Anderson Localization: A Floquet operator Krylov space perspective

This paper employs operator Krylov space methods within a Floquet framework to characterize Anderson localization and the Aubry-André transition by linking stroboscopic dynamics to an effective inhomogeneous Ising model, revealing distinct signatures such as Porter-Thomas distributions and multifractal scaling that differentiate localized, delocalized, and critical phases.

The Gist

Imagine you are trying to understand how a drop of ink spreads through a piece of paper. In physics, this is similar to studying how a particle (or information) moves through a material. Sometimes, the material is clean, and the ink spreads out smoothly. Other times, the paper is crumpled and full of obstacles, and the ink gets stuck in one spot. This "stuck" behavior is called Anderson Localization.

This paper introduces a new, clever way to study this problem using a mathematical tool called Krylov space. Think of Krylov space not as a physical place, but as a special "map" or "ladder" that physicists build to track how a system changes over time.

Here is the breakdown of what the authors did, using simple analogies:

1. The "Stroboscopic" Trick (Taking Snapshots)

Usually, when physicists study how things move, they watch the movie frame-by-frame in continuous time. The authors decided to try something different: they treated time like a stroboscope (like a flashing light at a concert). Instead of watching the smooth motion, they only looked at the system at specific, spaced-out moments (snapshots).

• Why do this? It turns out that looking at these "snapshots" makes the math much easier and faster to solve. It's like trying to understand a complex dance by watching a series of high-quality photos rather than trying to track every tiny muscle movement in real-time.
• The Result: They mapped the problem onto a "Floquet" model, which is like translating the dance into a different language where the steps are easier to count.

2. The "Krylov Ladder"

To analyze these snapshots, the authors built a "ladder" of operators (mathematical tools).

• The Seed: They start with one specific "seed" (like a single drop of ink).
• The Rungs: They ask, "If I wait one step, where is the ink? If I wait two steps, where is it?" Each answer becomes a new rung on their ladder.
• The Map: This ladder turns out to look exactly like a 1D Ising model (a chain of magnets). The authors realized that the complex quantum problem could be visualized as a single particle hopping along a chain of these magnets.

3. The Two Ways of Averaging (The "Recipe" Problem)

The materials they studied were "disordered," meaning they were full of random bumps and holes (like a bumpy road). To get a clear picture, they had to average the results over thousands of different random roads.

The paper discovered a crucial "recipe" difference:

• Method A (The Bad Recipe): Calculate the math for each bumpy road individually, then average the final numbers.
  • Result: This created a weird "dip" or hole in the data that didn't make physical sense. It was like averaging the taste of 100 different soups, but the math got confused and said the soup had a hole in the middle.
• Method B (The Good Recipe): First, average the "bumpy road" data itself (the autocorrelation), and then do the math.
  • Result: This produced a smooth, realistic spectrum. It turned out that for this specific problem, you must smooth out the noise before you build your ladder.

4. The Three States of Matter (Localized, Delocalized, and Critical)

The authors tested their method on two famous models: the Anderson Model (random disorder) and the Aubry-André Model (quasi-periodic disorder). They found three distinct behaviors:

• The Localized Phase (The Trap):

  • What happens: The particle gets stuck. It can't move far from where it started.
  • The Krylov View: On their "ladder," the wavefront of the particle stays right at the bottom rung. It doesn't climb up.
  • The Sound: The "spectrum" (the sound of the system) has sharp, distinct peaks, like a bell ringing.

• The Delocalized Phase (The Free Runner):

  • What happens: The particle spreads out freely across the whole system.
  • The Krylov View: The wavefront races up the ladder, moving ballistically (like a bullet).
  • The Sound: The spectrum is smooth and flat. Interestingly, the fluctuations in the data followed a Porter-Thomas distribution.
  • Analogy: This is a bit surprising because Porter-Thomas distributions usually appear in chaotic, complex systems (like a crowded room where everyone is shouting randomly). The authors found that even a simple, single-particle system acts like a chaotic crowd when it's delocalized.

• The Critical Point (The Edge):

  • What happens: The system is right on the edge between being stuck and being free.
  • The Krylov View: The wavefront spreads, but it does so in a "fractal" way—like a coastline that looks jagged no matter how much you zoom in.
  • The Sound: It shows a mix of behaviors, and the data suggests a "multifractal" scaling, meaning the complexity changes depending on how you look at it.

5. The "Renormalization" of the Ladder

As the authors climbed higher up their Krylov ladder (looking at longer times), they noticed something interesting about the "rungs" (the parameters of their math).

• The randomness of the rungs started to smooth out. The distribution of these parameters got narrower and narrower, approaching a "fixed point."
• Analogy: Imagine tuning a radio. At first, the static is loud and chaotic. As you turn the dial (recursion step), the static clears up, and you find a steady, clear frequency. The math naturally "renormalizes" itself, filtering out the noise as you go deeper.

Summary

The paper claims that by switching from continuous time to "stroboscopic" snapshots, physicists can build a more efficient and accurate map (Krylov space) to study how particles get stuck or move freely in disordered materials. They found that:

1. Order of operations matters: You must average the raw data before doing the complex math to get the right answer.
2. Simple can look complex: Even a single particle moving freely behaves like a chaotic crowd (Porter-Thomas distribution).
3. The map reveals the phase: You can tell if a system is "stuck" or "free" just by looking at how the wavefront travels up the Krylov ladder.

This work doesn't propose a new medical treatment or a new technology; rather, it refines the mathematical toolkit physicists use to understand the fundamental behavior of quantum matter.

Technical Summary

Technical Summary: Anderson Localization: A Floquet operator Krylov space perspective

Problem Statement
The paper addresses the problem of Anderson localization and the single-particle localization-delocalization transition in the Aubry-André (AA) model. While Operator Krylov space methods have emerged as a powerful tool for studying operator spreading in quantum many-body systems (including Hamiltonian, unitary, and dissipative dynamics), the authors identify specific computational and conceptual advantages in studying these problems via stroboscopic (discrete-time) dynamics rather than continuous-time evolution. The central challenge is to map the dynamics of a seed operator under a Hamiltonian $H$ to an effective operator Krylov space description that allows for efficient computation of spectral functions and the extraction of localization properties.

Methodology
The authors employ a Floquet operator Krylov space approach. Instead of analyzing the continuous time evolution $e^{-iHt}$, they discretize time into steps $t = nT$, generating a unitary operator $U = e^{-iHT}$. This maps the problem to a stroboscopic Floquet dynamics framework.

1. Krylov Space Construction:

   • Arnoldi Iteration: The standard approach involves generating a Krylov basis $\{|O_n)\}$ via Gram-Schmidt orthogonalization of the sequence $(U^\dagger)^n O_0 U^n$. In this basis, the unitary $U$ takes an upper-Hessenberg form.
   • Floquet ITFIM Mapping: The authors utilize a specific basis transformation (equivalent to a Jordan-Wigner transformation) that maps the Krylov dynamics to a 1D Floquet Inhomogeneous Transverse Field Ising Model (ITFIM). In this representation, the unitary matrix becomes sparse and five-diagonal (a CMV matrix structure). The parameters of this effective model are the "Krylov angles" $\theta_j$, which are determined recursively.
   • Moment Method: To avoid the computationally expensive Gram-Schmidt procedure, the authors utilize a "moment method." This allows for the direct extraction of Krylov angles from the discrete-time autocorrelation function $A(nT) = (O_0 | (U^\dagger)^n O_0 U^n)$ without constructing the full orthogonal basis.

2. Spectral Function Estimation:

   • The spectral function is computed using the Bernstein–Szegő approximation, which utilizes the orthogonal polynomials on the unit circle (OPUC) generated by the Krylov angles. This is compared against the standard truncated Fourier transform of the autocorrelation function.
   • Two averaging schemes are investigated: (i) averaging the Krylov parameters (angles) over disorder realizations, and (ii) extracting Krylov parameters from the disorder-averaged autocorrelation function.

3. Models Studied:

   • Anderson Model: A 1D non-interacting fermionic system with random on-site potentials $h_j \in [-h, h]$.
   • Aubry-André (AA) Model: A 1D system with a quasi-periodic potential $h_j = h \cos(2\pi\beta j)$, known to exhibit a sharp localization-delocalization transition at $h=2$.

Key Results

• Anderson Model (Localized Phase):

  • For any disorder strength $h$, the disorder-averaged autocorrelation saturates to a constant, confirming localization.
  • The Krylov angles $\theta_j$ approach $\pi/2$ (the critical point of the effective ITFIM) as the recursion step $j$ increases.
  • The decay of $\langle \cos \theta_j \rangle$ follows distinct power laws: $\langle \cos \theta_{2j-1} \rangle \sim 1/\sqrt{j}$ and $\langle \cos \theta_{2j} \rangle \sim -1/j$ for strong disorder.
  • Averaging Scheme: The authors demonstrate that extracting Krylov parameters from the disorder-averaged autocorrelation yields a more physical spectral function (smooth, no artificial dips) compared to averaging the parameters themselves, which produces unphysical dips near $\omega=0$ and $\pi$.
  • The localization in real space corresponds to a "stretched-exponential" localization of the 0-mode wavefunction in Krylov space.

• Aubry-André Model (Delocalized, Critical, and Localized Phases):

  • Delocalized Phase ($h < 2$): The autocorrelation decays, and the wavefunction spreads ballistically in Krylov space ($j \sim n$). The statistics of the autocorrelation follow a Porter-Thomas distribution, and the inverse participation ratio (IPR) scales as $L^{-1}$.
  • Critical Point ($h = 2$): The system exhibits multifractal scaling. The IPR scales as $L^{-D}$ with $D \approx 0.5$. The wavefunction spreads into the bulk but develops a tail. A Porter-Thomas distribution is also observed at the critical point.
  • Localized Phase ($h > 2$): The autocorrelation does not decay but exhibits persistent oscillations. The Krylov angles show strong fluctuations, and the spectral function displays pronounced peaks indicating long-lived modes. The wavefunction remains confined near the origin of the Krylov chain.
  • Wavefront Weight: A quantity $W(nT)$, defined as the weight of the wavefunction at the propagating front, serves as an indicator of the phase transition, dropping sharply around $h=2$.

• Krylov Angle Statistics:

  • The distribution of Krylov angles narrows as the recursion step increases, approaching a Gaussian distribution. The width of this distribution decays slowly as $\sigma \sim j^{-0.2}$, suggesting a slow logarithmic narrowing.
  • The Krylov angles are found to be essentially uncorrelated for large indices.

Significance and Claims
The paper claims that studying Hamiltonian dynamics via stroboscopic Floquet mapping offers distinct advantages over continuous-time Krylov methods:

1. Computational Efficiency: The moment method allows for the direct calculation of Krylov parameters from autocorrelation data, bypassing the need for full Gram-Schmidt orthogonalization.
2. Physical Insight: The mapping to an effective Floquet ITFIM provides a clear single-particle interpretation where the localization-delocalization transition corresponds to the appearance (delocalized) or absence (localized) of edge modes in the effective model.
3. Robustness of Averaging: The study establishes that averaging the physical observable (autocorrelation) before extracting Krylov parameters yields more physically sensible spectral functions than averaging the parameters themselves.
4. Universality: The observation of Porter-Thomas statistics in the delocalized and critical regimes of a non-interacting single-particle model suggests that operator spreading in Krylov space can mimic chaotic behavior even in integrable or non-interacting systems.

The authors conclude that the recursion procedure effectively performs a renormalization group flow, driving the system toward a fixed point in the Krylov chain where the angles approach $\pi/2$. They identify the emergence of multifractal scaling at the critical point and the specific power-law behaviors of Krylov angles as key signatures of the localization transition. Future work is suggested to extend these findings to higher dimensions and interacting systems.