Searcharxiv⌕ Search
Home/Archive/arXiv

arXiv · 1603.05988

Deflation as a Method of Variance Reduction for Estimating the Trace of a Matrix Inverse

Arjun Singh Gambhir·Andreas Stathopoulos·Kostas Orginos
View original source ↗Read source document ↗Export metadata ↓

Abstract

Many fields require computing the trace of the inverse of a large, sparse matrix. The typical method used for such computations is the Hutchinson method which is a Monte Carlo (MC) averaging over matrix quadratures. To improve its convergence, several variance reductions techniques have been proposed. In this paper, we study the effects of deflating the near null singular value space. We make two main contributions. First, we analyze the variance of the Hutchinson method as a function of the deflated singular values and vectors. Although this provides good intuition in general, by assuming additionally that the singular vectors are random unitary matrices, we arrive at concise formulas for the deflated variance that include only the variance and mean of the singular values. We make the remarkable observation that deflation may increase variance for Hermitian matrices but not for non-Hermitian ones. This is a rare, if not unique, property where non-Hermitian matrices outperform Hermitian ones. The theory can be used as a model for predicting the benefits of deflation. Second, we use deflation in the context of a large scale application of "disconnected diagrams" in Lattice QCD. On lattices, Hierarchical Probing (HP) has previously provided an order of magnitude of variance reduction over MC by removing "error" from neighboring nodes of increasing distance in the lattice. Although deflation used directly on MC yields a limited improvement of 30% in our problem, when combined with HP they reduce variance by a factor of over 60 compared to MC. For this, we pre-computated 1000 smallest singular values of an ill-conditioned matrix of size 25 million. Using PRIMME and a domain-specific Algebraic Multigrid preconditioner, we perform one of the largest eigenvalue computations in Lattice QCD at a fraction of the cost of our trace computation.

Explore related subjects

hep-latcs.NAmath.NA

Keep this discovery

BibTeXRIS

Arjun Singh Gambhir, Andreas Stathopoulos, Kostas Orginos. 2016-03-18. Deflation as a Method of Variance Reduction for Estimating the Trace of a Matrix Inverse. https://doi.org/10.1137/16m1066361

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related papers

arXiv2026

Properties of the positive and negative parity charm-strange and bottom-strange mesons $D_s$, $D_s^*$, $D_{s0}^*$, $D_{s1}$, $B_s$, $B_s^*$, $B_{s0}^*$, $B_{s1}$ from lattice QCD: masses, decay constants, and compositeness

We present a lattice-QCD determination of properties of the lightest scalar, pseudoscalar, vector, and axial-vector heavy-strange mesons. This includes the decay constants of all mesons, and the binding energies and Weinberg compositeness parameters of the positive-parity states. The calculations are performed with domain-wall fermions for the light and strange quarks and anisotropic clover actions for the charm and bottom quarks. We use seven ensembles generated by RBC/UKQCD with pion masses ranging from 431 MeV to 139 MeV and lattice spacings ranging from 0.114 fm to 0.073 fm, which allows us to perform combined chiral and continuum extrapolations. For the negative-parity mesons, we obtain $f_{D_s}=251.4(2.1)(0.4)(2.5)\:{\rm MeV}$, $f_{D_s^*}=272.5(4.3)(1.0)(2.7)\:{\rm MeV}$, $f_{B_s}=228.4(5.8)(0.5)(2.3)\:{\rm MeV}$, $f_{B_s^*}=229.2(4.4)(0.8)(2.3)\:{\rm MeV}$, $f_{D_s^*}/f_{D_s}=1.086(14)(11)$, and $f_{B_s^*}/f_{B_s}=1.003(22)(10)$. In the positive-parity sector, the finite-volume energies and decay constants are extracted using the GEVP from correlation matrices with three different types of hadron interpolating operators, including operators with covariant derivatives and meson-meson-scattering operators at both source and sink. After extrapolation to the physical point, we obtain $f_{D^*_{s0}}=136.6 (8.0)(4.0)(1.4)$ MeV, $f_{D_{s1}}=200 (33)(24)(2)$ MeV, $f_{B^*_{s0}}=207 (12)(8)(2)$ MeV, and $f_{B_{s1}}= 196 (16)(11)(2)$ MeV. Our results for $f_{B^*_{s0}}$ and $f_{B_{s1}}$ are the first from lattice QCD. L\"uscher's method is used to find the infinite-volume bound-state masses. At the physical point, we obtain $m_{D^*_{s0}}-m_D-m_K=-48 (14)(4)$ MeV, $m_{D_{s1}}-m_{D^*}-m_K=-61 (15)(2)$ MeV, $m_{B^*_{s0}}-m_B-m_K= -69 (13)(4)$ MeV, and $m_{B_{s1}}-m_{B^*}-m_K=-77 (10)(5)$ MeV. Our analysis shows consistency with the positive-parity states being predominantly molecular.

hep-lat↗
arXiv2026

Sector-Resolved Flow Sampling for Topologically Frozen Lattice Gauge Theories

Topological fluctuations are essential to nonperturbative gauge theories but become increasingly difficult to sample toward the continuum limit, where Markov chains can freeze in sectors of fixed topological charge. We introduce a generative sampler, a mixture of sector-resolved samplers (MSRS) that explicitly resolves these sectors and exploits a key advantage of generative models, the ability to directly evaluate the domain-restricted partition function and thereby determine the relative weights of disconnected sectors. We train a generative model in a reference topological sector combined with a bijective topological shift that deterministically maps its samples to other sectors. We demonstrate the method in two-dimensional compact U (1) lattice gauge theory, where it reproduces the topological-charge distribution and yields an unbiased susceptibility in a regime where hybrid Monte Carlo is frozen and overrelaxation gives inaccurate estimates. Our approach also outperforms existing flow-based samplers by orders of magnitude. These results demonstrate that explicit sector resolution provides a promising route to overcoming topological barriers in lattice gauge theory.

hep-lat↗
arXiv2026

Gauge field digitization in the Hamiltonian limit

Quantum computers can circumvent the numerical sign problem in gauge theories at finite density or in real time. Quantum simulations of gauge theories require a finite-dimensional representation of continuous gauge fields. Replacing a continuous gauge group by a finite subgroup can substantially reduce the required quantum resources, but introduces digitization errors that must be controlled in the Hamiltonian, or continuous-time, limit. Previous studies, using the isotropic Euclidean lattices showed that the freezing transition of the discrete subgroup can make it a bad approximation for the continuous group at large Euclidean couplings. Here, we study the digitization of U(1) by its Z($N$) subgroups in 2+1 dimensions using anisotropic Euclidean lattices. We derive the trajectories of the spatial and temporal gauge couplings along which the Hamiltonian limit is approached at fixed Hamiltonian coupling. While the temporal coupling exhibits power-law scaling in the continuous U(1) theory, it grows only logarithmically for finite Z($N$). Using classical lattice simulations and exact diagonalization, we verify that these trajectories reproduce the corresponding Hamiltonian theories. We find that the freezing transition persists in the Hamiltonian limit of discrete gauge groups and that finite-$N$ theories can differ substantially from U(1) even outside the frozen regime, in contrast to the behavior on isotropic Euclidean lattices, where for small couplings, the discrete group provides a very accurate approximation of the continuous group. Our results provide a classical benchmark for quantifying the systematic errors due to gauge-field digitization in quantum simulations.

hep-lat↗