1: \begin{abstract}

2: High-dimensional Bayesian optimization (BO) is challenging, as globally finding the optimum requires exponentially many samples.

3: To sidestep the difficulty, several local BO algorithms have been proposed recently.

4: Despite their strong empirical performance, little is known for their convergence property.

5: In this work, we present the first analysis of a recently proposed local BO algorithm \parencite{muller2021local}.

6: We prove that the algorithm converges to a stationary point with polynomial number of samples, in both noisy and noise-free settings.

7: Numerical simulations are provided to verify the results.

8: \end{abstract}

9: