Elements: FourPhonon: A Computational Tool for Higher-Order Phonon Anharmonicity and Thermal Properties

材料的导热系数在半导体器件的热管理、建筑绝缘材料、热障涂层和热电废热回收等许多新兴应用中都具有重要的意义。热由声子携带，声子是晶格振动的量子力学描述。传统上，热导率被认为是由涉及三个声子的散射过程控制的，但最近发现，涉及四个声子的散射过程可以起到重要甚至主导的作用。然而，即使对于最简单的材料，由于其复杂的公式和巨大的计算成本，预测四声子散射和由此产生的热导率是非常具有挑战性的。为了应对这些挑战，该项目旨在开发和优化一个开源计算程序包FourPhonon，使感兴趣的用户能够对其材料和应用程序进行这种计算。还将开发基于GPU和机器学习的方法，以显著加快计算速度。该项目将在未来十年将四声子散射从突破性转变为学术界和工业界的一种新的常规能力。该项目的目标是增强FourPhonon，这是一个由PI领导的团队部署的开源代码，可用于预测四声子散射率和由此产生的导热系数。自从FourPhonon的第一个版本发布以来，它已经被世界各地的许多研究人员用于他们的材料和应用。然而，为了跟上理论的进步，计算方法的升级是必要的，考虑到巨大的甚至无法负担的计算成本，加速计算是必要的。在这项提案中，调查人员将通过增强FourPhonon来满足这些需求。对于基本版本，该项目将：(1)开发一个可实现随温度变化的力常数的接口，这将使能够包括声子重整化和相变现象，以及(2)实现三声子和四声子散射通道的完全迭代方案。对于高级功能，该项目将：(1)通过异质计算使用GPU并行化来加速四声子散射的计算，以及(2)通过在一小部分散射过程的数据集上训练的机器学习模型来加速计算。改进的FourPhonon包将能够准确和负担得起地预测大量具有重要技术意义的材料的导热系数。这项由先进数字基础设施办公室颁发的奖项由工程局内的化学、生物工程、环境和运输系统司联合支持。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

Thermal conductivity of materials is important in many emerging applications, such as thermal management of semiconductor devices, insulation materials for buildings, thermal barrier coatings, and thermoelectric waste heat recovery. Heat is carried by phonons, the quantum mechanical description of lattice vibration. Conventionally, thermal conductivity was considered to be controlled by the scattering processes that involve three phonons, but recently it has been discovered that the scattering processes that involve four phonons can play a significant or even leading role. Predicting four-phonon scattering and the resulting thermal conductivity, however, is extremely challenging due to the complex formulation and tremendous computational cost even for the simplest materials. To address these challenges, this project is aimed at the development and optimization of an open-source computational package, FourPhonon, to enable interested users to perform such calculations for their materials and applications. Approaches based on GPU and machine learning will also be developed to significantly accelerate the speed of computation. The project will transform four-phonon scattering from a breakthrough to a new routine capability for academia and industry in the coming decade.The objective of this project is to enhance FourPhonon, an open-source code that was deployed by a team led by the PI and can be used to predict four-phonon scattering rates and the resulting thermal conductivity. Since the release of the first version of FourPhonon, it has been used by many researchers worldwide for their materials and applications. However, upgrades in computational methods are needed to keep up with theoretical advances, and acceleration of computation is necessary considering the large or even unaffordable computational cost. In this proposal, the investigators will fulfill these needs by enhancing FourPhonon. For the base version, the project will: (1) develop an interface that can implement temperature-dependent force constants, which will enable the capability of the inclusion of phonon renormalization and phase transition phenomena, and (2) enable the full iterative scheme of both three- and four-phonon scattering channels. For the advanced features, the project will: (1) accelerate the computation of four-phonon scattering using GPU parallelization via heterogeneous computing, and (2) accelerate the computation via machine learning models that are trained on datasets of a small fraction of the scattering processes. The improved FourPhonon package will enable accurate and affordable prediction of thermal conductivity of a large number of materials that are technologically significant.This award by the Office of Advanced Cyberinfrastructure is jointly supported by the Division of Chemical, Bioengineering, Environmental, and Transport Systems within the Directorate for Engineering.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.