Fig. 2: Numerical study of the effectiveness of information sharing.
From: Variational quantum algorithm with information sharing
Comparison between (a–b) independent vs (c–d) information shared VQE optimisation strategies for the quantum spin model, Eq. (2) with field h = hX = hZ. The sample distribution of 100 repetitions of the same optimisation are shown as density plots for each hα, with darker shades corresponding to more observations, and the solid curve showing the mean. Optimisation is run for thirty iterations and we consider fifteen values of hα. Data boxes show the total number of function evaluations per repetition and the number of evaluations seen by each optimiser \({\mathcal{B}}({h}_{\alpha })\), in addition to the numbers quoted each BO receives ten initialisation data points that are also either independent or shared (see “Practicals aspects of Bayesian optimisation” in the “Methods” section). a Independent BO’s at each hα. b As before, but each optimiser requests two additional energy evaluations (at random θ parameter points) at each iteration. c BOIS nearest-neighbour information sharing, i.e. BO’s at neighbouring h-field points (e.g. hα and hα±1) share device measurement results. d BOIS all-to-all information sharing, i.e., every BO sees all device measurement results. Insets of (c) and (d) show the same data on a log-scale.
