Adaptive quantum state tomography improves accuracy quadratically.

We introduce a simple protocol for adaptive quantum state tomography, which reduces the worst-case infidelity [1-F(ρ,ρ)] between the estimate and the true state from O(1/sqrt[N]) to O(1/N). It uses a single adaptation step and just one extra measurement setting. In a linear optical qubit experiment, we demonstrate a full order of magnitude reduction in infidelity (from 0.1% to 0.01%) for a modest number of samples (N ≈ 3 × 10(4)).