We propose a new class of vector transport based gradient descent enabling silver stepsize acceleration on Riemannian manifolds, yielding provable accelerated gradient methods for potential functional optimization in Wasserstein space.