This action might not be possible to undo. Are you sure you want to continue?

1 Theoreti
al ba
kground

1.1 Linear adve
tion equation . . . . . . . . . . . . . . . . . . . . . . . . .

1.1.1 Example: The equation of mass transport . . . . . . . . . . . .

1.1.2 Wave solutions of the Cau
hy problem for linear adve
tion equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1.1.3 Solution using the method of
hara
teristi
s . . . . . . . . . . .

1.1.4 Riemann problem of two gases. Conta
t dis
ontinuity . . . . .

1.1.5 Two-dimensional adve
tion type problems . . . . . . . . . . . .

1.2 Linear diusion equation . . . . . . . . . . . . . . . . . . . . . . . . . .

1.2.1 Mole
ular diusion: Fi
k's law . . . . . . . . . . . . . . . . . .

1.2.2 Solution of the diusion equation . . . . . . . . . . . . . . . . .

1.3 The steady-state linear adve
tion-diusion rea
tion problem . . . . . .

.

.

3

3

3

.

.

.

.

.

.

.

.

6

7

11

12

15

15

16

19

**2 Fun
tion-analyti
al ba
kground
**

2.1 Classi
al solutions . . . . . . . . . . . . . .

2.2 Sobolev spa
es and weak solutions . . . . .

2.3 Variational formulation for ellipti
problems

2.4 Galerkin method . . . . . . . . . . . . . . .

.

.

.

.

21

21

21

26

32

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

**3 Finite Element method
**

35

3.1 FEM for adve
tion-diusion problems . . . . . . . . . . . . . . . . . . . 35

3.2 Implementation of a

**nite element method . . . . . . . . . . . . . . . . . 40
**

4 Turbulent boundary layer
ows

4.1 Governing equations for vis
ous
uid
ow . . . . . . . .

4.2 Reynolds number. Laminar and turbulent
ow . . . . .

4.2.1 Flat plate boundary layer
ow . . . . . . . . . .

4.3 Laminar
ow . . . . . . . . . . . . . . . . . . . . . . . .

4.4 Turbulent
ow. . . . . . . . . . . . . . . . . . . . . . . .

4.5 Statisti
al approa
h . . . . . . . . . . . . . . . . . . . .

4.6 Universal solutions . . . . . . . . . . . . . . . . . . . . .

4.7 Statisti
al theory and modelling . . . . . . . . . . . . . .

4.7.1 The Reynolds averaged Navier-Stokes equations .

4.7.2 The Reynolds stress tensor . . . . . . . . . . . .

4.7.3 The turbulent boundary layer equations . . . . .

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

41

41

42

44

46

46

48

49

50

50

52

52

1

Contents

**4.7.4 Prandtl's mixing length model . . . . . . . . . . . . . . . . . . . 53
**

4.7.5 Universal form of the turbulent boundary layer equations . . . . 54

4.8 k-! turbulen
e model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55

**5 DG-FEM for s
alar
onservation laws
**

5.1 Weak formulation for DG-FEM . . . . . . . . .

5.2 Fun
tional setting, tra
es and numeri
al
uxes

5.3 Jump and averaging operators . . . . . . . . . .

5.4 DG-FEM formulation for hyperboli
problems .

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

57

57

58

59

61

**6 DG-FEM for diusion type problems
**

63

6.1 Weak formulation for DG-FEM for diusion type problems . . . . . . . 63

6.2 Consisten
y and
onservation . . . . . . . . . . . . . . . . . . . . . . . . 65

7 Finite-volume methods

67

7.1 Finite-volume formulation . . . . . . . . . . . . . . . . . . . . . . . . . . 67

7.2 Expli
it solution using low-storage Runge-Kutta method . . . . . . . . . 69

7.3 Interpretation of expli
it s
hemes. Downwind numbering . . . . . . . . . 71

2

3. temperature or on entration of a pollutant n under the a tion of an external adve tive velo ity . The linear adve tion equation des ribes the transport of a s alar valued quantity u like density .1 Theoreti al ba kground In this hapter the two basi type equations onsidered in this ourse are presented linear adve tion equation linear diusion equation (heat-transport equation) 1.1 Linear adve tion equation Let R d be a d-dimensional domain (d = 2.

We introdu e the notation d X v ~v = ~a r ai (1.eld ~a.1 Example: The equation of mass transport In this se tion we explain that the linear adve tion equation indeed des ribed the transport of a passive s alar .3) 1.1.1) xi i=1 Then the linear adve tion equation reads u ~u = 0 + ~a r t (1.2) In the parti ular ase of a 1-dimensional spa e-domain we have u u + a = 0 t x (1.

eld under onsideration. As an example. we study the equation of mass transport with Passive s alar quantity is uid density u Adve tive .

eld is the uid velo ity ~u ~a. 3 .

Thus for any arbitrary ontrol volume V R d (where the ontrol volume is at rest.4). Now we want to understand this equation physi ally.. 1. does not move) Time rate of hange of mass inside V = Mass ux through boundary of V Z Z ~u ~n d (1.e. We assume a ow . The mass ux through boundary of V is asso iated with the mass transported ( onve ted) by the velo ity ~u through boundary of V . For this purpose we onsider the following simple situation of a uniform ow in a ylindri al du t (see Fig.4) d~x = t V V Therein V denotes the boundary of ell V and ~n is the outer unit normal ve tor. i.1 Theoreti al ba kground The physi al prin iple of onservation of mass states that mass an neither be reated nor destroyed.

and therefore the in. Denote the uid density.eld parallel to the ylinder side walls.

nitesimal volume dV represents an in.

A2 and by the ylinder side walls.nitesimal " uid parti le" of mass dm = dV . on the ylinder side walls. Denote ~n1 and ~n2 the unit normal ve tor of A1 and A2 respe tively poinint in outward dire tion. Finally. Note that at the in ow boundary A1 we have ~u ~n1 < 0 and at the out ow boundary A2 ~u ~n2 > 0. the velo ity . We onsider a ontrol volume V bounded by the ross-se tions A1 .

i.1: Control volumes and surfa e uxes in FVM Change of mass m inside V = in time interval t Mass min in owing through A1 in t Mass mout out owing through A2 t Note that there is no mass ux through the ylinder side walls.eld is tangential to the wall ~u ~n = 0 (no penetration ondition). We start with the onservation prin iple dm= ρ dV 11 00 00 11 00 11 00 11 00 11 00 11 00 11 00 11 00 11 00 11 00 11 00 11 00 11 00 11 00 11 00 11 00 11 00 11 00 11 00 11 00 11 00 11 00 11 00 11 000 111 00 11 000 111 000 111 000 111 A1 A2 0 1 1 0 0 1 0 1 0 1 0 1 0 1 0 1 11 00 00 11 00 11 00 11 00 11 00 11 00 11 00 11 00 11 A : Area u u : velocity 11 00 00 11 00 11 00 11 00 11 00 11 00 11 00 11 00 11 00 11 00 11 00 11 00 11 00 11 00 11 00 11 00 11 00 11 00 11 00 11 00 11 00 11 00 11 00 11 00 11 00 11 V: control volume Figure 1. A uid parti le dm onve ted with the velo ity . So we have to ompute the mass ux m_ = m=t through a given surfa e A.e. the mass m rossing A in the time interval t.

A).eld ~u rosses A in t. 4 . A) denotes the distan e of dm from A. ~u ~n is the omponent of velo ity parallel to ~n and dist(dm. if ~u ~nt > dist(dm. where ~n is the surfa e normal ve tor of A.

Using the de.2: Mass passing through ross se tion A in time t volume V be omes Z d~x = t V Z A1 ~u ~n d Z A2 ~u ~n d Z W ~|{z} u ~n d =0 on W where W denotes the ylinder side wall.2 pass through ross se tion A in time t. n ∆t Figure 1. 1.1 Linear adve tion equation Thus all uid parti les ontained in the ylinder sket hed in Fig.1. The asso iated mass m in this ylinder of volume V = ~u ~nt A is given by m = V = ~u ~nt A: Thus the mass ux through A is given by m = ~u ~n A: t More generally the mass ux through A is given by the integral m_ = m_ = Z A ~u ~n d: Putting this all together the prin iple of mass onservation applied to the ontrol Cylinder of volume ∆ V cross section A u.

Using Gaussian Theorem we have Z V ~u ~n d = Z V r~ (~u) d~x 5 .nition of the boundary V of V we arrive at the desired form Z Z ~u ~n d (1.5) d~x = t V V This is alled the onservative (or integral) form of the equation des ribing the prin iple of mass onservation.

9) 1. Now for a moment we assume that and ~u are smooth.6) holds for an arbitrary ontrol volume ontaining ~x.9). where Ma is the speed of sound (whi h is about 340ms 1 for air at standard onditions). the so- alled Rankine-Hugoniot ondition. If su h a fun tion solves (1.9) in the weak sense.1 Theoreti al ba kground Therefore (1.2 Wave solutions of the Cau hy problem for linear adve tion equations We study the Cau hy problem de. Then from the fa t that (1. However.8) Air ows an be treated to be in ompressible if jj~ujj < 0:3Ma.9). the above mentionned Rankine-Hugoniot ondition.6) if and only if the data on both sides of the dis ontinuity satisfy a ertain ondition. Then using the relation r~ (~u) = |r~{z ~u} + r~ ~u =0 the linear adve tion equation be omes ~ = 0 + ~u r t (1.9) whi h are pie ewise C 1 ( ontinuously dierentiable) but whi h may have jumps (dis ontinuities). then the issue is "in whi h sense does this equation hold ?" This question will lead us to the on ept of weak solutions of (1.5) an be written as Z Z Z ~ (~u) d~x = 0 ~ (~u) d~x = + r (1. A ow is alled in ompressible if r~ ~u = 0 (1. We are interested in solutions of (1. then it does also solve (1. Using this on ept we will derive a ondition of fundamental importan e.6) d~x + r t V V V t whi h is alled divergen e form.1. we an also infer (1. In this le ture we are interested in to alled in ompressible uid ow. In the ase of non-smooth data. we infer that the integrand must vanish identi ally: ~ (~u) = 0 + r (1.7) t This form is alled dierential form of the prin iple of onservation of mass.

11) A Cau hy-problem is a partial dierential equation whi h is de. t > 0 g u u + a = 0 in R R + (1. t) 2 R 2 : x 2 R . t) = u0 (x) in R f 0 g (1.ned by this equation in the spa e-time domain R R + f (x. t 2 R .10) t x u(x.

ned on the entire R d (d 2 N ) for the spatial variable(s) and on the half plane of positive real numbers R >0 6 .

12) To see this..1. Therefore is is an initial value problem and we only have to impose an initial boundary ondition (i.1 Linear adve tion equation for the temporal variable.e. t) = u0 (x at) (1. a boundary ondition for the temporal variable at t = 0). we de. The solution of the Cau hy problem is given by u(x.

t) = u0 (x at) is often alled wave solution. We rewrite (1.12) has the following properties: A solution of the form u(x. t)t) (1. or u = u(x a(u. the solution u(x. 1). as su h a solution preserves the shape of the initial solution ( alled wave form) u0 (x) and propagates at a onstant speen a to the right (if a > 0) or left (if a < 0) as time evolves. 1.e. Both have wave solutions. If we .1.. t) = x at and then substitution gives u0 (z (x. Su h waves preserve the initial waveform u0 (x) although the initial waveform may be lo ally stret hed or ompressed due to variations in the wave speed a. x.10) as a formal s alar produ t in the spa e-time or x-t plane. a is alled wave speed: the wave-front. propagates with speed a. x.14) u We note that u t + a x is the dire tional derivative of u(x.3 Solution using the method of hara teristi s Now we study the linear adve tion problem with a = onst (linear) or a = a(u.13) an also be interpreted as wave solution. The solution u(x. t) (quasi-linear). t) whi h is onstant along ea h ray x at = x0 . t) is onstant along ea h ray x at = x0 whi h are alled hara teristi urves (or simply hara tersti s) or wave-fronts. t)) du z u (z (x. t x t x = 0 (1. t)) du z du + a 0 + a 0 = 0 = 0 (a a ) = 0 t x dz |{z} t dz x dz = a A solution of the form (1. i. u u u u + a = (1. Remark 1 A solution of the form u = u(x a(x)t) .ne z (x. a) . t) in the x-t plane in the dire tion (a.

x (x0 . t0 ) in the x t plane and step to (x0 + a. 1) 7 . t0 + 1) in (a.

15) dt In pra ti e it is more onvinient to use the x-t plane to sket h the hara teristi s.t) t u(x.3 (right) des ribes the more general ase of non- onstant a.t0 ) (t) = f(x(t).10).0) t=0 u 0 (x) a ( ξ. 2 ∆ t ) u(x.t0 ) (t)g dt (1. t). 1) in the x-t plane. 1. t) determine su h that x = + at 2. for ea h point (x. For a linear adve tion equation. For ea h (x. The hara teristi s provide method for det u(x. Figure 4.t0 ) (t) is the hara teristi urve rossing point (x0 .t0 ) (t) dx (x0 . (x0 . t0 ) in the x-t plane.16) Proof.t) u(x. t) = + = + a = 0 dt t x dt t x (x0 . t) 2 R 2 : = a . then the solution for u remains onstant. the slope of the urve x = x(t) is dx=dt. ∆ t ) t 1 3∆t (x 1 . and so we have dx = a (1.14) states that there is no hange in the solution u in the dire tion of (a. t) = u(. u u dx u u d u(x(t). 0) x Figure 1. Then u = onst along hara teristi urves (x0 .1 Theoreti al ba kground dire tion. 3 ∆ t ) u(x. t 2 R + .3: Illustration of hara teristi urves termining the solution expli itely. t) is given by 8 u(x. t0 ) 2 (x0 . 0) . Then (1. In the t x plane. Proposition 1 Let u be a solution of (1. In that ase the slope of the urve t = t(x) is dt=dx = 1=a.t1) slope dt/dx=1/a 2∆t 1 ∆t ∆x = a ∆t ∆x 2∆x 3∆x x u(x. t) in the x-t plane there is exa tly one hara teristi line passing through (x. The u(x. Here we des ribe the method for a = onst.

1.1 Linear adve tion equation

Remark 2 The method of hara teristi s allows us to al ulate the solution of a

partial dierential equation of adve tion type by

nding the solution of the
orresponding ordinary dierential equation for the
hara
teristi
urves. The
hara
teristi

urves are a one-parametri
family of
urves in the x-t plane whi
h are the integral

urves of the ordinary dierential equation dx=dt = a.

Example As an example we onsider the following quasi-linear partial dierential

equation

x

u

u

+

= 0

t

x

u(x; 0) =
(x)

u(0; t) = b(t)

R+

on R + f0g

on f0g R +

in

R+

**First we divide the PDE by x to obtain
**

1 u

u

+

= 0

t

x x

We apply (1.16) and see that a = a(x; t) = 1=x so that

dx

1

u =
onst for

=

dt

x

or u =
onst for dt=dx = x whi
h
an be integrated to give

1

u =
onst for t = x2 +
onst

2

The wavefronts are parabolas t = x2 =2 +
onst whi
h are shifted by the parameter

onst in t-dire
tion.

t=x 2 /2

t

b(t)

c(x)

x

**Figure 1.4: Sket
h of
hara
teristi
s for the example
**

Now we have to spe
ify the solution u(x; t) for ea
h point (x; t) in the spa
e-time domain R + R + . The value u(x; t) is
ompletely determined by the uniquely determined

9

1 Theoreti al ba kground

**hara
teristi
line passing through (x; t). Note that wavefronts (or
hara
teristi
s) with
**

t x2 =2 pass through the positive x-axis and hen
e for ea
h (x; t) on su
h a
hara
teristi
the solution is determined by
(x). The wavefronts with t x2 =2 pass through

the positive t-axis and thus the solution at (x; t) lo
ated on su
h a
hara
teristi
is

determined by b(t).

Now we
onsider an arbitrary point (x1 ; t1 ). The wavefront through (x1 ; t1 ) is

1 2

(x

2

t t1 =

x21 )

**Case 1. If t1 x21 =2 then this wavefront interse
ts the positive x-axis at x = x0 , say,
**

where t = 0 and therefore

1

t1 = (x20 x21 ):

2

Solving this expression for x0 gives

x0 =

q

x21 2t1 :

**Sin
e any two points (x0 ; t0 ), (x1 ; t1 ) on the same
hara
teristi
urve have the same
**

value of u, we
on
lude that u(x1 ; t1 ) = u(x0 ; 0) =
(x0 ) whi
h implies that

u(x1 ; t1 ) =

q

x21

2t1

:

**Case 2. If t1 > x21 =2, then the wavefront interse
ts the positive t axis at t = t0 , where
**

t0

1 2

x :

2 1

t1 =

**Solving this equation for t0 gives
**

1 2

x :

2 1

t0 = t1

**Sin
e any two points (x0 ; t0 ), (x1 ; t1 ) on the same
hara
teristi
urve have the same
**

value of u, we
on
lude that u(x1 ; t1 ) = u(0; t0 ) = b(t0 ) whi
h implies that

u(x1 ; t1 ) = b t1

1 2

x

2 1

After dropping the subs ripts we obtain the

we wish to summarize the following important notes 10 . t 21 x2 t > 21 x2 As a remark. b t 12 x2 .nal solution ( u(x. t) = p x2 2t .

t) ! 1 as x ! 0) we have a n < 0 at x = 0.1 Linear adve tion equation Initial and boundary onditions. Finally. we study the so- alled Riemann problem for the linear s alar adve tion equation. 0) = ur . t) = 1=x > 0 as x ! 0 (in fa t .4 Riemann problem of two gases. The outer unit normal ve tor n of the spatial domain = (0. Riemann problems are of fundamental importan e for the theoreti al and numeri al solution of hyperboli onservation laws. The ondition at x = 0 is alled boundary ondition as it imposes the ondition on the boundary of the spatial domain R + . As a(x. To ea h gas on the left and right hand side of the diaphragm. t) = ur . Conta t dis ontinuity A Riemann problem is a onservation law to be solved in R with pie ewise onstant initial data whi h have a single dis ontinuity. The Riemann problem for the linear s alar adve tion equation reads u u + a = 0 in R R + t x ( ul .1. x < at u(x. a(x. there orresponds a spe i. we onsider the situation of two gases in a one-dimensional tube initially separated by a diaphragm. we have to pres ribe onditions at t = 0 and where an < 0 (whi h means here: at x = 0). x > at In order to illustrate the physi al meaning of the solution of the Riemann problem. 1.1. Putting is together. x<0 u(x. The solution in R + R + is ompletely determined by the onditions at t = 0 and x = 0. x>0 Its solution is given by ( ul . This ondition says that "we have to impose initial and boundary onditions where hara teristi s enter the domain in the x-t plane". The ondition at t = 0 is alled initial ondition as it pres ribes the solution at the initial time t = 0. 1) at x = 0 is n = 1 and points into the negative x-dire tion.

t) = r . x>0 Its solution is given by ( l . x > ut 11 . This situation may be des ribed by the following Riemann problem for the equation of mass onservation with onstant velo ity u + u = 0 in R R + t x ( l . 0) = r . x<0 (x. onstant density l and r resp. x < ut (x.

(We assume that pressure and temperature are the same on both sides). see Fig. imagine the situation of a one-dimensional tube with two regions x < 0 Gas tube gas 1: O2 t diaphragm gas 2: N O 2 ρ l Contact discontinuity at x = a t ρ r x Figure 1. We apply a onstant velo ity . there are dierent gases.5.1 Theoreti al ba kground At t = 0. separated by a rigid diaphragm. Suppose at ea h side of the diaphragm. 1. and x > 0. say for x < 0 we have Oxygen O2 with density l and for x > 0 we have Nitrogen dioxide NO2 with density r .5: Illustration of solution of linear s alar Riemann problem and onta t dis ontinuity.

Now the velo ity . This is the situation at t = 0.eld u > 0 and assume that the diaphragm is instantaneously removed.

The dis ontinuity in (initially at x = 0 for t = 0) travels to the right with hara teristi speed u and is lo ated at x = ut at time t. the two gases remain in onta t along the hara terisi urve in the x t plane along whi h there is a jump in density. This dis ontinuity is a so alled onta t dis ontinuity: The gas initially on one side of the diaphragm never mixes with gas from the other side. We study a problem with onstant adve tion . This is the onta t dis ontinuity. As time evolves.1.5 Two-dimensional adve tion type problems In this se tion we onsider an adve tion type problem in a spatially two-dimensional domain.eld transports the gas mole ules with adve tion speed u > 0. 1.

we supply gas X1 for x > x0 and gas X2 for x < x0 . Lx ) (0. This is the situation sket hed in Fig.6 (left). At the in ow boundary. X2 = NO2 (Nitrogen dioxide) and X3 = N2 (Nitrogen).eld ~u = (ux . 1. say X1 = O2 (Oxygen). we pres ribe that ex lusively gas mole ules of type X3 are in . At t = 0. We onsider the orresponding initial-boundary value problem for the on entration 12 . uy )T in the spatial domain = (0. Ly ) with three gases involved.

1. otherwise For t large enough (here: t > maxfLx =ux .1 Linear adve tion equation t=0 y Gas X t = t1 y Gas X3 Gas X 2 Gas X3 2 x Gas X2 x Gas X2 Gas X1 Gas X1 Figure 1. y. y < uuxy (x x0 ) and x x0 < ux t .e. y. Ly =uy g). see Fig. y) 2 (~u) j x > x0 g R + (~u) j x < x0 g R + . y. t) n(x. t) = ( n0 0 = 0 in R + = 0 = n0 = 0 in f0g auf f(x.6: Adve tion type equation for on entration. n n n + ux + uy t x y n(x. t) Its solution is given by n(x. there is a onta t dis ontinuity. Adve tive transport of mole ules is along the integral urves of the velo ity . independent of t) ( n(x. y < uuxy (x x0 ) . Solution at t = 0 and t = t1 . 0) n(x. t) = n0 0 . n = nX1 of gas X1 . As time evolves. otherwise Now the aim is to illustrate what happens physi ally..10 (left and right). 1. y) 2 auf f(x. the mole ules of gas X1 and X2 are adve ted into the domain whereas the mole ules of gas X3 are adve ted out of the domain. Between ea h two dierent gases. the solution be omes stationary or steady-state (i.

This is alled a stationary solution of the orresponding time-dependent 13 . Fig. see Fig. At time t4 . whi h are straight lines for the ase ~u = onst. Thus at two arbitrary times t5 . f. 1. y) with y < uy =ux (x x0 ) we have ex lusively gas mole ules of type X1 and otherwise ex lusively mole ules of typ X2 .10 (left): For (x. X2 leave the domain at the out ow boundary. 1. the solution at t5 and t6 is exa tly the same.10 (right). see 1.10 (left). albeit the individual mole ules still hange. X2 enter the domain at the in ow boundary and the same number of mole ules of X1 resp.eld ~u. t6 > t4 . New mole ules of gas X1 resp. all mole ules of gas X3 have left the domain.

Right: Corresponding velo ity (adve tion) . t=t y t=0 y 4 uy ux Gas X3 Gas X 2 Gas X 2 x Gas X2 Gas X1 x Gas X2 Gas X1 Figure 1.8: Adve tion type equation for on entration. Left: At t = t4 the stationary solution is rea hed. Solution at t = 0 and t = t1 .7: Adve tion type equation for on entration.1 Theoreti al ba kground t = t2 y t = t3 y Gas X 3 Gas X Gas X3 Gas X 2 2 x Gas X2 x Gas X2 Gas X1 Gas X1 Figure 1.

eld. Remark 3 In the above example. problem. the gas mole ules are transported ex lusively along the integral urves of the velo ity .

there is a mixing of mole ules due to their thermal random motion. viz. In parti ular. this is an idealized situation. adve tion and diusion: An adve tion type solution: the mole ules are adve ted along the integral urves of the adve tion . However.eld. In a real uid.. The solution is a superposition of two solutions stemming from two dierent kinds of transport pro esses. there is a separation between the two gases at all times alled onta t dis ontinuity.

The next se tion is dedi ated to diusion type pro esses. Then diusion o urs in the dire tion in whi h the on entration de reases. Diusion tends to equalize the mole ular distribution of the diusing substan e until statisti al equilibrium is rea hed. Diusion an be observed whenever the on entration of a substan e (being the number of mole ules of the substan e per unit volume) varies from pla e to pla e.eld. 14 . Diusion is due to the individual random thermal motion of the mole ules. A diusion type solution: the mole ules diuse isotropi ally in all dire tions.

a few mole ules of gas X1 and X2 are found on the other's side after some while t1 (middle). 1. The on entration is the number of mole ules per unit volume and is in this situation a fun tion of the x- oordinate alone: = (x). At t = 0 the two gases X1 and X2 are separated by a diaphragm (left). des ribed by Fi k's law Diusion of heat (thermal energy).9 illustrates a diusion pro ess.1. the mixture of the two gases is homogeneous. des ribed by Fourier's law Diusion of momentum (internal fri tion in vis ous uids) 1. whi h is designated as the x-axis.2 Linear diusion equation 1. Then we de.2.9: Gaseous diusion.1 Mole ular diusion: Fi k's law Fig. Gas X1 11 00 11 00 11 00 111 000 111 000 111 000 111 000 11 00 11 00 11 00 11 00 00 11 11 00 00 11 111 000 000 111 000 111 11 00 00 11 00 11 111 000 000 111 000 111 000 111 11 00 11 00 11 00 t=0 11 00 11 00 11 00 11 00 11 00 11 00 00 11 11 00 00 11 00 11 00 11 11 00 00 11 111 000 000 111 000 111 000 111 Diaphragm 11 00 00 11 00 11 00 11 11 00 00 11 00 11 11 00 00 11 00 11 00 11 111 000 000 111 000 111 000 111 11 00 11 00 11 00 t = t1 111 000 111 000 111 000 11 00 11 00 11 00 11 00 11 00 11 00 11 00 11 00 11 00 11 00 000 111 111 000 000 111 000 111 11 00 00 11 00 11 11 00 00 11 00 11 00 11 11 00 00 11 00 11 11 00 00 11 00 11 00 11 11 00 00 11 00 11 111 000 000 111 000 111 000 111 11 00 11 00 11 00 111 000 111 000 111 000 000 111 111 000 000 111 000 111 111 000 111 000 111 000 111 000 111 000 111 000 111 000 111 000 111 000 111 000 111 000 111 000 111 000 111 000 00 11 11 00 00 11 00 11 111 000 000 111 000 111 000 111 111 000 111 000 111 000 11 00 00 11 00 11 00 11 11 00 00 11 00 11 00 11 111 000 111 000 111 000 000 111 000 111 111 000 000 111 111 000 111 000 111 000 11 00 11 00 11 00 111 000 000 111 000 111 11 00 11 00 11 00 00 11 00 11 11 00 00 11 00 11 00 11 11 00 00 11 111 000 111 000 111 000 111 000 111 000 111 000 00 11 00 11 11 00 00 11 00 11 11 00 00 11 111 000 000 111 000 111 111 000 111 000 111 000 00 11 00 11 11 00 00 11 11 00 11 00 11 00 11 00 11 00 11 00 11 00 00 11 11 00 00 11 11 00 00 11 00 11 000 111 000 111 111 000 000 111 11 00 11 00 11 00 11 00 11 00 11 00 111 000 111 000 111 000 111 000 111 000 111 000 111 000 000 111 000 111 111 000 000 111 11 00 00 11 00 11 11 00 00 11 00 11 000 111 000 111 111 000 000 111 00 11 11 00 00 11 00 11 000 111 111 000 000 111 000 111 00 11 00 11 11 00 00 11 11 00 11 00 11 00 11 00 00 11 11 00 00 11 Gas X 2 11 00 11 00 11 00 11 00 11 00 11 00 11 00 000 111 000 111 111 000 000 111 00 11 00 11 11 00 00 11 111 000 111 000 111 000 111 000 t = t 2 : statistical equilibrium is reached Figure 1.2 Linear diusion equation The (linear) diusion equation des ribes diusion type pro esses whi h is fundamental transport pro ess in physi s Diusion of gas mole ules ( on entration). and is onstant in the other dire tions. After removing the diaphragm at t = 0. We onsider the diusion of a substan e into itself. 1. A statisti al equilibrium has been rea hed and (on the ma ros opi level) no further diusion an be observed (right). We onsider a situation sket hed in Fig. After a ertain time t2 .10 where the on entration of a substan e varies in a ertain dire tion.

Then Fi k's law states that j = D d dx 15 . jE = Net number of parti les that diuse through a surfa e A in time t area(A) t where "net number" means the number of parti les rossing A in dire tion of the surfa e normal ~n of A minus the number of parti les rossing A in the dire tion of ~n.ne the parti le urrent density j as the net number of parti les that ross. a unit area pla ed perpendi ular to the dire tion of diusion. The parti le urrent density j is of dimension m2 s 1 . per unit time.

In the general situation.1 Theoreti al ba kground 000 111 000 111 000 111 00 11 00 11 11 00 00 11 00 11 00 11 00 11 00 11 11 00 00 11 00 11 00 11 00 11 11 00 00 11 000 111 111 000 000 111 11 00 00 11 00 11 000 111 111 000 000 111 11 00 00 11 00 11 00 11 11 00 00 11 00 11 11 00 00 11 00 11 000 111 00 111 11 000 11 00 111 00 000 11 000 111 111 000 000 111 000 111 000 111 00 11 11 00 00 11 00 11 00 11 00 11 11 00 00 11 111 000 000 111 000 111 11 00 11 00 11 00 000 111 111 000 000 11 111 00 00 11 00 11 00 11 00 11 00 11 00 11 00 11 00 11 00 11 00 11 00 11 00 11 00 11 00 11 000 111 000 111 000 111 000 111 000 111 000 111 000 111 11 00 111 000 000 111 000 111 000 111 11 00 111 000 00 11 000 111 11 00 111 000 00 11 000 111 00 11 000 111 00 11 000 111 00 11 00 11 000 111 00 11 000 111 000 111 00 11 000 111 000 111 000 111 11 00 000 111 000 111 000 111 11 00 000 111 000 111 11 00 000 111 000 111 00 11 00 11 00 11 00 11 11 00 00 111 11 00 11 000 111 11 00 000 00 11 000 111 11 00 000 111 000 111 000 111 000 111 000 11 111 00 111 000 11 00 000 111 11 00 111 000 11 00 000 111 11 00 111 000 11 00 000 111 000 111 000 111 00 11 000 111 000 111 00 11 000 111 00 11 00 111 11 000 111 000 00 11 000 111 00 11 000 111 00 11 000 111 000 111 000 111 00 11 000 111 000 111 000 111 00 11 000 111 000 111 000 111 00 11 000 111 000 111 000 111 000 111 Larger concentration 000 111 000 111 111 000 000 111 00 11 11 00 00 11 111 000 000 111 000 111 000 111 11 00 11 00 11 00 11 00 00 11 00 11 00 11 00 11 00 11 11 00 00 11 y x Smaller concentration Figure 1. where D is a oeÆ ient hara teristi of the substan e. The minus sign indi ates that the net ow due to diusion is in the dire tion in whi h de reases. Fi k's law states that the parti le urrent density is related to the on entration gradient by ~ ~j = Dr We onsider an arbitrary losed ontrol volume V with surfa e (boundary) V and outer normal ve tor ~n .10: Illustration of Fi k's law for mole ular diusion. alled the diusion oeÆ ient and d =dx is the on entration gradient in x-dire tion. Then in the absen e of an adve tion .

t) u( ax. at) . the stret hed fun tion p a = onst > 0 u~(x.eld the hange in the number of parti les in V is equal to the number of parti les diusing through the boundary of V via the parti le urrent density: Z Z Z d ~ ) ~nV d ~j ~nV d = ( Dr dV = dt V V V Appli ation of Gaussian Theorem gives Z Z d r~ Dr~ d dV = dt V V Assuming that .2 Solution of the diusion equation In this se tion we solve the one-dimensional diusion equation analyti ally ~2 u u = r t First we note that given a solution u(x.2. 16 (1.17).18) . is also a solution of (1.17) (1. t) of (1. D are suÆ iently smooth (at least in the sense of Sobolev fun tions) we obtain the dierential form of the diusion equation ~ Dr ~ = r t 1.17). sin e u~xx = auxx and u~t = aut .

x>0 . As H is invariant with respe t to stret hing. px : t z (x.2 Linear diusion equation A spe ial Cau hy problem We solve the Cau hy problem for (1. 0) = H (x) = 1 0 .18). t) = Then we have dg 1 dg z p = dz x dz t d2 g 1 dg 1 p = 2 Gxx = Gx = x x dz t dz t dg x dg z = Gt = dz x dz 2t3=2 Gx = From these relations we obtain the following ordinary dierential equation for g Gxx Gt = 0 2 x = 0 () ddzg2 1t + dg dz 2t3=2 d2 g dg x p =0 2 + dz dz 2 t p Now we substitute h = dg=dz and use z = x= t to obtain dh = dz d dh=dz = ln(h) = h dz =) ln(z ) = () z h 2 d z2 dz 4 ~2 z2 4 z 2 h(z ) = 2 e ( 2p ) 17 . t)) . t) = g(z (x.17) with initial ondition ( u(x.19) The fun tion H is often alled Heavyside fun tion. 1) t t t where we formally use a = t 1 in (1. t) = u( p . we make the ansatz x 1 1 u(x.1. t) = u( p x. Thus we seek G(x. x<0 (1.

The ru ial point is the integral Z x=p4t 2 e s ds I (x. First we .1 Theoreti al ba kground Integration of the last equation gives g(z ) resp. t) =) or Z z h(z ) dz = 1 + 2 0 Z x=pt z 2 G(x. Note that as t ! 0 we have x= t ! +1 (x > 0) and x= t ! 1 (x < 0) and Z 1 p p 2 e s ds = 1 + 2 x>0 : lim G(x. ) = 0 whi h depends on x.20) Now we want to visualize the solution (1. t) = 1 2 t!0 from whi h we infer that 1 = 1 . G(x. 2 2 = The solution G(x. t) = 1 + 2 2 t!0 | 0 {z } p = =2 x<0 : p lim G(x. t. Finally we determinepthe onstants 1 and 2 by p using the initial ondition. t) = + p 2 p1 2 R + then reads Z x=p4t 0 2 e s ds (1.20). t) = 1 + 2 2 g(z ) = 1 + 0 Z z 0 z 2 e ( 2p ) dz p p where in the last equation we substituted s = z=(2 ) and dz = 2 ds. t and the parameter . t) = 1 + 2 e ( 2p ) dz 0 Z x=p4t p 2 e s ds G(x. t) of the Cau hy problem on R 1 1 G(x.

x x > 0 (for the ase x < 0 thepargument is similar) and p . t) de reases due to diusion in the negative x-dire tion. t. re all that the gradient in u in the initial ondition auses a diusion of u in the negative x-dire tion. ) on x for . We onsider the dependen e on t. t. At t = 0 we have x= t = 1 and I (x. t. As a physi al intepretation. ) be omes smaller. As t in reases. u(x. Now we study the dependen e of I (x. the upper limit of the integral de reases and so the value of I (x. ) = =2. As t in reases.

If is de reased. 18 . Finally we study the ee t of the parameter on the diusion pro ess.xed t and . Thus I (x. t. ) be omes smaller as x ! 0. As a physi al interpretation. in the vi inity of x = 0 the gradient of u is maximal and hen e this region is ae ted largest by diusion. For small values of x the upper limit of the integral is smaller than for large values of x.

3 The steady-state linear adve tion-diusion rea tion problem then for .1.

For a larger value of . t the integral I (x. (for . ) be omes larger and the diusion pro ess is slower. t.xed x.

1. 1.11.12 we have a a relatively large diusivity and the initial dis ontinuity is smeared out rapidly. In 1. t0 ) ~u: onve tive velo ity . t) the integral I (x. t) at three dierent times t0 = 0. As an illustration. The initial dis ontinuity is smeared out slowly.12: History of u(x. Put in other words. Figures 1. Diusion means that the initial dis ontinuity at x = 0 is smeared out as time in reases.11: History of u(x. t0 ) T ~T + ~u r t with the following notation p r~ (r~ T ) = q_V in (0. a diusion type pro ess smoothens strong gradients/dis ontinuities in the initial data.x x. t1 and t2 . t) for three times t0 = 0 < t1 < t2 for large diusivity .11 we have a small diusivity .12 show snapshots of the solution u(x. t=0 u t = t1 u x t = t2 u x x Figure 1. In Fig. t. t) for three times t0 = 0 < t1 < t2 for small diusivity .3 The steady-state linear adve tion-diusion rea tion problem The linear diusion-rea tion type equation is of the form r~ ( r~ u) + ~b r~ u + u = f in (0. t=0 u t = t1 u x t=t 2 u x x Figure 1. ) be omes smaller and the diusion pro ess is faster. t0 ) A typi al example is the heat-transport equation. 1. whi h governs the spa e-time evolution of the temperature T in a spa e-time domain (0.

eld 19 .

But in a variety of situations of pra ti al relevan e this is approximately the ase. T = Tw D (0. in parti ular if the boundary is onne ted to an in. t0 ) on A Diri hlet boundary ondition for temperature states that the boundary temperature remains onstant and is not ae ted by the uid temperature. There are three types of boundary onditions Diri hlet boundary onditions. Of ourse the uid temperature does a t on the boundary temperature. N . and R . 0) = T0 (~x) Moreover we have to spe ify boundary onditions on . We assume a non-overlapping partitionning of into D .1 Theoreti al ba kground : density of the uid : heat ondu tion oeÆ ient q_V : volumetri heat sour e An initial ondition is imposed T (~x.

a heating system operating at.g. E. say Tw = 22Æ C or a old oor in the basement of a building without isolation may have the same temperature as the outside surrounding. ~ T ~n = q_ r on N (0. Neumann boundary onditions.nitely large energy reservoir.. t0) The right hand side term q_ is a surfa e-spe i.

For a perfe t insulator q_ = 0. 20 . t0 ) where Tw and T0 denote the wall-temperature and the free-stream uid temperature resp.e. whi h is a typi al quantity of engineering interest in indoor air ow simulation. i. heat ux per unit area.. and h is alled the onve tive heat-transfer oeÆ ient. Robin boundary onditions..g. given by Newton's ooling law ~ T ~n = h (Tw r T0 ) R (0. A Robin boundary ondition is e. heat ux.

.1) u = h on D (2.2 Fun tion-analyti al ba kground 2. r~ ( r~ u) + ~b r~ u + u = f in (2. h are smooth fun tions.2) is alled lassi al solution if u is in C 2 ( ) and satis. We assume that ~b.1)-(2.2) A solution u of (2.1 Classi al solutions We onsider the linear adve tion-diusion-rea tion problem with Diri hlet boundary ondition in a domain with smooth boundary .

In parti ular.es (2. C 2 -smooth solutions annot be expe ted in general for pra ti al problems.2 Sobolev spa es and weak solutions Solutions in the lassi al sense are very restri tive. De. In physi al experiments. the value of a fun tion at a ertain point ~x 2 is irrelevant. In a physi al sense. 2. A fundamental property of a solution is that it should be "bounded" at least in an integral sense. instead of pointwise values u(~x) only weighted averages an be measured.1) pointwise for ea h x 2 . whi h leads to the on ept of Lp -fun tions. u is also in C 1 ( ) and u = g on pointwise for ea h x 2 .

nition 1 We de.

ne by Lp ( the spa e of measurable and square-integrable fun tions Z p L ( ) = f u : ! R j ju(x)j2 d~x < 1 g Moreover on L2 ( ) we de.

p = For p = 1 the norm is given by Z ju(x)jp d~x 1=p jjujj0.ne the norm jj jj0.p by jjujj0.1 = ess: supx2 ju(x)j 21 .

whose p-th power annot not integrated. For d = 2. But for d = 3. the fun tion u(x) = ln(j~xj) is in L2 ( ). De. but u(x) = j~xj 1 is not in L2 ( ). Thus the regularity of a singularity also depends on the spa e dimension of . for any bounded domain .2 Fun tion-analyti al ba kground Thus in prin iple. u(x) = j~xj 1 does belong to L2 ( ). Example 1 Consider = f~x 2 R d : j~xj < 1g. a fun tion u does not belong to Lp ( ) only if it has a singularity.

Then we de.nition 2 Denote u : ! R .

ne its support supp(u) by supp(u) = f x 2 j u(x) 6= 0 g De.

nition 3 We denote by C01 ( ) the set of in.

We onsider the C01 (R )-fun tion ( 1 Ce 1 t2 0 !(x) = jtj < 1 jtj > 1 In R d we onsider the C01 (R d )-fun tion ( !(x) = 1 Ce 1 jj~xjj2 0 and we hoose C su h that Z jj~xjj < 1 jj~xjj > 1 Z !(~x) d~x = 1 j~xj1 Now we des ribe the averaging pro edure by Sobolev for smoothening Lp -fun tions. For u 2 Lp ( ) we de.nitely-times ontinuously dierentiable fun tions from to R for whi h supp(u) is a ompa t subset of . Example.

only the values ~z with jj~z ~xjj h ontribute to the integral. Then for ea h h > 0< the fun tion uh de. Now the following Theorem holds Theorem 1 Let u 2 Lp ( ) (1 p < 1) and let u be extended by zero in R d .3) h R For given ~x. Thus (2.3) an be interpreted as the average of u over the jj~x ~zjj h.ne Rd !(~x) d~x = Z d ~x ~z uh (x) = u(~z)! d~z (2.

ned by (2.3) is in.

uh 2 Lp ( ) and lim jju uh jj = 0 h!0 The averaging pro edure by Sobolev an be interpreted as a measurement of a physi al quantity: Ea h measurement devi e for determining a physi al quantity u does not return pointwise value at ~x but a spe i.nitesimal times dierentiable. Moreover.

ed sampling (averaged) value in the neighbourhood of ~x. Then ea h measurement may be des ribed by h ( ~x) where 22 . This average may be hara terised by its sampling fun tion .

2 Sobolev spa es and weak solutions : R d ! R is C 1 -smooth we de.2.

i = u(~x)(~x) d~x R We then say that u and v are equation in a weak-sense or integral-sense.ne h = (~x=h) and the support of h has radius h. From the point of physi al measurement. The statement that the measurement devi e is positionned at ~x orresponds to onsidering ( ~x) and therefore supp() f ~z : jj~x ~zjj < hg A parti ular example is !((~x ~z)=h) with sampling radius h. we say that u and v annot be distinguished if hu. Equation (2. The orresponding natural measure for the dieren e u v is then the L2 -norm Z jju vjj0. whi h an be interpreted as the sampling radius of the measurement instrument. i 8 2 C01 (R d ) with Z d hu. = ju(x) v(x)j2 d~x Then the on ept of a weak derivative of a fun tion is motivated by the rule of partial integration: For any smooth fun tion u 2 C 1 ( ).4) makes sense also for rough fun tions. whi h leads to the following De. i = hv. the rule of integration by parts be omes Z Z 8 2 C01 ( ) (2.4) u dx = ( 1) ( u) dx as and vanish on as 2 C 1 ( ).

: : : . and denote = (1 . d ) 2 N d a multi-index with jj = di=1 i . Note that L1lo ( ) is used to allow also for unbounded domains .nitionP4 Denote a domain. Now we are in a position to de. A fun tion v 2 L1lo ( ) is alled weak derivative of u 2 L1lo ( ) if Z u dx = ( 1) Z v dx 8 2 C01 ( ) We then all v the th-weak derivative of u and write = v.

ne the Sobolev spa e H m ( ) of fun tions whi h are m-times dierentiable in the weak sense. De.

nition 5 We de.

p ( ) by H m.ne the Sobolev spa e H m.p ( ) = f v 2 L2 ( ) j u 2 Lp ( ) 0 jj n g 23 .

1 ( ) is given by jjujjm. An example of a H 1 -fun tion for d = 2.p ( ) is given by 0 jjujjm. whi h are gently enough to be integrated in a ertain way. r~ u(x) = ln(j~xj 11 ) + 1 j~x~xj2 The dominator in the . For u(x) = ln(j~xj) ~ u(x) = j~xj 2~x. but we have r 1 u 2 H ( ) for d = 3. u does not belong to H 1 ( ).p = X 0jjm 1 p1 jj ujjp0. pie ewise smooth fun tions belong to H 1 ( ) The on ept of Lp -fun tions allows to onsider also fun tions with singularities.2 Fun tion-analyti al ba kground For 1 p < 1 the norm on H m.p = 0jmax jj ujj0. 3 is u(x) = ln(ln(j~xj 1 ) + 1) . Therefore for d = 2.1 jm Remark 4 The ontinuous.p A For p = 1 the norm on H m.

p ( ) onverges to an element in H m. Major advantage of the spa e H m.1 ( ) For p = 2 we an de. 3. Denote a bounded Lips hitz domain and denote C m.p ( ).1 ( ) = H m+1. The averaging pro edure by Sobolev demonstrated that C 1 ( ) is dense in H m. we need a fun tion to have H 2 ( )-regularity to be ontinuous (Relli h embedding theorem).p ( ) is that it is omplete: With respe t to the Lp -norm every Cau hy sequen e of elements in H m. Then C m.1 ( ) the set of Lips hitz ontinuous fun tions.p ( ) (for a C 0. For d = 2.0 smooth domain ).rst term goes very slowly to 1 as jxj ! 0 in a way that ~ u at x = 0 be omes integrable. the singularity of r Hen e we see that fun tions in H 1 ( ) are not ne essarily ontinuous.

2 ( ). Finally we have to take are regarding boundary onditions of a Sobolev fun tion in H m.2 ( ) are smooth (for d = 2.2 ( ) (u. v)H m.2( ) X Z 0m u v d~x and H m.2 ( ) be omes a Hilbert-spa e. For m 2 fun tions in H m.ne a s alar produ t on H m. 3) and values on 24 .

2 Sobolev spa es and weak solutions are well-de.2.

The question is if meaningful boundary onditions an be de. On the other hand. boundary onditions do not make sense for L2 -fun tions.ned.

C jju~jj1=2. As C 1 ( ) is dense in H 1 ( ). .2 ( ) H 1 ( ). for ea h u~ 2 H 1=2.ned for H 1 ( ) fun tions. the tra e operator is surje tive.2 ( ) there exists u 2 H 1. i. .2.2 ( ) su h that u = u~ and jjujj1. Then the so- alled tra e- operator : u 7! uj .2. Remark 5 In some de. We set H 1=2.2.2 ( ) su h that jj ujj1=2. u 2 H 1.2 ( ) where C is a onstant independent of u~. Moreover.e. C jjujj1..2. this an be a omplished as shown in the following. u 2 H 1. Theorem 2 Denote R d a bounded Lips hitz domain. u 2 C 1 ( ) has a unique and ontinuous extension from C 1 ( ) onto H 1.2 ( ) where C is a onstant independent of u.

y)j 0 Z 1 v 2 j (.nitions of the tra e operator C 1 ( ) is used instead of C 1 ( ) Proof First we onsider the spe ial ase = (0. y) v(0. for ea h two points (0. y) we have v(x. y)j j v 0 2 Z 1 v j (. y)j2 d + jv(x. Given v 2 C 1 ( ). 1)2 . Integration . y)d v (x. y)jd + jv(x. y) Z x 2 (. y) = whi h an be rearranged as Z x v(0. y)j2 0 Z x 0 v (x. y) and (x. y) = Therefore we have jv(0. y)jd + jv(x. y)d + v(x. y)j2 0 where we used the inequality (a + b)2 2a2 + 2b2 in the last step.

y)j2 d dy + 2 j v Z 1Z 1 0 0 jv(x.rst over y from 0 to 1 and then over x from 0 to 1 gives Z 1 0 jv(0. y)j2 dydx: 25 . y)jdy = 2 Z 1Z 1 0 0 (.

2 Fun tion-analyti al ba kground The left hand side integral is the L2 -norm of v along the boundary segment f0g (0. Putting together the ontributions from all four boundary segments we get jjvjj2L2 ( ) 8jjvjj2H 1 ( ) This argumentation an be modi. 1) and the right hand side term an be bounded from above by the full H 1 -norm of v.

we an transform the boundary lo ally to artesian oordinates. Using the tra e inequality. given v 2 H 1 ( ) we an de.ed to work also for polygonally bounded domains. For urvilinear boundaries.

The limit fun tion vj is then de. l ! 1.ne its tra e v = vj as a fun tion in L2 ( ). the tra es (vk j ) on is a Cau hy sequen e in the Lebesgue-spa e L2 ( ). k. l 2 N As the right hand side goes to zero for k. Given v 2 H 1 ( ) we onsider a sequen e (vk ) C 1 ( ) with jjvk vjjH 1 ( ) ! 0 as n ! 1. Then the tra e inequality implies that jjvk vl jjL2 ( ) ( )jjvk vl jjH 1( ) .

ned as tra e v of v 2 H 1 ( ).1)-(2. 2 Finally we give the on ept of so- alled strong solutions of the ellipti boundary value problem (2.2) r~ ( r~ u) + ~b r~ u + u = f in u = h on D For this purpose we interpret derivatives in the weak sense and boundary onditions are treated in the sense of the tra e operator De.

1)-(2.2 ( ) Then a fun tion u 2 H 2 ( ) is alled strong solution of (2.2).2).nition 6 We onsider the ellipti boundary value problem (2. if u satis. We assume the following regularity of the data . 2 L1 ( ) f 2 L2 ( ) h 2 H 1=2. bi 2 H 1.1 ( ).1)-(2.

and if u satis.1) in L2 ( )-sense.es (2.

r~ ( r~ u) + ~b r~ u + u = f in (2. D [ N = .6) ~ ru ~n = g on N (2.5) u = 0 on D (2. Neumann data are pres ribed. D \ N = .2) in L2 ( )-sense.7) 26 . 2.3 Variational formulation for ellipti problems We assume that on one part of the boundary D Diri hlet data are imposed and on the other N .es (2.

10) fv dx (2..3 Variational formulation for ellipti problems First we assume that the solution u of (2.9) (2. i.2..e.8) ~n d Then (2.5) with a so- alled test fun tion v 2 C 1 ( ) with v = 0 on D and integrate over the domain Z r~ ( r~ u)v dx + Z ~b r ~ uv dx + Z uv dx = We apply integration by parts to the Lapla ian term.5)-(2. i.5) holds pointwise for ea h x 2 . Then equation (2.e.11) Now the aim is to relax the regularity of the test.and ansatz fun tions: (1) If u 2 C01 ( ) satis. Z Z r~ ( r~ ) dx = ~r ~ dx r Z ~ r Z fv dx (2. Then we obtain Z ~ur ~ v dx + r Z ~b r ~ uv dx + For a moment assume that homogeneous Diri hlet data N Z uv dx = Z fv dx + u = 0 in on D with orresponding variational formulation ~ur ~ v dx + r gv d = .8) be omes Z ~ur ~ v dx r Z ~ uv ~n d + r Z Z ~b r ~ uv dx + uv dx = Z fv dx Now we onsider the boundary integral arising from integration by parts Z ~ uv ~n d = r = Z Z D ~ u v ~n d + r |{z} Z N =0 ~ u ~n v d | r {z } =g gv d This boundary integral does not depend on the unknown solution u and is therefore treated as a right hand side term.7) belongs to C 1 ( ). We multiply (2. we onsider the following problem with r~ ( r~ u) + ~b r~ u + u = f Z Z Z ~b r ~ u + u v dx = Z (2.

es equation (2. as C01 ( ) is a dense subspa e of H01 ( ).11) holds for all v 2 C01 ( ) then it will also hold for ea h v 2 H01 ( ). (2) We an then allow the solution u to belong to H01 ( ). 27 . as C01 ( ) is a dense subspa e of H01 ( ).

13) (2.12) is based on the framework of linear fun tional analysis. v) = f (v) = Z Z 8 v 2 H01 ( ) a(u.9)-(2.14) The theory for existen e and uniqueness of a solution of (2.10) reads: Find u 2 H01 ( ) su h that with a(u. v) = f (v) ~ur ~ v dx + r Z ~b r ~ uv dx + Z uv dx fv dx (2.12) (2.2 Fun tion-analyti al ba kground Then the variational form (or weak form) for the ellipti boundary value problem (2. We introdu e the on ept of a dualspa e V of a normed spa e V De.

nition 7 The dual spa e V of a spa e V with norm jj jjV is de.

e. ) is stri tly oer ive.. Ea h fun tion f an be asso iated one-to-one with a orresponding the linear fun tional (f.16) .jjvjjV =1 28 8 u. ) is bounded. v 2 V !R (2.jjvjjV =1 jf (v)j Remark 6 The dual spa e of H 1 ( ) is denoted by H 1 ( ). u 7! Z 2 L2 ( ) fu d~x: Existen e and uniqueness of the solution of the variational form of a ellipti boundary value problem are ensured by the famous Lax-Milgram Lemma. 9 > 0 su h that A(u. i. ) : H 1 ( ) ! R .15) (2.. v) : V V a bilinear form with the following properties A(. u) jjujj2V 8 u2V Then for ea h linear form f : V ! R whi h is bounded with jjf jjV sup jf (v)j v2V. Theorem 3 (Lax-Milgram) Denote V a real Hilbert-spa e and A(u. i. v)j C jjujjV jjvjjV A(.ned by V =f f : V with norm ! R j f is linear and bounded g jjf jjV sup v2V.e. 9 C > 0 su h that jA(u.

18) Some omments on the proof First we have to show that A(.20) Se ondly we have to show that A(. v) = 0 8v 2 V implies u=0 (2. we have to show that for ea h f 2 V there exists at least one u 2 V su h that A(u. v) = f (v) 8v 2 V . u) jjujj2V ) u=0 (2.19) This follows immediately from the stri t oer ivity of A as 0 = A(u.3 Variational formulation for ellipti problems there exists a uniquely determined solution u of the problem Find u 2 V su h that A(u. v) = f (v) 8v 2 V (2. A(u.e. v) is inje tiv..2. i.e. i.17) Moreover we have the following a-priori estimate (or stability estimate) jjujjV 1 jjf jjV (2. v) is surje tive. In a .

Consider un and um in V with (2.21) A(un . Then as A is ontinuous A(u. v) = nlim !1 A(un . as V is a Hilbert spa e: un ! u (n ! 1) with some u 2 V . un um )j jjun um jj2V Therefore (un ) is a Cau hy sequen e in V and onverges. v) = f (v) 8v 2 V g is losed. v) = nlim !1 fn (v) = f (v) 8v 2 V The . v) = A(nlim !1 un .22) A(um.rst step we show that the set Range(A) A(V ) f f 2 V j 9 u 2 V su h that A(u. v) = fn (v) 8v 2 V (2. v) = f (v) 8v 2 V m Then jjfn fm jjV jjun um jjV jA(un um . Let f be an element in A(V ) and let (fn ) be a sequen e in A(V ) with fn ! f (as n ! 1).

9)-(2. 2 Then existen e and uniqueness of a weak solution of the ellipti boundary value problem (2. we have that A(V ) is also omplete.. in parti ular. 29 .e. V is also a Hilbert spa e. The argument is now to onsider the proje tion operator from V to A(V ).10) an be proven by virtue of Theorem 3. it is omplete. A is surje tive. i. As ea h losed subset of a omplete subset is omplete. We have shown that A(V ) is losed.nal step is to show that Range(A) = V .

v) = 0 for ea h v 2 H01 ( )" has the unique solution u = 0. v)j jjujjH 1 jjvjjH 1 + (jjbjjL1 + jj 1 jj) jjujjH 1 jjvjjH 1 3 maxf jj jjL1 . f have the following regularity: .10). : : : . u) = (( r 2 30 . r ~ u)) + (( 1 r ~ ~b)u. bj .9)-(2. whi h show that a(. u) = (( r ~ u. : : : . bd ). 2 L1 ( ) (j = 1. v) = (( r f (v) = (f.2 Fun tion-analyti al ba kground Theorem 4 Assume that the domain R d is Lips hitz-bounded. u) a(u. v)j jjf (v)jj relation holds Z ~b r ~ u udx = = from whi h we infer Then we an obtain 2 Z Z Z r~ (~bu) udx + ~ ~b)u udx (r ~b r ~ u udx = Z Z Z ~b ~n d u| 2 {z } =0 ~b (r ~ u) udx ~ ~b)u2 dx (r ~ u. whi h is uniquely determined. r ~ v)) = ((r uv dx . d). ) is stri tly oer ive Note that the following useful ja(u. ~b = (b1 . u) + ( u. jj jjL1 g jjujjH 1 jjvjjH 1 jjf jjL2 jjvjjL2 jjf jjH 1 jjvjjH 1 Se ondly we have to show that a(. . v) Then we have the following estimates. jjbj jjL1 . Then for ea h f 2 H 1 ( ) there exists a solution u 2 H01 ( ) of (2. Proof First we introdu e the L2 -s alar produ t (u. ) and f () are bounded j(( r~ u. v) = Z ~ u. Additionally assume that at least one of the following two ondition holds r~ ~b 2 L1 ( ) and 21 r~ ~b 0 or: the homogeneous problem "Find u 2 H01 ( ) with a(u. Moreover assume that the data . r ~ u)) + (~b r ~ u. Z r~ u r~ v dx and write ~ u. r~ v))j + j(~b r~ u + u. r ~ v)) + (~b r ~ u + u. v) a(u.

we set V = f v 2 H 1 ( ) j v = 0 on D g Then the variational form of (2. i. Then under the assumtions of Theorem 5. 31 . v) = f (v) = Z fv dx + Z 8 v2V Z uv dx gv d Finally we summarize some results on the regularity of the solution. bj . 2 L1 ( ).e. Here we only onsider the ase of homogeneous Diri hlet boundary onditions. Then ~ ujj2L2 + jj 0 jjL1 jjujj2L2 a(u. we have regularity u 2 H ( ). and r k 1 k given f 2 H ( ).e. Moreover assume that 90 > 0 su h that 0 a. Assume that 9 0 > 0 su h that Re all that for in ompressible uid ows r 0 a. 2 H k 1.1 ( ).5)-(2. 21 ℄ if 2 C 0. and r~ ~b 2 L1 ( ) Moreover assume that the homogeneous problem "Find u su h that a(u. The regularity assumption 2 C 1. u 2 H 2 ( 0 ) for any ompa t set 0 .5)-(2. For non- onvex Lips hitz-domains this statement is wrong.3 Variational formulation for ellipti problems ~ ~b = 0. then for any and if 2 H k. The statement of the theorem an be "shifted" for k 1: If is a C k. u) jj0 jjL1 jjr minf jj0 jjL1 .7) reads Find u 2 H01 ( ) su h that a(u.1 ( ).2.1 ( ) bj . H 2 -regularity an be obtained even for onvex domains. Theorem 6 Denote a bounded and onvex domain.7) where N is non-empty. v) = f (v) where Z Z ~b r ~ uv dx + ~ur ~ v dx + r a(u.1-domain. Theorem 5 Denote a bounded C 1.e.1 is often too restri tive for problems of pra ti al relevan e: However. ~ ~b 2 H k 1. in .1 -domain and assume the following regularity of the data 2 H 1. in . Moreover we have the following results In the interior of . H 2 regularity of the solution an be obtained even for nonsmooth domains. But then it is still possible to get u 2 H 1+s ( ) with s 2 [0. v) = 0 for all v 2 H01 ( )" has the unique solution u = 0.1 ( ).t ( ) is Holder ontinuous. Then for any f 2 L2 ( ) the solution u is H 2 -regular. the solution u is in H 2 ( ). jj 0 jjL1 g jjujj2H 1 2 Remark 7 In the general ase (2.

) and a linear form f 2 V Find u 2 V su h that A(u.23) where V is an in. v) = f (v) 8v 2 V (2.2 Fun tion-analyti al ba kground 2.4 Galerkin method The Galerkin method is a proje tion method for solving the variational problem: Given a bounded and stri tly oer ive bilinear form A(.

De.nite-dimensional Hilbert spa e.

nition 8 Denote Vh V a .

v) = f (v) 8v 2 Vh (2. Find uh 2 Vh su h that A(uh.23) reads: Given a bounded and stri tly oer ive bilinear form A(. Then the Galerkin method for (2. v) = A(u. Re all that the best approximation u of an element x in a normed spa e X with respe t to a subset U X de. ) and a linear form f 2 V . ).nite-dimensional sub spa e of the Hilbert spa e V . v) = f (v) 8v 2 Vh we get the important Galerkin orthogonality relation A(uh u. v) = 0 8v 2 Vh whi h means that the error en = u uh is perpendi ular on the ansatz spa e Vh with respe t to the "s alar produ t" A(.24) Remark 8 From the fa t that the Galerkin solution uh and the ontinuous solution u satisfy A(uh.

ned by jjx ujj = vinf jjv xjj 2U satis.

i.23) and uh the solution of (2.) Denote u the solution of (2. ). 32 (2.24).es the orthogonality relation (u x. v) = 0 8v 2 U . The Galerkin proje tion immediately gives Cea's lemma. Theorem 7 (Lemma by Cea.e. x u ? U with respe t to (.. Then jjuh ujjV C vinf jjv ujj 2Vh with onstant C 1 depending on A but independent of Vh .25) . whi h shows that the error in the Galerkin method jju uh jjV is determined by how well the exa t solution u an be approximated by the subspa e Vh .

The . u uh ) = A(u uh .4 Galerkin method Proof. u v) C jju uh jjV jju vjjV 2 from whi h immediately follows the assertion. uh ) = A(u uh .2. From the Galerkin orthogonality we have A(u uh . v) = 0 Then we obtain the following estimate jju uh jj2V A(u uh .

fj = f (j ) Noti e that the Galerkin method is only a semidis rete method. j ) = (f. j ) = Pn i=1 ui i i=1 i=1 The Galerkin equations (2.24) are equivalent to the system of linear equations n X i=1 ui A(i .24) is equivalent to the solution of a system of linear equations. i ) . as an expli it evaluation of the integrals by using quadrature formulas is not spe i.nite dimensional Galerkin equations (2. We write for the dis rete solution uh of (2. j ) n X i=1 Aij ui = fj j = 1. Denote 1 . n (2. n where Aij = A(j . : : : . n a basis for Vh . then ea h uh 2 Vh an be written as a linear ombination v = n X i=1 vi i with oeÆ ients vi 2 R . j ) A( ui i . : : : . : : : .26) j = 1.24) uh = and obtain n n X X ui A(i .

33 . Remark 9 (Examples of global Galerkin ansatz spa es) If the spa e domain is arthesian = (0. y) = m X i. Ly ).j =0 ij xi yj g with h = 1=m. Lx ) (0.ed. then a tensor produ t ansatz an be made Vh = Qm ( ) = f p(x.

For i. In fundamental uid dynami s resear h.. Legendre polynomials or Ts hebys he polynomials) lead to a matrix Ah whi h is dense. lo al ansatz spa e) As mentioned above. Tensor produ t bases of L2 -orthogonal polynomials (e. that the Lapla e problem d2 u = f in[0.. This approa h leads to so- all Spe tralGalerkin methods: On geometri ally simple domains like re tangular box or ylindri domains this leads to very fast methods. x2 . this is a very popular method for solving the nonlinear Navier-Stokes equations for turbulent ows in simple geometries. xy.e. j = xj . j 1 we obtain Aij = Z 1 0 dxi dxj dx = dx dx Z 1 0 xi 1 xj 1 dx = 1 dx i+j 1 similar to the Hilbert matrix Aij = (i + j + 1) 1 . 1℄ . whi h are periodi in at least one dire tion. y. turbulent ow in a hannel/pipe or thermal onve tion problem in a re tangular box/ ylinder. : : : then gives that the Matrix Ah is similar to the Hilbert matrix with ondition number N = (m + 1)2 growing as ond2 (Ah ) = O(eN ). i. Remark 10 (Global vs. x. dx2 u(0) = u(1) = 0 is also ill- onditioned. the problem of approximating ontinuous fun tions by global polynomials is ill- onditioned. The Lemma by Cea (Lemma 7) shows that the Galerkin dis retisation is losely related by the approximation of the exa t solution by a . y. but with a very suitable ondition ond2 (Ah ) = O(N ). Very similarly it an be seen.g.2 Fun tion-analyti al ba kground Using the monomial basis 1. We set i = xi .

This is the underlying idea of the Finite-element method. 34 . the on ept of spline interpolation uses a pie ewise polynomial interpolation of low degree. In interpolation theory. Instead. it is well known that using global polynomials with in reasing degree does not always lead to an improvement in the approximation.nite-dimensional subspa e.

for all T 6= T 0 { The subs ript h denotes an appropriate s ale parameter regarding T . then an approximation for is used.e.1 FEM for adve tion-diusion problems Finite-element method for the adve tion-diusion-rea tion problem with homogeneous Diri hlet boundary onditions are pres ribed. i.. T \ T 0 is either a ommon point or a ommon entire fa e. { For two adja ent ells T and T 0 . vh ) = f (vh ) Z ~b r ~ uv dx + Z 8 vh 2 Vh uv dx fv dx Dis retisation of the boundary .e. r~ ( r~ u) + ~b r~ u + u = f u = 0 in on (3. whi h satisfy the following properties { Ea h T is of simple geometry (i. the (maximal or minimal) ell diameter. Put in other words. A triangulation of is a set of ells T . { Polygonal approximation of (linear splines) { Spline interpolation of (pie ewise higher order polynomials) Triangulation of . Th = f T j T g of . hanging nodes are not allowed.e.. i. 35 . If is urvilinear bounded. v) = f (v) = Z Z ~ur ~ v dx + r a(uh .1) The orresponding Galerkin method reads: Given Vh V Find uh 2 Vh with uh = 0 on su h that with a(u. or a tetrahedron or a hexaeder in 3D { = [T 2Th T { T \ T 0 = . a triangle or a quadrilateral in 2D..3 Finite Element method 3.

3 Finite Element method De.

ne the lo al degrees of freedom. On ea h ell T 2 Th we de.

For ea h xilo the orresponding nodal fun tional ilo is de.ne { a lo al basis of test. x 2 T . ilo ilo =1 The set of lo al degrees of freedom is the dual spa e of the set of lo al testand ansatz fun tions.and ansatz fun tions BT = f Nilo j Nilo : T ! Nlo :dof R gilo =1 and { a set of nodal fun tionals f ilo gNilo lo :dof =1 asso iated with the lo al degrees lo :dof of freedom f xilo gN .

lo =1 u(T.ilo ) (x) The ansatz fun tions are de. i. ilo : Njlo 7! ilo (Njlo ) = Njlo (xilo ) T We require that the basis is so- alled unisolvent..e. In the sequel we assume ( 1 0 i (Nj ) = Nj (xi ) = if if i=j i 6= j Then we make the following ansatz for the lo al solution (restri ted to T ) u(x)jT = Nlo :dof X i.ilo ) N(T. ea h Ni 2 BT is uniquely determined by the pres ribed values i (Nj ) = Nj (xi ).ned by ilo 2 B .

y0 )T . In 2D the referen e triangle is Tb = f (. N3 (. 0 . 1g: Given a triangle T with orner points (x0 . N2 (. y2 )T we onsider the following aÆne linear mapping FT : Tb !T . ) = 36 x0 y0 . FT ! FT = B + x0 y0 . the basis fun tions are N1 (. FT 10 = xy 1 1 . ) = (1 )(1 ) . ) j . (x2 .ned on the referen e element. B = x1 x0 x2 y1 y0 y2 whi h maps the orner points of Tb to the orner points of T FT 00 = xy 0 0 . On the referen e triangle Tb. FT 01 = xy 2 2 Linear ansatz. 0 . (x1 . ) = . y1 )T .

N2 (.3. ) = (1 )(1 2 2) = ( 1 if = = 0 0 if + = 1 or + = 1 2 1 if = 1 . = 0 0 if = 0 or = 1 1 if = 21 . ) = (2 1) . N4 (. ) = 4(1 ) = 1 if = 21 . ) = 4 N6 (.1 FEM for adve tion-diusion problems Quadrati ansatz. = 0 0 if = 0 or = 1 De. N1 (. ) = 4 = 1 2 1 if = 21 . = 1 0 if = 0 or = 21 N2 (. 0 = 4(1 ) The ansatz fun tions an be plotted easily by taking into a ount that ( N1 (. ) = (1 )(1 2 2) . ) = 4 (1 ) = ( N5 (. ) = (2 1) = ( 1 if = 1 . ) = 4 (1 ) N5 (. = 1 0 if = 0 or = N3 (. ) = (2 1) . = 21 0 if = 0 or = 1 ( N5 (. ) = (2 1) = ( N4 (. N3 (.

nition of global degrees of freedom. The aim is to de.

glob i.glob ui.ne global degrees of freedom su h that the approximative solution u an be written globally as u(x) = Nglob X:dof i.glob For this purpose we de.

ne De.

o.f. : : : . Nglob:dof g .o. ilo ) = iglob Moreover we have a mapping from the global degrees of freedom to the lo al degrees of freedom. we have a list of asso iated ells T (iglob) and a 37 . For ea h global degree of freedom iglob with nodal fun tion i.glob ).f. Nlo :dof g ! 7 f 1. has a ontribution from several lo al d. : : : . (T.glob (N ) = N (xi.s. ilo ) 7! lo !glob (T. : : : . Nelmt g f 1.ne a mapping from the lo al degrees of freedom to the global degrees of freedom lo !glob : f 1. in general ea h global d. However.

3 Finite Element method Table 3. list of asso iated lo al degrees of freedom L(iglob) T (iglob) = L(iglob) = f T 2 Th j xi. Nlo :dof number of global degrees of freedom global degree of freedom iglob with orresponding nodal fun tional iglob = (xiglob ) for some test fun tion . ilo in ell T g Ea h lo al ansatz fun tion on ell T is extended by zero to a fun tion on . Nelmt g f 1.o. : : : .f. ilo ) we de. : : : . asso iated with the ell id T 2 f 1.glob 2 T g f (T. ilo ) 2 f 1. Nlo :dof g : xilo = xiglob is lo al d. For lo al d. (T. : : : . : : : .f. Nelmt number of lo al degrees of freedom per ell (whi h is assumed to be onstant) lo al degree of freedom ilo 2 f 1.1: Notation lo al and global degrees of freedom Symbol Nelmt T Nlo :dof ilo Nglob:dof iglob Meaning number of ells (elements) ell T 2 Th .o.

ilo ) (x) = N(T.ne ( (T.ilo ) (x) 0 : : x2T else For ea h global degree of freedom iglob de.

j ) ) on ea h ell T .j ) ) f ((T.j ) ) Thus the remaining problem is to assemble a((T. see (2. vjglob ) = f (jglob ) = X T 2T X T 2T a((T.ilo ) (x) .ne the orresponding ansatz fun tion iglob (x) = (T.glob Assembling of the linear system.j ) ). with ilo given by xilo = xiglob Then we an make the following global ansatz for the solution u u(x) = Nglob X:dof i. (T. Then these lo al ontributions are added to the orresponding line/ olumn of the global stiness matrix and right hand side ve tor. f ((T.glob i.glob=1 ui.i) .i) . a(iglob . 38 .26). (T.

j++) { A_lo [j. i< N_lo _dof. j < N_lo _dof. j < N_lo _dof. ve tor ---------------------------------------------------*/ double A_lo [N_lo _dof℄[N_lo _dof℄. /*---------------------------------------------------| Loop over all elements ----------------------------------------------------*/ for(elmt = 0.s.s.h.1 FEM for adve tion-diusion problems /*--------------------------------------------------| Global stiffness matrix and r.h.s. i< N_elmt. ve tor ---------------------------------------------------*/ double A_glob[N_glob_dof℄[N_glob_dof℄.h. i< N_lo _dof. j++) 39 . for(j = 0. i++) { iglob = global_dof[elmt. set_entries_zero(f_glob). double f_glob[N_glob_dof℄. ve tor ----------------------------------------------------*/ for(i = 0. /*--------------------------------------------------| Lo al stiffness matrix and r.h.3. i++) { for(j = 0. } /*---------------------------------------------------| Add entries of lo al stiffness matrix/ r.i℄ = assemble_a(i. ve tor | to global stiffness matrix/ r.i℄.s.j). set_entries_zero(A_glob). elmt) { /*---------------------------------------------------| On the urrent ell | ompute lo al entries of stiffness matrix | and right-hand side ve tor ----------------------------------------------------*/ for(i = 0. } f_lo [i℄ = assemble_f(i). double f_lo [N_lo _dof℄.

i< N_elmt. }/**for(elmt = 0.j℄. } f_glob[iglob℄ = f_lo [i℄.2 Implementation of a .3 Finite Element method { } jglob = global_dof[elmt. elmt)**/ VERGLEICH QUADRAT element zu tetra el 3. A_glob[jglob℄[iglob℄ = A_lo [j℄[i℄.

nite element method 40 .

e. F v is related to the strain rate tensor T (~ u ) by Fv = 2T (~u ) where 1~ 1 ~ ~ ~u)T r ~u + (r T (~ u ) S(~u) r ~u I . ~u is the uid velo ity. i. We assume a onstant-property Newtonian uid. Vis osity is a diusion type transport me hanism for momentum. the prin iple of onservation of momentum for an arbitrary losed ontrol volume V states that Time rate of hange of momentum inside V = Flux of momentum through V due to onve tion + Flux of momentum through V due to vis ous for es + Flux of momentum through V due to pressure for es Z Z Z Z Fv~ n d p~n d (~u)~u ~n d ~u d~x = t V V V V where F v denotes the vis ous stress tensor.1 Governing equations for vis ous uid ow In vis ous uids. with S(~u) = 3 2 The onstant proportionality fa tor is the oeÆ ient of vis osity of the uid and is of dimension m 1 kgs 1 The kinemati vis osity is de. the density and p is the pressure.4 Turbulent boundary layer ows 4.

1) t Here we use the notation j (~u ~u)ij = ui uj . r~ ~u = u xi 41 . By appli ation of Gaussian Theorem for integration by parts the momentum equation be omes Z Z Z ~ (~u ~u) d~x + r~ p d~x = 0 r ~u d~x + t V V V The momentum equation an be written in dierential form as follows ~ (~u ~u) + r ~p r ~ (2T (~u )) = 0 ~u + r (4.ned by = =. For air under standard onditions = 1:5 5 m2 s 1 .

Then the momentum equation reads (in a omponentwise notation) d d X X p~ u~i u~j ~u~ + (~u~i u~j ) + x~ ~ + x ~ x ~ x~j x~i t~ i i j =1 j j =1 j We introdu e the following non-dimensional quantities u~ ~ x~ ui = i . whereas for large velo ity.2) (4. they are ontinuously rossing plane P whi h is atta hed to a frame of referen e moving with mean velo ity ~u. the non-linear hara ter of the equations dominates. As ea h mole ule arries a momentum of the mean uid motion in x-dire tion.2 Reynolds number. Lref Uref ref 42 = 0 (4. the ow solution behaves almost linear. Laminar and turbulent ow The momentum equations are non-linear. 4. Thus it an be expe ted that for small uid velo ity. For this purpose.4 Turbulent boundary layer ows We now onsider the in ompressible Navier-Stokes equations. The Reynolds number appears in a natural way when transforming the Navier-Stokes equations from dimensional to non-dimensional variables. The transition between both limiting ases is des ribed physi ally by the so- alled Reynolds number.. Mean ow is in x-dire tion with velo ity varying in y-dire tion aused by the no-slip ondition at the wall ~u = 0. = 1 xi = i .e.3) . denote dimensional quantities by a tilde.1: Illustration of vis osity. Due to the thermal statisti al motion of the mole ules. the equation of onservation of mass and momentum for an in ompressible uid r~ ~u = 0 ~ (~u ~u) + r ~p (~u) + r t r~ (2S(~u)) = 0 Rapid flow u(y) u(y) moving frame of reference Rapid flow net momentum transfer y y Slow flow x no−slip on the wall Flux of momentum through area A due to viscosity x Slow flow Figure 4. i. this results in a momentum ow a ross plane P .

2) an be written by substitution of (4.3) and by using the hain rule of dierentiation d X t x p~ xj 2 ref Uref ~ ui + ref Uref (ui uj ) + i x ~ x x~i xi t t j =1 j j d X xj ui xi uj xj ~ + Uref x ~ x x~j xj x~i xi j =1 j j = 0 The non-dimensionalisation of t follows from the de. Laminar and turbulent ow Then (4.2 Reynolds number.4.

Thus the other quantities are then non-dimensionalised a ording to t = t~Uref . ref Uref = ~ ~ref Uref xref By substitution of x= x~ we obtain d 2 ref Uref U2 X 1 p~ ui + ref ref (ui uj ) + xref t xref j =1 xj xref xi d ui uj Uref X = 0 ~ + x2ref j =1 xj xj xi Now we substitute p~ and ~ and obtain d 2 ref Uref U2 X U 2 p ui + ref ref (ui uj ) + ref ref xref t xref j =1 xj xref xi d ref Uref xref X ui uj = 0 + x2ref xj xi j =1 xj 2 x 1 and obtain the NavierWe then divide the equation by the ommon fa tor ref Uref ref Stokes equations in non-dimensional form d X p ui + (ui uj ) + t x x i j =1 j The parameter 1 d X ui uj + x x xi j j =1 j = 0 (4.4) is alled Reynolds number Re and is identi. xref p = p~ 2 .nition of x and u and the nondimensional variables for p and arise naturally from the Navier-Stokes equations.

5) The de.ed as Re = 1 ~u~ x~ = ref ref ~ (4.

nition of the Reynolds number relies on two referen e values. a referen e velo ity and a referen e length. 43 ..e. i.

Left: The hord length of an airfoil used as referen e length for (global) Reynolds number Re. Right: The distan e from the leading edge of a at plate L = x determines the lo al ow onditions and is therefore used for de.2: Choi e of hara teristi length for Reynolds number.4 Turbulent boundary layer ows Uoo Uoo L L Figure 4.

ning a lo al Reynolds number Rex . Referen e velo ity. { The referen e velo ity hara terises the global ow .

For external ows. the typi al hoi e is the so- alled far-.eld.

whi h is the velo ity at the far.eld velo ity u1 .

{ The referen e length for the global Reynolds number is the diameter of the on.eld boundary remote from the obsta le. { In order to hara terise the lo al ow behaviour. Referen e length. a lo al Reynolds number based on a lo al uid velo ity and a lo al length s ale might be useful in some ases.

These are ows with one domi44 . however. of fundamental importan e in aerodynami s: Boundary layer ows. for a ow over a at plate. the transition from laminar to turbulent ow is determined by the lo al Reynolds number. whi h allow to predi t the ow behaviour of large air rafts by performing measurements at small models of same geometry in a wind-tunnel. or the diameter of a hannel or pipe. { In some ases.2. provided the Reynolds number of the two ows are the equal.e. i. 4. this is the underlying prin iple of wind tunnels. the hord length of an airfoil. For the ow around an airfoil. E. the ow state at a ertain position is determined by its distan e in streamwise dire tion from the origin of the plate.1 Flat plate boundary layer ow In this le ture we fo us on a parti ular type of ow. whi h is omputed from the distan e to the stagnation point.g.guration.. The non-dimensional form of the Navier-Stokes equations also gives rise to the existan e of similarity solutions: Two ow solutions around the two obje ts of same geometry but at dierent sizes are the same. a lo al Reynolds number is used to determine the lo al ow state. whi h is.. the diameter of a sphere or ylinder.

4. the wall-normal dire tion is the y-dire tion (in a wall-. onveniently. whi h are bounded (at least in part) by one or more solid surfa es. Laminar and turbulent ow nant ow dire tion alled the stream-wise dire tion (by onvention: the x-dire tion).2 Reynolds number.

y = 0 g We onsider the ow in the bounded domain ( K. B ). w = Assume that the in ow boundary whi h is lo ated at x = K is at suÆ iently large distan e from the plate su h that we an assume undisturbed time-independent on ow onditions in ~ex -dire tio (so- alled far-. 00 < y < C . L 0. = f ~x 2 R 3 j K < x < L . We assume a at plate of length L denoted by W ("W" stands for wall) and lo ated in the half plane f ~x 2 R 3 j 0 x L . Su h ows may be internal su h as the ow through pipes and du ts. whi h shows a variaty of important features of the full Navier-Stokes equations. L) (0. C ) ( B. i.tted lo al oordinate system for urvilinear surfa es). Here we onsider the ow over a at plate. B < z < Bg with K. We assume periodi boundary onditions at z = B and z = B .e. or they may be external su h as the ow around air raft and ships' hulls..

6) r~ (2T (~u )) = 0 in (0.3: Sket h of ow over at plate in an (ideal) in.eld boundary onditions for x ! 1). T ) (4.9) in f0g (4.8) on W (0. At the wall. T ) (4. Moreover we use the initial ondition ~u = 0.10) For the dis ussion of this hapter. as the uid mole ules are adherent to the wall.7) ~u = U1~ex ~u = 0 ~u = 0 on (0. ~ (~u ~u) + r ~p (~u) + r t r~ ~u = 0 in (0. it is not ne essary to des ribe the boundary onditions at the other boundaries. the uid velo ity is zero (so- alled no-slip). T ) (4. T ) (4. Γ4 Uoo Γ3 Γ1 y y z z x x Γ2 lam ΓW turb ΓW Figure 4.

45 .nite domain (left) and in a bounded domain used for a numeri al simulation (right).

4 Turbulent boundary layer ows 4.3 Laminar ow The time tl required for the uid to travel at speed U1 over the distan e L (re all that L is the length of the plate) is L (4. at a .11) U1 Hen e.

t) = u(x. as the on.12) Noti e that the steady-state solution does not depend on the z- oordinate. it sounds reasonable that there exists a time t0 tl su h that a steady state solution is obtained whi h is of the form tl = ~u(~x. y)~ex t t0 (4.rst glan e.

then for ea h on ow velo ity U1 there exists a steady state solution whi h is stable and. t1 +dt). However. Of parti ular interest are so- alled similarity solutions. Consider the steady state solution u(x.guration is homogeneous in spanwise dire tion. is not too large. the same steady state solution as before the disturban e is re overed. moreover. su h steady state solutions are stable. the Reynolds number.e. if Re is small. U1 = U1 (t) or for a moving boundary W = W (t) we annot expe t to obtain a steady-state solution.. Remark 11 Of ourse.e. y) at two dierent streamwise positions x1 and x2 .. i. Then for small disturban es in the on ow onditions or in the shape of the geometry during a ertain time interval (t1 . Then pro. i. To be more pre ise. are damped out and for t t2 > t1 + dt. moreover. for time-dependent boundary onditions. whi h lead temporarily to small disturban es in the solution. a global attra tor. steady-state solutions an be observed only if the riti al parameter. Su h steady state solutions are referred to as steady laminar ow.

The steady state solution is then no longer stable and it be omes a repellor.4 Turbulent ow. then there is a riti al Reynolds number at whi h a bifur ation o urs. The attra tor is then a so- alled turbulent solution whi h is unsteady three-dimensional 46 .les at dierent streamwise positions x ollaps when they are plotted using the similarity transformation r u(x1 . y=g(x1 )) u(x2 . If the Reynolds number in in reased. g(x) = U1 U1 U1 independent of the Reynolds number. y=g(x2 )) x = . 4.

neither periodi nor quasi-periodi irregular.4.4 Turbulent ow. random-like. Here a quasi-periodi solution means that it is a superposition of a .

Remark 12 Note that ow an be laminar. Now we want to des ribed the unsteadyness of turbulent solutions in the framework of dynami al systems: the time-evolution of the velo ity-. This is the ase if no disturban es are present (neither in the geometry nor in the on ow onditions).nite number of periodi solutions with mutually dierent frequen ies. for a re-entry spa e vehi le (with a very smooth surfa e) in the outer part of the atmosphere (where the uid is almost perfe tly at rest). even if the Reynolds number is larger than the riti al Reynolds number. e.g.

3 . t)j2 d~x 2 De. : : : . For in ompressible ow. at ea h time t the solution belongs to ~ ~u = 0 g H1div ( ) = f ~u : ! R d j ui 2 H 1 ( ) i = 1. r We remark that the total kineti energy of the uid in the entire domain is given by jju(t)jjengy = Z j~u(~x.eld is interpreted as the orbit of the time-evolution operator of the initial-boundary value problem of the Navier-Stokes equations whi h wanders on random-like ( haoti ) urves on a fra tal(?) manifold in the spa e of admissible solutions.

fra tal (?) stru ture.nition 9 The time-evolution operator Gt is the mapping Gt : H1div ( ) 7! H1div ( ) . it des ribes a random-like urve whi h lies dense in a ertain subset of H1div.e0 ( ) = f ~u 2 H1div ( ) . ~u0 () 7! ~u(.e0 ( ) of ompli ated. we an hara terise the dierent types of solutions as follows. A turbulent solution is no longer a single losed urve of regular shape. t)jjengy = e0 g Using the framework of dynami al systems.6)-(4. there exists time t0 and a total energy e0 su h that f Gt~u0 gtt0 H1div. the total energy of the uid remains onstant. Instead. Hen e. t) where ~u(.10) for the initial ondition ~u(. jj~u(. After a transient initial period. 47 . whi h is a single losed urve of regular shape. t) is the solution of (4. A steady-state solution ~ust orresponds to f Gt~u0 gtt0 = f ~ust g A periodi or quasi-periodi solution orresponds to a losed orbit f Gt~u0 gtt0 . t) = ~u0 ().

As Re is in reased. At a ertain Reynolds number.5 Statisti al approa h For turbulent-boundary layer ows.4 Turbulent boundary layer ows Remark 13 A prominent example are the Reynolds number ee ts for the ow behind a ir ular ylinder. the stationary solution is a globally stable attra tor. the three stages of in reasing ow omplexity an be observed. albeit the solution looks haoti and irregular. For the ow around a ylinder. as Reynolds number is in reased. The solution be omes unsteady. it turns out that there are fundamental laws for the statisti s (like mean value and varian es) of the ow . 4. As Re is in reased. these frequen ies an no longer be distinguished and the ow be omes turbulent and appears fully random. the steady state solution be omes a repellor. For small Re. often alled van-Karman vortex street. but periodi . the solution be omes superposition of more and more frequen ies.

De.eld.

nition 10 The ensemble averaging .

lter over an ensemble of N ow experiments is de.

ned by h ~u iN = N1 ~u(i) N X i=1 ~u(i) (~x. The time averaging . t) Therein. denote the solution of the ith ow experiment where the in ow boundary onditions U1 and the shape of the boundary W are disturbed by small hanges whi h are random with a probability fun tion whi h is independent and identi ally distributed.

lter h i(t.t+t) over the interval (t. t + t) is de.

t) dt h ~u i(t.ned by Z t+t ~u(~x.t+t) = 1t t The spa e averaging .

or to run a numeri al simulation on N workstations with randomly perturbed in ow onditions. De. y.z+z) = 1z Z z +z z ~u(x. z. On the other hand. t) dz Remark 14 The ensemble averaging on ept is only of theoreti al interest: It would require to setup a large number of N very similar ow experiments.lter h i(z. In the following we will see that all three averaging on epts give the same average provided the ow is statisti ally onverged.z+z) in the homogeneous (spanwise) z- oordinate dire tion over the interval (z. the on ept of time-averaging and spa e-averaging (over homogeneous ow dire tions) allows to onsider only one ow experiment (or simulation). z + z ) is given by h ~u i(z.

t+t) . lim z !1 h ~u i(z.z+z) .nition 11 A at plate turbulent boundary layer ow is alled statisti ally steady (statisti ally onverged) if the following limit values exist lim h ~u iN . N !1 48 lim t!1 h ~u i(t.

6 Universal solutions After a ertain time (in the sense of (4. whi h is therefore two-dimensional and steady-state. then the ow be omes statisti ally steady: The ergodi theorem now states that averaging over time or over the homogeneous oordinate dire tion give the same solution.z+z) = u(x.4. y) z!1 As both time-averaging and spa e-averaging (over homogeneous dire tion) are the same. y)~ex ~u(x.t+t) = lim h ~u i(z. Then lim h ~u iN = N !1 lim t!1 h ~u i(t. we may de.11)) and if the spanwise extent of the domain is large enough. Proposition 2 (Ergodi theorem) Consider a fully developed turbulent boundary layer ow. whi h is homogeneous in (spanwise) z- oordinate dire tion.

13) !1 t!1 z!1 De.ne the statisti al averaging operator h i by ~u h ~u i = Nlim h ~u iN = lim h ~u i(t.z+z) (4.t+t) = lim h ~u i(z.

nition 12 The velo ity .

t) in a turbulent boundary layer ow an be de omposed ~u(~x. alled the turbulent u tuating part and denoted by ~u0 ~u0 (~x. t) ~u(x.14) h ~u i (4. y) = (4. t) = ~u(x.15) and into an unsteady random part with zero mean. y) + ~u0 (~x. ~u0 = 0 (4. y) .eld ~u(~x. t) = ~u(~x.16) The ensemble averaging . t) into a mean part ~u whi h is time-independent ~u(x.

lter has the important properties that it is linear and that .

Proposition 3 The ensemble averaging hiN .ltering and dierentiation ommute.

t: Proje tion identity: Non- ommutativity w.r. i.t.t.t.6 Universal solutions The question arises if for the mean ow u(x.e. xj : Commutativity with dierentiation w. 2 R h x iN = x h iN j j h i = t h iN t N h h iN iN = h iN h h iN iN 6= h iN h iN 4.r.r. if solutions at dierent streamwise positions and at dierent Reynolds numbers ollaps 49 .lter has the following properties Linearity (i): Linearity (ii): Commutativity with dierentiation w.. y) a similarity solution exists. nonlinear expressions: h + iN = h iN + h iN h iN = h iN .

provided the ow is fully developed turbulent. the ow is alled fully developed turbulent. We de. After a ertain distan e x from the leading edge of the plate (where the lo al Reynolds number Rex = U1 x= is large enough). Now the surprising answer is: Yes. y99 ) = 0:99U1 whi h is alled the boundary-layer thi kness.4 Turbulent boundary layer ows if plotted in suitably s aled oordinates. Denote y99 = y99 (x) the elevation where u(x.

as it is the for e exerted by the uid on the wall (or vi e versa by Newton's law a tio equals re-a tio). u = r w .e.7 Statisti al theory and modelling A ording to Hadamard (1923) problems in mathemati al physi s Hadamard (1923) are alled well-posed if the following three requirements are satis. In 1925 Prandtl found that in the near-wall region y / 0:2y99 the solution is universal (i. / 40. independent of streamwise position x and Reynolds number) and given by ( y+ : y+ / 5 + + u (u ) = 1 (4. w = u j y y=0 where w is alled wall-shear stress.17) + : y+ ' 40 ln(y ) + C with onstant = 0:41. In the intermediate range 5 / y+ no simple analyti formula an be given. 4. The quantity u has dimension of velo ity and is alled fri tion velo ity.ne so- alled universal oordinates ~u+ = u . and C = 5:1. u y+ = yu ..

but there are many important situations where the statisti ally averaged mean- ow is well-posed. the unsteady ow solution ~u(~x. The statisti ally averaged Navier-Stokes equations are often alled Reynolds averaged Navier-Stokes equations.ed: there exists a solution the solution is uniquely determined the solution depends ontinuously on the data In turbulent ows.1 The Reynolds averaged Navier-Stokes equations We apply the ensemble averaging . 4.7. t) is not well-posed.

lter hi to the Navier-Stokes equations. As .

ltering is a linear operation we get h t ~ui + hr~ (~u ~u)i + 1 hr~ pi 50 hr~ (2 S(~u))i = 0 .

7 Statisti al theory and modelling For the ensemble averaging operator is an be seen immediately. that .4.

hpi = P However. the non-linear term gives rise to a problem h~u ~ui 6= h~ui h~ui At least.ltering and dierentiation are ommutative operations (4. the non-linear term an be written in the form h~u ~ui = h(U~ + ~u0 ) (U~ + ~u0 )i = hU~ U~ i + 2hU~ ~u0 i + h~u0 ~u0 i = hU~ U~ i + 2U~ h~u0 i + h~u0 ~u0 i |{z} =0 = hU~ U~ i + h~u0 ~u0 i where in the last line we used the fa t that U~ is no longer random.18) h~ui + r~ h~u ~ui + 1 r~ hpi r~ (2 S(h~ui)) = 0 t We then substitute the mean (average) velo ity and pressure h~ui = U~ . We .

20) Remark 15 Equation (4. 2.. U~ =t = 0.nally substitute the relation h~u ~ui = hU~ U~ i + h~u0 ~u0 i into (4. 2.20) is the governing equation for the mean ow quantities U~ . Ui (i = 1. However. 3). Hen e the mean- ow equations are un losed. these are 4 equations with 10 unknowns P . The so- alled Reynolds averaged Navier-Stokes equations be ome r~ U~ = 0 (4.18) and obtain ~ ~ (U~ U~ ) + r ~ h~u0 u~0 i + 1 r ~P ~ 2 S(U~ ) = 0 U + r r t We already know that the mean solution for the at plate turbulent boundary layer is steady state. This extra term r 51 . ~ h~u0 u~0 i is alled the Reynolds stress tensor. i. but one-extra term in luded. The Reynolds averaged Navier-Stokes equations have the same form as the original Navier-Stokes equations. P . 3) and hu0i u0j i (i.e.19) r~ (U~ U~ ) + r~ h~u0 u~0 i + 1 r~ P r~ 2 S(U~ ) = 0 (4. j = 1.

7.4 Turbulent boundary layer ows 4. ~ (~u) for a linear adve tionWe re all the interpretation of the onve tive term r diusion-rea tion type problem.2 The Reynolds stress tensor ~ h~u0 ~u0 i are the averaged ee t of the quadrati . nonThe Reynolds stress tensor r linear turbulent onve tion term. From the orresponding onservation form we see that Z Z V (~u) ~nd = d X V j =1 uj nj d and therefore the omponent uj gives the transport of in xj -dire tion via the onve tive .

eld ~u. put in other words Z V uj nj d = Transport of due to omponent uj of the onve tive .

eld ~u in xj -dire tion a ross the part of the boundary of V with normal omponent ~n ~ej = nj This formulation an be used to give an interpretation of the Reynolds stress tensor. whi h an be written in onservation form as Z V h~u0 u~0 i ~nd = Z d X V j =1 hu0i u0j i nj d It is worthwhile pointing out that in the situation of the Reynolds stress tensor the onve tive .

eld is the random u tuating velo ity .

eld ~u0 the transported quantity is random u tuating momentum ~u0 (or to be more pre ise ~u0 Thus the omponent hui uj i gives the statisti ally averaged transport of random u0i momentum in xj -dire tion due to the random onve tive .

Put in other words Z V hu0iu0j i nj d = Statisti ally averaged transport ( ux) of random momentum in xi -dire tion u0i due to the omponent u0j of the random onve tive .eld u0j .

y)~ex . We then substitute the ansatz 0 1 U U~ = V A 0 52 .eld ~u0 in xj -dire tion a ross the part of the boundary of V with normal omponent ~n ~ej = nj 4.3 The turbulent boundary layer equations For the at plate turbulent boundary layer equation we already know that the steady mean- ow solution is of the form U~ (~x) = U (x.7.

7 Statisti al theory and modelling into the equation for the U omponent in (4.20) and obtain 2U V 0 0 1 P 2U U 2 = 0 U + U + h u u i + hu0 v0 i + 2 x y x y x x y This equation is alled boundary-layer equation be ause it is based on the ansatz designed for boundary layer ows.4. Now the following additional assumptions are made U U O(1) x V U O(1) y 0 0 hu u i O(1) x 2U 2 O(1) x 1 P O(1) (4.21) x whi h are all satis.

Then the boundary layer equations redu e to 2U 0 0 h uvi 2 = 0 y y This equation is often alled equilibrium stress balan e equation. y) Substitution into the equilibrium stress balan e equation gives U ( + t ) = 0 (4.23) y y Remark 16 Equation (4. (i) laminar diusion due to the thermal random motion of the mole ules and (ii) turbulent diusion due to the turbulent u tuation motion of the uid. the dominant pro ess is diusion of momentum in wall-normal dire tion. 53 .22) y together with an algebrai model for the so- alled turbulent vis osity or eddy vis osity t designed for wall-bounded ows t = t (U.23) states that in the near-wall region turbulent boundary layer ow.4 Prandtl's mixing length model In 1925.. U=y. Ludwig Prandtl proposed a on ept to model hu0 v0 i hu0 v0 i = t U (4.7. There are two dierent sour es of wall-normal diusion.ed provided the pressure gradient is negligible small. viz. 4. this equation is un losed as the turbulent u tuation term hu0 v0 i is unknown. however.

25) u Then the hain rule for dierentiation gives U u+ y+ = w u+ y+ y + 2 u u . = 2 w y+ u u+ .24) y Note that due to the no-slip ondition ~u(. ( + t ) + = w y + ( + t ) u 1 . ~u0 = 0 at the wall and hen e hu0 v0 i = 0 at the wall.4 Turbulent boundary layer ows The turbulent vis osity t is of dimension m2 s 1 . lm = y with a suitable onstant to be determined later. t = u y (4. 4. t) = 0. t+ = t (4. hu0 v0 i = t U .5 Universal form of the turbulent boundary layer equations Now we integrate the equilibrium stress balan e equation from y0 = 0 to y0 = y Z y U U U ( + t ) dy0 = ( + t ) jy jy=0 y y y y 0 as t = u y vanishes at the wall y = 0. y+ = .7. (1 + t+ ) + = 1 y In universal oordinates. Prandtl's model for t be omes u y t+ = t = = y+ ( + t ) 54 . and hen e U U ( + t ) jy = jy=0 w y y Now we write in universal oordinates U yu u+ = . and the natural length s ale for a wall-bounded ow is the wall-distan e: u = u . U~ = 0. This motivates the ansatz to write t as a produ t of a hara teristi velo ity u and a hara teristi length lm t = u lm The natural length s ale from the law of the wall is the fri tion velo ity u .

and the region (40=u .. by using the so- alled van Driest damping fun tion D(y+ ) = 1 e y =26 : + t+ = y+ 1: e y =26 Remark 18 Note that the non-dimensional wall-distan e y+ = yu = an be interpreted as a lo al Reynolds number. whi h reads in its original form 2 U . 0:2y99 ) where (4. it follows from U u+ y+ 1 u u U y = + + y = u y = y u y y yu = From this relation it an be seen that the Prandtl mixing length model is an algebrai turbulen e model t = t (U.g. An additional onsequen e is that hu0 v0 i and hen e t are mu h smaller for y+ / 10 then predi ted by the Prandtl relation. 4.8 k-! turbulen e model 55 . This ee t is alled blo king as it stems from the vis ous ee t of the proximity of the wall.26) is alled logarithmi law of the wall. due to hu0 v0 ijy=0 = 0 we t+ 1 + ) u = 1 ) u+ = y+ y+ Logarithmi boundary layer.26) t+ = y+ 1 ) + y y The onstants and C are determined by omparison with experimental data = 0:41. It is modelled by redu ing t very lose +to the wall e. Therefore for small y+ -values (very lose to the wall) we an assume laminar ow. Due to the logarithmi dependen e u+ ln y+ . y+ ' 40: At a suÆ ient distan e from the wall. U=y). y+ have t and thus / 5: Very lose to the wall. t and thus 1 1 u+ = + ) u+ = ln y+ + C (4.4. Vis ous sublayer. C = 5:1.8 k-! turbulen e model Remark 17 The fa t that the mean- ow and the turbulent u tuations are zero at the wall implies that hu0 v0 i = 0. whereas for large y+ -values fully developed turbulent ow onditions are expe ted.26) holds is alled logarithmi part of the boundary layer. Remark 19 We note that in the logarithmi part of the boundary layer U u = y y whi h is the relation originally used by Prandtl. t = lm m y y where lm is alled mixing length. l = y hu0 v0 i = t U . the relation (4.

4 Turbulent boundary layer ows 56 .

For .1) The general form of a s alar onservation law reads r~ F~ (u) = f in (5.5 DG-FEM for s alar onservation laws The linear adve tion equation is the non- onservative form of the linear s alar onservation law r~ (~bu) = f in (5. In the linear ase F~ (u) = ~bu.2) where F~ is alled ux fun tion.

4) 5.3) u = g on (5. hen e r~ F~ (u) = f in (5.rst order problems. X K 2Th Z K F~ (u) X Z K 2Th K r~ v dx + Z K r~ F~ (u) v dx = r~ F~ (u) v dx = F~ (u) ~n v ds = Z f v dx X K 2Th X K 2Th Z Z K K f v dx f v dx where ~n is the outward normal ve tor to K .1 Weak formulation for DG-FEM We multiply this equation with a fun tion v and integrate over the domain Z r~ F~ (u) v dx = Z f v dx We write the integral over as a sum over all ell K and apply the rule of integration by parts on ea h ell K Z . This is the weak formulation used to de. a ording to hara teristi theory only boundary onditions at the in ow boundary an be pres ribed. .

Now we introdu e the .ne the DG-methods.

nite element spa es asso iated to the triangulation Th = f K g of . We set Vh = f v 2 L2 ( ) j vjK 2 P (K ) 8K 2 Th g 57 .

We remark that for setting v 1 the lassi al .5 DG-FEM for s alar onservation laws where P (K ) is the spa e of polynomial fun tions of degree less than of equal p 1 on K.5) is not stable.5) Histori ally. The orresponding weak problem then reads Find uh 2 Vh su h that 8K 2 Th Z Z ~ v dx + F~ (u) ~n v ds = F~ (u) r Z K K K f v dx 8v 2 P (K ) (5. it is well known that the dire t numeri al solution of (5.

nite volume method (FVM) is re overed Find uh 2 Vh su h that Z K F~ (u) ~n ds = Z K f dx 8K 2 Th (5.6) In FVM the term F~ (u) in the surfa e integral is repla ed by a suitable approximation ~ (u) alled numeri al ux and denoted by F[ Z K F~ (u) ~n ds Z K ~ (u) ~n ds F[ (5. tra es and numeri al uxes The spa e of solution Vh is a subspa e of H 1 (Th ) whi h is de.2 Fun tional setting.8) 5. the symmetry of the stiness matrix.7) The hoi e of the numeri al ux determines stability and a ura y of the method and also ae ts e. Therefore the following general DG-FEM formulation is onsidered: Find uh 2 Vh su h that 8K 2 Th Z Z ~ (u) ~n v ds = ~ v dx + F[ F~ (u) r K K Z K f v dx 8v 2 P (K ) (5.g.

ned by H 1 (Th ) = f v : ! R : vjK 2 H 1 (K ) 8K 2 Th g The tra es of fun tions in H 1 (Th ) may de.

ned as follows: De.

ne the union of boundaries and inner boundaries of the elements K 2 Th Union of boundaries of the elements: = [K 2Th K 0 Union of inner boundaries of the elements: = n Then the tra e of fun tions in H 1 (Th ) is given by jTh : H 1 (Th ) 7! T ( ) K 2Th L2 (K ) u 7! jTh u = K 2Th jK u Fun tions in T ( ) are thus double-valued on 58 0 and single-valued on . .

5.3 Jump and averaging operators The spa e L2 ( ) an then be identi.

ed as the subspa e of T ( ) onsisting of fun tions for whi h the two values oin ide on all internal edges L2 ( ) = f jTh v 2 T ( ) : jKi v = jKj v 8Ki \ Kj 6= .9) ~ (u) is alled onsistent if De. F~ 7! Fb~ (5. g The general form of a s alar onservation law with ux fun tion F~ reads r~ F~ (u) = f in Then the numeri al ux are taken to be linear fun tions Fb~ : [H 1 (Th )℄d ! [T ( )℄d .

1) be omes Z S F (u) ~n ds = \ ~ Z S f dx sin e the integrals on ea h inner boundaries of S mutually an el out. For s alar fun tions ~ g q 2 T ( ) and ve tor valued fun tions ~ 2 [T ( )℄d we want to de. whi h holds whenever the numeri al ux ve tor is single-valued: Denote S the union of any olle tion of elements. then (7. 5.3 Jump and averaging operators In this se tion we introdu e some tra e operators that will help us to operate with the numeri al uxes and to obtain the desired DG-FEM formulation.10) ~ (u) is alled onservative if F[ ~ (u) is single-valued on F[ The term onservative omes from the following important property.nition 13 The numeri al ux ve tor F[ ~ (u) = F~ (u)j F[ (5.

ne average fqg. fphi 0 ~ and jump dqe. de on the skeleton of inner fa es as follows. De.

Let e 2 Eh0 be an interior edge share by elements K1 . K2 .nition 14 (Average and jump operator) Denote Eh the set of all edges and Eh0 the set of interior edges. De.

ne the unit normal ve tors ~n1 . We set qi qjKi and de. ~n2 on e pointing exterior to K1 and K2 . respe tively.

de : : T ( ) ! [L2 ( 0 )℄d dqe = q1~n1 + q2~n2 on e 2 Eh0 59 .ne on e f g : T ( ) ! L2 ( 0 ) fqg = 12 (q1 + q2 ) .

5 DG-FEM for s alar onservation laws For ~ 2 [T ( )℄d we de.

ne ~ 1 and ~ 2 analogously by f g : [T ( )℄d ! [L2 ( 0 )℄d de : : [T ( )℄d ! L2 ( 0 ) f~g = 12 (~1 + ~2 ) . f~g = ~ on e 2 Eh (5. For e 2 Eh we set dqe = q~n . d~e = ~1~n1 + ~2~n2 on e 2 Eh0 Denote Eh the set of boundary edges. This de.11) Noti e that the jump dqe of a s alar fun tion q is a ve tor parallel to the normal. and the jump d~ e of a ve tor fun tion is a s alar quantity.

nition has the advantage that it does not depend on assigning any ordering to the elements Ki . Moreover we noti e that on boundary edges fqg and d~ e need not to be de.

Lemma 1 For all q 2 T ( ) and for all ~ 2 [T ( )℄d the following identity holds X Z K 2Th K q~ ~n ds = Z dqe f~gds + Proof.ned as su h terms do not appear in the orresponding formulations.12) Z 0 fqgd~eds Due to the de. We write the right hand side as Z dqe f~gds + Z 0 fqgd~eds = Z dqe f~gds + Z 0 Z 0 fqgd~eds dqe f~gds + (5.

60 2 .11 we have Z dqe f~gds = Z q~ ~nds Now we onsider an inner fa e e 2 Eh0 ommon to K1 and K2 Z e dqe f~gds + Z Z e fqgd~eds Z ~ 1 + ~ 2 1 ds + (q1 + q2 )(~ 1 ~n1 + ~ 2 ~n2 )ds 2 e e2 Z Z Z Z ~ ~ ~ ~ q1~n1 1 ds + q1~n1 2 ds + q2~n2 1 ds + q2~n2 2 ds = 2 2 2 2 eZ eZ eZ eZ q1 ~ q2 ~ q2 ~ q1 ~ ~n ds + ~n ds + 1 ~n1 ds + 2 ~n2 ds + 2 1 1 2 2 2 e 2 e 2 Z e Z e q1 ~ 1 ~n1 ds + q2 ~ 2 ~n2 ds = = (q1~n1 + q2~n2 ) e e Summing over all K then gives the assertion.nition on boundary fa es 5.

vi = Z X K 2Th X Z e e6 Z K ~ v dx . v) + b(uh .16) ~ (g)g ~n dve ds fF[ (5. v) = b(u. vi 8v 2 Vh (5.4 DG-FEM formulation for hyperboli problems 5.17) 61 .5.14) (5. Then the variational problem be omes Find uh 2 Vh su h that a(uh .15) (5.13) with a(u. v) = f (v) = hg.4 DG-FEM formulation for hyperboli problems Now we apply this equation to X Z K 2Th K ~ (u) ~n v ds F[ ~ (u) and q = v. Sin e the numeri al ux is assumed to be onservative We set ~ = F[ ~ (u)e vanishes and we obtain the jump dfF[ X Z K 2Th K F (u) ~n v ds = [ ~ = XZ e2Eh e ~ (u)g ~n dve ds fF[ X Z e6 e fF (u)g ~n dve ds + [ ~ X Z e e ~ (u)g ~n dve ds fF[ On the in ow boundary the boundary ondition an be substituted. F~ (u) r ~ (u)g ~n dve ds fF[ f v dx X Z e e (5. v) = f (v) + hg.

5 DG-FEM for s alar onservation laws 62 .

6 DG-FEM for diusion type problems In this hapter the DG-FEM for diusion type problems is des ribed.1 Weak formulation for DG-FEM for diusion type problems In the previous hapter it was des ribed that DG-FEM is naturally applied to the ~ F~ (u) = f . For sake of simpli ity we onsider the lapla e problem u = f in u = 0 on where is assumed to be a polygonal domain and f is a given fun tion in L2 ( ). 6. This motivates to formulate the Lapla e onservative formulation r problem as a .

integration over the domain and writing the integral over as a sum over all ell K gives Z X X K 2Th K 2Th Z K K X ~ ~ dx = K 2Th X r~ ~ v dx = K 2Th Z Z K K r~ u ~ dx f v dx We apply formally the rule of integration by parts on ea h ell K Z X K 2Th X K 2Th Z K K ~ ~ dx = ~ v dx = ~ r X K 2Th X K 2Th Z Z K K ~ ~ dx + ur f v dx + Z K Z K u ~ ~n ds ~ ~n v ds 63 .rst-order system ~u ~ = r r~ ~ = f in u = 0 on Multipli ation of these equations by test fun tions ~ and v respe tively.

This is the weak formulation used to de.6 DG-FEM for diusion type problems where ~n is the outward normal ve tor to K .

ne the DG-methods. We de.

and ansatz-spa es Vh = dh = f v 2 L2 ( ) : vjK 2 P (K ) 8K 2 Th g f 2 [L2 ( )℄d : vjK 2 [P (K )℄d 8K 2 Th g The orresponding weak formulation is obtained by introdu ing the numeri al uxes ~ u and u on the boundary of K ~b and ub as suitable approximations for = r Find uh 2 Vh and h 2 h su h that 8 K 2 Th Z X K K 2Th Z X K 2Th X ~ v dx = ~h r K K 2Th X ~h ~ dx = Z K Z K K 2Th Z ~ ~ dx + uh r f v dx + Z K K ~ h u h ~ ~n ds ~n v ds 8~ 2 [P (K )℄2 8v 2 P (K ) We sum of intergrals over ells K an be written as integral over Z Z Z ~h ~ dx = Z ~ v dx = ~h r f v dx + ~h ~ dx = Z Z Z ~ v dx = ~h r Z f v dx + K K 2Th ~ ~ dx + uh r K 2Th K X Z Now we apply Lemma 1 and obtain Z X Z ~ ~ dx + uh r Z u h ~ ~n ds ~ h ~n v ds du h e f~ g ds + f~ h g dve ds + Z 0 8v 2 P (K ) Z 0 fu h g d~ e ds d~ he fvg ds (6.1) (6. so that the .ne the test.2) Now we want to express h solely in terms of uh .

whi h also follows from Lemma 1 Z Z r~ ~ v d x = Z = r~ ~ uh d x = Z Z ~ uh dx ~ r ~ ~nv ds f~ g dve ds + ~ vdx ~ r In this equation we set v = uh Z Z ~ vdx ~ r Z f~ g duh e ds + Substitution of this equation into (6.nal formulation will only involve uh .1) gives Z 64 ~h ~ dx = Z r~ uh ~ dx + Z du h uh e f~ g ds + Z 0 Z 0 Z 0 d~ e fvg ds d~ e fvg ds (6. For this purpose.3) fu h uh g d~ e ds . we onsider the following relation.

2 Consisten y and onservation Let u solve the boundary value problem u = f in u = 0 on 2 We onsider (6.4) i du h uh e fr~ vg f~ h g dve ds i fu h uh g dr~ ve 0 Z ~ h Bh (uh . v) = Z + Z Z f g dve ds + r~ uh r~ v dx + h Z Z 0 fv dx h 0 fu h uh g dr~ ve ds d~ h e fvg ds 8v 2 Vh (6.5) 6.3) with v 2 H (Th ) (re all that then by de. v) = Z Z du h uh e fr~ vg ds + d~ h e fvg ds (6.2) Z r~ uh r~ vdx + Z = f v dx + whi h is written as with B (uh .2 Consisten y and onservation ~ v and obtain In this equation we take ~ = r Z Z ~ v dx = ~h r r~ uh r~ vdx + Z du h uh e fr~ vg ds + Z 0 fu h uh g dr~ ve ds The left hand side of this equation is then substituted into (6.6.

v) = Z + = + + r~ uh r~ v dx + Z Z h 0 h Z Z fu h uh g dr~ ve dr~ uh e fvg ds i du h uh e fr~ vg f~ h g dve ds i d~ h e fvg ds fr~ uh g dve ds + du h uh e fr~ vg h 0 h fu h uh g dr~ ve uh v dx + Z Z d~ e fvg ds Z i 0 dr~ uh e fvg ds f~ h g dve g ds i d~ h e fvg ds 65 .nition v 2 H 2 (K ) 8K 2 Th ) Z Z r~ ~ v dx = ~ uh and obtain we set ~ = r Z Z uh v dx = Z r~ uh r~ vdx Z ~ vdx ~ r f~ g dve ds + Z 0 Z fr~ uh g dve ds 0 This is substituted into (6.5) and we obtain B (uh .

3) we see that = r If we assume that (ii) the ve tor ux b is onsistent. First we assume that the numeri al uxes are onsistent. due = 0. i. (i) ub = uj whi h ~ u.e. from (6.6) holds for all v numeri al ux must be onsistent. 66 (6. implies that dube = 0. fr B (uh . Then. As fug = u. dr ~ ue = 0 and u = f we get u 2 H 2 ( ). v) = Z + + f v dx + Z h Z r~ u dve ds i du h e fr~ vg h 0 Z f~ h g dve ds ( fu h g ~ ve u )dr i d~ h e fvg ds From this equation we see that onsisten y of the formulation and onservativity of the uxes are strongly related. fubg = u on . we know that ~ ug = r ~ u. fbg = = r~ u on . then we get that dbe = 0. Moreover.6 DG-FEM for diusion type problems Sin e f 2 L2 ( ) and is assumed to be a onvex polygonal domain. v) = Z f v dx On the other hand it an be shown that if (6. B (uh .6) 2 H 2 (Th ) then the . Substitution of there relations gives onsisten y.

7 Finite-volume methods We onsider the .

nite-volume method for paraboli problems.1 Finite-volume formulation This onservative formulation is the natural starting point for the . We remark that the dierential form of a onservation law u ~ F~ (u) + r ~ ( r ~ u) = f in + u + r t originates from the following onservative formulation: For any ontrol volume V Z d u dx + dt V Z V udx + Z V F~ (u) ~nd + Z V ~ u ~nd = r Z V f dx (7.1) 7.

Moreover denote Pi the bary enter of ell Vi .nite volume method. for i 6= j and [Ni=1 V i = . (1) De ompose a set of dual grid ells (or ontrol volumes) Th = f Vi gN i=1 with Vi \ Vj = . (3) We seek a dis ret solution whi h is onstant on ea h dual grid ell Vi : 8 VZi Th .

Z d ~ ui ~nd = r F~ (ui ) ~nd + ui dx + u dx + dt V i V V V In the following we onsider the situation sket hed in Figure 7. in .nd Z ui 2 P0 (Vi) Z s.1 Z V f dx (4) Compute (re over) the gradients: As the solution that we seek is pie ewise onstant. the gradients in the vis ous uxes would vanish. ~ u is "re oved" from u using the GreenAs a remedy.t.

In ve tor notation this equation be omes Z Z ~ru dx = u~n d Vi Vi 67 .1 denote Z Z u u~ek ~n d dx = x Vi k Vi where ~ek denotes the unit ve tor in the xk -dire tion.nite-volume methods r Gauss divergen e theorem. For the situation of Figure 7.

the surfa e integral is omputed via Z 4 1 1 X ui + uij u~n d = area(Si. i3 S i.1: Control volumes and surfa e uxes in FVM ~ u pie ewise onstant on ea h Vi .1.7 Finite-volume methods Pi2 00 11 11 00 00 11 000 111 111 000 000 111 000 111 111 000 000 111 S i. Then we obtain We seek r r~ ui = Vol(1V ) i Z Vi u~n d Using the notation in Figure 7.ij ) Vol(Vi ) Vi Vol(Vi ) j =1 2 r~ ui = ~ (u) for the onve tive ux F~ (u) and r ~ u (5) Introdu e a numeri al ux fun tion F[ ~u for the vis ous ux r 8 VZi Th . i1 S i. i2 P i3 Pi 00 11 11 00 00 11 Pi1 000 111 111 000 000 111 000 111 111 000 000 111 S i. i4 11 00 00 11 00 11 00 11 111 000 000 111 000 111 000 111 Pi4 111 000 000 111 000 111 000 111 Figure 7.

2) (6) For the numeri al ux fun tion for the vis ous uxes. d u dx + dt V i V udx + V ~ (ui ) ~nd + F\ Z V ~d r ui ~nd = Z V f dx (7.nd Z ui 2 P0 (Vi)Z s.3) (7) The design of a numeri al s heme for the onve tive uxes is the most riti al issue in .t.ij ) 2 (7.ij 4 X ~ j =1 ~d r ui ~nd ~ uij rui + r area(Si. a entral s heme is used Z V ~d r ui ~nd = = 4 Z X j =1 Si.

nite-volume methods. Here we restri t ourselves to the situation of 68 .

7.2 Expli it solution using low-storage Runge-Kutta method linear adve tion F~ (u) = ~bu with a onstant adve tion .

then the dis retisation sten il for the uxes of ell Vj takes into a ount only dire t neighbours of ell Vj . : : : .ij ) ( u where u = i uij ~b ~n > 0 ~b ~n < 0 (7. As we seek a solution whi h is pie ewise onstant w. : : : .t.r. F~ v (~u) = (Fv1 (~u). Z V ~bu i ~nd = = 4 Z X ~bu i ~nd j =1 Si. 7. therefore we have Z d u dx = uj Vol(Vj ) dt V j Moreover we de. (7.4) where ~n is the outer normal ve tor of ontrol volume Vi for fa e Si.3). un )T and the ve tor of invis id ( onve tive) and vis ous uxes F~ i (~u) = (Fi1 (~u). Z Z Vj Vj Fin (~u)) Fvn (~u)) F~d uj ~nd ~d r uj ~nd If we use (7. with Fij (~u) = Fvj (~u) = Fi2 (~u)T .ij 4 X j =1 ~bu ~n area(Si.4). hen e Fi does only depend on ui and the dire t neighbours uij . : : : . We introdu e the ve tor of unknowns ~u = (u1 .2 Expli it solution using low-storage Runge-Kutta method We solve the FV-s heme using an expli it Runge-Kutta method.ij . the ontrol volumes. Fv2 (~u)T .eld ~b. u2 .

Rea n (~u))T S~ (f~) = (Sr 1 (f~). Sr 2 (~u). : : : . : : : . Sr j = Z Vj f dx 69 . Rea 2 (~u). Sr n (~u))T with Rea j (~u) = Z Vj udx .ne R~ (~u) = (Rea 1 (~u).

7 Finite-volume methods Then the .

(7. : : : . un ) 2 R n s.t. u2 .2) be omes Find (u1 .5) d uj Vol(Vj ) + Fij (~u) + Fij (~u) + Rea j (~u) = Sr j dt In FVM. it is onvenient to de.nite volume formulation (7.

ne the so- alled residual for ell Vj as 1 Resj (~u) = Sr Fij (~u) Fvj (~u) Rea j (~u) (7.6) Vol(Vj ) j The .

8) by ! d ~u = Res(~u) (7. u2 .e. e...9) dt where t denotes a .8) Problem (7. : : : . i.7) then be omes Z V ~ u ~nd = r ! Z V f dx (7. Res(~u) = 0 (7. We repla e (7. a Runge-Kutta method.t.g.7) Find (u1 .8) is solved iteratively in pseudo-time using an expli it s heme. un ) 2 R n s.1) has a steady state solution.nite volume method then be omes d u = Resj (~u) dt j or in ve tor notation ! d ~u = Res(~u) dt R Now we onsider the situation where (7. limt!1 ddt Vi u dx = 0 forall ells Vi (or limt!1 ~u = 0) and u is then solution of the orresponding steady state problem Z V udx + Z V F~ (u) ~nd + The FV-formulation for (7.

ti ious pseudo time.9) satis. The steady state solution of (7.

es d~u=dt = 0 and hen e satis.

(1) Initial guess ~(u)0 = 0 (2) UNTIL CONVERGENCE DO: Given ~un from the previous pseudo time step.es (7. ! ~un1 = ~(u)n + Res(~un0 ) ! ~un2 = ~(u)n + Res(~un1 ) ::: = ::: ! ~unk = ~(u)n0 + Res(~unk 1 ) 70 . Using a suitable time step size t (whi h depends on the problem data and ~b) we introdu e an upper index n to denote the pseudo time step and a lower index k to denote the Runge-Kutta stage.8). The steady state solution is omputed by mar hing in pseudo time t using a so- alled low-storage K -stage Runge-Kutta s heme.

3 Interpretation of expli it s hemes. Downwind numbering and update the solution ~un+1 ~unk This modi.7.

ation of the lassi al Runge-Kutta s heme is alled low-storage version. as the intermediate results ~unk do not have to be stored for the .

then the arising linear system requires at least N iteration steps.nal update of the solution. Downwind numbering Finally it is worthwhile to ompare the expli it solution strategy and an impli it method using an iterative linear solver: Figure 7.2 illustrates the transport of the numeri al solution for a hyperboli problem through the omputational domain. Note that if the solution has to pass through N ells to travel from the in ow boundary to the out ow boundary. CCFL 2 (0. The onstant CCFL is alled Courant-Friedri hLevy number. of Ja obi-type or Gauss-Seidel type. For a 3-stage low storage RK-s heme. This illustrates the hyperbolic nature of the problem. For an expli it s heme the time-step t is limited by the Courant-Friedri h-Levy ondition: " # hi h2i t = C + jj~bjj1. 1:4). Figure 7.2: Transport of the numeri al solution using an expli it s heme 7. then at least N Runge-Kutta steps are required to obtain the steady-state solution. u oo u oo Propagation of the numerical solution through the dual grid.Vi 4 CFL where hi is the smallest size of ell Vi .g. e.3 Interpretation of expli it s hemes. If the same problem is solved using an impli it s heme. Rea hing a onverged solution for an iterative linear solver an be 71 .

7 Finite-volume methods measured in terms of the de.

The number of iterations required for an iterative linear solver to onverge an be redu ed signi. Usually the number of pseudo time iterations is mu h smaller then the number of iteration steps for the linear solver. sin e in the expli it FV-s heme. the information travels along the hara teristi s (whi h is the most-eÆ ient way) from the in ow boundary through all ells to the out ow boundary.nition of the residual from numeri al linear algebra. this is equivalent to the situation that a steady-state solution for the iteration in pseudo time has been obtained.

these improved linear solvers an be used also for ontinuous Finite-Element methods and DG-FEM. Of oarse. 72 . Hen e an understanding of the hara teristi transport of information helps arising in expli it s hemes gives valuable ideas for the design of linear-solvers for impli it methods. antly if the grid-points are also ordered with respe t to the transport of physi al information (downwind numbering).

- ContentServer (1)
- AI_EFEM1D
- Linux Refresher Presentation
- examples from FEM
- Mixed FEM for elastoplasticity
- Markov Chain Monte Carlo
- implementation of anisotropic damage model
- Bsc Thesis
- concepts of higher order finite elements using sparse grids
- multiscale method
- coupled damage plasticity model
- implementation of component based parallel finite element library
- 1620290802_ftp
- Web 5155
- space time FEM
- ms10_2012
- Flow
- cuda3
- Slide Chpt09
- week7_PolyGen
- Week2 ODEs
- week5_part1
- Lecture 4
- File Streams
- 15 Spherical Trigonometry

Sign up to vote on this title

UsefulNot useful- Callen Thermodynamics Solution
- Manuscript Gaseous Microflow
- Russo Phys Con
- j Food Proce Ng 2001
- WorldSci - Modern Many Particle Physics
- PHY1321 B Final Exam Answers Fall 2014
- Langlois
- Design Evaluation of Particulate Wet Scrubbing Systems
- Perfect Gas Law
- Revision on Important Concepts
- Denis Serre Systems of Conservation Laws 1_ Hyperbolicity, Entropies, Shock Waves 1999.pdf
- Journal TSEA 2014 Shaft
- Differentiation of Constraints
- Mach Reflection
- Report 5
- Mass Transfer
- Journal of Chemical Engg
- 9781447151395-c1
- Andrea Raspini- Massive Particles with Definite Chiralities
- 00180321
- Thermo Vle Part 02
- ThermoVid_4_01
- ANSI Z21.58-2007 CSA 1.6 -2007 Outdoor Cooking Appliances
- Chem 156 Syllabus
- Correlaciones nuevas para retención del gas a gran escala en reactores de columna
- Smad Assignment No 2
- Mariana Grana et al- Generalized structures of N = 1 vacua
- Direct Calculation of Unsteady state
- ChE422 Topic 8
- Graduation Thesis Ha Minh Tan Final v2
- Lecture Dgfem Ws2006.Ps