Abstract:Variational quantum algorithms have been widely demonstrated in both experimental and theoretical contexts to have extensive applications in quantum simulation, optimization, and machine learning. However, the exponential growth in the dimension of the Hilbert space results in the phenomenon of vanishing parameter gradients in the circuit as the number of qubits and circuit depth increase, known as the barren plateau phenomena. In recent years, research in non-equilibrium statistical physics has led to the discovery of the realization of many-body localization. As a type of floquet system, many-body localized floquet system has phase avoiding thermalization with an extensive parameter space coverage and have been experimentally demonstrated can produce time crystals. We applied this circuit to the variational quantum algorithms for the calculation of many-body ground states and studied the variance of gradient for parameter updates under this circuit. We found that this circuit structure can effectively avoid barren plateaus. We also analyzed the entropy growth, information scrambling, and optimizer dynamics of this circuit. Leveraging this characteristic, we designed a new type of variational ansatz, called the 'many-body localization ansatz'. We applied it to solve quantum many-body ground states and examined its circuit properties. Our numerical results show that our ansatz significantly improved the variational quantum algorithm.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is the "Barren Plateau (BP)" phenomenon in Variational Quantum Algorithms (VQA). Specifically, as the number of qubits and the circuit depth increase, the variance of the parameter gradient will decrease exponentially, which causes the gradient in the optimization process to almost disappear and makes training difficult. This phenomenon is very common in traditional variational quantum algorithms, especially more obvious in circuits using randomly initialized parameters.
To solve this problem, the author introduced the concept of Many - Body Localization (MBL) and applied it to variational quantum algorithms. By constructing a circuit structure with MBL characteristics, the author found that this circuit can effectively avoid the BP phenomenon, thereby improving the performance of variational quantum algorithms. Specifically, the main contributions of the paper include:
1. **Avoiding the BP phenomenon**: The author proved through numerical simulation that when the circuit parameters are in the MBL region, even in very deep circuits, the variance of the parameter gradient will not decrease exponentially with the increase of the number of qubits. This indicates that the MBL circuit can effectively avoid the BP phenomenon.
2. **Information entanglement and entropy growth**: The author studied the information entanglement and entropy growth under different circuit structures. The results show that in the MBL region, information entanglement follows the area law instead of the volume law, which further explains why the MBL circuit can avoid the BP phenomenon.
3. **Optimizer dynamic analysis**: The author also analyzed the changes of the cost function and entropy during the optimization process. The results show that the circuit initialized with MBL shows better stability and faster convergence speed in the early stage of optimization, which helps to improve the overall performance of variational quantum algorithms.
4. **Combination of theory and experiment**: The author not only theoretically analyzed the advantages of the MBL circuit, but also verified these theoretical predictions through numerical simulation. In addition, the author also discussed future research directions, including applying quantum optimal control methods to MBL circuits, understanding the impact of many - body dynamics on circuit performance, and using Random Matrix Theory (RMT) to further optimize the stochastic circuit variational algorithm.
In summary, this paper proposes an improved method for Variational Quantum Algorithms (VQA) by introducing the MBL concept, effectively solves the BP phenomenon, and provides new ideas and tools for research in the field of quantum computing.