Experimental studies of reverse annealing of p-spin problem on D-Wave quantum annealer
ORAL
Abstract
Iterated reverse annealing (IRA) is a formalism of reverse annealing implemented in current quantum annealers. The system is prepared in a trial solution state, annealed reversely to an inversion point, and then annealed forwardly. It may also be iterated with the last output state as new input.
We perform IRA experiments on the D-Wave 2000Q device, with a focus on the p-spin problem where p=2. We examine the dependence of IRA performance on parameters such as problem size, anneal time, inversion points; and also on D-Wave’s functionalities like state reinitialization, number of iterations, pausing. The performance is evaluated in total and partial success probabilities. We can find correct solutions with near certainty. Comparison with standard forward annealing is also made.
We also perform open-system simulations with dephasing in the instantaneous energy eigen-basis. The experimental total success probabilities are well matched. A difference between independent and collective SB coupling model is reported due to spin-symmetry breaking of the classical input. Simulations also show IRA fails in a closed system, suggesting the importance of thermal relaxation in the convergence to the correct solution.
We perform IRA experiments on the D-Wave 2000Q device, with a focus on the p-spin problem where p=2. We examine the dependence of IRA performance on parameters such as problem size, anneal time, inversion points; and also on D-Wave’s functionalities like state reinitialization, number of iterations, pausing. The performance is evaluated in total and partial success probabilities. We can find correct solutions with near certainty. Comparison with standard forward annealing is also made.
We also perform open-system simulations with dephasing in the instantaneous energy eigen-basis. The experimental total success probabilities are well matched. A difference between independent and collective SB coupling model is reported due to spin-symmetry breaking of the classical input. Simulations also show IRA fails in a closed system, suggesting the importance of thermal relaxation in the convergence to the correct solution.
–
Presenters
-
Ka Wa Yip
Univ of Southern California, University of Southern California
Authors
-
Yuki Bando
Tokyo Institute of Technology
-
Ka Wa Yip
Univ of Southern California, University of Southern California
-
Hidetoshi Nishimori
Tokyo Institute of Technology
-
Daniel Lidar
Univ of Southern California, University of Southern California