Superposition of Random Plane Waves in High Spatial Dimensions: Random Matrix Approach to Landscape Complexity

Abstract

Motivated by current interest in understanding statistical properties of random landscapes in high-dimensional spaces, we consider a model of the landscape in ${\mathbb{R}}^N$ obtained by superimposing $M>N$ plane waves of random wavevectors and amplitudes. For this landscape we show how to compute the "annealed complexity" controlling the asymptotic growth rate of the mean number of stationary points as $N\to \infty$ at fixed ratio $\alpha =M/N>1$. The framework of this computation requires us to study spectral properties of $N\times N$ matrices $W=KT{K}^T$, where $T$ is diagonal with $M$ mean zero i.i.d. real normally distributed entries, and all $MN$ entries of $K$ are also i.i.d. real normal random variables. We suggest to call the latter Gaussian Marchenko-Pastur Ensemble, as such matrices appeared in the seminal 1967 paper by those authors. We compute the associated mean spectral density and evaluate some moments and correlation functions involving products of characteristic polynomials for such and related matrices.

arXiv comment40 pages, 6 figures

References in corpus (29)

• Spin-Glass Theory for Pedestrians
• Large Deviations of Extreme Eigenvalues of Random Matrices
• On the stationary points of the TAP free energy
• The statistics of critical points of Gaussian fields on large-dimensional spaces
• Characteristic polynomials of random matrices
• Complexity of random smooth functions on the high-dimensional sphere
• Products and Ratios of Characteristic Polynomials of Random Hermitian Matrices
• An exact formula for general spectral correlation function of random Hermitian matrices
• Critical Behaviour of the Number of Minima of a Random Landscape at the Glass Transition Point and the Tracy-Widom distribution
• Characteristic polynomials of real symmetric random matrices
• Universal Results for Correlations of Characteristic Polynomials: Riemann-Hilbert Approach
• Complex energy landscapes in spiked-tensor and simple glassy models: ruggedness, arrangements of local minima and phase transitions
• Superbosonization of invariant random matrix ensembles
• Topology trivialization and large deviations for the minimum in the simplest random optimization
• Quenched complexity of the p-spin spherical spin-glass with external magnetic field
• High-Dimensional Random Fields and Random Matrix Theory
• Exponential number of equilibria and depinning threshold for a directed polymer in a random potential
• Complexity of Random Energy Landscapes, Glass Transition and Absolute Value of Spectral Determinant of Random Matrices
• Triviality of the geometry of mixed $p$-spin spherical Hamiltonians with external field
• The Loss Surfaces of Neural Networks with General Activation Functions
• Distribution of rare saddles in the $p$-spin energy landscape
• Concentration of the complexity of spherical pure $p$-spin models at arbitrary energies
• Inflation in multi-field random Gaussian landscapes
• On the correlation function of the characteristic polynomials of the hermitian Wigner ensemble
• Counting stationary points of the loss function in the simplest constrained least-square optimization
• Landscape Complexity for the Empirical Risk of Generalized Linear Models
• On the correlation functions of the characteristic polynomials of the sparse hermitian random matrices
• Glass--like transition described by toppling of stability hierarchy
• The Distribution of Vacua in Random Landscape Potentials