Modeling late-time sensitivity to initial conditions in Boussinesq Rayleigh-Taylor turbulence Journal: Physica D: Nonlinear Phenomena

Sébastien Thévenin, Benoît-Joseph Gréa · 2025

Read Modeling late-time sensitivity to initial conditions in Boussinesq Rayleigh-Taylor turbulence Journal: Physica D: Nonlinear Phenomena by Sébastien Thévenin, Benoît-Joseph Gréa (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2503-16244.txt.

Introduction

Sébastien Thévenin and Benoît-Joseph Gréa present a 2025 physics paper on Rayleigh-Taylor turbulence, where a heavy fluid pushes into a light one. How much the late flow still owes to its start was the open question. The start is summed up in four scale-free numbers that describe random bumps on the first interface.

The team joined direct simulations, a neural network trained on physics rules, and theory. The network let them push to very late times and new starts past the reach of simulations. Growth rate at late times looks universal, they find. But the virtual time origin, a shift in the growth clock, depends strongly on the start numbers. A mixing-layer model explains most of that link and predicts the origin well.

Catalog record

CREATORSébastien Thévenin, Benoît-Joseph Gréa
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE59 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2503-16244

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Modeling late-time sensitivity to initial conditions in Boussinesq Rayleigh-Taylor turbulence Journal: Physica D: Nonlinear Phenomena Sébastien Thévenin, Benoît-Joseph Gréa Abstract This article sheds light on the late-time influence of initial conditions in Boussinesq Rayleigh-Taylor turbulence using an approach combining direct numerical simulations, machine learning and theory. The initial conditions are characterized by four non-dimensional numbers describing the statistical properties of random-phase multi-mode perturbations of an initial diffuse interface. Based on high-fidelity data, a surrogate physics-informed neural network is used to extrapolate the dynamics to very late times and unseen initial conditions, beyond the reach of simulations. This enables uncertainty and global sensitivity analyses to be carried out, revealing the influence of initial conditions in the late-time regime. While the results support the idea of a universal self-similar growth rate, the virtual time origin is found to be strongly sensitive to the initial Reynolds, perturbation steepness and bandwidth numbers. An analytical model based on the phenomenology of Rayleigh-Taylor mixing layers explains most of this dependency, and provide accurate predictions for the virtual time origin. It turns out that when the initial perturbation reaches nonlinear saturation earlier, the mixing layer also re-accelerates earlier, while the virtual time origin is larger. † † graphicalabstract: † † highlights: Late-time sensitivity to initial conditions is imprinted in the virtual origin The initial Reynolds, perturbation steepness and bandwidth numbers are key factors An analytical model is built to predict the virtual time origin Early nonlinear saturation leads to early re-acceleration and larger virtual origin A work combining simulations, physics-informed machine learning, and theory 1 Introduction The Rayleigh-Taylor (RT) instability is encountered in various fields [ 66 , 67 ] , including geophysics [ 15 , 28 , 42 ] , astrophysics [ 4 , 49 , 34 ] and engineering applications such as inertial confinement fusion [ 31 , 5 , 45 , 68 ] . It appears at the interface separating two fluids of different densities under the effect of an acceleration directed toward the lighter fluid, in the opposite direction of the density gradient [ 43 , 59 , 55 , 7 ] . This is the case, for instance, when the heavy fluid is above the light fluid in a gravity field. The destabilizing acceleration amplifies any perturbation of the interface, leading to the development of a mixing layer. The shear created at the interface later triggers secondary instabilities, marking the start of the transition to turbulence. Since the pionneering experiments and simulations of Read [ 44 ] and Youngs [ 65 ] , it has been well established that the height h of the mixing layer upper half evolves toward a quadratic solution at late times, h(t)=\alpha_{\infty}\mathcal{A}g(t+t_{\infty})^{2}, (1) which is proportional to the Atwood number \mathcal{A}=(\rho_{h}-\rho_{l})/(\rho_{h}+\rho_{l}) , characterizing the density contrast between the heavy and light fluids of densities \rho_{h} and \rho_{l} , and the acceleration g , with t the time that is equal to zero at the moment of the disturbance. This evolution is supported by theoretical arguments involving self-similarity [ 46 ] or mass flux and energy budgets [ 14 ] . It is also solution to the buoyancy-drag equation, a simple model commonly used to describe RT flows throughout all their stages, which can be derived from first principles [ 41 , 16 , 21 , 53 ] . An important fundamental question in the study of RT flows is whether the self-similar parameters \alpha_{\infty} and t_{\infty} depend on initial conditions. If so, this has far-reaching implications for the development of turbulence models. On the one hand, their calibration relies in part on the reproduction of the late-time solution, Eq. ( 1 ), which makes the model coefficients dependent on the self-similar parameters. On the other hand, the models themselves should integrate this dependency to be able to describe the transition and bridge the gap between the early phases of RT instability and fully developed turbulence. Many studies are already moving in this direction, by modifying the model’s structure and closure [ 54 , 64 ] , its initialization [ 48 ] or the choice of variables and solutions to be reproduced [ 62 ] . The self-similar growth rate \alpha_{\infty} has long been a central concern, as large differences were reported in the literature [ 18 ] . Values measured in experiments, up to \sim 0.070 , were indeed much higher than those obtained in numerical simulations, around \sim 0.020-0.025 . Today, there seem to be a number of reasons for these differences. First, it is very difficult to perform RT experiments with controlled initial conditions and without adding extra effects such as tilt, shear or variable accelerations [ 2 ] , complicating a proper comparison with simulations. Recent experiments that achieve this compare very well with values found in simulations [ 47 ] . Second, the bounded nature of experiments and numerical simulations can induce large-scale confinement of the flow structures, which alters the dynamics. This may occur earlier if scales close to the lateral size of the system are present in the initial disturbance spectrum, which may be associated to the mode competition case of Dimonte [ 17 , 40 ] . As pointed out by Cabot and Cook [ 11 ] , another reason stems from the evaluation of \alpha_{\infty} , which was often calculated as h(t)/(\mathcal{A}gt^{2}) . This estimate, which neglects the linear term of the self-similar solution, converges very slowly to its asymptotic value, resulting in overestimates if evaluations are made too early. Finally, the definition of the height h can also lead to different values for \alpha_{\infty} . Many works now support the idea of a universal value for the asymptotic growth rate in the case of unbounded RT, from large direct numerical simulations [ 11 , 35 , 9 , 62 ] and well-controlled experiments [ 47 ] to theoretical arguments [ 37 , 58 ] . The virtual origin, expressed as the time t_{\infty} in Eq. ( 1 ), has a somewhat more obscure meaning. While it is a parameter characterizing the late asymptotic regime, it seems instinctively linked to what happens during the early stages of the instability, and hence to the initial conditions. However, unlike the growth rate, it is still poorly understood and has not received much attention, despite having a significant impact on modeling and predicting the RT dynamics. Perhaps this is because it is very difficult to estimate, both in experiments and in simulations. Indeed, it seems that the only way to compute it is to use Eq. ( 1 ) and its time derivative \dot{h} , which gives an expression that converges very slowly to its asymptotic value. As a result, experiments and simulations must run for a sufficiently long time, without confining the large scales. In practice, this is very difficult to achieve, as it would require initializing the perturbation spectrum with very small wavelengths, which implies very sophisticated experiments or very high-resolution simulations. This challenge is addressed in this work with the help of a machine learning surrogate model capable of realistic time extrapolation, enabling the calculation of converged estimates of the self-similar parameters. This allows to investigate the late-time sensitivity of RT to the initial conditions, focusing especially on the

Read the complete text