Rational Hybrid Monte Carlo Research Articles

Low-mode deflation for twisted-mass and RHMC reweighting in lattice QCD

We propose improved estimators to compute the reweighting factors which are needed for lattice QCD calculations that rely on twisted-mass reweighting for the light quark contribution and the Rational Hybrid Monte Carlo (RHMC) algorithm for non-degenerate quark masses. This is the case for a number of modern large-scale simulations based on O(a) improved Wilson fermions. We find a significant reduction of uncertainties for the reweighting factors at similar computational cost compared to the conventional estimation. This leads to a significant increase in precision for phenomenologically relevant observables with high correlation to the low eigenmodes of the Wilson-Dirac operator in the presence of exceptionally small eigenvalues. Supplementary details regarding the spectral gap of the light quark Dirac operator on the 2+1 flavor large-volume ensembles explored in this study can be found in an accompanying appendix.

Single flavour optimisations to Hybrid Monte Carlo

It has become increasingly important to include one or more individual flavours of dynamical fermion in lattice QCD simulations. This is due in part to the advent of QCD+QED calculations, where isospin symmetry breaking means that the up, down, and strange quarks must be treated separately. These single-flavour pseudofermions are typically implemented as rational approximations to the inverse of the fermion matrix, using the technique known as Rational Hybrid Monte Carlo (RHMC). Over the years, a wide range of methods have been developed for accelerating simulations of two degenerate flavours of pseudofermion, while there are comparatively fewer such techniques for single-flavour pseudofermions. Here, we investigate two different filtering methods that can be applied to RHMC for simulating single-flavour pseudofermions, namely polynomial filtering (PF-RHMC), and filtering via truncations of the ordered product (tRHMC). A novel integration step-size tuning technique based on the characteristic scale is also introduced. Studies are performed on two different lattice volumes, demonstrating that one can achieve significant reductions in the computational cost of single-flavour simulations with these filtering techniques.

Accelerating lattice QCD simulations with 2 flavors of staggered fermions on multiple GPUs using OpenACC—A first attempt

We present the results of an effort to accelerate a Rational Hybrid Monte Carlo (RHMC) program for lattice quantum chromodynamics (QCD) simulation for 2 flavors of staggered fermions on multiple Kepler K20X GPUs distributed on different nodes of a Cray XC30. We do not use CUDA but adopt a higher level directive based programming approach using the OpenACC platform. The lattice QCD algorithm is known to be bandwidth bound; our timing results illustrate this clearly, and we discuss how this limits the parallelization gains. We achieve more than a factor three speed-up compared to the CPU only MPI program.

Domain wall fermion QCD with the exact one flavor algorithm

Lattice QCD calculations including the effects of one or more non-degenerate sea quark flavors are conventionally performed using the Rational Hybrid Monte Carlo (RHMC) algorithm, which computes the square root of the determinant of $\mathscr{D}^{\dagger} \mathscr{D}$, where $\mathscr{D}$ is the Dirac operator. The special case of two degenerate quark flavors with the same mass is described directly by the determinant of $\mathscr{D}^{\dagger} \mathscr{D}$ --- in particular, no square root is necessary --- enabling a variety of algorithmic developments, which have driven down the cost of simulating the light (up and down) quarks in the isospin-symmetric limit of equal masses. As a result, the relative cost of single quark flavors --- such as the strange or charm --- computed with RHMC has become more expensive. This problem is even more severe in the context of our measurements of the $\Delta I = 1/2$ $K \rightarrow \pi \pi$ matrix elements on lattice ensembles with $G$-parity boundary conditions, since $G$-parity is associated with a doubling of the number of quark flavors described by $\mathscr{D}$, and thus RHMC is needed for the isospin-symmetric light quarks as well. In this paper we report on our implementation of the exact one flavor algorithm (EOFA) introduced by the TWQCD collaboration for simulations including single flavors of domain wall quarks. We have developed a new preconditioner for the EOFA Dirac equation, which both reduces the cost of solving the Dirac equation and allows us to re-use the bulk of our existing high-performance code. Coupling these improvements with careful tuning of our integrator, the time per accepted trajectory in the production of our 2+1 flavor $G$-parity ensembles with physical pion and kaon masses has been decreased by a factor of 4.2.

Green-Schwarz superstring on the lattice

We consider possible discretizations for a gauge-fixed Green-Schwarz action of Type IIB superstring. We use them for measuring the action, from which we extract the cusp anomalous dimension of planar $\mathcal{N}=4$ SYM as derived from AdS/CFT, as well as the mass of the two $AdS$ excitations transverse to the relevant null cusp classical string solution. We perform lattice simulations employing a Rational Hybrid Monte Carlo (RHMC) algorithm and two Wilson-like fermion discretizations, one of which preserves the global $SO(6)$ symmetry of the model. We compare our results with the expected behavior at various values of $g=\frac{\sqrt{\lambda}}{4\pi}$. For both the observables, we find a good agreement for large $g$, which is the perturbative regime of the sigma-model. For smaller values of $g$, the expectation value of the action exhibits a deviation compatible with the presence of quadratic divergences. After their non-perturbative subtraction the continuum limit can be taken, and suggests a qualitative agreement with the non-perturbative expectation from AdS/CFT. Furthermore, we detect a phase in the fermion determinant, whose origin we explain, that for very small $g$ leads to a sign problem not treatable via standard reweigthing. The continuum extrapolations of the observables in the two different discretizations agree within errors, which is strongly suggesting that they lead to the same continuum limit. Part of the results discussed here were presented earlier in arXiv:1601.04670.

A deflated conjugate gradient method for multiple right hand sides and multiple shifts

We consider the task of computing solutions of linear systems that only differ by a shift with the identity matrix as well as linear systems with several different right-hand sides. In the past, Krylov subspace methods have been developed which exploit either the need for solutions to multiple right-hand sides (e.g. deflation type methods and block methods) or multiple shifts (e.g. shifted CG) with some success. In this paper we present a block Krylov subspace method which, based on a block Lanczos process, exploits both features--shifts and multiple right-hand sides--at once. Such situations arise, for example, in lattice quantum chromodynamics (QCD) simulations within the Rational Hybrid Monte Carlo (RHMC) algorithm. We present numerical evidence that our method is superior compared to applying other iterative methods to each of the systems individually as well as, in typical situations, to shifted or block Krylov subspace methods.

New one-flavor hybrid Monte Carlo simulation method for lattice fermions with [formula omitted] hermiticity

We propose a new method for Hybrid Monte Carlo (HMC) simulations with odd numbers of dynamical fermions on the lattice. It employs a different approach from polynomial or rational HMC. In this method, γ5 hermiticity of the lattice Dirac operators is crucial and it can be applied to Wilson, domain-wall, and overlap fermions. We compare HMC simulations with two degenerate flavors and (1+1) degenerate flavors using optimal domain-wall fermions. The ratio of the efficiency, (number of accepted trajectories)/(simulation time), is about 3:2. The relation between pseudofermion action of chirally symmetric lattice fermions in four-dimensional (overlap) and five-dimensional (domain-wall) representation are also analyzed.

Testing Algorithms for Finite Temperature Lattice QCD

We discuss recent algorithmic improvements in simulating finite temperature QCD on a lattice. In particular, the Rational Hybrid Monte Carlo(RHMC) algorithm is employed to generate lattice configurations for 2+1 flavor QCD. Unlike the Hybrid R Algorithm, RHMC is reversible, admitting a Metropolis accept/reject step that eliminates the O ( δ t 2 ) errors inherent in the R Algorithm. We also employ several algorithmic speed-ups, including multiple time scales, the use of a more efficient numerical integrator, and Hasenbusch pre-conditioning of the fermion force.

Accelerating staggered-fermion dynamics with the rational hybrid Monte Carlo algorithm

Improved staggered-fermion formulations are a popular choice for lattice QCD calculations. Historically, the algorithm used for such calculations has been the inexact $R$ algorithm, which has systematic errors that only vanish as the square of the integration step size. We describe how the exact rational hybrid Monte Carlo (RHMC) algorithm may be used in this context, and show that for parameters corresponding to current state-of-the-art computations it leads to a factor of approximately seven decrease in cost as well as having no step-size errors.

Rational hybrid Monte Carlo algorithm for theories with unknown spectral bounds

The Rational Hybrid Monte Carlo (RHMC) algorithm extends the Hybrid Monte Carlo algorithm for lattice QCD simulations to situations involving fractional powers of the determinant of the quadratic Dirac operator. This avoids the updating increment (dt) dependence of observables which plagues the Hybrid Molecular-dynamics (HMD) method. The RHMC algorithm uses rational approximations to fractional powers of the quadratic Dirac operator. Such approximations are only available when positive upper and lower bounds to the operator's spectrum are known. We apply the RHMC algorithm to simulations of 2 theories for which a positive lower spectral bound is unknown: lattice QCD with staggered quarks at finite isospin chemical potential and lattice QCD with massless staggered quarks and chiral 4-fermion interactions ({chi}QCD). A choice of lower bound is made in each case, and the properties of the RHMC simulations these define are studied. Justification of our choices of lower bounds is made by comparing measurements with those from HMD simulations, and by comparing different choices of lower bounds.

The RHMC algorithm for 2 flavours of dynamical staggered fermions

We describe an implementation of the Rational Hybrid Monte Carlo (RHMC) algorithm for dynamical computations with two flavours of staggered quarks. We discuss several variants of the method, the performance and possible sources of error for each of them, and we compare the performance and results to the inexact R algorithm.

Comparing the R algorithm and RHMC for staggered fermions

The R algorithm is widely used for simulating two flavours of dynamical staggered fermions. We give a simple proof that the algorithm converges to the desired probability distribution to within O(δπ 2) errors, but show that the relevant expansion parameter is ( δτ/m) 2 , m being the quark mass. The Rational Hybrid Monte Carlo (RHMC) algorithm provides an exact (i.e., has no step size errors) alternative for simulating the square root of the staggered Dirac operator. We propose using it to test the validity of the R algorithm for simulations carried out with δτ ≈ m.