⚛️ quantum physics

Pseudospectral method for solving PDEs using Matrix Product States

This paper proposes a novel pseudospectral method that extends Hermite Distributed Approximating Functionals (HDAF) to Matrix Product States (MPS) to efficiently solve time-dependent PDEs, such as the Schrödinger equation, by achieving high accuracy and exponential memory advantages over traditional vector-based methods.

Original authors: Jorge Gidi, Paula García-Molina, Luca Tagliacozzo, Juan José García-Ripoll

Published 2026-03-18

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Jorge Gidi, Paula García-Molina, Luca Tagliacozzo, Juan José García-Ripoll

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are trying to predict the path of a single, invisible speck of dust floating in a room. In the quantum world, this isn't just a speck; it's a "wave" of possibilities that can stretch, shrink, and twist in complex ways. Scientists call this the Schrödinger equation, and solving it is like trying to predict the weather, but for a single particle that can be in two places at once.

The problem is that as this particle moves, the "map" needed to track it gets huge. Traditional computers try to draw this map by dividing space into tiny grid squares (like a chessboard). If the particle spreads out, the chessboard needs to get bigger and bigger. Eventually, the computer runs out of memory because the number of squares becomes too vast to handle. This is the "curse of dimensionality."

This paper introduces a clever new way to solve this problem using Matrix Product States (MPS) and a mathematical trick called HDAF. Here is the breakdown using everyday analogies:

1. The Old Way: The Giant Chessboard

Imagine you are tracking a drop of ink spreading in a glass of water.

Traditional Method: You take a photo of the glass and divide it into a million tiny pixels. To track the ink, you have to update the color of every single pixel. If the ink spreads to fill the whole glass, you need a massive grid. If the glass gets bigger, your computer crashes because it can't store all those pixels.
The Problem: In quantum physics, the "glass" can get so big that the number of pixels exceeds the number of atoms in the universe.

2. The New Way: The Smart Zipper (MPS)

Instead of storing every single pixel, the authors use Matrix Product States (MPS).

The Analogy: Think of the ink spreading not as a giant grid of pixels, but as a long, flexible zipper.
How it works: A zipper doesn't need to store the whole picture of the fabric at once. It only needs to know how the teeth connect to the next tooth. If the ink is smooth and predictable, the "zipper" stays short and simple. If the ink gets messy and chaotic, the zipper gets a bit longer, but it never needs to become a giant grid.
The Benefit: This allows the computer to track the particle even when it spreads out over a huge distance, using a fraction of the memory. It's like compressing a 4K video into a tiny file without losing the important details.

3. The Secret Sauce: HDAF (The Magic Lens)

To make this zipper work, you need a way to calculate how the particle moves (its speed and acceleration).

The Old Lens (Finite Differences): Imagine trying to guess the slope of a hill by looking at two very close points. If the points are too close, your ruler gets shaky (math errors). If they are too far apart, you miss the curve. It's a clumsy way to measure change.
The New Lens (HDAF): The authors use Hermite Distributed Approximating Functionals (HDAF).
- The Analogy: Think of HDAF as a magic lens or a smart filter. Instead of just looking at two points, it looks at a small neighborhood of points and uses a special mathematical "recipe" (based on bell curves and polynomials) to guess the shape of the hill perfectly.
- The Result: It is incredibly accurate. It can see the smooth curves of the quantum wave much better than the old "clunky ruler" method, and it does it without needing a super-computer.

4. The Race: Who Wins?

The authors tested four different "racing strategies" (algorithms) to see how fast and accurate they could move the particle forward in time:

Runge-Kutta: A standard, reliable runner.
Crank-Nicolson: A steady, cautious walker.
Arnoldi: A sprinter that takes big leaps.
Split-Step: The winner.

Why Split-Step Won:
Imagine you are moving a heavy box.

The Old Split-Step: You have to stop, take the box apart, carry the pieces to a different room (using a Fourier Transform, which is like a complex translation step), put them back together, and move again. It's accurate but slow.
The New HDAF Split-Step: Because the "magic lens" (HDAF) is so good at understanding the physics, you don't need to take the box apart. You can just push it forward directly in its current form. It's like having a teleportation device that moves the particle instantly without the heavy translation step.

5. The Grand Experiment: The Double-Well

To prove this works, they simulated a particle in a "double-well" potential.

The Scenario: Imagine a ball in a bowl that has a small hill in the middle, splitting it into two valleys. The ball starts in one valley, gets a sudden push, and spreads out.
The Result: The ball doesn't just roll; it splits into two waves, one going left and one going right, creating a "quantum superposition" (being in two places at once).
The Outcome: Their new method (MPS + HDAF) successfully predicted this complex behavior, keeping the "zipper" (memory) small while the "wave" (particle) got huge. It showed that this method can handle real-world research problems, like levitating nanoparticles in labs.

The Bottom Line

This paper is about building a super-efficient, memory-saving engine for simulating quantum particles.

Old way: Use a giant grid that crashes your computer.
New way: Use a "smart zipper" (MPS) and a "magic lens" (HDAF) to track the particle with high precision but low cost.

It's a bridge between the complex world of quantum physics and the limited memory of our current computers, allowing scientists to simulate larger, more complex systems than ever before without needing a quantum computer.

1. Problem Statement

The paper addresses the challenge of solving time-dependent partial differential equations (PDEs), specifically the time-dependent Schrödinger equation, for systems with large spatial domains and bandwidth-limited functions.

The Bottleneck: Traditional numerical methods (like finite differences or standard vector-based spectral methods) suffer from the "curse of dimensionality." As the spatial domain expands (e.g., a particle spreading in a potential well), the number of grid points required grows exponentially, making storage and computation intractable on classical computers.
The Quantum Context: While quantum computers offer exponential compression via amplitude encoding, scalable fault-tolerant hardware is not yet available.
The Goal: Develop quantum-inspired algorithms that run on classical computers but leverage the exponential compression capabilities of quantum encodings (specifically Matrix Product States, or MPS) to solve PDEs efficiently.

2. Methodology

The authors propose a novel framework combining Matrix Product States (MPS) with Hermite Distributed Approximating Functionals (HDAF).

A. MPS/QTT Encoding

The wavefunction $\psi(x)$ is encoded using MPS (also known as Quantized Tensor Trains, QTT).
This representation maps $n$ qubits to $2^n$ grid points, allowing for exponential compression of the state vector. This is particularly effective for bandwidth-limited functions where Fourier coefficients decay rapidly.
Operators (Hamiltonians, derivatives) are encoded as Matrix Product Operators (MPOs).

B. HDAF for Differential Operators

Instead of using standard finite difference methods (which have low-order accuracy and require small grid steps) or Fourier transforms (which require global operations), the authors extend HDAF to the MPS framework.
HDAF Principle: HDAF approximates functions and their derivatives using a linear combination of Hermite polynomials weighted by a Gaussian filter. It acts as a "nascent" delta function.
Implementation:
- Derivatives and functions of derivatives (like the free propagator) are constructed as MPOs.
- The method uses displacement operators ( $\hat{\Sigma}_\pm$ ) to create sparse, banded MPOs with low bond dimensions.
- Key Advantage: HDAF provides pseudospectral accuracy (exponential convergence for smooth functions) without requiring global Fourier transforms, which are difficult to implement efficiently in the MPS coordinate basis.

C. Time Evolution Algorithms

The paper evaluates four classes of time-evolution algorithms using the HDAF-MPO operators:

Explicit Runge-Kutta: Euler, Improved Euler, and 4th-order RK.
Implicit Methods: Crank-Nicolson.
Krylov Subspace Methods: Explicitly restarted Arnoldi iteration.
Split-Step Method: Decomposes the evolution into kinetic (free propagator) and potential steps.
- Innovation: The HDAF formalism allows the free propagator to be applied directly in the coordinate basis, avoiding the need for Fourier Transforms (FFT) typically required in split-step methods.

3. Key Contributions

HDAF-MPO Construction: The authors successfully derived MPO representations for arbitrary differential operators and the free propagator using HDAF, achieving high accuracy with low bond dimensions.
Metaheuristics for Parameter Tuning: They developed rules for selecting HDAF parameters (polynomial order $M$ and Gaussian width $\sigma$ ) to minimize error and manage round-off errors in finite-precision arithmetic.
Benchmarking Framework: A comprehensive comparison of time-evolution algorithms (RK, Arnoldi, Split-Step) within the MPS-HDAF framework against traditional finite difference and FFT-based vector methods.
Physical Validation: Demonstration of the method on a physically relevant "quantum quench" problem (particle expansion) and a complex double-well potential scenario.

4. Results

The methods were tested on a quantum quench scenario where a particle initially confined in a harmonic potential is suddenly released into a wider potential, causing a massive spatial expansion (100-fold).

Accuracy vs. Cost:
- HDAF vs. Finite Differences: HDAF-MPOs achieved significantly higher accuracy (orders of magnitude lower error) for the same computational cost compared to finite difference MPOs.
- HDAF vs. FFT: The HDAF split-step method performed comparably to state-of-the-art FFT-based vector methods in terms of error scaling and run time.
Memory Efficiency:
- The MPS approach offered an exponential advantage in memory. While vector methods hit memory limits as the grid size increases, MPS maintained manageable memory usage by compressing the state.
- This allowed for simulations with $2^{20}$ ( $\approx 10^6$ ) grid points, which would be prohibitive for standard vector representations in high-dimensional extensions.
Algorithm Performance:
- The HDAF Split-Step method emerged as the optimal choice, offering the best balance between accuracy and computational cost.
- It successfully avoided the "chirping" (rapid phase oscillations) induced by particle acceleration, maintaining low bond dimensions ( $\chi$ ) comparable to the theoretical limits.
Double-Well Potential: The method successfully simulated a particle expanding in a double-well potential, reproducing the physical separation of the wavefunction into two peaks, demonstrating viability for non-trivial research scenarios.

5. Significance

Bridging the Gap: This work provides a robust classical alternative to quantum computing for simulating large quantum systems. It achieves "quantum-like" efficiency (exponential compression) without requiring quantum hardware.
Overcoming Limitations: By replacing Fourier transforms with HDAF in the split-step method, the authors removed a major bottleneck for MPS-based simulations, enabling efficient time evolution directly in the coordinate basis.
Scalability: The approach is highly scalable. The authors note that while current simulations are 1D, the exponential memory savings of MPS make this method uniquely suited for future extensions to multi-dimensional PDEs, where traditional grid-based methods fail completely.
Practical Application: The successful reproduction of complex dynamics (expansion in double-well potentials) validates the method for use in real-world research areas like optomechanics and levitodynamics, where large spatial domains and non-linear potentials are common.

In conclusion, the paper establishes MPS-HDAF as a powerful, high-accuracy, and memory-efficient framework for solving time-dependent PDEs, outperforming traditional finite difference methods and matching spectral methods while offering superior scalability for large systems.