High-Accuracy Excited-State Reference Benchmark Dataset for Organic Semiconductors

The BenchmarkSet1500 resource theme provides a dataset of multireference excited states for 1500 small organic semiconductors, alongside a Python-based workflow used to generate the associated high-level excited-state calculations. It is designed for researchers in organic electronics and data-driven chemistry who require reliable and reproducible excited-state data, as well as those developing machine learning models or screening pipelines. By combining standardised computational workflows with multi-level electronic structure methods (TD-DFT, CASSCF, NEVPT2), the resource theme enables reproducible data generation and delivers an AI-ready dataset suitable for structure-property analysis, direct quantum chemistry method comparison, and molecular design.

Sort by

Data Sources

BenchmarkSet1500: High-Accuracy Excited-State Reference Benchmark Dataset for Organic Semiconductors

BenchmarkSet1500 is an open-access multireference excited-state database established to provide the first dedicated high-accuracy benchmark set for organic semiconductor research. The repository comprises 1,500 small organic molecules with consistently computed vertical excited-state properties obtained using state-averaged complete active space self-consistent field (SA-CASSCF) and strongly contracted N-electron valence state second-order perturbation theory (SC-NEVPT2), alongside the full reproducible workflow code used to generate the dataset. The dataset focuses on systems where single-reference approaches (e.g. TD-DFT) are known to fail, including molecules exhibiting strong static correlation and inverted singlet-triplet gaps. BenchmarkSet1500 is designed to support rigorous method benchmarking, systematic assessment of theory-level performance, development of predictive models, and screening for technologically relevant organic semiconductors. This dataset is available in two forms: (1) a data collection with one entry per molecule which contains curated metadata, optimised geometries (at B3LYP/6-31g* level of theory), complete electronic-structure output files, and computed excited-state energies and oscillator strengths for low-lying singlet and triplet states and (2) a consolidated machine-learning-ready CSV file which aggregates all molecules with their structural descriptors and excited-state properties to enable immediate integration into data-driven workflows.

Tools

PSDI_Benchmark_Set_1500 Code Github Repository

This resource provides a reproducible computational workflow for generating excited-state data for organic semiconductors, from initial molecular geometries to final compiled excited-state descriptors. It can be used for large-scale high-throughput screening, materials discovery, and machine learning. The workflow automates Gaussian input generation for ground state geometry optimisation and submission, error detection and resubmission for common calculation failures, conversion of completed jobs into ORCA-ready inputs for excited state calculations, and final extraction of key molecular and excited-state properties into a single CSV output.

What We Provide