Localization in matrix computations: Theory and applications

Many important problems in mathematics and physics lead to (nonsparse) functions, vectors, or matrices in which the fraction of nonnegligible entries is vanishingly small compared the total number of entries as the size of the system tends to infinity. In other words, the nonnegligible entries tend to be localized, or concentrated, around a small region within the computational domain, with rapid decay away from this region (uniformly as the system size grows). When present, localization opens up the possibility of developing fast approximation algorithms, the complexity of which scales linearly in the size of the problem.While localization already plays an important role in various areas of quantum physics and chemistry, it has received until recently relatively little attention by researchers in numerical linear algebra. In this chapter we survey localization phenomena arising in various fields, and we provide unified theoretical explanations for such phenomena using general results on the decay behavior of matrix functions. We also discuss computational implications for a range of applications.