1
$\begingroup$

I am reading the book: Statistical Mechanics of Neural Networks by Huang, Haiping.

I cannot figure out how to get the following step in (2.14), page 10, chapter 2, spin Glass Models and Cavity Method.

$$ Z^{new} = Z^{old} \sum_{\{\sigma_i|i\in\partial a\}} \left[\exp\left(\beta J_a \prod_{i\in \partial a} \sigma_i\right) \prod_{i \in \partial a} \left(\frac{1+ \sigma_i m_{i->a}}{2}\right)\right] \\= Z^{old} \cosh(\beta J_a) \left[1 + \tanh(\beta J_a) \prod_{i \in \partial a} m_{i->a}\right] $$

I don't know how many items in total in this summation and how to get the second part from the first part.

I knew the relation between $e^x$ and $\cosh(x)$ or $\tanh(x)$:

$$\cosh (x) = \frac{e^{x} + e^{-x}}{2}$$

Any help would be appreciated.

$\endgroup$

1 Answer 1

1
$\begingroup$

One way to get this is to group configurations of the neighbors according to their product.

Let's call the neighbors $\sigma_1, \sigma_2, \ldots, \sigma_k$ and the messages $m_1, m_2, \ldots, m_k$. $$\begin{aligned} \sum_{\sigma_1, \sigma_2, \ldots, \sigma_k} e^{\beta J_a \prod_{i=1}^k\sigma_i} \prod_{i=1}^k \frac{1+\sigma_i m_i}{2} &= \sum_{\sigma=\pm1} e^{\beta J_a \sigma} \sum_{\substack{\sigma_1, \sigma_2, \ldots, \sigma_k\\\prod_{i=1}^k\sigma_i=\sigma}} \prod_{i=1}^k \frac{1+\sigma_i m_i}{2} \end{aligned}$$

The inner sum is

$$\begin{aligned} \sum_{\substack{\sigma_1, \sigma_2, \ldots, \sigma_k\\\prod_{i=1}^k\sigma_i=\sigma}} \prod_{i=1}^k \frac{1+\sigma_i m_i}{2} &=\sum_{\sigma_1, \sigma_2, \ldots, \sigma_{k-1}} \prod_{i=1}^{k-1} \frac{1+\sigma_i m_i}{2} \frac{1+\sigma\left(\prod_{i=1}^{k-1}\sigma_i\right) m_k}{2}\\ &=\sum_{\sigma_1, \sigma_2, \ldots, \sigma_{k-1}} \left[\left(\prod_{i=1}^{k-1} \frac{1+\sigma_i m_i}{2}\right)+\frac{\sigma m_k}{2}\left(\prod_{i=1}^{k-1}\frac{1+\sigma_i m_i}{2}\sigma_i \right)\right]\\ &=\frac12 \prod_{i=1}^{k-1}\underbrace{\sum_{\sigma_i}\frac{1+\sigma_i m_i}{2}}_{1} + \frac12\sigma m_k \prod_{i=1}^{k-1} \underbrace{\sum_{\sigma_i}\frac{\sigma_i + m_i}{2}}_{m_i}\\ &= \frac{1 + \sigma\prod_{i=1}^k m_i}{2} \end{aligned}$$ Bringing everything together $$\begin{aligned} \sum_{\sigma = \pm 1} e^{\beta J_a \sigma}\frac{1 +\sigma \prod_{i=1}^k m_i} {2} = \cosh(\beta J_a) + \sinh\left(\beta J_a\right)\prod_{i=1}^k m_i = \cosh(\beta J_a) \left[1 + \tanh\left(\beta J_a\right)\prod_{i=1}^k m_i\right] \end{aligned}$$

EDIT: there is a simpler way. Using the identity $$e^x = \cosh(x)+\sinh(x)$$ on the term with $J_a$, one gets $$\begin{aligned} \sum_{\sigma_1, \sigma_2, \ldots, \sigma_k} e^{\beta J_a \prod_{i=1}^k\sigma_i} \prod_{i=1}^k \frac{1+\sigma_i m_i}{2} &= \sum_{\sigma_1, \sigma_2, \ldots, \sigma_k} \left[\cosh(\beta J_a)+\sinh(\beta J_a)\prod_{i=1}^k\sigma_i\right] \prod_{i=1}^k \frac{1+\sigma_i m_i}{2}\\ %&=\cosh(\beta J_a)\sum_{\sigma_1, \sigma_2, \ldots, \sigma_k}\prod_{i=1}^k \frac{1+\sigma_i m_i}{2} + \sinh(\beta J_a)\sum_{\sigma_1, \sigma_2, \ldots, \sigma_k}\prod_{i=1}^k\frac{\sigma_i+ m_i}{2}\\ &=\prod_{i=1}^k \sum_{\sigma_i} \left[\cosh(\beta J_a)+\sinh(\beta J_a)\sigma_i\right]\frac{1+\sigma_i m_i}{2}\\ &= \prod_{i=1}^k\sum_{\sigma_i}\frac{\cosh(\beta J_a)+m_i\sinh(\beta J_a)}{2}\\ &=\cosh(\beta J_a) \left[1 + \tanh\left(\beta J_a\right)\prod_{i=1}^k m_i\right] \end{aligned}$$

$\endgroup$

Your Answer

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

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