Regularised density-potential inversion for periodic systems: application to exact exchange in one dimension

This paper presents a convex analysis-based, Moreau--Yosida regularised formulation of density-functional theory for periodic systems that ensures numerical stability against perturbations, demonstrating through a one-dimensional Hartree--Fock implementation that it is feasible to recover local Kohn--Sham potentials reproducing exact exchange effects.

The Gist

Imagine you are trying to solve a massive, complex jigsaw puzzle. In the world of quantum chemistry, this puzzle is an atom or a molecule. The pieces are electrons, and they are constantly buzzing around, repelling each other, and interacting in ways that are incredibly hard to calculate directly.

For decades, scientists have used a clever shortcut called Density Functional Theory (DFT). Instead of trying to track every single electron (the hard way), DFT says, "Let's just look at the density of the electrons—where they are most likely to be found." It's like looking at a crowd of people from a helicopter: you don't need to know the name and face of every single person; you just need to know where the crowd is thickest.

However, there's a catch. To use this shortcut, you need to know the "rules of the game" (the potential energy landscape) that creates that specific crowd pattern. Usually, we guess these rules. But what if we wanted to work backward? What if we knew exactly where the electrons should be (perhaps from a super-accurate, but slow, calculation) and wanted to figure out exactly what the "rules" (the potential) were that created that pattern?

This is called Inverse Kohn-Sham (iKS). It's like looking at a finished cake and trying to reverse-engineer the exact recipe and oven temperature that baked it.

The Problem: A Wobbly Recipe

The problem with this "reverse engineering" is that it's mathematically unstable. It's like trying to balance a pencil on its tip. If you make a tiny, almost invisible mistake in the shape of the cake (the electron density), the recipe you calculate might be completely wrong. In math terms, the problem is "ill-posed" and sensitive to noise.

The Solution: The "Regularized" Lens

The authors of this paper introduce a mathematical tool called Moreau-Yosida Regularization.

Think of this as putting on a pair of special glasses or using a smoothing filter on a photo.

• Without the glasses: The image is sharp but full of static and noise. If you try to trace the edges, your hand shakes, and you get a jagged, unusable line.
• With the glasses: The image is slightly softer, but the noise is gone. The lines are smooth and stable. You can trace them confidently.

In this paper, the "glasses" are a parameter (let's call it $\epsilon$).

1. Large $\epsilon$ (Heavy Glasses): The image is very smooth. You get a very stable, easy-to-find answer, but it might not look exactly like the original cake. It's a rough approximation.
2. Small $\epsilon$ (Light Glasses): The image is sharper. You get closer to the true recipe.
3. Zero $\epsilon$ (No Glasses): You get the perfect, exact recipe, but the math breaks down and becomes unstable.

The genius of this paper is showing how to slowly turn the dial from "Heavy Glasses" to "Light Glasses" in a controlled way, allowing the computer to find the perfect recipe without the math crashing.

The Specific Experiment: One-Dimensional Crystals

To test this, the authors didn't try to solve a real 3D molecule immediately (which is too messy). Instead, they built a 1D model—imagine a single line of atoms, like a string of pearls.

They used a specific type of interaction between the electrons called the Yukawa potential.

• Analogy: Imagine the electrons are magnets. In real life, magnets have a force that drops off quickly but can be tricky to calculate over long distances. The Yukawa potential is like a magnet that has a "soft" force field that fades out smoothly, making the math much friendlier for their "smoothing glasses."

They asked the computer: "Here is the exact electron density for a string of atoms. Can you find the local potential (the local rules) that creates this, specifically accounting for the 'exact exchange' (a tricky quantum effect where electrons avoid each other in a specific way)?"

The Results: A Proof of Concept

1. It Works: They successfully reversed the process. They took a known electron density and found the potential that created it.
2. Stability: They proved that even if they added a little bit of "noise" (a tiny error) to the input density, the output potential didn't go crazy. The "glasses" kept the solution stable. This is called non-expansiveness—errors don't get amplified; they stay small or get smaller.
3. The "Non-Representable" Density: They even tried feeding the computer a density that was physically impossible (like a negative amount of electrons in a spot). The system didn't crash; instead, it found the closest possible valid density. It's like asking a chef to bake a cake with -5 eggs; the chef just says, "Okay, I'll bake the closest thing I can that actually works."

Why Does This Matter?

This paper is a proof of concept. It's like building a working prototype of a new engine in a garage before putting it in a race car.

• For Chemists: It shows a new, mathematically rigorous way to find the "exact" rules of the game from the electron density.
• For the Future: If this works for simple 1D lines, the authors hope to scale it up to complex 3D materials. This could lead to better, more accurate computer simulations for designing new batteries, solar cells, and drugs, because we can finally reverse-engineer the "perfect" quantum rules from the "perfect" electron clouds.

In summary: The authors built a mathematical "stabilizer" that allows scientists to safely reverse-engineer the quantum rules of matter from electron patterns, proving that we can do this without the math falling apart, even when the input data is slightly imperfect.

Technical Summary

1. Problem Statement

The paper addresses the inverse Kohn–Sham (iKS) problem: determining the local effective potential ($v_s$) that yields a specific interacting ground-state electron density ($\rho_{gs}$) within Density-Functional Theory (DFT).

• Mathematical Challenges: The standard Hohenberg–Kohn mapping from density to potential is ill-posed. It lacks coercivity, is highly sensitive to perturbations (non-expansive), and relies on the unverified assumption of $v$-representability (that a given density can be generated by a non-interacting system).
• Periodic Systems: While DFT is widely used for periodic materials, rigorous mathematical formulations for density-potential inversion in periodic boundary conditions (Born–von Kármán) are scarce. Existing methods often struggle with numerical stability, convergence, and the handling of non-$N$-representable densities.
• Goal: To develop a mathematically rigorous, numerically stable framework for density-potential inversion in periodic systems, specifically demonstrating its ability to recover the exact exchange potential from a Hartree–Fock (HF) density.

2. Methodology

The authors propose a framework based on Moreau–Yosida (MY) regularisation applied to the convex analysis formulation of DFT.

A. Theoretical Framework

• Periodic Setting: The theory is formulated for $d$-dimensional systems with $p$ periodic directions. It utilizes Born–von Kármán boundary conditions and defines function spaces (Sobolev spaces $H^{-1}$ and $H^1$) suitable for periodic densities and potentials.
• Yukawa Interaction: To harmonize with the function spaces and ensure finite interaction energy, the electron-electron repulsion is modeled using a Yukawa potential ($e^{-\gamma r}/r$) rather than the Coulomb potential. This choice ensures the interaction operator is bounded within the chosen Sobolev spaces.
• MY Regularisation: The universal density functional $F(\rho)$ is regularized to create a differentiable, convex functional:
  $$F^\varepsilon(\rho_0) = \min_{\rho} \left( F(\rho) + \frac{1}{2\varepsilon} \|\rho - \rho_0\|^2_X \right)$$
  where $\varepsilon$ is the regularisation parameter and $\|\cdot\|_X$ is a specific norm.
• The Inversion Scheme:
  1. Given a target density $\rho_{ref}$ (e.g., from an interacting HF calculation), the method finds the proximal density $\rho_\varepsilon$ that minimizes the regularized functional.
  2. The Kohn–Sham potential is recovered via the limit:
     $$v_s = \lim_{\varepsilon \to 0^+} \frac{1}{\varepsilon} J(\rho_\varepsilon - \rho_{ref})$$
     where $J$ is the duality map (essentially a Yukawa-Hartree potential operator).
  3. Model Functional: To improve convergence, the authors introduce a model functional $F^{mod}_{\alpha, \xi}$ that includes the external potential ($v_{ext}$) and Hartree energy ($v_H$) as guidance terms with parameters $\alpha$ and $\xi$.

B. Numerical Implementation

• System: A 1D periodic system (spin-restricted Hartree–Fock level) implemented in the open-source code Sable.
• Basis: Plane-wave basis set with a momentum cutoff.
• Optimization:
  • Standard DIIS (Direct Inversion in the Iterative Subspace) methods failed to converge for small $\varepsilon$ due to the large Hartree term introduced by the penalty.
  • The authors implemented a robust second-order orbital rotation method with line search, using the density from a larger $\varepsilon$ as an initial guess for smaller $\varepsilon$ (continuation method).
• Process: The algorithm iteratively decreases $\varepsilon$ (increasing the penalty weight) to force the non-interacting system to reproduce the target density, extracting the local exchange potential $v_x$ in the limit.

3. Key Contributions

1. Rigorous Periodic Formulation: Provided a complete convex analysis formulation of DFT for periodic systems, defining appropriate dual spaces and norms that accommodate band structures and translation symmetrization.
2. MY Regularisation for Periodic Systems: Extended the MY regularisation framework to periodic boundary conditions, proving that the proximal mapping is non-expansive (perturbations in input density do not amplify in the output density).
3. Proof of Concept for Exact Exchange: Demonstrated that the scheme can successfully recover the local Kohn–Sham exchange potential that reproduces the effects of non-local exact exchange (from HF theory) in 1D.
4. Handling Non-$N$-Representability: Showed that the method naturally handles non-$N$-representable densities by converging to the closest $N$-representable density in the least-squares sense, rather than failing.
5. Error Analysis: Performed a detailed theoretical and numerical error analysis, quantifying how perturbations in the input density propagate through the regularized map to the potential.

4. Key Results

• Convergence: The method reliably converges for regularisation parameters as small as $\varepsilon \approx 10^{-8} - 10^{-7}$.
• Role of Model Parameters:
  • Including the external potential ($v_{ext}$) in the model functional ($\alpha=1$) drastically improves convergence, reducing density errors by an order of magnitude compared to excluding it.
  • Including the Hartree potential ($\xi=1$) had a negligible effect on accuracy but was crucial for SCF stability in some cases.
• Recovery of Exchange Potential: The recovered local exchange potential ($v_x$) accurately reproduces the HF density. The resulting Kohn–Sham band structure showed a band gap reduction (0.227 a.u. in HF vs. 0.187 a.u. in KS), consistent with the known effect of localizing exchange.
• Non-$N$-Representable Densities: When tested with a density containing negative regions (non-physical), the algorithm produced the "best" physical approximation (closest $N$-representable density), with the error plateauing at a non-zero constant as $\varepsilon \to 0$.
• Error Propagation: The analysis confirmed the non-expansive nature of the proximal map: $\|\tilde{\rho}_\varepsilon - \rho_\varepsilon\| \leq \|\tilde{\rho}_{ref} - \rho_{ref}\|$. The error in the potential scales linearly with $\varepsilon$ for small perturbations.

5. Significance

• Theoretical Validation: This work provides a mathematically rigorous "proof-of-principle" that density-potential inversion is feasible for periodic systems using MY regularisation, overcoming the ill-posedness of the standard inverse problem.
• Functional Development: By recovering the exact exchange potential from a known density, the method generates high-quality data for training and testing approximate exchange-correlation functionals (e.g., for range-separated hybrids or machine learning potentials).
• Robustness: The demonstration that the method works for non-$N$-representable inputs makes it a powerful tool for analyzing densities from correlated wavefunction methods where $v$-representability is not guaranteed.
• Future Outlook: While currently applied to 1D exact exchange, the framework is general. The authors suggest future work will apply this to correlated systems (post-HF) to recover full exchange-correlation potentials, potentially bridging the gap between high-accuracy wavefunction methods and efficient DFT.

In summary, the paper establishes a stable, mathematically grounded numerical pipeline for inverting densities to potentials in periodic systems, successfully recovering exact exchange effects and providing a robust tool for future DFT functional development.