Lattice Field Theory at the Exascale Frontier

点阵场论（LFT）提供了使用数值模拟来研究自然界基本力的工具。LFT的传统应用领域是量子色动力学（QCD），该理论描述了粒子物理学标准模型（SM）中的强核力。这些计算现在包括了电磁效应，并且达到了低于百分之百的精度。其他应用跨越了广泛的主题，从超越标准模型的理论，到低维强耦合费米子模型，再到新的宇宙学范式。这一科学努力的核心在于进行复杂和苛刻的数值模拟的能力。因此，高性能计算的百亿亿次时代看起来是一个充满机遇的时代。三十多年来，英国LFT社区一直处于该领域的前沿，并开发了广泛的研究领域组合，与高性能计算协同联系，导致算法和代码性能方面的重大进步。成功的亮点包括：影响新硬件的设计（蓝色基因系统）；开发算法（混合蒙特卡罗），被许多其他社区广泛使用；最大限度地利用新技术（晶格QCD从业者是新平台的首批用户之一，包括用于科学计算的gpu）；将LFT技术应用于人工智能中的新问题。随着pre-Exascale和Exascale系统的出现，LFT的研究项目及其影响可以以一种变革性的方式扩大，但前提是关键挑战得到解决。随着每秒浮点运算次数的增加，计算节点之间的通信会滞后，这种不平衡将严重影响未来的LFT模拟。这些挑战对于所有LFT代码库来说都是共同的，对于其他大量使用HPC资源的社区来说更是如此。新架构的瓶颈需要仔细识别，并且必须设计最小化通信的软件，以便最大限度地利用即将到来的大型计算机。随着我们进入异构架构的时代，新软件的设计必须清楚地将算法进展与不同硬件上的实现细节隔离开来，这样我们的软件才能以有限的努力有效地部署在即将问世的机器上。EXA-LAT项目的目标是开发一套通用的最佳实践、kpi和价值指标，可在不久的将来被整个LFT社区使用，并将为未来系统的设计和采购提供信息。除了LFT社区的参与外，许多供应商和计算中心也加入了该项目，以及来自“邻近”学科的学者。因此，我们的目标是创建一个国家和国际焦点，以促进学者，工业合作伙伴和研究软件工程师（rse）的活动。这个协同的环境将为学者、rse和学生举办培训活动，这将有助于创造一支沉浸在由该领域领先供应商组成的网络中的熟练劳动力。EXA-LAT将为LFT社区的长期努力奠定基础，以充分受益于Exascale设施，并将我们科学工作的一些技能转移给跨学科的更广泛的用户群体。

Lattice Field Theory (LFT) provides the tools to study the fundamental forces of nature using numerical simulations. The traditional realm of application of LFT has been Quantum Chromodynamics (QCD), the theory describing the strong nuclear force within the Standard Model (SM) of particle physics. These calculations now include electromagnetic effects and achieve sub percent accuracy. Other applications span a wide range of topics, from theories beyond the Standard Model, to low-dimensional strongly coupled fermionic models, to new cosmological paradigms. At the core of this scientific endeavour lies the ability to perform sophisticated and demanding numerical simulations. The Exascale era of High Performance Computing therefore looks like a time of great opportunities. The UK LFT community has been at the forefront of the field for more than three decades and has developed a broad portfolio of research areas, with synergetic connections to High-Performance Computing, leading to significant progress in algorithms and code performance. Highlights of successes include: influencing the design of new hardware (Blue Gene systems); developing algorithms (Hybrid Monte Carlo) that are used widely by many other communities; maximising the benefits from new technologies (lattice QCD practitioners were amongst the first users of new platforms, including GPUs for scientific computing); applying LFT techniques to new problems in Artificial Intelligence. The research programme in LFT, and its impact, can be expanded in a transformative way with the advent of pre-Exascale and Exascale systems, but only if key challenges are addressed. As the number of floating point operations per second increases, the communications between computing nodes are lagging behind, and this imbalance will severely affect future LFT simulations across the board. These challenges are common to all LFT codebases, and more generally to other communities that are large users of HPC resources. The bottlenecks on new architectures need to be carefully identified, and software that minimises the communications must be designed in order to make the best usage of forthcoming large computers. As we are entering an era of heterogeneous architectures, the design of new software must clearly isolate the algorithmic progress from the details of the implementation on disparate hardware, so that our software can be deployed efficiently on forthcoming machines with limited effort. The goal of the EXA-LAT project is to develop a common set of best practices, KPIs and figures of merit that can be used by the whole LFT community in the near future and will inform the design and procurement of future systems. Besides the participation of the LFT community, numerous vendors and computing centres have joined the project, together with scholars from 'neighbouring' disciplines. Thereby we aim to create a national and international focal point that will foster the activity of scholars, industrial partners and Research Sotfware Engineers (RSEs). This synergetic environment will host training events for academics, RSEs and students, which will contribute to the creation of a skilled work force immersed in a network that comprises the leading vendors in the subject. EXA-LAT will set the foundations for a long-term effort by the LFT community to fully benefit of Exascale facilities and transfer some of the skills that characterise our scientific work to a wider group of users across disciplines.