ITR - (ASE) - (sim+dmc): Parallel Data Mining for Nanoscale Kinetic Monte Carlo Simulation Models

ITR - (ASE) - (sim dmc)：纳米级动力学蒙特卡罗模拟模型的并行数据挖掘

• 批准号: 0428826
• 负责人: Talat Rahman
• 金额: --
• 依托单位: Kansas State University
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 2004
• 资助国家: 美国
• 起止时间: 2004-10-01 至 2009-06-30
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=0428826&HistoricalAwards=false
• 关键词: ITR; ASE; sim+dmc; Parallel; Data

Intellectual Merit:Advances in computational science and technology have made the theoreticalmodeling of materials processes and properties viable, desirable, and a strong supplement to experimental work. Since the understanding and manipulation of the macroscopic properties of materials relies on information obtained at the microscopic level, one of the challenges in ASE is in developing the framework for a seamless, multiscale study of materials properties and related phenomena. The kinetic Monte Carlo method is one such technique which is suitable for simulations over a large range of length and time scales and which has the potential to connect atomistic details with macroscopic observations.The standard method is, however, handicapped because of the requirement of prior knowledge of theunderlying atomic mechanisms and their energetics. Typically only a few processes involving singleatom motion are provided as input to kinetic Monte Carlo simulations, thereby neglecting the role ofcollective atomic motion, vacancy creation, and complex atomic processes, as well as biasing the timeevolution of the system. In the proposed research in ASE with technical focus in sim and dmc, we planto overcome these limitations through the inclusion of unique and innovative pattern recognition schemesalong with automated procedures for the calculation of system energetics on the fly. This procedure willallow the development of an extensive database of possible atomic events. The database so collected willserve as input for further analysis and processing using machine learning and data mining for thedevelopment of efficient, robust, and accurate mapping functions which will be extensively tested throughsimulations of a variety of phenomena in epitaxial growth and validated through comparison with relevantexperimental data. The resulting mapping functions will serve to vastly increase the accuracy and speedof simulations.Broader Impact:Our goal of creating accurate and efficient computational algorithms for the simulationof phenomena such as thin-film growth will be a significant achievement in the technical focus areas ofsim and dmc, because of the innovative methodologies resulting from cross-disciplinary approaches. Thesuccessful implementation of the algorithms for computer design of materials, however, will be abreakthrough in ASE, as it will enable the development of technologically important materials with muchreduced cost and much greater control.The work will also provide us opportunities for educational and outreach activities with broad national, international and societal impact. Apart from the education and training of our graduate and undergraduate students in ITR, we will propose to work with the K-12 community in this endeavor. We intend to do so through the integration of research and education. Our team will collectively incorporate products of the research into courses on computational methods in physics, on data mining, on machine learning, and on adaptive parallelization techniques. A module for instructional and outreach purposes will also be developed. Two high school teachers will be recruited to spend summer sessions at KSU. Regular outreach activities with K-12 teachers and students will help broaden the pool of individuals in IT and nanoscale science literate individuals. Existing international collaborations of the PI with Prof. Alatalo, Finland, Dr. Trushin, Russia, and Dr. Durukanoglu, Turkey will help extend the outcomes of the proposed work internationally.

智力优势：计算科学和技术的进步使材料过程和性质的理论建模变得可行，可取，并成为实验工作的有力补充。由于材料的宏观性质的理解和操纵依赖于在微观水平上获得的信息，ASE的挑战之一是在开发一个无缝的，多尺度的材料特性和相关现象的研究框架。动力学Monte Carlo方法就是这样一种技术，它适合于在大范围的长度和时间尺度上进行模拟，并且有可能将原子细节与宏观观测联系起来，然而，标准方法由于需要基本原子机制及其能量学的先验知识而受到限制。通常只有少数涉及单原子运动的过程被提供作为动力学蒙特卡罗模拟的输入，从而忽略了集体原子运动、空位产生和复杂原子过程的作用，以及对系统时间演化的影响。在ASE的拟议研究中，技术重点是sim和dmc，我们计划通过纳入独特和创新的模式识别schemesalong自动化程序来克服这些局限性，用于系统能量学的计算。这个程序将允许开发一个可能的原子事件的广泛数据库。如此收集的数据库将作为输入，用于使用机器学习和数据挖掘进行进一步的分析和处理，以开发高效、稳健和准确的映射函数，这些映射函数将通过模拟外延生长中的各种现象进行广泛测试，并通过与相关实验数据的比较进行验证。由此产生的映射函数将大大提高模拟的准确性和速度。更广泛的影响：我们的目标是为模拟薄膜生长等现象创建准确有效的计算算法，这将是sim和dmc技术重点领域的一项重大成就，因为跨学科方法带来了创新方法。然而，材料计算机设计算法的成功实现将是ASE的一个突破，因为它将使技术上重要的材料的开发具有更低的成本和更大的控制。这项工作还将为我们提供具有广泛的国家、国际和社会影响的教育和推广活动的机会。除了我们的研究生和本科生在ITR的教育和培训，我们将建议与K-12社区在这奋进的努力。我们打算通过研究和教育相结合来做到这一点。我们的团队将共同将研究的产品纳入物理学计算方法，数据挖掘，机器学习和自适应并行化技术的课程中。还将开发一个教学和外联模块。两名高中教师将被招募到KSU参加暑期课程。与K-12教师和学生的定期外展活动将有助于扩大IT和纳米级科学素养人才库。PI与芬兰Alatalo教授、俄罗斯Trushin博士和土耳其Durukanoglu博士的现有国际合作将有助于在国际上推广拟议工作的成果。