Searcharxiv⌕ Search

arXiv subjects

M. L. Ndawana

Publications and source records attributed to M. L. Ndawana.

2 recordsLinked to original sources

Finite-Size Scaling of the Level Compressibility at the Anderson Transition

We compute the number level variance $Σ_{2}$ and the level compressibility $χ$ from high precision data for the Anderson model of localization and show that they can be used in order to estimate the critical properties at the metal-insulator transition by means of finite-size scaling. With $N$, $W$, and $L$ denoting, respectively, system size, disorder strength, and the average number of levels in units of the mean level spacing, we find that both $χ(N,W)$ and the integrated $Σ_{2}$ obey finite-size scaling. The high precision data was obtained for an anisotropic three-dimensional Anderson model with disorder given by a box distribution of width $W/2$. We compute the critical exponent as $ν\approx 1.45 \pm 0.12$ and the critical disorder as $W_{\rm c} \approx 8.59 \pm 0.05$ in agreement with previous transfer-matrix studies in the anisotropic model. Furthermore, we find $χ\approx 0.28 \pm 0.06$ at the metal-insulator transition in very close agreement with previous results.

cond-mat.dis-nn↗

The critical exponent of the localization length at the Anderson transition in 3D disordered systems is larger than 1

In a recent communication to the cond-mat archives, Suslov [cond-mat/0105325] severely criticizes a multitude of numerical results obtained by various groups for the critical exponent $ν$ of the localization length at the disorder-induced metal-insulator transition in the three-dimensional Anderson model of localization as ``entirely absurd'' and ``evident desinformation''. These claims are based on the observation that there still is a large disagreement between analytical, numerical and experimental results for the critical exponent. The author proposes, based on a ``simple procedure to deal with corrections to scaling'', that the numerical data support nu approx 1, whereas recent numerical papers find nu = 1.58 +/- 0.06. As we show here, these claims are entirely wrong. The proposed scheme does neither yield any improved accuracy when compared to the existing finite-size scaling methods, nor does it give nu approx 1 when applied to high-precision data. Rather, high-precision numerics with error epsilon approx 0.1% together with all available finite-size-scaling methods evidently produce a critical exponent nu approx 1.58.

cond-mat.dis-nn↗