Quantum Chemistry

Electronic Structure Theory

Our foundation draws from the full hierarchy of quantum chemistry methods, understanding the accuracy-cost tradeoffs across theoretical levels to inform machine learning model design:

• Semi-empirical Methods: Fast approximations for large-scale screening
• Density Functional Theory: Balance of accuracy and computational cost
• Wavefunction Methods: High-accuracy benchmarks and challenging systems
• Coupled-Cluster Theory: Gold-standard accuracy for validation

Density Functional Theory Applications

Our DFT work addresses diverse chemical problems:

Structure-Property Relationships

• Energetic materials: decomposition pathways and sensitivity
• Thermochemistry: reaction energies and molecular stability
• Organometallic bonding: transition metal complexes
• Reaction mechanisms: transition states and intermediates

Functional Selection

The choice of exchange-correlation functional critically impacts accuracy:

• Range-Separated Hybrids: Essential for charge-transfer states
• Dispersion Corrections: Required for non-covalent interactions
• Meta-GGA Functionals: Improved treatment of reaction barriers
• Double-Hybrid Methods: Approaching wavefunction accuracy

Computational Dynamics

Ab Initio Molecular Dynamics

Car-Parrinello molecular dynamics and Born-Oppenheimer approaches explore systems where classical methods prove inadequate:

• Nucleic acid flexibility and conformational changes
• Biomolecular hydration effects and solvation
• Thermal decomposition in energetic compounds
• Conformational sampling for flexible molecules

Applications

• Reaction mechanism elucidation
• Temperature effects on molecular properties
• Solvent effects and explicit solvation
• Excited state dynamics

High-Accuracy Methods

Coupled-Cluster Techniques

Coupled-cluster methods, especially CCSD(T) (“gold standard”), provide benchmark-quality data essential for:

• Validating density functional approximations
• Training and testing machine learning models
• Understanding systematic errors in approximate methods
• Critical evaluation of reaction energetics

AIQM3 Framework

Our AIQM3 (Artificial Intelligence Quantum Mechanical) framework combines quantum chemistry with machine learning, targeting coupled-cluster accuracy at semi-empirical computational cost across seven main-group elements (H, C, N, O, F, S, Cl).

ML Training Data Generation

High-quality reference quantum calculations—comprising millions of DFT and high-level wavefunction calculations—generate training sets for neural network potentials:

Data Generation Strategy

• Diverse Sampling: Coverage of relevant chemical space
• Appropriate Theory: Matching method to target accuracy
• Active Learning: Identifying regions requiring additional data
• Quality Control: Validation of convergence and basis set completeness

Computational Infrastructure

• Massively parallel DFT calculations on HPC systems
• Automated workflow management
• Error detection and recomputation
• Database management for millions of calculations

Benchmarking and Validation

Rigorous comparison of ML models against quantum chemical benchmarks:

• Standard test sets (GMTKN55, COMP6, etc.)
• Custom benchmarks for specific applications
• Systematic error analysis
• Transferability testing across chemical domains

Integration with Machine Learning

Our quantum chemistry expertise directly informs ML model development:

Physics-Based Representations

• Incorporating quantum mechanical insights into neural network architectures
• Physically motivated features and symmetries
• Energy conservation and invariances

Uncertainty Quantification

• Identifying regions where ML models may fail
• Active learning to target high-uncertainty regions
• Ensemble methods for robust predictions

Hybrid QM/ML Methods

• ML for expensive energy components
• QM for critical interactions
• Seamless integration in molecular simulations

Vision

Our goal is not to replace quantum chemistry but to democratize access to accurate electronic structure calculations. By combining the rigor of quantum mechanics with the efficiency of machine learning, we enable:

• Routine coupled-cluster quality predictions
• Exploration of million-molecule chemical libraries
• Real-time property predictions for experimental guidance
• Accessibility of accurate methods to non-specialists

Applications

Quantum chemistry underpins all our research directions:

• ML Potential Development: Generating training data and benchmarks
• Drug Discovery: Accurate property predictions for pharmaceuticals
• Materials Design: Electronic properties of functional materials
• Reaction Modeling: Mechanistic understanding and kinetics
• Catalyst Design: Electronic structure of transition metal complexes

Computational Resources

This work leverages:

• CMU high-performance computing infrastructure
• National supercomputing resources (XSEDE, NERSC)
• Cloud computing for high-throughput calculations
• GPU acceleration for density functional theory

Future Directions

Emerging priorities in our quantum chemistry research:

• Excited state methods for photochemistry and spectroscopy
• Relativistic effects for heavy elements
• Strongly correlated systems
• Machine learning acceleration of wavefunction methods
• Automated error estimation and correction
• Real-time quantum dynamics

Collaborations

Our quantum chemistry research involves:

• Method developers in the quantum chemistry community
• Experimental collaborators requiring accurate predictions
• High-performance computing centers
• Software development teams for quantum chemistry codes