CRII:OAC: Novel techniques for improving convergence and scalability of a Monte Carlo radiation solver for large-scale combustion simulations

燃烧一直是一个重要的能源的年龄，并将继续如此相当长的未来。 在高性能计算（HPC）的帮助下，预测性和准确的燃烧模拟具有巨大的潜力，可以成为实用系统（例如，燃气轮机、内燃机、熔炉等）。 燃烧系统的详细预测建模尤其需要对热辐射进行详细且准确的建模。 然而，用于燃烧模拟的热辐射模型通常过于简化。 在高性能计算中，使用详细辐射模型的主要瓶颈是计算成本高和并行效率差。 该项目探索了几种新颖的想法，以提高HPC燃烧模拟中高保真辐射求解器的效率和鲁棒性，从而有可能在现实的时间范围内对实际燃烧系统进行预测和准确的模拟。 进行这种大规模可靠的预测模拟的能力不仅在真实的燃烧装置的设计过程中是重要的，而且对于进一步理解燃烧过程的基本原理也是必不可少的。 考虑到对更清洁燃烧设备的需求不断增长，这种预测能力可能在学术研究以及能源和运输行业中产生重大影响。 该项目还将对在本科生中普及计算机编程产生影响。 因此，这项研究符合NSF的使命，以促进科学的进步和促进国家的健康，繁荣和福利。 本项目中选择的辐射求解器是基于蒙特-卡罗射线跟踪（MCRT）的求解器。 它是最精确的辐射求解器之一，并且随着问题复杂性的增加，它通常会超越所有其他辐射求解器。 为了提高MCRT求解器在大规模燃烧系统HPC模拟中的效率和可扩展性，本研究汇集了数学，统计理论和计算机科学等不同学科的思想，并将其应用于解决工程问题。 考虑到HPC中MCRT求解器的性能主要取决于底层统计算法和计算负载平衡，当前的研究分为三个主要任务。 首先，该项目正在开发新的算法，以改善收敛使用特殊的统计分布与低差异。 第二，MCRT负载管理的新策略，无论是在计算时间和内存利用率，正在探索提高可扩展性的求解器在HPC模拟。 第三，改进的MCRT解算器计划被创建为具有标准化接口的模块化、平台独立的解算器模块，以便其可以与任何燃烧和/或CFD解算器一起使用，而不会显著牺牲其性能。 通过提高MCRT的效率和可扩展性，这项工作旨在在现实的时间尺度内实现更准确的大规模燃烧系统预测HPC模拟。该奖项反映了NSF的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Combustion has been an important source of energy for ages and will continue to be so for considerable future. With the help of high-performance computing (HPC), predictive and accurate combustion simulations have a tremendous potential to emerge as a cost-effective and reliable design, assessment, and decision-making tool for practical systems (e.g., gas turbines, internal combustion engines, furnaces, etc.).  Detailed predictive modeling of combustion system requires, among other things, detailed and accurate modeling of thermal radiation. However, models for thermal radiation used in combustion simulations are usually over-simplified. The main bottlenecks in using detailed radiation model are its high computational cost and poor parallel efficiency in HPC.  This project explores several novel ideas to increase efficiency and robustness of a high-fidelity radiation solver in HPC combustion simulations, leading to the possibility of performing predictive and accurate simulations of practical combustion systems in a realistic time-frame.  The ability to perform such large-scale reliable predictive simulation is not only important in the design process of real combustion devices but also essential to further our understanding of fundamentals of combustion processes.  Considering the ever-increasing need for cleaner combustion devices, this predictive capability can potentially have a significant effect in academic research, as well as in energy and transportation industry. The project will also have an impact in popularizing computer programming in undergraduate students. Therefore, this research aligns with the NSF's mission to promote the progress of science and to advance the national health, prosperity, and welfare.  The radiation solver of choice in this project is a Monte-Carlo ray tracing-based (MCRT) solver. It is one of the most accurate radiation solver available, and typically outshines all other radiation solvers as the complexity of the problem increases.  To achieve improvements in efficiency and scalability of the MCRT solver in HPC simulations of large-scale combustion systems, this research brings together ideas from different disciplines of mathematics, statistical theory, and computer science and applies them to solve an engineering problem.  Considering the fact that the performance of an MCRT solver in HPC primarily depends on the underlying statistical algorithm and computational load-balancing, the current research is divided into three primary tasks. First, the project is developing new algorithms for improved convergence using special statistical distributions with low discrepancy. Second, novel strategies for MCRT load management, both in terms of computational time and memory utilization, are being explored to improve scalability of the solver in HPC simulations.  Third, the improved MCRT solver are planned to be created as a modular, platform-independent solver module with standardized interfaces so that it can be used with any combustion and/or CFD solver without significant sacrifice of its performance.  By enhancing efficiency and scalability of MCRT, this work aims to enable more accurate predictive HPC simulations of large-scale combustion systems in a realistic timescale.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.