Hermite processes are a family of random processes that generalize fractional Brownian motion, a well-known model for phenomena with long-range dependence and self-similarity. Unlike fractional Brownian motion, higher-order Hermite processes are not Gaussian, which makes them mathematically harder to analyze. A basic but important question about any random process is whether its finite-dimensional distributions have smooth probability density functions, meaning probabilities can be described by a well-behaved function rather than something more irregular. This paper answers that question affirmatively for Hermite processes of any finite order and for a broad class of related random variables built by integrating deterministic functions against these processes.
The main technical framework used is Malliavin calculus, a branch of probability theory that provides tools for differentiating random variables in a functional sense and is particularly powerful for proving the existence and smoothness of densities. The key concept is the Malliavin derivative, and the central challenge is showing that this derivative is nondegenerate, meaning it does not collapse to zero in a way that would obstruct the density from being smooth. The authors establish that not only do smooth densities exist, but they belong to the Schwartz space, the class of infinitely differentiable functions that decay faster than any polynomial. They also show that linear independence of the weight functions used in the integrals is exactly the right condition for absolute continuity, giving a clean and complete characterization.
The proof strategy is novel because it works directly with the non-Gaussian structure of the problem rather than approximating by Gaussian processes or borrowing smoothness from a Gaussian limit. The authors identify an analytic property of certain directional transforms related to the singular kernels that define Hermite processes, which ensures nondegeneracy holds at every level of the so-called Wiener chaos expansion. They then combine uniform estimates on Malliavin derivatives with an inductive argument using a classical inequality from linear algebra to propagate smoothness bounds across arbitrarily many derivatives. The result is a self-contained and general approach applicable to a wide range of non-Gaussian processes sharing this kernel structure.