RiuNet. Repositorio Institucional de la Universitat Politécnica de Valéncia
Publicación
Artículo científico (JournalArticle)
A Comprehensive Framework for the One-Point Probability Density Function of Solutions to Non-Autonomous Separable Random Differential Equations
[EN] We develop a probabilistic framework to determine and estimate the one-point probability density function (1-PDF) of the solution of a broad class of non-autonomous separable random differential equations with random initial conditions. The approach is based on the random variable transformation (RVT) technique and reduces to the study of the distribution of a stochastic integral involving the random input or driving process. By assuming separability of the ODE and several analytical assumptions (invertibility of the state transformation, independence between sources of randomness, absolute continuity of the input process, and integrability of the density formula), the framework is built on two main theoretical results. First, a general density representation theorem provides sufficient conditions for the solution to be absolutely continuous, together with an explicit integral formula for its 1-PDF. Second, a convergence theorem proves that, when the driving process is replaced by an approximating one, the resulting density converges in L1(R) to the exact density, with an explicit error bound controlled by the L1 error of the integrated input. We then specialize the framework to four representative settings for the driving process: a Gaussian process, for which a closed-form 1-PDF is obtained, and three parametric families-affine-in-parameters representations, truncated Karhunen-Lo & egrave;ve, and generalized polynomial chaos expansions-for which tractable semi-explicit formulas are derived. In each case the 1-PDF is shown to be consistent with the prescribed initial condition, and all expressions admit numerical evaluation by standard quadrature and sampling. Finally, the framework is illustrated by four representative stochastic growth models, one per specialized setting, where all underlying hypotheses are verified explicitly and the computed analytical and semi-analytical densities are validated against Monte Carlo simulations., C. Burgos, J.C. Cortés and C.L. Pérez-Diukina have been supported by the grant PAID-11-25 funded by Universitat Politècnica
de València and by grant PID2025-172735NB-I00 funded by MCIN/AEI. C. Burgos has been supported by the grant CIGE/2023/79
(Generalitat Valenciana). F. Rodríguez has been supported by the grant PID2021-125517OB-I00 funded by MCIN/AEI.