Positive representations of complex distributions on groups

Abstract

A normalizable complex distribution $P(x)$ on a manifold $\mathcal{M}$ can be regarded as a complex weight, thereby allowing to define expectation values of observables $A(x)$ defined on $\mathcal{M}$. Straightforward importance sampling, $x\sim P$, is not available for non positive $P$, leading to the well-known sign (or phase) problem. A positive representation $\rho (z)$ of $P(x)$ is any normalizable positive distribution on the complexified manifold ${\mathcal{M}}^c$, such that, $⟨A(x){⟩}_P=⟨A(z){⟩}_{\rho }$ for a dense set of observables, where $A(z)$ stands for the analytically continued function on ${\mathcal{M}}^c$. Such representations allow to carry out Monte Carlo calculations to obtain estimates of $⟨A(x){⟩}_P$, through the sampling $z\sim \rho$. In the present work we tackle the problem of constructing positive representations for complex weights defined on manifolds of compact Lie groups, both Abelian and non Abelian, as required in lattice gauge field theories. Since the variance of the estimates increase for broad representations, special attention is put on the question of localization of the support of the representations.

23 pages, 2 figures. Substantial revision in Secs. II,III, and VI

References in corpus (8)

• Computational complexity and fundamental limitations to fermionic quantum Monte Carlo simulations
• Does the complex Langevin method give unbiased results?
• A complex path around the sign problem
• Existence of positive representations for complex weights
• Beyond Complex Langevin Equations: positive representation of a class of complex measures
• Explicit positive representation for weights on $R^d$
• Representation of complex probabilities and complex Gibbs sampling
• Satisfying positivity requirement in the Beyond Complex Langevin approach

Cited by in corpus (2)

• Complex Langevin and other approaches to the sign problem in quantum many-body physics
• Towards learning optimized kernels for complex Langevin