🔢 mathematics

Guaranteed Lower Eigenvalue Bounds for Spectral Galerkin Methods with Application to Schrödinger Operators

This paper extends the framework for guaranteed lower eigenvalue bounds from finite element methods to conforming spectral Galerkin methods, providing rigorous two-sided spectral approximations for Schrödinger operators with both bounded and singular potentials while achieving high precision with significantly fewer degrees of freedom than traditional approaches.

Original authors: Xuefeng Liu

Published 2026-07-07

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Xuefeng Liu

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 find the exact pitch of a guitar string, but the string is made of a strange, wobbly material that changes its stiffness depending on where you pluck it. In the world of physics and math, this is like trying to calculate the energy levels of an electron in an atom (a "Schrödinger operator").

For a long time, mathematicians had a very good way to guess the highest possible pitch (an "upper bound"). It's like saying, "I'm sure the note is no higher than this." But finding a guaranteed lowest possible pitch (a "lower bound") was much harder. The old methods were like saying, "It's definitely higher than X, but only if you already know the answer to a completely different, unrelated question." If you didn't have that extra information, you were stuck.

This paper introduces a new, self-contained way to find that guaranteed lower bound for spectral methods (a high-precision way of solving these problems). Here is how it works, using some everyday analogies:

1. The "Perfect" vs. The "Real" Problem

Think of the "perfect" problem as a simple, empty room where sound waves bounce around. We know exactly how the sound behaves there.

The Paper's First Trick: If you try to solve the problem using only the "perfect" sound waves (eigenfunctions) as your building blocks, the math gives you a perfect, closed-form answer. It's like having a ruler that is perfectly calibrated. The paper proves that if you use these perfect blocks, you can calculate a "safety margin" (a constant) that tells you exactly how far off your answer might be, without needing any outside help.

2. The "Wobbly" Problem (The Potential)

Now, imagine you put a heavy, uneven rug in the middle of that room. This rug represents the "potential" (the $V$ in the Schrödinger equation). The sound waves now interact with the rug, making the math messy. The perfect sound waves from the empty room are no longer the perfect solution.

The Old Way: To get a lower bound here, you had to guess how heavy the rug was. If the rug was huge, your guess had to be huge, and your "safety margin" became so wide it was useless.
The Paper's Second Trick (The Projection Gap): The author introduces a clever "gap" measurement. Imagine you have two ways of looking at the room:
1. The Blindfolded View: You ignore the rug and just look at the empty room's waves.
2. The Real View: You look at the room with the rug.
  The paper calculates the "gap" between these two views. By measuring how much the rug distorts the waves relative to the empty room, they can create a new, tighter safety margin. This works well if the rug isn't too heavy.

3. The "Heavy Rug" Problem (The Composite Method)

What if the rug is massive? Like a mountain of sand? The previous trick fails because the distortion is too big.

The Paper's Third Trick (The Composite Discretisation): This is the paper's biggest innovation. Instead of trying to measure the whole heavy rug at once, they build a "fake, lighter rug" that sits underneath the real one.
- Think of it like this: You want to know the weight of a giant, irregular boulder. Instead of weighing the whole thing (which is hard), you build a smaller, perfect cube that fits inside the boulder. You know the weight of that cube exactly.
- The paper creates a mathematical "cube" (a simplified version of the potential) that is guaranteed to be lighter than the real potential. They then solve the problem for this lighter cube. Because this lighter cube is simpler, the math gives them a very tight, accurate lower bound.
- Finally, they add a small correction factor to account for the difference between the "cube" and the "boulder."
- The Result: Even with a massive, heavy potential, they can get a very precise lower bound without the answer getting "inflated" by the size of the potential.

4. Why This Matters (The "Spectral" Advantage)

The paper compares this new method to the old standard, which is like using a pixelated grid (Finite Element Methods) to measure the room.

The Grid Method: To get a better picture, you have to add more and more tiny pixels. It's slow, and the "safety margin" shrinks slowly.
The Spectral Method (This Paper): This method uses smooth, global waves (like a whole sine wave) instead of pixels. It's like using a high-definition lens.
The Payoff: The paper shows that with this new method, they can get the same (or better) accuracy as the grid method, but using 100 times fewer data points. It's like getting a 4K movie resolution using only the data required for a 480p video.

5. The "Coulomb" Bonus

The paper mentions that this same "gap" logic can be extended to handle "singular" potentials—mathematical objects that are infinitely sharp or heavy at a single point (like the center of an atom, where an electron feels a massive pull).

The Analogy: Imagine a black hole in the middle of the room. The old methods would break because the gravity is infinite. This new method uses a special "shield" (a Hardy inequality) to handle the infinite spike, allowing them to calculate the energy levels of atoms like Hydrogen with guaranteed accuracy.

Summary

In short, this paper solves a decades-old problem: How do we guarantee that our calculated energy levels aren't too low?

They did it by:

Finding a perfect "safety margin" for simple cases.
Creating a "gap" measurement to handle moderate complexity.
Inventing a "composite" strategy (using a simplified under-approximation) to handle massive complexity without losing precision.

The result is a way to calculate quantum energy levels that is incredibly fast, highly accurate, and mathematically "certified" to be correct, requiring far less computing power than previous methods.

Technical Summary: Guaranteed Lower Eigenvalue Bounds for Spectral Galerkin Methods with Application to Schrödinger Operators

Problem Statement
Spectral Galerkin methods are widely recognized for their high-precision approximation of eigenvalues, particularly for smooth problems where they exhibit spectral (super-algebraic) convergence. However, a rigorous, computable lower bound for these eigenvalues has remained elusive within the spectral framework. While classical enclosure methods (e.g., Kato, Weinstein–Temple, Lehmann–Goerisch) can provide lower bounds, they typically require a priori information about neighboring eigenvalues or specific problem structures. Conversely, existing guaranteed lower-bound frameworks, such as those based on finite element methods (FEM), rely on local mesh geometry and interpolation error estimates (constants $C_h$ ) that do not have a direct spectral counterpart. This paper addresses the gap by establishing a rigorous, unconditional lower-bound framework for conforming spectral Galerkin methods applied to Schrödinger operators ( $-\Delta + V$ ).

Methodology and Framework
The paper builds upon an abstract Hilbert-space framework (originally developed by Liu [16, 17]) which relates a discrete Galerkin eigenvalue $\lambda_{k,N}$ to the exact eigenvalue $\lambda_k$ via a projection error constant $C_N$ . The core inequality is:
$\lambda_k \ge \frac{\lambda_{k,N}}{1 + C_N^2 \lambda_{k,N}}$
The methodology proceeds through four distinct mechanisms to determine $C_N$ for spectral spaces:

Exact Eigenbasis Spaces: For trial spaces spanned by exact eigenfunctions of the operator, the paper derives a closed-form, optimal projection constant $C_N = \lambda_{M+1}^{-1/2}$ , where $\lambda_{M+1}$ is the first omitted eigenvalue. This is established via a Parseval argument, independent of mesh geometry.
Projection-Gap Mechanism (Standard Matrix): For Schrödinger operators with non-negative potentials ( $V \ge 0$ ) where the trial space (e.g., trigonometric polynomials) is not an eigenbasis, the standard Ritz projection $P_N$ differs from the Laplacian modal projection $P_N^0$ . The paper introduces a "projection-gap" analysis, controlling the difference $e = P_N^0 u - P_N u$ using an auxiliary discrete solution operator. This yields a constant $C_N^{(V)}$ dependent on a potential form factor $\eta_V$ . For bounded potentials, this factor initially depends on $\|V\|_{L^\infty}$ , leading to a constant $C_N^{(V)} = \sqrt{(1 + \min(X^2, X))/\nu^*}$ , where $X = \|V\|_{L^\infty}/\lambda_{1,N}$ and $\nu^*$ is the spectral gap of the Laplacian.
Composite Discretisation (Removing $\|V\|_{L^\infty}$ Dependence): To eliminate the inflation of the constant caused by large potentials (common in domain truncation for unbounded domains), the paper proposes a composite discretisation. This involves:
- Constructing a bandlimited under-approximation $W$ of the potential $V$ .
- Defining a reduced-degree weighted projection $\Pi_m^{\tilde{W}}$ onto a subspace $V_m$ .
- Utilizing a "slaving identity" (an exact aliasing property) where the weighted projection annihilates the interpolation residual of the higher-degree space.
  This construction restores the sharp, potential-independent constant $C_N = \nu^{*-1/2}$ , effectively decoupling the bound quality from the magnitude of the potential.
Domain Truncation: For operators on $\mathbb{R}^d$ , the paper combines the spectral lower bounds on a truncated box $\Omega$ with Neumann boundary conditions and a posteriori confinement checks (comparing the computed upper bound to the exterior potential minimum $\sigma(\Omega)$ ) to produce certified two-sided bounds on the whole space.

Key Contributions

Optimal Spectral Constant: The derivation of the closed-form constant $C_N = \lambda_{M+1}^{-1/2}$ for eigenbasis trial spaces, which serves as the theoretical benchmark.
Projection-Gap Analysis: The extension of the auxiliary-operator technique to spectral methods, providing explicit constants for the standard Galerkin matrix of Schrödinger operators with bounded potentials.
Composite Spectral Method: The development of a composite discretisation that achieves the optimal spectral-gap constant without dependence on $\|V\|_{L^\infty}$ , overcoming the limitations of the standard matrix approach for confining potentials.
Extension to Singular Potentials: A preview of how the auxiliary-projector mechanism extends to singular potentials (e.g., Coulomb singularities) via a localized Hardy inequality, removing the requirement for bounded $L^\infty$ norms.
Two-Sided Enclosures: The provision of fully computable, certified two-sided bounds for Schrödinger operators on $\mathbb{R}^d$ using Neumann domain truncation.

Numerical Results
The paper validates the theory on two benchmark potentials in two dimensions: a ring potential $V_1 = (|x|^2-1)^2$ and an anisotropic double well $V_2 = (x_1^2-1)^2 + x_2^2$ .

Efficiency: The spectral method achieves certified bounds comparable to or surpassing those of certified Crouzeix–Raviart enriched (CECR) finite element methods (from [18]) using approximately two orders of magnitude fewer degrees of freedom (e.g., ~2,400 DOFs vs. ~400,000 DOFs).
Accuracy: For the ring potential, the spectral enclosure at $N=48$ is strictly inside the FEM enclosure. For the double well, the spectral bounds are comparable to FEM bounds at $N=48$ .
Mechanism Comparison: In regimes with moderate potentials, the standard auxiliary-projector method (Theorem 4.9) is efficient and nearly optimal. However, for large confining potentials (where $X \gg 1$ ), the standard method's gap inflates significantly (e.g., 72% gap vs. 0.48% for the composite method). The composite method (Theorem 4.12) maintains the optimal convergence rate $O(N^{-2})$ regardless of potential magnitude.
Tail Refinement: The paper demonstrates that a "tail-refined" form factor (Theorem 4.5), computable from the standard matrix, significantly improves the standard method's bounds (reducing the gap from 72% to 7.7% for $V_1$ ), though it remains inferior to the composite method for very large potentials.

Significance and Claims
The paper claims to resolve the long-standing issue of obtaining rigorous lower bounds directly from spectral discretisations without a priori spectral information. By extending the projection-based framework to spectral methods, it provides:

Unconditional Bounds: Bounds that do not require knowledge of neighboring eigenvalues.
Optimal Constants: A constant determined by the spectral gap of the Laplacian rather than mesh-dependent interpolation constants.
High Efficiency: The ability to achieve certified accuracy with significantly fewer degrees of freedom compared to low-order FEM, leveraging the spectral convergence of the upper bounds.
Robustness: The composite method ensures that the certification quality does not degrade for large or confining potentials, a critical requirement for quantum chemistry applications involving domain truncation.

The authors note that while the upper bounds converge spectrally, the certified lower bounds close at an algebraic rate $O(N^{-2})$ . The transfer of exponential accuracy to the lower bounds remains an open problem. The extension to singular potentials (e.g., Coulomb systems) is acknowledged as a subject for a companion paper.