Consider the Ising Model in the $d$-dimensional discrete torus with side lengh $L$, denoted by $\mathbb{T}_L $, with nearest neighbors interaction (with interaction parameter $J$, no magnetic field, and inverse of temperature $\beta$

The Gaussian Domination Bound/ Infrared Bound states that for every non-zero $ p \in \mathbb{T}_L^*:=(\frac{2\pi}{L} \mathbb{Z}^d) \cap (-\pi,\pi]$, we have:

$$ \sum_{x \in \mathbb{T}_L} e^{ip \cdot x} \langle \sigma_0 \sigma_x \rangle_{L,\beta} \le \frac{1}{2\beta E(p)}, $$

where $\langle \cdot \rangle^{0}_{L,\beta} $ denotes the expected value of a random variable with respect to finite volume Gibbs Measure in $\mathbb{T}_L$; $i$ is the imaginary unit; $ip \cdot x$ denotes the inner product of $ip$ and $x$; and $$ E(p):= J \sum_{x: \|x\|_1=1} (1-e^{ip\cdot x}). $$

Multiple articles point out that the following statement is equivalent to that bound, $\forall (v_x) \in \mathbb{C}^{\mathbb{T}_L}$, we have $$ \sum_{x,y \in \mathbb{T}_L: \|x-y\|_1=1 }v_x \bar{v}_y \langle \sigma_x \sigma_y \rangle_{L,\beta} \le \frac{1}{2\beta}\sum_{x,y \in \mathbb{T}_L: \|x-y\|_1=1 } v_x \bar{v}_y G_L(x,y) + \frac{1}{L^d} \big| \sum_{x \in \mathbb{T}_L} v_x \big|^2 \sum_{x \in \mathbb{T}_L} \langle \sigma_0 \sigma_x \rangle_{L,\beta}, $$ where $$ G_L(x,y) := \sum_{p \in \mathbb{T}_L^* \backslash \{0\} } \frac{1}{L^d} \frac{e^{ip \cdot (x-y)}}{E(p)}. $$

Well, I tried to use the usual proof as in the Chapter 10 of the book Statistical Mechanics of Lattice Systems:a Concrete Mathematical Introduction. Following the proof in page 465 until the moment they choose the values of $\alpha_i$, I believe that this equivalence will come choosing new values for $\alpha_i$, but to this moment, I don't know what to choose.

  • $\begingroup$ Are you asking about the proof of the infrared bound (which is what is on p. 465) or about the equivalence between the two formulations? $\endgroup$
    – Ben
    May 6, 2017 at 20:31
  • $\begingroup$ My question is about the equivalence. I understand the proof in p. 465 and I was trying to somehow adapt it to the second statement. $\endgroup$
    – Kernel
    May 7, 2017 at 13:21

1 Answer 1


So what you actually prove is that $$ \langle (\sigma (-\Delta h))^2 \rangle \le \frac{1}{2\beta} (-h\Delta h) $$ When you take $h=e^{ikx}$, your first statement pops up. However, for any $v$ with $\sum_x v_x =0$, we can always find $h$ such that $-\Delta h =v$ (the kernel of $-\Delta$ is all constant functions). Plug in $h$, and your second statement follows for "neutral" $v$. If $\sum_x v_x\ne0$, we can always define $u=v-1/L^d\sum _x v$ and use $u$ instead.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.