L-9

Multifractality

In the last lecture we discussed that the eigenstates of the Anderson model can be localized, delocalized or multifractal. The idea is to look at the (generalized) IPR

$IPR(q)=\sum _{n}|\psi _{n}|^{2q}\sim L^{-\tau _{q}}$

The exponent $\tau _{q}$ is called multifractal exponent . Normalization imposes $\tau _{1}=0$ and the fact that the wave fuction is defined everywhere that $\tau _{0}=-d$. In general $\tau _{0}$ is the fractal dimension of the object we are considering and it is simply a geometrical property.

• Delocalized eigenstates

In this case, $|\psi _{n}|^{2}\approx L^{-d}$ for all the $L^{d}$ sites. This gives

$\tau _{q}^{\text{deloc}}=d(q-1)$

• Multifractal eigenstates.

This case correspond to more complex wave function for which we expect

$|\psi _{n}|^{2}\approx L^{-\alpha }\quad {\text{for}}\;L^{f(\alpha )}\;{\text{sites}}$

The exponent $\alpha$ is positive and $f(\alpha )$ is called multifractal spectrum . It is a convex function and its maximum is the fractal dimension of the object, in our case d. We can determine the relation between multifractal spectrum and exponent

$IPR(q)=\sum _{n}|\psi _{n}|^{2q}\sim \int d\alpha L^{-\alpha q}L^{f(\alpha )}$

for large L

$\tau (q)=\min _{\alpha }{(\alpha q-f(\alpha ))}$

This means that for $\alpha ^{*}(q)$ that verifies $f'(\alpha ^{*}(q))=q$ we have

$\tau (q)=\alpha ^{*}(q)q-f(\alpha ^{*}(q))$

A metal has a simple spectrum. Indeed, all sites have $\alpha =d$, hence $f(\alpha =d)=d$ and $f(\alpha \neq d)=-\infty$. Then $\alpha ^{*}(q)=d$ becomes $q$ independent. A multifractal has a smooth spectrum with a maximum at $\alpha _{0}$ with $f(\alpha _{0})=d$. At $q=1$, $f'(\alpha _{1})=1$ and $f(\alpha _{1})=\alpha _{1}$.

Larkin model

In your homewoork you solved a toy model for the interface. Consider a collection of L monomers $h_{1},h_{2},\ldots ,h_{L}$ in 1d with periodic boundary condition:

$\partial _{t}h_{i}(t)=h_{i+1}(t)+h_{i-1}(t)-2h_{i}(t)+F_{i}$

For simplicity, we assume $F_i$ iid Gaussian numbers with zero mean a variance D: ${\overline {F_{i}}}=0,\quad {\overline {F_{i}^{2}}}=\sigma ^{2}$. You proved that the roughness exponent of this model is $\zeta _{L}=(4-d)/2=3/2$ and the force per unit length acting of the interface is $f=\sigma /{\sqrt {L}}$

In the real model for depinning the disorder is however a non-linear function of h. The idea of Larkin is that linearization is correct only up, $r_{f}$ the length of correlation of the disorder along the h direction .