arXiv reader

Fixed points of an infinite dimensional operator related to Gibbs measures

U. R. Olimov and U. A. Rozikov

arXiv
2206.13822v1 · PDF · 28 June 2022
Categories
math-ph, math.MP
DOI
10.1134/S0040577923020125
Licence
CC BY 4.0 The site shows the text.

Abstract

We describe fixed points of an infinite dimensional non-linear operator related to a hard core (HC) model with a countable set \mathbb{N} of spin values on the Cayley tree. This operator is defined by a countable set of parameters \lambda_{i}>0, a_{ij}\in \{0,1\}, i,j \in \mathbb{N}. We find a sufficient condition on these parameters under which the operator has unique fixed point. When this condition is not satisfied then we show that the operator may have up to five fixed points. Also, we prove that every fixed point generates a normalisable boundary law and therefore defines a Gibbs measure for the given HC-model.

Items 13

Lemma

Lemma 1

3Y8Q

If \lambda=(\lambda_1,\lambda_2,\dots)\in l^{1}_{+} then A_{k} is an invariant with respect to operator (?), F:l^1_+\to l^1_+, i.e., F(A_{k})\subset A_{k}.

Proof of Lemma 1

For any x\in \ell^1_{+} from (?) we get

0<x'_{i}=\lambda_{i}\left(\frac{1+\sum_{j=1}^{\infty}a_{ij}x_j}{1+\|x\|}\right)^k\leq\lambda_{i}\left(\frac{1+\|x\|}{1+\|x\|}\right)^k=\lambda_{i}.

Consequently,

\|x'\|=\sum_{i=1}^{\infty}x'_i\leq\sum_{i=1}^{\infty}\lambda_i=\|\lambda\|.

Now we find a lower bound for \|x'\|. For x\in A_k, we have from (?) that

x'_{1}=\lambda_{1},
x'_{i}=\lambda_{i}\left(\frac{1+\sum_{j=1}^{\infty}a_{ij}x_j}{1+\|x\|}\right)^k\geq\frac{\lambda_{i}}{(1+\|x\|)^k}\geq\frac{\lambda_{i}}{(1+\|\lambda\|)^k}, \ \ i\geq 2.

Therefore,

\|x'\|\geq\lambda_{1}+\sum_{j=2}^{\infty}\frac{\lambda_{i}}{(1+\|\lambda\|)^k}=\lambda_1+\frac{\|\lambda\|-\lambda_1}{(1+\|\lambda\|)^k}.
Lemma

Lemma 2

4GBF

For any \lambda\in\ell^1_{+} with \|\lambda\|<\hat{\lambda}, there exists \kappa(\lambda,k)\in (0, 1) such that

\|F(x)-F(y)\|<\kappa\|x-y\|, \ \ \mbox{for all} \ \ x, y\in A_{k}.

Thus F:A_k\to A_k is a contraction.

Proof of Lemma 2

Recall that x_1'=\lambda_1. Take any (\lambda_{1},x_{2},x_{3},\dots,x_{n},\dots) and (\lambda_{1},y_{2},y_{3},\dots,y_{n},\dots). Then we have

F_{i}(x)-F_{i}(y)=\lambda_{i}\left(\left(\frac{1+\sum_{j=1}^{\infty}a_{ij}x_j}{1+\|x\|}\right)^k-\left(\frac{1+\sum_{j=1}^{\infty}a_{ij}y_j}{1+\|y\|}\right)^k\right)
=\lambda_{i}\left(\frac{1+\sum_{j=1}^{\infty}a_{ij}x_j}{1+\|x\|}-\frac{1+\sum_{j=1}^{\infty}a_{ij}y_j}{1+\|y\|}\right)U(x,y),

where

U(x,y):= \sum_{p=0}^{k-1}\left(\frac{1+\sum_{j=1}^{\infty}a_{ij}x_j}{1+\|x\|}\right)^{k-p-1}\cdot\left(\frac{1+\sum_{j=1}^{\infty}a_{ij}y_j}{1+\|y\|}\right)^{p}.

Consequently,

F_{i}(x)-F_{i}(y)= U(x,y)\lambda_{i} \frac{(1+\sum_{j=1}^{\infty}a_{ij}x_j)(1+\|y\|)-(1+\sum_{j=1}^{\infty}a_{ij}y_j)(1+\|x\|)}{(1+\|x\|)(1+\|y\|)}
=U(x,y)\lambda_{i}\frac{(1+\|x\|)\sum_{j=1}^{\infty}a_{ij}(x_j-y_j)+(\|y\|-\|x\|)\left(1+\sum_{j=1}^{\infty}a_{ij}x_j\right)}{(1+\|x\|)(1+\|y\|)}.

Now we use the following inequalities:

1) If x,y\in\ell^1_{+} then \|y\|-\|x\|=\sum_{j=1}^{\infty}(y_{j}-x_{j}) and \|y-x\|=\sum_{j=1}^{\infty}|y_{j}-x_{j}|, therefore,

-\|y-x\|\leq\|y\|-\|x\|\leq\|y-x\|.

2) For any x,y\in\ell^1_{+}, by a_{ij}\in{\{0,1\}} we get

\sum_{j=1}^{\infty}a_{ij}x_{j}\leq\|x\|, \ \ \sum_{j=1}^{\infty}a_{ij}(x_{j}-y_{j})\leq\|x-y\|.

3) We note also that

\|y\|-\|x\|+\sum_{j=1}^{\infty}a_{ij}(x_j-y_j)
=\sum_{j=1}^{\infty}(y_j-x_j)-\sum_{j=1}^{\infty}a_{ij}(y_j-x_j)=-\sum_{j=1}^{\infty}(1-a_{ij})(x_j-y_j).

Thus we have

-\|x-y\|\leq\sum_{j=1}^{\infty}(1-a_{ij})(x_j-y_j)\leq\|x-y\|.

Since a_{ij}\in\{0, 1\} and x_i>0 we have U(x,y)\leq k for any x,y. Using the above-mentioned inequalities we obtain:

|F_{i}(x)-F_{i}(y)|\leq \frac{k\lambda_{i}(1+2\|x\|)}{(1+\|x\|)(1+\|y\|)}\|x-y\|.
\|F(x)-F(y)\|=\sum_{j=1}^{\infty}|F_{i}(x)-F_{i}(y)| \leq\frac{k\|\lambda\|(1+2\|x\|)}{(1+\|x\|)(1+\|y\|)}\|x-y\|.

Denote

K(x,y)=\frac{k\|\lambda\|(1+2\|x\|)}{(1+\|x\|)(1+\|y\|)}=\frac{k\|\lambda\|(1+2\|x\|)}{1+\|x\|}\frac{1}{1+\|y\|}.

To find upper bound of K(x,y), we introduce

g(t)=\frac{1+2t}{1+t}, \ \ h(t)=\frac{1}{1+t}, \ \ t>0.

We find maximal values of these functions for t which satisfies:

\lambda_1+\frac{\|\lambda\|-\lambda_1}{(1+\|\lambda\|)^2}\leq t \leq\|\lambda\|.

It is clear that g(t) is an increasing function with maximal value

g_{\max}(\|\lambda\|)=\frac{1+2\|\lambda\|}{1+\|\lambda\|}.

Moreover, function h(t) is a decreasing function and

h_{\max}(\lambda_1+\frac{\|\lambda\|-\lambda_1}{(1+\|\lambda\|)^2})=\frac{1}{1+\lambda_1+\frac{\|\lambda\|-\lambda_1}{(1+\|\lambda\|)^2}}<\frac{1}{1+\frac{\|\lambda\|}{(1+\|\lambda\|)^2}}.

Using these values we obtain

K(x,y)<k\|\lambda\|\frac{(1+2\|\lambda\|)}{(1+\|\lambda\|)}\frac{1}{1+\frac{\|\lambda\|}{(1+\|\lambda\|)^2}}=k\frac{2\|\lambda\|^3+3\|\lambda\|^2+\|\lambda\|}{\|\lambda\|^2+2\|\lambda\|+2}.

Now we want to find \lambda such that

\frac{2k\|\lambda\|^3+3k\|\lambda\|^2+k\|\lambda\|}{\|\lambda\|^2+2\|\lambda\|+2}<1.

That is

\varphi(\|\lambda\|):=2k\|\lambda\|^3+(3k-1)\|\lambda\|^2+(k-2)\|\lambda\|-2<0.

Note that \varphi(0)=-2<0 and \varphi(2)>0. Therefore \varphi has at least one zero in (0,2). According to Descartes’ theorem, \varphi(\|\lambda\|)=0 has unique positive solution, since the signs of its coefficients change only once. By Cardano formula we obtain explicit form of the unique root, which is \hat \lambda defined above.

Thus \varphi(\|\lambda\|)<0 if \|\lambda\|<\hat\lambda.

Theorem

Theorem 1

5055

If \|\lambda\|<\hat{\lambda} then the operator (?) has unique fixed point z^{*} and for any initial point z^{(0)}\in A_{k} we have \lim\limits_{n \to \infty}F^n(z^{(0)})=z^{*}.

Theorem

Theorem 2

YMKF

Let \mathcal N be the number of fixed points of the operator (?). Then

\mathcal N=\left\{\begin{array}{lllllllll} 1, \ \ \mbox{if} \ \ (L,\lambda_1)\in A_1\setminus (B_1\cup B_2)\\[2mm] 2, \ \ \mbox{if} \ \ (L,\lambda_1)\in A_2\setminus B_2\\[2mm] 3, \ \ \mbox{if} \ \ (L,\lambda_1)\in B_1\\[2mm] 3, \ \ \mbox{if} \ \ (L,\lambda_1)\in B_2\setminus (A_2\cup A_3)\\[2mm] 3, \ \ \mbox{if} \ \ (L,\lambda_1)\in A_3\setminus B_2\\[2mm] 3, \ \ \mbox{if} \ \ (L,\lambda_1)\in A_1\cap B_1\\[2mm] 3, \ \ \mbox{if} \ \ (L,\lambda_1)\in A_1\cap B_2\\[2mm] 4, \ \ \mbox{if} \ \ (L,\lambda_1)\in A_2\cap B_2\\[2mm] 5, \ \ \mbox{if} \ \ (L,\lambda_1)\in A_3\cap B_2 \end{array}\right.
Definition

Definition 1

4POE

(see [Z1] and Chapter 12 of [Ge]) A family of vectors l=\left\{l_{xy}\right\}_{\langle x, y) \in L} with l_ {xy}= \left\{l_{xy}(i): i \in \mathbb{N}\right\} \in(0, \infty)^{\mathbb{N}} is called the boundary law for the Hamiltonian (?) if

1) for each \langle x, y\rangle \in L there exists a constant c_{x y}>0 such that the consistency equation

l_{xy}(i)=c_{xy} \prod_{z \in \partial x \backslash\{y\}} \sum_{j \in \mathbb{N}} \lambda_i a_{ij}\lambda_{j} l_{zx}(j)

holds for any i \in \mathbb{N}, where \partial x is the set of nearest neighbors of x.

2) The boundary law l is said to be normalisable if and only if

\sum_{i \in \mathbb{N}}\left(\prod_{z \in \partial x} \sum_{j \in \mathbb{N}}\lambda_i a_{ij}\lambda_{j} l_{zx} (j)\right)<\infty

for all x \in V.

Uses Set-up
Theorem

Theorem 3

GD54

[Z1] For any Gibbsian specification \gamma with associated family of transfer matrices (Q_b)_{b \in L} we have

  1. Each normalisable boundary law (l_{xy})_{x,y} for (Q_b)_{b \in L} defines a unique Gibbs measure \mu (corresponding to \gamma) via the equation given for any connected set \Lambda \subset V

    \mu(\sigma_{\Lambda \cup \partial \Lambda}=\omega_{\Lambda \cup \partial \Lambda}) = (Z_\Lambda)^{-1} \prod_{y \in \partial \Lambda} l_{y y_\Lambda}(\omega(y)) \prod_{b \cap \Lambda \neq \emptyset} Q_b(\omega_b),

    where for any y \in \partial \Lambda, y_\Lambda denotes the unique nearest-neighbor of y in \Lambda.

  2. Conversely, every Gibbs measure \mu admits a representation of the form (?) in terms of a normalisable boundary law (unique up to a constant positive factor).

Uses Set-up
Lemma

Lemma 3

UE7A

If \lambda\in l^{1}_{+} then any solution of the form (?) is normalisable.

Uses Set-up
Proof of Lemma 3

Recall Q(i,j):=\lambda_ia_{ij}\lambda_j. The normalisablity of boundary laws can be reduced (see [HRo]) to show that

\sum_{i\in\mathbb{N}}\sum_{j\in\mathbb{N}}z_{x, i}Q(i,j)z_{y, j}<\infty, \ \ \forall \langle x, y\rangle \in L.

Now we check this condition for solution (?):

\sum_{i\in\mathbb{N}}\sum_{j\in\mathbb{N}}z_{x, i}Q(i,j)z_{y, j}=\sum_{i\in\mathbb{N}}\sum_{j\in\mathbb{N}}\lambda_i a_{ij}\lambda_{j}z_{x,i}z_{y,j}\leq \sum_{i\in\mathbb{N}}\lambda_i z_{x,i} \sum_{j\in\mathbb{N}}\lambda_{j}z_{y,j}.

Since the solution is independent on the vertices of the Cayley tree, for the RHS of (Proof) we have:

\sum_{i\in\mathbb{N}}\lambda_i z_{x,i} \sum_{j\in\mathbb{N}}\lambda_{j}z_{y,j}=\left(\sum_{i\in\mathbb{N}}\lambda_i z_{i}\right)^2.

Therefore, the lemma follows from the following estimate (in the case of (?)):

\sum_{i\in\mathbb{N}}\lambda_i z_{i}=\lambda_1+ \sum_{n\in \mathbb{N}}\lambda_{2n}^2\left(\dfrac{1+M_2}{1+\lambda_1+M_1+M_2}\right)^2
+\sum_{n\in \mathbb{N}}\lambda_{2n+1}^2\left(\dfrac{1+M_1}{1+\lambda_1+M_1+M_2}\right)^2 < \lambda_1+\sum_{i\in\mathbb{N}_1} \lambda^{2}_i<+\infty.

Because, if \lambda\in l^{1}_{+} then \sum_{i\in\mathbb{N}_1}\lambda_i^2<\infty.

Theorem

Theorem 4

F1BG

Let \mathcal{N}_G be the number of translation invariant Gibbs measures for Hamiltonian (?), corresponding to graph G defined by (?), then

\mathcal N_G=\left\{\begin{array}{lllllllll} 1, \ \ \mbox{if} \ \ (L,\lambda_1)\in A_1\setminus (B_1\cup B_2)\\[2mm] 2, \ \ \mbox{if} \ \ (L,\lambda_1)\in A_2\setminus B_2\\[2mm] 3, \ \ \mbox{if} \ \ (L,\lambda_1)\in B_1\\[2mm] 3, \ \ \mbox{if} \ \ (L,\lambda_1)\in B_2\setminus (A_2\cup A_3)\\[2mm] 3, \ \ \mbox{if} \ \ (L,\lambda_1)\in A_3\setminus B_2\\[2mm] 3, \ \ \mbox{if} \ \ (L,\lambda_1)\in A_1\cap B_1\\[2mm] 3, \ \ \mbox{if} \ \ (L,\lambda_1)\in A_1\cap B_2\\[2mm] 4, \ \ \mbox{if} \ \ (L,\lambda_1)\in A_2\cap B_2\\[2mm] 5, \ \ \mbox{if} \ \ (L,\lambda_1)\in A_3\cap B_2 \end{array}\right.
Uses Set-up
Remark

Remark 1

XFT7

Lemma 3 can be proved for the case of uniqueness of the fixed point, for operator F, mentioned in the previous sections, therefore the corresponding Hamiltonian (?) has unique translation invariant Gibbs measure.

Source and licence

The text on this page is from arXiv:2206.13822v1 by U. R. Olimov and U. A. Rozikov, under the licence CC BY 4.0. The corpus made these changes: it read the items out of the LaTeX source, it expanded the macros of the paper, it put each proof after its statement, and it converted the text to HTML. The formulas are the TeX of the paper, and KaTeX renders them. A formula or a reference can look different from the PDF. The page is published under CC BY 4.0.

The record of each item is in analysis/objects/2206/2206.13822.jsonl.