Exact calculation of disconnected loops

TL;DR

This paper implements an exact calculation of disconnected loops in lattice QCD using GPU acceleration, comparing stochastic methods and noise reduction techniques to improve the precision of flavor-singlet meson mass computations.

Contribution

It introduces an efficient GPU-based implementation for exact disconnected loop calculations and evaluates various noise reduction and dilution schemes in lattice QCD.

Findings

Consistency among different methods for eta' mass determination.

Gauge noise remains large for the ensemble studied.

Spin dilution significantly affects some quantities.

Abstract

We present an implementation of the disconnected diagram contributions to quantities such as the flavor-singlet pseudoscalar meson mass which are accelerated by GPGPU technology utilizing the NVIDIA CUDA platform. To enable the exact evaluation of the disconnected loops we use a $16^3 \times 32$ lattice and $N_f=2$ Wilson fermions simulated by the SESAM Collaboration. The disconnected loops are also computed using stochastic methods with several noise reduction techniques. In particular, we analyze various dilution schemes as well as the recently proposed truncated s olver method. We find consistency among the different methods used for the determination of the $\eta^\prime$ mass, albeit that the gauge noise for the ensemble studied is large. We also find that the effect of 'dilution' d oes not go beyond that of optimal statistical noise in many cases. It has been observed, however, that spin dilution does have a significant effect for some quantities studied.