Projector, Neural, and Tensor-Network Representations of $\mathbb{Z}_N$ Cluster and Dipolar-cluster SPT States

This paper presents a unified framework for expressing $\mathbb{Z}_N$ cluster and dipolar-cluster symmetry-protected topological states through projector, neural, and tensor-network representations, deriving closed-form weight functions and demonstrating that tensor product states offer a potentially more efficient description for modulated SPT phases than conventional matrix product states.

The Gist

Imagine you are trying to describe a massive, intricate dance performed by thousands of dancers (quantum particles) on a stage. In the world of quantum physics, describing how these dancers move and interact is incredibly difficult because they are all connected in a web of "spooky" relationships called entanglement.

This paper is like a new instruction manual for writing down the choreography of a specific type of dance called a Cluster State. The authors, a team of physicists, have found a smarter, more efficient way to write down these instructions using three different "languages" that turn out to be secretly the same thing.

Here is the breakdown of their discovery using everyday analogies:

1. The Problem: The "Too Many Notes" Problem

Usually, to describe a quantum dance, you need a massive list of numbers. As the number of dancers grows, the list of numbers grows so fast that even the world's most powerful supercomputers can't handle it. It's like trying to write down every single possible move for a dance troupe of a million people; the paper would be longer than the universe.

Physicists have developed "shortcuts" to describe these dances:

• MPS (Matrix Product State): Like a chain of dominoes, where each domino only talks to its immediate neighbors.
• NQS (Neural Quantum State): Inspired by Artificial Intelligence (AI), this treats the dance like a neural network, using "hidden" variables to summarize the connections.

2. The New Tool: The "P-Representation" (The Projector)

The authors introduce a new, unifying language they call the P-representation.

• The Analogy: Imagine the dancers are wearing special masks (Projectors). The mask hides the dancer's identity but shows a specific symbol. The "interaction" between dancers isn't a direct handshake; it's mediated by a giant, invisible sheet of paper (the Interaction Matrix) that connects the masks.
• Why it's cool: This method separates the "dancer" (the physical particle) from the "connection" (the math). It turns a messy quantum problem into a clean, structured puzzle.

3. The Big Discovery: Three Ways to Say the Same Thing

The paper shows that you can translate between three different ways of describing the same dance:

• The "Projector" Way (P-rep): The raw, structural view.
• The "AI" Way (NQS): They show that the "connection sheet" can be broken down into Neural Network weights. Think of this as teaching an AI to learn the dance. The paper provides the exact "weights" (the settings on the AI knobs) needed to perfectly recreate the dance for any number of dancers ($Z_N$), not just the simple case of two.
• The "Chain" Way (MPS/TPS): They show how to rearrange those AI weights to look like a chain of dominoes (MPS).

The "Aha!" Moment:
For simple dances, the chain of dominoes (MPS) is perfect. But for more complex dances involving dipolar symmetry (where dancers interact with neighbors and neighbors-of-neighbors), the simple chain breaks.

• The Solution: They realized that for these complex dances, you need a 3D structure (called a Tensor Product State or TPS) instead of a 1D chain.
• The Metaphor: Imagine the simple dance is a line of people holding hands. The complex dance is a line of people holding hands, but also holding hands with the person two spots away. To describe this, you can't just use a line; you need a web. The authors built this web (TPS) directly from their AI (NQS) instructions.

4. The "Non-Invertible" Mystery (The Kramers-Wannier Operator)

The paper also tackles a weird symmetry called the Kramers-Wannier (KW) operator.

• The Analogy: Imagine a magic mirror that transforms the dance. Usually, if you look in a mirror and then look in a mirror again, you get back to the original image (invertible).
• The Twist: This specific quantum mirror is "non-invertible." If you look in it, you get a new dance, but you cannot look in it again to get the original back. Information is lost.
• The Explanation: The authors explain this by comparing the KW operator to a Fourier Transform (a math tool that turns a sound wave into a frequency chart).
  • A normal Fourier Transform looks at the "charge" (the position of the dancer).
  • This new "Dipolar Fourier Transform" looks at the difference between dancers (the "dipole").
  • Because the "difference" between dancers has a rule (the total difference must be zero), you lose information about the absolute positions. You can't reverse it because many different original dances could result in the same "difference" pattern.

Summary of the Impact

1. Universal Translator: They created a universal translator that converts between AI-based descriptions (NQS) and traditional physics descriptions (MPS/TPS) for a wide class of quantum states.
2. Efficiency: They proved that for certain complex quantum states, the "web" structure (TPS) is a much more compact and natural way to describe the physics than the "chain" structure (MPS), even if it's slightly harder to calculate.
3. Demystifying AI: They showed that the "hidden layers" in AI neural networks aren't just magic black boxes; they have a very clear, physical meaning in quantum mechanics.

In a nutshell: The authors took a complex quantum dance, realized it could be described as a neural network, used that to build a better 3D map (TPS) for complex interactions, and explained a weird "broken mirror" symmetry by comparing it to a specialized type of frequency analysis. It's a bridge between the world of Artificial Intelligence and the fundamental laws of quantum matter.

Technical Summary

1. Problem Statement

The paper addresses the challenge of efficiently representing and analyzing many-body quantum wavefunctions, specifically Symmetry-Protected Topological (SPT) states. While Neural Quantum States (NQS), particularly those based on Restricted Boltzmann Machines (RBM), have proven successful for specific models (like the $Z_2$ cluster state), there is a need to:

1. Generalize these representations to arbitrary $Z_N$ symmetries and modulated SPT (mSPT) states (e.g., dipolar and quadrupolar symmetries).
2. Clarify the mathematical relationship between NQS and traditional tensor network states (Matrix Product States - MPS, and Tensor Product States - TPS).
3. Develop a unified framework to handle complex symmetries, including non-invertible symmetries like the Kramers-Wannier (KW) duality, which are difficult to treat with standard MPS approaches.

2. Methodology

The authors introduce and utilize a novel framework called the P-representation (Projector representation) to bridge the gap between wavefunction definitions, NQS, and tensor networks.

• The P-Representation: The wavefunction is expressed as a trace of a product of local projectors ($P_s = |s\rangle\langle s|$) and interaction matrices ($\Omega$). The interaction matrix encodes the correlations between physical spins.
• Decomposition to NQS: The interaction matrix $\Omega$ is decomposed into a product of two weight matrices ($W$ and $W^t$), analogous to the factorization in RBMs. This allows the derivation of explicit closed-form weight functions $W(s, h)$ coupling physical spins ($s$) to hidden variables ($h$).
• Mapping to Tensor Networks: By grouping the weight matrices and projectors differently around each site, the authors derive:
  • MPS: For nearest-neighbor interactions (standard $Z_N$ cluster state).
  • TPS: For longer-range or modulated interactions (dipolar cluster states), where local tensors carry three virtual indices connecting to nearest and next-nearest neighbors.
• Symmetry Analysis: The P-representation is used to prove symmetry invariance and symmetry fractionalization via "push-through" conditions, where symmetry operators acting on physical indices are moved through the interaction matrices to act on virtual indices.
• Kramers-Wannier (KW) Operator: The authors construct the Matrix Product Operator (MPO) for the KW operator using the P-representation, interpreting it as a dipolar Fourier transform.

3. Key Contributions

A. Generalization to $Z_N$ and Closed-Form Weights

The paper provides the first closed-form RBM weight function for general $Z_N$ cluster states, generalizing previous $Z_2$ results. The weight function is given by:
$$W(s,h) = \kappa^{-1/2} \omega^{ah^2 + bs^2 + csh}$$
where coefficients $a, b, c$ depend on $N$. This allows for the exact encoding of $Z_N$ cluster states and their dipolar variants.

B. Unification of NQS, MPS, and TPS

The authors demonstrate that NQS, MPS, and TPS are not distinct paradigms but different organizational views of the same underlying structure derived from the P-representation:

• $Z_N$ Cluster State: The P-representation yields an MPS with bond dimension $N$.
• Type-I Dipolar Cluster: Also admits an MPS representation.
• Type-II Dipolar Cluster: Requires a Tensor Product State (TPS) with rank-3 tensors (three virtual indices per site) to naturally capture the interaction between a site, its nearest neighbor, and its next-nearest neighbor.

C. Efficient Representation of Modulated SPTs

For the Type-II dipolar cluster state (protected by two charge and two dipole symmetries), the authors show that the natural TPS representation is more compact than a conventional MPS.

• TPS Efficiency: The local TPS tensor has $N^4$ elements.
• MPS Inefficiency: To represent the same state via MPS, the bond dimension must scale as $\chi = N^2$, resulting in $N^5$ elements per local tensor.
• Note: While TPS is more compact locally, the authors note that exact contraction is computationally more expensive ($O(N^8 L)$ vs $O(N^7 L)$) due to loops introduced by the non-nearest-neighbor connections, suggesting TPS is primarily a superior representational tool for understanding the state's structure.

D. Kramers-Wannier as Dipolar Fourier Transform

The paper offers a new physical interpretation of the non-invertible Kramers-Wannier (KW) operator.

• Interpretation: The KW transformation is identified as a dipolar Fourier transform, where the target variable is the dipole moment $d_j = s_j - s_{j+1}$ rather than the charge $s_j$.
• Non-invertibility: The mapping is non-invertible because the dipole variables are constrained ($\sum d_j = 0$), covering only a $1/N$ fraction of the Hilbert space.
• MPO Construction: The authors derive a compact MPO for the KW operator in the P-representation (the $P_2$-representation), generalizing previous $Z_2$ results to arbitrary $Z_N$.

4. Key Results

1. Exact NQS Construction: Explicit derivation of NQS weights for $Z_N$ cluster states and two types of dipolar cluster states (Type-I and Type-II).
2. Symmetry Fractionalization: Successful demonstration of how global symmetries fractionalize into edge operators in open chains using the P-representation, providing simpler proofs than traditional MPS methods.
3. TPS Benchmarking: Numerical benchmarks using Density Matrix Renormalization Group (DMRG) confirm that the Type-II dipolar cluster state naturally requires a higher bond dimension in MPS, validating the necessity of the TPS structure for efficient representation.
4. Fusion Rules: Verification of the fusion rules for the non-invertible symmetry generators (e.g., $K_c^\dagger K_c = \sum C_1^m \sum C_2^n$), confirming the algebraic structure of the symmetries.

5. Significance

• Theoretical Unification: The work establishes a rigorous bridge between machine learning-based quantum states (NQS) and tensor network theory, showing that NQS can systematically generate appropriate tensor network ansatzes (MPS/TPS) for complex SPT phases.
• Handling Modulated Symmetries: It provides a systematic toolkit for analyzing mSPT states, which are crucial for understanding modern topological phases protected by spatially modulated symmetries (dipolar, quadrupolar).
• Insight into Non-Invertible Symmetries: By interpreting the KW operator as a dipolar Fourier transform, the paper offers an intuitive explanation for the non-invertibility of these symmetries, a topic of intense current interest in high-energy and condensed matter physics.
• Future Directions: The framework suggests that perturbations to these fixed-point states can be viewed as renormalizations of the interaction matrices, opening avenues for using AI to study non-fixed-point SPT phases and 2D topological states (like the toric code) with $Z_N$ degrees of freedom.

In summary, this paper provides a comprehensive mathematical framework that unifies neural network and tensor network approaches, offering new insights into the structure of $Z_N$ SPT phases and the nature of non-invertible symmetries.