Fault-tolerant quantum computations of vibrational wave functions

Marco Majland, Rasmus Berg Jensen, Patrick Ettenhuber, Irfansha Shaik, N. T. Zinner, Ove Christiansen

Abstract

Quantum computation of vibrational properties of molecules is a promising platform to obtain computational advantages for computational chemistry. However, fault-tolerant quantum computations of vibrational properties remain a relatively unexplored field in quantum computing. In this work, we present different algorithms for efficient encodings of vibrational Hamiltonians using qubitization. Specifically, we investigate different encoding representations, high order tensor decomposition to obtain low rank approximations for the vibrational Hamiltonian, rectilinear and polyspherical coordinate systems, parallelization and grouping algorithms. To investigate the performance of the different methods, we perform benchmark computations for both small and large molecules with more than one hundred vibrational modes.

Introduction

While quantum computing research has heavily focused on the electronic structure problem, the "nuclear motion problem"—the simulation of atomic vibrations—is equally critical for understanding chemical reactivity, thermodynamics, and spectroscopy. Classical methods for vibrational wave functions scale exponentially with the number of atoms, making large molecular systems intractable.

Previous quantum approaches often relied on Trotterization, which introduces approximation errors requiring high gate counts. This work addresses the challenge of fault-tolerant resource estimation by developing efficient encodings for vibrational Hamiltonians, specifically targeting future error-corrected hardware where Toffoli gate counts are the primary cost metric.

Methods: Qubitization and Tensor Decomposition

The core of the proposed framework is qubitization, which encodes the Hamiltonian as a Linear Combination of Unitaries (LCU). Unlike Trotterization, qubitization avoids time-evolution errors and generally requires fewer Toffoli gates.

Circuit diagram for the SELECT oracle used in qubitization.

To manage the massive number of interaction terms in the Potential Energy Surface (PES), the authors employ Canonical Polyadic (CP) and Tucker decompositions. These high-order tensor decompositions find low-rank approximations of the Hamiltonian, significantly reducing the number of operators that must be loaded into the quantum computer.

Furthermore, the study explores:

  • Coordinate Systems: Comparing standard rectilinear (normal) coordinates with curvilinear (polyspherical) coordinates, the latter being superior for large-amplitude motions like molecular torsions.
  • Parallelization: Circuit depth is reduced by grouping commuting operators using a weighted greedy graph-coloring algorithm, which provides near-optimal parallelization for systems with hundreds of modes.

Comparative Analysis of Representations

A significant finding of this research is the comparison between triangular and diagonal representations of one-mode operators. While diagonalizing operators reduces the total number of terms, the overhead of the basis transformations required to implement these diagonal forms often exceeds the savings.

Toffoli gate complexity for different Hamiltonian representations.

As shown in the complexity analysis, the triangular representation often yields lower Toffoli gate counts for the basis set sizes typical in vibrational problems. Additionally, the error introduced by tensor decomposition is found to be controllable and stochastic, allowing for a single threshold ($\epsilon_{LR}$) to balance gate reduction with chemical accuracy (defined as error $< 1 \text{ cm}^{-1}$).

Benchmarking and Results

The algorithms were benchmarked on systems ranging from small molecules (Formaldehyde, Hydrogen Peroxide) to large Polycyclic Aromatic Hydrocarbons (PAHs) with up to 156 vibrational modes (PAH8).

Key results include:

  • Term Reduction: Tensor decomposition reduced Hamiltonian terms by one to two orders of magnitude for large molecules.
  • Scaling: For PAH8 (156 modes), the estimated cost is approximately $2.01 \times 10^{11}$ Toffoli gates and ~10,000 qubits.
  • Parallelization Speedup: Grouping algorithms reduced circuit depth for large PAH chains by roughly two orders of magnitude compared to sequential encoding.

Conclusion

This work demonstrates that "chemically accurate" vibrational simulations are feasible on fault-tolerant hardware using optimized qubitization and tensor decomposition. While $10^{11}$ Toffoli gates still represent a significant temporal cost (potentially several days of runtime), the framework provides a clear path for further optimization. Future work may focus on basis set optimization for specific modes and extending these techniques to calculate transition intensities and time-resolved dynamics.

Reference: Majland, M., et al. (2025). Fault-tolerant quantum computations of vibrational wave functions. arXiv:2508.16253v1 [quant-ph].

Summary updated

Abstract

Quantum computation of vibrational properties of molecules is a promising platform to obtain computational advantages for computational chemistry. However, fault-tolerant quantum computations of vibrational properties remain a relatively unexplored field in quantum computing. In this work, we present different algorithms for efficient encodings of vibrational Hamiltonians using qubitization. Specifically, we investigate different encoding representations, high order tensor decomposition to obtain low rank approximations for the vibrational Hamiltonian, rectilinear and polyspherical coordinate systems, parallelization and grouping algorithms. To investigate the performance of the different methods, we perform benchmark computations for both small and large molecules with more than one hundred vibrational modes.
BibTeX
@article{majland2025fault,
  author = {Marco Majland and Rasmus Berg Jensen and Patrick Ettenhuber and Irfansha Shaik and N. T. Zinner and Ove Christiansen},
  title = {{Fault-tolerant quantum computations of vibrational wave functions}},
  journal = {Journal of Chemical Theory and Computation},
  year = {2025},
  doi = {10.1021/acs.jctc.5c01419},
}
Figure