We use overlap fermions as valence quarks to calculate meson masses in a wide quark mass range on the 2 + 1-flavor domain-wall fermion gauge configurations generated by the RBC and UKQCD Collaborations. The well-defined quark masses in the overlap fermion formalism and the clear valence quark mass dependence of meson masses observed from the calculation facilitate a direct derivation of physical current quark masses through a global fit to the lattice data, which incorporates O(a 2 ) and O(m (3)(6) GeV, respectively. Furthermore, we observe that the mass difference of the vector meson and the pseudoscalar meson with the same valence quark content is proportional to the reciprocal of the square root of the valence quark masses. The hyperfine splitting of charmonium, M J/ψ − Mη c , is determined to be 119(2)(7) MeV, which is in good agreement with the experimental value. We also predict the decay constant of Ds to be fD s = 254(2)(4) MeV. The masses of charmonium P -wave states χc0, χc1 and hc are also in good agreement with experiments.

We present a valence calculation of the electric polarizability of the neutron, neutral pion, and neutral kaon on two dynamically generated nHYP-clover ensembles. The pion masses for these ensembles are 227(2) MeV and 306(1) MeV, which are the lowest ones used in polarizability studies. This is part of a program geared towards determining these parameters at the physical point. We carry out a high statistics calculation that allows us to: (1) perform an extrapolation of the kaon polarizability to the physical point; we find αK = 0.269(43) × 10 −4 fm 3 , (2) quantitatively compare our neutron polarizability results with predictions from χPT, and (3) analyze the dependence on both the valence and sea quark masses. The kaon polarizability varies slowly with the light quark mass and the extrapolation can be done with high confidence.

We present a lattice QCD calculation of the polarizability of the neutron and other neutral hadrons that includes the effects of the background field on the sea quarks. This is done by perturbatively reweighting the charges of the sea quarks to couple them to the background field. The main challenge in such a calculation is stochastic estimation of the weight factors, and we discuss the difficulties in this estimation. Here we use an extremely aggressive dilution scheme to reduce the stochastic noise to a manageable level. The pion mass in our calculation is 300 MeV and the lattice size is 3 fm. For the neutron, we find that αE = 2.70(55) × 10 −4 fm 3 , which is the most precise lattice QCD determination of the polarizability to date that includes sea effects.

A direct calculation of the mixed-action parameter ∆mix with valence overlap fermions on a domainwall fermion sea is presented. The calculation is performed on four ensembles of the 2+1-flavor domain-wall gauge configurations: 24 3 × 64 (am l = 0.005, a = 0.114 fm) and 32 3 × 64 (am l = 0.004, 0.006, 0.008, a = 0.085 fm). For pion masses close to 300 MeV we find ∆mix = 0.030(6) GeV 4 at a = 0.114 fm and ∆mix = 0.033(12) GeV 4 at a = 0.085 fm. The results are quite independent of the lattice spacing and they are significantly smaller than the results for valence domain-wall fermions on Asqtad sea or those of valence overlap fermions on clover sea. Combining the results extracted from these two ensembles, we get ∆mix = 0.030(6)(5) GeV 4 , where the first error is statistical and the second is the systematic error associated with the fitting method.

Abstract-Lattice QCD calculations were one of the first applications to show the potential of GPUs in the area of high performance computing. Our interest is to find ways to effectively use GPUs for lattice calculations using the overlap operator. The large memory footprint of these codes requires the use of multiple GPUs in parallel. In this paper we show the methods we used to implement this operator efficiently. We run our codes both on a GPU cluster and a CPU cluster with similar interconnects. We find that to match performance the CPU cluster requires 20-30 times more CPU cores than GPUs.