Self-learning intellectual models for predicting the development of Rayleigh-Taylor turbulent mixing

Abstract

One of the key problems in laser fusion is the stably compression of targets. The efficiency with which a target is burnt out is strongly affected by turbulent flows generated during the compression of a light fusion fuel by a heavy shell under the action of inertial forces. In its simplified formulation (i.e., in plane geometry and under the assumption of uniform compression at a constant acceleration), the compression problem reduces to the classical problem of the Rayleigh-Taylor instability. A description of such turbulent flows is a fundamental problem of hydrodynamics. At present, there are two approaches to studying turbulence: through straightforward numerical calculations based on the hydrodynamic equations and by devising semi-empirical models of turbulent mixing. Direct numerical simulation of turbulent flows is a rather laborious computational task that requires exact specification of the initial conditions. Semi-empirical models, in turn, require the determination of a large number of constants. In this paper, the problem in question is proposed to be solved by a fundamentally new approach: the processing of large amounts of experimental (numerical) data using the neuronet analysis method.