Searcharxiv⌕ Search

arXiv · 2611.21992

Accelerating auxiliary-field quantum Monte Carlo with mixed-precision random propagators

Abstract

We develop an algorithm to propagate auxiliary-field quantum Monte Carlo (AFQMC) wavefunctions in mixed precision. We target NVIDIA Tensor Cores using the TensorFloat-32 (TF32) format. We approximate each full-precision array by a TF32 leading term and a TF32 residual, and each full-precision matrix product by three TF32 products accumulated in full precision. Benchmarks on hydrogen chains show that mixed-precision propagation is consistently faster than double-precision GPU propagation, with an increasing advantage as the system size grows. The resulting local-energy deviations remain small over the propagation windows considered here. These results identify the stochastic two-body propagator as a particularly effective target for reduced precision, while our method allows higher precision to be retained for accuracy-sensitive operations. Hence, our algorithm provides a practical way to speed up AFQMC calculations and demonstrates where reduced precision can be applied without adverse effects on the method's accuracy.

Explore related subjects

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Jacob Z. Williams, Yu Zhang. 2026-11-16. Accelerating auxiliary-field quantum Monte Carlo with mixed-precision random propagators. https://searcharxiv.com/abs/2611.21992

Citations retain the original authors and DOI and include this site’s record URL. If this discovery service contributed to your research, you can also cite Searcharxiv by Indiciae: BibTeX · RIS.

KEEP EXPLORING

Related papers

Efficient strategies for obtaining accurate kinetics from biased simulations

It has been demonstrated that one can recover the unbiased rate constants for slow conformational transitions by observing transitions driven by a bias potential added along one or several collective variables (CVs). However, the choice of CV, biasing scheme, biasing parameters and the rate estimator can all affect the results and it is not obvious which to choose for a given problem. In this study, we compare accuracy versus computational cost for a previously established benchmark system, a coarse-grained protein with computationally accessible unfolding kinetics. We apply bias to four CVs of decreasing quality in both a time-dependent and quasi-static manner, and compare different estimators for the unfolding rate constant in terms of absolute accuracy to the estimated reference unbiased rate. We show that in the case of gentle biasing, the exponential average time-dependent rate (EATR) framework, which simultaneously fits the unbiased rate constant and a coordinate quality metric $γ$, recovers accurate kinetics in nearly all cases. For poor CVs, estimators equivalent to assuming a good coordinate ($γ=1$) systematically fail regardless of biasing method or parameters. We introduce heuristics to show when we expect the rates to be inaccurate, and identify which combinations of biasing scheme, estimator and number of runs reach a given accuracy at the lowest cost.

physics.chem-ph↗

Neural Network Path Optimization for Finding Transition States on a Machine Learned Potential

Transition states govern chemical reaction rates and outcomes, yet locating them remains a central challenge in computational chemistry. Double-ended search methods find transition states given reactant and product structures, but generally rely on discretized path representations developed for density functional theory, where potential energy surface evaluations are expensive. Machine learned interatomic potentials (MLIPs) dramatically reduce this cost and leverage automatic differentiation to access higher-order derivatives efficiently. We introduce Popcornn, a framework to find transition states by optimizing continuous neural network paths atop an MLIP with a parallelized adaptive numerical integrator. This combination eliminates path discretization artifacts, reduces hyperparameter sensitivity, and enables non-local optimization to overcome local features using second-order potential-energy-surface information. Popcornn consistently recovers more correct transition states than standard methods such as nudged elastic band and growing string, across MLIPs on thousands of organic reactions, establishing it as a robust, automation-ready method for high-throughput transition state search. Popcornn's infrastructure is flexible in path objective, integration strategy, and optimization algorithm, making it a practical tool and platform for investigating open questions in MLIP-based path optimization.

physics.chem-ph↗

Quadrature-Free Radial Intracule Densities: A Bessel-Harmonic Formulation for Gaussian Basis Sets

The radial intracule density $I(R)$, the probability distribution of the interelectronic distance, is a compact and physically transparent picture of electron correlation, but its evaluation for molecules has always relied on a numerical angular quadrature over the orientation of the interelectronic vector. We show that this quadrature can be removed altogether for Cartesian Gaussian basis sets and arbitrary second-order density matrices. Starting from the Hermite Gaussian expansion of the products of basis functions, the coupling exponential $e^{2μ\mathbf{R}\cdot\mathbf{T}}$ that links radial and angular variables is expanded in modified spherical Bessel functions and real spherical harmonics, and the expansion terminates rigorously at $l=p_x+p_y+p_z$, where the $p$'s are the Cartesian powers arising from the Hermite expansion. The angular average is then exact, the atomic limit reduces to a finite sum of terms $C_iR^{n_i}e^{-μR^2}$, and the radial moments $\int_0^\infty R^nI(R)\,dR$ are given in closed form through the confluent hypergeometric function ${}_1F_1$. The scheme is exact in principle and fails only arithmetically: we identify the catastrophic cancellations in the angular-momentum coupling that ruin binary64 evaluations for heavy atoms and remove them by elevating the cancelling chain to binary128. Sum rules for the total, Coulomb and exchange-correlation intracules are recovered to seven to eight significant digits for systems from LiH to Kr$_2$, whereas Euler-Maclaurin and Lebedev quadratures err by up to 22% for stretched or multicenter molecules.

physics.chem-ph↗