High precision, fast computation and high accuracy? Linear scaling double hybrid DFT calculations with spin-polarization
5 affiliations
- 1Department of Chemistry
- 2California
- 3Chemical Sciences Division
- 4Lawrence Berkeley National Laboratory
- 5Applied Mathematics and Computational Research Division
Abstract
Double-hybrid density functional theory (DH-DFT) has emerged as the most accurate DFT for molecular thermochemistry and kinetics, yet its widespread adoption for molecules with many atoms, A, is restricted by the steep O(A5) computational scaling and large O(A3) memory footprint of conventional canonical basis second-order Møller–Plesset (MP2) perturbation theory. Our previously reported spin-restricted local MP2 framework based on numerical sparsity successfully eliminated the O(A5) scaling bottleneck of MP2 for closed-shell systems. In this work, we address the challenge of generalizing it to achieve a linear-scaling, spin-unrestricted local MP2 algorithm, which enables asymptotically linear scaling evaluation of double hybrids for open shell and reactive molecules. The O(A1) linear scaling wall time of our implementation is validated on model radical-cation chains. Shared-memory (OpenMP) and distributed-memory (MPI) parallel implementations are benchmarked on open-shell transition metal configurations from the ROST61, MME55 and 16OSTM10 datasets. Governed by a single, user-defined precision threshold (ϵ), the numerical sparsity algorithm recovers over 99.99% of the canonical correlation energy using ϵ = 10−6 (a reasonable choice for production calculations), and yields relative energies with root-mean-square errors well below the threshold of chemical accuracy. As a larger example, a 181-atom transition metal complex is evaluated in the large def2-QZVPP basis set (8, 792 basis functions) with ϵ = 10−6 within one day utilizing 128 MPI ranks. Finally, integration with different double hybrid functionals, including ωB97M(2), PBE-QIDH and revDSD-PBEP86-D4, is demonstrated.
Related
- Towards exascale fully relativistic pseudopotential density functional theory calculations enabled by mixed-precision computation and compressed-communication using residual based subspace iteration
- Prediction Accuracy Can Select the Scaling Law of Required Precision
- Improving the accuracy of 182 ,184 ,186W(n , n'γ) cross-section calculations
- High-precision ab initio calculations of nuclear binding energies: Tin isotopes from dripline to dripline
- Accurate Wichmann-Kroll vacuum polarization calculations with Gaussian basis set
BibTeX
@misc{shen2026high,
title = {{High precision, fast computation and high accuracy? Linear scaling double hybrid DFT calculations with spin-polarization}},
author = {Yao Shen and Zhenling Wang and Tianyi Shi and Zhe Wang and Hrishikesh Ram and Etienne Palos and Yang Liu and Xiaoye S. Li and Martin Head-Gordon},
howpublished = {ChemRxiv},
year = {2026},
month = sep,
doi = {10.26434/chemrxiv.15008144/v1},
url = {https://chemrxiv.org/doi/full/10.26434/chemrxiv.15008144/v1},
note = {CC BY},
}
RIS
TY - UNPB
TI - High precision, fast computation and high accuracy? Linear scaling double hybrid DFT calculations with spin-polarization
AU - Yao Shen
AU - Zhenling Wang
AU - Tianyi Shi
AU - Zhe Wang
AU - Hrishikesh Ram
AU - Etienne Palos
AU - Yang Liu
AU - Xiaoye S. Li
AU - Martin Head-Gordon
PY - 2026
DA - 2026/09/01
PB - ChemRxiv
DO - 10.26434/chemrxiv.15008144/v1
UR - https://chemrxiv.org/doi/full/10.26434/chemrxiv.15008144/v1
AB - Double-hybrid density functional theory (DH-DFT) has emerged as the most accurate DFT for molecular thermochemistry and kinetics, yet its widespread adoption for molecules with many atoms, A, is restricted by the steep O(A5) computational scaling and large O(A3) memory footprint of conventional canonical basis second-order Møller–Plesset (MP2) perturbation theory. Our previously reported spin-restricted local MP2 framework based on numerical sparsity successfully eliminated the O(A5) scaling bottleneck of MP2 for closed-shell systems. In this work, we address the challenge of generalizing it to achieve a linear-scaling, spin-unrestricted local MP2 algorithm, which enables asymptotically linear scaling evaluation of double hybrids for open shell and reactive molecules. The O(A1) linear scaling wall time of our implementation is validated on model radical-cation chains. Shared-memory (OpenMP) and distributed-memory (MPI) parallel implementations are benchmarked on open-shell transition metal configurations from the ROST61, MME55 and 16OSTM10 datasets. Governed by a single, user-defined precision threshold (ϵ), the numerical sparsity algorithm recovers over 99.99% of the canonical correlation energy using ϵ = 10−6 (a reasonable choice for production calculations), and yields relative energies with root-mean-square errors well below the threshold of chemical accuracy. As a larger example, a 181-atom transition metal complex is evaluated in the large def2-QZVPP basis set (8, 792 basis functions) with ϵ = 10−6 within one day utilizing 128 MPI ranks. Finally, integration with different double hybrid functionals, including ωB97M(2), PBE-QIDH and revDSD-PBEP86-D4, is demonstrated.
ER -