Data-driven reconstruction of band dispersion and quantum geometry via Koopman dynamic mode decomposition

summary

The gist

Data-driven reconstruction of band dispersion and quantum geometry via Koopman dynamic mode decomposition presents a framework to reconstruct band structures and topological properties directly from…

In short

This work uses Koopman Dynamic Mode Decomposition (DMD) to reconstruct band structures and quantum geometry directly from spatiotemporal data, bypassing the need for an explicit Hamiltonian. It shows DMD can extract band dispersion, topological invariants like Chern numbers, and the quantum metric by analyzing how different spatial modes evolve in time.

Key concepts

Hamiltonian Floquet-Bloch Decomposition (FBD)
This is a traditional method where physical properties like energy bands are calculated using an explicit mathematical model called a Hamiltonian. It relies on knowing the exact governing equations of the system, which can be complex, making it equation-based and requiring prior knowledge.
Koopman Dynamic Mode Decomposition (DMD)
This is a data-driven technique that approximates complex spatiotemporal data by decomposing it into fundamental spatial modes and their temporal dynamics. It works directly on observed data, allowing researchers to infer physical properties like band structure without needing the underlying equations.
Quantum Metric
The quantum metric is a geometric property of the system's quantum states that describes how distances and angles are measured in its state space. This paper shows it can be reconstructed from DMD modes using fidelity calculations between adjacent modes, providing insight into the system's underlying geometry.
Inverse Participation Ratio (IPR)
The IPR is a measure used to diagnose whether a state or mode is localized or extended in space. By analyzing how the IPR changes as disorder increases, researchers can distinguish between different physical regimes, such as extended states and exponentially localized states.

Terminology used across episodes

This episode discusses

The paper

Data-driven reconstruction of band dispersion and quantum geometry via Koopman dynamic mode decomposition · Read on arXiv

ShanghaiTech University · Newcastle University

We present a data-driven framework for reconstructing band structures using Koopman operator analysis and dynamic mode decomposition (Koopman-DMD). Instead of deriving spectra from an explicit Hamiltonian, the approach reconstructs band dispersion and modal dynamics directly from spatiotemporal data, including wavefunctions and observables. This framework establishes a correspondence between Hamiltonian Floquet-Bloch decomposition and Koopman-DMD, whereby the extracted DMD modes encode frequencies, decay or growth rates, spatial profiles and projection weights. These quantities allow the reconstruction of spectral functions, local density of states, and delocalized-to-localized measures such as the inverse participation ratio. Also, these extended DMD modes enable inference of quantum-geometric and topological properties, including the quantum metric, Berry curvature and geometric phases. Applications to prototypical one- and two-dimensional tight-binding models, including disordered Su-Schrieffer-Heeger model and its Floquet and non-Hermitian variants, graphene and Haldane models, demonstrate that Koopman-DMD provides a unified route for the data-driven analysis of wave propagation, localization, and topological phases in condensed matter, photonics, and related fields.

Transcript

Introduction to the show: ident: Physics. The day in general physics: instruments, methods and results across the field.

Kai: I'm Kai.

Mira: And I'm Mira, here with Lev.

Kai: We're looking at the paper: "Data-driven reconstruction of band dispersion and quantum geometry via Koopman dynamic mode decomposition".

Title and authors: Mira: So, to recap what we’ve covered, this paper introduces a data-driven framework using Koopman dynamic mode decomposition to reconstruct band structures and quantum geometry directly from spatiotemporal data, bypassing the need for an explicit Hamiltonian. We've touched on how this connects to Floquet-Bloch theory and how it can extract spectral properties.

Kai: Exactly, and the central question they are asking is whether an unknown system can be effectively described in terms of quasiparticle-like excitations like Bloch modes, localized modes, or collective oscillatory modes associated with pattern formation. The answer they give is yes.

Lev: That’s a strong claim for a data-driven method. What does that actually mean when we're dealing with real experimental noise and finite measurements? Does the reconstruction hold up under those kinds of conditions?

Mira: They demonstrate this connection numerically in prototypical one and two dimensional lattice models, specifically the SSH model, graphene, and Haldane models. Across these typical lattice models, they show that Koopman dynamic mode decomposition is computationally efficient for extracting these modes.

Kai: And beyond just the structure of the bands themselves, they show that you can infer topological invariants like winding numbers or Chern numbers through discrete gauge invariant formulas that are formally parallel to those computed in Hamiltonian Floquet-Bloch analysis.

Lev: Inferring topology from data is ambitious. If you’re building an error correction code, knowing the topological properties without solving the full Hamiltonian analytically would save us time, but what if the DMD modes are just noisy representations of something else?

Mira: They show that you can reconstruct the quantum geometric tensor by estimating the quantum metric from the fidelity between adjacent dynamic mode decomposition modes. This metric is insensitive to local phase fluctuations, which is a nice feature for robustness.

Kai: So, it’s about using data to get these topological and geometric properties, which we normally derive from a known Hamiltonian's structure. It bridges that gap between theoretical modeling and experimental observation in some way.

The paper's summary: Lev: If the authors are looking at this as an improvement over existing methods, what specific advantages are they claiming for their Koopman dynamic mode decomposition approach compared to traditional Floquet-Bloch methods?

Mira: They present a comparative formulation of Hamiltonian Floquet-Bloch decomposition and Koopman dynamics. The key difference they highlight is that the Floquet-Bloch method is equation based, while Koopman DMD is equation free but data driven.

Kai: So the improvement lies in this shift from top-down derivation to bottom-up inference, where you don't need a fully specified microscopic model to get these results. They claim that DMD can recover the band gap and group velocity, and group velocity dispersion directly from spatiotemporal data instead of an explicit Hamiltonian.

Lev: That’s appealing because it reduces the requirement for having perfect knowledge of the underlying physics at every single point in space or time. But what about systems where we only have limited measurements?

Mira: They show that you can reconstruct band dispersion and modal dynamics directly from spatiotemporal data including wavefunctions and observables, which is a key improvement over methods that rely on prior knowledge of the governing equations.

Kai: And they also claim this method is powerful enough to capture the crossover from spatially extended states to exponentially localized states in models like the Anderson model by analyzing things like the inverse participation ratio. That’s how you diagnose localization without solving for it first.

Lev: Diagnosing localization through an observable like the inverse participation ratio is helpful, but what if we're looking at something more complex than just simple localization? Like driven non-Hermitian systems where things are really messy?

Mira: They show that using the Center of Mass distribution of DMD modes can clearly distinguish between topological edge pinning, Non-Hermitian skin effect induced boundary accumulation, and random Anderson localization in these driven non-Hermitian systems.

The paper's improvements: Kai: So, to wrap up the paper "Data-driven reconstruction of band dispersion and quantum geometry via Koopman dynamic mode decomposition," the main point is that we can use data to reconstruct quasiparticle excitations and spatiotemporal coherent structures from data without needing a fully specified microscopic model.

Mira: And they show that this framework, by bridging Hamiltonian Floquet-Bloch decomposition and Koopman-DMD, allows us to infer band structure features, topological invariants like the quantum metric, and even the quantum geometric tensors directly from experimental observations.

Lev: From an error correction standpoint, it’s interesting because it offers a way to probe topological properties without requiring a full Hamiltonian diagonalization upfront. But I still have my concerns about how well this scales when we move this from the numerical demonstrations on lattices to actual hardware with limited measurement capability.

Kai: I agree, but the demonstration shows that for models like SSH and graphene, the extracted DMD modes agree closely with exact Floquet-Bloch results. It’s a strong link between data and theory in those specific settings.

Mira: So, while it's not a replacement for every theoretical tool, this paper provides a way to look at complex systems from the bottom up through observation rather than starting top down with an equation. We should definitely keep an eye on how this is applied to driven, disordered systems where we don't have perfect knowledge of the system dynamics.

Lev: Yeah, so it’s a complementary framework, where Hamiltonian methods derive structure from equations top-down and Koopman dynamic mode decomposition infers dynamics bottom-up from observation. That complementarity is really what makes this work useful for driven or disordered systems where the underlying model isn't perfectly known yet.

Kai: That's the picture we have now for this paper. It gives us a data-driven route to understanding band dispersion and quantum geometry by analyzing spatiotemporal data through Koopman dynamic mode decomposition.

Conclusion: Kai: So, we're wrapping up on "Data-driven reconstruction of band dispersion and quantum geometry via Koopman dynamic mode decomposition." The main idea is that you can figure out band structure and quantum geometry by looking at time-resolved data without needing a full equation for the system.

Mira: Exactly. They show that this Koopman dynamic mode decomposition method lets us extract things like the band gap and even the quantum metric from just observing spatiotemporal data, rather than starting with an explicit Hamiltonian model.

Lev: From an error-correction standpoint, that's a big deal because it means we can probe topological phases without having to solve for them analytically first. But I still wonder about the overhead for real hardware; how much measurement record do you actually need before this method becomes practical?

Kai: They tested it on models like the SSH and graphene systems, and those results matched what we get from standard Floquet-Bloch analysis. It shows the method is robust enough to work across different types of lattice physics.

Mira: The paper also highlights how you can reconstruct topological invariants like winding numbers or Chern numbers using formulas that are formally similar to the ones used in traditional Hamiltonian methods. That's a direct link between data and topology.

Lev: Inferring those topological properties from data is ambitious, but if the fidelity holds up under noisy measurements, it could be useful for characterizing real-world materials where we don't have perfect control over every parameter.

Kai: They also showed you can diagnose localization regimes by looking at how the DMD modes are distributed in energy versus the inverse participation ratio plane. That helps separate extended states from localized ones clearly.

Mira: And they even found a way to reconstruct the quantum metric and Berry curvature of those dynamic modes using gauge invariant formulas derived from link variables. That’s getting into genuine quantum geometry inference here.

Lev: That geometric tensor reconstruction is what I find most interesting for hardware applications; if we can map that to physical observables, it could give us new metrics to monitor the health of our qubit states.

Kai: So, in summary, this paper shows that Koopman dynamic mode decomposition provides a data-driven route to extract effective quasiparticle excitations and spatiotemporal coherent structures from data without needing a fully specified microscopic model.

Mira: It’s a powerful tool because it bridges the gap between rigorous theoretical modeling and mode-resolved inference from experimental data, especially in driven or disordered systems.

Lev: It's a good framework for checking what we think is happening in complex systems, but we still need to figure out the practical measurement limits before we can really rely on this for designing new hardware architectures.

Kai: That's right. We’ve seen how this Koopman dynamic mode decomposition approach lets us reconstruct band dispersion and quantum geometry directly from spatiotemporal data. Next up, we’re looking at how AI is changing the game in inverse design of free energy landscapes with diffusion models.

More episodes

← Home