Circuittree Enables Approximate Quantum State Preparation with Tree-Based Bayesian Optimization

The challenge of preparing complex quantum states on today’s limited quantum computers demands innovative approaches that balance accuracy with achievable circuit complexity. Nicholas S. DiBrita, Jason Han, and colleagues from Rice University, along with Younghyun Cho from Santa Clara University and Hengrui Luo and Tirthak Patel from Rice University, now present a new method for approximate state preparation that significantly reduces the resources needed to create desired quantum outcomes. Their work introduces CircuitTree, a framework that uses intelligent algorithms to navigate the complex parameter space of quantum circuits, effectively finding solutions that closely match target states. This advance overcomes limitations of existing techniques by employing a structured optimisation strategy and demonstrates the potential to unlock more complex quantum computations with shallower, more practical circuits.

Ry-CX Ansatz for Quantum Synthesis

Researchers have developed CircuitTree, a new method for designing efficient quantum circuits. This approach aims to find circuits that accurately approximate a desired quantum operation while minimizing the number of operations and circuit depth. CircuitTree utilizes a specific type of circuit, built from layers of Ry (rotation around the Y-axis) and CX (controlled-NOT) gates, well-suited for current quantum hardware. The method employs Bayesian Optimization to intelligently search for the best settings for the Ry gates, forming the core of the synthesis process. A key feature is its compatibility with quantum computers that have limited connections between qubits, avoiding complex and error-prone operations to move information around the chip.

This design prioritizes practicality, focusing on the limitations of real-world quantum hardware and striving for efficiency in both circuit depth and the number of two-qubit gates. The layered structure and optimization strategy are intended to allow the method to scale to larger and more complex circuits. Experiments were conducted using both a local computing cluster and IBM Quantum’s ibm_nazca computer, employing Python libraries for analysis and optimization. Performance was evaluated using metrics such as Total Variation Distance, alongside metrics for synthesis time, circuit depth, and the number of two-qubit gates. The team tested CircuitTree on three types of circuits, randomly generated circuits, circuits designed for preparing specific quantum states, and circuits used in quantum algorithms, demonstrating its versatility. Results indicate that CircuitTree can generate circuits with lower depth and fewer gates compared to existing methods, offering a potentially valuable tool for quantum computing researchers and developers.

Circuit Parameter Optimization with Gradient-Boosted Trees

Researchers addressed the challenge of preparing quantum states on near-term quantum computers, framing it as an optimization problem. They developed CircuitTree, a framework that uses tree-based models, specifically gradient-boosted regression trees, to navigate the complex landscape of quantum circuit parameters. This approach differs from traditional Bayesian Optimization methods that rely on Gaussian Processes, which can struggle with the scale and smoothness requirements of this problem. To improve efficiency, the team decomposed the parameter space, aligning it with the layered architecture of variational quantum circuits.

This layerwise decomposition allows for a block coordinate optimization strategy, where parameters within each layer are optimized independently, ensuring global convergence and enhancing optimization stability and efficiency compared to random parameter partitioning. The objective function, measuring the discrepancy between the circuit output and the target transformation, is defined using the Total Variation Distance, calculated from probability vectors obtained through quantum measurements. Due to the inherent stochasticity of quantum measurements and potential device imperfections, the team employed surrogate models to approximate the true cost function, training these models on observed evaluations to create datasets of parameter-cost pairs. Researchers formally analyzed the convergence of this surrogate-guided optimization process, providing theoretical guarantees under mild assumptions regarding noise and model fidelity. Empirical validation on widely-used quantum benchmarks demonstrated that CircuitTree achieves low Total Variation Distance and high fidelity with significantly shallower circuits than existing methods, highlighting the effectiveness of the proposed framework and its potential for advancing near-term quantum computation.

Gradient Boosted Trees Enhance Quantum State Preparation

Researchers have developed CircuitTree, a new framework for preparing quantum states on near-term quantum computers, addressing challenges posed by limited circuit depth and complex parameter spaces. They achieved significant improvements in both the fidelity and efficiency of preparing target quantum states using a surrogate-guided optimization approach based on tree-based models. Experiments demonstrate that CircuitTree consistently outperforms existing methods, achieving lower Total Variation Distance between prepared and target states while requiring fewer quantum hardware measurements. The team systematically compared different surrogate models, revealing that Gradient Boosted Regression Trees consistently delivered the fastest convergence and lowest Total Variation Distance.

Specifically, Gradient Boosted Regression Trees outperformed both Quantile Regression Forests and Gaussian Processes in optimizing Random Quantum Circuits. Further analysis showed that CircuitTree’s structured layerwise decomposition strategy significantly enhances optimization stability and fidelity. Measurements confirm that CircuitTree achieves substantial reductions in the number of shots required for optimization, a critical metric for resource-constrained near-term devices. The team evaluated performance across three representative target state families, Random Quantum Circuits, Quantum State Preparation, and Variational Quantum Eigensolver, demonstrating consistent improvements in all cases.

In one set of experiments, CircuitTree achieved a Total Variation Distance below 0. 1 after 4000 iterations, while baseline methods struggled to reach comparable levels of accuracy. These results demonstrate the potential of CircuitTree to enable more efficient and reliable quantum computation on near-term hardware. Experiments were conducted on IBM’s ibm_nazca device, a 127-qubit quantum computer, utilizing AMD EPYC 7702P processors with 64 cores and 2. 0GHz clock speed.

CircuitTree Optimizes Quantum State Preparation Efficiently

CircuitTree, a novel framework for approximate quantum state preparation, achieves significant advances in utilizing near-term quantum computers. Researchers developed a surrogate-guided optimization approach based on Bayesian Optimization and employing tree-based models, overcoming limitations of existing methods that struggle with high-dimensional, non-smooth parameter spaces. This new method systematically decomposes the optimization problem, enabling efficient and scalable solutions without requiring gradient information or complete access to the quantum system’s operations. The team demonstrated both the efficacy and theoretical convergence of CircuitTree through rigorous testing on simulated and real quantum hardware.

Results show that the framework achieves low error and high fidelity in state preparation while requiring shallower circuits than previously established techniques. This improvement is achieved by exploiting the inherent architectural structure within quantum circuits to enhance the optimization process. Future work will focus on automating the hyperparameter tuning process and exploring methods for more effectively leveraging prior knowledge to improve the efficiency of the optimization. This research contributes to the growing field of structured black-box optimization and its application to quantum algorithm design, demonstrating the potential of non-Gaussian surrogates to deliver both scalability and provable convergence in near-term quantum computing.