Linear stability

In mathematics, in the theory of differential equations and dynamical systems, a particular stationary or quasistationary solution to a nonlinear system is called linearly unstable if the linearization of the equation at this solution has the form dr/dt = A r, where r is the perturbation to the steady state, A is a linear operator whose spectrum contains eigenvalues with positive real part. If all the eigenvalues have negative real part, then the solution is called linearly stable. Other names for linear stability include exponential stability or stability in terms of first approximation. If there exist an eigenvalue with zero real part then the question about stability cannot be solved on the basis of the first approximation and we approach the so-called "centre and focus problem". Examples Ordinary differential equation The differential equation \frac{dx}{dt} = x - x^2 has two stationary (time-independent) solutions: x&nb

Simulation of Time-Dependent Viscoelastic Fluid Flows by Spectral Elements

Azadeh Jafari

The research work reported in this dissertation is aimed to develop efficient and stable numerical schemes in order to obtain accurate numerical solution for viscoelastic fluid flows within the spectral element context. The present research consists in the transformation of a large class of differential constitutive models into an equation where the main variable is the logarithm of the conformation tensor or a quantity related to it in a simple way. Particular cases cover the Oldroyd-B fluid and the FENE-P model. Applying matrix logarithm formulation in the framework of the spectral element method is a new type of approach that according to our knowledge no one has implemented before. The reformulation of the classical constitutive equation using a new variable namely the logarithmic formulation, enforces the eigenvalues of the conformation tensor to remain positive for all steps of the simulation. However, satisfying the symmetric positive definiteness of the conformation tensor during the simulation is the necessary condition for stability; but definitely, it is not the sufficient condition to reach meaningful results. The main effort of this research is devoted to introduce a new algorithm in order to overcome the drawback of direct reformulating the classical constitutive equation to the logarithmic one. To evaluate the capability of the extended matrix logarithm formulation, comprehensive studies have been done based on the linear stability analysis to show the influence of this method on the resulting eigenvalue spectra and explain its success to tackle high Weissenberg numbers. With this new method one can treat high Weissenberg number flows at values of practical interest. One of the worst obstacles for numerical simulation of viscoelastic fluids is the presence of spurious modes during the simulation. At high Weissenberg number, many schemes suffer from instabilities and numerical convergence may not be attainable. This is often attributed to the presence of solution singularities due to the geometry, the dominant non-linear terms in the constitutive equations, or the change of type of the underlying mixed-form differential system. Refining the mesh proved to be not very helpful. In this study, to understand more deeply the mechanism of instability generation a comprehensive study about the growth of spurious modes with time evolution, mesh refinement, boundary conditions and Weissenberg number or any other affected parameters has been performed. Then to get rid of these spurious modes the filter based stabilization of spectral element methods proposed by Boyd was applied with success.

EPFL2011

Simulation numérique de phénomènes MHD

Gilles Steiner

The purpose of this thesis is the study, from the numerical simulation point of view, of the aluminum electrolysis process. Navier-Stokes equations for the computation of a two fluids flow with free interface are coupled with Maxwell equations describing the electric current repartition and the magnetic induction field in an electrolysis reduction cell. The emphasis is set on an efficient method for the computation of the magnetic induction in an unbounded domain. The algorithm is based on a Schwarz domain decomposition method and on the Poisson integral representation formula for harmonic functions. The partial differential equations that rule the phenomena are discretized in space and time and implemented in an existing numerical simulation software. This code is then tested on an academic test case and also in a more realistic situation. The key parts of the mathematical model are emphasized. Finally the time-evolution model is compared with another approach, dealing with stationary situations and their linear stability.

EPFL2009

Simulation of Time-Dependent Viscoelastic Fluid Flows by Spectral Elements

Azadeh Jafari

EPFL2011

Simulation of Time-Dependent Viscoelastic Fluid Flows by Spectral Elements

Effect of E x B flows on the linear stability of ion temperature gradient modes, using a global and spectral gyrokinetic model

Simulation numérique de phénomènes MHD

Simulation of Time-Dependent Viscoelastic Fluid Flows by Spectral Elements

Effect of E x B flows on the linear stability of ion temperature gradient modes, using a global and spectral gyrokinetic model

Simulation numérique de phénomènes MHD