• Register
PhysicsOverflow is a next-generation academic platform for physicists and astronomers, including a community peer review system and a postgraduate-level discussion forum analogous to MathOverflow.

Welcome to PhysicsOverflow! PhysicsOverflow is an open platform for community peer review and graduate-level Physics discussion.

Please help promote PhysicsOverflow ads elsewhere if you like it.


New printer friendly PO pages!

Migration to Bielefeld University was successful!

Please vote for this year's PhysicsOverflow ads!

Please do help out in categorising submissions. Submit a paper to PhysicsOverflow!

... see more

Tools for paper authors

Submit paper
Claim Paper Authorship

Tools for SE users

Search User
Reclaim SE Account
Request Account Merger
Nativise imported posts
Claim post (deleted users)
Import SE post

Users whose questions have been imported from Physics Stack Exchange, Theoretical Physics Stack Exchange, or any other Stack Exchange site are kindly requested to reclaim their account and not to register as a new user.

Public \(\beta\) tools

Report a bug with a feature
Request a new functionality
404 page design
Send feedback


(propose a free ad)

Site Statistics

204 submissions , 162 unreviewed
5,024 questions , 2,178 unanswered
5,345 answers , 22,682 comments
1,470 users with positive rep
815 active unimported users
More ...

  Numerical solution of Parisi equations for the SK spin-glass model

+ 3 like - 0 dislike

I am trying to reproduce the numerical solution of the so-called Parisi equations for the Sherrington-Kirkpatrick (SK) model. There are at least two main methods to solve these equations, and unfortunately I am confused by both.

First, the equations themselves. They are:

$$ q(x) = \int_{-\infty}^{\infty} dy \, P(x,y) m(x,y)^2 \,,$$

$$ \dot{m}(x,y) = - \frac{\dot{q}(x)}{2} \left[ m''(x,y) + 2 \beta x m(x,y) m'(x,y) \right] \,,$$

$$ \dot{P}(x,y) = \frac{\dot{q}(x)}{2} \left[ P''(x,y) - 2 \beta x \left[ m(x,y) P(x,y) \right]' \right] \,,$$

and the initial conditions are:

$$ m(1,y) = \tanh(\beta y), \qquad P(0,y) = \delta(y) \,.$$

Primes denote $y$ derivatives and a dot denotes an $x$ derivative. The domain is $x \in [0,1]$, $y \in [-\infty, \infty]$. In any numerical attempt, the range of $y$ is approximated as finite, $y \in [-y_{\text{max}}, y_{\text{max}}]$. 

In each of the ways to solve these equations that I know about, one proceeds by first supplying a guess for $q(x)$, then solving the $m$ equation, then the $P$ equation. Then $q$ is updated according to the first equation. These steps are iterated until convergence, which typically takes hundreds of cycles .

The two main methods I know of are an integral kernel method by Nemoto and a pseudo-spectral method by Crisanti and Rizzo. In the Nemoto method kernels are used, and these require an initial guess for not only $q(x)$ but also $m(x,y), P(x,y)$. I have no idea what a good guess would be, and I have tried many and failed to get convergence. Presumably the guess should agree with the initial conditions for $m, P$ for $x=1, 0$ respectively. The Crisanti and Rizzo method takes a Fourier transform in the $y$-direction, and then numerically integrates in the $x$-direction using an Adams-Bashforth scheme. Implicit in their approach is that they are setting $P, m$ to have Dirichlet boundary conditions in the $y$-direction, although this is inconsistent with the initial condition for $m$. I have tried to reproduce their method using using both Mathematica's NDSolve, and by adapting this pseudo-spectral code by Trefethen for the KdV equation, but I cannot get either code to converge. From the shape of the curves it is clear I am not handling the boundary conditions properly.

I would greatly appreciate any guidance, as well as any references that might shed light or offer a more easy-to-understand numerical method to solve these equations.

asked Jan 15, 2018 in Computational Physics by gshartnett (15 points) [ revision history ]
recategorized Jan 15, 2018 by Dilaton

1 Answer

+ 0 like - 0 dislike

I have a partial answer and a question. 

I am also trying to compute a numerical solution to the Parisi formula for the SK model (with the hope of simulating Montanari's algorithm; is this your motivation as well?). While I can roughly solve for any fixed $\beta$ (see below if you're curious for how to do that), I do not know how to take a limit as $\beta$ goes to infinity as one must do to find the ground states. This seems numerically difficult. I was wondering if you had any thoughts on this / have succeeded in the simulation since posting the question?

Anyway, I think a simple way to solve for the Parisi measure at a fixed temperature is to break the problem into two parts: for any given measure, one can use the Matlab "pdepe" solver (applicable since the PDE is hyperbolic and just in 1 space variable). This allows us to evaluate the Parisi functional for any measure, so then we can use stochastic optimization algorithms, e.g. simulated annealing / genetic algorithms, to search over the space of feasible measures.

answered Apr 18, 2019 by DA [ no revision ]

Your answer

Please use answers only to (at least partly) answer questions. To comment, discuss, or ask for clarification, leave a comment instead.
To mask links under text, please type your text, highlight it, and click the "link" button. You can then enter your link URL.
Please consult the FAQ for as to how to format your post.
This is the answer box; if you want to write a comment instead, please use the 'add comment' button.
Live preview (may slow down editor)   Preview
Your name to display (optional):
Privacy: Your email address will only be used for sending these notifications.
Anti-spam verification:
If you are a human please identify the position of the character covered by the symbol $\varnothing$ in the following word:
Then drag the red bullet below over the corresponding character of our banner. When you drop it there, the bullet changes to green (on slow internet connections after a few seconds).
To avoid this verification in future, please log in or register.

user contributions licensed under cc by-sa 3.0 with attribution required

Your rights