Large deviations theory

Large deviations theory
Name	Large deviations theory
Field	Probability theory
Introduced	Early 20th century
Key contributors	Srinivasa Ramanujan, Harold Jeffreys, Harald Cramér, Richard H. S. Cox, S. R. Srinivasa Varadhan, Mark Kac, Yuri L. Levitin, Rudolf Carathéodory, Jakob Bernoulli, Andrey Kolmogorov

Large deviations theory is a branch of probability theory concerned with the asymptotic estimation of rare events, quantifying probabilities that stochastic systems deviate significantly from typical behavior. It connects to statistical mechanics, information theory, and dynamical systems through rate functions and exponential decay, providing rigorous tools for analysing tail probabilities, atypical fluctuations, and phase transitions. The theory builds on foundational work in probability and analysis and has spawned a wide range of mathematical results and applications across science and engineering.

History and motivations

Early motivations arose from attempts to formalize results in the law of large numbers and central limit theorem, with roots in the works of Jakob Bernoulli and later developments by Andrey Kolmogorov, Harald Cramér, and Harold Jeffreys. Influential mid-20th century figures such as Mark Kac and Richard H. S. Cox framed probabilistic and statistical problems that demanded precise asymptotics, while later contributions by S. R. Srinivasa Varadhan and contemporaries unified disparate strands. Historical connections run through seminal institutions and events including the mathematical milieu of Cambridge University, Princeton University, Courant Institute, and international gatherings like the International Congress of Mathematicians. Motivations also came from applied problems studied at laboratories and institutes such as Bell Laboratories, Los Alamos National Laboratory, and École Normale Supérieure, and from scientific programs at organizations including the National Science Foundation and NASA.

Mathematical foundations

Foundational mathematical structures draw on convex analysis developed by figures associated with École Polytechnique, functional analysis linked to Stefan Banach and John von Neumann, and measure theory advanced by Henri Lebesgue and Andrey Kolmogorov. The formalism uses topological vector spaces prominent in work at University of Göttingen and Berlin Mathematical School, and variational principles echo research from Princeton University and Massachusetts Institute of Technology. Key foundational results integrate contributions by Rudolf Carathéodory on convexity, insights from Sofia Kovalevskaya and Emmy Noether in analysis, and probabilistic rigor developed in papers circulated at Institute for Advanced Study and Royal Society meetings.

Main results and principles

Central results include sample path large deviations and rate function formulation by scholars associated with Columbia University and Stanford University, the {\it Gartner–Ellis theorem} whose lineage touches seminars at Université Paris-Sud, and the contraction principle influenced by work at University of Oxford. Theorems provide exponential tightness, Varadhan's lemma connected to research at Rutgers University, and Freidlin–Wentzell theory developed through collaborations involving University of Chicago and University of Michigan. Principles are applied in settings studied by researchers from University of California, Berkeley, Imperial College London, and ETH Zurich, and inform modern developments at research centers like Max Planck Society.

Examples and applications

Canonical examples include Cramér's theorem for sums of independent variables studied at Trinity College, Cambridge and occupancy problems explored in projects at Harvard University; queueing models used at AT&T Bell Labs and telecommunications research; random matrix extremes investigated in collaborations involving Institute for Advanced Study and Courant Institute; and spin systems and phase transitions in statistical physics examined at Princeton University and Los Alamos National Laboratory. Applications range across hypothesis testing and information theory work at Bell Labs and IBM Research, finance and risk management models in institutions like Goldman Sachs and J.P. Morgan, reliability engineering at General Electric research centers, and rare event simulation projects at Sandia National Laboratories and Lawrence Berkeley National Laboratory.

Methods and techniques

Analytical and probabilistic techniques trace to methods developed at University of Cambridge, Moscow State University, and École Normale Supérieure. Convex duality and Legendre–Fenchel transforms draw on traditions from Sorbonne University and ETH Zurich; entropy methods and Sanov's theorem build on statistical work at Columbia University and University of Chicago; importance sampling for efficient simulation evolved in applied research at Bell Laboratories and Los Alamos National Laboratory. PDE approaches and Hamilton–Jacobi methods link to studies at Massachusetts Institute of Technology and University of California, Berkeley, while combinatorial and enumeration techniques relate to results from Princeton University and Cambridge University Press-supported authors.

Extensions and related theories

Extensions include infinite-dimensional large deviations researched at Steklov Institute of Mathematics and Imperial College London, dynamical large deviations pursued at Max Planck Institute for Mathematics and Weizmann Institute of Science, and connections to information geometry explored at University of Pennsylvania and Tokyo University. Related theories intersect with concentration of measure studied at Microsoft Research and IBM Research, moderate deviations developed in seminars at Northwestern University and University of Texas at Austin, and metastability theory associated with investigators at École Polytechnique Fédérale de Lausanne and Université Paris-Dauphine. Cross-disciplinary influence reaches into fields represented by National Institutes of Health, European Research Council, and industry laboratories such as Siemens and Siemens AG.

Category:Probability theory