2,443
个编辑
更改
跳到导航
跳到搜索
==Naïve Mean-Field Approximation==
For a general coupling distribution, the macroscopic observables associated with the Boltzmann distribution are usually hard to solve, thus we usually need approximations. The simplest mean-field approximation is called ''Naïve Mean-Field'' or ''Variational Mean-Field''. It also has a strong relation to the variational inference approaches, which will be shown later.
First of all we see that once a configuration <math>\textstyle\underline\sigma^t</math> is given, a spin ''i'' experiences a ''local field'' as
:<math> h_i=\sum_jJ_{ij}\sigma^t_j.</math>
As in the Glauber dynamics, the expectation of value that spin ''i'' in the "next time" when we flip spin ''i'' is given by
:<math> m_i^{t+1}=\tanh(\beta\sum_iJ_{ij}\sigma_j^t).</math>
Then a simple approximation says that let us forget the time indices, and replace the spin values by its mean, this results to the ''Naïve Mean-Field'' approximation:
:<math> m_i=\tanh(\beta\sum_iJ_{ij}m_j).</math>
==Curie-Weiss model and the phase transition==
The Curie-Weiss model is the most simple model for ferromagnets, where <math>\textstyle J_{ij}=J/n</math>, a constant, and system is fully connected.
Without the external field (which is considered in the whole lecture) In the model there is a symmetry that for any configuration, flipping each spin to its opposite value does not alter the energy of the configuration, i.e.
:<math>E(\underline\sigma)=E(-\underline\sigma).</math>
It means that the magnetization of each spin should be always zero, as marginal distribution for spin ''i'' has to satisfy <math>\textstyle p_i(\sigma_i)=-p_i(\sigma_i).</math>
This is indeed the case at high temperature where <math>\textstyle \beta</math> is small. This is called the ''Paramagnetic'' state. However at a low temperature we can not see this state, because of
the spontaneous symmetry breaking, at which point the paramagnetic state becomes unstable. We can show the phenomenon using the Naïve Mean-Field Approximation.
Here the couplings are identical, the mean-field equation for the magnetization simply becomes
:<math>m=\tanh(\beta Jm).</math>
The fixed-points of above equation can be shown in the following figure where <math>\beta J=0.2, 0.5, 1, 2, 20.</math>
[[File:CurrieWeiss.png|400px]]
[[category:旧词条迁移]]
Mean-field approximation and the Curie-Weiss model (查看源代码)
2020年10月15日 (四) 16:19的版本
、 2020年10月15日 (四) 16:19创建页面，内容为“ ==Naïve Mean-Field Approximation== For a general coupling distribution, the macroscopic observables associated with the Boltzmann distribution are usually hard to…”
==Naïve Mean-Field Approximation==
For a general coupling distribution, the macroscopic observables associated with the Boltzmann distribution are usually hard to solve, thus we usually need approximations. The simplest mean-field approximation is called ''Naïve Mean-Field'' or ''Variational Mean-Field''. It also has a strong relation to the variational inference approaches, which will be shown later.
First of all we see that once a configuration <math>\textstyle\underline\sigma^t</math> is given, a spin ''i'' experiences a ''local field'' as
:<math> h_i=\sum_jJ_{ij}\sigma^t_j.</math>
As in the Glauber dynamics, the expectation of value that spin ''i'' in the "next time" when we flip spin ''i'' is given by
:<math> m_i^{t+1}=\tanh(\beta\sum_iJ_{ij}\sigma_j^t).</math>
Then a simple approximation says that let us forget the time indices, and replace the spin values by its mean, this results to the ''Naïve Mean-Field'' approximation:
:<math> m_i=\tanh(\beta\sum_iJ_{ij}m_j).</math>
==Curie-Weiss model and the phase transition==
The Curie-Weiss model is the most simple model for ferromagnets, where <math>\textstyle J_{ij}=J/n</math>, a constant, and system is fully connected.
Without the external field (which is considered in the whole lecture) In the model there is a symmetry that for any configuration, flipping each spin to its opposite value does not alter the energy of the configuration, i.e.
:<math>E(\underline\sigma)=E(-\underline\sigma).</math>
It means that the magnetization of each spin should be always zero, as marginal distribution for spin ''i'' has to satisfy <math>\textstyle p_i(\sigma_i)=-p_i(\sigma_i).</math>
This is indeed the case at high temperature where <math>\textstyle \beta</math> is small. This is called the ''Paramagnetic'' state. However at a low temperature we can not see this state, because of
the spontaneous symmetry breaking, at which point the paramagnetic state becomes unstable. We can show the phenomenon using the Naïve Mean-Field Approximation.
Here the couplings are identical, the mean-field equation for the magnetization simply becomes
:<math>m=\tanh(\beta Jm).</math>
The fixed-points of above equation can be shown in the following figure where <math>\beta J=0.2, 0.5, 1, 2, 20.</math>
[[File:CurrieWeiss.png|400px]]
[[category:旧词条迁移]]