TL;DR
This paper introduces novel, hardware-efficient quantum circuit designs (ansatze) for solving portfolio optimization problems on near-term quantum computers. By combining these ansatze with a hybrid quantum-classical computing scheme and error mitigation techniques, they demonstrate practical applications on superconducting quantum hardware, paving the way for quantum advantage in finance.
What Problem Does This Paper Address?
The advent of Noisy Intermediate-Scale Quantum (NISQ) computers presents a unique opportunity to tackle complex problems that are intractable for classical computers. However, realizing this potential is challenging due to the limitations of NISQ devices, such as limited qubit counts, short coherence times, and noise. A key hurdle is designing efficient quantum algorithms, particularly variational quantum algorithms (VQAs) like the Variational Quantum Eigensolver (VQE), that can run on these constrained machines. VQE relies on a "trial state" preparation circuit, known as an ansatz, which needs to be both expressive enough to capture the problem's solution space and shallow enough to avoid excessive noise. For combinatorial optimization problems like portfolio optimization, existing ansatze often fall into two categories: hardware-efficient (HE) ansatze that are shallow but may suffer from issues like barren plateaus and poor expressibility, or problem-inspired ansatze that are more expressive but tend to be too deep for NISQ devices. This paper aims to bridge this gap by developing ansatze that are both hardware-efficient and problem-inspired, specifically for portfolio optimization.
Methods
The authors propose two novel ansatze, the CCC (Controlled-Controlled-R_y) and CC (Controlled-R_y) ansatze, inspired by the structure of Dicke states and W states. These ansatze feature a "staircase structure" and have a linear circuit depth with a constant factor of at most 2 on CNOT gates, making them hardware-efficient. Crucially, they are designed to span a specific subspace relevant to portfolio optimization, thus avoiding the barren plateau problem often encountered with HE ansatze.
To address the limitations of NISQ hardware, they introduce a Hybrid Distributed Computing (HDC) scheme. This scheme partitions the quantum circuit into smaller, manageable subcircuits (fragments) using techniques like circuit cutting. This allows for the execution of larger problems by distributing the computation across multiple smaller quantum circuits or even classical resources.
Furthermore, they incorporate several techniques to enhance performance and accuracy:
- Simultaneous Sampling: A method to improve the efficiency of sampling from quantum circuits.
- Problem-Specific Measurement Error Mitigation: Techniques tailored to reduce the impact of noise on measurement outcomes, leveraging the symmetry properties of the prepared states.
- Fragment Reuse Technique: A method to reduce the number of circuit submissions to the hardware by reusing computed results.
The cost function used for optimization is the Conditional Value-at-Risk (CVaR), which, with a small confidence level, acts as a local cost function, further mitigating issues related to barren plateaus.
Datasets
The paper uses financial data generated using qiskit_finance.data_providers for simulating and experimenting with portfolio optimization. No new benchmark datasets are introduced; rather, the focus is on applying the proposed methods to a standard financial optimization problem.
Key Insights and Ideas
- Balancing Expressibility and Trainability: The core insight is that for NISQ-era combinatorial optimization, a restricted expressibility that targets the problem's relevant subspace is more beneficial than a highly expressive but noisy ansatz. This is because it helps avoid barren plateaus and local minima.
- Dicke State Structure for Finance: The paper demonstrates that the structure of Dicke states, which are known for their entanglement properties, can be effectively adapted to represent financial optimization problems.
- HDC for Scalability: The HDC scheme, by breaking down large circuits into smaller fragments, is crucial for scaling quantum algorithms to larger problem sizes on current NISQ hardware.
- Symmetry for Error Mitigation: Exploiting the inherent symmetries of the prepared Dicke states significantly simplifies and improves the effectiveness of measurement error mitigation techniques.
- CVaR as a NISQ-Friendly Cost Function: Using CVaR with a small confidence level acts as a local cost function, making the optimization landscape smoother and more amenable to NISQ devices.
Key Results
- The proposed CCC and CC ansatze achieve linear circuit depth and a reduced number of parameters compared to existing methods for preparing Dicke states.
- Numerical simulations show that the proposed ansatze consistently outperform hardware-efficient ansatze in terms of time consumption and the probability of obtaining optimal and feasible solutions, especially as the number of assets increases.
- Experiments were successfully conducted on a superconducting quantum computer with up to 55 qubits, demonstrating the practical feasibility of the HDC scheme.
- The HDC scheme, combined with error mitigation and fragment reuse, significantly reduces the sampling complexity and improves the accuracy of results on NISQ hardware.
- The paper shows that the proposed methods can achieve high probabilities of obtaining optimal solutions even for complex scenarios, outperforming HE ansatze.
Limitations and Future Work
- Noise Resilience: While error mitigation techniques are employed, the inherent noise of NISQ computers remains a limiting factor. The authors suggest that more advanced error mitigation methods, such as zero-noise extrapolation and probabilistic error cancellation, could yield even better results.
- Ergodic Search: For achieving true quantum advantage, ergodically searching the partitioned subspaces might not always be efficient. The authors suggest that selecting appropriate columns or partial elements of these columns based on asset correlations could be a more efficient approach, though this might be cumbersome.
- Hardware Limitations: The current experiments are limited by the available qubit count and connectivity of NISQ devices. Future work could explore larger-scale problems as quantum hardware improves.
- Generalization: While the paper focuses on portfolio optimization, the proposed ansatze and HDC scheme have the potential to be applied to other combinatorial optimization problems and fields like chemistry and machine learning.
Summary updated