On the determination of path signature from its unitary development

We establish an explicit, constructive approach to determine any element X in the tensor algebra T (Rᵈ) = ₍=₀^ (Rᵈ) ^ n from its moment generating function. The only assumption is that X has a nonzero radius of convergence, which relaxes the condition of having an infinite radius of convergence in the literature. The key building block of our approach is tridiagonal antisymmetric matrices, whose sparsity offers a considerable advantage for dimension reduction in applications. In particular, specialising X to the signature space of bounded p-variation paths in Rᵈ with 1 p <2, we show that the developments of such sparse matrices are sufficient to separate points over the space of signatures, which yields a refined answer to the "moment problem" concerning the signature. Based on the above theoretical investigations, we propose a new distance function for probability measures on the path space, termed as the "Restricted Path Characteristic Function Distance" (RPCFD), and validate its effectiveness via numerical experiments on hypothesis testing for examples of fractional Brownian motions.