⚛️ quantum physics

Frustration-Induced Expressibility Limitations in Variational Quantum Algorithms

This paper demonstrates that geometric frustration in quantum many-body systems limits the expressibility of standard variational quantum algorithm ansätze, necessitating deeper circuits or bond-resolved parameters to accurately capture inhomogeneous correlations and near-degenerate spectra, a challenge distinct from optimization issues like barren plateaus.

Original authors: Sandip Maiti

Published 2026-04-14

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Sandip Maiti

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are trying to teach a robot to solve a complex puzzle. This puzzle represents a quantum system—a tiny world where particles interact in strange ways. The robot uses a specific set of instructions (an "ansatz") to guess the solution. Usually, this works great. But sometimes, the puzzle has a special feature called "frustration," and suddenly, the robot gets stuck, no matter how hard it tries.

This paper, written by Sandip Maiti, investigates exactly why this happens and how to fix it. Here is the story in simple terms:

1. The Puzzle: Geometric Frustration

Imagine a group of friends trying to sit at a round table.

The Rule: Everyone wants to sit next to their best friend, but they also want to sit opposite their rival.
The Problem: If you have three friends in a triangle, and everyone wants to sit next to their best friend, you can't satisfy everyone at once. If A sits next to B, and B sits next to C, then C is forced to sit next to A, even if they are rivals.

In physics, this is called geometric frustration. The particles (spins) are like those friends. They have competing rules that make it impossible to find a single "perfect" arrangement where everyone is happy. This creates a chaotic, messy energy landscape with many almost-equal solutions, rather than one clear winner.

2. The Robot's Mistake: The "One-Size-Fits-All" Approach

The researchers tested a standard robot instruction set called the Hamiltonian Variational Ansatz (HVA).

How it works: Think of this robot as a painter who uses a single brush size for the whole painting. If the painting has a smooth blue sky, this works perfectly.
The Failure: When the robot tried to paint the "frustrated" puzzle, it failed. Why? Because the frustrated puzzle isn't smooth. Some parts of the triangle need a "red" connection, while others need a "blue" connection.
The Analogy: The robot was trying to paint a complex, patchwork quilt using only one color of thread. It kept trying to force the whole pattern to look the same, but the pattern needed to be different in every spot.

The result? The robot had to make the painting incredibly deep and complex (increasing the "circuit depth") just to get a mediocre result. It was like trying to fix a broken watch by hitting it with a sledgehammer; the more you hit it, the worse it gets.

3. The Real Culprit: Not "Stuck," Just "Blind"

A common fear in quantum computing is the "Barren Plateau." This is like a robot getting lost in a foggy field where it can't see any direction to go (the gradients vanish).

The Discovery: The authors found that this wasn't the problem. The robot could see the direction (the gradients were fine).
The Real Issue: The robot was blind to the details. Its instruction manual was too rigid. It lacked expressibility. It simply didn't have the "tools" to describe the messy, patchwork reality of the frustrated system. It wasn't lost; it was just wearing the wrong glasses.

4. The Solution: The "Custom-Tailored" Robot

To fix this, the researchers gave the robot a new set of instructions: the Bond-Resolved Ansatz.

The Change: Instead of using one brush for the whole painting, they gave the robot a different brush for every single connection between friends.
The Result: Now, the robot could paint the "red" connections red and the "blue" connections blue. It could adapt to the specific messiness of the triangle.
The Outcome: Suddenly, the robot solved the puzzle with a much simpler, shallower circuit. It didn't need to hit the watch with a sledgehammer anymore; it just needed the right screwdriver.

5. The Bigger Picture: Excited States and Future Challenges

The paper also looked at what happens when the system isn't in its "resting" state but is excited (like a friend jumping up from the chair).

In frustrated systems, these excited states are very close together (like a crowded room where everyone is standing shoulder-to-shoulder).
The robot struggles to tell them apart. While the new "custom-tailored" instructions helped with the main puzzle, separating these crowded excited states remains a tough challenge, especially as the system gets bigger.

Summary

The Main Takeaway:
When quantum systems are "frustrated" (conflicted), they become messy and unique. Standard quantum algorithms that use a "one-size-fits-all" approach fail because they can't capture this messiness. They aren't failing because the math is too hard to solve; they are failing because their design is too rigid.

The Lesson for the Future:
To build better quantum computers for these difficult problems, we shouldn't just make the circuits deeper (more complex). Instead, we need to design smarter circuits that respect the local details of the problem. We need to give the algorithm a "local map" rather than a "global map."

By doing this, we can solve these complex quantum puzzles much faster and with less energy, bringing us closer to using quantum computers for real-world problems like designing new materials or drugs.

1. Problem Statement

Variational Quantum Algorithms (VQAs), such as the Variational Quantum Eigensolver (VQE) and Quantum Approximate Optimization Algorithm (QAOA), are leading candidates for simulating strongly correlated quantum many-body systems on Noisy Intermediate-Scale Quantum (NISQ) devices. However, their performance degrades significantly when applied to geometrically frustrated systems (e.g., antiferromagnetic models on lattices with triangular plaquettes).

While previous studies have attributed these failures to optimization issues like barren plateaus (vanishing gradients) or insufficient circuit depth, the specific physical origin of this degradation in frustrated regimes remains unclear. Furthermore, existing work often focuses on ground states, leaving the behavior of variational methods for low-energy excitations in frustrated systems under-explored. The core problem is understanding why standard, Hamiltonian-inspired ansätze fail to capture the complex correlation structures induced by frustration and how to overcome these limitations without prohibitive circuit depth.

2. Methodology

The author investigates these issues using the Transverse-Field Ising Model (TFIM) on a 2D square lattice with frustrated diagonal couplings ( $J_1 = J_2 = J$ ). This model creates triangular plaquettes, inducing geometric frustration in the antiferromagnetic regime ( $J < 0$ ).

Key Methodological Components:

Models & Regimes: Simulations cover various transverse field strengths ( $h$ $h$ ):
- Frustrated ordered regime ( $h=0.5$ ).
- Crossover/near-critical region ( $h=1.0, 1.5$ ).
- Paramagnetic regime ( $h=4.0$ ).
Ansätze Comparison:
- Standard Hamiltonian Variational Ansatz (HVA): Uses globally shared parameters for all interaction terms (e.g., a single $\gamma$ for all $ZZ$ interactions per layer). This assumes uniform correlation structures.
- Bond-Resolved HVA: A novel modification where independent variational parameters are assigned to each bond/interaction term. This allows the circuit to adapt to local, inhomogeneous correlation patterns.
- Hardware-Efficient Ansatz (HEA): Used as a baseline for generic expressibility without physics-informed structure.
Algorithms for Excited States:
- Variational Quantum Deflation (VQD): Uses a penalty term to enforce orthogonality to lower-energy states.
- Symmetry-Resolved Approach: Exploits the global $Z_2$ symmetry of the TFIM to restrict optimization to specific parity sectors (even/odd), avoiding penalty terms.
Numerical Benchmarks:
- Exact Diagonalization (ED): Used for small systems ( $L=2, 3, 4$ ) to provide ground truth.
- Krylov Subspace Diagonalization: Used for larger systems ( $L=5$ ) where ED is intractable, serving as a reference for low-energy spectra.
- Gradient Analysis: Monitoring the Euclidean norm of the energy gradient to distinguish between barren plateaus and expressibility limits.

3. Key Contributions

Identification of Expressibility Limitations: The paper demonstrates that the failure of VQAs in frustrated regimes is not primarily caused by barren plateaus (optimization landscape issues) but by insufficient expressibility of standard ansätze. Standard HVA circuits with global parameters cannot capture the spatially inhomogeneous correlations induced by frustration.
Bond-Resolved Ansatz: The introduction of a bond-resolved HVA, where parameters are decoupled for each bond, significantly recovers accuracy. This approach adapts the variational manifold to the specific local correlation structure, achieving high fidelity at much shallower circuit depths compared to the standard HVA.
Excitation Gap Challenges: The study reveals that near-degenerate spectra in frustrated regimes pose severe challenges for excited state calculations. While symmetry-resolved methods outperform penalty-based VQD in these regimes, the dense spectrum still requires high expressibility to resolve.
Physical Origin of Degradation: The work establishes a direct link between geometric frustration, the emergence of heterogeneous bond correlations, and the inability of uniform variational structures to represent the ground state efficiently.

4. Key Results

Inhomogeneous Correlations: Exact Diagonalization (ED) of a $3 \times 3$ lattice at low fields ( $h=0.5$ ) shows that frustration leads to a heterogeneous pattern of correlations. Some bonds are energetically satisfied (antiferromagnetic), while others remain frustrated, creating a non-uniform correlation profile that global parameters cannot replicate.
Performance Degradation:
- In the frustrated regime, the standard HVA requires a dramatic increase in circuit depth to reach high fidelity ( $f > 0.99$ ). For a $3 \times 3$ lattice at $h=0.5$ , the standard HVA requires depth $p=24$ and still shows deviation, whereas the bond-resolved HVA achieves similar accuracy at $p=8$ .
- Gradient Analysis: In the frustrated regime, gradients do not vanish (ruling out barren plateaus) but saturate at finite values while fidelity remains low. This confirms the system is trapped in a suboptimal solution due to the restricted variational manifold, not a flat landscape.
Resource Efficiency: Table I shows that the standard HVA requires a massive increase in CNOT gate counts in the frustrated regime compared to the paramagnetic regime. The bond-resolved HVA significantly reduces these resource requirements.
Excitation Spectra:
- In the paramagnetic regime (large gaps), both VQD and symmetry-resolved methods work well.
- In the frustrated regime (near-degenerate states), VQD becomes unstable due to the difficulty of enforcing orthogonality. Symmetry-resolved approaches provide a more robust path but still struggle as system size increases and spectral crowding intensifies.
Scalability: Using Krylov subspace methods for a $5 \times 5$ lattice, the authors show that the expressibility limitations persist beyond the reach of exact diagonalization, confirming that frustration-induced complexity is an intrinsic feature of the system, not a finite-size artifact.

5. Significance

This paper provides a physical explanation for the limitations of variational quantum algorithms in frustrated systems, shifting the focus from optimization heuristics to ansatz design.

Design Principles: It suggests that for frustrated systems, ansätze must incorporate local flexibility (bond-resolved parameters) rather than relying on global symmetries or uniform structures.
NISQ Relevance: By showing that bond-resolved ansätze can achieve high accuracy at reduced circuit depths, the work offers a practical strategy for simulating frustrated magnets on near-term quantum hardware where circuit depth is a critical constraint.
Future Directions: The results highlight the need for structured ansätze that balance local flexibility with parameter control to prevent exponential scaling of parameters as system size grows.

In summary, the paper argues that geometric frustration creates a correlation landscape that is fundamentally incompatible with uniform variational structures, and overcoming this requires explicitly modeling the spatial inhomogeneity of the system through bond-resolved parameterization.