Part A Fluid Dynamics & Waves

Draft date: 7 February 2010 21

2 Two-dimensional incompressible irrotational ow

2.1 Velocity potential and streamfunction

We now focus on purely two-dimensional ows, in which the velocity takes the form u(x, y, t) = u(x, y, t)i + v (x, y, t)j . With the velocity given by (2.1), the vorticity takes the form =u= u v x y k. (2.2) (2.1)

We assume throughout that the ow is irrotational, i.e. that u 0 and hence v u = 0. x y (2.3)

We have already shown in Section 1 that this condition implies the existence of a velocity potential such that u , that is u= , x v= . y (2.4)

We also recall the denition of as

x x

(x, y, t) = 0 (t) +
0

u dx = 0 (t) +
0

(u dx + v dy ) ,

(2.5)

where the scalar function 0 (t) is arbitrary, and the value of (x, y, t) is independent of the integration path chosen to join the origin 0 to the point x = (x, y ). This fact is even easier to establish when we restrict our attention to two dimensions. If we consider two alternative paths, whose union forms a simple closed contour C in the (x, y )-plane, Greens Theorem implies that (u dx + v dy )
C S

u v x y

dxdy,

(2.6)

where S is the region enclosed by C , and the right-hand side of (2.6) is zero by virtue of (2.3).

22

OCIAM Mathematical Institute

University of Oxford

(i)
6

(ii)
6 6 4 4

(iii)

6 6 4 2

6 6 4 2

Figure 2.1: Streamline plots for (i) uniform ow; (ii) stagnation point ow; (iii) line source. We also assume that the ow is incompressible, so that u = 0, which in two dimensions reduces to u v + . (2.7) x y By substituting (2.4) into (2.7), we nd that satises Laplaces equation, that is 2 2 + 2 = 2 = 0. x2 y (2.8)

As noted in Section 1, Laplaces equation is linear, and therefore much easier to solve than the nonlinear Euler equations. We will see that many ows may be constructed using well-known simple solutions of Laplaces equation. Once we have solved for , the velocity components are easily recovered from (2.4), and the pressure may be found using Bernoullis Theorem. We recall from Section 1 that this reads p 1 + ||2 + is constant in steady irrotational ow, 2 or p 1 + + ||2 + = F (t) in unsteady irrotational ow. t 2 (2.9)

(2.10)

Example 2.1 Uniform ow

A linear function of x and y trivially satises Laplaces equation. The velocity potential = U x cos + U y sin , where U and are constants, represents the velocity eld (u, v ) = U (cos , sin ). (2.12) (2.11)

This corresponds to uniform ow, with speed U , in a direction making an angle with the x-axis, as shown in Figure 2.1(i).

Part A Fluid Dynamics & Waves Example 2.2 Stagnation point ow

The velocity potential

Draft date: 7 February 2010 23

1 2 x y2 2 clearly satises Laplaces equation and corresponds to the stagnation point ow = (u, v ) = (x, y )

(2.13)

(2.14)

considered previously in Section 1. We recall that the streamlines are hyperbolae, as illustrated in Figure 2.1(ii).

Example 2.3 Line source

We recall that the two-dimensional Laplaces equation may be written as 2 = 1 r r r r + 1 2 = 0, r2 2 (2.15)

in terms of plane polar coordinates (r, ). If we seek a radially-symmetric solution, with = (r), we nd that Q = log r, (2.16) 2 where Q is a constant of integration, and the additional arbitrary constant may be set to zero without loss of generality. The corresponding velocity eld is given by u= d Q er = er , dr 2r (2.17)

where er is the unit-vector in the r-direction. The velocity (2.17) may be rewritten in terms of Cartesian variables as Q u = (u, v ) = (x, y ). (2.18) 2 2 (x + y 2 ) We see that the uid ows radially outward, at a speed that increases without bound as we approach the origin. The streamlines are straight rays pointing outward from the origin, as shown in Figure 2.1(iii). This ow is called a line source: a source because uid is being squirted out of the origin, and a line because the point (0, 0) in the (x, y )-plane really corresponds to a line (the z -axis) in three-dimensional space. The constant Q is called the source strength. To see why, let us calculate the rate at which uid crosses a contour C containing the origin, namely u n ds,
C

where n is the outward-pointing unit normal to C . The integral is particularly straightforward if we choose C to be the circle r = a, for some constant radius a, in which case n er and hence
2

u n ds =
C 0

Q er er a d = Q. 2a

(2.19)

It is easily shown that any other simple closed curve C containing the origin gives the same value of the ux. Therefore Q is the rate at which uid is produced by the source. If Q is negative, then uid is being consumed rather than produced, and the origin is called a line sink, of strength |Q|.

24

OCIAM Mathematical Institute

6

University of Oxford

y0
2

6 6 4 2 0 2 4 6

Figure 2.2: Streamlines for the ow produced by a line source of strength Q in a uniform ow at speed U in the x-direction. (In this plot, Q = 2U .) One of the great advantages of dealing with a linear partial dierential equation is that the result of combining two ows may be found by simply adding the corresponding velocity potentials. This is emphatically not the case for the nonlinear Euler equations in general. Example 2.4 Line source in a uniform ow
Consider the ow produced by placing a line source of strength Q at the origin in a uniform ow of speed U in the x-direction. The velocity potentials due to the source and the uniform ow are U x and (Q/2 ) log r respectively, and the potential of the combined ow is thus = Ux + The corresponding velocity components are u=U+ Qx , 2 ( x2 + y 2 ) v= Qy , 2 (x2 + y 2 ) (2.21) Q log r. 2 (2.20)

and the streamlines are shown in Figure 2.2. It is easily seen that there is just one stagnation point (i.e. point where the velocity is zero) in the ow, at (x, y ) = (Q/2U, 0). The separatrix crossing the x-axis through this point separates the uid emanating from the source from the uid carried by the uniform ow.

By analogy with (2.3) and (2.4), we could instead start from (2.7) and deduce the existence of a streamfunction (x, y, t) satisfying u= , y v= . x (2.22)

Part A Fluid Dynamics & Waves

y

Draft date: 7 February 2010 25

n C 1

Figure 2.3: Schematic of a curve C joining two streamlines on which = 1 and = 2 . A handy shorthand for (2.22) is u = ( k). By comparison with (2.5), we infer that may be dened as
x

(2.23)

(x, y, t) = 0 (t) +
0

(u dy v dx) ,

(2.24)

where the function 0 (t) is arbitrary: just as for , any function of t may be absorbed into without inuencing the velocity components given by (2.22). Again, the integral in (2.24) is independent of the path taken from the origin to the point x, and the proof follows quickly from Greens Theorem. The representation (2.22) implies that satises u = , y x , x y 0, (2.25)

and it follows that the streamfunction is constant along streamlines. The contours of therefore give us a handy tool for plotting the streamlines of a given ow. Now let us consider two neighbouring streamlines, on which takes the constant values = 1 and = 2 , say. The ux, i.e. the net ow of uid, between the two streamlines, is given by the integral ux =
C

u n ds,

(2.26)

where C is a smooth path joining the two streamlines, with unit normal n, as shown schematically in Figure 2.3. Substituting for the velocity components from (2.4) and

26

OCIAM Mathematical Institute

University of Oxford

identifying n ds with (dy, dx), we can rewrite (2.26) as ux =

C

, y x

(dy, dx) =
C

dx + dy x y

= 2 1 ,

(2.27)

where the notation [ ]C respresents the change in the value of from one end of C to the other. Therefore, the change in the value of from one streamline to another is equal to the net ux of uid between them. (2.28)

By substituting (2.22) into (2.3), we nd that , like , satises the two-dimensional Laplaces equation: 2 2 + = 2 = 0. x2 y 2 (2.29)

Again, we can pose familiar solutions of Laplaces equation for and read o the corresponding velocity elds from (2.22). We begin by showing how the ows considered in Examples 2.12.4 may be described using the streamfunction. Example 2.5 Uniform ow
The streamfunction corresponding to the uniform ow (2.12) is easily seen to be = U y cos U x sin , up to an arbitrary constant. Hence the streamlines are given by = constant, that is, y = x tan + constant, in agreement with Figure 2.1(i). (2.31) (2.30)

Example 2.6 Stagnation point ow

The streamfunction corresponding to the stagnation point ow (2.14), satises the dierential equations = y, x We easily deduce that = xy (2.33) up to an arbitrary constant. The streamlines plotted in Figure 2.1(ii) are the contours of , namely the hyperbolae xy = constant. = x. y (2.32)

The representation (2.23) is helpful if we wish to use the streamfunction in plane polar coordinates (r, ). Using the formulae given in 2.5.2, we nd that ( k) 1 er e , r r . r (2.34)

and the polar velocity components ur and u are therefore related to by ur = 1 , r u = (2.35)

Part A Fluid Dynamics & Waves Example 2.7 Line source

Draft date: 7 February 2010 27

By substituting the velocity eld (2.17) due to a line source into equation (2.35), we nd that the corresponding streamfunction must satisfy = 0, r Hence, up to the addition of an arbitrary constant, = Q . 2 (2.37) 1 Q = . r 2r (2.36)

The contours of are the rays = constant, in agreement with the streamlines plotted in Figure 2.1(iii).

Example 2.7 demonstrates that the streamfunction due to a point source is multivalued. We should not be too surprised by this, since our uniqueness proof relied on Greens Theorem, which does not apply to a velocity eld like (2.18), which is not dierentiable (or even bounded) at the origin. Furthermore, the jump of Q in as increases from 0 to 2 corresponds to the ux between these two streamlines, according to (2.28). This agrees with the result that Q is the rate at which the source produces uid. Example 2.8 Line source in a uniform ow
The streamfunction corresponding to the ow (2.21) may be found by combining the streamfunctions (2.30) and (2.37) found above, to obtain = Uy + Q . 2 (2.38)

The streamlines plotted in Figure g:unso are the contours of this function, which may be written in the form Q C , (2.39) r= 2U sin where dierent streamlines correspond to dierent choices of the constant C .

As our nal example, we examine the implications of a radially-symmetric streamfunction, by analogy with (2.3). Example 2.9 Line vortex
If the streamfunction depends only upon the radial coordinate r, then, as in equation (2.16), it must take the form = log r, (2.40) 2 up to the addition of an arbitrary constant. The minus sign in front of the integration constant is included for later convenience. We see that this will give circular streamlines r = constant, as shown in Figure 2.4. By using (2.35), we nd that the velocity eld is given by u= d e = e . dr 2r (2.41)

Hence the uid rotates around the origin, with a speed that increases without bound as r 0. This ow is called a line vortex, the line in question being the z -axis in three-dimensional space.

28

OCIAM Mathematical Institute

6

University of Oxford

y0
2

6 6 4 2 0 2 4 6

Figure 2.4: Streamlines due to a vortex at the origin.

The constant is called the vortex strength. We recall that the strength of a source measures the rate at which it produces uid. In contrast, we will see that the strength of a vortex corresponds to the circulation around a contour C containing the origin, namely u dx
C C

(u dx + v dy ).

We make the integral as simple as possible by choosing C to be the circle r = a, for some constant radius a, in which case (dx, dy ) e a d and hence
2

u dx =
C 0

e e a d = . 2a

(2.42)

It is easily shown that the circulation around any other simple closed curve C containing the origin is also equal to . If is positive, the vortex rotates in an anticlockwise sense, as in Figure 2.4; a negative value of the circulation corresponds to clockwise rotation.

We recall that the circulation is supposed to be identically zero in an irrotational ow. However, this result relies on Greens Theorem, which does not apply to a singular velocity eld such as (2.41). A vortex may be thought of as a source of vorticity in an otherwise irrotational ow.

2.2

Complex potential

The velocity potential and streamfunction give us the two alternative representations (2.4) and (2.22) of the velocity. By comparing these expressions, we see that and must satisfy the equations = , x y = . y x (2.43)

Part A Fluid Dynamics & Waves

Draft date: 7 February 2010 29

We recognise these as the CauchyRiemann equations, which relate the real and imaginary parts of a holomorphic function. Provided and are continuously dierentiable, we can deduce from (2.43) that (x, y ) + i (x, y ) w(z ), (2.44)

where w is a holomorphic function of z = x +iy , known as the complex potential. We can use many of the powerful techniques of complex analysis to determine and manipulate the complex potential, and this allows us to solve for and simultaneously. We recall that the holomorphic function w(z ) has a well-dened derivative which is independent of the direction of dierentiation in the complex plane, that is dw 1 ( + i ) ( + i ). dz x i y (2.45)

Hence we nd that the velocity components (u, v ) may be recovered from w using the formula dw u iv. (2.46) dz We illustrate the use of w(z ) by listing the complex potentials corresponding to each of the examples introduced in 2.1. Example 2.10 Uniform ow
The complex potential w(z ) = U ei z (2.47) corresponds to uniform ow at speed U in a direction making an angle with the x-axis. We can easily see this by calculating dw/dz and substituting into (2.46).

Example 2.11 Stagnation point ow

The velocity eld (u, v ) = (x, y ) corresponds to the complex potential w(z ) = z2 . 2 (2.48)

We can easily see that the real and imaginary parts agree with equations (2.13) and (2.33).

Example 2.12 Line source

A line source of strength Q at the origin is represented by the complex potential w(z ) = Q log z. 2 (2.49)

We note that this is a multifunction, with a branch point at the origin. This is no surprise when we recall that a line source has a multivalued streamfunction.

It is easy to generalise (2.49) to a source at an arbitrary point, say (a, b), in the (x, y )-plane. The required complex potential is w (z ) = Q log(z c), 2 (2.50)

where c = a + ib is the complex number corresponding to the point (a, b) in the complex plane.

210 OCIAM Mathematical Institute

University of Oxford

(i) y y

(ii)

11 00 00 11 00 11 00 11 00 11 00 11 00 11 00 11

Q x d

Q x d d

Figure 2.5: (i) Schematic of a source in the half-plane x > 0 bounded by a wall at x = 0. (ii) The same ow mimicked by inserting an image source. Example 2.13 Line vortex
A line vortex of strength at the origin is represented by the complex potential w(z ) = i log z. 2 (2.51)

This again is a multifunction, and we can infer from (2.51) that the velocity potential is the multivalued function = . (2.52) 2 The jump in that occurs upon a circuit of the origin corresponds to the circulation .

A vortex at an arbitrary point (a, b) is represented by the complex potential w (z ) = i log(z c), 2 (2.53)

where c = a + ib. These basic ows can all be combined by simply superimposing the corresponding complex potentials.

2.3
2.3.1

Method of images
Plane boundaries

Now we consider how some of the ows that we have considered so far are modied in the presence of boundaries. As a rst example, suppose uid occupies the half-space x > 0 and there is a rigid impermeable wall along the y -axis x = 0. Now we wish to nd the ow caused by a source of strength Q placed in the uid at the point (d, 0), where d > 0. We have to impose the boundary condition of zero normal velocity through the wall, that is u = 0 at x = 0. (2.54)

Part A Fluid Dynamics & Waves

Draft date: 7 February 2010 211

Im z z r2 2 d d 1 r1 Re z

Figure 2.6: The lengths r1 , r2 , angles 1 , 2 and branch cuts in the denition of the complex potential (2.59). Alternatively, we could insist that x = 0 be a streamline for the ow, that is Im w(z ) = = constant at x = 0. (2.55)

The two boundary conditions (2.54) and (2.55) are equivalent by virtue of (2.22). The idea of the method of images is to insert a ctitious image source at the point (d, 0) behind the wall. By symmetry we would expect the ows caused by the two sources to cancel each other out on the symmetry line x = 0, and thereby satisfy the condition of zero normal velocity there. Of course, we will have introduced a new singularity, but it is outside the physical region x > 0 in which we are interested. By superimposing the real and image sources, we obtain the velocity potential w= Q Q log(z d) + log(z + d), 2 2 (2.56)

and the corresponding velocity components are given by u iv = dw Qz = . 2 dz (z d2 ) (2.57)

In particular, on the x-axis we have z = iy and hence u iv

x=0

Qiy , (y 2 + d2 )

(2.58)

so the boundary condition (2.54) of zero normal velocity is indeed satised. To extract the velocity potential and streamfunction from the velocity potential (2.56), we have to confront the fact that w(z ) is a multifunction. We can expand each log in (2.56) to obtain w(z ) = + i = Q iQ log r1 + log r2 + (1 + 2 ) , 2 2 (2.59)

where the lengths r1 , r2 and angles 1 , 2 are dened as shown in Figure 2.6. There is a branch cut extending to innity from each of the branch points z = d, and it

212 OCIAM Mathematical Institute

2

University of Oxford

y0

-1

-2 -1 0 1 2 3

Figure 2.7: Streamlines for the ow produced by a source at the point (1, 0) with an impermeable wall at x = 0. is convenient to take them to lie along the x-axis as shown, so that 1 [0, 2 ) and 2 (, ]. We easily see that 1 + 2 on the y -axis, and hence the boundary condition (2.55) is satised by the complex potential (2.56). We can read o from (2.59) the streamfunction for the ow and deduce that the streamlines are given by 1 + 2 = constant. This may be rearranged to show that the streamlines are hyperbolae satisfying the equation x2 2xy y 2 = d2 , (2.60)

where = cot(2/Q) is constant. Thee resulting curves are plotted in Figure 2.7, with d = 1. Notice that the competition between the real and image sources causes a stagnation point at the origin. This idea is easily extended to describe the ow due to a vortex in a half-space. The only dierence is that, to achieve the necessary cancellation, the strength of the image vortex must be equal and opposite to that of the original vortex, as shown schematically in Figure 2.8. Thus, the complex potential due to a vortex of strength at the point (d, 0) in the half-space x > 0 with an impermeable boundary at x = 0 is w (z ) = i i log(z d) + log(z + d). 2 2 (2.61)

We can verify that this satises the required boundary condition at x = 0 by calculating the velocity eld, using dw id u iv = = , (2.62) dz (z 2 d2 )

Part A Fluid Dynamics & Waves

Draft date: 7 February 2010 213

(i) y y

(ii)

11 00 00 11 00 11 00 11 00 11 00 11 00 11 00 11

x d

x d d

Figure 2.8: (i) Schematic of a vortex in the half-plane x > 0 bounded by a wall at x = 0. (ii) The same ow mimicked by inserting an image vortex.

y0

-1

-2 -1 0 1 2 3

Figure 2.9: Streamlines for the ow produced by a vortex at the point (1, 0) with an impermeable wall at x = 0.

214 OCIAM Mathematical Institute

University of Oxford

(i) Im z c 111111 000000 000000 111111 000000 111111 000000 111111 000000 111111 c

(ii) Im z c

Re z

Re z

Figure 2.10: (i) Schematic of a source in the positive quadrant bounded by walls along the real and imaginary axes. (ii) The same ow mimicked by inserting three image sources. which is clearly pure imaginary when z = iy . The streamfunction for this ow is found from (2.61) to be given by = log 2 r2 r1 , (2.63)

where r1 and r2 are again dened as in Figure 2.6. Hence the streamlines are curves on which r2 /r1 = constant, and it is easily shown that these are circles centred on the x-axis, as shown in Figure 2.9. The same ideas are easily extended to describe ow in a quadrant. In Figure 2.10(i), we show a source at the point representing the complex number c = a +ib (with a, b > 0) in the Argand diagram, with impermeable boundaries along the real and imaginary axes. The condition of zero ow across the real axis may be satised by inserting an image source at z = a ib = c, where a bar is used to denote the complex conjugate. Now, to impose the equivalent boundary condition on the imaginary axis, we need to insert an image corresponding to each of the sources now present in Re z > 0, at z = a + ib = c and at z = a ib = c respectively. We therefore end up with three image sources altogther, as shown in Figure 2.10(ii), and the resulting ow, with the required symmetry in both the real and imaginary axes, is represented by the complex potential Q Q Q Q log (z c) + log (z + c) + log (z c) + log (z + c) . (2.64) 2 2 2 2 An analogous approach works for a vortex in a quadrant, provided the signs of the images are adjusted as shown in Figure 2.11. The resulting complex potential is w= i i i i log (z c) log (z + c) + log (z c) + log (z + c) . (2.65) 2 2 2 2 In principle, the method of images may be extended to several other domains with straight boundaries. However, an innite number of images will often be required to achieve symmetry in every boundary. w=

Part A Fluid Dynamics & Waves

Draft date: 7 February 2010 215

(i) Im z 111111 000000 000000 111111 000000 111111 000000 111111 000000 111111

(ii) Im z

Re z

Re z

Figure 2.11: (i) Schematic of a vortex in the positive quadrant bounded by walls along the real and imaginary axes. (ii) The same ow mimicked by inserting three image vortices.

4 ai (i) Im z (ii) 2 ai

11111111 00000000
a a Q Re z

11111111 00000000

2ai

4ai

Figure 2.12: (i) Schematic of a source in the strip bounded by walls along the lines Im z = a. (ii) The same ow mimicked by inserting an innite system of image sources.

216 OCIAM Mathematical Institute Example 2.14 A source in a strip

University of Oxford

Fluid occupies the region a < y < a between two rigid boundaries at y = a. We will use the method of images to nd the ow caused by a source placed in the uid at the origin (x, y ) = (0, 0), as shown in Figure 2.11(i). At rst glance, we can take care of the walls by inserting image sources at z = 2ai. However, then we realise that the image at (say) z = 2ai needs its own image in the wall Im z = ai, and we are therefore forced to include another image at z = aai. By continuing this argument, we nd that an innite system of image sources along the imaginary axis is needed, at the points z = 2nai for all integers n, as shown in Figure 2.11(ii). The resulting velocity potential is apparently w (z ) = Q lim 2 N
N

log(z 2nai),
n=N

(2.66)

although it is far from clear that this series converges! However, we can use it to calculate the velocity components, namely u iv = Q dw = dz 2 1 1 + 2z 2 z z + 4 n2 a 2 n=1

(2.67)

This series does converge, and may be evaluated using the identity 1 1 coth(b ) 2 + . 2 + b2 n 2 b 2b n=1 We therefore nd that Q z coth . 4a 2a On the upper wall, where z = ia + x, we therefore nd that u iv = u iv = Q x tanh , 4a 2a (2.69)

(2.68)

(2.70)

so the boundary condition v = 0 on y = a is indeed satised.

Example 2.14 illustrates that in more complicated domains, although the method of images may work in principle, it is often unweildy in practice. As we will see, a far neater approach in such cases is to use conformal mapping. However, another example where the image system may be found relatively easily is where the boundary is a circle.

2.3.2

The Circle Theorem

Theorem 2.1 Milne-Thomsons Circle Theorem Suppose a velocity potential w(z ) = f (z ) is given such that any singularities in f (z ) occur in |z | > a. Then the potential w(z ) = f (z ) + f a2 z (2.71)

(where the bar again denotes complex conjugate) has

Part A Fluid Dynamics & Waves

Draft date: 7 February 2010 217

(i)
4 4

(ii)

4 4 2 0 2 4

4 4 2 0 2 4

Figure 2.13: (i) Uniform ow in the x-direction. (ii) Uniform ow past a circular cylinder. 1. the same singularities as f (z ) in |z | > a; 2. the circle |z | = a as a streamline. Proof To prove part 1, we note that, if |z | > a, then a2 /z < a. Since f (z ) is assumed to have no singularities in |z | a, it follows that the second term in (2.71) has no singularities in |z | > a and the result follows. For part 2, note that, when |z | = a, we have z = a2 /z and hence w (z )
|z |=a

= f (z ) + f (z ) = 2 Re f (z ) .

The right-hand side is evidently real, and it follows that Im w(z ) = = 0 on |z | = a, and the circle |z | = a is therefore a streamline. This theorem allows us to nd the resulting ow when a circular cylinder is placed in a background ow with complex potential f (z ). Example 2.15 Uniform ow past a circular cylinder
Our starting point is uniform ow at speed U in the x-direction, as shown in Figure 2.13(i), with complex potential w(z ) = U z . If we insert an impermeable circular obstacle, at |z | = a say, then the ow will be disturbed as illustrated in Figure 2.13(ii). Our task is to calculate the disturbed ow. Note that f (z ) = U z has a pole at innity but is holomorphic in the disc |z | a and thus satises the hypotheses of the Circle Theorem. We can therefore obtain the solution by substituting f (z ) = U z into equation (2.71): w(z ) = U z + U a2 U a2 Uz + . z z (2.72)

218 OCIAM Mathematical Institute

University of Oxford

x
Figure 2.14: Streamlines caused by a doublet in a uniform ow.
The streamfunction is just the imaginary part of this function, namely = Uy 1 a2 + y2 , (2.73)

x2

and we see that the circle x2 + y 2 = a2 is indeed a streamline, with = 0. The resulting ow (with a = 1) is shown in Figure 2.13(ii).

The Circle Theorem may be viewed as an extension of the method of images to circular boundaries. In Example 2.15, we see that application of the Circle Theorem results in a singularity being inserted inside the cylinder. The ow due to the image singularity then repels the uniform ow, such that the circle |z | = a ends up being a streamline. In this case, the required singularity turns out to be a so-called doublet, with complex potential w(z ) = U a2 /z . The singularity at the origin is inside the obstacle and thus does not adversely aect the external ow. The full streamline pattern, including the doublet inside the cylinder, is shown in Figure 2.14. The Circle Theorem allows us to construct a velocity potential that shares the same singularities with the given external ow f (z ) and has the circle |z | = a as a streamline. However, it does not guarantee that the resulting potential is unique. Consider the ow due to a vortex, as illustrated in Figure 2.4. This is a velocity eld that has |z | = a as a streamline, has no singularities in |z | > a and decays to zero as |z | . Hence, we could superimpose the ow due to a vortex at the origin onto any ow outside a circular cylinder, without introducing any further singularities in the uid or disturbing the streamline at |z | = a. The most general ow produced by a inserting a circular

Part A Fluid Dynamics & Waves

(i) 0 < < 1
4 4

Draft date: 7 February 2010 219

(ii) = 1
4

(iii) > 1

2 2 2

4 4 2 0 2 4

4 4 2 0 2 4

4 4 2 0 2 4

Figure 2.15: Streamlines for uniform ow at speed U ow past a cylinder pf radius a with with circulation . (i) 0 < < 1; (ii) = 1; (iii) > 1, where = /4U a. cylinder into an external ow w(z ) = f (z ) is thus w(z ) = f (z ) + f a2 z i log z, 2 (2.74)

where is arbitrary, and represents the circulation about the cylinder. Example 2.16 Uniform ow past a circular cylinder with circulation
If we incorporate circulation in the complex potential (2.72) corresponding to uniform ow past a circular cylinder, we obtain w (z ) = U z + U a2 i log z. z 2 (2.75)

We can read o the streamfunction from the imaginary part of this function, namely = Uy 1 a2 x2 + y 2 log x2 + y 2 . 4 (2.76)

Hence the circle x2 + y 2 = a2 is still a streamline, with = (/2 ) log a. Any stagnation points in the ow satisfy 0 = u iv = dw U a2 i =U 2 , dz z 2z (2.77)

which can be rearranged to the quadratic equation z 2 2iaz a2 = 0, where for convenience we introduce the shorthand = The roots of equation (2.78) are given by z = i a 1 2. (2.80) . 4U a (2.79) (2.78)

220 OCIAM Mathematical Institute

p p Im z p

University of Oxford

p p B dz Re z

p p

i dz

Figure 2.16: Schematic of an obstacle B , with boundary B , subject to a pressure p.

When = 0, there is no circulation and the ow is as illustrated in Figure 2.13(ii), with stagnation points at z = a. As increases, the anticlockwise circulation causes the stagnation points to move upwards around the cylinder, as shown in Figure 2.15(i). When reaches the value 1, the two stagnation points coalesce at the top of the cylinder z = ia, as shown in Figure 2.15(ii). If > 1, then one stagnation point moves into the ow, as shown in Figure 2.15(iii); the other one is inside the cylinder. Now there is a region of uid, bounded by the separatrix through the stagnation point, that simply rotates around the cylinder, and is completely cut o from the external ow.

2.4

Blasius Theorem

Suppose we know the velocity potential w(z ) corresponding to ow past a given obstacle B , and we wish to determine the force exerted on the obstacle by the uid owing past. We will show how to achieve this for an arbitrarily-shaped obstacle. Although so far we have learnt how to determine w(z ) only when B is circular, we will nd that the Circle Theorem may be generalised to other shapes by using conformal mapping. Lemma 2.2 Suppose uid ows steadily past an obstacle B with simple closed boundary B . If gravity is neglected, the net force (Fx , Fy ) exerted on B by the uid (per unit length out of the plane) is given by Fx + iFy = i 2 dw dz
2

dz.

(2.81)

Proof We recall that the pressure is assumed to act in the inward normal direction, so

Part A Fluid Dynamics & Waves

Draft date: 7 February 2010 221

the net force experienced by the body B is given by (Fx , Fy ) =

B

pn ds =
B

p(dy, dx).

(2.82)

We translate the components (Fx , Fy ) of F into a complex scalar by writing Fx + iFy =

B

p(dy + idx) =
B

pi dz,

(2.83)

where dz = dx + idy . The directions of dz and i dz are illustrated in Figure 2.16. To evaluate the pressure from the complex potential, we turn to Bernoullis Theorem. Restricting our attention to steady ow and neglecting gravity, we substitute the velocity components from (2.46) into (2.9) to obtain dw p=P 2 dz
2

(2.84)

where P is a constant reference pressure. Equation (2.81) follows from substitution of (2.84) into (2.83); the constant term P in the pressure integrates to zero by Cauchys Theorem. Although equation (2.81) gives us formulae for the force components, it is very inconvenient to use it as it stands. The modulus signs mean that the integrand in (2.81) will in general be holomorphic nowhere. Hence, none of the standard tools of complex analysis is any use in evaluating the right-hand side of (2.81). Fortunately, there is a neat trick that eliminates the modulus signs in (2.81) and thus transforms the integrand into a meromorphic function. Theorem 2.3 Blasius Theorem Suppose uid ows steadily past an obstacle B with simple closed boundary B . If gravity is neglected, the net force (Fx , Fy ) exerted on B by the uid is given by Fx iFy = i 2 dw dz
2

dz.

(2.85)

Proof By taking the complex conjugate of (2.81), we obtain Fx iFy = i 2 dw dz

2

dz,

(2.86)

where dz = dx idy . By factorising the integrand in the form dw dz we can rewrite (2.86) as Fx iFy = i 2 dw dw i dz dz dz 2 dw dw. dz (2.88)
2

dw dw , dz dz

(2.87)

222 OCIAM Mathematical Institute

University of Oxford

Now we recall that B is supposed to be a streamline for the ow. Hence = Im w is constant on B and it follows that dw dw on B . We therefore obtain Fx iFy = i 2 dw dw, dz (2.89)

and equation (2.85) follows. (For an alternative proof of Blasius Theorem, see Acheson, section 4.10.) Example 2.17 Uniform ow past a circular cylinder We illustrate the use of Blasius Theorem for the ow considered in Example 2.15. Substituting the velocity potential (2.72), into equation (2.85), we nd the force components
Fx iFy = i 2 U
|z |=a

U a2 z2

dz.

(2.90)

This integral is easily evaluated using Cauchys Residue Theorem. There is just one pole, of order 4, at the origin. However, the Laurent expansion of the integrand, namely U 2 a4 2U a2 + U2 4 z z2 has no term of order 1/z , and the residue of the pole is therefore zero.

Example 2.17 shows that the net force exerted on a circular cylinder by a uniform ow is zero! This counter-intuitive result is known as DAlemberts Paradox. In practice, we would certainly expect a cylinder subject to a uniform ow to experience a nonzero force. The discrepancy between the theoretical prediction and experimental reality is now known to be due to our neglect of viscous eects. Example 2.18 Uniform ow past a circular cylinder with circulation
Next we apply Blasius Theorem to the ow considered in Example 2.16, with circulation included. Substituting the velocity potential (2.75), into equation (2.85), we nd the force components 2 i U a2 i Fx iFy = U 2 dz. (2.91) 2 |z|=a z 2z Again, there is just one pole, of order 4, at the origin. When we expand out the integrand, we easily nd that the coecient of 1/z is iU / , and we therefore deduce from Cauchys Residue Theorem that Fx iFy = iU . (2.92) Hence the so-called drag force Fx , parallel to the ow, is zero, while the lift force, transverse to the ow, is given by Fy = U . (2.93)

We can interpret the lift formula (2.93) as follows. If is positive (as in Figure 2.15), then the circulation opposes the uniform ow above the cylinder and reinforces it below. Hence the average speed along the bottom of the cylinder will be higher than that along the top. Bernoullis Theorem tells us that higher speed is associated with lower pressure,

Part A Fluid Dynamics & Waves

Draft date: 7 February 2010 223

and the cylinder therefore experiences a net downwards force. On the other hand, if is negative, then the circulation is in the clockwise sense and the cylinder experiences an upwards force. It may be shown that equation (2.93) applies to uniform ow past any obstacle, whether or not it is circular. Theorem 2.4 KuttaJoukowski Lift Theorem Consider steady uniform ow at speed U past an obstacle B , where there is a circulation about B but there are no singularities in the ow. Then the obstacle experiences a drag force D parallel to the ow and lift force L perpendicular to the ow given by D = 0, L = U . (2.94)

Proof Since there are assumed to be no singularities in the ow, dw/dz must be holomorphic outside B and must therefore have a Laurent expansion of the form dw b1 U ei + +O dz z 1 z2 as z . (2.95)

The rst term in (2.95) corresponds to the imposed uniform ow at innity, which we allow to approach at an arbitrary inclination to the x-axis. We recall from Blasius Theorem that the components of the force on B are given by (2.85). Since there are no singularities in the ow, we can use Cauchys Theorem to deform the contour of integration from B to a large circle C of radius R, as shown schematically in Figure 2.17(i): Fx iFy = i 2 dw dz
2

dz =

i 2

dw dz

dz.

(2.96)

We parametrise C using z = Rei and use (2.95) to obtain Fx iFy = i 2

2 0

U ei +

b1 ei +O R

1 R2

iRei d,

(2.97)

and, letting R , we nd that Fx iFy = 2b1 U ei . (2.98)

The drag and lift forces are dened to be the components parallel and perpendicular to the ow, as shown in Figure 2.17(ii). Simple trigonometry tells as that they are related to Fx and Fy by D + iL = (Fx + iFy ) ei , (2.99) and we deduce from (2.98) that D iL = 2b1 U. (2.100)

224 OCIAM Mathematical Institute

(i) Im z (ii)

University of Oxford

L R B Re z

Fy D Fx

Figure 2.17: (i) Schematic showing deformation of the boundary B of an obstacle B being deformed to a large circular contour of radius R. (ii) The relations between the drag and lift forces and the components Fx and Fy . To evaluate the constant b1 , we note that dw dz = dz
2 0

U ei +

b1 ei +O R

1 R2

iRei d = 2 ib1 .

(2.101)

On the other hand, the real and imaginary parts of the left-hand side give us dw dz = dz (u iv ) (dx +idy ) =
C C

(u dx + v dy )+i
C

(u dy v dx) = +iQ, (2.102)

where is the circulation around C , while Q is the ux through C . (These are equal to the jumps in the values of and respectively as the closed circuit C is traced.) The ux must be zero, as there are not supposed to be any sources or sinks in the ow, and we deduce from (2.101) and (2.102) that b1 = Hence equation (2.100) becomes D iL = iU , and the result (2.94) follows. (2.104) i . 2 (2.103)

2.5
2.5.1

Background material
Integral theorems

Greens Theorem Let the functions P (x, y ) and Q(x, y ) be continuously dierentiable in the region D bounded by the simple closed curve D (traversed in the anticlockwise sense) with outward unit normal n. Then the Divergence Theorem restricted

Part A Fluid Dynamics & Waves to two dimensions tells us that (P, Q) dxdy
D D

Draft date: 7 February 2010 225

(P, Q) n ds,

(2.105)

which may be rewritten as

D

P Q + x y

dxdy
D

(P dy Q dx).

(2.106)

Fundamental Theorem of Calculus for line integrals ferentiable scalar function (x, y ), we have dx =
C C

For any continuously dif-

dx + dy x y

= (b) (a)

(2.107)

where C is a smooth path joining the two points with position vectors x = a and x = b.

2.5.2

Plane polar coordinates

(r, ) are related to (x, y ) by x r cos , y r sin . (2.108)

Denition

The corresponding unit basis vectors are er i cos + j sin , e i sin + j cos , (2.109)

in terms of the usual Cartesian basis vectors {i, j }. Grad, div and curl Given a two-dimensional scalar eld (r, ) and a vector eld u(r, ) = ur (r, )er + u (r, )e , we have u 2 u (k) 1 er + e , r r 1 1 u (rur ) + , r r r 2 1 1 2 + + , r2 r r r2 2 1 ur 1 (ru ) k, r r r 1 er e , r r (2.110) (2.111) (2.112) (2.113) (2.114)

where k i j er e is a unit vector perpendicular to the (x, y )-plane.

226 OCIAM Mathematical Institute

University of Oxford

2.5.3

Complex analysis
Given a complex number z = x + iy , we dene:

Standard notation

Re z = x, the real part of z ; Im z = y , the imaginary part of z ; z = x iy , the complex conjugate of z ; |z | = x2 + y 2 , the modulus of z .

Note that the complex conjugate commutes with all standard mathematical operations, that is z 1 + z 2 z1 + z2 , and so forth. Derivatives and holomorphic functions A function f (z ) is holomorphic (or analytic ) if it has a well-dened derivative df f (z + h) f (z ) = lim (2.116) h0 dz h which is independent of the direction in which the complex variable h approaches zero. Any holomorphic function has derivatives of all orders, and a power series expansion, with positive radius of convergence. A complex-valued function may be decomposed into its real and imaginary parts, i.e. f (z ) (x, y ) + i (x, y ). Then f (z ) is holomorphic if and only if the rst partial derivatives of and are continuous and satisfy the CauchyRiemann equations = , = . (2.117) x y y x Laurent expansions If a function f (z ) is holomorphic on an annulus, say < |z a| < R, then it has a convergent Laurent expansion

z 1 z2 z1 z2 ,

ez ez ,

(2.115)

f (z ) =
n=

bn (z a)n .

(2.118)

The singularity at z = a is 1. removable if bn = 0 for all n < 0; 2. a pole of order N if bN = 0 but bn = 0 for all n < N ; 3. an essential singularity if there does not exist an N such that bn = 0 for all n < N . In case 2, the residue Res (f (z ); a) of the pole is the coecient b1 of (z a)1 in the Laurent expansion about z = a.

Part A Fluid Dynamics & Waves

Draft date: 7 February 2010 227

Multifunctions Functions such as log z and z for non-integer cannot be dened both uniquely and continuously when z is complex. For example, log z = log r + i, (2.119)

where (r, ) are the usual plane polar coordinates, also known as the modulus and argument of z . To dene log z uniquely, we have to choose a branch of . For example, if we choose to lie in the range [0, 2 ), then log z is uniquely dened, but discontinuous across the branch cut at = 0, i.e. along the postive real axis. By writing z r ei , (2.120) we can also dene z uniquely once we have chosen a branch for . For more complicated multifunctions, see Priestley Chapter 9. Contour integrals A closed curve C in the complex plane may be parametrised as z = ( ), where the real parameter lies in some interval [0, c] and (0) = (c). Then we dene the contour integral of f (z ) around C as
c

f (z ) dz :=
C 0

f ( )

d d. d

(2.121)

Cauchys Theorem then

If f (z ) is holomorphic inside and on the simple close curve C , f (z ) dz 0.

C

(2.122)

lies inside the contour C , and Deformation Theorem Suppose that the contour C , on C and on the region contained between that the function f (z ) is holomorphic on C without changing the contour them. Then the contour C may be deformed onto C integral of f (z ): f (z ) dz
C C

f (z ) dz.

(2.123)

Cauchys Residue Theorem Suppose f (z ) is holomorphic on and inside a contour C , except for a nite number of poles at the points a1 , . . . , an inside C . Then the contour integral of f (z ) around C is equal to 2 i times the sum of the residues of all the poles inside C , i.e.
n

f (z ) dz 2 i
C k=1

Res (f (z ); ak ) .

(2.124)