For over a century now, the field of hydrodynamic instabilities has been constantly and abundantly renewed, and enriched by a fruitful dialogue with other fields of physics: phase transitions, nonlinear optics and chemistry, plasma physics, astrophysics and geophysics. Observation and analysis have been stimulated by new experimental techniques and numerical simulations, as well as by the development and adaptation of new concepts, in particular, those related to asymptotic analysis and the theory of nonlinear dynamical systems. Ever since the observations of Osborne Reynolds in 1883, there has been unflagging interest in the fundamental problem of the transition to turbulence. This topic has been given new life by concepts such as convective instabilities, transient growth, and by the recognized importance of unstable nonlinear solutions. New problems have emerged, such as flows involving fluid–structure interactions, granular flows, and flows of complex fluids – non-Newtonian and biological fluids, suspensions of particles, bubbly flows – where constitutive laws play an essential role.

This book has been written over the course of 10 years of teaching postgraduate students in fluid dynamics at the University of Toulouse. It is intended for any student, researcher, or engineer already conversant with basic hydrodynamics, and interested in the questions listed above.

Introduction

As we have discussed in earlier chapters, a dissipative physical system in a uniform, stationary state can become linearly unstable when a control parameter R exceeds a critical value Rc. This occurs, for example, in the instability of a fluid layer heated from below governed by the Rayleigh number, or in the instability of plane Poiseuille flow governed by the Reynolds number. The instability can be manifested as the appearance of a stationary, spatially periodic structure (Rayleigh–Bénard convection rolls, for example), or as a growing traveling wave (Tollmien–Schlichting waves). This type of situation was studied in Chapter 8 for spatially confined systems, or systems with imposed periodicity, where the dynamics can be reduced to a system of differential equations for the amplitudes of a few spatial harmonics. When the physical system is spatially extended, that is, when its size is large compared to the wavelength of the periodic structure, the wave number spectrum tends to become continuous, and spatial modulations of the amplitudes can arise. The appropriate formalism for describing these modulations is that of envelope equations.

In the present chapter we shall present this formalism and study the conditions for saturation of the primary instability arising at R = Rc, as well as for secondary instabilities which arise when the bifurcation parameter exceeds a second threshold. First we study the case where the periodic structure is stationary, that is, where the eigenvalue of the linearized system crossing the imaginary axis at threshold R = Rc is real, i.e., the bifurcation is of the saddle–node or pitchfork type for a system possessing the reflection symmetry x → -x.

Introduction

A useful mathematical framework for studying linear and nonlinear stability is the theory of ordinary differential equations (ODEs), also known as the theory of dynamical systems when the focus is on geometric and qualitative representations of the ideas and solutions. An informal presentation of this theory was given in Chapter 1. The goal of this chapter is to give a more systematic account of it from a mathematical point of view. In particular, we shall show how to reduce the number of degrees of freedom of a problem to obtain the “normal forms” of elementary bifurcations referred to often in the previous chapters.

In spite of its restrictive nature relative to the theory of partial differential equations, the theory of dynamical systems has revealed the extraordinary richness and complexity of the types of behavior that can arise when nonlinear effects play a role. This theory originated in the work of Henri Poincaré, in particular, in his book Méthodes nouvelles de la mécanique céleste. Poincaré's ideas were further developed during the first half of the twentieth century by the Russian school of mathematics (Kolmogorov, Arnold). The discovery that a system with a small number of degrees of freedom can display unpredictable, chaotic behavior then led to a great deal of research in this area beginning in the 1960s. Qualitatively new concepts such as deterministic chaos and sensitivity to the initial conditions were introduced, and these have significantly modified our understanding of deterministic models and their use in the description of natural phenomena.

Introduction

The inertial instability of parallel flows described in the preceding chapter is associated with the existence of an inflection point in the velocity profile. This is the principal instability of parallel or quasi-parallel shear flows at large Reynolds number and far from walls or interfaces, such as mixing layers, jets, and wakes. We have seen that the order of magnitude of the growth rate is U/δ, where U is the difference of the speeds on either side of the vorticity layer of thickness δ, and that viscosity plays only a diffusive role tending to attenuate the growth rate. The instability of flow profiles without inflection points is profoundly different. Let us consider two fundamental flows: plane Poiseuille flow and boundary layer flow. Observation shows that plane Poiseuille flow is unstable beyond a certain Reynolds number. Similarly, a boundary layer on a surface becomes unstable at some distance from the leading edge. However, these two flows do not possess an inflection point, and so, ignoring viscosity, they are stable according to the Rayleigh theorem. On the other hand, the growth rate of the observed instabilities is much smaller than would be expected for an inertial instability. It is therefore clearly important to investigate the role played by viscosity, which is the goal of the present chapter. We shall see that viscosity has two effects: the expected stabilizing dissipative effect, and also a destabilizing effect.

In Chapter 6 we discussed the fundamentals of turbulence and reviewed the mixing length and eddy diffusivity models. As was mentioned there, these classical models do not treat turbulence as a transported property, and as a result they are best applicable to equilibrium turbulent fields. In an equilibrium turbulent field at any particular location there is a balance among the generation, dissipation, and transported turbulent energy for the entire eddy size spectrum, and as a result turbulence characteristics at each point only depend on the local parameters at that point.

Our daily experience, however, shows that turbulence is in general a transported property, and turbulence generated at one location in a flow field affects the flow field downstream from that location. One can see this by simply disturbing the surface of a stream and noting that the vortices resulting from the disturbance move downstream.

In this chapter, turbulence models that treat turbulence as a transported property are discussed. Turbulence models based on Reynolds–averaged Navier–Stokes [(RANS)-type] models are first discussed. These models, as their title suggests, avoid the difficulty of dealing with turbulent fluctuations entirely. We then discuss two methods that actually attempt to resolve these turbulent fluctuations, either over the entire range of eddy sizes [direct numerical simulation (DNS) method] or over the range of eddies that are large enough to have nonuniversal behavior [largeeddy simulation (LES) method].

An integral method is a powerful and flexible technique for the approximate solution of boundary-layer problems. It is based on the integration of the boundary-layer conservation equations over the boundary-layer thickness and the assumption of approximate and well-defined velocity, temperature, and mass-fraction profiles in the boundary layer. In this way, the partial differential conservation equations are replaced with ODEs in which the dependent variable is the boundary-layer thickness. The solution of the ODE derived in this way then provides the thickness of the boundary layer. Knowing the boundary-layer thickness, along with the aforementioned approximate velocity and temperature profiles, we can then easily find the transport rates through the boundary layer. The integral technique is quite flexible and, unlike the similarity solution method, can be applied to relatively complicated flow configurations.

Integral Momentum Equations

Let us first consider the velocity boundary layer on a flat plate that is subject to the steady and uniform parallel flow of a fluid, as shown in Fig. 5.1. We define a control volume composed of a slice of the flow field that has a thickness dx and height Y. We choose Y to be large enough so that it will be larger than the boundary-layer thickness throughout the range of interest. The inflow and outflow parameters relevant to momentum and energy are also depicted in Fig. 5.1.