Datasets:
tensorkelechi
/
arxiv-cstext

Dataset card Data Studio Files Community
1
text
stringlengths
0
89.3k
KAt
i1n1Mx0tN 28
Invoking Lemma 2 checking that 27 implies 28 reduces to
check whether there exists βtRn
such that
KAt
i1n1Mβ
tNtN
that is to say with Ninvertible
β
tKAt
i1n1MN10tN 29
To conclude the proof we show that 29 is equivalent to
check the positivity of a solution of a discrete time dynamics
Indeed as the invertibility of A1andA2implies that Mis
invertible we have
β
t1KAt1
i1n1MN1
KAt
i1n1MN1NM1Ai1n1MN1
β
tNM1Ai1n1MN1
Lβt
This theorem states that to verify the nonnegativity of a
single solution to the auxiliary linear discretetime system
described in 26 is enough to ensure that any solution to 14
switches no more than ntimes This property is different from
the classical nonnegativity problem where any solution to a
system must remain nonnegative componentwise for any
nonnegative initial condition In this classical setting it is
well known that for a discretetime system to be positive a
necessary and sufficient condition is that the entries of the
dynamical matrix have to be positive 19 This does not
correspond to our problem and we will see in Section VI
that the entries of the dynamical matrix Lare not necessarily
nonnegative so that we cannot invoke the results from 19
in general This means new results are needed to ensure the
nonnegativity of system 26 for a given initial condition this
is addressed in the next section
V N ONNEGATIVITY ANALYSIS
We derive in this section conditions under which the solu
tion to 26 initialized with a given β00is nonnegative The
results are based on a function analysis and allows deriving
numerically tractable conditions
A Nonnegativity theorem
The next theorem provides easytotest conditions to ensure
the satisfaction of item ii of Theorem 4 The result of
Theorem 5 is based on the necessary condition given in
Proposition 2 which is given below
Theorem 5 Consider system 26 with a given β00and
assume that σAi1n1λ1 λ n with λ1
λn0 Then Lis diagonalizable and we set v1 v ntheeigenvectors of L and γ1 γ nRsuch that β0n
i1γivi
Let us finally set ρiγiviRnandµilnλiλ1with
i1 n and assume that ρ10 Then the solution βtis
nonnegative for any tNif for every ℓ1 n one of
the conditions of Proposition 2 is satisfied with zitheℓth
component of ρi
Proof According to the definition of Lin 26 the eigenval
ues of Lare also eigenvalues of Ai1n1 They are distinct
and the neigenvectors viare independent The solution
of the linear discretetime system 26 is a weighted linear
combination of modal solutions
βtn
i1γiviλt
i
To conclude the proof of this theorem we have to find
conditions on ρ1 ρ nsuch that
βtn
i1γiviλt
iλt
1ρ1 λt
nρn0tN30
We recall that ρiRnand inequality 30 has to be understood
componentwise ie βtℓ0for every ℓ1 n
According to the definition of µi for every i2 n
we have 0µ2 µn Hence the problem is equivalent
with finding a condition on ρ1 ρ nRnsuch that
ρ1n
i2eµitρitN 31
Note that 31 holds if we have
ρ1n
i2eµisρisR
We conclude the proof using Proposition 2
Proposition 2 LetnN0µ2µnand
z1 z nR Ifz10z1 zn0and one of the
following conditions holds
iz2 z n0
iiz2 z n0
iii there exists j2 nsuch that
z1n
i2
ij1µi
µjzieµissR 32