A Bloch decomposition based split-step pseudo spectral method for quantum dynamics with periodic potentials

Zhongyi Huang, Shi Jin, Peter A. Markowich, Christof Sparber

Introduction

In the realm of solid-state physics, modeling the movement of electrons through a crystal lattice is a foundational task. Mathematically, this involves solving the one-dimensional Schrödinger equation where the Hamiltonian includes a periodic potential representing the lattice and a non-periodic potential representing external electric fields. While numerical methods for these equations have existed for decades, simulating the "semiclassical regime"—where the lattice's spatial oscillations are extremely rapid compared to the overall movement—remains a significant computational challenge.

This paper introduces a novel numerical framework that leverages the physical properties of periodic structures to achieve unprecedented efficiency. By employing a Bloch decomposition-based split-step pseudo-spectral method, the authors demonstrate that it is possible to simulate quantum dynamics with time steps that are independent of the small semiclassical scale.

The Computational Bottleneck

The standard approach for solving time-dependent Schrödinger equations is the Time-Splitting Spectral (TS) method. This technique alternates between solving the kinetic energy (dispersion) in the frequency domain and the potential energy in the spatial domain. However, in the semiclassical regime, characterized by a small parameter $\epsilon$, the wave function oscillates at a frequency of $1/\epsilon$.

To maintain accuracy and avoid spurious numerical artifacts, traditional TS methods require the time step ($\Delta t$) and the spatial mesh size ($\Delta x$) to be proportional to $\epsilon$. When $\epsilon$ is small (e.g., $1/1024$), the number of required time steps becomes so large that long-term simulations become computationally prohibitive. Traditional methods often produce "extremely spurious" results unless an incredibly fine mesh is used.

The Bloch Decomposition Approach

The authors' primary innovation is a "smart" splitting of the Hamiltonian operator. Instead of the traditional kinetic-versus-potential split, they group the kinetic energy operator and the periodic lattice potential together.

1. Diagonalization via Bloch Theory

By using classical Bloch decomposition, the combined periodic operator can be diagonalized. This allows the dominant effects of the lattice and the dispersion to be computed simultaneously and exactly in a single step of the splitting algorithm. The non-periodic external potential is then treated as a perturbation in the second step.

2. Preprocessing and Energy Bands

The method relies on the calculation of "energy bands" ($E_m(k)$), which are the eigenvalues of the periodic Hamiltonian. A key advantage of this method is that these bands are computed in a one-time preprocessing stage. Once the bands are established, the simulation can proceed without the need to resolve the fine-scale oscillations of the lattice at every time step. While there are technically infinite bands, the authors show that using a truncated set (e.g., 8 to 32 bands) is sufficient for high accuracy.

3. Commutator Error Reduction

The accuracy of split-step methods is governed by the "commutator" of the two split operators. Because the external potential typically varies slowly compared to the lattice, the commutator between the periodic Hamiltonian and the external field is relatively small. This physical separation of scales—treating the "fast" degrees of freedom analytically and the "slow" degrees numerically—is what allows the time step $\Delta t$ to be chosen much larger than in traditional methods: $\Delta t = O(1)$ rather than $O(\epsilon)$.

Numerical Validation

The researchers tested the Bloch Decomposition (BD) method against two primary models:

  • Mathieu’s Model: A smooth periodic potential based on a cosine function, representing an idealized crystal.
  • Kronig-Penney Model: A non-smooth potential consisting of rectangular barriers. This is a rigorous test for spectral methods, as sharp edges often cause numerical instabilities.

The BD method proved remarkably robust. Even in the Kronig-Penney model, the preprocessing of the Bloch basis effectively "absorbed" the non-smoothness of the potential, allowing for stable results where traditional methods struggled.

Key Findings and Performance

The results demonstrate a massive leap in performance and reliability:

  • Massive Speedup: For a semiclassical parameter of $\epsilon = 1/1024$, the BD method is effectively a thousand times more efficient per time step than the traditional TS method.
  • Unconditional Stability: The method is shown to be unconditionally stable, meaning the numerical solution will not "blow up" even when using large time steps.
  • Handling Non-Smoothness: Because the lattice potential is handled in the Bloch basis, the lack of regularity in models like Kronig-Penney does not degrade the method's performance.

Limitations and Future Work

While the paper focuses on one-dimensional linear systems, it identifies several paths for future research. The current implementation assumes "isolated" energy bands; if bands cross (the Landau-Zener phenomena), energy can jump between bands, requiring more sophisticated handling of the adiabatic splitting. Additionally, the authors aim to extend this framework to non-linear dynamics, such as the Gross-Pitaevskii equation, and higher-dimensional 3D systems.

By aligning the numerical architecture with the underlying physics of Bloch waves, this method provides a powerful and efficient tool for simulating quantum transport in periodic media.

Summary updated

Abstract

We present a new numerical method for accurate computations of solutions to (linear) one dimensional Schrödinger equations with periodic potentials. This is a prominent model in solid state physics where we also allow for perturbations by non-periodic potentials describing external electric fields. Our approach is based on the classical Bloch decomposition method which allows to diagonalize the periodic part of the Hamiltonian operator. Hence, the dominant effects from dispersion and periodic lattice potential are computed together, while the non-periodic potential acts only as a perturbation. Because the split-step communicator error between the periodic and non-periodic parts is relatively small, the step size can be chosen substantially larger than for the traditional splitting of the dispersion and potential operators. Indeed it is shown by the given examples, that our method is unconditionally stable and more efficient than the traditional split-step pseudo spectral schemes. To this end a particular focus is on the semiclassical regime, where the new algorithm naturally incorporates the adiabatic splitting of slow and fast degrees of freedom.
BibTeX
@inproceedings{huang2007bloch,
  author = {Zhongyi Huang and Shi Jin and Peter A. Markowich and Christof Sparber},
  title = {{A Bloch decomposition based split-step pseudo spectral method for quantum dynamics with periodic potentials}},
  booktitle = {SIAM},
  year = {2007},
  doi = {10.1137/060652026},
}
Figure