Coupling of the continuum and semiclassical limit. Part I: convergence of eigenvalues

This paper establishes the convergence of eigenvalues from a discretized $d$-dimensional Schrödinger operator to its continuum counterpart under a coupled semiclassical limit governed by a mesh spacing and parameter scaling, while also fully characterizing the spectral asymptotics for the harmonic oscillator across all possible scaling regimes.

The Gist

The Big Picture: Bridging Two Worlds

Imagine you are trying to understand how a smooth, flowing river (the Continuum) behaves when you look at it through a grid of square tiles (the Discrete).

In physics, we often model the universe as smooth and continuous (like water). But computers and digital simulations can only handle "pixels" or a grid of points (like a mosaic). The big question this paper asks is: If we make our grid finer and finer, does the mosaic eventually look exactly like the smooth river?

Usually, the answer is "yes," but only if you change the rules of the game at the same time. This paper is about finding the perfect recipe for changing those rules so that the digital simulation perfectly mimics the real, smooth physics.

The Characters in Our Story

1. The Smooth River (Continuum): This is the real-world physics described by the Schrödinger equation. It represents how particles (like electrons) move in a smooth landscape.
2. The Mosaic (Discrete): This is the computer model. It breaks space into a grid of points. The particles can only sit on the dots, not in between.
3. The Zoom Knob ($N$): This is how fine our grid is. A small $N$ means big, chunky tiles. A huge $N$ means tiny, microscopic tiles.
4. The Energy Scale ($\lambda$): This is a "volume knob" for the energy of the system. Turning it up makes the particles move faster and behave more like classical objects (like billiard balls) rather than fuzzy waves.

The Secret Sauce: The "Coupling"

The authors discovered that you can't just make the grid finer ($N$ goes up) while keeping the energy scale fixed. If you do that, the simulation breaks.

Instead, you have to turn the Zoom Knob ($N$) and the Energy Scale ($\lambda$) at the same time, in a very specific relationship. They call this relationship $\gamma$ (gamma).

Think of $\gamma$ as a dial on a mixing board. Depending on where you set this dial, you get a completely different result:

1. The "Sweet Spot" ($\gamma$ is between -1 and 1)

The Analogy: Imagine you are watching a high-definition movie. As you zoom in (making the pixels smaller), you also increase the brightness and contrast just enough so the image stays sharp.
The Result: This is the main discovery of the paper. In this range, the digital mosaic perfectly matches the smooth river. The energy levels (notes) the particles sing in the digital world are exactly the same as in the real world. The authors proved that if you tune your dial to this "Sweet Spot," your computer simulation is a valid representation of reality.

2. The "Too Fuzzy" Zone ($\gamma > 1$)

The Analogy: Imagine you are zooming in on a photo, but you forget to increase the resolution. The pixels get smaller, but the image becomes blurry and loses all its detail.
The Result: The energy of the particles drops to zero. The simulation forgets about the "hills and valleys" of the landscape and just acts like a flat, empty space. It's a boring, useless simulation.

3. The "Too Rigid" Zone ($\gamma < -1$)

The Analogy: Imagine you are zooming in, but you turn the brightness up so high that the image burns out. The particles get so energetic that they stop caring about the smooth landscape and just bounce off the individual grid points like pinballs on a pegboard.
The Result: The smooth river disappears entirely. The particles only "feel" the specific points where they are sitting. The physics becomes purely digital and loses its connection to the smooth, continuous world.

4. The "Edge Cases" ($\gamma = 1$ and $\gamma = -1$)

These are the transition points.

• At $\gamma = 1$, the simulation matches the smooth river, but without the "quantum fuzziness" (the semiclassical part).
• At $\gamma = -1$, the simulation is stuck in a weird middle ground where it's neither fully smooth nor fully rigid.

The "Harmonic Oscillator" Test Drive

To prove their theory, the authors used a classic physics problem called the Harmonic Oscillator.

• The Metaphor: Think of a ball rolling back and forth in a smooth, U-shaped bowl.
• The Test: They tried to simulate this ball on their digital grid using different settings for their "dial" ($\gamma$).
• The Finding: They showed mathematically that only when the dial is in the "Sweet Spot" does the digital ball roll with the exact same rhythm and energy as the real ball in the real bowl. In all other zones, the rhythm is wrong.

Why Does This Matter?

This paper is like a user manual for physicists and engineers who want to simulate quantum mechanics on computers.

• For Scientists: It tells them exactly how to set up their simulations so they don't get garbage results. If they want to study how electrons move in a new material, they now know the precise mathematical relationship between the grid size and the energy scale to use.
• For the Future: The authors mention this is "Part I." They have proven the energy levels match. In "Part II," they plan to prove that the shapes of the waves (how the particles spread out) also match.

Summary in One Sentence

The authors found the exact "recipe" for mixing a digital grid with high-energy physics, proving that if you tune the variables just right, a pixelated simulation can perfectly recreate the smooth, continuous laws of the quantum universe.

Technical Summary

1. Problem Statement

The paper investigates the spectral asymptotics of a $d$-dimensional discrete Schrödinger operator defined on a lattice $\mathbb{Z}^d$ as the lattice spacing vanishes and the semiclassical parameter diverges simultaneously.

The operator is given by:
$$H_N f(x) := -\frac{N^2}{2} \sum_{|x-y|=1} (f(y) - f(x)) + N^{2(1-\gamma)} V_N(x) f(x)$$
where:

• $N \in \mathbb{N}$ is a scaling parameter.
• The mesh spacing is $\delta_N = 1/N$.
• The potential is discretized as $V_N(x) = V(x/N)$.
• The semiclassical parameter is coupled to the mesh size via $\lambda_N = N^{1-\gamma}$, where $\gamma \in \mathbb{R}$.

The Core Question: How do the eigenvalues $E_n(H_N)$ behave as $N \to \infty$ for different values of the coupling parameter $\gamma$? Specifically, does the discrete operator converge to the semiclassical limit of the continuous operator $H_{cont} = -\frac{1}{2}\Delta_{\mathbb{R}^d} + \lambda_N^2 V$?

2. Methodology

The authors employ a combination of variational methods, asymptotic analysis, and spectral localization techniques.

• Test Functions (Upper Bounds): To establish upper bounds on eigenvalues, the authors construct approximate eigenfunctions using Hermite polynomials. These functions are scaled to match the local behavior of the potential near its minima. The Min-Max Principle is then applied to these test functions.
• IMS Localization (Lower Bounds): To prove lower bounds, the authors utilize the IMS (Ismagilov-Morgan-Simon) localization formula. This technique allows them to decompose the operator into localized regions (intervals or cubes) around the potential minima and a remainder term. By comparing the discrete operator to a modified harmonic oscillator on these localized domains, they derive rigorous lower bounds.
• Agmon-Allegretto-Piepenbrink Theorem: For the lower bound analysis of the harmonic oscillator, the authors use this theorem, which relates the existence of positive superharmonic functions to the ground state energy. They construct specific intervals based on the zeros of Hermite polynomials to apply this criterion effectively.
• Tensorization: For dimensions $d > 1$, the results are derived by tensorizing the one-dimensional results, leveraging the separability of the harmonic oscillator and general potentials with non-degenerate minima.

3. Key Contributions and Results

The paper's primary contribution is the complete characterization of the spectral asymptotics across all possible values of the coupling parameter $\gamma$. The authors identify five distinct asymptotic regimes:

A. The Main Result: The Semiclassical-Continuum Coupling ($\gamma \in (-1, 1)$)

This is the central focus of the paper. The authors prove that if $\gamma \in (-1, 1)$, the discrete operator converges to the semiclassical limit of the continuous operator.

• Theorem 1.2: For a potential $V$ satisfying standard assumptions (smooth, non-negative, non-degenerate minima, strictly positive at infinity), the eigenvalues satisfy:
  $$\lim_{N \to \infty} \frac{E_n(H_N)}{\lambda_N} = e_n(V)$$
  where $e_n(V)$ are the eigenvalues of the continuous semiclassical operator $-\frac{1}{2}\Delta_{\mathbb{R}^d} + \lambda_N^2 V$.
• Significance: This rigorously validates the specific scaling relation $\lambda_N = N^{1-\gamma}$ as the correct bridge between the discrete lattice model and the continuous semiclassical limit. It confirms that the discrete approximation captures the classical limit of the continuous model.

B. The Harmonic Oscillator Analysis

The authors first solve the problem for the harmonic oscillator potential ($V(x) \propto |x|^2$) in detail (Section 2).

• They prove that for $\gamma \in (-1, \infty)$, the scaled eigenvalues converge to the standard harmonic oscillator energy levels $e_n = \frac{1}{2}\sum \omega_i(2n_i+1)$.
• They establish precise error estimates using the properties of Hermite polynomials and the IMS localization formula to handle boundary effects in the discrete setting.

C. Other Scaling Regimes

The paper extends the analysis beyond the semiclassical domain to classify four other regimes:

1. $\gamma > 1$: The potential term becomes negligible compared to the kinetic term. The limit is the free discrete Laplacian (purely continuous spectrum), and eigenvalues tend to zero.
2. $\gamma = 1$: The operator converges to the standard continuum Schrödinger operator $-\frac{1}{2}\Delta + V$ (without the large $\lambda$ factor). This corresponds to the norm-resolvent convergence studied in previous literature.
3. $\gamma = -1$: The system behaves as a purely discrete model. The eigenvalues scale as $N^2$ and converge to the values of the potential at lattice points ($E_n \to V(n)$).
4. $\gamma < -1$: This regime corresponds to a semiclassical approximation of a discrete model. The kinetic energy term vanishes relative to the potential.
   • The ground state energy scales as $N^{2|\gamma|}$ and converges to 0.
   • Excited states ($n \ge 1$) become degenerate (even and odd levels merge) and converge to the potential values at integer points: $E_{2n} \approx E_{2n-1} \approx \frac{\omega^2 n^2}{2}$.

4. Significance and Impact

• Unified Framework: The paper provides a unified understanding of how discrete lattice models approximate continuous quantum systems. It clarifies exactly when a discrete model recovers the semiclassical physics of the continuum and when it deviates into purely discrete or free-particle behaviors.
• Parameter Space Characterization: By mapping the behavior of eigenvalues across the entire real line of $\gamma$, the authors identify the precise "sweet spot" ($\gamma \in (-1, 1)$) where the discrete approximation is valid for semiclassical analysis.
• Foundation for Future Work: This paper (Part I) focuses on eigenvalues. The authors note that the regime $\gamma \in (-1, 1)$ is particularly important because it allows for the derivation of Agmon-type estimates for eigenfunctions (decay rates), which will be the subject of their companion paper (Part II).
• Methodological Rigor: The use of the IMS localization formula adapted to discrete lattices and the careful handling of the error terms in the discrete Laplacian expansion provide a robust toolkit for analyzing discrete spectral problems in the semiclassical limit.

In summary, the paper rigorously establishes that the specific coupling of the mesh size and the semiclassical parameter defined by $\lambda_N = N^{1-\gamma}$ with $\gamma \in (-1, 1)$ is the necessary and sufficient condition for a discrete Schrödinger operator to faithfully reproduce the semiclassical spectral asymptotics of its continuous counterpart.