Guang Lin holds multiple titles at Purdue: Associate Dean for Research and Innovation at College of Science; Moses Cobb Stevens Professor in Mathematical Sciences; Full Professor in the School of Mechanical Engineering; and courtesy professorships in the Department of Earth, Atmospheric, and Planetary Sciences and the Department of Statistics. This reflects his broad interests in applying computational science to many different fields.

This specific heat transfer collaboration has its roots all the way back in 2012, when Professor of Mechanical Engineering Xiulin Ruan and his Ph.D. student Tianli Feng started tackling four-phonon scattering, a phenomenon which is fundamental to how a material conducts heat. In 2016, they finally reported these findings in a groundbreaking paper on four-phonon scattering. In the years since, Ruan and his team have received numerous awards for their work in this field.

However, the difficulty with studying four-phonon scattering is the computational power required to run simulations. Their calculations required a parallel computing strategy, essentially utilizing a computing cluster with one terabyte of memory. Ph.D. student Zherui Han developed his own thermal conductivity solver called FourPhonon, eventually writing a paper on it and publishing it open-source on GitHub.

Then Ruan enlisted the help of Lin and a Ph.D. student, Ziqi Guo, co-advised by him and Ruan. Guo developed a machine-learning framework to speed up these calculations.

Guo has now published his latest advancement, a branch of the FourPhonon solver called FourPhonon_GPU. This strategy offloads massive parallelizable tasks to the GPU, while using the CPU for process enumeration and control-heavy operations. This approach achieves up to 26-times acceleration versus using CPU alone.

Their research has been published in Cell Reports Physical Science.

“The new method significantly cuts down the time and computing power needed, making it faster and easier to identify materials that could be used for everything from electronics to energy-efficient buildings,” said Guo. “It also enables us to study new physics phenomenon with unprecedented accuracy, which was inaccessible due to the high computational cost.”

“This research has required a unique combination of fundamental heat transfer science, and cutting-edge computational techniques,” said Lin. “We need to run these simulations in order for Xiulin and his team to test and verify their theories. But they needed our expertise in computation to streamline the process and make it possible for scientists to expand their studies moving forward.”

The thermal radiative properties portion of the work was supported by National Science Foundation (Award No. 2102645). The thermal conductivity portion was supported by National Science Foundation (Award No. 2321301), and the methodology is implemented into an open-source code FourPhonon and that effort was supported by National Science Foundation (Award No. 2311848). The US Department of Energy also supports this work through the Office of Science Advanced Scientific Computing Research program (DE-SC0023161) and the Office of Fusion Energy Sciences (DE-SC0024583). Simulations were performed at the Rosen Center for Advanced Computing (RCAC) at Purdue University.

Source: Guang Lin, guanglin@purdue.edu; Xiulin Ruan, ruan@purdue.edu

GitHub: https://github.com/FourPhonon

FourPhonon_GPU, a CPU-GPU heterogeneous computing scheme for phonon scattering and thermal conductivity calculations
Ziqi Guo, David McLoughlin, Xiulin Ruan, Guang Lin
https://doi.org/10.1016/j.xcrp.2026.103491
ABSTRACT: Accurate prediction of phonon scattering is crucial for understanding thermal transport properties. However, the computational cost of such calculations, especially for four-phonon scattering, becomes prohibitive as the number of phonon branches and scattering processes increases. Here, we report FourPhonon_GPU, a GPU-accelerated framework for phonon scattering and thermal conductivity calculations. Leveraging OpenACC and a heterogeneous CPU-GPU computing strategy, we offload massive parallelizable tasks to the GPU while using the CPU for process enumeration and control-heavy operations. Under the relaxation-time approximation (RTA), our approach achieves up to 26x acceleration without sacrificing physical accuracy. Benchmarking on various GPU architectures confirms the method’s scalability and highlights the importance of aligning parallelization strategies with hardware capabilities. This work provides an efficient and accurate computational tool for phonon transport modeling and opens pathways for accelerated materials discovery.