General exponential basis set parametrization: Application to time-dependent bivariational wave functions

Mads Greisen Højlund, Alberto Zoccante, Ove Christiansen

Abstract

We present equations of motion (EOMs) for general time-dependent wave functions with exponentially parametrized biorthogonal basis sets. The equations are fully bivariational in the sense of the time-dependent bivariational principle (TDBVP) and offer an alternative, constraint-free formulation of adaptive basis sets for bivariational wave functions. We simplify the highly non-linear basis set equations using Lie algebraic techniques and show that the computationally intensive parts of the theory are in fact identical to those that arise with linearly parametrized basis sets. Our approach thus offers easy implementation on top of existing code with minimal computational overhead in the context of both nuclear dynamics and time-dependent electronic structure. Computationally tractable working equations are provided for single and double exponential parametrizations of the basis set evolution. The EOMs are generally applicable for any value of the basis set parameters, unlike the approach of transforming the parameters to zero at each evaluation of the EOMs. We show that the basis set equations contain a well-defined set of singularities, which are identified and removed by a simple scheme. The exponential basis set equations are implemented in conjunction with time-dependent vibrational coupled cluster with time-dependent modals (TDMVCC) and we investigate the propagation properties in terms of the average integrator step size. For the systems we test, the exponentially parametrized basis sets yield slightly larger step sizes compared to linearly parametrized basis sets.

Introduction

In the field of time-dependent quantum chemistry, simulating the evolution of wave functions—such as a molecule's vibrational response to laser excitation—requires high computational efficiency. To manage the complexity of these simulations, researchers often employ "adaptive" basis sets (modals or orbitals) that evolve alongside the wave function to capture the most relevant physics.

Traditionally, these basis sets are updated using linear parametrizations. However, maintaining the physical integrity of the simulation requires enforcing strict constraints, such as biorthonormality, which can lead to complex equations and numerical instabilities. While exponential parametrizations are a staple in time-independent problems (e.g., Coupled Cluster theory) due to their naturally constraint-free nature, they have historically been avoided in time-dependent contexts. The prevailing assumption was that their Equations of Motion (EOMs) would be too non-linear and computationally prohibitive to solve. This work challenges that assumption by providing a mathematically rigorous and computationally efficient framework for exponential basis sets.

Methodology

The Time-Dependent Bivariational Principle (TDBVP)

The authors apply the TDBVP to wave functions where basis transformations are represented as operator exponentials ($e^{\hat{\kappa}}$). By using a bivariational approach, the framework allows for independent evolution of the "bra" and "ket" spaces, which is essential for non-Hermitian theories like Coupled Cluster.

Lie Algebraic Simplification

The primary innovation of this research is the application of Lie algebraic techniques to simplify the non-linear derivatives inherent in exponential maps. By utilizing similarity-transformed derivatives, the authors demonstrate that the most computationally demanding aspect of the theory—solving for the basis set evolution—is mathematically equivalent to the operations performed in traditional linear theories. This "hidden simplicity" means that exponential basis sets can be integrated into existing software architectures with minimal additional programming effort.

Parametrization Flavors

The paper details two specific implementations:

  1. Single Exponential: A direct analogue to linear parametrization, providing a straightforward transition to the exponential framework.
  2. Double Exponential: Based on polar decomposition, this approach helps "lock" the bra and ket spaces together, preventing numerical drift during long-duration simulations.

Managing Singularities

Exponential maps possess inherent singularities, analogous to the distortion at the poles of a Mercator projection. The authors identify these well-defined points where the mathematical description fails and implement a "reset" scheme. When the system approaches a singularity, the Hamiltonian integrals are transformed, and the parameters are reset to zero, effectively establishing a fresh local coordinate system to continue the propagation smoothly.

Numerical Results and Discussion

The framework was implemented within the MidasCpp software package and tested on two molecular benchmarks using Time-Dependent Vibrational Coupled Cluster (TDMVCC) theory:

  • Water ($H_2O$): A simulation of intramolecular vibrational energy redistribution (IVR) following the excitation of the symmetric stretch.
  • 5D trans-bithiophene: A model for Franck-Condon emission ($S_1 \to S_0$).

Performance and Accuracy

The exponential methods achieved perfect numerical equivalence with linear reference models in calculating Autocorrelation Functions (ACFs), validating the theoretical derivation. More importantly, the exponential parametrization demonstrated improved numerical efficiency. In both the water and bithiophene tests, the exponential methods allowed for average integrator step sizes that were 1% to 3% larger than those of linear methods.

Computational Overhead

The additional mathematical operations required for the exponential framework, such as matrix diagonalizations, scale as $O(N^3)$ per mode. In vibrational dynamics, where the number of basis functions per mode ($N$) is typically small, this cost is negligible compared to the overall simulation time.

Conclusion

This work provides a robust, constraint-free alternative to linear basis set evolution in quantum dynamics. By proving that exponential parametrizations are computationally comparable to linear ones while offering superior integrator stability, the authors open new avenues for time-dependent electronic structure and nuclear dynamics.

Future research will focus on overcoming the limitations of the current "full-space matrix representation" (FSMR). By developing polynomial-scaling implementations, the authors aim to extend this exponential framework to larger electronic systems where the number of basis functions is significantly higher, potentially offering even greater advantages in stability and efficiency for complex molecular simulations.

Summary updated

Abstract

We present equations of motion (EOMs) for general time-dependent wave functions with exponentially parametrized biorthogonal basis sets. The equations are fully bivariational in the sense of the time-dependent bivariational principle (TDBVP) and offer an alternative, constraint free formulation of adaptive basis sets for bivariational wave functions. We simplify the highly non-linear basis set equations using Lie algebraic techniques and show that the computationally intensive parts of the theory are in fact identical to those that arise with linearly parametrized basis sets. Our approach thus offers easy implementation on top of existing code with minimal computational overhead in the context of both nuclear dynamics and time-dependent electronic structure. Computationally tractable working equations are provided for single and double exponential parametrizations of the basis set evolution. The EOMs are generally applicable for any value of the basis set parameters, unlike the approach of transforming the parameters to zero at each evaluation of the EOMs. We show that the basis set equations contain a well-defined set of singularities, which are identified and removed by a simple scheme. The exponential basis set equations are implemented in conjunction with time-dependent vibrational coupled cluster with time-dependent modals (TDMVCC) and we investigate the propagation properties in terms of the average integrator step size. For the systems we test, the exponentially parametrized basis sets yield slightly larger step sizes compared to linearly parametrized basis set.
BibTeX
@article{hjlund2023general,
  author = {Mads Greisen Højlund and Alberto Zoccante and Ove Christiansen},
  title = {{General exponential basis set parametrization: Application to time-dependent bivariational wave functions}},
  journal = {The Journal of Chemical Physics},
  year = {2023},
  doi = {10.1063/5.0152601},
}
Figure