Order Parameter Discovery for Quantum Many-Body Systems

arXiv:2408.01400 · quant-ph · Submitted 2024-08-02 · Read on arXiv

Listen

Radio episode about this paper

Transcript

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

Kai: Today's paper: "Order Parameter Discovery for Quantum Many-Body Systems".

Mira: This work introduces a method for constructing phase diagrams using a vector field derived from the reduced fidelity susceptibility (RFS) and demonstrates how information encoded in this vector field can…

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

Title and authors: Kai: So, we're looking at this paper titled "Order Parameter Discovery for Quantum Many-Body Systems," and the authors are Mariella, Murphy, Di Marcantonio, Najafi, Vallecorsa, Zhuk, and Rico. It seems like they're tackling a really fundamental problem in quantum physics: how to find out if a system is going into an ordered phase without already knowing what that order parameter even looks like.

Mira: I agree with Kai; the title itself suggests they are moving away from traditional methods where you have to guess the symmetry first, and instead, they're using some geometric information derived from fidelity susceptibility to map out phase diagrams and discover those hidden order parameters automatically.

Lev: From a hardware standpoint, that sounds incredibly efficient because it bypasses the need for exhaustive state characterization; if this method works on small subsystem measurements, it opens up possibilities for running simulations on systems too big for full state tomography.

Kai: Exactly, Lev; the paper seems to be proposing a new way to build these phase diagrams by constructing a vector field from the reduced fidelity susceptibility, which they then use to find observables that actually distinguish between phases without prior knowledge.

Mira: It's interesting how they connect this vector field directly to the second derivative of the fidelity susceptibility, defining a function g(lambda) based on those derivatives, which then yields a vector field P(lambda) = -grad g(lambda) for parameter space X.

Lev: If we think about running this on real hardware, we'd need to be careful about the smoothing assumptions they mention in constructing that vector field P(lambda), because noise in experimental measurements could easily corrupt that gradient calculation.

Kai: That's a valid point, Lev; the authors do acknowledge those smoothing assumptions are necessary for the construction of P(lambda): X to R squared, but they are trying to make it robust enough for general many-body Hamiltonians.

Mira: The implication here is that instead of relying on established theories like Landau-Ginzburg-Wilson theory, which requires you to know the symmetry beforehand, this approach lets the system suggest its own character through geometric mapping of quantum correlations.

The paper's summary: Kai: Moving past what we just touched on, the core summary of "Order Parameter Discovery for Quantum Many-Body Systems" is that they introduce a method where you construct a vector field from the reduced fidelity susceptibility, and then use information encoded in that vector field to identify observables corresponding to order parameters without needing prior knowledge of symmetry or transition type.

Mira: That really boils down to using the RFS vector field, which they define on a two-dimensional parameter space X R squared parameterized by control parameters (lambda one lambda two) T, and showing how the angle of this vector field theta(lambda) can map out phase transitions as sources in that field.

Lev: I'm thinking about the practical application; if they can identify transition lines purely from this geometric structure, it drastically simplifies the task of finding where a system moves from one ordered state to another on a phase diagram.

Kai: Precisely; and what's powerful is that they then generalize existing pairwise optimization methods to simultaneously handle all pairs of phase labels identified by the vector field, which allows for discovering multi-phase order parameters in a single optimization step twenty-six.

Mira: They also demonstrate this efficacy on three specific models: the ANNNI model, the cluster Hamiltonian, and a Rydberg atom chain, showing that this approach is applicable across different types of quantum systems.

Lev: If we want to run this on real hardware, the idea of finding observables via a QCQP optimization problem based on these vector field labels sounds like a very structured way to design experiments rather than just blind measurements.

Kai: Right, so the main point is that they use the RFS vector field and subsequent optimization to simultaneously map out phase boundaries and discover the observables that characterize those phases directly.

The paper's improvements: Mira: The authors highlight several areas where this method could be improved, such as incorporating a more robust way for handling non-local order, perhaps by using larger subsystem sizes k as suggested in the paper's notes on increasing 'k'.

Kai: I think integrating that "source/sink" pattern analysis they show in Figure eighteen into a machine learning classifier would be a smart way to automate the identification of phase transition lines, instead of just relying on pre-defined theoretical boundaries.

Lev: From an error correction perspective, if we could have an adaptive strategy for selecting the subsystem size k based on the indefiniteness condition of matrix A, that would be crucial for handling cases like topological transitions where a linear order parameter might fail initially.

Mira: Another improvement mentioned is developing a generalized optimization routine for the non-convex objective function described in equation (fifteen), specifically looking at leveraging the closed-form solution from the Singular Value Decomposition of matrix A when it's applicable.

Kai: That would make the optimization step much more tractable computationally, especially since that QCQP formulation can be quite demanding when dealing with all those identified pairs of phase labels.

Lev: If we could implement that adaptive scaling strategy for k, it would give us a formal way to ensure that even if the linear order parameter doesn't work, the system automatically scales up to capture the necessary higher-order correlations.

Mira: Essentially, these suggestions aim to make the method more general and computationally efficient while ensuring it correctly identifies complex ordering patterns in systems that aren't just simple Ising-like transitions.

Conclusion: Kai: So, to wrap up on "Order Parameter Discovery for Quantum Many-Body Systems," this paper shows a framework using the RFS vector field to construct phase diagrams and simultaneously discover observables corresponding to order parameters without needing prior knowledge of symmetry or transition type.

Mira: The implication is that we can move toward discovering the physical characteristics of many-body quantum systems through geometric analysis derived from fidelity, rather than starting with symmetry assumptions.

Lev: For real hardware, the immediate value lies in using this framework to guide experiment design and designing optimized measurement strategies based on the labels derived from the vector field.

Kai: And we can also use it for rigorous validation by applying finite-size scaling analysis to verify that the discovered observable is indeed an order parameter for models like ANNNI.

Mira: The paper suggests further work involves making the method more robust by improving handling of non-local order and generalizing the optimization routine, which would help apply this technique to a wider variety of quantum models.

Lev: My final thought is that if we can reliably implement an adaptive strategy for subsystem size k, it could provide a formal way to handle situations where standard linear order parameters don't suffice, giving us better tools for simulating complex phases.

Kai: It’s a solid piece of work that shows how geometric information from reduced density matrices can be used as a powerful tool for exploring the landscape of quantum many-body systems.

IBM Research - Dublin · Cavendish Laboratory, University of Cambridge, J.J. Thomson Avenue, Cambridge CB3 0HE, United Kingdom · Department of Physical Chemistry and EHU Quantum Center, University of the Basque Country UPV/EHU, Box 644, 48080 Bilbao, Spain · IBM Research · European Organization for Nuclear Research (CERN), Geneva 1211, Switzerland · Donostia International Physics Center · IKERBASQUE, Basque Foundation for Science

quant-ph

Submitted: 2024-08-02

Updated: 2026-09-04

Journal ref: Quantum 10, 2217 (2026)

DOI: 10.22331/q-2026-09-29-2217

Code: https://github.com/Fradm98/qs-mps

License: http://creativecommons.org/licenses/by-nc-sa/4.0/

Importance score: 81/100

The gist: This work introduces a method for constructing phase diagrams using a vector field derived from the reduced fidelity susceptibility (RFS) and demonstrates how information encoded in this vector field

Key concepts

Reduced Fidelity Susceptibility (RFS)
This is a geometric information derived from quantum correlations used to construct a vector field. This vector field helps in mapping out phase diagrams and identifying hidden order parameters in many-body systems automatically.
Order Parameter Discovery
The method aims to find observables that characterize phases without needing prior knowledge of the system's symmetry or transition type. It uses the RFS vector field to suggest which observables correspond to these order parameters.
Vector Field P(lambda)
This is a vector field constructed from the second derivative of the fidelity susceptibility, defined as P(lambda) = -grad g(lambda) for parameter space X. This field is used to map out phase transitions by observing where its angle changes.
Pairwise Optimization
The paper generalizes existing pairwise optimization methods to simultaneously handle all pairs of phase labels identified by the vector field, allowing for the discovery of multi-phase order parameters in one step.

Terminology

Summary

This work introduces a method for constructing phase diagrams using a vector field derived from the reduced fidelity susceptibility (RFS) and demonstrates how information encoded in this vector field can be used to discover observables corresponding to order parameters without prior knowledge of symmetry or transition type.

The core methodology involves:

  1. Constructing a vector field from the RFS defined on a two-dimensional parameter space, where the Hamiltonian is parameterized as a general many-body Hamiltonian with control parameters:

"We consider a general many-body Hamiltonian parameterized by the space X ⊆ R squared, of which its decomposition in terms of base and driving components reads H(λ) = H0 + λ1H1 + λ2H2, (8) with control parameters (λ1 λ2) T ∈ X."

The RFS is defined relative to a bipartition of the Hilbert space, resulting in the reduced density matrix (RDM), denoted as:

Let ρ0(λ) denote the reduced density matrix (RDM) resulting from tracing out the subsystem B for the ground state, so ρ0(λ) = TrB (ψ0 (λ)⟩ ⟨ψ0 (λ)). (9)

A key function, g(λ), is defined based on the second derivative of the fidelity susceptibility:

We introduce one of the key functions for our method5, that is g(λ):= − ∂2f(λ, δ) / ∂δ2 + ∂2f(λ, δ) / ∂δ2, where f(λ, δ) = F (ρ0(λ), ρ0(λ + δ)) ∈ [0, 1], for λ ∈ X and δ a perturbation in the latter space.

This leads to a vector field P:

We obtain the vector field P: X → R squared (under sufficient smoothing assumptions) defined by; P(λ):= −∇g(λ). (12)

The angle of this vector field, θ(λ), is also defined:

We obtain a scalar function mapping the parameters λ to the angles of the vectors in the image of P, that is θ(λ) = Arg e T 1 P(λ) + i e T 2 P(λ), (13).

The paper asserts that phase transitions materialize as sources in this vector field:

We expect that phase transitions materialize as sources in the vector field (12), i.e., the loci where the susceptibility is maximized.

The process for order parameter discovery involves:

"Given representative RDMs from two phases, the observable that maximally distinguishes them is found by solving a simple optimization problem. Our work builds upon and extends this approach in two key ways: first, we use the RFS vector field to identify all phase boundaries simultaneously, without requiring prior selection of representative states; second, we generalize the pairwise optimization of [25] to simultaneously handle all pairs of phase labels identified from the vector field, allowing multi-phase order parameters to be discovered in a single optimization step [26]."

The method is applied to three well-established models:

We demonstrate the efficacy of the method on three models: the ANNNI model [27, 28], the cluster Hamiltonian, and a Rydberg atom chain.

For order parameter discovery, an optimization problem (QCQP) is formulated using labels derived from the vector field:

"Given a finite set of parameters I in the neighbor of λc, we define a label yi ∈ [-1, 1] for each parameter λi as yi = sin (θ(λi) + η). From (14), we see that distinct phases will be assigned opposite signs, sign(yi). We define the index sets I+ = [iyi > 0] and I− = [iyi < 0], partitioning the indices for the parameters determining the ground states laying on the ordered and disordered phases, respectively. Under the assumption of non-degeneracy of the problem (see Appendix B for more details), we devise the following (non-convex) quadratically constrained quadratic program (QCQP) [38]: min M∈S m X(i,j)∈I+×I− −⟨M⟩ 2 i pi + ⟨M⟩ 2 j pj!, s.t. M 2 F ≤ 1, (15)."

The paper validates the discovered order parameters using finite-size scaling analysis:

"We apply finite-size scaling and verify that the obtained observable M is indeed an order parameter for the ANNNI Model. Finite-size scaling is a technique used to study phase transitions by examining how physical quantities change with system size.

Improvements for AI systems

As a fastidious researcher, I have analyzed this paper, Order Parameter Discovery for Quantum Many-Body Systems, which introduces a novel method based on the Reduced Fidelity Susceptibility (RFS) vector field for automated order parameter discovery in quantum many-body systems.

The core contribution is moving beyond traditional Landau-Ginzburg-Wilson (LGW) theory by using geometric information derived from subsystem fidelity to map out phase diagrams and simultaneously discover the specific Hermitian observables that characterize those phases.

Here are the specific improvements I propose for AI systems, categorized by application:

I will focus on enhancing AI systems in three key areas: Quantum Simulation/Discovery, Materials Science/Condensed Matter Prediction, and Quantum Metrology.

A. Specific Improvements for AI Systems (Methodological Enhancements)

  1. Improve the structure of the phase diagram construction by incorporating a more robust method for handling non-local order (e.g., using larger subsystem sizes, as suggested by the paper's note on increasing 'k').

  2. Integrate the source/sink pattern analysis (Figure 18) into a machine learning classifier to automatically identify phase transition lines rather than relying solely on pre-defined theoretical boundaries.

  3. Implement a generalized optimization routine (Eq. 15/45) that can handle non-convex objectives efficiently, specifically leveraging the closed-form solution derived from the Singular Value Decomposition (SVD) of the operator matrix A when applicable.

  4. Develop an adaptive strategy for selecting subsystem size 'k' based on the indefiniteness condition of matrix A, ensuring that when a linear order parameter fails (as in topological transitions), the system automatically scales up to capture higher-order correlations.

B. Capabilities of the Improved AI System

The improved AI system can perform the following specific tasks:

  1. Predict phase diagrams for complex quantum Hamiltonians (like ANNNI or Cluster models) without requiring prior knowledge of symmetry or order parameters, by simply analyzing the RFS vector field data derived from small subsystem measurements (RDMs).

  2. Automatically discover and characterize new, previously unknown order parameters (e.g., string order parameters in cluster models or crystalline phases in Rydberg chains) for a given quantum system by performing the optimization described in Section 4.

  3. Perform high-throughput, automated classification of quantum many-body states into distinct phases (e.g., FM vs PM vs FP) using only local reduced density matrices, circumventing the need for full state tomography or global overlaps which are computationally intractable at scale.

  4. Design optimal measurement strategies for quantum experiments by identifying the sparse set of Pauli operators that maximally distinguish between target phases, as suggested in Section 6.3 and 6.1.

  5. Quantify critical exponents and determine the universality class (e.g., Ising vs KT transition) for a system by analyzing the finite-size scaling behavior of the discovered order parameter observable, allowing for rigorous validation of theoretical predictions with high precision derived from DMRG data analysis (as shown in Section 5.3).

  6. Develop Quantum Phase Transition Explorer tools that can visualize the geometric structure of critical points (sources and sinks) in parameter space, providing an intuitive understanding of the physical mechanism driving the transition.

Abstract

Quantum phase transitions reveal deep insights into the behavior of many-body quantum systems, but identifying these transitions without prior knowledge of order parameters remains a significant challenge. In this work, we introduce a method for constructing phase diagrams using the vector field of the reduced fidelity susceptibility (RFS), and demonstrate how information encoded in this vector field can be used to discover observables corresponding to order parameters. We apply our approach to well-established models: the Axial Next Nearest Neighbour Interaction (ANNNI) model, a cluster state model, and a chain of Rydberg atoms; and validate the discovered order parameters using eigendecomposition and finite-size scaling analysis, confirming the expected universality classes. Our results demonstrate that the RFS vector field offers a unified framework for phase characterization and order-parameter discovery that requires no prior knowledge of symmetry or transition type, while relying only on reduced density matrices of small subsystems.

Sources

Related papers