Shaping Maximally Localized Wannier Functions via Discrete Adiabatic Transport

This paper introduces a non-variational, deterministic algorithm for constructing Maximally Localized Wannier Functions by unifying gauge smoothing with the projected position operator eigenvalue problem through discrete adiabatic transport, thereby eliminating the need for iterative spread minimization while revealing the geometric origin of mesh-dependent spread scaling in systems like graphene.

The Gist

The Big Picture: Organizing a Messy Crowd

Imagine you are trying to organize a massive, chaotic crowd of people (electrons) inside a giant, repeating city grid (a crystal). Your goal is to group these people into small, tight-knit neighborhoods (called Wannier Functions) that are as compact as possible.

In the world of physics, the standard way to do this is like trying to find the perfect arrangement by guessing, checking, and adjusting thousands of times. You tweak the positions slightly, see if the crowd gets tighter, and repeat. This is a "variational" method—it's like searching for the bottom of a valley in the dark by feeling your way down. It works, but it can be slow, and sometimes you get stuck in a local dip that isn't the true bottom.

This paper proposes a new, smarter way. Instead of guessing and checking, the authors built a "deterministic" machine. It's like having a GPS that tells you exactly which way to walk to get to the center, step-by-step, without needing to guess.

The Core Idea: The "Adiabatic Transport" Elevator

The authors' method relies on a concept called Discrete Adiabatic Transport.

• The Analogy: Imagine the electrons are passengers on a train moving through a tunnel. The tunnel has different sections (energy bands). Sometimes, the tracks merge or split (degeneracies).
• The Old Way: If you just look at the tracks locally, you might get confused about which passenger belongs to which train car when the tracks cross. You might swap passengers by mistake, creating a messy, jumbled neighborhood.
• The New Way: The authors use a "smooth elevator" (adiabatic transport). As the train moves, this elevator gently carries the passengers from one section of the track to the next, ensuring they stay in the correct order and don't get swapped. It "peels" the layers of the crowd apart smoothly, even when the tracks get messy.

By doing this, the "phase" (the internal rhythm or timing) of the electrons becomes a straight, flat line instead of a jagged, bumpy one.

The "Sinc-Loop": A Self-Correcting Compass

Once the crowd is smoothed out, the authors need to find the exact center of each neighborhood.

• The Old Way: You would calculate a "spread score" (how messy the neighborhood is) and try to minimize it. This is like trying to find the center of a room by measuring the distance to every wall and hoping the numbers get smaller.
• The New Way: The authors discovered a mathematical trick called the "sinc-loop."
  • The Analogy: Imagine you are trying to find the center of a room, but you have a special compass. You point the compass, it tells you "You are off by X amount," you move X amount, and the compass tells you again.
  • The paper shows that if you follow this compass, it doesn't just wander; it locks onto the center with incredible speed (mathematically, it converges cubically). You don't need to calculate a "messiness score" to know you're getting closer; the compass is the solution.

The Big Discovery: Why Graphene is "Frustrated"

The authors tested their method on Graphene (a material made of a single layer of carbon atoms shaped like a honeycomb).

• The Problem: When other scientists tried to calculate the size of these neighborhoods in Graphene using a very fine grid (high resolution), the neighborhoods seemed to get larger as the grid got finer. This was confusing. Usually, a finer grid gives a more precise answer, not a bigger error.
• The Paper's Explanation: The authors realized this wasn't a mistake or a computer glitch. It was a fundamental geometric truth.
  • The Analogy: Imagine trying to lay a flat sheet of paper over a ball. You can't do it perfectly without crumpling the edges. The "crumpling" (geometric frustration) has to go somewhere.
  • In 2D materials like Graphene, the math forces this "crumpling" to pile up along the very edges of the grid (the boundary seam).
  • Because the "crumpling" is stuck on the edge, and the edge gets longer as you make the grid finer, the total "messiness" (spread) grows linearly with the size of the grid.

The Takeaway: The authors didn't just fix the calculation; they proved why the calculation behaves this way. They showed that the "messiness" is an intrinsic feature of the geometry of the material, forced to accumulate on the boundary because the rules of the universe (non-commuting position operators) prevent it from being smoothed out everywhere at once.

Summary of the Workflow

1. Smooth the Crowd: Use the "elevator" (adiabatic transport) to move electrons smoothly across the grid, preventing them from getting swapped at crossing points.
2. Align the Rhythm: This smoothing makes the internal timing of the electrons a straight line.
3. Find the Center: Use the "Sinc-Loop" compass to pinpoint the exact center of the neighborhood using simple, repetitive steps.
4. Reveal the Truth: The method clearly shows that in 2D materials, the "messiness" is forced to the edges, explaining why the size of the neighborhoods appears to grow with grid resolution.

In short, the paper replaces a slow, guessing game with a direct, step-by-step construction kit that not only builds the neighborhoods faster but also reveals the hidden geometric rules that govern how they behave.

Technical Summary

Technical Summary: Shaping Maximally Localized Wannier Functions via Discrete Adiabatic Transport

Problem Statement
Maximally Localized Wannier Functions (MLWFs) are a standard tool in condensed matter physics for constructing localized basis sets, analyzing polarization, and modeling topological materials. The canonical approach, established by Marzari and Vanderbilt (MV), involves the variational minimization of a spread functional over a high-dimensional gauge manifold. While effective, this method relies on iterative global optimization. In two or higher dimensions, finding a globally smooth gauge is fundamentally obstructed by the non-commutativity of projected position operators, leading to "geometric frustration." Standard algorithms resolve this by distributing the unavoidable gauge mismatch across the entire Brillouin zone. Furthermore, in systems like graphene, this approach has been observed to yield spreads that scale linearly with the mesh resolution ($O(L)$), a phenomenon whose physical origin requires rigorous analytical isolation.

Methodology
The authors propose a non-variational, constructive algorithm that unifies gauge smoothing with the eigenvalue problem of the projected position operator (equivalently, the projected translation operator $e^{i\delta\mathbf{k}\cdot\hat{x}}$). The core of the methodology consists of three distinct components:

1. Discrete Adiabatic Transport (Band Peeling): Instead of variational disentanglement, the authors employ a deterministic procedure based on Kato's adiabatic theorem. They transport the periodic parts of Bloch functions across band degeneracies using a projection operator. This "band peeling" process aligns the Bloch functions such that they exhibit an approximately linear phase dependence across the Brillouin zone, effectively flattening the gauge along chosen one-dimensional strings. This process is shown to be formally equivalent to the eigenvalue equation of the projected translation operator to first order in the mesh spacing.
2. Deterministic Center Fixing (Sinc-Loop): In the transport-aligned gauge, the Wannier centers are not found by minimizing a spread functional. Instead, the authors derive a trigonometric equation relating the Wannier center to the overlap of Bloch functions. This equation is solved via a deterministic fixed-point iteration, termed the "sinc-loop," which converges cubically. This replaces the variational search with a self-consistent update of explicit center equations and projected-position matrices.
3. Sequential Extraction: For composite bands, the algorithm sequentially extracts individual MLWFs. By flattening the interior gauge along specific directions, the method isolates the geometric frustration inherent in 2D systems rather than globally distributing it.

Key Results
The paper validates the framework through benchmark calculations in one-dimensional (1D) and two-dimensional (2D) systems:

• 1D and 2D Square Potential: In 1D systems and a 2D square potential model, the method yields Wannier centers, orbital shapes, and spreads in excellent agreement with standard variational minimization schemes (such as those implemented in WANNIER90) and direct spatial matrix solvers. The overlap phases of the transported Bloch functions are shown to be approximately planar, consistent with the theoretical minimum spread condition.
• Graphene Analysis: The method is applied to a five-band subspace of graphene. The results confirm that the orbital shapes and Wannier centers align with standard results. However, the study rigorously analyzes the spread scaling. The authors demonstrate that the observed $O(L)$ mesh-dependent spread scaling is not a numerical artifact.
  • The sequential extraction forces the gauge to be flat along 1D strings, pushing the geometric frustration entirely onto a one-dimensional boundary seam.
  • The local gauge defects on this seam remain finite ($O(L^0)$), but because the number of grid points along the seam scales as $O(L)$, the macroscopic integration of these defects results in a linear divergence of the total spread.
  • This confirms that the $O(L)$ scaling is an intrinsic geometric manifestation of the non-commutativity of projected position operators in 2D systems.

Significance and Claims
The paper claims to provide a transparent, analytical framework for constructing MLWFs that avoids the variational complexities of the standard approach. By treating gauge alignment and center determination as a deterministic sequence of fixed-point iterations, the method offers a different perspective on the geometric obstructions in 2D systems.

The primary significance lies in the rigorous isolation of the physical origin of the $O(L)$ spread scaling in graphene. The authors demonstrate that this scaling is a direct consequence of forcing a 1D gauge flattening in a 2D system, which concentrates geometric defects on the boundary seam. While standard variational methods distribute these defects to minimize the total spread, the constructive approach presented here mathematically forces the accumulation of frustration, thereby revealing the intrinsic geometric nature of the non-commuting projected position operators. The work does not claim to replace variational methods for all applications but offers a constructive alternative that provides deeper analytical insight into the geometric structure of Wannier functions.