arXiv subjects

James Richings

Publications and source records attributed to James Richings.

8 recordsLinked to original sources

arXiv2026

Efficiently unquenching QCD+QED at O($α$)

We outline a strategy to efficiently include the electromagnetic interactions of the sea quarks in QCD+QED. When computing iso-spin breaking corrections to hadronic quantities at leading order in the electromagnetic coupling, the sea-quark charges result in quark-line disconnected diagrams which are challenging to compute precisely. An analysis of the variance of stochastic estimators for the relevant traces of quark propagators helps us to improve the situation for certain flavour combinations and space-time decompositions. We present preliminary numerical results for the variances of the corresponding contributions using an ensemble of $N_\mathrm{f}=2+1$ domain-wall fermions generated by the RBC/UKQCD collaboration.

hep-lat↗

arXiv2025

Introducing CQ: A C-like API for Quantum Accelerated HPC

In this paper we present CQ, a specification for a C-like API for quantum accelerated HPC, as well as CQ-SimBE, a reference implementation of CQ written in C99, and built on top of the statevector simulator QuEST. CQ focuses on enabling the incremental integration of quantum computing into classical HPC codes by supporting runtime offloading from languages such as C and Fortran. It provides a way of describing and offloading quantum computations which is compatible with strictly and strongly typed compiled languages, and gives the programmer fine-grained control over classical data movement. The CQ Simulated Backend (CQ-SimBE) provides both a way to demonstrate the usage and utility of CQ, and a space to experiment with new features such as support for analogue quantum computing. Both the CQ specification and CQ-SimBE are open-source, and available in public repositories.

cs.DC↗

arXiv2024

2BP: 2-Stage Backpropagation

As Deep Neural Networks (DNNs) grow in size and complexity, they often exceed the memory capacity of a single accelerator, necessitating the sharding of model parameters across multiple accelerators. Pipeline parallelism is a commonly used sharding strategy for training large DNNs. However, current implementations of pipeline parallelism are being unintentionally bottlenecked by the automatic differentiation tools provided by ML frameworks. This paper introduces 2-stage backpropagation (2BP). By splitting the backward propagation step into two separate stages, we can reduce idle compute time. We tested 2BP on various model architectures and pipelining schedules, achieving increases in throughput in all cases. Using 2BP, we were able to achieve a 1.70x increase in throughput compared to traditional methods when training a LLaMa-like transformer with 7 billion parameters across 4 GPUs.

cs.LG↗

arXiv2024

Benchmarking Machine Learning Applications on Heterogeneous Architecture using Reframe

With the rapid increase in machine learning workloads performed on HPC systems, it is beneficial to regularly perform machine learning specific benchmarks to monitor performance and identify issues. Furthermore, as part of the Edinburgh International Data Facility, EPCC currently hosts a wide range of machine learning accelerators including Nvidia GPUs, the Graphcore Bow Pod64 and Cerebras CS-2, which are managed via Kubernetes and Slurm. We extended the Reframe framework to support the Kubernetes scheduler backend, and utilise Reframe to perform machine learning benchmarks, and we discuss the preliminary results collected and challenges involved in integrating Reframe across multiple platforms and architectures.

cs.DC↗

arXiv2022

Isospin-breaking corrections to light leptonic decays in lattice QCD+QED at the physical point

We report on the physical-point RBC/UKQCD calculation of the leading isospin-breaking corrections to light-meson leptonic decays. This is highly relevant for future precision tests in the flavour physics sector, in particular the first-row unitarity of the Cabibbo-Kobayashi-Maskawa matrix containing the elements $V_{us}$ and $V_{ud}$. The simulations were performed using Domain-Wall fermions for $2+1$ flavours, and with isospin-breaking effects included perturbatively in the path integral through order $α$ and $(m_u - m_d)/Λ_{\mathrm{QCD}}$. We use QED$_{\mathrm{L}}$ for the inclusion of electromagnetism, and discuss here the non-locality of this prescription which has significant impact on the infinite-volume extrapolation.

hep-lat↗

arXiv2022

Isospin-breaking corrections to light-meson leptonic decays from lattice simulations at physical quark masses

The decreasing uncertainties in theoretical predictions and experimental measurements of several hadronic observables related to weak processes, which in many cases are now smaller than $\mathrm{O}(1\%)$, require theoretical calculations to include subleading corrections that were neglected so far. Precise determinations of leptonic and semi-leptonic decay rates, including QED and strong isospin-breaking effects, can play a central role in solving the current tensions in the first-row unitarity of the CKM matrix. In this work we present the first RBC/UKQCD lattice calculation of the isospin-breaking corrections to the ratio of leptonic decay rates of kaons and pions into muons and neutrinos. The calculation is performed with $N_\mathrm{f}=2+1$ dynamical quarks close to the physical point and domain wall fermions in the Möbius formulation are employed. Long-distance QED interactions are included according to the $\mathrm{QED_L}$ prescription and the crucial role of finite-volume electromagnetic corrections in the determination of leptonic decay rates, which produce a large systematic uncertainty, is extensively discussed. Finally, we study the different sources of uncertainty on $|V_\mathrm{us}|/|V_\mathrm{ud}|$ and observe that, if finite-volume systematics can be reduced, the error from isospin-breaking corrections is potentially sub-dominant in the final precision of the ratio of the CKM matrix elements.

hep-lat↗

arXiv2022

Lattice QCD and the Computational Frontier

The search for new physics requires a joint experimental and theoretical effort. Lattice QCD is already an essential tool for obtaining precise model-free theoretical predictions of the hadronic processes underlying many key experimental searches, such as those involving heavy flavor physics, the anomalous magnetic moment of the muon, nucleon-neutrino scattering, and rare, second-order electroweak processes. As experimental measurements become more precise over the next decade, lattice QCD will play an increasing role in providing the needed matching theoretical precision. Achieving the needed precision requires simulations with lattices with substantially increased resolution. As we push to finer lattice spacing we encounter an array of new challenges. They include algorithmic and software-engineering challenges, challenges in computer technology and design, and challenges in maintaining the necessary human resources. In this white paper we describe those challenges and discuss ways they are being dealt with. Overcoming them is key to supporting the community effort required to deliver the needed theoretical support for experiments in the coming decade.

hep-lat↗

arXiv2021

Near-Physical Point Lattice Calculation of Isospin-Breaking Corrections to $K_{\ell2}/π_{\ell2}$

In recent years, lattice determinations of non-perturbative quantities such as $f_K$ and $f_π$, which are relevant for $V_{us}$ and $V_{ud}$, have reached an impressive precision of $\mathcal{O}(1\%)$ or better. To make further progress, electromagnetic and strong isospin breaking effects must be included in lattice QCD simulations. We present the status of the RBC/UKQCD lattice calculation of isospin-breaking corrections to light meson leptonic decays. This computation is performed in a (2+1)-flavor QCD simulation using Domain Wall Fermions with near-physical quark masses. The isospin-breaking effects are implemented via a perturbative expansion of the action in $α$ and $(m_u-m_d)$. In this calculation, we work in the electro-quenched approximation and the photons are implemented in the Feynman gauge and $\text{QED}_\text{L}$ formulation.

hep-lat↗