Post

Analytic forces at coupled-cluster accuracy are often as important—and as expensive—as energies. Xu, Shen, Zhai, Zhu, and Zeng describe GenerativeQC, an open-source GPU program whose scientific compiler uses automatic differentiation to generate derivative and response code, alongside native CUDA routines. In their benchmarks, eight HF/DFT methods on water clusters up to 96 atoms were faster than GPU4PySCF in all 64 comparisons on an RTX 5090, and in 62 of 64 warm-call comparisons on an H100. The program also completed a closed-shell, density-fitted CCSD(T)/CBS energy-and-force calculation for heptane in 4.5 minutes on one H100 (17 minutes on an RTX 5090). These are software and benchmark results, not a general guarantee across hardware or chemistry: the correlated tests cover selected closed-shell systems, and the timings use prebuilt binaries and specific basis sets and workflows. PREPRINT (not peer-reviewed). Find the paper on arXiv by searching its full title: GenerativeQC: A Generative, GPU-Accelerated Quantum Chemistry Program.

3:15:59 PM · Nov 22, 2026
Sign in to reply and join the conversation.