4
$\begingroup$

In Anderson localized systems, a great portion of the system's properties are governed by the localization length. These phenomena are well understood and have been studied for ages. However, I could not find an (even approximate) formula for the localization length in dependence on the disorder strength. I am especially interested in 1D. I know that calculating this numerically is not too hard but I would rather have a formula (or at least a table) at hand.

Is there such a formula for a spin chain or a corresponding chain model? If so, where can I find it?

$\endgroup$
2
  • 1
    $\begingroup$ Perhaps you may take a look at the book of Efetov. Supersymmetry in disorder and Chaos. Cambridge University Press. $\endgroup$ Commented Apr 22, 2020 at 16:15
  • $\begingroup$ The book by Imry is another place to look at: furet.com/livres/… $\endgroup$
    – Roger V.
    Commented Apr 30, 2021 at 14:12

1 Answer 1

4
+50
$\begingroup$

Localization in one dimension is often studied with techniques from random matrix theory. In particular, one can find the Lyapunov exponent corresponding to the random matrices that describe a given model, and under certain circumstances (described at great length in the first link below) this exponent is the inverse of the localization length.

With that in mind, here are estimates of the localization length for two models prevalent in the literature, which use this Lyapunov exponent technique.

First, the Anderson model. This is a tight-binding model of the form (in 1D): $$H=\sum_{n} \varepsilon_{n}|n\rangle\langle n|+V \sum_{n}| n\rangle\langle n+1|+h.c. $$ with random disorder $\epsilon_n$.

For a disorder drawn randomly between $-W/2$ and $W/2$ (a uniform distribution), and in the middle of the band, the localization length $\xi$ in units of the lattice constant is estimated (in the second reference below) as:

$$\xi^{-1}(E) = \gamma(E)=\frac{\left\langle\varepsilon^{2}\right\rangle}{V^{2}} F\left(\frac{E V}{\left\langle\varepsilon^{2}\right\rangle}\right) \quad(W,|E| \ll V)$$ where $F$ is a smooth function that goes from about 8.7 at $F(0)$ to 8 at $F(\infty)$. Actually, I think this is valid for other disorder distributions characterized by some width $W$, but it isn't completely clear to me what the necessary conditions are for this distribution. There is also a more complex scaling relation near the edges of the band:

$$\gamma(E)=\frac{\left\langle\varepsilon^{2}\right\rangle^{1 / 3}}{V^{2 / 3}} H\left(\frac{V^{1 / 3}(|E|-2 V)}{\left\langle\varepsilon^{2}\right\rangle^{2 / 3}}\right)$$ where $H$ is yet another smooth function that varies from $H(x\rightarrow -\infty)=-1/8x$ to $H(0)=0.289$ to $H(x\rightarrow \infty)=\sqrt{x}$.

A second model, studied because of its experimental relevance, is a 1D Bose-Einstein condensate in a disordered speckle field. This is a continuous model that is probably less relevant to the OP, but worth mentioning for completeness. The model definition uses the Gross-Pitaevskii equation for a BEC in a harmonic potential with random speckle:

$$ i \hbar \partial_{t} \psi=\left[\frac{-\hbar^{2}}{2 m} \partial_{z}^{2}+V_{\mathrm{ho}}(z)+V(z)+g|\psi|^{2}-\mu\right] \psi $$

with the speckle field $V(x)$ characterized by average intensity $V_R$ and correlation length $\sigma_R$. Then, the localization length is estimated (in ref. 3) as:

$$\xi^{-1}(k)=\gamma(k)=\frac{\pi m^{2} V_{\mathrm{R}}^{2} \sigma_{\mathrm{R}}}{2 \hbar^{4} k^{2}} \left(1-k \sigma_{\mathrm{R}}\right) \Theta\left(1-k \sigma_{\mathrm{R}}\right)$$

Note that this model has an effective mobility edge even in 1D, which is due to the finite correlation length of the disorder.

If you are interested in other models, searching for results for the Lyapunov exponent might help you to find more information.

References:

Lyapunov exponents, one-dimensional Anderson localisation and products of random matrices

B. A. Van Tiggelen, "Localization of waves," from Diffuse waves in complex media

Anderson Localization of Expanding Bose-Einstein Condensates in Random Potentials

$\endgroup$
2
  • $\begingroup$ This answer is a pleasure to read! It is nice to see the weak dependence on energy in the middle of the band in the Anderson model. For the edges of the band, no worries if you don't have the answer off the top of your head, but does the number of states with $|E|>2V$ grow with the number of sites or stay roughly constant? Such states yield the $H(x\rightarrow \infty)$ limit in the weak disorder limit, and hence would have localization length independent of disorder after plugging in the asymptotics. To stave off mobility edges, I expect the latter to be true, but I don't quite see it. $\endgroup$
    – user196574
    Commented May 2, 2021 at 7:23
  • $\begingroup$ Thanks! For $|E|>2V$ the density of states is low but finite, I believe, corresponding to your option two. The density of states just past the band edge generally has a universal form called a 'Lifschitz tail,' see for example: arxiv.org/abs/math-ph/0405020 . The fact that the disorder drops out might be precisely a reflection of this universality, but I'm not too familiar with the subject. $\endgroup$
    – Rococo
    Commented May 2, 2021 at 21:35

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.