Study with the several resources on Docsity
Earn points by helping other students or get them with a premium plan
Prepare for your exams
Study with the several resources on Docsity
Earn points to download
Earn points by helping other students or get them with a premium plan
diffusion mass transfer equation
Typology: Cheat Sheet
1 / 30
This page cannot be seen from the preview
Don't miss anything!
In this Lecture, we shall present the main ideas behind the simulation of fluid turbulence. We firts discuss the case of the direct numerical simulation, in which all scales of motion within the grid resolution are simulated and then move on to turbulence modeling, where the effect of unresolved scales on the resolved ones is taken into account by various forms of modeling,
Turbulence is the peculiar state of matter, typical of gases and liquids, charac- terized by the simultaneous interaction of a broad spectrum of scales of motion, both in space and time. Such simultaneous interaction gives rise to a host of morpho-dynamical complexity which makes the mid/long-term behavior of tur- bulent flows very hard to predict, weather forecasting possibly offering the most popular example in point. The importance of turbulence, from both theoretical and practical points of view, cannot be overstated. Besides the intellectual challenge associated with the predictability of the dynamics of complex nonlinear systems, the practical relevance of turbulence is even more compelling as one thinks of the pervasive presence of fluids across most natural and industrial endeavours: air and blood flow in our body, gas flows in car engines, geophysical and cosmological flows, to name but a few. Even though the basic equations of motion of fluid turbulence, the Navier-Stokes equations, are known for nearly two centuries, the problem of predicting the behaviour of turbulent flows, even only in a statistical sense, is still open to this day. In the last few decades, numerical simulation has played a leading role in advancing the frontier of knowledge of this over-resilient problem (often called the last unsolved problem of classical physics). To appraise the potential and limitations of the numerical approach to fluid turbulence, it is instructive to revisit some basic facts about the physics of turbulent flows. The degree of turbulence of a given flow is commonly expressed in terms of a single dimensionless parameter, the Reynolds number, defined as
Re =
ν
where U is a typical macroscopic flow speed, L the corresponding spatial scale and ν is the kinematic molecular viscosity of the fluid. The Reynolds number measures the relative strength of advective over dissipative phenomena in a fluid flow: Re ∼ u∇u/(ν∆u). Given the fact that many fluids feature a viscosity around 10−^6 m^2 /s and many flows of practical interest work at speeds around and above U = 1 m/s within devices sized around and above L = 1 m, it is readily checked that Re = 10^6 is commonplace in real life applications. In other words, non-linear inertia far exceeds dissipation, a hallmark of macroscopic phenomena. The basic physics of turbulence is largely dictated by the way energy is transferred across scales of motion. To date, turbulence energetics is best understood in terms of an energy cascade from large scales (l ∼ L) where energy is fed into the system, down to small scales where dissipation takes central stage (see Fig. ??). This cascade is driven by the nonlinear mode–mode coupling in momentum space associated with the advective term of the Navier–Stokes equations, ~u · ∇~u.
movie...), we conclude that the number of degrees of freedom involved in a turbulent flow at Reynolds number Re is given by:
Ndof ∼ Re^9 /^4. (4)
According to this estimate, even a standard flow with Re = 10^6 , features more than 10^13 degrees of freedom, enough saturate the most powerful present- day computers! This sets the current bar of Direct Numerical Simulation (DNS) of turbulent flows, manifestly one falling short of meeting the needs raised by many real life applications. The message comes down quite plain: computers alone won’t do! Of course, this does not mean that computer simulation is useless. Quite the contrary, it plays a pivotal role as a complement and sometimes even an alternative to experimental studies.^1 Yet, the message is that sheer increase of raw compute power must be ac- companied by a corresponding advance of computational methods. Fluid turbulence is very sensitive to the space dimensionality. For instance, three-dimensional fluids support finite dissipation even in the (singular) limit of zero viscosity, while two-dimensional ones do not. This is rather intuitive, since three dimensional space offers much more morpho-dynamical freedom than two or one-dimensional ones, hence the flow can go correspondingly ”wilder”. Therefore, we shall begin our discussion with two-dimensional turbulence.
As noted above, two-dimensional turbulence differs considerably from three- dimensional turbulence. In particular, two-dimensional turbulence supports an infinite number of (Casimir) invariants which can only exist in ‘Flatland’. These read as follows:
Ω 2 p =
V
|ω|^2 p^ V, p = 1, 2 ,... , (5)
where ~ω = ∇ × ~u (6)
is the vorticity of the fluid occupying a region of volume V. By taking the curl of the Navier–Stokes equations, one obtains:
Dt~ω = ν∆~ω + ~ω · ∇~u. (7)
where Dt is the material derivative. It is easily seen that the second term on the right-hand side, acting as a source/sink of vorticity, is identically zero in two dimensions because the vorticity is orthogonal to the flow field.
(^1) As pointed out by P. Moin and K. Mahesh, ‘DNS need not attain real life Reynolds numbers to be useful in the study of real life applications.’ (P. Moin and K. Mahesh, Direct numerical simulation: a tool in turbulence research, Ann. Rev. Fluid Mech. 30 , 539, 1998.)
As a result, in the inviscid limit ν → 0, vorticity is a conserved quantity (topological invariant), and so are all its powers. The fact that vorticity, and particularly enstrophy Ω 2 , is conserved has a profound impact on the scaling laws of 2D turbulence. It can be shown that the enstrophy cascade leads to a fast decaying (k−^3 ) energy spectrum, as opposed to the much slower k−^5 /^3 fall-off of 3D turbulence. This regularity derives from the existence of long lived metastable states, vortices, that manage to escape dissipation for quite long times. The dynamics of these long-lived vortices has been studied in depth by various groups and reveals a number of fascinating aspects whose description goes however beyond the scope of this Lecture.
Most real-life flows of practical interest exhibit Reynolds numbers far too high to be amenable to direct simulation by present-day computers, and for many years to come. This raises the challenge of predicting the behavior of highly turbulent flows without directly simulating all scales of motion but only those that fit the avail- able computer resolution. The effect of the unresolved scales of motion on the resolved ones must therefore be modeled. To this purpose, it proves expedient to split the actual velocity field into large-scale (resolved) and short-scale (unresolved) components:
ua = Ua + ˜ua (8)
and seek turbulent closures yielding the Reynolds stress tensor:
σab = ρ〈˜ua u˜b〉 (9)
in terms of resolved field Ua. Here, brackets denote ensemble-averaging, typi- cally replaced by space-time averaging on the grid. This task makes the object of an intense area of turbulence research, known as turbulence modeling, or sub-grid scale (SGS) modeling. One of the most powerful heuristics behind SGS is the concept of eddy vis- cosity. This idea, a significant contribution of kinetic theory to fluid turbulence, assumes that the effect of small scales on the large ones can be likened to a diffusive motion caused by random collisions. Small eddies are kinematically transported without distortion by the large ones, while the large ones expe- rience diffusive Brownian-like motion due to erratic collisions with the small eddies. One of the simplest and most popular models in this class is due to Smagorin- ski [?]. This model is based on the following representation of the Reynolds stress tensor: σab = ρνe (|S|) Sab, (10)
The k − model often provides an improvement over Smagorinski, but it is still liable to criticism. In particular, it does not account for the higher directional nature of turbulence near solid boundaries. To cope with this problem, a further level of sophistication is introduced, by formulating dynamic equations for the Reynolds stress itself σab. These equations tends to be rather cumbersome, espe- cially in connection with the formulation of proper boundary conditions. Even though they enjoy significant popularity in the CFD engineering community, we shall not delve any further into this option.
The modeling of fluid turbulence still stands as one of the major open topics in modern science and engineering. Current computer capabilities allow the full simulation of flows up to Reynolds numbers of the order of 10^4 , which is short by several orders of magnitude for many engineering applications, let alone environmental and geophysical ones. To fill thsi gap, teh current practice is extensive resort to various forms of turbulence modeling.
