A Study of Entanglement and Ansatz Expressivity for the Transverse-Field Ising Model using Variational Quantum Eigensolver

This paper investigates the effectiveness of various variational quantum eigensolver ansatzes, including hardware-efficient and physics-inspired approaches, in simulating the transverse-field Ising model across one, two, and three dimensions by benchmarking their performance on systems up to 27 qubits using metrics such as energy variance, entanglement entropy, and spin correlations.

The Gist

Imagine you are trying to find the absolute lowest point in a vast, foggy mountain range. This lowest point represents the "ground state" of a complex quantum system—the most stable, energy-efficient way particles can arrange themselves.

In the world of quantum computing, finding this point is the job of an algorithm called VQE (Variational Quantum Eigensolver). Think of VQE as a hiker equipped with a map (the quantum computer) and a compass (a classical computer). The hiker's goal is to adjust their path until they hit the bottom of the valley.

However, there's a catch: the hiker can only walk on specific trails laid out by a circuit design (called an ansatz). If the trails don't go near the bottom of the valley, the hiker will never find the true lowest point, no matter how good their compass is.

This paper is a study by researchers at the Tata Institute of Fundamental Research (TIFR) in India. They wanted to figure out: Which trail design works best for finding the bottom of the valley in a specific quantum mountain range called the "Transverse-Field Ising Model" (TFIM)?

Here is the breakdown of their adventure using simple analogies:

1. The Mountain Range (The TFIM)

The TFIM is a model used to study how tiny magnets (spins) interact with each other.

• The "Wind" (Transverse Field): Imagine a strong wind blowing across the mountains.
  • Weak Wind: The magnets want to line up together (like soldiers standing in formation). This state is very "entangled," meaning the magnets are deeply connected; you can't describe one without describing all the others.
  • Strong Wind: The wind blows them apart, and they spin randomly. This state is "less entangled" and easier to describe.
• The Challenge: The researchers wanted to see if their quantum hikers could find the perfect arrangement of magnets, especially when the wind was weak and the connections were complex.

2. The Trail Designs (The Ansatzes)

The researchers tested three different "trail maps" (circuit designs) to see which one could reach the bottom of the valley best:

• The "Swiss Army Knife" (HEA / EfficientSU2):
  • What it is: A generic, flexible trail that uses standard hardware gates. It's like a Swiss Army knife—versatile and easy to carry.
  • Pros: It has many adjustable knobs (parameters), so it can theoretically reach almost anywhere.
  • Cons: Because it's so flexible, the path is full of confusing bumps and dead ends (local minima). The hiker often gets stuck in a small dip, thinking it's the bottom, when it's not.
• The "Specialized Guide" (HVA):
  • What it is: A trail built specifically for this mountain. It follows the physics of the magnets step-by-step.
  • Pros: It stays on the right path and avoids many dead ends.
  • Cons: It's rigid. If the mountain changes slightly, this trail might not be able to bend enough to reach the true bottom.
• The "Guide with a Detour" (HVA-SB):
  • What it is: The Specialized Guide, but with a special "symmetry-breaking" layer added.
  • Why it helps: Sometimes the mountain has two identical valleys (degenerate states). The rigid guide gets stuck trying to pick one. This version adds a "detour" (a symmetry-breaking gate) that forces the hiker to pick a side, helping them settle into the correct valley.

3. The Experiment

The researchers simulated these hikers on computers (using NVIDIA GPUs) for mountains of different sizes:

• 1D: A single line of magnets (easy to visualize).
• 2D: A grid of magnets (like a checkerboard).
• 3D: A cube of magnets (very complex).

They measured two things:

1. Energy Accuracy: Did they find the true bottom of the valley?
2. Entanglement: Did they capture the deep "connection" between the magnets?

4. What They Found (The Results)

• The "Swiss Army Knife" (HEA) is expressive but tricky:
  It has the most potential to find the right answer because it's so flexible. However, it often gets lost in the fog. In the 3D simulations, it tended to pick a "symmetry-broken" state (choosing a side) which is actually correct for a huge, infinite system, but slightly wrong for the small, finite systems they were testing. It also tended to underestimate how connected the magnets were.

• The "Specialized Guide" (HVA) is stable but rigid:
  It was very good at finding the energy in the "easy" parts of the mountain (strong wind). But when the wind was weak and the magnets were deeply entangled, this guide failed. It couldn't bend enough to reach the true bottom.

• The "Guide with a Detour" (HVA-SB) was the hero:
  By adding that small symmetry-breaking layer, it managed to combine the best of both worlds. It stayed on the physics-based path but had just enough flexibility to handle the tricky, highly entangled states. It gave the most accurate results for the energy.

5. The Big Takeaway

The paper concludes that there is no perfect trail.

• If you want flexibility, you need a complex circuit (like HEA), but you risk getting stuck in optimization traps.
• If you want stability, you need a physics-based circuit (like HVA), but you might miss the most complex, entangled states.

The researchers found that entanglement is hard to capture. Even the best quantum computers today (NISQ era) struggle to perfectly simulate the deep connections between particles in 3D.

In summary: To solve these quantum puzzles, we need to build better "maps" (ansatzes) that are smart enough to follow the physics but flexible enough to handle the messiness of real-world noise. The researchers suggest that in the future, we might use AI to help design these maps or to help the hikers navigate the foggy valleys more efficiently.

Technical Summary

1. Problem Statement

The Variational Quantum Eigensolver (VQE) is a leading hybrid quantum-classical algorithm for simulating many-body systems in the Noisy Intermediate-Scale Quantum (NISQ) era. However, its effectiveness is heavily contingent on two factors:

1. Ansatz Expressivity: The ability of the parametric quantum circuit (PQC) to represent the true ground state, particularly in regimes of strong entanglement or degeneracy.
2. Optimization Stability: The ability of the classical optimizer to navigate the loss landscape to find the optimal parameters.

The Transverse-Field Ising Model (TFIM) serves as an ideal testbed for this problem because its ground state undergoes a quantum phase transition. In the low transverse-field regime ($h_x \ll h_c$), the ground state is highly entangled (superposition of $|00\dots0\rangle$ and $|11\dots1\rangle$), while in the high-field regime ($h_x \gg h_c$), it is a product state with low entanglement. The authors aim to benchmark different ansatz architectures to determine how well they capture these entanglement properties and critical behaviors across 1D, 2D, and 3D lattices.

2. Methodology

Model and Setup

• Hamiltonian: The TFIM Hamiltonian is defined as $H = J_z \sum_{\langle ij \rangle} \sigma^z_i \sigma^z_j + h_x \sum_i \sigma^x_i$, with $J_z = -1.0$ and periodic boundary conditions.
• System Sizes: Simulations were performed on systems ranging up to 27 qubits (specifically 10 sites in 1D, $4\times4$ in 2D, and various sizes in 3D).
• Simulation Framework: Exact VQE simulations were conducted using the NVIDIA CUDA-Q simulator on GPUs, allowing for exact state-vector simulations to benchmark against exact diagonalization.

Ansatz Architectures

Three distinct classes of ansatzes were compared:

1. Hardware-Efficient Ansatz (HEA): Specifically the EfficientSU2 from Qiskit. It uses native hardware gates ($R_z, R_x, CNOT$) and is designed for high expressivity.
2. Hamiltonian Variational Ansatz (HVA): Constructed by Trotterizing the TFIM Hamiltonian terms. It uses $R_{ZZ}$ gates for coupling and $R_x$ gates for the transverse field. It is physically inspired but restricted to a specific subspace.
3. HVA with Symmetry Breaking (HVA-SB): An extension of HVA that adds an $R_z$ layer to explicitly break the $Z_2$ parity symmetry, allowing the circuit to access states that are superpositions of parity sectors.

Optimization Strategies

• HEA: Optimized using L-BFGS due to its smooth optimization landscape.
• HVA and HVA-SB: Optimized using COBYLA (derivative-free) because their landscapes are more rugged and prone to local minima.
• 3D Extension: A "Real Amplitude" ansatz (removing $R_z$ rotations to restrict to real amplitudes) was used for 3D to reduce parameters and maintain a smooth landscape.

Metrics for Evaluation

• Expressivity: Measured via Frame Potential ($F_t$), where lower values indicate higher expressivity (closer to a unitary t-design).
• Energy Accuracy: Measured via Energy Variance ($Var(E) = \frac{\langle H^2 \rangle - \langle H \rangle^2}{\langle H \rangle^2}$).
• Physical Observables:
  • Magnetization ($M$): $\frac{1}{N}\sum \langle \sigma^z_i \rangle$.
  • Spin Correlation ($M_{Corr}$): $\frac{1}{N/2}\sum \langle \sigma^z_i \sigma^z_{i+N/2} \rangle$ (used to probe criticality in finite systems).
  • Entanglement Entropy: Von Neumann entropy ($S_{EE}$) calculated via reduced density matrices.

3. Key Contributions

1. Systematic Benchmarking of Expressivity vs. Optimization: The study quantifies the trade-off between the expressivity of an ansatz (HEA) and the stability of its optimization landscape (HVA).
2. Role of Symmetry Breaking: The paper demonstrates that adding symmetry-breaking layers (HVA-SB) is crucial for accessing the correct ground state in finite-size systems where the true ground state is a superposition of degenerate symmetry-broken states.
3. Dimensionality Scaling: The work extends VQE benchmarks from 1D to 2D and 3D, highlighting how increased connectivity and entanglement in higher dimensions exacerbate optimization difficulties.
4. Entanglement Underestimation: The authors identify a systematic limitation where VQE, particularly with shallow circuits, tends to underestimate entanglement entropy in highly entangled regimes.

4. Results

1D System (10 Sites)

• Energy Variance: HEA showed gradual improvement with increasing circuit depth. In contrast, HVA and HVA-SB exhibited a sharp transition from poor to accurate energy estimation once sufficient depth was reached. This suggests HVA benefits from being restricted to the Hamiltonian subspace.
• Entanglement & Correlations:
  • HVA produced high single-site entanglement entropy ($S \approx 1$) in the low-field regime because it failed to spontaneously break symmetry in finite systems.
  • HEA and HVA-SB correctly captured the qualitative curvature changes near the critical point, though HEA tended to select symmetry-broken states (low entanglement) which is physically correct only in the thermodynamic limit.

2D System ($4\times4$ Lattice)

• Optimization Instability: Optimization became significantly more unstable due to increased connectivity.
• Performance: HVA performed poorly in low-entanglement regimes. HEA underestimated entropy in highly entangled low-field regions, again tending toward symmetry-broken states.
• Criticality: Both spin correlation and entanglement entropy successfully signaled the phase transition near the critical field $h_x \approx 3.0$.

3D System

• Challenges: Optimization challenges intensified further.
• Solution: A modified HEA (Real Amplitude) with parameter initialization from nearby transverse-field points allowed for reasonable ground-state energy accuracy.
• Findings: Ground-state energy per site remained independent of lattice size, while entanglement entropy continued to show clear signatures of critical behavior.

5. Significance and Conclusion

The paper concludes that expressivity and optimization stability are competing requirements in VQE:

• HEA offers high expressivity but suffers from irregular optimization landscapes and can get stuck in local minima or select incorrect symmetry sectors in finite systems.
• HVA offers a smoother path to the ground state within its restricted subspace but lacks the flexibility to capture states outside that subspace without symmetry-breaking modifications.
• HVA-SB bridges this gap by allowing access to parity-violating states, improving energy estimation.

Implications:

• For finite-size simulations, ansatz selection is critical for correctly interpreting entanglement and phase transitions.
• Shallow circuits struggle to capture highly correlated states, leading to systematic underestimation of entanglement.
• Future work should focus on adaptive ansatz strategies and machine-learning-based optimizers to mitigate these limitations and scale VQE to larger, strongly correlated systems.

This study provides a rigorous framework for selecting ansatzes based on the specific physical regime (entanglement level) and system dimensionality, offering practical guidance for NISQ-era simulations of quantum phase transitions.