Tensor Cross Interpolation of Purities in Quantum Many-Body Systems

This paper demonstrates that the entanglement feature, which encodes subsystem purities of quantum many-body states, can be efficiently learned from a polynomial number of samples using the tensor cross interpolation algorithm, enabling applications such as quantifying entanglement pattern distances and optimizing physical index ordering.

The Gist

The Big Problem: The "Infinite Library"

Imagine a quantum system (like a collection of tiny magnets or atoms) as a massive library. In a normal library, if you want to know everything about a book, you read it. But in a quantum library, the number of possible "books" (states) grows so fast that if you add just a few more shelves (particles), the library becomes larger than the number of atoms in the universe.

Physicists usually try to understand these systems by looking at small, specific sections (like "how entangled is the left half with the right half?"). But this is like trying to understand a whole novel by only reading the first and last sentences of every chapter. You miss the complex connections in the middle.

The Solution: The "Entanglement Feature"

The authors propose a clever way to store the "purity" (a measure of how mixed or pure a quantum state is) of every possible section of the system.

Think of the quantum state as a giant, complex tapestry. Usually, describing every thread is impossible. The authors suggest encoding the information about how "tangled" every possible cut of the tapestry is into a single, special "shadow" or "feature map." They call this the Entanglement Feature.

Surprisingly, even for very messy, complex quantum states, this "feature map" isn't actually that messy. It often has a simple, hidden structure, much like a complex song might actually be built from a simple repeating melody.

The Tool: "Tensor Cross Interpolation" (TCI)

The big question is: How do we find this simple structure without reading the entire, impossible library?

The authors use a technique called Tensor Cross Interpolation (TCI).

• The Analogy: Imagine you are trying to guess the plot of a massive, 1,000-page mystery novel, but you are only allowed to read a few pages.
• The Old Way: You read page 1, then page 2, then page 3... all the way to the end. This takes forever and is impossible for huge books.
• The TCI Way: The algorithm acts like a super-smart detective. It reads a few strategic pages (pivots). Based on those, it guesses the structure of the rest. Then, it checks its guess against a few new pages. If the guess is good, it stops. If not, it adjusts.
• The Result: Instead of reading 1,000 pages, the detective only needs to read a handful (a polynomial number) to understand the whole story. The paper shows that for many quantum systems, this "detective" can reconstruct the entire "feature map" using very few samples.

What They Found

The researchers tested this method on different types of quantum "stories":

1. Random Chaos (Haar States): These are like pure noise. You might think they are too messy to compress. However, the authors found that even for these chaotic states, the "feature map" is surprisingly simple and easy to learn once the system gets big enough.
2. Ordered States (Area-Law): These are like well-organized libraries. As expected, their feature maps are very simple and easy to compress.
3. The "Goldilocks" Zone (Phase Transitions): They looked at systems right on the edge of changing phases (like water turning to ice). Here, the feature map is tricky. Sometimes it's easy to learn; other times, it remains complex and hard to compress, revealing that these states have a unique, stubborn complexity.

What You Can Do With This

The paper demonstrates two specific ways to use this "feature map" once you have learned it:

1. The "Similarity Test": You can compare two different quantum states not just by how much they are entangled on average, but by comparing their entire "feature maps." It's like comparing two people not just by their height, but by comparing their entire fingerprints. This helps group similar quantum states together and spot the weird outliers.
2. The "Reordering Puzzle": Imagine you have a deck of cards that has been shuffled randomly. The connections between the cards look chaotic. The authors show that by looking at the "feature map," you can figure out the original order of the cards. If you rearrange the physical parts of the quantum system into this "optimal order," the chaos disappears, and the system becomes much easier to describe and store.

Summary

The paper introduces a new way to "compress" the overwhelming complexity of quantum systems. By treating the purity of all possible sections as a single, learnable object (the Entanglement Feature) and using a smart sampling algorithm (TCI), they can reconstruct the whole picture from just a few data points. This allows physicists to compare complex quantum states and even find the best way to arrange them to make them simpler.

Technical Summary

Technical Summary: Tensor Cross Interpolation of Purities in Quantum Many-Body Systems

Problem Statement
The complete characterization of quantum many-body systems is fundamentally hindered by the exponential scaling of the Hilbert space with the number of degrees of freedom ($L$). While standard probes often focus on local observables or specific bipartite cuts (e.g., contiguous left/right partitions), a full description requires knowledge of all bipartite purities (or Renyi entropies) for all possible disjoint regions. Storing this information naively requires $O(2^L)$ resources. Although the "entanglement feature" (EF)—a fictitious wave function encoding all purities as amplitudes—has been proposed as a compact representation, its practical utility for general quantum states remains unexplored. Specifically, it is unclear if the EF for complex, highly entangled states can be efficiently learned from a limited set of data without prior knowledge of the full wave function.

Methodology
The authors propose using the Tensor Cross Interpolation (TCI) algorithm to learn and interpolate the entanglement feature.

• Entanglement Feature (EF): Defined as $|EF\rangle = \sum_{b \in \{0,1\}^L} e^{-S(b)} |b\rangle$, where $S(b)$ is the second Renyi entropy for a bipartition defined by bitstring $b$. The authors utilize a "dual basis" to remove $Z_2$ redundancy, mapping the problem to $L-1$ virtual spins.
• TCI Algorithm: Instead of computing all $2^L$ purities and performing a full Singular Value Decomposition (SVD) sweep (which is exponentially expensive), TCI iteratively selects a small subset of "pivot" bitstrings (purity values) to construct a low-rank approximation of the full tensor.
• Efficiency: If the EF can be represented as a Matrix Product State (MPS) with bond dimension $\chi$, TCI requires only $O(L\chi^2)$ purity evaluations to reconstruct the full feature. The authors employ the TensorCrossInterpolation.jl package in "Mode II," which adaptively increases the bond dimension until a specified error tolerance is met.

Key Contributions and Results
The paper systematically benchmarks the complexity of the EF across various quantum states using TCI:

1. Haar Random States:

   • Analytic Expectation: The average EF of Haar states is known to be an MPS with bond dimension $\chi=2$.
   • Numerical Finding: For individual Haar realizations, the TCI bond dimension $\chi$ exhibits a size-dependent transition. For small $L$, sample-to-sample fluctuations in purity are large, requiring an exponentially large $\chi$ to interpolate the "noise." As $L$ increases, fluctuations decay exponentially ($\sim d^{-L}$), allowing TCI to ignore noise and converge to $\chi=2$.

2. Random Matrix Product States (MPS):

   • Analytic Insight: While a direct construction of the EF from an MPS of bond dimension $\phi$ yields $\chi = \phi^4$, the authors argue that for large $\phi$, the EF spectrum should collapse toward the Haar limit ($\chi=2$).
   • Numerical Finding: The gap between the second and third singular values of the EF grows with $\phi$, and the ratio of higher singular values decays as $\sim \phi^{-4}$, confirming that random MPS with large $\phi$ possess a compressible EF.

3. Many-Body Quantum Eigenstates:
   The authors apply TCI to eigenstates of various 1D Hamiltonians, identifying three distinct complexity regimes:

   • Volume-Law (ETH-like) States: Highly excited eigenstates of chaotic Ising models and SYK ground states. Similar to Haar states, these show an initial exponential growth in $\chi$ followed by a drop to a constant (or slowly growing) value at large system sizes, indicating compressibility.
   • Area-Law States: Highly excited eigenstates in the Many-Body Localized (MBL) phase. These exhibit a constant, small bond dimension $\chi$ across system sizes, consistent with their low entanglement.
   • Intermediate/Complex States:
     • MBL Transition (MBLT) & Critical Ground States: While these states have logarithmic entanglement scaling, their EF complexity varies. MBLT eigenstates show an exponential increase in $\chi$ with system size, suggesting their EF is not compressible despite the sub-volume law entanglement.
     • Motzkin/Fredkin Chains: The EF of the two-color Motzkin chain grows exponentially, whereas the one-color (colorless) Motzkin and Fredkin states are efficiently learned. This highlights that entanglement scaling alone does not predict EF compressibility.

Applications Demonstrated
The paper demonstrates two practical applications of the learned EF:

1. Quantifying Entanglement Distance: The authors define a distance metric between states based on the $L_2$-norm of the difference in their EFs. Using stress majorization, they visualize a "entanglement map" where states cluster according to their global purity structure (e.g., volume-law states near Haar, area-law states near product states). This reveals outliers (like the two-color Motzkin state) that are distinct from standard volume/area-law clusters.
2. Optimal Spatial Ordering: The authors propose an algorithm to find the optimal ordering of physical indices to minimize bipartite entanglement. By using the EF as a "purity oracle" and a divide-and-conquer strategy with TTOpt (a TCI-based optimizer), they successfully reorder randomly shuffled MPS states. This recovers area-law scaling from apparent volume-law behavior, reducing the mid-cut entropy by a factor of $\sim 3.4$ for $L=40$.

Significance and Claims
The authors conclude that the entanglement feature is a powerful framework for characterizing quantum states beyond standard entanglement scaling.

• Compressibility: They demonstrate that for a broad class of states (including chaotic eigenstates and MBL states), the EF is efficiently learnable via TCI, requiring only polynomial resources in system size.
• Information Content: The EF contains more information than simple entanglement scaling, capable of distinguishing states with similar scaling but different internal structures (e.g., MBLT vs. Area-law).
• Utility: The method provides a scalable way to process data from quantum simulators (snapshots) to reconstruct global properties.
• Future Directions: The paper speculates on applications to multipartite entanglement quantification (via log-EF and mutual information) and quantum error correction (identifying correctable erasure sets), though these are presented as conceptual possibilities rather than implemented results. They also suggest TCI could be adapted for searching resonances in MBL phases or interpolating observables from noisy intermediate-scale quantum (NISQ) devices.

The work establishes that the "complexity" of a quantum state's full purity profile is often much lower than the Hilbert space dimension suggests, provided the state admits a low-rank MPS representation of its entanglement feature.