Exploring the Phase Diagram of the quantum one-dimensional ANNNI model

arXiv:2402.11022 · cond-mat.str-el, cond-mat.dis-nn, cond-mat.stat-mech, quant-ph · Submitted 2024-02-16 · 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: "Exploring the Phase Diagram of the quantum one-dimensional ANNNI model".

Mira: This manuscript explores the intersection of Quantum Machine Learning (QML) and Tensor Networks (TNs) to reconstruct the phase diagram of the one-dimensional Axial Next-Nearest-Neighbour Ising (ANNNI) model with a transverse…

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

Title and authors: Mira: Before we get into the specifics of how they built this analysis, I want to talk about what this paper is actually about—the title "Exploring the Phase Diagram of the quantum one-dimensional ANNNI model" and who wrote it.

Kai: It’s an exploration of how Quantum Machine Learning and Tensor Networks intersect specifically to reconstruct the phase diagram for a one-dimensional Axial Next-Nearest-Neighbour Ising model with a transverse field.

Lev: That specific setup sounds like something that could be hard to map onto current superconducting qubit architectures because of the competing interactions involved.

Kai: Right, precisely; the ANNNI model itself represents quantum fluctuations and frustrated exchange interactions, which makes it a good testing ground for studying magnetic ordering and frustration in general.

Mira: The authors are using this specific model because they note that it’s the simplest model combining quantum fluctuations from the transverse field and frustrated exchange interactions, which is very useful for initial studies.

Lev: If we were to run this on real hardware, the challenge would be setting up a system that can accurately represent both nearest-neighbor ferromagnetic and next-nearest-neighbor antiferromagnetic interactions simultaneously.

Kai: The authors are using Tensor Networks, specifically Matrix Product States, as their primary tool for representing these one-dimensional quantum states efficiently.

Mira: MPS is the relevant class of Tensor Networks for 1D systems because it allows them to encode the entanglement structure of a quantum state very succinctly through these matrices.

Lev: I’d be interested in knowing how they handle the complexity introduced by that frustration parameter κ, as that directly impacts the resulting phase diagram.

Kai: They use the adimensional ratios κ = -J2/J1 and h = B/J1 to characterize the model, which lets them systematically study how those parameters shift the phase boundaries.

Mira: That systematic approach is what makes this work strong; they aren't just looking at one case but exploring how changing those ratios dictates the entire structure of the phase diagram.

Lev: If we were developing an error correction scheme, knowing exactly where those transition lines are located based on kappa and h would give us a very specific target for our simulation parameters.

Kai: So, in short, they’re using these tools to systematically map out the regions of magnetic ordering versus disordered phases in this quantum model.

Mira: It sets up a really interesting framework because it shows how abstract concepts from condensed matter physics can be made tangible through computational methods like QML and TNs.

Lev: It’s about taking a complex physical problem, like frustration, and breaking it down into manageable computational steps for analysis.

The paper's summary: Kai: Now that we’ve talked about the setup, I want to summarize what the paper actually found regarding their methodology and the results for this "Exploring the Phase Diagram of the quantum one-dimensional ANNNI model."

Mira: Essentially, they used a hybrid approach: they use Tensor Networks to generate states from DMRG simulations, then feed those into QML classifiers to sort out which phase each state belongs to.

Lev: So, the TN analysis provides the numerical pieces of evidence about changes in the ground state properties before the QML does the final classification based on those input data.

Kai: That’s right; they use these numerical pieces of evidence from TNs to give input to their QML models, which then classify different phases like ferromagnetic, paramagnetic, floating phase and antiphase.

Mira: The key finding is that this pipeline effectively reconstructs the phase diagram by using both supervised and unsupervised QML techniques.

Lev: It’s interesting that they use the Anomaly Detection architecture for the unsupervised part because it helps them identify phases beyond what they might have labeled in their initial training set.

Kai: The results show that increasing system size to twenty spins allowed them to track classifier behavior, and the floating phase became increasingly evident as the system size grew.

Mira: That scaling behavior is important because it hints at the nature of that floating phase, suggesting its existence isn't just an artifact of small systems but a genuine feature of the model.

Lev: If we could run this on hardware, seeing that a phase becomes more evident with larger system sizes would be a very useful validation point for our simulation setup.

Kai: So they’ve confirmed that this hybrid QML pipeline provides a robust way to map out the phase diagram using these specific techniques.

Mira: It’s really about connecting the mathematical structure of the quantum state representation directly to the physical phases observed in the model, which is a significant connection.

The paper's improvements: Lev: The authors suggest that leveraging Tensor Networks to prepare initial states via DMRG states can be transposed into Parameterized Quantum Circuits, which is a bridge between classical simulation and quantum computation.

Kai: This means the AI system can automate the conversion of large, correlated classical ground states obtained via DMRG into quantum circuit inputs.

Mira: That’s a direct improvement because it significantly speeds up the state preparation phase for variational quantum algorithms compared to generating purely random states.

Lev: From an error correction standpoint, this state preparation method is much more structured than just random input, which is what we need to test on noisy hardware.

Kai: Furthermore, using the Anomaly Detection architecture allows the AI system to be trained without needing exhaustive pre-labeled training sets for every possible configuration.

Mira: That’s a huge win for experimentalists; it means they can discover new phases, like the floating phase, just by running the unsupervised model and observing what it flags as unusual.

Lev: If an error correction researcher could use this method to probe new states, that would be very valuable for designing codes that are resilient to unexpected behaviors.

Kai: The paper also highlights using Finite-Size Scaling analysis with Binder's cumulant as a way to locate phase transition points by plotting them against system size N.

Mira: That FSS technique is a standard tool, but when combined with QML classification, it gives us an enhanced capability to pinpoint those critical points more accurately.

Lev: Pinpointing the exact frustration values, like kappa KT = zero point eight one four, would help us reduce the uncertainty in our predictions about where these transitions occur.

Kai: So these improvements focus on making the AI pipeline more general and applicable to various physical systems beyond just this one ANNNI model.

Mira: It’s a method for building a more adaptable system that can handle new, complex physics by leveraging the structure of TNs as its foundation.

Conclusion: Kai: So, wrapping up the discussion on this paper "Exploring the Phase Diagram of the quantum one-dimensional ANNNI model," we’ve seen how QML and Tensor Networks work together to give us a solid method for phase diagram reconstruction.

Mira: It’s clear that this hybrid approach is powerful because it connects the structural representation of quantum states with observable physical phases in a very concrete way.

Lev: For me, the biggest implication is that this provides a clear pathway for using these AI tools to automate the extraction of critical parameters from simulation data.

Kai: We’re talking about automating the process of analyzing complex simulation results to find those key values like correlation lengths and critical exponents with high precision.

Mira: I think this will help experimentalists move away from purely manual analysis toward a more systematic, data-driven approach when studying these types of systems.

Lev: If we can automate the extraction of those parameters, it means we can test the limits of our error correction theories much more rigorously against physical observables.

Kai: So, to finish up on this paper "Exploring the Phase Diagram of the quantum one-dimensional ANNNI model," this work shows a robust method for connecting classical simulation with advanced quantum machine learning.

Mira: It’s a testament to how different mathematical fields can collaborate to reveal new physical insights from these complex quantum models.

Lev: It gives us concrete tools that we can actually use to push the boundaries of what we expect from quantum simulations in condensed matter physics modeling.

M. Cea, * M. Grossi † S. Monaco ‡ E. Rico, L. Tagliacozzo ¶ and S. Vallecorsa **

Max-Plank-Institut f¨ur Quantenoptik · Munich Center for Quantum Science and Technology (MCQST) · European Organization for Nuclear Research (CERN) · RWTH Aachen University · Deutsches Elektronen-Synchrotron (DESY) · Department of Physical Chemistry, University of the Basque Country UPV/EHU · Donostia International Physics Center · EHU Quantum Center, University of the Basque Country UPV/EHU · IKERBASQUE, Basque Foundation for Science · Institute of Fundamental Physics IFF-CSIC

cond-mat.str-el, cond-mat.dis-nn, cond-mat.stat-mech, quant-ph

Submitted: 2024-02-16

Updated: 2026-09-29

Comments: 15 pages, 15 figures

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Importance score: 83/100

The gist: This manuscript explores the intersection of Quantum Machine Learning (QML) and Tensor Networks (TNs) to reconstruct the phase diagram of the one-dimensional Axial Next-Nearest-Neighbour Ising

Key concepts

ANNNI model: The Axial Next-Nearest-Neighbour Ising (ANNNI) model is a quantum model used to study magnetic ordering and frustration. It combines quantum fluctuations from a transverse field with frustrated exchange interactions, making it useful for initial studies of these physical phenomena.
Tensor Networks (TNs): Tensor Networks, specifically Matrix Product States (MPS), are used as the primary tool to efficiently represent one-dimensional quantum states. They encode the entanglement structure of these states succinctly through matrices, which is crucial for analyzing 1D systems.
QML and TNs Hybrid Approach: The paper uses a hybrid approach where Tensor Networks generate initial quantum states from DMRG simulations, and Quantum Machine Learning (QML) classifiers then sort these states into different phases. This connects the mathematical structure of quantum states directly to observable physical phases.
Floating Phase: This is a phase identified by the QML analysis. The results show that this phase becomes increasingly evident as the system size grows, suggesting it is a genuine feature of the model rather than just an artifact of small systems.

Terminology

ANNNI model — 28× in this paper · explained in 1 episode
Frustration — 17× in this paper · explained in 2 episodes
Phase Transitions — 10× in this paper · explained in 2 episodes
Entanglement — 9× in this paper · explained in 3 episodes
Machine Learning — 9× in this paper
Quantum Circuits — 8× in this paper · explained in 2 episodes
Quantum Machine Learning — 8× in this paper
Accuracy — 7× in this paper · explained in 1 episode
Finite-Size Scaling — 6× in this paper
Generalization — 6× in this paper · explained in 13 episodes
Quantum Physics — 5× in this paper · explained in 10 episodes
Entanglement Entropy — 4× in this paper · explained in 2 episodes
Loss Function — 4× in this paper · explained in 2 episodes
Matrix Product States — 4× in this paper
Quantum Convolutional Neural Network — 4× in this paper
Axial Next-Nearest-Neighbour Ising — 3× in this paper
Quantum Anomaly Detection — 3× in this paper
Hybrid Approach — 2× in this paper · explained in 4 episodes
Universality Class — 2× in this paper · explained in 2 episodes
According to Conformal Field Theory — 1× in this paper
Clustering — 1× in this paper · explained in 2 episodes
Core — 1× in this paper · explained in 2 episodes
Matrix Product States (MPS) — 1× in this paper · explained in 2 episodes
Symmetry Breaking — 1× in this paper · explained in 3 episodes
Trace — 1× in this paper · explained in 2 episodes
Variational Quantum Eigensolver — 1× in this paper
Variational Quantum Eigensolver (VQE) — 1× in this paper · explained in 5 episodes

Summary

This manuscript explores the intersection of Quantum Machine Learning (QML) and Tensor Networks (TNs) to reconstruct the phase diagram of the one-dimensional Axial Next-Nearest-Neighbour Ising (ANNNI) model with a transverse field. This study is significant because it connects QML and TNs in various stages of algorithm construction, focusing on phase diagram reconstruction using both supervised and unsupervised techniques. The ANNNI model is important as it represents quantum fluctuations and frustrated exchange interactions, making it a paradigm for studying magnetic ordering, frustration, and the presence of a floating phase.

Model Description

The ANNNI model with a transverse field consists of ferromagnetic interactions between nearest neighbors (J1 > 0) and antiferromagnetic interactions between next-nearest neighbors (J2 < 0), competing against each other. The Hamiltonian is given by Equation (1):

H ANNNI = − J1 Σ i σ x i σ x(i+1 − J2 Σ i σ x i σ x(i+2 − B Σ z) i. This model is characterized by the adimensional ratios κ = −J2/J1 (the frustration parameter) and h = B/J1 (the transverse magnetic field). The paper notes that this model is the simplest model combining the effect of quantum fluctuations (owing to the presence of a transverse magnetic field), and frustrated exchange interactions.

Tensor Network Analysis

Tensor Networks serve as versatile mathematical constructs for representing high-dimensional data, with Matrix Product States (MPS) being relevant for depicting quantum states in one-dimensional systems. DMRG is an iterative numerical algorithm used to obtain the best MPS for encoding the ground state of quantum systems in one and two dimensions. In the context of phase diagrams, finite-size scaling (FSS) techniques are used to make predictions about the thermodynamic limit by studying local order parameters of finite systems. The analysis reveals four phases: the ferromagnetic phase, the paramagnetic phase, the floating phase and the antiphase, with a disordered line separating them.

Phase Transitions

The paper identifies several types of transitions within the ANNNI model:

Ising (I) transition, which is expected to be Ising-like for κ < 0.5.
Kosterlitz-Thouless (KT) transition between the paramagnetic phase and the floating phase, where a critical point is associated with the Luttinger liquid exponent taking the value K = 1/2.
Pokrovsky-Talapov (PT) transition between the floating phase and the antiphase, which is expected to be in this universality class.

The analysis of correlation length provides insights into these transitions:

(A) Ferromagnetic Phase and Paramagnetic Phase:

In the ferromagnetic phase, the magnetization along x is non-zero. The quantum phase transition to a disordered paramagnetic phase occurs at a critical value of frustration where magnetization approaches zero. Binder’s cumulant (Eq. 6) is used to locate this transition point by plotting it as a function of system size N.

(B) Floating Phase:

The floating phase is described by a Luttinger liquid with algebraic incommensurate correlations, stable when the Luttinger exponent lies within the interval 1/4 < K < 1/2. The Kosterlitz-Thouless transition occurs when K = 1/2, and the Pokrovsky-Talapov transition arises when incommensurability vanishes.

Quantum Machine Learning Analysis

QML is employed using two main architectures to classify the phases:

Quantum Convolutional Neural Network (QCNN): This supervised model uses labels encoded in a 2-qubit state (Ferromagnetic: 00⟩, Paramagnetic: 01⟩, Antiphase: 10⟩, Floating phase: 11⟩). The training minimizes a cross-entropy loss function. One analysis uses only analytical points on the axes, while another includes the entire phase diagram.
Quantum Anomaly Detection (AD): This unsupervised architecture functions as a quantum Autoencoder, aiming to compress information into trash qubits. It is particularly suitable for identifying phases beyond analytical ranges, such as the Floating phase. Training involves minimizing a loss function related to Pauli-Z expectation values (Eq. 15).

Summary and Conclusion

The QML pipeline, combining TN analysis (for state preparation) and QCNN/AD (for classification), provides a robust method for phase diagram reconstruction. The TN analysis provides numerical pieces of evidence of changes in the properties of the ground state. The paper confirms that increasing system size from 12 to 20 spins allows tracking classifier behavior, with the presence of the floating phase becoming increasingly evident with a higher number of spins, suggesting it might be classified as paramagnetic rather than antiphase in some QCNN analyses.

Improvements for AI systems

As a fastidious and diligent AI researcher, I have analyzed this paper, Exploring the Phase Diagram of the quantum one-dimensional ANNNI model, which explores connecting Quantum Machine Learning (QML) and Tensor Networks (TNs) to reconstruct the phase diagram of a 1D quantum spin model.

The improvements derived from this work can be applied across several domains, primarily where complex, high-dimensional, or frustrated systems are involved.

Here are the specific improvements and what they enable the improved AI system to do:

)

Improve QML Training and Generalization in Noisy Environments:

The paper highlights challenges like the barren plateau phenomenon in QML (Section II.B).

By leveraging TNs to prepare initial states (as done in Section IV.A), the system can utilize highly structured, physically relevant data inputs instead of purely random quantum circuits.
By employing the Anomaly Detection (AD) architecture, which is unsupervised and bypasses label acquisition, the AI system can be trained to identify new phases (like the floating phase) without needing exhaustive pre-labeled training sets for every possible configuration.
By increasing system size to 20 spins and using Finite-Size Scaling (FSS) analysis via Binder's cumulant, the AI model gains better generalization capabilities, allowing it to predict behavior in regions of the phase diagram not explicitly seen in the training data.

Enhanced Phase Transition Detection for Complex Physical Systems:

The paper provides a robust framework for mapping complex physical models (ANNNI) onto a phase diagram using both TN analysis and QML classification (QCNN).

An improved AI system can be used to rapidly classify the state of a given physical system (e.g., a material under specific magnetic fields/interactions) by performing an efficient hybrid TN/QML pipeline.
Specifically, it can distinguish between different phases—Ferromagnetic, Paramagnetic, Antiphase, and Floating—with high accuracy as the system size increases.

Accurate Characterization of Quantum Critical Phenomena:

The paper provides tools to extract key critical parameters from finite-size simulations (e.g., correlation length scaling in Appendix A and Luttinger liquid parameters in Appendix B).

An improved AI system can be trained to perform automated analysis on simulation data (like entanglement entropy or correlation functions) derived from quantum simulators.
This enables the AI to accurately extract critical exponents (like the central charge, e.g., extracting 'c=1' for a Luttinger liquid) and locate precise phase transition points (e.g., identifying critical frustration values like κKT = 0.814) with high precision, reducing experimental error in condensed matter physics modeling.

Efficient State Preparation for Quantum Simulations:

The method of transposing DMRG states into Parameterized Quantum Circuits (PQCs) provides a bridge between classical simulation and quantum computation.

An improved AI system can automate the conversion of large, correlated classical ground states (obtained via DMRG) into quantum circuit inputs, significantly speeding up the state preparation phase for variational quantum algorithms compared to purely random state generation.