Excitation spectra and rank tomography of finite MPS tangent spaces

arXiv:2607.05269 · cond-mat.quant-gas, quant-ph · Submitted 2026-07-06 · Read on arXiv

Listen

Transcript

Introduction to the show: ident: Quantum Radio. Generated commentary on the latest quantum physics and condensed matter papers.

Kai: Today's paper: "Excitation spectra and rank tomography of finite MPS tangent spaces".

Mira: We formulate a tangent-space method for algebraic varieties of matrix product states (MPS) to study excitation spectra of non-uniform quantum many-body systems with open boundary conditions,

Kai: First, who's behind it and why it matters.

Title and authors: Kai: So Mira and I were looking at "Excitation spectra and rank tomography of finite MPS tangent spaces," and it sounds like this paper is taking a pretty deep dive into using algebraic geometry to study how quantum systems described by Matrix Product States behave when we perturb them.

Mira: Exactly, Kai, the title suggests they are connecting the geometry of these MPS representations—which are essentially varieties defined by polynomial relations based on bond dimensions—to the actual excitation spectra of non-uniform systems. It’s ambitious because it moves away from just using standard variational methods for finding ground states and starts looking at how well those approximate states can capture dynamics.

Lev: From a computational standpoint, I wonder if this geometric approach actually translates to something feasible for running on current hardware, especially since we're talking about linearizing around a variational ground state and projecting onto that horizontal tangent space.

Kai: That’s the core question, Lev; the paper lays out how they define this tangent space as a projective variety defined by rank constraints on tensor flattenings, specifically stating "rank(T(k)) ≤ Dk for all k = one <ref:2607.05269#pg0>..., N − one" (<ref:2607.05269#pg0>). This structure is what they use to define the directions we can move in the parameter space.

Mira: And that's where the real theoretical meat is; they introduce a way to characterize this tangent space’s expressivity using particle-number resolved Schmidt-rank distribution, or PRSR, which provides "a direct measure of parametric deficiency in each particle-number sector" (<ref:2607.05269#pg1>). This links the internal structure of the state directly to its spectral accuracy.

Lev: That sounds like it might help us understand where we’re failing in our DMRG calculations, because if we know the deficiency per sector, we can aim our bond dimension choices more intelligently instead of just picking a large arbitrary number.

Kai: Exactly; and they show how this method reproduces low-lying excitations using the Bose–Hubbard model as a benchmark, capturing finite-size precursors to things like the Mott insulator to superfluid transition (<ref:2607.05269#pg0>). That’s a concrete result we can test.

Mira: And it goes further by showing how this approach captures those finite-size precursors, which is important because standard methods often struggle to see those subtle changes near critical points (<ref:2607.05269#pg1>). The connection between the rank profiles and the deficiency helps explain why states with the same coarse bond dimensions can have different spectral accuracy.

Lev: If we can diagnose that deficiency, it gives us a roadmap for what kind of higher-order ansatz we might need to employ when our linear method starts showing trouble, which is something I think would be very useful for error correction studies.

Kai: So, to recap the main idea of "Excitation spectra and rank tomography of finite MPS tangent spaces," it’s using algebraic geometry to define the tangent space of MPS as a variety constrained by rank conditions, and then using PRSR to measure how expressive that state is in terms of its particle-sector rank profiles.

Title and authors: Mira: Precisely; it’s about moving beyond just finding the ground state and instead quantifying the geometric constraints on the space of states near that ground state, which directly impacts what we can learn about its excitations.

Lev: I see how this relates to our work on error correction; if we know where the parametric deficiency is highest, we might be able to construct a more robust subspace for error-corrected simulations.

Kai: Speaking of robustness, the paper suggests that by using this tangent-space method, we can dynamically determine the minimum required bond dimension needed for a specific spectral accuracy target, using that parametric deficiency metric (<ref:2607.05269#pg1>).

Mira: That dynamic selection of bond dimensions based on sector-specific needs sounds like a significant practical improvement over picking a fixed, large number just to be safe.

Lev: For running this on real hardware, I'd need to see how computationally demanding the rank tomography part is; if it adds too much overhead to the simulation loop, it won't be practical for larger systems.

Kai: The authors address that by focusing on linear perturbations around a stationary state psi bar, which leads to a generalized eigenvalue problem Bx = omega Ax, where omega is the phase rotation frequency of the tangent vector corresponding to those linear changes (<ref:2607.05269#pg1>).

Mira: That eigenvalue omega gives us a direct way to approximate the excitation spectrum, which is what we were hoping for when looking at how these varieties relate to dynamical principles.

Lev: If the resulting frequencies are reliable, it could give us a new way to probe the stability of phases that standard time-evolution methods might miss due to truncation errors.

Kai: So, looking at the overall implication of this work on "Excitation spectra and rank tomography of finite MPS tangent spaces," it’s providing a rigorous geometric framework to diagnose the limitations of MPS approximations in complex regimes like phase transitions.

Mira: It gives us a quantifiable measure—the PRSR—of how much information is missing in each particle sector, which should help us predict where our spectral reconstruction will fail before we even run the full simulation.

Lev: For me, the implication is that this method offers a systematic way to move from just observing results to understanding the underlying geometric constraints that govern those results in quantum many-body problems.

Kai: So, as we wrap up on "Excitation spectra and rank tomography of finite MPS tangent spaces," it seems they’ve given us a powerful tool to not only find excitations but also characterize the expressivity of the states themselves through that particle-sector analysis.

Mira: That’s right; it ties together geometry, state structure, and spectral accuracy in a way that allows us to diagnose limitations based on underlying physics rather than just numerical error.

Lev: It really solidifies the idea that understanding the shape of the variational manifold is key to advancing simulation techniques for complex quantum systems.

Kai: We'll keep an eye on how this method evolves, and I think we’ve got some really exciting things on our hands with this new perspective on MPS analysis.

The paper's summary: Kai: So, to kick things off, we're talking about this paper that uses algebraic geometry to study how Matrix Product States behave when we perturb them, specifically focusing on finding their excitation spectra.

Mira: That’s right; it essentially maps the space of possible MPS states onto a geometric variety defined by rank constraints and then uses tools like particle-number resolved Schmidt rank distribution to tell us how expressive those states actually are.

Kai: What I find really interesting is the idea of defining a tangent space for this variety, splitting it into directions that correspond to simple gauge changes and directions that are orthogonal to them, which is what they call the horizontal tangent space.

Mira: And that distinction between vertical and horizontal directions is crucial because it tells us exactly how we can apply time-dependent variational principles to approximate the actual dynamics of the system.

Kai: It sounds like this could help us get a much clearer picture of how ground state structure dictates whether a variational approximation will actually capture low-energy excitations accurately.

Mira: Exactly, and their quantification of parametric deficiency across different particle-number sectors gives us a concrete metric to judge the success or failure of our ansatz in specific regimes.

Kai: This could mean we can stop picking bond dimensions arbitrarily and start selecting them based on where the "missing directions" are most concentrated, which seems like a huge step for practical simulation design.

Mira: It moves us from just guessing parameters to having a theoretically grounded method for diagnosing the limitations of any MPS representation before we even start calculating full dynamics.

Kai: And that leads directly into how this could help us find those subtle precursors to phase transitions in systems where standard methods often get stuck.

Mira: Because if we can track the spectral softening as parameters approach a critical point using this tangent-space approach, it gives us a way to see those features even when the system is far from equilibrium.

Kai: It’s really exciting because it suggests that the internal structure of the ground state itself carries information about its dynamical response, which isn't always obvious from just looking at the bond dimensions.

Mira: And this opens up avenues for understanding systems with broken particle number conservation as well, which is a major hurdle in many realistic setups.

Kai: So, essentially, they’ve provided a new geometric language to interrogate MPS states for both their static structure and their dynamic properties.

Mira: It’s about linking the abstract mathematical description of the state's space to observable physical quantities like excitation energies in 1D systems <ref:2607.05269#pg0>.

Kai: We need to see if this framework can actually handle the complexity of realistic models, given how much detail they're trying to capture with those rank constraints.

The paper's improvements: Tom: So, we're looking at how the authors suggest extending their work on finite MPS tangent spaces to get even more useful tools for analyzing these quantum states.

Kai: What I found compelling is their proposal to use that rank tomography metric, the PRSR, not just as a diagnostic tool but as a direct guide for optimizing the simulation parameters.

Mira: That’s right; they suggest that if you know where the parametric deficiency is most severe in a specific sector, you can dynamically adjust the bond dimension to target spectral accuracy more efficiently.

Kai: It’s smart because it moves away from guesswork when choosing how large your tensor network needs to be for a particular problem.

Mira: And they also introduce ideas about diagnosing rank frustration, which is that situation where different particle-number sectors demand very different bond patterns for the same ground state.

Kai: That sounds like a real practical problem; if we can diagnose that frustration early, it gives us a roadmap for when we should consider using more complex ansatzes than just the linear approximation.

Mira: It suggests that this geometric approach is not just descriptive but predictive, allowing us to anticipate where our linear method will fall short based on the state's inherent structure.

Kai: I’m thinking about what this means for running these simulations on real quantum hardware; if we can pre-calculate the optimal bond dimensions, it could drastically cut down on unnecessary computational time during the actual experiment.

Mira: Indeed, and it connects back to our background work on error correction; knowing the parametric deficiency per sector might even inform how we structure parity checks in a fault-tolerant simulation.

Kai: So, this isn't just about getting better spectral results; it’s about building a more intelligent pipeline for simulating these many-body systems using tensor networks.

Mira: Precisely, and their future work points toward applying these concepts to more complex Hamiltonians where particle number conservation is explicitly broken.

Kai: That would be a big step because most of our current DMRG tools rely heavily on those fixed sectors, so tackling non-conserved dynamics with this geometric framework would be a real test.

Conclusion: Kai: So, to wrap things up on "Excitation spectra and rank tomography of finite MPS tangent spaces," we can summarize that the authors developed a geometric method using rank constraints and particle-sector analysis to precisely map out the expressivity of MPS states and how that structure influences their excitation spectra.

Mira: That's right; they successfully translated abstract algebraic geometry into a tool that quantifies how much information is missing in each particle number sector, which is a significant step for any theorist working with tensor networks.

Kai: It really shows how fundamental the underlying mathematical structure of the state dictates its dynamical properties, which is something I think we need to keep focused on when building our next generation of quantum hardware.

Lev: For me, what stands out is the potential for this method to provide a theoretical blueprint for designing better error-corrected simulations because it helps identify exactly where the truncation errors are most likely to manifest.

Mira: And their future work focusing on non-conserved dynamics is really important because that’s where many of our most interesting physical phenomena occur, moving beyond simple ground state analysis.

Kai: I'm looking forward to seeing how this translates into something we can actually implement in the lab; it's one thing to have a neat mathematical framework and another to see it perform reliably on a real superconducting qubit setup.

Lev: If the method is robust enough for small systems, it could definitely inform how we approach scaling up simulations for larger, more complex quantum many-body problems.

Mira: It gives us a clearer picture of the limitations of MPS approximations in regimes like phase transitions and non-equilibrium processes that are currently very difficult to study accurately.

Kai: We’ll keep an eye on how this geometric framework evolves because understanding the shape of the variational manifold is definitely key to advancing our simulation techniques.

INO-CNR Pitaevskii BEC Center and Dipartimento di Fisica, Universita di Trento · Max-Planck-Institute for the Mathematics in the Sciences

cond-mat.quant-gas, quant-ph

Submitted: 2026-07-06

Updated: 2026-10-07

Comments: 19 double-sided pages, 6 figures. v2: Updated reference to completed work; otherwise unchanged. v3: Updated manuscript for publication

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 83/100

The gist: We formulate a tangent-space method for algebraic varieties of matrix product states (MPS) to study excitation spectra of non-uniform quantum many-body systems with open boundary conditions, and

Key concepts

Projective MPS Variety: This is the geometric space containing all possible pure MPS states, defined by polynomial relations involving bond dimensions. It is characterized by rank constraints on tensor flattenings, specifically that the rank of these flattenings must not exceed a certain limit for each level k.
Horizontal Tangent Space: Within the tangent space of the MPS variety, this subspace represents directions that are orthogonal to gauge orbits. These directions are defined by specific Hermitian conditions involving matrices Yi and measure how perturbations affect the system's dynamics in a way that is independent of simple rephrasing (gauge transformations).
Particle-Resolved Schmidt Rank Distribution (PRSR): This concept analyzes the tangent space within specific particle-number sectors. By fixing a reference state and defining cuts, PRSR measures the actual rank profile for each sector, revealing how states with identical coarse bond dimensions can still have different tangent ranks and spectral accuracy.

Terminology

Hilbert Space — 13× in this paper · explained in 2 episodes
Accuracy — 9× in this paper · explained in 1 episode
Matrix Product States (MPS) — 5× in this paper · explained in 2 episodes
Benchmark — 4× in this paper · explained in 2 episodes
Frustration — 4× in this paper · explained in 2 episodes
For Matrix Product States — 3× in this paper
Gauge Freedom — 3× in this paper · explained in 2 episodes
Hubbard Model — 3× in this paper · explained in 3 episodes
Phase Transitions — 3× in this paper · explained in 2 episodes
Consistency — 2× in this paper · explained in 3 episodes
Effective Hamiltonian — 2× in this paper · explained in 2 episodes
Core — 1× in this paper · explained in 2 episodes
Curse of Dimensionality — 1× in this paper · explained in 6 episodes
Generalization — 1× in this paper · explained in 13 episodes
Robustness — 1× in this paper · explained in 5 episodes
Stiefel manifold — 1× in this paper · explained in 2 episodes

Summary

We formulate a tangent-space method for algebraic varieties of matrix product states (MPS) to study excitation spectra of non-uniform quantum many-body systems with open boundary conditions, and introduce rank tomography to characterize the expressivity of this method.

The gist

The construction defines an MPS tangent space as a projective variety defined by rank constraints on tensor flattenings, and uses particle-number resolved Schmidt-rank distribution (PRSR) to define particle-resolved rank profiles inside the fixed coarse MPS rank stratum, explaining how internal ground state structure controls parametric deficiency and expressivity.

Variational Manifolds of MPS

The paper begins by defining pure quantum states as points in a projective Hilbert space, and a variational class of pure states as an algebraic variety where the variational space is defined by polynomial relations. For Matrix Product States (MPS), this variety is characterized by bond dimensions and physical dimensions. The affine MPS contraction map, denoted as the polynomial map, defines the parametrization of this space:

The projective MPS variety, denoted as the Zariski closure of the image of the projective contraction map, admits a description in terms of flattening ranks: rank(T(k)) ≤ Dk for all k = 1,..., N − 1.
The full-rank MPS stratum is defined as the subset where each flattening attains maximal rank: rank(T(k)) = Dk, k = 1,..., N − 1. This set is shown to be a smooth Zariski open subset of the projective variety VD,d.

Tangent Space of MPS Variety

The tangent space of the parameter manifold TpPD,d naturally splits into directions tangent to gauge orbits and directions orthogonal to them. The horizontal tangent space, denoted as ker(DpΦD,d)⊥, is defined by orthogonality with respect to a product metric on the Stiefel factors and the Fubini–Study metric on the projective factor.

The vertical tangent space consists precisely of the tangent directions along the fiber of ΦD,d(p), equivalently along the gauge orbit through p. Its dimension is given by dimR ker(DpΦD,d) = N X−1 i=1 D2 i.
A tangent vector (δM, δC) is horizontal if it satisfies specific Hermitian conditions involving matrices Yi: Yi:= X di ji=1 (δMi ji)∗Mi ji − d Xi+1 ji+1=1 Mi+1 ji+1 (δMi+1 ji+1)∗, for i = 1,..., N − 2 and a similar condition for YN−1.

Linearised MPS Excitations on the Tangent Space

Excitation spectra are approximated using vectors in the horizontal tangent space ker(DpΦD,d)⊥ via the time-dependent variational principle (TDVP).

The TDVP replaces the exact time derivative with a tangent vector [δψ] that minimizes the residual error: δψ∗ = arg min δψ∥δψ + iHψ∥2. This leads to the Dirac-Frenkel condition: ⟨δψ,(i∂t − H)ψ⟩ = 0 for all δψ ∈ T[ψ]V=D,d.
For linear perturbations around a stationary state ψ¯, the dynamics are governed by the generalized eigenvalue problem: Bx = ωAx, where A and B are matrices derived from the overlap matrix and the projected Hamiltonian H'. This yields a first-order approximation to the excitation spectrum: The eigenvalue ω is the phase rotation frequency of a tangent vector corresponding to linear perturbations around the GS ψ¯.

Tangent-Space Tomography

To understand expressivity, rank tomography is introduced based on particle-number sectors.

The physical tangent space T = Im(U) = T[ψ¯]V=D,d ⊂ X is analyzed by decomposing the Hilbert space into particle-number sectors H = M HM, and defining the projected sector-M tangent space T proj M:= PMT = P M T.
The parametric deficiency of the tangent space in sector M is defined as: δpar M:= d target M − dim(T proj M), which quantifies the number of missing independent directions in each sector.
The particle-resolved Schmidt rank distribution (PRSR), rl(m), is derived by fixing a reference state ψ¯ ∈ HM0 and a cut l, providing a direct measure of parametric deficiency in each particle-number sector. This profile explains how "states with the same coarse bond dimensions can nevertheless have different tangent ranks and different spectral accuracy.

Improvements for AI systems

As a fastidious researcher, I have analyzed this paper, Excitation spectra and rank tomography of linear matrix product tangent spaces. The core contribution is a novel geometric approach—using algebro-geometric tools (varieties and tangent spaces) on Matrix Product States (MPS) to study excitation spectra beyond standard variational methods.

Here are the specific improvements I can propose for AI systems, categorized by the capability they unlock:

The improved AI system will be able to perform high-fidelity, low-energy spectral analysis of quantum many-body systems using tensor network representations, overcoming limitations in current Variational Quantum Eigensolver (VQE) or Density Matrix Renormalization Group (DMRG) approaches when probing phase transitions.

Here are the specific improvements:

[Linear MPS Tangent-Space Method for Low-Energy Excitations]: The AI system will be able to reconstruct the low-energy excitation spectrum of 1D quantum many-body systems (like the Bose–Hubbard model) by linearizing around a variational ground state and diagonalizing an effective Hamiltonian projected onto the horizontal tangent space of the MPS variety.
[Phase Transition Precursor Detection]: The system can precisely identify finite-size precursors of quantum phase transitions (e.g., Mott insulator to superfluid transition) by observing the softening of specific particle-type modes in the excitation spectrum as system parameters approach critical points, a feature captured accurately by the method even with modest bond dimensions (D=8).
[Expressivity and Accuracy Quantification via Rank Tomography]: The AI can quantify the expressivity of any given MPS state or variational ansatz. It will use particle-resolved Schmidt rank distribution (PRSR) to diagnose exactly how the internal structure of a ground state controls its susceptibility to linear perturbations, allowing researchers to predict where spectral reconstruction will fail based on the state's underlying particle-number correlations.
[Adaptive Bond Dimension Selection]: The system can dynamically determine the minimum required bond dimension for achieving a target spectral accuracy, moving beyond arbitrary choices. It will use the parametric deficiency metric to calculate exactly how many independent directions are missing in critical sectors and recommend bond dimensions that maximize precision for specific excitation types (e.g., single-particle vs. two-particle excitations).
[Symmetry Breaking Dynamics]: The AI can analyze the excitation spectra of Hamiltonians where particle number conservation is explicitly broken (e.g., using a chemical potential term), providing a continuous color scale representation of the expectation value of the particle number operator, offering insights into non-conserved dynamics that are inaccessible to standard fixed-sector DMRG methods.
[Automated Rank Frustration Diagnosis]: The system can diagnose rank frustration in ground states where different particle-number sectors require different optimal bond patterns, providing a roadmap for researchers to use higher-order (double tangent) ansatzes when the linear method shows deficiency in specific regions of the parameter space.

This improved AI system will enable:

The accurate calculation of low-lying excitation energies for 1D quantum systems with high fidelity, even under finite-size constraints.
A rigorous, theoretically grounded method to diagnose the limitations (expressivity and accuracy) of MPS approximations in complex regimes like phase transitions.
The ability to select the optimal computational resources (bond dimension) needed for a given level of spectral precision, leading to more efficient quantum simulation workflows.

Abstract

We formulate the matrix product state (MPS) tangent-space excitation construction for finite, non-uniform systems with open boundary conditions, using the smooth full-rank stratum of the MPS variety as the variational manifold. The resulting linear tangent ansatz yields a projected-Hamiltonian eigenproblem for approximating excitation spectra. We further introduce a rank tomography that characterizes the particle-sector expressivity of the MPS tangent space. For number-conserving systems, we resolve the Schmidt ranks of reference states by particle number and derive an explicit relation between the resulting rank profile and the dimensions of the tangent-space sectors. This allows sector completeness and parametric deficiency to be determined directly from the reference state, without explicitly constructing a tangent basis, and provides a direct diagnostic of the sectorwise expressivity of the MPS excitation ansatz. We benchmark the finite-system construction on Bose--Hubbard chains against exact diagonalization, finding accurate low-lying excitation branches while identifying sector-dependent limitations of the linear tangent ansatz.