1. Scientific motivation for
this project
1.1 Introduction
Fluid Mechanics is a very active research field
involving scientists from various disciplines, especially physicists, mathematicians,
engineers and computer scientists.
It has a long history which goes back to L. Euler
who introduced the celebrated Euler equations (both the compressible and
the incompressible ones) in 1755. It is worth recalling that L. Euler mentioned
immediately the analytical (we would say mathematical nowadays) difficulties
associated with these models. And, two centuries and a half later, some
of these difficulties are still important issues which generate a lot of
activity in Mathematics and Scientific Computing: let us mention, in particular,
the questions of the formation of singularities in three-dimensional solutions
of the incompressible Euler equations, of the mathematical structure of
solutions of compressible Euler questions, etc.
Despite fundamental contributions by outstanding
scientists and mathematicians like Riemann, von Karman, Leray, von Neumann,
Hoff, Lax, etc., basic issues concerning the mathematical understanding
of Fluid Mechanics equations is still far from being satisfactory.
In addition, computers and the possibility
of numerical simulations since the late 40's and von Neumann's work have
changed drastically the field where the theoretical mathematical analysis
is now intimately connected with numerical experiments and simulations.
Furthermore, the progress in power and speed of modern computers has led
engineers and scientists into considering more and more complex fluid flows
and mathematical models.
Therefore, the complexity of realistic fluid
flows together with the need of advanced tools for their numerical simulation
demand for a systematic investigation of the fundamental mathematical models.
The latter include not only the classical nonlinear
partial differential equations, such as Euler or Navier-Stokes equations,
but also their modifications which are motivated by the need of reducing
their numerical complexity. Let us mention, for instance, the various turbulence
models, or coupled models such as those involving the interaction between
viscous and inviscid flows, fluids and structures, etc.
Our goal in this programme is to improve the
mathematical understanding of these models as well as of the various
existing numerical methods and approaches. In addition we expect to propose
new methods based upon the already known fundamental mathematical properties
of the Fluid Mechanics equations.
Research on such challenging mathematical and
computational issues would greatly benefit from a concerted action among
several European groups.
We describe below, in particular, the objectives
of the proposed programme, the problems and methods that will be investigated,
as well as the activities that we are going to undertake. Before we do
so, we wish to make a few more specific remarks on the subject which better
motivate this proposal and outline why this is a good time to promote this
concerted action.
1.2 Remarks on some recent mathematical progress
on fluid flows
We shall not attempt to review all recent
progress on fluid flows but we shall only give two examples indicating
some possible research directions for the groups concerned with this proposal.
The first example concerns compressible Navier-Stokes equations. In the
past ten years, much progress has been made on the one-dimensional case
by Kezhilzhov and his collaborators, D. Hoff and D. Serre, and, very recently,
global weak solutions have been obtained by P.L. Lions. These mathematical
achievements bring the subject to a situation somewhat comparable to the
one for incompressible flows after J. Leray's work. In view of the
impact of J. Leray's solutions and, more importantly, proofs and methods
from the numerical analysis of incompressible Navier-Stokes equations,
we may expect considerable progress on the mathematical understanding of
numerical methods for compressible Navier-Stokes equations.
The second example we want to mention concerns gas-dynamics
equations and compressible Euler equations, which represent one of the
fundamental examples of hyperbolic conservation laws. This is obviously
a subject where the interplay between mathematical theory, the construction
of numerical methods, and computer experiments, has played a crucial role
in the development of the field. The mathematical theory of compensated
compactness of L. Tartar (and F. Murat, R. Di Perna) combined with the
earlier idea of "Lax entropies" has had and is still having many applications
to the numerical analysis of finite differences and finite volumes methods.
More recently, the kinetic formulation of such equations and the related
idea of the so-called Boltzmann schemes are beginning to receive much attention
leading to truly multidimensional schemes.
Let us emphasise the fact that the above two
examples illustrate some of our goals in this proposal and the strong connections
between progress in the mathematical understanding, the understanding of
existing numerical methods and the design of new numerical approaches.
In the next subsection, we present more examples of efficient numerical
approaches that require a better mathematical understanding.
1.3 Remarks on some recent and perspective approaches
for the numerical simulation of fluid flows
Numerical methods have found widespread application
in the simulation of a large variety of flows of both scientific and industrial
interest. Engineers and physicists have brought remarkable ideas to the
subject, and adopted advanced mathematical tools for its investigation.
We will describe two families of problems which,
although not exhaustive of the field of our interest, are nonetheless of
paramount importance and useful to illustrate the kind of contribution
that this project could bring to the subject.
1.3.1 Modeling and simulation of turbulent and
reactive flows
Direct numerical simulation (DNS) is a method
in which all scales of motion of a turbulent flow are computed. A DNS must
include everything from the large integral scale to the dissipative (viscous,
or Kolmogoroff) scale.
It is well known that, apart from flows at moderate
Reynolds number, turbulent behaviour can hardly be reproduced by the DNS.
Indeed, for flows with high Reynolds numbers, the Kolmogoroff turbulent
scales are much smaller than the affordable time-step and mesh- spacings,
which are the time and length scales of the discretized problem. Thus the
turbulent diffusion cannot be captured by DNS. The level of the physical
model needs therefore to be reduced through the introduction of suitable
turbulence models. Instances are provided by the Baldwin-Lomax algebraic
method, the k-e method (Launder, Schumann, Spalding, Pironneau,...) for
the conservation of kinetic energy and its dissipation rate, the Reynolds
stress model, the renormalization group theory (Orszag, Yakot,...), and
the large eddy simulation (LES) method (Ferziger, Rogallo, Germano, ..).
LES provides a direct numerical simulation of large vortical structures
but models the exchange mechanism that transfers energy from high to low
frequencies. Thus, when LES is adopted, the largest scales of motion are
represented explicitly while the small scales are treated by some approximate
parameterization or model. Large eddy simulations are three dimensional
and time-dependent, and therefore costly, although they are often much
less costly than DNS of the same flow.
In turbulence modeling and in LES, we are facing
a difficult problem with the correct treatment of the boundary conditions.
Experiments reveal that coherent structures exist there and herefore, a
major modeling effort has to be carried out in order to represent the corresponding
physical phenomena.
More generally, the use of advanced multiscale
numerical methods which can take care of flows with multiple scales, is
of great help not only for their simulation but also for the investigation
of the computed flow-fields (e.g., by wavelet transformation), and,
consequently, provide a tool for carrying out a-posteriori error analysis
and adaptivity.
Chemical reactions or combustion (e.g.
premixed turbulent combustion or diffusion controlled combustion) add further
complexity to the flow structure. Montecarlo techniques for the transport
of the probability density function (pdf) are usually adopted for the simulation
of turbulent combustion, in alternative to the DNS.
One of the aims of our project is bringing
together mathematicians, physicists and engineers to cooperate on the following
items: modeling the physics of turbulent and reactive flows, develop a
mathematical analysis of the associated partial differential equations,
simulate them by numerical methods, and investigate their application
to real life problems.
1.3.2 Shock-capturing methods for nonlinear conservation
laws
As pointed out by P. Lax in a recent historical
retrospective, interest in shock waves in the United States started during
the Second World War, culminating in the well-known, influential treatise
on supersonic flow and shock waves by Courant and Friedrichs. A systematic
effort to compute flows with shock waves was initiated during the war by
von Neumann at Los Alamos. Starting with an idea conceived in 1944, von
Neumann and Richtmyer in 1950 invented a numerical scheme that, with
the aid of artificial viscosity, captured shocks as rapid internal transitions.
Upon Richtmyer's influential monograph on numerical
methods for solving initial value problems, Lax's contribution laid in
the convergence and accuracy of difference approximations of solutions
of hyperbolic equations. For linear equations the main results are that
stability implies convergence, and that stability follows from the stability
of the "frozen" difference schemes with constant coefficients, something
that can be checked by Fourier transformation and matrix analysis. For
nonlinear equations proof of convergence is much harder, and for realistic,
complicated problems analysis furnishes "design principles". One of the
important design principles is that conservation laws should be approximated
by difference equations in conservation form, employing a numerical flux
that approximates the physical flux. It was proven that if approximate
solutions constructed by such a scheme converge strongly, say in the L1
sense, then the limit satisfies the original conservation laws.
Another useful design principle is that if an entropy
function can be defined for a given system of conservation laws, then the
numerical method employed to construct approximate solutions should avoid
increasing entropy.
Nowadays, after McCormack, Baldwin, Harten and Jameson
in the 70's, almost all the efficient numerical schemes are nonlinear,
i.e. they introduce nonlinear dissipation. Numerical dissipation provides
stability, damps unwanted oscillations generated by second and higher order
schemes around discontinuities and stiff layers. It is introduced either
by centered approximations (using second and fourth order smoothers, or
even spectral viscosity), or resorting to uncentered, upwind like schemes.
Upwind techniques introduce physical information
in the numerical discretization (the direction of propagation of
waves) (Courant, Isaacson and Rees, Steger and Warming, Van Leer,..). At
a first stage of their evolution they have been designed so to be conservative,
and later on to take into account information on exact solutions as well,
by generalizing Godunov's idea of solving one-dimensional Riemann's
problems at interfaces (also numerically, after Roe and Osher).
The total variation diminishing schemes (TVD), and
the essentially non oscillatory schemes (ENO), have then been introduced
(Harten , Engquist, Osher, ...) with the purpose of getting high
resolution schemes without loosing the control on the total variation of
the numerical solution (smoothing out the potential oscillations), and
guaranteeing high order away from the shock (including the extrema of the
flow flux). The extension of these techniques to multidimensional vector
problems is still an active area of investigation, especially when using
finite volume or finite element (unstructured) grids, where multidimensional,
cartesian splitting is no longer straightforward (Baines, Hughes, Deconinck,
Dervieux, ...). In our days, there is a cross fertilization of ideas
between the finite element approach using nonlinear stabilization a' la
SUPG (Johnson, Szepessy, Tezduyar, Deconinck, Roe,..) or the so-called
discontinuous finite element method (Raviart, Lesaint, Cockburn, Shu, ...)
and the finite volume method using node centered unknowns and dual cells.
In particular, this allows the use of the rich finite element functional
framework (including the compensated compactness technique, or the use
of equivalent principles between SUPG methods and Galerkin methods with
bubble corrections recently introduced by Brezzi, Franca, Hughes,...) to
analyze shock capturing schemes on unstructured grids.