The Steepest Slope to a Quantum Few-Body Solution

Paolo Recchia, Debabrota Basu, Mario Gattobigio, Christian Miniatura, Stéphane BressanResearch Square 2024 · DOI

TL;DR

Researchers have introduced a family of Gradient Variational Methods (GVM) that replace traditional stochastic "trial-and-error" searches with gradient-based optimization (specifically the Adam optimizer) to solve the Schrödinger equation for few-body systems. Their hybrid approach significantly outperforms the state-of-the-art, offering more stable and efficient solutions for clusters of up to six particles.

What Problem Does This Paper Address?

In quantum mechanics, the Schrödinger equation is the "holy grail," but it is notoriously difficult to solve. While we can solve it analytically for two particles (like a hydrogen atom), adding even a third particle makes an exact solution impossible. Physicists rely on the Stochastic Variational Method (SVM), which uses a "gambling" approach: it randomly generates parameters for a set of basis functions and keeps the ones that result in the lowest energy.

However, as the number of particles ($N$) grows, the search space for these parameters expands quadratically, leading to the "curse of dimensionality." SVM becomes computationally expensive, prone to numerical instabilities (singularities), and often gets stuck in local minima. This paper asks: Can we stop gambling and start following the gradient?

Methods

The authors propose moving from stochastic search to Gradient-Based Optimization. They utilize the Adam optimizer—a staple in machine learning—to tune the nonlinear parameters of correlated Gaussian basis functions.

They introduce several variations:

  • Baseline GVM: Optimizes a fixed batch of basis functions all at once.
  • Static-GVM: Iteratively adds and optimizes small batches while keeping previous ones fixed.
  • Dynamic-GVM: Optimizes all previous batches simultaneously every time a new batch is added.
  • Hybrid SVM-GVM: Uses SVM as a "warm-up" to find a decent starting point, then lets Adam fine-tune the entire set.

To handle numerical "singularities" (where basis functions become too similar, making the math break down), they developed a clever workaround: if a singularity is detected, they slightly perturb the gradient using a normal distribution to "nudge" the optimizer into a safer region.

Datasets

The methods were evaluated on clusters of Helium (He) atoms, ranging from two-body (dimer) to six-body systems.

  • These systems are modeled using Gaussian-shaped two-body and three-body potentials.
  • Helium clusters are ideal benchmarks because they exhibit complex phenomena like the Efimov effect (a range of universal quantum states) and have well-documented experimental and theoretical energy values for comparison.

Key Insights and Ideas

  • Following the Slope vs. Jumping Blindly: The authors argue that randomly jumping into a parameter space (SVM) is "riskier" than smoothly following the steepest slope (GVM). This is why GVM is less prone to the singularities that plague SVM in small systems.
  • The Efficiency of Hybridization: The most successful approach wasn't purely gradient-based. The SVM-GVM hybrid proved best, suggesting that stochastic search is good for global exploration, while gradient descent is superior for local refinement.
  • The "Edge of Stability": The paper discusses how high learning rates in gradient descent can cause energy oscillations. While this looks like instability, the optimizer can often still converge over long timescales—a phenomenon known in ML as "unstable convergence."
  • Scalability: As the number of particles increases from 4 to 6, the performance gap between SVM and GVM widens, proving that gradient methods scale much better to higher-dimensional quantum problems.

Key Results

  • Stability in Two-Body Systems: SVM frequently failed to predict the energy of the Helium dimer due to singularities (only a 7% success rate). In contrast, GVM-based methods achieved a 100% success rate.
  • Superior Accuracy for Large Clusters: For 4, 5, and 6-body systems, the hybrid SVM-GVM method consistently found lower (more accurate) ground-state energies than the standard SVM.
  • Quantifiable Gains: The improvement in average energy was approximately 0.3‰ for 4-body, 0.8‰ for 5-body, and a significant 8‰ for 6-body systems.
  • Speed: GVM methods reached lower energy states faster than SVM as the system complexity increased.

Limitations and Future Work

The authors acknowledge that gradient methods require careful hyperparameter tuning, particularly the learning rate. If the learning rate is too high, the system oscillates; if too low, it converges too slowly.

Future directions include: * Exploring alternative basis sets, such as wavelets or neural network ansaetze. * Refining the gradient calculations by incorporating analytical expressions for the gradients rather than numerical approximations. * Applying these methods to even more complex physics scenarios beyond identical bosons, such as systems with different particle masses or spins.

Summary updated

Abstract

<title>Abstract</title> Quantum few-body systems are deceptively simple. Indeed, with the notable exception of a few special cases, their associated Schrödinger equation cannot be solved analytically for more than two particles. One has to resort to approximation methods to tackle quantum few-body problems. In particular, variational methods have been proposed to ease numerical calculations and obtain precise solutions. One such method is the Stochastic Variational Method, which employs a stochastic search to determine the number and parameters of correlated Gaussian basis functions used to construct an ansatz of the wave function. Stochastic methods, however, face numerical and optimization challenges as the number of particles increases. We introduce a family of gradient variational methods that replace stochastic search with gradient optimization. We comparatively and empirically evaluate the performance of the baseline Stochastic Variational Method, several instances of the gradient variational method family, and some hybrid methods for selected few-body problems. We show that gradient and hybrid methods can be more efficient and effective than the Stochastic Variational Method. We discuss the role of singularities, oscillations, and gradient optimization strategies in the performance of the respective methods.
BibTeX
@inproceedings{recchia2024steepest,
  author = {Paolo Recchia and Debabrota Basu and Mario Gattobigio and Christian Miniatura and Stéphane Bressan},
  title = {{The Steepest Slope to a Quantum Few-Body Solution}},
  booktitle = {Research Square},
  year = {2024},
  doi = {10.21203/rs.3.rs-4907430/v1},
}
Figure