arXiv subjects

Dario Baum

Publications and source records attributed to Dario Baum.

3 recordsLinked to original sources

arXiv2025

Transfer learning of GW-Bethe-Salpeter Equation excitation energies

A persistent challenge in machine learning for electronic-structure calculations is the sharp imbalance between abundant low-fidelity data like DFT or TDDFT results and the scarcity of high-fidelity data like many-body perturbation theory labels. We show that transfer learning provides an effective route to bridge this gap: graph neural networks pretrained on DFT and TDDFT properties can be finetuned with limited qs$GW$ and qs$GW$-BSE data to yield accurate predictions of quasiparticle and excitation energies. Assessing both full-model and readout-only finetuning across chemically diverse test sets, we find that pretraining improves accuracy, reduces reliance on costly qs$GW$ data, and mitigates large predictive outliers even for molecules larger or chemically distinct from those seen during finetuning. Our results demonstrate that multi-fidelity transfer learning can substantially extend the reach of many-body-level predictions across chemical space.

physics.chem-ph↗

arXiv2025

qs$GW$ quasiparticle and $GW$-BSE excitation energies of 133,885 molecules

Machine learning applications in the chemical sciences, especially when based on neural networks, critically depend on the availability of large quantities of high quality data. As they provide excellent accuracy for both charged and neutral excitations, a large dataset containing quasiparticle self-consistent GW (qs$GW$) and Bethe-Salpeter equation (BSE) data would be highly desirable to model excited state energies and properties. In this work, we introduce a dataset for qs$GW$-BSE excitation energies and qs$GW$ quasiparticle energies of unprecedented size. Our dataset, denoted QM9GWBSE, supplies $GW$-BSE singlet-singlet and singlet-triplet excitation energies, corresponding transition dipole moments and oscillator strengths as well as qs$GW$ quasiparticle energies for all molecules from the popular QM9 dataset. We anticipate that QM9GWBSE will provide a solid foundation to train highly accurate machine learning models for the prediction of molecular excited state properties.

physics.chem-ph↗

arXiv2025

Predicting complete basis set limit quasiparticle energies from triple-$\zeta$ calculations

We present a simple linear model to estimate the basis set incompleteness errors (BSIE) of (vertex-corrected) $GW$ QP energies based on the kinetic energy of the corresponding orbital only. We parametrise the model for $G_0W_0$, quasi-particle self-consistent $GW$ (qs$GW$), and vertex-corrected ($\Sigma^{BSE}@L^{BSE}$) QP energies on a large set of molecules containing 10 different elements for which we calculate complete basis set (CBS) limit extrapolated reference values with correlation-consistent basis sets ranging from triple- to hextuple-$\zeta$ (TZ/6Z). Based on these accurate reference values, we obtain model parameters for Gaussian-type and Slater-type orbital (GTO/STO) basis sets which allow for the extrapolation of QP energies calculated with TZ basis sets to the CBS limit with errors of 20 to 30 meV. Analysing extrapolation errors, we show the commonly used extrapolation method which assumes an inverse linear dependence of the BSIE on the inverse number of basis functions to be valid, but to produce larger errors, even when a quadruple-$\zeta$ calculation is used in the extrapolation.

physics.chem-ph↗