The paper studies a type of random process called weighted sub-fractional Brownian motion, which is a generalization of standard Brownian motion (the classic mathematical model of random, jittery movement). Unlike ordinary Brownian motion, this process has memory: what happens now depends in a complicated way on the entire past, not just the present moment. This "non-Markovian" property makes the process mathematically difficult to work with, because standard tools for studying random systems rely on the convenient assumption that the present fully captures all relevant information about the past. The authors address this by constructing a "Markovian lift," a way of expanding the description of the system into a richer, infinite-dimensional state space where the memory is explicitly encoded, so that the process once again satisfies the simpler Markovian property.
Having established this theoretical framework, the authors analyze differential equations driven by this noise. They prove that, under reasonable conditions on the equations' coefficients (such as Lipschitz continuity, meaning the functions do not change too wildly), the solutions form well-behaved random processes with smooth probabilistic transitions. The technical challenge is that the noise can be rough or have varying intensity over time, so the authors carefully distinguish between cases requiring different levels of regularity. The lift lives in a Hilbert space, a kind of infinite-dimensional geometric setting, and the authors show the state captures enough information to restart the system from any point.
Because infinite-dimensional objects are impractical to simulate, the paper also develops finite-dimensional approximations. The idea is to replace the infinite memory structure with a finite collection of simpler processes called Ornstein-Uhlenbeck factors, which are individually memoryless and driven by the same underlying noise. The authors prove these approximations converge to the true solution uniformly over time, and they work out explicit convergence rates depending on how many approximating factors are used. These rates are expressed in terms of a parameter controlling the number of discretization levels, giving practitioners concrete guidance on how many factors are needed to achieve a desired level of accuracy.