Blogs
#Quantum Technology
Quantum computers are leading the future of information technology. Its computational power has the potential to reach realms far beyond what traditional computers can achieve. By pursuing and continuously challenging these possibilities, we can envisage a future where our lives, society, and industries are profoundly transformed. SoftBank's Research Institute of Advanced Technology is at the forefront of these challenges, paving the way toward such a future.
Moreover, success requires collaboration with partners willing to learn, challenge, and achieve together. Hence, we continue to venture into the unexplored territories of quantum computing in collaboration with the academic community, corporations, and individual researchers.
Quantum computers have the potential to surpass traditional computers in terms of computational power and information capacity by leveraging the fundamental quantum mechanical properties of superposition and quantum entanglement. Specifically, superposition allows a single quantum bit to exist in multiple states simultaneously.
Furthermore, quantum entanglement allows two qubits to influence each other instantaneously, even when physically separated. This enables rapid information transfer between quantum computers in different locations.
By utilizing the states of entanglement and superposition, along with combining quantum algorithms, quantum computers are expected to operate significantly faster (with fewer computational steps) than conventional algorithms for specific computational problems. However, it should be noted that quantum computers do not surpass traditional computers in all aspects
In classical computing, as problems become more complex, the computational time increases exponentially. However, by utilizing quantum computing, there is the potential to find solutions in a much shorter computational time.
*NISQ (Noisy Intermediate-Scale Quantum) is the current quantum computer that possesses tens to hundreds of quantum bits but lacks error correction functionalities.
*FTQC (Fault-Tolerant Quantum Computing), which is more advanced and capable of error correction:
*QAOA (Quantum Approximate Optimization Algorithm) is a quantum algorithm for combinatorial optimization problems.
*VQE (Variational Quantum Eigensolver) is a hybrid algorithm that combines classical and quantum computation. It calculates the ground state energy of molecules and other systems.
*QSCI (Quantum-Selected Configuration Interaction) is a hybrid quantum-classical algorithm that calculates multi-electron Hamiltonians' ground and excited state energies on noisy quantum devices.
*Shor's Algorithm is a quantum algorithm used to factor large numbers into their prime factors efficiently.
*Grover's Algorithm is a quantum algorithm that efficiently solves unstructured search problems within an unsorted database.
*Surface code is an approach to quantum error correction where qubits are arranged in a 2D lattice and combined in specific patterns to achieve error correction.
*Toric code is an approach to quantum error correction where qubits are arranged on a donut-shaped structure (torus) surface, and loop-like patterns are utilized on the torus surface to achieve error correction.
Quantum computers that leverage the unique properties of quantum mechanics are currently treated as "NISQ (Noisy Intermediate-Scale Quantum) devices" within the existing technological framework. However, they face a significant challenge known as errors. Quantum bits are extremely sensitive to errors caused by weak environmental noise or collisions of subatomic particles, making it a critical challenge to perform quantum calculations accurately while mitigating these effects. Developing solutions to these challenges has become an urgent task.
SoftBank's Research Institute of Advanced Technology is intensifying collaborations with domestic and international research institutions such as the "Quantum Innovation Initiative Consortium" operated by the University of Tokyo, and the Quantum Computing Research Center (KQCC) at Keio University. The aim is to deepen and elevate research in quantum computing to further its practical application and implementation in society.
Considering that quantum computers' quantum nature is widely regarded as suitable for quantum chemistry problems, one potential area where quantum superiority can be discovered, we researched to uncover new possibilities for solving chemical problems. We pursued this by more accurately simulating interactions between electrons and atomic nuclei (Nuclear Quantum Effects, NQEs) through an approach beyond the Born-Oppenheimer approximation (Beyond Born-Oppenheimer).
In this research, unlike previous studies that only used state vector simulators, we have employed not only shot-based simulators (simulators that mimic actual devices) but also actual NISQ (Noisy Intermediate-Scale Quantum) devices. For the first time, we have applied the Variational Quantum Eigensolver (VQE) to the NEO (Nuclear-Electronic Orbital) Hamiltonian.
Additionally, we explored efficient methods considering the properties of quantum computers, such as choosing a wavefunction (Ansatz) as a hardware-efficient initial estimate, selecting the starting point for optimization, and methods for initial point optimization. As a result, we found that selecting the appropriate initial values can improve both the quality of the computational results and the convergence speed.
In particular, in the VQE algorithm, by choosing an initial value optimized in advance for the NEO Hamiltonian using a state vector simulator, as opposed to the typically used random initial values that are susceptible to noise, we were able to drastically reduce errors caused by accumulated shot noise and hardware errors, thereby obtaining much more accurate results.
VQE, a Hybrid Algorithm of Quantum and Classical Computing
An Example of Comparing Advanced and Ordinary Initialization for Optimizers in State Vector Simulators:
Improvements in Computational Results Quality and Convergence Speed through Proper Initial Value Selection
Evaluation Results with a State Vector Simulator:
Confirming Improved Accuracy by Increasing the Number of Shots
The "Red Line" Represents the Threshold of Chemical Accuracy.
Through this approach, it has been confirmed that the convergence speed of the computation process improves, and the problem of the Barren Plateau can be avoided. This demonstrates the potential for precise simulation of chemical problems on real quantum devices.
However, in the calculation of actual computations for the research theme, the influence of gate operation noise, measurement noise, and shot noise caused by the large scale of the problem has been observed. This has shed light on new research topics, such as circuit optimization of trial wavefunctions and improvements in classical optimization methods for variational quantum circuits. This research is an essential step in exploring new possibilities in the field, showcasing the potential for chemical calculations using quantum computers in the future.
*State Vector Simulator: A simulator used on classical computers to mathematically represent and simulate the state vector of a quantum system in quantum mechanics.
*Barren Plateau: A state in adjusting parameters in a quantum circuit where the gradient approaches zero, resulting in a flat region where learning and optimization do not progress effectively.
SoftBank has initiated a project in partnership with RIKEN (Institute of Physical and Chemical Research) to commercialize quantum computers. This project focuses on developing a hybrid platform of quantum computers and supercomputers, combining quantum calculations with the remarkable power of classical computation of supercomputers to maximize computational capacity and value.
SoftBank's Research Institute of Advanced Technology is researching early practical applications as part of this project. This includes exploring the utility of new quantum algorithms through the quantum-classical hybrid platform, taking steps against the challenges of Noisy Intermediate-Scale Quantum (NISQ) devices, such as minimizing errors caused by noise and optimizing quantum circuits, and validating the usefulness of the hybrid platform through these applications.
It aims to apply the advantages of collaboration with supercomputers, such as error suppression/mitigation, algorithm optimization, and coordinated computation, to practical applications.
As the next-generation social infrastructure supporting a society coexisting with AI, mobile networks are required to have more advanced communication quality and optimized base station resources. To efficiently execute the immense computations needed for this, the use of quantum computing is being explored.
In particular, quantum optimization algorithms such as the Quantum Approximate Optimization Algorithm (QAOA) and the Variational Quantum Eigensolver (VQE) can efficiently explore solution spaces that cannot be achieved with classical computers. Even for large-scale optimization problems, it is theoretically possible to solve combinatorial optimization problems more efficiently using quantum computers than classical computers. However, there are challenges in applying quantum optimization algorithms to real-world problems within the limitations of the current quantum computers' qubit count and quality. Hence, research is being conducted to reduce variables and optimize quantum circuits.
Furthermore, network control flexibility has become crucial as the need for telecommunications infrastructure as a next-generation social infrastructure becomes more complex. Towards the sophistication of machine learning algorithms for operations such as fault diagnosis and automation of recovery functions in network operations using AI/ML (Machine Learning), we are engaged in research to apply quantum computing.
Quantum kernel learning, one of the quantum machine learning algorithms, is capable of representing features in ultra-high dimensions by leveraging the principles of quantum mechanics. It is believed to have advantages in learning efficiency compared to classical computers. We are aiming to explore the superiority of quantum computers and achieve early practical application in learning and inference performance for failure events on datasets extracted from commercial system logs.
*Quantum Kernel Learning: One of the machine learning methods that utilizes quantum computers. It can efficiently calculate the similarity (kernel function) between high-dimensional data, enabling fast processing of vast datasеts or complex computations, leading to more accurate predictions and learning.
As research and development progress in quantum computing, we aim to harness this remarkable computational power to tackle previously challenging problems and create new value. For instance, fields where quantum computers can showcase their capabilities are wide-ranging, including the advancement of distributed computing networks, the development of new materials, and the development of next-generation AI.
We will continue to pursue these future visions made possible by quantum computing. By doing so, we believe we can contribute to allowing quantum computers to create new value across society while allowing ourselves to continue challenging the possibilities and future of quantum computing.
Quantum computing transcends the framework of conventional computing methods and holds the potential to discover new solutions to problems previously deemed unsolvable. Quantum computers pioneer a new realm beyond conventional wisdom and undiscovered horizons. SoftBank's Research Institution of Advanced Technology will continue to create the future at the forefront of this exploration and challenge.
