SCF Convergence Data: Time-Series Features for Quantum Chemistry Workflows

Lechen Dong published this dataset on 2026-05-11 to support the reproducibility of machine-learning models for detecting Self-Consistent Field convergence failures. It contains structural coordinates and extracted time-series features for over 1,400 molecules from QM9 and QM40 parent sets. The data is pre-partitioned into training, validation, and test splits for direct use with a Gradient Boosting Classifier.

Use Cases

• Training classifiers for early SCF failure detection based on time-series descriptors.
• Benchmarking ML models on out-of-domain test sets like iQM9-default and iQM40-default.
• Analyzing convergence difficulty based on molecular structure diversity curated via PCA on Morgan fingerprints.

Strengths

• Data is pre-partitioned into train, val, and test splits, ready for model training.
• Includes over 1,400 molecules across multiple curated subsets (e.g., QM9-H1000).
• Validation and test sets are pre-scaled with a MinMaxScaler fitted on their respective training sets.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Data may reflect bias inherent to the specific quantum chemistry calculations and parent sets used.

Provenance

Source

figshare

Collection Method

Derived from quantum chemistry calculations on molecules from the QM9 and QM40 datasets.

Freshness

Last updated 2026-05-11 14:51:40; freshness should be verified.