General-order open-shell coupled-cluster method with partial-spin adaptation II: further formulations, simplifications, implementations, and numerical results

Basic Information

• Paper ID: 2412.11029
• Title: General-order open-shell coupled-cluster method with partial-spin adaptation II: further formulations, simplifications, implementations, and numerical results
• Author: Cong Wang
• Classification: physics.chem-ph
• Publication Date: December 15, 2024
• Paper Link: https://arxiv.org/abs/2412.11029

Abstract

This paper is a continuation of previous work (arXiv:2403.10128), describing additional aspects including projected linear combinations and hash table normalization. An overview of the implementation of the general-order partial-spin adaptation (PSA) coupled-cluster (CC) method is provided, along with numerical results.

Research Background and Motivation

Core Problems

The coupled-cluster method for open-shell systems faces significant challenges in spatial orbital adaptation. Compared to spin-orbital methods, spatial orbital adaptation can in principle reduce the prefactor of computational cost, since the vector space dimension of spatial indices is smaller than that of spin-orbital indices for a given basis set.

Significance

1. Accuracy Requirements: Accuracy beyond chemical precision (1 kcal mol⁻¹) is crucial for predicting reaction selectivity
2. Method Limitations: High-order CC methods (beyond triple excitations) are among the few methods currently capable of achieving such precision
3. Technical Challenges: Contraction between active orbitals in open-shell systems complicates the formulation, making open-shell spatial-orbital CC methods still a challenge in quantum chemistry

Limitations of Existing Methods

• Closed-shell methods cannot be directly applied to open-shell systems
• Spin-orbital methods have higher computational costs
• Open-shell spatial-orbital methods face difficulties in formula derivation and implementation

Core Contributions

1. Extended Linear Combinations of Projected Manifolds: Proposed a linear combination scheme of projection operators applicable to general-order PSA-CC
2. Developed Hash Table Normalization Algorithm: Provided an efficient normalization method with O(1) computational complexity
3. Implemented General-order PSA-CC Method: Completed full implementation from equation generation to numerical computation
4. Provided Numerical Validation: Verified method accuracy through comparison with FCI and spin-orbital CC methods

Methodology Details

Theoretical Framework

The PSA-CC method is based on the following fundamental equations:

⟨Φ|He^{T_PSA}|Φ⟩_c = E                    (1)
⟨μ^PSA|He^{T_PSA}|Φ⟩_c = 0                (2)

where ⟨μ^PSA| and T_PSA are the projected manifold and excitation operator, respectively.

Linear Combinations of Projected Manifolds

To reduce the number of equations and accelerate CC convergence, linear combinations of projected manifolds are introduced:

⟨μ̌^PSA|He^{T_PSA}|Φ⟩_c = 0               (3)

Key linear combinations include:

• Single excitations: Ě^A_I = (1/2)E^A_I
• Double excitations: Ě^{AB} = (1/3)E^{AB} + (1/6)E^{AB}_
• Triple excitations: Complex combinations with coefficients such as 17/120, -1/120, etc.

Hash Table Normalization Algorithm

Core Concept:

• Keys: Symmetry-equivalent minimized virtual index equations (without prefactors)
• Values: Normalized equations (without prefactors)
• Average search complexity: O(1)

Algorithm Flow:

1. Start with an empty set of symmetry-equivalent equations
2. Iterate through all unnormalized equations
3. Check whether each belongs to an existing hash table key
4. If not present, add to hash table; if present, return normalized form
5. Merge equations differing only in prefactors

Working Equation Generation Procedure

1. Select CC Method and Spin Adaptation Level
2. Generate Terms:
   • Cluster operator T_PSA terms
   • Hamiltonian H terms
   • Excitation manifold R terms
3. Tensor Contraction Calculation: Using DECC algorithm
4. Normalization and Merging: Apply hash table algorithm
5. Add Amplitude Expressions
6. Collect Relevant Permutation Equations

Computational Setup

Computational Details

• Reference Calculations: ROHF and FCI calculations using PySCF 2.2.1
• Spin-orbital Calculations: Using NWChem 7.2.0
• PSA-CC Calculations: Using locally developed software
• Basis Sets: Standard basis sets including cc-pCVDZ, cc-pVDZ, etc.

Test Molecules

Including lithium atom and a series of diatomic molecules:

• Li (X²S state)
• BeH (X²Σ⁺ state)
• BH (a³Π state)
• CH (X²Π state)
• NH (X³Σ⁻ state)
• OH (X²Π state)

Spin Adaptation Levels

Multiple PSA levels were tested:

• PSA-T1R1
• PSA-T1|2R1|2
• PSA-T1|2|11R1|2
• PSA-T1|2|3R1|2|3
• PSA-T1|2|11|3R1|2|3

Experimental Results

Convergence Verification

Lithium Atom Results (correlation energy, atomic units):

The deviation of PSA-T1|2|11|3R1|2|3-CCSDT from FCI is only 2.2×10⁻¹¹ a.u., demonstrating excellent accuracy.

Comparison with Spin-orbital Methods

Energy Difference Analysis:

• Energy differences between PSA and spin-orbital CC methods are approximately 10⁻⁴ to 10⁻⁵ a.u.
• Corresponding errors are below 0.1 kcal mol⁻¹, satisfying chemical accuracy requirements
• The effect of T11 spin adaptation is negligible within the current scope

OH Molecule Example:

• PSA-T1|2R1|2-CCSD vs. spin-orbital CCSD difference: ~6×10⁻⁵ a.u.
• PSA-T1|2|3R1|2|3-CCSDT vs. spin-orbital CCSDT difference: ~8×10⁻⁵ a.u.
• PSA-T1|2|3R1|2|3-CCSDTQ vs. spin-orbital CCSDTQ difference: ~7×10⁻⁵ a.u.

Key Findings

1. Importance of CCSD Level: Differences between PSA and spin-orbital methods primarily appear at the CCSD level; improving CCSD-level spin adaptation may be more effective than improving higher-order CC expansions
2. Faster Convergence of Spin-orbital Methods: Numerical results indicate that spin-orbital methods converge to FCI faster than current PSA levels
3. Computational Efficiency Potential: Although the current implementation may have many equations affecting computational speed, the finite-dimensional vector space of spatial orbitals remains smaller than spin-orbitals; after further optimization, PSA-CC is expected to be more efficient than spin-orbital methods

Related Work

Historical Development

• Early Work: Lindgren (1978), Nakatsuji & Hirao (1977-1979), etc. laid the foundation for open-shell CC theory
• Method Development: Janssen & Schaefer (1991), Knowles et al. (1993-2000) developed practical open-shell CC methods
• Recent Progress: Datta & Gauss (2013-2019), Herrmann & Hanrath (2020-2022) further refined theory and implementation

Positioning of This Work

This work, based on the existing PSA-CC framework, specifically addresses:

1. Systematic treatment of linear combinations of projected manifolds
2. Efficient equation normalization algorithms
3. Complete implementation of general-order methods

Conclusions and Discussion

Main Conclusions

1. Method Validity: The PSA-CC method achieves accuracy comparable to spin-orbital CC methods, with energy differences below chemical precision
2. Implementation Feasibility: Through hash table normalization and projected linear combinations, general-order PSA-CC method has been successfully implemented
3. Optimization Potential: Further factorization and active index equation merging are expected to significantly improve computational efficiency

Limitations

1. Number of Equations: High-order PSA-CC and spin adaptation generate numerous equations, potentially affecting computational speed
2. Convergence Speed: Current PSA levels converge slower than spin-orbital methods
3. Implementation Complexity: Method implementation is more complex than standard CC methods

Future Directions

1. Factorization Optimization: Develop more efficient tensor contraction strategies
2. Spin Adaptation Improvement: Particularly for T2|2-level CCSD spin adaptation
3. Application of Orthogonality Relations: Utilize orthogonality relations to accelerate convergence of different spin states
4. Active Index Merging: Improve efficiency by merging tensors with small active index dimensions

In-depth Evaluation

Strengths

1. Theoretical Completeness: Provides a complete solution from theoretical derivation to numerical implementation
2. Algorithmic Innovation: Hash table normalization algorithm with O(1) complexity significantly improves efficiency
3. Sufficient Numerical Verification: Method accuracy verified through multiple molecular systems and different CC levels
4. Practical Value: Provides new tools for high-precision calculations of open-shell systems

Weaknesses

1. Computational Efficiency: Computational efficiency advantages of current implementation not yet fully demonstrated
2. Scope of Applicability: Testing primarily on small molecular systems; performance on larger molecules remains to be verified
3. Insufficient Theoretical Analysis: Limited theoretical explanation for precision differences between PSA and spin-orbital methods

Impact

1. Academic Contribution: Provides important progress for open-shell CC theory development
2. Practical Value: Likely to become an effective tool for high-precision quantum chemistry calculations
3. Reproducibility: Provides detailed implementation details and supplementary materials

Applicable Scenarios

1. High-precision Requirements: Reaction selectivity predictions requiring accuracy beyond chemical precision
2. Open-shell Systems: Transition metal compounds, free radicals, and other open-shell molecules
3. Method Development: Foundation for developing more efficient open-shell CC methods

References

This paper cites 102 references covering the historical development of CC theory, open-shell methods, spin adaptation techniques, and other aspects, providing readers with comprehensive background knowledge and information on related work. Key references include Shavitt & Bartlett's monograph on CC theory, Helgaker et al.'s textbook on molecular electronic structure theory, and important works by Knowles, Gauss, Kállay, and others on open-shell CC methods.