⚛️ quantum physics

Variational Quantum Eigensolver: A Comparative Analysis of Classical and Quantum Optimizer Methods

This study evaluates classical and quantum optimizers for the Variational Quantum Eigensolver applied to the Ising model, proposing and validating a novel hybrid QN-SPSA+PSR method that combines computational efficiency with precise gradient estimation to enhance convergence and stability on NISQ devices.

Original authors: Duc-Truyen Le, Vu-Linh Nguyen, Cong-Ha Nguyen, Quoc-Hung Nguyen, Van-Duy Nguyen

Published 2026-07-07

📖 4 min read🧠 Deep dive

CC BY 4.0

Original authors: Duc-Truyen Le, Vu-Linh Nguyen, Cong-Ha Nguyen, Quoc-Hung Nguyen, Van-Duy Nguyen

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 find the lowest point in a vast, foggy, and mountainous landscape. This landscape represents a complex quantum problem (specifically, the Ising Model, which describes how tiny magnets interact). Your goal is to find the absolute bottom of the valley (the "ground state energy") because that tells you the most stable, efficient state of the system.

This paper is about building the best possible hiking guide to help you find that bottom quickly and accurately, even when your hiking boots are a bit worn out (representing today's noisy, imperfect quantum computers).

Here is the breakdown of their journey, explained simply:

1. The Problem: The Foggy Mountain

The researchers are using a tool called VQE (Variational Quantum Eigensolver). Think of VQE as a hybrid team:

The Quantum Hiker: A small, noisy quantum computer that can take a quick peek at the terrain but gets tired easily and makes mistakes (noise).
The Classical Coach: A regular supercomputer that analyzes the data from the hiker and tells them which direction to step next.

The challenge is that the "terrain" (the math behind the quantum problem) is incredibly complex. If the coach gives bad directions, the hiker gets stuck on a small hill instead of reaching the deep valley.

2. The Map (The Ansatz)

Before hiking, you need a map. In quantum terms, this is called an Ansatz. It's a specific route or pattern of steps the quantum computer takes.

The authors looked at the specific rules of the "Ising Mountain" (symmetries and how the magnets interact).
Instead of drawing a map that tries to cover every possible path (which would be too heavy and slow), they drew a smart, streamlined map.
They found that a specific type of map called RealAmplitudes (which uses simple, real-number turns) worked best for this specific mountain, saving energy and time compared to more complex maps.

3. The Hiking Guides (Optimizers)

The core of the paper is testing different "hiking guides" (optimizers) to see which one gets the team to the bottom fastest. They tested three main styles:

The Blind Explorer (Classical/COBYLA): This guide doesn't look at the slope at all. It just guesses a direction, checks the height, and tries again. It's cheap and fast but can be slow to find the exact bottom.
The Stochastic Guide (SPSA): This guide takes two quick, random steps to guess which way is down. It's efficient but can be a bit jittery and unstable.
The Precise Surveyor (PSR): This guide measures the slope exactly using a special quantum rule. It's very accurate but takes a long time to measure every single step.
The Quantum Geometer (QN): This guide uses the "shape" of the quantum landscape itself to know the best path. It's powerful but usually requires a massive amount of data to calculate the shape, which is too slow for today's computers.

4. The New Super-Guide: QN-SPSA+PSR

The authors invented a new hybrid guide called QN-SPSA+PSR. Here is the creative analogy for how it works:

Imagine you are navigating a ship in a storm.

The Metric (The Map Shape): You need to know the shape of the ocean floor to steer correctly. Calculating the entire ocean floor is too hard. So, this new guide uses a fast, rough sketch of the ocean floor (using a method called SPSA). This is the "Quantum Natural Gradient" part—it understands the geometry without doing all the heavy math.
The Compass (The Direction): Once you know the general shape, you need to know exactly which way to turn. Instead of guessing randomly (like the Stochastic guide), this guide uses a laser-precise compass (using the Parameter-Shift Rule or PSR) to measure the exact slope.

The Result: By combining the fast, rough map (to save time) with the laser-precise compass (to ensure accuracy), the new guide moves much faster and more stably than the others.

5. The Results

When they tested these guides on a 12-qubit quantum computer simulation:

The new hybrid guide (QN-SPSA+PSR) reached the bottom of the valley faster and more reliably than the "Blind Explorer" or the "Stochastic Guide."
It performed almost as well as the "Precise Surveyor" (which is the gold standard) but didn't require nearly as much computing power.
They also found that for this specific mountain, the simple map (RealAmplitudes) was actually better than the fancy, complex maps.

Summary

The paper claims that by mixing a fast, approximate understanding of the quantum landscape's shape with exact measurements of the slope, they created a new method that helps today's imperfect quantum computers solve problems faster and more accurately. It's a smarter way to navigate the foggy quantum mountains without getting stuck or running out of battery.

Technical Summary: Variational Quantum Eigensolver: A Comparative Analysis of Classical and Quantum Optimizer Methods

Problem Statement
Variational Quantum Algorithms (VQAs), particularly the Variational Quantum Eigensolver (VQE), are currently the most viable approach for leveraging Noisy Intermediate-Scale Quantum (NISQ) devices to solve problems such as finding the ground state energy of quantum systems. However, the performance of VQAs is heavily constrained by the efficiency and stability of the optimization subroutine used to update variational parameters. The optimization landscape often suffers from issues like barren plateaus, noise sensitivity, and high computational costs associated with gradient estimation. While classical optimizers (e.g., COBYLA) and derivative-free methods are robust, they may lack the convergence speed of gradient-based methods. Conversely, quantum-informed optimizers, such as those utilizing the Quantum Natural Gradient (QN), offer geometric advantages but often incur prohibitive computational costs due to the need to estimate the Fubini-Study metric (Quantum Fisher Information Matrix). The paper addresses the need to balance computational efficiency, convergence speed, and stability in VQE optimization, specifically for the Transverse Ising Model (TIM).

Methodology
The study employs the 1D Transverse Ising Model (TIM) as a testbed to evaluate and compare various optimization strategies within the VQE framework. The methodology involves three primary components:

Ansatz Construction: The authors construct parameterized quantum circuits (PQCs) based on the symmetries of the TIM Hamiltonian. Specifically, they utilize the real representation of eigenstates and the $Z_2$ spin-flip symmetry to reduce the parameter space. Two hardware-efficient ansätze are tested:
- RealAmplitudes: Uses $R_y$ rotation gates with linear entanglement, respecting the real-valued nature of the TIM ground state.
- EfficientSU2: Uses a full entanglement scheme with $R_x, R_y, R_z$ gates, allowing for complex-valued Hilbert space representation.
  The study investigates how these structures interact with different optimizers.
Optimizer Classification and Combination: The authors categorize optimizers into classical and quantum operations and create a comparative grid by combining different gradient estimation methods with different metric (Hessian/FIM) approximations:
- Gradient Estimation:
  - Parameter-Shift Rule (PSR): Provides exact analytical gradients.
  - Finite Difference (FD): A classical numerical approximation.
  - Simultaneous Perturbation Stochastic Approximation (SPSA): A stochastic method requiring only two function evaluations per iteration.
- Metric Approximation ( $g(\theta)$ ):
  - Identity (Constant): Standard gradient descent.
  - QN-BDA (Block Diagonal Approximation): Approximates the Fubini-Study metric using block-diagonal structures derived from circuit layers.
  - QN-SPSA: Stochastically approximates the metric using a 2-SPSA approach, requiring only four quantum evaluations per iteration.
Proposed Hybrid Method (QN-SPSA+PSR): A novel combinatorial scheme is introduced where the Fubini-Study metric is approximated via the QN-SPSA framework (using stochastic estimation for the metric), while the gradient of the cost function is calculated exactly using the Parameter-Shift Rule (PSR). This aims to integrate the computational efficiency of QN-SPSA with the precision of PSR.

Key Contributions

Development of QN-SPSA+PSR: The paper proposes a new optimization algorithm that replaces the stochastic gradient estimation in QN-SPSA with the exact PSR. This hybrid approach aims to stabilize the updates driven by the stochastically estimated metric while maintaining low computational overhead.
Comprehensive Comparative Analysis: The study provides a systematic evaluation of a $3 \times 3$ grid of optimizer combinations (varying gradient and metric methods) alongside classical baselines like COBYLA.
Symmetry-Informed Ansatz Design: The paper demonstrates how leveraging physical symmetries (real representation and spin-flip symmetry) of the TIM can guide the selection of efficient ansätze (RealAmplitudes) that reduce the number of trainable parameters and circuit depth.
Resource vs. Accuracy Trade-off Quantification: The authors quantify the computational cost (in terms of quantum circuit evaluations) versus the final relative error for various methods, offering a practical guide for NISQ applications.

Results
Numerical simulations were conducted on a 12-qubit TIM system with a two-layer RealAmplitudes ansatz. Key findings include:

Performance of QN-SPSA+PSR: The proposed method demonstrated faster convergence per parameter update compared to QN-SPSA+SPSA and achieved a final accuracy comparable to the more computationally expensive QN-BDA+PSR. It exhibited greater stability than purely stochastic methods.
Ansatz Efficiency: The RealAmplitudes ansatz with linear entanglement outperformed the EfficientSU2 ansatz in convergence speed and consistency for the TIM problem, validating the approach of tailoring the ansatz to the problem's symmetries.
Classical Baseline (COBYLA): The derivative-free COBYLA optimizer showed remarkable efficiency, requiring only one quantum evaluation per iteration. It achieved relative errors comparable to advanced quantum-informed methods in this specific benchmark, highlighting its viability for near-term devices.
Scalability: While QN-BDA+PSR theoretically offers the best convergence, its cost scales with the number of layers and parameters. QN-SPSA+PSR maintained a lower cost profile (scaling with the number of layers and a constant number of stochastic evaluations for the metric) while delivering competitive results.
System Size and Field Strength: The QN-SPSA+PSR method consistently delivered reliable ground state energy estimates across varying system sizes (6 to 12 qubits) and external field strengths, closely aligning with exact results.

Significance and Claims
The paper claims that the QN-SPSA+PSR method offers a practical pathway for improving the optimization subroutines of VQAs on NISQ devices. By combining the low metric-estimation cost of QN-SPSA with the precision of PSR, the method achieves a balance between stability, convergence speed, and computational resource consumption. The authors suggest that this approach is particularly beneficial for VQAs and Quantum Machine Learning (QML) tasks where computational overhead is a limiting factor.

The study modestly concludes that while the results are promising for the TIM model, the performance of these methods may vary with more complex problem structures and stronger quantum correlations. The work serves as a foundational comparison, suggesting that quantum-informed metric updates can improve stability for certain optimizer combinations, motivating further investigation into larger and more complex systems. The paper does not claim universal superiority over all classical methods but highlights the specific conditions under which quantum-informed optimizers provide a distinct advantage in terms of convergence direction and stability.