Collaborative Research: ITR: (ASE)-(sim+dmc): Algorithms for Large-Scale Simulations of Turbulent Combustion

大规模湍流燃烧模拟的算法nsf -ITR资助项目：Stephen B. Pope，康奈尔大学，peyman Givi，匹兹堡大学这个ITR合作项目的重点是开发和使用创新的计算算法来模拟湍流燃烧。这是一个极具智力挑战的课题，因为它将高度复杂的非线性燃烧化学与湍流的多尺度和随机方面相结合。为了应对这一挑战，该项目的四个组成部分是(1)适用于燃烧化学的降维算法(2)存储-检索算法（包括使用广泛分布的数据库）(3)在大规模并行系统上实现高效性能的算法实现，以及(4)湍流燃烧模拟的性能。在燃烧（和其他应用）中，如果问题的维度可以降低，计算成本可以显著降低。目前正在探索和发展两种新的降维方法。这些都是基于预像曲线和迭代泰勒级数。存储检索算法在紊流燃烧计算中已经被证明是非常有效的，并且有许多其他的应用已经成熟。这些算法的基础是重用直接计算成本很高的数据（例如，刚性ODE控制化学反应的解）。在模拟早期生成的数据可以在模拟后期有效地重用。这一思想被扩展到广泛分布的计算和数据库中，这样在世界范围内产生的所有以前的模拟数据都可以使用。为了实现对湍流燃烧的精确和高效的模拟，结合了几种先进的方法：通过大涡模拟（LES）来处理流动，从而显式地表示大规模，非定常的3D运动；子网格尺度组成的统计分布完全由其联合概率密度函数（PDF）表示，其演化方程由拉格朗日粒子法求解；采用降维与检索相结合的方法，结合了真实的燃烧化学。这方面工作的目标是开发这些方法的全面实现，在大规模并行系统上有效地执行。最后，作为正在进行的国际合作研讨会的一部分，对存在高质量实验数据的几个“目标火焰”进行了模拟。除了测试和演示所开发的方法之外，这些模拟还用于调查物理子模型的性能，并阐明所涉及过程的物理和化学。现在，以及未来几十年，湍流燃烧是一个对社会和几个主要行业具有巨大意义的话题。能源的使用（在电力生产、运输、加工工业和其他地方）主要是通过在湍流中燃烧燃料而发生的。虽然目前人们对燃料电池和可能重新出现的核能很感兴趣，但现实情况是，燃烧技术在未来几十年仍将占据主导地位。有令人信服的理由寻求燃烧装置的改进，环境和经济，工业界越来越多地将计算机模拟作为实现改进设计的手段。更高的燃烧效率直接导致二氧化碳排放量的减少（对于给定的产量）；与此同时，减少氮氧化物和微粒等污染物的排放也在不断得到寻求。随着计算机功能的不断增强和仿真精度的提高，作为设计过程中不可或缺的一部分的计算机仿真将不可避免地变得越来越重要。在这个项目中，正在开发计算机算法，以大大提高我们模拟燃烧过程的能力，从而影响改进的燃烧装置的设计。虽然该项目的重点是湍流燃烧模拟，但所开发的算法（特别是降维和存储检索）在计算科学和工程中具有广泛的适用性。

ABSTRACTAlgorithms for Large-Scale Simulation of Turbulent CombustionNSF-ITR GrantPI's: Stephen B. Pope, Cornell UniversityPeyman Givi, University of PittsburghThe focus of this collaborative ITR project is the development and use of innovative computational algorithms for the simulation of turbulent combustion. This is a topic of extreme intellectual challenge as it combines highly complex and non-linear combustion chemistry with the multi-scale and stochastic aspects of turbulence. Addressing this challenge, the four components of the project are (1) Dimension Reduction Algorithms suitable for combustion chemistry (2) Storage-Retrieval Algorithms including the use of widely-distributed databases (3) Algorithm Implementation for efficient performance on large-scale parallel systems, and (4) performance of Turbulent Combustion Simulations. In combustion (and other applications) the computational cost can be dramatically decreased if the dimensionality of the problem can be reduced. Two new approaches to dimension reduction are being explored and developed. These are based on pre-image curves and iterated Taylor series. Storage-retrieval algorithms have proved extremely effective in turbulent combustion calculations, and there are many other applications ripe for their use. The basis of these algorithms is to re-use data that are costly to compute directly (e.g., the solutions to the stiff ODE's governing chemical reactions). Data generated early in a simulation are efficiently re-used later in the simulation. This idea is extended to widely distributed computing and databases, so that data generated worldwide in all previous simulations can be used. To achieve accurate and efficient simulations of turbulent combustion, several advanced methodologies are combined: the flow is treated by large-eddy simulation (LES) so that the large-scale, unsteady, 3D motions are explicitly represented; the statistical distribution of the subgrid scale compositions is fully represented by its joint probability density function (PDF) whose evolution equation is solved by a Lagrangian particle method; and realistic combustion chemistry is incorporated using the combination of dimension reduction and storage-retrieval. The objective of this aspect of the work is to develop a comprehensive implementation of these methodologies that performs efficiently on large-scale parallel systems. Finally, as part of an ongoing international collaborative workshop, simulations are performed for several "target flames" for which there exist high-quality experimental data. In addition to testing and demonstrating the methodology developed, these simulations serve to investigate the performance of the physical sub-models, and to shed light on the physics and chemistry of the processes involved. Now, and for many decades to come, turbulent combustion is a topic of tremendous significance to society and to several major industries. Energy usage (in power production, transportation, process industry and elsewhere) occurs predominantly through the combustion of fuels in turbulent flows. While there is, appropriately, great current interest in fuel cells and the possible re-emergence of nuclear power, the reality is that combustion technologies will remain dominant for many decades. There are compelling reasons to seek improvements in combustion devices, environmental and economic, and the industry is looking increasingly to computer simulations as a means of achieving improved designs. Higher combustion efficiencies lead directly to reduced CO2 emissions (for given output); at the same time, lower emissions of pollutants such as NO and particulates are continually being sought. It is inevitable that computer simulation, already an integral part of the design process, will grow in importance, as computers continually increase in power and the fidelity of the simulations improves. In this project, computer algorithms are being developed to increase substantially our abilities to simulate combustion processes and hence to impact the design of improved combustion devices. While the focus of the project is on turbulent combustion simulations, the algorithms developed (especially for dimension reduction and storage-retrieval) have broad applicability in computational science and engineering in general.