Abstract
Optimizing the architecture of variational quantum circuits (VQCs) is crucial for advancing quantum computing (QC) towards practical applications. Current methods range from static ansatz design and evolutionary methods to machine learned VQC optimization, but are either slow, sample inefficient or require infeasible circuit depth to realize advantages. Quality diversity (QD) search methods combine diversity-driven optimization with user-specified features that offer insight into the optimization quality of circuit solution candidates. However, the choice of quality measures and the representational modeling of the circuits to allow for optimization with the current state-of-the-art QD methods like covariance matrix adaptation (CMA), is currently still an open problem. In this work we introduce a directly matrix-based circuit engineering, that can be readily optimized with QD-CMA methods and evaluate heuristic circuit quality properties like expressivity and gate-diversity as quality measures. We empirically show superior circuit optimization of our QD optimization w.r.t. speed and solution score against a set of robust benchmark algorithms from the literature on a selection of NP-hard combinatorial optimization problems.
Introduction
In the current Noisy Intermediate-Scale Quantum (NISQ) era, designing efficient Variational Quantum Circuit (VQC) architectures is a primary bottleneck. VQCs function similarly to neural networks, where parameters are tuned to minimize a cost function. However, they often suffer from "barren plateaus"—regions where the optimization landscape becomes too flat for gradient-based solvers—and extreme sample inefficiency in traditional evolutionary searches. This paper proposes a novel framework using Quality Diversity (QD) optimization to bridge this gap, creating a diverse archive of high-performing circuits that avoid over-parameterization.
Methodology
The core of the proposed approach is a Matrix-Based Encoding scheme. Quantum circuits are mapped 1:1 to a matrix of scalars where the integer component defines the gate type (e.g., CNOT, RX, or Identity) and the decimal component defines parameters or target qubits. This allows the entire circuit structure to be optimized as a continuous numerical problem using Covariance Matrix Adaptation Map-Annealing (CMA-MAE).
To guide the search, the authors employ two heuristic quality measures:
- Sparsity: The count of non-identity gates, encouraging shorter circuits less prone to noise.
- Gate Diversity: The variety of unique gates used, preventing the search from collapsing into easily simulable classical structures.
Experimental Evaluation
The framework was tested on four NP-hard combinatorial optimization problems: MaxCut, Minimum Vertex Cover (MinVER), Maximum Independent Set (MaxIND), and Maximum Clique (MaxCLI). These were evaluated across challenging graph topologies, including Barbell, Ladder, and Caveman graphs, as well as Erdos-Reny random graphs.
Results and Discussion
Performance Benchmarking
The QD methods, particularly CMA-MAE, demonstrated significant efficiency gains over established baselines. The algorithms frequently identified near-optimal solutions within just 20 optimization steps, consistently outperforming the analytic Quantum Approximate Optimization Algorithm (QAOA) and the evolutionary QNEAT in terms of approximation ratios.
Key Insights
- Sparsity as a Shield: By rewarding sparsity, the QD algorithm naturally avoids the barren plateau problem. Simpler circuits provide a more navigable optimization landscape, allowing the solver to find high-quality solutions that dense, complex circuits might obscure.
- The "Small Gate Set" Advantage: Experiments revealed that basic gate sets (e.g., RX, CNOT, and Identity) often yielded better global search results than more complex universal sets like Clifford+T. Simplicity in the building blocks facilitated more effective exploration.
Conclusion
This work introduces a robust, task-agnostic framework for quantum circuit design that leverages the strengths of Quality Diversity algorithms. By treating circuit architecture as a continuous optimization problem, the method achieves superior speed and solution quality on combinatorial benchmarks. Future work will focus on extending these results to Quantum Machine Learning (QML) and validating the designs on physical NISQ hardware to test their resilience against real-world decoherence.
Summary updated