Quantum Computing Center

Development and real-device evaluation of new Monte Carlo algorithms: First, we will improve the amplitude estimation algorithm developed in the previous fiscal year and devise a method to calculate amplitude estimates with the highest possible accuracy in the presence of noise. In particular, we aim to develop an algorithm that is resistant to noise specific to real devices. We will also develop a method to efficiently implement the target Monte Carlo integration in an algorithm. At the same time, we will apply this to financial engineering and clarify the requirements for solving meaningful problems with a quantum computer in the future.

Development and real-device evaluation of variational quantum algorithms for chemical reaction calculations: We will continue to improve the VQE algorithm for real-device use. We will implement the improved algorithm on an updated IBM Q real device and adapt it to other chemical reaction systems. In doing so, we will also consider the calculation of molecular excited states and quantum dynamics calculations. In addition, we will develop the theory of natural gradient VQE, explore the mechanism for avoiding plateaus, extend it to be applicable to large-scale systems, and confirm its effectiveness through simulation.

Development and real-device evaluation of quantum machine learners: We will conduct a detailed study of the kernel function, which is the core of the quantum pattern classifier. In particular, we will characterize kernel functions where quantum mechanical effects are prominent. We will also continue to study methods for designing kernel functions. Based on this foundation, we will apply it to multi-class classification problems and develop a kernel density function method. We will also continue to analyze the quantum reservoir computing method with the aim of improving its performance. In particular, from the perspective that this is a method that actively utilizes noise, we will consider more effective ways to apply real devices.

Development of a quantum computer interface: We will continue to develop a qubit mapping compiler with the use of real quantum computers in mind. We will also develop an efficient method for implementing controlled-NOT gates that act between spatially separated qubits, and a methodology for efficiently implementing typical operations such as the Toffoli gate.

Development and real-device evaluation of new Monte Carlo algorithms: Monte Carlo calculation is a process required for many industrial problems, such as in materials science and financial engineering. We aim to accelerate this process with quantum computation. In FY2019, as a new quantum amplitude estimation algorithm (including Monte Carlo algorithms), we developed a method that executes Grover's algorithm with different numbers of operations in parallel and performs statistical processing on the obtained execution results to make an estimation. Unlike conventional methods, this method does not make heavy use of quantum operations, making it suitable for NISQ. We also analyzed this algorithm in a noisy environment and evaluated its performance on a real device using IBM Q. Statistical analysis confirmed that the experimental results were consistent with the theoretical model that takes noise into account.

Development and real-device evaluation of variational quantum algorithms for chemical reaction calculations: We aim to build a quantum computing environment for useful chemical reactions by improving the Variational Quantum Eigensolver (VQE), a quantum algorithm for finding the ground state of a Hamiltonian. In FY2019, we used a quantum computer to analyze the reaction of lithium-air batteries. In particular, while confirming the reproducibility of reaction energies by VQE, we found that the following are important for maximizing the performance of NISQ: selection of an active space that balances computational cost and accuracy, error mitigation based on accurate calibration of the real quantum computer, analysis of the energy profile and avoidance of local stable structures, and evaluation and selection of an appropriate combination of entanglers and optimization methods. On the other hand, we devised a new optimization algorithm based on the "natural gradient method," which can solve a problem inherent in the VQE optimization update rule (the plateau problem).

Development and real-device evaluation of quantum machine learners: We aim to accelerate machine learning algorithms such as pattern classifiers and time-series analyzers using quantum computers. In FY2019, with the goal of obtaining design guidelines for functions that map data to quantum states, we developed a method to visualize data embedded in quantum states for a 2-qubit quantum pattern classifier. This enabled the pre-evaluation of classifier performance and the design of high-precision classifiers by combining complementary classifiers (kernels). We confirmed the effectiveness of these methods using the IBM Q simulator. On the other hand, we began the development of "quantum reservoirs" for performing time-series data analysis on a quantum computer. In FY2019, we first developed a reservoir with universal approximation capability for time-series regression and prediction problems, and confirmed regression performance consistent with theory using a 10-qubit IBM Q real device.

Development of a quantum computer interface: We aim to develop software (a qubit mapping compiler) that efficiently connects quantum algorithms and real quantum computers. In FY2019, we developed a qubit mapping compiler for the 20-qubit real device IBM Q. Specifically, we developed a compiler that automatically extracts a desirable qubit layout for executing short quantum operations on IBM Q. We also developed methods for efficiently implementing the Grover operator and the Toffoli gate.