We can write $$\begin{aligned}
\label{4.1}\mathbb{E}^{\{\sim k\}}(f^q;W_{k})=
\frac{1}{2}\mathbb{E}^{\{\sim k\}}\otimes\mathbb{\tilde
E}^{\{\sim k\}}\left((f^q-\tilde f^{q})(W_{k}-\tilde W_k )\right)\end{aligned}$$ where $\mathbb{\tilde
E}^{\{\sim k\}}$ is an isomorphic copy of $\mathbb{E}^{\{\sim k\}}$. If we define the function $F$ to be $F(s)=sf+(1-s)\tilde f$ then $$\begin{aligned}
& ~\eqref{4.1}=\frac{1}{2}\mathbb{E}^{\{\sim k\}}\otimes\mathbb{\tilde
E}^{\{\sim k\}}\left(\left(\int_0^1ds\frac{d}{ds}
F(s)^{q}\right)(W_{k}-\tilde W_k) \right)\nonumber \\ \nonumber& \ \ \
\ \ \ \ = \frac{1}{2}\mathbb{E}^{\{\sim k\}}\otimes\mathbb{\tilde
E}^{\{\sim k\}}\left(\left(\int_0^1dsq
F(s)^{q-1}\frac{d}{ds}F(s)\right)(W_{k}-\tilde W_k ) \right) \\ &\ \ \
\ \ \
\ =\frac{1}{2}\mathbb{E}^{\{\sim k\}}\otimes\mathbb{\tilde
E}^{\{\sim k\}}\left(\left(q\int_0^1dsF(s)^{q-1}(f-\tilde f)\right)(W_{k}-\tilde W_k )\right) \nonumber\end{aligned}$$ If we use the Holder inequality for the conjugate numbers $p$ and $q$, then the last quantity can be bounded by
$ \ \ \ \ \ \ \ \ \ \frac{q}{2}\left\{\mathbb{E}^{\{\sim k\}}\otimes\mathbb{\tilde
E}^{\{\sim k\}}\left( \int_0^1dsF(s)^{q-1} \right)^p\right\}^{\frac{1}{p}}\times$
$$\label{4.2}\ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \left\{\mathbb{E}^{\{\sim k\}}\otimes\mathbb{\tilde
E}^{\{\sim k\}}\left\vert(f-\tilde f)(W_{k}-\tilde W_k) \right\vert^q\right\}^{\frac{1}{q}}$$
For the first term in the above product, by Jensen’s Inequality and $\frac{1}{q}+\frac{1}{p}=1$, we obtain $$\begin{aligned}
& \left\{ \mathbb{E}^{\{\sim k\}}\otimes\mathbb{\tilde
E}^{\{\sim k\}} \left(\int_0^1dsF(s)^{q-1}\right)^p\right\} ^{\frac{1}{p}}\nonumber \\ &\nonumber \ \ \ \ \ \ \ \ \ \leq \left\{\mathbb{E}^{\{\sim k\}}\otimes\mathbb{\tilde
E}^{\{\sim k\}}\int_0^1dsF(s)^{q}\right\}^{\frac{1}{p}}= \left(\int_0^1ds\mathbb{E}^{\{\sim k\}}\otimes\mathbb{\tilde
E}^{\{\sim k\}}F(s)^{q}\right)^{\frac{1}{p}} \\ & \label{4.3}
\ \ \ \ \ \ \ \ \ \leq \left( 2^q\int_0^1ds\mathbb{E}^{\{\sim k\}}\otimes\mathbb{\tilde
E}^{\{\sim k\}} \left(sf^{q}+(1-s)\tilde
f^{q} \right) \right)^{\frac{1}{p}}= 2^{\frac{q}{p}}(\mathbb{E}^{\{\sim k\}} f^{q} )^\frac{1}{p}\end{aligned}$$ If we plug into we finally get $$\begin{aligned}
\mathbb{E}^{\{\sim k\}}(f^q;W_{k})\leq &\frac{2^{\frac{q}{p}}q}{2}(\mathbb{E}^{\{\sim k\}}f^{q} ) ^\frac{1}{p}\left\{\mathbb{E}^{\{\sim k\}}\otimes\mathbb{\tilde
E}^{\{\sim k\}}\left(\vert f-\tilde f\vert \vert W_{k}-\tilde W_k \vert \right)^q\right\}^{\frac{1}{q}}\\
\leq& 2^62^{\frac{q}{p}}q(\mathbb{E}^{\{\sim k\}} f^{q} ) ^\frac{1}{p}\left\{\mathbb{E}^{\{\sim k\}} \left(\vert f-\mathbb{E}^{\{\sim k\}} f\vert ^q(\left\vert W_{k}\right\vert^q+\mathbb{E}^{\{\sim k\}}\left\vert W_{k}\right\vert^q) \right)\right\}^{\frac{1}{q}}\end{aligned}$$ The lemma follows for constant $c_0=2^62^{\frac{q}{p}}q$.
Define now the quantity $$A(k)=\nu\left(\mathbb{E}^{\{\sim k\}} \vert f\vert^q\right)^{-\frac{q}{p}}\left\vert\mathbb{E}^{\{\sim k\}} (\vert f\vert^q;W_{k})\right\vert^q$$The next lemma presents an estimate of $A(k)$ involving $Q(k,k)$.
\[lem4.2\] Suppose that that hypothesis (H0)-(H2) are satisfied. Then $$A(k)\leq\frac{c^{q}_{0}C}{\epsilon } Q(k,k)+\frac{c^{q}_{0}\hat cK}{ \epsilon}\sum_{i=k-1,k+1}\nu\left\vert \nabla_{i} f\right\vert^q$$ where the constants $\epsilon$ and $K$ are as in hypothesis (H2).
We can initially bound $A(k)$ with the use of Lemma \[lem4.1\] $$\begin{aligned}
\nonumber A(k)=&\nu\left(\mathbb{E}^{\{\sim k\}}f^q\right)^{-\frac{q}{p}}\left\vert\mathbb{E}^{\{\sim k\}}(f^q;W_{k})\right\vert^q\leq c^{q}_{0}\nu \mathbb{E}^{k-1}\mathbb{E}^{k+1}(\vert f-\mathbb{E}^{k-1}\mathbb{E}^{k+1}f\vert^qU_k)\\ = & \label{4.4}c^{q}_{0}\nu_{\Lambda(k)}\left((\mathbb{E}^{M(k)}\vert f-\mathbb{E}^{\{\sim k\}}f\vert^q)U_k\right) \end{aligned}$$ because $U_k$ is localized in $\Lambda(k)$. If we use the entropy inequality and hypothesis (H1) for $\nu_{\Lambda(k)}$ as well as $(H2)$, as we did in Lemma \[lem3.1\], then for $K$ as in $(H2)$, we can bound by $$\begin{aligned}
~\eqref{4.4}\leq &\frac{c^{q}_{0}C}{\epsilon } Q(k,k)+ \frac{c^{q}_{0}K}{ \epsilon}\nu_{\Lambda(k)}\mathbb{E}^{\{\sim k\}}\vert f-\mathbb{E}^{\{\sim k\}}f\vert^q \\ \leq& \frac{c^{q}_{0}C}{\epsilon } Q(k,k)+\frac{c^{q}_{0}\hat cK}{ \epsilon} \sum_{i=k-1,k+1}\nu\left\vert \nabla_{i} f\right\vert^q\end{aligned}$$where above we used that $\mathbb{E}^{\{\sim
k\}}=\mathbb{E}^{k-1}\mathbb{E}^{k+1}$ satisfies the $SG_q$ with constant $\hat c$ uniformly on the boundary conditions, by hypothesis (H0) and Remark \[rem1.1\].
\[lem4.3\]The following inequality holds $$\nu\left\vert \nabla_{i}(\mathbb{E}^{i-1}\mathbb{E}^{i+1}\vert f\vert^q)^\frac{1}{q}
\right\vert^q \leq c_1\nu\left\vert \nabla_i f
\right\vert^q+\frac{J^qc_1}{q^q} A(i)$$
We have $$\begin{aligned}
\nonumber\nu\left\vert \nabla_{i}(\mathbb{E}^{i-1}\mathbb{E}^{i+1}f^q)^\frac{1}{q}
\right\vert^q=&\nu\left\vert \frac{1}{q}(\mathbb{E}^{i-1}\mathbb{E}^{i+1}f^q)^{\frac{1}{q}-1}
\nabla_{i}(\mathbb{E}^{i-1}\mathbb{E}^{i+1}f^q)\right\vert^q \\ =&\label{4.5}\frac{1}{q^{q}}\nu (\mathbb{E}^{i-1}\mathbb{E}^{i+1}f^q)^{-\frac{q}{p}}
\left\vert\nabla_{i}(\mathbb{E}^{i-1}\mathbb{E}^{i+1}f^q)\right\vert^q\end{aligned}$$ But from relationship of Lemma \[3.3\], for $\rho_i$ being the density of $\mathbb{E}^{\{\sim i\}}$ we have
$\left\vert\nabla_{i}(\mathbb{E}^{i-1}\mathbb{E}^{i+1}f^q)\right\vert^{q}=\left\vert\nabla_{i}(\int \int \rho_{i} f^q dx_{i-1}dx_{i+1})\right\vert^{q}\leq $ $$\ \ \ \ \label{4.6}2^{2q-2}\left\vert\int \int \nabla_{i}(f^q) \rho_{i} dx_{i-1}dx_{i+1}\right\vert^{q}+ 2^{2q-2}\left\vert\int \int f^q(\nabla_{i}\rho_{i} )dx_{i-1}dx_{i+1}\right\vert^{q}$$ For the second term in we have $$\label{4.7} \left\vert\int \int f^q(\nabla_{i}\rho_{i} )dx_{i-1}dx_{i+1}\right\vert^{q}\leq J^{q}\left\vert\mathbb{E}^{i-1}\mathbb{E}^{i+1}(f^q; \nabla_{i}V(x_{i-1},x_{i})+\nabla_{i}V(x_{i+1},x_{i}))\right\vert^q$$ While for the first term of the following bound holds $$\begin{aligned}
\nonumber \left\vert\int \int \nabla_{i}(f^q) \rho_{i} dx_{i-1}dx_{i+1}\right\vert^{q}&=q^{q}\left\vert\mathbb{E}^{i-1}\mathbb{E}^{i+1}(f^{q-1}( \nabla_{i}f)) \right\vert^{q}\\ \nonumber & \leq
q^{q}\left( \mathbb{E}^{i-1}\mathbb{E}^{i+1} f^{(q-1)p} \right)^{\frac{q}{p}}\left(\mathbb{E}^{i-1}\mathbb{E}^{i+1}\left\vert\nabla_{i}f\right\vert^q\right)\\
&\label{4.8}=q^{q}\left( \mathbb{E}^{i-1}\mathbb{E}^{i+1} f^{q} \right)^{\frac{q}{p}}\left(\mathbb{E}^{i-1}\mathbb{E}^{i+1}\left\vert\nabla_{i}f\right\vert^q\right)\end{aligned}$$ where above we used the Hölder inequality and that $p$ is the conjugate of $q$. If we plug and in we get $$\begin{aligned}
\left\vert\nabla_{i}(\mathbb{E}^{i-1}\mathbb{E}^{i+1}f^q)\right\vert^q\leq & 2^{2q-2}q^{q}\left( \mathbb{E}^{i-1}\mathbb{E}^{i+1} f^{q} \right)^{\frac{q}{p}}\left(\mathbb{E}^{i-1}\mathbb{E}^{i+1}\left\vert\nabla_{i}f\right\vert^q\right)\\
&+2^{2q-2}J^{q}\left\vert\mathbb{E}^{i-1}\mathbb{E}^{i+1}(f^q; \nabla_{i}V(x_{i-1},x_{i})+\nabla_{i}V(x_{i+1},x_{i}))\right\vert
^q\end{aligned}$$ From the last relationship and the lemma follows.
Now we can prove Proposition \[prp2.2\].
***Proof of Proposition \[prp2.2\]**.* We have $$\begin{aligned}
\nonumber \nu\left\vert \nabla_{\Gamma_1}(\mathbb{E}^{\Gamma_{0}} f^q)^\frac{1}{q}
\right\vert^q=&\sum_{i\in \Gamma_1} \nu\left\vert \nabla_{i}(\mathbb{E}^{\Gamma_{0}} f^q)^\frac{1}{q}
\right\vert^q\leq\sum_{i\in \Gamma_1}\nu\left\vert \nabla_{i}(\mathbb{E}^{\{ \sim i\}} f^q)^\frac{1}{q}
\right\vert^q \\ \leq & \sum_{i\in \Gamma_1}c_1\nu\left\vert \nabla_i f
\right\vert^q+\frac{J^qc_1}{q^q} \sum_{i\in \Gamma_1} A(i)\nonumber\end{aligned}$$ where the last inequality is due to Lemma \[lem4.3\]. If we use Lemma \[lem4.2\] to bound $A(i)$ we get $$\begin{aligned}
\nonumber \nu\left\vert \nabla_{\Gamma_1}(\mathbb{E}^{\Gamma_{0}} f^q)^\frac{1}{q}
\right\vert^q\leq&\sum_{i\in \Gamma_1}c_1\nu\left\vert \nabla_i f
\right\vert^q+\frac{c^{q}_{0}\hat c K}{ \epsilon}\frac{J^qc_1}{q^q} \sum_{i\in \Gamma_1}\sum_{r=i-1,i+1}\nu\left\vert \nabla_{r} f\right\vert^q\\ &+\frac{J^qc_1}{q^q} \frac{c^{q}_{0}C}{\epsilon }\sum_{i\in \Gamma_1} Q(i,i)\nonumber\end{aligned}$$ Furthermore, if we use Lemma \[lem3.2\] to bound $Q(i,i)$ we obtain $$\begin{aligned}
\nu\left\vert \nabla_{\Gamma_1}(\mathbb{E}^{\Gamma_0}
f^q)^\frac{1}{q}
\right\vert^q\leq & \nonumber\sum_{i\in \Gamma_1}c_1\nu\left\vert \nabla_i f
\right\vert^q+\frac{c^{q}_{0}\hat cK}{ \epsilon}\frac{J^qc_1}{q^q} \sum_{i\in \Gamma_1}\sum_{r=i-1,i+1}\nu\left\vert \nabla_{r} f\right\vert^q
\\ & \nonumber +\frac{J^qc_1}{q^q} \frac{c^{q}_{0}CD}{\epsilon }\sum_{i\in \Gamma_1} \sum_{r=k-2}^{k+2}\nu\left\vert \nabla_{r}f
\right\vert^q
\\ & +\frac{J^qc_1}{q^q} \frac{c^{q}_{0}CD}{\epsilon }\sum_{i\in \Gamma_1}\sum_{ n=0 }^{\infty} J^{(n+1)(q-1)}\sum_{r=0}^3\nu\left\vert \nabla_{i+3+4n+r} f
\right\vert^q\nonumber \\ & \label{4.9} +\frac{J^qc_1}{q^q} \frac{c^{q}_{0}CD}{\epsilon }\sum_{i\in \Gamma_1}\sum_{ n=0 }^{\infty} J^{(n+1)(q-1)}\sum_{r=0}^3\nu\left\vert \nabla_{i-3-4n-r} f
\right\vert^q\end{aligned}$$ If we set $R=c_1+\frac{c_1}{q^q} (\frac{c^{q}_{0}CD}{\epsilon }+\frac{c^{q}_{0}\hat cK}{ \epsilon})$ and we choose $J<1$, relationship gives $$\nu\left\vert \nabla_{\Gamma_1}(\mathbb{E}^{\Gamma_{0}} f^q)^\frac{1}{q}
\right\vert^q\leq (R+RJ^{q}4+\frac{R8J^{q}}{1-J^{q-1}})\nu\left\vert \nabla_{ \Gamma_1} f
\right\vert^q+RJ^{q}(4+\frac{8}{1-J^{q-1}})\nu\left\vert \nabla_{ \Gamma_0} f
\right\vert^q$$ For $J$ sufficiently small (H3) such that $RJ^{q}(4+\frac{8}{1-J^{q-1}})<1$ the lemma follows for constants
$C_1=R+RJ^{q}4+\frac{R8J^{q}}{1-J^{q-1}}$ and $C_2=RJ^{q}(4+\frac{8}{1-J^{q-1}})<1$.
Proof of Lemma \[lem3.2\]
=========================
This section is dedicated in the proof of Lemma \[lem3.2\] under the assumptions (H0)-(H3). We begin by showing the weaker result of Lemma \[50\] under the weaker assumptions (H1)-(H3).
\[50\]Suppose that hypothesis (H1)-(H3) are satisfied. Then $$\begin{aligned}
Q(k,k)\leq & J^{q}S\nu\left\vert f-\mathbb{E}^{k-1}\mathbb{E}^{k+1}f\right\vert^q+
S\sum_{r=k-2}^{k+2}\nu\left\vert \nabla_{r}
f
\right\vert^q\\&+S\sum_{ n=0 }^{\infty} J^{(n+1)(q-1)}\sum_{r=0}^3\left(\nu\left\vert \nabla_{k+3+4n+r} f
\right\vert^q+\nu\left\vert \nabla_{k-3-4n-r} f
\right\vert^q\right)\end{aligned}$$for some positive constant $S$.
Lemma \[lem3.2\] follows for some constant $D>0$ directly from the last lemma and the Spectral Gap inequality implied from (H0). The remaining of this section is dedicated to the proof of Lemma \[50\]. At first we prove some lemmata. To start, for any $k \in \mathbb{Z}$, we define the sets $M_s(k)$ for $s=k-3,k+3$ as $$\label{Ms(k)}
M_s(k)=\begin{cases}\{j\in\mathbb{Z}:j\geq k+3\}=\{k+3,k+4,...\}\ & \text{if \ } s=k+3\\
\{j\in\mathbb{Z}:j\leq\ k-3\}=\{...,k-4,k-3\}\ & \text{if \ } s=k-3 \\
\end{cases}$$
\[remarkMs(k)\] Since $\Lambda(k)=\{k-2,k-1,k,k+1,k+2\}$ and $M(k)=\mathbb{Z}\smallsetminus\Lambda(k)$, with the use of the definition (\[Ms(k)\]) we can write $$M(k)= \{j\in\mathbb{Z}:j\leq\ k-3\}\cup\{j\in\mathbb{Z}:j\geq k+3\}=M_{k-3}(k)\cup M_{k+3}(k)$$ Since the sets $M_{k-3}(k)$ and $M_{k+3}(k)$ are disjoint we obtain that $\mathbb{E}^{M(k)}$ is a product measure, and for every function $f$ we can write$$\label{productmeasurein1dim}\mathbb{E}^{M(k)}f=\mathbb{E}^{M_{k-3}(k)}\otimes\mathbb{E}^{M_{k+3}(k)}f$$ Accordingly, for functions, say $f_{k-3}$ and $f_{k+3}$, that depend on variables $x_i$ with $i\notin\ M_{k+3}(k)$ and $i\notin\ M_{k-3}(k)$ respectively, we obtain $$\mathbb{E}^{M(k)}f_{k-3}=\mathbb{E}^{M_{k-3}(k)}f_{k-3}$$ and $$\mathbb{E}^{M(k)}f_{k+3}=\mathbb{E}^{M_{k+3}(k)}f_{k+3}$$ For instance, for $r=k-2,k+2$ and $ s\in \{k-3,k+3\}:s\sim r$, that is for the couples $(r,s)=(k-2,k-3)$ and $(r,s)=(k+2,k+3)$, we have $$\label{5.2}\mathbb{E}^{M(k)}\nabla_rV(x_{s},x_{r})=\mathbb{E}^{M_s(k)}\nabla_rV(x_{s},x_{r})$$
\[remarkMs(k)2\] Consider couples $(r,s)$ that take the values $ (k-2,k-3)$ and $(k+2,k+3)$. We then have that $\nabla_rV(x_{s},x_{r})$ is localised in $\Lambda(k-4)$ when $(r,s)=(k-2,k-3)$ and in $\Lambda(k+4)$ when $(r,s)=(k+2,k+3)$. Furthermore, from Remark \[remarkMs(k)\], for $(r,s)=(k-2,k-3)$ we get that $$\mathbb{E}^{M_{k-3}(k)}\nabla_{k-2}V(x_{k-3},x_{k-2})=\mathbb{E}^{\{...,k-4,k-3\}}\nabla_{k-2}V(x_{k-3},x_{k-2})$$ is localised in $\Lambda(k-4)$, while for $(r,s)=(k+2,k+3)$ we get that $$\mathbb{E}^{M_{k+3}(k)}\nabla_{k+2}V(x_{k+3},x_{k+2})=\mathbb{E}^{\{k+3,k+4,...\}}\nabla_{k+2}V(x_{k+3},x_{k+2})$$ is localised in $\Lambda(k+4)$. So, if we set $$Y_{s}(x_{s},x_{r})=\vert\nabla_rV(x_{s},x_{r})-\mathbb{E}^{M_s(k)}\nabla_rV(x_{s},x_{r})\vert$$ we then have that $Y_{k+3}(x_{k+2},x_{k+3})$ and $Y_{k-3}(x_{k-2},x_{k-3})$ are localised in $\Lambda(k+4)$ and $\Lambda(k-4)$ respectively. Thus, we have $$\begin{aligned}
\nu(f^qY_{k+3}^q(x_{k+3},x_{k+2}))=\nu_{\Lambda(k+4)}\left((\mathbb{E}^{M(k+4)}f^q)Y_{k+3}^q(x_{k+3},x_{k+2})\right)
\end{aligned}$$ $$\begin{aligned}
\nu(f^qY_{k-3}^q(x_{k-3},x_{k-2}))=\nu_{\Lambda(k-4)}\left((\mathbb{E}^{M(k-4)}f^q)Y_{k-3}^q(x_{k-3},x_{k-2})\right)
\end{aligned}$$ If we combine the last two together we can write $$\begin{aligned}
\nu(f^qY_{s}^q(x_{s},x_{r}))=\nu_{\Lambda(t)}\left((\mathbb{E}^{M(t)}f^q)Y^q_{s}(x_{s},x_{r})\mathcal{I}_{t\in
\{k-4,k+4\}\cap M_s(k)}\right)
\end{aligned}$$ for $(r,s)\in \{(k+2,k+3),(k-2,k-3)\}$.
\[lem5.1\]Suppose conditions (H1) and (H2) are satisfied. Then for $r=k-2,k+2$ and $ s\in \{k-3,k+3\}:s\sim r$ the following inequality is true $$\begin{aligned}
\nu_{\Lambda(k)}\left(\mathbb{E}^{M(k)}
\vert f\vert^q\right)^{-\frac{q}{p}}&\left\vert\mathbb{E}^{M(k)}(\vert
f\vert^q;\nabla_rV(x_{s},x_{r}))\right\vert^q \leq \\ &
\frac{C}{ \epsilon }\nu_{\Lambda(t)}\left\vert \nabla_{\Lambda(t)}
(\mathbb{E}^{M(t)}\vert f\vert^q )^{\frac{1}{q}}\right\vert^q\mathcal{I}_{t\in
\{k-4,k+4\}\cap M_s(k)}+\frac{K}{\epsilon}\nu \vert f\vert^q\end{aligned}$$ where $\mathcal{I}_A$ denotes the characteristic function of a set $A$ and the set $M_s(k)$ as in (\[Ms(k)\]).
For any two function $f$ and $g$ the covariance with respect to a measure $\mu$ can be computed as bellow $$\begin{aligned}
\mu (f;g)&=\mu\left((f-\mu f)(g-\mu g)\right)=\mu\left (f(g-\mu g)\right)-\mu\left(\mu f(g-\mu g)\right)\\ &=\mu\left(f(g-\mu g)\right)-(\mu f)\mu\left(g-\mu g\right)=\mu\left (f(g-\mu g)\right)\end{aligned}$$ Using this expression we can write $$\begin{aligned}
\label{5.1} \mathbb{E}^{M(k)}(f^q;\nabla_rV(x_{s},x_{r}))= \mathbb{E}^{M(k)}(
f^q(\nabla_rV(x_{s},x_{r})-\mathbb{E}^{M(k)}\nabla_rV(x_{s},x_{r})))\end{aligned}$$ If we use from Remark \[remarkMs(k)\], becomes $$\label{5.3}\mathbb{E}^{M(k)}(f^q;\nabla_rV(x_{s},x_{r}))=\mathbb{E}^{M(k)}(f^q(\nabla_rV(x_{s},x_{r})-\mathbb{E}^{M_s(k)}\nabla_rV(x_{s},x_{r})))$$ If we set $$Y_{s}(x_{s},x_{r})=\vert\nabla_rV(x_{s},x_{r})-\mathbb{E}^{M_s(k)}\nabla_rV(x_{s},x_{r})\vert$$ then for we can write $$\begin{aligned}
\nonumber\left\vert \mathbb{E}^{M(k)}(f^q;\nabla_rV(x_{s},x_{r})) \right\vert&\leq\mathbb{E}^{M(k)}(f^{q-1+1}Y_{s}(x_{s},x_{r}))
\\ & \leq\left(\mathbb{E}^{M(k)}f^{(q-1)p}\right)^\frac{1}{p}\left (\mathbb{E}^{M(k)}(f^qY^q_{s}(x_{s},x_{r}))\right
)^{\frac{1}{q}}\nonumber\\
&\label{5.4}=\left(\mathbb{E}^{M(k)}f^{q}\right)^\frac{1}{p}\left(\mathbb{E}^{M(k)}(f^qY^q_{s}(x_{s},x_{r}))\right
)^{\frac{1}{q}}\end{aligned}$$ where above we used the Hölder inequality and that $\frac{1}{p}+\frac{1}{q}=1$. So, for $s=k+3,k-3$ from relationship we obtain $$\begin{aligned}
\nonumber\nu_{\Lambda(k)}\left(\mathbb{E}^{M(k)}f^q\right)^{-\frac{q}{p}}\left\vert\mathbb{E}^{M(k)}(f^q;\nabla_rV(x_{s},x_{r}))\right
\vert^{q}& \leq\nu_{\Lambda(k)}\mathbb{E}^{M(k)}(f^qY_{s}^q(x_{s},x_{r}))\\ & =\nu(f^qY_{s}^q(x_{s},x_{r}))\end{aligned}$$ If we combine the last inequality together with Remark \[remarkMs(k)2\] we finally obtain $$\begin{aligned}
\nonumber\nu_{\Lambda(k)}\left(\mathbb{E}^{M(k)}f^q\right)^{-\frac{q}{p}}&\left\vert\mathbb{E}^{M(k)}(f^q;\nabla_rV(x_{s},x_{r}))\right
\vert^{q} \leq\\ &
\label{5.5}\nu_{\Lambda(t)}\left((\mathbb{E}^{M(t)}f^q)Y^q_{s}(x_{s},x_{r})\mathcal{I}_{t\in
\{k-4,k+4\}\cap M_s(k)}\right)\end{aligned}$$ If in we use the Entropy Inequality and the $LS_q$ for $\nu_{\Lambda(s)}$ from hypothesis (H1) and (H2), we get $$\begin{aligned}
\nu_{\Lambda(k)}\left(\mathbb{E}^{M(k)}
f^q\right)^{-\frac{q}{p}} &\left(\mathbb{E}^{M(k)}(
f^q;\nabla_rV(x_{s},x_{r}))\right)^q \leq \\ &
\frac{C}{ \epsilon }\nu_{\Lambda(t)}\left\vert \nabla_{\Lambda(t)}
(\mathbb{E}^{M(t)}f^q )^{\frac{1}{q}}\right\vert^q\mathcal{I}_{t\in
\{k-4,k+4\}\cap M_s(k)}+\frac{K}{\epsilon}\nu f^q \end{aligned}$$ and $ s\in \{k-3,k+3\}:s\sim r$ and $K$ and $\epsilon$ as in hypothesis (H2).
\[lem5.2-1\]Suppose $P$ and $G$ are positive functions with domain on $\mathbb{N}$ such that for constants $J,K'>0$ $$\label{l6-5.6}P(4)\leq G(4)+J^qK'P(8)$$ and for $n=4k$ for $k\in\mathbb{N}\cap [2,\infty)$ $$\label{l6-5.7}P(n)\leq G(n)+J^qK'P(n-4)+ J^qK'P(n+4)$$ Then for $J$ sufficiently small such that $$\label{l6-5.8}J\leq1 \ \text{\;
and \;}JK'+J^qK'J^{q-1}\leqslant 1$$ the following inequality holds $$\begin{aligned}
P(4n)\nonumber \leq&\frac{1}{1-J^qK'J^{q-1}}\sum_{m=0}^{n-2} J^{mq-m}G(4n-4m)+J^{(n-1)q-(n-1)}G(4) \\ &\label{l6-5.9}+ J^{q-1}P(4n+4) \end{aligned}$$ for any $n\in \mathbb{N},n\geq2$ .
In order to show we will work inductively.
Step 1: The base case of the induction (n=2).
We prove for $n=2 $. For $k=8$ in (\[l6-5.7\]) we have $$P(8)\leq G(8)+J^qK'P(12)+ J^qK'P(4)$$ If we bound $P(4)$ in the above inequality by (\[l6-5.6\]) we obtain $$P(8) \leq G(8)+J^qK'P(12)+ J^qK'G(4)+(J^{q}K')^2P(8)\Rightarrow$$ $$\label{5.10}P(8)\leq \frac{1}{1-(J^{q}K')^2}G(8)+ \frac{J^qK'}{1-(J^{q}K')^2}G(4)+\frac{J^qK'}{1-(J^{q}K')^2}P(12)$$ For $J$ satisfying properties , we have $JK'+J^qK'J^{q-1}\leq 1 $ and $JK'<1$ which implies $$\begin{aligned}
\label{5.11}JK'+(J^{q}K')^2&\leq 1 \Rightarrow
\frac{J^qK'}{1-(J^{q}K')^2}\leq{J^{q-1}}\end{aligned}$$ From and we have $$\begin{aligned}
P(8)& \leq\frac{1}{1-(J^{q}K')^2}G(8)+ J^{q-1}G(4)+J^{q-1}P(12)\\&\leq\frac{1}{1-J^{q}K'J^{q-1}}G(8)+ J^{q-1}G(4)+J^{q-1}P(12)\end{aligned}$$ because of . This proves for $n=2$.
Step 2: The induction step. Suppose the inequality is true for $n=k$. Then we will show it is also true for $n=k+1$.
If we use for $n=4k+4$ we have $$\label{5.12}P(4k+4)\leq G(4k+4)+J^qK'P(4k)+ J^qK'P(4k+8)$$ If we use for $n=k$ to bound $P(4k)$ in we get $$\begin{aligned}
P(4k+4)\leq& G(4k+4)+ \frac{J^qK'}{1-J^qK'J^{q-1}}\sum_{m=0}^{k-2} J^{mq-m}G(4k-4m)\\&
+J^qK'J^{(k-1)q-(k-1)}G(4)+J^qK'J^{q-1}P(4k+4)+ J^qK'P(4k+8) \end{aligned}$$ This implies $$\begin{aligned}
\nonumber P(4k+4)\leq & \frac{1}{1-J^qK'J^{q-1}}G(4k+4)+ \frac{J^qK'}{1-J^qK'J^{q-1}}\sum_{m=0}^{k-2} \frac{J^{mq-m}}{1-J^qK'J^{q-1}}G(4k-4m)\\ &+
\label{5.13} \frac{J^qK'J^{(k-1)q-(k-1)}}{1-J^qK'J^{q-1}}G(4)+ \frac{J^qK'}{1-J^qK'J^{q-1}}P(4k+8)\end{aligned}$$ If we use condition for $J$, becomes
$P(4k+4)\leq \frac{1}{1-J^qK'J^{q-1}}\sum_{m=0}^{k-1} J^{mq-m}G(4k+4-4m)+J^{kq-k}G(4)$
$\ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ +J^{q-1}P(4k+8)$
which proves for $n=k+1$. This finishes the proof of .
\[lem5.2\]Suppose $P$ and $G$ are positive functions with domain on $\mathbb{N}$ such that for constants $J,K'>0$ one has$$\label{newlemma5.4}\sup_{n\in\mathbb{N}}P(n)<\infty$$ as well as (\[l6-5.6\]) and (\[l6-5.7\]) for $n=4k$ for $k\in\mathbb{N}\cap [2,\infty)$. Then for $J$ sufficiently small such that (\[l6-5.8\]) is true, the following inequality holds $$P(4)\leq\frac{1}{1-J^{2q-2}}\sum_{n=0}^{+\infty} J^{nq-n}G(4n+4)$$
We can use relationship from Lemma \[lem5.2-1\] to prove the lemma. We first replace the bound of $P(8)$ from in , to obtain $$\begin{aligned}
P(4)&\leq G(4)+J^qK'\frac{1}{1-J^qK'J^{q-1}}G(8)+J^qK'J^{q-1}G(4)+J^qK'J^{q-1}P(12) \\ & \leq (1+J^qK'J^{q-1})G(4)+J^{2q-2}G(8)+JK'J^{2q-2}P(12) \end{aligned}$$ where at the last inequality we used . If we now bound in the above expression $P(12)$ from , then $P(16)$ from and so on, we will finally obtain $$\begin{aligned}
P(4)\leq& (1+J^{q}K'\sum_{n=0}^{+\infty}J^{(2n+1)q-(2n+1)} )G(4) \\ &+
\frac{J^qK'}{1-J^qK'J^{q-1}}\sum_{n=1}^{+\infty} J^{(n-1)q-(n-1)}(\sum_{s=0}^{+\infty}J^{2sq-2s})G(4n+4)\\
= &
(1+\frac{J^{2q-1}K'}{1-J^{2q-2}} )G(4)+ \frac{J^qK'}{1-J^qK'J^{q-1}}\frac{1}{1-J^{2q-2}}\sum_{n=1}^{+\infty} J^{(n-1)q-(n-1)}G(4n+4)\end{aligned}$$ where above we used that $J<1$, as well as that $$\lim_{n\rightarrow\infty}J^{nq-n}P(8+4n)=0$$ since (\[newlemma5.4\]) is true. Furthermore, if we use again we then get $$P(4)\leq \frac{1}{1-J^{2q-2}}\sum_{n=0}^{+\infty} J^{nq-n}G(4n+4)$$
The next lemma presents a bound for $$Q(u,k)=\nu_{\Lambda(u)}\left\vert \nabla_{\Lambda(u)}\left( \mathbb{E}^{M(u)}
\vert h_k\vert ^q \right)^{\frac{1}{q}}\right\vert^q$$ in terms of $Q(t,k)\mathcal{I}_{dist(u,t)=4}$.
\[lem5.3\]Under hypothesis (H1) and (H2) the following bound for $Q(u,k)$ holds $$\begin{aligned}
Q(u, k)\leq&\nu_{\Lambda(u)}\left\vert \nabla_{u}
h_k
\right\vert^q+\sum_{r= u-1,u+1}\nu_{\Lambda(u)}\left\vert \nabla_{r}
h_k
\right\vert^q+\frac{J^qc_{1}2cK}{\epsilon}\nu\left\vert h_{k}\right\vert^q \\ &+c_{1}\sum_{r= u-2,u+2}\nu_{\Lambda(u)}\left\vert \nabla_{r} h_{k}
\right\vert^q+ \frac{J^qc_{1}C}{ \epsilon}\sum_{dist(u,t)=4}Q(t,k)\end{aligned}$$ where $h_{k}=f-\mathbb{E}^{\{\sim k\}}f$.
We have $$\begin{aligned}
Q(u,k)=&\nonumber\nu_{\Lambda(u)}\left\vert \nabla_{\Lambda(u)}\left( \mathbb{E}^{M(u)}
\vert h_k\vert^q \right)^{\frac{1}{q}}\right\vert^q=\nu_{\Lambda(u)}\left\vert \nabla_{u} (\mathbb{E}^{M(u)}
\vert h_k\vert^q)^{\frac{1}{q}}
\right\vert^q\\ &
\label{5.14}+\sum_{r= u-1,u+1}\nu_{\Lambda(u)}\left\vert \nabla_{r} (\mathbb{E}^{M(u)}
\vert h_k\vert^q)^{\frac{1}{q}}
\right\vert^q+\sum_{r= u-2,u+2}\nu_{\Lambda(u)}\left\vert \nabla_{r} (\mathbb{E}^{M(u)}
\vert h_k\vert^q)^{\frac{1}{q}}
\right\vert^q \end{aligned}$$ For $r=u-1,u,u+1$ $$\label{5.15}\nu_{\Lambda(u)}\left\vert \nabla_{r} (\mathbb{E}^{M(u)}
\vert h_k\vert^q)^{\frac{1}{q}}
\right\vert^q\leq\nu_{\Lambda(u)}\left\vert \nabla_r
h_k
\right\vert^q$$ For $r=u-2,u+2$ $$\begin{aligned}
\nonumber\nu_{\Lambda(u)}&\left\vert \nabla_{r} (\mathbb{E}^{M(u)}
\vert h_k\vert^q)^{\frac{1}{q}}
\right\vert^q\leq c_1\nu_{\Lambda(u)}\left\vert \nabla_{r} h_{k}
\right\vert^q+\\ &
\label{5.16}\frac{J^qc_1}{q^q}\nu_{\Lambda(u)}\left(\mathbb{E}^{M(u)}\vert h_k\vert^{q}\right)^{-\frac{q}{p}}\left(\mathbb{E}^{M(u)}(\vert h_k\vert^q;\nabla_rV(x_{r},x_{s}))\right)^q
\mathcal{I}_{s\in \{u-3,u+3\}:s\sim r}\end{aligned}$$ For $s\in \{u-3,u+3\}:s\sim r$, if we use Lemma \[lem5.1\] we obtain $$\begin{aligned}
\nonumber\nu_{\Lambda(u)}&\left(\mathbb{E}^{M(u)}\vert h_k\vert^{q}\right)^{-\frac{q}{p}}\left\vert\mathbb{E}^{M(u)}(\vert h_k\vert^q;V'(x_{r},x_{s}))\right\vert^q
\mathcal{I}_{s\in \{u-1,u+1\}:s\sim r}\leq\\ &\frac{C}{ \epsilon }\nu_{\Lambda(t)}\left\vert \nabla_{\Lambda(t)}
(\mathbb{E}^{M(t)}\vert h_k\vert^q )^{\frac{1}{q}}\right\vert^q\mathcal{I}_{(s,t)=(u+1,u+4)\cup(u-1,u-4):s
\sim r}+\frac{K}{\epsilon}\nu \vert h_k\vert^q
\label{5.17}\end{aligned}$$ From and we get $$\begin{aligned}
\nonumber\nu_{\Lambda(u)}&\left\vert \nabla_{r} (\mathbb{E}^{M(u)}
\vert h_k\vert^q)^{\frac{1}{q}}
\right\vert^q\leq c_{1}\nu_{\Lambda(u)}\left\vert \nabla_{r} h_{k}
\right\vert^q\\&
\nonumber+\frac{J^qc_{1}C}{ \epsilon }Q(t,k)\mathcal{I}_{(s,t)=(u+1,u+4)\cup(u-1,u-4):s
\sim r}\mathcal{I}_{s\in \{u-1,u+1\}:s\sim r}\\ &\label{5.18}+\frac{J^qc_{1}K}{\epsilon}\nu\left\vert h_k\right\vert^q\mathcal{I}_{s\in \{u-1,u+1\}:s\sim r}\end{aligned}$$ To summarise, if we plug and in we finally obtain $$\begin{aligned}
Q(u,k)\leq&\frac{J^qc_1C}{ \epsilon}\sum_{dist(u,t)=4}Q(t,k)+\frac{J^qc_12K}{\epsilon}\nu\left\vert f-\mathbb{E}^{\{\sim k\}}f\right\vert^q+\nu\left\vert \nabla_u
h_k
\right\vert^q\\ &
+\sum_{r= u-1,u+1}\nu_{\Lambda(u)}\left\vert \nabla_{r}
h_k
\right\vert^q+c_{1}\sum_{r= u-2,u+2}\nu_{\Lambda(u)}\left\vert \nabla_{r} h_k
\right\vert^q\end{aligned}$$
\[lem5.4\]Suppose conditions (H1) is satisfied. Then for $r\in \Lambda(k)$, the following statements are true
\(a) When $r=\{k-2,k,k+2\}$ $$\nu\left\vert \nabla_{r} h_{k}
\right\vert^q\leq c_1\nu\left\vert \nabla_{r}f
\right\vert^q+\frac{J^qCc_{1}}{\epsilon }Q(k,k)+\frac{J^qc_{1}K}{\epsilon }\nu\left\vert f-\mathbb{E}^{\{\sim k\}}f
\right\vert^q$$ (b) When $r\in \{k-1,k+1\}$ $$\nu_{}\left\vert \nabla_{r} h_{k}
\right\vert^q= \nu\left\vert \nabla_{r}f
\right\vert^q$$where $h_k=f-\mathbb{E}^{\{\sim k\}}f$.
We will show (a). For general $r \in \Lambda(k)\smallsetminus\{k-1,k+1\}$ we have $$\label{5.19}\nu\left\vert \nabla_r h_k
\right\vert^q\leq2^{q-1}\nu\left\vert \nabla_r
f
\right\vert^q+2^{q-1}\nu\left\vert \nabla_r
\mathbb{E}^{\{\sim k\}}f
\right\vert^q$$ We will now compute $\nu\left\vert \nabla_r
\mathbb{E}^{\{\sim k\}}f
\right\vert^q$ for the separate cases of $r\in \{k-2,k+2\}$ and $r=k$.
Consider $r=\{k-2,k+2\}$. In this case $$\begin{aligned}
\nonumber\nu\left\vert \nabla_r
\mathbb{E}^{\{\sim k\}}f
\right\vert^q\leq & 2^{q-1}\nu\left\vert \nabla_{r} f
\right\vert^q\\ &
\label{5.20}+J^q2^{q-1}\nu\left\vert\mathbb{E}^{\{\sim k\}}(f;\nabla_rV(x_s,x_r))\right\vert^q\mathcal{I}_{s\in \{k-1,k+1\}:s\sim r}\end{aligned}$$ If we use Lemma \[lem3.1\] (a) to bound the second term on the right hand side of we obtain $$\begin{aligned}
\nonumber\nu\left\vert \nabla_{r}
\mathbb{E}^{\{\sim k\}}f
\right\vert^q\leq&2^{q-1}\nu\left\vert \nabla_{r} f
\right\vert^q+\frac{J^q2^{q-1}C}{\epsilon }Q(k,k) \\&
\label{5.21}+\frac{J^q2^{q-1}K}{\epsilon }\nu\left\vert f-\mathbb{E}^{\{\sim k\}}f
\right\vert^q \end{aligned}$$ Combining and together we derive $$\nu\left\vert \nabla_r h_k
\right\vert^q\leq c_1\nu\left\vert \nabla_rf
\right\vert^q+
\frac{J^qCc_1}{\epsilon }Q(k,k)+\frac{J^qc_1K}{\epsilon}\nu\left\vert f-\mathbb{E}^{\{\sim k\}}f
\right\vert^q$$for $K$ as in (H2).
Consider $r=k$. In this case $$\begin{aligned}
\nonumber\nu\left\vert \nabla_k
\mathbb{E}^{\{\sim k\}}f
\right\vert^q\leq & 2^{q-1}\nu_{}\left\vert \nabla_{k} f
\right\vert^q+J^q2^{q-1}\nu\left(\mathbb{E}^{\{\sim k\}}(f;W_{k})\right)^q \\ \leq&
\label{5.22} 2^{q-1}\nu\left\vert \nabla_{k}f
\right\vert^q+\frac{J^qC2^{q-1}}{\epsilon}Q(k,k)+\frac{J^q2^{q-1}K}{\epsilon}\nu\left\vert f-\mathbb{E}^{\{\sim k\}}f
\right\vert^q\end{aligned}$$ where in the last inequality the Lemma \[lem3.1\] (a) was used for $K$ as in (H2). From and $$\nu\left\vert \nabla_{k} h_{k}
\right\vert^q\leq c_1\nu\left\vert \nabla_{k}f
\right\vert^q+\frac{J^qCc_{1}}{\epsilon }Q(k,k)+\frac{J^qc_{1}K}{\epsilon }\nu\left\vert f-\mathbb{E}^{\{\sim k\}}f
\right\vert^q$$
We can now prove Lemma \[50\].
***Proof of Lemma \[50\].*** If we combine the bound for $Q(k,k)$ from Lemma \[lem5.3\], together with the bounds for $\nu\left\vert \nabla_{r} h_{k}\right\vert^q,r=k-2,k-1,k,k+1,k+2$ from Lemma \[lem5.4\], we obtain $$\begin{aligned}
Q(k,k)\leq& \sum_{r=k-1,k+1}\nu\left\vert \nabla_r f
\right\vert^q+c_1\nu\left\vert \nabla_kf
\right\vert^q+\frac{J^qCc_1}{\epsilon }Q(k,k)+\frac{J^qc_1K}{\epsilon}\nu\left\vert f-\mathbb{E}^{\{\sim k\}}f
\right\vert^q \nonumber \\ & \nonumber
+c_1\sum_{r=k-2,k+2}\left( c_1\nu\left\vert \nabla_rf
\right\vert^q+\frac{J^qCc_1}{\epsilon }Q(k,k)+\frac{J^qc_1K}{\epsilon }\nu\left\vert f-\mathbb{E}^{\{\sim k\}}f\right\vert^q \right)\\ & \nonumber
+\frac{J^qc_1C}{\epsilon}\sum_{dist(k,t)=4}Q(t,k)+\frac{J^qc_12K}{\epsilon}\nu\left\vert f-\mathbb{E}^{\{\sim k\}}f\right\vert^q \\= & \nonumber \sum_{r=k-1,k+1}\nu\left\vert \nabla_r f
\right\vert^q+c_1\nu\left\vert \nabla_kf
\right\vert^q+\frac{J^q(c_13+c^2_12)K}{\epsilon}\nu\left\vert f-\mathbb{E}^{\{\sim k\}}f\right\vert^q\\ & + \label{5.23}c^2_1\sum_{r=k-2,k+2} \nu\left\vert \nabla_rf
\right\vert^q+\frac{J^q2C(c_1+c^2_1)}{\epsilon}Q(k,k)+\frac{J^qc_1C}{\epsilon}\sum_{dist(k,t)=4}Q(t,k)\end{aligned}$$ In order to bound $\sum_{dist(k,t)=4}Q(t,k)$ in the above quantity the lemma bellow will be used.
\[lem5.5\]Under conditions (H1)-(H3) the following inequality $$\begin{aligned}
\sum_{t:dist(t,k)=4}Q(t,k)\leq & J^qTQ(k,k)+J^qT\nu\left\vert f-\mathbb{E}^{\{\sim k\}}f\right\vert^q+T\sum_{r=k-2,k+2}\nu\left\vert \nabla_r
f
\right\vert^q\\ &
+T\sum_{n=0}^{\infty} J^{n(q-1)}\sum_{r=0}^3\left(\nu\left\vert \nabla_{k+3+4n+r} f
\right\vert^q+\nu\left\vert \nabla_{k-3-4n-r} f
\right\vert^q\right)\end{aligned}$$ is satisfied for some positive constant $T$ independent of $k$.
The proof of Lemma \[lem5.5\] will be presented later in the section. If we use the bound of Lemma \[lem5.5\] in , we obtain $$\begin{aligned}
Q(k,k)\leq&\nonumber J^q\left(\frac{TJ^qc_{1}C}{ \epsilon}+\frac{c_13cK}{\epsilon}+\frac{c^2_12cK}{\epsilon}\right)\nu\left\vert f-\mathbb{E}^{\{\sim k\}}f
\right\vert^q\\ &
\nonumber +J^q\left(\frac{2Cc^2_1}{\epsilon}+\frac{Cc_1}{\epsilon}\right )Q(k,k)+\frac{J^qc_1C}{ \epsilon}J^qTQ(k,k)\\ &\nonumber
+\sum_{r=k-1,k+1}\nu\left\vert \nabla_r f
\right\vert^q+c_1\nu\left\vert \nabla_kf
\right\vert^q+( \frac{J^qc_1C}{ \epsilon }T+c_1^2)\sum_{r=k-2,k+2}\nu\left\vert \nabla_r f
\right\vert^q \\ &
\label{5.24}+\frac{J^qc_1C}{ \epsilon}T\sum_{ n=0 }^{\infty} J^{n(q-1)}\sum_{r=0}^3\left(\nu\left\vert \nabla_{k+3+4n+r} f
\right\vert^q+\nu\left\vert \nabla_{k-3-4n-r} f
\right\vert^q\right)\end{aligned}$$ If we choose $J$ sufficiently small such that $$1-J^q\left(\frac{2Cc^2_1}{\epsilon}+\frac{J^qc_1CT}{ \epsilon}+\frac{Cc_1}{\epsilon }\right )>\frac{1}{2}$$ then from we have $$\begin{aligned}
Q(k,k)\leq&2J^q\left(\frac{TJ^qc_{1}C}{ \epsilon}+\frac{c_13cK}{\epsilon}+\frac{c^2_12cK}{\epsilon}\right)\nu\left\vert f-\mathbb{E}^{\{\sim k\}}f
\right\vert^q+2c_1\nu\left\vert \nabla_kf
\right\vert^q\\ &
\nonumber
+2\sum_{r=k-1,k+1}\nu\left\vert \nabla_r f
\right\vert^q+2( \frac{J^qc_1C}{ \epsilon }T+c_1^2)\sum_{r=k-2,k+2}\nu\left\vert \nabla_r f
\right\vert^q \\ &+
\frac{2J^qc_1C}{ \epsilon}T\sum_{ n=0 }^{\infty} J^{n(q-1)}\sum_{r=0}^3\left(\nu\left\vert \nabla_{k+3+4n+r} f
\right\vert^q+\nu\left\vert \nabla_{k-3-4n-r} f
\right\vert^q\right)\end{aligned}$$ and the lemma follows for an appropriate positive constant $D$.
It remains to show Lemma \[lem5.5\]. For this we will need the following lemmata.
\[lem5.6\]Under conditions (H1)-(H3) the following two bounds for $Q(u,k)$ hold.
\(a) For $u$ such that $dist(u,k)\geq 8$ $$\begin{aligned}
Q(u,k)\leq & c_1\nu_{\Lambda(u)}\left\vert \nabla_u
f
\right\vert^q+c_1\sum_{r= u-1,u+1}\nu_{\Lambda(u)}\left\vert \nabla_r
f
\right\vert^q+c^2_1\sum_{r=u-2,u+2}\nu_{\Lambda(u)}\left\vert \nabla_r f
\right\vert^q \\ &
+
\frac{J^qc_1C}{\epsilon}\sum_{dist(u,t)=4}Q(t,k)+\frac{J^qc_{1}2K}{\epsilon}\nu\left\vert f-\mathbb{E}^{\{\sim k\}}f
\right\vert^q\end{aligned}$$ (b) For $u$ such that $dist(u,k)=4$ $$\begin{aligned}
Q(u,k)\nonumber\leq &c_1\nu\left\vert \nabla_u
f\right\vert^q+c^2_1\sum_{r=u-2,u+2}\nu\left\vert \nabla_r f
\right\vert^q
\nonumber +J^q\left(\frac{c_12K}{\epsilon}+\frac{c^2_1K}{\epsilon }\right)\nu\left\vert f-\mathbb{E}^{\{\sim k\}}f
\right\vert^q \\ &
\nonumber +c_1\sum_{r=u-1,u+1}\nu\left\vert \nabla_r
f
\right\vert^q+\frac{J^qCc^2_1}{\epsilon}Q(k,k)+\frac{J^qc_1C}{\epsilon}\sum_{dist(u,t)=4,t\neq k}Q(t,k)\end{aligned}$$
The lemma follows from the bound of $Q(u,k)$ in Lemma \[lem5.3\]. In the case where $dist(u,k)\geq 8$, for $r=u-2,u-1,u,u+1,u+2$ we have that $$\label{5.25}\nu\left\vert \nabla_{r}
h_k
\right\vert^q\leq2^{q-1}2\nu\left\vert \nabla_{r}
f
\right\vert^q$$ Substituting in the expression from Lemma \[lem5.3\] we immediately obtain (a). Consider the case where $dist(u,k)=4$. Then for $r=u-1,u,u+1$ $$\label{5.26}\nu\left\vert \nabla_{r}
h_k
\right\vert^q\leq2^{q-1}2\nu\left\vert \nabla_{r}
f
\right\vert^q$$ While for $r=\{u-2,u+2\}$ we can bound $\nu\left\vert \nabla_{r} h_{k}
\right\vert^q$ from Lemma \[lem5.4\] (a). If we plug the bounds from and Lemma \[lem5.4\] (a) into the expression from Lemma \[lem5.3\], we obtain $$\begin{aligned}
Q(u,k)\leq &
J^q\left(\frac{c_12K}{\epsilon}+\frac{c^2_1K}{\epsilon }\right)\nu\left\vert f-\mathbb{E}^{\{\sim k\}}f
\right\vert^q+
\frac{J^qCc^2_1}{\epsilon}Q(k,k)+c_1\nu\left\vert \nabla_u
f
\right\vert^q\\ &
+c_1\sum_{r= u-1,u+1}\nu\left\vert \nabla_r
f
\right\vert^q+c^2_1\sum_{r=u-2,u+2}\nu\left\vert \nabla_r f
\right\vert^q+ \frac{J^qc_1C}{ \epsilon}\sum_{dist(u,t)=4}Q(t,k)\end{aligned}$$
Before proving Lemma \[lem5.5\], we will also need to show that for any $k\in \mathbb{N}$$$\sup_{n\in\mathbb{N}}\sum_{dist(u,k)=n} Q(u,k)<C_{f}<\infty$$for $C_f$ a constant which depends on the function $f$ but not on $n,u$ and $k$. To show this we first need the following lemma.
\[newlemma0section5\]For any $r,k \in \mathbb{Z}$ we have $$\nu\left\vert \nabla_r h_k
\right\vert^q\leq\tilde C_f<\infty$$ where $\tilde C_f$ depends on the function $f$ but not on $r$ and $k$.
For general $r \in \{k-2,k,k+2\}$ $$\label{neweq1ofnewlemma10}\nu\left\vert \nabla_r h_k
\right\vert^q\leq2^{q-1}\nu\left\vert \nabla_r
f
\right\vert^q+2^{q-1}\nu\left\vert \nabla_r
\mathbb{E}^{\{\sim k\}}f
\right\vert^q$$ since $h_k=f-\mathbb{E}^{\{\sim k\}}f$. For the second term on the right hand side of (\[neweq1ofnewlemma10\]) we have $$\begin{aligned}
\label{neweq2ofnewlemma10}\nu\left\vert \nabla_r
\mathbb{E}^{\{\sim k\}}f
\right\vert^q\leq & 2^{q-1}\nu\left\vert \nabla_{r} f
\right\vert^q+J^q2^{q-1}\nu\left\vert\mathbb{E}^{\{\sim k\}}(f;Z_k)\right\vert^q \end{aligned}$$ where $$Z_k=\nabla_{k-2}V(x_{k-2},x_{k-1})\mathcal{I}_{r=k-1}+\nabla_{k+2}V(x_{k+2},x_{k+1})\mathcal{I}_{r=k+1}+W_{k}\mathcal{I}_{r=k}$$ where $W_k$ as in (\[defineW\_k\]). We will now compute the last term on the right hand side of (\[neweq2ofnewlemma10\]) $$\begin{aligned}
\nu\left\vert\mathbb{E}^{\{\sim k\}}(f;Z_k)\right\vert^q =&\nu\left\vert\mathbb{E}^{\{\sim k\}}(f-\mathbb{E}^{\{\sim k\}}f)(Z_{k}-\mathbb{E}^{\{\sim k\}}Z_k)\right\vert^q \\ =&\nu\left\vert\mathbb{E}^{\{\sim k\}}\left(f(Z_{k}-\mathbb{E}^{\{\sim k\}}Z_k)\right)\right\vert^q \leq \nu f^{q}\vert Z_{k}-\mathbb{E}^{\{\sim k\}}Z_k\vert^q\end{aligned}$$ If we use the entropic inequality (\[3.3\]) we obtain $$\begin{aligned}
\label{neweq3ofnewlemma10}\nu\left\vert\mathbb{E}^{\{\sim k\}}(f;Z_k)\right\vert^q \leq\frac{1}{\epsilon} &\nu f^q\log\frac{f^q}{\nu f^q}+\frac{1}{\epsilon}\nu f^q\log \nu e^{\epsilon\vert Z_{k}-\mathbb{E}^{\{\sim k\}}Z_k\vert^q}\nonumber\\ \leq\frac{1}{\epsilon} &\nu f^q\log\frac{f^q}{\nu f^q}+\frac{K}{\epsilon}\nu f^q\end{aligned}$$ where $K$ as in (H2). If we combine (\[neweq1ofnewlemma10\]), (\[neweq2ofnewlemma10\]) and (\[neweq3ofnewlemma10\]) we get that for $r \in \{k-2,k,k+2\}$ $$\begin{aligned}
\label{neweq3-4ofnewlemma10} \nu\left\vert \nabla_r h_k
\right\vert^q\leq2^{q}\nu\left\vert \nabla_r
f
\right\vert^q+\frac{J^q 2^{2q-2}}{\epsilon} \nu f^q\log\frac{f^q}{\nu f^q}+\frac{J^q 2^{2q-2}}{\epsilon}\nu f^q\end{aligned}$$ For $r \notin \{k-2,k,k+2\}$ we have $$\label{neweq3-4+1ofnewlemma10}\nu\left\vert \nabla_r h_k
\right\vert^q\leq2^{q} \nu f^q$$ From (\[neweq3-4ofnewlemma10\]) and (\[neweq3-4+1ofnewlemma10\]) the lemma follows since functions $f$ are as in Remark \[newremark1\].
\[newlemmasection5\]If (H2) is satisfied, then for any $k\in \mathbb{N}$$$\sup_{n\in\mathbb{N}}\sum_{dist(u,k)=n} Q(u,k)<C_{f}<\infty$$where $C_f$ is a constant which depends on the function $f$ but not on $u$ and $k$.
Since we work on the one dimensional lattice, it is sufficient to show that $$\sup_{n\in\mathbb{N}} Q(u,k)<C'_{f}<\infty$$ for $C'_f$ depends only on the functions $f$. To compute $Q(u,k)$ we can use (\[5.14\]) and (\[5.15\]) to obtain$$\begin{aligned}
\label{neweq4ofnewlemma10}Q(u,k)\leq\sum_{r= u-1,u,u+1}\nu\left\vert \nabla_r
h_k
\right\vert^q
+\sum_{r= u-2,u+2}\nu_{\Lambda(u)}\left\vert \nabla_{r} (\mathbb{E}^{M(u)}
\vert h_k\vert^q)^{\frac{1}{q}}
\right\vert^q \end{aligned}$$ Furthermore, from (\[5.16\]) for $r=u-2,u+2$ we have $$\begin{aligned}
\label{neweq6ofnewlemma10}\nu_{\Lambda(u)}&\left\vert \nabla_{r} (\mathbb{E}^{M(u)}
\vert h_k\vert^q)^{\frac{1}{q}}
\right\vert^q\leq c_1\nu\left\vert \nabla_{r} h_{k}
\right\vert^q+
\frac{J^qc_1}{q^q}I_0\end{aligned}$$ where $$I_0 :=\nu_{\Lambda(u)}\left(\mathbb{E}^{M(u)}\vert h_k\vert^{q}\right)^{-\frac{q}{p}}\left(\mathbb{E}^{M(u)}(\vert h_k\vert^q;\nabla_rV(x_{r},x_{s}))\right)^q
\mathcal{I}_{s\in \{u-3,u+3\}:s\sim r}$$ In order to bound the second term on the right hand side of (\[neweq6ofnewlemma10\]) we compute $$\begin{aligned}
\mathbb{E}^{M(u)}&(\vert h_k\vert^q;\nabla_rV(x_{r},x_{s}))=\mathbb{E}^{M(u)}\left(\vert h_k\vert^{(q-1)+1}\left(\nabla_rV(x_{r},x_{s})-\mathbb{E}^{M(u)}\nabla_rV(x_{r},x_{s})\right)\right)\\&\leq\left(\mathbb{E}^{M(u)}\vert h_k\vert^{pq-p}\right)^{\frac{1}{p}} \left( \mathbb{E}^{M(u)}\left(\vert h_k\vert^{q}\left\vert\nabla_rV(x_{r},x_{s})-\mathbb{E}^{M(u)}\nabla_rV(x_{r},x_{s})\right\vert^q\right) \right)^\frac{1}{q}\end{aligned}$$ From the last bound, since $p$ and $q$ are conjugate, we get $$\begin{aligned}
I_{0}& \leq\nu_{\Lambda(u)}\mathbb{E}^{M(u)}\left(\vert h_k\vert^{q}\left\vert\nabla_rV(x_{r},x_{s})-\mathbb{E}^{M(u)}\nabla_rV(x_{r},x_{s})\right\vert^q\right)
\mathcal{I}_{s\in \{u-3,u+3\}:s\sim r}\\&=\nu(\vert h_k\vert^{q}N_r)\leq 2^{q-1}\nu (f^{q}N_r)+2^{q-1}\nu( (\mathbb{E}^{\{\sim k\}} f^{q})N_r)\end{aligned}$$ where above we denoted $N_{r}=\left\vert\nabla_rV(x_{r},x_{s})-\mathbb{E}^{M(u)}\nabla_r V(x_{r},x_{s})\right\vert^q
\mathcal{I}_{s\in \{u-3,u+3\}:s\sim r}$. If we use again the entropic inequality (\[3.3\]) we obtain $$\begin{aligned}
\nonumber\label{neweq7ofnewlemma10}I_{0}\nonumber \leq&\frac{2^{q-1} }{\epsilon}\nu f^{q}\log \frac{f^{q}}{\nu f^{q}}+\frac{2^{q-1} }{\epsilon}\nu f^{q}\log \nu e^{\epsilon N_r}+\frac{2^{q-1} }{\epsilon}\nu\mathbb{E}^{\{\sim k\}} f^{q}\log\frac{ \mathbb{E}^{\{\sim k\}} f^{q}}{\nu \mathbb{E}^{\{\sim k\}} f^{q}}\\ \nonumber &+\frac{2^{q-1} }{\epsilon}\log \nu e^{\epsilon N_r}\nu \mathbb{E}^{\{\sim k\}} f^{q}\\ \leq& \frac{2^{q-1} }{\epsilon}\nu f^{q}\log \frac{f^{q}}{\nu f^{q}}+\frac{2^{q} K }{\epsilon}\nu f^{q}+\frac{2^{q-1} }{\epsilon}\nu\mathbb{E}^{\{\sim k\}} f^{q}\log\frac{ \mathbb{E}^{\{\sim k\}} f^{q}}{\nu \mathbb{E}^{\{\sim k\}} f^{q}}\end{aligned}$$ where $K$ as in (H2). For the last term on the right hand side of (\[neweq7ofnewlemma10\]) we can write $$\begin{aligned}
\label{neweq7+1ofnewlemma10}\nu\mathbb{E}^{\{\sim k\}} f^{q}\log\frac{ \mathbb{E}^{\{\sim k\}} f^{q}}{\nu \mathbb{E}^{\{\sim k\}} f^{q}}=\nu f^{q}\log\frac{ \mathbb{E}^{\{\sim k\}} f^{q}}{\nu \mathbb{E}^{\{\sim k\}} f^{q}}\leq\nu f^{q}\log\frac{ f^{q}}{\nu f^{q}}\end{aligned}$$ Combining together (\[neweq7ofnewlemma10\]) and (\[neweq7+1ofnewlemma10\]) we obtain $$\begin{aligned}
\label{neweq8ofnewlemma10} I_{0}\leq\frac{2^{q} }{\epsilon}\nu f^{q}\log \frac{f^{q}}{\nu f^{q}}+\frac{2^{q} K }{\epsilon}\nu f^{q}\end{aligned}$$ From (\[neweq6ofnewlemma10\]), and (\[neweq8ofnewlemma10\]) we then get that for $r=u-2,u+2$ $$\begin{aligned}
\label{neweq9ofnewlemma10}\nu_{\Lambda(u)}\left\vert \nabla_{r} (\mathbb{E}^{M(u)}
\vert h_k\vert^q)^{\frac{1}{q}}
\right\vert^q\leq c_1\nu\left\vert \nabla_{r} h_{k}
\right\vert^q+
\frac{J^q2^{q}c_1}{q^q\epsilon}\nu f^{q}\log \frac{f^{q}}{\nu f^{q}}+\frac{J^qc_12^{q} K }{q^{q}\epsilon}\nu f^{q}\end{aligned}$$ If we combine (\[neweq4ofnewlemma10\]) and (\[neweq9ofnewlemma10\]) together with Lemma \[newlemma0section5\] we conclude that for any function $f$ there is a bound of $\nu_{\Lambda(u)}\left\vert \nabla_{r} (\mathbb{E}^{M(u)}
\vert h_k\vert^q)^{\frac{1}{q}}
\right\vert^q$ uniformly with respect to the set $M(u)$ depending only on $\nu f^{q}$, $\max_{i\in \mathbb{Z}}\nu \left \vert \nabla _i f \right\vert^q$ and $\nu f^{q}\log \frac{f^{q}}{\nu f^{q}}$.
We can now prove Lemma \[lem5.5\].
***Proof of Lemma \[lem5.5\].*** For every $u$ s.t. $dist(u,k)\geq8$ define $$\begin{aligned}
G(u,k):=& c_1\nu_{\Lambda(u)}\left\vert \nabla_u
f
\right\vert^q+c_1\sum_{r= u-1,u+1}\nu_{}\left\vert \nabla_r
f
\right\vert^q \\ &
+c^2_1\sum_{r=u-2,u+2}\nu_{\Lambda(u)}\left\vert \nabla_r f
\right\vert^q+\frac{J^qc_12K}{\epsilon}\nu\left\vert f-\mathbb{E}^{\{\sim k\}}f
\right\vert^q\end{aligned}$$ and for every $u$ s.t. $dist(u,k)=4$ define $$\begin{aligned}
G(u,k):=&c_1\nu\left\vert \nabla_{u}
f
\right\vert^q+c_1\sum_{r= u-1,u+1}\nu\left\vert \nabla_u
f
\right\vert^q+\frac{J^qCc^2_1}{\epsilon}Q(k,k)\\ &
+c^2_1\sum_{i=u-2,u+2}\nu\left\vert \nabla_r f
\right\vert^q+J^q\left(\frac{c_12K}{\epsilon}+\frac{c^2_1K}{\epsilon }\right)\nu\left\vert f-\mathbb{E}^{\{\sim k\}}f
\right\vert^q\end{aligned}$$ If we set $K'=\frac{c_{1}C}{ \epsilon}$, then from Lemma \[lem5.6\] (a) and (b) respectively we can write $$\label{5.28}Q(u,k)\leq G(u,k)+ J^qK'\sum_{dist(u,t)=4}Q(t,k) \text{,
\; for } dist(u,k)\geq8$$ and $$\label{5.29}Q(u,k)\leq G(u,k)+ J^qK'Q(t,k)\mathcal{I}_{dist(t,u)=4,t\neq k} \text{, \;
for } dist(u,k)=4$$ From equation we obtain $$\sum_{dist(u,k)=n} Q(u,k)\leq \sum_{dist(u,k)=n}G(u,k)+ J^qK'\sum_{dist(u,k)=n}\sum_{dist(t, u)=4}Q(t,k)$$ or equivalently $$\begin{aligned}
\sum_{dist(u,k)=n} Q(u,k)\leq &\sum_{dist(u,k)=n}G(u,k)+ J^qK'\sum_{dist(t,k)=n+4}Q(t,k)\\
& +J^qK'\sum_{dist(t,k)=n-4}Q(t,k)\end{aligned}$$ which implies $$\label{5.30}\tilde{Q}(n)\leq \tilde{G}(n)+J^q K'\tilde{Q}(n-4)+J^qK'\tilde{Q}(n+4)$$ where we denote $$\tilde{Q}(n)=\sum_{dist(u,k)=n} Q(u,k)\text{\; and \;}\tilde{G}(n)= \sum_{dist(u,k)=n}G(u,k)$$ While from equation , we have $$\sum_{dist(u,k)=4} Q(u,k)\leq \sum_{dist(u,k)=4}G(u,k)+ J^qK'\sum_{dist(u,k)=4}Q(t,k)\mathcal{I}_{dist(t,u)=4,t\neq k}$$ This implies $$\sum_{dist(u,k)=4} Q(u,k)\leq \sum_{dist(u,k)=4}G(u,k)+ J^qK'\sum_{dist(t,k)=8}Q(t,k)$$ which is equivalent to $$\label{5.31}\tilde{Q}(4)\leq \tilde{G}(4)+J^qK'\tilde{Q}(8)$$ Choose $J$ in (H3) sufficiently small such that hypothesis of Lemma \[lem5.2\] is satisfied. Then, since relationships , and Lemma \[newlemmasection5\] are true, the conditions of Lemma \[lem5.2\] are satisfied for $P=\tilde{ Q}$ and $G=\tilde{ G}$ and so we obtain $$\nonumber \tilde{Q}(4)\leq \hat{ J}\sum_{n=0}^{+\infty} J^{nq-n}\tilde{G}(4n+4)$$ where $\hat{ J}=\frac{1}{1-J^{2q-2}}$. This is equivalent to $$\begin{aligned}
\sum_{t:dist(t,k)=4}Q(t,k)\leq &\hat{ J} \sum_{dist(u,k)=4}G(u,k)\nonumber\\
& \label{5.32}+ \hat{ J}\sum_{n=1}^{+\infty} J^{nq-n}\sum_{dist(u,k)=4n+4}G(u,k)\end{aligned}$$ Substituting $G(u,k)$ leads to $$\begin{aligned}
\nonumber\sum_{t:dist(t,k)=4}Q&(t,k)\leq\frac{J^qCc^{2}_{1}}{\epsilon }\hat J \sum_{dist(u,k)=4}Q(k,k)\\ &
\nonumber + \hat{ J}c_1\sum_{n=0}^{+\infty} J^{nq-n}\sum_{dist(u,k)=4n+4} \nu_{\Lambda(u)}\left\vert \nabla_{u}
f
\right\vert^q\\ &
+ \hat{ J}c_1\sum_{n=0}^{+\infty} J^{nq-n}\sum_{dist(u,k)=4n+4}\sum_{r= u-1,u+1}\nu_{\Lambda(u)}\left\vert \nabla_{r}
f
\right\vert^q\nonumber\\ \nonumber &
+\hat{ J}c^2_1\sum_{n=0}^{+\infty} J^{nq-n}\sum_{dist(u,k)=4n+4}\sum_{r= u-2,u+2}\nu_{\Lambda(u)}\left\vert \nabla_{r} f
\right\vert^q\\ &
\label{5.33}+J^q\hat J\frac{c_1K}{\epsilon}(c_{1}+2)\sum_{n=0}^{+\infty} J^{nq-n}\sum_{dist(u,k)=4n+4}\nu\left\vert f-\mathbb{E}^{\{\sim k\}}f
\right\vert^q\end{aligned}$$ But for $J$ in (H3) we have $J^{q-1}<1$ which implies $\tilde{J}=\sum_{n=0}^{+\infty} J^{nq-n}<\infty$. then implies
$\sum_{t:dist(t,k)=4}Q(t,k)\leq\frac{J^qCc^{2}_{1}}{\epsilon}2\hat{ J} Q(k,k)$
$ \ \ \ \ \ \ \ \ \ +\hat{J}c_1\tilde{J}\sum_{dist(u,k)=4n+4}\sum_{r= u-1,u,u+1}\nu\left\vert \nabla_{r}
f
\right\vert^q$
$ \ \ \ \ \ \ \ \ \ +\hat{J}c^2_1\tilde{J}\sum_{dist(u,k)=4n+4}\sum_{r= u-2,u+2}\nu\left\vert \nabla_{r} f
\right\vert^q$
$ \ \ \ \ \ \ \ \ +J^{q}\frac{c_1K}{\epsilon}(c_{1}+2)\hat{ J}\frac{2}{1-J^{q-1}}\nu\left\vert f-\mathbb{E}^{\{\sim k\}}f
\right\vert^q $
and the lemma follows for appropriate constant $T>0$.
Conclusion
==========
In the present work, we have determined conditions for the infinite volume Gibbs measure to satisfy the Log-Sobolev Inequality. As explained in the introduction, the criterion presented in Theorem \[thm2.1\] can in particular be applied in the case of local specifications $\{\mathbb{E}^{\Lambda,\omega}\}_{\Lambda\subset\subset \mathbb{Z} ,\omega \in
\Omega}$ with no quadratic interactions for which $$\left\Vert \nabla_i \nabla_j V(x_i,x_j) \right\Vert_{\infty}=\infty$$ Thus, we have shown that our results can go beyond the usual uniform boundness of the second derivative of the interactions considered in \[Z1\], \[Z2\], \[M\] and \[O-R\].
Concerning the additional conditions (H1) and (H2) placed here to handle the exotic interactions, they refer to finite dimensional measures with no boundary conditions which are easier to handle than the $\{\mathbb{E}^{\Lambda,\omega}\}_{\Lambda\subset\subset \mathbb{Z} ,\omega \in
\Omega}$ measures or the infinite dimensional Gibbs measure $\nu$.
In fact, the following results concerning the conditions can be proven. This is a work in progress that will consist the material of a forthcoming paper.
\[conclProp\]The hypothesis (H0), (H3) and (H2) imply hypothesis (H1).
Consequently, the main result of Theorem \[thm2.1\] is then reduced to the following
\[conclTheor\]If hypothesis (H0), (H3) and (H2) are satisfied, then the infinite dimensional Gibbs measure $\nu$ for the local specification $\{\mathbb{E}^{\Lambda,\omega}\}_{\Lambda\subset\subset \mathbb{Z},\omega \in
\Omega}$ satisfies the $q$ Log-Sobolev inequality $$\nu \left\vert f\right\vert^q log\frac{\left\vert f\right\vert^q}{\nu \left\vert f\right\vert^q}\leq \mathfrak{C} \ \nu \left\vert \nabla f
\right\vert^q$$ for some positive constant $\mathfrak{C}$ independent of $f$.
Concerning examples of measures that satisfy the above conditions, one can consider measures with phase $\phi (x)=\vert x\vert^t$ with $t\geq \frac{q}{q-1}$ and interaction $V(x,y)=\vert x-y\vert^r$, with $\max\{r,(r-1)q\}<t$. The main idea of the proof of the Proposition \[conclProp\] follows in main lines the method followed in the current paper. Although some of the details are more involved because of the lack of hypothesis (H1), the fact that in Proposition \[conclProp\] the Gibbs measure is localised and thus the approximation procedure starts from a finite set compensates for the loss of the LSq for $\nu_{\Lambda (i)}$.
In this paper we have been concerned with the $q$ Logarithmic Sobolev inequality for measures on the 1 dimensional Lattice $\mathbb{Z}$. It is interesting to try to extend the current result to a higher dimensional lattice on $\mathbb{Z}^d,d\geq 2$, although this does not appear to be immediate. In a different direction, we can consider the following class of modified Logarithmic Sobolev inequalities presented in \[G-G-M\]: $$\label{modeq}\nu \left\vert f\right\vert^2 log\frac{\left\vert f\right\vert^2}{\nu \left\vert f\right\vert^2}\leq \mathfrak{C}\ \mathfrak{}\ \int H_{a,c} \left( \frac{ \nabla f
}{f}\right)f^2d\nu$$ for some positive constant $\mathfrak{C}$, where $$H_{a,c}( x)=\begin{cases}\frac{x^2}{2} & \text{if \ } \vert x\vert \leq a \\
a^{2-\beta} \frac{\vert x \vert ^{\beta}}{\beta}+a^2\frac{\beta-2}{2\beta}& \text{if \ }\vert x\vert \geq\ a \text{\ and \ }c \neq1\\
+\infty & \text{if \ }\vert x\vert \geq\ a \text{\ and \ }c =1
\end{cases}$$ for $c\in [1,2], a>0$ and $\beta$ satisfying $\frac{1}{c}+\frac{1}{\beta}=1$ ($\beta\geq 2$). This new class of inequalities is an interpolation between Log-Sobolev (LS2) and Spectral Gap inequalities (SG2), which retains the basic properties of the Log-Sobolev inequalities mentioned in Remark \[rem1.1\]. Some preliminary results suggest that on $\mathbb{Z}^d,d\geq2$, the infinite dimensional Gibbs measure satisfies a \[G-G-M\] type inequality with $\beta=2q$, under hypothesis (H0) for LSq ($1< q< 2$) and some hypothesis stronger than (H2). This is work in early stages, but hopefully a modified LS inequality comparable to the \[G-G-M\] inequalities can be obtained in the case of the higher dimensional lattice.
In addition, it is interesting to investigate whether the result presented in this paper can be extended to the family of weaker inequalities presented in \[G-G-M\], assuming (H0) and (H1) for the (\[modeq\]) inequality instead of the LSq. However, this does not seem to be immediate especially in showing the sweeping out relationships and so more work needs to be done towards this direction.
Furthermore, concerning the hypothesis on the single-site measure, the main hypothesis (H0) for $\mathbb{E}^{\{i\},\omega}$ can be reduced to the same assumption for the boundary free single-site measure, that is
**(H0$'$)**: The single-site measure $\frac{e^{-\phi (x)}dx}{\int e^{-\phi (x)}dx}$ satisfies the LSq Inequality.
Measures as in ($H0'$) do not involve boundary conditions and for this reason it is easier to show that they satisfy the Log-Sobolev inequality. For instance, when in $\mathbb{R}$ one can think of phases that are convex and increase sufficiently fast, like $\phi(x)=\left\vert x\right\vert^p$ for $p>2$ (see \[B-Z\]). In the case of the Heisenberg group $\mathbb{}\mathbb{H}$ one can consider $\phi(x)=\beta d(x)^p$ with $p$ conjugate of $q$ (see \[H-Z\]).
However, that does not mean that condition (H0$'$) is in general weaker than condition (H0) as there are examples of single-site boundary free measures $\frac{e^{-\phi (x)}dx}{\int e^{-\phi (x)}dx}$ that do not satisfy the LS$q$ inequality, which when perturbed with interactions, give new measures $\mathbb{E}^{\{i\},\omega}$ that satisfy the Log-Sobolev-q inequality uniformly on the boundary conditions, that is condition (H0) is satisfied. In addition, in the case of hypothesis ($H0'$), it seems that the analogues of Proposition \[conclProp\] and Theorem \[conclTheor\] will be more to difficult to be shown.
**Acknowledgements:** The author would like to thank Prof. Boguslaw Zegarlinski for his valuable comments and suggestions.
[Dillo 83]{}
, [*Correlation Asymptotics of Classical Lattice spin Systems with Nonconvex Hamilton function at Low Temperature*]{}, Ann. Henri Poincare, 1, 59-100 (2000) , [*Diffusions hypercontractives* ]{},Séminaire de Probabilités XIX, Springer Lecture Notes in Math., 1123, 177-206 (1985) , Compactness and the maximal Gibbs state for random fields on the Lattice, Commun. Math. Phys., 84, 297-327 (1982). , [*Exponential integrability and transportation cost related to logarithmic sobolev inequalities* ]{}, J of Funct Anal., 163, 1-28 (1999)
, [*From Brunn-Minkowski to Brascamp-Lieb and to Logarithmic Sobolev Inequalities*]{}, Geom. funct. anal., 10, 1028-1052 (2000).
, [*Entropy Bounds and Isoperimetry.*]{} Memoirs of the American Mathematical Society, 176, 1 - 69 (2005)
, [*Log-Sobolev inequality for unbounded spin systems* ]{}, J of Funct Anal., 166, 168-178 (1999) , [*Large Deviations* ]{}, Academic Press, San Diego, (1989).
, [*The problem of uniqueness of a Gibbs random field and the problem of phase transition*]{}, Funct. Anal. Apll., 2, 302-312 (1968). , [*Modified logarithmic Sobolev inequalities and transportation inequalities*]{}, Probab.Theor. Relat. Fields 133, 409-436 (2005).
, [*Spectral Gaps for Spin Systems: Some Non-convex Phase Examples*]{}, J. Func. Anal., 180, 66-84 (2001). , [*Logarithmic Sobolev inequalities*]{}, Am. J. Math., 97, 1061-1083 (1976)
, [*Lectures on Logarithmic Sobolev Inequalities*]{}, IHP Course 98, Séminaire de Probabilités XXVI, Lecture Notes in Mathematics 1801, Springer, 1-134 (2003).
, [*Semiclassical Analysis, Witten Laplacians and Statistical Mechanics*]{}. Partial Differential Equations and Applications. World Scientific, Singapore (2002)
, Coercive inequalities on metric measure spaces. J of Funct Anal. (to appear) , [*Logarithmic Sobolev Inequalities for Infinite Dimensional Hörmander Type Generators on the Heisenberg Group*]{}. Potential Anal., 31, 79-102 (2009) , [*Concentration of measure and logarithmic Sobolev inequalities*]{}, Séminaire de Probabilités, XXXIII, Lecture Notes in Math. 1709, Springer-Verlag, 120-216. (1999) , [*Logarithmic Sobolev Inequality for Weakly Dependent Random Variables.*]{} (preprint)
, [*On the spectrum of semi-classical Witten-Laplacians and Schrödinger operators in large dimension*]{}, J of Funct Anal, Volume 220 (2), 243-264 (2005)
, [*A new criterion for the Logarithmic Sobolev Inequality and two Applications* ]{}, J. Func. Anal., 243, 121-157 (2007).
,[*Random Fields*]{}, LNM 534, Springer (1976) , [*Orlicz-Sobolev inequalities for sub-Gaussian measures and ergodicity of Markov semi-groups* ]{}, J. Func. Anal., 243 (1), 28–66 (2007).
, [*The log-Sobolev inequality for weakly coupled lattice field*]{}, Probab.Theor. Relat. Fields 115 , 1-40 (1999)
, [*On log-Sobolev Inequalities for Infinite Lattice Systems*]{}, Lett. Math. Phys. 20, 173-182 (1990)
, [*The strong decay to equilibrium for the stochastic dynamics of unbounded spin systems on a lattice*]{}, Comm. Math. Phys. 175, 401-432 (1996)
[^1]: Department of Mathematics, Imperial College London, 180 Queen’s Gate, London, SW7 2AZ. Email: [email protected]
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We define one-dimensional particles as non-abelian representations of the symmetric group $S_N$. The exact solution of an $XXZ$ type Hamiltonian built up with such particles is achieved using the coordinate Bethe Ansatz. The Bethe equations show that fractional statistics, effectively, accounts for coupling an external gauge field to an integer statistics’ system.'
address:
- '$^{\dagger}$Institut für Physik,Universität Augsburg, D-86135 Augsburg, Germany'
- '$^{\natural}$ Istituto di Fisica, Facoltá di Ingegneria, Università di Catania $\&$ $INFM$, viale A. Doria 6, I-95129 Catania, Italy'
author:
- 'Luigi Amico$^{\dagger,\natural}$, Andreas Osterloh$^{\dagger}$, and Ulrich Eckern$^{\dagger}$'
title: 'One-Dimensional ${\bf XXZ}$ Model for Particles Obeying Fractional Statistics'
---
PACS Numbers: 71.10.Pm, 71.27.+a, 75.10.Jm
[2]{} Physical behaviour of quantum systems is deeply affected by the statistics of the constituting effective degrees of freedom. Quasi-particles and quasi-holes in condensed matter physics may obey statistics interpolating between fermionic and bosonic behaviour. Examples are the excitations of two-dimensional electron systems exhibiting Fractional Quantum Hall effect [@HALL]. These excitations are called [*anyons*]{}. They have been a subject of intense study also in connection with superconductivity [@WIL] and superfluidity [@VOLOVIK]. Fractional statistics of such particles arises from the trajectory-dependence of the particle exchange procedure in the two-dimensional configuration space. This feature makes the concept of anyons purely two-dimensional. The Fock space formulation of anyon operator algebras takes into account these characteristics. The creation and annihilation operators (introduced as Jordan-Wigner transforms of usual fermions on a two-dimensional lattice[@LERDASCIUTO] or as unitary representations of the diffeomorphism group of $\RR^2$ [@GOLDIN]) obey deformed commutation relations if the exchange involves anyons at different spatial positions (see Appendix). $N$-anyon-states are abelian representations of the braid group $B_{N}$ [@LERDA] (whereas bosons and fermions furnish, respectively, the identical and alternating abelian representations of the symmetric group $S_{N}$). These features make anyons different from $q$-oscillators, the latter providing a realization of Gel’fand-Farlie quantum group, which is a [*local*]{} deformation of the Weyl-Heisenberg (bosons) or Clifford algebra (fermions) [@GELFAND]. The path dependence implies that the one-particle state is inextricably related with the complete state of the many body configuration. This intrinsic non-locality makes anyon physics very difficult. Even statistical properties of a free anyon gas are only partially established using the virial expansion [@VIRIAL].
Haldane [@HALDANE] formulated the notion of fractional statistics without any reference to the spatial dimension $D$. The generalized Pauli principle is expressed in terms of the reduction of the single-particle Hilbert space when particles are added to a many body system keeping boundary conditions fixed. Another way to introduce dimensionality-independent fractional statistics has been formulated in Ref. where [*quons*]{} have been introduced. Quons’ fractional statistics results from the “superposition” of statistical properties of bosons and fermions [@WU]. In $D>2$, this is consistent with spin-statistics theorem. In $2D$, Haldane particles and quons capture the essential features of anyons [@HALDANE; @GOLDIN].
Recently, an outgrowing interest has been devoted to generalized statistics in one dimension. A specific way to introduce $D=1$ fractional statistics has been proposed in connection to the quantization of the solutions of the Calogero model [@CALOGERO]. There, the potential $1/x^{2}$ is interpreted as “statistics interaction”. The same notion of fractional statistics applies also to anyons in a strong magnetic field that restricts the allowed energies to the lowest Landau level. The anyon gas, then, is described by an effective field theory on a ring where the dynamics of particles is one-dimensional [@1DANY]. It is worthwhile noting that such one-dimensional particles obey fractional statistics, but they are not “true anyons” since in $D \neq 2$ trajectories in the particle configuration space have no meaningful braiding property. Instead, nonlocal “deformations” of the commutation relations furnish [*non-abelian*]{} representations of the symmetric group $S_{N}$.
In this paper, we deal with particles in $D=1$ that preserve the intrinsic non-locality of two-dimensional anyons, but which are still representations of $S_{N}$. This representation is no longer abelian. The second quantized formalism and the Fock space representation is developed. The $XXZ$ model for such particles is formulated and solved exactly using coordinate Bethe Ansatz (BA) [@BETHE] in $D=1$.
For $D=1$ we define a set of creation/annihilation operators $\mbox{$\{f_i^{\dagger}\, ,\, f_i\}$}$ for a spinless particle at site $i$. They obey the deformed relations
$$\begin{aligned}
f^\dagger_j f_{{ k}} + q_{{ j}, {k}}\, f_{{ k}} f^\dagger_j &=&
\delta_{{j}, {k}} \quad ,
\label{any1}
\\
f_{ j} f_{ k} + q^{-1}_{j,k} \, f_{k } f_{ j} &=& 0 \qquad ,
\label{any2} \end{aligned}$$
where $q^{-1}_{j,k}\doteq (q_{j,k})^{-1}$. Since the operators are path-independent in $D\neq 2$ (compare Appendix), (\[any1\]) and (\[any2\]) have to constitute a representation of $S_N$, and not of $B_N$. This is ensured by the “consistency relations” $$q_{{ j}, { k}} = q^{-1}_{{ k}, { j}} = q^{\dagger}_{k,j} \quad ,
\label{CONS1}$$
$$[ {f}_j f^{\dagger}_k , q_{j,k} ] =0 \quad .
\label{CONS2}$$
Such a representation is non-abelian [@LERDA]. For $j=k$, (\[CONS1\]) gives $q_{j,j}=\pm\id $. Hence (\[any1\]), (\[any2\]) are an extension of anyon commutation relations [@LERDASCIUTO; @GOLDIN] to $D \neq 2 $. In contrast to the true anyonic case (see Appendix), $q_{j,k}$ has no relation with the configuration space geometry, but is a free “external” parameter.
Relations (\[any1\]) and (\[any2\]) are formally analog to quon commutation rules [@WU]. Note that the deformation parameter here depends on two indices $(j,k)$, whereas it does not in quon commutation rules. Without this index-dependence, relation (\[CONS1\]) directly implies $q^2=\id$. As a consequence, quons obey integer statistics in $1D$ if $q$ is a $\CC$-number [@WU] (as an operator, it has eigenvalues $\pm 1$). If $\mbox{$q_{j,k}=\pm \id \quad \forall \ (j,k)$}$, then (\[any1\]), (\[any2\]) describe spinless fermions or bosons, respectively. However, for application we choose $q_{j,k}$ being $\CC$-numbers and $q_{j,j} = 1 $, see (\[our-q\]), which implies the Pauli exclusion principle, as for spinless electrons or hard core bosons.
Relations (\[CONS1\]), (\[CONS2\]) hold if $q_{j,k}$ is an operator commuting or anticommuting with both, ${f}_j$ and $f^{\dagger}_k $. For this reason we add this as a postulate [@NOTECOMMUTE]
$$\label{postulate-rel}
[ f^\dagger_{k}\, ,\, q_{j,k}] \ =\ \left [ f_j \, ,\, q_{j,k}\right ] \ = \ 0 \quad .$$
The introduction of two indices for the deformation parameter allows the construction of consistent commutation relations even for $q_{j,k}$ being $\CC$–numbers. We make use of this possibility in (\[our-q\]).
To develop a Fock representation of the algebra (\[any1\]), (\[any2\]), we take $\nu_{{ j}}\doteq f^\dagger_j f_{j}$ as number operators. Relations (\[CONS1\]), (\[CONS2\]), and (\[postulate-rel\]) yield commutators of $\nu_{{ j}}$ and $f^{\dagger}_{j}$, $f_j$ being unaffected by the deformation parameter $q_{j,k}$:
$$\label{NUMBERCOMM}
[ \nu_{j}, \nu_{k} ]=0 \: ,\;
[ \nu_{j}, f^{\dagger}_{k} ]=
\delta_{j,k} f^\dagger_{k} \: ,\;
[ \nu_{j}, f_{k}]=-
\delta_{j,k} f_{k}\;.$$
Moreover, the property $q_{j,j}=\id $ implies that number operators are idempotent: $(\nu_{j})^2=\nu_{j}$. Because of (\[NUMBERCOMM\]) the one-particle Fock representation of the algebra (\[any1\]), (\[any2\]) is unaffected by $q_{j,k}$. Instead, the action of $ f_l, \, f_l^\dagger, \, \nu_l $ on the $N$-particle state $|n_{1}\dots n_{N}\rangle$ is deformed according to $$\begin{aligned}
f_l &|n_{1} .. n_{N}\rangle = & {\scriptstyle (-)^{l-1}\delta_{n_l,1}}
\left [ \prod_{k=1}^{l-1} q_{l,k} \right ]
|n_{1}..n_l-1..n_{N}\rangle\ , \nonumber \\
f_l^{\dagger} &|n_{1} .. n_{N}\rangle = & {\scriptstyle (-)^{l-1}
\delta_{n_l,0}}
\left [ \prod_{k=1}^{l-1} q_{k,l} \right ]
|n_{1}..n_l+1..n_{N}\rangle \ , \nonumber \\
\nu_l &|n_{1} .. n_{N}\rangle = & n_l\, |n_{1}.. n_{N}\rangle \ ,
\label{FOCKREPRE}\end{aligned}$$ where $n_l\in \{ 0,1\}$. Equations (\[FOCKREPRE\]) generalize the corresponding relations fulfilled by integer statistics particles [@NEGELE] characterized by $\prod_{k=1}^{l-1} q_{l,k}=(\pm)^{l-1}$ (for fermions/bosons). An explicit realization of the operators $f_{j}$ in terms of spinless fermionic operators $a_{j}$ is $\mbox{$ f_{{ j}}\doteq {a_{j}}\,\exp\left (- i \sum_{ l}\Phi_{ l} n_{{l}} \right ) $}$, where $\Phi_{l}$ are hermitean operators commuting with fermionic degrees of freedom. By direct calculation, the relations (\[any1\]), (\[any2\]) are obtained by setting $\mbox{$ q_{jk}\doteq\exp{\left [i(\Phi_k-\Phi_j )\right ]}$}$. This realization has been suggested in the Ref. where $\Phi_{k}\equiv p_{k}$, $p_{k}$ being momenta of a phononic bath coupled to fermionic degrees of freedom.
In the following we consider the 1$D$ anisotropic Heisenberg model ($XXZ$ model) of spinless fermions
$$H_{XXZ} = - {t}\, \sum_{i } \, \left ( f^\dagger_{i} f_{{i+1}} +
f^\dagger_{i+1} f_{{i}} \right ) + U\, \sum_{ i}\nu_{i} \nu_{i+1} \; .
\label{DEFXXZ}$$
The $f$-operators obey relations (\[any1\]), (\[any2\]) with $q_{j,k}$ defined in close analogy with anyonic relations (see Appendix)
$$\begin{aligned}
\label{our-q}
q_{j,k}=\left \{ \begin{array}{ll} q \quad & j>k \\
1 \quad & j=k \\
q^{-1} \quad & j<k \end{array} \right. \, , \, \quad q\in\CC\ \, . \end{aligned}$$
We point out that the postulate (\[postulate-rel\]) is fulfilled since $q_{j,k}$ is a $\CC$–number for arbitrary, fixed $(j,k)$. Relations (\[CONS1\]) and (\[our-q\]) imply that $q$ is on the unit circle, [*i.e.*]{} $|q|=1$. Additionally, Periodic Boundary Conditions (PBC) $f_{i+L} \equiv f_i $ are chosen, where $L$ denotes the period. The parameters $t$ and $U$ are the hopping amplitude and the Coulomb interaction strength, respectively.
It is worthwhile mentioning that (\[our-q\]) implies fixing an order on the infinite periodic chain. This order can only be defined locally on the manifold $S^1$, which needs two charts for its description. We choose two “charts” $C_1\doteq \{j_1,..,j_N\}$ and $C_2\doteq\{j_{N-l},..,j_{N},j_1,..,j_{N-l-1}\}$ on the ring thought as a discrete subset of $S^1$. On each chart, the given order is well defined by interpreting them as ordered sets. The intersections between $C_1$ and $C_2$ are $\{j_{N-l}\dots j_N\}$ and $\{j_1\dots j_{N-l-1}\}$. In such sets, the orders defined on $C_1$, $C_2$ are identical. Now, in $D=1$, only nearest neighbour ($n.n.$) exchanges can take place. Thus $q_{j_k,j_l}$ is connected to the $n.n.$ exchange $j_k\leftrightarrow j_l$. On the chart $C_1$, where $j_1<j_N$, the exchange $j_N\leftrightarrow j_1$ is not a $n.n.$ exchange. To allow for $n.n.$ hopping $j_N\leftrightarrow j_1$, we must use $C_2$ on which $j_N<j_1$. This implies $q_{j_N,j_1}=q^{-1}$. The picture depicted above is equivalent to fixing a period $P_0 \doteq \{\,j_1\, ,\, \dots\, ,\, j_1 + L\, \}$ on the infinite periodic chain. Consistency of the PBC with this induced order is given if the results are independent of $P_0$. In the sequel it will be seen that this condition is fulfilled.
The correspondence between (\[DEFXXZ\]) and the deformed anisotropic Heisenberg model can be established by $\mbox{$S_j^{(+)} \doteq f^{\dagger}_j$}$, $\mbox{$S_j^{(-)}\doteq f_j$}$ and $\mbox{$ S_j^{(z)}\doteq 1/2 -\nu_{j}$}$. On-site, the operators $S_j^{(z)}, S_j^{(\pm)}$ generate the fundamental representation (spin $s=1/2$) of $su(2)$, but for $j \neq k$
$$\begin{aligned}
\left [ S_j^{(+)}, S_k^{(-)} \right ]& =&
(1+q_{j,k}) \; S_k^{(+)}\, S_j^{(-)} -\delta_{j,k} \ ,\nonumber \\
\left [ S_j^{(-)}, S_k^{(-)} \right ]& =&
(1+q_{j,k}) \; S_k^{(-)}\, S_j^{(-)} -\delta_{j,k} \ ,\nonumber \\
\left [ S_j^{(z)}, S_k^{(\pm )} \right ]& =& 0 \ .\end{aligned}$$
We now show that the $XXZ$ model (\[DEFXXZ\]) is exactly solvable by means of the coordinate BA. The general $N$-particle state on a chain with $L$ sites can be written as
$$|\Psi \rangle \doteq \sum_{1 \le j_1 < ... < j_N \le L}
\psi \left ( j_1, j_2,... ,j_N \right )\;
f_{j_{1}}^{\, \dagger}f_{j_{2}}^{\, \dagger}
...f_{j_{N}}^{\, \dagger}|0 \rangle \, .$$
The action of $H_{XXZ}$ on $|\Psi \rangle $ , [*i.e.*]{} the eigenvalue equation, then reads
$$\begin{aligned}
&U&\sum_{l=1}^{N-1}\delta_{j_l+1,j_{l+1}} \psi \left (j_1,\dots ,j_N\right )\nonumber \\
&-&t \sum_{l=1}^{N} Q_{l}^{(+)} \psi \left (j_1..j_l -1..j_N\right )
+ Q_{l}^{(-)} \psi \left (j_1..j_l+1..j_N\right )\nonumber \\& &
= E\, \psi \left (j_1,\dots ,j_N\right )\quad ,\end{aligned}$$
where $
Q_l^{(\pm)} \doteq
\prod_{s=1}^{l-1} q_{j_l, j_s} q_{j_s,j_l \mp 1}
$. Using (\[our-q\]) as definition for $q_{j,k}$, it follows that $$Q_{l}^{(-)} = Q_{l}^{(+)} = 1\; .$$ Hence, exactly the same eigenvalue equation as in the ordinary $XXZ$ model is obtained. Consequently, inserting the BA
$$\psi(j_1,...j_N)=\sum_{\pi \in {\it S}_N} A(\pi)\exp\left (i\sum_{m=1}^N j_{m}k_{\pi(m)}\right )\quad ,$$
the energy $E$ and the two–body scattering matrix $S(k\, ,\, k') \doteq -\exp{[-i\theta( k \, , \, k')]}$ for the “deformed” model are identical with the known terms occuring in the usual $XXZ$ model [@YANGYANG]. Imposing PBC, however, yields
$$\begin{aligned}
\psi \left ( j_2,\dots , j_{N}, j_1\right ) &=& \left [\prod_{l=2}^{N} (- q_{j_1,j_l}) \right ]
\psi \left ( j_{1},j_{2},\dots , j_{N}\right ) \nonumber \\
&=&\ (-q)^{N-1}\, \psi \left ( j_{1},j_{2},\dots , j_{N}\right ) \quad .
\label{TWISTPBC}\end{aligned}$$
Equation (\[TWISTPBC\]) shows that the fractional statistics produces a twist in the PBC that modifies the periodicity of the Bethe wave function.
Since the twisting factor $q^{N-1}$ does not depend on $j_1$, the starting point of the chosen period $P_0$, the boundary condition is consistent with our choice of $q_{j,k}$. The twist $q^{N-1}$ does not depend on the particles’ configuration, but on the number of particles only. This is crucial because configuration-dependence of the twisting factor would destroy the solvability of the model, since it modifies the structure of the exponential functions in imposing PBC on the BA wave function (making it impossible to extract from (\[TWISTPBC\]) a relation for the amplitudes $A(\pi)$). So, the coordinate BA solvability of the model (with PBC) demands a careful choice of the functional form of $q_{j,k}$. The choice $q_{j,k}\doteq \exp{[i \delta \, (j-k)]}$ for the $XXZ$ model, for instance, leads to the same structure of the $S$ matrix, but produces incompatible boundary conditions.
Since $|q|=1$, $q=\exp{[i\arg(q)]}$. So a phase shift by multiples of $\arg(q)$ occurs in the BA wave function on the [*r.h.s.*]{} of eq. (\[TWISTPBC\]). The Bethe Equations (BE) are obtained as
$$k_j L\ =\ \arg([-q]^{N-1})\ +\ 2\pi I_j\ -\ \sum_{m=1}^{N}\theta ( k_j\, , k_m ) \ ,
\label{BETHEEQ}$$
where $I_j \in \ZZ$. In the fermionic case, one obtains $k_j L=2\pi (I_j+\frac{N-1}{2})-\sum_{m=1}^{N}\theta ( k_j\, , k_m )$, whereas the hard core bosonic case yields $k_j L=2\pi I_j-\sum_{m=1}^{N}\theta ( k_j\, , k_m )$. In all cases, $I_j \in \ZZ$. Equation (\[BETHEEQ\]) differs from the BE for the ordinary $XXZ$ model in the additive term $\arg([-q]^{N-1})$. This term has its origin in the fractional statistics of the particles, and vanishes for integer statistics. Equations (\[BETHEEQ\]) were obtained for the $1D$ $XXZ$ model on a ring threaded by an external magnetic flux [@BATCHELOR; @SHASTRY]. We recover the BE of identifying $\Phi\equiv \alpha \doteq \arg([-q]^{N-1}) $, $\Phi$ being the magnetic flux in units of $\hbar c/ e$.
The limit $k_N\rightarrow \pi-\mu$, where $\cos(\mu)\doteq U/2t$ for $|U|\leq 2|t|$, in the $N$-th equation of (\[BETHEEQ\]) relates the statistics with the ratio $U/2t$ for the ground state, $$\alpha= \pi \left ( L-N-2I_N+1 \right ) -\mu \left ( L-2N+2 \right ) \; .$$ At half filling ($N/L=1/2$, $I_N=(N-1)/2$) the energy and the total momentum, $\mbox{$ P\doteq \sum_{l=1}^N k_l/L $}$, of the ground state are affected by the statistics’ factor $\alpha$ [@SHASTRY; @YANGYANG] as follows: $$\begin{aligned}
E_0 (\alpha)- E_{0}(0)&=& {{\pi \sin( \mu)}\over{4\mu(\pi- \mu)L}}\; \alpha^2 \ ,\nonumber \\
P&=&\frac{1}{2} \; \alpha \ ,
\label{GROUND}\end{aligned}$$ where $E_{0}(0)$ denotes the ground state energy of the undeformed $XXZ$ model. The same structure of BE has been obtained in . In , the two species of particles, up-spin ($ \sigma=+ $) and down-spin ($\sigma = - $), have a dynamics governed by two distinct $XXZ^{({\sigma})}$ Hamiltonians coupled only via a local gauge field, included in $XXZ^{(\sigma )}$ by a Peierls-like substitution $t\rightarrow t \, W^{({\sigma})}_{m}$, where $W^{({\sigma})}_{m}\doteq \exp{ [i \sigma \sum_{l=1}^L \alpha_{m-l}\,n_{l,-\sigma} ]}$ (determined by the position of all particles of opposite species); $\alpha_l \in \RR$, $\alpha_{m+L}\equiv \alpha_{m}$. A comparison of the BE in with (\[BETHEEQ\]) shows that our deformation parameter $q$ can be interpreted as the “global” coupling constant of the gauge potential by setting $\alpha \sim \sum_{m=1}^L\alpha_{m-l}$. Vice versa, such an interaction produces statistics transmutation. In this sense, our deformed $XXZ$ model belongs to the same class of integrable models introduced in .
In conclusion, we have given a formulation of fractional statistics in one dimension realized by an anyonic-type deformation of the second quantized commutation rules. Coordinate BA solvability of the deformed $XXZ$ model demands a proper choice of the functional form of $q_{j,k}$. The statistics we have chosen in the present paper preserves the Yang-Baxter equation as well as the BA solvability of the undeformed model. The resulting BE are, however, modified. They show that fractional statistics plays the same role as a gauge field coupled to the undeformed model. Systematic investigations of fractional statistics seem interesting for at least two reasons. First, fractional statistics may be an alternative approach to handle complicated interactions between particles obeying integer statistics. Such interactions could be modelled deforming the particles’ statistics. Second, the study of “compatible statistics” could be relevant in order to find integrable Hamiltonians characterized by [*braided*]{} Yang–Baxter Equations (YBE) [@KUNDU]. Such a feature of the YBE could be closely related to actual braiding of particles in two dimensions.
A further development of the present approach is to take spin into account. A preliminary analysis of the “deformed” Hubbard model [@MORA] shows that fractional statistics modifies the $S$ matrix; the $R$ matrix obeys a braided YBE. We will report on this subject in a forthcoming paper.
We thank M. Rasetti for suggesting this line of research as well as for many discussions. We thank D. Bormann, R. Fazio, G. Falci, G. Giaquinta, P. Schwab, and M. Takahashi for fruitful discussions. L.A. acknowledges financial support of “Fondazione A. della Riccia”, and EU TMR Programme (Contract no. FMRX CT 960042). This paper has been written while the authors were members of the [*Graduiertenkolleg*]{} “Nonlinear Problems in Analysis, Geometry, and Physics" (GRK 283) financed by the German Science Foundation (DFG) and the State of Bavaria.
Here we summarize the commutation properties of two dimensional anyons [@LERDASCIUTO; @GOLDIN]. The creation/annihilation operators obey
$$\begin{aligned}
b^\dagger ({\bf x}_C)\, b ({\bf y}_C) +\; q({\bf x}_{C},{\bf y}_{C}) b
({\bf y}_C) \, b^\dagger({\bf x}_C) &=& \delta _{{\bf x}_C,{\bf y}_C}\ ,
\nonumber \\
{b({\bf x}_C)} \, b ({\bf y}_C) + q^{-1}({\bf x}_{C},{\bf y}_{C}) \, b ({\bf y}_C) \, {b} ({\bf x}_C) &=&0 \;.\end{aligned}$$
The operators ${b^{\dagger}({\bf x}_C)}$ (${ b({\bf x}_C)}$) create (annihilate) an anyon at site ${\bf x_C}\doteq (x_1, x_2)$. $C$ denotes the path running from $+\infty$ to ${\bf x_C}$ keeping $x_2$ constant. The relations above hold if ${\bf x}_{C} > {\bf y}_{C}$; in the case ${\bf x}_{C} < {\bf y}_{C}$, they are satisfied substituting $q \leftrightarrow q^{-1}$. Note that ${\bf x}_C > {\bf y}_C \Leftrightarrow \{ x_2 > y_2 \; \vee \;
x_1> y_1 \; (\hbox{if} \; x_2=y_2) \} $. The function $q({\bf x}_{C},{\bf y}_{C}) = q(|{\bf x}_C-{\bf y}_C|)$ can be simplified (see e.g. ) to $q= e^{i\pi\nu}$ ($\nu\in\RR $), where $\nu$ denotes the [*statistics*]{}. If two anyons are at the same position ${\bf x}_{C} = {\bf y}_{C}$, then $q=1$. Otherwise, the standard bosonic or fermionic algebras are deformed by the parameter $q$.
[100]{}
R.E. Prange and S.M. Girvin, [*The Quantum Hall Effect*]{}, Springer, Berlin (1990) F. Wilczek, in [*Fractional Statistics and Anyon Superconductivity*]{}, F. Wilczek (ed.), World Scientific, Singapore (1993) G.E. Volovik and V.M. Yakovenko, J. Phys. Cond. Mat. [**1**]{}, 5263 (1989) A. Lerda and S. Sciuto, Nucl. Phys. B [**401**]{}, 613 (1993) G.A. Goldin and D.H. Sharp, Phys. Rev. Lett. [**76**]{}, 1183 (1996) A. Lerda, [*Anyons*]{}, Springer, Berlin (1992) I.M. Gel’fand and D.B. Fairlie, Commun. Math. Phys. [**136**]{}, 487 (1991); A.J. MacFarlane, J. Phys. A [**22**]{}, 4581 (1989); L.C. Bidenharn, J. Phys. A [**22**]{}, L873 (1989) D.P. Arovas, R. Schrieffer, F. Wilczek, and A. Zee, Nucl. Physics B [**251**]{}, 117 (1985); A. Comet, Y. Georgelin, and S. Ouvry, J. Phys. A [**22**]{}, 3917 (1989); A. Kristoffersen, S. Mashkevich, J. Myrheim, and K. Olaussen, [*cond-mat/9711169*]{} F.D.M. Haldane, Phys. Rev. Lett. [**67**]{}, 937 (1991) O.W. Greenberg, Phys. Rev. Lett. [**64**]{}, 705 (1990); Phys. Rev. D [**43**]{}, 4111 (1991) L.-A. Wu, Z.-Y. Wu, and J. Sun, Phys. Lett. A [**170**]{}, 280 (1992); R. Scipioni, Phys. Lett. B [**327**]{}, 56 (1994); Z.-T. Yan, [*quant-ph/9711038*]{} J.M. Leinaas and J. Myrheim, Phys. Rev. B [**37**]{}, 9286 (1988); Int. J. Mod. Phys. B [**5**]{}, 2573 (1991); Int. J. Mod. Phys. A [**8**]{}, 3649 (1991); T.H. Hanson, J.M. Leinaas, and J. Myrheim, Nucl. Phys. B [**384**]{}, 559 (1992) H. Bethe, Z. Phys. [**71**]{}, 205 (1931) A. Montorsi and M. Rasetti, Phys. Rev. Lett. [**72**]{} 1730 (1994); in [*Superconductivity and Strongly Correlated Electronic Systems*]{}, A. Romano and G. Scarpetta (eds.), World Scientific, Singapore (1994) We note that relation (\[postulate-rel\]) is sufficient but not necessary for relation (\[CONS2\]), but it is crucial to obtain (\[NUMBERCOMM\]). J. Negele and H. Orland, [*Quantum Many Particle Systems*]{}, Addison-Wesley, California (1987) C.N. Yang and C.P. Yang, Phys. Rev. [**147**]{}, 303 (1966); [**150**]{}, 327 (1966) N. Byers and C.N. Yang, Phys. Rev. Lett. [**7**]{} 46 (1961); F.C. Alcaraz, M.N. Barber, and M.T. Batchelor, Phys. Rev. Lett. [**58**]{} 771 (1987); Ann. Phys. N.Y. [**182**]{} 280 (1988) B.S. Shastry and B. Sutherland, Phys. Rev. Lett. [**65**]{}, 243 (1990); B. Sutherland and B.S. Shastry, Phys. Rev. Lett. [**65**]{}, 1833 (1990) A.A. Borovik, A.A. Zvyagin, V.Yu. Popkov, and Yu.M. Strzhemechnyi, JETP Lett. [**55**]{} 292 (1992); A.A. Zvyagin, Sov. J. Low Temp. Phys. [**18**]{} 723 (1992) H.J. Schulz and B.S. Shastry, Phys. Rev. Lett. [**80**]{}, 1924 (1998) A. Kundu, [*cond-mat/9710033*]{}; L. Freidel and J.M. Maillet, Phys. Lett. B [**262**]{}, 278 (1991); [**263**]{}, 403 (1991)
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We present magnetization studies as a function of time, temperature and magnetic field for $H$ $\parallel$ c-axis, in a hole-doped pnictide superconductor, La$_{0.34}$Na$_{0.66}$Fe$_2$As$_2$, with, $T_c$ $\approx$ 27 K. The obtained vortex phase-diagram shows that the magnetic irreversibility line is very close to the mean-field superconducting transition line, similar to the low $T_c$ superconductors, evidencing a strong pinning behavior. The irreversibility line does not follow a power law behavior with ($T_c$-$T$), however, it is well described using an expression developed in the literature considering the effect of disorder in the system. The critical current density estimated using the Bean’s critical-state model is found to be of the order of 10$^5$ A/cm$^2$ below 12 K in the limit of zero magnetic field. A plot of the normalized pinning force density as a function of the reduced magnetic field at different temperatures shows a good scaling and the analysis suggests that the vortex pinning is due to normal point like pinning centers. The temperature dependence of the critical current density suggests that the pinning due to the variation in charge carrier mean free path alone is not sufficient to explain the experimental data. Magnetic relaxation rate as a function of temperature and magnetic field is also studied.'
author:
- 'Shyam Sundar$^{1,\ddagger}$, S. Salem-Sugui, Jr.$^1$, A.D. Alvarenga$^2$, M. M. Doria$^1$, Yanhong Gu$^{3,4}$, Shiliang Li$^{3,4,5}$, Huiqian Luo$^{3,5}$ and L. Ghivelder$^1$'
title: 'Strong pinning in the hole-doped pnictide superconductor La$_{0.34}$Na$_{0.66}$Fe$_2$As$_2$'
---
Introduction
============
Superconductivity in Fe-pnictides is a field of great interest for fundamental science as well as for technological advancement [@fa11; @pen18; @hid18; @pal15]. Since the advent of iron-based superconductors (IBS), many new superconductors have been discovered in different families of this interesting class of high $T_c$ superconducting compounds [@hid18]. Among the different families of IBS, the 122-class is more interesting in-terms of technological purpose and also due to the availability of good quality sizable single crystals [@pal15; @hid18; @asw12; @abd14; @mah15; @gri13]. Recently, a new platform in the 122-pnictide family was discovered with a chemical formula (La$_{0.5-x}$Na$_{0.5+x}$)Fe$_2$As$_2$ ((La,Na)-122 family) [@2; @yan15]. It is an interesting and unique member of the 122-family, which allows the investigation of the electron-hole asymmetry, because in this system the doping occurs at the La-Na site with no changing in the Fe-As layers, as opposed to the doping in other 122-pnictide systems [@1; @2; @yan15].
Vortex dynamics and pinning mechanism in IBS is quite interesting due to its salient features [@baruch2; @FeNi; @AsP; @pra13] such as, low Ginzburg number ($G_i$) [@ele17], small anisotropy [@yua09], high intergrain connectivity [@wei12], high upper critical field [@sen08] and moderate $T_c$ [@ren08]. These properties are vital for technological applications of type-II superconductors [@pal15; @hid18]. In-spite of low Ginzburg number ($G_i$) the vortex phase-diagram of IBS and low-$T_c$ superconductors (LTS) are quite different in the sense that the upper critical-field, $H_{c2}$(T), and the irreversibility line, $H_{irr}$(T), are separated by a broad vortex-liquid region in IBS [@shi18], whereas, in LTS, no appreciable vortex-liquid region exist. Our motivation in the present study is to explore the vortex-dynamics in this recently discovered 122-type of IBS [@1].
In this work, we report a study of vortex dynamics in a single crystal of the hole-doped La$_{0.34}$Na$_{0.66}$Fe$_2$As$_2$ superconductor using isofield temperature dependence of the magnetization, $M$($T$), isothermal magnetic field dependence of the magnetization, $M$($H$), and magnetic relaxation, $M$($t$) measurements for $H$ $\parallel$ c-axis. The obtained vortex phase-diagram shows that the irreversibility line, $H_{irr}$, is very close to the mean field $T_c$($H$)-line, similar to the behavior in LTS. Critical current density, $J_c$, and the pinning mechanism are analysed using the models developed by Griessen et al [@wen] and Dew-Hughes [@dew] respectively. Magnetic relaxation rate as a function of temperature and magnetic field is discussed.
Experimental
============
The hole doped La$_{0.34}$Na$_{0.66}$Fe$_2$As$_2$ crystal was obtained accidentally when growing LaFeAsO single crystal with NAs and NaF flux. Details of the crystal growth are given in Ref. [@1]. The obtained La$_{0.34}$Na$_{0.66}$Fe$_2$As$_2$ crystals were characterized using different experimental techniques (see Ref.[@1]), which confirm the good quality of the crystals. These crystals have the same structure as of A$_e$Fe$_2$As$_2$ (A$_e$ = alkaline earth metal, e.g. Ca, Sr, Ba) with La and Na occupying the A$_e$ site. A sample from the same batch as in Ref. [@1] with mass $m$= 0.1348 mg and dimensions 1.54 mm $\times$ 1.52 mm $\times$ 0.016 ($\pm$ 0.002) mm was used in the present study. The magnetic measurements were made using a Quantum Design vibrating sample magnetometer (VSM) built in a 9 T physical property measurement system (PPMS), with magnetic fields applied parallel to the $c$-axis of the sample. $M$($T$) measurement at $H$ = 10 Oe shows a superconducting transition $T_c$ $\approx$ 27 K and a transition width $\Delta{T_c}$ $\approx$ 4 K. All magnetic measurements were obtained after a standard zero field cooled (ZFC) procedure. Isothermal $M$($H$) were measured with a ramp field d$H$/d$t$ = 50 Oe/sec, and obtained in five quadrants. Isofield $M$($T$) were continuously measured during a slow heating (and cooling in the case of field cooled (FC) curves) of d$T$/d$t$ = 0.3 K/sec. $M$($t$) curves were obtained for a span time of 1.5 hours at various temperatures with $H$ = 30 kOe and at various magnetic fields at $T$ = 16 K.
Results and discussion
======================
Figure 1a shows selected isofield $M(T)$ curves obtained in both ZFC and FC modes. The ZFC and FC curves separate just below the temperature region where the magnetization appears to be flat, with a subtle diamagnetic inclination (normal background). Figure 1b shows a detail of the $M(T)$ curve obtained for $H$ = 1 T, where the background selected in the normal region follows the expression $M_{back}$ = $a$($H$) - $b$($H$)$T$, where $a$($H$) and $b$($H$) $>$ 0 are constants for each magnetic field. For curves with $H$ $>$ 3 kOe, $a$($H$) is negative, which is due to the contribution from sample holder. This background contribution was observed for all $M$($T$) curves and subtracted in the following analysis. The inset of Fig. 1b shows a detail of the main plot near the transition which clearly shows that it is difficult to resolve the difference between $T_c$($H$), the mean field transition temperature defined as the temperature for which magnetization become diamagnetic with decreasing temperature, and $T_{irr}$, which marks the irreversible temperature at which the ZFC and FC curves separate. This behavior has been observed for all $M$($T$) curves.
![(a) Selected isofield $M$($T$) curves as measured. (b) Isofield $M$($T$) curves with and without background subtraction for $H$ = 1 T. The inset shows an enlarged plot at temperatures near $T_c$(H).[]{data-label="fig1"}](fig1.eps){height="14"}
![(a) Isothermal $M$($H$) curves as measured. (b) Isothermal $J_c$($H$) curves as obtained from the Bean model. The inset shows detail of $J_c$($H$) curve at 16 K, showing a change in curvature as $J_c$($H$) decays.[]{data-label="fig2"}](fig2.eps){height="14"}
Therefore, we decided to obtain the irreversible field, $H_{irr}$, from isothermal hysteresis $M$($H$) curves as shown in Fig. 2a, which was limited to our maximum magnetic field of 9 T. It is possible to see from Fig. 2a that all $M$($H$) curves show a small asymmetry due to a rotation of the x-axis which is an effect due to the sample holder signal. This rotation can be fixed by calculating the equilibrium magnetization, $M_{eq}$ = ($M^+$+$M^-$)/2 where $M^+$ is the increasing field branch of a $M$($H$) and $M^-$ is the decreasing field branch. The new increasing and decreasing fields branches can then be obtained after subtracting the measured $M$($H$) from $M_{eq}$ which produces symmetric $M$($H$) curves with respect to the $x$-axis, evidencing that bulk pinning dominates in the sample. We define $H_{irr}$ as the magnetic field where the field increasing and field decreasing branches of the $M(H)$ curves merge together within the equipment resolution, $\sim$1x10$^{-5}$ emu. The critical current density associated to each $M$($H$) curve is obtained through the Bean’s model [@bean] using expression [@ume87], $J_c$ = 20$\Delta{M}$/$a$(1-$a$/3$b$) where $b$$>$$a$ (cm) are the single crystal dimensions defining the area perpendicular to the magnetic field, $\Delta{M}$(emu/cm$^3$) corresponds to the width of the hysteresis curves and the resulting $J_c$ is given in A/cm$^2$ units. Figure 2b shows the resulting $J_c$($H$) curves as obtained. The curves for $T$$<$12 K show $J_c$(0) values, the critical current at zero field, above 10$^5$ A/cm$^2$ which is considered the required threshold value for applications [@pal15]. However, $J_c$ suppresses relatively fast as magnetic field increases and shows $J_c$ of the order of $\sim$10$^4$ A/cm$^2$ up to 9 T (max. field of measurements) for $T$ $<$ 12 K. We believe that the observed critical current density may be improved by introducing artificial defects as observed in other iron-pnictide superconductors [@tos16]. Among the different families of pnictide superconductors, the 122-class is the most relevant for high field application purposes [@pal15]. For comparison, in the case of K-doped (x = 0.3) 122-pnictide superconductor, the observed $J_c$ is over 10$^5$ A/cm$^2$ in a wide temperature and magnetic field range (up to 0.8$T_c$, 6 T) [@don16; @shi17]. Whereas, in case of Co-doped (x = 0.057) and P-doped (x = 0.3) 122-pnictide superconductor, $J_c$ $>$ 10$^5$ A/cm$^2$ is observed for T = 0.5 $T_c$ and $H$ = 6 T [@shi17]. Similarly, near optimal doped Ni-122 pnictide superconductors shows $J_c$ $>$ 10$^5$ A/cm$^2$ in a wide temperature and magnetic field range [@ss19]. The effect of particle irradiation on the vortex dynamics and $J_c$ would be interesting to explore in the present sample. A detailed review of the critical current density and pinning in bulk, thin films, tapes and wires of IBS for technological importance is provided in Refs. [@pal15; @hid18].
The inset of Fig. 2b shows a selected $J_c$($H$) curve at 16 K, evidencing a change in curvature as $J_c$($H$) approaches zero which was observed in all curves at higher temperatures. The same effect can be observed in the respective $M$($H$) curves. This change in curvature (downward to upward curvature as field increases) is usually a precursor of the peak effect occurring in $J_c$($H$) curves [@shyam], which is absent in our sample. The interesting point of this effect is that $J_c$($H$) approaches zero with a downward curvature, while usually it approaches zero exponentially (upward curvature).
![(a) A typical magnetic relaxation curve measured during 1.5 hours, showing usual linear time dependence. (b) Relaxation rate, $R$ = $-$d\[ln($M$)\]/d\[ln$t$\], for $T$ = 16 K, plotted as a function of magnetic field.[]{data-label="fig3"}](fig3.eps){height="15"}
In order to better understand this effect we have obtained magnetic relaxation measurements, $M$($t$), as a function of magnetic field and temperature. The resulting $M$($t$) curves plotted as ln($M$) vs. ln$t$ produced the usual linear behavior allowing us to obtain the relaxation rate $R$ = $-$d\[ln($M$)\]/d\[ln$t$\], as shown in Fig. 3a. Figure 3b shows a plot of $R$ vs $H$ at $T$ = 16 K, where $R$ decreases monotonically as field increases, evidencing the strength of the pinning as the field increases.
![(a) A plot of $U_0$ vs $T$ for $H$ = 3 T. (b) Apparent activation energy, $U_0$ obtained from magnetic relaxation for $H$ = 3 T, plotted against 1/$J_c$($T$). (c) Activation energy curve obtained from Maley’s approach for $H$ = 3 T plotted against magnetization.[]{data-label="fig4"}](fig4.eps){height="15"}
Figure 4a shows a plot of $U_{0}$ vs. $T$ for $H$ = 3 T. Two different slopes at intermediate temperature suggest a change in the pinning mechanism. Figure 4b shows a plot of the apparent activation pinning energy, $U_{0}$ = $R$/$T$, against 1/$J_c$, obtained for $H$ = 3 T, which despite the fact that the values of the exponents do not correspond to the expected ones [@wei] ($\mu$ $\sim$ 1 and $p$ $\sim$ -0.5), the overall behavior suggests the existence of a crossover in the pinning mechanism of the type elastic to plastic occurring at some temperature between 10 K and 14 K [@wei]. This crossover has been associated to the fish-tail effect found in $M$($H$) curves [@wei] which is absent in ours curves. Similarly, in K-doped BaFe$_2$As$_2$, a pinning crossover is suggested through $U_0$ vs. 1/$J$ plot, however, no fish-tail effect is observed for that temperature range [@shyam]. Therefore, it would be misleading to interpret the behaviour of $R$ vs. $T$ and $U_0$ vs. 1/$J$ plots as a pinning crossover, because, there is no effect of such crossover in the measured isothermal $M(H)$ curves. Figure 4c shows a plot of the activation energy $U$= -$T$ln(d$M$/d$t$)+$C$$T$ as obtained in Ref.[@maley], where the smooth curve was obtained with a constant $C$=10. Parameter $C$ depends on the attempt frequency, hoping distance and sample dimension. It must be mentioned that the smooth curve following a log($M$) behavior, as in Ref.[@maley] for YBaCuO, was obtained without the need of a temperature scaling. Also, the values of $U$ in Fig. 4c are of the same order of magnitude of the values found for YBaCuO in Ref.[@maley] at similar reduced temperatures. The scaling of $U$($M$) with magnetic field for data obtained at $T$ = 16 K, as performed in Refs.[@abulafia; @FeCo; @BaK], would point to a possible pinning crossover, as suggested in Fig. 3b. Nevertheless, we could not find any reasonable scaling at this temperature.
![Vortex phase-diagram of La$_{0.34}$Na$_{0.66}$Fe$_2$As$_2$. The dotted line is a fitting to the irreversibility line curve (see text for details).[]{data-label="fig5"}](fig5.eps){height="12"}
Figure 5 shows the vortex phase-diagram where the values of $H_{irr}$ are very close to values of $T_c$($H$). Such a wide irreversible region, as shown in Fig. 2a, suggests that the studied system has potential for applications. In the phase-diagram (Fig. 5), at high temperatures, $T_c$($H$) increases linearly with d$H_{c2}$($T$)/d$T$ = -3.1 T/K, which from the Werthamer-Helfand-Hohenberg (WHH) formula [@whh], renders a $H_{c2}$(0) = -0.695 $T_c$ d$H_{c2}$($T$)/d$T$ = 58 T. This value is higher than the Pauli paramagnetic limit for $H_{c2}$(0) given by $H_p$ = 1.84 $T_c$ = 50 T[@pauli]. The Maki parameter[@whh] defined as $\alpha$ = $H_{c2}$(0)/$\sqrt{2}$$H_p$ = 0.82 for the studied sample. Systems possessing $\alpha$$>$ 1 are candidates to exhibit the exotic Fulde-Ferrell-Larkin-Ovchinnikov (FFLO) phase[@fflo1; @fflo2]. Since our system has an anisotropy $\gamma$= 1.9 [@1] the FFLO phase is a candidate to appear for $H$$\parallel$$ab$ planes. We observed that the irreversibility line does not follow a power law behavior with ($T_c$-$T$) as is commonly observed in many superconductors. In addition, the proximity of the irreversibility line to the $T_c$($H$) line resembles the behavior observed in low $T_c$ systems, as for instance in NbSe$_2$ [@nbse]. $H_{irr}(T)$ has been described in Refs. [@baruch0; @baruch1] by an expression which takes in to account the disorder in the system. Even for NbSe$_2$ the $H_{irr}(T)$ line is visibly further apart from the $H_{c2}(T)$ line than what is observed in our sample. The dotted line in Fig. 5, represents the best fit of the $H_{irr}(T)$ data to the expression provided in Refs. [@baruch0; @baruch1], which is mentioned below.
1-$t$-$b$+3$n_p$(1-$t$)$^2$ 4$\pi$-2 $\sqrt{2G_i}$ $t$$b$=0
where $t$ = $T$/$T_c$ is the reduced temperature, $b$ = $H$/$H_{c2}$(0), $n_p$ measures the disorder in the system and $G_i$ is a different definition of the Ginzburg number which measures the importance of thermal fluctuations. In that expression, $b$, $n_p$ and $G_i$ are fitting parameters, and the dotted line shown in Fig.5 was obtained for a slightly higher $T_c$ = 28 K, $n_p$ = 0.002, $G_i$$\sim$10$^{-8}$ and $H_{c2}$(0) = 60 T. It is observed that both $G_i$ and $n_p$ show a good fit within 10 % of the fitted values, whereas, $H_{c2}$(0) may vary within 4% of the value obtained from the fitting. The value of $n_p$ is similar to that obtained in Ref. [@baruch0] for NbSe$_2$, $H_{c2}$(0) is similar to the value obtained from the WHH expression, but $G_i$ is three orders of magnitude lower than the value obtained for NbSe$_2$. We do not have explanation for such a low value of $G_i$ in the present case. The importance of thermal fluctuations is sometimes associated to the extension of the reversible region, which in our case is virtually absent from the $M(T)$ curves.
![(a) Normalized pinning force density plotted against the reduced magnetic field. The experimental data taken at different temperatures collapsed into a single scaled curve, which is fitted using the expression, $F_p$/$F_p$$_{max}$ = $A$$h^p$(1-$h$)$^q$, where, $p$ and $q$ are fitting parameters, which define the pinning characteristics. (b) Normalized critical current density, $J_c$($T$)/$J_c$(0) as a function of reduced temperature, $T$/$T_c$. The solid lines represents the $\delta l$ and $\delta T_c$ pinning behavior.[]{data-label="fig6"}](fig6.eps){height="10"}
We present in Fig. 6a, the plot of the normalized pinning force density, $F_p$/$F_p$$_{max}$ as a function of the reduced magnetic field $h$=$H$/$H_{irr}$, where, $F_p$=$J_c$$\times$$H$. The collapse of the many different isothermal pinning force curves is evident on the graph. This plot allows us to obtain the reduced magnetic field ($h$) for which the pinning force reaches its maximum, $h_{max}$ = 0.45, and a fitting of the resulting curve to the Dew-Hughes formula [@dew] $F_p$/$F_p$$_{max}$ = $A$$h^p$(1-$h$)$^q$ produced $A$ = 5.14, $p$ = 1.08 and $q$ = 1.32, where $p$/($p$+$q$) = $h_{max}$ = 0.45 as obtained from the experimental data. This fitting has been applied to examine the pinning force on many systems over the years[@kobli] and the different values obtained for the parameters $h_{max}$, $p$ and $q$, are used to determine the type of the dominant pinning[@dew; @kobli; @mat13; @sha13]. In a classic paper, Dew-Hughes discussed the various scenarios of pinning centers involved in different pinning mechanisms [@dew]. For a system having pinning due to the variation in charge carrier mean free path ($\delta$$l$-pinning) and the pinning centers are point-like, the maximum in the normalized pinning force density occurs at $h$ = 0.33, with $p$ = 1 and $q$ = 2. In the case of the pinning due to the variation in the superconducting transition temperature ($\delta$$T_c$-pinning), the maximum in the normalized pinning force density is found to be at much higher $h$ values, in short, for point pinning centers, $h$ = 0.67, with $p$ = 1, $q$ = 2; for surface pinning centers, $h$ = 0.6, with $p$ = 1.5 and $q$ = 1 and for volume pinning centers, $h$ = 0.5, with $p$ = $q$ =1. Therefore, in the present study, $h$ = 0.45 with $p$ = 1.08 and $q$ = 1.32 shows that a single type of pinning is not sufficient to explain the results. However, similar values of $h$ has also been observed in many other studies of iron-pnictide superconductors, such as, for BaFe$_{1.9}$Ni$_{0.1}$As$_2$, $h$ = 0.4 [@sha13], for Ca$_{0.8}$La$_{0.2}$Fe$_{1-x}$Co$_x$As$_2$, $h$ = 0.44 [@wei], for Ba$_{0.68}$K$_{0.32}$Fe$_2$As$_2$, h = 0.43 [@sun09], for Ba(Fe$_{1-x}$Co$_x$)$_2$As$_2$, $h$ = 0.45 [@yam09]. In such studies, the value of $h$ $\approx$ 0.45, is attributed to an inhomogeneous distribution of dopants or Arsenic deficiency [@sun09]. Shahbazi et al [@sha13] argued that $h$ = 0.4, in case of BaFe$_{1.9}$Ni$_{0.1}$As$_2$ is due to the $\delta$$l$-type pinning. Zhou et al [@wei] related $h$ = 0.44 with the randomly distributed nano-scale point-like defects, which is common in the case of iron pnictide superconductors [@hua08; @sun09; @wei]. Therefore, $h$ = 0.45 in our case is attributed to the $\delta$$l$-type pinning due to point pinning centers. Whereas, for $\delta$$T_c$ pinning, the maximum in normalized pinning force density would occur at $h$ higher than 0.5. The type of pinning can be further examined following an approach developed in Ref.[@wen] where the temperature dependent critical current at zero field $J_c$($T$), normalized by the critical current at zero field at $T$ = 0 is plotted against $t$ = $T$/$T_c$. The equivalent plot for our data is shown in Fig. 6b, which is compared to the theoretical expression for $\delta$$l$-type of pinning, where $J_c$($T$)/$J_c$(0) = (1 + $t^2$)$^{-1/2}$(1- $t^2$)$^{5/2}$ and for $\delta$$T_c$-type pinning, $J_c$($T$)/$J_c$(0) = (1-$t^2$)$^{7/6}$(1 + $t^2$)$^{5/6}$ [@wen]. It is evident from Fig. 6b that the $\delta$$l$-type pinning alone can not explain the experimental data adequately.
Conclusions
===========
In conclusion, the vortex phase-diagram of the newly synthesized iron-based superconductor La$_{0.34}$Na$_{0.66}$Fe$_2$As$_2$ shows an irreversibility line very close to the mean field transition temperature $T_c$($H$) evidencing a strong pinning. The irreversibility line does not follow the usual power law with ($T_c$-$T$) but it was successfully fitted by an expression developed in Refs. [@baruch0; @baruch1] considering the effect of disorder, where a considerable low disorder similar to that observed for NbSe$_2$ [@baruch0; @baruch1] was found for our system. We observed that the upper critical field at zero temperature exceeds the prediction of the Pauli paramagnetic limit, suggesting that the system is a candidate to show the FFLO phase for $H$$\|$$ab$-planes. The critical current density at zero magnetic field reaches the threshold value $J_c$$>$10$^5$ A/cm$^2$ for temperatures below 12 K, which along with the fact that the irreversibility line is very close to $T_c$($H$) makes the system technologically relevant. Magnetic relaxation obtained as a function of field for a fixed temperature shows that the relaxation rate monotonically decreases as the field increases, while magnetic relaxation obtained for a fixed field as a function of temperature suggests a crossover in the pinning mechanism. The latter results allowed us to obtain a smooth curve of the isofield activation energy with magnetization, as first done by Maley[@maley], where the observed log($M$) behavior and values are similar to the ones obtained in Ref. [@maley] for YBaCuO. The pinning analysis using the Dew-Hughes model suggests a $\delta$$l$-type pinning due to the point pinning centers. However, the temperature dependence of the critical current density indicates that $\delta$$l$ pinning alone can not explain the data adequately. To explore this compound for technological use, the effect of grain boundaries on the critical current density and vortex dynamics in polycrystalline and thin film samples are yet to be investigated.
Acknowledgments {#acknowledgments .unnumbered}
===============
SS acknowledges financial support from a post doctoral fellowship by Fundação Carlos Chagas Filho de Amparo à Pesquisa do Estado do Rio de Janeiro - FAPERJ (Project E-26/202.848/2016). LG was also supported by FAPERJ (Projects E-26/202.820/2018 and E-26/010.003026/2014). SSS and ADA were partially supported by the Brazilian federal agency Conselho Nacional de Desenvolvimento Científico e Tecnológico - CNPq. The work at IOP, CAS is supported by the National Key R$\&$D Program of China (Nos. 2017YFA0302900, 2016YFA0300502, 2017YFA0303103), the National Natural Science Foundation of China (Nos. 11674406, 11674372, 11774399), the Strategic Priority Research Program (B) of the Chinese Academy of Sciences (XDB25000000, XDB07020000). HL is supported by the Youth Innovation Promotion Association of CAS (No. 2016004).
References {#references .unnumbered}
==========
[10]{}
Fa Wang and Dung-Hai Lee. . , 332:200, 2011.
Peng Zhang, Zhijun Wang, Xianxin Wu, Koichiro Yaji, Yukiaki Ishida, Yoshimitsu Kohama, Guangyang Dai, Yue Sun, Cedric Bareille, Kenta Kuroda, and et al. . , 2018.
Hideo Hosono, Akiyasu Yamamoto, Hidenori Hiramatsu, and Yanwei Ma. . , 21:278, 2018.
Ilaria Pallecchi, Michael Eisterer, Andrea Malagoli, and Marina Putti. . , 28:114005, 2015.
S. Aswartham, M. Abdel-Hafiez, D. Bombor, M. Kumar, A. U. B. Wolter, C. Hess, D. V. Evtushinsky, V. B. Zabolotnyy, A. A. Kordyuk, T. K. Kim, S. V. Borisenko, G. Behr, B. B[ü]{}chner, and S. Wurmehl. . , 85:224520, 2012.
M. Abdel-Hafiez, P. J. Pereira, S. A. Kuzmichev, T. E. Kuzmicheva, V. M. Pudalov, L. Harnagea, A. A. Kordyuk, A. V. Silhanek, V. V. Moshchalkov, B. Shen, Hai-Hu Wen, A. N. Vasiliev, and Xiao-Jia Chen. . , 90:054524, 2014.
Mahmoud Abdel-Hafiez, Yuanyuan Zhang, Zheng He, Jun Zhao, Christoph Bergmann, Cornelius Krellner, Chun-Gang Duan, Xingye Lu, Huiqian Luo, Pengcheng Dai, and Xiao-Jia Chen. . , 91:024510, 2015.
V. Grinenko, S.-L. Drechsler, M. Abdel-Hafiez, S. Aswartham, A. U. B. Wolter, S. Wurmehl, C. Hess, K. Nenkov, G. Fuchs, D. V. Efremov, B. Holzapfel, J. van den Brink, and B. B[ü]{}chner. . , 250:593, 2013.
Akira Iyo, Kenji Kawashima, Shigeyuki Ishida, Hiroshi Fujihisa, Yoshito Gotoh, Hiroshi Eisaki, and Yoshiyuki Yoshida. . , 140:369, 2018.
J.-Q. Yan, S. Nandi, B. Saparov, P. Cermak, Y. Xiao, Y. Su, W. T. Jin, A. Schneidewind, and Th. Br[ü]{}ckel. . , 91:024501, 2015.
Yanhong Gu, Jia-Ou Wang, Xiaoyan Ma, Huiqian Luo, Youguo Shi, and Shiliang Li. . , 31:125008, 2018.
B. Rosenstein, B. Ya. Shapiro, I. Shapiro, Y. Bruckental, A. Shaulov, and Y. Yeshurun. . , 72:144512, 2005.
S [Salem-Sugui Jr]{}, L Ghivelder, A D Alvarenga, L F Cohen, Huiqian Luo, and Xingye Lu. . , 26:025006, 2013.
S [Salem-Sugui Jr]{}, J Mosqueira, A D Alvarenga, D Sóñora, E P Herculano, Ding Hu, Genfu Chen, and Huiqian Luo. . , 28:055017, 2015.
G Prando, R Giraud, S Aswartham, O Vakaliuk, M Abdel-Hafiez, C Hess, SWurmehl, A U B Wolter, and B B[ü]{}chner. . , 25:505701, 2013.
S. Eley, M. Miura, B. Maiorov, and L. Civale. . , 16:409, 2017.
H. Q. Yuan, J. Singleton, F. F. Balakirev, S. A. Baily, G. F. Chen, J. L. Luo, and N. L. Wang. . , 457:565, 2009.
J. D. Weiss, C. Tarantini, J. Jiang, F. Kametani, A. A. Polyanskii, D. C. Larbalestier, and E. E. Hellstrom. . , 11:682, 2012.
C. Senatore, R. Fl[ü]{}kiger, M. Cantoni, G. Wu, R. H. Liu, and X. H. Chen. . , 78:054514, 2008.
Ren Zhi-An, Lu Wei, Yang Jie, Yi Wei, Shen Xiao-Li, Li Zheng-Cai, Che Guang-Can, Dong Xiao-Li, Sun Li-Ling, Zhou Fang, and Zhao Zhong-Xian. . , 25:2215, 2008.
Shiv J. Singh, Matthew Bristow, William R. Meier, Patrick Taylor, Stephen J. Blundell, Paul C. Canfield, and Amalia I. Coldea. . , 2:074802, 2018.
R. Griessen, Wen Hai-hu, A. J. J. van Dalen, B. Dam, J. Rector, H. G. Schnack, S. Libbrecht, E. Osquiguil, and Y. Bruynseraede. . , 72:1910, 1994.
D Dew-Hughes. . , 30:293, 1974.
Charles P. Bean. . , 36:31, 1964.
A. Umezawa, G. W. Crabtree, J. Z. Liu, H. W. Weber, W. K. Kwok, L. H. Nunez, T. J. Moran, and C. H. Sowers. . , 36:7151, 1987.
Toshinori Ozak, Lijun Wu, Cheng Zhang, Jan Jaroszynski, Weidong Si, Juan Zhou, Yimei Zhu, and Qiang Li. . , 7:13036, 2016.
Dongjoon Song, Shigeyuki Ishida, Akira Iyo, Masamichi Nakajima, Jun ichi Shimoyama, Michael Eisterer, and Hiroshi Eisaki. . , 6:26671, 2016.
Shigeyuki Ishida, Dongjoon Song, Hiraku Ogino, Akira Iyo, Hiroshi Eisaki, Masamichi Nakajima, Jun ichi Shimoyama, and Michael Eisterer. . , 95:014517, 2017.
Shyam Sundar, Jr Said Salem-Sugui, Edmund Lovell, Alexander Vanstone, Lesley F Cohen, Dongliang Gong, Rui Zhang, Xingye Lu, Huiqian Luo, and Luis Ghivelder. . , 2019.
Shyam Sundar, S. Salem-Sugui Jr., H. S. Amorim, Hai-Hu Wen, K. A. Yates, L. F. Cohen, and L. Ghivelder. . , 95:134509, 2017.
Wei Zhou, Xiangzhuo Xing, Wenjuan Wu, Haijun Zhao, and Zhixiang Shi. . , 6:22278, 2016.
M. P. Maley, J. O. Willis, H. Lessure, and M. E. McHenry. . , 42:2639, 1990.
Y. Abulafia, A. Shaulov, Y. Wolfus, R. Prozorov, L. Burlachkov, Y. Yeshurun, D. Majer, E. Zeldov, H. Wuhl, V. B. Geshkenbein, and V. M. Vinokur. . , 77:1596, 1996.
Shyam Sundar, J Mosqueira, A D Alvarenga, D Sóñora, A S Sefat, and S Salem-Sugui Jr. . , 30:125007, 2017.
S. [Salem-Sugui Jr.]{}, L. Ghivelder, A. D. Alvarenga, L. F. Cohen, K. A. Yates, K. Morrison, J. L. Pimentel Jr., Huiqian Luo, Zhaosheng Wang, and Hai-Hu Wen. . , 82:054513, 2010.
N. R. Werthamer, E. Helfand, and P. C. Hohenberg. . , 147:295, 1966.
A. M. Clogston. . , 9:266, 1962.
Peter Fulde and Richard A. Ferrell. . , 135:A550, 1964.
A. I. Larkin and Yu. N. Ovchinnikov. . , 20:762, 1965.
S. S. Banerjee, S. Saha, N. G. Patil, S. Ramakrishnan, A. K. Grover, S. Bhattacharya, G. Ravikumar, P. K.Mishra, T. V. C. Rao, V. C. Sahni, C. V. Tomy, G. Balakrishnan, D. Mck. Paul, and M. J. Higgins. . , 308:25, 1998.
B. Rosenstein and V. Zhuravlev. . , 76:014507, 2007.
Baruch Rosenstein and Dingping Li. . , 82:109, 2010.
Michael R Koblischka and Miryala Muralidhar. , 30:1630017, 2016.
Md. Matin, L. S. Sharath Chandra, M. K. Chattopadhyay, R. K. Meena, Rakesh Kaul, M. N. Singh, A. K. Sinha, and S. B. Roy. . , 113:163903, 2013.
M. Shahbazi, X. L. Wang, K. Y. Choi, and S. X. Dou. . , 103:032605, 2013.
D. L. Sun, Y. Liu, and C. T. Lin. . , 80:144515, 2009.
A. Yamamoto, J. Jaroszynski, C. Tarantini, L. Balicas, J. Jiang, A. Gurevich, D. C. Larbalestier, R. Jin, A. S. Sefat, M. A. McGuire, B. C. Sales, D. K. Christen, and D. Mandrus. . , 94:062511, 2009.
Huan Yang, Huiqian Luo, Zhaosheng Wang, and Hai-Hu Wen. . , 93:142506, 2008.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We describe a model of cytoskeletal mechanics based on the force-induced conformational change of protein cross-links in a stressed polymer network. Slow deformation of simulated networks containing cross-links that undergo repeated, serial domain unfolding leads to an unusual state—with many cross-links accumulating near the critical force for further unfolding. Thermal activation of these links gives rise to power-law rheology resembling the previously unexplained mechanical response of living cells. Moreover, we hypothesize that such protein cross-links function as biochemical mechano-sensors of cytoskeletal deformation.'
author:
- 'Brenton D. Hoffman'
- 'Gladys Massiera$^\dagger$'
- 'John C. Crocker'
title: |
Power-law rheology and mechano-sensing in a cytoskeleton\
model with forced protein unfolding
---
The importance of mechanical cues for understanding cell behavior is increasingly recognized. Stem cell differentiation [@mcbeath], tissue morphogenesis [@pascek] as well as cell growth [@wang] and death [@chen] are known to be affected by cell shape or the stiffness of the surrounding extra-cellular matrix. The molecular mechanisms by which such mechanical cues produce biochemical responses, however, remain essentially unknown [@janmeyc]. Similarly, the anomalous mechanical response of cells defies explanation despite the identification of the cytoskeleton’s major structural constituents and considerable modeling effort [@gardela; @wachsstock; @ningwang]. Rheological measurements on living cells yield a frequency dependent shear modulus that scales as a weak power-law, $|G^*(\omega)|\sim\omega^\beta$, with $0.1 < \beta < 0.25$, spanning many decades of frequency [@fabry; @feneberg; @alcaraz]. Such a mechanical response is rather unusual [@sollicha], having only been observed previously in seemingly unrelated materials such as foams and pastes. While networks of purified filamentary biopolymers (e.g. F-actin) generally have a frequency-independent plateau elasticity [@gardelb], studies of networks containing the cross-link proteins filamin [@gardel-PNAS; @gardelc] or $\alpha$-actinin [@tseng] have reproduced the weak power-law rheology as well as other aspects of the cell response, such as stress-induced stiffening. These results suggest that the power-law rheology of cells may be due to cross-link conformational change or unbinding.
In this Letter, we propose a simple cytoskeletal architecture that may explain both cell rheology and mechano-sensing, based on the force-induced serial unfolding of domains in protein cross-links. We show, using simulation, that shearing a simplified model network leads spontaneously to the accumulation of many cross-links at tensions on the cusp of unfolding. Thermally activated unfolding of these cross-links readily reproduces cells’ power-law frequency dependent shear modulus with physically plausible model parameters. Comparable models based on forced [*unbinding*]{} (e.g. of cross-links’ actin binding domains from actin) rather than unfolding do not produce power-law rheology. Moreover, this unusual, near-critical arrangement of cross-links is suggestive of a mechano-sensory function. We hypothesize that, by modulating the binding of signaling species, unfolding cross-link domains function as the fundamental biochemical transducers of cytoskeletal deformation.
![\[fig.one\] Schematic representation of serial unfolding and the cytoskeleton. (a) As cross-links are extended, domains serially unfold with typical force vs. extension curves having abrupt transitions at a critical force, $F_c$, corresponding states are labeled. (b) Our model of the cytoskeleton consisting of a network of generic semi-flexible polymers and partially unfolded, extensible cross-links.](figone.eps){width="2.75in"}
Many cross-linking proteins (e.g. filamin, $\alpha$-actinin, spectrin, plakin, etc.) have a structure consisting of similar, repeated domains that can be serially unfolded by applied force [@furuike; @riefa; @riefb], see Figure \[fig.one\](a). The unfolding time is typically described by the Bell model: [@bell] $$\tau_B(F)=\tau_a
\exp[E_B(1-\frac{F}{F_c})],E_B=\frac{F_c r_o}{k_BT}
\label{Eq.one}$$ where $\tau_a$ is a molecular attempt time, $F_c$ is a critical force, $r_o$ is a characteristic bond length-scale, and $k_BT$ is the thermal energy. The exponential form causes small changes in tension to yield rather large changes in unfolding time. As domains serially unfold, the molecule becomes progressively longer and longer. The entropic elasticity of the unfolded protein causes a spring-like response between unfolding events, producing a ‘sawtooth’ force-extension profile, Figure \[fig.one\]a.
We hypothesize that some cross-links unfold under physiological stresses, as sketched in Figure \[fig.one\]b, and that such unfolding is the source of microscopic stress relaxation causing weak power-law rheology. Simple estimates [@envelope] suggest that physiologically plausible stresses of order $\sim$100 Pa yield tensions sufficient to unfold some crosslink species. While the exponential sensitivity of the Bell model naturally gives rise to very broadly distributed characteristic unfolding and relaxation times, a broad distribution alone is not sufficient. As we shall see, the power-law form of the rheology requires an exponential distribution of cross-link tension. Just such force distributions emerge spontaneously when simulated model networks are slowly sheared.
For our simulations, we constructed two-dimensional networks having periodic connectivity in the left/right direction, whose nodes are initially offset from a triangular lattice by a Gaussian-distributed amount, Figure \[fig.two\](a,b). Force carrying mechanical links connect these nodes (15% of the nearest-neighbor bonds are left empty to introduce topological disorder). Networks larger than roughly 25 by 35 nodes ($\sim$$10^4$ links) gave results that were essentially independent of system size and changes in network disorder. To model serial unfolding, each link has a linear ‘sawtooth’ force-extension curve: $F = k[(x-x_0) \bmod x_c]$, where $k$ is a spring constant, $x$ is the instantaneous link extension, $x_c \equiv F_c/k$ is the critical extension for bond rupture and $x_0$ is the link length at the beginning of the simulation (typically 5-100 $x_c$). Our network does not contain separate rod-like and cross-link elements; each of our simulated links may be considered conceptually equivalent to a single inextensible rod of length $x_0$ connected in series with a cross-link having an infinite number of unfoldable domains. Compressed links generate a compressive force (for $x<x_o, F= -k[(x_o-x) \bmod x_c] $), but this is not critical to our results.
The network is sheared by translating the nodes clamped to the upper boundary in a series of small steps ($\Delta \gamma ~<10^{-3}$) at a constant strain rate, $\gamma(t)=\dot{\gamma}t$. All nodes are first displaced according to an affine shear deformation, and then relaxed to mechanical equilibrium (zero total force on each node) by moving the non-boundary nodes using an ‘over-damped’ steepest descent algorithm. During relaxation, links move freely between branches of the force extension curve until an equilibrium configuration is reached. This is equivalent to assuming that domains unfold and refold instantaneously when the tension reaches $F_c$ and zero, respectively.
To model Bell-type thermally-activated unfolding a Kinetic Monte Carlo (KMC) algorithm is used. The expected unfolding rate of each tensed ($F>0$) link is computed from its tension and the Bell model (rate $= 1/\tau_B(F))$. Consistent with the KMC algorithm, an exponentially distributed time step is generated which is inversely proportional to the total unfolding rate, and the link to unfold is selected in a rate weighted manner. To unfold the selected sub-critical link, its force extension curve is temporarily modified to be zero in the $x_c$-wide extension interval it occupies. The network is then relaxed, with the selected link typically moving into an adjacent, non-zero interval.
![\[fig.two\] Representations of our simplified model network and its force distribution without activated unfolding. (a) and (b) the network at strain $\gamma = $0.3 and 1.5 respectively. (c) Probability distributions of scaled force, curves for strains $\gamma = 0.3, 1.0, 1.5$, and $2.0$ (thin to thick) respectively. Line represents exponential behavior.](figtwo.eps){width="2.75in"}
It is illustrative to first consider the $T=0$ limit of the unfolding simulation ([*i.e.*]{} without Bell unfolding). In that case, the network configuration depends only on the strain, $\gamma$, and the network’s geometrical parameters. For small strains, a roughly uniform distribution of link tension develops. For strains, $\gamma > 0.4$, links accumulate at tensions somewhat smaller than $F_c$, Figure \[fig.two\](c). The distribution of link tensions approximates an increasing exponential function of force over a small range, $P(F) \sim exp(F/F_e)$, the qualitative form required to produce power-law rheology. As the strain is increased further, the range of forces showing exponential behavior broadens, but maintains the same slope: $F_e/F_c \approx 0.4$.
In principle, the evolution of thermalized networks is more complicated, depending additionally on the Bell parameters $E_B$ and $\tau_a$ and the strain rate $\dot{\gamma}$, (assumed to be slow, $\dot{\gamma} \tau_a \ll 1)$. In practice, however, the qualitative form of $P(F)$ is quite similar to the athermal case, but with a somewhat smaller critical force, see Figure \[fig.three\](a). Indeed, the $P(F)$’s from all of our thermalized unfolding simulations could be accurately scaled onto one another by renormalizing their forces by an effective critical force, $F_c'$. Important for later discussion, the number of ‘unfolded’ domains in the network is a monotonic function of network strain, with little dependence on strain rate or hysteresis upon strain reversal, Figure \[fig.three\](b).
![\[fig.three\] Force distributions and scaling behavior with Bell-type activated unfolding. (a) Probability distributions of scaled force, determined for strain $\gamma = 1.5$, $E_B=25$ and $\tau_a=50$ nsec. Solid lines are from unfolding simulations with $\dot{\gamma} = 7, 1000$ sec$^{-1}$ and the athermal simulation (thin to thick) respectively. The dotted line is from the unbinding simulation at $\gamma= 0.5$, $E_B=25$, $\tau_a=50$ nsec, and $\dot{\gamma} = 10$ sec$^{-1}$ (b) The number of unfolded domains as a function of strain. The solid is line is during strain application and the dotted line is during strain reversal. (c) Most probable unfolding force as a function of dimensionless strain rate. ](figthree.eps){width="2.75in"}
The effective critical force behavior has a simple explanation. Bell-type molecules subjected to a constant force loading rate have a well-defined most probable force for unfolding, which is a logarithmic function of the loading rate [@riefa]. The constant strain rate, $\dot{\gamma}$, of our model slowly stretches links, leading to a constant force loading rate. We can identify the resulting most probable link unfolding force with $F_c'$, which, as expected, scales logarithmically with $\dot{\gamma}$, Figure \[fig.three\](c). Links with tensions even slightly smaller than $F_c'$ have an almost negligible Bell unfolding rate (since $E_B\gg1$), leading $P(F)$ to evolve much as an athermal model with ‘all or nothing’ unfolding at critical force $F_c'$.
The frequency-dependent mechanical response of our network can be estimated by superposing Maxwell modes [@winter]. Formally, such a modulus describes the network’s differential response to a small oscillatory applied stress superposed on the network’s static stress. This is compatible with the emerging view that cell measurements actually report such a differential shear modulus at a cell-generated prestress [@gardel-PNAS]. As expected, the computed shear modulus, Figure \[fig.four\](top), has a roughly power-law form, $|G^*(\omega)|\sim\omega^\beta$, over a finite frequency range, with a non-universal exponent $\beta \approx 2.5 F_c / F_c' E_B $. The modulus has a terminal mode at frequencies below the zero-force unfolding rate, $\tau_a^{-1} \exp(-E_B)$, and approaches a plateau value, $G_o$ at frequencies higher than $\omega_{max} \approx 1/\tau_B(F_c')$. Surprisingly, the frequency range with power-law rheology varies quite slowly with the strain rate: $\omega_{max} \sim \dot{\gamma}^b \tau_a^{b-1}$, with $b \approx 0.18$.
![\[fig.four\] The computed frequency dependent mechanical response of our model. (top) Dimensionless differential storage (solid) and loss (dashed) moduli computed for $E_B = 25,30,35,40$ (thick to thin). (bottom) The experimental data of Fabry [*et al.*]{} for untreated HASM cells, compared to our model (solid lines). At high frequencies, a $G^* \sim
\omega^{0.75}$, contribution was added (dashed line). Model rheology generated using $E_B = 25$, $G_o=5$ kPa, $ \tau_a= 70$ nsec and $\dot{\gamma}=7$ sec$^{-1}$. ](figfour.eps){width="2.75in"}
In a separate study of forced unbinding, (rather than unfolding), force-extension curves with a single period were used ($F = k(x-x_o)$ for $|F|<F_c$, zero otherwise). Any link at zero force in mechanical equilibrium after a time/strain step corresponds to a link that has exceeded its critical force and unbound. We then reset its $x_o$ parameter to its new equilibrium length to simulate rapid rebinding at zero force with the same network connectivity. Activated unbinding is handled as before, with the KMC-selected link having its force set to zero, followed by network relaxation and resetting of the $x_o$’s of any unbound links. Unlike the unfolding simulations, the unbinding simulations never showed any accumulation of near critical links, Figure \[fig.three\](a), indicating such models do not produce power-law rheology.
Before proceeding, some discussion of the model’s strain rate, $\dot{\gamma}$, is in order. We suppose that in cells $\dot{\gamma}$ is caused by molecular motor sliding or filament treadmilling at a roughly constant velocity. Rather than a uniform, pure shear deformation on the cellular scale, a spatially random deformation field with a typical, mesoscopic strain rate $\dot{\gamma}$ would serve just as well. We further suppose that other active processes continuously ‘remodel’ the network with a turnover rate comparable to $\dot{\gamma}$, such that typical network segments are strained to $\gamma \sim 1$ in a dynamic steady state that replicates the non-steady behavior of our model at $\gamma \sim 1$. To be consistent with 100-1000 second estimates for cytoskeletal turnover, $\dot{\gamma}$ should thus be in the range $10^{-3}$-$10^{-2}$ sec$^{-1}$.
The computed rheology reproduces the cell response qualitatively, but falls short of quantitative replication. The experimental literature reports at least five frequency decades of power-law scaling with exponents in the range $0.10 < \beta < 0.25$, while our model only yields five decades for $\beta < 0.17$ (or higher exponents over a narrower frequency range). This is due to the modest ‘height’ of our exponential force distribution, $P(F_c')/P(F$=$0) \approx 3$, as in Figure \[fig.three\](a). Recent athermal simulations using a more realistic network structure yield a more pronounced exponential $P(F)$ [@didonna-preprint], suggesting our limited frequency range may be an artifact of our simplified network geometry. To estimate the model parameters corresponding to the physical case, we dimensionalized our model, added a high-frequency contribution $G^* \sim \omega^{0.75}$, and compared it to literature data [@fabry]. Our model reproduces the response of normal HASM cells, Figure \[fig.four\](bottom), with physically plausible parameter values: $E_B=25$, $\tau_a=70$ nsec and $G_o=5$ kPa. The simulated strain rate, $\dot{\gamma} \approx 7$ sec$^{-1}$, is higher than our earlier estimate; smaller dimensionless strain rates were too computationally intensive to be accessible. Because of the slow variation of model rheology with $\dot{\gamma}$, however, a more realistic value $\dot{\gamma} = 10^{-2}$ sec$^{-1}$, should yield a similar response with slightly smaller exponent, $\beta$$\approx$$0.15$, and narrow the frequency range with power law rheology by $(700)^{0.18} = 3.2\times$, or half a decade.
The arrangement of structural molecules on the cusp of conformational change suggests an optimal configuration for a chemical mechano-sensor—suggesting that cells may maintain a metabolically costly dynamic cytoskeleton as much for its sensory as its structural functions. Unlike other proposed sensor mechanisms [@janmeyc; @tamada] that transduce molecular stress, the earlier noted correspondence of unfolding and network strain indicates the sensing of deformation on the supramolecular scale. Many cross-linking species specifically bind a number of signaling proteins, including heat shock proteins, protein kinase C, Ral A, PIP2, PIP3, PI3-kinase, and MEKK1 (for reviews [@stossel; @otey]). Any of these proteins that specifically binds (or unbinds) cross-link domains upon forced unfolding would then transduce the shear or extensional strain of the network, which is presumably a prerequisite for shape or matrix compliance sensing. Furthermore, the myriad, multi-domain cross-link proteins and their isoforms localized to different parts of the cell suggest that each different cell sub-structure could have its own mechano-sensing capability.
Our cross-link unfolding model reproduces the cell response and makes biochemically testable predictions: that some cross-link species should be partially unfolded under normal physiological conditions, and that their unfolding should increase with cell deformation. Such biochemical studies, combined with rheology simulations having more realistic network structure, hold the prospect of a cytoskeleton model grounded in polymer and single-molecule biophysics, one that can be integrated with mechano-sensory signaling pathways.
We thank A Bausch, B DiDonna, JJ Fredberg, M Gardel, A Levine, T Lubensky, P Janmey and D Weitz, for useful conversations. Support came from the David and Lucile Packard Foundation, the Bourse Lavoisier and Penn’s Ashton Fellowship.
$^\dagger$Present Address: Laboratoire de Spectometrie Physique, Université Joseph Fourier, 38402 St Martin d’Hères, France.
[26]{} natexlab\#1[\#1]{}bibnamefont \#1[\#1]{}bibfnamefont \#1[\#1]{}citenamefont \#1[\#1]{}url \#1[`#1`]{}urlprefix\[2\][\#2]{} \[2\]\[\][[\#2](#2)]{}
, , , , , ****, ().
, ****, ().
, , , ****, ().
, , , , , ****, ().
, ****, ().
, ****, ().
, , , ****, ().
, ****, ().
, ****, ().
, ****, ().
, ****, ().
, ****, ().
, ****, ().
, ****, ().
, ****, ().
, ****, ().
, ****, ().
, ****, ().
, ****, ().
, ****, ().
, ****, ().
,
, ****, ().
, ****, ().
, ****, ().
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We present a robust scheme by which fractional quantum Hall states of bosons can be achieved for ultracold atomic gases. We describe a new form of optical flux lattice, suitable for commonly used atomic species with groundstate angular momentum $J_g = 1$, for which the lowest energy band is topological and nearly dispersionless. Through exact diagonalization studies, we show that, even for moderate interactions, the many-body groundstates consist of bosonic fractional quantum Hall states, including the Laughlin state and the Moore-Read (Pfaffian) state. Under realistic conditions, these phases are shown to have energy gaps that are larger than temperature scales achievable in ultracold gases.'
author:
- 'Nigel R. Cooper$^1$ and Jean Dalibard$^{2,3}$'
date: 14 December 2012
title: Reaching Fractional Quantum Hall States with Optical Flux Lattices
---
There is intense interest in finding new settings in which topological phases of matter analogous to fractional quantum Hall (FQH) states appear. Ultracold atomic gases are ideal systems with which to achieve this goal: they allow studies of strong correlation phenomena for both fermions and bosons, and FQH physics can be approached for homogeneous fluids [@advances] as well as for atoms confined in optical lattices [@lewensteinrev].
While existing theories of FQH-like phases in lattices have focussed on tight-binding models [@sorensen:086803; @mollercooper-cf; @mazzamr; @PhysRevLett.106.236804; @shenggusunsheng; @regnaultlattice; @hormozi], one of the most promising routes to topological flat bands for ultracold atoms is through optical flux lattices (OFLs) [@ofl; @cooperdalibard; @coopermoessner]. An OFL uses a set of laser beams to produce a spatially periodic atom-laser coupling that induces resonant transitions between two (or more) internal atomic states. The resulting energy bands, in particular the lowest one, have non-zero Chern numbers, and can be made narrow in energy [@coopermoessner]. This opens the path to experimental studies of novel strong correlation phenomena in topological flat bands, notably the FQH effect of bosons.
We present in this paper the first characterization of the many-body ground state of bosons in an OFL. We start with the design of a novel type of OFL, which fully exploits the structure of the most commonly used (bosonic) atomic species. The scheme is robust since, by contrast to some other OFL proposals [@ofl], it does not require phase locking between the various optical beams composing the lattice. For optimized parameters its lowest band has Chern number 1 and is nearly dispersionless, closely analogous to the lowest Landau level for charged particles moving in a uniform magnetic field. We use exact diagonalization to determine the many-body spectrum of a bosonic gas in this OFL. We show that FQH ground states appear for relatively weak atom interaction at the same filling factors as for a continuum Landau level [@advances]. Our work provides a concrete experimental scheme by which FQH states of bosons can be realized with large energy scales. Furthermore, it provides the first example of a non-Abelian quantum Hall state (the $\nu=1$ Moore-Read state [@MooreR91]) in a lattice model at high particle density with only two-body interactions.
We focus in this paper on the case of atoms whose internal ground level has angular momentum $J_g=1$. This is the case for several stable bosonic isotopes of alkali metal species, namely $^7$Li, $^{23}$Na, $^{39}$K, $^{41}$K, $^{87}$Rb. We denote $|X\rangle$, $|Y\rangle$, $|Z\rangle$ a basis of this level, defined such that $\hat J_X|X\rangle=0$ (and similarly for $Y$ and $Z$). Here the set of directions $X,Y,Z$ represents an orthogonal trihedron of the physical space \[see Figure \[fig:Fig1\](a)\] and $\hat J_X$ stands for the component of the angular momentum operator along the $X$ direction. Note that one can replace $|X\rangle$, $|Y\rangle$, $|Z\rangle$ by a triplet of internal states selected among a more complex level scheme, as proposed [*e.g.*]{} in [@Campbell:2011]. Our scheme will apply as long as each pair of states can be coupled by a resonant two-photon Raman transition with a negligible spontaneous emission rate [@external].
We assume that $|X\rangle$, $|Y\rangle$, $|Z\rangle$ are the eigenstates of the atomic Hamiltonian in the absence of atom-laser coupling. We suppose that these three states are non-degenerate and non-equally spaced, and their energies are such that $E_X < E_Y<E_Z$, with $E_Z-E_Y \neq E_Y-E_X$. For alkali atoms this situation can be reached by illuminating the atomic sample with microwaves close to the hyperfine resonance (see supplementary information). We denote by $z$ the $(1,1,1)$ direction of the $X,Y,Z$ trihedron, and assume that the centre-of-mass motion of the atoms along the direction $z$ is frozen. Therefore we consider in the following only the atomic motion in the perpendicular $xy$ plane \[see Figure \[fig:Fig1\](a)\].
-6mm -5mm{width="4.2cm"} {width="135mm"} -30mm [(a) (b) (c) (d) ]{}
The atoms are irradiated with laser beams propagating in the $xy$ plane along three directions making an angle of $2\pi/3$ with each other. The three wave vectors are ${\boldsymbol}k_1=k/2\, (\sqrt 3\, {\boldsymbol}u_x +{\boldsymbol}u_y)$, ${\boldsymbol}k_2=k/2\, (-\sqrt 3\, {\boldsymbol}u_x +{\boldsymbol}u_y)$ and ${\boldsymbol}k_3=-k{\boldsymbol}u_y$, where $\{{\boldsymbol}u_x,{\boldsymbol}u_y\}$ is an orthogonal unit basis of the $xy$ plane. Here $k$ stands for the typical wave number of the laser beams [@Note1]. We choose the frequency components in each laser beam so that an atom can undergo resonant Raman transitions between the three internal states, by absorbing a photon in one wave and emitting a photon in a stimulated manner in another wave. The momentum change in such a transition is $\pm {\boldsymbol}q_i$, where ${\boldsymbol}q_i={\boldsymbol}k_{i}-{\boldsymbol}k_{i+1}$. Here we set ${\boldsymbol}k_4\equiv {\boldsymbol}k_1$ and take $\hbar=1$ for simplicity.
Suppose first that each beam $i$ consists only of a monochromatic plane wave with frequency $\omega_i$ and phase $\varphi_i$, and that the $\omega_i$’s are chosen such the three Raman conditions are fulfilled: $\omega_1-\omega_2=E_Y-E_X$, $\omega_2-\omega_3=E_Z-E_Y$ \[and thus $\omega_1-\omega_3=E_Z-E_X$, see Figure \[fig:Fig1\](b)\]. Each family of momentum eigenstates ${\cal F}({\boldsymbol}p)=\{ |X,{\boldsymbol}p-{\boldsymbol}k_1\rangle, |Y,{\boldsymbol}p-{\boldsymbol}k_2\rangle, |Z,{\boldsymbol}p-{\boldsymbol}k_3\rangle \}$ generates a manifold that is globally stable with respect to atom-laser coupling. The three states of ${\cal F}({\boldsymbol}p)$ form an equilateral triangle in momentum space \[Figure \[fig:Fig1\](c)\]. The coupling between the atom and the laser field can be written (see supplementary material) $$\begin{aligned}
\hat{V}=- \Omega &\Big(& |Y\rangle \langle X| e^{i ({\boldsymbol}q_1\cdot {\boldsymbol}r +\varphi_1-\varphi_2)}+|Z\rangle \langle Y| e^{i ({\boldsymbol}q_2\cdot {\boldsymbol}r +\varphi_2-\varphi_3)}\nonumber \\
&+&|X\rangle \langle Z| e^{i ({\boldsymbol}q_3\cdot {\boldsymbol}r +\varphi_3-\varphi_1)}\Big)\ +\ \mbox{H.c.},
\label{eq:coupling}\end{aligned}$$ where H.c. stands for Hermitian conjugate. The amplitude and sign of the coupling strength $\Omega$ can be adjusted by tuning the intensity of the coupling lasers, and their detuning with respect to the atomic resonance. The fact that all three Raman transitions in Eq. (\[eq:coupling\]) have the same amplitude is ensured by (i) taking the same intensity for each laser beam, (ii) choosing in-plane linear polarizations. A similar ring-coupling scheme has been used in [@2012:Jimenez] to implement the Peierls substitution in a 1D optical lattice. However in [@2012:Jimenez] only two laser Raman transitions were used and the ring was closed using radio frequency transitions, which is not appropriate for our purpose.
With only one triplet of laser frequencies as in Fig. \[fig:Fig1\](b), we do not produce the desired infinite periodic lattice for the atomic motion in momentum space [@coopermoessner]. However this goal can be reached by adding inside the beams $i$ two other triplets of frequency components $\omega'_i$ and $\omega''_i$, $i=1,2,3$. Here the roles are circularly exchanged with respect to the first triplet $\omega_i$: The $\omega'_i$ (resp. $\omega''_i$) are such that $\omega'_2-\omega'_3=E_Y-E_X$ and $\omega'_3-\omega'_1=E_Z-E_Y$ (resp. $\omega''_3-\omega''_1=E_Y-E_X$ and $\omega''_1-\omega''_2=E_Z-E_Y$). In the following we suppose that the differences between the average frequencies $\bar \omega, \bar \omega', \bar \omega''$ of the triplets are much larger than the splittings $E_\alpha-E_\beta$. Processes involving the absorption of a photon from a frequency triplet and the stimulated emission of a photon in another triplet thus play a negligible role.
With the three frequency triplets acting simultaneously on an atom, the family of states that are coupled to a given initial state can be represented by the infinite lattice in momentum space shown in Fig. \[fig:Fig1\](d). Since there are 3 possible Raman transitions and 3 possible pairs of beams to induce a given transition, the atom-laser coupling $\hat V$ generalizing (\[eq:coupling\]) is now characterized by 9 matrix elements. These elements depend on the 9 phases $\varphi_i,\varphi'_i,\varphi''_i$ and are summarized in Table 1. From this Table, it is straightforward to write down explicitly the coupling $\hat V$. For example the three terms appearing in (\[eq:coupling\]) correspond to the diagonal terms of the array of Table \[table\].
$X \rightarrow Y$ $Y \rightarrow Z$ $Z \rightarrow X$
-------------------- ---------------------------------- ---------------------------------- ----------------------------------
${\boldsymbol}q_1$ $e^{i(\varphi_1-\varphi_2)}$ $e^{i(\varphi''_1-\varphi''_2)}$ $e^{i(\varphi'_1-\varphi'_2)}$
${\boldsymbol}q_2$ $e^{i(\varphi'_2-\varphi'_3)}$ $e^{i(\varphi_2-\varphi_3)}$ $e^{i(\varphi''_2-\varphi''_3)}$
${\boldsymbol}q_3$ $e^{i(\varphi''_3-\varphi''_1)}$ $e^{i(\varphi'_3-\varphi'_1)}$ $e^{i(\varphi_3-\varphi_1)}$
: Phases of the Raman coupling matrix elements. Each line corresponds to a given momentum kick ${\boldsymbol}q_i={\boldsymbol}k_i-{\boldsymbol}k_{i+1}$, and each column to a given pair of internal atomic states. This $3\times 3$ array can be understood as a determinant: each of the six terms appearing in the calculation of this determinant corresponds to one of the six types of triangles in Fig. \[fig:Fig1\](d). The terms with positive (resp. negative) sign in the determinant calculation are for the upwards (resp. downwards) pointing triangles. []{data-label="table"}
In order to characterize the possible non-trivial topology associated with the lattice in momentum space, we now evaluate the total phase gained by an atom when it undergoes a series of Raman transitions $X\to Y\to Z \to X$ and performs a closed loop in momentum space. This corresponds to traveling around the three sides of one of the triangles of Fig. \[fig:Fig1\](d). The resulting phase is different for upwards pointing triangles \[such as the one of Fig. \[fig:Fig1\](c)\] and downwards pointing ones \[like the triangles labelled $\alpha,\beta,\gamma$ in Fig. \[fig:Fig1\](d)\]. For an upwards pointing triangle, the global phase is always zero. Indeed moving around the sides of such a triangle involves absorption and stimulated emission of photons whose frequencies belong to the same triplet, [*e.g.*]{} the $\omega_i$ triplet for the triangle of Fig. \[fig:Fig1\](c). Therefore each laser phase $\varphi_i$ (or $\varphi'_i$, $\varphi''_i$) enters both with a $+$ and a $-$ sign in the total accumulated phase around such a triangle, leading to a null result.
Downwards pointing triangles on the other hand correspond to a non-trivial phase. Consider for example the clockwise oriented path around the sides of the triangle labelled $\alpha$ in Fig. \[fig:Fig1\](d): (i) The $X\to Y$ transition is accompanied by a change of atomic momentum ${\boldsymbol}q_2$, and it corresponds to a phase change $\varphi'_2-\varphi'_3$ (see Table \[table\]); (ii) the $Y\to Z$ transition is along ${\boldsymbol}q_1$, with the phase change $\varphi''_1-\varphi''_2$; (iii) the $Z\to X$ transition is along ${\boldsymbol}q_3$, with the phase change $\varphi_3-\varphi_1$. As a result, the phase accumulated when traveling around the sides of triangle $\alpha$ is $$\Phi_\alpha=\varphi''_1-\varphi_1+\varphi'_2-\varphi''_2+\varphi_3-\varphi'_3.
\label{eq:triangle_phase}$$ We can similarly calculate the phases $\Phi_{\beta,\gamma}$ for the two other downwards pointing triangles. Although the sum $\Phi_{\alpha}+\Phi_{\beta}+\Phi_{\gamma}$ is always zero, we can identify configurations such that each of these three phases takes a non-trivial value. For example the choice $\varphi_1=2\pi/3$, $\varphi_3=-2\pi/3$, and all other phases equal to zero yields $$\Phi_{\alpha}=\Phi_{\beta}=\Phi_{\gamma}=2\pi/3 \quad \mbox{mod }2\pi.
\label{eq:triangle_phase_choice}$$ From now on, we will stick to this choice, together with the assumption that $\Omega>0$, which is obtained for an alkali-metal atom by tuning the lasers between the $D_1$ and $D_2$ resonance lines.
The OFL formed in this way has a reciprocal lattice spanned by the basis vectors ${\bm G}_1 = 3 {\bm q}_1$ and ${\bm G}_2 = {\bm
q}_2$. The real space lattice vectors are ${\bm a}_1 =
\frac{2\pi}{3\sqrt{3}q}(\sqrt{3}\,{\boldsymbol}u_x + {\boldsymbol}u_y)$ and ${\bm a}_2 =
\frac{4\pi}{\sqrt{3}q}{\boldsymbol}u_y$, where $q = |{\bm q}_i| =
\sqrt{3}k$. This geometry is equivalent to that of the three-state triangular flux lattice of Ref. . However, the pattern of phases in the reciprocal space tight binding model differs: here we have fluxes of $0$ and $2\pi/3$ in the upwards and downwards pointing triangles, as opposed to $\pi/3$ for each [@coopermoessner]. Nevertheless, the physical properties of the OFLs are very similar: in each unit cell of the real space lattice the lowest energy dressed state experiences $N_\phi=1$ flux quantum; the resulting bandstructure shows low energy bands that are analogous to Landau levels. In particular, the lowest energy band has Chern number of $1$, and very narrow energy width, $W$, over a broad range of lattice depths $\Omega$. Here we focus on a lattice of depth $\Omega = 3E_{\rm R}$ \[where $E_{\rm R} \equiv q^2/(2m)$ is the recoil energy for atomic mass $m$\] close to which this bandwidth has a (local) minimum of $W\simeq 0.015 E_{\rm R}$. In view of this very small bandwidth, the system is highly susceptible to the formation of strongly correlated phases even for relatively weak interactions.
We have used exact diagonalization to study the groundstates of interacting bosons occupying the lowest energy band of the OFL for $\Omega = 3 E_{\rm R}$. (We neglect the population of higher bands, since the gap to the next band is very large, $\Delta \simeq 46\, W$.) We consider the bosons to interact via spin-independent contact interactions, which is a good approximation for $^{87}$Rb. We write the two-dimensional coupling constant as $g_{\rm 2D} =
\frac{\hbar^2}{m} {\tilde g}$, where ${\tilde g}$ is dimensionless. For atoms with 3D scattering length $a_{\rm s}$ restricted to 2D by a harmonic confinement of oscillator length $a_0$, and neglecting (sub)band mixing, this is ${\tilde
g} = \sqrt{8\pi}a_{\rm s}/a_0$ [@hadzibabicdalibard]. We study a finite system in a periodic geometry, with sides ${\bm
L}_1 = N_1 {\bm a}_1$ and ${\bm L}_2 = N_2 {\bm a}_2$, where $N_{1,2}$ are integers. The total flux is then $N_\phi =
N_1 N_2$, so for $N$ particles the Landau level filling factor is $\nu\equiv N/N_\phi$. The interacting many-particle states can be classified by a conserved crystal momentum at $N_\phi$ points in the Brillouin zone. We construct the Hamiltonian at each crystal momentum and use a standard Lanczos method to determine the low energy spectrum.
For very weak interactions, ${\tilde g}\ll 1$, the bosons form a condensate in the minima of the band dispersion. However, our numerical results show that, even for moderate interaction strength ${\tilde g}\gtrsim 0.2$, this (compressible) condensed phase is replaced by strongly correlated (incompressible) FQH states at filling factors $\nu=1/2,2/3,3/4$ and 1. Here, we focus on the FQH states at $\nu=1/2$ and $1$. (Results for $\nu=2/3, 3/4$ are described in the Supplementary Material.)
Evidence for the appearance of incompressible phases is found by calculating the discontinuity in the chemical potential $\Delta \mu$ for the groundstate: the difference between the chemical potential for adding a particle and that for removing a particle. A non-zero and positive $\Delta \mu$ indicates that the system is incompressible. To minimize finite-size effects we define[@cwg] $\Delta\mu \equiv N\left[E_{N+1}/(N+1)+E_{N-1}/(N-1) -
2E_N/N\right]$, where $E_N$ is the groundstate energy for $N$ particles.
In Fig. \[fig:incompressible\] we plot the dependence of $\Delta
\mu$ on interaction strength ${\tilde g}$ at filling factors $\nu=1/2$ and $1$. For $\nu=1/2$ there is an onset of incompressibility for ${\tilde g} \gtrsim 0.2$, and for $\nu=1$ incompressibility appears for ${\tilde g}\gtrsim 0.4$. In the thermodynamic limit, $N\to
\infty$, the transitions from compressible $\Delta \mu=0$ to incompressible $\Delta \mu>0$ should be sharp, and can even be discontinuous for first-order transitions, but are rounded in Fig. \[fig:incompressible\] by finite-size effects. The observed rises of $\Delta \mu$ are indications of the approximate values of ${\tilde g}$ at which there are transitions into the incompressible phases.
![\[fig:incompressible\] Incompressibility, as measured by the discontinuity in the chemical potential $\Delta \mu$ defined in the text, as a function of interaction strength ${\tilde g}$ at several filling factors: $\nu=1/2$ (for $N=9$ bosons in a system of size $N_\phi = N_1 N_2 = 6\times 3$, circles); and $\nu=1$ ($N=12$ in $N_\phi = 6\times 2$, triangles). The inset shows $\Delta \mu$ as a function of filling factor for a series of interaction strengths ${\tilde g}$.](fig2){width="0.9\columnwidth"}
To explore the nature of these incompressible phases it is instructive to study their (neutral) excitation spectra in the strong-interaction limit ${\tilde g}\to \infty$. These spectra, Fig. \[fig:spectrum\], show all the expected properties of the bosonic Laughlin ($\nu=1/2$) and Moore-Read ($\nu=1$) states. On this periodic geometry, these topologically ordered incompressible phases should show groundstate degeneracies (of 2 and 3 respectively) in the thermodynamic limit, separated by an energy gap from the remaining excitations. As shown in Fig. \[fig:spectrum\], even for these finite systems these groundstate degeneracies appear clearly. Results on other system sizes and geometries (not shown) are consistent with these results, confirming that these near degeneracies are robust features, not imposed by symmetries, that characterize these topological phases.
The FQH states that we find for the OFL (at $\nu = 1/2,2/3,3/4,1$) are the same as those found for contact interacting bosons in the continuum lowest Landau level (LLL) [@advances]. We have established the equivalence of the phases of these two models by studying the evolution of the many-body spectrum for a series of Bloch wavefunctions that interpolate between those of the lowest band of the OFL and those of the LLL. To do so, we consider a fictitious atom with $N_{\rm s} =12$ internal states, and represent the LLL by the $N_{\rm
s} = 12$ triangular OFL of Ref. , the lowest band of which has properties that are indistinguishable from those of the LLL for suitable coupling $\Omega'\simeq 10 E_{\rm
R}$ [@corrections]. We place 9 additional internal states at the midpoints of the bonds of Fig.\[fig:Fig1\](d), coupled to each other and to the original states $X,Y,Z$ by bonds of strength $\Omega'$ and with $\pi/12$ flux through each new triangular plaquette [@flow]. Choosing $(\Omega,\Omega') = (3(1-\lambda), 10\lambda)E_{\rm R}$ and varying $\lambda$ leads to smooth interpolation of the lowest energy band and the many-body spectrum, from those of the present model ($\lambda=0$) to those of the LLL ($\lambda=1$). In all cases ($\nu =
1/2,2/3,3/4,1$) the energy gap remains open, showing that the phases of these two models are the same [@scaffidimoller; @wujainsun]. Indeed, there is very little change in the spectrum, showing that this OFL (with $\Omega = 3 E_{\rm R}$) is a very close representation of the LLL. For example, for the LLL the $\nu=1/2$ state has zero interaction energy, as the two-body correlation function vanishes exactly at zero range. Here, the $\nu=1/2$ state in the OFL has $E_N/N\simeq 7\times 10^{-5} {\tilde g} E_{\rm R}$ \[see Fig. \[fig:spectrum\](a)\] showing that the zero-range two-body correlation function nearly vanishes.
![\[fig:spectrum\] Low-energy spectra for the OFL with $\Omega/E_{\rm R} = 3$ in the strong-interaction limit ${\tilde g}\to \infty$ at filling factors (a) $\nu=1/2$ ($N=9$ bosons in $N_\phi = N_1N_2 = 6\times 3$) and (b) $\nu=1$ ($N=12$ bosons in $N_1N_2 = 6\times 2$). The crystal momentum ${\bm k} \equiv \alpha_1 {\bm G}_1/N_1 + \alpha_2 {\bm G}_2/N_2$ is labelled by the index, $i=1+\alpha_1 + N_1 \alpha_2$ for $\alpha_1=0,\ldots,N_1-1$ and $\alpha_2=0,N_2-1$. The quasi-degenerate groundstates have the expected multiplicities and crystal momenta for the Laughlin state ($\nu=1/2$), and the Moore-Read state ($\nu =1$).](fig3 "fig:"){width="1.0\columnwidth"} -0.5cm -3.5cm (a)(b)
To summarize we have proposed a robust atom-laser configuration that can lead to FQH states of bosons in a well-accessible range of parameters. The robustness of the setup is ensured by the absence of need for a stabilization of the relative phases of different beams [@Note2]. The only phase difference to be controlled is within each single-mode laser beam ($\varphi_i,
\varphi'_i,\varphi''_i$) and can be set by acoustic-optic modulators driven by programmable function generators. The lowest energy band is insensitive to fluctuations in the laser amplitudes around $\Omega = 3E_{\rm R}$, its bandwidth increasing by less than $10^{-3} E_{\rm R}$ within the range $\Omega/E_{\rm R} = 2 - 4$, and its topology remaining unchanged. The minimal interaction strength ${\tilde g} \approx 0.2$ for obtaining FQH states corresponds to a 2D confinement frequency of $\gtrsim 7\,\mbox{kHz}$ for Rb, which is readily achieved in an optical lattice. A clear signal of the formation of strongly correlated phases would be the appearance of density plateaus (wedding cake structure) in in-situ images of the gas, arising from incompressibility $\Delta \mu>0$. This requires the temperature to be smaller than $\Delta \mu$, which for the Laughlin state we find from Fig. \[fig:incompressible\] to be $\approx 0.02 E_{\rm R}$ for ${\tilde g} = 0.4$, that is $10$nK for $^{87}$Rb.
0.1cm
[10]{}
N. R. Cooper, Advances in Physics [**57**]{}, 539 (2008).
M. Lewenstein [*et al.*]{}, Advances in Physics [**56**]{}, 243 (2007).
A. S. S[ø]{}rensen, E. Demler, and M. D. Lukin, Phys. Rev. Lett. [**94**]{}, 086803 (2005).
G. Möller and N. R. Cooper, Phys. Rev. Lett. [**103**]{}, 105303 (2009).
L. Mazza, M. Rizzi, M. Lewenstein, and J. I. Cirac, Phys. Rev. A [**82**]{}, 043629 (2010).
T. Neupert, L. Santos, C. Chamon, and C. Mudry, Phys. Rev. Lett. [**106**]{}, 236804 (2011).
D. Sheng, Z.-C. Gu, K. Sun, and L. Sheng, Nat. Commun. [**2**]{}, 389 (2011).
N. Regnault and B. A. Bernevig, Phys. Rev. X [**1**]{}, 021014 (2011).
L. Hormozi, G. M[" o]{}ller and S. H. Simon, Phys. Rev. Lett. [**108**]{}, 256809 (2012).
N. R. Cooper, Phys. Rev. Lett. [**106**]{}, 175301 (2011).
N. R. Cooper and J. Dalibard, Europhysics Letters [**95**]{}, 66004 (2011).
N. R. Cooper and R. Moessner, Phys. Rev. Lett. [**109**]{}, 215302 (2012).
G. Moore and N. Read, Nucl. Phys. B [**360**]{}, 362 (1991).
D. L. Campbell, G. Juzeli |[u]{}ūnas, and I. B. Spielman, Phys. Rev. A [**84**]{}, 025602 (2011).
The three states $|X\rangle$, $|Y\rangle, |Z\rangle$ can also stand for different motional states, such as the first three vibrational states along the strongly confined direction $z$. However in this case, the hypothesis of a single, state-independent interaction strength should be relaxed.
Although these beams do not have exactly the same frequency, the difference between wave numbers is negligible.
K. Jiménez-Garcia [*et al.*]{}, Phys. Rev. Lett. [**108**]{}, 225303 (2012).
Z. Hadzibabic and J. Dalibard, Rivista del Nuovo Cimento [**34**]{}, 389 (2011).
N. R. Cooper, N. K. Wilkin, and J. M. F. Gunn, Phys. Rev. Lett. [**87**]{}, 120405 (2001).
Deviations from the continuum Landau level fall exponentially with $N_{\rm s}$ [@coopermoessner], and are negligible for $N_{\rm s}=12$.
This new model has net flux $\pi/3$ through each triangle of Fig. \[fig:Fig1\](d). In connecting the new ($\Omega'$) bonds to those ($\Omega$) of the original model (with fluxes of $0$ and $2\pi/3$ through upwards and downwards pointing triangles), we assign flux $\pi/3$ through the loop formed by the two sets of bonds spanning each link ${\bm q}_3$.
T. [Scaffidi]{} and G. [M[ö]{}ller]{}, Phys. Rev. Lett. [**109**]{}, 246805 (2012).
Y.-H. Wu, J. K. Jain, and K. Sun, Phys. Rev. B [**86**]{}, 165129 (2012).
This robustness shows up in the evaluation of the phase around the sides of any triangle of Fig. \[fig:Fig1\](d): each beam $i$ enters both with a positive sign ([*i.e.*]{}, absorption) and a negative sign ([*i.e.*]{}, stimulated emission), so that its path length cancels out.
**Supplementary material**
The basis set $|X\rangle,|Y\rangle,|Z\rangle$
=============================================
The components $\hat J_X$ and $\hat J_Y$ of a spin 1 angular momentum operator read in the standard eigenbasis $\{|m_Z=+1\rangle, |m_Z=0\rangle, |m_Z=-1\rangle\}$ of $\hat J_Z$: $$\hat J_X=\frac{1}{\sqrt 2}\begin{pmatrix}
0& 1 & 0\\
1 &0 &1\\
0 & 1 & 0
\end{pmatrix}
\quad
\hat J_Y=\frac{1}{\sqrt 2}\begin{pmatrix}
0& -i & 0\\
i &0 &-i\\
0 & i & 0
\end{pmatrix} .
\label{eq:JX_JY}$$ In the standard eigenbasis $|m_Z=0,\pm 1\rangle$, we define the eigenvectors $|\alpha\rangle$ of $\hat J_\alpha$ ($\alpha=X,Y,Z$) with zero eigenvalue: $$|X\rangle =\frac{1}{\sqrt 2}\begin{pmatrix}
-1 \\ 0 \\ 1
\end{pmatrix}
\quad
|Y\rangle =\frac{i}{\sqrt 2}\begin{pmatrix}
1 \\ 0 \\ 1
\end{pmatrix}
\quad
|Z\rangle =\begin{pmatrix}
0 \\ 1 \\ 0
\end{pmatrix}.
\label{eq:X_Y_Z}$$ These three states also form an orthonormal basis set (denoted hereafter the Cartesian basis) and are such that $$\hat J_X|Y\rangle = i |Z\rangle, \quad \hat J_Y|Z\rangle = i |X\rangle, \quad \hat J_Z|X\rangle = i|Y\rangle .
\label{eq:circular}$$ For any real 3-vector ${\boldsymbol}u$, the state $u_X|X\rangle +u_Y|Y\rangle +u_Z|Z\rangle$ is the eigenstate of ${\boldsymbol}u\cdot \hat {{\boldsymbol}J}$ with eigenvalue 0.
Microwave dressing of a spin-1 ground state
===========================================
We consider an atom from the alkali-metal family with a nuclear spin $I=3/2$, so that its ground level is split in two hyperfine levels with angular momentum $F=1$ and $F=2$. The atom is irradiated with a linearly polarized microwave (mw), whose frequency is detuned by $\Delta_{\rm mw}$ from the frequency $\omega_{\rm hf}$ of the $F=1 \leftrightarrow F=2$ hyperfine transition. The coupling between the mw and the atom (half-Rabi frequency) is defined by $\kappa_{\rm mw}=\mu_{\rm B} B_{\rm mw}/2$, where $\mu_{\rm B}$ is the Bohr magneton and $B_{\rm mw}$ is the amplitude of the oscillating mw magnetic field. We restrict to the case where $\kappa_{\rm mw} \ll \Delta_{\rm mw} \ll \omega_{\rm hf}$. We can then treat the atom-mw coupling using the rotating wave approximation and evaluate the shifts of the Zeeman states of the $F=1$ level using second-order perturbation theory. Taking the quantization axis parallel to the polarization axis ($Z$) of the microwave, the shifts of the states $m_Z=0,\pm 1$ read [@Gerbier:2006] $$\Delta E({m_Z})=\frac{\kappa_{\rm mw}^2}{\Delta_{\rm mw}} \left(1-\frac{m_Z^2}{4}\right).
\label{eq:mw_shift}$$ The action of the mw on the $F=1$ hyperfine level can thus be described by the operator $\alpha +\beta \hat J_Z^2$, where $\beta=-\kappa_{\rm mw}^2/(4\Delta_{\rm mw})$.
Suppose now that a second mw, at a different detuning $\Delta'_{\rm mw}$ and with a linearly polarization along $X$, also irradiates the atom. In the perturbative framework used above, the combined action of the two mws is described by the effective Hamiltonian $H_{\rm mw,eff}=\beta \hat J_Z^2 + \beta' J_X^2$, up to an additive constant. The eigenstates of $H_{\rm mw,eff}$ are the states $|X\rangle, |Y\rangle,|Z\rangle$ with the energies $\beta$, $\beta+\beta'$, $\beta'$, respectively.
To estimate the required value for $B_{\rm mw}$, we start by noticing that when writing the atom-laser coupling \[Eqn (1) of the main text\], we only take into account the resonant elements. For example we assume that the atom can undergo a transition $|X\rangle\to|Y\rangle$ via the absorption of a photon ${\boldsymbol}k_1$ and the emission of a photon ${\boldsymbol}k_2$, and we neglect the transitions $|X\rangle\to|Y\rangle$ occurring via the absorption of ${\boldsymbol}k_2$ and emission of ${\boldsymbol}k_1$. This is legitimate when the two-photon coupling $\Omega$ is small compared to the energy difference between $|X\rangle$, $|Y\rangle$, $|Z\rangle$: $\Omega\ll \beta,\beta'$. We also note that the perturbative approach leading to (\[eq:mw\_shift\]) is valid only when $\kappa_{\rm mw}\ll \Delta_{\rm mw}$. Requiring that the two sides of each strong inequality differ by at least a factor 5, we find $\Omega \lesssim \kappa_{\rm mw}/100$. Consider as an illustration the case of Rubidium atoms for which the optimal two-photon coupling $\Omega=3 q^2/(2m)=9 k^2/(2m) \approx 2\pi \times 30$kHz. The microwave coupling $\kappa_{\rm mw}$ has to be $\gtrsim 3$MHz, corresponding to a microwave magnetic field of $\gtrsim 4$G. This is a large, but still realistic value, especially if one uses small resonant loops to increase the value of $B_{\rm mw}$ at the location of the atomic sample.
The light-shift operator for an alkali-metal atomic scheme
==========================================================
![Amplitude of the couplings for the two components D$_1$ (top) and D$_2$ (bottom) of the resonance line of an alkali atom. Here the spin of the nucleus is supposed to be zero. The coupling amplitudes for the $g_-$ state take symmetric values with respect to those indicated for the $g_+$ state.[]{data-label="Fig:CG"}](CG_12.pdf "fig:"){height="2.5cm"} ![Amplitude of the couplings for the two components D$_1$ (top) and D$_2$ (bottom) of the resonance line of an alkali atom. Here the spin of the nucleus is supposed to be zero. The coupling amplitudes for the $g_-$ state take symmetric values with respect to those indicated for the $g_+$ state.[]{data-label="Fig:CG"}](CG_32.pdf "fig:"){height="2.5cm"}
We consider an atom that is irradiated by a monochromatic laser beam of frequency $\omega$. The light-shift operator $\hat{V}$ gives the restriction of the atom-laser coupling to the ground atomic level $g$ at first order in laser intensity [@cct:1972]. In the absence of electron and nuclear spins, the ground and excited states that are involved in the resonant transition of an alkali-metal atom have an angular momentum $J_g=0$ and $J_e=1$, respectively. The atom-laser coupling can be written $$\begin{aligned}
\hat{U} &=& \sum_{\alpha=X,Y,Z} \kappa_\alpha |e, \alpha \rangle \langle g| + \mbox{H.c.}\\
&=& \sum_{m=0,\pm 1} \kappa_m |e, m \rangle \langle g| + \mbox{H.c.}
\label{}\end{aligned}$$ Here the $\kappa_\alpha$, $\alpha=X,Y,Z$, (resp. $\kappa_m$, $m=0,\pm 1$) denote the atom-laser coupling strengths (half-Rabi frequencies) in the Cartesian (resp. standard) basis with $$\kappa_\pm=\frac{1}{\sqrt 2}(\mp \kappa_X + i\kappa_Y),\qquad \kappa_0=\kappa_Z.
\label{}$$ In this “no-spin" case, the light-shift operator simply describes the displacement of the ground state by the quantity $|{\boldsymbol}\kappa|^2/\Delta$, where $\Delta$ is the detuning of the laser with respect to the atomic transition, ${\boldsymbol}\kappa=(\kappa_X,\kappa_Y,\kappa_Z)$ and $$|{\boldsymbol}\kappa|^2= \sum_{\alpha=X,Y,Z} |\kappa_\alpha|^2=\sum_{m=0,\pm 1} |\kappa_m|^2 .
\label{}$$
We now take the spin of the valence electron into account, but still assume that the nucleus has spin zero. The ground level is then a two-dimensional manifold (angular momentum $J_g=1/2$). Because of the fine-structure Hamiltonian, the resonance line of the atom is split into two components D$_1$ and D$_2$, corresponding to the transition from the ground state to the excited states with total angular momentum $J_e=1/2$ and $J_e=3/2$, respectively.
Let us focus first on the light-shift operator associated with the D$_1$ transition, which is dominant if the detuning $\Delta_1=\omega-\omega_1$ of the laser excitation from the D$_1$ line (frequency $\omega_1$) is much smaller than the detuning $\Delta_2$ from the D$_2$ line. Using the well-known Clebsch-Gordan coefficients (see [*e.g.*]{} Figure \[Fig:CG\]) and taking $Z$ as quantization axis, we get in the corresponding $\{ |g_+\rangle, |g_-\rangle \}$ basis: $$\hat V= \frac{1}{3\Delta_1}
\begin{pmatrix}
|\kappa_0|^2+2|\kappa_-|^2 & -\sqrt 2(\kappa_0\kappa_-^* + \kappa_+ \kappa_0^*)\\
-\sqrt 2(\kappa_-\kappa_0^* + \kappa_0 \kappa_-^* ) &
|\kappa_0|^2+2|\kappa_+|^2
\end{pmatrix} .
\label{eq:first_express_U}$$ This coupling can be written in a compact form $$\hat V= \frac{|{\boldsymbol}\kappa|^2}{3\Delta_1} \hat 1+ {\boldsymbol}B_1\cdot \hat{{\boldsymbol}S},
\label{eq:second_express_U}$$ where $\hat 1$ and $\hat S$ are the identity and spin operators for the spin 1/2 ground state, respectively, and where the (real) effective field ${\boldsymbol}B_1$ for the D$_1$ transition is: $${\boldsymbol}B_1=- \frac{2i}{3\Delta_1} {\boldsymbol}\kappa \times {\boldsymbol}\kappa^*.
\label{eq:effective_field}$$
We now take into account both the D$_1$ and D$_2$ transitions. A straightforward generalization of the preceding calculation leads to $$\hat V= A\, \hat 1+ {\boldsymbol}B\cdot \hat{{\boldsymbol}S},
\label{eq:third_express_U}$$ with $$A=\frac{|{\boldsymbol}\kappa|^2}{3} \left( \frac{1}{\Delta_1}+\frac{2}{\Delta_2} \right)
\label{eq:scalar}$$ and $${\boldsymbol}B= \frac{2i}{3} \left( \frac{1}{\Delta_2}- \frac{1}{\Delta_1}\right) {\boldsymbol}\kappa \times {\boldsymbol}\kappa^*.
\label{eq:effectiveB}$$
As a final step we take into account the nucleus spin $I$. As above we consider the case $I=3/2$, which leads to a splitting of the ground level into two hyperfine states of angular momentum $F=1$ and $F=2$. Here we consider the $F=1$ state, with the three Zeeman sublevels $$\begin{aligned}
|F=1, m_F=\pm 1\rangle &=& \mp \frac{\sqrt 3}{2}|\mp \frac{1}{2};\pm\frac{3}{2}\rangle \pm \frac{1}{2}|\pm \frac{1}{2};\pm\frac{1}{2}\rangle, \\
|F=1, m_F=0\rangle &=& -\frac{1}{\sqrt 2} |-\frac{1}{2};\frac{1}{2}\rangle +\frac{1}{\sqrt 2}|\frac{1}{2};-\frac{1}{2}\rangle,\end{aligned}$$ where the state $|m_e;m_n\rangle$ is labelled by the quantum numbers for the projection along the $Z$ axis of the electron spin ($m_e$) and nuclear spin ($m_n$). We suppose that the detunings $\Delta_1$ and $\Delta _2$ of the laser with respect to the excited states $J_e=1/2$ and $J_e=3/2$ are large compared to the hyperfine splittings of these levels. The calculation of the restriction of the coupling (\[eq:third\_express\_U\]) to the $F=1$ ground level then gives: $$\hat V= A\, \hat 1 + {\boldsymbol}B'\cdot \hat{{\boldsymbol}F},
\label{eq:fourth_express_U}$$ where $${\boldsymbol}B'=-\frac{1}{4}{\boldsymbol}B=\frac{i}{6} \left( \frac{1}{\Delta_1}- \frac{1}{\Delta_2}\right) {\boldsymbol}\kappa \times {\boldsymbol}\kappa^*.
\label{eq:effectiveBfinal}$$ Using (\[eq:circular\]) the vector part of the coupling $\hat U_{\rm vec}=\hat U-A\,\hat 1$ can be written in the Cartesian basis for the $F=1$ ground state: $$\hat V_{\rm vec} = i \left( B'_Z |Y\rangle \langle X| + B'_X |Z\rangle \langle Y| + B'_Y |X\rangle \langle Z|\right) +\mbox{H.c.}
\label{eq:expanded_coupling}$$
![Laser scheme providing a resonant Raman coupling between two sublevels of the $F=1$ ground state.[]{data-label="fig:laser_scheme"}](laser_scheme_v2.pdf){width="4cm"}
Hamiltonian for resonant Raman transitions
==========================================
Consider now a scheme such as the one of Fig. \[fig:laser\_scheme\], where an external field lifts the degeneracy between the three states of the Cartesian basis, and where two monochromatic light waves with the same amplitude (characterized by $\kappa>0$) and with frequency, phase, wave vector and polarizations $\omega_i,\varphi_i,{\boldsymbol}k_i, {\boldsymbol}\epsilon_i$, $i=1,2$ induce a resonant Raman coupling between $|X\rangle$ and $|Y\rangle$.
The transition from $|X\rangle$ to $|Y\rangle$ occurs resonantly via the absorption of a photon in wave 1 and the stimulated emission of a photon in wave 2. Keeping only this resonant process, the relevant contribution to the cross-product ${\boldsymbol}\kappa \times {\boldsymbol}\kappa^*$ entering in (\[eq:effectiveBfinal\]) is $\kappa^2 \left({\boldsymbol}\epsilon_1 e^{i({\boldsymbol}k_1\cdot{\boldsymbol}r+\varphi_1)}\right)\times \left({\boldsymbol}\epsilon_2 e^{i({\boldsymbol}k_2\cdot {\boldsymbol}r +\varphi_2)} \right)^*$, where ${\boldsymbol}r$ is the position of the atom.
In this work we restrict to waves with linear (real) polarizations in the $xy$ plane, making an angle of $2\pi/3$ with each other. In this case ${\boldsymbol}\epsilon_1\times {\boldsymbol}\epsilon_2=(\sqrt{3}/2){\boldsymbol}u_z$. The projection of ${\boldsymbol}u_z$ on each basis vector of the trihedron $X,Y,Z$ is $1/\sqrt 3$, and the value of $B'_Z$ that is relevant for the resonant coupling between $|X\rangle$ and $|Y\rangle$ (see (\[eq:expanded\_coupling\])) is thus $$B'_Z=\frac{i \kappa^2}{12} \left( \frac{1}{\Delta_1}- \frac{1}{\Delta_2}\right) e^{i({\boldsymbol}q_1\cdot{\boldsymbol}r+\varphi_1-\varphi_2)},
\label{}$$ with ${\boldsymbol}q_1={\boldsymbol}k_1-{\boldsymbol}k_2$. This leads to the resonant part of the vector light-shift operator: $$\hat V_{\rm vec}^{\rm (res)}= - \Omega\, e^{i({\boldsymbol}q_1\cdot{\boldsymbol}r+\varphi_1-\varphi_2)}\, |Y\rangle \langle X| +\mbox{H.c.},
\label{}$$ where we have set $$\Omega= \frac{\kappa^2}{12} \left( \frac{1}{\Delta_1}- \frac{1}{\Delta_2}\right) .
\label{}$$ The coupling strength $\Omega$ is positive when the laser is tuned between the D$_1$ and the D$_2$ lines ($\Delta_1>0>\Delta_2$), and negative otherwise.
Consider for example the case of Rubidium atoms and choose the detuning such that the scalar part of the atom laser coupling (\[eq:scalar\]) vanishes: $\Delta_2=-2\Delta_1$, corresponding to the wavelength $\lambda=790$nm ($\lambda_1=795$nm, $\lambda_2=780$nm). The optimal two-photon coupling is $\Omega=3
E_{\rm R}=9 k^2/(2m) \approx 2\pi \times 30$kHz and the photon scattering rate is $\gamma\approx 9\times \Gamma
\kappa^2/(2\Delta_1^2) \approx 20$s$^{-1}$, where the factor 9 accounts for the number of monochromatic beams shining on the atoms. The time scale for establishing a many-body state such as the Laughlin state can be estimated as the inverse of the corresponding gap $\Delta\mu^{-1}$. Taking $\Delta \mu=0.02\,E_{\rm R}$ as a typical value (see Fig. 2 of the main text), we obtain $\Delta\mu^{-1}\approx 1\,$ms. The heating due to photon scattering should then play a minor role during this time.
Additional Numerical Results
============================
We present in this section some additional numerical results alluded to in the main text.
In Fig. \[fig:spectrumall\], we present the excitation spectra for a system of size $N_1\times N_2 = 6\times 2$ at all filling factors ($\nu=1/2,2/3,3/4$ and $1$) for which we find evidence for incompressible FQH states. These show the expected features of the Laughlin state ($\nu=1/2$), composite fermion states ($\nu = 2/3,3/4$) and Moore-Read state ($\nu=1$) of bosons. On this periodic geometry, these topologically ordered incompressible phases should show groundstate degeneracies (of 2,3,4 and 3 respectively) in the thermodynamic limit, separated by an energy gap from the remaining excitations. This is indeed the case, as may be seen clearly in Fig. \[fig:spectrumall\].
![\[fig:spectrumall\] Low-energy spectra for the OFL with $\Omega/E_{\rm R} = 3$ in the strong-interaction limit ${\tilde g}\to \infty$ for a system of size $N_\phi = N_1N_2 =
6\times 2$ at filling factors for which the groundstate is incompressible. The crystal momentum ${\bm k} \equiv \alpha_1 {\bm
G}_1/6 + \alpha_2 {\bm G}_2/2$ is labelled by the index, $i=1+\alpha_1 + 6 \alpha_2$ for $\alpha_1=0,\ldots,5$ and $\alpha_2=0,1$. The quasi-degenerate groundstates have the expected multiplicities and crystal momenta for the Laughlin state ($\nu=1/2$), composite fermion states ($\nu=2/3$, $3/4$), and the Moore-Read state ($\nu =1$).](figS3){width="0.9\columnwidth"}
![\[fig:lll\] Low-energy spectra for contact-interacting bosons in the LLL for a system of the same size and geometry as Fig. \[fig:spectrumall\] ($N_\phi=12$), at filling factors $\nu=1/2,2/3,3/4,1$ for which the groundstate is incompressible. The crystal momentum is labelled as for Fig. \[fig:spectrumall\].](figS4){width="0.9\columnwidth"}
In Fig. \[fig:lll\], we present the equivalent excitation spectra for a lowest band formed from states in the lowest Landau level (LLL). As described in the main text, we represent the LLL by the $N_{\rm s}=12$ OFL lattice of Ref. , the lowest band of which has properties that are indistinguishable from those of the LLL for a lattice coupling of $\Omega' = 10 E_{\rm R}$. This very close equivalence arises from the very rapid convergence of the properties of the $N_{\rm s}$-state OFL lattice of Ref. to those of a charged particle in a uniform magnetic field, the spatial fluctuations of the energy and effective magnetic field experienced by the lowest energy dressed state of the OFL falling exponentially with increasing $N_{\rm s}$, and already negligible for $N_{\rm s}=12$ [@coopermoessner2]. Interpolation between the $N_{\rm s}=3$ OFL of this paper and the LLL leads to a smooth evolution both of the single-particle levels of the lowest bands, and of the many body spectrum for bosons occupying this lowest band. This establishes that the FQH phases of these two systems are the same. Moreover, the spectra of the OFL (Fig. \[fig:spectrumall\]) differ only very slightly from those of the LLL (Fig. \[fig:lll\]). The main qualitative difference is that some approximate degeneracies in Fig. \[fig:spectrumall\] become exact degeneracies connected to the many-body translational symmetry of the LLL [@haldanemtm]. Furthermore, the approximate 3-fold groundstate degeneracy of Fig. \[fig:spectrumall\](d) becomes an exact degeneracy in the LLL Fig. \[fig:lll\](d). This is due to a $\pi/3$ rotational symmetry of the LLL in this geometry of $N_1\times N_2 = 6\times 2$ (for which the sides of the simulation cell have equal length $|{\bm L}_1| =
|{\bm L}_2|$), which transforms the three groundstates of the Moore-Read state in the LLL
{width="0.7\columnwidth"}
into each other. We emphasize that, in general, this three-fold quasi-degeneracy is unrelated to any symmetry, but is an emergent quasi-degeneracy in the thermodynamic limit. This is evidenced by studies on other system sizes and geometries, just as in the LLL [@advances2]. For example, Fig. \[fig:nu1again\] shows the spectrum for the OFL for $N=14$ particles in a system of size $N_1\times N_2 = 7\times 2$ for which no symmetry relates the three quasi-degenerate groundstates.
[10]{}
F. Gerbier [*et al.*]{}, Phys. Rev. A [**73**]{}, 041602 (2006).
C. Cohen-Tannoudji and J. Dupont-Roc, Phys. Rev. A [**5**]{}, 968 (1972).
N. R. Cooper and R. Moessner, Phys. Rev. Lett. [**109**]{}, 215302 (2012).
F. D. M. Haldane, Phys. Rev. Lett. [**55**]{}, 2095 (1985).
N. R. Cooper, Advances in Physics [**57**]{}, 539 (2008).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The Daya Bay Reactor Neutrino Experiment discovered an unexpectedly large neutrino oscillation related to the mixing angle $\theta_{13}$ in 2012. This finding paved the way to the next generation of neutrino oscillation experiments. In this article, we review the history, featured design, and scientific results of Daya Bay. Prospects of the experiment are also described.'
address:
- 'Institute of High Energy Physics, Beijing, China'
- 'Department of Physics, University of California and Lawrence Berkeley National Laboratory, Berkeley, California, USA'
author:
- Jun Cao
- 'Kam-Biu Luk'
title: An overview of the Daya Bay Reactor Neutrino Experiment
---
neutrino oscillation ,neutrino mixing ,reactor ,Daya Bay 14.60.Pq ,29.40.Mc ,28.50.Hw ,13.15.+g
Introduction
============
Neutrino oscillation was firmly established by 2002. Around that time, atmospheric and accelerator neutrino experiments, e.g. Super-K [@superk] and K2K [@k2k], have determined the oscillation parameters $\theta_{23}$ and $|\Delta m^2_{32}|$ whereas solar and reactor neutrino experiments, such as SNO [@sno] and KamLAND [@kamland], measured $\theta_{12}$ and $\Delta m^2_{21}$. However, the mixing angle $\theta_{13}$, the CP violating phase $\delta_{\rm CP}$, and the sign of $\Delta m^2_{32}$ (aka the mass hierarchy) were unknown. In addition, $\theta_{13}$, unlike the other two mixing angles, was expected to be small [@chooz; @paloverde].
Among the three unknown quantities, $\theta_{13}$ plays a critical role in defining the future experimental program on neutrino oscillation. It is known that the CP-violating effect is proportional to $$J = \sin 2\theta_{12} \sin 2\theta_{23} \sin 2\theta_{13} \cos\theta_{13} \sin\delta_{\rm CP} \approx 0.9 \sin2\theta_{13} \sin\delta_{\rm CP} \,.$$ Resolution of the mass hierarchy problem also relies on the size of $\theta_{13}$. If it is too small, current technologies may not be able to determine $\delta_{\rm CP}$ and the mass hierarchy.
The mixing angle $\theta_{13}$ can be measured by accelerator-based or reactor-based experiments. However, the appearance probability of $\nu_{\mu} \to \nu_e$ in an accelerator neutrino experiment also depends on the yet unknown $\delta_{\rm CP}$ and the mass hierarchy. Hence, this type of experiments can only provide evidence for a non-zero $\theta_{13}$ but cannot measure its value unambiguously. On the other hand, reactor-based experiments can unambiguously determine $\theta_{13}$ via measuring the survival probability of the electron antineutrino $\overline{\nu}_{e}$ at short distance (${\cal O}(km)$) from the reactors. In the three-neutrino framework, the survival probability is given by $$\label{eq:psurv}
P = 1 - \cos^4\theta_{13}\sin^2 2\theta_{12}\sin^2\Delta_{21} - \sin^2 2\theta_{13}\sin^2
\Delta_{ee}\,,$$ where $\sin^2\Delta_{ee} \equiv \cos^2\theta_{12}\sin^2
\Delta_{31}+\sin^2\theta_{12}\sin^2{\Delta_{32}}\,$ and $\Delta_{ji}\equiv 1.267 {\Delta}m^2_{ji} L/E$. ${\Delta}m^2_{ji}$ is the mass-squared difference in eV$^2$, $E$ is the energy of the $\overline{\nu}_{e}$ in MeV, and $L$ is the distance in meters from the production point.
Pinning down $\theta_{13}$ by performing a relative measurement with a set of near and far detectors was suggested at the beginning of this millennium [@mikaelyan]. This method allows cancellation of most systematic uncertainties due to the reactor and the detector that previous experiments suffered. Since 2002, eight reactor experiments were proposed [@whitepaper]; three of them, Daya Bay [@dybcdr], Double Chooz [@doublechooz], and RENO [@reno], were finally constructed.
Among the eight proposals, the Daya Bay experiment is most sensitive for measuring $\theta_{13}$. The nuclear-power complex is among the top five most powerful in the world, providing a very intense flux of antineutrinos. In addition, it is very close to a mountain range in which an array of horizontal tunnels can be built, providing sufficient overburden to attenuate cosmic rays and space to accommodate a relatively large-scale experiment.
The Daya Bay nuclear-power complex is located on the southern coast of China, 55 km to the northeast of Hong Kong and 45 km to the east of Shenzhen. As shown in Fig. 1, the nuclear complex consists of six reactors grouped into three pairs with each pair referred to as a nuclear power plant (NPP). All six cores are functionally identical pressurized water reactors, each with a maximum of 2.9 GW of thermal power. The last core started commercial operation on 7 August 2011, a week before the start-up of the Daya Bay experiment. The distance between the cores for each pair is 88 m. The Daya Bay cores are separated from the Ling Ao cores by about 1100 m, while the Ling Ao-II cores are around 500 m away from the Ling Ao cores.
![Layout of the Daya Bay experiment. The dots represent the reactor cores, labelled as D1, D2, L1 to L4. Eight antineutrino detectors, labelled as AD1 to AD8, are installed in three underground experimental halls (EH1-EH3). The bottom sub-panel shows the survival probability as a function of the effective baseline $L$. The near and far detectors locate in the shaded area. \[fig:layout\]](DYB_8ADs.pdf){width="60.00000%"}
The Daya Bay experiment consists of three underground experimental halls (EHs) connected with horizontal tunnels. The overburden for the Daya Bay near hall (EH1), the Ling Ao near hall (EH2) and the far hall (EH3) are 250, 265, and 860 equivalent meters of water, respectively. Eight antineutrino detectors (ADs) are installed in the three halls, with two in EH1, two in EH2, and four in EH3. Each AD has 20-ton target mass to catch the reactor antineutrinos. The sensitivity to $\sin^22\theta_{13}$ was designed to be better than 0.01 at 90% confidence level in 3 years.
History of the Daya Bay Experiment {#history}
==================================
The idea of determining $\theta_{13}$ using the Daya Bay reactor complex was proposed in 2003. The first dedicated workshop for the Daya Bay experiment was held in the University of Hong Kong in November 2003 [@hk2003]. It was immediately followed by the second one in January 2004 at the Institute of High Energy Physics [@ihep2004], at which a preliminary experimental design was presented, including the unique multiple-detector scheme and the reflective panel design for light collection. In response to the recommendation of measuring $\sin^22\theta_{13}$ to the level of 0.01 by the APS Neutrino Study [@numatrix] and NuSAG [@nusag], the target mass of the detectors was enlarged from 8 ton to 20 ton. The Conceptual Design Report (CDR) was released at the end of 2006 [@dybcdr].
The Daya Bay Collaboration with international participation was formed in February 2006. The project was approved by the Chinese Academy of Sciences in May 2006 and by the Ministry of Science and Technology of China in January 2007. It passed the CD-2 review as required by the US Department of Energy in January 2008.
Ground breaking of the experiment took place in October 2007. Detectors were assembled onsite in parallel to the civil construction. After 4 years of construction, the first of the three underground experimental halls, EH1, started data taking on August 11, 2011. Data was used to study the detector performance, resulting a paper submitted on 28 February 2012 [@ad12]. This publication reported that the relative detection uncertainty of two ADs was only 0.2%, much better than the designed value of 0.38% documented in the CDR.
Since the detector fabrication was out of sync with the civil construction, the collaboration decided to operate the experiment with two phases to maximize the scientific reach. The first phase was run with 6 ADs out of a total of 8, with two in EH1, one in EH2, and three in EH3. The second near hall, EH2, was ready on 5 November 2011, and EH3 started data taking on 24 December 2011. On 8 March 2012, the Daya Bay collaboration announced the discovery of a new disappearance of reactor antineutrinos at 5.2 standard deviations ($\,\sigma$) and measured $\sin^22\theta_{13}=0.092\pm0.016{\rm (stat)}\pm0.005{\rm (syst)}$ with 55 days data [@DB_discovery]. It was further verified at 7.7$\,\sigma$ with 139 days of data [@DB_CPC] and at 4.6$\,\sigma$ with a statistically independent data set using antineutrino events tagged by neutron capture on hydrogen [@DB_nH].
The six-detector phase terminated on 28 July 2012. The last two ADs were installed. The full configuration of the Daya Bay experiment started data taking on 19 October 2012, running reliably to present. Two additional results on neutrino oscillation were reported subsequently. With all 217 days of data acquired in the first phase, a spectral and rate analysis improved the precision of $\theta_{13}$ and measured the effective mass splitting $\Delta m^2_{ee}$ for the first time [@DB_spectana]. A new analysis with 621 days of data, including the 6-AD phase and the full 8-AD configuration, was released recently; the measured $\sin^22\theta_{13}$ has reached a precision of 6% [@DB_2015]. These results are summarized in Table \[tab:sens\].
Release time Data Config $\sin^22\theta_{13}$ $\Delta m^2_{ee} $
--------------------------- ---------- ------------ --------------------------- ------------------------
2012/3/8 [@DB_discovery] 55 days 6 ADs $0.092\pm0.016\pm0.005$ -
2012/10/23 [@DB_CPC] 139 days 6 ADs $0.089\pm0.010\pm0.005$ -
2013/10/24 [@DB_spectana] 217 days 6 ADs $0.090^{+0.008}_{-0.009}$ $2.59^{+0.19}_{-0.20}$
2014/6/24 [@DB_nH] 217 days 6 ADs (nH) $0.083\pm0.018$ -
2015/5/13 [@DB_2015] 621 days 6+8 ADs $0.084\pm0.005$ $2.42\pm0.11$
: Daya Bay measurements on $\sin^22\theta_{13}$ and $\Delta m^2_{ee}$. The first uncertainty of $\sin^22\theta_{13}$ in the first two rows is the statistical error and the second is the systematic one. The unit of $\Delta m^2_{ee}$ is $10^{-3}$ eV$^2$. \[tab:sens\]
Daya Bay has accumulated the largest reactor antineutrino sample in the world, which enables many precision measurements. The most precise reactor antineutrino spectrum has been measured [@DB_reactor]. A search for a sterile neutrino has significantly extended the exclusion area in the low-mass region of the $\sin^22\theta_{14}$-$\Delta m^2_{41}$ parameter space [@DB_sterile]. Many exotic searches are ongoing. The Daya Bay experiment plans to operate until 2020. A 3%-precision measurement on both $\sin^22\theta_{13}$ and $\Delta m^2_{ee}$ is expected.
Design and Features {#design}
===================
In a reactor neutrino experiment, the sensitivity or precision in $\theta_{13}$ depends on how well the rate deficit and distortion in the energy spectrum are determined. When the exposure, defined as the product of the target mass of the far site detector in tons, the thermal power of the reactor in GW, and the live time in years, is larger than 10,000 GW-ton-yr, distortion in the energy spectrum, thus statistics, will dominate the sensitivity [@huber2002]. Such exposure corresponds to about 8 years for Daya Bay. Therefore, for most cases, the rate deficit will dominate the sensitivity, and related systematic uncertainties, including the detection efficiency, target proton number, and backgrounds, should be controlled to $\lesssim 0.5$% to measure $\sin^22\theta_{13}$ to 0.01 at 90% confidence level.
When multiple reactor cores are spread out over a large area, a single near site can only constrain the antineutrino flux from the nearby cores. In this case, the reactor-related uncertainties cannot be completely cancelled by the near-far relative measurement. Moving the near site farther away from the cores will improve the cancellation but lose sensitivity due to an increase in the oscillation effect. To obtain the best sensitivity, Daya Bay is configured with one far site for observing the maximal oscillation effect, and two near sites for determining the flux of the reactor antineutrinos from the Daya Bay and Ling Ao NPPs. The best locations of the three halls were determined with a $\chi^2$ method [@sunyx], with the projected uncertainties and estimated background at the candidate sites derived from geological survey information. For the optimal configuration, the uncertainty related to the reactors is reduced to 5% of the uncorrelated uncertainty of a single core (0.8%), which is totally insignificant.
Experience and lessons learned in CHOOZ [@chooz], Palo Verde [@paloverde], and KamLAND [@kamland] were taken into account in designing the Daya Bay detectors. Some unique features of the Daya Bay design significantly improve the detector performance; indeed, the built-in redundancy is crucial for precision measurements. Details of the Daya Bay detector design and fabrication can be found in Ref. [@longdet]. In the following we will briefly describe the experimental design highlighting the concept of multiple muon taggers, multiple antineutrino detectors, a remote-controlled calibration system, photon reflectors and shields, as well as the optimization of the detector dimensions.
Each AD has three nested cylindrical volumes separated by concentric acrylic vessels as shown in Fig. \[fig:det\]. Serving as the target for the inverse beta-decay reaction, the innermost volume holds 20 t of gadolinium-doped liquid scintillator (Gd-LS) with 0.1% Gd by weight [@ding; @ding2; @yeh; @lsproduction]. The middle volume is called the gamma catcher and is filled with 20 t of undoped liquid scintillator (LS) for detecting gamma-rays that escape the target volume. The outer volume contains 37 t of mineral oil (MO) to provide optical homogeneity and to shield the inner volumes against radiation from the detector components. There are 192 20-cm PMTs (Hamamatsu R5912) installed in the MO volume and around the circumference of the stainless steel vessel (SSV). The top and the bottom surfaces are not instrumented with PMTs; instead, there are two highly reflective panels. The PMTs are recessed in a 3-mm-thick black cylindrical shield located at the equator of the PMT bulb. In each hall, the ADs are submerged in a water pool that provides at least 2.5 m of water to degrade radiation from the rock. The water pool is optically divided into the inner (IWS) and outer (OWS) regions, both equipped with 20-cm PMTs to serve as water Cherenkov detectors. On the top of the water pool, there are Resistive Plate Chambers (RPCs) serving as another muon detector.
![Schematic diagram of the Daya Bay detectors. \[fig:det\]](AD+muon-PRL_pub.pdf){width="60.00000%"}
The multiple-module scheme is the most prominent feature of Daya Bay. The current generation of reactor neutrino experiments for determining $\theta_{13}$ planned to achieve a relative detector uncertainty of (0.38-0.6)%. Such uncertainties need very careful validation. With at least two ADs in each experimental hall, comparison of performance among the ADs at the same site can actually “measure" the relative uncertainty between them. By comparing the components of the detector responses, the relative uncertainty of two ADs was found to be 0.2% [@ad12]. The measured ratio of the antineutrino rates in two ADs was $0.981\pm0.004~(1.019\pm0.004)$ while the expected ratio was 0.982 (1.012) for the Daya Bay (Ling Ao) near site [@DB_2015], validating the uncertainty estimation. The deviation from unity is due to slightly different baselines of the two ADs to the reactor cores. Furthermore, the uncertainties of the ADs are found to be largely uncorrelated. Therefore, the total relative detector uncertainty is statistically reduced by $1/\sqrt{N}$, where $N$ is the number of ADs at a given site.
The water pool is divided into two water Cherenkov detectors, as shown in Fig. \[fig:det\]. The outer one is 1 m thick and the inner one is 1.5 m. The top of the water pool is covered by RPC tiles. The outer Cherenkov counter and the RPC play important roles in determining the fast neutron background originating from muon spallation in the surrounding rock. The efficiency of the water Cherenkov detectors for detecting muons was designed to be 95%, and 90% for the RPCs. The combined efficiency was aimed to be ($99.5\pm0.25$)%. As it turned out, due to high reflectivity of the custom-made Tyvek^^ composite film lining the partitions and the well-designed water purification system, the efficiency of the inner Cherenkov detector was 99.98% [@DB_muon]. Again, the multiple-detector design provides robust tagging of the cosmic-ray muons, which is essential for rejecting muon-induced background.
The geometry of the AD was optimized with extensive Monte Carlo simulation. The MO shielding is thin and the distance from the apex of the PMT to the liquid scintillator is only 20 cm. This distance is driven by the requirement of uniform detector response instead of radiation shielding. As a result, Daya Bay has a relatively high singles rate and accidental coincidence to maximize the target mass. Since the accidental-coincidence background can be determined accurately to high precision, it has negligible impact to the sensitivity or precision.
To obtain good light yield with fewer number of PMTs, reflective panels are used. A specular reflective film with reflectivity $>98\%$ in the scintillation light region, ESR (3M^^), is sandwiched between two 1-cm-thick acrylic panels 4.5 m in diameter. The space between the two acrylic panels is evacuated when bonding together on the edge. The sandwich structure is maintained by vacuum pressure with the least mechanical connections between two panels to keep the optical surface intact. A reflector is put on the top of the gamma catcher and another at the bottom. Such a design reduces the number of PMTs by about one half while achieving almost the same energy and vertex resolution. Furthermore, the mechanical structure of the AD is simplified with the adoption of reflective panels, enabling transportable ADs.
Three water-proof automated calibration units (ACUs) are mounted on the top of each AD. Each ACU is equipped with a LED, a $^{68}$Ge source, and a composite source of $^{241}$Am-$^{13}$C and $^{60}$Co. Deployment of the source into the liquid scintillator, one at a time, is controlled remotely. With the ACUs, the AD can be fully submerged in water without a chimney for calibration. Therefore, Daya Bay does not experience backgrounds coming from external radioactivity or the Michel electrons from decays of stopped muon.
The inner wall of the stainless steel tank is painted black with a fluor-carbon paint. To reduce the stray light reflected from the PMT glass, cable, and other components, another light shield made of black ABS is installed at the equator of the PMTs. This design has an unforeseen advantage of suppressing an instrumental background, the PMT flasher events, which appeared in many neutrino experiments using PMTs but become difficult to reject for relative small detectors. The electronic components or connections on the base of the Hamamatsu PMT may occasionally discharge and produce a flash of light. The detected energy of these faked events ranges from sub-MeV to a hundred MeV in Daya Bay. Although only a small fraction of the PMTs spontaneously discharge infrequently, these flasher events significantly increase the accidental-coincidence background. With the black shield, the flasher events always appear as a characteristic PMT-hit pattern; thus they can be easily identified and rejected [@DB_CPC]. Without this unique PMT-hit pattern, we would have to either turn off the flashing PMTs, or bear a larger uncertainty in the selection efficiency and a larger accidental-coincidence background.
Signal and Background {#signal}
=====================
The reactor antineutrinos are detected via the inverse $\beta$-decay (IBD) reaction, $\overline{\nu}_e + p \to e^+ + n$, in the Gd-LS. The coincidence of the prompt scintillation from the $e^+$ and the delayed neutron capture on Gd provides a distinctive signature. The positron carries almost all of the kinetic energy of the antineutrino, thus the positron energy deposited in the liquid scintillator is highly correlated with the antineutrino energy. The neutron thermalizes before being captured on either a proton or a gadolinium nucleus with a mean capture time of $\sim$30 $\mu$s in Gd-LS or $\sim$200 $\mu$s in normal LS. When a neutron is captured on Gd, it releases several gamma-rays with a total energy of $\sim$8 MeV, and is thus easily distinguished from the background coming from natural radioactivity. The capture on H suffers from a larger background but provides an independent measurement and can improve the precision of the $\theta_{13}$ measurement.
The ADs are calibrated with sources in the ACUs weekly, and with spallation products, IBDs, and natural radiation from materials inside the detectors. Two independent calibration algorithms are utilized. The energy scale is determined using the $^{60}$Co or Am-C neutron source in the ACUs, or spallation neutrons. The uncertainty in the energy scale is determined by comparing more than 10 known references in the 8 ADs and by studying their stabilities over time. The energy scale uncertainty has reduced from 0.5% reported in the initial publications [@ad12; @DB_discovery] to 0.2% in the latest [@DB_2015]. The reduction comes from the improvements in the correction of position and time dependence.
Nonlinearity in the energy response of an AD originates from two dominant sources: particle-dependent nonlinear light yield of the scintillator and charge-dependent nonlinearity in the PMT readout electronics. Each effect is at the level of 10%. We have constructed a semi-empirical model that predicts the reconstructed energy for a particle assuming a specific energy deposited in the scintillator. The model contains four parameters: Birks’ constant, the relative contribution to the total light yield from Cherenkov radiation, and the amplitude and scale of an exponential correction describing the non-linear electronics response. This exponential form of the electronics response is motivated by Monte Carlo (MC) and data; it has been confirmed with an independent FADC measurement. Besides the calibration references used in the energy scale studies, a broad set of calibration sources were deployed into the two ADs of EH1 using the ACUs and a manual calibration system [@huanghx] during the shutdown in the summer of 2012. The energy nonlinearity, i.e. the absolute energy scale, is determined to $<1$% above 2 MeV.
To select reactor antineutrino events, the PMT flasher background is rejected first. The prompt and delayed signals are required to be 0.7-12 MeV and 6-12 MeV, respectively. The temporal separation between a pair of prompt and delayed signals should be within 1-200 $\mu$s. To reject cosmogenic backgrounds, the delayed signal is required to be 600 $\mu$s, 1000 $\mu$s, or 1 s later than a muon, depending on the deposited energy of the muon. Finally, no other signal should occur in 200 $\mu$s before the prompt signal and after the delayed signal. Two independent algorithms are developed following these criteria with minor differences. The selected samples differ by less than 10%, mostly due to the different energy calibration used.
A detailed treatment of the absolute and relative efficiencies was reported in Refs. [@ad12; @DB_CPC]. The uncertainties of the absolute efficiencies are correlated among the ADs and are thus negligible in oscillation analyses. The determination of all relative uncertainties is data-driven. The dominant ones come from the energy calibration and the neutron capture fraction on Gd, both at the 0.1% level. The total relative uncertainty is conservatively estimated to be 0.2%, uncorrelated among ADs.
Five kinds of background are considered in Daya Bay. The accidental background is the largest one but contributes negligible to the total systematic uncertainty. The most serious background is the cosmogenic $^9$Li/$^8$He, which contributes an uncertainty of $\sim0.2$%. The remaining three kinds of background have an uncertainty of $\sim 0.01$%.
The accidental background, from accidental correlation of two unrelated signals, is determined by measuring the rate of both prompt- and delayed-like signals, and then estimating the probability that two signals randomly satisfy the time coincidence for the IBD selection.
The $^9$Li/$^8$He background comes from the $\beta$-n decay of $^9$Li/$^8$He produced by muons in the ADs. The rate is evaluated from the distribution of the time since the last muon using the known decay times for these isotopes. A 50% systematic uncertainty is assigned to account for the extrapolation to zero deposited muon energy.
An energetic neutron entering an AD can form a fast-neutron background by recoiling off a proton before being captured. It is estimated by extrapolating the prompt energy spectrum into the IBD energy region, or by studying the muon-tagged fast neutron sample.
The $^{13}$C($\alpha$,n)$^{16}$O background was determined using MC after estimating the amount of $^{238}$U, $^{232}$Th, $^{227}$Ac, and $^{210}$Po in the from their cascade decays, or by fitting their $\alpha$-particle energy peaks in the data.
A neutron emitted from the neutron source in an ACU could generate a gamma-ray via inelastic scattering in the stainless steel vessel before subsequently being captured on Fe/Cr/Mn/Ni. An IBD is mimicked if both gamma-rays from the scattering and capture processes enter the scintillator. This correlated background is estimated using MC. The normalization is constrained by the measured rate of single delayed-like candidates from this source.
Oscillation analyses {#oscillation}
====================
Early results of Daya Bay were based on rate analysis when the statistics were low. The rate deficit at the far site was $\sim6$% compared to the prediction based on a weighted combination of two near site measurements. The value of $\sin^22\theta_{13}$ is determined with a $\chi^2$ constructed with pull terms accounting for the correlation of the systematic errors, $$\begin{aligned}
\label{eqn:chi2}
\chi^2 &=&
\sum_{d=1}^{6}
\frac{\left[M_d-T_d\left(1+ \varepsilon
+ \sum_r\omega_r^d\alpha_r
+ \varepsilon_d\right) +\eta_d\right]^2}
{M_d+B_d} \nonumber \\
&+&
\sum_r\frac{\alpha_r^2}{\sigma_r^2}
+ \sum_{d=1}^{6} \left(
\frac{\varepsilon_d^2}{\sigma_d^2}
+ \frac{\eta_d^2}{\sigma_{B}^2}
\right)
\,,\end{aligned}$$ where $M_d$ are the measured number of IBD events of the $d$-th AD with its backgrounds subtracted, $B_d$ is the corresponding background, $T_d$ is the prediction from antineutrino flux, including MC corrections for energy responses and neutrino oscillations, $\omega_r^d$ is the fraction of IBD contribution of the $r$-th reactor to the $d$-th AD determined by the baselines and antineutrino fluxes. The uncorrelated reactor uncertainty is $\sigma_r$ (0.8%). The parameter $\sigma_d$ (0.2%) is the uncorrelated detection uncertainty. The parameter $\sigma_{B}$ ($\sim0.2$%) is the quadratic sum of the background uncertainties, which are site dependent. The corresponding pull parameters are ($\alpha_r, \varepsilon_d, \eta_d$). The absolute normalization $\varepsilon$, which absorbs the detector- and reactor-related correlated uncertainties, is a free parameter determined from the fit to the data. While keeping $\varepsilon$ free, the reactor antineutrino flux is determined by the near site measurements. The model-dependent reactor flux prediction enters the fit only at secondary order.
With increased statistics, the latest Daya Bay analyses are based on rate and shape analysis. While rate deficit still dominates the $\theta_{13}$ sensitivity, the spectral information starts to contribute. To take advantage of the spectral information, an analysis with a similar $\chi^2$ expression as defined in Eq. \[eqn:chi2\] but with energy bins and relevant uncertainties is used. Additional inputs to the fit include the background shape uncertainties and energy nonlinearity model described in Sec. \[signal\]. Besides improving the precision of the $\sin^22\theta_{13}$, the effective mass splitting $\Delta m^2_{ee}$ has been measured for the first time [@DB_spectana].
Another approach in extracting the oscillation parameters is to construct a $\chi^2$ expression using a covariance matrix $$\label{eq:chi22}
\chi^{2} = \sum_{i,j}(N_j^{\mathrm{f}} - w_j \cdot N_j^{\mathrm{n}}) (V^{-1})_{ij} (N_i^{\mathrm{f}} - w_i \cdot N_i^{\mathrm{n}}),$$ where $N_i$ is the observed number of events after background subtraction in the $i$-th bin of reconstructed positron energy. The superscript $f~(n)$ denotes a far (near) detector. The symbol $V$ represents a covariance matrix that includes known systematic and statistical uncertainties. The quantity $w_i$ is a weight that accounts for the differences between the near- and far-site measurements. In this method, the flux and spectrum of the antineutrinos at the reactor play a negligible role [@DB_2015]. The results from the fit shown in the $\sin^22\theta_{13}$-$\Delta m^2_{ee}$ plane are depicted in Fig. \[fig:contours\].
![Error contours corresponding to the $68.3\%$, $95.5\%$ and $99.7\%$ confidence levels in the $|\Delta m^{2}_{\mathrm{ee}}|$-$\sin^{2}2\theta_{13}$ plane. The contours were obtained with a least-squares fit given in Eq. \[eq:chi22\] using the measured $\overline{\nu}_e$ rates and energy spectra at the near and far sites. The adjoining panels show the dependence of $\Delta\chi^{2}$ on $\sin^{2}2\theta_{13}$ (top) and $|\Delta m^2_{\mathrm{ee}}|$ (right). Figure adapted from [@DB_2015]. \[fig:contours\]](farNearAllowedOscillation_621days_wAccel_v5.pdf){width="60.00000%"}
Daya Bay has also accumulated an IBD sample with neutron capture on hydrogen (nH) which has comparable statistics to that of neutron capture on gadolinium (nGd). Since the delayed signal of 2.2 MeV falls into the energy region totally dominated by natural radiation, the coincidence background is huge. By requiring the reconstructed distance between the positron and delayed-neutron vertices be $<50$ cm, 98% of this background can be rejected while losing 25% of the signal. A spectral subtraction is further needed to remove the accidental backgrounds. An analysis using the 217-day data set yielded $\sin^22\theta_{13}=0.080\pm0.018$. Since the IBD sample and the systematic uncertainties are largely independent from the nGd analysis, the nH analysis provides an independent measurement of $\theta_{13}$. The correlation between the nH and nGd analyses is evaluated to be 0.05. When the nH and nGd analyses of this data set are combined, the $\sin^22\theta_{13}$ sensitivity is improved by 8% [@DB_nH].
Reactor Antineutrino Spectrum and Exotic searches {#reactor}
=================================================
Although the oscillation analyses of Daya Bay have negligible dependence on the external prediction of the reactor antineutrino flux and spectrum, the knowledge is a crucial factor for many reactor experiments. In early studies, the estimation relied on calculations or other indirect means, such as the $\beta$ spectrum measurements made on reactor fuels, based on the understanding of the complex fission processes in the reactor core. These methods have rather strong dependence on theoretical models. Daya Bay has accumulated the world’s largest sample of reactor antineutrinos, at a rate of $\sim 1$ million per year. A direct measurement will provide the most precise and model independent reactor antineutrino spectrum.
In reactor cores, antineutrinos are emitted from four primary fuel isotopes: $^{235}$U, $^{238}$U, $^{239}$Pu, and $^{241}$Pu. Each fission releases about 200 MeV energy ((0.2-0.5)% uncertainty). Fission rates can be estimated with the thermal power measurement (0.5% uncertainty) and core simulation of the evolution of the fuel composition (0.6% uncertainty with the constraint of the total thermal power) [@cjflux]. The most uncertain part is the rate and spectrum of antineutrinos emitted from each fission of the four isotopes. By fitting the measured $\beta$ spectrum of the $^{235}$U, $^{239}$Pu, and $^{241}$Pu fuel [@illschr; @illvonf; @illhahn] with hypothesized virtual $\beta$ decays, plus theoretical calculation for $^{238}$U, two models, the ILL-Vogel and the Huber-Mueller models, have been developed [@vogel238; @mueller; @huber] with an uncertainty of (2-3)%.
The latest Daya Bay measurement on the absolute reactor antineutrino rate is $\sigma_f=(5.92\pm0.14)\times 10^{-43} {\rm cm}^2\, {\rm fission}^{-1}$, where the dominant uncertainty comes from the absolute efficiency (2.1%) [@DB_reactor]. Comparing to the Huber-Mueller model, there is a $\sim6$% deficit, consistent with the past 19 short baseline ($<100$ m) measurements.
With the nonlinearity model described in Sec. \[signal\], Daya Bay has measured the prompt-energy spectrum with a precision ranging from 1.0% at 3.5 MeV to 6.7% at 7 MeV. An excess of about 10% at $\sim 5$ MeV compared with expectations is observed, leading to a discrepancy of up to $4~\sigma$.
Such a deviation shows the importance of the direct measurement of the reactor antineutrino spectrum, particularly for next-generation reactor experiments such as JUNO [@JUNO], and may indicate the need to revisit the models underlying the calculations. The prompt-energy spectrum is unfolded, i.e. removing the detector response effect, to an antineutrino spectrum, as shown in Fig. \[fig:generic\].
Daya Bay will improve the measurement with much larger statistics and better energy nonlinearity. Recently a Flash Analog-to-Digital Converter (FADC) system has been installed on one AD to pin down the nonlinearity from the electronics. Another calibration and systematic study campaign is being planned. We expect to measure the reactor antineutrino spectrum to 1% precision in a large energy range, and improve the precision especially for the very high- and very low-energy regions which are outside the ranges of current models.
![Top panel: The extracted reactor antineutrino spectrum and its correlation matrix. Bottom panel: Ratio of the extracted reactor antineutrino spectrum to the Huber+Mueller prediction. Figure adapted from [@DB_reactor]. \[fig:generic\]](generic_pre){width="60.00000%"}
The high precision antineutrino spectrum is also excellent for light sterile neutrino searches. If light sterile neutrinos mix with the three active neutrinos, their presence could be detected by looking for the fast oscillatory behavior in the spectrum. Daya Bay has significantly extended the exclusion area in $10^{-3} {\rm eV}^2 \lesssim |\Delta m^2_{41}|\lesssim 0.1 {\rm eV}^2$ [@DB_sterile]. Further improvements with increasing statistics are expected.
Besides the sterile neutrino studies, more exotic searches are in progress, e.g. for the non-standard interaction, decoherence effect, mass-varying neutrino, Lorentz-violation and CPT violation, etc.
Summary and Prospect {#summary}
====================
With an almost ideal experimental site and unique design, the Daya Bay experiment has excellent capability for high precision measurements of reactor antineutrinos. We have reviewed our design experience, which may help future reactor neutrino experiments. The measurements on $\theta_{13}$ and effective mass splitting are reviewed. Current precision on $\sin^22\theta_{13}$ and $|\Delta m^2_{\mathrm{ee}}|$ are 6% and 4.5%, respectively. The projected precisions are shown in Fig. \[fig:future\]. The Daya Bay experiment is expected to operate until 2020; by then, the precision is $\sim3$% for both $\sin^22\theta_{13}$ and $|\Delta m^2_{\mathrm{ee}}|$. Daya Bay has also obtained the most precise reactor antineutrino spectrum, which will be very valuable for designing the next-generation reactor neutrino experiments that depend on this input, such as JUNO.
![Projected precision of $\sin^22\theta_{13}$ (black thick lines) and $\Delta m^2_{ee}$ (red thin lines) in Daya Bay, where the solid lines present the precision estimated with current systematics and the dashed lines show the statistical limit with zero systematic uncertainty. The points on the curves show the precision of published Daya Bay results.\[fig:future\]](dyb_theta_delta.pdf){width="60.00000%"}
Acknowledgement {#acknowledgement .unnumbered}
===============
We would like to thank Jie Zhao for preparing the Fig. \[fig:future\]. J.C. is partially supported by the National Natural Science Foundation of China (11225525) and K.B.L. is partially supported by the U.S. Department of Energy, OHEP DE-AC02-05CH11231.
[00]{}
Y. Fukuda [*et al.*]{}, Phys. Rev. Lett. [**81**]{}, 1562 (1998).
M. H. Ahn [*et al.*]{}, Phys. Rev. Lett. [**90**]{}, 041801 (2003).
Q.R. Ahmad [*et al.*]{}, Phys. Rev. Lett. [**89**]{}, 011301 (2002).
K. Eguchi [*et al.*]{}, Phys. Rev. Lett. [**90**]{}, 021802 (2003).
M. Apollonio [*et al.*]{}, Eur. Phys. J. C27, 331 (2003)
F. Boehm [*et al.*]{}, Phys. Rev. D[**62**]{}, 072002 (2000).
L. A. Mikaelyan and V. V. Sinev, Phys. Atom. Nucl. [**63**]{}, 1002 (2000).
K. Anderson [*et al.*]{}, [*White paper report on using nuclear reactors to search for a value of $\theta_{13}$*]{}, arXiv:hep-ex/0402041.
X.H. Guo [*et al.*]{} (Daya Bay Collaboration), arXiv:hep-ex/0701029.
F. Ardellier [*et al.*]{} (Double Chooz Collaboration), arXiv:hep-ex/0606025.
S. B. Kim (RENO Collaboration), AIP Conf. Proc. [**981**]{}, 205 (2008).
http://dayabay.ihep.ac.cn/hk2003/
Workshop webpage: http://dayabay.ihep.ac.cn/ihep2004/
, http://www.aps.org/policy/reports/multidivisional/neutrino/index.cfm
http://www.science.doe.gov/hep/NuSAG2ndRptFeb2006.pdf
F. P. An [*et al.*]{} (Daya Bay Collaboration), Nucl. Instr. and Meth. A [**685**]{}, 78 (2012).
F. P. An [*et al.*]{} (Daya Bay Collaboration), Phys. Rev. Lett. [**108**]{}, 171803 (2012).
F. P. An [*et al.*]{} (Daya Bay Collaboration), Chin. Phys. C[**37**]{}, 011001 (2013).
F. P. An [*et al.*]{} (Daya Bay Collaboration), Phys. Rev. D [**90**]{}, 071101 (2014).
F. P. An [*et al.*]{} (Daya Bay Collaboration), Phys. Rev. Lett. [**112**]{}, 061801 (2014).
F. P. An [*et al.*]{} (Daya Bay Collaboration), Phys. Rev. Lett. [**115**]{}, 111802 (2015).
F. P. An [*et al.*]{} (Daya Bay Collaboration), Phys. Rev. Lett. [**116**]{}, 061801 (2016).
F. P. An [*et al.*]{} (Daya Bay Collaboration), Phys. Rev. Lett. [**113**]{}, 141802 (2014).
P. Huber, M. Lindner, T. Schwetz, and W. Winter, Nulc. Phys. [**B665**]{}, 487 (2003).
Y. X. Sun [*et al.*]{}, HEPNP [**29**]{}, 543 (2005).
F. P. An [*et al.*]{} (Daya Bay Collaboration), Nucl. Instrum. Meth. A [**811**]{}, 133 (2016).
Y. Y. Ding [*et al.*]{}, Journal of Rare Earths [**25**]{}, 310 (2007).
Y. Y. Ding [*et al.*]{}, Nucl. Instr. and Meth. A [**584**]{}, 238 (2008).
M. Yeh [*et al.*]{}, Nucl. Instr. and Meth. A [**578**]{}, 329 (2007).
W. Beriguete [*et al.*]{}, Nucl. Instr. and Meth. A [**763**]{}, 82 (2014).
H. X. Huang [*et. al.*]{}, JINST [**8**]{}, P09013 (2013).
F. P. An [*et al.*]{} (Daya Bay Collaboration), Nucl. Instr. and Meth. A [**773**]{}, 8 (2015).
J. Cao, Nucl. Phys. B Proc. Supp. [**229-232**]{}, 205 (2012).
W. G. K. Schreckenbach, G. Colvin and F. von Feilitzsch, Phys. Lett. B [**160**]{}, 325 (1985).
A. F. von Feilitzsch and K. Schreckenbach, Phys. Lett. B [**118**]{}, 162 (1982).
A. A. Hahn [*et al.*]{}, Phys. Lett. B [**218**]{}, 365 (1989).
P. Vogel, G. K. Schenter, F. M. Mann, and R. E. Schenter, Phys. Rev. C[**24**]{}, 1543 (1981).
T. Mueller [*et al.*]{}, Phys. Rev. C[**83**]{}, 054615 (2011).
P. Huber, Phys. Rev. C[**84**]{}, 024617 (2011) \[Erratum-ibid. [**85**]{}, 029901(E) (2012)\].
F. P. An [*et al.*]{}, J. Phys. G: Nucl. Part. Phys. [**43**]{}, 030401 (2016).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Visual commonsense reasoning task aims at leading the research field into solving cognition-level reasoning with the ability of predicting correct answers and meanwhile providing convincing reasoning paths, resulting in three sub-tasks i.e., Q$\rightarrow$A, QA$\rightarrow$R and Q$\rightarrow$AR. It poses great challenges over the proper semantic alignment between vision and linguistic domains and knowledge reasoning to generate persuasive reasoning paths. Existing works either resort to a powerful end-to-end network that cannot produce interpretable reasoning paths or solely explore intra-relationship of visual objects (homogeneous graph) while ignoring the cross-domain semantic alignment among visual concepts and linguistic words. In this paper, we propose a new Heterogeneous Graph Learning (HGL) framework for seamlessly integrating the intra-graph and inter-graph reasoning in order to bridge vision and language domain. Our HGL consists of a primal vision-to-answer heterogeneous graph (VAHG) module and a dual question-to-answer heterogeneous graph (QAHG) module to interactively refine reasoning paths for semantic agreement. Moreover, our HGL integrates a contextual voting module to exploit long-range visual context for better global reasoning. Experiments on the large-scale Visual Commonsense Reasoning benchmark demonstrate the superior performance of our proposed modules on three tasks (improving 5% accuracy on Q$\rightarrow$A, 3.5% on QA$\rightarrow$R, 5.8% on Q$\rightarrow$AR)[^1].'
author:
- |
Weijiang Yu$^{1}$, Jingwen Zhou$^{1}$, Weihao Yu$^{1}$, Xiaodan Liang$^{2, }$[^2], Nong Xiao$^{1}$\
$^1$School of Data and Computer Science, Sun Yat-sen University\
$^2$School of Intelligent Systems Engineering, Sun Yat-sen University\
[[email protected], [email protected], [email protected],]{}\
[[email protected], [email protected]]{}
bibliography:
- 'egbib.bib'
title: Heterogeneous Graph Learning for Visual Commonsense Reasoning
---
Introduction
============
Visual and language tasks have attracted more and more researches, which contains visual question answering (VQA) [@lu2016hierarchical; @schwartz2017high; @shih2016look], visual dialogue [@guo2019image; @das2017learning], visual question generation (VQG) [@jain2017creativity; @mostafazadeh2016generating], visual grounding [@hu2017modeling; @deng2018visual; @xiao2017weakly] and visual-language navigation [@wang2018reinforced; @ke2019tactical]. These tasks can roughly be divided into two types. One type is to explore a powerful end-to-end network. Devlin et al. have introduced a powerful end-to-end network named BERT [@devlin2018bert] for learning more discriminative representation of languages. Anderson et al [@anderson2018bottom]. utilized the attention mechanisms [@showtell] and presented a bottom-up top-down end-to-end architecture. While these works resorted to a powerful end-to-end network that cannot produce interpretable reasoning paths. The other type is to explore intra-relationship of visual objects (homogeneous graph). Norcliffe-Brown et al. [@norcliffe2018learning] presented a spatial graph and a semantic graph to model object location and semantic relationships. Tang et al. [@tang2018learning] modeled the intra-relationship of visual objects by composing dynamic tree structures that place the visual objects into a visual context. However, all of them solely consider intra-relationship limited to homogeneous graphs, which is not enough for visual commonsense reasoning (VCR) due to its high demand of the proper semantic alignment between vision and linguistic domain. In this paper, we resolve the challenge via heterogeneous graph learning, which seamlessly integrates the intra-graph and inter-graph reasoning to bridge vision and language domain.
Current approaches mainly fall into homogeneous graph modeling with same domain (e.g. vision-to-vision graph) to mine the intra-relationship. However, one of the keys to the cognition-level problem is to excavate the inter-relationship of vision and linguistics (e.g. vision-to-answer graph) by aligning the semantic nodes between two different domains. As is shown in Figure \[fig:pic1\], we show the difference of homogeneous graphs and heterogeneous graphs. The homogeneous graph may lead to information isolated island (dotted portion in Figure \[fig:pic1\](a)(b)), such as vision-to-vision graph that is limited to object relationship and is not related to functionality-based information (e.g. get, pours into) from linguistics, which would hinder semantic inference. For instance, “person1 pours bottle2 into wineglass2”, this sentence contains functionality-based information “pours into” which is different from the semantic visual entities such as “person1”, “bottle2” and “wineglass2”. It may not be enough to bridge the vision and language domain via homogeneous graph as shown in Figure \[fig:pic1\](a)(b). However, we can connect the visual semantic node (e.g. “person1”) with functional node (e.g. “pours into”) from linguistics via a heterogeneous graph (Figure \[fig:pic1\](c)), which can support proper semantic alignment between vision and linguistic domain. Benefit from the merits of the heterogeneous graph, we can seamlessly connect the inter-relationship between vision and linguistics, which can refine the reasoning path for semantic agreement. Here, we propose to use the heterogeneous graph learning for VCR task to support the visual representation being aligned with linguistics.
A heterogeneous graph module including a primal vision-to-answer heterogeneous graph (VAHG) and a dual question-to-answer heterogeneous graph (QAHG) is the core of Heterogeneous Graph Learning (HGL), which contains two steps: (1) build a heterogeneous graph and evolve the graph; (2) utilize the evolved graph to guide the answer selection. First, given the generated node representation of vision and linguistics as input, the confident weights are utilized to learn the correlation of each node. Then, to locate relevant node relationship in the heterogeneous graph conditioned by the given question and image, we utilize heterogeneous graph reasoning to get the evolved heterogeneous graph representation. Finally, given evolved graph representation, a guidance mechanism is utilized to route the correct output content.
Moreover, there exists some ambiguous semantics (e.g. rainy day) that lack of specific labels for detection and can not benefit from the labeled object bounding boxes and categories such as “person” and “dog” during training in VCR task. To solve this problem, our HGL integrates a contextual voted module (CVM) for visual scene understanding with a global perspective at the low-level features.
The key merits of our paper lie in four aspects: a) a framework called HGL is introduced to seamlessly integrate the intra-graph and inter-graph in order to bridge vision and linguistic domain, which consists of a heterogeneous graph module and a CVM; b) a heterogeneous graph module is proposed including a primal VAHG and a dual QAHG to collaborate with each other via heterogeneous graph reasoning and guidance mechanism; c) a CVM is presented to provide a new perspective for global reasoning; d) extensive experiments have demonstrated the state-of-the-art performance of our proposed HGL on three cognition-level tasks.
![(a) Answer-to-answer homogeneous graph is to model intra-relationship of each word in all answers of linguistics; (b) Vision-to-vision homogeneous graph is to mine intra-relationship of each object from images; (c) Vision-to-Answer heterogeneous graph is to excavate inter-relationship between object and answer. The dotted portion in (a)&(b) means information isolated island. The concept of information isolated island in our paper refers to the independent of different semantic nodes can not achieve semantic inference in a homogeneous graph that connects similar semantic nodes by attribute (e.g. Figure 1(a)) or grammar (e.g. Figure 1(b)).[]{data-label="fig:pic1"}](figure/pic1){width="1\linewidth"}
Related Work
============
**Visual Comprehension** Recently, visual comprehension has made significant progress in many tasks, such as visual question answering [@hudson2019gqa; @ben2017mutan; @antol2015vqa], visual dialog [@guo2019image; @tang2018learning; @das2017visual] and visual question generation [@krishna2019information; @jain2017creativity]. There are mainly two aspects to methodology in the domain of visual comprehension. On the one hand, an attention-based approach was usually applied and raised its superior performance. Anderson et al. [@anderson2018bottom] presented a powerful architecture driven via bottom-up and top-down attention for image captioning and visual question answering. In multi-hop reasoning question answering task, a bi-directional attention mechanism [@cao2019bag] was proposed that was combined with entity graph convolutional network to obtain the relation-aware representation of nodes for entity graphs. On the other hand, a graph-based approach was developing rapidly recently, combining graph representation of questions and topic images with graph neural networks. Wu et al. [@wu2018image] incorporated high-level concepts such as external knowledge into the successful CNN-RNN approach for image captioning and visual question answering. A graph-based approach [@norcliffe2018learning] to model object-level relationships conditioned by the given question and image, including spatial relationship and object semantics relationship. In contrast, our proposed HGL differs in that inter-relationship is built via different domains (e.g. vision-to-answer graph) to align vision and linguistic domains.
**Graph Learning** Some researchers effort to model domain knowledge as homogeneous graph for excavating correlations among labels or objects in images, which has been proved effective in many tasks [@niepert2016learning; @lu2016visual]. Graph convolution approaches with spectral variants [@bruna2013spectral] and diffusion approaches [@duvenaud2015convolutional] have well developed and been applied into semi-supervised node classification [@kipf2017semi]. Some researches utilize the adjacency matrix to model the relationship of all node pairs [@wang2016structural; @cao2016deep], while others incorporate higher-order structure inspired by simulated random walks [@abu2017learning; @grover2016node2vec]. Li et al. [@li2018factorizable] solved scene graph generation via subgraph-based model using bottom-up relationship inference of objects in images. Liang et al. [@liang2018dynamic] modeled semantic correlations via neural-symbolic graph for explicitly incorporating semantic concept hierarchy during propagation. Yang et al. [@yang2016end] built a prior knowledge-guide graph for body part locations to well consider the global pose configurations. In this work, we firstly propose a heterogeneous graph learning to seamlessly integrate intra-graph and inter-graph reasoning to generate persuasive reasoning paths and bridge cross-domain semantic alignment.
![Overview of our HGL framework. Taking the image, question and candidate answers with four-way multiple choices as input, we use HGL to predict the right choice of candidate answers. We firstly use CNN (ResNet50 [@he2015deep]) tailed with the CVM to obtain the visual representation with global reasoning. Then we utilize the shared BERT [@devlin2018bert] to extract question representation and candidate answer representation, respectively. Then taking the three representations as the input of heterogeneous graph module, a primal VAHG module with a dual QAHG module is used to construct heterogeneous graph relationship to align semantics among vision, question and answer via heterogeneous graph reasoning and guidance, which outputs two evolved representations. The two representations are fed into a parser and classification to classify the final result.[]{data-label="fig:framework"}](figure/pic2){width="1\linewidth"}
Methodology {#method}
===========
Our Heterogeneous Graph Learning (HGL) framework is shown in Figure \[fig:framework\]. The HGL consists of a heterogeneous graph module (e.g. a primal VAHG module and a dual QAHG module) and a contextual voting module (CVM). The heterogeneous graph module is to align semantic nodes between vision and linguistic domain. The goal of CVM is to exploit long-range visual context for better global reasoning.
Definition
----------
Given the object or region set $\mathcal{O}=\{o_i\}_{i=1}^N$ of an input image $\mathcal{I}$, the word set $\mathcal{Q}=\{q_i\}_{i=1}^M$ of a query and the word set $\mathcal{A}=\{a_i\}_{i=1}^B$ of the candidate answers, we seek to construct heterogeneous graph node set $\mathcal{V}^o=\{v_i^o\}_{i=1}^N$, $\mathcal{V}^q=\{v_i^q\}_{i=1}^M$ and $\mathcal{V}^a=\{v_i^a\}_{i=1}^B$, correspondingly. Each node $v_i^o\in\{\mathcal{V}^o\}$ corresponds to a visual object $o_i\in\mathcal{O}$ and the associated feature vector with $d$ dimensions indicates $\mathbf{v}_i^o\in\mathbb{R}^d$. Similarly, the associated query word vector and associated answer word vector can separately be formulated as $\mathbf{v}_i^q\in\mathbb{R}^d$ and $\mathbf{v}_i^a\in\mathbb{R}^d$. By concatenating the joint embedding $\mathbf{v}_i$ together into a matrix $\mathbf{X}$, we separately define three matrices, such as vision matrix $\mathbf{X}_o\in\mathbb{R}^{N\times d}$, query matrix $\mathbf{X}_q\in\mathbb{R}^{M\times d}$ and answer matrix $\mathbf{X}_a\in\mathbb{R}^{B\times d}$, where $N$, $M$ and $B$ denote separately the visual object number, query word number and answer word number. We define the final output of our framework as $\mathbf{Y}_{p}\in\mathbb{R}^{4}$, which is a vector with 4 dimensions according to four-way multiple choice of candidate answers as shown in Figure \[fig:framework\].
![Implementation details of the primal VAHG module and the dual QAHG module by taking the representation of image, question and answer as inputs.[]{data-label="fig:framework2"}](figure/pic5){width="1\linewidth"}
Heterogeneous Graph Learning
----------------------------
### Vision-to-answer heterogeneous graph (VAHG) module
This module aims to align the semantics between vision and candidate answer domains via a heterogeneous graph reasoning, then generate a vision-to-answer guided representation $\mathbf{Y}^{v}\in\mathbb{R}^{B\times d}$ for classification via a guidance mechanism (Figure \[fig:framework2\]). We firstly introduce the reasoning of vision-to-answer heterogeneous graph, then show the guidance mechanism. We define the vision-to-answer heterogeneous reasoning as: $$\begin{aligned}
\mathbf{Y}_o=\delta(\mathbf{A}_{}^{T}\mathbf{X}_o\mathbf{W}_o), \label{vision-gcn}\end{aligned}$$ where $\mathbf{Y}_o\in\mathbb{R}^{B\times d}$ is the visual evolved representation and $\mathbf{W}_{o}\in\mathbb{R}^{d\times d}$ is a trainable weighted matrix. $\delta$ is a non-linear function. The $\mathbf{A}_{}\in\mathbb{R}^{N\times B}$ is a heterogeneous adjacency weighted matrix by calculating the accumulative weights of answer nodes to vision nodes, which is formulated as: $$\begin{aligned}
A_{ij}^{}=\frac{\mathrm{exp}(A_{ij}^{'})}{\sum_{ij}^{}\mathrm{exp}(A_{ij}^{'})}, \quad A_{ij}^{'}=\mathbf{v}_i^{oT}\mathbf{v}_j^a, \label{va-relation}\end{aligned}$$ where $A_{ij}^{}\in\mathbf{A}_{}$ is a scalar to indicate the correlations between $\mathbf{v}_i^{o}$ and $\mathbf{v}_j^a$. The $\mathbf{A}_{}\in\mathbb{R}^{N\times B}$ is the heterogeneous adjacency weighted matrix normalized by using a softmax at each location. In this way, different heterogeneous node representation can adaptively propagate to each other.
We obtain the visual evolved representation $\mathbf{Y}_o$ via vision-to-answer heterogeneous graph reasoning (Equation (\[vision-gcn\])). Given the $\mathbf{Y}_o$ and answer matrix $\mathbf{X}_a\in\mathbb{R}^{B\times d}$ as input, we present a guidance mechanism for producing the vision-to-answer guided representation $Y^{v}$. We divide the guidance mechanism into two steps. We first generate a middle representation $\mathbf{X}_{middle}$ that is enhanced by word-level attention values, then we propose a final process to generate the target representation $\mathbf{Y}^{v}$. The first step is formulated as following: $$\begin{aligned}
\mathbf{X}_{a'}^{}&=\mathcal{F}_{}^{}(\mathbf{X}_a),\label{self-va1}\\
\mathbf{x}_{m}^{}=a_{n}^{}\mathbf{x}_{a'}^{}, &\quad
a_{n}^{}=\frac{\mathrm{exp(\mathbf{X}_{a'}^{}\mathbf{W}_{a'}^{})}}{\sum_{n\in B}\mathrm{exp}(\mathbf{X}_{a'}^{}\mathbf{W}_{a'}^{})},\label{self-va2}\\
\mathbf{X}_{middle}^{}&=f_{}([\mathbf{X}_{m}^{},\mathbf{Y}_o]),\end{aligned}$$ where $\mathcal{F}_{}$ is a MLP used to encode the answer feature, $a_{n}^{}$ is a word-level attention value by utilizing weighted product $\mathbf{W}_{a'}^{}$ on the encoded answer feature $\mathbf{X}_{a'}^{}$ with $B$ number words. These attention values are normalized among all word-level linguistic representation via a softmax operation,which is shown in Equation (\[self-va2\]). Then we apply the attention value $a_n$ on $\mathbf{x}_{a'}\in\mathbf{X}_{a'}$ to produce an attention vector $\mathbf{x}_{m}^{}\in\mathbf{X}_{m}^{}$. We concatenate the vector $\mathbf{x}_{m}$ together into a matrix $\mathbf{X}_{m}$ that is enhanced by the attention value. After that, to combine the relationship between $\mathbf{X}_{m}$ and $\mathbf{Y}_o$ with $\mathbf{Y}_o$ instead of simply combining $\mathbf{Y}_o$ with $\mathbf{X}_{m}$, we concatenate the $\mathbf{X}_{m}^{}$ with $\mathbf{Y}_o$, then utilize a MLP $f$ on the concatenated result to get a middle representation $\mathbf{X}_{middle}^{}\in\mathbb{R}^{B\times d}$ for generating our final vision-to-answer guided representation $\mathbf{Y}^{v}$.
At the second step, the $\mathbf{X}_{middle}^{}$ and the visual evolved representation $\mathbf{Y}_o$ are utilized for producing the vision-to-answer guided representation $\mathbf{Y}^{v}$, which is formulated as: $$\begin{aligned}
\mathbf{Y}^{v}&=\Psi_{}(\phi(\mathbf{Y}_o\mathbf{W}_{o'}+\mathbf{X}_{middle}^{}\mathbf{W}_a^{})\mathbf{W}^{}),\end{aligned}$$ where $\mathbf{W}_{o'}^{}$ and $\mathbf{W}_{a}^{}$ are both learnable mapping matrixes to map the different embedding features into a common space to better combine the $\mathbf{Y}_o$ and $\mathbf{X}_{middle}^{}$, and the senior weighted matrix $\mathbf{W}^{}$ is to map the combination result into the same dimensionality. The $\Phi_{}$ and $\phi_{}$ are both vision-guided functions such as MLPs to get the final vision-to-answer guided representation $\mathbf{Y}^{v}$.
### Question-to-answer heterogeneous graph (QAHG) module
In this section, we produce a question-to-answer heterogeneous graph module that is similar to VAHG. The implementation details of the QAHG module are shown in Figure \[fig:framework2\]. This module aims to support the proper semantic alignment between question and answer domains. Given the query word vector and answer word vector as input, we aim to generate the question-to-answer guided representation $\mathbf{Y}^{q}\in\mathbb{R}^{B\times d}$ as the final output of this module. Specifically, taking the answer vector as input, we utilize a question-to-answer heterogeneous graph reasoning to produce a query evolved representation $\mathbf{X}_{q}\in\mathbb{R}^{B\times d}$. Then a symmetric guidance mechanism is utilized for generating the question-to-answer guided representation $\mathbf{Y}^{q}$.
After getting the $\mathbf{Y}^{v}$ and $\mathbf{Y}^{q}$ from VAHG module and QAHG module, respectively, we utilize a parser at the end of the HGL framework as shown in Figure \[fig:framework\] to adaptively merge $\mathbf{Y}^{v}$ and $\mathbf{Y}^{q}$ to get an enhanced representation $\mathbf{Y}^{a}$ for final classification. The parser can be formulated as: $$\mathbf{Y}^{a}=F(w^{o}\mathbf{Y}^{v} + w^{q}\mathbf{Y}^{q}).$$ where $w^{o}$ and $w^{q}$ are derived from the original visual feature, query feature and answer feature to calculate the importance of the task. We use a simple dot product to merge the two representations ($\mathbf{Y}^{v}$ and $\mathbf{Y}^{q}$). Then we use linear mapping function $F$ such as FC to produce the enhanced representation $\mathbf{Y}^{a}$ for final classification. The $w^{o}$ and $w^{q}$ can be calculated as: $$\begin{aligned}
w^{o}=&\frac{\mathrm{exp}(\varphi_v([\mathbf{X}_o,\mathbf{X}_a]\mathbf{W}_{oa})}{\mathrm{exp}(\varphi_v([\mathbf{X}_o,\mathbf{X}_a]\mathbf{W}_{oa})+\mathrm{exp}(\varphi_q([\mathbf{X}_q,\mathbf{X}_a]\mathbf{W}_{qa})},\\
w^{q}=&\frac{\mathrm{exp}(\varphi_q([\mathbf{X}_q,\mathbf{X}_a]\mathbf{W}_{qa})}{\mathrm{exp}(\varphi_v([\mathbf{X}_o,\mathbf{X}_a]\mathbf{W}_{oa})+\mathrm{exp}(\varphi_q([\mathbf{X}_q,\mathbf{X}_a]\mathbf{W}_{qa})},\end{aligned}$$ where $\mathbf{W}_{oa}$ and $\mathbf{W}_{qa}$ are both trainable weighted matrices, and $\varphi_v$ and $\varphi_q$ indicate different MLP networks. $[.]$ means concatenation operation.
### Contextual voting module
This module aims to replenish relevant scene context into local visual objects to give the objects with a global perspective via a voting mechanism, which is formulated as: $$\begin{aligned}
y_{i}^l&=\frac{1}{\mathcal{C}(x)} \sum_{\forall j} f\left(x_{i}^l, x_{j}^l\right) g\left(x_{j}^l\right), \\\label{cvm1}
a_{x_{j}\rightarrow x_{i}}&=\frac{\mathrm{exp}(W_n^{aT}\phi(x_i^l, x_{j}^l))}{\sum_{n\in\mathcal{N}}^{} \mathrm{exp}(W_n^{aT} \phi(x_{i}^l, {x}_{j}^l))}, \\\label{cvm2}
y_{i}^{l+1}&=a_{x_{j}\rightarrow x_{i}}{y}_{i}^{l}{W}^a + {x}_{i}^{l},\end{aligned}$$ where $y_{i}^l$, $x_{i}^l$ denote the output and the input at position $i$ of $l$-th convolution layer, and $x_{j}^l$ denote the input at position $j$ in relevant image content. The Equation (\[cvm1\]) represents that each output has collected global input information from relevant positions. In Equation (\[cvm2\]), $a_{x_{j}\rightarrow x_{i}}$ denotes the learnable voting weight from position $x_{j}$ to $x_{i}$ for adaptively select relevant contextual information into local visual feature via weighted sum and element-wise product. The $W^a$ and $W_n^a$ are both trainable weights and $\phi,f,g$ are non-linear functions with conv 1$\times$1 operation. The output of this module is the $y_{i}^{l+1}$, which denotes the residual visual feature maps via residual addition between input ${x}_{i}^{l}$ and the enhanced feature $a_{x_{j}\rightarrow x_{i}}{y}_{i}^{l}{W}^a$.
Experiments
===========
Task Setup
----------
The visual commonsense reasoning (VCR) task [@zellers2019vcr] is a new cognition-level reasoning consists of three sub-tasks: Q$\rightarrow$A, QA$\rightarrow$R and Q$\rightarrow$AR. The VCR is a four-way multi-choice task, and the model must choose the correct answer from four given answer choices for a question, and then select the right rationale from four given rationale choices for that question and answer.
Dataset and Evaluation
----------------------
We carry out extensive experiments on VCR [@zellers2019vcr] benchmark, a representative large-scale visual commonsense reasoning dataset containing a total of 290k multiple choice QA problems derived from 110k movie scenes. The dataset is officially split into a training set consisting of 80,418 images with 212,923 questions, a validation set containing 9,929 images with 26,534 questions and a test set made up of 9,557 with 25,263 queries. We follow this data partition in all experiments. The dataset is challenge because of the complex and diverse language, multiple scenes and hard inference types as mentioned in [@zellers2019vcr]. Of note, unlike many VQA datasets wherein the answer is a single word, there are more than 7.5 words for average answer length and more than 16 words for average rationale length. We strictly follow the data preprocessing and evaluation from [@zellers2019vcr] for fairly comparison.
Implementation
--------------
We conduct all experiments using 8 GeForce GTX TITAN XP cards on a single server. The batch size is set to 96 with 12 images on each GPU. We strictly follow the baseline [@zellers2019vcr] to utilize the ResNet-50 [@he2015deep] and BERT [@devlin2018bert] as our backbone and implement our proposed heterogeneous graph learning on it in PyTorch [@paszke2017automatic]. The hyper-parameters in training mostly follow R2C [@zellers2019vcr]. We train our model by utilizing multi-class cross entropy between the prediction and label. Each task is trained separately for question answering and answer reasoning via the same network. For all training, Adam [@kingma2014adam] with weight decay of 0.0001 and beta of 0.9 is adopted to optimize all models. The initial learning rate is 0.0002, reducing half ($\times$0.5) for two epochs when the validation accuracy is not increasing. We train 20 epochs for all models from scratch in an end-to-end manner. Unless otherwise noted, settings are the same for all experiments.
Comparison with state-of-the-art
--------------------------------
**Quantitative results.** During this section, we show our state-of-the-art results of validation and test on VCR [@zellers2019vcr] dataset with respect to three tasks in Table \[tab:results\]. Note the label of the test set is not available, and we get the test predictions by submitting our results to a public leaderboard [@zellers2019vcr]. As can be seen, our HGL achieves an overall test accuracy of 70.1% compared to 65.1% by R2C [@zellers2019vcr] on Q$\rightarrow$A task, 70.8% compared to 67.3% on QA$\rightarrow$R task, and 49.8% compared to 44.0% on Q$\rightarrow$AR task, respectively. To compare with the state-of-the-art text only methods on Q$\rightarrow$A task, our HGL performs 16.2% test accuracy improvement better than BERT [@devlin2018bert], and even outperform ESIM+ELMo [@Chen2017EnhancedLF] by 24.2% test accuracy. Compared with the several advanced methods of VQA, our model improves around 23.9% test accuracy at least on three tasks. The superior performance further demonstrates the effectiveness of our model on the cognition-level task.
**Qualitative results.** Figure \[fig:result2\] shows the qualitative result of our HGL. It also shows the primal learned vision-to-answer heterogeneous graph (VAHG) and a dual learned question-to-answer heterogeneous graph (QAHG) to further demonstrate the interpretability of our HGL. As shown in Figure \[fig:result2\], we utilize HGL on the four candidate responses to support proper semantic alignment between vision and linguistic domains. For better comprehensive analysis, we show the weighted connections of the VAHG and QAHG according to our correct predictions on different tasks. In Figure \[fig:result2\](d), our VAHG successfully aligns the visual representation “person5 (brown box)” to the linguistic word “witness”, and also successfully connects “person1 (red box)” with “person5” by the linguistic word “witness” to infer the right answer. Because the “person5 (brown box)” is the “witness” in this scenario. Visual representation “person1 (red box)” is connected with the emotional word “angry” to achieve a heterogeneous relationship. Based on the right answer, in Figure \[fig:result2\](g), our QAHG can connect the word “angry” with “gritting his teeth” for successfully reasoning the rationale. Moreover, in Figure \[fig:result2\](c), “feeling” from question can be aligned with the most suitable semantic word “anger” from the answer for right answer prediction, which demonstrated the effectiveness of our QAHG. These results can demonstrate that the VAHG and QAHG can really achieve proper semantic alignment between vision and linguistic domains for supporting cognition-level reasoning. The CVM is suitable to apply to visual context, because there are more information can be obtained from the visual evidence. For instance, the example of Figure \[fig:result\], the felling of the “person2” must get information from visual evidence (e.g. raindrop) instead of the question to predict the right answer and reason. The arrow is pointing at raindrop and snow. More results are shown in supplementary material.
-- ---------------------------------------- ---------- ---------- ---------- ---------- ---------- ----------
Val Test Val Test Val Test
Chance 25.0 25.0 25.0 25.0 6.2 6.2
BERT [@devlin2018bert] 53.8 53.9 64.1 64.5 34.8 35.0
BERT (response only) [@zellers2019vcr] 27.6 27.7 26.3 26.2 7.6 7.3
ESIM+ELMo [@Chen2017EnhancedLF] 45.8 45.9 55.0 55.1 25.3 25.6
LSTM+ELMo [@peters2018deep] 28.1 28.3 28.7 28.5 8.3 8.4
RevisitedVQA [@jabri2016revisiting] 39.4 40.5 34.0 33.7 13.5 13.8
BottomUpTopDown[@anderson2018bottom] 42.8 44.1 25.1 25.1 10.7 11.0
MLB [@Kim2017] 45.5 46.2 36.1 36.8 17.0 17.2
MUTAN [@ben2017mutan] 44.4 45.5 32.0 32.2 14.6 14.6
R2C [@zellers2019vcr] 63.8 65.1 67.2 67.3 43.1 44.0
HGL (Ours) **69.4** **70.1** **70.6** **70.8** **49.1** **49.8**
Human 91.0 93.0 85.0
-- ---------------------------------------- ---------- ---------- ---------- ---------- ---------- ----------
: Main results of validation and test dataset on VCR with respect to three tasks. Note that we do not need any extra information such as additional data or features.[]{data-label="tab:results"}
Model $Q\rightarrow A$ $QA\rightarrow R$ $Q\rightarrow AR$
-- ------------------ ------------------ ------------------- -------------------
Baseline 63.8 67.2 43.1
Baseline w/ CVM 65.6 68.4 45.4
Baseline w/ QAHG 66.1 68.2 45.8
Baseline w/ VAHG 66.4 69.1 46.4
HGL w/o CVM 68.4 69.7 48.3
HGL w/o QAHG 67.8 69.9 48.2
HGL w/o VAHG 68.0 68.8 48.0
HGL **69.4** **70.6** **49.1**
: Ablation studies for our HGL on three tasks over the validation set.[]{data-label="tab:ablations"}
Ablation Studies
----------------
**The effect of CVM.** Note that our CVM learns a more enhanced object feature with a global perspective, respectively. The advantage of CVM is shown in Table \[tab:ablations\], the CVM increases overall validation accuracy by 1.8% compared with baseline on Q$\rightarrow$A task, 1.2% on QA$\rightarrow$R and 3.1% on Q$\rightarrow$AR. In Figure \[fig:result\], the model w/ CVM shows the superior ability to successfully parse the semantics of rain and snow in the image for better understand the rainy/snowy scene by highlighting the relevant context as indicated by the red arrows in Figure \[fig:result\](b).
**The effect of QAHG.** The effect of QAHG module is apparently to boost the validation accuracy by around 1.0% from baseline on all tasks. In Figure \[fig:result2\](c), the “feeling” from the question is connected with the “getting angry” from the right answer choice, and the “getting angry” from the question is connected with “gritting his teeth” from the rationale in Figure \[fig:result2\](g), which demonstrates the effectiveness of QAHG that can generate persuasive reasoning paths.
**The effect of VAHG.** We analyze the effect of VAHG on VCR. The VAHG module can promote the baseline by 2.6% (Q$\rightarrow$A), 2.1% (QA$\rightarrow$R) and 3.3% (Q$\rightarrow$AR) accuracy. In Figure \[fig:result2\](h), the visual representation “person1 (red box)” is semantically aligned to the word “person1”. The visual representation “person5 (brown box)” is semantically aligned to the word “witness” in Figure \[fig:result2\](d), and “person1 (red box)” and “person5 (brown box)” are connected by the word “witness”. Based on these relationships, the HGL can refine reasoning paths for semantic agreement.
![Qualitative results of VAHG and QAHG. (a)(b)(e)(f) are the learned VAHG and QAHG of four-way multiple choices on answer task and reason task, respectively. (c)(d)(g)(h) show the weighted connection of VAHG and QAHG according to the prediction from four choices. The predicted result is shown as **bold** font, and the ground truth (GT) is shown as $\checkmark$. Please zoom in the colored PDF version of this paper for more details.[]{data-label="fig:result2"}](figure/pic4){width="1.0\linewidth"}
![Qualitative results of our CVM. (a) The model w/o CVM. (b) The model w/ CVM.[]{data-label="fig:result"}](figure/pic3){width="1.0\linewidth"}
**HGL w/o CVM.** As can be seen in Table \[tab:ablations\], the effect of combination between VAHG module and QAHG module can reach a high performance to 68.4%, 69.7% and 48.3%, which can demonstrate the effectiveness of building VAHG and QAHG to bridge vision and language domains.
**HGL w/o QAHG.** The proposed CVM collaborated with VAHG module is evaluated on three tasks and performs 67.8%, 69.9% and 48.2%, correspondingly. It can support the feasibility of incorporating CVM with VAHG module.
**HGL w/o VAHG.** The ability of CVM+QAHG is shown in Table \[tab:ablations\], which gets great scores on overall tasks as validating the availability of the combination.
Conclusion
==========
In this paper, we proposed a novel heterogeneous graph learning framework called HGL for seamlessly integrating the intra-graph and inter-graph reasoning in order to bridge the proper semantic alignment between vision and linguistic domains, which contains a dual heterogeneous graph module including a vision-to-answer heterogeneous graph module and a question-to-answer heterogeneous graph module. Furthermore, the HGL integrates a contextual voting module to exploit long-range visual context for better global reasoning.
Acknowledgements
================
This work was supported in part by the National Natural Science Foundation of China (NSFC) under Grant No.1811461, in part by the National Natural Science Foundation of China (NSFC) under Grant No.61976233, and in part by the Natural Science Foundation of Guangdong Province, China under Grant No.2018B030312002.
[^1]: Our code is released in
[^2]: Corresponding author is Xiaodan Liang
| {
"pile_set_name": "ArXiv"
} |
---
title: 'Nuclear $k_{T}$ in d+Au Collisions from Multiparticle Jet Reconstruction at STAR'
---
Introduction {#intro}
============
Many analyses of Au+Au collisions at RHIC probe the dense medium using jets by studying medium modifications of the jet yields and jet shapes [@Medium; @Mod]. The results from d+Au and p+p collisions provide the baseline for medium modification studies in Au+Au collisions. For these simpler systems, jets can be examined in great detail via full jet reconstruction.
This study presents the most recent d+Au nuclear $k_{T}$ measurements from full jet reconstruction at RHIC based upon year 2003 STAR data. The transverse and longitudinal jet shapes characterized by the $j_T$, and $z$ distributions from p+p collisions are also presented. $k_T$ measures the transverse momentum of a parton within a nucleon or nucleus, and its growth from p+p to nuclear collisions is believed to contribute to the Cronin effect [@cronin].
Experiment
==========
The STAR detector is well suited for investigating jet production at RHIC, due to the complete $\phi$ coverage and large $\eta$ coverage of the TPC and EMC. The STAR Barrel EMC data used in this study provide neutral energy measurements including $\pi^{0}$ decay photons. During the 2003 RHIC run, only half of the full barrel EMC was installed. This first half of the detector consists of 2400 towers of $0.05 \otimes 0.05$ in $\eta \otimes \phi$ each, leading to full azimuthal coverage in $0< \eta <1$. The barrel EMC was read out in minimum bias events. It was also used to trigger on “high tower” events, where one of the towers was above a nominal energy threshold of $E_{T}>2.5$ GeV. This “high tower” triggered event sample contains a much larger fraction of jets than the minimum bias event sample.
[0.8]{}
This analysis uses two different algorithms that reconstruct a jet by capturing the spray of fragmenting particles in a geometric cone. One centers the cone on the most energetic hadrons in the event, while the other optimizes the cone direction to maximize the included energy. Only the hadron centered cone algorithm is robust enough at high multiplicity to be used for reconstructing jets in d+Au collisions.
Measurement of jet energy requires corrections for charged particle energy deposition in the EMC, for the finite efficiencies of the TPC and EMC, and for the unmeasured energy carried by long-lived neutral particles ($n$, $K_L$, ...). It is also possible for the jet reconstruction algorithm to miss soft particles.
A sample of [pythia]{}[^1] events processed by the STAR detector simulator was analyzed in the same fashion as the data. Since the originating parameters for each parton were recorded in the monte-carlo [pythia]{} sample, the energy scale response of the STAR detector could be quantified. Figure 1 shows the total energy scale correction for jet radii of 0.7 and 0.5 as a function of $E_{T(Uncorrected)}$. From here on, $E_{T}$ will exclusively refer to reconstructed transverse jet energy, corrected for detector energy response using the curves in Figure 1.
Inclusive Jet Studies
=====================
[0.8]{}
Jets in this study are required to pass the following cuts: number charged hadrons $>1$, Cone Radius=0.5, $\frac{E_{neutral}}{E_{charged}}<.65$, and jet thrust axis $>40(70)cm$ from the inside(outside) of the BEMC (roughly $.2<\eta<.65$ depending on z-Vertex). When analyzing jet shape, the perpendicular component of the hadron momentum ($j_T$), and the fragmentation fraction $z$, are generally used. Figure 2 shows jet $j_{T}$ and $p_{h}/E_T$ distributions using these cuts. Jets with high hadron count are biased toward higher detector energy response. The $p_{h}/E_{T}$ distribution, since it is a per hadron distribution, is biased toward jets with high hadron count. This means $p_{h}/E_{T}$ is not directly comparable to the theoretical fragmentation function Dh(z), due to the bias introduced by the high hadron count jets. However, the substantial agreement between [pythia]{} and the data demonstrates that the phenomenological parameterizations embodied by [pythia]{} continue to be valid at these jet energies.
The di-jet opening angle is another basic observable obtained from jet events. For di-jets, $\Delta\phi=\pi$ in leading order QCD. Gluon radiation broadens $\Delta\phi$, which is measured by the per-parton intrinsic $k_{T}\equiv E_{T}\sin{\Delta\phi}$. When reconstructing di-jets, one “trigger” jet with observed $E_T > 7$ GeV is reconstructed from both neutral and charged hadrons using the high tower which triggered the event. The other “away” jet direction is reconstructed using charged particles only, allowing the “away” jet to range $-0.5<\eta<0.5$. The “away” jet is not used to estimate jet energy scale. Figure 3 shows the di-jet $k_{T}$ distribution comparison for p+p and d+Au collisions. The p+p collisions have been supplemented with d+Au minbias background such that the multiplicity distributions of the supplemented p+p events matches that of the d+Au “high tower” sample. Taking $\sigma_{k_{T}(obs)}^{2}=\sigma_{k_{T}(pp)}^{2}+\sigma_{k_{T}(nucl)}^{2}$, the $\sqrt{\sigma_{k_{T}(nucl)}^{2}}=1.21\pm0.35\pm^{0.65}_{0.57}$ GeV/c in d+Au collisions. The major systematic uncertainties are the jet energy scale in p+p and d+Au, the detector resolution, fit uncertainties, and d+Au background multiplicity. The fit uncertainties are largest. The systematic error on the d+Au $\sigma_{k_{T}(nucl)}$ due to the detector $k_{T}$ resolution is small. In contrast, it is potentially large for the absolute measurement of $\sigma_{k_{T}(pp)}$, and this effect is currently under study.
[0.8]{}
[ ]{}
Conclusion
==========
Jets have been reconstructed in both p+p and d+Au collisions. The p+p jets have been used to measure $j_{T}$, $z$ and $k_{T}$. The inclusive jet $j_{T}$ and $z$ results compare well with the [pythia]{} monte-carlo simulation studies. The d+Au $k_{T_{nucl}}$ value, though having large uncertainty, appears to be smaller than previous experiments [@E609; @Collaboration] at much lower energy. The $\langle j_T \rangle$ and $\sqrt{\langle k_T^2 \rangle}$ results from full jet reconstruction agree well with similar measurements using di-hadron correlations [@subhasis].
[99]{} M. Gyulassy et al., nucl-th$/0302077$. D. Antreasyan et al., Phys. Rev. D 19, 764 (1979). S. Chattopadhyay et al. (STAR Collaboration), nucl-ex$/0403013$. M.D. Corcoran et al. (E609 Collaboration), Phys. Lett. B259, 209 (1991).
[^1]: PYTHIA version 6.203 and 6.205.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'In this paper, we introduce a class of stochastic partial differential equations (SPDEs) with fractional time-derivatives, and study the $L_2$-theory of the equations. This class of SPDEs can be used to describe random effects on transport of particles in medium with thermal memory or particles subject to sticking and trapping.'
address:
- 'Department of Mathematics, University of Washington, Seattle, WA 98195, USA'
- 'Department of Mathematics, Korea University, 1 Anam-dong, Sungbuk-gu, Seoul, 136-701, Republic of Korea'
- 'Department of Mathematical Sciences and Research Institute of Mathematics, Seoul National University, San56-1 Shinrim-dong Kwanak-gu, Seoul 151-747, Republic of Korea'
author:
- '[**Zhen-Qing Chen**]{}'
- '[**Kyeong-Hun Kim**]{}'
- '[**Panki Kim**]{}'
title: Fractional time stochastic partial differential equations
---
[^1]
[^2]
[^3]
**Introduction**
================
Fractional calculus has attracted lots of attentions in several fields including mathematics, physics, chemistry, engineering, hydrology and even finance and social sciences. The classical heat equation $\partial_t u=\Delta u$ describes heat propagation in homogeneous medium. The time-fractional diffusion equation $\partial^\beta_t u=\Delta u$ with $0<\beta<1$ has been widely used to model the anomalous diffusions exhibiting subdiffusive behavior, due to particle sticking and trapping phenomena. Here the fractional time derivative $\partial^{\beta}_t$ is the Caputo derivative of order $\beta\in (0,1)$, defined by $$\label{e:1.1}
\frac{\partial^\beta f(t)}{\partial t^\beta}=\frac{1}{\Gamma (1-\beta)} \frac{d}{dt}
\int_0^t (t-s)^{-\beta} \left(f(s)-f(0)\right) ds$$ where $\Gamma$ is the Gamma function defined by $\Gamma(\lambda):=
\int^{\infty}_0 t^{\lambda-1} e^{-t}dt$.
Fractional diffusion equations are becoming popular in many areas of application [@GM; @KS; @LS; @MK; @P; @SBMW; @V]. So far, on the basis of either deterministic or probabilistic methods, the study of fractional calculus is mainly restricted to deterministic equations; see [@EIK; @MS; @P; @Za] and the references therein. In this paper, we introduce and investigate a class of stochastic partial differential equations (SPDEs) with fractional time derivatives.
The SPDEs with fractional time derivative that we are going to study in this paper naturally arise from the consideration of the heat equation in a material of thermal memory. Let $u(t,x), e(t,x)$ and $\vec F(t,x)$ denote the body temperature, internal energy and flux density, respectively. Then the relations $$\label{heat 1}
\frac{\partial e}{\partial t}(t,x)=-{\rm div} \, \vec F,$$ $$e(t,x)=\beta u(t,x), \quad \vec F(t,x)=-\lambda \nabla u(t,x) , \quad\quad
\beta, \lambda >0$$ yield the classical heat equation $\beta \frac{\partial u}{\partial t}=\lambda \Delta u$. According to the law of the classical heat equation, the speed of the heat flow is infinite. However in real modeling, the propagation speed can be finite because the heat flow can be disrupted by the response of the material. It has been proved (see e.g., [@V; @LS]) that in a material with thermal memory $$\label{2014.5}
e(t,x)=\bar{\beta} u(t,x)+\int^t_0 n(t-s)u(s,x)ds$$ holds with some appropriate constant $\bar{\beta}$ and kernel $n$. Typically, $n(t)$ is a positive decreasing function which blows up near $t=0$, for instance $n(t)=t^{-\alpha}$ for $\alpha\in (0,1)$. In this case the convolution implies that nearer past affects the present more. If in addition the internal energy $e(t,x)$ depends also on past random effects, then becomes $$\label{2014.6}
e(t,x)=\bar{\beta} u(t,x)+\int^t_0 n(t-s)u(s,x)ds+\int^t_0
\ell(t-s)h(s, u(s,x))dW_s,$$ where $W$ is a random process, such as Brownian motion, modeling the random effects. If $u(0, x)=0$, $\bar{\beta}=0$, $n(t)=\Gamma ( 1-\beta_1 )^{-1} t^{-\beta_1}$ and $\ell(t)=\Gamma (2- \beta_2 )^{-1}t^{1-\beta_2}$ for some constants $\beta_i \in (0,1)$, then (\[2014.6\]) (after differentiation in $t$) becomes $$\begin{aligned}
\label{e:1.5}
-{\rm div} \, \vec F = \frac{\partial e}{\partial t}(t,x)&=&\frac{1}{\Gamma (1- \beta_1 )}\frac{\partial }{\partial t} \int^t_0 (t-s)^{-\beta_1} u(s,x)ds
\nonumber \\
&&+\frac1{\Gamma (2- \beta_2 )}\frac{\partial }{\partial t}\int^t_0
(t-s)^{1-\beta_2}h(s, u(s,x))dW_s.\end{aligned}$$ Since $$\begin{aligned}
\int_0^t (t-s)^{-{\beta_2}} \int_0^s
h(a, u(a,x))
dW_ads&= \int_0^t
\int_a^t (t-s)^{-{\beta_2}}ds
h(a, u(a,x))
dW_a
\\&=\frac{1}{1-{\beta_2}} \int_0^t
(t-a)^{1-{\beta_2}}
h(a, u(a,x))
dW_a,\end{aligned}$$ by the definition of Caputo derivative (\[heat 1\]) we have $$\begin{aligned}
\partial^{\beta_2}_t \int_0^t
h(s, u(s,x))
dW_s&=
\frac{1}{\Gamma (1-{\beta_2})}\frac{\partial }{\partial t}
\int_0^t (t-s)^{-{\beta_2}} \int_0^s
h(a, u(a,x))
dW_ads\\
&=
\frac{1}{\Gamma (2-{\beta_2})}\frac{\partial }{\partial t}
\int_0^t
(t-s)^{1-{\beta_2}}
h(s, u(s,x))
dW_s.\end{aligned}$$ Thus can be rewritten as the following stochastic partial differential equation involving fractional time-derivative $$\label{e:1.4}
\partial^{\beta_1}_t u = {\rm div} \, \vec F + \partial^{\beta_2}_t \int_0^t
h(s, u(s,x)) dW_s.$$ It is this type of stochastic equations and its natural extensions that will be studied in this paper.
Now let $(\Omega,{\mathcal{F}},{\mathbb{P}})$ be a complete probability space, $\{{\mathcal{F}}_{t},t\geq0\}$ be an increasing filtration of $\sigma$-fields ${\mathcal{F}}_{t}\subset{\mathcal{F}}$, each of which contains all $({\mathcal{F}},{\mathbb{P}})$-null sets. We assume that on $\Omega$ we are given an independent family of one-dimensional Wiener processes $W^{1}_{t}$, $W^{2}_{t},...$ relative to the filtration $\{{\mathcal{F}}_{t},t\geq0\}$.
Motivated by , in this paper we consider the following quasi-linear SPDEs of the non-divergence form type $$\begin{aligned}
\nonumber
\partial^{\beta}_t u&=&\left(a^{ij}u_{x^ix^j}+b^iu_{x^i}+cu+f(u) \right)\\
&& +\sum_{k=1}^\infty\partial^{\gamma}_t \int^t_0 (\sigma^{ijk}u_{x^ix^j} +\mu^{ik}u_{x^i}+\nu^ku+g^k(u))\, d
W^k_s
\label{eqn 2014.1}
\end{aligned}$$ as well as of the divergence form type $$\begin{aligned}
\nonumber
\partial^{\beta}_t u &=&\left( D_i\left(
a^{ij}u_{x^j}+b^iu+f^i(u)\right)+cu+h(u)
\right) \\
&& + \sum_{k=1}^\infty\partial^{\gamma}_t \int^t_0 (\sigma^{ijk}u_{x^ix^j} +\mu^{ik}u_{x^i}+\nu^ku+g^k(u))\,
dW^k_s, \label{2014.2}
\end{aligned}$$ given for $\omega\in \Omega, \, t\geq0$ and $x\in {\mathbb{R}}^d$, and study the $L_2$-theory of the equations. The constants $\beta,\gamma\in (0,1)$ are assumed to satisfy the condition $$\label{e:1.7}
\gamma<\beta+1/2.$$ The indices $i$ and $j$ go from $1$ to the dimension $d$ with the summation convention on $i,j$ being enforced. The coefficients $a^{ij}, b^i, c, \sigma^{ijk}, \mu^{ik}, \nu^k$ are functions depending on $(\omega,t,x)$ and the functions $f,f^i, h, g^k$ depend on $(\omega,t,x)$ and the unknown $u$. By considering infinitely many independent Brownian motions $W^k_t$ we cover equations driven by measure-valued processes, for instance, driven by space-time white noise (see Section 3.3). It is worth mentioning that unlike the classical SPDE, we allow the second-order derivatives of the unknown solution $u$ to appear in the stochastic part when $\gamma<1/2$.
As for stochastic differential equations (SDEs), SPDE should be interpreted by its integral form $$\begin{aligned}
&&u(t,x)-u(0,x)\\
&=&\frac{1}{\Gamma(\beta)}\int^t_0
(t-s)^{\beta-1}\left(a^{ij}(s,x)u_{x^ix^j}(s,x)+\cdots+f(s,
u(s,x))\right)ds\\
&&+\frac{1}{\Gamma(1+\beta-\gamma)}
\int^t_0(t-s)^{\beta-\gamma}\left(\sigma^{ijk}(s,x)u_{x^ix^j}(s,x)+\cdots+g^k(s,
u(s,x))\right)dW^k_s.\end{aligned}$$ Similarly one can write down the integral version of SPDE but in the distributional sense with respect to $x$ variable.
We next explain the constraint . A special case of the SPDEs for both and is $$\label{e:1.8}
\partial^\beta_t u (t,x) =\Delta u(t,x) + \partial^\gamma_t \int_0^t g(s,x) dW_s,$$ where $W$ is a one-dimensional Brownian motion. For functions $h_1$ and $h_2$ on $\R_+$, we define its convolution $h_1*h_2$ by $$h_1*h_2 (t)=\int^t_0 h_1(t-s)h_2(s)ds.$$ Let $$k_{\beta}(t):=\Gamma(\beta)^{-1} t^{\beta-1},$$ and define $$I^{\beta} \varphi =k_{\beta}*\varphi := \int_0^t k_\beta (t-s)\varphi (s) ds.$$ One can easily check for any $\beta, \gamma \in (0,1)$, $k_\beta * k_\gamma = k_{\beta+\gamma}$. So in particular, we have $$\label{e:1.10}
k_{\beta}*k_{1-\beta} (t)\equiv 1 .$$ Suppose $u(x, t)$ is a solution of (\[e:1.8\]). In view of the definition of Caputo derivative , equation is understood by its integral version $$k_{1-\beta}*(u(t,x)-u(0, x))=\int^t_0 \Delta u(s,x)ds+k_{1-\gamma}*\int^t_0
g(s)dW_s.$$ By taking convolution with $k_{\beta}$ on both sides, we get from and $$\int^t_0
\left(u(s,x)-u(0,x)\right)\,ds=\left(I^{\beta}*\int^\cdot_0 \Delta
u(s,x)ds\right)(t)+\left(I^{\beta+1-\gamma}* \int^\cdot_0 g(s,x)dW_s\right)(t).$$ Since the first two terms are differentiable in $t$, the last term above should be differentiable in $t$. Recall that $$I^{a}: C^{b}\to C^{a+b},$$ and $\int^t_0 g(s)dW_s \in C^{1/2-\varepsilon}$ for any $\varepsilon>0$. Thus we must have $$\beta+1-\gamma>1/2,$$ which is equivalent to .
The main results of this paper are Theorems \[thm divtype\] and \[main\], on the unique solvability of SPDEs and , and $L_2$-estimates of their solutions. For SPDE , we establish in Theorem \[main\] the unique solvability in the space $L_2(\Omega\times[0,T], H^{\sigma}_2)$ for any $\sigma\in {\mathbb{R}}$ under appropriate differentiability assumption on $x$-variable of the coefficients. On the other hand, the unique solvability of SPDE in the space $L_2(\Omega\times[0,T], H^{1}_2)$ is obtained in Theorem \[thm divtype\] under the merely measurability condition of the coefficients $a^{ij}$.
The rest of the paper is organized as follows. In Section 2 we present some preliminary results on the fractional derivatives and in Section 3 we introduce stochastic Banach spaces and few key estimates. Our main results for (\[2014.2\]) and (\[eqn 2014.1\]) are presented and proved in Section 4 and Subsection 5.1, respectively. Subsection 5.2 contains an application to an equation driven by space-time white noise.
We close this section with some notation. As usual, ${\mathbb{R}}^{d}$ stands for the Euclidean space of points $x=(x^{1},...,x^{d})$. For $i=1,...,d$, multi-indices $\alpha=(\alpha_{1},...,\alpha_{d})$, $\alpha_{i}\in\{0,1,2,...\}$, and functions $u(x)$ we set $$u_{x^{i}}=\partial u/\partial x^{i}=D_{i}u,\quad
D^{\alpha}u=D_{1}^{\alpha_{1}}\cdot...\cdot D^{\alpha_{d}}_{d}u,
\quad|\alpha|=\alpha_{1}+...+\alpha_{d}.$$ We also use the notation $D^m_x$ for a partial derivative of order $m$ with respect to $x$. By $C^{\infty}_0({\mathbb{R}}^d)$ we denote the collection of smooth functions having compact support in ${\mathbb{R}}^d$. For $p\geq 1$, let $$L_p=L_p({\mathbb{R}}^d):=\{ u : {\mathbb{R}}^d \to {\mathbb{R}}, \| u \|^p_{L_p}:=\int_{{\mathbb{R}}^d} |u(x)|^p dx < \infty \}$$ and we use the notation $(f,g)_{L_2}:= \int_{{\mathbb{R}}^d} f(x)g(x) dx$. We denote $${\mathcal{F}}(g)(\xi):=\frac{1}{(2\pi)^{d/2}}\int_{{\mathbb{R}}^d}e^{-i\xi\cdot
x}g(x)dx \quad \text{and}\quad {\mathcal{F}}^{-1}(f)(\xi):=\frac{1}{(2\pi)^{d/2}}\int_{{\mathbb{R}}^d}e^{i\xi\cdot
x}f(x)dx$$ the Fourier transform of $g$ in ${\mathbb{R}}^d$ and the inverse Fourier transform of $f$ in ${\mathbb{R}}^d$, respectively.
If we write $N=N(a,b,\cdots)$, this means that the constant $N$ depends only on $a,b,\cdots$. Throughout this paper, for functions depending on $(\omega,t,x)$, usually the argument $\omega \in \Omega$ will be omitted.
**Preliminary results** {#section 1}
=======================
First we introduce a few elementary facts on the fractional derivatives. The reader can find further details in [@EIK] and references therein. Recall that $\beta\in (0,1)$ and $$k_{\beta}(t):=t^{\beta-1}\Gamma(\beta)^{-1}, \quad t>0.$$ Let $T>0$. If $f$ is absolutely continuous on $[0, T]$ with $f(0)=0$ then $$\label{2014.1.23.2}
\frac{d}{dt} (k_{\beta}*f)=k_{\beta}* \frac{d}{dt}f, \quad t \in [0, T].$$ For functions $\varphi\in L_1([0,T])$, the Riemann-Liouville fractional integral of the order $\beta\in (0,1)$ is defined by $$I^{\beta} \varphi (t)=k_{\beta}*\varphi (t)=\frac{1}{\Gamma(\beta)}\int^t_0 (t-s)^{\beta-1}\varphi(s)ds.$$ Note that by Jensen’s inequality $$|I^\beta \varphi (t)|^p \leq \frac{1}{(\Gamma (\beta))^p} \left(\frac{t^\beta}{\beta}\right)^{p-1}
\int_0^t (t-s)^{\beta-1} |\varphi (s)|^p ds.$$ Thus it follows that for any $p\in [1,\infty]$, $$\label{eqn 7.03.1}
\|I^{\beta}\varphi\|_{L_p([0,T])}\leq N(T,\beta) \|\varphi\|_{L_p([0,T])}.$$ Consequently, if $\varphi_n \to \varphi$ in $L_p([0,T])$ then $I^{\beta}\varphi_n$ also converges to $I^{\beta} \varphi$ in $L_p([0,T])$. Also one can prove that if $f_n(\omega,t)$ converges in probability uniformly in $[0,T]$ then so does $I^{\beta}f_n$.
If $I^{1-\beta}\varphi$ is absolutely continuous, then Riemann-Liouville derivative of order $\beta$ is defined by $$\label{Riemann-Liouville}
D^{\beta}_t\varphi (t)=\frac{d}{dt} (I^{1-\beta}\varphi)(t).$$ If $\varphi$ is continuous and $I^{1-\beta}\varphi$ is absolutely continuous, the generalized functional derivative (or the Caputo derivative) of order $\beta$ is given by $$\begin{aligned}
\label{caputo}
\partial^{\beta}_t\varphi (t)&=&D^{\beta}_t(\varphi-\varphi(0))
= D^{\beta}_t\varphi (t)-\frac{\varphi(0)}{t^{\beta}\Gamma(1-\beta)}.\end{aligned}$$ It is easy to check for any $\varphi\in L_1([0,T])$, $$\begin{aligned}
\label{e:DI}
D^{\beta}_tI^{\beta}\varphi =\varphi.\end{aligned}$$ Furthermore, if $\varphi$ is absolutely continuous on $[0,T]$ then by $$\begin{aligned}
\label{e:pbetav}
\partial^{\beta}_t\varphi=I^{1-\beta}\frac{d}{dt}\varphi.\end{aligned}$$ Thus by $$\begin{aligned}
\label{e:Dbetadt}
D^{1-\beta}_t\partial^{\beta}_t \varphi=\frac{d}{dt}\varphi, \quad \text{a.e.}.\end{aligned}$$
Denote by $$E_{\beta}(z)=\sum_{k=0}^{\infty}\frac{z^k}{\Gamma(\beta k+1)}, \quad z \in {\mathbb{C}}$$ the Mittag-Leffler function. We will also use the generalized Mittag-Leffler function $$E_{\beta,\gamma}(z):=\sum_{k=0}^{\infty}\frac{z^k}{\Gamma(\beta k+\gamma)}, \quad z \in {\mathbb{C}}.$$
We assume $\beta, \gamma \in (0,1)$. It is well known (see e.g. (12) in [@D Theorem 1.3-4]) that, when $-1 < \gamma -\beta < 1$, $E_{\beta,\gamma}(t)$ is bounded on $(-\infty,0]$ and $$\label{mittag}
\lim_{{\mathbb{R}}\ni t \to \infty} t E_{\beta,\gamma}(-t) =\frac{1}{\Gamma (\gamma-\beta)}.$$ Furthermore, for any constant $\lambda$, $$\label{frac heat}
\varphi(t):=E_{\beta}(\lambda t^{\beta})$$ is a solution of the equation $$\partial^{\beta}_t \varphi=\lambda \varphi, \quad t>0$$ with the initial condition $\varphi(0)=1$.
The following is a classical result. We provide a proof for the readers’ convenience.
\[l:DaE\] Let $a\in (0,1)$ and $b \ge 0$, then $$\label{seattle1}
D_t^a E_{\beta}(-b
t^{\beta})=t^{-a}E_{\beta,1-a}(-bt^{\beta}),$$ $$\label{seattle2}
I^aE_{\beta}(-b
t^{\beta})=t^{a}E_{\beta,1+a}(-bt^{\beta}).$$
We first prove (\[seattle1\]). One can easily check (see e.g. [@EIK (5.1.2)]) $$\label{seattle3}
D_t^at^{b-1}=\frac{\Gamma(b)}{\Gamma(b-a)}t^{b-a-1}, \quad a,b>0.$$ Thus, $$\begin{aligned}
D_t^a E_{\beta}(-b t^{\beta})&=&D_t^a\left(\sum_{k=0}^\infty
\frac{(-1)^kb^{2k}t^{\beta k}}{\Gamma(\beta k+1)}\right)\\
&=&\sum_{k=0}^\infty\frac{(-1)^kb^{2k}}{\Gamma(\beta
k+1)}\frac{\Gamma(\beta k+1)}{\Gamma(\beta k+1-a)} t^{\beta
k-a}\\
&=&t^{-a}\sum_{k=0}^\infty\frac{(-bt^{\beta})^k}{\Gamma(\beta
k+1-a)}\\
&=&t^{-a}E_{\beta,1-a}(-bt^{\beta}).\end{aligned}$$
To prove (\[seattle2\]), in place of (\[seattle3\]), it is enough to use $$I^at^{b-1}=\frac{\Gamma(b)}{\Gamma(b+a)}t^{b+a-1}, \quad a,b>0.$$ The lemma is proved.
Define $$p(t,x)={\mathcal{F}}^{-1} (E_{\beta}(-|\xi|^{2}t^{\beta})), \quad q(t,x)=D^{1-\beta}_t p.$$ Actually (\[mittag\]) shows that if $d>1$ then $E_{\beta}(-|\xi|^2t^{\beta})\not\in L_1({\mathbb{R}}^d)$ for fixed $t>0$. Thus we understand $p(t,x)$ as the inverse transform of a radial function in the sense of improper integral, or we can define $p(t,x)$ first as in [@EIK Section 5.2.2] so that $p(t,\cdot)\in L_1({\mathbb{R}}^d)$ and ${\mathcal{F}}(p(t,x))(\xi)=E_{\beta}(-|\xi|^2t^{\beta})$.
Since, for $x\neq 0$, $p(t,x)\to 0$ as $t\downarrow 0$ (see [@EIK]), the Riemann-Liouville derivative of $p(\cdot, x)$ coincides with the Caputo derivative of $p(\cdot , x)$ for every $x\not= 0$.
Here is a list of other useful properties of $p$ and $q$.
\[lemma list\] (i) For $(t,x)\in (0,T]\times {\mathbb{R}}^d \setminus \{0\}$, we have $\partial^{\beta}_t p=D^{\beta}_t p=\Delta p$.
\(ii) For each $x\neq 0$ and $m\leq 3$, $$\label{eqn 9.21.1}
\lim_{t\to 0+} D^m_x p(t,x)=\lim_{t\to 0+}D^m_x q(t,x)=0.$$
\(iii) $D^m_x p(t,\cdot)$ is integrable for each $t>0$ if $m\leq 1$.
\(iv) For each $t>0$ and $m\leq 3$, $D^m_x q(t,\cdot)$ is integrable in ${\mathbb{R}}^d$ uniformly on $[\varepsilon,T]$ for any $\varepsilon>0$.
\(v) For each $x\neq 0$, $p(\cdot,x)$ is absolutely continuous and $\frac{
\partial}{ \partial t}p(t,x)\to 0$ as $t\downarrow 0$. Moreover, $\frac{\partial}{ \partial t}p(t,\cdot)$ is integrable in ${\mathbb{R}}^d$ uniformly on $[\varepsilon,T]$ for any $\varepsilon>0$.
\(vi) For any compact set $K\subset {\mathbb{R}}^d\setminus \{0\}$ and $m\leq 3$, the functions $p, q, \frac{\partial}{ \partial t}p(t,\cdot), D^m p$ and $D^m q$ are continuous and bounded on $[0,T]\times K$.
\(i) is a consequence of (\[frac heat\]). See (5.2.44) and (A.19) of [@EIK] for (v). Others can be found in Propositions 5,1 and 5.2 of [@EIK]. In particular, (vi) is a consequence of (5.2.6) and (5.2.13) of [@EIK].
We need the following integration by parts formula.
\[L:2.6\] Suppose that $f$ is absolutely continuous on $[0, T]$ with $f(T)=0$ and $g$ is continuous and $I^{1-\beta}g$ is absolutely continuous on $[0, T]$. Then $$\int_0^T f(t)D^\beta_t g(t) dt = \int_0^T G(t) D^\beta_t F(t) dt,$$ where $F(t)=f(T-t)$ and $G(t)=g(T-t)$.
Note that $$\begin{aligned}
\int_0^T |f'(t)| \left( \int_0^t \theta^{-\beta}
|g(t-\theta )| d\theta\right) dt \le \|g\|_{L_\infty([0,T])}
\int_0^T |f'(t)| dt \int_0^T \theta^{-\beta}
d\theta <\infty.\end{aligned}$$ Thus, using $f(T)=0$, the integration by part and the Fubini’s Theorem, we get $$\begin{aligned}
\int_0^T f(t) D^\beta_t g(t) dt
&=& \int_0^T f(t) \frac{d}{dt} (I^{1-\beta} g) (t) dt \nonumber\\
&=& f(t) (I^{ 1-\beta } g) (t) \Big|_0^T -\int_0^T f'(t) (I^{1-\beta } g)(t) dt \nonumber\\
&=& - \frac{1}{\Gamma (1-\beta)} \int_0^T f'(t) \left( \int_0^t \theta^{-\beta}
g(t-\theta ) d\theta\right) dt \nonumber\\
&=& - \frac{1}{\Gamma (1-\beta)} \int_0^T \theta^{-\beta} \left( \int_\theta^T
f'(t) g(t-\theta ) dt \right) d\theta \label{e:sd} .\end{aligned}$$ As $F(0)=f(T)=0$ and $f$ is absolutely continuous, by and we have $D^\beta_t F = \partial^\beta_t F=(I^{1-\beta} F')$. So by the integration by part and the Fubini’s Theorem $$\begin{aligned}
\int_0^TG(t) D^\beta_t F(t) dt
&=& \int_0^T G(t) (I^{1-\beta} F') (t) dt \\
&=& \frac{1}{\Gamma (1-\beta )} \int_0^T G(t) \left( \int_0^t \theta^{-\beta}
F'(t-\theta ) d\theta \right) dt \\
&=& \frac{1}{\Gamma (1-\beta )} \int_0^T \theta^{-\beta} \left( \int_\theta^T G(t)
F'(t-\theta ) dt\right) d\theta \\
&=& - \frac{1}{\Gamma (1-\beta )} \int_0^T \theta^{-\beta} \left( \int_\theta^T
f' (T-t+\theta ) g(T-t) dt\right) d\theta \\
&=& - \frac{1}{\Gamma (1-\beta )} \int_0^T \theta^{-\beta} \left( \int_\theta^T
f' (s) g(s-\theta) ds\right) d\theta,\end{aligned}$$ which is $
\int_0^T f(s) D^\beta_t g(s) ds$ by . This proves the lemma.
\[L:2.3\] For each $(t, x)\in (0,
\infty) \times {\mathbb{R}}^d\setminus\{0\}$, $$\label{eqn lap}
\Delta q (t,x)=\frac{\partial}{\partial t}p (t,x).$$
Since both are continuous it is enough to prove that the equality holds for almost all $(t, x)$. Let $\phi (t)$ and $\psi (x)$ be smooth functions with compact support in $(0, T)$ and ${\mathbb{R}}^d\setminus\{0\}$ respectively. Define $\Phi (s):=\phi (T-s)$. Since $\phi (t)$ is smooth function with compact support in $(0, T)$, using Lemma \[lemma list\](i), Lemma \[L:2.6\] and , for every $x \in {\mathbb{R}}^d$ $$\begin{aligned}
\int^T_0 D^{1-\beta}_t \Phi(s) \Delta p(T-s,x) \,ds
&=&\int^T_0 D^{1-\beta}_t \Phi(s) \left(\partial^{\beta}_t p(\cdot, x)\right) (T-s) \,ds \nonumber\\
&=& \int^T_0 \phi (s ) \left( D^{1-\beta}_t \partial^{\beta}_t p(\cdot, x) \right) (s) \,ds \nonumber \\
&=&\int^T_0 \phi( s ) \frac{\partial}{\partial s} p(s,x) \,ds. \label{e:fnnew}\end{aligned}$$ Recall that $D^2_x q$, $D^2_x p$ and $\frac{\partial}{\partial s}p$ are locally integrable in ${\mathbb{R}}^d\setminus\{0\}$ uniformly on the support of $\phi(t)$. By the integration by parts in $x$, Lemma \[lemma list\](vi), Lemma \[L:2.6\], and the Fubini’s Theorem, $$\begin{aligned}
\int_{{\mathbb{R}}^d}\int^T_0 \phi(s) \psi (x) \Delta q (s,x) \,dsdx &=&\int^T_0\int_{{\mathbb{R}}^d} \phi(s) \Delta \psi (x) q (s,x) \,dxds\\
&=&\int_{{\mathbb{R}}^d} \Delta \psi(x)\left( \int_0^T \phi (s) \, D^{1-\beta}_t p(s,x) \,ds \right)dx\\
&=&\int_{{\mathbb{R}}^d} \Delta \psi (x) \left( \int^T_0 D^{1-\beta}_t \Phi (s) p(T-s,x) \,ds \right)dx\\
&=&\int_{{\mathbb{R}}^d} \psi (x) \left( \int^T_0 D^{1-\beta}_t \Phi(s) \Delta p(T-s,x) \,ds \right)dx \\
&=&\int_{{\mathbb{R}}^d}\int^T_0 \phi( s ) \psi (x) \frac{\partial}{\partial s} p(s,x) \,dsdx.\end{aligned}$$ Since $\phi(t)$ and $\psi(x)$ are arbitrary, the lemma is proved.
**Key Estimates**
=================
In this section, we first define a stochastic Banach space and establish key lemmas. Then we study the $L_2$-theory of a model equation for SPDEs with fractional time-derivatives.
For $n=0,1,2,...$, define the Banach spaces $$H^n_2:=H^n_2({\mathbb{R}}^d) =\left\{u: u, D_xu, \cdots, D_x^n u \in L_2\right\}.$$ In general, for $\sigma \in {\mathbb{R}}$ define the space $H^{\sigma}_2=H^{\sigma}_2({\mathbb{R}}^d)=(1-\Delta)^{-\sigma/2}L_2$ as the set of all distributions $u$ on ${\mathbb{R}}^d$ such that $(1-\Delta)^{\sigma/2}u\in L_2$. For $u\in H^\sigma_2$, we define $$\label{e:2.22}
\|u\|_{H^{\sigma}_2}:=\|(1-\Delta)^{\sigma/2}u\|_{L_2}
:=\|{\mathcal{F}}^{-1}[(1+|\xi|^2)^{\sigma/2}{\mathcal{F}}(u)(\xi)]\|_{L_2}.$$ Similarly for $\ell_2$-valued $g=(g^1, g^2, \dots)$, $$\label{e:2.22n}
\|g\|_{H^{\sigma}_2(\ell_2)}:=\||(1-\Delta)^{\sigma/2}g|_{\ell_2}\|_{L_2}
:=\||{\mathcal{F}}^{-1}[(1+|\xi|^2)^{\sigma/2}{\mathcal{F}}(g)(\xi)]|_{\ell_2}\|_{L_2}.$$
Let ${\mathcal{P}}$ be the predictable $\sigma$-field and ${\mathcal{P}}^{d{\mathbb{P}}\times dt}$ be the completion of ${\mathcal{P}}$ with respect to $d{\mathbb{P}}\times dt$. For each $\sigma \in {\mathbb{R}}$, define the Banach space $${\mathbb{H}}^{\sigma}_2(T):=L_2(\Omega\times [0,T],{\mathcal{P}}, H^{\sigma}_2).$$ That is, $u\in {\mathbb{H}}^{\sigma}_2(T)$ if $u$ is an $H^{\sigma}_2$-valued ${\mathcal{P}}^{d{\mathbb{P}}\times dt}$-measurable process defined on $\Omega \times [0,T]$ so that $$\|u\|_{{\mathbb{H}}^{\sigma}_2(T)}:= \left( \E\int^T_0\|u (t, \cdot )\|^2_{H^{\sigma}_2}\,dt\right)^{1/2}<\infty.$$ For an $\ell_2$-valued process $g=(g^1,g^2,...)$, we write $g\in {\mathbb{H}}^{\sigma}_2(T,\ell_2)$ if $g^k\in {\mathbb{H}}^{\sigma}_2(T)$ for every $k\geq 1$ and $$\|g\|_{{\mathbb{H}}^{\sigma}_2(T,\ell_2)}:=
\left( \E \int^T_0\|g\|^2_{H^{\sigma}_2(\ell_2)}\,dt \right)^{1/2}<\infty.$$ Denote ${\mathbb{L}}_2(T)={\mathbb{H}}^0_2(T)$ and ${\mathbb{L}}_2(T,\ell_2)={\mathbb{H}}^0_2(T,\ell_2)$. Write $g\in {\mathbb{H}}^{\infty}_0(T,\ell_2)$ if $g^k=0$ for all sufficiently large $k$, and each $g^k$ is of the type $$g^k=\sum_{i=1}^nI_{(\tau_{i-1},\tau_i]}(t) g^{ik}(x)$$ where $\tau_i$ are bounded stopping times with respect to ${\mathcal{F}}_t$ and $g^{ik} \in C^{\infty}_0 ({\mathbb{R}}^d)$. It is known ([@Kr99 Theorem 3.10]) that ${\mathbb{H}}^{\infty}_0(T,\ell_2)$ is dense in ${\mathbb{H}}^{\sigma}_2(T,\ell_2)$ for any $\sigma$.
Finally we use $U^{\sigma}_2$ to denote the family of $H^{\sigma}_2({\mathbb{R}}^d)$-valued ${\mathcal{F}}_0$-measurable random variables $u_0$ having $$\|u_0\|_{U^{\sigma}_2}:=\left( \E \|u_0\|^2_{H^{\sigma}_2}
\right)^{1/2}<\infty.$$
\[lemma 2\] Suppose $a>0$. [(i)]{} Let $h=(h^1,h^2,\cdots) \in L_2(\Omega\times [0,T],{\mathcal{P}}, \ell_2)$. Then the equality $$I^{a}(\sum_{k=1}^{\infty} \int^{\cdot}_0 h^k (s) dW^k_s)(t)=\sum_{k=1}^{\infty} (I^{a} \int^{\cdot}_0 h^k (s) dW^k_s)(t)$$ holds for all $t\leq T$ (a.s.) and also in $L_2(\Omega\times [0,T])$.
[(ii)]{} Suppose $
h_n
=(h^1_n,h^2_n,\cdots)$ converges to $h=(h^1,h^2,\cdots)$ in $L_2(\Omega\times [0,T],{\mathcal{P}}, \ell_2)$ as $n\to \infty$. Then, as $n\to \infty$, $$\label{eqn 8.30.2}
\sum_{k=1}^\infty I^{a}\int^{\cdot}_0 h^k_n dW^k_s \quad
\hbox{converges to} \quad \sum_{k=1}^\infty I^{a}\int^{\cdot}_0 h^k (s) dW^k_s$$ in probability uniformly on $[0,T]$.
Since the series $\sum_{k=1}^{\infty} \int^t_0 h^k (s) dW^k_t$ converges in $L_2(\Omega \times [0,T])$, by (\[eqn 7.03.1\]) we have $$\begin{aligned}
I^{a}\Big( \sum_{k=1}^{\infty} \int^{\cdot}_0 h^k (s) dW^k_s \Big)(t)&=&\lim_{n\to \infty} I^{a}\Big( \sum_{k=1}^{n} \int^{\cdot}_0 h^k (s) dW^k_s \Big)(t) \\
&=&\lim_{n\to \infty} \sum_{k=1}^{n} I^{a} \Big( \int^{\cdot}_0 h^k (s) dW^k_s \Big)(t) \quad \quad \text{in}\,\,\,L_2(\Omega \times [0,T]).\end{aligned}$$ Thus, the series $\sum_{k=1}^{\infty} I^{a} \int^t_0 h^k (s) dW^k_s$ converges in $L_2(\Omega\times [0,T])$ and $$\label{eqn 7.04.1}
I^{a}\sum_{k=1}^{\infty} \int^t_0 h^k (s) dW^k_s=\sum_{k=1}^{\infty} I^{a} \int^t_0 h^k (s) dW^k_s$$ in $L_2(\Omega\times [0,T])$, and thus the equality holds (a.e.). Also by Burkholder-Davis-Gundy inequality $$\begin{aligned}
&&{\mathbb{E}}\Big[ \sup_{t\leq T} \Big|\sum_{k=n}^{m} I^{a} \int^t_0 h^k (s) dW^k_s
\Big|^2 \Big] \nonumber\\
&\leq& N{\mathbb{E}}\Big[ \sup_{t\leq T} \Big|\sum_{k=n}^{m} \int^t_0 h^k (s) dW^k_s
\Big|^2 \Big] \leq N \sum_{k=n}^{m} {\mathbb{E}}\Big[ \int^T_0 |h^k(s)|^2 ds \Big]
\to 0 \label{eqn 8.29.1}\end{aligned}$$ as $n,m\to \infty$. Therefore the series $\sum_{k=1}^{\infty} I^{a} \int^t_0 h^k (s) dW^k_s$ converges in probability uniformly on $[0,T]$. It follows that both sides of (\[eqn 7.04.1\]) are continuous, and therefore the equality above holds for all $t\leq T$ (a.s.).
\[remark 1\] Let $\sigma\in {\mathbb{R}}$. Suppose $g_n \to g$ in ${\mathbb{H}}^{\sigma}_2(T,\ell_2)$ as $n\to \infty$, and $\phi\in C^{\infty}_0 ({\mathbb{R}}^d)$. Then $(g_n (t, \cdot),\phi)_{L_2} \to (g(t, \cdot),\phi)_{L_2}$ in $L_2(\Omega\times [0,T],{\mathcal{P}}, \ell_2)$, and therefore Lemma \[lemma 2\](ii) holds with $h_n(t):=(g_n(t, \cdot),\phi)_{L_2}$ and $h(t):=(g(t, \cdot),\phi)_{L_2}$.
\[derivative\] Let $\alpha>1/2$ and $g\in {\mathbb{H}}^{\infty}_0(T,\ell_2)$. Then $I^{\alpha} \sum_{k=1}^\infty\int^{\cdot}_0 g^k (s) dW^k_s$ is differentiable in $t$ and (a.s.) for all $t\leq T$ $$\frac{\partial}{ \partial t}(I^{\alpha} \sum_{k=1}^\infty \int^{\cdot}_0 g^k (s) dW^k_s)(t)=\frac{1}{\Gamma(\alpha)}\sum_{k=1}^\infty \int^t_0 (t-s)^{\alpha-1}g^k(s)dW^k_s.$$
We integrate the right hand side and then use the stochastic Fubini’s theorem (see e.g. [@Kr11]) to get $$\begin{aligned}
\frac{1}{\Gamma(\alpha)} \sum_{k=1}^\infty\int^t_0\int^s_0 (s-r)^{\alpha-1}g^k(r)dW^k_r ds &=&\frac{1}{\Gamma(\alpha)}\sum_{k=1}^\infty \int^t_0\int^t_r (s-r)^{\alpha-1} ds g^k(r)dW^k_r\\
&=&\frac{1}{\alpha \Gamma(\alpha)} \sum_{k=1}^\infty\int^t_0 (t-r)^{\alpha} g^k(r)dW^k_r.\end{aligned}$$ Similarly, by Lemma \[lemma 2\](i) $$\begin{aligned}
I^{\alpha} \sum_{k=1}^\infty \Big(\int^{\cdot}_0 g^k (s) dW^k_s \Big) (t)&=&\frac{1}{\Gamma(\alpha)}\sum_{k=1}^\infty\int^t_0 (t-s)^{\alpha-1}\left(\int^s_0 g^k(r)dW^k_r\right) ds\\
&=&\frac{1}{\alpha \Gamma(\alpha)} \sum_{k=1}^\infty\int^t_0 (t-r)^{\alpha} g^k(r)dW^k_r.\end{aligned}$$ The lemma is proved.
Lemma \[derivative\] can be easily extended for any $g\in {\mathbb{L}}_2(T,\ell_2)$.
For the remainder of this paper, we assume that holds. For $a\in {\mathbb{R}}$, denote $a_+=\max\{a,0\}$. Define $$\label{gamma0}
\gamma_0:=\frac{(2\gamma-1)_+}{\beta}<2.$$ Note that since $\gamma<\beta+1/2$ we have $$\label{gamma1}
\gamma_0<2, \quad \text{and}\quad \gamma_0=0 \quad \text{if}\,\, \gamma\leq 1/2.$$
\[def 3.23\] We write $u\in {\mathcal{H}}^{\sigma+2}_2(T)$ if $u\in {\mathbb{H}}^{\sigma+2}_2(T)$, $u(0)\in U^{\sigma+1}_2$, and for some $f\in {\mathbb{H}}^{\sigma}_2(T)$ and $g\in {\mathbb{H}}^{\sigma+\gamma_0}_2(T,\ell_2)$ it holds that $$\label{eqn 7.08.1}
\partial^{\beta}_tu (t,x)=f (t,x)
+\sum_{k=1}^{\infty} \partial^{\gamma}_t \int^t_0 g^k (s,x)\, dW^k_s$$ in the sense of distributions. That is, for any $\phi\in C^{\infty}_0 ({\mathbb{R}}^d)$ the equality $$\label{sense of solution}
(I^{1-\beta}(u-u(0))(t), \phi)_{L_2}=\int^t_0(f(s, \cdot), \phi)_{L_2}ds+\sum_{k=1}^{\infty} (I^{1-\gamma} \int^{\cdot}_0 (g^k(s, \cdot),\phi)_{L_2}\, dW^k_s)(t)$$ holds for all $t\leq T$ (a.s.). In this case we write $$f={\mathbb{D}}u, \quad g={\mathbb{S}}u$$ and define $$\label{e:3.10}
\|u\|_{{\mathcal{H}}^{\sigma+2}_2(T)}=\|u(0)\|_{U^{\sigma+1}_2}+\|u\|_{{\mathbb{H}}^{\sigma+2}_2(T)}+\|{\mathbb{D}}u\|_{{\mathbb{H}}^{\sigma}_2(T)}+\|{\mathbb{S}}u\|_{{\mathbb{H}}^{\sigma+\gamma_0}_2(T,\ell_2)}.$$ Finally define $$\begin{aligned}
\label{e:ch0}
{\mathcal{H}}^{\sigma+2}_{2,0}(T)={\mathcal{H}}^{\sigma+2}_2(T) \cap \{u: u(0)=0\}.
\end{aligned}$$
By (\[e:1.10\]), (\[2014.1.23.2\]) and Lemma \[derivative\], (\[sense of solution\]) is equivalent to $$\begin{aligned}
(u(t, \cdot)-u(0, \cdot), \phi)_{L_2}&=&\frac{1}{\Gamma(\beta)}\int^t_0(t-s)^{\beta-1}(f(s, \cdot), \phi)_{L_2}ds\\
&&+\frac{1}{\Gamma(1+\beta-\gamma)}\sum_{k=1}^{\infty} \int^t_0 (t-s)^{\beta-\gamma} (g^k(s, \cdot),\phi)_{L_2}\, dW^k_s.
\end{aligned}$$
\[Banach\] [(i)]{} ${\mathcal{H}}^{\sigma+2}_2(T)$ is a Banach space.
[(ii)]{} Let $u\in {\mathcal{H}}^{\sigma+2}_2(T)$. Then $u$ is a continuous $H^{\sigma}_2$-valued process.
[(iii)]{} Assume that $u\in {\mathcal{H}}^{2}_2(T)$ and (\[eqn 7.08.1\]) holds. Then (a.s.) $$\label{eqn 8.27.1}
(k_{1-\beta}*\|u-u(0)-v\|^2_{L_2})(t)\leq 2\int^t_0 (f(s, \cdot),u(s, \cdot)-u(0, \cdot)-v(s, \cdot))_{L_2}ds \quad \hbox{for } t\leq T ,$$ where $$\begin{aligned}
\label{e:defv}
v(t,x)=\Gamma(1+\beta-\gamma)^{-1}\sum_{k=1}^\infty \int^t_0 (t-s)^{\beta-\gamma} g^k(s,x) dW^k_s.\end{aligned}$$
\(i) We only need to prove the completeness of the space. Let $u_n$, $n=1,2,\cdots$, be a Cauchy sequence in ${\mathcal{H}}^{\sigma+2}_2(T)$ with $$f_n={\mathbb{D}}u_n, \quad g_n={\mathbb{S}}u_n.$$ Then there exist $u, f, g, u_0$ so that $u_n, f_n, g_n, u_n(0)$ converge to $u,f,g,u_0$ respectively in their corresponding spaces. To prove $u_n \to u$ in ${\mathcal{H}}^{\sigma+2}_2(T)$, it suffices to show (\[sense of solution\]) holds for all $t\leq T$ (a.s.). Since the series $\sum_{k=1}^\infty \int^t_0 (g^k_n (s),\phi) dW^k_s$ converges in probability uniformly on $[0,T]$, so does $(I^{1-\gamma} \sum_{k=1}^\infty \int^\cdot_0 (g^k_n(s),\phi) dW^k_s)(t)$. By Remark \[remark 1\], considering the limit of $$(I^{1-\beta}(u_n-u_n(0)))(t), \phi)_{L_2}=\int^t_0(f_n(s, \cdot), \phi)_{L_2}ds+\sum_{k=1}^{\infty} (I^{1-\gamma} \int^{\cdot}_0 (g^k_n(s, \cdot),\phi)_{L_2}\, dW^k_s)(t)$$ for $t\leq T$, we get (\[sense of solution\]) for all $t\leq T$ (a.s.) since both sides of (\[sense of solution\]) are continuous in $t$.
\(ii) We only prove the case $\sigma=0$. The general case is covered by applying $(1-\Delta)^{\sigma/2}$ to (\[eqn 7.08.1\]). Denote $f={\mathbb{D}}u$ and $g={\mathbb{S}}u$. Notice that as an $L_2({\mathbb{R}}^d)$-valued process, $u(t)-u(0)$ satisfies $$k_{1-\beta}*(u(\cdot,x)-u(0,x))(t)=\int^t_0 f(s,x) ds+(k_{1-\gamma}* (\sum_{k=1}^\infty \int^{\cdot}_0 g^k(s,x)dW^k_s))(t) \quad \hbox{all } t\leq T \, (a.s.).$$ Taking the convolution with $k_{\beta}$ and using $$\frac{\partial}{ \partial t}(k_{\beta}*\int^{\cdot}_0 f(s,x)ds)(t)=(k_{\beta}*f)(t,x),$$ $$\frac{\partial}{ \partial t}(k_{\beta}*k_{1-\gamma}*(\sum_{k=1}^\infty \int^{\cdot}_0 g^k (s,x)dw_s)) (t)=\sum_{k=1}^\infty \frac{1}{\Gamma(1+\beta-\gamma)}\int^{t}_0 (t-s)^{\beta-\gamma}g^k(s,x)dW^k_s,$$ where the second equality is from Lemma \[derivative\], we get for all $t\leq T$ (a.s.) $$\label{eqn 10}
u(t,x)-u(0,x)=(k_{\beta}*f)(t,x)+\sum_{k=1}^\infty \frac{1}{\Gamma(1+\beta-\gamma)}\int^{t}_0 (t-s)^{\beta-\gamma}g^k(s,x)dW^k_s.$$ Hence the claim follows.
\(iii) Denote $w(t,x)=u(t,x)-u(0,x)-v(t,x)$. Then we have $\partial^{\beta}_tw(t,x)=f(t,x)$. Now we use the fact (see e.g. [@Za Lemma 2.1]) that if $\kappa$ is a positive decreasing function on $[0,T]$ then $$\kappa*\|w\|^2_{L_2} (t)\leq 2\int^t_0 (\frac{\partial}{ \partial s}(\kappa*w)(s, \cdot),w(s, \cdot))_{L_2}ds.$$ We take (see [@Za]) a sequence of such functions $\kappa_n\in H^1_1([0,T])$ so that $\kappa_n \to k_{1-\beta}$ in $L_1([0,T]$ and $\frac{\partial}{ \partial t}(\kappa_n*w)\to \frac{\partial}{ \partial t}(k_{1-\beta}*w)$ in $L_2([0,T],L_2)$. Finally for (\[eqn 8.27.1\]) it is enough to note that $$\kappa_n *\|w\|^2_{L_2} \to k_{1-\beta}*\|w\|^2_{L_2} \quad \text{in} \quad L_1([0,T])$$ and both sides of (\[eqn 8.27.1\]) are continuous in $t$. The lemma is proved.
Recall that for any $\sigma$, $$\|u\|^2_{{\mathbb{H}}^{\sigma}_2(t)}:={\mathbb{E}}\int^t_0\|u(s)\|^2_{H^{\sigma}_2}ds.$$
\[lemm 8.28.6\] Let $u\in {\mathcal{H}}^{\sigma+2}_2(T)$. Then for any $t\leq T$, $$\label{eqn 1.27.1}
(k_{1-\beta}*{\mathbb{E}}\|u\|^2_{H^{\sigma}_2})(t)\leq N ({\mathbb{E}}\|u(0)\|^2_{H^{\sigma}_2}+\|{\mathbb{D}}u\|^2_{{\mathbb{H}}^{\sigma}_2(t)}+\|{\mathbb{S}}u\|^2_{{\mathbb{H}}^{\sigma}_2(t,\ell_2)}+\|u\|^2_{{\mathbb{H}}^{\sigma}_2(t)})
\leq N \|u\|^2_{{\mathcal{H}}^{\sigma+2}_2(t)},$$ where $N$ depends only on $T, \beta$ and $\gamma$. In particular, for any $t\leq T$, $$\label{eqn 1.27.2}
\|u\|^2_{{\mathbb{H}}^{\sigma}_2(t)}\leq N \int^t_0 (t-s)^{-1+\beta}\|u\|^2_{{\mathcal{H}}^{\sigma+2}_2(s)}\, ds.$$
We only consider the case $\sigma=0$. In general, one can consider $\Delta^{\sigma/2}u$ in place of $u$. Denote $v$ as in Lemma \[Banach\]. Then by (\[eqn 8.27.1\]), $$(k_{1-\beta}*{\mathbb{E}}\|u\|^2_{L_2})(t)\leq 2 (k_{1-\beta}*{\mathbb{E}}\|u(0)+v\|^2_{L_2})(t)+2{\mathbb{E}}\int^t_0(f(s, \cdot),u(s, \cdot)-u(0, \cdot)-v(s, \cdot))_{L_2} ds.$$ Note that by Fubini’s theorem and Davis’s inequality $$\begin{aligned}
\nonumber
&&{\mathbb{E}}\int^t_0\|v(s)\|^2_{L_2}ds = \int^{t}_0 {\mathbb{E}}\|v(s)\|^2_{L_2}ds\\
&\leq& N {\mathbb{E}}\int^t_0 \int^s_0 (s-r)^{2(\beta-\gamma)}
\|g(r, \cdot)\|^2_{L_2(\ell_2)} drds \leq N\|g\|^2_{{\mathbb{L}}_2(t,\ell_2)}. \label{eqn 8.28.5}\end{aligned}$$ For the last inequality we use the fact $2(\beta-\gamma)>-1$. Therefore, by young’s inequality $${\mathbb{E}}\int^t_0(f(s, \cdot),u(s, \cdot)-u(0, \cdot)-v(s, \cdot))_{L_2} ds\leq N({\mathbb{E}}\|u(0)\|^2_{L_2}+\|f\|^2_{{\mathbb{L}}_2(t)}+\|g\|^2_{{\mathbb{L}}_2(t,\ell_2)}+\|u\|^2_{{\mathbb{L}}_2(t)}).$$ Also, $$(k_{1-\beta}*{\mathbb{E}}\|u(0)+v\|^2_{L_2})(t)\leq N {\mathbb{E}}\|u(0)\|^2_{L_2}+ N k_{1-\beta}*{\mathbb{E}}\|v\|^2_{L_2}(t),$$ $$k_{1-\beta}*{\mathbb{E}}\|v\|^2_{L_2}(t)\leq N \int^t_0 (t-s)^{-\beta}\int^s_0 (s-r)^{2(\beta-\gamma)}{\mathbb{E}}\|g(r, \cdot)\|^2_{L_2(\ell_2)} drds\leq N\|g\|^2_{{\mathbb{L}}_2(t,\ell_2)}.$$ This proves .
Note that the second equality in just follows from the definition of $\|u\|^2_{{\mathcal{H}}^{\sigma+2}_2(t)}$. Hence to prove (\[eqn 1.27.2\]) it is enough to consider a convolution with $k_{\beta}$ and use , which implies $$(k_{\beta}*k_{1-\beta}*{\mathbb{E}}\|u\|^2_{L_2})(t)=\int^t_0 {\mathbb{E}}\|u\|^2_{L_2}ds=\|u\|^2_{{\mathbb{L}}_2(t)}.$$ Hence the theorem is now proved.
Define $$P_{\beta,\gamma}(t,x)=\begin{cases}I^{\beta-\gamma}p (\cdot,x) (t)\quad &\text{if} \quad\beta\geq \gamma\\
\partial^{\gamma-\beta}_t p (\cdot,x) (t) \quad &\text{if}\quad \beta<\gamma.
\end{cases}$$
\[lemma main estimate\] Let $g\in {\mathbb{H}}^{\infty}_0(T,\ell_2)$ and $u$ be defined by $$u(t,x)=\sum_{k=1}^{\infty}\int^t_0 \int_{{\mathbb{R}}^d}P_{\beta,\gamma}(t-s, x-y)g^k(s,y)dy dW^k_s.$$ Let $\sigma \leq 2\wedge (\frac{1-2\gamma}{\beta} +2)$ if $\gamma \neq 1/2$, and $\sigma<2$ if $\gamma=1/2$. Then it holds that $${\mathbb{E}}\int^T_0 \|\Delta^{\sigma/2}u(t, \cdot)\|^2_{L_2} dt\leq N
\|g\|^2_{{\mathbb{L}}_2(T, \ell_2)}.$$ In general, for any $\gamma_1 \in {\mathbb{R}}$, $${\mathbb{E}}\int^T_0 \|u(t, \cdot)\|^2_{H^{\gamma_1+\sigma}_2} dt\leq N
\|g\|^2_{{\mathbb{H}}^{\gamma_1}_2(T, \ell_2)}.$$
Let $a:=\gamma-\beta<1/2$. Recall that $E_{\alpha,\gamma}(t)$ is bounded on $(-\infty,0]$. By the Fourier transform and Lemma \[l:DaE\], we have for any $\sigma\geq 0$ $$\begin{aligned}
&&{\mathbb{E}}\int^T_0\|\Delta^{\sigma/2}u(t, \cdot)\|^2_{L_2} dt\\ &\leq&N \sum_{k=1}^\infty {\mathbb{E}}\int^T_0 \int^t_0\int_{{\mathbb{R}}^d} |\xi|^{2\sigma}(t-s)^{-2a}E^2_{\beta, 1-a}(-|\xi|^2(t-s)^{\beta})|\hat{g}^k(s, \xi)|^2 d\xi ds\,dt\\
&\leq&N \|g\|^2_{{\mathbb{L}}_2(T, \ell_2)}\\
&&+N {\mathbb{E}}\int^T_0 \int^T_s\int_{|\xi|\geq 1} |\xi|^{2\sigma}(t-s)^{-2a}E^2_{\beta, 1-a}(-|\xi|^2(t-s)^{\beta})
|\hat{g}(s, \xi)|^2_{\ell_2} d\xi \,dt\,ds.\end{aligned}$$ By the substitution $r=|\xi|^{2/\beta} (t-s)$, the last term above is bounded by constant times of $${\mathbb{E}}\int_{|\xi|\geq 1}\int^T_0 |\hat{g}(s,\xi)|_{\ell_2}^2\int^{T|\xi|^{\frac{2}{\beta}}}_0 |\xi|^{2(\sigma+(2a-1)/{\beta})} r^{-2a}E^2_{\beta, 1-a}(-r^{\beta}) dr ds\,d\xi.$$ Let $\gamma>1/2$. Then, since $E_{\beta, 1-a}(-r^{\beta})$ is bounded on $[0,\infty)$ and $E_{\beta, 1-a}(-r^{\beta}) \le N r^{-\beta}$ if $r \geq 1$ (see (\[mittag\])), we have $$\begin{aligned}
&&\int^{T|\xi|^{\frac{2}{\beta}}}_0 |\xi|^{2(\sigma+(2a-1)/{\beta})} r^{-2a}E^2_{\beta, 1-a}(-r^{\beta}) dr\\
&\leq& \int^{\infty}_0 r^{-2a} E^2_{\beta, 1-a}(-r^{\beta}) dr\\
&\le & N \left(\int_0^1 r^{-2a} dr+ \int_1^\infty r^{-2 \gamma} dr \right)<\infty.\end{aligned}$$ If $\gamma=1/2$, then since $|\xi|\geq 1$ and $\sigma<2$, $$\begin{aligned}
&&\int^{T|\xi|^{\frac{2}{\beta}}}_0|\xi|^{2(\sigma+(2a-1)/{\beta})} r^{-2a}E^2_{\beta, 1-a}(-r^{\beta}) dr\\
&=&\int^{T|\xi|^{\frac{2}{\beta}}}_0|\xi|^{-2(2-\sigma)} r^{-2a}E^2_{\beta, 1-a}(-r^{\beta}) dr\\
&\leq& \int^1_0r^{-2a} E^2_{\beta, 1-a}(-r^{\beta})dr+N
|\xi|^{-2(2-\sigma)} \int^{T|\xi|^{\frac{2}{\beta}}}_1 r^{-1} dr \\
&\leq&N \int_0^1 r^{-2a} dr+N|\xi|^{-2(2-\sigma)} \ln|\xi|\leq N<\infty.\end{aligned}$$ The case $\gamma<1/2$ is treated similarly using $\sigma\leq 2$. Indeed, $$\begin{aligned}
&&\int^{T|\xi|^{\frac{2}{\beta}}}_0|\xi|^{2(\sigma+(2a-1)/{\beta})} r^{-2a}E^2_{\beta, 1-a}(-r^{\beta}) dr\\
&\leq& \int^1_0 r^{-2a}E^2_{\beta, 1-a}(-r^{\beta})dr+N
|\xi|^{2(\sigma+(2a-1)/{\beta})} \int^{T|\xi|^{\frac{2}{\beta}}}_1 r^{-2\beta} dr\\
&\leq&N\int_0^1 r^{-2a} dr+N |\xi|^{2\sigma-4}\leq N<\infty.\end{aligned}$$ Therefore the lemma is proved.
Lemma \[lemma main estimate\] says that $u$ (which is a solution of (\[time-space\]) below) is smoother than $g$ by order $2\wedge ((1-2\gamma)\beta^{-1}+2)$ if $\gamma\neq 1/2$ and $2-\varepsilon$ if $\gamma=1/2$, where $\varepsilon>0$. Thus, for example, to estimate the second derivative of solution $u$ we need to assume $$\|g\|_{{\mathbb{H}}^{\gamma_0}_2(T,\ell_2)}<\infty \quad \text{if}\,\, \gamma\neq 1/2, \quad \text{and}\quad \|g\|_{{\mathbb{H}}^{\varepsilon}_2(T,\ell_2)}<\infty \quad \text{if}\,\, \gamma=1/2.$$ Recall $\gamma_0=(2\gamma-1)_+/{\beta}<2$, which is defined in (\[gamma0\]).
We first consider the equation $$\label{time-space}
\partial^{\beta}_tu (t,x)=\Delta u(t,x)+f(t,x)
+\sum_{k=1}^{\infty} \partial^{\gamma}_t \int^t_0 g^k(s,x)\, dW^k_s.$$ Note that by letting $\beta\to 1$ and $\gamma \to 1$ we get the classical stochastic partial differentia equations.
\[lemma formula\] Let $f\in {\mathbb{L}}_2(T)$, $g\in {\mathbb{H}}^{\infty}_0(T,\ell_2)$ and $u\in {\mathbb{H}}^2_2(T)$. Then $u$ satisfies (\[time-space\]) with initial data $u_0\in U^1_2$ in the sense distributions (see Definition \[def 3.23\]) if and only if $$\begin{aligned}
u(t,x)&=&\int_{{\mathbb{R}}^d}p(t,x-y)u_0(y)dy+\int^t_0\int_{{\mathbb{R}}^d} q(t-s,x-y)f(s,y)dyds \nonumber\\
&+&\sum_{k=1}^\infty \int^t_0 \int_{{\mathbb{R}}^d}P_{\beta,\gamma}(t-s,x-y)g^k(s,y)dydW^k_s. \label{right term}\end{aligned}$$
Suppose $u$ satisfies (\[time-space\]). Recall that for the solution of the (deterministic) equation $$\partial^{\beta}_t\bar{u}=\Delta \bar{u}+f, \quad u(0)=u_0$$ is given by the formula $$\label{eqn deterministic}
\bar{u}(t,x):=\int_{{\mathbb{R}}^d}p(t,x-y)u_0(y)dy+ \int^t_0\int_{{\mathbb{R}}^d} q(t-s,x-y)f(s,y)dyds.$$ In fact, in [@EIK Section 5.2] the representation (\[eqn deterministic\]) is proved for sufficiently smooth $f$. This and the estimate of the solution obtained in [@Za Theorem 3.1] allow us to use an approximation (see the proof of Theorem \[thm laplace\]) and get (\[eqn deterministic\]) for general $f\in {\mathbb{L}}_2(T)$.
Thus by considering $u-\bar{u}$, where $\bar{u}$ is defined above, we may assume without loss of generality that $u_0=0$ and $f=0$. Suppose first $\beta \leq \gamma$. Set $a=\gamma-\beta$, $$v(t,x):=\sum_{k=1}^\infty\int^t_0 g^k(s,y)dW^k_s,$$ and $$w(t,x):=D^{a}_tv(t,x)=\frac{1}{\Gamma(1-a)}\sum_{k=1}^\infty\int^t_0(t-s)^{-a}g^k(s,x)dW^k_s.$$ Then $u-w$ satisfies the following fractional diffusion equation $$\partial^{\beta}_t(u-w)=\Delta u=\Delta (u-w)+ \Delta w, \quad \quad (u-v)(0)=0.$$ Thus by (\[eqn deterministic\]) with $\Delta w$ in place of $f$, we have $$u(t,x)=
w(t, x)
+\int^t_0\int_{{\mathbb{R}}^d}q(t-s,x-y)\Delta w(s,y)dyds.$$ Nota that for any $s<t$ $$\begin{aligned}
\int_{{\mathbb{R}}^d}q(t-s,x-y)\Delta w(s,y)dy
&=&\int_{{\mathbb{R}}^d} \Delta q(t-s,x-y) D_t^a v
(s,y) dy\\&=&\int_{{\mathbb{R}}^d}\frac{\partial}{\partial t}p(t-s,x-y)D^a_tv(s,y)dy\\
&=&\frac{\partial}{\partial t}\int_{{\mathbb{R}}^d}p(t-s,x-y)D^a_tv(s,y)dy.\end{aligned}$$ Indeed, the first equality is from Lemma \[lemma list\](iv) and the integration by parts, the second equality is from Lemma \[L:2.3\] and the third equality is from Lemma \[lemma list\](v). Therefore $u(t,x)$ is equal to $$\begin{aligned}
&&D_t^a v(t,x)+ \int^t_0\frac{\partial}{\partial t}\int_{{\mathbb{R}}^d}
p(t-s,x-y) D_t^a v (s,y) dyds\\
&=&\frac{\partial}{\partial t}\int^t_0\int_{{\mathbb{R}}^d} p(t-s,x-y) D_t^a v
(s,y) dyds\\
&=&\frac{1}{\Gamma(1-a)}\sum_{k=1}^\infty\frac{\partial}{\partial t}\int^{t}_0\int_{{\mathbb{R}}^d} p(t-s,x-y)
\int^s_0(s-r)^{-a}g^k(r,y)dW^k_r dyds\\
&=&\frac{1}{\Gamma(1-a)}\sum_{k=1}^\infty\frac{\partial}{\partial t}\int^{t}_0\int_{{\mathbb{R}}^d} p(s,x-y)
\int^{t-s}_0(t-s-r)^{-a}g^k(r,y)dW^k_r dyds.\end{aligned}$$
Hence it is enough to prove $$\begin{aligned}
\nonumber
&&\frac{1}{\Gamma(1-a)}\sum_{k=1}^\infty \int^{t}_0\int_{{\mathbb{R}}^d} p(s,x-y) \int^{t-s}_0(t-s-r)^{-a}g^k(r,y)dW^k_r
dyds\\
& =&\sum_{k=1}^\infty\int^t_0 \int^s_0 \int_{{\mathbb{R}}^d} D_t^ap (s-r,x-y)g^k(r,y)dy dW^k_r ds.
\label{imp}\end{aligned}$$ The latter equals $$\begin{aligned}
&&\sum_{k=1}^\infty\int^t_0 \int_{{\mathbb{R}}^d}\int^{t-r}_0 D_t^ap (s,x-y) ds g^k(r,y)dy dW^k_r \nonumber\\
&=&\sum_{k=1}^\infty\int^t_0 \int_{{\mathbb{R}}^d}I^{1-a}p (t-r,x-y) g^k(r,y)dy dW^k_r \label{new 10}\\
&=&\frac{1}{\Gamma(1-a)}\sum_{k=1}^\infty\int^t_0\int_{{\mathbb{R}}^d}\int^{t-r}_0(t-r-s)^{-a}p(s,x-y)ds g^k(r,y) dy dW^k_r.
\nonumber\end{aligned}$$ For (\[new 10\]) above we used the fact that $\int^t_0 D_t^a p ds=I^{1-a}p$. We thus get using the stochastic Fubini’s theorem (see [@Kr11]).
We now consider the case $\beta\geq \gamma$. Put $a=\beta-\gamma$ and define $$v(t,x)=\int^t_0 g^k(s,x)dW^k_s, \quad w(t,x)=I^{a}v(t,x)=\frac{1}{a\Gamma(a)}\int^{t}_0
(t-s)^a g^k(s,x)dW^k_s.$$ From this point on it is enough to repeat the case $\beta\leq \gamma$. Indeed, following the previous steps, we get $$u(t,x)=\frac{1}{a\Gamma(a)}\sum_{k=1}^\infty\frac{\partial}{\partial t}\int^{t}_0\int_{{\mathbb{R}}^d} p(s,x-y)
\int^{t-s}_0(t-s-r)^{a}g^k(r,y)dW^k_r dyds.$$ Note that $$\begin{aligned}
&&\sum_{k=1}^\infty\int^t_0 \int^s_0 \int_{{\mathbb{R}}^d} I^ap (s-r,x-y)g^k(r,y)dy dW^k_r ds\\
&=&\sum_{k=1}^\infty\int^t_0 \int^{t-r}_0 \int_{{\mathbb{R}}^d} I^ap (s,x-y)g^k(r,y)dy dsdW^k_r\\
&=&\frac{1}{a\Gamma(a)}\sum_{k=1}^\infty\int^t_0 \int^{t-r}_0 \int_{{\mathbb{R}}^d} (t-s-r)^ap (r,x-y)g^k(r,y)dy
dsdW^k_r.\end{aligned}$$ This clearly proves the case $\gamma\geq \beta$.
On the other hand, going backward of the above equalities one easily finds that if $u$ is given as in (\[right term\]), then it satisfies (\[time-space\]). The proof of the lemma is now complete.
Recall $\gamma_0=(2\gamma-1)_+/{\beta}<2$. Fix $\varepsilon_0\in (0,1)$ and define $$\label{e:3.23}
\sigma_0:=\gamma_0+ (\varepsilon_0 1_{\gamma=1/2})=
\begin{cases}
(2\gamma-1) /\beta \quad &\hbox{if } \gamma >1/2\\
\varepsilon_0 &\hbox{if } \gamma =1/2\\
0 &\hbox{if } \gamma <1/2.
\end{cases}$$
\[thm laplace\] For any $f\in {\mathbb{H}}^{\sigma}_2(T)$, $g\in {\mathbb{H}}^{\sigma+\sigma_0}_2(T,\ell_2)$ and $u_0\in U^{\sigma+1}_2$, equation (\[time-space\]) has a unique solution $u\in {\mathcal{H}}^{\sigma+2}_2(T)$, and for this solution we have $$\label{main estimate1}
\|u\|_{{\mathbb{H}}^{\sigma+2}_2(T)}\leq N\left(\|u_0\|_{U^{\sigma+1}_2}+\|f\|_{{\mathbb{H}}^{\sigma}_2(T)}+\|g\|_{{\mathbb{H}}^{\sigma+\sigma_0}_2(T,\ell_2)}\right),$$ where $N$ depends only on $d$ and $T$.
Without loss of generality we only need to prove the case $\sigma=0$.
For the deterministic equation, the theorem including the estimate is proved in [@Za]. Thus the uniqueness for equation (\[time-space\]) easily follows.
Define $w$ as in (\[eqn deterministic\]). Then by considering $u-w$, we may assume without loss of generality that $u_0=0$ and $f=0$.
First, assume $g\in {\mathbb{H}}^{\infty}_0(T,\ell_2)$. Then by Lemmas \[lemma main estimate\] and \[lemma formula\], equation (\[time-space\]) has a unique solution $u\in {\mathbb{H}}^2_2(T)$ and estimate holds.
For general $g\in {\mathbb{H}}^{\sigma_0}_2(T,\ell_2)$, take a sequence of $g_n\in {\mathbb{H}}^{\infty}_0(T,\ell_2)$ so that $g_n \to g$ in ${\mathbb{H}}^{\sigma_0}_2(T,\ell_2)$. Define $u_n$ as the solution of equation (\[time-space\]) with $g_n$ in place of $g$, that is, $$\label{approx}
(I^{1-\beta}u_n)(t)=\int^t_0\Delta u_n (s)dt+\sum_{k=1}^{\infty} (I^{1-\gamma} \int^{\cdot}_0 g^k_n(s)\, dW^k_s)(t).$$ Then by Lemmas \[lemma main estimate\] and \[lemma formula\] $$\label{eqn 7.03.2}
\|u_n\|_{{\mathbb{H}}^2_2(T)}\leq N \|g_n\|_{{\mathbb{H}}^{\sigma_0}_2(T,\ell_2)},$$ $$\|u_n-u_m\|_{{\mathbb{H}}^2_2(T)}\leq N \|g_n-g_m\|_{{\mathbb{H}}^{\sigma_0}_2(T,\ell_2)}.$$ Thus $u_n\to u$ in ${\mathbb{H}}^2_2(T)$ for some $u$. Letting $n\to \infty$ in (\[approx\]) and using Remark \[remark 1\], we see that $u$ is a solution of (\[time-space\]). Also we easily get (\[main estimate1\]) from (\[eqn 7.03.2\]). The theorem is proved.
The following lemma is taken from [@YGD Corllary 2].
(Gronwall’s lemma) \[gronwall\] Suppose $b>0$ and $a(t)$ is a nonnegative nondecreasing locally integrable function on $[0,T)$, and suppose $\eta(t)$ in nonnegative locally integrable on $[0,T)$ with $$\eta(t)\leq a(t)+b\int^t_0 (t-s)^{\beta-1}\eta(s)ds, \quad \forall t<T.$$ Then it holds that $$\eta(t)\leq a(t)E_{\beta}(b \Gamma(\beta)t^{\beta}).$$
**SPDE of divergence form type**
================================
In this section, we study the equation of divergence type $$\begin{aligned}
\nonumber
\partial^{\beta}_t u&=&D_i\left[
a^{ij}u_{x^j}+b^iu+f^i(u)\right]+cu+h(u)
\\
&&+ \sum_{k=1}^\infty\partial^{\gamma}_t \int^t_0 (\sigma^{ijk}u_{x^ix^j}+\mu^{ik} u_{x^i}
+\nu^ku +g^k(u))\, dW^k_s
\label{eqn 7.04.5}
\end{aligned}$$ where the coefficients $a^{ij}, b^i, c, \sigma^{ijk}, \mu^{ik}, \nu^k$ are functions depending on $(\omega,t,x)$ and the functions $f^i, h, g^k$ depend on $(\omega,t,x)$ and the unknown $u$.
For a $\ell_2$-valued continuous function $v$ in ${\mathbb{R}}^d$, we define the space $C^{\alpha}(\ell_2)$, $\alpha\in [0,1]$ by the norm $$|v|_{C^{\alpha}(\ell_2)}=\sup_x |v|_{\ell_2}+ \sup_{x\neq y}\frac{|v(x)-v(y)|_{\ell_2}}{|x-y|^{\alpha}}.$$
\[ass 1\] (i) The coefficients $a^{ij}, b^i, c, \sigma^{ijk}, \mu^{ik},
\nu^{ik}$ are $ {\mathcal{P}}\otimes {\mathcal{B}}([0, T] \times {\mathbb{R}}^d)$-measurable.
\(ii) There exist constants $\delta, K_1>0$ so that for any $\xi\in
{\mathbb{R}}^d$ $$\delta |\xi|^2 \leq a^{ij}\xi_i\xi_j \leq K_1|\xi|^2 \quad \forall i,j,\omega,t,x.$$ $$\begin{aligned}
\label{e:aii}
|b^i|+|c|+|\sigma^{ij}|_{\ell_2}+|\mu^i|_{\ell_2}+|\nu|_{\ell_2}
\leq K_1 \quad \forall i,j,\omega,t,x.\end{aligned}$$
\(iii) $\sigma^{ijk}=0$ if $\gamma\geq 1/2$, and $\mu^{ik}=0$ if $\gamma \geq 1/2+ \beta/2$ for every $i,j, k, \omega,t,x$.
Recall that $\sigma_0$ is the constant defined in .
\[ass 2\] (i) There exist constant $\kappa, K_2>0$, $$|\sigma^{ij}(t,\cdot)|_{C^1(\ell_2)}+|\mu^i(t,\cdot)|_{C^{|\sigma_0-1|+\kappa}(\ell_2)}+
|\nu(t,\cdot)|_{C^{|\sigma_0-1|+\kappa}(\ell_2)}\leq K_2, \quad \quad \forall i,j,\omega,t.$$
\(ii) For any $\varepsilon>0$ there exists $K_3=K_3(\varepsilon, T)$ so that $$\begin{aligned}
&\|f^i(t, \cdot, u(\cdot))-f^i(t, \cdot, v(\cdot))\|_{L_2}+\|h(t, \cdot, u(\cdot))-h(t, \cdot, v(\cdot))\|_{H^{-1}_2}\\
&\qquad +\|g(t, \cdot, u(\cdot))-g(t, \cdot, v(\cdot))\|_{H^{-1+\sigma_0}_2(\ell_2)}\leq \varepsilon \|u\|_{H^{1}_2}+ K_3\|u\|_{L_2},\end{aligned}$$ for any $u,v \in H^{1}_2$ and $\omega, t$.
See Example \[example 7\] for examples satisfying Assumption \[ass 2\](ii).
Denote $$f^i_0=f^i(t,x,0), \quad h_0=h(t,x,0), \quad g_0=g(t,x,0).$$ We will use a well-known inequality (eg. [@Kr99 Lemma 5.2]) $$\begin{aligned}
\label{e:Kr9}
\|au\|_{H^{\sigma}_2}\leq N( \sigma ,d) |a|_{C^{\gamma}}\|u\|_{H^{\sigma}_2},\end{aligned}$$ where $\gamma \geq |\sigma|$ is $\sigma$ is an integral, and otherwise $\gamma >|\sigma|$.
The following is the one of the two main results of this paper.
\[thm divtype\] Suppose Assumptions \[ass 1\] and \[ass 2\] hold. There exists $\kappa_0>0$ depending only on $K,\gamma,\beta, d, T$ so that if $\sup_{\omega,i,j, t \le T}|\sigma^{ij}(t, \cdot)|_{C^1(\ell_2)}\leq \kappa_0$ then equation (\[eqn 7.04.5\]) with initial data $u_0\in U^0_2$ has a unique solution $u\in {\mathbb{H}}^1_2(T)$, and for this solution $$\label{estimate div}
\|u\|^2_{{\mathcal{H}}^1_2(T)}\leq
N\left({\mathbb{E}}\|u_0\|^2_{L_2}+\|f^i_0\|^2_{{\mathbb{L}}_2(T)}+\|h_0\|^2_{{\mathbb{H}}^{-1}_2(T)}
+\|g_0\|^2_{{\mathbb{H}}^{-1+\sigma_0}_2(\ell_2)}\right),$$ where $N$ depends only on $\gamma,\beta, \delta, d, K$ and $T$.
[**[A: Linear case]{}**]{}. Let $f^i$, $h$ and $g^k$ depend only on $(\omega, t, x)$. Due to the method of continuity and solvability result of Lemma \[thm laplace\], it is enough to show that there exists $\kappa_0>0$ so that if $|\sigma^{ij} (t, \cdot)|_{C^1(\ell_2)}\leq \kappa_0$ and $u\in {\mathcal{H}}^1_2(T)$ is a solution of , then the estimate (\[estimate div\]) holds. We refer the reader to the proof of [@Kr99 Theorem 5.1] for details.
[*Step 1.*]{} Assume $b^i=c=\mu^{ik}=\nu^k=0$.
By Theorem \[thm laplace\], the equation $$\partial^{\beta}_t v=\Delta v +D_if^i+h+
\sum_{k=1}^\infty\partial^{\gamma}_t \int^t_0 (\sigma^{ijk}u_{x^ix^j}+g^k)
dW^k_s$$ with initial data $u_0$ has a unique solution $v\in {\mathcal{H}}^1_2(T)$, and $$\label{01.24.11}
\|v\|^2_{{\mathbb{H}}^1_2(T)}\leq
N\left({\mathbb{E}}\|u_0\|^2_{L_2}+\|D_if^i+h\|^2_{{\mathbb{H}}^{-1}_2(T)}+\|\sigma^{ij}u_{x^ix^j}+g\|^2_{{\mathbb{H}}^{-1+\sigma_0}_2(T,\ell_2)}\right).$$
Note that for each $\omega$, $\bar{u}=u-v$ satisfies the deterministic equation $$\partial^{\beta}_t \bar{u}=D_i(a^{ij}
u_{x^j})-\Delta
v=D_i(a^{ij}
\bar{u}_{x^j} +a^{ij}v_{x^j}-v_{x^i}) =
D_i(a^{ij}\bar{u}_{x^j}+\bar{f}^i), \quad \bar{u}(0)=0$$ where $\bar{f}^i=\sum_{j=1}^d a^{ij}v_{x^j}- v_{x^i}$. By a result for the deterministic equations (see [@Za]), $$\|\bar{u}\|_{{\mathbb{H}}^1_2(T)}\leq N\|\bar{f}^i\|_{{\mathbb{L}}_2(T)}\leq
N\|v\|_{{\mathbb{H}}^1_2(T)}.$$ Note that $\sigma^{ij}=0$ if $\gamma \geq 1/2$. If $\gamma<1/2$ then $-1+\sigma_0=-1$ and so by for every $t \le T$ $$\begin{aligned}
\|\sigma^{ij} (t, \cdot) u_{x^ix^j}(t, \cdot) \|_{H^{-1}_2(\ell_2)}&\leq
N|\sigma^{ij}(t, \cdot)|_{C^1(\ell_2)}\|u_{x^ix^j}(t, \cdot)\|_{H^{-1}_2}\\
&\leq
N\sup_{\omega,i,j, t \le T}|\sigma^{ij}(t, \cdot)|_{C^1(\ell_2)}\|u(t, \cdot)\|_{H^1_2}.\end{aligned}$$
This and (\[01.24.11\]) certainly lead to $$\begin{aligned}
\label{N_0}
\|u\|^2_{{\mathbb{H}}^1_2(T)}&\leq& N_0
\sup_{\omega,i,j, t \le T}|\sigma^{ij}(t, \cdot)|_{C^1(\ell_2)}\|u\|^2_{{\mathbb{H}}^1_2(T)}\\
&&+N_0\left(\|u_0\|^2_{U^0_2}+\|D_if^i+h\|^2_{{\mathbb{H}}^{-1}_2(T)}
+\|g\|^2_{{\mathbb{H}}^{-1+\sigma_0}_2(T,\ell_2)}\right), \nonumber\end{aligned}$$ where $N_0$ depends only on $\beta, \gamma, \delta, d, T$ and $K$. Note that $\|D_if^i\|_{H^{-1}_2}\leq N\|f^i\|_{L_2}$. Hence for the desired estimate it is enough to take $$\kappa_0=(2N_0)^{-1/2}.$$
[*Step 2.*]{} Take $\kappa_0$ from [*Step 1*]{}, and assume $\sup_{\omega,i,j, t \le T}|\sigma^{ij}(t, \cdot)|_{C^1(\ell_2)}\leq \kappa_0$. Then by the result of [*Step 1*]{}, for each $t\leq T$, $$\begin{aligned}
\|u\|^2_{{\mathbb{H}}^1_2(t)} &\leq&
N\|b^iu+f^i\|^2_{{\mathbb{L}}_2(t)}+N\|cu+h\|^2_{{\mathbb{H}}^{-1}_2(t)}
+ N\|\mu^iu_{x^i}+\nu
u+g\|^2_{{\mathbb{H}}^{-1+\sigma_0}_2(t,\ell_2)}. \label{01.26.1}\end{aligned}$$ Note that, by $$\|b^iu\|_{L_2}+\|cu\|_{H^{-1}_2} \leq \|b^iu\|_{L_2}+\|cu\|_{L_2}\leq N\|u\|_{L_2}\leq \varepsilon \|u\|_{H^1_2}+N\|u\|_{H^{-1}_2}.$$ Also, since $-1+\sigma_0<1$, by a Sobolev embedding theorem and , for any $\varepsilon>0$ $$\|\nu u\|_{H^{\sigma_0-1}_2}\leq N|\nu|_{C^{|\sigma_0-1|+\kappa}(\ell_2)}\|u\|_{H^{\sigma_0-1}_2}\leq N \varepsilon \|u\|_{H^1_2}+N(\varepsilon)\|u\|_{H^{-1}_2}.$$ By the assumption, $\nu^i=0$ unless $-1+\sigma_0<0$. Therefore, $$\begin{aligned}
\|\mu^i u_{x^i}\|_{H^{-1+\sigma_0}_2}\leq N |\mu^i|_{C^{|\sigma_0-1|+\kappa}(\ell_2)}\|u_x\|_{H^{-1+\sigma_0}_2}
\leq N\|u\|_{H^{\sigma_0}_2}\leq N \varepsilon \|u\|_{H^1_2}+N(\varepsilon)\|u\|_{H^{-1}_2}.\end{aligned}$$ Taking sufficiently small $\varepsilon > 0$ and using (\[01.26.1\]), we get for any $t\leq T$, $$\label{e:4.8}
\|u\|^2_{{\mathbb{H}}^1_2(t)}\leq N \|u\|^2_{{\mathbb{H}}^{-1}_2(t)}+ N (\|u_0\|^2_{U^0_2}+\|f^i\|^2_{{\mathbb{L}}_2(t)}+\|h\|^2_{{\mathbb{H}}^{-1}_2(t)}+\|g\|^2_{{\mathbb{H}}^{-1+\sigma_0}_2(t,\ell_2)}).$$ Since $${\mathbb{D}}u = D_i (a^{ij}u_{x_j}+b^i u +f)+cu+h, \quad {\mathbb{S}}u = \sigma^{ij}u_{x_ix_j} + \mu^{i}u_{x_i}+\nu u +g$$ and $D_i: H^{\gamma}_2 \to H^{\gamma-1}_2$ is a bounded operator, we have $$\label{e:4.9}
\| {\mathbb{D}}u \|^2_{{\mathbb{H}}^{-1}_2(t)} \leq N \Big( \| u\|^2_{{\mathbb{H}}^1_2 (t)}
+ \| f\|^2_{{\mathbb{L}}_2(t)} \Big)
\quad \hbox{and} \quad
\| {\mathbb{S}}u \|^2_{{\mathbb{H}}^{-1}_2(t, \ell_2)} \leq N \Big( \| u\|^2_{{\mathbb{H}}^1_2 (t)}
+ \| g\|^2_{{\mathbb{H}}^{-1}_2(t, \ell_2)} \Big) .$$ This, and Proposition \[lemm 8.28.6\] (with $\sigma=-1$) yield $$(k_{1-\beta}*{\mathbb{E}}\|u\|^2_{H^{-1}_2})(t)\leq N (\|u_0\|^2_{U^0_2}+\|f^i\|^2_{{\mathbb{L}}_2(t)}+\|h\|^2_{{\mathbb{H}}^{-1}_2(t)}+\|g\|^2_{{\mathbb{H}}^{-1+\sigma_0}_2(t,\ell_2)}+\|u\|^2_{{\mathbb{H}}^{-1}_2(t)}).$$ Denote $\eta(t)={\mathbb{E}}\int^t_0\|u(s, \cdot)\|^2_{H^{-1}_2}ds$. Then taking the convolution with $k_{\beta}$ (recall $k_{\beta}*k_{1-\beta}=1$), we get $$\label{gronwall18}
\eta(t)\leq N M+ N(k_{\beta}*\eta)(t),$$ where $$M:=\|u_0\|^2_{U^{0}_2}+\|f^i\|^2_{{\mathbb{L}}_2(T)}+\|h\|^2_{{\mathbb{H}}^{-1}_2(T)}+\|g\|^2_{{\mathbb{H}}^{-1+\sigma_0}_2(T,\ell_2)}.$$ Consequently (\[e:4.8\]), (\[gronwall18\]) and Gronwall’s lemma (Lemma \[gronwall\]) finish the proof for the linear case.
[**[B: Non-linear case]{}**]{}. The proof is identical to that of the non-divergence type case. See the proof of Theorem \[main\] below (it is enough to replace $\sigma$ by $-1$).
**SPDE of non-divergence form type**
====================================
$L_2$-theory for fractional time SPDE of non-divergence form type
-----------------------------------------------------------------
In this subsection, we study the equation of non-divergence type $$\begin{aligned}
\nonumber
\partial^{\beta}_t u&=&\left(a^{ij}u_{x^ix^i}+b^iu+cu+f(u)\right)
\\
&&+ \sum_{k=1}^\infty\partial^{\gamma}_t \int^t_0 (\sigma^{ijk}u_{x^ix^j}+\mu^{ik} u_{x^i}
+\nu^ku +g^k(u))\, dW^k_s
\label{non-div eqn}
\end{aligned}$$ where the coefficients $a^{ij}, b^i, c, \sigma^{ijk}, \mu^{ik}, \nu^k$ are functions depending on $(\omega,t,x)$ and the functions $f, g^k$ depend on $(\omega,t,x)$ and the unknown $u$. Recall that $\sigma_0$ is the constant defined by
\[ass 3\]
\(i) The coefficients $a^{ij}$ are uniformly continuous in $x$, that is for any $\varepsilon>0$, there exists $\delta>0$ so that $$|a^{ij}(t,x)-a^{ij}(t,y)| < \varepsilon, \quad \forall i,j,\omega,t$$ whenever $|x-y|<\delta$.
\(ii) Hölder continuity of $a^{ij}$ when $\sigma \neq 0$ : if $\sigma \neq 0$, there exists constants $\kappa, K_1>0$ so that $$\label{weaken}
|a^{ij}(t,\cdot)|_{C^{|\sigma|+\kappa}}<K_1, \quad \forall i,j,\omega,t.$$
\(iii) For any $i,j,\omega$ and $t$ $$\label{weaken2}
|b^i(t,\cdot)|_{C^{|\sigma|+\kappa}}+| c(t,\cdot)|_{C^{|\sigma|+\kappa}} +|\sigma^{ij}(t,\cdot), \mu^i(t,\cdot),\nu(t,\cdot)|_{C^{|\sigma+\sigma_0|+\kappa}(\ell_2)} \leq K_2 < \infty.$$
\(iv) $|\sigma^{ij}(t,x)|_{\ell_2}\leq \kappa_0$, where $\kappa_0$ is the constant in Theorem \[thm divtype\].
\(v) For any $\varepsilon>0$ there exists $K_3=K_3(\varepsilon)$ so that $$\label{Sobolev}
\|f(t, \cdot, u(\cdot))-f(t, \cdot, v(\cdot))\|_{H^{\sigma}_2}+\|g(t, \cdot, u(\cdot))-g(t, \cdot, v(\cdot))\|_{H^{\sigma+\sigma_0}_2}\leq \varepsilon \|u\|_{H^{\sigma+2}_2}+ K_3\|u\|_{H^{\sigma+1}_2},$$ for any $u,v \in H^{\sigma+2}_2$.
If $\sigma$ is integer then one can slightly weaken (\[weaken\]) and (\[weaken2\]) and take $\kappa=0$ as is done in [@Kr99].
\[example 7\] (i) Let $\delta:=\sigma+2-d/2>0$ and $f_0=f_0(x)\in H^{\sigma}_2$. Take $$f(x,u)=f_0(x) \sup_x |u|.$$ Then by a Sobolev embedding $$\begin{aligned}
\nonumber
\|f(u)-f(v)\|_{H^{\sigma}_2} &\leq& \|f_0\|_{H^{\sigma}_2} \sup_x |u-v|\leq N\|u-v\|_{H^{\sigma+2-\delta/2}_2}\\
&\leq&\varepsilon \|u-v\|_{H^{\sigma+2}_2}+K \|u-v\|_{H^{\sigma}_2}. \label{same}
\end{aligned}$$
\(ii) Fix $\varepsilon>0$ and take $\delta\in (0,1)$ and a random $C^{|\sigma|+\varepsilon}$-function $a(t,x)$. Let $$f(t,x,u)=a(t,x)(-\Delta)^{\delta}u, \quad |a(t,\cdot)|_{C^{|\sigma|+\varepsilon}}\leq K, \quad \forall \omega,t.$$ Then the argument used to prove (\[same\]) easily leads to (\[Sobolev\]).
Denote $$f_0=f(t,x,0), \quad g_0=g(t,x,0).$$
Here is the second main result of this paper.
\[main\] Let $\sigma \in {\mathbb{R}}$ and Assumptions \[ass 1\] and \[ass 3\] hold. Then for any $f\in {\mathbb{H}}^{\sigma}_2(T)$, $g\in {\mathbb{H}}^{\sigma+\sigma_0}_2(T,\ell_2)$ and $u_0\in U^{\sigma+1}_2$, equation (\[non-div eqn\]) admits a unique solution $u\in {\mathcal{H}}^{\sigma+2}_2(T)$, and for this solution we have $$\label{main estimate}
\|u\|_{{\mathbb{H}}^{\sigma+2}_2(T)}\leq N\left(\|u_0\|_{U^{\sigma+1}_2}+\|f_0\|_{{\mathbb{H}}^{\sigma}_2(T)}+\|g_0\|_{{\mathbb{H}}^{\sigma+\sigma_0}_2(T,\ell_2)}\right),$$ where $N$ depends only on $d, \beta, \gamma, \delta, K$ and $T$.
By considering $u-v$ if needed, where $v$ is the solution of $ \partial^{\beta}_t v= \Delta v$ with $v(0)=u_0$, we may assume without loss of generality that $u_0=0$.
. Let $f^i$, $h$ and $g^k$ depend only on $(\omega, t, x)$. Due to the method of continuity we only need to prove that the estimate holds given that a solution already exists.
[*Step 1.*]{} Assume that all the coefficients are independent of $x$, so that equation (\[non-div eqn\]) is of type (\[eqn 7.04.5\]). By applying the operator $(1-\Delta)^{(\sigma+1)/2}$ to equation (\[non-div eqn\]), one can simplify the problem to the case $\sigma=-1$. In this case all the claims follow from Theorem \[estimate div\].
[*Step 2.*]{} Next, we weaken the condition in [*Step 1*]{} by proving that there exists a $\varepsilon_1\in (0,\kappa_0]$ so that the theorem holds if $$\label{purbation}
|a^{ij}(t,x)-a^{ij}(t,y)|+|\sigma^{ij}(t,x)-\sigma^{ij}(t,y)|_{\ell_2}\leq \varepsilon_1 \quad \forall i,j,\omega,t,x,y.$$
Fix $x_0\in {\mathbb{R}}^d$ and denote $$a^{ij}_0(t,x)=a^{ij}(t,x_0), \quad \sigma^{ij}_0(t,x)=\sigma^{ij}(t,x_0).$$ Note that equation (\[non-div eqn\]) can be written as $$\partial^{\beta}_t u=\left(a^{ij}_0u_{x^ix^i}+\bar{f}\right)
+ \sum_{k=1}^\infty\partial^{\gamma}_t \int^t_0 (\sigma^{ijk}_0u_{x^ix^j}+\bar{g}^k)\, dW^k_s,$$ where $$\bar{f}:=(a^{ij}-a^{ij}_0)u_{x^ix^j}+b^iu_{x^i}+cu+f,$$ $$\bar{g}^k:=(\sigma^{ijk}-\sigma^{ijk}_0)u_{x^ix^j}+\mu^{ik}u_{x^i}+\nu^k u +g^k.$$ Note that the coefficients $a^{ij}_0$ and $\sigma^{ij}_0$ are independent of $x$. By the result of [*Step 1*]{}, for each $t\leq T$, $$\label{middle estimate}
\|u\|_{{\mathbb{H}}^{\sigma+2}_2(t)}\leq N\left(\|u_0\|_{U^{\sigma+1}_2}+\|\bar{f}\|_{{\mathbb{H}}^{\sigma}_2(t)}+\|\bar{g}\|_{{\mathbb{H}}^{\sigma+\sigma_0}_2(t,\ell_2)}\right).$$ To estimate $\bar{f}$ and $\bar{g}$ we use the following well known embedding result: for $0\leq \alpha_1 \leq \alpha_2$ and $\alpha_2>0$ $$\label{holder}
|v|_{C^{\alpha_1}}
\le N |v|^{1-\alpha_3}_{C^0}|v|^{\alpha_3}_{C^{\alpha_2}}, \quad \alpha_3:=\frac{\alpha_1}{\alpha_2}.$$ If $\sigma=0$, then $$\|(a^{ij}(t, \cdot)-a^{ij}_0(t))u_{x^ix^j}(t, \cdot)\|_{L_2}\leq N \sup_x|a^{ij}(t, \cdot)-a^{ij}_0(t)| \cdot \|u(t, \cdot)\|_{H^2_2},$$ and otherwise, by first using and then taking $\alpha_1=|\sigma|+\kappa/2$ and $\alpha_2=|\sigma|+\kappa$ in (\[holder\]), $$\begin{aligned}
\|(a^{ij}(t, \cdot)-a^{ij}_0(t))u_{x^ix^j}(t, \cdot)\|_{H^{\sigma}_2}
&\leq N |a^{ij}(t, \cdot)-a^{ij}_0 (t)|_{C^{|\sigma|+\kappa/2}} \cdot \|u(t, \cdot)\|_{H^{\sigma+2}_2}\\
&\leq N \sup_x|a^{ij}(t, \cdot)-a^{ij}_0 (t)|^{\delta} \cdot \|u(t, \cdot)\|_{H^{\sigma+2}_2}.\end{aligned}$$ where $$\delta:=1-\frac{|\sigma|+\kappa/2}{|\sigma|+\kappa}>0.$$
The term $\|(\sigma^{ij}(t, \cdot)-\sigma^{ij}_0(t))u_{x^ix^j}(t, \cdot)\|_{H^{\sigma+\sigma_0}_2}$ and others can be handled similarly. For instance, by $$\|b^i(t, \cdot)u_{x^i}(t, \cdot)\|_{H^{\sigma}_2}
\le N |b^i(t, \cdot)|_{C^{|\sigma|+\kappa}}\cdot \|u(t, \cdot)\|_{H^{\sigma+1}_2}\leq \varepsilon \|u(t, \cdot)\|_{H^{\sigma+2}_2}+N\|u(t, \cdot)\|_{H^{\sigma}_2},$$ and, since $\mu^i=0$ unless $\sigma_0 <1$, $$\begin{aligned}
\|\mu^i (t, \cdot) u_{x^i}(t, \cdot)\|_{H^{\sigma+\sigma_0}_2(\ell_2)}&\leq N |\mu(t, \cdot)|_{C^{|\sigma+\sigma_0|+\kappa}(\ell_2)} \|u(t, \cdot) \|_{H^{\sigma+\sigma_0+1}_2}
\\&\leq \varepsilon \|u(t, \cdot)\|_{H^{\sigma+2}_2}+N\|u(t, \cdot)\|_{H^{\sigma}_2}.\end{aligned}$$ Hence, from (\[middle estimate\]) it follows $$\begin{aligned}
\|u\|_{{\mathbb{H}}^{\sigma+2}_2(t)} &\leq& N\left(\sup_{x,t \le T} |a^{ij}-a^{ij}_0|^{\delta}+ \sup_{x,t \le T} |\sigma^{ij}-\sigma^{ij}_0|^{\delta}_{\ell_2} +\varepsilon\right)\cdot \|u\|_{{\mathbb{H}}^{\sigma+2}_2(t)} \\
&&+N\|u\|_{{\mathbb{H}}^{\sigma}_2(t)}+N\left(\|u_0\|_{U^{\sigma+1}_2}+\|f\|_{{\mathbb{H}}^{\sigma}_2(t)}+\|g\|_{{\mathbb{H}}^{\sigma+\sigma_0}_2(t,\ell_2)}\right).\end{aligned}$$ Take $\varepsilon, \varepsilon_1>0$ so that $\varepsilon, \varepsilon_1< (4N)^{-1}$. If we assume (\[purbation\]) then for each $t\leq T$, $$\label{eqn 1.26.5}
\|u\|_{{\mathbb{H}}^{\sigma+2}_2(t)}\leq N \|u\|_{{\mathbb{H}}^{\sigma}_2(t)}+\left(\|u_0\|_{U^{\sigma+1}_2}+\|f\|_{{\mathbb{H}}^{\sigma}_2(t)}+\|g\|_{{\mathbb{H}}^{\sigma+\sigma_0}_2(t,\ell_2)}\right).$$ Just as and the rest of the argument in the proof of Theorem \[thm divtype\], this and Gronwall’s lemma (Lemma \[gronwall\]) lead to the desired estimate.
[*Step 3.*]{} General linear case without condition . Extension of [*Step 2*]{} to the general case is quite straightforward and can be found for example in the proof of [@Kr99 Theorem 5.1]. One introduces a partition of unity $\{\zeta_n :n=1,2,\cdot\}$ of $C^{\infty}_0({\mathbb{R}}^d)$-functions so that (\[purbation\]) holds on each support of $\zeta_n$. Then one estimates $u\zeta_n$ using the result of [*Step 2*]{} and by summing up these estimate one easily gets (\[eqn 1.26.5\]), which is sufficient for our estimate.
[**[B: Non-linear case]{}**]{}. We modify the proof of [@Kr99 Theorem 5.1]. Recall that ${\mathcal{H}}^{\sigma+2}_{2,0}(T)$ is defined in . For each $u\in {\mathcal{H}}^{\sigma+2}_{2,0}(T)$ consider the equation $$\begin{aligned}
\nonumber
\partial^{\beta}_t v&=&\left(a^{ij}
v_{x^ix^i}+b^iv+cv+f(u)\right)
\\
&&+ \sum_{k=1}^\infty\partial^{\gamma}_t \int^t_0 (\sigma^{ijk}v_{x^ix^j}+\mu^{ik} v_{x^i}
+\nu^kv +g^k(u))\, dW^k_s
\nonumber
\end{aligned}$$ with initial data $v(0)=0$. By the above results, this equation has a unique solution $v\in {\mathcal{H}}^{\sigma+2}_{2, 0}(T)$. By denoting $v={\mathcal{R}}u$ we can define an operator ${\mathcal{R}}: \,\,{\mathcal{H}}^{\sigma+2}_{2,0}(T) \to {\mathcal{H}}^{\sigma+2}_{2,0}(T)$.
Note that due to the interpolation $\|\xi\|_{H^{\sigma+1}_2}\leq \varepsilon \|\xi\|_{H^{\sigma+2}_2}+N\|\xi\|_{H^{\sigma}_2}$, (\[Sobolev\]) is equivalent to $$\label{Sobolev2}
\|f(t, \cdot, u(\cdot))-f(t, \cdot, v(\cdot))\|_{H^{\sigma}_2}+\|g(t, \cdot, u(\cdot))-g(t, \cdot, v(\cdot))\|_{H^{\sigma+\sigma_0}_2}\leq \varepsilon \|u\|_{H^{\sigma+2}_2}+ K\|u\|_{H^{\sigma}_2}$$ for some $K=K(\eps)>0$.
By the results for the linear case and (\[Sobolev2\]) and Proposition \[lemm 8.28.6\], for each $t\leq T$, $$\begin{aligned}
\|{\mathcal{R}}u-{\mathcal{R}}v\|^2_{{\mathcal{H}}^{\sigma+2}_2(t)}&\leq& N \|f(u)-f(v)\|^2_{{\mathbb{H}}^{\sigma}_2(t)}+N\|g(u)-g(v)\|^2_{{\mathbb{H}}^{\sigma+\sigma_0}_2(t,\ell_2)}\\
&\leq& N\varepsilon^2\|u-v\|^2_{{\mathcal{H}}^{\sigma+2}_2(t)}+NK^2\|u-v\|^2_{{\mathbb{H}}^{\sigma}_2(t)}\\
&\leq& N_0\varepsilon^2\|u-v\|^2_{{\mathcal{H}}^{\sigma+2}_2(t)}+N_1\int^t_0 (t-s)^{-1+\beta}\|u-v\|^2_{{\mathcal{H}}^{\sigma+2}_2(s)}\,ds,\end{aligned}$$ where $N_1$ depends also on $\varepsilon$. Next, we fix $\varepsilon$ so that $\theta:=N_0\varepsilon^2<1/4$. Then repeating the above inequality and using the identity $$\begin{aligned}
\int^t_0 (t-s_1)^{-1+\beta}\int^{s_1}_0 (s_1-s_2)^{-1+\beta}\cdots \int^{s_{n-1}}_0 (s_{n-1}-s_n)^{-1+\beta} ds_n \cdots ds_1
=\frac{\Gamma(\beta)}{\beta\Gamma(n\beta+1)} t^{n\beta},\end{aligned}$$ we get $$\begin{aligned}
&&\|{\mathcal{R}}^m u-{\mathcal{R}}^m v\|^2_{{\mathcal{H}}^{\sigma+2}_2(t)}
\\
&&\leq \sum_{k=0}^m \begin{pmatrix} m\\ k\end{pmatrix}
\theta^{m-k} (T^{\beta}N_1)^k \frac{\Gamma(\beta)}{\beta\Gamma(k\beta+1)} \, \|u-v\|^2_{{\mathcal{H}}^{\sigma+2}_2(t)}\\
&&\leq 2^m \theta^m \left[\max_k \left( (\theta^{-1}T^{\beta}N_1)^k\frac{\Gamma(\beta)}{\beta\Gamma(k\beta+1)}\right)\right]\, \|u-v\|^2_{{\mathcal{H}}^{\sigma+2}_2(t)}\\
&&\leq
\frac{1}{2^m} N_2 \|u-v\|^2_{{\mathcal{H}}^{\sigma+2}_2(t)}.\end{aligned}$$ For the second inequality above we use $\sum_{k=0}^m \begin{pmatrix} m\\ k\end{pmatrix}=2^m$. It follows that if $m$ is sufficiently large then ${\mathcal{R}}^m$ is a contraction in ${\mathcal{H}}^{\sigma+2}_{2,0}(T)$, and this yields all the claims. The theorem is proved.
An application to SPDE driven by space-time white noise
-------------------------------------------------------
In this subsection, we consider a SPDE driven by space-time white noise. We consider $$\label{space-time}
\partial^{\beta}_t u=\left(a^{ij}u_{x^ix^j}+b^iu_{x^i}+cu+f(u)\right)
+ \sum_{k=1}^\infty\partial^{\gamma}_t \int^t_0 h(u) \, dB_t$$ where the coefficients $a^{ij}, b^ic$ and are functions depending on $(\omega,t,x)$, the functions $f$ and $h$ depends on $(\omega,t,x)$ and the unknown $u$, and $B_t$ is a space-time white noise.
Let $\{\eta^k:k=1,2,\cdots\}$ be an orthogonal basis of $L_2({\mathbb{R}}^d)$. Then (at least formally) $$B_t=\sum_{k=1}^{\infty} \eta^k W^k_t$$ where $W^k_t:=(B_t,\eta^k)_{L_2}$are independent one dimensional Wiener processes. Hence one can rewrite (\[space-time\]) as $$\partial^{\beta}_t u= \left(a^{ij}u_{x^ix^j}+b^iu_{x^i}+cu+f(u)\right)
+ \sum_{k=1}^\infty\partial^{\gamma}_t \int^t_0 h(u) \eta^k \, dW^k_t.$$
Denote $$g^k(t,x,u)=h(t,x,u)\eta^k(x).$$ To apply Theorem \[main\], we only need to find $\sigma$ and conditions on $h$ so that (\[Sobolev\]) holds. The following lemma is a consequence of [@Kr99 Lemma 8.4].
\[lemma 1.26\] Let $\gamma<-1/2$. Then $$\|g(t, \cdot, u(\cdot))-g(t, \cdot, v(\cdot))\|_{H^{\gamma}_2(\ell_2)}\leq N \|h(t, \cdot, u(\cdot))-h(t, \cdot, v(\cdot))\|_{L_2}.$$
The following is an easy consequence of Theorem \[main\] and Lemma \[lemma 1.26\]. Recall that $f_0=f(t,x,0)$. We also denote $
h_0=h(t,x,0).
$
\[thm space-time\] Let $$\label{eqn last}
\sigma+\sigma_0<-1/2, \quad \sigma+2>0.$$ Assume $$|f(t,x,v_1)-f(t,x,v_2)|+|h(t,x,v_1)-h(t,x,v_2)|\leq K|v_1-v_2| \quad \forall \omega,t, x, v_1, v_2,$$ and Assumptions \[ass 1\] and \[ass 3\] hold with $\sigma$ satisfying (\[eqn last\]). Then equation (\[space-time\]) with initial data $u_0\in U^{\sigma+1}_2$ has a unique solution $u$, and for this solution we have $$\|u\|_{{\mathbb{H}}^{\sigma+2}_2(T)}\leq N
\left(\|f_0\|_{{\mathbb{H}}^{\sigma}_2(T)}+\|h_0\|_{{\mathbb{L}}_2(T)}+\|u_0\|_{U^{\sigma+1}_2}\right).$$
It is enough to note that, since $\sigma+2>0$, $$\begin{aligned}
&&\|f(t, \cdot, u(\cdot))-f(t, \cdot, v(\cdot))\|_{H^{\sigma}_2}+\|g(t, \cdot, u(\cdot))-g(t, \cdot, v(\cdot))\|_{H^{\sigma+\sigma_0}_2(\ell_2)}\\
&\leq& N\|u-v\|_{L_2}\leq
\varepsilon \|u-v\|_{H^{\sigma+2}_2}+K\|u-v\|_{H^{\sigma}_2}.
\end{aligned}$$ The corollary is proved.
The constant $\sigma+2$ gives the regularity of solution $u$. To see how smooth the above solution is, we recall $$\sigma_0:=\gamma_0+ (\varepsilon_0 1_{\gamma=1/2})
=\begin{cases}
(2\gamma-1) /\beta \quad &\hbox{if } \gamma >1/2\\
\varepsilon_0 &\hbox{if } \gamma =1/2\\
0 &\hbox{if } \gamma <1/2.
\end{cases}$$ Since $\sigma+2=(\sigma+\sigma_0)+(2-\sigma_0)<-1/2+(2-\sigma_0)$, it follows $$\sigma+2 < \begin{cases}
\frac{3}{2}-\frac{2\gamma-1}{\beta} \quad &\hbox{if } \gamma >1/2\\
\frac{3}{2} &\hbox{if } \gamma \leq 1/2 .
\end{cases}$$ Since we are assuming $\sigma+2>0$, we need $$\gamma<\frac{1}{2}+\frac{3}{4}\beta,$$ which is slightly stronger than (\[e:1.7\]).
[**Acknowledgement.**]{} We thank T. Kumagai for discussions on topics related to deterministic fractional time equations.
[mm]{}
M. M. Djrbashian, [*Harmonic analysis and boundary value problems in the complex domain*]{}, translated from the manuscript by H. M. Jerbashian and A. M. Jerbashian \[A. M. Dzhrbashyan\]. Operator Theory: Advances and Applications [**65**]{}. Birkhäuser Verlag, Basel, 1993
S.D. Eidelman, S.D. Ivasyshen and A.N. Kochubei, [*Analytic methods in the theory of differential and pseudo-differential equations of parabolic type.*]{} Operator theory: Advances and Applications [**[152]{}**]{}, 2004.
R. Gorenflo and F. Mainardi, [*Fractional diffusion processes: Probability distribution and continuous time random walk*]{}, Lecture Notes in Physics [**621**]{} (2003), 148–166.
J. Klafter and I. M. Sokolov, [*Anomalous diffusion spreads its wings*]{}, Physics Work August (2005), 29–32.
N.V. Krylov, An analytic approach to SPDEs, [*Stochastic Partial Differential Equations: Six Perspectives*]{}, Mathematical Surveys and Monographs [**[64]{}**]{}, AMS, Providence, RI. 1999
N.V. Krylov, [*On the Itô-Wentzell formula for distribution-valued processes and related topics*]{}. Probab. Theory Related Fields [**150**]{} (2011), no. 1–2, 295–319.
A. Lunardi and E. Sinestrari, [*An inverse problem in the theory of materials with memory*]{}, Nonlin. Anal. Theory Meth. Appl. [**12**]{} (1988), 1317–1355,
M. M. Meerschaert and A. Sikorskii, [*Stochastic Models for Fractional Calculus*]{}, De Gruyter Studies in Mathematics [**43**]{}. Walter de Gruyter, Berlin/Boston, 2012.
R. Metzeler and J. Klafter, [*The random walk’s guide to anomalous diffusion: a fractional dynamics approach*]{}, Phys. Peps. [**339(1)**]{} (2000), 77.
J. Pr$\ddot{u}$ss, [*Evolutionary integral equations and applications*]{}, Monogr. Math. [**87**]{}, Birkh$\ddot{a}$user, Basel, 1993.
R. Schumer, D.A. Benson, M.M. Meerschaert and S. W. Wheatcraft, [*Eulerian derivation of the fractional advection-dispersion equation*]{}, J. Contaminant Hydrol. [**48**]{} (2001), 69–88.
L. von Wolfersdorf, [*An identification of memory kernels in linear theory of heat equation*]{}, Math. Method. appl. sci. [**17**]{} (1994) 919—932.
H. Ye, J. Gao and Y. Ding, [*A generalized Gronwall inequality and its application to a fractional differential equation*]{}, J. Math. Anal. Appl. [**[328]{}**]{} (2007), 1075–1081.
R. Zacher, [*Weak solutions of abstract evolutionary integro-differential equations in Hilbert spaces*]{}, Funkcialaj Ekvacioj [**[52]{}**]{} (2009), 1–18.
[^1]: Research of ZC is partially supported by NSF Grants NSF Grant DMS-1206276.
[^2]: Research of KK was supported by Basic Science Research Program through the National Research Foundation of Korea(NRF) funded by the Ministry of Education, Science and Technology (2013020522)
[^3]: Research of PK was supported by the National Research Foundation of Korea(NRF) grant funded by the Korea government(MEST) (2013004822).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The experimental observation of the long-sought quantum anomalous Hall effect was recently reported in magnetically doped topological insulator thin films \[Chang *et al*., Science **340**, 167 (2013)\]. An intriguing observation is a rapid decrease from the quantized plateau in the Hall conductance, accompanied by a peak in the longitudinal conductance as a function of the gate voltage. Here, we present a quantum transport theory with an effective model for magnetic topological insulator thin films. The good agreement between theory and experiment reveals that the measured transport originates from a topologically nontrivial conduction band which, near its band edge, has concentrated Berry curvature and a local maximum in group velocity. The indispensable roles of the broken structure inversion and particle-hole symmetries are also revealed. The results are instructive for future experiments and transport studies based on first-principles calculations.'
author:
- 'Hai-Zhou Lu, An Zhao, and Shun-Qing Shen'
date: '20 June, 2013'
title: Quantum Transport in Magnetic Topological Insulator Thin Films
---
In some metallic ferromagnets, a transverse current can be induced by a longitudinal electric field, known as the anomalous Hall effect [@Nagaosa10rmp; @Xiao11rmp]. The phenomenon does not need an external magnetic field, thus it is distinct from the ordinary Hall effect. It has been perceived that in some insulating ferromagnets the anomalous Hall conductance could be quantized in units of the conductance quantum $e^{2}/h$, meanwhile, the longitudinal conductance vanishes [@Haldane-88prl], leading to the quantum anomalous Hall effect, the last and long-sought family member of the Hall effects. In the quantum anomalous Hall system, the nontrivial topology of the bulk states and broken time-reversal symmetry give rise to chiral edge states in the energy gap. The dissipationless transport of the topologically protected edge states gives the quantized conductances, and is believed to have promising applications in quantum electronic devices with low power consumption. Solutions and mechanisms to realize the quantum anomalous Hall effect have attracted tremendous efforts in past decades [@Qi06prb; @Onoda06prl; @Liu08prl; @Yu10sci; @Nomura11prl; @Qiao10rc; @Chu11prb; @Chang13am; @Checkelsky12natphys]. One of the most promising schemes [@Yu10sci] is based on the magnetically doped topological insulators [@Hor10prb; @Chen10sci; @Wray11natphys], where the interplay of strong spin-orbit coupling and magnetic exchange interaction gives rise to the band inversion required by the quantum anomalous Hall effect.
![(color online). Calculated Hall (square) and longitudinal (circle) conductances as functions of the Fermi energy $E_{F}$. Parameters: $\Delta=0.01$ eV, $B=-30$ eV$\cdot$Å$^{2}$, $\gamma=3$ eV$\cdot$Å, $m=-0.06$ eV, $D=28$ eV$\cdot$Å$^{2}$, $V=0.01$ eV, $nu_{1}^{2}=nu_{2}^{2}=100$ (eV$\cdot$Å)$^{2}$, and $nu_{3}^{2}=nu_{4}^{2}=10$ (eV.Å)$^{2}$. The energy-related parameters are up to a scaling compared to the experiment. Inset: the Hall and longitudinal conductances measured in the experiment (adopted from Ref. [@Chang13sci]).[]{data-label="fig:sigma"}](sigma){width="8.5cm"}
Recently, the experimental observation on the quantum anomalous Hall effect was reported in Cr-doped (Bi,Sb)$_{2}$Te$_{3}$ ultra thin films [@Chang13sci]. The measured Hall conductance exhibits a quantized plateau while the longitudinal conductance decreases drastically at lower temperatures. A more subtle behavior appeared on the positive gate voltage side of the quantized plateau: the Hall conductance shows a sudden drop, accompanied by a peak in the longitudinal conductance (the inset of Fig.\[fig:sigma\]). Understanding the mechanisms beneath the subtle behavior is crucial because they are closely related to the topological origin of the quantized plateau. In this Letter, we present a quantum transport theory, based on an effective microscopic model for the topological insulator thin films grown on substrates. With the single model, both the calculated Hall and longitudinal conductances match the measured data very well (see Fig. \[fig:sigma\]). The good agreement between theory and experiment reveals the following. (1) The conduction and valence bands closest to the quantized plateau are *always* topologically nontrivial. The transport features in Fig. \[fig:sigma\] come from the nontrivial conduction band, which has concentrated Berry curvature and a local maximum in group velocity near its band edge. (2) To induce the band inversion for the quantum anomalous Hall effect, a stronger magnetic exchange field is required to overcome not only the finite-size gap of the thin film but also the effect of structural inversion asymmetry (SIA), which is caused by the potential difference between the top and bottom surfaces. Usually SIA is stronger in thicker films, resulting in a possible obstacle to realize the quantum anomalous Hall effect in thicker samples. (3) The transport is in the diffusive regime: the longitudinal transport depends on the group velocity rather than the density of states, which will be helpful for further transport studies based on first-principles calculations. (4) The broken particle-hole symmetry is also indispensable, which gives rise to the local maximum in the group velocity and the asymmetry around the quantized plateau.
![(color online). (a) Solid and dashed curves depict the top and bottom surface states, respectively, of a thin film of topological insulator on the substrate before magnetic doping. \[(b)-(d)\] The exchange field from magnetic dopants can lift the degeneracies at $k=0$ and induce a band inversion, whereas SIA induced by the substrate is competing against the exchange field. (d) No band crossing at finite $k$ if $\Delta B<0$. (e) Band crossings at finite $k$ are expected if the band inversion is achieved by doping a thin film with $\Delta B>0$. (f) SIA can turn the band crossings to anticrossings while exchanging the topological properties between inner and outer bands. $0,\pm1$ indicate the contribution to $\sigma_{xy}$ (Chern number) of a band if the band is fully occupied. []{data-label="fig:evolution"}](picture_SIA){width="8.5cm"}
Ferromagnetic insulators were predicted to form in magnetically doped topological insulators such as Cr or Fe doped Bi$_{2}$Te$_{3}$ and Sb$_{2}$Te$_{3}$, and their thin films have a topological band structure, leading to the quantum anomalous Hall effect [@Yu10sci]. Starting from the three-dimensional effective model for topological insulators [@Zhang09np; @Shen11spin] with the exchange field splitting, and using the solution of the confined quantum states at the $\Gamma$ point ($k_{x},k_{y}=0$) in a thin film as a basis [@Lu10prb; @Shan10njp], we obtain the effective model [@supp] $$\begin{aligned}
H=H_{0}+\frac{m}{2}\tau_{0}\otimes\sigma_{z}.\label{H}\end{aligned}$$ $m$ is the exchange field from the magnetic dopants [@Liu08prl; @Yu10sci], which acts effectively like a Zeeman field. $\tau_{0}$ is a $2\times2$ unit matrix. $\sigma_{z}$ is the $z$ Pauli matrix. $H_{0}$ is the effective model for the thin films of the topological insulator [@Lu10prb; @Shan10njp] $$\begin{aligned}
& & H_{0}=-Dk^{2}\nonumber \\
& & +\left(\begin{array}{cccc}
\frac{\Delta}{2}-Bk^{2} & i\gamma k_{-} & V & 0\\
-i\gamma k_{+} & -\frac{\Delta}{2}+Bk^{2} & 0 & V\\
V & 0 & -\frac{\Delta}{2}+Bk^{2} & i\gamma k_{-}\\
0 & V & -i\gamma k_{+} & \frac{\Delta}{2}-Bk^{2}
\end{array}\right)\nonumber \\\end{aligned}$$ where $(k_{x},k_{y})$ is the wave vector, $k^{2}=k_{x}^{2}+k_{y}^{2}$. The $D$ term breaks the particle-hole symmetry, and the band gap opening requires $|D|<|B|$. $\Delta$ is the hybridization of the top and bottom surface states of the thin film [@Lu10prb; @Linder09prb; @Liu10rc], which becomes relevant for thin films, e.g., Bi$_{2}$Se$_{3}$ thinner than 5 nm [@Zhang10natphys; @Sakamoto10prb]. Both $\Delta$ and $B$ are functions of the thickness of thin film, and approach zero simultaneously for a thicker film. $\gamma=v\hbar$, with $v$ the effective velocity. $V$ measures the SIA between the top and bottom surfaces of the thin film. The inclusion of SIA here is a natural consequence for a realistic thin film grown on a substrate, which always induces a potential distribution along the film growth direction [@Shan10njp]. The potential shifts the gapless Dirac cones on the top and bottom surfaces, and gives the Rashba-like splitting in the band structure when the top-bottom hybridization $\Delta$ is present [\[]{}Fig. \[fig:evolution\] (a)[\]]{} as observed by ARPES [@Zhang10natphys].
It can be explicitly shown that the SIA increases the exchange field required by the quantum anomalous Hall effect. Under a unitary transformation [@supp], the Hamiltonian can be diagonalized into two $2\times2$ blocks $$\begin{aligned}
h_{s}= & -Dk^{2}+\sigma_{z}\left(\Gamma+s\Lambda\right)+s\gamma(k_{x}\sigma_{y}-k_{y}\sigma_{x})\cos\Theta,\end{aligned}$$ where $s=\pm1$ for the outer and inner blocks, respectively. The outer (inner) block has a larger (smaller) band gap at $k=0$. We have defined $\Gamma$ = $\sqrt{(m/2)^{2}+\gamma^{2}k^{2}\sin^{2}\Theta}$, $\Lambda$ = $\sqrt{(\Delta/2-Bk^{2})^{2}+V^{2}}$, and $\cos\Theta$ = $(\Delta/2-Bk^{2})/\Lambda$. $\sigma_{x,y,z}$ are the Pauli matrices. For $s=+1$, the dispersions of the two bands (denoted as the outer bands) are $E_{i}=-Dk^{2}\pm\sqrt{\left(\Lambda+\Gamma\right)^{2}+(\gamma k)^{2}\cos^{2}\Theta}$ ($i=1$ for $-$, and 4 for $+$), and the outer energy gap at $k=0$ is $\left|m\right|+\sqrt{\Delta^{2}+4V^{2}}$, which is always positive. For $s=-1$, the dispersions of the two inner bands are $E_{i}=-Dk^{2}\pm\sqrt{\left(\Lambda-\Gamma\right)^{2}+(\gamma k)^{2}\cos^{2}\Theta}$ ($i=2$ for $-$, and 3 for $+$). The inner energy gap at $k=0$ is $\left|m\right|-\sqrt{\Delta^{2}+4V^{2}}$. Without $m$, the inner bands and outer bands touch at $k=0$ \[Fig. \[fig:evolution\] (a)\]. A finite $m$ can lift the degeneracies at $k=0$ \[Fig. \[fig:evolution\] (b)\]. Increasing $m$ then produces a band inversion for bands 2 and 3 at $k=0$, when $$\begin{aligned}
|m|=\sqrt{\Delta^{2}+4V^{2}},\label{m-Delta-V}\end{aligned}$$ and changes their topological properties from trivial to nontrivial or vice versa \[Figs. \[fig:evolution\] (c) and \[fig:evolution\] (d)\]. We have shown in Eq. (\[m-Delta-V\]) that, in the presence of SIA, the band inversion requires the exchange field $m$ to exceed not $\Delta$ but a larger value $\sqrt{\Delta^{2}+4V^{2}}$.
![(color online). (a) There are band crossings if the band inversion is achieved by magnetically doping a thin film with $\Delta B>0$ . (b) The Berry curvature of the corresponding bands in (a). We have adopted polar coordinates so $\Omega_{i}^{z}(\mathbf{k})\rightarrow k\Omega_{i}^{z}(k)$. (c) The band crossings in (a) turn to anticrossings in the presence of SIA ($V\neq0$). (d) The Berry curvature of the bands in (c). Parameters: $\Delta=-0.01$ eV, $B=-30$ eV$\cdot$Å$^{2}$, $\gamma=3$ eV$\cdot$Å, $m=-0.06$ eV, $D=28$ eV$\cdot$Å$^{2}$. Left: $V=0$. Right: $V=0.01$ eV. The band with the lowest (highest) energy at $k=0$ is indexed as band 1 (4). Band 1: red dashed. Band 2: blue solid. Band 3: green solid. Band 4: black dashed. []{data-label="fig:SIA"}](SIA){width="8.5cm"}
Moreover, we find that SIA also leads to qualitative differences: it makes the bands closest to the gap always topologically nontrivial. Without SIA ($V=0$), the model reduces to the one proposed by Yu *et al* [@Yu10sci]. From the band structure, we can show that if $\Delta B>0$, there will be band crossings at a finite $k=\sqrt{\Delta/2B}$ \[Fig. \[fig:evolution\] (e)\]. We find that SIA can turn the band crossings at finite $k$ into band anticrossings, leading to a change of topological properties [\[]{}Fig. \[fig:evolution\] (f)[\]]{}. As a result, the inner two bands, 2 and 3, become nontrivial. To show this, we investigate the topological properties of the bands with the help of the Berry curvature $$\begin{aligned}
\Omega_{i}^{z}(\mathbf{k})=-2\sum_{j\neq i}\frac{\mathrm{Im}\langle i|\partial H/\partial k_{x}|j\rangle\langle j|\partial H/\partial k_{y}|i\rangle}{(E_{i}-E_{j})^{2}},\end{aligned}$$ where for given $\mathbf{k}$, $\left|i\right\rangle $ is the eigenstate in band $i$ with energy $E_{i}$. Usually, $\Omega_{i}^{z}(\mathbf{k})$ of band $i$ is a function of $k$, and can have either positive or negative values. Its integral over all the $k$ in the first Brillouin zone is always an integer, i.e., the Chern number. If the Chern number is zero, the energy band is topologically trivial. However, if $\Omega_{i}^{z}(\mathbf{k})$ is always positive or negative, its integral must be a nonzero integer, meaning that the band is nontrivial. Figure \[fig:SIA\] shows a case in which the quantum anomalous Hall phase is created from a thin film with $\Delta B>0$ and $m$ has already overcome $\sqrt{\Delta^{2}+4V^{2}}$ to invert bands 2 and 3 at $k=0$ \[Fig. \[fig:SIA\] (a)\]. Without SIA ($V=0$), the Berry curvature in Fig. \[fig:SIA\] (b) shows that bands 2 and 3 are trivial while bands 1 and 4 are nontrivial. With SIA ($V\neq0$), the band crossings in Fig. \[fig:SIA\] (a) turn into anticrossings in Fig. \[fig:SIA\] (c). Meanwhile, bands 2 and 3 exchanged their Berry curvature with bands 1 and 4 for $k>\sqrt{\Delta/2B}$ \[Fig. \[fig:SIA\] (d)\]. As a result, bands 2 and 3 become nontrivial. Therefore, we have shown that the inner bands are always nontrivial. In contrast, if $\Delta B<0$ there is no band crossing-anticrossing transition (see Fig. 1 of the Supplemental Material [@supp]), and the inner two bands are topologically nontrivial only because of the band inversion at $k=0$. Thus, in the presence of SIA, the bands closest to the gap are always topologically nontrivial. We shall show that the nontrivial bands account for the measured longitudinal and Hall conductances in the following calculation.
*Hall conductance*. - According to the linear response theory, the intrinsic Hall conductance is given by the integral of the Berry curvature over occupied states in $k$ space [@Xiao11rmp] $$\begin{aligned}
\sigma_{xy} & = & -\frac{e^{2}}{\hbar}\sum_{i}\int\frac{d^{2}\mathbf{k}}{(2\pi)^{2}}f(E_{i}-E_{F})\Omega_{i}^{z}(\mathbf{k}),\end{aligned}$$ where $f(x)=[\mathrm{exp}(x/k_{B}T)+1]^{-1}$ is the Fermi function and $E_{F}$ is the Fermi energy. Without SIA, it has been shown that [@Lu10prb] the in-gap Hall conductance $\sigma_{xy}^{\mathrm{gap}}=-\frac{e^{2}}{2h}[\mathrm{sgn}(\Delta+m)+\mathrm{sgn}(-\Delta+m)]$. For the quantized anomalous Hall effect, $|m|$ must be larger than $|\Delta|$, and then $\sigma_{xy}^{\mathrm{gap}}=-\frac{e^{2}}{h}\mathrm{sgn}(m)$, which only depends on the sign of $m$. A positive quantized plateau of $\sigma_{xy}$ indicates that $m<0$. In the presence of SIA, the value of the quantized Hall conductance remains the same but the quantum anomalous Hall phase requires $|m|>\sqrt{\Delta^{2}+4V^{2}}$. In the experiment [@Chang13sci], the drop of the Hall conductance (inset of Fig. \[fig:sigma\]) from the quantized plateau is rather prompt on the positive gate voltage side. This behavior implies that the band closest to the gap is topologically nontrivial (band 3). The nonzero Berry curvature of the band is mainly distributed near its band edge, leading to the prompt drop of $\sigma_{xy}$. The concentrated Berry curvature distribution is probably given by a relatively narrow bandwidth, and small band gap. Another feature in $\sigma_{xy}$ is a small dip in $\sigma_{xy}$ (marked by the arrow in Fig. \[fig:sigma\]). This is given by a trivial band at higher energy (band 4). If the band 4 is fully occupied, it has no contribution to the total Hall conductance. But before being fully occupied, with increasing $E_{F}$, it first reduces then enhances the Hall conductance, giving rise to the dip. By definition, the intrinsic Hall conductance will always vanish at higher energy. However, $\sigma_{xy}$, in the experiment, saturates at some finite values, in particular, rather high ($\sim0.5e^{2}/h$) on the negative gate voltage side. The pinning of the Fermi surface probably does not account for the saturation, because $\sigma_{xx}$ is still increasing meanwhile. The impurity scattering induced extrinsic Hall conductance [@Nagaosa10rmp] may be one of the reasons. We calculated the extrinsic contribution from the side-jump mechanism [@supp]. However, it turns out that the perturbation theory assuming weak impurity scattering [@Sinitsyn07prb; @Yang11prb; @Culcer11prb; @Lu13arXiv] may not be enough, because the experiment was in a strong disordered or bad-metal regime [@Chang13sci; @Chang13am].
![(color online). Group velocity $dE/dk$ and the density of states of bands 1-4. Parameters: $\Delta=0.01$, $B=-30$ eV$\cdot$Å$^{2}$, $\gamma=3$ eV$\cdot$Å, $m=-0.06$ eV, and $V=0.01$ eV. From left to right: $D=28$, 15, and 0 eV$\cdot$Å$^{2}$. The colors and styles follow those in Fig. \[fig:SIA\]. The case with $\Delta=-0.01$ shows a similar result.[]{data-label="fig:group velocity"}](dE_dk_DOS_compare){width="8.5cm"}
*Longitudinal conductivity*. - Considering the low mobility in the experimental sample, the electronic transport outside the gap is diffusive. To the leading order, the longitudinal conductivity in the diffusive regime is given by the Einstein relation [@Ashcroft-Mermin] $\sigma_{xx}=e^{2}N_{F}\mathcal{D}$, where $N_{F}$ is the density of states at the Fermi energy, the diffusion coefficient $\mathcal{D}=v_{F}^{2}\tau/2$, with the Fermi velocity defined as $v_{F}=(\partial E/\partial k)/\hbar$, the scattering time $\tau=\hbar/(2\pi N_{F}nu^{2})$ where $n$ is the impurity concentration, $u$ measures the scattering strength. For simplicity, we use $nu^{2}$ to account for the overall effect from both the nonmagnetic and magnetic scattering. Both $\sigma_{xx}$ and $\tau$ depend on $N_{F}$, indicating that a large density of states also means strong scattering on the Fermi surface. As a result, the density of states cancels out in $\sigma_{xx}$, and we arrive at $$\begin{aligned}
\sigma_{xx}=\frac{e^{2}}{h}\sum_{i=1}^{4}\frac{1}{2}\left|\frac{\partial E_{i}}{\partial k}\right|_{E_{i}=E_{F}}^{2}\frac{1}{nu_{i}^{2}}\end{aligned}$$ for multiple bands. Suppose that $nu_{i}^{2}$ varies slowly with the Fermi energy; the peak in $\sigma_{xx}$, therefore, corresponds to a local maximum in the group velocity. This behavior can be captured by the parameters of the same orders as those in the experiments. In Fig. \[fig:group velocity\], we show the group velocity and the density of states [\[]{}given by $k_{F}/(2\pi|dE/dk|_{k_{F}})$ in polar coordinates, $k_{F}$ the Fermi wave vector[\]]{}. Compared with Fig. \[fig:sigma\], it is clear that $\sigma_{xx}$ is dominantly determined by the group velocity rather than the density of states. It is found that the peak is better produced by comparable $|B|$ and $|D|$. Although Fig. \[fig:group velocity\] shows only a case with $\Delta B<0$, similar $\sigma_{xx}$ can be produced with $\Delta B>0$.
*Thick films*. - For those thicker films (e.g., typically more than 5 nm for Bi$_{2}$Se$_{3}$), the hybridization between the top and bottom surfaces becomes negligibly small, and both $\Delta$ and $B$ may be abandoned. In this case, the Hamiltonian can be projected to the top and bottom surfaces [@supp] $$\begin{aligned}
U^{\dag}HU=-Dk^{2}\pm V+\frac{m}{2}\sigma_{z}+\gamma(k_{x}\sigma_{y}-k_{y}\sigma_{x}),\end{aligned}$$ where $U=(\sigma_{0}-i\sigma_{y})\otimes\sigma_{0}/\sqrt{2}$, and $\pm V$ shows that SIA can shift the relative positions of the bands from the top ($+$) surface with respect to those from the bottom ($-$) surface. As massive Dirac fermion systems, each of the top and bottom surfaces contributes $-(e^2/2h)\mathrm{sgn}(m)$ to the Hall conductance when the Fermi level is located in the gap [@Redlich84prd; @Jackiw84prd], so the total Hall conductance is $-(e^2/h)\mathrm{sgn}(m)$. For a smaller $\left|V\right|<\left|m\right|/2$, there is still an energy gap that protects the quantized Hall conductance. However, for a larger $V>|m|/2$, the energy gap closes and the Hall conductance is not quantized. Since the SIA comes from the substrate-induced electric field along the growth direction, a thicker film has a larger $V$ [@Zhang10natphys]. Thus, if no other mechanism is taken into account, it may be hard to realize a quantized anomalous Hall conductance in thicker films, although the finite-size $\Delta$ is smaller. Also, in this case, the surface states of the lateral sides of the film contribute to the longitudinal conductance [@Chu11prb], which forces the Hall resistance to deviate from the quantized value.
We thank Michael Ma for helpful discussions on the longitudinal transport. This work was supported by the Research Grant Council of Hong Kong under Grant No. HKU7051/11P.
[10]{} N. Nagaosa, J. Sinova, S. Onoda, A. H. MacDonald, and N. P. Ong, Rev. Mod. Phys. **82**, 1539 (2010).
D. Xiao, M. C. Chang, and Q. Niu, Rev. Mod. Phys. **82**, 1959 (2010).
F. D. M. Haldane, Phys. Rev. Lett. **61**, 2015 (1988).
C. X. Liu, X. L. Qi, X. Dai, Z. Fang, and S. C. Zhang Phys. Rev. Lett. **101**, 146802 (2008).
R. Yu, W. Zhang, H. J. Zhang, S. C. Zhang, X. Dai, and Z. Fang, Science **329**, 61 (2010).
Z. Qiao, S. A. Yang, W. Feng, W. K. Tse, J. Ding, Y. Yao, J. Wang, and Q. Niu, Phys. Rev. B **82**, 161414(R) (2010).
X. L. Qi, Y. S. Wu, and S. C. Zhang, Phys. Rev. B **74**, 085308 (2006).
S. Onoda, N. Sugimoto, and N. Nagaosa, Phys. Rev. Lett. **97**, 126602 (2006).
K. Nomura and N. Nagaosa, Phys. Rev. Lett. **106**, 166802 (2011).
R. L. Chu, J. R. Shi, and S. Q. Shen, Phys. Rev. B **84**, 085312 (2011).
C. Z. Chang, J. Zhang, M. Liu, Z. Zhang, X. Feng, K. Li, L. L. Wang, X. Chen, X. Dai, Z. Fang, X. L. Qi, S. C. Zhang, Y. Wang, K. He, X. C. Ma, and Q. K. Xue, Adv. Mater. **25**, 1065 (2013).
J. G. Checkelsky, J. Ye, Y. Onose, Y. Iwasa, and Y. Tokura, Nat. Phys. **8**, 729 (2012).
Y. S. Hor, P. Roushan, H. Beidenkopf, J. Seo, D. Qu, J. G. Checkelsky, L. A. Wray, D. Hsieh, Y. Xia, S.-Y. Xu, D. Qian, M. Z. Hasan, N. P. Ong, A. Yazdani, and R. J. Cava, Phys. Rev. B **81**, 195203 (2010).
Y. L. Chen, J.-H. Chu, J. G. Analytis, Z. K. Liu, K. Igarashi, H.-H. Kuo, X. L. Qi, S. K. Mo, R. G. Moore, D. H. Lu, M. Hashimoto, T. Sasagawa, S. C. Zhang, I. R. Fisher, Z. Hussain, and Z. X. Shen, Science **329**, 659 (2010).
L. A. Wray, S. Y. Xu, Y. Xia, D. Hsieh, A. V. Fedorov, Y. S. Hor, R. J. Cava, A. Bansil, H. Lin, and M. Z. Hasan, Nat. Phys. **7**, 32 (2011).
C. Z. Chang, J. Zhang, X. Feng, J. Shen, Z. Zhang, M. Guo, K. Li, Y. Ou, P. Wei, L. L. Wang, Z. Q. Ji, Y. Feng, S. Ji, X. Chen, J. Jia, X. Dai, Z. Fang, S. C. Zhang, K. He, Y. Wang, L. Lu, X. C. Ma, and Q. K. Xue, Science **340**, 167 (2013).
H. J. Zhang, C. X. Liu, X. L. Qi, X. Dai, Z. Fang, and S. C. Zhang, Nat. Phys. **5**, 438 (2009).
S. Q. Shen, W. Y. Shan, and H. Z. Lu, SPIN **01**, 33 (2011).
H. Z. Lu, W. Y. Shan, W. Yao, Q. Niu, and S. Q. Shen, Phys. Rev. B **81**, 115407 (2010).
W. Y. Shan, H. Z. Lu, and S. Q. Shen, New J. Phys. **12** 043048 (2010).
See Supplementary Material for the calculation details.
J. Linder, T. Yokoyama, and A. Sudb[ø]{}, Phys. Rev. B **80**, 205401 (2009).
C. X. Liu, H. J. Zhang, B. Yan, X. L. Qi, T. Frauenheim, X. Dai, Z. Fang, and S. C. Zhang, Phys. Rev. B **81**, 041307(R) (2010).
Y. Zhang, K. He, C. Z. Chang, C. L. Song, L. L. Wang, X. Chen, J. F. Jia, Z. Fang, X. Dai, W. Y. Shan, S. Q. Shen, Q. Niu, X. L. Qi, S. C. Zhang, X. C. Ma, and Q. K. Xue, Nat. Phys. **6**, 584 (2010).
Y. Sakamoto, T. Hirahara, H. Miyazaki, S. I. Kimura, and S. Hasegawa, Phys. Rev. B **81**, 165432 (2010).
N. A. Sinitsyn, A. H. MacDonald, T. Jungwirth, V. K. Dugaev, and J. Sinova, Phys. Rev. B **75**, 045315 (2007).
S. A. Yang, H. Pan, Y. Yao, and Q. Niu, Phys. Rev. B **83**, 125122 (2011).
D. Culcer and S. Das Sarma, Phys. Rev. B **83**, 245441 (2011).
H. Z. Lu and S. Q. Shen, Phys. Rev. B **88**, 081304(R) (2013).
N. W. Ashcroft and N. D. Mermin, *Solid State Physics* (Saunders, Philadelphia, 1976).
A. N. Redlich, Phys. Rev. D **29**, 2366 (1984).
R. Jackiw, Phys. Rev. D **29**, 2375 (1984).
| {
"pile_set_name": "ArXiv"
} |
---
author:
- 'Brendan Hassett [^1]'
date: August 2004
title: Classical and minimal models of the moduli space of curves of genus two
---
Introduction
============
This paper is an introduction to the minimal model program, as applied to the moduli space of curves. Our long-term goal is a geometric description of the canonical model of the moduli space when it is of general type. This entails proving that the canonical model exists and interpreting it as a parameter space in its own right.
Work of Eisenbud, Harris, and Mumford shows that $M_g$ is of general type when $g\ge 24$ (see [@HaMu] and subsequent papers). A standard conjecture of birational geometry–the finite generation of the canonical ring–would imply the canonical model is $${\mathrm{Proj\ }}\oplus_{n\ge 0} \Gamma({\overline{M}}_g,nK_{{\overline{M}}_g}).$$ Unfortunately, this has yet to be verified in a single genus!
There is some cause for optimism: Shepherd-Barron [@SB] has recently shown that the canonical model of the moduli space of principally polarized abelian surfaces of dimension $g\ge 12$ is the first Voronoi compactification.
Another possible line of attack is to consider [*log*]{} canonical models of the moduli space. The moduli space is best regarded as a pair $({\overline{M}}_g,\Delta)$, where ${\overline{M}}_g$ is Deligne-Mumford compactification by stable curves and $\Delta$ is its boundary. It is implicit in the work of Mumford [@Mu1] that the moduli space of stable curves is its own log canonical model (see Theorem \[thm:Mumford\]). Our basic strategy it to interpolate between the log canonical model and the (conjectural) canonical model by considering $${\mathrm{Proj\ }}\oplus_{n\ge 0} \Gamma({\overline{M}}_g,n(K_{{\overline{M}}_g}+\alpha\Delta)),$$ where $\alpha \in {{\mathbb Q}}\cap [0,1]$ is chosen so that $K+\alpha\Delta$ is effective.
This program is a subject of ongoing work, inspired by correspondence with S. Keel, in collaboration with D. Hyeon. Future papers will address the stable behavior of these spaces for successively smaller values of $\alpha$. It is remarkable that their behavior is largely independent of the genus (see, for example, Remark \[rem:fullforce\].)
However, for small values of $g$ special complexities arise. When $g=2$ or $3$, the locus in ${\overline{M}}_g$ of curves with automorphism has codimension $\le 1$. To include these spaces under our general framework, we must take into account the properties of the moduli stack ${\overline{\mathcal M}}_g$. In particular, it is necessary to use the canonical divisor of the moduli stack rather than its coarse moduli space. These differ substantially, as the natural morphism ${\overline{\mathcal M}}_g {\rightarrow}{\overline{M}}_g$ is ramified at stable curves admitting automorphisms.
Luckily, we have inherited a tremendously rich literature on curves of small genus. The invariant-theoretic properties of $M_2$ were extensively studied by the 19th century German school [@Cl], who realized it as an open subset of the weighted projective space ${{\mathbb P}}(1,2,3,5)$. Theorem \[thm:main\] reinterprets this classical construction using the modern language of stacks and minimal models.
We work over an algebraically closed field $k$ of characteristic zero. We use the notation $\equiv$ for ${{\mathbb Q}}$-linear equivalence of divisors. Throughout, a [*curve*]{} is a connected, projective, reduced scheme of dimension one. The [*genus*]{} of a curve is its arithmetic genus.
The moduli stack of smooth (resp. stable) curves of genus $g$ is denoted ${{\mathcal M}}_g$ (resp. ${\overline{\mathcal M}}_g$); the corresponding coarse moduli scheme is denoted $M_g$ (resp. ${\overline{M}}_g$). The boundary divisors in ${\overline{M}}_g$ (resp. ${\overline{\mathcal M}}_g$) are denoted $$\Delta_0,\Delta_1,\ldots,\Delta_{\lfloor g/2 \rfloor}
(\text{resp. }
\delta_0,\delta_1,\ldots,\delta_{\lfloor g/2 \rfloor}).$$ Let $Q:{\overline{\mathcal M}}_g {\rightarrow}{\overline{M}}_g$ denote the natural morphism from the moduli stack to the coarse moduli space, so that $$Q^*\Delta_i=\delta_i, i\neq 1 \quad \quad
Q^*\Delta_1=2\delta_1.$$ We write $\delta=\sum_{0\le i \le g/2}\delta_i$; abusing notation, we also use $\delta$ for the corresponding divisor $$\Delta_0+1/2\Delta_1+\Delta_2+\ldots +\Delta_{\lceil g/2\rceil}$$ on ${\overline{M}}_g$.
[**Acknowledgments:**]{} The key ideas underlying this program were worked out in correspondence with S. Keel. They have been further developed in collaboration with D. Hyeon. The author also benefited from conversations with James Spencer about rigidification, quotient stacks, and moduli of curves of genus two, and from comments on the manuscript by Michael van Opstall. Part of this paper was prepared during a visit to the Mathematisches Institut of the Georg-August-Universität, Göttingen.
Classical geometry
==================
Elementary facts about curves of genus two {#subsect:element}
------------------------------------------
We recall results from standard textbooks, e.g., [@Ha] IV Ex 2.2 and §5. Let $C$ denote a smooth curve of genus two with sheaf of differentials $\omega_C$. The global sections of $\omega_C$ give the canonical morphism $$j:C {\rightarrow}{{\mathbb P}}(\Gamma(C,\omega_C))\simeq {{\mathbb P}}^1$$ which is finite of degree two. The corresponding covering transformation $\iota:C{\rightarrow}C$ is called the hyperelliptic involution. By the Hurwitz formula, $j$ is branched over six distinct points $$\{b_1,\ldots,b_6 \} \subset {{\mathbb P}}(\Gamma(C,\omega_C)).$$ On fixing an identification ${{\mathbb P}}(\Gamma(C,\omega_C))\simeq {{\mathbb P}}^1$, we can write down a nontrivial binary sextic form vanishing at the branch points $$F \in \Gamma({{\mathbb P}}^1,{{\mathcal O}}_{{{\mathbb P}}^1}(6)),$$ determined by $\{b_1,\ldots,b_6\} \in {{\mathbb P}}^1$ up to a scalar.
Conversely, suppose we have a binary sextic form $F$ with six distinct zeros $b_1,\ldots, b_6 \in {{\mathbb P}}^1$. Then there is a unique degree-two cover of ${{\mathbb P}}^1$ branched over these points. This is a smooth curve of genus two and the map to ${{\mathbb P}}^1$ is the canonical morphism. Moreover, the isomorphism class of $C$ depends only on the orbit of $F$ under the action of ${\mathrm{GL}}_2$.
To summarize: There is a one-to-one correspondence between isomorphism classes of curves of genus two and ${\mathrm{GL}}_2$-orbits of binary sextic forms with distinct zeros.
We will need a relative version of this dictionary, following [@Vi]. Let $\pi:{{\mathcal C}}{\rightarrow}S$ be a smooth morphism to a scheme of finite type over $k$, with fibers curves of genus two. Since the relative dualizing sheaf $\omega_{\pi}$ is globally generated, there is a relative double cover $$\begin{array}{rcccl}
{{\mathcal C}}& & \stackrel{j}{\longrightarrow} & &
{{\mathbb P}}^1(\pi_*\omega_{\pi}):={{\mathbb P}}\\
&{\scriptstyle \pi} \searrow & & \swarrow {\scriptstyle \psi} & \\
& & S & &
\end{array}$$ with associated involution $\iota$. Using the trace we decompose $j_*{{\mathcal O}}_{{{\mathcal C}}}\simeq {{\mathcal O}}_{{{\mathbb P}}}\oplus {{\mathcal L}}$, where ${{\mathcal L}}$ has relative degree $-3$ on the fibers of $\psi$. The ${{\mathcal O}}_{{{\mathbb P}}}$-algebra structure on ${{\mathcal O}}_{{{\mathcal C}}}$ is thus determined by an isomorphism ${{\mathcal L}}^2{\rightarrow}{{\mathcal O}}_{{{\mathbb P}}}$, i.e., by a nonvanishing section of ${{\mathcal L}}^{-2}$ with zeros along the branch locus of $j$. By relative duality $$j_*\omega_{\pi}={\mathscr{H}\!\mathit{om}}_{{{\mathcal O}}_{{{\mathbb P}}}}(j_*{{\mathcal O}}_{{{\mathcal C}}},\omega_{\psi})
\simeq \omega_{\psi} \oplus (\omega_{\psi}\otimes {{\mathcal L}}^{-1})$$ which yields $$\pi_*\omega_{\pi} \simeq \psi_*(\omega_{\psi}\otimes {{\mathcal L}}^{-1}).$$ Using the identifications $$\psi_*{{\mathcal O}}_{{{\mathbb P}}}(+1)=\pi_*\omega_{\pi}, \quad
\omega_{\psi}=(\psi^*\det \pi_*\omega_{\pi})(-2), \quad
{\mathrm{Pic}}({{\mathbb P}})={\mathrm{Pic}}(S)\oplus {{\mathbb Z}}c_1({{\mathcal O}}_{{{\mathbb P}}}(+1)),$$ we find $${{\mathcal L}}^{-1}\simeq {{\mathcal O}}_{{{\mathbb P}}}(+1)\otimes \omega_{\psi}^{-1}=
{{\mathcal O}}_{{{\mathbb P}}}(3)\otimes (\psi^* \det \pi_*\omega_{\pi})^{-1}.$$ The class of the branch divisor is thus $$\label{eq:Chern}
-2c_1({{\mathcal L}})=c_1({{\mathcal O}}_{{{\mathbb P}}}(6))-2\psi^* c_1( \det \pi_*\omega_{\pi}).$$
This has practical implications: An isomorphism of $C$ induces a linear transformation on $\Gamma(C,\omega_C)$, which respects the binary sextic $F$ up to a scalar. Formula (\[eq:Chern\]) allows us to keep track of this scalar. For each $M\in {\mathrm{GL}}_2$, we have the linear action $$(x,y) \mapsto (x,y)\left( \begin{matrix} m_{11} & m_{12} \\
m_{21} & m_{22}
\end{matrix} \right)$$ which induces a natural left action on binary sextic forms $$F \mapsto
F(xm_{11}+ym_{21}, xm_{12}+ym_{22}).$$ We normalize this action using formula (\[eq:Chern\]) $$\label{eq:action}
F \mapsto (M,F):=(\det M)^{-2}F(xm_{11}+ym_{21}, xm_{12}+ym_{22}),$$ so that $(M,F)=F$ if and only $M$ is induced from an automorphism of $C$.
A smooth curve is [*bielliptic*]{} if it admits a degree-two morphism $i: C{\rightarrow}E$ to an elliptic curve; the covering transformation is called a [*bielliptic involution*]{}. For curves of genus two any bielliptic involution commutes with the hyperelliptic involution, which yields a diagram $$\begin{array}{ccc}
C & \stackrel{i} {\rightarrow}& E \\
{\scriptstyle j} \downarrow \quad & &
\quad \downarrow
{\scriptstyle \bar{j}} \\
{{\mathbb P}}^1 & \stackrel{\bar{i}}{\rightarrow}& {{\mathbb P}}^1
\end{array}$$ where $\bar{i}$ and $\bar{j}$ are the double covers induced on quotients. The branch locus of $j$ is preserved by the covering transformation for $\bar{i}$, which is conjugate to $[x,y] \mapsto [y,x]$. The resulting involution of the branch locus will also be called a bielliptic involution. Thus $C$ is isomorphic to a double cover branched over $$\{[\alpha_1,1],[1,\alpha_1],[\alpha_2,1,][1,\alpha_2],[\alpha_3,1],
[1,\alpha_3]\}$$ for some $\alpha_1,\alpha_2,\alpha_3\in k$. Conversely, each such curve admits a diagram as above and thus is bielliptic.
Invariant theory of binary sextics {#subsect:invariant}
----------------------------------
We observe the classical convention for normalizing the coefficients of a binary sextic $$F=ax^6+6bx^5y+15cx^4y^2+20dx^3y^3+15ex^2y^4+6fxy^5+gy^6.$$ The action (\[eq:action\]) induces an action of ${\mathrm{GL}}_2$ on $k[a,b,c,d,e,f,g]$. Recall that a polynomial $P \in k[a,b,c,d,e,f,g]$ is [*${\mathrm{SL}}_2$-invariant*]{} if for each $M\in {\mathrm{SL}}_2$, we have $$(M,P) = P.$$ We write $$R:=k[a,b,c,d,e,f,g]^{{\mathrm{SL}}_2}$$ for the ring of such invariants. If $P$ is ${\mathrm{SL}}_2$-invariant then each homogeneous component of $P$ is as well, so $R$ is a graded ring. Every homogeneous invariant satisfies the functional relation $$\label{eq:functional}
(M,P)=(\det M)^{\deg(P)}P, \quad M \in {\mathrm{GL}}_2;$$ here it is essential that the action (\[eq:action\]) include the factor $(\det M)^2$. The transformation $$(x,y) {\rightarrow}(y,x)$$ thus reverses the sign of invariants of odd degree. These are called [*skew invariants*]{} in the classical literature.
Explicit generators for $R$ were first written down in the nineteenth century, e.g., [@Cl], pp. 296, in symbolic notation, [@Ca] and [@Sa] as explicit polynomials–the second edition of Salmon’s [*Higher algebra*]{} has the most detailed information, and also [@El] pp. 322. A nice early twentieth-century discussion is [@Sc] pp. 90 and a modern account invoking the representation theory of ${\mathrm{SL}}_2$ is [@Sp].
For our purposes, the symmetric function representation of the invariants in [@Ig2] pp. 176 and 185 is the most useful. Let $\xi_1,\ldots,\xi_6$ denote the roots of the dehomogenized form $F(x,1)$, and write $(ij)$ as shorthand for $\xi_i-\xi_j$. We write $$\begin{aligned}
A&=&a^2 \sum_{\text{fifteen}} (12)^2(34)^2(56)^2 \\
B&=&a^4 \sum_{\text{ten}} (12)^2(23)^2(31)^2(45)^2(56)^2(64)^2 \\
C&=&a^6 \sum_{\text{sixty}} (12)^2(23)^2(31)^2(45)^2(56)^2(64)^2
(14)^2(25)^2(36)^2 \\
D&=& a^{10}\prod_{ij}(ij)^2 \\
E&=& a^{15}\prod_{\text{fifteen}} \det \left(
\begin{matrix} 1 & \xi_1+\xi_2 & \xi_1\xi_2 \\
1 & \xi_3+\xi_4 & \xi_3\xi_4 \\
1 & \xi_5+\xi_6 & \xi_5\xi_6
\end{matrix}
\right) = a^{15} \prod_{\text{fifteen}} \left(
(14)(36)(52)-(16)(32)(54)\right)\end{aligned}$$ where the summations are chosen to make the expressions $\mathfrak{S}_6$-symmetric. Consequently, $A,B,C,D,$ and $E$ can all be expressed as polynomials in ${{\mathbb Q}}[a,b,c,d,e,f,g]$, e.g., $$\begin{aligned}
A&=&-240(ag-6bf+15ce-10d^2)\\
B&=&-162000\det \left(\begin{matrix} a & b & c & d \\
b & c & d & e \\
c& d & e & f \\
d & e & f & g
\end{matrix} \right) + 1620 (ag-6bf+15ce-10d^2)^2.\end{aligned}$$ In classical terminology, $(ag-6bf+15ce-10d^2)$ is the [*sixth transvectant of $F$ over itself*]{}; transvection is one of the main operations in Gordan’s proof of finiteness for invariants of binary forms. The determinantal expression is the [*catalecticant*]{} of $F$: It vanishes precisely when $F$ can be expressed as a sum of three sixth powers [@El] pp. 276.
The following facts will be useful for subsequent analysis:
\[prop:inv\]
1. [The expressions $A,B,C,D,$ and $E$ are invariant and generate $R$ [@Hi], pp. 100, [@Cl], etc.]{}
2. [$D$ is the discriminant and vanishes precisely when the binary form has a multiple root.]{}
3. [$B,C,D,$ and $E$ vanish whenever the binary form has a triple root; $A$ vanishes when the form has a quadruple root.]{}
4. [$E$ vanishes if and only if the form admits a bielliptic involution, as defined in §\[subsect:element\] [@El], pp. 327 and [@Cl], pp. 457.]{}
5. [The unique irreducible relation among the invariants is $$E^2=G(A,B,C,D),$$ where $G$ is weighted-homogeneous of degree $30$ [@Cl], pp. 299.]{}
The notation used for the generating invariants is not consistent among authors. Our notation is consistent with that of Igusa, but inconsistent with Clebsch’s and Salmon’s. Of course, the invariants of degree two and fifteen are unique up to scalar.
The projective invariant-theory quotient
----------------------------------------
We consider $$X:={\mathrm{Proj\ }}R= {\mathrm{Proj\ }}\frac{k[A,B,C,D,E]}{\left<E^2-G(A,B,C,D)\right>}.$$ If $A=B=C=D=0$ then $E=0$ as well, so $X$ is covered by the distinguished affine open subsets $$\{A\neq 0 \} \quad \{B \neq 0 \} \quad \{C \neq 0 \}
\quad \{ D\neq 0 \}.$$ However, in each localization $$(k[A,B,C,D,E][A^{-1}])_0 \quad
(k[A,B,C,D,E][B^{-1}])_0$$ $$(k[A,B,C,D,E][C^{-1}])_0 \quad
(k[A,B,C,D,E][D^{-1}])_0$$ only even powers of $E$ appear, so all the functions over these distinguished open subsets can be expressed in terms of $A,B,C,D$. In light of Proposition \[prop:inv\], we find
\[prop:weight1\]
1. [$X\simeq
{\mathrm{Proj\ }}k[A,B,C,D] \simeq {{\mathbb P}}(2,4,6,10)\simeq {{\mathbb P}}(1,2,3,5)$ [@Ig2], pp. 177.]{}
2. [A binary sextic with a zero of multiplicity three, admitting a nonvanishing invariant of positive degree, is mapped to $p:=[1,0,0,0,0] \in X$.]{}
3. [All positive-degree invariants vanish at binary sextics with a zero of multiplicity four; they do not yield points of $X$. ]{}
Geometric Invariant Theory gives an interpretation of the points of $X$:
\[prop:GIT\]
1. [a binary sextic is stable (resp. semistable) if and only if its zeros have multiplicity $\le 2$ (resp. $\le 3$) [@GIT], ch. 4 §1;]{}
2. [$X - \{[1,0,0,0,0]\}$ is a geometric quotient for binary sextics with zeros of multiplicity $\le 2$ [@GIT] 1.10.]{}
The ‘only if’ part of the first assertion can be deduced from Proposition \[prop:weight1\]. As $X-\{D=0\}$ is a geometric quotient for binary sextics with distinct zeros, our analysis of genus two curves in §\[subsect:element\] yields
\[prop:m2\] The moduli scheme $M_2$ can be identified with $X-\{D=0\}$, where $D$ is the discriminant.
This construction definitely fails in characteristic two. If the double cover $j:C{\rightarrow}{{\mathbb P}}^1$ is wildly ramified, the branch divisor may have multiplicities $>3$. These curves correspond to unstable points under the ${\mathrm{SL}}_2$-action, and thus are not represented in the invariant-theory quotient. [@Ig1] has a detailed account of what must be done in this case.
Invariant-theory quotient as a contraction
------------------------------------------
We sketch the relationship between the invariant-theory quotient and the moduli space of stable curves.
A birational map of normal projective varieties $$\beta:Y \dashrightarrow X$$ is a [*contraction*]{} if $\beta^{-1}$ has no exceptional divisors, i.e., the proper transform of each codimension-one subset in $X$ has codimension one in $Y$.
\[prop:bircont\] There exists a birational contraction $\beta:{\overline{M}}_2\dashrightarrow X$ restricting to the identity along the open subset $M_2$. $\beta$ is an isomorphism over ${\overline{M}}_2 - \Delta_1$ and contracts $\Delta_1$ to the point $p$.
[*proof:*]{} To produce the birational contraction, we exhibit a morphism $$\beta^{-1}:U \hookrightarrow {\overline{M}}_2$$ where $U\subset X$ is open with complement of codimension $\ge 2$ and $\beta^{-1}|_{M_2\cap U}$ is the identity. We shall take $U=X-p$, where $p$ corresponds to the binary forms with a triple zero (cf. Proposition \[prop:weight1\].)
The universal binary sextic is a hypersurface $$W:=\{ax^6+6bx^5y+15cx^4y^2+20dx^3y^3+15ex^2y^4+6fxy^5+gy^6=0\}
\subset {{\mathbb A}}^7 \times {{\mathbb P}}^1.$$ Its class in $\mathrm{Pic}({{\mathbb A}}^7 \times {{\mathbb P}}^1)$ is divisible by two, so there exists a double cover ${{\mathcal C}}'{\rightarrow}{{\mathbb A}}^7\times {{\mathbb P}}^1$ simply branched over $W$. Composing with the projection onto the first factor, we obtain a morphism $$\pi':{{\mathcal C}}' {\rightarrow}{{\mathbb A}}^7.$$ Let $S\subset {{\mathbb A}}^7$ denote the open subset corresponding to forms whose zeros all have multiplicity $\le 2$ and $$\pi:{{\mathcal C}}{\rightarrow}S$$ the restriction of $\pi'$ to $S$. Since $\pi$ is a composition of flat morphisms, it is also flat.
Consider the fiber of $\pi$ over a given binary sextic $F$: It is a double cover $j:{{\mathcal C}}_F{\rightarrow}{{\mathbb P}}^1$ branched over the zeros of $F$. We claim ${{\mathcal C}}_F$ is a stable curve of genus two, not contained in $\Delta_1$. Evidently ${{\mathcal C}}_F$ is smooth and simply branched over the zeros with multiplicity one. Over the double zeros ${{\mathcal C}}_F$ has local equation $y^2=x^2$, which defines a node. We have $j^*{{\mathcal O}}_{{{\mathbb P}}^1}(+1)=\omega_{{{\mathcal C}}_F}$, which is ample on ${{\mathcal C}}_F$, so ${{\mathcal C}}_F$ is stable. The normalization $\nu:{{\mathcal C}}^{\nu}_F{\rightarrow}{{\mathcal C}}_F$ is the double cover branched along $F'=0$, where $F'$ is the product of the factors of $F$ with multiplicity one; ${{\mathcal C}}_F$ is obtained from ${{\mathcal C}}^{\nu}_F$ by gluing the pairs of points over the each double root of $F$. There are three possibilities:
1. [$\deg(F')=4$, in which case ${{\mathcal C}}^{\nu}_F$ is connected of genus one;]{}
2. [$\deg(F')=2$, in which case ${{\mathcal C}}^{\nu}_F$ is connected of genus zero;]{}
3. [$\deg(F')=0$, in which case ${{\mathcal C}}^{\nu}_F$ has two connected components of genus zero.]{}
Since ${{\mathcal C}}_F$ cannot be expressed as the union of two subcurves of genus one meeting at a point, the resulting curve is not in $\Delta_1$.
The classifying morphism $$S{\rightarrow}{\overline{\mathcal M}}_2-\delta_1$$ is equivariant with respect to the ${\mathrm{GL}}_2$-action on binary sextics, and therefore descends to a morphism $$\Phi:S/{\mathrm{GL}}_2 {\rightarrow}{\overline{\mathcal M}}_2-\delta_1.$$ We remark that this is a morphism of stacks. Since $U$ is a geometric quotient for binary sextics with zeros of multiplicity $\le 2$ (see Proposition \[prop:GIT\]), $U$ is also the coarse moduli space for $S/{\mathrm{GL}}_2$. We define $\beta^{-1}$ to be the induced morphism on coarse moduli spaces.
It remains to show this is bijective onto its image. Suppose we are given a stable curve $C$ of genus two not contained in $\Delta_1$. Quite generally, $\omega_C$ is globally generated for any stable curve without disconnecting nodes; the only curves in ${\overline{M}}_2$ with disconnecting nodes lie in $\Delta_1$. Thus the sections of $\omega_C$ give a double cover $$j:C {\rightarrow}{{\mathbb P}}^1$$ branched along a sextic, with zeros of multiplicity $\le 2$ because $C$ is nodal. The analysis above shows that every such sextic arises in this way. $\square$
Blowing up the invariant-theory quotient
----------------------------------------
We recall the principal result of [@Ig2]. Let $A_g$ denote the moduli space of principally polarized abelian varieties of dimension $g$, ${\overline{A}}_g$ its Satake compactification. Recall that ${\overline{A}}_g={\mathrm{Proj\ }}S$, where $S$ is the ring of $\mathrm{Sp}(g,{{\mathbb Z}})$-modular forms; we use $\lambda$ to denote the resulting polarization on ${\overline{A}}_g$. Let $t:M_g \hookrightarrow A_g$ denote the Torelli morphism, associating to each curve its Jacobian.
Now assume $g=2$. Regarding $M_2$ as an open subset of $X$ (see Proposition \[prop:m2\]), $t$ extends to a rational map $$\tau:X \dashrightarrow {\overline{A}}_2.$$ The inclusion and the Torelli morphism induce $$\label{eq:compact}
M_2 \hookrightarrow \widetilde{X}:={\overline{\mathrm{Graph}(\tau)}} \subset
X \times {\overline{A}}_2.$$ In particular, $\widetilde{X}$ compactifies $M_2$.
\[thm:Igusa\][@Ig2] The indeterminacy of $\tau$ is the point $p=[1,0,0,0,0]\in X$ corresponding to binary sextic forms with a zero of multiplicity three. If we choose local coordinates at $p$ $$x_1=2^43^2B/A^2 \quad x_2=2^63^3(3C-AB)/A^3 \quad x_3=2\cdot3^5D/A^5$$ then $\tau$ is resolved by a weighted blow-up centered at $p$ $$b:\widetilde{X} {\rightarrow}X$$ with weights $$\text{weight}(x_1)=2 \quad
\text{weight}(x_2)=3 \quad
\text{weight}(x_3)=6.$$ The exceptional divisor of $b$ is mapped isomorphically to the locus of principally polarized abelian surfaces that decompose as a product of two elliptic curves (with the induced product polarization).
\[rem:Igusa\] Igusa’s result is considerably more precise: He explicitly computes the correspondence between the ring of invariants $R$ and the ring of modular forms $S$. In particular, $S$ is a polynomial ring with generators in degrees $4,6,10,12$ and the locus of products is given by the vanishing of a form of weight $10$.
Comparing the blow-up with moduli space {#subsect:comp}
---------------------------------------
\[prop:codimone\] The open imbedding $M_2 \hookrightarrow \widetilde{X}$ extends to a birational map $$\gamma: {\overline{M}}_2 \dashrightarrow \widetilde{X}$$ which is an isomorphism in codimension one.
In particular, $\gamma$ and $\gamma^{-1}$ are both birational contractions. In Proposition \[prop:nef2\] we will prove that $\gamma$ is an isomorphism.
[*proof:*]{} The Torelli morphism admits an extension $\overline{t}: {\overline{M}}_g {\rightarrow}{\overline{A}}_g$ [@Na1], Theorem 3. This is not an isomorphism for $g>1$: The divisor $\Delta_0 \subset {\overline{M}}_g$ is mapped to a boundary stratum of ${\overline{A}}_g$, which has codimension $\ge 2$. However, in genus two $\overline{t}$ is an isomorphism at the generic point of $\Delta_1$. Indeed, the Jacobian of a curve $[E_1\cup_q E_2] \in \Delta_1$, with $E_1$ and $E_2$ smooth of genus one, is the abelian surface $E_1\times E_2$.
The following diagram summarizes the various birational maps and morphisms: $$\begin{array}{ccc}
{\overline{A}}_2 &\stackrel{\pi}{\leftarrow} & \widetilde{X} \\
{\scriptstyle \overline{t}} \uparrow \quad & &
\quad \downarrow {\scriptstyle b} \\
{\overline{M}}_2 &\stackrel{\beta}{\dashrightarrow} & X
\end{array}.$$ By Theorem \[thm:Igusa\], $\pi$ is also an isomorphism over the generic point of $\overline{t}(\Delta_1)$, so $\gamma$ is an isomorphism at the generic point of $\Delta_1$. $\beta$ and $b$ are both isomorphisms over the generic point of the divisor $\beta(\Delta_0)$ (see Proposition \[prop:bircont\]), so $\gamma$ is an isomorphism at the generic point of $\Delta_0$. Since $\gamma$ is regular along $M_2={\overline{M}}_2-\Delta_0-\Delta_1$, the result follows. $\square$
The proper transforms of $\Delta_0$ and $\Delta_1$ in $\widetilde{X}$ are denoted $\widetilde{\Delta}_0$ and $\widetilde{\Delta}_1$. Thus $\widetilde{\Delta}_1$ is the exceptional divisor of $b:\widetilde{X} {\rightarrow}X$.
There are a number of partial desingularizations of ${\overline{A}}_g$ through which $\overline{t}$ factors, e.g., the ‘Igusa monoidal transform’ [@Ig3], [@Na1] and the toroidal compactification associated to the 2nd Voronoi fan [@Na2]. When $g=2$, these approaches coincide [@Na2] Remark 2.8 and yield a partial desingularization $\widetilde{A}_2 {\rightarrow}{\overline{A}}_2.$ See [@Ig3] Theorem 5 for a blow-up representation, expressed in terms of modular forms; the center of this blow-up is in the boundary ${\overline{A}}_2 - A_2$. Namikawa [@Na1] §9 has shown that the factorization ${\overline{M}}_2 \stackrel{\sim}{{\rightarrow}} \widetilde{A}_2$ is an isomorphism. Notwithstanding Igusa’s explicit formulas for $\tau:X \dashrightarrow {\overline{A}}_2$ and $\widetilde{A}_2 {\rightarrow}{\overline{A}}_2$, it is not entirely obvious how to extract an isomorphism $\widetilde{A}_2 \stackrel{\sim}{{\rightarrow}} \widetilde{X}.$
Stack geometry
==============
A stack-theoretic quotient
--------------------------
Proposition \[prop:weight1\] might suggest that the invariant $E$ is irrelevant to the geometry of the quotient. However, we have so far ignored possible [*stack structures*]{} on the quotient, which are intertwined with the geometry of $E$. There are a number of natural stacks to consider, including the ${\mathrm{GL}}_2$-quotient stack. Our choice is dictated by pedagogical imperatives, i.e., to exhibit concretely the nontrivial inertia along the bielliptic locus where $E$ vanishes.
The ring of invariants $R$ is graded by degree, so we have a natural ${{\mathbb G}}_m$-action on the affine variety $Y={\mathrm{Spec\ }}R$. Now ${{\mathbb G}}_m$ acts on the open subset $Y-(0,0,0,0,0)$ with finite stabilizers and closed orbits, so the quotient stack $${{\mathcal X}}:=\left( Y-(0,0,0,0,0) \right)/{{\mathbb G}}_m$$ is a separated Deligne-Mumford stack with coarse moduli space $q:{{\mathcal X}}{\rightarrow}X$ (see [@LM] 10.13.2,7.6,8.1 for more information). The points of $Y-(0,0,0,0,0)$ with nontrivial stabilizer map to the points of ${{\mathcal X}}$ with nontrivial inertia groups; this is the ramification locus of $q$.
We collect some geometric properties of ${{\mathcal X}}$:
\[prop:ramq\]
1. [The closed imbedding ${\mathrm{Spec\ }}R \hookrightarrow {\mathrm{Spec\ }}k[A,B,C,D,E] $ induces a closed imbedding $i:{{\mathcal X}}\hookrightarrow {{\mathcal P}}(2,4,6,10,15)$, where $${{\mathcal P}}(2,4,6,10,15):= \left({\mathrm{Spec\ }}k[A,B,C,D,E]-(0,0,0,0,0)\right)/{{\mathbb G}}_m,$$ with ${{\mathbb G}}_m$ acting with weights $(2,4,6,10,15)$ $$t\cdot (A,B,C,D,E) \mapsto (t^2A,t^4B,t^6C,t^{10}D,t^{15}E);$$]{}
2. [the dualizing sheaf of ${{\mathcal X}}$ is given via adjunction $$\omega_{{{\mathcal X}}}=i^*\omega_{{{\mathcal P}}(2,4,6,10,15)}({{\mathcal X}})\simeq
i^*{{\mathcal O}}_{{{\mathcal P}}(2,4,6,10,15)}(-7),$$ where ${{\mathcal O}}_{{{\mathcal P}}(2,4,6,10,15)}(+1)$ is the invertible sheaf associated with the principal ${{\mathbb G}}_m$-bundle classified by the identity character of ${{\mathbb G}}_m$;]{}
3. [the ramification divisor of $q$ is $${{\mathcal E}}:=\{ E=0 \} \subset {{\mathcal X}}$$ and we have $$q^*\omega_X\simeq \omega_{{{\mathcal X}}}(-{{\mathcal E}})\simeq i^*{{\mathcal O}}_{{{\mathcal P}}(2,4,6,10,15)}(-22).$$]{}
[*proof:*]{} The first two assertions do not require proof. As for the third, it follows from the classification of possible automorphisms of binary sextics [@Bo], [@Ig1] §8 The only automorphism type occuring in codimension one is the bielliptic involution (cf. Proposition \[prop:inv\]). $\square$
Notwithstanding Proposition \[prop:m2\], ${{\mathcal M}}_2$ is [*not*]{} contained in ${{\mathcal X}}$ as an open substack. Using the functional relation (\[eq:functional\]), the inertia group of $[F]\in {{\mathcal X}}-p$ is the quotient $$\{M\in {\mathrm{GL}}_2: (M,F)=F \}/\{M\in {\mathrm{SL}}_2: (M,F)=F \},$$ which is trivial for generic binary forms.
The inertia group at $[C]\in {{\mathcal M}}_2$ is ${\mathrm{Aut}}(C)$, and the presence of the hyperelliptic involution $\iota$ means this is always nontrivial. Now ${\mathrm{Aut}}(C)$ has a natural representation on $\Gamma(C,\omega_C)$ and an induced representation on $\wedge^2\Gamma(C,\omega_C)$ that is not faithful: We do not see elements of ${\mathrm{Aut}}(C)$ acting on $\Gamma(C,\omega_C)$ with determinant one, e.g., $\iota$, which acts on $\Gamma(C,\omega_C)$ by $-I$. The corresponding quotient of ${\mathrm{Aut}}(C)$ is the inertia picked up by ${{\mathcal X}}$.
Analysis of boundary divisors in the moduli stack {#subsect:stackgeo}
-------------------------------------------------
\[lem:involution\] Every stable curve of genus two admits a canonical hyperelliptic involution, which is central in its automorphism group.
[*proof:*]{} We claim that every stable curve of genus two is canonically a double cover of a nodal curve of genus zero $$j:C {\rightarrow}R, \quad
R=\begin{cases} {{\mathbb P}}^1 & \text{ if } [C]\not \in \Delta_1\\
{{\mathbb P}}^1 \cup_r {{\mathbb P}}^1 & \text{ if } [C] \in \Delta_1
\end{cases}$$ branched over six smooth points $b_1,\ldots,b_6 \in R$ and the node $r\in R$. The covering transformation $\iota$ therefore commutes with each automorphism of $C$.
The cover is induced by $$C {\rightarrow}{{\mathbb P}}(\Gamma(C,\omega^2_C))\simeq {{\mathbb P}}^2$$ which factors $$C \stackrel{j}{{\rightarrow}} R \subset {{\mathbb P}}^2,$$ where $R$ is a plane conic and $j$ is finite of degree two. Indeed, for curves not in $\Delta_1$ this is the double cover $C{\rightarrow}{{\mathbb P}}^1$ discussed in the proof of Proposition \[prop:bircont\]. For curves $C=E_1\cup_q E_2 \in \Delta_1$ with $q=E_1\cap E_2$ the disconnecting node joining the genus-one components $E_1$ and $E_2$, we have a double cover $$j:E_1\cup_q E_2 \longrightarrow {{\mathbb P}}^1 \cup_r {{\mathbb P}}^1,$$ with $j(q)=r$ and $j$ mapping the genus one components two-to-one onto rational components, with ramification at $q$ along each component. $\square$
Consider stable curves of genus two with automorphisms beyond $\iota$. There are two possibilities: Either $(R,b_1,\ldots,b_6)$ admits automorphisms permuting the $b_i$ or $j:C {\rightarrow}R$ admits covering transformations other than the canonical hyperelliptic involution. The classification of automorphism groups ([@Bo] or [@Ig1]§8) yields the following possibilities in codimension one:
1. [the curves $C$ in $\Delta_1$; here $j:C {\rightarrow}{{\mathbb P}}^1 \cup {{\mathbb P}}^1$ admits involutions fixing each component of $C$;]{}
2. [the closure of the locus of curves $j:C {\rightarrow}{{\mathbb P}}^1$ branched over six points admitting a bielliptic involution.]{}
At each point $[C]\in {\overline{M}}_2$, the moduli space is étale-locally isomorphic to the quotient $T_{[C]}{\overline{\mathcal M}}_2/{\mathrm{Aut}}(C)$ at the origin, where $$T_{[C]}{\overline{\mathcal M}}_2=\mathrm{Ext}^1(\Omega^1_C,{{\mathcal O}}_C))$$ is the tangent space with the induced automorphism action. When $C$ is smooth, Serre duality gives $$T_{[C]}{\overline{\mathcal M}}_2=\Gamma(C,\omega^2_C)^*;$$ equation (\[eq:action\]) shows that the hyperelliptic involution acts trivially and a bielliptic involution acts by reflection.
This local isomorphism can be chosen so the divisors $\Delta_0$ and $\Delta_1,$ correspond to the images of unions of distinguished hyperplanes in $T_{[C]}{\overline{\mathcal M}}_2$. The local-global spectral sequence gives an exact sequence $$\label{ex:locglob}
0 {\rightarrow}H^1({\mathscr{H}\!\mathit{om}}(\Omega^1_C,{{\mathcal O}}_C)) {\rightarrow}\mathrm{Ext}^1(\Omega^1_C,{{\mathcal O}}_C) {\rightarrow}\Gamma({\mathscr{E}\!\mathit{xt}}^1(\Omega^1_C,{{\mathcal O}}_C)){\rightarrow}0.$$ A local computation implies $${\mathscr{E}\!\mathit{xt}}^1(\Omega^1_C,{{\mathcal O}}_C) \simeq
\oplus_{\text{nodes }p\in C} k.$$ If $C=E_1\cup_q E_2$ is in $\Delta_1$, let $\widehat{\delta_1}\subset T_{[C]}{\overline{\mathcal M}}_2$ denote the hyperplane corresponding to the node disconnecting the genus-one components of $C$, i.e., the kernel of the projection onto the direct summand corresponding to $q$. The extra covering involutions of $j:C{\rightarrow}{{\mathbb P}}^1\cup {{\mathbb P}}^1$ mentioned above act on $T_{[C]}{\overline{\mathcal M}}_2$, trivially on $\widehat{\delta_1}$ and by multiplication by $(-1)$ on ${\mathscr{E}\!\mathit{xt}}^1(\Omega^1_C,{{\mathcal O}}_C)_q$. This corresponds to reflection across $\widehat{\delta_1}$. Let $\widehat{\delta_0}$ denote the union of the hyperplanes corresponding to each of the non-disconnecting nodes of $C$. The surjectivity of the last arrow in (\[ex:locglob\]) means that $\widehat{\delta}=\widehat{\delta_1}\cup \widehat{\delta_0}$ is normal crossings.
Let $\xi \subset {\overline{\mathcal M}}_2$ (resp. $\Xi \subset {\overline{M}}_2$) denote the closure of the smooth curves admitting a bielliptic involution.
These are irreducible of codimension one, e.g., by Proposition \[prop:inv\] and the characterization of bielliptic curves. This can also be seen infinitesimally: Under the local identification with $T_{[C]}{\overline{\mathcal M}}_2/{\mathrm{Aut}}(C)$, each branch of $\xi$ is identified with the hyperplane $T_{[C]}{\overline{\mathcal M}}_2$ fixed by the corresponding bielliptic involution (acting by reflection on the tangent space). The union of these hyperplanes is denoted $\widehat{\xi}$.
The bielliptic divisor has more complicated local geometry, as $\Xi$ may have quite a few local branches. For example, $$F(x,y)=xy(x-\xi y)(x+\xi y)(x - \xi^{-1}y)(x+\xi^{-1}y)$$ has automorphism group isomorphic to the Klein four-group and admits two involutions $$[x,y]\mapsto [y,x] \quad
[x,y] \mapsto [-y, x].$$ The cyclotomic form $$F(x,y)=x^6+y^6=(x-\zeta y)(x-\zeta^3 y)(x-\zeta^5 y)(x-\zeta^7 y)
(x-\zeta^9 y)(x-\zeta^{11}y) \quad \zeta^{12}=1$$ has automorphism group isomorphic to the dihedral group with $12$ elements and admits [*four*]{} distinct involutions $$[x,y]\mapsto [y,x] \quad
[x,y]\mapsto [-x,y] \quad
[x,y] \mapsto [\zeta^4 y, x] \quad
[x,y] \mapsto [\zeta^8 y, x] .$$ In particular, $\widehat{\xi}$ is [*not*]{} normal crossings.
\[prop:ramify\] $$K_{{\overline{\mathcal M}}_2}+\alpha \delta \equiv Q^*(K_{{\overline{M}}_2}+ \alpha \Delta_0+
\frac{1+\alpha}{2}\Delta_1+\frac{1}{2}\Xi)$$
[*proof:*]{} Lemma \[lem:involution\] allows us to consider the rigidification ([@ACV] §5) of ${\overline{\mathcal M}}_2$ with respect to the group $\left<\iota\right>$ generated by the canonical involution $$r:{\overline{\mathcal M}}_2 {\rightarrow}{{\overline{\mathcal M} }^{<\iota >}_2}.$$ Given a scheme $T$, $T$-valued points of ${{\overline{\mathcal M} }^{<\iota >}_2}$ correspond to families of genus two stable curves over $T$, where two families are identified if they differ by a $\left<\iota\right>$-valued cocycle over $T$. The inertia group of ${{\overline{\mathcal M} }^{<\iota >}_2}$ at $[C]$ is the quotient $${\mathrm{Aut}}(C)/\left< \iota \right>.$$ We have a factorization $${\overline{\mathcal M}}_2 \stackrel{r}{{\rightarrow}} {{\overline{\mathcal M} }^{<\iota >}_2}\stackrel{\bar{q}}{{\rightarrow}} {\overline{M}}_2$$ so that $r$ is étale of degree two and ${\overline{M}}_2$ is the coarse moduli space of ${{\overline{\mathcal M} }^{<\iota >}_2}$ [@ACV] Theorem 5.1.5. We therefore obtain the following formula for dualizing sheaves $$r^*\omega_{{{\overline{\mathcal M} }^{<\iota >}_2}}\simeq \omega_{{\overline{\mathcal M}}_2}.$$
Let $\delta_1^{<\iota>}$ and $\xi^{<\iota>}$ denote the corresponding Cartier divisors in ${{\overline{\mathcal M} }^{<\iota >}_2}$. As $\bar{q}$ has simple ramification along these divisors, we obtain $$\begin{aligned}
\bar{q}^*\Delta_1=2\delta_1^{<\iota >} & &
\bar{q}^*\Xi=2\xi^{<\iota >} \\
{\bar{q}}^*K_{{\overline{M}}_2}&=&K_{{{\overline{\mathcal M} }^{<\iota >}_2}}-\delta_1^{<\iota>}-\xi^{<\iota>},\end{aligned}$$ which together imply the formula.$\square$
Analysis of $b:\widetilde{X}{\rightarrow}X$ along the exceptional divisor
-------------------------------------------------------------------------
By Theorem \[thm:Igusa\], the exceptional divisor $\widetilde{\Delta}_1$ is mapped isomorphically to the locus in ${\overline{A}}_2$ parametrizing abelian surfaces decomposing into products of elliptic curves (as a principally polarized abelian variety); this is isomorphic to ${{\mathbb P}}(2,3,6)$.
\[prop:symsquare\] Let $\widetilde{\Xi}$ denote the proper transform of $\Xi$ in $\widetilde{X}$. We have the following formulas: $$\begin{aligned}
K_{\widetilde{X}}&\equiv& b^*K_X+10\widetilde{\Delta}_1\\
b^*\{G=0\}&\equiv& \widetilde{\Xi}
+12\widetilde{\Delta}_1\\
b^*\{D=0\}& \equiv & \widetilde{\Delta}_0
+6\widetilde{\Delta}_1 \\
\widetilde{\Xi}&\equiv&
3\widetilde{\Delta}_0+12\widetilde{\Delta}_1.\end{aligned}$$
We pause to explore the geometry of $\widetilde{\Delta}_1$. Naively, one might expect this to be the symmetric square of the moduli space of elliptic curves. However, in taking symmetric squares we should be mindful of the stack structures. The coarse moduli space of the symmetric square need not be isomorphic to the symmetric square of the coarse moduli space.
The standard theory of modular forms implies $${\overline{\mathcal M}}_{1,1}\simeq {{\mathcal P}}(4,6)=
\left({\mathrm{Spec\ }}k[g_2,g_3] - (0,0)\right)/{{\mathbb G}}_m,$$ with ${{\mathbb G}}_m$ acting with weights $(4,6)$ $$t\cdot (g_2,g_3) \mapsto (t^4g_2,t^6g_3).$$ The coarse moduli space is $${\overline{M}}_{1,1}={\overline{A}}_1 \simeq {\mathrm{Proj\ }}k[g_2,g_3]\simeq {{\mathbb P}}^1.$$ The symmetric square of the stack has the following quotient-stack presentation: $$\begin{aligned}
({\overline{\mathcal M}}_{1,1} \times {\overline{\mathcal M}}_{1,1})/{\mathfrak{S}}_2
&=&\left({\mathrm{Spec\ }}k[g_2,g_3,h_2,h_3]-Z\}\right/H \\
Z&=&\{(g_2,g_3,h_2,h_3): g_2=g_3=0 \text{ or } h_2=h_3=0\}.\end{aligned}$$ Here $H$ is the group generated by the torus $$(t,u)\cdot(g_2,g_3,h_2,h_3)\mapsto
(t^4g_2,t^6g_3,u^4h_2,u^6h_3)$$ and the involution $$(g_2,g_3,h_2,h_3) \mapsto (h_2,h_3,g_2,g_3),$$ i.e., $H={\mathfrak{S}}_2 \ltimes {{\mathbb G}}_m^2$ where ${\mathfrak{S}}_2$ acts on ${{\mathbb G}}_m^2$ by interchanging the factors.
The coarse moduli space of the stack is the invariant-theory quotient for the action of $H$. Consider the elements $p\in k[g_2,g_3,h_2,h_3]$ with the following properties:
1. [$p(g_2,g_3,h_2,h_3)=p(h_2,h_3,g_2,g_3)$;]{}
2. [$p(t^2g_2,t^3g_3,u^2h_2,u^3h_3)=(tu)^N p(g_2,g_3,h_2,h_3)$ for some $N$.]{}
This ring is generated by $g_2h_2,g_3h_3,$ and $g_2^3h_3+g_3^2h_2^3$ and $${\mathrm{Proj\ }}k[g_2h_2,g_3h_3,g_2^3h_3^2+g_3^2h_2^3] \simeq {{\mathbb P}}(4,6,12)
\simeq {{\mathbb P}}(2,3,6),$$ which explains why the weights of $b$ are $(2,3,6)$. See [@Ig2], Theorem 3, for a discussion in terms of the modular forms for $\mathrm{Sp}(2,{{\mathbb Z}})$ (see Remark \[rem:Igusa\]).
[*proof of proposition:*]{} The first equation follows because $b:\widetilde{X}{\rightarrow}X$ has weights $(2,3,6)$. As for the second, $\widetilde{\Xi}$ is the proper transform of the divisor $\{G=0\} \subset X$ parametrizing forms admitting an bielliptic involution. When a bielliptic curve of genus two specializes to a stable curve in $\Delta_1$, the bielliptic involution specializes to a morphism exchanging the elliptic components. Therefore, $\widetilde{\Xi} \cap \widetilde{\Delta}_1 \subset
\widetilde{\Delta}_1 $ is the diagonal in the symmetric square, which is cut out by a form of weighted-degree twelve.
For the third equation, $\widetilde{\Delta}_0$ is the proper transform of $\{D=0\}$. The intersection $\widetilde{\Delta}_0 \cap \widetilde{\Delta}_1 \subset
\widetilde{\Delta}_1$ is the locus where the discriminant $\Delta=g_2^3-27g^2_3$ vanishes, and thus has weighted-degree six. The last equation follows because $G$ has weighted degree thirty and $D$ has weighted degree ten.$\square$
One consequence of this analysis is worth mentioning.
\[prop:blowpol\] For sufficiently small $\epsilon>0$, the divisor $$\widetilde{\Delta}_0+(6-\epsilon)\widetilde{\Delta}_1$$ is ample on $\widetilde{X}$.
[*proof:*]{} The divisor can be expressed $$b^*(\text{ample divisor})-\epsilon(\text{$b$-exceptional divisor}),$$ which yields a polarization of the blow-up $b:\widetilde{X}{\rightarrow}X$. $\square$
Birational geometry
===================
Divisor classes and birational contractions of ${\overline{M}}_2$ {#subsect:canM2}
-----------------------------------------------------------------
It is well known that the rational divisor class group of ${\overline{M}}_2$ is freely generated by the boundary divisors $\Delta_0$ and $\Delta_1$; see [@HaMo] for a nice account of divisors on ${\overline{M}}_g$ for arbitrary $g$. When $g=2$, Proposition \[prop:bircont\] gives an elementary proof of this fact: Since $X\simeq {{\mathbb P}}(1,2,3,5)$ its divisor class group has rank one and is generated by the discriminant divisor $\{D=0\}$; the same holds true for $X-p$. By Proposition \[prop:codimone\] ${\overline{M}}_2$ is isomorphic to $\widetilde{X}$ up to codimension $\le 1$, so these have isomorphic class groups. It follows that the rational divisor class group of ${\overline{M}}_2$ is generated by $\Delta_1$ and the proper transform of the discriminant, which is just $\Delta_0$.
The nef cone of ${\overline{M}}_2$ is also well-known. We will not give a self-contained proof here, but rather rely on the general result of Cornalba-Harris [@CH]:
\[thm:CH\] The line bundle $a\lambda -b\delta$ is nef on ${\overline{M}}_g,g\ge 2,$ if and only if $a\ge 11b \ge 0$.
Here $\lambda$ is the pull-back of the polarization on ${\overline{A}}_g$ via the extended Torelli map $\overline{t}:{\overline{M}}_g {\rightarrow}{\overline{A}}_g$ (see §\[subsect:comp\]).
To apply this in our situation, we observe that $$\lambda\equiv \frac{1}{10}(\Delta_0+\Delta_1)$$ over ${\overline{M}}_2$ (see [@HaMo] pp. 175). The factor $10$ can be explained by the fact that $\overline{t}(\Delta_1)$ is defined by the vanishing of a modular form of weight ten (see Remark \[rem:Igusa\] and [@Ig2]). Substitution gives the first part of
\[prop:nef2\] The nef cone of ${\overline{M}}_2$ is generated by the divisors $\Delta_0+\Delta_1$ and $\Delta_0+6\Delta_1$, respectively. These are both semiample, inducing the birational contractions $$\overline{t}:{\overline{M}}_2 {\rightarrow}{\overline{A}}_2 \quad \beta:{\overline{M}}_2 {\rightarrow}X.$$ The rational map $\gamma:{\overline{M}}_2 \dashrightarrow \widetilde{X}$ is an isomorphism.
[*remainder of proof:*]{} Of course, $\Delta_0+\Delta_1$ is semiample and induces the birational contraction morphism $\overline{t}:{\overline{M}}_2 {\rightarrow}{\overline{A}}_2$. As $\widetilde{X}$ and ${\overline{M}}_2$ are isomorphic in codimension one, Proposition \[prop:blowpol\] says that $\widetilde{\Delta}_0+(6-\epsilon)\widetilde{\Delta}_1$ is ample on $\widetilde{X}$ and the corresponding divisor $\Delta_0+(6-\epsilon)\Delta_1$ is ample on ${\overline{M}}_2$. It follows that $\widetilde{X}$ and ${\overline{M}}_2$ are each isomorphic to ${\mathrm{Proj\ }}$ of $$\oplus_{n\ge 0} \Gamma({{\mathcal O}}_{\widetilde{X}}(
n(\widetilde{\Delta}_0+(6-\epsilon)\widetilde{\Delta}_1)))
\simeq \oplus_{n\ge 0} \Gamma({{\mathcal O}}_{{\overline{M}}_2}(
n(\Delta_0+(6-\epsilon)\Delta_1)))
.$$ In particular, the rational map $\gamma$ is an isomorphism. Thus the contractions $b:\widetilde{X}{\rightarrow}X$ and $\beta:{\overline{M}}_2 \dashrightarrow X$ coincide. $\square$
Canonical class of ${\overline{M}}_2$
-------------------------------------
The canonical class $K_{{\overline{M}}_2}$ can also be computed by elementary methods. We know that $$\omega_{{{\mathbb P}}(1,2,3,5)}\simeq {{\mathcal O}}_{{{\mathbb P}}(1,2,3,5)}(-11),$$ which also follows from the third part of Proposition \[prop:ramq\]. Since the discriminant has degree ten, we find $$K_X=-\frac{11}{5}\{D=0\}.$$ Applying the formulas of Proposition \[prop:symsquare\], we obtain $$K_{\widetilde{X}}\equiv
-\frac{11}{5}\widetilde{\Delta}_0-\frac{16}{5}\widetilde{\Delta}_1$$ $$K_{\widetilde{X}}+\frac{1}{2}\widetilde{\Delta}_1+\frac{1}{2}\widetilde{\Xi}
\equiv
-\frac{7}{10}\widetilde{\Delta}_0+\frac{3}{10}\widetilde{\Delta}_1.$$ Since $\widetilde{X}$ and ${\overline{M}}_2$ agree in codimension one, they have the same canonical class $$K_{{\overline{M}}_2}\equiv -\frac{11}{5}\Delta_0-\frac{16}{5}\Delta_1.$$ In particular, we obtain $$\label{logcancomp}
K_{{\overline{M}}_2}+ \alpha \Delta_0+
\frac{1+\alpha}{2}\Delta_1+\frac{1}{2}\Xi\equiv
(-\frac{7}{10}+\alpha)\Delta_0
+(\frac{3}{10}+\alpha/2)\Delta_1.$$
The importance of this divisor stems from the ramification equation of Proposition \[prop:ramify\]. This divisor class pulls back to the class $$K_{{\overline{\mathcal M}}_2}+\alpha\delta$$ on the moduli stack. We shall interpret log canonical models of the moduli stack using this divisor.
Using the computations of Proposition \[prop:symsquare\], we obtain a discrepancy equations for $\beta:{\overline{M}}_2{\rightarrow}X$ $$\label{M2discrep1}
K_{{\overline{M}}_2}+\alpha\Delta_0+\frac{1}{2}\Xi=
\beta^*(K_X+\alpha\{D=0\}+\frac{1}{2}\{G=0\})+(4-6\alpha)\Delta_1$$ $$\label{M2discrep2}
K_{{\overline{M}}_2}+\alpha\Delta_0+\frac{1+\alpha}{2}\Delta_1+\frac{1}{2}\Xi=
\beta^*(K_X+\alpha\{D=0\}+\frac{1}{2}\{G=0\})+\frac{9-11\alpha}{2}\Delta_1.$$
Generalities on log canonical models
------------------------------------
See [@FA] §2 for definitions of the relevant terms and technical background. Let $V$ be a normal projective variety, $D=\sum_{i=1}^n a_iD_i$ a ${{\mathbb Q}}$-divisor such that $0\le a_i\le 1$ and $K_V+D$ is ${{\mathbb Q}}$-Cartier. Abusing notation, we write $V-D$ for $V-\cup_i D_i$.
\[defn:strict\] $(V,D)$ is a [*strict log canonical model*]{} if $K_V+D$ is ample, $(V,D)$ has log canonical singularities, and $V-D$ has canonical singularities.
The idea here is to realize $(V,D)$ as a log canonical model without introducing boundary divisors over $V-D$. This is natural if we want to respect the geometry of the open complement.
The following recognition criterion for strict log canonical models is based on [@FA] §2:
\[prop:strict\] Consider birational projective contractions $\rho:\widetilde{V}{\rightarrow}V$ where the exceptional locus of $\rho$ is divisorial, $D'_i$ denotes the proper transform of $D_i$, and $E_j$ (resp. $F_k$) denotes the exceptional divisors of $\rho$ with $\rho(E_j)\subset D$ (resp. $\rho(F_k) \not \subset D$).
The following are equivalent:
1. [$(V,D)$ is a strict log canonical model.]{}
2. [For some resolution of singularities $\rho:\widetilde{V} {\rightarrow}V$, with the union of the exceptional locus and $\cup_i D'_i$ normal crossings, there exist $b_j\in {{\mathbb Q}}\cap [0,1]$ so that $$\widetilde{D}=\sum_i a_i D'_i + \sum_j b_j E_j$$ satisfies the formula $$\label{eq:discrep}
K_{\widetilde{V}}+\widetilde{D} \equiv \rho^*(K_V + D)+
\sum_j d_j E_j + \sum_k e_k F_k, \quad d_j,e_k \ge 0.$$ For each such choice of $b_j$ $$K_{\widetilde{V}}+\widetilde{D} - \sum_j d_j E_j - \sum_k e_k F_k$$ is semiample and induces $\rho$. Furthermore, we may take the $b_j=1$.]{}
3. [For some contraction $\rho:\widetilde{V} {\rightarrow}V$, there exist $b_j\in {{\mathbb Q}}\cap [0,1]$ so that $$\widetilde{D}=\sum_i a_i D'_i + \sum_j b_j E_j$$ satisfies the formula $$K_{\widetilde{V}}+\widetilde{D} \equiv \rho^*(K_V + D)+
\sum_j d_j E_j + \sum_k e_k F_k, \quad d_j,e_k \ge 0,$$ $(\widetilde{V},\widetilde{D})$ is log canonical, and $\widetilde{V}-\widetilde{D}$ has canonical singularities. The divisor $$K_{\widetilde{V}}+\widetilde{D} - \sum_j d_j E_j - \sum_k e_k F_k$$ is semiample and induces $\rho$.]{}
Some general facts are worth mentioning before we indicate the proof. First, we can decide whether a pair is canonical or log canonical by computing discrepancies on [*any*]{} resolution. Second, discrepancies increase as the coefficients of the log divisor are decreased [@FA] 2.17.3. Third, in situations (2) and (3) the pair $(V,D)$ is the [*log canonical model*]{} of $(\widetilde{V},\widetilde{D})$; such models are unique [@FA] 2.22.1.
[*proof:*]{} It is trivial that the second statement implies the third. To see that the third implies the first, take a resolution for $(\widetilde{V},\widetilde{D})$ so that the union of the exceptional locus and all the proper transforms of the $D'_i,E_j,$ and $F_k$ is normal crossings. Comparing discrepancies for $(\widetilde{V},\widetilde{D})$ and $(V,D)$, using the fact the coefficients of components in $\widetilde{D}$ are at least as large as the coefficents of the corresponding components appearing in $\rho^*(K_V+D)$, we find that $(V,D)$ is log-canonical and has canonical singularities along $V-D$. Since $\rho^*(K_V+D)$ induces $\rho$, $K_V+D$ must be ample on $V$.
For the remaining implication, since $(V,D)$ has log canonical singularities and canonical singularities away from $D$, the discrepancy equation (\[eq:discrep\]) follows. Since $K_V+D$ is ample on $V$, its pull-back to $\widetilde{V}$ is semiample and induces $\rho$. $\square$
We shall also need the following basic fact, a special case of [@FA] 20.2, 20.3:
\[prop:ramifylc\] Let $W$ be a smooth variety and $h:W{\rightarrow}V$ be a finite dominant morphism to a normal variety. Let $D=\sum_i a_i D_i, 0\le a_i \le 1$ be a ${{\mathbb Q}}$-divisor on $V$ containing all the divisorial components of the branch locus of $h$. Let $\bar{D}$ be a ${{\mathbb Q}}$-divisor on $W$ so that ${\mathrm{supp}}(h^{-1}(D))={\mathrm{supp}}(\bar{D})$ and $h^*(K_V+D)=K_W+\bar{D}$. Then $(V,D)$ has log canonical singularities along $D$ iff $(W,\bar{D})$ has log canonical singularities along $\bar{D}$.
If $D$ and $\bar{D}$ have multiplicity one at each component then $h^*(K_V+D)=K_W+\bar{D}$ follows from the other assumptions.
Example of ${\overline{M}}_g$
-----------------------------
The standpoint of this section owes a great deal to Mumford [@Mu1] [@Mu2]:
\[thm:Mumford\] For $g\ge 4$, the pair $({\overline{M}}_g,\Delta)$ is a strict log canonical model.
\[rem:stacklc\] This is also the natural log canonical model from the point of view of the moduli stack. Indeed, for $g\ge 4$ the locus in $M_g$ of curves with automorphisms has codimension $\ge 2$, so the branch divisor of $Q:{\overline{\mathcal M}}_g {\rightarrow}{\overline{M}}_g$ is just $\Delta_1$; over $\Delta_1$, we have simple ramification. We therefore have ([@HaMu] pp. 52) $Q^*K_{{\overline{M}}_g}=K_{{\overline{\mathcal M}}_g}-\delta_1$ and thus $$Q^*(K_{{\overline{M}}_g}+\Delta)=K_{{\overline{\mathcal M}}_g}+\delta.$$
[*sketch proof:*]{} We first check that $K_{{\overline{M}}_g}+\Delta$ is ample. The formula from [@HaMu] §2 (or [@HaMo]) $$K_{{\overline{M}}_g}=13\lambda-2\Delta_0-3/2\Delta_{1}-2\sum_{2\le i\le g/2}\Delta_i$$ gives $K_{{\overline{M}}_g}+\delta =13\lambda-\delta$, which is ample by Theorem \[thm:CH\] (see also [@Mu1]).
The singularity analysis follows [@HaMu]. ${\overline{M}}_g$ has canonical singularities by Theorem 1 of [@HaMu]. To show that $({\overline{M}}_g,\Delta)$ has log canonical singularities, we use the fact that ${\overline{M}}_g$ is étale-locally a quotient of a smooth variety by a finite group. At $[C]$ it has a local presentation $$h:T_{[C]}{\overline{\mathcal M}}_g \twoheadrightarrow T_{[C]}{\overline{\mathcal M}}_g/{\mathrm{Aut}}(C)$$ in terms of its tangent space $$T_{[C]}{\overline{\mathcal M}}_g=\mathrm{Ext}^1(\Omega^1_C,{{\mathcal O}}_C).$$
We analyze the quotient morphism using Proposition \[prop:ramifylc\]. The preimage of the boundary divisor corresponds to a union of hyperplanes $\widehat{\delta}\subset T_{[C]}{\overline{\mathcal M}}_g$, meeting in normal crossings. The pair $(T_{[C]}{\overline{\mathcal M}}_g,\bar{\Delta})$ then has log canonical singularities. An application of Proposition \[prop:ramifylc\], utilizing the ramification discussion in Remark \[rem:stacklc\], implies $(K_{{\overline{\mathcal M}}_g},\Delta)$ is log canonical. $\square$
\[rem:fullforce\] Using the full force of Theorem \[thm:CH\] we get a sharper statement. Consider the pair $$({\overline{M}}_g,\alpha\Delta_0 +\frac{1+\alpha}{2}\Delta_1 +
\alpha(\Delta_2+\ldots+\Delta_{\lfloor g/2 \rfloor})),$$ with log canonical divisor pulling back to $K_{{\overline{\mathcal M}}_g}+\alpha \delta$ on the moduli stack. The ${{\mathbb Q}}$-divisor $K_{{\overline{\mathcal M}}_g}+\alpha\delta$ is the pull-back of an ample line bundle if and only if $9/11<\alpha\le 1$. Since ${\overline{M}}_g$ is a locally a quotient of a smooth variety by a finite group, all divisors on ${\overline{M}}_g$ are ${{\mathbb Q}}$-Cartier. An easy computation with the discrepancy equation (\[eq:discrep\]) then shows that the pair remains log canonical even as the coefficients are reduced.
Application to ${\overline{\mathcal M}}_2$
------------------------------------------
The source of the special difficulties in this case is the fact that $K_{{\overline{M}}_2}+\Delta$ is not effective. Indeed, in §\[subsect:canM2\] we computed $$K_{{\overline{M}}_2}\equiv -\frac{11\Delta_0+6\Delta_1}{5}$$ so $K_{{\overline{M}}_2}+\Delta=-\text{effective divisor}$.
In order to recover a result analogous to Theorem \[thm:Mumford\], we must take the ‘log canonical model of the moduli stack’, as interpreted on ${\overline{M}}_2$ via Proposition \[prop:ramify\]:
\[thm:main\] Consider the log canonical model of ${\overline{\mathcal M}}_2$ with respect to the $K_{{\overline{\mathcal M}}_2}+\alpha\delta$, i.e., the log canonical model of ${\overline{M}}_2$ with respect to $$K_{{\overline{M}}_2}+\alpha \Delta_0+\frac{1+\alpha}{2}\Delta_1+\frac{1}{2}\Xi.$$
1. [For $9/11<\alpha\le 1$, we recover ${\overline{M}}_2$.]{}
2. [For $7/10 < \alpha \le 9/11$ we recover the invariant theory quotient $X\simeq {{\mathbb P}}(1,2,3,5)$.]{}
3. [For $\alpha =7/10$ we get a point; the log canonical divisor fails to be effective for $\alpha<7/10$.]{}
[*proof:*]{} The necessary ampleness results have already been stated. Proposition \[prop:nef2\] and Equation (\[logcancomp\]) imply the log canonical divisor on ${\overline{M}}_2$ is ample if and only if $\alpha>9/11$. Proposition \[prop:ramq\] implies that $$K_X+1/2\{G=0\}+\alpha\{D=0\}$$ is positive on $X$ if and only if $\alpha>7/10$. When $\alpha=7/10$ it is zero and when $\alpha<7/10$ it is negative.
It remains to verify the singularity conditions: First, we check that ${\overline{M}}_2$ has canonical singularities away from $\Delta_0,\Delta_1,$ and $\Xi$. Suppose that $C$ is not in the boundary and does not admit admit a bielliptic involution. In Proposition \[prop:m2\] we saw $$M_2 \simeq X-\{D=0\} \subset {{\mathbb P}}(1,2,3,5),$$ so we need to analyze the singularities of ${{\mathbb P}}(1,2,3,5)-\{D=0\}$. A point in weighted projective space is nonsingular when the weights corresponding to its non-vanishing coordinates are relatively prime, so the only possible singularity occurs when $A=B=C=0$. The corresponding binary sextic form is $$x(x^5+y^5),$$ the unique form with an automorphism group of order five [@Bo] pp. 51 [@Ig1] pp. 645. At this point, ${{\mathbb P}}(1,2,3,5)$ is locally isomorphic to the cyclic quotient singularity $\frac{1}{5}(1,2,3)$, i.e., the quotient of ${{\mathbb A}}^3$ under the action $$(a,b,c) \mapsto (\zeta a, \zeta^2 b, \zeta^3 c) \quad
\zeta\neq 1 \in \mu_5.$$ This is canonical by the Reid-Tai criterion; see [@HaMu] pp. 28 for a general result.
Second, we address the singularities along the boundary. For $\alpha>9/11$ we need that $$K_{{\overline{M}}_2}+\alpha \Delta_0 + \frac{1+\alpha}{2}\Delta_1+1/2\Xi$$ is log canonical. When $\alpha\le 9/11$, Proposition \[prop:strict\] and the discrepancy computation (\[M2discrep2\]) reduce us to showing that this is log canonical. Since ${\overline{M}}_2$ is ${{\mathbb Q}}$-factorial and discrepancies increase as coefficients of log divisors decrease [@FA] 2.17.3, it suffices to verify that $$\label{logmain}
K_{{\overline{M}}_2}+\Delta_0 + \Delta_1+\frac{1}{2}\Xi$$ is log canonical.
The proof relies on the description of the boundary divisors in terms of the local presentation $$T_{[C]}{\overline{\mathcal M}}_2/{\mathrm{Aut}}(C),$$ as sketched in §\[subsect:stackgeo\]. The key observation is that $\Xi$ does not play a rôle in the analysis. Each bielliptic involution acts on $T_{[C]}{\overline{\mathcal M}}_2$ by reflection across the corresponding hyperplane in $\widehat{\xi}$, so the quotient $$h:T_{[C]}{\overline{\mathcal M}}_2 {\rightarrow}T_{[C]}{\overline{\mathcal M}}_2/H,
\quad H={\mathrm{Aut}}(C)/\left<\iota\right>,$$ has simple ramification along $\widehat{\xi}$. Since $\Xi$ has coefficient $1/2$ in (\[logmain\]), $\widehat{\xi}$ does not appear in the pull-back of the log canonical divisor to $T_{[C]}{\overline{\mathcal M}}_2$.
Thus (\[logmain\]) pulls back to $$K_{T_{[C]}{\overline{\mathcal M}}_2}+\widehat{\delta},$$ and we have seen that $\widehat{\delta}$ is normal crossings. It follows that $(T_{[C]}{\overline{\mathcal M}}_2,\widehat{\delta})$ is log canonical and Proposition \[prop:ramifylc\] gives the desired result.$\square$
D. Abramovich, A. Corti, and A. Vistoli, Twisted bundles and admissible covers, [*Comm. Algebra*]{} [**31**]{} (2003), no. 8, 3547–3618.
O. Bolza, On binary sextics with linear transformations into themselves, [*Amer. J. Math.*]{} [**10**]{} (1887) 47-70.
A. Cayley, Tables for the binary sextic, [*Amer. J. Math.*]{} [**4**]{} (1881), 379–384. Reprinted in: [*Collected Mathematical Papers*]{}, vol. XI, pp. 372-376, Cambridge University Press, 1896.
A. Clebsch, [*Theorie der binären algebraischen Formen*]{}, Verlag von B.G. Teubner, Leipzig, 1872.
M. Cornalba and J. Harris, Divisor classes associated to families of stable varieties, with applications to the moduli space of curves, [*Ann. Sci. École Norm. Sup. (4)* ]{}[**21** ]{} (1988), no. 3, 455–475.
E.B. Elliott, [*An Introduction to the Algebra of Quantics*]{}, Oxford University/Clarendon Press, 1895.
J. Kollár, [*Flips and abundance for algebraic threefolds*]{}, [*Astérisque*]{} [**211**]{}, 1992.
J. Harris and I. Morrison, [*Moduli of Curves*]{}, Springer-Verlag, New York, 1998.
J. Harris and D. Mumford, On the Kodaira dimension of the moduli space of curves, [*Invent. math.*]{} [**67**]{} (1982), 23–86.
R. Hartshorne, [*Algebraic geometry*]{}, Springer-Verlag, New York-Heidelberg, 1977.
D. Hilbert, [*Theory of algebraic invariants*]{}, Cambridge University Press, 1993.
J. Igusa, Arithmetic variety of moduli for genus two, [*Ann. of Math. (2)*]{} [**72**]{} (1960) 612–649.
J. Igusa, On Siegel modular forms of genus two, [*Amer. J. Math.*]{} [**84**]{} (1962) 175–200.
J. Igusa, A desingularization problem in the theory of Siegel modular functions, [*Math. Ann.*]{} [**168**]{} (1967) 228–260
G. Laumon and L. Moret-Bailly, [*Champs algébriques*]{}, Springer-Verlag, Berlin-Heidelberg, 2000.
D. Mumford, Stability of projective varieties, [*Enseignement Math. (2)*]{} [**23**]{} (1977), no. 1-2, 39–110.
D. Mumford, Hirzebruch’s proportionality theorem in the noncompact case, [*Invent. Math.*]{}[**42**]{} (1977), 239–272.
D. Mumford, J. Fogarty, and F. Kirwan, [*Geometric Invariant Theory*]{}, third enlarged edition, Springer-Verlag, Berlin-Heidelberg, 1994.
Y. Namikawa, On the canonical holomorphic map from the moduli space of stable curves to the Igusa monoidal transform, [*Nagoya Math. J.*]{} [**52**]{} (1973), 197–259.
Y. Namikawa, A new compactification of the Siegel space and degeneration of Abelian varieties I/II [*Math. Ann.*]{} [**221**]{} (1976), no. 2, 97–141/ no. 3, 201–241.
G. Salmon, [*Lessons introductory to the modern higher algebra,*]{} 2nd edition, Hodges Smith, Dublin, 1866.
N. I. Shepherd-Barron, Canonical rings for moduli spaces of abelian varieties, preprint (2003).
I. Schur, [*Vorlesungen über Invariententheorie*]{}, Bearbeitet und herausgegeben von Helmut Grunsky, Springer-Verlag, Berlin-New York 1968
T.A. Springer, [*Invariant theory*]{}, Lecture Notes in Mathematics, Vol. 585, Springer-Verlag, Berlin-New York, 1977.
A. Vistoli, The Chow ring of ${{\mathcal M}}_2$, Appendix to “Equivariant intersection theory” by D. Edidin and W. Graham, Invent. Math. [**131**]{} (1998), no. 3, 635–644.
Rice University, MS 136\
Department of Mathematics\
PO Box 1892\
Houston, TX 77251-1892\
USA\
`[email protected]`
[^1]: Partial support was provided by National Science Foundation grants 0196187 and 0134259 and the Sloan Foundation.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We study the [*propagation of electromagnetic waves*]{} in a spacetime devoid of a metric but equipped with a [*linear*]{} electromagnetic spacetime relation $H\sim\chi\cdot F$. Here $H$ is the electromagnetic excitation $({\cal D},{\cal H})$ and $F$ the field strength $(E,B)$, whereas $\chi$ (36 independent components) characterizes the electromagnetic permittivity/permeability of spacetime. We derive analytically the corresponding Fresnel equation and show that it is always quartic in the wave covectors. We study the ‘Fresnel tensor density’ ${\cal G}^{ijkl}$ as (cubic) function of $\chi$ and identify the leading part of $\chi$ (20 components) as indispensable for light propagation. Upon requiring electric/magnetic reciprocity of the spacetime relation, the leading part of $\chi$ induces the [*light cone*]{} structure of spacetime (9 components), i.e., the spacetime metric up to a function. The possible existence of an Abelian [*axion*]{} field (1 component of $\chi$) and/or of a [*skewon*]{} field (15 components) and their effect on light propagation is discussed in some detail. The newly introduced skewon field is expected to be T-odd and related to dissipation. [*file nonsym37.tex, 2002-03-24*]{}'
address: |
Institute for Theoretical Physics, University of Cologne\
50923 Köln, Germany
author:
- 'Guillermo F. Rubilar[^1], Yuri N. Obukhov[^2] , and Friedrich W. Hehl[^3]'
title: |
GENERALLY COVARIANT FRESNEL EQUATION\
AND THE EMERGENCE OF THE LIGHT CONE STRUCTURE\
IN LINEAR PRE-METRIC ELECTRODYNAMICS
---
Introduction
============
In pre-metric electrodynamics, the axioms of electric charge and of magnetic flux conservation manifest themselves in the Maxwell equations for the excitation $H=({\cal D},{\cal H})$ and the field strength $F=(E,B)$, see [@gentle]: $$\label{me}
dH=J, \qquad dF=0 \,.$$ Here $J=(\rho,j)$ is the electric current. These equations keep their form in all reference frames and, since they are metric independent, in all gravitational fields as well. In order to develop this scheme into a predictive physical theory, one needs additionally a [ *spacetime relation*]{} $H= H(F)$, which is a functional in general but is assumed here to be of a local nature. It should be well understood, this spacetime relation is a [*universal*]{} constitutive law for the [*vacuum*]{}. We do [*not*]{} treat the constitutive behavior of material media here, we rather search for a universal law for the vacuum, i.e., for the spacetime manifold itself[^4].
The simplest assumption is that the spacetime relation is [*local and linear*]{}. If local coordinates $x^i$ are given, with $i,j,... =0,1,2,3$, we can decompose the excitation and field strength 2-forms into their components according to $$H = {\frac 1 2}\,H_{ij}\,dx^i\wedge dx^j,\qquad
F = {\frac 1 2}\,F_{ij}\,dx^i\wedge dx^j.\label{geo1}$$ Then the [*linear*]{} spacetime relation, see [@Post; @HO02], can be written as $$\label{cl}
H_{ij}=\frac{1}{4}\,{\hat \epsilon}_{ijkl}\,\chi^{klmn}\,F_{mn} \,.$$ Here ${\hat\epsilon}_{ijkl}$ is the Levi-Civita symbol, with ${\hat\epsilon}_{0123}=1$. The quantity $\chi^{ijkl}(x)$ characterizes the electromagnetic properties of the vacuum and is as such of universal importance. It is an untwisted tensor density of weight $+1$ with 36 independent components. If we decompose it into irreducible pieces with respect to the 6-dimensional linear group, then we find $$\label{decomp'}\chi^{ijkl}={}^{(1)}\chi^{ijkl}
+ {}^{(2)}\chi^{ijkl} + {}^{(3)}\chi^{ijkl}\,, \quad{\rm with} \quad
36= 20\oplus 15 \oplus 1$$ independent components, respectively. The irreducible pieces of $\chi$ are defined as follows: $$\begin{aligned}
\label{chiirr2}
{}^{(2)}\chi^{ijkl}&:=&\frac{1}{2}\left( \chi^{ijkl}-
\chi^{klij}\right)=-{}^{(2)}\chi^{klij}\,,\quad
{}^{(3)}\chi^{ijkl}:=\chi^{[ijkl]}\,, \nonumber\\{}^{(1)}
\chi^{ijkl}&:=& \chi^{ijkl} -\,
{}^{(2)}\chi^{ijkl}-\,{}^{(3)}\chi^{ijkl}={}^{(1)} \chi^{klij}\,. \end{aligned}$$ Since no metric is available, we cannot form traces. The Abelian axion piece $^{(3)}\chi^{ijkl}=:\alpha(x)\,\epsilon^{ijkl}$ has been introduced by Ni [@Ni73; @Ni77]. Constitutive laws (for matter) with $^{(2)}\chi^{ijkl}\ne0$ (non-vanishing “skewon fields”[^5] have been discussed by Nieves and Pal [@NP89; @NP94]. They yield P- and CP-violating terms in the field equations. Thus all constitutive functions in (\[decomp’\]) can claim respectability from a physical point of view in the framework of a linear response theory. We will come back to this question in Sec.3.
Our strategy will now be the following: After having specified the spacetime relation in (\[cl\]), we have a predictive theory. Using Hadamard’s method [@Hadamard; @Lichnerowicz] for determining the propagation of waves in the spacetime specified so far, namely a 4-dimensional differentiable manifold carrying the tensor density $\chi^{ijkl}$ of (\[decomp’\]), we will derive, in Sec.2, the [ *Fresnel equation*]{} controlling the waves. It will turn out that the axion $\alpha(x)$ drops out and does not affect the waves, in spite of showing up in the Maxwell equations. In other words, from $\chi^{ijkl}$ in (\[decomp’\]), only $^{(1)} \chi^{ijkl}$, the leading piece, and ${}^{(2)} \chi^{ijkl}$, the skewon field, enter the Fresnel equation. We will give (for the first time) a generally covariant and explicit analytical derivation of the corresponding Fresnel equation and will show that its wave surfaces are in general of [*quartic*]{} order.
We will study in detail the conditions which must be fulfilled in order to factorize the quartic wave surfaces into two equal quadratic wave surfaces, the [*light cone*]{}. Up to a function, this amounts to a new [*derivation of the metric of spacetime from electromagnetic data*]{}.
In this article, we will restrict ourselves to [*linear*]{} pre-metric electrodynamics. Nevertheless, our methods and results are also useful in media with [*non*]{}linear relations between $H$ and $F$, provided one studies electromagnetic [*perturbations*]{} on top of a background solution fulfilling the field equations. Then one can define, see Appendix A, an effective constitutive tensor density $\chi^{ijkl}_{\rm eff}$ that is analogous to $\chi^{ijkl}$ of (\[decomp’\]). In this way, our results, in particular our Fresnel equation, also apply to the work done in geometrical optics by De Lorenci, Novello, et al.[@nov1] in nonlinear electrodynamics à la Born-Infeld, see also [@BLV01; @LP00].
Wave propagation: Fresnel equation
==================================
We will study the propagation of a discontinuity of the electromagnetic field following the lines of Ref. [@OFR00]. The surface of discontinuity $S$ is defined locally by a function $\Phi$ such that $\Phi=$ const. on $S$. Across $S$, the geometric Hadamard conditions are satisfied: $$\begin{aligned}
&& [F_{ij}] = 0,\qquad [\partial_i F_{jk}] = q_i\, f_{jk},
\label{had1}\\
&& [H_{ij}] = 0,\qquad [\partial_i H_{jk}] = q_i\, h_{jk}.
\label{had2} \end{aligned}$$ Here $\left[{\cal F}\right](x)$ denotes the discontinuity of a function ${\cal F}$ across $S$, $q_i:=\partial_i\Phi$ is the wave covector, and $f_{ij}$ and $h_{ij}$ are tensors describing the corresponding jumps of the derivatives of field strength and excitation. If we use Maxwell’s vacuum equations $dH =0$ and $ dF =0$, then (\[had1\]) and (\[had2\]) yield $$\label{4Dwave}
{\epsilon}^{\,ijkl}\, q_{j}\,h_{kl}=0 \,,\qquad
{\epsilon}^{\,ijkl}\, q_{j}\,f_{kl}=0\,.$$ These equations admit non-vanishing wave covectors $q_i$, provided the constraints $$\label{integrability}
{\epsilon}^{\,ijkl}\, f_{ij}\, f_{kl}=0, \qquad
{\epsilon}^{\,ijkl}\, h_{ij}\, h_{kl}=0, \qquad
{\epsilon}^{\,ijkl}\, f_{ij}\, h_{kl}=0$$ are fulfilled.
We assume that $\chi^{ijkl}$ is continuous across $S$. Then, we find from (\[cl\]), (\[decomp’\]) and (\[had1\]), (\[had2\]), $$\label{clhf}
h_{ij}=\frac{1}{4}\,{\hat \epsilon}_{ijkl}\,{\chi}{}^{klmn}
\,f_{mn}\,.$$ With this spacetime relation for the jumps, the conditions (\[4Dwave\]) can be rewritten as $$\label{4Dwave2}
{\chi}{}^{\,ijkl}\, q_{j}\,f_{kl}=
\left(^{(1)}{\chi}{}^{\,ijkl}+\,^{(2)}{\chi}{}^{\,ijkl}\right)
q_{j}\,f_{kl}=0\,,\qquad {\epsilon}^{\,ijkl}\, q_{j}\,f_{kl}=0\,,$$ since $^{(3)}{\chi}{}^{\,ijkl}\, q_{j}\,f_{kl}=
\alpha\,\epsilon^{ijkl}\, q_{j}\,f_{kl}=0$. Thus the axion field $\alpha$ drops out even though it enters the Maxwell equations.
Technically, the system (\[4Dwave2\]) was analyzed in Ref. [@OFR00] in the framework of a 3-(co)vector decomposition of the electromagnetic discontinuity tensor $f_{ij}$. Here we present a different, generally covariant derivation of the Fresnel equation by applying ideas of Tamm [@Tamm].
As a first step, eq.(\[4Dwave2\])$_2$ is solved by $$f_{ij} = q_ia_j - q_ja_i \,.\label{fqa}$$ The covector $a_i$ is only defined up to a ‘gauge’ transformation $$\label{agt}
a_i\rightarrow a_i + \lambda q_i \,,$$ with an arbitrary scalar $\lambda$. We substitute (\[fqa\]) into (\[4Dwave2\])$_1$: $$\label{4Dwave3}
{\chi}{}^{\,ijkl}\,q_{j}q_ka_l=0 \,.$$ Because of (\[agt\]), not all the equations in (\[4Dwave3\]) are independent. In order to isolate the trivial parts of (\[4Dwave3\]), it is convenient to pick a specific (anholonomic) [*coframe*]{} $\vartheta^\alpha=e_i{}^\alpha\, dx^i$, here $\alpha,\beta,\ldots=\hat 0, \hat 1, \hat 2, \hat 3$: Namely, we identify the zeroth leg of the coframe with the wave covector, that is, $\vartheta^{\hat 0}=q$ or, in components, $e_i{}^{\hat 0}=q_i$ or $q_\alpha=(1,0,0,0)$. Then $$\label{eval}
{\chi}^{\alpha\beta\gamma\delta}\,q_{\beta}q_\gamma a_\delta=
{\chi}^{\alpha\hat{0}\hat{0}\delta}\, a_\delta=
{\chi}^{\alpha\hat{0}\hat{0}b}\, a_b=0 \,.$$ ‘Spatial’ anholonomic indices run over ${a},b,...=\hat 1, \hat 2, \hat
3$. The zeroth component of (\[eval\]) vanishes. Thus we find $$\label{e3}
{\chi}^{\hat{0}a\hat{0}b}\, a_b=W^{ab}\,a_b =0\,,\quad{\rm
with}\quad W^{ab}:= {\chi}^{\hat{0}a\hat{0}b}\,.$$
The necessary and sufficient condition for the existence of non-trivial solutions for $a_{b}$ is the vanishing of the determinant of the $3\times 3$ matrix $W$: $$\label{det01}
{\cal W}:=\det W
=\frac{1}{3!}\epsilon_{ a b c}
\epsilon_{ d e f}W^{ a d}W^{ b e}W^{ c f}
=\frac{1}{3!}\epsilon_{ a b c}
\epsilon_{ d e f}
{\chi}^{\hat 0 a \hat 0 d}
{\chi}^{\hat 0 b \hat 0 e}
{\chi}^{\hat 0 c \hat 0 f} \,.$$ This yields the Fresnel equation.
We can rewrite $\cal W$ in a fully 4-dimensional covariant manner. First, we observe that the 3-dimensional Levi-Civita symbol $\epsilon_{ a b c}$ is related to the 4-dimensional one by means of $\epsilon_{ a b c}\equiv \epsilon_{\hat 0 a b c}$. Then we can extend one $\hat 0$-component of the constitutive tensors to a fourth summation index. As a result, we find the identity $$\label{id01}
\epsilon_{\hat 0 a b c} {\chi}{}^{\, \hat 0 a \hat 0 d} {\chi}{}^{\,
\hat 0 b \hat 0 e} {\chi}{}^{\, \hat 0 c \hat 0 f}= \epsilon_{\hat
0 \beta\gamma\delta} {\chi}{}^{\, \hat 0 \beta \hat 0 d}
{\chi}{}^{\, \hat 0 \gamma \hat 0 e} {\chi}{}^{\, \hat 0 \delta \hat
0 f}= \frac{1}{2}\, \epsilon_{\alpha\beta\gamma\delta}
{\chi}{}^{\, \alpha\beta \hat 0 d} {\chi}{}^{\, \hat 0 \gamma \hat 0
e} {\chi}{}^{\, \hat 0 \delta \hat 0 f} \,,$$ which holds true due to the (anti)symmetry properties of the Levi-Civita symbol and of the constitutive tensor. This allows us to rewrite (\[det01\]) as $$\label{det02}
{\cal W}=\frac{1}{3!}\,\frac{1}{2}\epsilon_{\alpha\beta\gamma\delta}\,
\epsilon_{\hat 0 d e f}\,
{\chi}{}^{\, \alpha\beta \hat 0 d}
{\chi}{}^{\, \hat 0 \gamma \hat 0 e}
{\chi}{}^{\, \hat 0 \delta \hat 0 f} \,.$$ Now we apply the same procedure to the second Levi-Civita symbol and finally obtain $$\label{det03}
{\cal W}=\frac{1}{4!}\epsilon_{\alpha\beta\gamma\delta}\,
\epsilon_{\lambda\rho\sigma\tau}
{\chi}{}^{\, \alpha\beta \hat 0\rho}
{\chi}{}^{\, \hat 0 \gamma \hat 0\sigma}
{\chi}{}^{\, \hat 0 \delta \lambda\tau} \,.$$ Since $e_i{}^{\hat 0}=q_i$, this can, in coordinate components, be written as $$\label{wgen}
{\cal W}=\frac{\theta^2}{4!}\,{\epsilon}_{mnpq}\,
{\epsilon}_{rstu}\, {\chi}{}^{\,mnri}\,
{\chi}{}^{\,jpsk}\,{\chi}{}^{\,lqtu } \, q_iq_jq_kq_l\, ,$$ with $\theta:=\det(e_i{}^\alpha)$. We define the fourth order tensor density of weight $+1$ $$\label{G4}
{\cal G}^{ijkl}(\chi):=\frac{1}{4!}\,{\epsilon}_{mnpq}\,
{\epsilon}_{rstu}\, {\chi}^{mnr(i}\, {\chi}^{j|ps|k}\,
{\chi}^{l)qtu }\,.$$ It is totally symmetric, ${\cal G}^{ijkl}(\chi)= {\cal
G}^{(ijkl)}(\chi)$, and thus has 35 independent components (see [@Schouten]). With (\[G4\]), we find the Fresnel equation in the generally covariant form $$\label{Fresnel}
{\cal G}^{ijkl}(\chi)\,q_i q_j q_k q_l = 0 \,.$$
The general Fresnel equation (\[Fresnel\]) is always a [*quartic*]{} equation in $q_i$ despite the fact that it was derived from a determinant of a $3\times 3$ matrix quadratic in the wave covectors. This corrects Denisov and Denisov [@DD99] who claim that a particular case of the general linear constitutive law may yield a sixth order Fresnel equation; in [@Lam99], even a Fresnel equation of eighth order is claimed to hold.
The rest of our paper is devoted to finding conditions that will make the quartic Fresnel equation [*factorize*]{} into two quadratic ones and to determine under which circumstances the Fresnel equation turns out to be a perfect square.
For practical calculations, it is convenient to use a 1+3 coordinate decomposition similar to that of Refs. [@OH99; @HOR00; @OFR00]. Correspondingly, we can rewrite our general result (\[Fresnel\]) as $${\cal W}=q_0^2\left(q_0^4 M + q_0^3q_a\,M^a + q_0^2q_a q_b\,M^{ab} +
q_0q_a q_b q_c\,M^{abc} + q_a q_b q_c
q_d\,M^{abcd}\right)=0\,,\label{fresnel}$$ with $$\label{compare01}
M:={\cal G}^{0000}\,,\quad M^a:=4{\cal G}^{000a}\,,
\quad M^{ab}:=6{\cal G}^{00ab}\,,\quad$$ $$\label{compare02}
M^{abc}:=4{\cal G}^{0abc}\,, \quad M^{abcd}:={\cal G}^{abcd} \,.$$ In terms of the $3\times 3$ matrices $${\cal A}^{ba}:={\chi}{}^{0a0b}\,,
\qquad {\cal B}_{ba}:=\frac{1}{4}\,{\hat\epsilon}_{acd}\,
{\chi}{}^{cdef}\,{\hat\epsilon}_{efb}\,,$$ $${\cal C}_{\ a}^{b}:=\frac{1}{2} \,{\hat\epsilon}_{acd}\,
{\chi}{}^{cd0b}\,,
\qquad {\cal D}^{\ a}_{b}:=\frac{1}{2} \, {\chi}{}^{0acd}\,
{\hat\epsilon}_{cdb} \,,$$ the tensor density $\chi^{ijkl}$ can be written as a $6\times 6$ matrix $${\chi}^{IK}= \left( \begin{array}{cl} {\cal B}_{ab}& {\cal D}_a^{\ b}
\\ %\hline
{\cal C}^a_{\ b} & {\cal A}^{ab} \end{array} \right)\,,\quad{\rm
with}\quad I,K,...=1,2,...,6\,.\label{almost}$$ For the $M$’s, we have explicitly: $$\begin{aligned}
M&=&\det{\cal A} \,, \\
M^a&=& -\hat{\epsilon}_{bcd}\left( {\cal A}^{ba}\,{\cal A}^{ce}\,
{\cal C}^d_{\ e} + {\cal A}^{ab}\,{\cal A}^{ec}\,{\cal D}_e^{\ d}
\right)\,,\label{ma1}\\
M^{ab}&=& \frac{1}{2}\,{\cal A}^{(ab)}\left[({\cal C}^d{}_d)^2 +
({\cal D}_c{}^c)^2 - ({\cal C}^c{}_d + {\cal D}_d{}^c)({\cal C}^d{}_c +
{\cal D}_c{}^d)\right]\nonumber\\
&&+({\cal C}^d{}_c + {\cal D}_c{}^d)({\cal A}^{c(a}{\cal C}^{b)}{}_d +
{\cal D}_d{}^{(a}{\cal A}^{b)c}) - {\cal C}^d{}_d
{\cal A}^{c(a}{\cal C}^{b)}{}_c \nonumber\\
&& - {\cal D}_c{}^{(a}{\cal A}^{b)c}{\cal D}_d{}^d
- {\cal A}^{dc}{\cal C}^{(a}{}_c {\cal D}_d{}^{b)}+
\left({\cal A}^{(ab)}{\cal A}^{dc}-
{\cal A}^{d(a}{\cal A}^{b)c}\right){\cal B}_{dc}\,,\label{ma2}\\
M^{abc} &=& \epsilon^{de(c|}\left[{\cal B}_{df}(
{\cal A}^{ab)}\,{\cal D}_e^{\ f} - {\cal D}_e^{\ a}{\cal A}^{b)f}\,)
+ {\cal B}_{fd}({\cal A}^{ab)}\,{\cal C}_{\ e}^f -
{\cal A}^{f|a}{\cal C}_{\ e}^{b)})\,
\right. \nonumber \\
&& \left. +{\cal C}^{a}_{\ f}\,{\cal D}_e^{\ b)}\,{\cal D}_d^{\ f}
+ {\cal D}_f^{\ a}\,{\cal C}^{b)}_{\ e}\,{\cal C}^{f}_{\ d} \right]
\, ,\label{ma3}\\
M^{abcd} &=& \epsilon^{ef(c}\epsilon^{|gh|d}\,{\cal B}_{hf}
\left[\frac{1}{2} \,{\cal A}^{ab)}\,{\cal B}_{ge}
- {\cal C}^{a}_{\ e}\,{\cal D}_g^{\ b)}\right] \,.\label{ma4} \end{aligned}$$ Here the $M$’s, as totally symmetric tensors in 3 dimensions, carry $1\oplus 3\oplus 6\oplus 10 \oplus15=35$ independent components. All these results have been verified by using the computer algebra system Maple together with the tensor package GrTensor[^6].
Axion and skewon
================
Before we draw explicitly conclusions from the Fresnel equation (\[Fresnel\]) or (\[fresnel\]), it is instructive to study the structure of ${\cal G}^{ijkl}$. As we saw, this tensor density has 35 independent components, the same number as $^{(1)}\chi^{ijkl}+\,^{(2)}\chi^{ijkl}$. This can be understood as follows: Because of (\[4Dwave2\])$_1$, we have $$\label{drei}
{\cal G}^{ijkl}(^{(3)}\chi)=0\,.$$ This can also be directly verified by substituting $^{(3)}\chi\sim
\epsilon $ into (\[G4\]). Accordingly, if the spacetime relation (\[cl\]) contained only the axion piece $^{(3)}\chi$, the propagation of electromagnetic waves would not be well-behaved. Therefore, in nature, it is excluded that $\chi$ consists merely of $^{(3)}\chi\sim \alpha$. On the other hand, if we substitute $\chi$ into ${\cal G}$, we find $$\label{einspluszwei}
{\cal G}^{ijkl}(\chi)={\cal G}^{ijkl}(^{(1)}\chi+{}^{(2)}\chi+
{}^{(3)}\chi)= {\cal G}^{ijkl}(^{(1)}\chi+{}^{(2)}\chi)\,.$$ Note that $\cal G$ depends on $\chi$ is a cubic way, that is, our last equation is by no means trivial. Consequently, if either $^{(1)}\chi$ or $^{(2)}\chi$ or both pieces exist in nature — and this is for sure — then the axion piece doesn’t interfere with the local laws of the propagation of electromagnetic waves as determined by the Fresnel equation. Or expressed in a positive manner: The Abelian axion could be around, the laws of electrodynamics are compatible with it and the light propagation wouldn’t be affected locally. There have been intensive experimental searches for axions, so far without success [@Cooper; @Cheng; @Stedman]. However, the axion remains a serious candidate for a particle search in experimental high energy physics.
The axion piece has one further interesting property. The kinetic electromagnetic energy-momentum current in pre-metric electrodynamics reads, see [@HO01; @HO02], $$^ {\rm k} \Sigma_\alpha :={\frac 1 2}\left[F\wedge(e_\alpha\rfloor
H) - H\wedge (e_\alpha\rfloor
F)\right]\,.\label{simax}$$ If we substitute the spacetime relation (\[cl\]) into it, we find, in reminiscence of (\[einspluszwei\]), $$\label{axenergy}
^{\rm k}\Sigma_\alpha(\chi) = \,^{\rm k}\Sigma_\alpha
\left(^{(1)}\chi +{}^{(2)}\chi +{}^{(3)}\chi \right)=\,^{\rm
k}\Sigma_\alpha \left(^{(1)}\chi +{}^{(2)}\chi \right)\,.$$ Moreover, $^{\rm k}\Sigma_\alpha(^{(3)}\chi)=0$. Clearly then, the axion does not contribute to the electromagnetic energy-momentum current. Nevertheless, as Ni [@Ni73; @Ni77] has shown, it is possible to develop a reasonable theory for the Abelian axion. Accordingly, the axion piece $^{(3)}\chi$ of the constitutive tensor density $\chi$ cannot be dismissed a priori.
Let’s turn then to the skewon piece $^{(2)}\chi$. The name we derived from the skew- or antisymmetric $6\times 6$ matrix $^{(2)}\chi$ can be mapped to. Commonly, this piece is not considered seriously. The conventional argument runs as follows, see Post [@Post]. Suppose a Lagrangian 4-form $L$ exists for the electromagnetic field. In general $H\sim \partial L/\partial F$. If $H$ is assumed to be linear in $F$, as is done in (\[cl\]), then $L$ reads $L\sim H\wedge
F\sim\chi\cdot F\wedge F={}^{(1)}\chi \cdot F\wedge F+ {}^{(3)}\chi \cdot
F\wedge F$, since the piece with $^{(2)}\chi$ drops out of $L$ because of the antisymmetry $^{(2)}\chi^{ijkl}= -\,^{(2)}\chi^{klij}$, see (\[chiirr2\]). Hence, $$\label{lagrangian}
L(\chi)= L(^{(1)}\chi+{}^{(3)}\chi)= L(^{(1)}\chi)+ L(^{(3)}\chi)\,.$$ The term with $^{(1)}\chi$ eventually becomes the Maxwell Lagrangian, the term with $^{(3)}\chi$ part of the axion Lagrangian. Since we conventionally assume that all information of a physical system is coded into its Lagrangian, we reject $^{(2)}\chi^{ijkl}\ne 0$ as being unphysical.
However, as we saw already in (\[axenergy\]), if a piece drops out from a certain expression, it does not necessarily imply that this piece lost its right of existence. Eq.(\[lagrangian\]) only shows that $L$ is ‘insensitive’ to the skewon piece. In contrast, the electromagnetic energy-momentum current (\[simax\]) definitely ‘feels’ the contribution from the skewon, $$\label{skenergy}
^ {\rm k} \Sigma_\alpha(^{(2)}\chi)\ne 0\,,$$ that is, the skewon piece does carry electromagnetic energy-momentum. Thereby, it could displays its presence. Clearly then, we expect that $^{(2)}\chi$ influences light propagation. And, indeed, it does. By inspection, we find $$\label{gskenergy}
{\cal G}^{ijkl}(^{(1)}\chi+{}^{(2)}\chi)= {\cal G}^{ijkl}
(^{(1)}\chi) + \overline{{\cal
G}}^{ijkl}(^{(1)}\chi,{}^{(2)}\chi)\,,$$ with $$\overline{{\cal G}}^{ijkl}(^{(1)}\chi,{}^{(2)}
\chi)\ne 0,\label{einsundzw}$$ even though $$\label{zwei}
{\cal G}^{ijkl}(^{(2)}\chi)=0\,.$$ In other words, like the axion, the skewon cannot propagate light decently unless the $^{(1)}\chi$ piece participates. Moreover, $$\label{zweidrei}
{\cal G}^{ijkl}(^{(2)}\chi+{}^{(3)}\chi)=0\,,$$ i.e., the skewon and the axion piece cannot exist alone or together, the ‘leading’ $^{(1)}\chi$ piece is indispensable. If we knew more about the detailed structure of $ \overline{{\cal G}}^{ijkl}
(^{(1)}\chi,{}^{(2)}\chi)$, then we could better judge which modes of the skewon are compatible with present-day experiments.
In any case, in linear pre-metric electrodynamics a Lagrangian is not assumed to exist a priori. Therefore, it is not alarming that $^{(2)}\chi$ drops out from the proto-Lagrangian $L$. Since it carries electromagnetic energy-momentum, we will take [*the possible existence of*]{} $^{(2)}\chi^{ijkl}$ [*for granted*]{}.
What is then the possible physical meaning of $^{(2)}\chi$? This has already been discussed by Nieves and Pal [@NP89; @NP94]. But let us first collect some formalism. Consider a certain vector field $\xi=\xi^\alpha e_\alpha$, with the basis $e_\alpha$ of the tangent vector space at each point of spacetime. Then we can transvect the energy-momentum current (\[simax\]) with $\xi^\alpha$: $$\label{QQ}
{\cal Q}:=\xi^\alpha\,^{\rm k}\Sigma_\alpha=
\frac{1}{2}\left[F\wedge(\xi\rfloor H)- H\wedge(\xi\rfloor
F)\right]\,.$$ The scalar-valued 3-form $\cal Q$ is expected to be related to conserved quantities provided we can find suitable (Killing type) vector fields $\xi$. Therefore we determine its exterior derivative and find after some algebra, $$\label{dQQ}
d{\cal Q}=(\xi\rfloor F)\wedge J+\frac{1}{2}\left( F\wedge {\cal
L}_\xi H-H\wedge{\cal L}_\xi F \right)\,,$$ or, in holonomic components, with ${\cal Q}^i:=
\epsilon^{ijkl}Q_{jkl}/6$, $\,{\cal J}^i:= \epsilon^{ijkl}J_{jkl}/6$, and ${\cal H}^{ij} :=\epsilon^{ijkl}H_{kl}/2$, $$\label{diqq}
\partial_i {\cal Q}^i=\xi^kF_{kl}\,{\cal J}^l+\frac{1}{4}\left(
F_{kl}\,{\cal L}_\xi {\cal H}^{kl} -{\cal H}^{kl}\, {\cal L}_\xi
F_{kl}\right)\,.$$ Here ${\cal L}_\xi$ denotes the Lie derivative along the vector $\xi$. Now we substitute the linear spacetime relation (\[cl\]), or ${\cal
H}^{kl}=\chi^{klmn}F_{mn}/2$, and find $$\label{diqqlin}
\partial_i {\cal Q}^i=\xi^kF_{kl}\,{\cal J}^l+\frac{1}{8}\left[
F_{kl}\,{\cal L}_\xi (\chi^{klmn}F_{mn}) -\chi^{klmn}F_{mn}\,
{\cal L}_\xi F_{kl}\right]\,.$$ We apply the Leibniz rule of the Lie derivative and rearrange a bit: $$\label{diqqlin1}
\partial_i {\cal Q}^i=\xi^kF_{kl}\,{\cal J}^l+\frac{1}{8}\left[
({\cal L}_\xi \chi^{ijkl})\,F_{ij}F_{kl}+(\chi^{ijkl}-
\chi^{klij})\,F_{ij}\, {\cal L}_\xi F_{kl}\right]\,.$$ We substitute the irreducible pieces of $\chi^{ijkl}$. Then we have $$\label{diqqlin2}
\partial_i {\cal Q}^i=\xi^kF_{kl}\,{\cal J}^l+\frac{1}{8}\, {\cal
L}_\xi \left(^{(1)}\chi^{ijkl}+\,^{(3)}\chi^{ijkl}
\right)\,F_{ij}F_{kl}+ \frac{1}{4}\,^{(2)}\chi^{ijkl} \,F_{ij}\,
{\cal L}_\xi F_{kl}\,.$$
Here it is manifest that $^{(1)}\chi$ and the axion $^{(3)}\chi$ behave qualitatively different as compared to the skewon $^{(2)}\chi$. If $^{(1)}\chi$ and $^{(3)}\chi$ carry a symmetry along $\xi$ such that $ {\cal L}_\xi{}^{(1)}\chi^{ijkl}=0$ and $ {\cal L}_\xi
{}^{(3)}\chi^{ijkl}=0$, then, in vacuum, i.e., for ${\cal J}^i=0$, we still have [*non-*]{}conservation because of the offending term $^{(2)}\chi\, F\dot{ F}$. Here the dot symbolizes the Lie derivative along $\xi$. If $\xi$ can be interpreted as a ‘time’ direction, then $\cal Q$ represents the electromagnetic energy density.
In any case we see that $^{(2)}\chi$ induces a [*dissipative*]{} term with a first ‘time’ derivative. This is what we might have expected since, in general, dissipative phenomena cannot be described conveniently in a Lagrangian framework. It is then our hypothesis that the skewon piece $^{(2)}\chi$ can represent a field which is [ *odd under T transformations*]{}. This is also what had been discussed by Nieves and Pal [@NP89; @NP94]. Of course, we must investigate how this [*skewon*]{}, as we may call it in a preliminary way, disturbs the light cone and whether there is perhaps only a viable subclass of the 15 independent components of the skewon. An isotropic skewon $s$ could be constructed by putting in (\[almost\]) ${\cal
C}^a{}_b=s\,\delta^a_b$ and ${\cal D}_a{}^b=-s\,\delta_a^b$, see Nieves and Pal [@NP94].
Let us try then to collect our results in order to get a rough picture of the physical meaning of these different irreducible pieces. The piece $^{(1)}\chi^{ijkl}$ is indispensable for an appropriate propagation of light, as we saw in (\[drei\]), (\[zwei\]), and (\[zweidrei\]). Thus we call it the leading piece of $\chi^{ijkl}$.
If $^{(1)}\chi^{ijkl}$ exists alone and for the spacetime relation (\[cl\]) [electric/magnetic reciprocity]{} is assumed to hold additionally, see Sec.4, then, up to an unknown function, the metric of spacetime can be derived including its Lorentzian signature [@OH99; @HOR00; @GR01]. One can think of this reduction in the way that electric/magnetic reciprocity cuts the 20 components of $^{(1)}\chi^{ijkl}$ into half, that is, only 10 components are left for the metric. Modulo an undetermined function, we have then 9 remaining components. These 9 remaining components of $^{(1)}\chi^{ijkl}$ determine the [*light cone*]{} at each point of spacetime. Accordingly, in $^{(1)}\chi^{ijkl}$ the light cone of spacetime is hidden and thereby conventional Maxwell-Lorentzian vacuum electrodynamics as well. To put it more geometrically, the first irreducible piece $^{(1)}\chi$, via electric/magnetic reciprocity, yields the [*conformal*]{} structure of spacetime. In this sense, there is no doubt that $^{(1)}\chi^{ijkl}$ is the principal part of the constitutive tensor density $\chi$ of the vacuum. But, as we argued above, there are no a priori reasons for excluding either the axion $^{(3)}\chi$ or the skewon $^{(2)}\chi$. In particular, the Fresnel equation and the electromagnetic energy-momentum current can accommodate both pieces in a plausible way.
Almost complex structure {#complex}
========================
Let us now search for a condition restricting the constitutive tensor density $\chi^{ijkl}$ of the spacetime relation (\[cl\]). Since the times of Maxwell and Heaviside, in the equations of electrodynamics a certain symmetry was noticed between the electric and the magnetic quantities and was used in theoretical discussions. We formulate electric/magnetic reciprocity as follows [@HO02]: Given the energy-momentum current (\[simax\]). It is electric-magnetic reciprocal, i.e., it remains invariant $\Sigmakin\rightarrow
\Sigmakin$ under the [transformation]{} $$\label{duality1}
H\rightarrow \zeta F\,,\quad F\rightarrow
-\frac{1}{\zeta}\,H\,,$$ with the dimensionful pseudo-scalar function $\zeta=\zeta(x)$.
We now postulate [*electric/magnetic reciprocity of the spacetime relation (\[cl\])*]{}. Under (\[duality1\]), the spacetime relation transforms into $$\label{duality2}
\zeta\,F_{ij}=-\frac{1}{4\,\zeta}\,{\hat
\epsilon}_{ijkl}\,\chi^{klmn}\,H_{mn} \,.$$ If we substitute this into (\[cl\]), we find, after some algebra, the consistency condition $$\label{duality3}
-\frac{1}{8\,\zeta^2}\,({\hat \epsilon}_{ijmn}\,\chi^{mnpq})\,( {\hat
\epsilon}_{pqr\!s}\,\chi^{r\!skl})=\delta^{kl}_{ij}\,,$$ with the function $$\label{duality4}
\zeta^2:= -\frac{1}{96}\,({\hat \epsilon}_{ijmn}\,\chi^{mnpq})\,(
{\hat \epsilon}_{pqr\!s}\,\chi^{r\!sij})\,.$$ In order to allow for a more compact notation, we introduce the dimensionless density $\stackrel{\rm
o}{\chi}{\!}^{ijkl}:={\chi}^{ijkl}/\zeta$ and define $$\label{circle}
\kappa_{ij}{}^{kl}=\frac{1}{2}\,{\hat \epsilon}_{ijmn}\,\stackrel{\rm
o}{\chi}{\!}^{mnkl}\,.$$ Then the [*closure relation*]{} reads $$\label{duality5}
\kappa_{ij}{}^{mn} \kappa_{mn}{}^{kl}=-2\delta_{ij}^{kl}$$ or, even more compactly, as $6\times 6$ matrix equation, $$\label{close1a}
{\kappabf}\,{\kappabf}=-\,\mathbf{1}_6.$$ Mathematically this means that the operator ${\kappabf}$ represents an almost complex structure on the space of 2-forms. For the special case of $^{(2)}\chi={}^{(3)}\chi\equiv 0$, such closure relations were first discussed by Toupin [@Toupin], Schönberg [@Schoenberg], and Jadczyk [@Jadczyk].
Let us now make the closure relation explicit. We turn back to the constitutive $6\times 6$ matrix (\[almost\]). We define dimensionless $3\times 3$ matrices $\stackrel{\;{\rm o}}{\cal
A}\;:={\cal A}/\zeta$, etc. In terms of these dimensionless matrices (we immediately drop the small circle for convenience), the closure relation reads, $$\begin{aligned}
{\cal A}^{ac}{\cal B}_{cb} + {\cal C}^a{}_c{\cal C}^c{}_b &=&
- \delta^a_b,\label{almostclose1} \\
{\cal C}^a{}_c {\cal A}^{cb} + {\cal A}^{ac}{\cal D}_c^{\ b}
&=& 0,\label{almostclose2}\\
{\cal B}_{ac}{\cal C}^c{}_b + {\cal D}_a^{\ c}{\cal B}_{cb}
&=& 0, \label{almostclose3}\\
{\cal B}_{ac}{\cal A}^{cb} + {\cal D}_a^{\ c}{\cal D}_c^{\ b}
&=& -\delta^b_a.\label{almostclose4} \end{aligned}$$ Assume $\det {\cal B}\neq 0$. Then we can find the general non-degenerate solution. We define the matrix $K_{ab}$ by $$K:={\cal BC}\,,\qquad{\rm i.e.}\qquad {\cal C}= {\cal
B}^{-1}K\,,\label{CDKK}$$ and substitute it into (\[almostclose3\]), $${\cal D} = - K{\cal B}^{-1}.\label{KK}$$ Next, we solve (\[almostclose1\]) with respect to $\cal A$: $${\cal A}=-{\cal B}^{-1}-{\cal B}^{-1}K{\cal B}^{-1}K{\cal
B}^{-1}.\label{ABK}$$ We multiply (\[ABK\]) by $\cal C$ from the left and by $\cal D$ from the right, respectively, and find with (\[CDKK\]) and (\[KK\]), $$\begin{aligned}
{\cal CA} &=& - {\cal B}^{-1}K{\cal B}^{-1}
- {\cal B}^{-1}K{\cal B}^{-1}K{\cal B}^{-1}K{\cal B}^{-1},\\
{\cal AD} &=& + {\cal B}^{-1}K{\cal B}^{-1}
+ {\cal B}^{-1}K{\cal B}^{-1}K{\cal B}^{-1}K{\cal B}^{-1}. \end{aligned}$$ Thus, (\[almostclose2\]) is automatically satisfied. Accordingly, only (\[almostclose4\]) has still to be checked. We compute its first and second term of its left side, $$\begin{aligned}
{\cal BA} &=& -1 - K{\cal B}^{-1}K{\cal B}^{-1}\,,\\
{\cal D}^2 &=& K {\cal B}^{-1} K {\cal B}^{-1}\,, \end{aligned}$$ and find that it is fulfilled, indeed.
Summing up, we have derived the general solution of the closure relation (\[close1a\]) in terms of two arbitrary matrices ${\cal B}$ and $K$ as $$\begin{aligned}
{\cal A} &=& - {\cal B}^{-1} - {\cal B}^{-1}K{\cal B}^{-1}K{\cal
B}^{-1}\,,\\
{\cal C} &=& {\cal B}^{-1}K\,,\\
{\cal D} &=& - K{\cal B}^{-1}\,. \end{aligned}$$ The solution thus has $2\times 9 = 18$ independent components. Alternatively, one can write the solution as $$\begin{aligned}
{\cal A} &=&% - {\cal B}^{-1} - {\cal C}^{2}{\cal B}^{-1} =
- (1+{\cal C}^2){\cal B}^{-1}\, ,\\ {\cal D} &=& - {\cal BCB}^{-1}\,, \end{aligned}$$ which is parametrized by the arbitrary matrices ${\cal B}$ and ${\cal C}$ with altogether 18 independent components.
Spacetime metric {#metric}
================
Can we construct the spacetime metric by using the new general solution for the generalized closure relation? Technically, this is equivalent to the question: Can the general Fresnel equation be reduced to the light-cone structure? Here we study the conditions for such a reduction.
Let us decompose the arbitrary matrix ${\cal B}$ into its symmetric and antisymmetric parts, $${\cal B}_{ab} = b_{ab} + \hat{\epsilon}_{abc}n^c\,,\quad{\rm
with}\quad b_{ab}:= {\cal B}_{(ab)}\,,\quad n^c:= \epsilon^{cab}
\,B_{[ab]}\,.\label{Bn}$$ Note that $ b_{ab}$ contributes to $^{(1)}\chi$ and $n^c$ to $^{(2)}\chi$. Now we can lower the index of $n^a$ by means of $b_{ab}$, namely, $n_a:=b_{bc}\,n^c$ and $n^2 := n^cn_c =
b_{ab}\,n^an^b$, and, provided $\det b\ne 0$, we can raise an index by $b^{ab}$ which denotes the inverse of $b_{ab}$. We find $\det {\cal B}
= \det b + n^2$, and the inverse of (\[Bn\]) reads $$\label{invBn}
{\cal B}^{ab} = {\frac 1 {\det{b}+n^2}}\left({\bar b}^{ab} + n^an^b -
\epsilon^{abc}\,n_c\right).$$ Here the symmetric matrix ${\bar b}^{ab}$ is the matrix of the minors of $b_{ab}$. If $\det b\ne 0$, then ${\bar b}^{ab}=b^{ab}\det{b}$.
Let us find out what is qualitatively new in the asymmetric case as compared to the previously studied symmetric case [@OFR00]. For this purpose we consider the particular solution of the closure relation for $K=0$. Consequently, ${\cal C} = {\cal D} =0$ and ${\cal
A} = - {\cal B}^{-1}$. Then, by substituting (\[Bn\]) into (\[fresnel\]), we obtain: $${\cal W} = -\,{\frac {q_0^2}{(\det{b}+n^2)}}\left[q_0^4
- 2q_0^2\left(q^2\,\det{b} - (qn)^2\right) + \left(q^2\,\det{b} +
(qn)^2\right)^2 \right]. \label{freC0}$$ Here we used the abbreviation $(qn):=q_a\,n^a$.
In general, this expression is neither a square of a quadratic polynomial nor a product of two quadratic polynomials. In other words, neither a light cone nor a birefringence (double light cone) structure arises generically. In order to study the reduction conditions, let us assume that the Fresnel equation is a product of two quadratic equations for $q_i$, i.e., the spacetime ‘medium’ is birefringent. Accordingly, for (\[freC0\]) we make the general ansatz $${\cal W} = -\,\frac{q_0^2}{(\det{b}+n^2)} (q_0^2-\alpha)(q_0^2-\beta)=
-\frac{q_0^2}{(\det{b}+n^2)}\left[q_0^4 - (\alpha+\beta) q_0^2
+\alpha\beta
\right], \label{freC02}$$ with some polynomials $\alpha$ and $\beta$ of order 2 in $q_a$. This implies the relations $$\label{eqab}
\alpha+\beta=2\left({\bar q}^2 - (qn)^2\right) , \qquad
\alpha\beta= \left({\bar q}^2 + (qn)^2\right)^2 \,,$$ with $\bar{q}^2 := q_a q_b \bar{b}^{ab}$. Since $\alpha$ and $\beta$ enter symmetrically in (\[eqab\]), the solutions of this nonlinear system can be given in the form $$\label{solalphabeta}
\alpha= -\,\left[(qn) +\sqrt{-\overline{q}^2}\right]^2 , \qquad
\beta= -\,\left[(qn) - \sqrt{-\overline{q}^2}\right]^2 \, .$$
Thus, the question of the reducibility of the Fresnel equation translates into the algebraic problem of whether the square root $\sqrt{-\overline{q}^2}$ is a real linear polynomial in $q_a$. There are three cases, depending on the rank of the $3\times 3$ matrix $b_{ab}$.
\(i) When $b_{ab}$ has rank 3, in other words, when $\det b \neq 0$, then we can write $\overline{q}^2 = q_aq_b \,b^{ab}\det b$, and the general conclusion is that [*no factorization into light-cones*]{} is possible (the roots $\alpha$ are complex), unless $n^a=0$. This latter condition implies that the constitutive tensor is symmetric, and the results of [@OH99; @OFR00; @GR01] are recovered.
\(ii) When $b_{ab}$ has rank 2, i.e., $\det b =0$, but at least one of the minors is nontrivial. Then, without loss of generality, we can assume the following structure of the matrix $b$: $$b_{ab} = \left(\begin{array}{ccc}b_{11}&b_{12}&0\\ b_{12}&b_{22}& 0\\
0 & 0 & 0\end{array}\right).\label{b-deg}$$ Its only non-vanishing minor is $$\overline{b}^{33} = \left|\begin{array}{cc}b_{11}&b_{12}\\ b_{21}&b_{22}
\end{array}\right| = b_{11}b_{22}-b_{12}^2 \neq 0.$$ Note that (\[b-deg\]) is the most general form of a rank 2 matrix $b_{ab}$, up to a renaming of the coordinates. In order to avoid complex solutions, we have to assume that the minor $\overline{b}^{33}
= -\,\mu^2 < 0$, so that $\sqrt{-{\bar q}^2}=\mu q_3$. Then (\[solalphabeta\]) leads to $$\alpha = -\,\left[q_1n^1 + q_2n^2 + q_3(n^3+\mu)\right]^2, \qquad
\beta = -\,\left[q_1n^1 + q_2n^2 + q_3(n^3-\mu)\right]^2\,.$$ The interpretation is clear: we have birefringence, i.e., two light-cones. In this case, the Fresnel equation is found to be $$\begin{aligned}
{\cal W} = -\,{\frac {q_0^2}{b_{11}(n^1)^2+2b_{12}n^1n^2+b_{22}(n^2)^2}}
\,&& (q_0^2+\left[q_1n^1 + q_2n^2 + q_3(n^3+\mu)\right]^2) \nonumber \\
&&\times (q_0^2+\left[q_1n^1 + q_2n^2 + q_3(n^3-\mu)\right]^2)=0\,. \end{aligned}$$ Then we can read off, up to conformal factors, the components of the two corresponding ‘metric’ tensors defining the light-cones: $$g_1^{ij}=\left( \begin{array}{cccc}
1 & 0 & 0 & 0 \\
0 & (n^1)^2 & n^1 n^2 & n^1(n^3+\mu) \\
0 & n^1n^2 & (n^2)^2 & n^2n^3 \\
0 & n^1(n^3+\mu) & n^2n^3 & (n^3+\mu)^2
\end{array}\right) \, ,$$ $$g_2^{ij}=\left( \begin{array}{cccc}
1 & 0 & 0 & 0 \\
0 & (n^1)^2 & n^1 n^2 & n^1(n^3-\mu) \\
0 & n^1n^2 & (n^2)^2 & n^2n^3 \\
0 & n^1(n^3-\mu) & n^2n^3 & (n^3-\mu)^2
\end{array}\right) \, .$$ We can verify that $\det (g_1^{ij})=\det (g_2^{ij})=
(n^1)^2(n^2)^2{\bar{b}}^{33}=-(n^1)^2(n^2)^2\mu^2<0$, so that both metrics have the correct Lorentzian signature.
\(iii) When the $3\times 3$ matrix $b_{ab}$ has rank 1: In this case all the minors are zero, i.e., $\overline{q}^{ab}=0$, which corresponds to the case 2 for $\mu=0$. We then see that the Fresnel equation reduces to a single light cone, but the resulting metric is degenerated, since $\det (g^{ij})=0$.
Discussion and conclusion
=========================
In this paper we extended appreciably our earlier results [@OH99; @HOR00; @OFR00; @GR01] of linear pre-metric electrodynamics by relaxing the symmetry and the closure property of the constitutive tensor $\chi^{ijkl}$ of spacetime. In the most general case of a linear spacetime relation (when both [*closure*]{} and [*symmetry*]{} are absent), the wave propagation is governed by the newly derived Fresnel type equation (\[Fresnel\]) which is still of quartic order, contrary to the claims in Ref. [@DD99]. We have deduced the tensor density ${\cal G}^{ijkl}$ in (\[G4\]) which induces the generally covariant form of the Fresnel equation (\[Fresnel\]). We studied some properties of ${\cal G}^{ijkl}$ and showed that $^{(1)}\chi^{ijkl}$, as the leading part of $\chi^{ijkl}$, is indispensable for a decent propagation of electromagnetic waves. However, also each of the two remaining parts, the axion $^{(3)}\chi^{ijkl}$ and the skewon $^{(2)}\chi^{ijkl}$, has a well-defined physical meaning. The axion drops out from light propagation. High energy physicist search for such type of particles. The skewon seems to be T-odd and related to dissipative processes. Its influence on light propagation, see (\[einsundzw\]), deserves further study.
In Sec.\[complex\], the general solution of the closure relation is presented for the case of an asymmetric linear $\chi^{ijkl}$. Within a particular class of the asymmetric solutions of the closure relation, the reduction of the Fresnel equation was analyzed in detail. We have demonstrated that some of these solutions can yield birefringence. A preliminary study of the general case with $K\neq 0$ shows that the conclusions remain qualitatively the same as those presented in Sec.\[metric\].
Thus, we can conclude that the conditions of closure and symmetry of $\chi^{ijkl}$ are sufficient for the existence of a well-defined light cone structure. If any of these conditions is violated, the light cone structure seems to be lost. The necessary conditions have still to be found. Of course, if one wants to study possible violations of Lorentz invariance by means of the skewon $^{(2)}\chi^{ijkl}$, e.g., then the light cone structure cannot be considered as sacrosanct any longer.
[**Acknowledgments**]{} GFR would like to thank the German Academic Exchange Service (DAAD) for a graduate fellowship (Kennziffer A/98/00829). YNO is grateful to the Alexander von Humboldt Foundation for support.
An effective constitutive tensor {#ab}
================================
Consider an arbitrary local spacetime relation $H=H(F)$. In components, the corresponding Maxwell equation, for $J=0$, reads $$\label{mehc}
\epsilon^{ijkl}\,\partial_j\, H_{kl}=0 \,.$$ A small perturbation $\Delta F$ of the electromagnetic field around some background $\bar F$ can be written as $F={\bar F}+\Delta F$. Then, to first order in the perturbation, we have for the excitation $$\label{expH}
H_{kl}(F)=H_{kl}({\bar F})+\frac{1}{2} \left.\frac{\partial
H_{kl}}{\partial F_{mn}}\right|_{\bar F} \Delta F_{mn} \,.$$ Inserting (\[expH\]) into (\[mehc\]) and assuming that the background field $\bar F$ is a solution of (\[mehc\]), i.e. $\epsilon^{ijkl}\,\partial_j\,H_{kl}({\bar F})=0$, we obtain an equation for the perturbation: $$\partial_j\left(\chi^{ijkl}_{\rm eff} \Delta F_{kl}\right)=0 \,,\qquad
\chi^{ijkl}_{\rm eff}:=\frac{1}{2}\epsilon^{ijmn}
\left.\frac{\partial H_{mn}}{\partial F_{kl}}\right|_{\bar F} \,.$$ The effective constitutive tensor density $\chi^{ijkl}_{\rm eff}$ will, in general, depend on the local constitutive law and on the background field $\bar F$.
[99]{}
F.W. Hehl, Yu.N. Obukhov, [*A gentle introduction to the foundations of classical electrodynamics: The meaning of the excitations (D,H) and the field strengths (E, B)*]{}. Los Alamos Eprint Archive physics/0005084 (2000).
E.J. Post, [*Formal Structure of Electromagnetics – General Covariance and Electromagnetics*]{} (North Holland: Amsterdam, 1962, and Dover: Mineola, New York, 1997).
F.W. Hehl, Yu.N. Obukhov, [*Foundations of classical electrodynamics*]{} (Birkhäuser: Boston, MA, 2002) To be published.
L. Cooper and G.E. Stedman, [*Axion detection by ring lasers*]{}, [*Phys. Lett.*]{} [**B357**]{} (1995) 464-468. S.L. Cheng, C.Q. Geng, W.-T. Ni, [*Axion–photon couplings in invisible axion models,*]{} [*Phys. Rev.*]{} [**D52**]{} (1995) 3132-3135. G.E. Stedman, [*Ring-laser tests of fundamental physics and geophysics*]{}, [*Rep. Progr. Phys.*]{} [**60**]{} (1997) 615-688.
W.-T. Ni, [*A non-metric theory of gravity*]{}, [*Dept. Physics, Montana State University, Bozeman. Preprint December 1973*]{}. The paper is available via [http://gravity5.phys. nthu.edu.tw/]{}.
W.-T. Ni, [*Equivalence principles and electromagnetism*]{}, [*Phys. Rev. Lett.*]{} [**38**]{} (1977) 301-304.
R.B. Mann, J.W. Moffat, J.G. Taylor, [*Particle spectrum in a new theory of gravitation*]{}, [*Phys. Lett.*]{} [**B97**]{} (1980) 73-76.
H.H. Soleng, J.O. Eeg, [*Skewons and gravitons*]{}, [*Acta Phys. Polon.*]{} [**B23**]{} (1992) 87-114.
J.F. Nieves and P.B. Pal, [*P and CP-odd terms in the photon self-energy within a medium*]{}, [*Phys. Rev.*]{} [ **D39**]{} (1989) 652-659.
J.F. Nieves and P.B. Pal, [*The third electromagnetic constant of an isotropic medium*]{}, [*Am. J. Phys.*]{} [**62**]{} (1994) 207-216.
J. Hadamard, [*Leçons sur la propagation des ondes et les équations de l’hydrodynamique*]{} (Hermann: Paris, 1903).
A. Lichnerowicz, [*Relativity theory and mathematical physics*]{}, in: [*“Astrofisica e cosmologia gravitazione quanti e relatività", Centenario di Einstein*]{} (Giunti Barbera: Firenze, 1979).
M. Novello, V.A. De Lorenci, J.M. Salim, and R. Klippert, [*Geometrical aspects of light propagation in nonlinear electrodynamics*]{}, [*Phys. Rev.*]{} [**D61**]{} (2000) 045001 (10 pages). C. Barcelo, S. Liberati and M. Visser, [*Analog gravity from field theory normal modes?*]{} [*Class. Quantum Grav.*]{} [**18**]{} (2001) 3595-3610.
U. Leonhardt and P. Piwnicki, [*Relativistic effects of light in moving media with extremely low group velocity*]{}, [*Phys. Rev. Lett.*]{} [**84**]{} (2000) 822-825; [*Phys. Rev. Lett.*]{} [**85**]{} (2000) 5253;\
M. Visser, [*Phys. Rev. Lett.*]{} [**85**]{} (2000) 5252.
Yu.N. Obukhov, T. Fukui, and G.F. Rubilar, [*Wave propagation in linear electrodynamics*]{}, [*Phys. Rev.*]{} [**D62**]{} (2000) 044050 (5 pages). gr-qc/0005018
I.E. Tamm, [*Relativistic crystal optics in relation with the geometry of bi-quadratic form*]{}, [*Zhurn. Ross. Fiz.-Khim. Ob.*]{} [**57**]{} (1925) 209-224 (in Russian). Reprinted in: I.E. Tamm, [*Collected Papers*]{} (Nauka: Moscow, 1975) Vol. 1, pp. 33-61 (in Russian). See also: I.E. Tamm, [ *Electrodynamics of an anisotropic medium in special relativity theory*]{}, ibid, pp. 19-31; a short version in German, written jointly with L.I. Mandelstam, ibid. pp. 62-67.
J.A. Schouten, [*Tensor Analysis for Physicists.*]{} 2nd ed. reprinted (Dover: Mineola, New York 1989), page 22.
V.I. Denisov and M.I. Denisov, [*Verification of Einstein’s principle of equivalence using a laser gyroscope in terrestrial conditions*]{}, [*Phys. Rev.*]{} [**D 60**]{} (1999) 047301 (4 pages).
C. Lämmerzahl et al., [*Reasons for the electromagnetic field to obey Maxwell’s equations.*]{} Preprint, University of Konstanz, August 1999.
Yu.N. Obukhov and F.W. Hehl, [*Spacetime metric from linear electrodynamics*]{}, [*Phys. Lett.*]{} [**B458**]{} (1999) 466-470. gr-qc/9904067
F.W. Hehl, Yu.N. Obukhov, and G.F. Rubilar, [ *Spacetime metric from linear electrodynamics II*]{}, [*talk given at Internat. European Conf. on Gravitation “Journées Relativistes 99", Weimar, Germany, 12-17 Sep 1999.*]{} [*Ann. d. Phys. (Leipzig)*]{} [**9**]{} (2000) Special issue, SI-71-78. Guest editors: G. Neugebauer and R. Collier, Jena. gr-qc/9911096.
A. Gross and G.F. Rubilar, [*On the derivation of the spacetime metric from linear electrodynamics*]{}, [*Phys. Lett.*]{} [**A 285**]{} (2001) 267-272. gr-qc/0103016.
F.W. Hehl, Yu.N. Obukhov, [*On the energy-momentum current of the electromagnetic field in a pre-metric axiomatic approach*]{}. Los Alamos Eprint Archive gr-qc/0103020.
R.A. Toupin, [*Elasticity and electro-magnetics*]{}, in: [*Non-Linear Continuum Theories, C.I.M.E. Conference, Bressanone, Italy 1965*]{}. C. Truesdell and G. Grioli coordinators. Pp.203-342.
M. Schönberg, [*Electromagnetism and gravitation*]{}, [*Rivista Brasileira de Fisica*]{} [**1**]{} (1971) 91-122.
A.Z. Jadczyk, [*Electromagnetic permeability of the vacuum and light-cone structure*]{}, [*Bull. Acad. Pol. Sci., Sér. sci. phys. et astr.*]{} [**27**]{} (1979) 91-94.
[^1]: Email: [email protected].
[^2]: Email: [email protected]. Permanent address: Department of Theoretical Physics, Moscow State University, 117234 Moscow, Russia.
[^3]: Email: [email protected].
[^4]: Nevertheless, since Maxwell’s equations for material media have the same appearance as the vacuum equations (\[me\]), a constitutive relation for a material medium can be reminiscent of the [*spacetime relation*]{} — and we will profit from this analogy.
[^5]: The name ‘skewon’ has been used earlier in a post-Einsteinian theory of gravity by Mann, Moffat, and Taylor [@MMT80], see also Soleng and Eeg[@SE92]. We are using this name in a different context since its old meaning doesn’t seem to be in use any longer, see a title search in the data banks of SLAC and DESY.
[^6]: See [ http://grtensor.org]{}.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The Large Hadron Collider (LHC) provides data which give information on dark matter. In particular, measurements related to the Higgs sector lead to strong constraints on the invisible sector which are competitive with astrophysical limits. Some recent LHC results on dark matter coming from the Higgs sector in the Inert Doublet Model (IDM) are presented.'
author:
- 'M. Krawczyk, M. Matej, D. Sokołowska, B. Świeżewska'
date: 'December 30, 2014'
title: 'Universe in the light of LHC [^1]'
---
Introduction
============
In the autumn of the year 2014 one can safely conclude that the SM-like Higgs scenario [@BE; @Higgs; @Kibble] is being observed at the LHC [@ATLAS-Higgs; @CMS-Higgs]. Such a scenario can be realized in various models beyond the Standard Model (SM). It was found recently that the LHC is very effective in constraining models with so called Higgs-portal to the dark matter (DM), see eg. [@portalH]. In particular, LHC results on the Higgs boson properties can give stronger limits on the Higgs-DM couplings than the astrophysical DM experiments. In addition, some important constraints are coming from the dedicated search of dark matter at the LHC [@LHCinv]. Let us start with a little bit of history. It was only 50 years ago when the Quark Model as well as the mass generation mechanism had been proposed. These were crucial steps towards building a theory of elementary particles known as the Standard Model. The first idea of a global SU(3) symmetry arose from an observation and classification of a plethora of hadrons, which were being discovered copiously in early 1950s. It led to our current understanding of the structure of matter at the fundamental level, probed up to a distance of 10$^{-18}$ cm, in the form of 3 generations of quarks and leptons.
The masses of these particles show no clear pattern, besides the fact that the second generation of fermions is heavier than the first one, and the third one is the heaviest. Among the particles which are carriers of fundamental interactions, the photon (electromagnetic interaction) and the gluons (strong interaction) are massless, while $W^{\pm}$ and $Z$, the carriers of the electroweak force, are massive, actually very massive as compared to the proton. It was already known in the 1960s that this may create a serious problem in describing a very short range weak interaction (a point-like interaction according to Fermi) in a theoretical approach based on a local symmetry.
Brout-Englert-Higgs mechanism in the SM
---------------------------------------
The Brout-Englert-Higgs mechanism (B-E-H), proposed already in 1960s, is based on spontaneous breaking of the EW symmetry $SU(2) \times U(1)$ to $U(1)_{\mathrm{QED}}$[@BE; @Higgs; @Kibble]. One SU(2) doublet $\Phi$ of spin 0 fields with a non-zero vacuum expectation value $v$ (VEV) is introduced, and the gauge bosons and fermions acquire mass thanks to interaction with this constant field component. [ Mass of $W^\pm$ generated in this way is equal to $ M_W = g v/2$, and at tree level $\rho=\frac{M_W}{M_Z \cos \theta_W} =1$. Masses of fermions are generated due to Yukawa interaction with $\Phi$.]{}
The Higgs boson $h$ which arises in the B-E-H mechanism has spin 0, even CP parity and no electric charge. Its couplings to the SM particles are all fixed, being proportional to their masses. The only unknown parameter is (was) $M_h$ related to the strength of Higgs self-interaction. Long term hunting for a Higgs boson seems to have reached its finale in the summer 2012, when at the LHC the Higgs boson with mass around 125 GeV has been discovered. Up to now, with all collected data (already 1 million of Higgses!), the signal strengths in various channels (defined with respect to the SM prediction) are close to 1, and the observed scenario can be described as a SM-like Higgs scenario.
SM-like Higgs scenarios at the LHC
----------------------------------
Although the SM is in very good agreement with existing data, there are many serious arguments to go beyond it. The SM has many free parameters, contains massless neutrinos, does not have a DM candidate, cannot describe baryon asymmetry of the Universe, etc. The recently discovered 125 GeV scalar has properties very close to those predicted by the SM. But how close? As long as other new particles are not seen at the LHC the only relevant BSM models are those allowing for SM-like scenario, i.e., with a SM-like Higgs boson and other new particles too heavy or too weakly interacting to be observed in existing experiments. The main production channel of the Higgs particle at the LHC is gluon-gluon fusion. The channels allowing most precise measurements are Higgs decays to $\gamma \gamma$ and $ZZ$. Loop couplings of the Higgs to gauge bosons $gg$, $\gamma \gamma $, $\gamma Z$ are sensitive to new physics (even to contribution of very heavy particles due to nondecoupling effects). The overall signal strength is equal to $\mu= 1.00 \pm 0.13$ (CMS) [@CMS-Higgs], $1.30\pm 0.12 \mathrm{(stat)}^{+0.14}_{-0.11}\mathrm{(syst)}$ (ATLAS) [@ATLAS-Higgs].
Dark Matter
-----------
Throughout the years much evidence for the existence of DM has been collected: rotation curves of galaxies, gravitational lensing, etc. [@dm-corfu]. A typical candidate for DM is the so-called WIMP (weakly interacting massive particle). The DM relic density is inferred from the measurements made by WMAP and Planck with a good accuracy [@dm-relic]. There are other astrophysical experiments searching for DM, either directly (via scattering off nuclei) or indirectly (search for products of DM annihilation or decay). Unfortunately, the picture given by these experiments is not entirely consistent. However, some information about DM can be drawn from the LHC measurements, and hopefully it can shed some light on its nature.
TheInert Doublet Model
======================
Among the simplest extensions of the Higgs sector in the SM are models with two SU(2) doublets (Two Higgs Doublet Models – 2HDMs). In the non-supersymmetric 2HDMs a special role is played by the Inert Doublet Model (IDM) – the only version of 2HDM with a stable particle (scalar) [@Ma; @Barbieri].
In the IDM The scalars’ interactions are defined by the following potential $$\begin{aligned}
V=&-\frac{1}{2}\Big[m_{11}^{2}(\phi_{S}^{\dagger}\phi_{S})+m_{22}^{2}(\phi_{D}^{\dagger}\phi_{D})\Big]+\frac{1}{2}\Big[\lambda_{1}(\phi_{S}^{\dagger}\phi_{S})^{2}+\lambda_{2}(\phi_{D}^{\dagger}\phi_{D})^{2}\Big]\\
&+\lambda_{3}(\phi_{S}^{\dagger}\phi_{S})(\phi_{D}^{\dagger}\phi_{D})+\lambda_{4}(\phi_{S}^{\dagger}\phi_{D})(\phi_{D}^{\dagger}\phi_{S})+\frac{1}{2}\lambda_{5}\Big[(\phi_{S}^{\dagger}\phi_{D})^{2}+(\phi_{D}^{\dagger}\phi_{S})^{2}\Big].\nonumber\end{aligned}$$ This potential possesses a global discrete $\mathbb{Z}_2$-type symmetry $D$ under an action of which the field $\phi_D$ changes sign, while $\phi_S$ remains untouched. The interactions with fermions are chosen in the IDM such as to preserve this symmetry, i.e., only $\phi_S$ couples to fermions. In this way the whole IDM lagrangian is $D$-symmetric, and moreover, the vacuum state of this model is such that $D$ is not broken spontaneously. The VEVs of the two doublets read $$\langle\phi_S\rangle= \left( \begin{array}{c} 0\\
\frac{v_S}{\sqrt 2}\end{array} \right),\quad
\langle\phi_D\rangle= \left( \begin{array}{c} 0\\
0 \end{array} \right).\label{vacu}$$ With these choices the model possesses an exact $D$-symmetry, which leads to a conserved quantum number ($D$ parity). Because of that the lightest $D$-odd particle is stable, and constitutes a good candidate for a DM particle.
The particle spectrum of the IDM consists of the Higgs boson $h$ which follows from the $\phi_S$ doublet, and the dark scalars $H$, $A$ and $H^{\pm}$ coming from $\phi_D$. The Higgs boson has all tree-level couplings to fermions and gauge bosons equal to the SM ones. Nonetheless, some non-SM effects can occur at the loop level, due to the existence of new scalars. The dark scalars do not couple to fermions at tree level but they do interact with the gauge bosons (through the covariant derivative) and the Higgs particle. The lightest one among them that is neutral plays the role of the DM particle. Here we assume that $M_H<M_A, {M_{H^{\pm}}}$, hence $H$ is the DM candidate in our model.
Deviations from the SM properties of the Higgs boson can be observed in two ways, because of decays of the Higgs into invisible dark particles or because of additional loop effects thereof. In the following we will first discuss invisible decays of the Higgs boson, and then loop induced decay of the Higgs boson to a pair of photons.
Invisible Higgs decays
----------------------
The Higgs boson of the IDM has additional, non-SM decay channels leading to dark particles: $h\to
AA, HH {\rm{\,\,or}}\,\, H^{\pm}H^{\mp}$. The last channel is excluded (at tree-level) by the LEP limits for ${M_{H^{\pm}}}$: ${M_{H^{\pm}}}\gtrsim 70{\,\mathrm{GeV}}$. The partial decay width for the process $h\to HH$ reads (see e.g. Ref. [@jhep]) \[part\] (hHH)=, where $\lambda_{345}=\lambda_3+\lambda_4+\lambda_5$ is proportional to the coupling between the Higgs boson and a pair of DM particles. For the decay $h\to AA$ the parameters $\lambda_{345}$ and $M_H$ have to be replaced by $\lambda_{345}^-=\lambda_3+\lambda_4-\lambda_5$ and $M_A$, respectively.
Since the decay width (\[part\]) depends on the mass of the product of the decay and its coupling to the Higgs boson, these quantities can be constrained with the use of the LHC results on the branching ratio of the Higgs boson decay to invisible particles. In the same way the measurement of the total Higgs decay width can be used, since $\Gamma (h\to \mathrm{inv})$ contributes significantly to it (see next section). Below, for the sake of simplicity we will assume that $A$ is too heavy for the $h\to AA$ process to be allowed, i.e., $M_H <M_h/2$ and $M_A>M_h/2$. In Fig. \[invisible\] the constraints on $\lambda_{345}$ and $M_H$, coming from experimental constraints on Br$(h\to \mathrm{inv})<0.37$ [@invisible] and on the total width $\Gamma(h)<5.4\ \Gamma(h)^{\mathrm{SM}}$ [@gamma], are presented. From Fig. \[invisible\] one can see that the coupling $\lambda_{345}$ is constrained by Br$(h\to\mathrm{inv})$ to a small value, $|\lambda_{345}|\lesssim0.05$ for $M_H<62{\,\mathrm{GeV}}$.
![Constraints on $\lambda_{345}$ and $M_H$ following from the LHC measurement of Br$(h\to \mathrm{inv})$ and $\Gamma(h)$. The region between the corresponding curves is allowed. We assume that the decay $h\to AA$ is kinematically forbidden. \[invisible\]](rysunki/inv.pdf){width=".6\textwidth"}
Higgs decays to $\gamma\gamma$
------------------------------
The differences between the SM and the IDM can also be observed in the loop induced decays of the Higgs boson, $h\to \gamma\gamma$ and $h\to Z\gamma$. The first of these decays, being measured very precisely, recently gained much attention, since the first measurements showed some deviation from the SM expectation giving a hint on the existence of new physics. Nowadays, these measurements converged to the SM, as the observed signal strengths (often denoted by $\mu_{\gamma \gamma}$) are ${R_{\gamma\gamma}}= 1.17\pm 0.27$ (ATLAS) [@diphoton-ATLAS], $1.14^{+0.26}_{-0.23}$ (CMS) [@diphoton-CMS], where the expectation of the SM is ${R_{\gamma\gamma}}=1$. We see, than new physics effects are still acceptable within the experimental bounds. Note that the $Z\gamma$ signal strength has not yet been measured with enough precision to constrain new physics.
Let us consider $R_{\gamma\gamma}$ for the 125 GeV-$h$ in the IDM (see e.g. Refs. [@rgg; @Posch; @Arhrib]) R\_:=, where we have used the narrow-width approximation and the fact that the main production cross section $gg\to h$ is in the IDM the same as in the SM.
In the formula above $\textrm{Br}(h\to\gamma\gamma)^{\textrm {SM}}$ is known, and $\textrm{Br}(h\to\gamma\gamma)^{\textrm {IDM}}={\Gamma(h\to\gamma\gamma)^{\textrm {IDM}}}/{\Gamma(h)^{\textrm {IDM}}}$. All the tree-level decay widths of the Higgs boson to SM particles are in the IDM the same as in the SM. Only the existence of the invisible decay channels, and the $\gamma\gamma$, and $Z\gamma$ decays can modify the total decay width. However, branching ratios of the latter are very small, at the order of $10^{-3}-10^{-2}$ so they can be ignored, and to a good approximation, only invisible channels modify $\Gamma(h)$ (we used this fact already in Sec. 2.1). The branching ratios in the IDM are presented in Fig. \[br\] as functions of $\lambda_{345}$. Note that once the invisible channels are kinematically allowed, they dominate over the SM channels, so in general they tend to suppress ${R_{\gamma\gamma}}$.
![Branching ratios of the Higgs boson in the IDM as functions of ${\lambda_{345}}$. Left: invisible channels open ($M_H=50{\,\mathrm{GeV}}$, $M_A=58{\,\mathrm{GeV}}$). Right: invisible channels closed ($M_H=75{\,\mathrm{GeV}}$, $M_A>M_H$). Figure from Ref. [@rgg]. \[br\]](rysunki/br1.png "fig:"){width=".35\textwidth"} ![Branching ratios of the Higgs boson in the IDM as functions of ${\lambda_{345}}$. Left: invisible channels open ($M_H=50{\,\mathrm{GeV}}$, $M_A=58{\,\mathrm{GeV}}$). Right: invisible channels closed ($M_H=75{\,\mathrm{GeV}}$, $M_A>M_H$). Figure from Ref. [@rgg]. \[br\]](rysunki/br2.png "fig:"){width=".35\textwidth"}
If invisible channels are closed, the the partial decay width $\Gamma(h\to \gamma\gamma)$ although small can be a valuable source of information. In the SM the $h\to \gamma\gamma$ decay is induced by the a $W^{\pm}$ boson loop and fermionic loops (the top quark dominates). In general, in the IDM $\Gamma(h\to \gamma\gamma)$ differs from the SM one because of an extra contribution due to the charged scalar, $H^{\pm}$. This contribution can interfere either constructively or destructively with the SM part. Already in Fig. \[br\] (right panel) it is visible that Br$(h\to \gamma\gamma)$ can be enhanced or suppressed with respect to the SM.
### Enhanced diphoton signal strength
Let us first analyse the consequences of enhanced signal strength (we follow Ref. [@rgg]). In the left panel of Fig. \[rgg-gtr-1\] the dependence of ${R_{\gamma\gamma}}$ on $M_H$ is shown. One can clearly see that for $M_H<M_h/2\approx 62.5{\,\mathrm{GeV}}$ the diphoton signal strength is always suppressed with respect to the SM. This means that if enhancement of $R_{\gamma \gamma}$ is observed, DM with mass below 62.5 GeV is excluded.
In the right panel of Fig. \[rgg-gtr-1\] the allowed $(m_{22}^2, {M_{H^{\pm}}})$ region, obtained by scanning the parameter space subject to relevant theoretical and experimental constraints[^2], is presented. The parameter $m_{22}^2$ is important for ${R_{\gamma\gamma}}$ because the coupling between the Higgs boson and the charged scalar is proportional to $2{M_{H^{\pm}}}^2+m_{22}^2$. In the region marked by light green (gray) ${R_{\gamma\gamma}}>1$, while the (purple) lines indicate constant values of ${R_{\gamma\gamma}}$. Note that for ${R_{\gamma\gamma}}\geq1$ the viable region is unconstrained, however for substantial enhancement of ${R_{\gamma\gamma}}$ the allowed region is bounded. For example for ${R_{\gamma\gamma}}>1.2$, only fairly light charged scalar (and since $M_H<{M_{H^{\pm}}}$ also DM) is allowed, ${M_{H^{\pm}}}, M_H \lesssim 154{\,\mathrm{GeV}}$. The case where ${R_{\gamma\gamma}}$ goes below 1 will be analysed in the next section, and combined with the DM astrophysical measurements.
![Left: ${R_{\gamma\gamma}}$ dependence on $M_H$. Right: region allowed by the experimental and theoretical constraints in the $(m_{22}^2,\ {M_{H^{\pm}}})$ plane. Light green (gray) indicates the region where ${R_{\gamma\gamma}}\geqslant1$, the lines correspond to the constant values of ${R_{\gamma\gamma}}$. Plots are made for $-25\cdot10^4 {\,\mathrm{GeV}}^2\leqslant m_{22}^2\leqslant 9\cdot 10^4{\,\mathrm{GeV}}^2$. From Ref. [@rgg] \[rgg-gtr-1\]](rysunki/Hr "fig:"){width=".45\textwidth"} ![Left: ${R_{\gamma\gamma}}$ dependence on $M_H$. Right: region allowed by the experimental and theoretical constraints in the $(m_{22}^2,\ {M_{H^{\pm}}})$ plane. Light green (gray) indicates the region where ${R_{\gamma\gamma}}\geqslant1$, the lines correspond to the constant values of ${R_{\gamma\gamma}}$. Plots are made for $-25\cdot10^4 {\,\mathrm{GeV}}^2\leqslant m_{22}^2\leqslant 9\cdot 10^4{\,\mathrm{GeV}}^2$. From Ref. [@rgg] \[rgg-gtr-1\]](rysunki/m22Hp "fig:"){width=".45\textwidth"}
DM constraints from the Higgs LHC and Planck data
--------------------------------------------------
The current Planck $3\sigma$ limit for DM relic density is $0.1118<\Omega_{DM} h^2<0.1280$ [@dm-relic]. $\Omega_{DM} h^2$ depends on DM annihilation and production channels, so this measurement constrains the mass and couplings of the DM candidate. The IDM is a so-called “Higgs-portal” DM model, i.e., in a wide range of masses the DM candidate couples to fermions mainly through the exchange of $h$. Therefore the coupling ${\lambda_{345}}$ between the Higgs and the DM candidate is constrained by relic density measurement. On the other hand the same coupling, as was shown before, is important for the diphoton signal strength. This gives us an opportunity to combine these two types of constraints. In the following we will examine the case ${R_{\gamma\gamma}}>0.7$ (with agreement with $3 \sigma$ LHC limit), we studied other cases in [@jhep].
Fig. \[l345-constraints\] shows how the constraints arise. In the left panel ${R_{\gamma\gamma}}$ as a function of ${\lambda_{345}}$ is shown (for fixed values of masses). If we require that ${R_{\gamma\gamma}}>0.7$, upper and lower bounds on ${\lambda_{345}}$ arise.
![Left: ${R_{\gamma\gamma}}$ as a function of ${\lambda_{345}}$ for $M_H=55{\,\mathrm{GeV}}, M_A=60{\,\mathrm{GeV}}, {M_{H^{\pm}}}=120{\,\mathrm{GeV}}$ (from Ref. [@jhep]). Right: relic density of DM as a function of ${\lambda_{345}}$ for different DM mass. The WMAP 3-$\sigma$ bound is marked by the dashed black horizontal lines (from Ref. [@dm]). \[l345-constraints\]](rysunki/ogr-l345-rg07 "fig:"){width=".45\textwidth"} ![Left: ${R_{\gamma\gamma}}$ as a function of ${\lambda_{345}}$ for $M_H=55{\,\mathrm{GeV}}, M_A=60{\,\mathrm{GeV}}, {M_{H^{\pm}}}=120{\,\mathrm{GeV}}$ (from Ref. [@jhep]). Right: relic density of DM as a function of ${\lambda_{345}}$ for different DM mass. The WMAP 3-$\sigma$ bound is marked by the dashed black horizontal lines (from Ref. [@dm]). \[l345-constraints\]](rysunki/omega "fig:"){width=".45\textwidth"}
In the right panel the relic density of the DM as a function of ${\lambda_{345}}$ is presented for different values of $M_H$. To fall within the $\Omega_{DM}h^2$ experimental limits (i.e., between the black dashed horizontal lines)[^3] the value of ${\lambda_{345}}$ should be between the upper and lower limits. These two types of bounds will be combined in the following.
It has been shown in previous works [@dm-inne; @dm] that DM in the IDM can have the correct relic abundance only in three regions: for very light DM ($M_H\lesssim 10{\,\mathrm{GeV}}$), intermediate DM ($40{\,\mathrm{GeV}}\lesssim M_H \lesssim 160 {\,\mathrm{GeV}}$), and heavy DM ($M_H\gtrsim 500{\,\mathrm{GeV}}$). We will analyse these cases separately, following Ref. [@jhep].
As can be seen in the right panel of Fig. \[l345-constraints\] the right $\Omega_{DM}h^2$ of very light DM is obtained for $|{\lambda_{345}}|\sim \mathcal{O}(0.5)$. Smaller coupling means that DM does not annihilate efficiently enough, and the relic abundance is too big. As ${\lambda_{345}}$ in agreement with the LHC limit ${R_{\gamma\gamma}}>0.7$ is around $|{\lambda_{345}}|<0.04$, those two requirements cannot be reconciled, and the very light DM is excluded.
![Maps of the values of ${R_{\gamma\gamma}}$ in the $(M_H,\ {\lambda_{345}})$ plane for the intermediate DM with $h\to HH$ channel open (left), $h\to HH$ channel closed (middle), and for heavy DM (right) in comparison with the allowed by Planck (red) bands. See Ref. [@jhep].\[fig:dm\]](rysunki/RggPlanckOpen "fig:"){width=".32\textwidth"} ![Maps of the values of ${R_{\gamma\gamma}}$ in the $(M_H,\ {\lambda_{345}})$ plane for the intermediate DM with $h\to HH$ channel open (left), $h\to HH$ channel closed (middle), and for heavy DM (right) in comparison with the allowed by Planck (red) bands. See Ref. [@jhep].\[fig:dm\]](rysunki/RggPlanckClosed "fig:"){width=".32\textwidth"} ![Maps of the values of ${R_{\gamma\gamma}}$ in the $(M_H,\ {\lambda_{345}})$ plane for the intermediate DM with $h\to HH$ channel open (left), $h\to HH$ channel closed (middle), and for heavy DM (right) in comparison with the allowed by Planck (red) bands. See Ref. [@jhep].\[fig:dm\]](rysunki/RggPlanckHigh "fig:"){width=".32\textwidth"}
Results for intermediate and heavy masses are presented in Fig. \[fig:dm\]. The shades (of blue) indicate the values of ${R_{\gamma\gamma}}$. On this, constraints from Planck are superposed. The dark gray region is excluded ($\Omega_{DM}h^2$ is too big). The two regions indicated by arrows/red bands are in agreement with Planck data (correct relic density), and in the remaining region the relic density is too low (another DM component would be necessary in order to comply with relic density data). In the left panel a plot for intermediate DM, with $M_H<M_h/2$ is presented. One can see that relic density constraints (Planck) are in agreement with the assumption ${R_{\gamma\gamma}}>0.7$ only for $M_H>53{\,\mathrm{GeV}}$. In the middle panel intermediate DM with $M_H>M_h/2$ is analysed. Here all the points that are in agreement with the Planck measurement also give ${R_{\gamma\gamma}}>0.7$. However, if Planck constraints are to be met, no enhancement in ${R_{\gamma\gamma}}$ is possible. For the heavy DM (right panel of Fig. \[fig:dm\]) we can get correct relic density for all values of masses. Note that ${R_{\gamma\gamma}}$ is very close to 1 for this case.
Comparison with direct DM detection experiments
-----------------------------------------------
The constraints obtained above can be compared with the results of the direct experimental search of the DM [@jhep], where the DM is supposed to scatter off the nuclei. In the Higgs-portal models, among them IDM, the cross section $\sigma_{DM,N}$ is proportional to the square of the coupling of Higgs to DM (${\lambda_{345}}^2$) because the DM interacts with the nucleus through the exchange of the Higgs boson, $\sigma_{DM,N}\sim {\lambda_{345}}^2 f_N^2 /(M_N+M_H)^2 $, where $f_N$ is a formfactor, and $M_N$ is the mass of the nucleon. In Fig. \[lux\] a comparison of our results coming from the limit ${R_{\gamma\gamma}}>0.7$ , and the constraints from direct DM search experiments (LUX and XENON100), and from constraints on the invisible Higgs branching ratio (LHC ATLAS) are presented. Note, that our upper limits, represented by the line $R_{\gamma \gamma}>$ 0.7, are competitive with the upper limits from the mentioned dedicated DM experiments.
![Our results for upper limits on $\sigma_{DM,N}$ coming from the limit ${R_{\gamma\gamma}}>0.7$ (and $f_n=0.326$) compared with upper limits from LUX and XENON100 experiments, and from the LHC constraints (ATLAS) on the Higgs invisible Br. \[lux\]](rysunki/dm-rg07cm){width=".5\textwidth"}
Similarly, the constraints coming from the DM relic density measurements (red bands in Fig. \[fig:dm\]) can be translated to constraints on the DM-nucleon scattering cross section. In Fig. \[lux2\] the allowed regions of $\sigma_{DM,N}$ (red bands) as a function of $M_H$ are shown. They are coming from $\lambda_{345}$ regions allowed by the Planck data and ${R_{\gamma\gamma}}>0.7$ ($f_N=0.326$). Comparison with upper limits from LUX is shown. We see, that the direct detection limits (LUX) stay in agreement with these constraints [@Stal], however loop corrections can bring the model close to the future experiments reach [@Klasen].
![Allowed regions of $\sigma_{DM,N}$ coming from the Planck limit on the DM (red bands), compared with upper limits from LUX (and $f_N=0.326$). For lower mass part of allowed regions ($M_H < 53$ GeV) is excluded by ${R_{\gamma\gamma}}>0.7$ condition. \[lux2\]](rysunki/mm-lux1b "fig:"){width=".45\textwidth" height="0.195\textheight"} ![Allowed regions of $\sigma_{DM,N}$ coming from the Planck limit on the DM (red bands), compared with upper limits from LUX (and $f_N=0.326$). For lower mass part of allowed regions ($M_H < 53$ GeV) is excluded by ${R_{\gamma\gamma}}>0.7$ condition. \[lux2\]](rysunki/mm-lux2c "fig:"){width=".45\textwidth" height="0.2\textheight"}
Conclusions
===========
The discovery of the Higgs boson was awaited for a long time since it was the last component needed to complete the Standard Model. Moreover, it also opens door to exploration of new phenomena. The search for new particles at the LHC gives exciting perspectives, but we can also use available data, e.g., the measurements of the Higgs boson properties, especially the $\gamma\gamma$ signal strength, to shed light on such issues as the properties of the DM. Other dedicated analysis of the DM in the IDM are ongoing, e.g. on the lepton pair production at the LHC in the processes $q\bar{q}\to HA$ followed by $A\to HZ$ or $H \to l \bar{l}$ [@s-il]. Finally, models like IDM can shed some light also on the problem of the thermal evolution of the Universe [@ginz].
Acknowledgments {#acknowledgments .unnumbered}
===============
MK would like to thank organizers of this interesting symposium for invitation, warm hospitality and exceptional atmosphere. We thank P. Swaczyna for his contribution to this work. The work of MK and DS was partially supported by the grant NCN OPUS 2012/05/B/ST2/03306 (2012-2016). The work of BŚ was supported by the Polish National Science Centre grant PRELUDIUM, under the decision number DEC-2013/11/N/ST2/04214.
F. Englert, R. Brout, [*Phys. Rev. Lett.*]{} [**13**]{} 321 (1964). P. W. Higgs, [*Phys. Lett.*]{} [**12**]{} 132 (1964); [*Phys. Rev. Lett.*]{} [**13**]{} 508 (1964), [*Phys. Rev.*]{} [**145**]{} 1156 (1966). G. S. Guralnik, C. R. Hagen, T. W. B. Kibble, [*Phys. Rev. Lett.*]{} [**13**]{} 585 (1964). The CMS Collaboration, CMS-PAS-HIG-14-009 (2014). The ATLAS Collaboration, ATLAS-CONF-2014-009 (2014). Belanger, G. and Dumont, B. and Ellwanger, U. and Gunion, J.F. and Kraml, S., [*Phys.Lett.*]{} [**B723**]{} 340 (2013). The ATLAS Collaboration, [*Phys.Rev.Lett.*]{}, [**112**]{} 201802 (2014). A. Morselli, Experimental Astroparticle Physics talk at CORFU 2014. The Planck Collaboration, arXiv:1303.5076 \[astro-ph\]. N. G. Deshpande, E. Ma, [*Phys. Rev.*]{} [**D18**]{} 2574 (1978). R. Barbieri, L. J. Hall, V. S. Rychkov, [*Phys. Rev.*]{} [**D74**]{} 015007 (2006). The ATLAS Collaboration, ATLAS-CONF-2014-010 (2014). The CMS Collaboration, [*Phys. Lett.*]{} [**B 736**]{}, 64 (2014). The ATLAS Collaboration, CERN-PH-EP-2014-198 (2014). The CMS Collaboration, [*Eur. Phys. J.*]{} [**C 74**]{} 3076 (2014). B. Świeżewska, M. Krawczyk, [*Phys. Rev.* ]{} [**D 88**]{} 035019 (2013). P. Posch, [*Phys. Lett.*]{} [**B696**]{} 447 (2011). A. Arhrib, R. Benbrik, N. Gaur, [*Phys. Rev.*]{} [**D 85**]{} 095021 (2012). M. Krawczyk, D. Soko[ł]{}owska, P. Swaczyna, B. Świeżewska, [*JHEP*]{} [**1309**]{} 055 (2013). E. M. Dolle, S. Su, [*Phys. Rev.*]{} [**D 80**]{} 055012 (2009), L. Lopez Honorez, E. Nezri, F. J. Oliver, M. Tytgat, [*JCAP*]{} [**028**]{} 0702 (2007). D. Soko[ł]{}owska, arXiv:1107.1991 \[hep-ph\]. A. Goudelis, B. Herrmann, O. St[å]{}l, [*JHEP*]{} [**1309**]{} 106 (2013). M. Klasen, C. E. Yaguna, J. D. Ruiz-Alvarez, [*Phys. Rev.*]{} [**D 87**]{} 075025 (2013). P. Swaczyna, Msc Thesis, University of Warsaw 2013. I. Ginzburg at al., [*Phys.Rev.*]{} [**D82**]{} (2010) 123533.
[^1]: Presented by M. Krawczyk at the Applied Nuclear Physics and Innovative Technologies 2014 Symposium.
[^2]: Such as perturbative unitarity, stability of the Inert vacuum as well as the LEP limits and the EW precision data ($S,T$ parameters).
[^3]: On this illustrative plot the WMAP limits are presented but later on we will use the more accurate Planck results.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We describe a $q$-deformed dynamical system corresponding to the quantum free particle moving along the circle. The algebra of observables is constructed and discussed. We construct and classify irreducible representations of the system.'
author:
- |
Tomasz Brzeziński [^1]\
University of Cambridge\
Department of Applied Mathematics\
and Theoretical Physics\
CB3 9EW Cambridge, U.K.
- |
Jakub Rembieliński & Kordian A. Smoliński\
University of Łódź\
Department of Theoretical Physics\
Pomorska 149/153, 90–236 Łódź, Poland
date: August 1992
title: ' Quantum particle on a quantum circle [^2]'
---
[H]{}
Introduction
============
Non-commutative geometry [@connes2; @manin2] has attracted much attention of theoretical physicists. It is based on the idea, that the commutative algebra of functions on a manifold can be replaced by an abstract non-commutative algebra. This is in the similarity to the usual quantisation procedure, when one considers a non-commutative algebra of operators in place of the commutative algebra of real-valued functions as a set of observables. In particular usual quantum mechanics can be understood as a non-commutative symplectic geometry [@dimakis1]. Another point of view is presented in [@rembielinski1] where quantum dynamics is treated as a non-commutative diferential calculus (quantum deRham complex). The ideas of non-commutative geometry lead to the concept of non-commutative or $q$-deformed physics recently realised as a number of simple quantum mechanical models with the $q$-deformed phase-space structure [@alekseev1; @arefeva1; @rembielinski1; @rembielinski2; @wess2].
In this paper we construct a simple toy model of the non-commutative quantum mechanics, i.e. we describe the motion of the quantum particle on a quantum circle. Our goal is to describe the unitary time evolution in the non-commutative phase-space. The similar problem has been stated in [@rembielinski1], where the motion of the particle on quantum line has been considered. There are two possible schemes of the construction of quantum mechanical non-commutative models described by the so called Faddeev’s rectangle. One can begin with the classical system, quantise it and then deform or one can first perform deformation and then quantisation. Both procedures do not necessary lead to the same quantum mechanical system. In our case however, the Faddeev’s rectangle appears to be commutative.
In the quantum theory one constructs the algebra of observables $\H(I,x,p)$ which in our case is generated by the hermitean angular momentum $p$ and unitary position operator $x$, interpreted as $x = e^{-i\phi}$ where $\phi$ is the angle. They obey the Heisenberg commutation relation $[x,p]=\hbar x$. Having defined the algebra of observables one can construct the notions of states, measurement, mean value etc. which are related to the irreducible Hilbert space representations of the algebra of observables. The dynamics of the system is given by the unitary time evolution which is provided by the Heisenberg equations of motion $\dot{\mit\Omega} = {i \over
\hbar}[H,{\mit\Omega}] + \partial_t{\mit\Omega}$. The possible convenient description of the dynamics is given by the suitable deRham complex. The demanding of the unitary time evolution forces the Heisenberg equations of motion to be unchanged on the $q$-deformed level too. This implies that the $q$-deformation leaves the probabilistic interpretation of the system unchanged. The only part in that scheme which can be deformed is the algebra of observables. This is in remarkable contrast to [@arefeva1] where the Heisenberg equations of motion were deformed. In our case we can demand that the phase-space is given by the quantum cylinder rather than the classical one.
The paper is organized as follows. In Section 2 we describe the usual quantum particle on a circle from the algebraic point of view. In Section 3 we deform the algebra of observables of the particle on circle and we solve the Heisenberg equations of motion showing that the unitary time evolution is possible in the case of a free particle. Section 4 is devoted to the construction and classification of the irreducible representations of the deformed algebra of observables. Finally in Section 5 we discuss the classical limits and the invariance of constructed algebra under space and time inversions.
Quantum mechanics of the particle on a circle
=============================================
In the standard approach, quantum free particle on a circle is described by the unitary operator $x$, corresponding to the position of the particle and an hermitean operator $p$—canonical momentum (angular momentum). The dynamics of the system is given by the hermitean Hamiltonian $H = {p^2
\over{2B}}$, where $B$ denotes the moment of inertia of the particle. The algebra of observables ${\cal H} (I,x,p)$ can be defined as $${\cal H} (I,x,p) = {\C}[I,x,p] / J(I,x,p).$$ Here ${\C}[I,x,p]$ is an associative, involutive (i.e. equipped with $^*$ str ucture, which is represented as the hermitean conjugation on Hilbert space) free algebra over $\bf
C$ generated by $I,x,p$ ($I$ is the identity) and $J(I,x,p)$ is a two-sided ideal in ${\C}[I,x,p]$ generated by the relation: $$xp = px +\hbar x.$$ Notice that the parameters of the theory (e.g. $B$ or $\hbar I$) can be treated as independent of time operators belonging to the center of the algebra of observables. Namely we can extend the algebra $\cal H$ to the algebra ${\cal H}'$ defined by $${\cal H}' = {\C}[I,x,p,K,{\mit\Lambda}]/J(I,x,p,K,{\mit\Lambda})$$ where the new generators ${\mit\Lambda}, K$ are hermitean and $J(I,x,p,K,{\mit\Lambda})$ is a two-sided ideal defined by the relations $$\begin{aligned}
xp & = & px + \hbar{\mit\Lambda} x \nonumber \\
x{\mit\Lambda} & = & {\mit\Lambda} x \nonumber \\
p{\mit\Lambda} & = & {\mit\Lambda} p \nonumber \\
xK & = & Kx \label{def.j}\\
pK & = & Kp \nonumber \\
{\mit\Lambda} K & = & K {\mit\Lambda}. \nonumber\end{aligned}$$ ${\mit\Lambda}$ and $K$ are assumed both invertible and ${\mit\Lambda}$ is positive definite. The Hamiltonian reads $$H = p^2K^2$$ i.e. $K^2$ is related to the moment of inertia $B$. Now the irreducibility demanded on the representation level implies that ${\mit\Lambda}$ and $K$ are multiplies of the identity. To obtain the standard quantum-mechanical limit we can choose $${\mit\Lambda} = I, \quad K = {1 \over {\sqrt{2B}}}I.$$ The Hamiltonian form of the Heisenberg equations of motion reads: $$\begin{aligned}
&&\dot{{\mit\Lambda}} = \dot{K} = 0 \nonumber \\
&&\dot{x} =
-{i \over {2B}}x(2p-\hbar) \label{qm.heisenberg}\\
&&\dot{p} = 0 \nonumber\end{aligned}$$ Eqs. (\[qm.heisenberg\]) have the well known solution $$p(t) = p_0, \quad x(t) = x_0e^{-{i \over {2B}}(2p_0 - \hbar)t}
\label{qm.heisenberg.sol}$$ where $p_0$ denotes initial angular momentum and $x_0$ denotes the initial position of a particle. Equations (\[qm.heisenberg\]) as well as (\[qm.heisenberg.sol\]) are just identical to the equations obtained from considerations of the algebra $\cal H$, therefore the algebras ${\cal H}'$ and $\cal H$ describe the same physical situation. In the next section however we will see that the algebra $\H'$ will be suitable to the non-commuative extension of the described quantum mechanical model.
$q$-deformed quantum particle on a circle
=========================================
Quantum cylinder is defined as a free involutive algebra generated by the identity $I$, unitary $x$ and hermitean $p$ modulo the relations $$xp = qpx, \quad x^* = x^{-1}, \quad p^* = p,
\label{qcyll}$$ where $q$ is a positive real number. This space can be considered as a phase-space of the $q$-deformed particle on a circle. To quantise this system we have to replace the first of equations (\[qcyll\]) by the equation $$xp =qpx + \hbar {\mit\Lambda} x.$$ It means that we have to deform consistently the extended algebra of observables $\H'$ to the algebra $\H_{q\varepsilon\xi}$ i.e. we have to deform the ideal $J(I,x,p,K,{\mit\Lambda})$ to the ideal $J_{q\varepsilon\xi}(I,x,p,K,{\mit\Lambda})$ in such a way that both $K$ and ${\mit\Lambda}$ are no longer commutative in the algebra $\H_{q\varepsilon\xi}
=
{\C}[I,x,p,K,{\mit\Lambda}]/J_{q\varepsilon\xi}$. This can be done by the replacement of equations (\[def.j\]) by the set of the following relations $$\begin{aligned}
xp & = & qpx + \hbar {\mit\Lambda} x \nonumber \\
x {\mit\Lambda} & = & \varepsilon {\mit\Lambda} x \nonumber \\
p {\mit\Lambda} & = & {\mit\Lambda} p \nonumber \\
xK & = & \xi Kx \label{def.qj} \\
pK & = & Kp \nonumber \\
K{\mit\Lambda} & = & {\mit\Lambda} K, \nonumber\end{aligned}$$ where all parameters $q, \varepsilon , \xi$ are real and positive. We use the same $^*$ structure as in the commutative case, i.e. $p,K,{\mit\Lambda}$ are hermitean and $x$ is unitary. Now we can consider the unitary time evolution of the system with the Hamiltonian: $$H = p^2K^2 + V(K, {\mit\Lambda})
\label{q.hamiltonian}$$ where $V$ is an arbitrary element of ${\cal H}_{q\varepsilon\xi}$ constructed only from $K$ and ${\mit\Lambda}$. It leads to the Hamilton equations of the form $$\begin{aligned}
\dot{{\mit\Lambda}}&=&\dot{K}=0
\nonumber \\
\dot{x} & = & ix[-2\varepsilon ^{-1} {\mit\Lambda} K^2p + \hbar
\varepsilon ^{-2} ({\mit\Lambda} K)^2 + \hbar^{-1}
(V(qK,\varepsilon^{-1}{\mit\Lambda} ) - V(K,{\mit\Lambda}))]
\label{qheisenberg}\\
\dot{p} &= &0. \nonumber\end{aligned}$$ Here we have used the natural condition, that $\dot{x}$ is linear in the moment um $p$. This condition is satisfied if $$\xi = q^{-1}. \label{ksi}$$ This reduces algebra $\H_{q\varepsilon\xi}$ to the algebra $\H_{q\varepsilon}$ given by the following relations: $$\begin{aligned}
xp &=& qpx + \hbar {\mit\Lambda} x \nonumber \\
x{\mit\Lambda} &=& \varepsilon {\mit\Lambda} x \nonumber \\
p{\mit\Lambda} &=& {\mit\Lambda} p \nonumber \\
xK &=& q^{-1} Kx \label{def.qj1}\\
pK &=& Kp \nonumber \\
K{\mit\Lambda} &=& {\mit\Lambda} K. \nonumber\end{aligned}$$ Note that the function $V(K,{\mit\Lambda})$ which plays a role of a scale of energy in the non-deformed case now comes into equations of motion, giving a correction to the velocity of a particle. In principle this correction can be quite large since it is proportional to the inverse of the Planck constant. If we demand the existence of classical limit ($\hbar =0$) we have to estimate $q = 1 + c_1\hbar + O(\hbar^2)$, $\varepsilon =
1 + c_2\hbar +O(\hbar^2)$. One can interpret this term as an internal or not related to the motion contribution to the angular velocity. We will see in Section 5 that the dependence of $\dot{x}$ on $V(K,{\mit\Lambda})$ vanishes, when we assume that the Hamiltonian is invariant under the time inversion. Equations (\[qheisenberg\]) have the solution $$p(t)=p_0, \quad x(t)=x_0e^{i[-2\varepsilon ^{-1} {\mit\Lambda} K^2p_0 +
\hbar
\varepsilon ^{-2} ({\mit\Lambda} K)^2 + \hbar^{-1}
(V(qK,\varepsilon^{-1}{\mit\Lambda} ) - V(K,{\mit\Lambda}))]t}$$ from what we see immediately, that the time evolution of the system is given by the unitary operator $U=\exp
({i\over\hbar}Ht)$.
Representations of $\H_{q\varepsilon}$ in the Hilbert space
===========================================================
In this section we will find the Hilbert space of the representations of the $q$-deformed algebra of observables $\H_{q\varepsilon}$. The simplest way to do this is to define the action of the operators $x,p, {\mit\Lambda}$ and $K$ on the orthonormal set of vectors. The operation $^*$ is represented by the hermitean conjugation. Since $K,p$, and ${\mit\Lambda}$ are hermitean and commute, they can be diagonalised simultanously. Let us denote the eigenvectors of $p,K$ and ${\mit\Lambda}$ by $| k,\kappa,
\lambda\rangle$ where $k,\kappa,\lambda$ belong to the spectrum of $p,K,{\mit\Lambda}$ respectively, i.e.$$\begin{aligned}
p | k,\kappa,\lambda \rangle & = & k | k,\kappa,\lambda \rangle \nonumber \\
K | k,\kappa,\lambda \rangle & = & \kappa | k,\kappa,\lambda \rangle \\
{\mit\Lambda} | k,\kappa,\lambda \rangle & = & \lambda | k,\kappa,\lambda
\rangle.
\nonumber\end{aligned}$$ By means of equations (\[def.qj1\]) we see that $$\begin{aligned}
x^n | k,\kappa,\lambda\rangle & = & | q^{-n}(k-n\hbar\varepsilon^{-1}\lambda),
q^n \kappa,\varepsilon^{-n}\lambda \rangle \label{rep.x}\\
& = & | k',\kappa ',\lambda ' \rangle \nonumber\end{aligned}$$ are again eigenvectors of $p,K$ and ${\mit\Lambda}$. Therefore the Hilbert space of representations of the described system is spanned by a lattice of vectors defined by (\[rep.x\]). Let us now classify representations of $\H_{q\varepsilon}$. We normalize $K$ in such a way that it becomes $I$ in a non-deformed case. Assume first that $q \geq 1$ and $\varepsilon \geq 1$. Then each set of numbers $\kappa_0, \lambda_0, k_0$ such that $$q^{-\h } \leq
\kappa_0 < q^\h , \quad \varepsilon^{-\h } \leq \lambda_0 < \varepsilon^\h ,
\quad 0 \leq k_0 < \hbar \varepsilon^{-1}\lambda_0$$ defines the irreducible representation of $\H_{q\varepsilon}$. Similarly if we assume that $q <1$ and $\varepsilon \geq 1$, we obtain that the irreducible representations of $\H_{q\varepsilon}$ can be labelled by the numbers $k_0,\kappa_0 ,\lambda_0$ such that $$q^\h \leq \kappa_0 < q^{-\h} , \quad \varepsilon^{-\h } \leq \lambda_0
< \varepsilon^\h ,
\quad - \hbar \varepsilon^{-1}\lambda_0 < k_0 \leq 0$$ The same classification can be repeated for $\varepsilon <1$ but now $\varepsilon^\h \leq \lambda_0 <\varepsilon^{-\h}$.
Immediately from the equation (\[rep.x\]) we see that the angular momentum $p$ is quantised. Moreover the classification of irreducible representations suggests that in general free particle on a $q$-deformed circle has anyonic rather than usual Bose-Fermi statistics.
Let us now put $V(K,{\mit\Lambda}) = 0$. Using equation (\[rep.x\]) we can obtain the energy spectrum of a free particle moving along a quantum circle $$E_n = (k_0-n\hbar \varepsilon^{-1}\lambda_0)^2\kappa_0^2.$$
Symmetries
==========
It is an easy exercise to check that the algebra $\H_{q\varepsilon}$ is invariant under the transformation $x\to -x$, $p
\to p$, $K\to K$, ${\mit\Lambda}\to {\mit\Lambda}$. This transformation corresponds to the space inversion in the non-deformed case. The representation (\[rep.x\]) is obviously invariant under this transformation. In the non-deformed case however the algebra (2) is also invariant under the time inversion $x
\to x^{-1}$, $p\to -p$ which is no longer a symmetry of $\H_{q\varepsilon}$. We can try to find the symmetry of $\H_{q\varepsilon}$ which generalizes the time-inversion. To do this let us consider the general, anti-unitary transformation $\T$, $\T^2=I$ defined as $$\begin{aligned}
\T x\T^{-1} = x^{-1}, \quad \T p\T^{-1} = -pf(K,{\mit\Lambda}),\\
\T K\T^{-1} = g(K,{\mit\Lambda}), \quad \T {\mit\Lambda} \T^{-1}=
h(K,{\mit\Lambda})\nonumber\end{aligned}$$ where $f,g,h$ are arbitrary elements of $\H_{q\varepsilon}$ depending only on $K,{\mit\Lambda}$. The algebra $\H_{q\varepsilon}$ is invariant under this transformation if and only if $\varepsilon = q$ and $\T$ has the form $$\begin{aligned}
\T x\T^{-1} & = & x^{-1} \nonumber \\
\T p\T^{-1} & = & -pf(K,{\mit\Lambda}) \nonumber \\
\T K\T^{-1} & = & f^{-1}(K,{\mit\Lambda}) K \\
\T {\mit\Lambda} \T^{-1} & = & f(K,{\mit\Lambda}) {\mit\Lambda} \nonumber\end{aligned}$$ where function $f$ fulfils the condition $$q^2f(q^{-1}K, q{\mit\Lambda}) = f(K,{\mit\Lambda}).
\label{eqn.f}$$ The algebra $\H_{q\varepsilon}$ reduces now to $\H_q$ with only one deformation parameter $q$ and the commutation relations $$\begin{aligned}
xp &=& qpx + \hbar {\mit\Lambda} x \nonumber \\ x{\mit\Lambda} &=& q
{\mit\Lambda}
x \nonumber \\ p{\mit\Lambda} &=& {\mit\Lambda} p \nonumber \\ xK &=& q^{-1}
Kx \label{def.qj2}\\ pK &=& Kp \nonumber \\ K{\mit\Lambda} &=& {\mit\Lambda}
K. \nonumber\end{aligned}$$ From the quations (\[def.qj2\]) it follows that $I$ and $K{\mit\Lambda}$ generate the center of $\H_q$. Focusing on the irreducible representations of $\H_q$ we can put $$K = c{\mit\Lambda}^{-1},$$ where the constant $c$ can be derived from the classical limit, i.e. $c = (2B)^{-\h}$. Using this identification and the classical limit we can easily solve equation (\[eqn.f\]), namely $$f(K) = {\mit\Lambda}^{-2},$$ i.e.$$\T x\T^{-1} = x^{-1} , \quad \T p \T^{-1} = -p{\mit\Lambda}^{-2},
\quad \T{\mit\Lambda}\T^{-1} = {\mit\Lambda}^{-1}.$$ In this case the action of the position operator $x$ on the basis $ | k,\lambda \rangle$ takes the form $$x^n | k, \lambda \rangle = | q^{-n} (k - n\hbar q^{-1}\lambda ),
q^{-n}\lambda \rangle.$$ An irreducible representation of $\H_q$ contains vectors numbered by the knots of the lattice generated by $x^n$, $$\{ q^{-n}(k- n\hbar q^{-1}\lambda), q^{-n} \lambda \}$$ with integer $n$. Now $$\T | k,\lambda\rangle = | -\lambda^{-2}k,\lambda^{-1} \rangle$$ thus the lattice should contain point $(q^{-n}(k-n\hbar
q^{-1}\lambda),q^{-n}k)$ together with the point $(-q^{-n}\lambda^{-2}(k-n\hbar q^{-1}),q^n\lambda^{-1})$. It means that we have two kinds of irreducible representation. One is given by the choice $\lambda_0 =1$, $k_0=0$ while the second one can be generated from $\lambda_0= q^\h$, $k_0 =\h q^\h \hbar$. They correspond to Bose and Fermi statistics respectively like in the non-deformed case.
If we now consider the motion given by the Hamiltonian (\[q.hamiltonian\]) and demand that $\T H \T^{-1} = H$ we will see that $V({\mit\Lambda} ) = {\rm const}$. $I$ and the equations of motion will take the form $$\begin{aligned}
\dot{{\mit\Lambda}} & = & 0 \nonumber \\
\dot{x} & = & ix [-2q^{-1}{\mit\Lambda} K^2p + \hbar q^{-2}({\mit\Lambda} K)^2]
\\
\dot{p} & = & 0. \nonumber\end{aligned}$$ i.e. the additional term $V$ does not contribute to the angular velocity of the particle.
If we now introduce new variables: $$P = qp{\mit\Lambda}^{-1}, \quad X = x
\label{trans.class}$$ then we obtain the canonical commutation relation $[X,P] =
\hbar X$ of the standard quantum mechanics on a circle. This is in a contrast to the case $\varepsilon \neq q$, where such a reparametrization is not possible. Note however that even in the case $q = \varepsilon$ the deformed theory can not be treated just as a standard quantum mechanics with non-commutative moment of inertia, since transformation (\[trans.class\]) is not unitary.
Conclusions
===========
We described the $q$-deformation of the quantum mechanics of a particle on a circle. To describe the unitary time evolution of the system we had to deform the algebra of observables leaving the Heisenberg equations of motion unchanged. We were able to reduce number of free parameters requiring the existence of classical limit and finally the symmetry of irreducible representations on the time-inversion. In this last case we found a non-unitary transformation of variables which allowed us to replace deformed cannonical commutation relations by the standard ones. This possibility was the straightforward consequence of the additional symmetry (time-inversion) in a contrast to the situation described in [@rembielinski1].
Acknowledgement {#acknowledgement .unnumbered}
===============
T. Brzeziński would like to thank St.John’s College, Cambridge for a Benefactors’ Studentship.
[10]{}
A.Yu. Alekseev, L.D. Faddeev: [*Commun. Math. Phys.*]{} [**141**]{}:413, (1991); L.D. Faddeev: Quantum symmetry in conformal field theory by Hamiltonian methods, [*Helsinki University preprint*]{}, (1991).
I.Ya. Aref’eva, I.V. Volovich: Quantum group particles and non-Archimedean geometry, [*Phys. Lett.*]{}, [**B268**]{}:179, (1991).
A. Connes: [*Géometrie non commutative*]{}, Inter Editions, (1990).
A. Dimakis, F. Müller-Hoissen: Quantum mechanics as a non-commutative symplectic geometry, [*Göttingen Univeristy preprint*]{}, (1991).
Yu.I. Manin: [*Quantum groups and non-commutative geometry*]{}, Montreal Notes, (1989).
J. Rembieliński: Non-commutative dynamics, [*Łódź Univeristy preprint*]{}, (1992).
J. Rembieliński: Non-commutative relativistic kinematics, [*Phys. Lett.*]{} [**B287**]{}:145, (1992).
J. Wess: A talk given at XIX ICGTMP Salamanca (Spain), 1992 and J. Schwenk, J. Wess: A $q$-deformed quantum mechanical toy model, [*Phys. Lett.*]{} [**B291**]{}:273, (1992).
[^1]: On leave from University of [Ł]{}ódź, Institute of Mathematics, ul. Banacha 22, 90–238 [Ł]{}ódź, Poland
[^2]: The work supported by the KBN grant 2 0218 91 01
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Establishing the emergence of evolutionary behavior as a defining characteristic of ‘life’ is a major step in the Artificial life (ALife) studies. We present here an abstract formal framework for this aim based upon the notion of high-level observations made on the ALife model at hand during its simulations. An observation process is defined as a computable transformation from the underlying dynamic structure of the model universe to a tuple consisting of abstract components needed to establish the evolutionary processes in the model. Starting with defining entities and their evolutionary relationships observed during the simulations of the model, the framework prescribes a series of definitions, followed by the axioms (conditions) that must be met in order to establish the level of evolutionary behavior in the model. The examples of Cellular Automata based Langton Loops and $\lambda$ calculus based Algorithmic Chemistry are used to illustrate the framework. Generic design suggestions for the ALife research are also drawn based upon the framework design and case study analysis.'
author:
- |
Janardan Misra\
HTS Research, Bangalore, India 560076\
[Email: [email protected]]{}
bibliography:
- 'alife.bib'
title: Towards a Framework for Observing Artificial Evolutionary Systems
---
[Keywords: Artificial Life, Evolution, Observations, Formal Framework, Evolutionary Processes. ]{}
Background
==========
The phenomenon of “life" on earth is one of the most intriguing one with vast variety and complexity of forms in which it is found on multiple levels ranging from microbiological scale to higher taxa exhibiting a wide array of characteristics. Although we can explain several aspects of life around us in the light of existing theories for real-life evolution, we do not yet have a comprehensive understanding of the principles underlying the emergence of life and the conditions that led to the diversity and complexity of life on earth [@Fut98]. Experimental methods to understand biological processes are usually difficult and error prone because living systems are by nature complex in design and usually hard to manipulate. Evolution is even more difficult to study experimentally since experiments may span over several generations and are usually difficult to control.
*Artificial life* (ALife) is an elegant methodology to complement real life theories to study the principles underlying the complex phenomena of life without directly working with the real-life organisms. For example, ALife studies can complement theoretical biology by uncovering detailed dynamics of evolution where real life experiments are not possible, and by developing generalized formal models for life to determine criterions so that life in any arbitrary model can be observed.
Cellular Automata based models are one of the earliest attempts of synthesis to understand the underlying logic of self reproduction [@Sipper98]. Later attempts in the field have considered several new kinds of synthetic structures including programs, $\lambda$ terms, strings, graphs, automata’s etc (see for overview [@ac:Dittrich01]) and have demonstrated that one or the other observable properties of real-life are shared by all of these models, though the parallel diversity and robust evolving structures which we find in real-life are yet to be designed. One of the guiding principles of ALife research behind these novel class of synthetic structures is that - “life is a property of form and organization rather than the matter used to build it" [@Langton95]. This criterion to identify life in these novel synthetic structures in turn poses further questions as to which kind of organizational structures possess life? Which properties should we be looking at in those structures? and most importantly how can we recognize life in any arbitrary model?
To partly address these questions, in this paper, *we proceed with the hypothesis that one of the possible ways life can be recognized in an arbitrary ALife model during its simulations is by observing population of entities undergoing evolution in the sprit of evolution by natural selection, which demands the presence of reproduction, heredity, variation owing to mutational changes, and finally natural selection based reproductive success. (See also [@Dawkins82]).* Though the criterion to equate life with the presence of evolutionary processes excludes other plausible properties including metabolism [@ac:BFF92], complexity [@ac:adami00], self organization [@ac:Kau93], autonomy and autopoiesis [@Zeleny81], yet captures a wide class of interesting phenomena related to population level evolution of entities [@SS00]. Such a identification of population level evolutionary phenomena in arbitrary ALife models critically depends upon the observations carried out over simulations as we discuss next.
Motivations
-----------
Observations play a fundamental role both in real life studies as well as in ALife research. In case of real-life studies the role of observations is usually limited to an experimental analysis to uncover the specific dynamics underlying the observed life forms and their properties using natural observations or controlled experiments. On the other hand, in case of ALife studies, in general there is no known method to decide beforehand the kind of entities, which might demonstrate non-trivial life-like behavior, without closely observing the simulations of the model.
The very identification of life is thus an existential problem for ALife studies and we need some sound formal framework to address this problem. In absence of a formal framework, we often encounter intuitive and informal arguments, which remain useful only to specific models and do not always have the generic perspective. We question whether these model-specific arguments are sufficient to support the presence of an extremely complex phenomenon such as evolution in ALife models. Without formal foundations to ascertain these (informally presented) claims, there is always a danger to run into conflicting arguments, which might, for example, be based upon observations of the smulations on different levels. Nehaniv and Dautenhahn [@ND98] specifically discuss that identification of time varying entities is a deep rooted problem in the context of formal definitions for self-reproduction and add that in absence of observers it is problematic to decide whether an instance of artificial self-replication be treated at all a life-like one.
In an attempt to provide a formal platform for observations in ALife studies, a high level abstraction mechanism is presented for characterizing the observations needed to establish the evolutionary behavior in ALife studies. Initial concepts in this direction appeared in [@Misra06a; @Misra07]. The central concept of the framework is the formalization of the observation process, which we believe is essential, but most often remains implicit in ALife studies. The observation process leads to abstractions on the model universe, which are consequently used for establishing the necessary elements and the level of evolution in that model. Examples of Cellular Automata based Langton Loops (Section \[chap:langton\]) and $\lambda$ calculus based Algorithmic Chemistry (Section \[chap:lambda\]) are used to demonstrate the applicability of the formalism. Importantly the framework does not build upon the low-level dynamics or the “physical laws" of the underlying universe of the particular ALife model at hand, and thus permits the study of higher-level observationally “emergent" phenomena as a basis of evolution.
Contributions
-------------
The paper brings the implicitly assumed notion of *observations to be carried out independent of the underlying structure of the model* into main focus of ALife studies. It was not clear before that observational processes can be independently studied in their own right and the work presented in this paper makes it clear by placing observations into distinct formal platform. The work can also be seen as an attempt to fulfill the need for explicitly separating the design of the ALife models from the abstractions used to describe their dynamic progression.
The approach has helped us to formalize certain aspects of life including recognition of reproductive relationships under parental mutations as well as reproductive mutations in children along with their epigenetic developments, which were believed to be difficult to formalize before [@ND98; @Nehaniv05]. The formalism captures wide range of reproductive instances including the case of multi parent reproduction (without resorting to the concept of species), and the case of reproduction without overall growth of the population (cf. [@ND98]). Finally framework design and analysis of the case studies are used to draw useful design suggestions for the ALife research so that interesting evolutionary phenomena involving life-like entities can be better synthesized and analyzed.\
The paper is organized as follows: In Section \[chap:framework\], we will formally elaborate the framework. Case studies will follow in Section \[sec:case-studies\] – Section \[chap:langton\] applies the framework on cellular automata based Langton’s Loops and Section \[chap:lambda\] on $\lambda$ calculus based Algorithmic chemistry. Section \[chap:related\] presents a discussion of related work, and is followed by concluding remarks in Section \[sec:concluding\], along with the discussion on design suggestions for the ALife researchers in Section \[sec:design-suggst\]. Limitations of the framework and pointers for further work are discussed in Sections \[sec:limit\] and \[sec:further-work\] respectively.
The Framework {#chap:framework}
=============
In the ensuing discussion, we will use “ALife model" and “model", “Observation process" and “Observer" interchangeably to add convenience in presentation. Similarly “real-life" is used in the paper to refer to organic life on earth in contrast with the “artificial-life". Also, *Observer Abstractions* will refer to specific observations and corresponding abstractions made upon the ALife model during its simulations. *Axioms* are used to specify conditions which need to be satisfied in order to infer various components of evolution. Thus for each fundamental component of evolution: self reproduction, mutation, heredity, and natural selection, framework specifies certain Axioms constraining what is needed to be observed and consequently inferred in a formal way if any claim towards presence of any of these evolutionary components has to be substantiated. The aim is to define these formal Axioms such that only valid claims for evolutionary processes in a model can be entertained. Auxiliary formal structures are used in the intermediate stages of analysis. E.g., distance measure for determining dissimilarity between entities for their specific characteristics (see Section \[sec:distance\]).
The Formal Structure of the Framework {#chap:formalstruct}
-------------------------------------
To illustrate the framework, we will use a simple example of a binary string based chemistry whenever required in the discussion to assist the intuition behind the formalism. The chemistry will be referred as $\mathbf{CBS}$ (Chemistry of Binary Strings). Specifics of the design and structure of the chemistry will be explained as we proceed.
### Observation Process and the Model Universe {#sec:obproc}
We define the observation process as a transformation from the underlying universe of the ALife model to a set of observed abstractions as follows:\
\
**Observation Process.** [ *$\Gamma \mapsto_{Obj} \Pi$: An observation process $Obj$ is defined as a computable transformation from the underlying model structure $\Gamma =
(\Sigma, \mathcal{T})$ to observer abstractions $\Pi = (E$, $F$, $\Upsilon$, $D$, $\delta_{mut}$, $\delta_{rep\_mut}$, $C)$ and represented as $\Gamma \mapsto_{Obj} \Pi$. $\Gamma$, and $\Pi$ are defined below.*]{}\
The condition of computability is to ensure that the framework is decidable (or feasible [@Bedau99]), that is, the observation process only involves feasible computable steps, which can also be algorithmically programmed by the designer of the model and that infeasible observations defined in terms of non verifiable claims (e.g., ‘meta - information’ based claims) can be avoided.\
\
**States**. [*$\Sigma$: set of observed states of the model in a simulation.*]{}\
The exact definition of a “state" would vary from one model to the other due to their irreducible design differences as well as the level at which observations are being made. A multiset [^1] can sometimes be used to represent state of a model by defining it as a collection of observable basic structures and their corresponding multiplicities in the model at any instance during its simulation. As an example, we can consider an observed state in the case of our example chemistry of binary strings, $\mathbf{CBS}$, as a multiset - such that some specific state could be - $$\{ (00101, 2), (10101, 1), (010, 1), (0100, 1) (10100, 1)\}$$
Further illustrative examples can be seen in the case studies appearing in Section \[sec:case-studies\].\
\
**Observed Run**. [*$\mathcal{T}$: set of observed sequences of states, ordered with respect to the temporal progression of the model. Each such sequence represents one *observed run* of the model. A sequence of states is formally represented as a mapping: $N \rightarrow \Sigma$, where $N$ is the set of non negative integers acting as a set of indexes for the states in the sequence.*]{}\
A temporally ordered state sequence is one of the basic building blocks in the framework upon which all other observed abstractions are made. Such a definition of a run of model implicitly implies that the framework is fundamentally based upon the dynamic simulations of the model and not upon static analytical inferences. This is in accord with the notion of “weak emergence" [@ac:BMPRAGIKR00], which is a generic characteristic of most of the ALife studies.
For a state $S$, $S-1$, and $S+1$ would denote the the states just before and after $S$ in a state sequence.\
$\Sigma\ \textit{and}\ \mathcal{T}$ thus define the underlying dynamic structure of the model $\Gamma = (\Sigma, \mathcal{T})$. Using $\Sigma\ \textit{and}\ \mathcal{T}$, sometimes, a state machine model can also be used to define $\Gamma$.
### Entities and Their Characteristics {#sec:observe}
$E$: set of entities observed and uniquely identified by the observer within a state and across the states of the model.
The criterion to select the set of uniquely identifiable entities in a given state of the ALife model is entirely dependent on the observation process as specified by the ALife researcher. Thus for the same set of simulations of a model, there may exist very different observed states as well as entities. Nonetheless, same observation process must not yield different sets of entities in two identical states.
Defining sound criterion to identify entities often requires a careful a attention since arbitrariness in defining entities might well lead to the problem of false positives as discussed later (see Section \[false+ve\].)
“Tagging" can be sometimes used as a mechanism for the identification of individual entities whenever there exist multiple entities in the same state which are otherwise indistinguishable. Thus an observer may associate and correspondingly identify every entity in a state using a unique tag. In cases, where tags are selected such that they remain invariant under time progression of the model (i.e., do not change owing to reactions or interactions of the entities), the tags can as well be used for recognition of the persistence of these entities across the states of the model.
For example, in case of $\mathbf{CBS}$, an observer might identify individual strings as entities such that to distinguish syntactically identical strings, we can associate with every string an integer tag such that with tag $i$, an entity corresponding to the binary string $s$ can be represented as $[s]_i$. Thus a possible set of entities corresponding to the example state given above becomes $$\{ [00101]_1, [00101]_2, [10101]_3, [010]_4, [0100]_5,
[10100]_6\}$$ Alternately another observer may choose to define entities as a tuples consisting of strings with *three identical leftmost bits* - giving the set of entities for the same state as $$\{ [\underline{001}01, \underline{001}01]_1,
[\underline{101}01, \underline{101}00]_2, [\underline{010},
\underline{010}0]_3\}$$
$F \subseteq E\ \times \Sigma$ returns the state(s) in the state sequences in which a particular entity is observed. For a specific state sequence $F$ can be treated as a function.
The state information provided by $F$ for entities will be used later to define valid evolutionary relationships among them. In general observers may use different mechanisms based upon the nature of model as well as the entities defined, to determine the state for a given entity. For example, as a simple mechanism, in case of $\mathbf{CBS}$, the observer can maintain a table mapping entities to their corresponding states in order to define $F$.\
Having defined the sequence of states with temporal ordering and the entities identified by their tags, we will now proceed to discuss how an observer might define the detailed observable characteristics for such entities. Using these characteristics it can draw descendent relationship, as well as can establish presence of other components of evolution, e.g., heredity and variation. To this aim, we will define ‘character space’, as set of values for the observed characteristics. These values might be purely symbolic without any relative ordering or can be ordered using suitable ordering relation.\
The observer should define the set of all possible orthogonal and [measurable]{} characteristics for possible entities in the model as a multi dimensional character space $\Upsilon =
\mathit{Char}_1 \times \mathit{Char}_2 \times \ldots \times
\mathit{Char}_n$, where each of $\mathit{Char}_i$ is the set of values for $i^{th}$ characteristic. Each of $\mathit{Char}_i$ make one dimension in the space $\Upsilon$. Each entity $e \in E$ is thus a point in $\Upsilon$, say $e\ =\ (v_1, v_2, \ldots v_n)$, where $v_i \in \mathit{Char}_i$.
For a vector $x = (a_1, a_2, \ldots, a_r), i^{th}$ element ($a_i$) will be denoted as $x[i]$. For some of the characteristics observer might define a ‘partial ordering’ ($\leq_i$ for $Char_i
\in \Upsilon$), which can be used to compare values for those characteristics. The absence of any characteristics in an entity is represented by special zero element $0_{char_i}$ such that if $\mathit{Char}_i$ is(partially) ordered then $\forall v \in
\mathit{Char}_i$. $0_{char_i} \leq_i v$.
Notice that, observable characteristics need not to be limited to syntactic level or *structural properties* and can also include semantic properties - *observable patterns of behaviors*. Though semantic properties are much more difficult to observe and measure than the syntactic ones since they require abstracting the patterns of reactions over a range of states. In case of $\mathbf{CBS}$, for simplicity we may assume that model consists of binary strings of size $n$. In that case each position of the string can represent one orthogonal dimension and we have only two binary values ({0, 1}) at any position in a string for corresponding dimension. Thus character space $\Upsilon$ in $\mathbf{CBS}$ is $n$ dimensional binary hypercube with each string occupying a possible diagonal end point. We will represent this hypercube as $\{0,\ 1\}^n$. The ordering relation $\leq$ for all dimensions is the same and defined as $0 < 1$.
In terms of such character space $\Upsilon$, an entity set $E$ at any state can be defined by annotating the points in $\Upsilon$ with integer constants denoting the multiplicity of the entities present in $E$ with characteristics defined by the point.
### Distance Measures {#sec:distance}
Another important structure in the framework is the “dissimilarity measure" ($D$) to define the “observable differences" ($\mathbf{Diff}$) between the characteristics of the entities in a population. The distance measure defined below can be used by the observer to distribute entities into separate clusters such that entities in the same cluster are sufficiently similar while entities from different clusters are distinguishably different in their characteristics. Again exact definition of distance function is model dependent.\
An observer defines a decidable clustering distance measure $D: E \times E \rightarrow
\mathbf{Diff}$, where $\mathbf{Diff}$ is the set of values to characterize the observable “differences" between entities in $E$.
Examples include the Hamming distance to define distance between genomic strings in the Eigen’s model of molecular evolution [@ps01], set of points where two computable functions differ in their function graphs, or the set of instructions where two programs may differ. One of the known criterions to define the concept of species is “phenotype similarity" [@Ridley96], which can also be seen as another example for distance measure.
In case of $\mathbf{CBS}$, we can define an auxiliary function $\oplus: \{0, 1\} \times \{0, 1\} \rightarrow \{0, 1\}$ as a binary $XOR$ such that we have $0\oplus0 = 1\oplus1 = 0$, and $1\oplus 0 = 0\oplus1 = 1$. Thus the clustering distance measure $D: E \times E \rightarrow \{0, 1\}^n$ is defined such that $\forall i. D(e_1, e_2)[i] = e_1[i] \oplus e_2[i]$, which implies that $\mathbf{Diff}$ = $\{0, 1 \}^n$. For example in case of two $n = 3$ bit binary entities $e_1 = [001]_1$ and $e_2 = [101]_2$, $D(e_1, e_2) = 100$. Other alternatives may include Hamming distance measure $D(e_1, e_2) = \sum_{i=1}^n (e_1[i] \oplus
e_2[i])$ with $\mathbf{Diff}$ = $\{0, 1, \ldots, n\}$.
### Observable Limits on Mutational Changes {#sec:limits}
The observer needs to specify the limits under which it can recognize an entity across states even in the presence of mutational changes in the entity owing to its interactions with the environment. This is an inherent limiting property on the part of the observer and could vary among observers. Based upon the limit referred here as $\delta_{mut}$, an observer can establish whether two entities in different successive states are indeed the same with differences owning to mutations or not. The smaller the limit, the harder it will be for an observer to keep recognizing entities across states and he would be counting mutated entities as the new entities. As entities are observed in more and more refined levels of details, their apparent similarities melt away and differences become sharply noticeable.
Another type of mutations arise during reproduction, in which case an observer has to identify whether an entity is indeed an descendent of another entity even though they might not be similar. This necessitates us to introduce another bound on observable reproductive mutations as $\delta_{rep\_mut}$. This limit on observable reproductive mutations is indeed crucial while working with models where epigenetic development in the entities can be observed [@Mah97]. This is because in such chemistries including examples from real life, the “child" entity and the “parent" entities do not resemble with each other at the beginning and observer has to wait until whole epigenetic developmental process gets unfolded and then compare the entities for similarities in their characteristics. $\delta_{rep\_mut}$ assists an observer to establish whether a particular entity could be treated as a “descendent" of another entity or not.
Another reason for introducing the limit $\delta_{rep\_mut}$ is that from the view point of an high level observation process not recording every micro level details, it is quite essential to distinguish between parent entities and other secondary entities involved in the reproductive process. Consider, for example, a model where entity $A$ reproduces according to reaction $A + B
\rightarrow 2A' +C$, where $A'$ is mutant child entity of $A$, which can be determined by an observational process only when it can establish that $A$ and $A'$ are sufficiently similar with respect to their characteristics, while $A'$ and $B$ are not. These limits on observable differences are formally defined as follows:\
Based upon the choice of clustering distance measure $D$, the observer selects some suitable $\delta_{mut},\ \delta_{rep\_mut} \in \mathbf{Diff}$, which will be used later to bound mutational changes (both reproductive and otherwise) for proper recognition. $\delta_{mut}$ and $
\delta_{rep\_mut}$ are vectors such that each element specifies an observer-defined threshold on the recognizable mutational changes for corresponding characteristics.
It is important to note that the choice of $\delta_{mut},\
\delta_{rep\_mut}$ critically affects further inferences. For example, a choice of very large values would result in the lack of identification of variability in characteristics and thus make it difficult to infer natural selection (discussed later). On the other hand if an observer decides to select very small values for $\delta_{mut}$ then it cannot recognize persistence of an entity across states under changes, similarly small values for $\delta_{rep\_mut}$ make it harder to establish reproductive relationship among entities and for such an observer every new entity would seem to be appearing *de novo* in the model.
Evolutionary Components {#chap:evocomponents}
-----------------------
Having defined the observation process as a computable transformation from the underlying sequence of observed states of the model to the set of components involving entities and their observable characteristics with measurable differences as well as observable limits on such differences, we will now proceed with formalization of the fundamental evolutionary components: mutations, reproduction, heredity and natural selection.
### Mutations {#mutformalized}
For evolution to be effective entities should change (mutate) over the course of their interaction with the environment (or other entities.) Moreover, there can also be observable differences between the child and the parent entities arising out of reproductive processes. These changes in the characteristics of the entities may or may not be inheritable based upon the design of the model and the simulation instance.
Mutations can be considered of carrying two kinds of effects in the entities: one where mutations change the values for specific characteristics, secondly where after mutation an entity has at least one new character not present before or when certain characteristics are lost. We define a *Recognition relation* to establish the non reproductive mutational changes in the entities:\
The observer establishes recognition of entities across states of the model with (or without) mutations by defining the function $\mathbf{R_{\delta_{mut}}}$: $E \rightsquigarrow E$, which is a partial function and satisfies the following axioms:\
$\forall e, e' \in E\ .\ \mathbf{R_{\delta_{mut}}}$$(e) = e'
\Rightarrow F(e') = F(e) + 1$.
Informally, the axiom states that entities to be recognized as the same even with mutational changes have be observed in successive states. $\mathbf{R_{\delta_{mut}}}$ is defined anti symmetric to ensure that entities are recognized based upon the time progression of the model not in any other arbitrary order.\
$\mathbf{R_{\delta_{mut}}}$ is an injective function, that is, $\forall e, e' \in E$. $\mathbf{R_{\delta_{mut}}}$$(e) =$ $\mathbf{R_{\delta_{mut}}}$$(e')$ $\Rightarrow e = e'$
Informally, the axiom states that no two different entities in one state can be recognized as the same in the next state.\
$\forall e, e' \in E$. $\forall \mathit{Char}_i \in \Upsilon$. $\mathbf{R_{\delta_{mut}}}$$(e) = e' \Rightarrow
0_{\mathit{diff}_i} \preceq_i D(e, e')[i] \preceq_i
\delta_{mut}[i]$
Informally $\mathbf{R_{\delta_{mut}}}$$(e)$ is that $e' \in E$, which is recognized in the next state by the observer as $e$ in the previous state with possible mutations bounded by $\delta_{mut}$. In other words if entity $e$ mutates and changes in the next state and identified as $e'$, then observer might be able to recognize $e$ and $e'$ as the same if these changes (between $e$ and $e'$) are bounded by $\delta_{mut}$.
### Reproduction {#reproformalized}
Reproduction is one of the fundamental components of evolution. Through reproduction, entities pass on their characteristics to the next generation and increase the population size. Reproduction is possibly the only way by which abstract entity structures can persist across generations in case of those Alife models, where entities do not persist forever. In our framework, the way an observer establishes reproduction is by providing observed evidence for it. This is done by defining causal descendence relationships among the entities across states. The parent and the child entities are recognized by the observer as being sufficiently similar and “causally” connected across the states:\
$C \subseteq E \times E\ .\ C$ establishes the observed causality among the entities appearing in the successive states. $C$ satisfies the following axiom:\
$\forall e, e' \in E\ .\ (e, e') \in C \Rightarrow [F(e') = F(e)
+ 1] \wedge [\not\!\exists e'' \in E\ .\ F(e'') = F(e) \wedge
\mathbf{R_{\delta_{mut}}}$ $(e'') = e']$
Informally, the axiom on causal relationship $C$ states that, if an entity $e$ is causally connected to another entity $e'$, then the observer must observe $e'$ in the next state of $e$ and never before. This is to ensure that mutations are not confused by the observer with reproductions. Notice that in order to establish causal relation between entities, observers need not necessarily know the underlying reaction semantics or the micro level dynamics of the model. Only requirement is that the observer’s claimed causality conforms with the stated axiom. In essence, this formulation of causality is an abstract specification which demands observers to identify the entities which have been observed to be causal sources for the appearance of a new entity. Only then proper descendance relation for the new entity can be established.
Apart from causality $C$ we also need auxiliary relation $\Delta$ to determine that the differences due to the reproductive mutations are also bounded by $\delta_{rep\_mut}$.\
$\Delta \subseteq E \times E$ such that $\forall e, e' \in E \ .\
(e, e') \in \Delta \Leftrightarrow \forall \mathit{Char}_i \in
\Upsilon\ .$ if $\mathit{Char}_i$ has an ordering then $D(e,
e')[i] \preceq_i \delta_{rep\_mut}[i]$.
Informally for $(e, e'$) to be in $\Delta$, their differences for each single characteristic $Char_i$ must be bounded by $\delta_{rep\_mut}[i]$.\
Based on the thus established notion of “causal” relationships between entities and $\Delta$, we will define $\mathbf{AncestorOf}$ relation, which connects entities for which an observer can establish descendence relationship across generations.\
$\mathbf{AncestorOf}$ = $(\ (C \ \cup \
\mathbf{R_{\delta_{mut}}})^+ \ \cap \ \Delta)^+$
In this definition the (inner) transitive closure of $(C \ \cup \
\mathbf{R_{\delta_{mut}}})$ captures the observed causality ($C$) across multiple states even in cases when “parent" entities might undergo mutational changes ($\mathbf{R_{\delta_{mut}}}$) before “child" entities complete their “epigenetic" maturation with possible reproductive mutations. Intersection with $\Delta$ ensures that causally related parent and child entities are not too different from each other, that is, reproductive mutational changes are under observable limit. Outer transitive closure is to make $\mathbf{AncestorOf}$ relationship transitive in nature so that entities in the same lineage can be related with each other. For $e, e' \in E$, $(e, e') \in$ $\mathbf{AncestorOf}$, describes that $e$ is observed as an ancestor of $e'$.
![Graphical view of the relationships between entities in successive states. Recognition relation $\mathbf{Rec}$, Causal relation $C$, and $\mathbf{AncestorOf}$.[]{data-label="fig:hassediagram"}](1.pdf)
Figure \[fig:hassediagram\] depicts graphically the relationships between entities in successive states. Vertical lines represent the states ($S_0, S_1, S_2, S_3, S_4$). Various kinds of arrows represent different relationships: recognition relation $\mathbf{R_{\delta_{mut}}}$, causal relation $C$, and $\mathbf{AncestorOf}$. The end points of the arrows on state lines represent entities.
Case of Reflexive Autocatalysis.
In the simplest form, a reflexive autocatalytic cycle is represented as a system of reaction equations: $$\begin{aligned}
A + X_1 &=& A_1 + Y_1 \\
A_1 + X_2 &=& A_2 + Y_2 \\
\vdots \\
A_{n-1} + X_{n} &=& mA' + Y_n\end{aligned}$$ where $m$ copies of entity $A'$ are produced at the end and that entity $A'$ is a variation of entity $A$, i.e., $(A, A') \in
\Delta$. Such autocataltic cycles are supposed to be the chemical basis of biological growth and reproduction. Examples include the Calvin cycle, reductive citric acid cycle, and the formose system. Competing cycles of this sort can even undergo limited evolution, though they are supposed to have very limited heredity [@ss97].
In the current framework suppose an observer could determine the causal relations - $(A, A_1)$, $(A_1, A_2)$, $\ldots$, $(A_{n-1},
A')$. Also assume that entity $A$ does not undergo any changes before $A'$ is produced, that is, $(A, A) \in R_{mut}$. Then $(C \
\cup \ \mathbf{R_{\delta_{mut}}})^+$ would contain $(A, A')$ so also would $(\ (C \ \cup \ \mathbf{R_{\delta_{mut}}})^+ \ \cap \
\Delta)$ establishing the reproduction of $A$ through reflexive autocatalytic cycle and with variation.
Recognition of reproductive relationships under parental mutations together with reproductive mutations and epigenetic developments in the child entities.
Let us see what it requires for establishing reproductive relationship when (parent) entities might be undergoing changes across states and child entities not only differ from the parent entities owing to reproductive mutational changes but also that there exist epigenetic developments in the child entities, which make it harder for any observer to establish similarities between child and parent entities by observing the child entities only in the beginning (i.e., in the state when child entities were observed for the first time.) Naturally it would require that an observer observes child entities so long that their epigenetic development unfolds completely - since in general there cannot be any fixed limit on the number of states required for such epigenetic development, we capture this requirement of observations across states using transitive closure - $(C \ \cup \
\mathbf{R_{\delta_{mut}}})^+$, where $\mathbf{R_{\delta_{mut}}}$ ensures that (mutational) changes in the parent entities and also the changes in the child entities during epigenetic development are accounted for.
Lets us assume that in a state $S_i$, a child entity $c$ was observed for the first time and (parent) entity $p$ present in the state $S_{i-1}$ was observed to be casually connected to it. Suppose that for entity $e$ its epigenetic development unfolds through states $S_{i+1}, S_{i+2}, \ldots, S_{i+r}$ such that with changes owing to the development $c$ was observed as $c_{1},
c_{2}, \ldots, c_{r}$ in these states with $(c, c_1), (c_1, c_2),
\ldots, (c_{r-1}, c_r) \in \mathbf{R_{\delta_{mut}}}$. Similarly suppose that parent entity $p$ undergoes mutations in these successive states and observed as $p_{1}, p_{2}. \ldots, p_{r}$ such that $(p, p_1), (p_1, p_2), \ldots, (p_{r-1}, p_r) \in
\mathbf{R_{\delta_{mut}}}$. It is clear that $(C \ \cup \
\mathbf{R_{\delta_{mut}}})^+$ would contain $(p, c), (p, c_1),
\ldots, (p, c_r)$, $\ldots, (p_r, c), (p_r, c_1), \ldots, (p_r,
c_r)$ among other tuples implying that the intersection of $(C \
\cup \ \mathbf{R_{\delta_{mut}}})^+$ with $\Delta$ would result in those tuples $(p_m, c_n)$, where $p_m$ and $c_n$ are sufficiently similar in their characteristic. Therefore if the resultant set $(\ (C \ \cup \ \mathbf{R_{\delta_{mut}}})^+ \ \cap \ \Delta)^+$ is not empty, the observer can establish the reproductive relationship between entities $p$ and $c$ even under parental mutational changes and the epigenetic changes and reproductive mutations in the child entity.
Using $\mathbf{AncestorOf}$ relation, we now can consider the cases of *entity level reproduction* and *Fecundity*:
### Case 1: Entity Level Reproduction {#case-1-entity-level-reproduction .unnumbered}
We consider the case where instances of individual entities can be observed as reproducing even though there might not be any observable increase in the size of the whole population.
For a given simulation of the model, an observer defines the following $\mathbf{Parent_{\Delta}}$ relation:\
$$\begin{split}
\mathbf{Parent_{\Delta}} = \{ & (p, c) \in \mathbf{AncestorOf} \mid\\
& \not\!\exists e \in E\ .\ [(p, e) \in \mathbf{AncestorOf}
\wedge (e, c) \in \mathbf{AncestorOf}]\}
\end{split}$$
The condition in defining $\mathbf{Parent_{\Delta}}$ is used to ensure that $p$ is the immediate parent of $c$ and thus there is no intermediate ancestor $e$ between $p$ and $c$. Using $\mathbf{Parent_{\Delta}}$ relation, in order for the observer to establish reproduction in the model, the following axiom should be satisfied:\
$\exists \textit{state sequence}\ T \in \mathcal{T} \ .\
\mathbf{Parent_{\Delta}} \neq \emptyset$
This means, if there is reproduction in the model, then there should exists some simulation $T \in \mathcal{T}$ of the model, where at least one instance of reproduction is observed.
In case of $\mathbf{CBS}$, we consider a very simple model of reproduction, where at any state of the model some of the strings are randomly chosen and are copied with some random errors. How it is done remains hidden from the observer but the observer can observe which parent entities are chosen for copying and can establish causal relation between these parent and their copied child entities if the random errors occur only at even positions as the way $\delta_{rep\_mut}$ has been defined in Section \[sec:limits\]. It can be easily seen that under such construction scheme *Axiom of Reproduction* will be satisfied.
### Case 2: Population Level Reproduction - Fecundity {#case-2-population-level-reproduction---fecundity .unnumbered}
Though entity level reproduction is essential to be observed, for natural selection it is the population level collective reproductive behavior (fecundity), which is significant owing to the *carrying capacity* of the environment. Since carrying capacity is an limiting constraint on the maximum possible size of population, an observer needs to establish that there is no perpetual decline in the size of the population. In other terms for all generations, there exists a future generation that is of the same size or larger. This allows cyclic population sizes where the cycle mean grows (or stays steady) over time. Also in case of fecundity, an observer need not to observe all the parents in the same state, nor do children need to be observed in the same states of the model. Formally we require the observer to establish Fecundity by satisfying the following axiom:\
There exist infinitely many different generations of entities in temporal ordering $G_1, G_2, \ldots$ such that $(\forall G_i
\subseteq E)(\exists G_{j>i} \subseteq E)\ .\ |G_j| \geq |G_i|$ where $G_j = \{c \in E \mid\ \exists a \in G_i\ .\ (a, c) \in
\mathbf{AncestorOf}\}$, (operator $|.|$ returns the size of a set.)\
Informally, the axiom states that for every generation of entities ($G_i$), in future there exist generation of its descendent entities ($G_j$) such that the size of descendent generation must be equal or more than current generation. Note that the granularity of the time for determining generations is entirely dependent on the design of the model and the observation process.
We can now formulate another important axiom from evolutionary perspective, which asserts that reproduction in the model should not entirely cease because of the (harmful) mutations.\
\[continuation axiom\] Some mutations do preserve reproduction. Formally, $\exists e \in E\ .\ Ch_e = \{e' \in E:
(e, e') \in \mathbf{Parent_{\Delta}} \cup
\mathbf{R_{\delta_{mut}}}$ $ \} \neq \emptyset \Rightarrow \exists
e'' \in Ch_e\ .\ \{e' \in E: (e'', e') \in
\mathbf{Parent_{\Delta}} \}$ $ \neq \emptyset$
Informally, this means, there exists entity $e \in E$, which reproduces (with mutations) and one of those (mutant) children of $e$ can also further reproduce. $Ch_e$ denotes the set of children of $e$.
In case of $\mathbf{CBS}$, since copying mechanisms do not work differently based upon selected entities, hence the errors during copying process do preserve the above axiom of *Preservation of Reproduction under Mutations*.
### Heredity {#heredityformalized}
Heredity, yet another precondition for evolution, can in general be observed in two different levels: Syntactic level and Semantic level. On *syntactic level*, entity level inheritance is implied by the structural proximity between parents and their progenies ranging over several generations - though in case of continuous structural changes in the parental entities and epigenetic development in progenies, this would require an observer to establish structural similarities over a range of states as discussed earlier with the definition of $\mathbf{AncestorOf}$ relation. Also for syntactic inheritance to persist, design of the model needs to ensure that environment, which controls the reaction semantics of entities, remains approximately constant over a course of time so that structural similarities also result into continued reproductive behavior.
Difficulty arises primarily on the level of multi parental reproduction - in this situation an observer might have to stipulate some kind of gender types and might have to relax the mechanism of recognizing the parent-child relationship in a way as happens for example in case of organic life, where male-female reproductive process (often) gives birth to a progeny belonging to “only" one gender type. In such a case, for heredity, an observer need to ensure that, [over a course of time all the gender types are sufficiently produced in the population.]{}
On the other hand it is also possible to observe inheritance on the semantic level (ignoring structural differences) in terms of *semantic relatedness* between entities, whereby an observer can observe that progenies and their parental entities exhibit similarities in their (reproductive) behaviors under near identical set of environments. This in turn would require an observer to identify the possible sequences of observable reactions between existing entities, which appear to be yielding new set entities (children) and in the child generation as well there exist a similar observable reproductive process, which enables the (re)production of entities. Such an observation would enable the observer to abstract the reproductive processes currently operational in the model. The inherent difficulties in this view are obvious - in essence an observer needs to abstract the reproductive semantics from observable reactions in the model, which in turn might require non trivial inferences in absence of the knowledge of the actual design of the model.
Considering the case of real-life from an observational view point, semantic view is in fact an abstraction over all the reproductive processes existing across various species and levels including the case of bacterial organisms, where next generation of bacteria may contain a mix of genetic material from various parental bacterium of previous generation through the process of horizontal transmission. So while in case of syntactic inheritance an observer would only be able establish inheritance across organisms belonging to same species, using semantic view, he could expand his horizon to the all organic life as a whole.
However, heredity as a mechanism of preservation of syntactic structures, appears to be crucial for those ALife models where entities have very limited set of reproductive variations possible, that is, where environment supports only rare forms of entities to reproduce and any changes in the syntactic structure of these reproductive entities may result in the elimination of the reproductive capability. Real-life on earth as well as the model of the Langton loops (as discussed further in Section \[chap:langton\]) are definitive examples where most of the variations in the genetic structure, or the loops geometry/transition rules result in the loss of reproductive/replicative capabilities. Also heredity usually requires further mechanisms to reduce possible undoing of current mutations in future generations owing to new mutations. Therefore, in order to establish inheritance in ALife models, sufficiently many generations of reproducing entities need to be observed to determine that the number of parent-child pairs where certain characteristics (both syntactic and semantic) were inherited by child entities without further mutations is significantly larger than those cases where mutations altered the characteristics in the child entities. We can express it as the following axiom:\
Let a statistically large observed subsequence of a run $T$: $$\Omega = lim_{N \rightarrow \infty} \langle S_n, \ldots S_N
\rangle , n \ll N$$ Consider $\mathbf{Parent_{\Delta}^{\Omega}}$ = $\{(e, e') \in \mathbf{Parent_{\Delta}}$ $| F(e) \in \Omega \wedge
F(e') \in \Omega \}$ to be the set of all parent - child pairs observed in $\Omega$. Again let $\mathbf{Inherited_{\Omega}^i}$ = $\{(e, e') \in \mathbf{Parent_{\Delta}^{\Omega}}$ $| \exists
\mathit{Char}_i \in \Upsilon\ .\ D(e, e')[i] =
0_{\mathit{diff}_i} \}$ be the set of those cases of reproduction where $i^{th}$ characteristics were inherited without (further) mutation. Then high degree of inheritance for $i^{th}$ characteristics $\mathit{Char}_i$ implies that $|\mathbf{Parent_{\Delta}^{\Omega}}|$/$|\mathbf{Inherited_{\Omega}^i}|$ $\simeq 1$. For syntactic inheritance to be observed in a population of entities, we should have some such characteristics which satisfy this condition.
The axiom of heredity together with the axiom of preservation of reproduction under mutation ensures that reproductive variation is maintained and propagated across generations.
### Natural Selection {#natselformalized}
There are several existing notions of selection in the literature on evolutionary theory [@Fut98; @Ridley96; @Ridley97; @SS00; @Mah97; @Kimura83]. In case of our observation based framework we choose to define natural selection as a *statistical inference* of *average reproductive success*, which should be established by an observer on the population of self reproducing entities over an evolutionary time scale i.e., over statistically large number of states in a state sequence. Other notions of selection using fitness, adaptedness, or traits etc. are rather intricate in nature because these concepts are relative to the specific abstraction of “common environment” shared by entities and “the environment-entity interactions”, which are the most basic processes of selection. Nonetheless selecting appropriate generic abstraction for these from the point of view of an observation process is not so simple. Therefore we consider more straightforward approach based upon the idea that on evolutionary scale the relative reproductive success is an effective measure, which is also an indicator of better adaptedness or fitness. We thus define the following (necessary) axioms for the natural selection:\
An Observer must observe statistically significant population of different reproducing entities, say $\Lambda$ ($|\Lambda| \gg 1$), for statistically large number of states in a state sequence $T \in
\mathcal{T}$. That is, for a statistically large subsequence $\Omega$ of $T$, $\Omega = lim_{N \rightarrow \infty} \langle S_n,
\ldots S_N \rangle , n \ll N$, the observer defines the set of reproducing entities $\Lambda \subseteq \bigcup_{S_j \in
\Omega}SR(S_j)$, where $SR(S_j)$ = $\{e \in E |$ $[F(e) = S_j]
\wedge [\exists e' \in E\ .\ (e, e') \in
\mathbf{Parent_{\Delta}}]\}$ is the set of all reproducing entities in state $S_j \in \Omega$.\
Entities in $\Lambda$ should be different with respect to characteristics in $\Upsilon$ and there should exist differential rate of reproduction among these reproducing entities. Rate of reproduction for an entity is the number of child entities it reproduces before undergoing any mutations beyond observable limit.\
In other words, $Rate_{rep}: E \rightarrow N^+$ defined as $\forall e \in E\ .\ Rate_{rep}(e) = |Child_e|$ where $Child_e =
\{e' \in E | \exists e'' \in E\ .\ (e'', e') \in
\mathbf{Parent_{\Delta}}$ and $[\mathbf{R_{\delta_{mut}}^+}$$(e) =
e'' \wedge \forall \mathit{Char}_i \in \Upsilon\ .\ D(e, e'') =
0_{\mathit{diff}_i}]\}$.
The above two axioms though necessary are not sufficient to establish natural selection since these cannot be use as such to distinguish between natural selection with neutral selection [@SS00]. The following axioms are therefore needed to sufficiently establish natural selection.\
There must be variation in heritable mutations in population of $\Lambda$. Formally, let $$Child_{mut} = \{e \in \Lambda | \exists e' \in \Lambda\ .\ (e, e')
\in \mathbf{Parent_{\Delta}} \wedge [\exists \mathit{Char}_i \in
\Upsilon\ .\ 0_{\mathit{diff}_i} \prec D(e, e')[i]]\}$$ be the set of child entities carrying reproductive mutations. Let $Var\_Child_{mut} \subseteq Child_{mut}$ be the set of those child entities which carry different mutations with respect to characteristics in $\Upsilon$, that is, $$\forall e, e' \in
Var\_Child_{mut} \textit{ we have } \exists \mathit{Char}_i \in
\Upsilon \ .\ 0_{\mathit{diff}_i} \prec D(e, e')[i]$$ Then axiom of heritable mutation demands that $|Var\_Child_{mut}|$ $\gg 1$, that is, there are significantly many child entities carrying different mutations.\
There must be non zero correlation between heritable variation and differential rate of reproduction. Formally, $$\begin{split}
\forall \mathit{Char}_i \in \Upsilon\ .\ & \forall e, e' \in
Var\_Child_{mut}\ . \ \textit{the following two conditions should hold:}\\
& \begin{split}
i)\ e[i] <_i e'[i] \Leftrightarrow & [Rate_{rep}(e) < Rate_{rep}(e')]\ \vee\ [Rate_{rep}(e) > Rate_{rep}(e')]
\end{split} \\
& ii)\ e[i] =_i e'[i] \Leftrightarrow Rate_{rep}(e) = Rate_{rep}(e')
\end{split}$$
Informally, this means as the value of characteristics inherited by the child entity changes, rate of reproduction also changes. Based upon the environmental pressures with respect to a particular characteristics, rate of reproduction might either increase or decrease as the characteristic changes.
The last two axioms state that there must be significant variation in population (in characters) of entities which must be maintained for evolutionarily significant periods and that this variation must be caused by the differences in inheriting mutations from the parent entities, which in turn directly affect the rate of reproduction. Having formalized the fundamental component of evolutionary processes to be observed in a model, we will illustrate the framework on two important ALife models in the following Section. These illustrations will later be used in concluding Section \[sec:concluding\] to extract generic design principles for ALife research.
Case Studies {#sec:case-studies}
============
General Considerations
----------------------
Having described the generic formal framework in Section \[chap:framework\], which formalizes the concept of observations and consequent axiomatic inferences to establish the level of evolution for ALife studies, in the following sections, we will apply the formalism to different models as case studies. These case studies include Cellular Automata based Langton Loops [@Langton87] and $\lambda$ Calculus based Algorithmic Chemistry [@ac:Fon92]. The case studies elaborate the steps and technical details specific to the example universe of the model, which remained implicitly defined in the generalized description of the framework.
For a given model, the steps to instantiate the framework can be described as follows: The observation process works on the simulations of the model which iteratively change the underlying states based upon the application of the updation rules of the model. The observation process starts with the identification of states of the model ($\Sigma$) during its simulations (i.e., state sequences $\mathcal{T})$). Usually any change in the model (i.e. the changes in the set of basic units) may give rise to a change of the observed state. It is important to note that in some cases there might be any changes in the observable state of the model even tough there is ongoing underlying activity in the model, that is, when model reaches, for example, a fix point.
For every state in the state sequence, the observation process (or the observer) needs to identify a set of well defined entities with suitable tagging for individual identification ($E$). These entities need to be described in terms of their characteristics ($\Upsilon$). Next important task is to define the limits on the observable mutational changes in individual characteristics of the entities ($\delta_{mut}$, $\delta_{rep\_mut}$), which will in turn define the recognition relation ($\mathbf{R_{\delta_{mut}}}$) to relate entities persisting across states of the model as well to determine whether two entities might be considered related under descendent relationship.
Once the sets of entities in various successive states of the model as well as their characteristics are known, important evolutionary relationships need to be established between them. These evolutionary relationship depend upon the intermediate causal relation ($C$) between the entities as observed under the mechanics of observation process. Using the limits on mutational changes as well as causal relationship between entities, we proceed to define the Ancestor ($\mathbf{AncestorOf}$) and the Parent sets ($\mathbf{Parent_{\Delta}}$). These sets determine whether there are entities which might be potentially reproducing in the model, even with observable changes between parent and child entities ($\Delta$).
Next stage of the observation process is to ascertain the level of effectiveness of evolution in the model. Using the long term observations on the model for statistically large number of generations, one can infer some statistical patterns for degree of heredity and variation. For natural selection to be effective, there should exist large number of reproducing entities with significant variation in their characteristics such that there exists correlation of this variation in the characteristics with the reproductive success of the entities.
This process at the end establishes the validity of all or some axioms of the framework for the given model which provides clues to the degree upto which evolutionary processes might be effective in that model universe. The case studies in the following sections will illustrate this process in detail.
In these case studies, constructs not explicitly defined are assumed to be same as what is defined in the framework.
Case Study **[1]{}: Langton Loops** {#chap:langton}
-----------------------------------
Research on the self reproduction has a long cherished history starting in early fifties [@Burks70; @Sipper98; @Freitas04]. After the pioneering work of Alan Turing in early 40s to define the mechanical meaning of ‘computation’ as a Turing machine transitions, John von Neumann defined Cellular Automata (CA) [@Neumann66] to explain the generic logic of self reproduction in mechanical terms. His synchronous cellular automata model was a two dimensional grid divided into cells, where each cell would change in parallel its state based upon the states of its neighborhood cells, its own state and its transition rule. For such CA model, von Neumann defined a virtual configuration space where he demonstrated analytically that there exists some universal replicator configuration which could replicate other configurations as well as itself. Though universal replicators are not found in nature and such self replicator was extremely large in its size, nonetheless the underlying logic of treating states of cells in the grid both as ‘data’ as well as ‘instruction’ was very fundamental contribution of this model and that was exactly was was discovered later in case of real life where DNA sequences specify both transcription as well as translation for their own replication in a cell. Another strength of von Neumann’s formulation was its ability to give rise to unlimited variety of self replicators [@McMullin00a; @McMullin00b]. Over the years this model was simplified and reduced in size considerably [@Codd68 81-105].
Finally Langton introduced loop like self replicating structures in [@ac:Lan84], which retained the ‘transcription - translation’ property of von Neumann’s model excluding the capability of universal replication and symbolic computation. Langton’s original self-replicating structure is a 86-cell loop constructed in two-dimensional, 8-state, 5-neighborhood cellular space consisting of a string of core cells in state $1$, surrounded by sheath cells in state $2$. These loops have since then, been extended into several interesting directions including evolving Evoloops in [@Samaya98b].
These cellular automata based ALife models offer the ideal example for our observer (observation process) based framework since these replicating loops and their variations evolve only with respect to some high level observation process, which can be used to define entities (loops) and their evolution. We will illustrate the formal framework by instantiating it on the Cellular automata based Langton loop model. Further details on the model itself can be found in the above references.
### Instantiating the Framework {#sec:langton_inst .unnumbered}
We consider the case of two dimensional CA lattice based model.An observation is defined on the CA model by assuming an underlying coordinate system such that each cell in a two dimensional cellular automata (CA) lattice can be associated with unique coordinates (represented as $(x, y)$.) A cell is then completely represented as ${{\langle}}(x, y), s{{\rangle}}$, where $s \in [0..7]$ is the state of the cell. When a cell is in state $0$, it is also known as a *quiescent* cell. Let us denote the set of all cells of a CA model as $Cell$, which is a potentially infinite set.
For a given cell ${{\langle}}(x, y), s{{\rangle}}\in Cell$, its coordinates can be accessed as follows: $co_x({{\langle}}(x, y), s{{\rangle}}) = x$, $co_y({{\langle}}(x, y), s{{\rangle}}) = y$, which can be extended to the set of cells: $\forall Z \subseteq Cell$, $co_x^+(Z) = \bigcup_{c \in Z}
co_x(c)$, $co_y^+(Z) = \bigcup_{c \in Z} co_y(c)$.
$Neigh: Cell \rightarrow 2^{Cell}$ gives the coordinate wise non quiescent cells in the surrounding neighborhood of a cell. Formally, $\forall (c = {{\langle}}(x, y), s{{\rangle}}) \in Cell$ we have $$Neigh(c) = \{ {{\langle}}(x \pm 1, y), s'{{\rangle}}, {{\langle}}(x, y \pm 1), s'{{\rangle}}\
|\ s' \neq 0\}$$
### The model Structure {#the-model-structure .unnumbered}
A CA-based model is usually initialized by setting some finite number of selected cells to non-quiescent states. At each step, state of every cell of the model is changed as per the state transition rules. Therefore we define for an observer *state* of the Langton’s model as the subset of $Cell$ consisting of only non quiescent cells. It is clear that for the observer change in a state is observable only if there is a change in the set of non quiescent cells. The *state* of the model for the observer will also be referred to as $configuration$. Thus $\Sigma$ denotes the set of all possible different configurations and a state sequence in $\mathcal{T}$ is a sequence of configurations observed in temporal order by the observer starting from some specific configuration. In the following discussion we will consider a fixed sequence given as $T \in \mathcal{T}$, starting with a specific initial state given in Figure \[ftr1\] (Time 0). For the fact that there exist a temporal (total) ordering of states in $T$, we can also associate an integer sequence $I = [0,
1, 2, \ldots]$ with $T$, which works as an indexing for the states. With the above structure of Langton’s CA model, the observer takes the following decisions.
### Entities {#entities .unnumbered}
Each entity in some state is characterized by two values - the connected set of non quiescent cells and the associated *pivot*. Two cells are connected only if there exists a consecutive sequence of neighboring non quiescent cells joining them in the lattice. The (function) *pivot* gives the coordinates for a cell uniquely associated with an entity in CA lattice in a particular state. Formally, the set of entities (loops) in the model is defined as follows: $$\begin{split}
E = \{ & [Z, pivot(Z)] \mid \exists\ \textit{a configuration }
S \in T\ .\ \\
& [Z \subseteq S \wedge Z \neq \emptyset] \wedge
[\forall c \in Z\ .\ \exists c' \in Neigh(c)\ .\ c' \in Z] \}
\end{split}$$ To define $pivot$, an observer may choose the coordinates of top left hand corner cell of an entity as the pivot for it. Formally $$pivot(Z) = (min\{co_x^+(Z)\}, max\{co_y^+(Z)\})\ \forall (e =
[Z, pivot(Z)]) \in\ E$$ This gives an obvious characterization for a two dimensional character space $\Upsilon = \mathit{Char}_1
\times \mathit{Char}_2$ with $\mathit{Char}_1$ being the set of all non quiescent connected set of cells and $\mathit{Char}_2$ being the set of corresponding pivots. We do not associate additional tags with entities because pivots can be used to uniquely identify them in any state of the model.
### State Function {#state-function .unnumbered}
$F: E \mapsto I$ is defined using a table which associates with each entity $e \in E$, the index $i \in I$ for the state in which $e$ is observed.
### Distance Measure {#distance-measure .unnumbered}
Distance function $D: E \times E \rightarrow \{0, 1\} \times \{0,
1\}$ is defined such that $\forall e, e' \in E\ .\ D(e, e') =
[d_g, d_p]$ where $d_g$ and $d_p$ are defined as follows: $d_g$ is $0$ only if both entities have the same number of cells arranged identically or else it returns $1$. $d_p$ is $0$ when the pivots for both the entities are same and $1$ otherwise.
### Limits on Observable Mutations {#limits-on-observable-mutations .unnumbered}
The observer next selects $\delta_{mut} = [1, 0]$, which means that observer can recognize an entity in future states even with mutations (changes in the states, number, or the arrangement of cells comprising the entity) provided that the pivot remains the same. Select $\delta_{rep\_mut}= [0,1]$ which implies that for reproduction observer strictly demands identical geometrical structure of the parent and child entities, though may have different pivots - this is essential to capture exact replication of the loops.
### Observing Reproduction and Fecundity {#observing-reproduction-and-fecundity .unnumbered}
**Recognition relation** $\mathbf{R_{\delta_{mut}}}$ $: E
\rightarrow E$ is defined as follows: $$\forall e, e' \in E, \mathbf{R_{\delta_{mut}}} (e) = e'
\Leftrightarrow\ [F(e') = F(e) + 1] \wedge [D(e, e') \leq
\delta_{mut}]
$$ Informally this means two entities in consecutive states are recognized same only if they have the same pivots. Which also means observer can recognize entity even with change in the number, state, and geometrical arrangement in the cells of an entity across states provided that entity does not shift in CA lattice altogether (which would result in the change of the pivot.)\
$\mathbf{R_{\delta_{mut}}}$ satisfies Axiom $1$, Axiom $2$, and Axiom $3$.
Axiom $1$ and Axiom $3$ are satisfied by definition. Axiom $2$, which states that $\mathbf{R_{\delta_{mut}}}$ is an injective function holds because no two entities in the same state share the same pivot. This is because pivot as defined above is connected to all other cells of the entity and all the non quiescent cells which are connected in any state are taken together as one entity. Thus two different entities in the same state always consist of cells such that cells in one entity are not connected with the cells of second entity, and hence always have different pivots.
**Causal relation** The relation $C$ between entities in consecutive states is defined as follows: $C \subseteq E \times E$ such that $\forall\ e, e' \in E$ where $e = [Z_e,
\mathit{pivot}(Z_e)]$ and $e' =[Z_{e'}, \mathit{pivot}(Z_{e'})]$ we require $$(e, e') \in C \iff \left\{ \begin{array}{ll} 1.\, & co_x^+(Z_e) \supset co_x^+(Z_{e'})\\
2.\, & co_y^+(Z_e) \supset co_y^+(Z_{e'})\\
3.\, & \mathit{pivot}(Z_e) \neq \mathit{pivot}(Z_{e'})\\
4. & F(e') = F(e) + 1
\end{array} \right.$$ Intuitively what we demand with above definition of causal relation $C$ is that child entity was part of the parent entity and at certain stage it “breaks off” from the parent entity, as can be seen in Figure \[ftr1\] at time step $127$.\
Causal relation $C$ defined above satisfies the Causality Axiom.
Condition $F(e') = F(e) + 1$ insures that $e$ and $e'$ are not observed in the same state. To establish that $e'$ is not the result of mutations in some other entity $e''$ observed in past (i.e., $[F(e'') = F(e)] \wedge [\mathbf{Rec}(e'') = e']$) we note that because of the definition of $\mathbf{Rec}$, $e''$ and $e'$ would otherwise have the same pivots, which means pivot of $e''$ will be included in the set of cells in $e$ (since $[co_x^+(Z_e)
\supset co_x^+(Z_{e'})] \wedge [co_y^+(Z_e) \supset
co_y^+(Z_{e'})]$), which is not possible because $e$ and $e''$ being different entities in the same state cannot have cells in common including pivot as argued above in the proof of previous lemma.
**Axiom of Reproduction** and the **Axiom of Fecundity** are satisfied by the entities and abstractions on Langton Loops described above.
These two axioms can be established by the observer in a specific state sequence as exemplified in Figure \[ftr1\] and Figure \[ftr2\] by repeatedly applying the recognition relation $\mathbf{Rec}$ when entities are changing in number and states of cells (retaining the pivots) and applying the causal relation when a parent entity splits (e.g. at Time=$127$). The relation $\Delta$ connects the initial parent entity and the child entity at Time=$151$.
With respect to Figure \[ftr1\], an entity is identified at Time=$0$ with associated pivot. Between time steps $[1\ldots126]$ entity changes in number and states of its cells but the pivot remains the same, hence as per the definition of $\mathbf{Rec}$, the observer can recognize the entity in these successive states. At Time=$127$, the (parent) entity is observed to be splitting into two identical copies. One of these is again recognized as the original parent entity because of its pivot and the second entity would be claimed to be causally related with the parent entity as per the definition of $C$. To see this, notice that the parent entity at Time=$126$ contains all the cells of the child entity appearing at Time=$127$, which satisfies the definition of $C$. Between time steps $128$ and $151$ both parent and child entities undergo changes in the number and states of their cells but their pivots remain fixed. Hence they can again be recognized. Finally at Time=$151$ the child entity becomes identical to the original parent entity, therefore the parent entity at Time=$0$ and the child entity at Time=$151$ are related using $\Delta$. The transitive closure finally give us the final descendence relationship between the parent and the child entity.
### Mutations, Inheritance, and Natural Selection {#mutations-inheritance-and-natural-selection .unnumbered}
Primary focus of Langton while defining the CA based replicating loop model was to demonstrate that genotype - phenotype based coding decoding scheme can be captured in CA universe as well [@Langton87]. And we have seen that this can be observed by the observer as defined above. Nonetheless, Langton loops do not exhibit mutations and indeed if we analyze the underlying state transitions defined for the cells in the model, it becomes clear that the transition behavior required for the reproduction changes immediately if any changes are introduced in an entity and resulting entity is no longer capable of reproduction or in other terms, none of the mutations in existing replicating loops preserve reproduction and in terms of the current framework *Axiom* of *Preservation of Reproduction under Mutations* is not valid. Because of the enormity of possible configurations and transition dynamics it is not easy to analyze which kind of replicating loops can ever withstand certain mutations and can preserve replicating functionality. Heredity of course is worth considering only when entities mutate and continue reproduction. Thus with existing Langton loops, an observer cannot observe heredity and subsequent natural selection.
The extension of Langton loops defined by Sayama as *Evoloops* in [@Samaya98b] is one such attempt, where not all the loops in the model are of the same type with respect to the number and geometrical arrangement of cells and final population witnesses (small) variety of different kinds (in size) of reproducing loops scattered on the lattice forming colonies. The Evoloops and their evolution can be formulated in the framework by suitably modifying the definition of the distance measure $D$ to measure the differences between the entities in the number and geometric arrangement of cells and by changing limit $\delta_{rep\_mut}$ such that the observer is able to establish descendence relationship even when the parent and the child entities (loops) are not identical. Since evoloops of different types replicate at different rates, where rate of replication is measured in terms of number of state transitions, we can infer that the loops satisfy the axiom of sorting. Indeed in a weak sense with available simulation results it appears that evoloops can be observed demonstrating heredity as well as selection.
### Conclusion {#conclusion .unnumbered}
We have seen that we can formally define an observation process on the CA universe which discovers the self replication of so called Langton loops during the simulation of model. The specific observer presented here follows the intuition that Langton implicitly stated when describing the loops. We also noted that mutations, heredity, and selection based axioms are not met in the model where this limitation can be attributed to underlying transition rules of the model. Evoloops, which were designed as extensions of Langton loops with mutations can be seen to be evolving with variation in the sizes and rates of reproduction.
Case Study **[2]{}: Algorithmic Chemistry** {#chap:lambda}
-------------------------------------------
Algorithmic Chemistry (AlChemy) was introduced in [@ac:Fon92] and further discussed in [@ac:FB94arrival; @ac:FWB94; @FB94twice; @ac:FB96]. The main focus of the AlChemy is to study the principles behind the emergence of biological organizations with the approximate abstraction of real chemistry as $\lambda$ calculus with finite reductions. Starting with a random population of $\lambda$ terms (molecules), using different filtering conditions on reactions, authors describe the emergence of different kinds of organizations: *Level $0$* organization consisting of a set of self copying $\lambda$ terms and hypercycles with mutually copying $\lambda$ terms, *Level $1$* *self maintaining* organizations consisting of $\lambda$ terms such that every term is effectively produced as a result of reaction between some other terms in the same organization and lastly *Level $2$* organization consisting of two or more Level $1$ sub organizations such that molecules migrate between these self maintaining sub-organizations. They also provide detailed algebraic characterization of Level $1$ and Level $2$ organizations without referring to the underlying syntactical structure of the $\lambda$ terms (molecules) or the micro dynamics (reduction semantics and filtering conditions) governing the output of reactions.
### Instantiating the Framework {#instantiating-the-framework .unnumbered}
In view of the proposed observer based framework, characterization of self replicating molecules and hypercycles consisting of mutually copying molecules is achieved by defining an observation process, which focusses on individual $\lambda$ terms as entities and identifies hypercycles as a set of individually replicating $\lambda$ terms in a sequence of reaction steps (reflexive autocatalysis).
Since Level $1$ and Level $2$ organizations emerge only when self copying reactions are filtered out (i.e., self reproduction is not allowed) to ensure that Level $0$ organizational structures do not become the fixed points, these cannot be analyzed under the current framework design because we only consider reproduction, mutation, inheritance, and selection based evolution and emergence of organizations.
### The Chemistry Structure {#the-chemistry-structure .unnumbered}
A chemical soup of AlChemy consisting of $\lambda$ terms as molecules is usually initialized with a population of large number of randomly generated $\lambda$ terms. A *state* of the chemistry could, therefore, be considered as the collection of all these $\lambda$ terms (with multiplicity). Since every non elastic reaction results into introduction of output $\lambda$ term into the soup and possible removal of some other randomly chosen terms, it is natural to consider such succession of states after every reaction step as a state sequence $T \in \mathcal{T}$.
The components of the observation process defined next are based upon the assumption that it is possible to observe the inputs terms for a reaction (collision), resultant output term to be added to the soup, and the randomly deleted terms from the soup, without knowing the actual reaction details or the reduction semantics.
### Entities {#entities-1 .unnumbered}
For a given state of the chemistry, let the observation process identify each $\lambda$ term as a separate entity associating an unique integer tag with it. Each such entity is represented as $[w, i]$ where $i$ is the tag uniquely associated with $\lambda$ term $w$. $E$ is the set of all such entities in the chemistry.
[Tagging]{}: Suitable tagging mechanism needs to be defined by the observer to recognize whether two $\lambda$ terms in successive states are the same and to distinguish between multiple syntactically identical copies of a $\lambda$ term in the soup at any state. We can associate tags of the form ${{\langle}}i_{size},
i_{lex}, i_{mul}{{\rangle}}$ ($i_{size}, i_{lex}, i_{mul} \in N$) with the individual molecules in the following way: for the initial population of $\lambda$ terms, they are arranged with respect to their sizes and we assign the size of these terms as the first component in their tags ($i_{size}$) and for terms of same size arrange them lexicographically and assign in increasing order second component of their tags ($i_{lex}$) such that multiple copies of a term have the same first two components of their tags and then assign increasing integers to each of these as their third component of the tag ($i_{mul})$. Under such tagging scheme a small population of $\lambda$ terms $\{\lambda x. x, \lambda x.
x, \lambda x_1. \lambda x_2.x_2\}$ defines the state - $\{[\lambda
x.x, {{\langle}}3, 1, 1{{\rangle}}], [\lambda x. x, {{\langle}}3, 1, 2{{\rangle}}], [\lambda
x_1. \lambda x_2. x_2, {{\langle}}5, 1, 1{{\rangle}}]\}$. For a given tag $tg =
{{\langle}}i, j, k {{\rangle}}$ its components are accessed as $i = tg[1], j =
tg[2]$, and $k = tg[3]$.
Next we discuss the mechanism for *updating* these tags after reaction and elimination steps. We increment by one the third component of the tags for each entity , which was not deleted from the soup from previous state and give new unique tag to the new terms added to the soup with respect to their position in the list of terms based on their size and lexicographic order such that third component of the newly added terms is always given value $1$. This numbering scheme reliably maintains the recognition of terms across states of the chemistry.
### Distance Measure {#distance-measure-1 .unnumbered}
Distance function $D: E \times E \rightarrow \{0, 1\} \times \{0,
1\}$ is defined such that $\forall (e=[w, t_g], e'=[w', t'_g]) \in
E\ .\ D(e, e')[1] = 0$ if $w$ and $w'$ are the same with respect to $\alpha$ renaming implying that entity $e'$ is the same entity $e$ in the previous state; otherwise $D(e, e')[1] = 1$. $D(e,
e')[2] = 0$ if $t'_g[3] - t_g[3] = 1$ indicating that entity $e$ is observed in the next state as entity $e'$, otherwise $D(e,
e')[2] = 1$. The distance function $D$ has been defined keeping in mind the use of these distances in defining recognition relation later.
### The Limits on Observable Mutations {#the-limits-on-observable-mutations .unnumbered}
Let $\delta_{mut} = [0, 0]$, indicating that syntactically different $\lambda$ terms (under $\alpha$ renaming) are treated as different entities. Also let $\delta_{rep\_mut} = [0, 1]$ indicating that reproductive mutations resulting into syntactically different term are not observable. This is primarily because under $\beta$ reduction semantics of Alchemy, even changes in the syntactical representations result into very different reaction behaviors.
### Observing Self Replicating Hypercycles {#observing-self-replicating-hypercycles .unnumbered}
We can observe the self-replicating elementary hypercycles as sets of self-replicating entities. Let us define, for that purpose, the recognition relation $\mathbf{R_{\delta_{mut}}}$ $: E \rightarrow
E$ as follows: $\forall e, e' \in E$, $\mathbf{R_{\delta_{mut}}}$ $(e) = e' \Leftrightarrow [F(e') = F(e) + 1] \wedge D(e, e') \leq
\delta_{mut}$. Informally this means two entities in consecutive states are recognized same only using their tags.\
$\mathbf{R_{\delta_{mut}}}$ satisfies Axiom $1$, Axiom $2$, and Axiom $3$.
Axiom $1$ and Axiom $3$ are satisfied by definition. Axiom $2$, which states that $\mathbf{R_{\delta_{mut}}}$ is an injective function holds because of the specific construct of tagging mechanism and the definition of Distance function $D$ which is such that two entities in successive states are recognized as same only when the difference between their third components of tags is $1$, and we know that the observer selects new tags in such a way that this difference is $1$ only when same entity was present in the previous state.
Next let us defines $\Delta \subseteq E \times E$ such that $\forall e, e' \in E . (e, e') \in \Delta \Leftrightarrow D(e, e')
\leq \delta_{rep\_mut}$. In order to define causal relation between entities in the AlChemy, we assume that observer has the knowledge of the reacting entities and the output term at any state. Therefore if entities $e_1$ and $e_2$ react in some state and yield $e_o$, the observer defines causal relation $C$ so that $(e_1, e_o) \in C$ and $(e_2, e_o) \in C$ with $F(e_1) = F(e_2) =
F(e_o) -1$.\
Causal relation $C$ defined above satisfies Axiom $4$.
First condition of Axiom $4$ is satisfied by definition since $F(e_o) = F(e_1) + 1 = F(e_2) + 1$. The second condition $[\not
\exists e' \in E. F(e_1) = F(e') \wedge \mathbf{R_{\delta_{mut}}}$ $(e') = e_o]$, that is, there does not exist any third entity $e'$ in the previous state, which has mutated into $e_o$, again follows from the specific construct of tagging as well as the distance function because as per the tagging mechanism explained before $e_o$ being newly added entity in the chemistry will have the $3^{rd}$ component of its tag as $1$ and all previously present entities, including $e_1, e_2$, in the chemistry would have their tags in new states updated such that their $3^{rd}$ components are always greater than $1$.
Relations $\mathbf{AncestorOf}$ and $\mathbf{Parent}$ can be defined same as in the framework.\
**Axiom of Reproduction** and the **Axiom of Fecundity** are satisfied by the entities and corresponding abstractions discussed above.
These two axioms depend upon the examples of self replicating $\lambda$ terms as well as elementary hypercycles. In case of hypercycles, the observer establishes multi-step reproduction using transitive closure of causal relation for each of the entities in the hypercycle. A quite well known example of self replicating $\lambda$ term is $\lambda x. (x)(x)$ since $(\lambda
x. (x)(x))(\lambda x.(x)(x))\ {\Rightarrow}_{\beta}\ (\lambda x.
(x)(x))(\lambda x. (x)(x))$. Though in case of Alchemy, the level $0$ organization consists of self-copiers like $\lambda x. x$ and hypercycles like $\{\lambda x_1. \lambda x_2. x_2, \lambda x. x$} as illustrated in Figure 3. As per the definition of causal relation, entity instances of $\lambda x_1. \lambda x_2. x_2$ and of $\lambda x. x$ are causally related to past instances of each other and therefore of themselves.
### Mutations, Inheritance, and Natural Selection {#mutations-inheritance-and-natural-selection-1 .unnumbered}
As emphasized in [@ac:FB94arrival], primary goal of AlChemy is to study alternative pathways in which higher level organizations (i.e., hypercycles, self maintaining organizations) can emerge starting with a random set of molecules. Therefore it appears that there is no explicit notion of mutations present in the chemistry. To see this notice that every new entity in the population is the result of reaction between two other entities. Therefore if one particular observer decides that one of the reacting entities is mutating into the resulting entity, it is still difficult to decide which of the two reacting entities should be considered as mutating into the new one. Even if such a view is adopted, the observer will observe that if a self-copying entity at any reaction step mutates into another entity then most often the new entity can no longer self-copy. Thus *Axiom* $7$ (Preservation of Reproduction under Mutation) would be violated. Finally as discussed at the beginning of the section, owing to the focus of our framework on the evolutionary processes, self-maintaining organization of the kind that arise in AlChemy are beyond the scope.
### Conclusion {#conclusion-1 .unnumbered}
Thus we have demonstrated that, based upon the knowledge of reacting terms and outputs, a precise observation process can be defined to work with AlChmey, which can be used to discover the self replicating $\lambda$ terms as well as hypercycles in the model. We also noted that mutations, heredity, and selection based axioms are not met in the chemistry where this limitation should be attributed to underlying reaction semantics of the chemistry as well as its design. This study highlights the fact that not all interesting dynamic processes are evolutionary in nature and therefore some of these non evolutionary processes are out of scope of the framework at present.
Related Work {#chap:related}
============
Because of the presence of sufficiently many biology-specific criterion (e.g., morphological characters, bio-molecular structures etc.) to distinguish life from non-life, in biological literature there is little formal work on recognizing life [*per se*]{}. There is, however some recent work on defining and developing methods to analyze genotype space structure based upon the macroscopic observations on phenotype characteristics (mainly morphological and reactive characteristics) [@Garay98; @Gamez03; @Lopez04].
To the authors’ knowledge, there is not much work focussing on the observation process for ALife studies reported in literature. Though there exist proposals to define ‘numerical parameters’ or ‘statistics’ [@Bedau99] to recognize life in a model. However, it is not clear whether there can be simple numerical definitions capturing the essence of life in arbitrary models and even if so does not seem to be the case with the existing proposals. The difficulty arises out of intricate nature of reproduction and selection inevitably involving non trivial identification of the population of evolving entities. Langton defined in [@Langton91] a quantitative matric, called *lambda* parameter to detect life in any generic one dimensional cellular automata model based upon the characteristics of its transition rules. This lambda parameter based analysis is based upon the assumption that any self organizing system can be treated as living and does not consider population centric evolutionary behavior as characteristic of life. In [@Bedau98] there is a discussion on the classification of long term adaptive evolutionary dynamics in natural and artificially evolving systems. This they achieve by defining activity statistics for the components, which quantifies the adaptive value of components (characteristics in our model). They employ similar mechanism as of ours by associating activity counters (tags) with all the components present in the system during simulation.
Self-reproduction, which has a long history of research starting from the late 1950s [@Burks70; @Sipper98; @Freitas04] has evaded precise formal definition applicable to a wide range of models [@ND98] in the sense of observable characterization of the reproducing entities. Though there is enough work on mathematical analysis of replication dynamics (fecundity) in various natural systems or the systems where environmental constraints governing the rate of reproductions are known (see for overview [@Freitas04 Chap5].) In some of the discussions related to self-replication in cellular automata models [@Samaya98b; @Morita98], formalizations of reproducing structures are presented, but they do not attempt to provide a general framework for observing reproduction or other components of evolutionary processes. These attempts at formalizing reproduction in CA models are reminiscent of our definition of entities (loops) in Section \[sec:langton\_inst\].
In other work [@Misra06], we proposed a multi-set theoretic framework to formalize self reproduction (with mutations) in dynamical hierarchies in terms of hierarchal multi-sets and corresponding inductively defined meta-reactions. The “self" in “self-reproduction" was defined in terms of [*observed structural equivalences*]{} between entities. We also introduced constraints to distinguish a simple “collection" of reacting entities from genuine cases of “emergent" organizational structures consisting of [*semantically coupled*]{} multi-set of entities.
Conclusion {#sec:concluding}
==========
General Remarks
---------------
This paper formalizes an implicit underlying component of ALife studies, namely the observation process, by which entities are identified and their evolution is observed in a particular ALife simulation. Under the assumption that the essence of life-like phenomena is their evolutionary behavior, we developed a framework to formally capture basic components of evolutionary phenomena. This work, in essence, brings insights from evolutionary theory for real-life into the realm of artificial-life for defining a formal framework for observational processes, which are needed for the identification of life-like phenomena in the ALife studies. We have argued that without such a formalism, claims pertaining to the evolutionary behavior in ALife studies will remain inconclusive.
We formally elaborate in algebraic terms the necessary and sufficient steps for an observational process, to be employed by an ALife researcher upon the time progressive model of his model universe, to uncover (hidden) life-like phenomena in the light of Darwinian evolution as defining characteristics of life. The observation process as specified in our framework may be carried out manually or can be alternatively algorithmically programmed and integrated within the model.
To define inference process we specify necessary conditions, as axioms, which must be satisfied by the outcomes of observations made upon the model universe in order to infer whether life-like phenomena is present in the model (Section \[chap:evocomponents\]). These axioms also specify the experimental work necessary in order to observe and lay claims for the presence of life in the model universe. The case studies on Langton loops (Section \[chap:langton\]) and Algorithmic Chemistry (Section \[chap:lambda\]) highlight the contributions that such an approach can make to the discussion of specific ALife experiments. An important property of such a study is to make explicit “multi-level observations", where entities and their relationship can be observed and defined on separate organizational levels.
The framework design and the case study analysis also provide us clues for ALife research designs so that to be better able to witness evolutionary phenomena in the model during its simulations. This is discussed next:
Design Suggestions for ALife Researchers {#sec:design-suggst}
----------------------------------------
As the framework is based upon the Darwinistic concepts of defining life in terms of evolutionary processes, the design suggestions we describe here are rather more suitable for those studies which aim to complement real life studies in an evolutionary framework.
- **Sufficient Reproduction with Variation:** The model must be designed such that there exist potentially large set of reproducing entities with significant variation in their characteristics. Quite often this hinges upon the choice of reaction rules or the semantics of the model and indeed it is a serious challenge for any model designer to define the reaction semantics which permits potentially large set of reproducers with significant variation. Another interesting aspect is that these reproducers must be relatively closely related to each other under the reaction semantics. This means that sufficiently many variations of reproducers should also be reproducers in themselves otherwise the axiom of preservation of reproduction under mutation will not effectively hold in the model and most of the reproducers would have to appear *de novo* during simulations. We encounter this problem in both of the case studies discussed in Section \[sec:case-studies\]. In case of Langton loops, any kind of change in the loop structure would cause caseation of replication. The work on designing Evoloops is therefore based upon the redefinition of the reaction semantics or transition rules which permit variation in replicating loops. Similarly in the case of Algorithmic chemistry, almost all of the single replicating $\lambda$ terms arise *de novo* and their variations do not replicate under $\beta$ reaction semantics.
- **Measurable Rates of Reproduction:** The model should be designed such that it is possible to impose some valid measure of determining the rates of reactions which in turn can be used to estimate differences in the rates of reproduction of different entities. This measurement of reproductive rates must be independent of the updation algorithm which selects entities for reaction. Therefore it can be argued that the models, where all (reproductive) reactions take place in a single step would be difficult to observe for natural selection, which works only when different entities reproduce at different rates. For example, it is not possible to infer differences in the rates of reproduction among different reproducing elementary hypercycles in the Algorithmic Chemistry consisting of the same number of $\lambda$ terms because every reaction between any two $\lambda$ terms occurs in a single step. On the other hand natural selection can be observed in case of Evoloops precisely because different types of loops consisting of different number of cells reproduce at different rates based upon the number of state transitions.
Limitations {#sec:limit}
-----------
The decision to equate life with evolutionary processes also excludes some of the interesting complex phenomena that are not evolutionary in nature from the scope of this work. Indeed, we have shown in Section \[chap:lambda\] that the framework cannot account for the dynamic non-evolutionary behavior of Level $1$ and Level $2$ organizations emerging in the Algorithmic Chemistry. We limit our attention to only those observations having evolutionary significance, though other observations can also be made upon the model including metabolism [@ac:BFF92], emergence of complexity [@ac:adami00], self organization [@ac:Kau93], and autonomous and autopoitic nature of life [@Zeleny81] etc.
We have not placed direct emphasis on certain concepts widely associated with ALife studies including the notion of “emergence”. In our current setting the notion of “strong emergence” is only implicitly present and indeed “the element of surprise” [@Bass97] often associated with emergence is not immediate in the framework. Similarly “the element of autonomy” of emergent processes with respect to the underlying micro-level dynamics is not addressed in our framework. Indeed, the spirit of the high level of observations and corresponding abstractions upon which the framework rests, may preclude such inferences. Nonetheless the idea of “weak emergence" [@Bedau97], which lays emphasis on the simulations of the model for the emergence of high level macro-states is fundamental to our framework, where the observation process is by default based upon the simulations of the model and not on analytical derivations.
Another limitation of the framework in its current state is that it cannot be used effectively to make predictions regarding the possible observable evolutionary dynamics in a ALife model during simulations. This limitation though carries forward from the nature of Darwinian theory which is too generic in its conceptualization and based upon random sources of change that make it difficult to derive useful predictions.
Similarly analysis of G$\ddot{o}$delian type conjunctures to counter possibility of strong Alife, stating the impossibility of formalizing life in general because that would imply formalizing “mathematically intelligent" entities like ourselves, which could in tern prove the G$\ddot{o}$del theorems in their own “mathematical universe" having correspondence with ours, is also beyond the scope of the current limits of the framework. See [@sullins97; @Rasmussen92].
### Problem of False Positives {#false+ve .unnumbered}
Terms ‘false positive’ and ‘false negative’ are used in general to highlight the limitations of ‘observation - inference’ based methodologies. *False positive* refers to a situation where observations and consequent inferences on a model result into a claim of the presence of certain property in the model which actually does not exist, while *false negative* is used to refer the situation where observations do not yield required support for the presence of certain property, which is actually present in the model. False negatives are usually the result of incomplete observations while false positives indicate arbitrariness in the observation/inference process.
Like any other generic specification framework, current framework also suffers from the weakness of administering false positives. False negatives are also possible, whereby an observation process is defined such that it does not infer evolution, even though there might actually be evolution present in the model. The case of false negatives, however will not concern us since our focus is to establish the presence of evolution in a given ALife model and not whether it is absent with respect to certain observations. The problem of false positives stems due to the fact that the framework permits arbitrariness in the definition of entities and their causal relationships. In case of causal relationships, they are defined in the framework as observation dependent and might not be consistent with the underlying micro-level dynamics of the model (Section \[reproformalized\]). This arbitrariness might give rise to false claims on the presence of evolution in the model though there might be none actually.
For example, an observer (say $ob$) might decide to “ignore" entities in some states in the beginning and then choose later on to observe them in some other states so that to use them for establishing (false) evolutionary relationships, which would not have been possible had he not preferred to ignore them earlier. This problem of selectively observing entities in various states requires additional constraints in the framework. We may add the following constraint by considering another observer $ob'$ with same universe of observation as $ob$. Let us consider a particular simulation of a model as a state sequence $T$. For a state subsequence $S$ of $T$, let $E_{ob}^S$ and $E_{ob'}^S$ denote the set of entities observed by $ob$ and $ob'$ respectively. Consider that $ob'$ observes some entities $\mathcal{X} \subseteq E_{ob'}^S$, which were ignored by $ob$, that is, $\mathcal{X}\ \not\subseteq E_{ob}^S$. Now consider the case when $ob$ chooses to observe $\mathcal{X}$ in some later subsequence $S'$ of $T$, $S \neq S'$, that is, $\mathcal{X}$ $\subseteq E_{ob}^{S'}$, and also $\mathcal{X}$ $\subseteq
E_{ob'}^{S'}$, where $E_{ob}^{S'}$, and $E_{ob'}^{S'}$ are the sets of entities observed by $ob$ and $ob'$ in $S'$. Now if $ob$ establishes evolutionary relationships using entities in $\mathcal{X}$, which cannot be established by $ob'$, then we say that $ob$ has drawn *illegitimate* conclusions.
Further work {#sec:further-work}
------------
Framework can be further extended in several interesting directions, including the following: We need to capture the essence of *strong emergence* by considering several observation processes at different organizational levels of the model. We can also study overlapping evolutionary processes - examples from real life include co-evolution, and sexual selection versus environmental selection. Framework ought to be extended so that fruitful predictions for a given ALife model regarding the nature of evolutionary dynamics can be made. We also need to introduce more strict constraints to overcome the problem of false positives by limiting as to what could be claimed as observed. Further insights can be gained by applying the framework to novel classes of ALife models to refine the framework further, which we are currently involved with.
[^1]: A *multiset* $M$ on a set $E$ is a mapping associating nonnegative integers (representing multiplicities) with each element of $E$, $M: E \rightarrow \mathcal{N}$. Informally a multiset may contain multiple copies of its elements.
| {
"pile_set_name": "ArXiv"
} |
---
author:
- Liguang Liu and Peter Sjögren
date: '13 April, 2015'
title: '**On the global Gaussian Lipschitz space** '
---
[**Abstract.**]{} A Lipschitz space is defined in the Ornstein-Uhlenbeck setting, by means of a bound for the gradient of the Ornstein-Uhlenbeck Poisson integral. This space is then characterized with a Lipschitz-type continuity condition. These functions turn out to have at most logarithmic growth at infinity. The analogous Lipschitz space containing only bounded functions was introduced by Gatto and Urbina and has been characterized by the authors in [@LS].
Introduction and main result
============================
Consider the Euclidean space $\rn$ endowed with the Gaussian measure $\gz$, given by $$d\gz(x)=\pi^{-n/2}e^{-|x|^2}.$$ The Gaussian analogue of the Euclidean Laplacian is the *Ornstein-Uhlenbeck operator* $$\cl =-\frac12 \Delta +x\cdot\nabla ,$$ where $\nabla = (\partial_{x_1}, \dots, \partial_{x_n}).$ The heat semigroup generated by $\cl$ and defined in $L^2(\gz)$ is the so-called *Ornstein-Uhlenbeck semigroup* $$T_t = e^{-t\cl},\quad t\ge0.$$ The *Ornstein-Uhlenbeck Poisson semigroup* $P_t = e^{-t\sqrt{-\cl}}, \;t\ge0$, can be defined from $\{T_t\}_{t\ge0}$ by subordination as $$P_tf(x)=\frac1{\sqrt \pi}\int_0^\infty \frac{e^{-u}}{\sqrt u}
\,T_{t^2/(4u)}f(x)\,du,\qquad x\in\rn,$$ for $f\in L^2(\gz)$. As explained in Section 2, $P_tf$ is given by integration against a kernel $P_t(x,y)$.
Via $\{P_t\}_{t\ge0}$, Gatto and Urbina [@GU] introduced the *Gaussian Lipschitz space* $\glip_\az$ for all $\az>0$. We shall always have $\az\in(0,1)$. Then the definition says that a function $f$ in $\rn$ is in $\glip_\az$ if it is bounded and satisfies $$\label{GLip}
\|t\partial_t P_tf\|_{L^\infty}\le A t^\az,\qquad t>0,$$ for some $A>0$. These spaces and also Gaussian Besov spaces were studied in a series of works; see [@GPU; @GU; @PU] and also the authors’ paper [@LS].
In [@LS], the authors characterized $\mathrm{GLip}_\az,\;0<\az<1$, in terms of a Lipschitz-type continuity condition. Indeed, Theorem 1.1 of [@LS] says that $f\in \mathrm{GLip}_\az$ if and only if there exists a positive constant $K$ such that $$\label{strong-lip}
|f(x)-f(y)|\le K \min\lf\{|x-y|^\az,\;\,
\lf(\frac{|x-y_x|}{1+|x|}\r)^{\frac\az2} +|y_x^\prime|^{\az}\,
\r\},
\qquad x,\:y\in\rn.$$ Here and in what follows, we use a decomposition of $y$ as $y=y_x+y_x^\prime$, where $y_x$ is parallel to $x$ and $y_x^\prime $ orthogonal to $x$; however, if $x=0$ or $n=1$, we let $y_x=y$ and $y_x^\prime=0$.
As is well known, a condition analogous to for the standard Poisson integral characterizes the ordinary Lipschitz space; see [@S1 Sect. V.4]. If only bounded functions are considered, one obtains the inhomogeneous Lipschitz space, and without the boundedness assumption the larger, homogeneous Lipschitz space.
In our setting, we shall see that the condition without the boundedness condition defines a Gaussian analogue of the homogeneous Lipschitz space. Since here no homogeneity is involved, we shall call it the *global Gaussian Lipschitz space*.
In , an a priori assumption is needed to assure that $P_t f$ exists. Here we apply a recent result by Garrigós, Harzstein, Signes, Torrea and Viviani [@GHSTV]. Clearly, a measurable function $f$ in $\rn$ has a well-defined Gaussian Poisson integral if $$\int P_t(x,y) |f(y)|\,dy<\infty,$$ for all $x\in \rn$ and $t>0$. Theorem 1.1 of [@GHSTV] says that this is equivalent to the growth condition $$\label{cond:GHSTV}
\int_{\rn} \frac{e^{-|y|^2}}{\sqrt{\ln(e+|y|)}}\,|f(y)|\,dy<\infty.$$ Moreover, ensures that $P_t f(x) \to f(x)$ as $t\to 0$ for a.a. $x\in \rn$.
We can now define the global Gaussian Lipschitz space.
Let $\az\in(0,1)$. A measurable function $f$ defined in $\rn$ and satisfying belongs to the global Gaussian Lipschitz space $\mathrm{GGLip}_\az$ if holds. The corresponding norm is $$\|f\|_{\mathrm{GGLip}_\az}
=\inf\{A>0:\, A \,\;\textup{satisfies}\,\; \eqref{GLip}\}.$$
Strictly speaking, this space consists of functions modulo constants. A natural question is now what continuity condition characterizes this space. To state the answer, we start in one dimension and introduce a distance by $$\label{d}
d(x,y) = \left| \int_{x}^y \frac{d \xi} {1+|\xi|}\right|, \qquad x, y \in \rr.$$ Then $$d(x,y) = \left| \ln(1+ |x|) - \mathrm{sgn}\, xy \; \ln(1+ |y|) \right|$$ for all $x, y \in \rr$, provided we define $\mathrm{sgn}\, 0= 1$. In several dimensions, we use this distance on the line spanned by $x$, defining $$d(x,y_x) = \left| \ln(1+ |x|) - \mathrm{sgn} \langle x,y\rangle \ln(1+ |y_x|) \right|, \qquad x, y \in \rn,$$ with $y_x$ as before.
Our result reads as follows.
\[thm1.2\] Let $\az\in(0,1)$ and let $f$ be a measurable function in $\rn$. The following are equivalent:
1. $f$ satisfies and $f\in \rm{GGLip}_\az$;
2. There exists a positive constant $K$ such that $$\label{eq:gglip}
|f(x)-f(y)|
\le K\min\lf\{
|x-y|^\alpha,\; d(x, y_x)^{\frac{\az} 2}
+|y_x'|^\alpha
\r\},
\qquad x,\:y\in\rn,$$ after correction of $f$ on a null set.
Moreover, $$\label{norm}
\|f\|_{\mathrm{GGLip}_\az} \simeq
\inf\{K:\, K\,\;\textup{satisfies}\,\; \eqref{eq:gglip}\}.$$
The meaning of the symbol $\simeq$ is explained below.
To compare and , one easily verifies that if $\langle x,y\rangle > 0$ and $1/2 < |x|/|y_x| < 2$, then $$\label{comp}
d(x,y_x) \simeq \frac{|x-y_x|}{1+|x|}.$$ Moreover, the space $\rm{GLip}_\alpha$ can be described in terms of the distance function $d$. Indeed, as implies boundedness (see [@LS Lemma 2.1]), it is easy to check that holds if and only if there exists a constant $K'>0$ such that $$|f(x)-f(y)|
\le K'\min\lf\{1,\;
|x-y|^\alpha,\; d(x, y_x)^{\frac{\az} 2}
+|y_x'|^\alpha
\r\}$$ for all $x,\,y\in\rn$. This also tells us that for bounded functions, is equivalent to . But implies only that $$f(x) = O((\ln |x|)^{\alpha/2}) \qquad \mathrm{as} \qquad |x| \to \infty.$$ This condition is sharp, as shown by an example in Section 5; observe that it is much stronger than .
The paper is organized as follows. Section 2 contains a needed improvement of the estimate for $P_t(x,y)$ and its derivatives in [@LS]. Some properties of the Gaussian Poisson integral are obtained in Section 3. Then Theorem \[thm1.2\] is proved in Section 4. Finally, we give in Section 5 an example of a function in $\rm{GGLip}_\az$ with logarithmic growth.
**Notation.** Throughout the paper, we shall write $C$ for various positive constants which depend only on $n$ and $\az$, unless otherwise explicitly stated. Given any two nonnegative quantities $A$ and $B$, the notation $A\lesssim B$ stands for $A\le C B$ (we say that $A$ is controlled by $B$), and $A\gtrsim B$ means $B\ls A$. If $B\ls A\ls B$, we write $A\simeq B$.
For positive quantities $X$, we shall write $\exp^*(-X),$ meaning $\exp(-cX)$ for some constant $c=c(n,\az)>0$.
The Ornstein-Uhlenbeck Poisson kernel {#sect2}
=====================================
It is known that for $f\in L^2(\gz)$, $$T_tf(x) =\frac1{\pi^{n/2}}\int_\rn M_{e^{-t}}(x,y)f(y)\,dy, \qquad x\in\rn,\;\;t>0,$$ where $M_{e^{-t}}$ is the *Mehler kernel* defined by $$M_r(x,y)=\frac{e^{-\frac{|y-rx|^2}{1-r^2}}}{(1-r^2)^{n/2}},
\qquad x,\,y\in\rn,\quad 0<r<1.$$ The Gaussian Poisson integral $ P_tf$ is given by an integral kernel called the *Ornstein-Uhlenbeck Poisson kernel* and denoted by $P_t(x,y)$; thus $$P_tf(x)=\int_{\rn} P_t(x,y) f(y)\, dy, \qquad x\in\rn,\;\;t>0.$$ Because of the subordination formula, $P_t(x,y)$ is given by $$\begin{aligned}
P_t(x,y) &=
\frac1{ \pi^{(n+1)/2}}\int_0^\infty \frac{e^{-u}}{\sqrt u}
\,M_{e^{-t^2/(4u)}}(x,y)\,du \notag \\
&= \frac1{2\pi^{(n+1)/2}} \int_0^\infty
\frac{t}{s^{3/2}} \,e^{-\frac{t^2}{4s}}\,
\frac{\exp(-\frac{|y-e^{-s}x|^2}{1-e^{-2s}})}{(1-e^{-2s})^{n/2}} \,ds.
\label{poisson-ker1}\end{aligned}$$ Here we inserted the expression for the Mehler kernel and transformed the variable.
The following estimate for $P_t$ and its first derivatives is established in [@LS Theorems 1.2 and 1.3].
\[prop2.1\] For all $ t>0$, $x,\:y\in \rn$ and $i\in\{1,2,\dots,n\}$, the kernel $P_t$ satisfies $$\begin{aligned}
P_t(x, y) +|t\partial_t P_t(x,y)|+|t\partial_{x_i} P_t(x,y)|
\le C \lf[K_1(t,x,y)+K_2(t,x,y)+K_3(t,x,y)+K_4(t,x,y)\r],\end{aligned}$$ where $$\begin{aligned}
K_1(t,x,y) &=& \frac t{(t^2+|x-y|^2)^{(n+1)/2}}\,\exp^\ast\lf(-t(1+|x|)\r); \\
K_2(t,x,y) &=&\frac t{|x|} \left(t^2+ \frac{|x-y_x|}{|x|}+|y_x^\prime |^2\right)^{
-\frac{n+2}2}\,\exp^\ast\lf(-\frac{(t^2+|y_x'|^2)|x|}{|x-y_x|}\r)\,
\chi_{\{|x|>1,\; x\cdot y>0,\, |x|/2\le |y_x|<|x|\}};\\
K_3(t,x,y) &=& \mathrm{min}(1,t)\, \exp^\ast(-|y|^2);\\
K_4(t,x,y) &=&\frac t{|y_x|} \left(\ln \frac {|x|} {|y_x|}\right)^{ -\frac32} \,
\exp^\ast\left(-\frac {t^2}{\ln \frac {|x|} {|y_x|}}\right) \,
\exp^\ast(-|y_x^\prime |^2) \, \chi_{\{x \cdot y>0,\:1<|y_x|<|x|/2\}}.
\end{aligned}$$
We need a slight sharpening of this lemma. The term $K_3$ will be modified to decay for large $x$.
\[lem2.2\] The estimate of Proposition \[prop2.1\] remains valid if the kernel $K_3(t,x,y)$ is replaced by $$\begin{aligned}
\wz{K_3}(t,x,y)= \min\lf\{1,\,\frac{t }{ [\ln(e+|x|)]^{1/2}}\r\}\exp^\ast(-|y|^2).\end{aligned}$$
From the proof of [@LS Theorem 1.3], we see that $|t\partial_t P_t(x,y)|$ and $|t\partial_{x_i} P_t(x,y)|$ can be controlled by an integral similar to the right-hand side of (only with $\exp$ in replaced by $\exp^\ast$). Thus, we only need to consider $P_t(x,y)$.
When $|x|\le 4+2|y_1|$, we have $
\exp^\ast(-|y|^2) \ls \exp^\ast(-|y|^2) \exp^\ast(-|x|^2)
$ and hence $K_3(t,x,y)\ls \wz{K_3}(t,x,y).$
Thus we assume from now on that $|x| > 4+2|y_1|$. We shall sharpen a few arguments in the proof of [@LS Proposition 4.1]. By the rotation invariance of $P_t(x,y)$ and $\wz{K_3}(t,x,y)$, we may assume that $x=(x_1,0,\dots,0)$ with $x_1 > 0$. The decomposition of $y$ will then be written $y = (y_1,0,\dots,0) + (0,y')$, and $|y_1| < x_1/2$.
[*Case 1.*]{} $-x_1/2<y_1\le 0$. Using the notation from the proof of [@LS Proposition 4.1(i)], we see that we only need to verify that $J_2 \ls \wz{K_3}$. By [@LS formula (4.9)] and the fact that $y_1 \le 0< x_1$, we have $ $ $$\begin{aligned}
J_2 &&\simeq \exp^\ast(-|y'|^2)\,\int_{\ln 2}^\infty
\frac {t}{s^{3/2}}\,\exp^\ast\lf(-\frac{t^2}{s}\r)\,
\exp^\ast(-|y_1-e^{-s}x_1|^2) \,ds \notag\\
&&\ls \exp^\ast(-|y|^2)\,\int_{\ln 2}^\infty
\frac {t}{s^{3/2}}\,\exp^\ast\lf(-\frac{t^2}{s}\r)\,
\exp^\ast(-e^{-2s}x_1^2) \,ds. \label{j2}\end{aligned}$$ Note that $$\begin{aligned}
\int_{\frac12\ln x_1}^\infty
\frac {t}{s^{3/2}}\,\exp^\ast\lf(-\frac{t^2}{s}\r)\,
\exp^\ast(-e^{-2s}x_1^2) \,ds
&&\simeq \int_{\frac12\ln x_1}^\infty
\frac {t}{s^{3/2}}\,\exp^\ast\lf(-\frac{t^2}{s}\r)\,
\,ds\\
&&\ls \min\lf\{1,\, t (\ln x_1)^{-1/2}\r\} \notag\end{aligned}$$ and $$\begin{aligned}
\quad\int_{\ln 2}^{\frac12\ln x_1}
\frac {t}{s^{3/2}}\,\exp^\ast\lf(-\frac{t^2}{s}\r)\,
\exp^\ast(-e^{-2s}x_1^2) \,ds
&&\le \exp^\ast(-x_1) \int_{\ln 2}^{\frac12\ln x_1}
\frac {t}{s^{3/2}}\,\exp^\ast\lf(-\frac{t^2}{s}\r)\,
\,ds\\
&&\ls\exp^\ast(-x_1) \min\{1,\,t\}, \notag\end{aligned}$$ from which the required estimate follows.
[*Case 2:*]{} $0<y_1<x_1/2$. Considering now the proof of [@LS Proposition 4.1(iii)], we only need to estimate the terms $J_{2,1}^{(2)}$ and $J_{2,3}$, and also $J_{2,2}$ when $y_1\in(0,1]$.
From [@LS formula (4.16)], we get for $y_1\in(0,1]$, $$\begin{aligned}
J_{2,2}
&&\simeq \frac t{(\ln \frac{x_1}{y_1})^ {3/2}}
\exp^*{ \left(-\frac{t^2}{\ln \frac{x_1}{y_1}}\right)} \exp^*{(-|y' |^2)}\\
&&\ls
\min\lf\{\frac t{(\ln \frac{x_1}{y_1})^{3/2}} ,\, \frac 1{\ln \frac{x_1}{y_1}} \r\}
\exp^*{(-|y|^2)}\\
&&\ls \wz{K_3}(t,x,y),\end{aligned}$$ since here $\ln \,(x_1/y_1) \gs \ln\,(e+|x|)$. Further, $$\begin{aligned}
\label{eq2.4}
J_{2,1}^{(2)}+J_{2,3}
&& \le \exp^\ast(-|y'|^2)\,
\int \frac {t}{s^{3/2}}\,\exp^\ast\lf(-\frac{t^2}{s}\r)\,
\exp^\ast(-|y_1-e^{-s}x_1|^2) \,ds,\end{aligned}$$ where the integral is taken over the set $\{s> \ln 2: |s-\ln\, (x_1/y_1)| > c_0 \}$, for some $c_0>0$. Thus the quotient $e^{-s}x_1/y_1$ stays away from $1$ in this integral, so that $|y_1-e^{-s}x_1| \simeq \max\{e^{-s} x_1,\, y_1 \}\simeq e^{-s}x_1+y_1$. This implies that the right-hand side of is controlled by the expression in and thus by $\wz{K_3}$.
Lemma \[lem2.2\] is proved.
Auxiliary lemmas {#sect3}
================
\[lem3.1\] There exists a constant $C>0$ such that for all $x,\,y\in\rn$ and $t>0$, $$|\partial_tP_t(x,y)|
\le C\, \frac1{t}\,P_{t/2}(x,y).$$
Differentiating , we get $$\partial_{t} P_t(x,y)
= \frac1{2\pi^{(n+1)/2}}\,\frac1{t} \int_0^\infty \frac {t}{s^{3/2}}\,e^{-\frac{t^2}{4s}}\,
\lf(1-\frac{t^2}{2s}\r)
\frac{e^{-\frac{|y-e^{-s}x|^2}{1-e^{-2s}}}}{(1-e^{-2s})^{n/2}} \,ds.$$ It is now enough to observe that $$e^{-\frac{t^2}{4s}}\,
\lf|1-\frac{t^2}{2s} \r| \ls e^{-\frac{{(t/2)}^2}{4s}}$$ and compare with .
\[lem3.2\] Fix $i\in\{1,2,\dots,n\}$ and let $R>0$. Then there exists a constant $C>0$, depending only on $n$ and $R$, such that for all $x,\:y\in\rn$ with $|x|< R$, $$\label{a}
|\partial_{x_i}P_t(x,y)|
\le C\, (1 + t^{-4-n}) P_{t/2}(x,y), \qquad t>0,$$ and $$\label{b}
|\partial_{x_i}P_t(x,y)|
\le C t^{-1/2}e^{-|y|^2} [\ln (e+|y|)]^{-3/4}, \qquad t>1.$$
In this proof, all constants denoted $C$ will depend only on $n$ and $R$, and the same applies to the implicit constants in the $\lesssim$ and $\simeq$ symbols. We let $|x|< R$, and we can clearly assume that $R>1$.
Differentiating , we get $$\label{xj}
\partial_{x_i}P_t(x,y)= \frac1{\pi^{(n+1)/2}} \int_0^\infty
\frac{t}{s^{3/2}} \,e^{-\frac{t^2}{4s}}\, \frac{e^{-s}(y_i-e^{-s}x_i)}{1-e^{-2s}}\,
\frac{\exp(-\frac{|y-e^{-s}x|^2}{1-e^{-2s}})}{(1-e^{-2s})^{n/2}} \,ds.$$ Compared with , the integral has now an extra factor ${e^{-s}(y_i-e^{-s}x_i)}/{(1-e^{-2s})}$.
With $\gamma>0$, we shall use repeatedly the simple inequality $$\label{simple}
e^{-\frac{t^2}{4s}} \le C_\gamma \left(\frac{s}{t^2}\right)^\gamma e^{-\frac{(t/2)^2}{4s}}$$ for some $ C_\gamma>0$, and here we sometimes drop the last factor.
We start with the simple case of bounded $y$; more precisely we assume $|y|\le e^{12}\,R$. Then the extra factor is no larger than $Ce^{-s}/{(1-e^{-2s})}$. An application of with $\gamma = 1+n/2$ yields $$|\partial_{x_i}P_t(x,y)| \lesssim t^{-2-n} \int_0^\infty \frac{t}{s^{3/2}} \,
e^{-\frac{(t/2)^2}{4s}}\frac{e^{-s}s^{1+n/2}}{(1-e^{-2s})^{1+n/2}}\,
\exp\left(-\frac{|y-e^{-s}x|^2}{1-e^{-2s}}\right)\,ds.$$ Comparing with , one sees that this estimate implies . If we choose instead $\gamma = 2+n/2$, will also follow, since $y$ stays bounded.
From now on, we assume that $|y|> e^{12}\,R$. Then implies $$\label{xjj}
|\partial_{x_i}P_t(x,y)| \lesssim \int_0^\infty
\frac{t}{s^{3/2}} \,e^{-\frac{t^2}{4s}}\, \frac{e^{-s}|y|}{1-e^{-2s}}\,
\frac{\exp(-\frac{|y-e^{-s}x|^2}{1-e^{-2s}})}{(1-e^{-2s})^{n/2}} \,ds.$$
We first estimate the exponent $$E(s,x,y) = -\frac{|y-e^{-s}x|^2}{1-e^{-2s}}$$ from . It satisfies $$E(s,x,y) \le \frac{-|y|^2 + 2e^{-s} y\cdot x}{1-e^{-2s}}
\le \frac{-|y|^2 + \frac12 e^{-2s}|y|^2 + 2 |x|^2}{1-e^{-2s}},$$ where we applied the inequality between the geometric and arithmetic means. If $e^{-s}<1/2$, then $$E(s,x,y) \le \frac{-|y|^2 + \frac12 e^{-2s}|y|^2}{1-e^{-2s}} + C.$$ If instead $e^{-s}\ge 1/2$, we have $2 |x|^2 < e^{-2s}|y|^2/4 $ since $|y|>e^{12}|x|$, and thus $$E(s,x,y) \le \frac{-|y|^2 + \frac34 e^{-2s}|y|^2}{1-e^{-2s}}.$$ In both cases, $$E(s,x,y) \le -|y|^2 \: \frac{1 - \frac34 e^{-2s}}{1-e^{-2s}} + C
\le -|y|^2\left(1 + \frac14 e^{-2s}\right) + C,$$ and this implies $$\label{expon}
e^{E(s,x,y)} \lesssim e^{-|y|^2} \min\left(1, \frac{e^{2s}}{|y|^2} \right).$$
We also need a converse inequality, under the assumption that $s>\ln|y|$. Then $$\label{converse}
E(s,x,y) \ge \frac{-|y|^2 - 2e^{-s} |y||x| -e^{-2s}|x|^2}{1-e^{-2s}}
\ge \frac{-|y|^2}{1-|y|^{-2}} - C \ge-|y|^2 - C.$$
Now split the integral in as $$\left(\int_0^3 + \int_3^{\ln{|y|}} +\int_{\ln{|y|}}^\infty\right) \frac{t}{s^{3/2}} \,e^{-\frac{t^2}{4s}}\, \frac{e^{-s}|y|}{1-e^{-2s}}\,
\frac{\exp(-\frac{|y-e^{-s}x|^2}{1-e^{-2s}})}{(1-e^{-2s})^{n/2}} \,ds
= I_1 + I_2 + I_3,$$ say; observe that $\ln|y|>12$. We shall prove that these three integrals satisfy the bounds in and .
In $I_3$, we have ${e^{-s}|y|}/{(1-e^{-2s})} \lesssim 1$. Comparing with , we conclude that $$I_3 \lesssim P_{t}(x,y) \lesssim P_{t/2}(x,y),$$ which is part of . Aiming at , we apply with $\gamma = 3/4$ and , where the minimum is 1, to conclude that $$I_3 \lesssim \int_{\ln{|y|}}^\infty t^{-1/2}\, s^{-3/4}\, e^{-s}\,|y|\, e^{-|y|^2} \,ds
\lesssim t^{-1/2} \, (\ln |y|)^{-3/4}\, e^{-|y|^2},$$ as desired.
To deal with $ I_2$, we apply , now with the second quantity in the minimum, and obtain $$\label{middle}
I_2 \lesssim \int_3^{\ln|y|} \frac{t}{s^{3/2}} \,
e^{-\frac{t^2}{4s}} \, \frac{e^s}{|y|}\,e^{-|y|^2} \,ds.$$ Using , again with $\gamma = 3/4$, we can estimate this integral by $$t^{-1/2} e^{-|y|^2} \int_3^{\ln|y|} s^{-3/4}\, \frac{e^s}{|y|} \,ds,$$ which gives the bound in for $ I_2$. Thinking of , we write the integral in as $$t e^{-|y|^2}|y|^{-1} \int_3^{\ln|y|} \phi(s)
e^{s/2} \,ds,$$ where $$\phi(s) = \frac{ e^{s/2}}{s^{3/2}} \,
e^{-\frac{t^2}{4s}}.$$ Here both the factors are increasing functions of $s$ in $(3,\infty)$, and so is $\phi$. Thus for any $\eta \in (0,1)$, $$\sup_{(3,\ln|y|)} \phi(s) \le \phi(\eta + \ln|y|),$$ and so $$I_2 \lesssim t e^{-|y|^2}|y|^{-1} \phi(\eta +\ln|y|) \int_3^{\ln|y|} e^{s/2} \,ds
\simeq t e^{-|y|^2} \frac{ 1}{(\eta +\ln|y|)^{3/2}} \,
e^{-\frac{t^2}{4(\eta +\ln|y|)}}.$$ Integrating in $\eta,$ we see that $$\label{dd}
I_2 \lesssim \int_{\ln|y|}^{1+\ln|y|} \frac{t}{s^{3/2}}\, e^{-\frac{t^2}{4s}} \,
e^{-|y|^2}\,ds.$$ Because of , this integral is dominated by the one defining $P_t(x,y)$ in . Since $P_t(x,y) \lesssim P_{t/2}(x,y)$, it follows that $ I_2 \lesssim P_{t/2}(x,y)$.
Finally, we estimate $I_1 $ by means of . Since here $1-e^{-2s} \simeq s$, we get $$I_1 \lesssim \int_0^{3} \frac{t}{s^{3/2}} \, e^{-\frac{t^2}{4s}} \,\frac 1{|y| \,s^{1+n/2}}\,
e^{-|y|^2}\,ds.$$ Using with $\gamma = 2+n/2$, we conclude that $$\label{03}
I_1 \lesssim t^{-3-n} \int_0^{3} s^{-1/2} e^{-\frac{(t/2)^2}{4s}} \frac 1{|y|}\,e^{-|y|^2} \,ds.$$ This leads immediately to the bound in . For , we can estimate the right-hand side in by $$t^{-3-n} \frac 1{|y|}\, e^{-\frac{t^2}{48}}\, e^{-|y|^2}
\lesssim t^{-4-n}\, \frac{t}{(\eta + \ln|y|)^{3/2}} \, e^{-\frac{t^2}{4(\eta + \ln|y|)}} \,
e^{-|y|^2}$$ with $\eta \in (0,1)$ as before, since $\ln|y|>12$. As a result, we get a bound for $I_1 $ similar to but with an extra factor $ t^{-4-n}$, and thus also the bound in .
Lemma \[lem3.2\] is proved.
\[prop3.3\] Let $f$ be a measurable function on $\rn$ satisfying . Then for all $i\in\{1,2,\dots,n\}$ and $x\in\rn$ , $$\label{eq3.15}
\partial_{x_i}\partial_tP_{s+t}f(x)
=\int_{\rn} \partial_{x_i}P_s(x,y)\,\partial_tP_tf(y)\,dy, \qquad s,\,t > 0,$$ and $$\label{eq3.16}
\lim_{t\to\infty}\partial_{x_i}P_tf(x)=0.$$
We can assume $|x|< R$ for some $R>0$ and thus apply the estimates from Lemma \[lem3.2\]. First we verify the absolute convergence of the integral in , by showing that $$\begin{aligned}
\int_{\rn}\int_{\rn} |\partial_{x_i}P_s(x,y)| |\partial_{t}P_t(y,z)||f(z)|\,dy\,dz
<\infty.\end{aligned}$$ Lemmas \[lem3.2\] and \[lem3.1\] imply that this integral is, up to a factor $C(n,R)$, no larger than $$\begin{aligned}
\frac{1+s^{-4-n}}{t}\int_{\rn}\int_{\rn} P_{s/2}(x,y) P_{t/2}(y,z)|f(z)|\,dy\,dz
= \frac{1+s^{-4-n}}{t}\int_{\rn} P_{(s+t)/2}(x,z)|f(z)|\,dz < \infty,\end{aligned}$$ where the equality comes from the semigroup property. The last integral here is finite because of ; indeed, [@GHSTV formula (6.4)] says that $P_t(x,y)$ is controlled by $
e^{-|y|^2}/\sqrt{\ln(e+|y|)
}
$, locally uniformly in $x$ and $t$.
Our next step consists in integrating the right-hand side of along intervals in the variables $x_i$ and $t$. We choose two points $x',\,x''\in\rn$ with $|x'|,\,|x''| < R$ which differ only in the $i$:th coordinate, and also two points $t',\,t''>0$. Fubini’s theorem applies because of the above estimates, and we get $$\begin{aligned}
&&\int_{x_i'}^{x_i''}\int_{t'}^{t''} \lf(\int_{\rn}\int_{\rn} \partial_{x_i}P_s(x,y)\partial_{t}p_t(y,z)f(z)\,dy\,dz\r)\,dt\,dx_i\\
&&\quad = \int_{\rn} \int_{\rn} [P_s(x'',y)-P_s(x',y)]\,[P_{t''}(y,z)-P_{t'}(y,z)]f(z)\,dy\,dz\\
&&\quad = P_{s+t''}f(x'')-P_{s+t''}f(x')- P_{s+t'}f(x'')+P_{s+t'}f(x').\end{aligned}$$ From this, we obtain by differentiating with respect to $x_i''$ and $t''$.
Finally, is a direct consequence of and .
Proposition \[prop3.3\] now allows us to apply the method of proof of [@LS Proposition 3.2] and obtain the same estimates as there.
\[cor:3.4\] Let $\az\in(0,1)$ and let $f\in\mathrm{GGLip}_\az$ with norm 1.
1. For all $i\in\{1,2,\dots,n\}$, $t>0$ and $x\in\rn$, $$\begin{aligned}
|\partial_{x_i} P_tf(x)|
\le C t^{\az-1}.\end{aligned}$$
2. For all $t>0$ and $x=(x_1,0,\dots,0)\in\rn$ with $x_1\ge 0$, $$\begin{aligned}
|\partial_{x_1} P_tf(x)|
\le C t^{\az-2}(1+x_1)^{-1}.\end{aligned}$$
Proof of Theorem \[thm1.2\] {#sect4}
===========================
$\rm (i) \Longrightarrow \rm (ii)$: We assume that $f$ satisfies and . According to [@GHSTV Theorem 1.1], $P_tf(x)\to f(x)$ as $t\to 0$ for a.a. $x\in\rn$, and thus we can modify $f$ on a null set so that this convergence holds for all $x$.
Now fix $x,\,y\in\rn$. For all $t>0$, we write $$\begin{aligned}
\label{eq4.1}
|f(x)-f(y)|\le |f(x)-P_{t}f(x)|+\lf| P_{t}f(x)-P_{t}f(y)\r|+
\lf|P_{t}f(y)-f(y)\r|.
\end{aligned}$$ Using Corollary \[cor:3.4\] (i) and arguing as in the verification of [@LS formula (3.7)], we get $$\begin{aligned}
\label{eq4.2}
|f(x)-f(y)|\ls |x-y|^\az.
\end{aligned}$$ To obtain , it is then enough to prove that $$|f(x)-f(y)| \ls d(x,y_x)^{\frac{\az} 2} + |y_x'|^\alpha.$$ By writing $$|f(x)-f(y)|\le |f(x)-f(y_x)|+ |f(y_x)-f(y)|$$ and applying to the last term here, we see that we need only verify that $$|f(x)-f(y_x)| \ls d(x,y_x)^{\frac{\az} 2}.$$ Making a rotation, we can assume that $x=(x_1,0,\dots,0)$ with $x_1\ge0$ and $y_x=(y_1,0,\dots,0)$.
We estimate $|f(x)-f(y_x)|$ as in . Of the three terms we then get, the first and third are controlled by $t^\alpha$. To the second term, we apply Corollary \[cor:3.4\] (ii) and the one-dimensional integral expression for $d$. As a result, $$|f(x)-f(y_x)| \ls t^\alpha + t^{\alpha-2} d(x,y_x),$$ and here we choose $t =d(x,y_x)^{1/2} $. This leads to , and the implication $\rm (i) \Longrightarrow \rm (ii)$ is proved.
$\rm (ii) \Longrightarrow \rm (i)$: Letting $y=0$, we see that implies that $f(x) = O((\ln|x|)^{\alpha/2})$ as $|x| \to \infty$ and thus also . We must verify .
Using the fact that $\int_\rn \partial_tP_t(x,y)\,dy=0$ and Lemma \[lem2.2\], we can write $$\begin{aligned}
|t\partial_tP_tf(x)|
&&= \lf|\int_\rn t \partial_t P_t(x,y)[f(y)-f(x)]\,dy\r|\\
&&\ls \int_\rn [K_1(t,x,y)+K_2(t,x,y)+ \wz{K_3}(t,x,y)+K_4(t,x,y)] |f(y)-f(x)|\,dy \notag.\end{aligned}$$
We thus get four integrals to control by $t^\alpha$. For $\int_{\rn} K_1(t,x,y) |f(y)-f(x)|\,dy $, we can apply the same simple argument as in [@LS end of Section 3], since it uses only the quantity $|x-y|^\az$ in .
The integral involving $K_2(t,x,y)$ can also be estimated as in [@LS], because applies in the support of $K_2(t,x,y)$.
For the integral with $\wz{K_3}(t,x,y)$, we apply the inequality $(a+b)^\kz \le a^\kz + b^\kz$ with $a,b >0$ and $\kz = \az/2 \in (0,1)$ to the expression in and get $$\begin{aligned}
&&\int_\rn \wz{K_3}(t,x,y) |f(y)-f(x)|\,dy\\
&& \quad \ls \min\lf\{1,\,\frac{t}{\sqrt{\ln(e+|x|)}}\r\}\int_{\rn}
\lf(\lf(\ln(1+|x|)\r)^{\frac{\az} 2}+\lf(\ln(1+|y_x|)\r)^{\frac{\az} 2}
+|y_x'|^\alpha\r)
\exp^\ast(-|y|^2)\,dy\notag\end{aligned}$$ The minimum here is no larger than $t^\az/[\ln(e+|x|)]^{\frac{\az} 2}$, which leads immediately to the bound $t^\az$ for the whole expression.
Finally, $$\begin{aligned}
\label{eq4.3}
\qquad\int_\rn K_4(t,x,y)|f(y)-f(x)|\,dy
&& \le \int_{\gfz{x \cdot y>0}{1<|y_x|<|x|/2}}
\frac t{|y_x|} \left(\ln \frac {|x|} {|y_x|}\right)^{ -\frac32} \,
\exp^\ast\left(-\frac {t^2}{\ln \frac {|x|} {|y_x|}}\right) \,\\
&&\quad\times
\exp^\ast(-|y_x^\prime |^2)
\lf(\lf[\ln(1+|x|)-\ln(1+|y_x|)\r]^{\frac{\az} 2}
+|y_x'|^\alpha\r)dy\notag\end{aligned}$$ When $1<|y_x|<|x|/2$, we have $$|\ln(1+|x|)-\ln(1+|y_x|)|=\ln\frac{1+|x|}{1+|y_x|} \simeq \ln\frac{|x|}{|y_x|}.$$ After a rotation, we can assume that $x = (x_1,0,\dots,0)$ with $x_1>0$, so that $y_x = (y_1,0,\dots,0)$ and $y'_x = (0,y') $ and we have $1<y_1<x_1/2$. The right-hand integral in is bounded by a constant times $$\int_1^{x_1/2} \int_{\rr^{n-1}}
\frac t{y_1} \left(\ln \frac {x_1} {y_1}\right)^{ -\frac32} \,
\exp^\ast\left(-\frac {t^2}{\ln \frac {x_1} {y_1}}\right)
\exp^\ast(-|y^\prime |^2) \lf(\lf[\ln\frac{x_1}{y_1}\r]^{\frac{\az} 2}
+|y'|^\alpha\r)\,dy'\,dy_1.$$ Integrating in $y'$ and noticing that $\ln\,(x_1/{y_1})\gs 1$, we can control this double integral by $$\int_1^{x_1/2}
\frac t{y_1} \left(\ln \frac {x_1} {y_1}\right)^{ \frac{\az} 2-\frac32} \,
\exp^\ast\left(-\frac {t^2}{\ln \frac {x_1} {y_1}}\right)
\,dy_1.$$ The transformation of variable $s = t^{-2}(\ln {x_1} - \ln {y_1})$ now gives the desired bound $t^\az$.
Summing up, we have verified and $\rm (i)$. The norm equivalence also follows, and this ends the proof of Theorem \[thm1.2\].
An example of a function in $\mathrm{GGLip}_\az$ {#sect5}
================================================
With $\az\in(0,1)$, we consider the function
$$f(x)=[\ln(e+|x|)]^{\az/2},\qquad x\in\rn.$$ We shall verify that $f$ belongs to $\mathrm{GGLip}_\az$, using Theorem \[thm1.2\].
The estimate $$\label{eq5.1}
|f(x)-f(y)| \ls |x-y|^\az$$ is easy and left to the reader.
To show that $$\label{eq5.2}
|f(x)-f(y)|\ls \lf|\ln(e+|x|)- \mathrm{sgn} \langle x,y\rangle \ln(e+|y_x|)\r|^{\frac{\az} 2}
+|y_x'|^\alpha,$$ write $$|f(x)-f(y)|
\le |f(x)-f(y_x)|+|f(y_x)-f(y)|.$$ The last term here is controlled by $|y_x'|^\alpha$, because of . To the first term on the right, we apply the inequality $|a^\kz-b^\kz|\le |a-b|^\kz, \; \; a,b>0$, with $\kz = \az/2 \in(0,1)$, getting $$|f(x)-f(y_x)|
= \lf|[\ln(e+|x|)]^{\az/2}-[\ln(e+|y_x|)]^{\az/2} \r|
\le \lf|\ln(e+|x|)-\ln(e+|y_x|)\r|^{\frac{\az} 2}.$$ This implies , and it follows that $f\in \mathrm{GGLip}_\az$.
[99]{}
G. Garrigós, S. Harzstein, T. Signes, J. L. Torrea and B. Viviani, Pointwise convergence to intial data of heat and Laplace equations, *Trans. Amer. Math. Soc.* (to appear).
A. E. Gatto, E. Pineda, and W. Urbina, Riesz potentials, Bessel potentials, and fractional derivatives on Besov-Lipschitz spaces for the Gaussian measure, *Recent Advances in Harmonic Analysis and Applications Springer Proceedings in Mathematics $\&$ Statistics*, Volume 25, 2013, pp 105–130.
A. E. Gatto and W. Urbina, On Gaussian Lipschitz spaces and the boundedness of fractional integrals and fractional derivatives on them, arXiv:0911.3962.
L. Liu and P. Sjögren, A characterization of the Gaussian Lipschitz space and sharp estimates for the Ornstein-Uhlenbeck Poisson kernel, arXiv:1401.4288v2.
E. Pineda and W. Urbina, Some results on Gaussian Besov-Lipschitz spaces and Gaussian Triebel-Lizorkin spaces, [*J. Approx. Theory*]{} 161 (2009), 529–564.
E. M. Stein, Singular Integrals and Differentiability Properties of Functions, *Princeton University Press, Princeton, N.J.* 1970. xiv+290 pp.
[Liguang Liu]{}
Department of Mathematics, School of Information\
Renmin University of China\
Beijing 100872\
China
[*E-mail*]{}: `[email protected]`
[Peter Sjögren]{}
Mathematical Sciences, University of Gothenburg\
and\
Mathematical Sciences, Chalmers\
SE-412 96 Göteborg\
Sweden
[*E-mail*]{}: `[email protected]`
| {
"pile_set_name": "ArXiv"
} |
---
author:
- Raphael Bousso
bibliography:
- 'all.bib'
title: 'The Cosmological Constant Problem, Dark Energy, and the Landscape of String Theory'
---
The Cosmological Constant Problem {#sec-ccp}
=================================
A Classical Ambiguity {#sec-classical}
---------------------
In the field equation for General Relativity, $$R_{\mu\nu} - \frac{1}{2} R g_{\mu\nu} + \Lambda g_{\mu\nu} =
8\pi G T_{\mu\nu}~,
\label{eq-gr}$$ there is an ambiguity: the cosmological constant, $\Lambda$, is not fixed by the structure of the theory.[^1] There is no formal reason to set it to zero, and in fact, Einstein famously tuned it to yield an (unstable) static cosmological solution—his “greatest blunder”.
After Hubble’s discovery that the universe is expanding, the cosmological term was widely abandoned. But setting $\Lambda=0$ was never particularly satisfying, even from a classical perspective. The situation is similar to a famous shortcoming of Newtonian gravity: nothing prevents us from equating the gravitational charge with inertial mass, but nothing forces us to do so, either.
A nonzero value of $\Lambda$ introduces a length scale and time scale $$r_\Lambda= c t_\Lambda=\sqrt{3/|\Lambda|}
\label{eq-rlam}$$ into General Relativity. An independent length scale arises from the constants of Nature: the Planck length[^2] $$l_{\rm P} = \sqrt{\frac{G\hbar}{c^3}}
\approx 1.616 \times 10^{-33} {\rm cm}~.$$ It has long been known empirically that $\Lambda$ is very small in Planck units (i.e., that $r_\Lambda$ is large in these natural units). This can be deduced just from the fact that the universe is large compared to the Planck length, and old compared to the Planck time.
First, consider the case of positive $\Lambda$. If no matter is present ($T_{\mu\nu}=0$), then the only isotropic solution to Einstein’s equation is de Sitter space, which exhibits a cosmological horizon of radius $r_\Lambda$ [@HawEll]. A cosmological horizon is the largest observable distance scale, and the presence of matter will only decrease the horizon radius [@GibHaw77a; @Bou00a]. But we observe scales that are large in Planck units ($r\gg 1$). Since $r_\Lambda$ must be even larger, Eq. (\[eq-rlam\]) implies that the cosmological constant is small.
Negative $\Lambda$ causes the universe to recollapse independently of spatial curvature, on a timescale $t_{\Lambda}$ [@Edw72]. Thus, the large age of the universe (in Planck units) implies that $(-\Lambda)$ is small. Summarizing the above arguments, one finds $$-3t^{-2}\lesssim\Lambda\lesssim 3r^{-2}~, \label{eq-trlam}$$ where $t$ and $r$ are any time scale and any distance scale that have been observed. We can see out to distances of billions of light years, so $r>10^{60}$; and stars are billions of years old, so $t>10^{60}$. With these data, known for many decades, Eq. (\[eq-trlam\]) implies roughly that $$|\Lambda|\lesssim 3\times 10^{-120}~.
\label{eq-small}$$ Thus, in Planck units, $\Lambda$ is very small indeed.
This result makes it tempting to set $\Lambda=0$ in the Einstein equation; and at the level of the classical gravity theory, we are free to do so. However, in Eq. (\[eq-gr\]), the $\Lambda$-term is not the only term proportional to the metric. Another, much more problematic contribution enters through the stress tensor on the right hand side.
Quantum Contributions to $\Lambda$ {#sec-quantum}
----------------------------------
In quantum field theory, the vacuum is highly nontrivial.[^3] In the Standard Model, the vacuum is responsible for physical phenomena such as confinement and the Higgs mechanism. Like any physical object, the vacuum will have an energy density. Lorentz invariance requires that the corresponding energy-momentum-stress tensor be proportional to the metric, $$\langle T_{\mu\nu} \rangle = -\rho_{\rm vacuum} g_{\mu\nu} ~.
\label{eq-lrho}$$ This is confirmed by direct calculation. (See any introductory textbook on quantum field theory, such as Ref. [@Zee].) The form of the stress tensor ensures that the vacuum looks the same to all observers independently of orientation or velocity. This property (and not, for example, vanishing energy density) is what distinguishes the vacuum from other objects such as a table.
Though it appears on the right hand side of Einstein’s equation, vacuum energy has the form of a cosmological constant, and one might as well absorb it and redefine $\Lambda$ via $$\Lambda=\Lambda_{\rm Einstein} +8\pi\rho_{\rm vacuum}~.$$ Equivalently, one may absorb the “bare” cosmological constant appearing in Einstein’s equation, $\Lambda_{\rm Einstein}$, into the energy density of the vacuum, defining $$\rho_\Lambda\equiv \rho_{\rm vacuum}+\frac{\Lambda_{\rm Einstein}}{8\pi}~.$$
![Perturbative and nonperturbative contributions to vacuum energy. (a) Virtual particle-antiparticle pairs, the zero-point fluctuations of the quantum fields (b) Effective scalar field potentials, such as the potential for the Higgs field shown here schematically. Before electroweak symmetry breaking in the early universe the vacuum energy was about 56 orders of magnitude greater than todays value (dashed line).[]{data-label="fig-loop"}](fig-loop){width=".8\textwidth"}
Eqs. (\[eq-rlam\]), (\[eq-trlam\]), and (\[eq-small\]) apply to the total cosmological constant, and can be restated as an empirical bound on the total energy density of the vacuum: $$|\rho_\Lambda|\lesssim 10^{-121}~.$$ But in the Standard Model, the energy of the vacuum receives many contributions much larger than this bound. Their value depends on the energy scale up to which we trust the theory. It is enormous even with a conservative cutoff.
This would be true already in free field theory. Like a harmonic oscillator in the ground state, every mode of every free field contributes a zero-point energy to the energy density of the vacuum. In a path integral description, this energy arises from virtual particle-antiparticle pairs, or “loops” (Fig. \[fig-loop\]a). For example, consider the electron, which is well understood at least up to energies of order $M=$ 100 GeV [@Pol06]. Dimensional analysis implies that electron loops up to this cutoff contribute of order $(100$ GeV$)^4$ to the vacuum energy, or $10^{-68}$ in Planck units.
Similar contributions are expected from other fields and from interactions. The real cutoff is probably of order the supersymmetry breaking scale, giving at least $(1$ TeV$)^4\approx 10^{-64}$. It may be as high as the Planck scale, which would yield $|\rho_\Lambda|$ of order unity.[^4] Thus, quantum field theory predicts multiple perturbative contributions to $|\rho_\Lambda|$. Each contribution is some 60 to 120 orders of magnitude larger than the experimental bound, Eq. (\[eq-small\]).
Additional contributions come from the effective potentials of scalar fields, such as the potential giving rise to symmetry breaking in the electroweak theory (Fig. \[fig-loop\]b). The vacuum energy of the symmetric and the broken phase differ by approximately $(200$ GeV$)^4\approx 10^{-67}$. Other symmetry breaking mechanisms at higher or lower energy, such as chiral symmetry breaking of QCD with $(300$ MeV$)^4\approx 10^{-79}$, will also contribute. There is no reason why the total vacuum energy should be small in the symmetric phase, and even less so in the broken phase that the universe is in now.
I have exhibited various known contributions to the vacuum energy. They are uncorrelated with one another and with the (unknown) bare cosmological constant appearing in Einstein’s equation, $\Lambda_{\rm
Einstein}$. Each contribution is dozens of orders of magnitude larger than the empirical bound today, Eq. (\[eq-small\]). In particular, the radiative correction terms from quantum fields are expected to be at least of order $10^{-64}$. They come with different signs, but it would seem overwhelmingly unlikely for such large terms to cancel to better than a part in $10^{120}$, in the present era.
This is the cosmological constant problem: [*Why is the vacuum energy today so small?*]{} It represents a serious crisis in physics: a discrepancy between theory and experiment, of 60 to 120 orders of magnitude. What makes this problem hard is that it arises from two otherwise extremely successful theories—the Standard Model and General Relativity—in a regime where both theories have been reliably and precisely tested and hence cannot be dramatically modified.
The Cosmological Constant {#sec-duck}
=========================
In exhibiting the cosmological constant problem, I made use only of a rather crude, and old, upper bound on the magnitude of the cosmological constant. The precise value of $\Lambda$ is irrelevant as far as the cosmological constant [*problem*]{} is concerned: we have known for several decades that $\Lambda$ is certainly much smaller than typical contributions to the vacuum energy that can be estimated from the Standard Model of particle physics. In this section, I will discuss the observed value and its implications.
Observed Value of $\Lambda$
---------------------------
The actual value of $\Lambda$ was first determined in 1998 from the apparent luminosity of distant supernovae [@Rie98; @Per98]. Their dimness indicates that the expansion of the universe has recently begun to accelerate, consistent with a positive cosmological constant $$\rho_\Lambda= (1.35\pm 0.15) \times 10^{-123}~,
\label{eq-duck}$$ and inconsistent with $\rho_\Lambda =0$. The quoted value and error bars are recent (WMAP7 + BAO + $H_0$ [@WMAP7]) and thus significantly improved relative to the original discovery.
Cross-checks have corroborated this conclusion. For example, the above value of $\rho_\Lambda$ also explains the observed spatial flatness of the universe [@WMAP7], which cannot be accounted for by baryonic and dark matter alone. And surveys of the history of structure formation in the universe [@Rei09] reveal a recent disruption of hierarchical clustering consistent with accelerated expansion driven by the cosmological constant of Eq. (\[eq-duck\]).
Why Dark Energy is Vacuum Energy
--------------------------------
The observed vacuum energy, Eq. (\[eq-duck\]), is sometimes referred to as “dark energy”. This choice of words is meant to be inclusive of other possible interpretations of the data, in which $\Lambda=0$. Dark energy might be a form of scalar matter (quintessence) which mimics a fixed cosmological constant closely enough to be compatible with observation, but retains some time-dependence that could in principle be discovered if it lurks just beyond current limits. Another frequently considered possibility is that General Relativity is modified at distances comparable to the size of the visible universe, so as to mimic a positive cosmological constant even though $\Lambda=0$. In both cases, model parameters can be adjusted to lead to predictions for future experiments that differ from those of a fixed cosmological constant.
Consideration of these theoretical possibilities, however, is at best premature. It conflicts with a basic tenet of science: adopt the simplest interpretation of the data, and complicate your model only if forced to by further observation.
Scenarios like quintessence or modified gravity are uncalled for by data and solve no theoretical problem.[^5] In particular, they do not address the cosmological constant problem. But such models contain adjustable parameters in addition to $\Lambda$. Therefore, they are less predictive than the standard $\Lambda$CDM model. Worse, in phenomenologically viable models, these additional parameters must be chosen small and fine-tuned in order to evade existing constraints.[^6] Again, such tunings are strictly [*in addition*]{} to the tuning of the the cosmological constant, which must be set to an unnaturally small or zero value in any case.
Therefore, dynamical dark energy should not be considered on the same footing with a pure cosmological constant. The discovery of any deviation from a cosmological constant in future experiments is highly unlikely, as is the discovery of a modification to General Relativity on large scales.
A frequent misconception that appears to underlie the consideration of “alternatives” to $\Lambda$ is the notion that vacuum energy is somehow optional. The idea is that the cosmological constant problem only arises if we “assume” that vacuum energy exists in the first place. (This flawed argument is found in surprisingly prominent places [@DETF].) It would be wonderful indeed if we could solve the cosmological constant problem with a single stroke, by declaring that vacuum energy just does not exist and setting $\Lambda$ to zero.
But in fact, we know that vacuum energy exists in Nature. We can manipulate the amount of vacuum energy in bounded regions, in Casimir-type experiments. And if $\Lambda$ had turned out to be unobservably small today, we would still know that it was large and positive in the early universe before electroweak symmetry breaking, according to the Standard Model of particle physics.[^7] More generally, the notion that the vacuum has energy is inseparable from the experimental success of the Standard Model as a local quantum field theory [@Pol06].
Contributions to $\Lambda$ from Standard Model fields are large, so the most straightforward theoretical estimate of its magnitude fails. But just because $\Lambda$ should be much larger than the observed value does not imply that it must be zero. In fact, no known extension or modification of the Standard Model predicts that $\Lambda=0$ without violently conflicting with other observations (such as the facts that the universe is not empty, and that supersymmetry, if it exists, is broken).
Thus, the cosmological constant problem is present either way, whether we imagine that $\Lambda$ is small (which is consistent with data) or that $\Lambda=0$ (which is not, unless further considerable complications are introduced). Dark energy is experimentally indistinguishable from vacuum energy, and definitely distinct from any other previously observed form of matter. The only reasonable conclusion is that dark energy is vacuum energy, and that its density is given by Eq. (\[eq-duck\]).
The Coincidence Problem {#sec-coincidence}
-----------------------
The observed value of $\Lambda$ does raise an interesting question, usually referred to as the coincidence problem or “why now” problem. Vacuum energy, or anything behaving like it (which includes all options still allowed by current data) does not redshift like matter. In the past, vacuum energy was negligible, and in the far future, matter will be very dilute and vacuum energy will dominate completely. The two can be comparable only in a particular epoch. It is intriguing that this is the same epoch in which we are making the observation.
Note that this apparent coincidence involves us, the observers, in its very definition. This constrains possible explanations (other than those involving an actual coincidence). In the following section, I will outline a framework which can solve both the coincidence problem and the (far more severe) cosmological constant problem of Sec. \[sec-quantum\].
The Landscape of String Theory and the Multiverse {#sec-theory}
=================================================
The string landscape is the only theoretical framework I am aware of that can explain why $\Lambda$ is small without conflicting with other data.[^8] (It is worth stressing, however, that the ideas I am about to discuss are still speculative, unlike those of the previous two sections.) The way in which string theory addresses the cosmological constant problem can be summarized as follows:
- Fundamentally, space is nine-dimensional. There are many distinct ways (perhaps $10^{500}$) of turning nine-dimensional space into three-dimensional space by compactifying six dimensions.[^9]
- Distinct compactifications correspond to different three-dimensional metastable vacua with different amounts of vacuum energy. In a small fraction of vacua, the cosmological constant will be accidentally small.
- All vacua are dynamically produced as large, widely separated regions in spacetime
- Regions with $\Lambda\sim 1$ contain at most a few bits of information and thus no complex structures of any kind. Therefore, observers find themselves in regions with $\Lambda\ll 1$.
The Landscape of String Theory {#sec-landscape}
------------------------------
String theory is naturally formulated in nine or ten spatial dimensions [@GSW; @Polchinski]. This does not contradict observation but implies that all but three of these dimensions are (effectively) compact and small, so that they would not have been observed in high-energy experiments. I will discuss the case of six compact extra dimensions for definiteness.
Simple examples of six-dimensional compact manifolds include the six-sphere and the six-dimensional torus. A much larger class of manifolds are the Calabi-Yau spaces, which have a number of useful properties and have been extensively studied. They are topologically complex, with hundreds of distinct cycles of various dimensions. Cycles are higher-dimensional analogues of the handles of a torus. A rubber band that wraps a handle cannot be removed, or wrapped around a different handle, without ripping it apart. A more pertinent example are electrical field lines, which can wrap a one-cycle (such as one of the cycles on a two-dimensional torus).
String theory contains a certain set of nonperturbative objects known as $D$-branes, which act as sources of $D+2$ flux. For example, a zero-brane is a pointlike object and sources a Maxwell field, much like an electron would. Higher-dimensional objects such as membranes act as sources of higher-dimensional analogues of the Maxwell field. Unlike in the Standard Model, however, the values of $D$ for which $D$-branes exist, their energy density, and their charge are all determined by consistency requirements. They are set by the string scale and are not adjustible parameters.
$D$-branes and their associated fluxes can wrap topological cycles the same way that rubber bands and electric field lines can wrap the handles of a torus. In string theory, the shape and size of the compact extra dimensions is determined by (among other things) the fluxes that wrap around the various topological cycles. The geometry of spacetime is dynamical and governed by equations that limit to Einstein’s equations in the appropriate limit. The presence of matter will deform the compact manifold correspondingly; in particular, one expects that each cycle can at most support a few units of flux before gravitational backreaction causes it to pinch off (changing the topology of the compact manifold) or grow to infinite size (“decompactify”).
Based on these arguments, we may suppose that there are on the order of 500 cycles, and that each can support between 0 and 9 units of flux. Then there are $10^{500}$ different, distinct choices for the matter content, shape, and size of the extra dimensions. This argument is a vast oversimplification, but it helps clarify how numbers like $10^{500}$ arise: by exponentiation of the number of topological cycles in a typical six-dimensional compact manifold.[^10]
A useful way of picturing the set of three-dimensional vacua of string theory is as a potential function in a 500-dimensional discrete parameter space. (Of course, as far as actual pictures go, two parameters will have to suffice, as in a real landscape.) Each metastable configuration of fluxes corresponds to a local minimum in the landscape. In any one-dimensional cross-section of the parameter space, there will only be a handful of minima, but overall the number of minima can be of order $10^{500}$.
The Spectrum of $\Lambda$ {#sec-spectrum}
-------------------------
Each vacuum has distinct matter and field content at low energies, determined by the matter content of the extra dimensions. (Pictorially, the field spectrum corresponds to the details of each valley’s shape near the minimum.) In particular, the energy of each vacuum is essentially a random variable that receives positive and negative contributions from all particle species. If we select one vacuum completely at random, the arguments of Sec. \[sec-quantum\] tell us that its cosmological constant will probably be large, presumably of order unity in Planck units (Fig. \[fig-spec3\])—as if we had thrown a dart at the interval $(-1,1)$, with an accuracy not much better than $\pm 1$.
![The spectrum of the cosmological constant (vacuum energy, dark energy) in the string landscape (schematic). Each blue line represents one three-dimensional vacuum. With $10^{500}$ vacua, the spectrum will be very dense, and many vacua will have values of $\Lambda$ compatible with observation (red/shaded region).[]{data-label="fig-spec3"}](spec3){width=".2\textwidth"}
But this is true for every vacuum, so the overall spectrum of $\Lambda$ will be quite dense, with an average spacing of order $10^{-500}$. This means that there will be a small fraction ($10^{-123}$) but a large number ($10^{377}$, in this example) of vacua with cosmological constant $|\Lambda|\lesssim 10^{-123}$. Given enough darts, even a poor player will eventually hit the bullseye.
This is progress: at least, the theory contains vacua whose cosmological constant is compatible with observation. But why is the universe in such a special, rare vacuum? Did the universe start out in this particular valley of the landscape at the big bang, and if so, why? In fact, there is no need to assume that initial conditions selected for a vacuum with small cosmological constant. As we shall now see, such vacua are dynamically produced during cosmological evolution.
de Sitter Expansion and Vacuum Decay
------------------------------------
Suppose that the universe began in some vacuum with $\Lambda>0$. Since about half of all vacua have positive energy, this is not a strong restriction. We will not assume that the initial vacuum energy is particularly small; it may be of order one in Planck units.
The universe evolves as de Sitter space, with metric $$ds^2=-dt^2+e^{2Ht}(dr^2+r^2 d\Omega_2^2)~,
\label{eq-desitter}$$ where the Hubble constant $H$ is given by $(\Lambda/3)^{1/2}$, and $d\Omega_2^2$ denotes the metric on the unit two-sphere. This is an exponentially expanding homogeneous and isotropic cosmology. In the following, it is not important that the universe looks globally like Eq. (\[eq-desitter\]). It suffices to have a finite initial region larger than one horizon volume, of proper radius $e^{Ht_0} r> H^{-1}$.
Classically, this evolution would continue eternally, and no other vacua would ever come into existence anywhere in the universe. This is because the vacuum itself is set by topological configurations of fluxes in the extra dimensions, which cannot change by classical evolution. Quantum mechanically, however, it is possible for fluxes to change by discrete amounts. This happens by a process completely analogous to the Schwinger process.
The Schwinger process is the spontaneous pair production of electrons and positrons in a strong electric field between two capacitor plates. It can be treated as a tunneling process in the semi-classical approximation. The two particles appear at a distance at which the part of the field that their charges cancel out compensates for their total rest mass, so that energy is conserved. Then the particles move apart with constant acceleration, driven by the remaining electric field, until they hit the plates (or in the case where the field lines wrap a topological circle, until they hit each other). The final result is that the electric flux has been lowered by a discrete amount, corresponding to removing one unit of electric charge from each capacitor plate.
Similarly, the amount of flux in the six extra dimensions can change as a result of Schwinger-like processes, whereby branes of appropriate dimension are spontaneously nucleated. (The Schwinger process itself is recovered in the case of zero-branes, i.e., charged point particles.) Again, this is a nonperturbative tunneling effect. Its rate is suppressed by the exponential of the brane action and is generically exponentially small.
Let us now give a description of this process from the 3+1 dimensional viewpoint. The effect of the six extra dimensions is to provide an effective potential landscape. Each minimum corresponds to a metastable vacuum with three large spatial dimensions. (Recall that the hundreds of dimensions of the landscape itself correspond to the topological cycles of the extra dimensions, not to actual spatial directions.)
The decay of a unit of flux, in this picture, corresponds to a transition from a higher to a lower-energy minimum in the potential landscape of string theory.[^11] This transition does not happen simultaneously everywhere in three-dimensional space, because that process would have infinite action. Rather, a bubble of the new vacuum appears spontaneously, as in a first-order phase transition. Like in the Schwinger process, the initial size of the bubble is controlled by energy conservation. The bubble wall is a domain wall that interpolates between two vacua in the effective potential. The gradient and potential energy in the domain wall are compensated by the vacuum energy difference in the enclosed volume.
The bubble expands at constant acceleration. As it moves outward, it converts the old, higher energy parent vacuum into a new, lower-energy vacuum. The vacuum energy difference pays not only for the ever-expanding domain wall but can also lead to the production of matter and radiation inside the new vacuum.
The symmetries of a first-order phase transition in a relativistic theory dictate that the region inside the bubble is an open (i.e., negatively curved) Friedmann-Robertson-Walker universe. In particular, time slices of constant density are infinitely large, even though the bubble starts out at finite size. (This is possible because the choice of time variable in which we see the bubble expand is different from, and indeed inconsistent with, a choice in which constant time corresponds to hypersurfaces of constant density within the bubble.) For this reason, the interior of the bubble is sometimes referred to as a “universe”, “pocket universe”, or “bubble universe”, even though it does not constitute all of the global spacetime.
Eternal Inflation
-----------------
We now turn to a crucial aspect of the decay of a metastable vacuum with positive energy: despite the decay and the expansion of the daughter bubble, the parent vacuum persists indefinitely. This effect is known as eternal inflation [@GutWei83; @Lin86a].
The volume occupied by the parent vacuum expands exponentially at a rate set by its own Hubble scale $3H=3(3/\Lambda)^{1/2}$. Some volume is lost to decay, at a rate $\Gamma$ per unit Hubble volume. As long as $\Gamma\ll 3H$ (which is generic due to the exponentially suppressed nature of vacuum decay), the exponential expansion wins out, and the parent vacuum region grows on average.
The fact that the new vacuum expands after it first appears does not affect this result, since different regions in de Sitter space are shielded from one another by cosmological event horizons. A straightforward analysis of light propagation in the metric of Eq. (\[eq-desitter\]) shows that any observer (represented by a timelike geodesic) is surrounded by a horizon of radius $H^-1$. The observer cannot receive any signals from any point $p$ beyond this horizon, by causality, no matter how long they wait. A bubble of a new vacuum that forms at $p$ cannot expand faster than the speed of light (though it does expand practically at that speed). Therefore it can never reach an observer who is initially more than a distance $H^{-1}$ from $p$ at the time of bubble nucleation.
Because the parent vacuum continues to grow in volume, it will decay not once but infinitely many times. Infinitely many bubble universes will be spawned; yet, the overall volume of parent vacuum will continue to increase at a rate set by $3H-\Gamma\approx 3H$. If the parent vacuum has multiple decay channels, then each decay type will be realized infinitely many times. For example, in the string landscape we expect that a de Sitter vacuum can decay to any one of its hundreds of immediate neighbor vacua in the high-dimensional potential landscape. All of these vacua will actually be produced as bubble universes, in exponentially distant regions, over and over.
The Multiverse
--------------
Let us now turn our attention to one of the daughter universes. It is useful to distinguish three cases, according to the sign of its cosmological constant. First, suppose that its vacuum energy is positive and that the vacuum is sufficiently long-lived (greater than about $t_\Lambda$). In this case, the daughter universe will enter a phase of exponential de Sitter expansion, beginning at a time of order $t_\Lambda$ after its nucleation. It will give rise to eternal inflation in its own right, decaying in infinitely many places and producing daughter universes, while persisting globally.
![Conformal diagram of an eternally expanding multiverse (schematic). Light travels at 45 degrees. Different colors/shades represent different vacua in the string landscape. Bubble universes have a triangular shape in this diagram. They are bounded by domain walls whose expansion is so rapid that they look like future light-cones. Event horizons shield different regions from one another: a hypothetical observer who survives multiple vacuum decays (black vertical line) would still only be able to probe a finite region in the infinite multiverse (black diamond).[]{data-label="fig-global"}](fig-global){width=".8\textwidth"}
Thus, the entire landscape of string theory can in principle be populated. All vacua are produced dynamically, in widely separated regions of spacetime, and each is produced infinitely many times. This can be illustrated in a conformal diagram (or “Penrose diagram”), which rescales the spacetime metric to render it finite but preserves causal relations (Fig. \[fig-global\]). By convention, light-rays propagate at 45 degrees. Bubbles look like future light-cones because they expand nearly at the speed of light. Bubble universes that form at late times are shown small due to the rescaling, even though their physical properties are independent of the time of their production. As a result of eternal inflation, the future boundary of the diagram has a fractal structure.
Vacua with nonpositive cosmological constant are “terminal”. They do not give rise to eternal inflation. If $\Lambda<0$, then the bubble universe begins to contract and collapses in a big crunch on a timescale of order $t_\Lambda$ [@CDL]. The spacelike singularity does not reach outside the bubble universe with $\Lambda<0$; it does not affect global eternal inflation.
One expects that the case $\Lambda=0$ arises only in vacua with unbroken supersymmetry. They are completely stable and do not end in a crunch. In the conformal diagram, they correspond to the “hat regions” near the future boundary (not shown in Fig. \[fig-global\]).
Why Observers are Located in Regions With $|\Lambda|\ll 1$
----------------------------------------------------------
I have argued that the string landscape contains vacua with very small cosmological constant, such as ours. Moreover, such vacua will be dynamically produced by inflation, starting from generic initial conditions. But the bubble universes with $|\Lambda\ll 1$, such as ours, are surely very atypical regions in the large multiverse. Typical regions (by almost any conceivable definition of “typical”) would have cosmological constant of order one in Planck units, since almost all vacua have this property. Why, then, do we find ourselves in one of the rare locations with $\Lambda \ll 1$?
Before addressing this question, it is worth noting that the same question could not be asked in a theory that failed to contain vacua with $\Lambda\ll 1$, or that failed to produce such vacua as spacetime regions. But in a theory that dynamically produces highly variable environments in different locations, it is important to understand correlations between environmental properties and the location of observers. What is typically observed depends on where one is observing, so these correlations will affect the predictions of the theory.
In Sec. \[sec-ccp\], I discussed that the cosmological constant sets a largest observable length or time scale, of order $|\Lambda^{-1/2}|$. A more precise result can be stated in terms of the maximum area on the past light-cone of an arbitrary point (event) $p$ in a universe with nonzero cosmological constant [@BouFre10a]. If $\Lambda>0$, the past light-cone of any point $p$ has maximum area of order $\Lambda^{-1}$; if $\Lambda<0$, it has maximum area of order $|\Lambda|^{-1}$ (if the universe is spatially flat), or $\Lambda^{-2}$ (if the universe is open).
The maximum area on the past light-cone of $p$, in units of the Planck length squared, is an upper bound on the entropy in the causal past of $p$: $$S\lesssim A$$ This follows from the covariant entropy bound [@CEB1; @RMP]. It implies that regions with $\Lambda\sim 1$ do not contain more than a few bits of information in any causally connected region. Whatever observers are made of, they presumably require more than one or two particles.
This means that observers can only be located in regions with $|\Lambda|\ll 1$. Because of cosmological horizons, they will not typically be able to see other regions. Though typical regions have $\Lambda=1$, observations are made in regions with $|\Lambda|\ll 1$.
Predicted Value of $\Lambda$
----------------------------
The argument shows only that $|\Lambda|\ll 1$ is a prediction of the string landscape; it does not explain why we see the particular value $\Lambda\sim 10^{-123}$. In order to make this, or any other quantitative prediction, we would need to begin by regulating the infinities of eternal inflation. This is known as the “measure problem”, and it has little to do with the string landscape.
The measure problem arises in any theory that gives rise to eternal inflation. For this, one long-lived metastable de Sitter vacuum is enough. We appear to live in such a vacuum, so the measure problem needs attention independently of the number of other vacua in the theory. A discussion of this problem and of current approaches to its solution would go beyond the scope of the present paper. The reader is referred to Ref. [@BouFre10d] and references therein; here we quote only the main result of this paper (see also Ref. [@BouHar07; @BouHar10]).
![The vertical bar indicates the observed amount of vacuum energy (“dark energy”). The solid line shows the prediction from the causal patch measure applied to the landscape of string theory, with the central 1$\sigma$ region indicated. This plot is from Ref. [@BouHar07]. The agreement remains good independently of any assumptions about the nature of the observers. The only relevant input parameter is the time when the observers emerge, $t_{\rm obs}\approx 13.7$ Gyr.[]{data-label="fig-stars_PS_hopkins"}](stars_PS_hopkins){width=".8\textwidth"}
Consider a class of observers that live at the time $t_{\rm obs}$ after the nucleation of their bubble universe. Restricting attention to positive values of $\Lambda$, the causal patch measure [@Bou06] predicts that such observers will find a cosmological constant $$\Lambda\sim t_{\rm obs}^{-2}~.$$ Using the observed value for the age of the universe, $t_{\rm obs}\approx 13.7$ Gyr, this result is in excellent agreement with observed value for the cosmological constant (see Fig. \[fig-stars\_PS\_hopkins\]).
The successful prediction (or postdiction, in this case) of $\Lambda$ obtains independently of the nature of the observers. For example, it applies to observers that do not require galaxies and even in vacua with very different low-energy physics. In addition to the cosmological constant problem, it also addresses the coincidence problem discussed in Sec. \[sec-coincidence\], since it predicts that observers should find themselves at the onset of vacuum domination, $t_\Lambda \sim t_{\rm obs}$. Thus the prediction is more robust, and quantitatively more successful, than the seminal arguments of Weinberg [@Wei87] and other early arguments requiring specific assumptions about observers [@DavUnw82; @Ban85; @BarTip; @Efs95]. (The dashed line in Fig. \[fig-stars\_PS\_hopkins\] shows the prediction from the assumption that observers require galaxies, with an earlier measure developed in Ref. [@MarSha97].
There are currently no fully satisfactory measures for regions with nonpositive cosmological constant [@BouFre10c]. This remains a major outstanding challenge. More broadly, it will be important to establish a solid theoretical basis for understanding both the landscape of string theory and the measure problem of eternal inflation.
Connecting with Standard Cosmology
----------------------------------
How is the picture of a multiverse compatible with the one universe we see? The multiverse is quite irregular, with different vacua in different places. This appears to conflict with the observed homogeneity and isotropy of the visible universe. We have not detected any other pocket universes. As far as we can see, the vacuum seems to be the same, with the same particles, forces, and coupling constants. Another concern is the claimed metastability of vacua. If vacua can decay, how come our own vacuum is still around after billions of years?
In fact, all of these observations are generic predictions of the model, and all arise from the fact that vacuum decay is an exponentially suppressed tunneling effect. This has three important consequences:
- Individual pocket universes, including ours, can have very long lifetimes easily exceeding 10 Gyr [@BP].
- When a bubble of new vacuum does form, it will be highly symmetric [@Col77]. The symmetry of the decay process translates into the prediction that each pocket universe is a negatively curved, spatially homogeneous and isotropic universe [@CDL]. (The spatial curvature radius can be made unobservably large, as usual, by a period of slow-roll inflation at early times in our own pocket universe.)
- Our parent vacuum need not produce many bubbles that collide with ours. For such collisions to be visible, they would have to occur in our past light-cone, and the expected number of collisions can be $\ll 1$ for natural parameters.
Thus, the fact that we observe only one vacuum is not in contradiction with the string landscape.
However, this does not mean that other vacua will never be observed. We would have to be somewhat lucky to observe a smoking gun signal of bubble collisions in the sky [@FreKle09; @CzeKle10; @KleLev11; @GobKle12]; for a review, see Ref. [@Kle11]. But it is a possibility, so the computation of its signature in the CMB for future searches such as PLANCK is of great interest [@FeeJoh10a; @FeeJoh10b; @JohPei11; @McEFee12].
Slow-roll inflation tends to wipe out signals from any era preceding it by stretching them to superhorizon scales. If slow-roll inflation occurred after the formation of our bubble (as seems plausible), and if it lasted significantly longer than the 60 e-foldings necessary for explaining the observed flatness, then any imprints of bubble collisions or of our parent vacuum will have been stretched to superhorizon scales.
The decay of our own parent vacuum plays the role of what we used to call the big bang. The vacuum energy of the parent vacuum is converted in part to the energy of the expanding domain wall bubble that separates our pocket universe from the parent vacuum. But some of this energy can be dissipated later, inside our pocket universe. It can drive a period of slow-roll inflation followed by the production of radiation and matter.
The decay of our parent vacuum will have taken place in an empty de Sitter environment, so all matter and radiation in our vacuum must come from the vacuum energy released in the decay. In order to connect with standard cosmology, the energy density of radiation produced must be at least sufficient for nucleosynthesis. This constrains the vacuum energy of our parent vacuum: $$\Lambda_{\rm parent}\gg 10^{-88}~.
\label{eq-parent}$$
This constraint is very powerful. Historically, it has ruled out one-dimensional potential landscapes such as the Abbott [@Abb85] or Brown-Teitelboim [@BT1; @BT2] models, which were explicitly invented for the purpose of solving the cosmological constant problem. In such models, neighboring vacua have nearly identical vacuum energy, $\Delta\Lambda<10^{-123}$. Each decay lowers $\Lambda$ by an amount less than the observed value, so a very dense spectrum of $\Lambda$ is scanned over time. This eventually produces a universe with $\Lambda$ as small as the observed value. But because Eq. (\[eq-parent\]) is not satisfied, the universe is predicted to be empty, in conflict with observation. One could invent one-dimensional landscapes in which the vacuum energy is random, but in natural models decay paths would end in terminal vacua with $\Lambda<0$ before reaching one of the rare vacua with $\Lambda\ll 1$.
In the string landscape, neighboring vacua typically have vastly different vacuum energy, with $\Lambda$ differing by as much as O(1) in Planck units (Sec. \[sec-spectrum\]). Thus, matter and radiation can be produced in the decay of our parent vacuum. Because the landscape is high-dimensional, there are many decay paths around terminal vacua. Thus, all de Sitter vacua in the landscape can be cosmologically produced by eternal inflation from generic initial conditions.
It is interesting that string theory, which was not invented for the purpose of solving the cosmological constant problem, thus evades a longstanding obstruction.
[^1]: This paper aims at a level that would be accessible to a graduate student. It is based on colloquia given at Caltech, MIT, and the University of Michigan, Ann Arbor, and on a lecture presented at [*Subnuclear Physics: Past, Present and Future*]{}, Pontificial Academy of Sciences, Vatican (October 2011, to appear in the proceedings). In parts, I closely follow Refs. [@Bou06b; @TASI07].
[^2]: Here $G$ denotes Newton’s constant and $c$ is the speed of light. In this paper Planck units are used unless other units are given explicitly. For example, $t_{\rm P}=l_{\rm P} /c \approx .539 \times 10^{-43}
{\rm s}$ and $M_{\rm P} = 2.177\times 10^{-5} {\rm g}$.
[^3]: Further details can be found in Weinberg’s classic review [@Wei89]. Among more recent reviews, I recommend Polchinski’s concise discussion of the cosmological constant problem [@Pol06], which I follow in parts of this subsection.
[^4]: Recall that Planck units are used throughout. $\rho_\Lambda=1$ would correspond to a density of $10^{94}$ g/cm$^3$.
[^5]: Some models have been claimed to address the coincidence problem described in Sec. \[sec-coincidence\] below. Aside from unsolved technical problems [@Car00], what would be the point of addressing the (relatively vague) coincidence problem with a model that ignores the logically prior and far more severe cosmological constant problem (Sec. \[sec-quantum\])?
[^6]: For example, quintessence models require exceedingly flat scalar field potentials which must be fine-tuned against radiative corrections, and their interaction with other matter must be tuned small in order to be compatible with observational limits on a long-range fifth force [@Car98; @Car00]. More natural models [@HalNom05] have become difficult to reconcile with observational constraints.
[^7]: The theory of electroweak symmetry breaking is supported by overwhelming experimental evidence (chiefly, the $W$ and $Z$ bosons, and soon perhaps the Higgs). It allows us to compute that $\Lambda\sim (200$ GeV$)^4$ at sufficiently high temperatures, when electroweak symmetry is unbroken [@Pol06]. Aside from the early universe, small regions with unbroken symmetry could be created in the laboratory, at least in principle.
[^8]: For alternative classes of approaches to the cosmological constant problem, and the obstructions they face, see Refs. [@Pol06; @TASI07].
[^9]: Amazingly, this idea was anticipated by Sakharov [@Sak84] before string theory became widely known.
[^10]: For a more detailed nontechnical version of this argument, see Ref. [@BouPol04]. Despite early results that the number of compactifications could be large [@LerLus87], the significance of this possibility was obscured by the unsolved problem of moduli stabilization and supersymmetry breaking [@DouKac06]; see, however, Ref. [@Sch98]. The argument that string theory contains sufficiently many metastable vacua to solve the cosmological constant problem, and that vacua with $\Lambda\sim 10^{-123}$ are cosmologically produced and reheated was presented in Ref. [@BP]. An explicit construction of a large class of nonsupersymmetric flux vacua was first proposed in Ref. [@KKLT]. (Constructions in noncritical string theory were proposed earlier [@Sil01; @MalSil02].) More advanced counting methods [@DenDou04b] bear out the quantitative estimates of Ref. [@BP] for the number of flux vacua. See Ref. [@DouKac06] for a review of flux vacua and further references.
[^11]: The following description of vacuum decay is a straightforward application of seminal results of Coleman for a one-dimensional potential with two vacua [@Col77; @CDL]. More complicated decay channels can arise in multidimensional potentials [@BroDah10]; they do not affect the conclusions presented here.
| {
"pile_set_name": "ArXiv"
} |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.