GFN Methods as Geometry Engines for Organic Photovoltaic Databases: A Benchmark Against BP86 /def2‐ SVP

ABSTRACT Molecular simulation workflows based on the GFN family of semiempirical methods (GFN1‐xTB, GFN2‐xTB, GFN0‐xTB, and GFN‐FF) are benchmarked as geometry engines against the original BP86/def2‐SVP DFT level of theory of the Harvard Clean Energy Project (CEP) database for high‐throughput organic photovoltaic (OPV) molecule screening. Two representative datasets are evaluated: a 73‐molecule subset of the CEP database of extended ‐conjugated systems and a 56‐molecule subset of the QM9 database of small organic molecules, selected via variance‐based Neyman allocation from stratified and clustered chemical spaces. Structural fidelity is assessed through heavy‐atom RMSD, radius of gyration, equilibrium rotational constants, bond lengths, and bond angles, and topological consistency via the VF2 graph isomorphism algorithm. GFN1‐xTB and GFN2‐xTB achieve the highest structural fidelity for both datasets, with mean hRMSD values of 0.765 and 0.771 Å for CEP and bond length MADs of 0.018 and 0.021 Å, respectively. GFN2‐xTB achieves the best angle accuracy (MAD = 0.70°) among all methods. VF2 topological failure rates are minimal for CEP (1.4% for both GFN1/2‐xTB workflows) but substantial for QM9 with GFN0‐xTB (26.8%) and GFN‐FF (16.1%), indicating that self‐consistent‐charge methods (GFN1/2‐xTB) provide substantially higher topological reliability than non‐self‐consistent methods for chemically diverse molecular spaces. Meanwhile, GFN‐FF provides the largest computational speedup (up to ∼177× relative to BP86/def2‐SVP) with an empirical ‐like apparent scaling over the sampled molecular size range. Sensitivity analyses across solvation environments, basis sets (def2‐SVP vs. def2‐TZVP), and DFT levels of theory (BP86/def2‐SVP vs. B3LYP/6‐31G(2df,p) for QM9) confirm that method rankings are preserved in all cases (rank correlation ρ > 0.98), validating the primary benchmark as robust for relative method comparison. For OPV database generation, GFN1‐xTB and GFN2‐xTB protocols provide the best overall balance of geometric fidelity, topological reliability, and computational throughput for bond‐resolved and graph‐based ML features, while GFN‐FF's workflow is the preferred geometry engine for first‐pass screening of large molecular libraries where topological validation can be applied. These results provide quantitative guidance for the deployment of GFN workflows in high‐throughput computational pipelines for ML‐oriented organic semiconductor discovery.

Authors

Institutions

Publication Details

Journal
International Journal of Quantum Chemistry
Published
2026-11-16
DOI
https://doi.org/10.1002/qua.70342
Primary Topic
Advanced Chemical Physics Studies
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

GFN Methods as Geometry Engines for Organic Photovoltaic Databases: A Benchmark Against BP86 /def2‐ SVP

Raoult Teukam Dabou, J.-P. Tchapet Njafa, Jean Pierre Nguenang, S. G. Nana Engo et al.
2 links
International Journal of Quantum Chemistry
Advanced Chemical Physics Studies
article

GFN Methods as Geometry Engines for Organic Photovoltaic Databases: A Benchmark Against BP86 /def2‐ SVP

Raoult Teukam Dabou, J.-P. Tchapet Njafa, Jean Pierre Nguenang, S. G. Nana Engo, Steve Cabrel Teguia Kouam, Patrick Sorrel Mvoto Kongo
article en

Abstract

ABSTRACT Molecular simulation workflows based on the GFN family of semiempirical methods (GFN1‐xTB, GFN2‐xTB, GFN0‐xTB, and GFN‐FF) are benchmarked as geometry engines against the original BP86/def2‐SVP DFT level of theory of the Harvard Clean Energy Project (CEP) database for high‐throughput organic photovoltaic (OPV) molecule screening. Two representative datasets are evaluated: a 73‐molecule subset of the CEP database of extended ‐conjugated systems and a 56‐molecule subset of the QM9 database of small organic molecules, selected via variance‐based Neyman allocation from stratified and clustered chemical spaces. Structural fidelity is assessed through heavy‐atom RMSD, radius of gyration, equilibrium rotational constants, bond lengths, and bond angles, and topological consistency via the VF2 graph isomorphism algorithm. GFN1‐xTB and GFN2‐xTB achieve the highest structural fidelity for both datasets, with mean hRMSD values of 0.765 and 0.771 Å for CEP and bond length MADs of 0.018 and 0.021 Å, respectively. GFN2‐xTB achieves the best angle accuracy (MAD = 0.70°) among all methods. VF2 topological failure rates are minimal for CEP (1.4% for both GFN1/2‐xTB workflows) but substantial for QM9 with GFN0‐xTB (26.8%) and GFN‐FF (16.1%), indicating that self‐consistent‐charge methods (GFN1/2‐xTB) provide substantially higher topological reliability than non‐self‐consistent methods for chemically diverse molecular spaces. Meanwhile, GFN‐FF provides the largest computational speedup (up to ∼177× relative to BP86/def2‐SVP) with an empirical ‐like apparent scaling over the sampled molecular size range. Sensitivity analyses across solvation environments, basis sets (def2‐SVP vs. def2‐TZVP), and DFT levels of theory (BP86/def2‐SVP vs. B3LYP/6‐31G(2df,p) for QM9) confirm that method rankings are preserved in all cases (rank correlation ρ > 0.98), validating the primary benchmark as robust for relative method comparison. For OPV database generation, GFN1‐xTB and GFN2‐xTB protocols provide the best overall balance of geometric fidelity, topological reliability, and computational throughput for bond‐resolved and graph‐based ML features, while GFN‐FF's workflow is the preferred geometry engine for first‐pass screening of large molecular libraries where topological validation can be applied. These results provide quantitative guidance for the deployment of GFN workflows in high‐throughput computational pipelines for ML‐oriented organic semiconductor discovery.

International Journal of Quantum ChemistryVol. 126(23)
Université de Yaoundé I (CM), University of Douala (CM), Université Laval (CA)
Advanced Chemical Physics Studies
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon