Species Thermodynamics Dataset Generation

[craabreu]

Species Thermodynamics Dataset Generation

Summary

This PR adds a script (scripts/thermo/species_thermodynamics.ipynb) that processes quantum mechanical calculation results to generate a comprehensive species thermodynamics dataset. The script extracts, validates, and consolidates thermodynamic properties from DFT-optimized geometries with DLPNO single-point energy calculations.

Dataset Output

File: data/thermo/quantumpioneer_species_thermo_dataset.csv

The generated dataset contains thermodynamic properties for unique chemical species, including:

• Basic Properties: SMILES representation, H298, S298, Cp300
• Wilhoit Parameters: Cp0, CpInf, a0, a1, a2, a3, H0, S0, B
• Quantum Chemical Data: DLPNO single-point energies, scaled DFT zero-point energies (ZPE)

Processing Workflow

1. Data Loading

   • Imports raw thermodynamic data from quantum_green_species_data_24august14_dft_opted_dlpno_sp_thermos.csv
   • Loads species reference data from quantum_green_species_data_24march12b.pkl

2. Data Filtering

   • Filters out species with missing thermodynamic data (p_thermo field)
   • Canonicalizes SMILES representations using RDKit for consistent molecular identification

3. Wilhoit Parameter Extraction

   • Implements custom parser to extract Wilhoit polynomial parameters from string representations
   • Regex pattern: (\w+)=\(?([-+]?\d+\.?\d*) handles both param=(value,'unit') and param=value formats
   • Extracts 9 Wilhoit parameters per species (Cp0, CpInf, a0-a3, H0, S0, B)

4. Dataset Assembly

   • Combines molecular identifiers, thermodynamic properties, and Wilhoit parameters
   • Includes DLPNO single-point energies (Hartree)
   • Adds scaled DFT zero-point energies (scaling factor: 0.972387)

5. Validation

   • Cross-validates against reaction dataset (quantum_green_ts_data_24july2c.pkl)
   • Verifies consistency of DLPNO energies and ZPE values between species and reaction datasets
   • Calculates maximum absolute differences to ensure data integrity

Technical Details

• Canonical SMILES: All molecular structures are converted to canonical SMILES with atom map numbers removed for uniqueness
• Data Deduplication: Ensures only unique species are included based on canonical SMILES
• Progress Tracking: Uses swifter for parallelized pandas operations with progress bars

Key Functions

• canonical_smiles(): Converts SMILES to canonical form with memoization
• parse_wilhoit_string(): Extracts numerical parameters from Wilhoit polynomial string representations

Dependencies

• pandas, swifter (data manipulation)
• RDKit (molecular structure handling)
• pathlib, re (file operations and parsing)