arXiv subjects

Guillaume Balarac

Publications and source records attributed to Guillaume Balarac.

14 recordsLinked to original sources

arXiv2026

Nonlocal energy transfer mechanism in three-dimensional quantum turbulence

We investigate the kinetic energy cascade in zero-temperature quantum turbulence. Using simple theoretical arguments and unprecedented numerical simulations, we unveil an universal mechanism transferring energy directly from large to very small scales, thus bypassing the Kolmogorov-like local energy cascade and resulting in nonclassical energy spectra. This mechanism rests both on the vast separation of scales typical of superfluid helium-4 flows and on the alignment between quantum vortices and large-scale velocity gradients, in direct analogy with vortex stretching in classical flows.

physics.flu-dyn↗

arXiv2023

Gradient-free online learning of subgrid-scale dynamics with neural emulators

In this paper, we propose a generic algorithm to train machine learning-based subgrid parametrizations online, i.e., with $\textit{a posteriori}$ loss functions, but for non-differentiable numerical solvers. The proposed approach leverages a neural emulator to approximate the reduced state-space solver, which is then used to allow gradient propagation through temporal integration steps. We apply this methodology on a chaotic two-timescales Lorenz-96 system and a single layer quasi-geostrophic system with zonal dynamics, known to be highly unstable with offline strategies. Using our algorithm, we are able to train a parametrization that recovers most of the benefits of online strategies without having to compute the gradient of the original solver. We found that training the neural emulator and parametrization components separately with different loss quantities is necessary in order to minimize the propagation of approximation biases. Experiments on emulator architectures with different complexities also indicates that emulator performance is key in order to learn an accurate parametrization. This work is a step towards learning parametrization with online strategies for climate models.

physics.comp-ph↗

arXiv2022

A posteriori learning for quasi-geostrophic turbulence parametrization

The use of machine learning to build subgrid parametrizations for climate models is receiving growing attention. State-of-the-art strategies address the problem as a supervised learning task and optimize algorithms that predict subgrid fluxes based on information from coarse resolution models. In practice, training data are generated from higher resolution numerical simulations transformed in order to mimic coarse resolution simulations. By essence, these strategies optimize subgrid parametrizations to meet so-called $\textit{a priori}$ criteria. But the actual purpose of a subgrid parametrization is to obtain good performance in terms of $\textit{a posteriori}$ metrics which imply computing entire model trajectories. In this paper, we focus on the representation of energy backscatter in two dimensional quasi-geostrophic turbulence and compare parametrizations obtained with different learning strategies at fixed computational complexity. We show that strategies based on $\textit{a priori}$ criteria yield parametrizations that tend to be unstable in direct simulations and describe how subgrid parametrizations can alternatively be trained end-to-end in order to meet $\textit{a posteriori}$ criteria. We illustrate that end-to-end learning strategies yield parametrizations that outperform known empirical and data-driven schemes in terms of performance, stability and ability to apply to different flow configurations. These results support the relevance of differentiable programming paradigms for climate models in the future.

physics.flu-dyn↗

arXiv2021

A posteriori learning of quasi-geostrophic turbulence parametrization: an experiment on integration steps

Modeling the subgrid-scale dynamics of reduced models is a long standing open problem that finds application in ocean, atmosphere and climate predictions where direct numerical simulation (DNS) is impossible. While neural networks (NNs) have already been applied to a range of three-dimensional flows with success, two dimensional flows are more challenging because of the backscatter of energy from small to large scales. We show that learning a model jointly with the dynamical solver and a meaningful \textit{a posteriori}-based loss function lead to stable and realistic simulations when applied to quasi-geostrophic turbulence.

cs.LG↗

arXiv2021

A finite-size correction model for two-fluid Large-Eddy Simulation of particle-laden boundary layer flow

In this paper the capabilities of the turbulence-resolving Eulerian-Eulerian two-phase flow model to predict the suspension of mono-dispersed finite-sized solid particles in a boundary layer flow are investigated. For heavier-than-fluid particles, having settling velocity of the order of the bed friction velocity, the two-fluid model significantly under-estimates the turbulent dispersion of particles. It is hypothesized that finite-size effects are important and a correction model for the drag law is proposed. This model is based on the assumption that the turbulent flow scales larger than the particle diameter will contribute to the resolved relative velocity between the two phases, whereas eddies smaller than the particle diameter will have two effects: (i) they will reduce the particle response time by adding a sub-particle scale eddy viscosity to the drag coefficient, and (ii) they will contribute to increase the production of granular temperature. Integrating finite-size effects allows us to quantitatively predict the concentration profile for heavier-than-fluid particles without any tuning parameter. The proposed modification of the two-fluid model extends its range of applicability to tackle particles having a size belonging to the inertial range of turbulence and allows us to envision more complex applications in terms of flow forcing conditions, i.e. sheet flow, wave-driven transport, turbidity currents and/or flow geometries, i.e. ripples, dunes, scour.

physics.flu-dyn↗

arXiv2021

Physical invariance in neural networks for subgrid-scale scalar flux modeling

In this paper we present a new strategy to model the subgrid-scale scalar flux in a three-dimensional turbulent incompressible flow using physics-informed neural networks (NNs). When trained from direct numerical simulation (DNS) data, state-of-the-art neural networks, such as convolutional neural networks, may not preserve well known physical priors, which may in turn question their application to real case-studies. To address this issue, we investigate hard and soft constraints into the model based on classical transformation invariances and symmetries derived from physical laws. From simulation-based experiments, we show that the proposed transformation-invariant NN model outperforms both purely data-driven ones as well as parametric state-of-the-art subgrid-scale models. The considered invariances are regarded as regularizers on physical metrics during the a priori evaluation and constrain the distribution tails of the predicted subgrid-scale term to be closer to the DNS. They also increase the stability and performance of the model when used as a surrogate during a large-eddy simulation. Moreover, the transformation-invariant NN is shown to generalize to regimes that have not been seen during the training phase.

physics.flu-dyn↗

arXiv2021

Spatio-temporal correlations in 3D homogeneous isotropic turbulence

We use Direct Numerical Simulations (DNS) of the forced Navier-Stokes equation for a 3-dimensional incompressible fluid in order to test recent theoretical predictions. We study the two- and three-point spatio-temporal correlation functions of the velocity field in stationary, isotropic and homogeneous turbulence. We compare our numerical results to the predictions from the Functional Renormalization Group (FRG) which were obtained in the large wavenumber limit. DNS are performed at various Reynolds numbers and the correlations are analyzed in different time regimes focusing on the large wavenumbers. At small time delays, we find that the two-point correlation function decays as a Gaussian in the variable $kt$ where $k$ is the wavenumber and $t$ the time delay. The three-point correlation function, determined from the time-dependent advection-velocity correlations, also follows a Gaussian decay at small $t$ with the same prefactor as the one of the two-point function. These behaviors are in precise agreement with the FRG results, and can be simply understood as a consequence of sweeping. At large time delays, the FRG predicts a crossover to an exponential in $k^2 t$, which we were not able to resolve in our simulations. However, we analyze the two-point spatio-temporal correlations of the modulus of the velocity, and show that they exhibit this crossover from a Gaussian to an exponential decay, although we lack of a theoretical understanding in this case. This intriguing phenomenon calls for further theoretical investigation.

physics.flu-dyn↗

arXiv2020

Analysis of the dissipative range of the energy spectrum in grid turbulence and in direct numerical simulations

We present a statistical analysis of the behavior of the kinetic energy spectrum in the dissipative range of scales of fully developed three-dimensional turbulence. We analyze spectra recorded in experiments of grid turbulence generated in the Modane wind tunnel, and spectra obtained from high-resolution direct numerical simulations of the forced Navier-Stokes equation. We focus on wave-number scales in the near-dissipative range. We find that in this domain, the spectra behave as a stretched exponential with an exponent $α= 0.68 \pm 0.19$. This result corroborates previous DNS studies which found that the spectrum in the near-dissipative range is best modeled by a stretched exponential with $α<1$. Moreover, we here give an estimate for the actual value of $α$, which turns out to be in precise agreement with the theoretical prediction $α=2/3$ from the non-perturbative renormalisation group.

physics.flu-dyn↗

arXiv2019

Numerical study of the cryogenic cooling of amplifiers for high power lasers

The French collaborative Trio4CLF project aims to understand and control the cryogenic cooling of amplifiers for high power (~1 PetaWatt) and high repetition rate (1-10 Hertz) lasers. In such amplifiers, the fluid (low temperature gaseous helium) evacuates the thermal power absorbed by the solid amplifying plates. A precise knowledge of the heat exchange and of the turbulent fluid flow in the amplifier is requested to evaluate its impact on the laser beam quality. Large Eddy Simulations representative of the amplifier cooling are performed using TrioCFD, a code developed by the CEA. First, validation simulations are carried out for heated periodic channel flows, allowing comparisons with Direct Numerical Simulation results from the literature. They are conducted using conjugate heat transfer calculation between the fluid and the solid. The channel flow is turbulent, with a bulk Reynolds number (based and the bulk velocity and the total height of the channel) of about 14000. Large Eddy Simulation of a heated open turbulent cryogenic helium flow with developing thermal boundary layers is then presented.

physics.flu-dyn↗

arXiv2019

Simulation des Grandes Échelles d'un écoulement d'air turbulent pour le refroidissement d'amplificateurs de lasers

The French collaborative Trio4CLF project aims to understand and control the cryogenic cooling of amplifiers for high power (~1 PetaWatt) and high repetition rate (1-10 Hertz) lasers. In such amplifiers, the fluid evacuates the thermal power absorbed by the solid amplifying plates. A precise knowledge of the heat exchange and thus of the turbulent fluid flow in the amplifier is requested to evaluate its impact on the laser beam quality. As a first step, a Large Eddy Simulation is carried out in air without heating to study the development of the turbulent flow. The CFD code used is TrioCFD, a code developed by the CEA. For validation purpose, the simulation is carried out in the experimental setup configuration: a closed-loop wind tunnel called TRANSAT. Two horizontal plates, separated by 0.05 m, are put in the airflow to represent the amplifier plates. Turbulent boundary layers develop from the plates edges. Numerically, the entrance flow is a homogeneous planar flow with a constant velocity at 10 m/s. The results of this Large Eddy Simulation are presented in this paper as a study of the development of the turbulent boundary layers created by the plates.

physics.app-ph↗

arXiv2017

Spatiotemporal velocity-velocity correlation function in fully developed turbulence

Turbulence is an ubiquitous phenomenon in natural and industrial flows. Since the celebrated work of Kolmogorov in 1941, understanding the statistical properties of fully developed turbulence has remained a major quest. In particular, deriving the properties of turbulent flows from a mesoscopic description, that is from Navier-Stokes equation, has eluded most theoretical attempts. Here, we provide a theoretical prediction for the {\it space and time} dependent velocity-velocity correlation function of homogeneous and isotropic turbulence from the field theory associated to Navier-Stokes equation with stochastic forcing. This prediction is the analytical fixed-point solution of Non-Perturbative Renormalisation Group flow equations, which are exact in a certain large wave-number limit. This solution is compared to two-point two-times correlation functions computed in direct numerical simulations. We obtain a remarkable agreement both in the inertial and in the dissipative ranges.

physics.flu-dyn↗

arXiv2015

Non-Kolmogorov cascade of helicity driven turbulence

We solve the Navier-Stokes equations with two simultaneous forcings. One forcing is applied at a given large-scale and it injects energy. The other forcing is applied at all scales belonging to the inertial range and it injects helicity. In this way we can vary the degree of turbulence helicity from non helical to maximally helical. We find that increasing the rate of helicity injection does not change the energy flux. On the other hand the level of total energy is strongly increased and the energy spectrum gets steeper. The energy spectrum spans from a Kolmogorov scaling law $k^{-5/3}$ for a non-helical turbulence, to a non-Kolmogorov scaling law $k^{-7/3}$ for a maximally helical turbulence. In the later case we find that the characteristic time of the turbulence is not the turnover time but a time based on the helicity injection rate. We also analyse the results in terms of helical modes decomposition. For a maximally helical turbulence one type of helical mode is found to be much more energetic than the other one, by several orders of magnitude. The energy cascade of the most energetic type of helical mode results from the sum of two fluxes. One flux is negative and can be understood in terms of a decimated model. This negative flux is however not sufficient to lead an inverse energy cascade. Indeed the other flux involving the least energetic type of helical mode is positive and the largest. The least energetic type of helical mode is then essential and cannot be neglected.

physics.flu-dyn↗

arXiv2014

A systematic bias in the calculation of spectral density from a 3D spatial grid

The energy spectral density $E(k)$, where $k$ is the spatial wave number, is a well-known diagnostic of homogeneous turbulence and magnetohydrodynamic turbulence. However in most of the curves plotted by different authors, some systematic kinks can be observed at $k=9$, $k=15$ and $k=19$. We claim that these kinks have no physical meaning, and are in fact the signature of the method which is used to estimate $E(k)$ from a 3D spatial grid. In this paper we give another method, in order to get rid of the spurious kinks and to estimate $E(k)$ much more accurately.

physics.comp-ph↗

arXiv2011

Cross-helicity effects and turbulent transport in magnetohydrodynamic flow

In the presence of large-scale vortical motions and/or magnetic-field strains, the turbulent cross helicity (velocity--magnetic-field correlation in fluctuations) may contribute to the turbulent electromotive force and the Reynolds stress. These effects of cross helicity are considered to balance the primary effects of turbulence such as the turbulent magnetic diffusivity in magnetic-field evolution and the eddy viscosity in the momentum transport. The cross-helicity effects may suppress the enhanced transports due to turbulence. Physical interpretation of the effects is presented with special emphasis on the difference between the cross-helicity effect and the usual $α$ or helicity effect in the dynamo action. The relative importance of the cross-helicity effect in dynamo action is validated with the aid of a direct numerical simulation (DNS) of the Kolmogorov flow with an imposed magnetic field. Several mechanisms that provide turbulence with the cross helicity are also discussed.

astro-ph.SR↗