[Paper Review] Fast numerical evaluation of time-derivative nonadiabatic couplings for mixed quantum-classical methods

Ilya G. Ryabinkin, Jayashree Nagesh+1 more|arXiv (Cornell University)|Aug 6, 2015

Spectroscopy and Quantum Chemical Studies6 citations

TL;DR

This paper presents a numerically efficient method to compute time-derivative nonadiabatic couplings (TDNACs) in mixed quantum-classical dynamics by reformulating the calculation in terms of molecular orbitals instead of Slater determinants, eliminating computationally expensive overlap determinant evaluations. The approach achieves speedups of up to 1,372× in benchmark systems, transforming TDNAC evaluation from a bottleneck into a tractable step in nonadiabatic dynamics simulations.

ABSTRACT

We have developed a numerical differentiation scheme which eliminates evaluation of overlap determinants in calculating the time-derivative non-adiabatic couplings (TDNACs). Evaluation of these determinants was the bottleneck in previous implementations of mixed quantum-classical methods using numerical differentiation of electronic wave functions in the Slater-determinant representation. The central idea of our approach is, first, to reduce the analytic time derivatives of Slater determinants to time derivatives of molecular orbitals, and then to apply a finite-difference formula. Benchmark calculations prove the efficiency of the proposed scheme showing impressive several-order-of-magnitude speedups of the TDNAC calculation step for midsize molecules.

Motivation & Objective

• To address the computational bottleneck in mixed quantum-classical nonadiabatic dynamics caused by evaluating Slater-determinant overlap matrices during TDNAC calculations.
• To eliminate the need for explicit evaluation of overlap determinants in numerical differentiation schemes for TDNACs.
• To develop a numerically stable and efficient alternative to conventional finite-difference methods that rely on many-electron wave function overlaps.
• To enable practical simulation of nonadiabatic dynamics in medium- and large-sized molecules by drastically reducing the cost of TDNAC evaluation.
• To validate the method’s accuracy and performance across diverse molecular systems and basis sets.

Proposed method

• The method reformulates the time-derivative nonadiabatic coupling (TDNAC) calculation by expressing the electronic wave functions in terms of molecular orbitals rather than Slater determinants.
• It applies finite-difference formulas—specifically first-order forward and second-order central differences—directly to the time-derivatives of molecular orbitals instead of the full many-electron wave functions.
• The central innovation is to reduce the analytic time derivative of a Slater determinant to the time derivative of its constituent molecular orbitals, thereby avoiding the need to compute overlap determinants between non-orthogonal Slater determinants.
• The approach leverages matrix-matrix multiplication of orbital coefficients and avoids screening procedures, relying on the orthogonality of molecular orbitals to simplify the calculation.
• The method is implemented using standard quantum chemistry software (Gaussian) and is compatible with spin-restricted formalisms, reducing computational cost by avoiding redundant calculations for α and β spin channels.
• The TDNACs are computed via the chain rule as τ_KJ = Ṙ · d_KJ, where d_KJ is derived from orbital derivatives, and the final coupling is obtained through finite-difference approximation on the orbital level.

Experimental results

Research questions

• RQ1 Can the evaluation of time-derivative nonadiabatic couplings be accelerated by avoiding explicit Slater-determinant overlap calculations?
• RQ2 Does an orbital-based finite-difference scheme for TDNACs maintain sufficient numerical accuracy compared to determinant-based methods?
• RQ3 To what extent can the proposed method reduce the computational cost of mixed quantum-classical nonadiabatic dynamics simulations?
• RQ4 How does the performance of the orbital-based scheme scale with system size and basis set size?
• RQ5 Can the method be effectively applied to midsize organic molecules with standard electronic structure methods like CIS?

Key findings

• The proposed orbital-based method achieves speedups of up to 1,372× in TDNAC evaluation for midsize organic molecules like C18H14O and C25H18, with the highest gains in small basis sets such as STO-3G.
• For a 50 fs FSSH trajectory of C18H14O using the 6-31G** basis set, the total simulation time was reduced from 99 hours (original Newton-X) to 45 hours with the new method, with TDNAC computation time dropping from 54 hours to just 12 minutes.
• The method maintains high numerical accuracy, with errors in TDNACs below 1×10⁻⁷ for time steps as small as 0.05 fs, and second-order central differences show superior convergence compared to first-order schemes.
• Even with increasing basis set size, the orbital-based scheme outperforms conventional determinant-based methods by over two orders of magnitude, particularly in systems where the number of occupied orbitals is comparable to the total basis size.
• The method is robust across different electronic states and molecular systems, with consistent performance improvements observed for both C18H14O and C25H18 across multiple basis sets.
• The elimination of overlap determinant evaluation enables efficient use of spin-restricted formalisms, avoiding the double computational cost associated with spin-unrestricted implementations in previous methods.