Flow Around a Cylinder

Lecture Notes for

Jeffrey R. Chasnov
 The Hong Kong University of Science and Technology
Department of Mathematics
Clear Water Bay, Kowloon
Hong Kong

Copyright © 2022 by Jeffrey Robert Chasnov

This work is licensed under the Creative Commons Attribution 3.0 Hong Kong License. To view
a copy of this license, visit http://creativecommons.org/licenses/by/3.0/hk/ or send a letter to
Creative Commons, 171 Second Street, Suite 300, San Francisco, California, 94105, USA.
Preface
View the Deep Dive video Flow Around a Cylinder on YouTube

These are my lecture notes for Mathematics for Engineers: The Capstone Course. Stu-
dents will learn how to solve for the steady and unsteady two-dimensional flow around
an infinite cylinder. Before students take this course, they should have some basic knowl-
edge of single-variable calculus, vector calculus, differential equations, matrix algebra and
numerical methods. Students should also be able to program in Matlab.
I have divided these notes into chapters called Lectures, with each Lecture correspond-
ing to a video on Coursera. I have also uploaded all my Coursera videos to YouTube, and
links are placed at the top of each Lecture.
The course is divided into three weeks. The first week introduces the Navier-Stokes
equations and defines the flow geometry. The second week details the numerical methods
we use to compute steady flow solutions. And the third week details the methods we use
to compute unsteady flow solutions and the Kármán vortex street.
After every Lecture of Week One, there are problems to solve. There are problems
in Week Two and Week Three also, but these weeks culminate in a substantial Matlab
program to write. Solutions to the problems and Learner Templates for the Matlab
programs can be found in the Appendix.

Jeffrey R. Chasnov
Hong Kong
Feb 2022
Contents

I Governing Equations 1

1 Navier-Stokes equations 2

2 Vorticity equation 4

3 Geometry of the flow 6

4 Two-dimensional flow 8

5 Stream function 10

6 Streamlines 12

7 Reynolds number 14

8 Log-polar coordinates 16

II Steady Flows 18

9 Finite difference method 19

10 Iteration equations 21

11 Free-stream boundary conditions 23

12 Cylinder boundary conditions 25

13 Summary of the boundary conditions 27

14 Matlab program (steady) (Part A) 29

15 Matlab program (steady) (Part B) 31

III Unsteady Flows 33

16 Periodic boundary conditions 34

17 Finite difference equations 36

18 Stream-function computation 38

19 Stream-function boundary conditions 40

20 Vorticity computation 42

iv
 CONTENTS v

21 Matlab program (unsteady) (Part A) 44

22 Matlab program (unsteady) (Part B) 46

Problem solutions and Matlab learner templates 48

 Week I

Governing Equations

In this week’s lectures, we introduce the Navier-Stokes equations and the flow around an infinite
circular cylinder. Our flow field will be two dimensional and we write the Navier-Stokes equations
using a scalar vorticity and stream function. To take advantage of the circular symmetry, we
formulate the equations in polar coordinates. A further simplification of the Laplacian operator is
made by defining log-polar coordinates.

1
Lecture 1 | Navier-Stokes
equations
View this lecture on YouTube

For an incompressible fluid, the continuity and Navier-Stokes equations are written in
vector notation as

∂𝑢 1
∇ · 𝑢 = 0, + (𝑢 · ∇)𝑢 = − ∇p + ν∇2 𝑢.
∂t ρ

The first equation states that the velocity field is a divergence-free (or solenoidal) vector
field. It is an expression of the incompressibility condition—that the density of the fluid
is everywhere constant. The second equation expresses Newton’s law F = ma applied to
a fluid, and its derivation is usually presented in a first course on Fluid Mechanics.
Some students may have trouble deciphering the compact vector notation. To illustrate
how to unpack these equations, we use Cartesian coordinates and write

𝑢 = u( x, y, z)𝑖 + v( x, y, z)𝑗 + w( x, y, z)𝑘,

where the three components also depend on the time t. The continuity equation expands
to
∂u ∂v ∂w
+ + = 0;
∂x ∂y ∂z
and the Navier-Stokes equation expands into the three-component equations
 2
∂ u ∂2 u ∂2 u
  
∂u ∂u ∂u ∂u 1 ∂p
+ u +v +w =− +ν + 2+ 2 ,
∂t ∂x ∂y ∂z ρ ∂x ∂x2 ∂y ∂z
 2
∂2 v ∂2 v
  
∂v ∂v ∂v ∂v 1 ∂p ∂ v
+ u +v +w =− +ν + + ,
∂t ∂x ∂y ∂z ρ ∂y ∂x2 ∂y2 ∂z2
 2
∂ w ∂2 w ∂2 w
  
∂w ∂w ∂w ∂w 1 ∂p
+ u +v +w =− +ν + + .
∂t ∂x ∂y ∂z ρ ∂z ∂x2 ∂y2 ∂z2

The goal of our capstone course is to solve numerically a two-dimensional version of these
equations for the flow around a cylinder.

2
 WEEK I. GOVERNING EQUATIONS 3

Problems for Lecture 1

1. Plane Couette flow consists of an incompressible fluid flowing between two infinite
plates separated by a distance d. The lower plate is stationary, and the upper plate is
moving to the right with velocity U. The pressure p is constant. Choose a coordinate
system so that the flow is in the x-direction and the y-axis is perpendicular to the two
plates. Find the solution for the velocity field of the form 𝑢( x, y, z) = u(y)𝑖.

2. Channel flow, or Poiseuille flow, consists of an incompressible fluid flowing between

two infinite plates separated by a distance d, but with both plates stationary. Here, the
fluid flows between the two plates in the direction of a constant pressure gradient. Choose
a coordinate system so that the fluid flow is in the x-direction and the y-axis is perpendic-
ular to the two plates. Find the solution for the velocity field of the form 𝑢( x, y, z) = u(y)𝑖,
with p = p( x ), dp/dx = − G and G is a positive constant.

3. Pipe flow consists of an incompressible fluid flowing through a pipe of circular cross-
section radius R, with a constant pressure gradient along the pipe length. Choose a
coordinate system so that the pressure gradient is in the x-direction and the y- and z-axes
are perpendicular to the sides of the pipe. Find the solution for the velocity field of the
form 𝑢 = u(y, z)𝑖, with p = p( x ), dp/dx = − G and G is a positive constant. Use polar
coordinates.

Solutions to the Problems

Lecture 2 | Vorticity equation
View this lecture on YouTube

The equation for the vorticity, 𝜔 = ∇ × 𝑢, may be found by taking the curl of the Navier-
Stokes equation; that is
   
∂𝑢 1 2
∇× + (𝑢 · ∇)𝑢 = ∇ × − ∇p + ν∇ 𝑢 .
∂t ρ

Examining each term, we have

 
∂𝑢 ∂ ∂𝜔 n o
∇× = (∇ × 𝑢) = and ∇ × ν∇2 𝑢 = ν∇2 (∇ × 𝑢) = ν∇2 𝜔.
∂t ∂t ∂t

And because the curl of a gradient is zero,

∇ × {−∇p/ρ} = 0.

To simplify the curl of the convection term, one writes

 
1
∇ × {(𝑢 · ∇)𝑢} = ∇ × ∇ ( 𝑢2 ) + 𝜔 × 𝑢 = ∇ × (𝜔 × 𝑢) ;
2

and the vorticity equation becomes

∂𝜔
+ ∇ × (𝜔 × 𝑢) = ν∇2 𝜔.
∂t

An alternative form expands the convection term to obtain

∂𝜔
+ (𝑢 · ∇)𝜔 = (𝜔 · ∇)𝑢 + ν∇2 𝜔.
∂t

Compared to the Navier-Stokes equation for 𝑢, there is an extra term on the right-hand
side of this equation, (𝜔 · ∇)𝑢. This term is called the vortex stretching term and is often
used to explain the energy cascade in three-dimensional turbulence.

4
 WEEK I. GOVERNING EQUATIONS 5

Problems for Lecture 2

1. Prove the following equalities:

a) ∇ × {(𝑢 · ∇)𝑢} = ∇ × (𝜔 × 𝑢);

b) ∇ × (𝜔 × 𝑢) = (𝑢 · ∇)𝜔 − (𝜔 · ∇)𝑢.

You can use the facts that the curl of a gradient and the divergence of a curl are equal to
zero, and the general vector identity

∇ × (𝑢 × 𝑣 ) = 𝑢(∇ · 𝑣 ) − 𝑣 (∇ · 𝑢) + (𝑣 · ∇)𝑢 − (𝑢 · ∇)𝑣.

You will also need to use 𝜔 = ∇ × 𝑢 and ∇ · 𝑢 = 0. However, you will need to prove that

1
𝑢 × (∇ × 𝑢) = ∇(𝑢 · 𝑢) − (𝑢 · ∇)𝑢.
2

Solutions to the Problems

Lecture 3 | Geometry of the flow
View this lecture on YouTube

We consider flow around an infinite cylinder with cross-sectional radius R. We define

our x-y coordinates to lie in the cross-sectional plane, and orient the axes so that the free-
stream velocity, 𝑢 f = U𝑖, is in the x-direction. The geometry of the flow in the plane is
shown in the figure below.

For the nonturbulent flows considered here, we assume that the velocity field vectors
lie in the x-y plane and have no z-component, and that the flow field is independent of the
z coordinate. And for steady flows only, we assume that the flow field in the bottom half
of the plane (y < 0) is a mirror image of the flow field in the top half of the plane (y > 0).
That is, with velocity field given by 𝑢 = u( x, y)𝑖 + v( x, y)𝑗, we assume u( x, −y) = u( x, y)
and v( x, −y) = −v( x, y).

6
 WEEK I. GOVERNING EQUATIONS 7

Problems for Lecture 3

1. In the cross-sectional plane, express the cylinder boundary in Cartesian and polar
coordinates. Express the free-stream velocity in polar coordinates.

Solutions to the Problems

Lecture 4 | Two-dimensional flow
View this lecture on YouTube

The vorticity equation is given by

∂𝜔
+ (𝑢 · ∇)𝜔 = (𝜔 · ∇)u + ν∇2 𝜔,
∂t

where 𝜔 = ∇ × 𝑢. For nonturbulent flow past an infinite cylinder, one expects that the
velocity field will lie in the cross-sectional plane of the cylinder and be independent of the
z coordinate.
To best model the boundary conditions on the cylinder, we will use polar coordinates
and write
𝑢 = ur (r, θ )r̂ + uθ (r, θ )θ̂,

where the components of the velocity may also depend on time.

In polar coordinates, the vorticity field becomes
 
1 ∂ ∂ur
𝜔 = ∇×𝑢 = (ruθ ) − 𝑘,
r ∂r ∂θ

and has only a single component in the z-axis direction. We can then define a scalar
vorticity field by  
1 ∂ ∂ur
ω= (ruθ ) − .
r ∂r ∂θ
And since
∂𝑢
(𝜔 · ∇)𝑢 = ω = 0,
∂z
and
∂2
   
∂ 1 ∂ 1
2 ∂ ∂
∇ = r̂ + θ̂ , ∇ = 2 r r + 2 ,
∂r r ∂θ r ∂r ∂r ∂θ
the vorticity vector equation in polar coordinates becomes the scalar equation

∂2 ω
   
∂ω ∂ω 1 ∂ω ν ∂ ∂ω
+ ur + uθ = 2 r r + .
∂t ∂r r ∂θ r ∂r ∂r ∂θ 2

8
 WEEK I. GOVERNING EQUATIONS 9

Problems for Lecture 4

1. With
𝜔 = ω ( x, y)𝑘, 𝑢 = u( x, y)𝑖 + v( x, y)𝑗,

show by direct calculation that

 
∂ω ∂ω
∇ × (𝜔 × 𝑢) = u +v 𝑘.
∂x ∂y

2. Define the scalar vorticity in Cartesian coordinates.

3. Write the scalar vorticity equation in Cartesian coordinates.

Solutions to the Problems

Lecture 5 | Stream function
View this lecture on YouTube

The incompressibility condition in polar coordinates is given by

 
1 ∂ ∂u
∇·𝑢 = (rur ) + θ = 0.
r ∂r ∂θ

To automatically satisfy this condition, we make use of the equality of mixed partials and
define a stream function ψ by

∂ψ ∂ψ
rur = , uθ = − .
∂θ ∂r

The definition of the scalar vorticity then becomes

 
1 ∂ ∂ur
ω= (ruθ ) −
r ∂r ∂θ
∂2 ψ
   
1 ∂ ∂ψ
=− 2 r r + 2 ,
r ∂r ∂r ∂θ

which is simply the equation

∇2 ψ = − ω

in polar coordinates.

10
 WEEK I. GOVERNING EQUATIONS 11

Problems for Lecture 5

1. Define the stream function in Cartesian coordinates.

2. Using Cartesian coordinates, write the scalar vorticity in terms of the stream function.

3. In general, a solenoidal vector field 𝑢 that satisfies ∇ · 𝑢 = 0 admits a vector potential

𝐴 such that 𝑢 = ∇ × 𝐴. Using Cartesian coordinates in two dimensions, show that the
stream function can be identified as the third component of a vector potential 𝐴.

Solutions to the Problems

Lecture 6 | Streamlines
View this lecture on YouTube

The streamlines in a steady flow show the direction of fluid motion. A streamline is
everywhere tangent to the velocity field of the flow (see figure).

With infinitesimal vector displacement d𝑟 = drr̂ + rdθ θ̂ along a streamline, and with the
velocity field parallel to d𝑟 so that their cross-product is zero, we have

0 = 𝑢 × d𝑟 = (ur r̂ + uθ θ̂ ) × (drr̂ + rdθ θ̂ ) = (−uθ dr + rur dθ ))𝑘.

But since the stream function ψ is defined by

∂ψ ∂ψ
rur = , uθ = − ,
∂θ ∂r

we have determined that

 
∂ψ ∂ψ
0= dr + dθ 𝑘 = dψ𝑘,
∂r ∂θ

or dψ = 0 along streamlines. The streamlines are therefore curves of constant ψ, and

a plot of the contour lines of the stream function thus provides us an excellent visual
representation of a steady two-dimensional flow.

12
 WEEK I. GOVERNING EQUATIONS 13

Problems for Lecture 6

1. Use Cartesian coordinates to show that the stream function is constant along stream-
lines.

Solutions to the Problems

Lecture 7 | Reynolds number
View this lecture on YouTube

The governing equations for a two-dimensional flow in terms of the stream function and
scalar vorticity field are given by
 
2 ∂ω 1 ∂ψ ∂ω ∂ψ ∂ω
∇ ψ = −ω, + − = ν∇2 ω,
∂t r ∂θ ∂r ∂r ∂θ

where ∇2 is the two-dimensional Laplacian in polar coordinates.

We now consider flow past an infinite circular cylinder of radius R, with free-stream
velocity 𝑢 = U𝑖. The units of our problem are time t and length l, and [ R] = l, [U ] = lt−1 ,
and the various terms in the equations have units
   
∂ ∂
= t −1 , = l −1 , [∇2 ] = l −2 , [ ν ] = l 2 t −1
∂t ∂r
[ ψ ] = l 2 t −1 , [ ω ] = t −1 .

If we nondimensionalize using R and U, the net result is the replacement of the viscosity
ν by a dimensionless grouping of parameters, that is,

ν
ν→ .
UR

It is standard here to define the Reynolds number Re in terms of the diameter of the
cylinder instead of the radius, so that with

2UR
Re = ,
ν

the nondimensional governing equations (with all variables now dimensionless) become
 
∂ω 1 ∂ψ ∂ω ∂ψ ∂ω 2 2
∇2 ψ = −ω, + − = ∇ ω.
∂t r ∂θ ∂r ∂r ∂θ Re

14
 WEEK I. GOVERNING EQUATIONS 15

Problems for Lecture 7

1. From the nondimensional equations for the stream function and the scalar vorticity,
determine a single equation for the stream function when Re = 0.

Solutions to the Problems

Lecture 8 | Log-polar coordinates
View this lecture on YouTube

The recurring factor r∂/∂r in the polar coordinate Laplacian is awkward to discretize
and we look for a change of variables r = r (ξ ), where

∂ ∂
r = .
∂r ∂ξ

Now,
∂ dr ∂
= ,
∂ξ dξ ∂r
so that we require
dr
= r.
dξ
This simple differential equation can be solved if we take as our boundary condition ξ = 0
when r = 1, corresponding to points lying on the boundary of the circular cylinder. The
solution of the differential equation is therefore given by

r = eξ .

The Laplacian in these log-polar coordinates then becomes

∂2
   
1 ∂ ∂
∇2 = r r + 2
r2 ∂r ∂r ∂θ
 2 2 
∂ ∂
= e−2ξ + 2 ;
∂ξ 2 ∂θ

and the governing equations become

∂2 ψ ∂2 ψ
 
+ 2 = −e2ξ ω,
∂ξ 2 ∂θ
∂2 ω ∂2 ω
   
∂ω ∂ψ ∂ω ∂ψ ∂ω 2 −2ξ
+ e−2ξ − = e + .
∂t ∂θ ∂ξ ∂ξ ∂θ Re ∂ξ 2 ∂θ 2

16
 WEEK I. GOVERNING EQUATIONS 17

Problems for Lecture 8

1. Determine the governing equations for a steady flow in log-polar coordinates. Try to
write the equations as neatly as possible.

Solutions to the Problems

 Week II

Steady Flows

In this week’s lectures, we show how to solve the stream function and scalar vorticity equations to
obtain steady solutions. These steady solutions for the fields do not depend on time. We discretize
the equations using a second-order finite difference method and the SOR iteration method. Low
Reynolds number iterations can be initialized with zero flow fields except for the free-stream con-
dition on the stream function. We derive a key boundary condition for the scalar vorticity on the
cylinder.

For the computational project, students will compute and plot the stream function for Re = 10.
Experiments have shown that the steady solutions are stable for Re < 46. The steady solutions still
exist for higher Reynolds numbers, but they are unstable and not observed in Nature.

18
Lecture 9 | Finite difference
method
View this lecture on YouTube

A finite difference approximation requires a grid in (ξ, θ ) space. With velocity field
𝑢 = u( x, y)𝑖 + v( x, y)𝑗, we assume that u is even in y and v is odd; that is,

u( x, −y) = u( x, y), v( x, −y) = −v( x, y).

If a function is even in a variable, then its partial derivative in that variable is odd, and
vice-a-versa. Since u = ∂ψ/∂y and ω = ∂v/∂x − ∂u/∂y, both the stream function and
scalar vorticity must be odd in y:

ψ( x, −y) = −ψ( x, y), ω ( x, −y) = −ω ( x, y).

We need then define a grid only for the upper half of the x-y plane, and take as our
boundary conditions at y = 0 the values ψ = ω = 0.
In log-polar coordinates, we can then define our grid as 0 ≤ ξ ≤ ξ max and 0 ≤ θ ≤ π,
and our computational domain forms a rectangle without any holes. The sides of this
rectangle correspond to the cylinder boundary (ξ = 0), the free stream (ξ = ξ max ), the
midline behind the cylinder (θ = 0), and the midline in front of the cylinder (θ = π).
We discretize the computational domain using square grid cells, and write

ξ i = (i − 1)h, i = 1, 2, . . . , n; θ j = ( j − 1)h, j = 1, 2, . . . , m,

where n and m are the number of grid points (including boundary points) in the ξ and θ
directions, and h is the side length of a grid cell. Because 0 ≤ θ ≤ π, the value of h must
satisfy h = π/(m − 1), and the maximum value of ξ is given by ξ max = (n − 1)π/(m − 1).
The radius of the computational domain is therefore given by eξ max , which is to be com-
pared to a cylinder radius of one. The choice n = m yields eξ max ≈ 23, and this will be
large enough for the relatively low Reynolds number solutions we seek here.

19
 WEEK II. STEADY FLOWS 20

Problems for Lecture 9

1. An even function of x satisfies f (− x ) = f ( x ). An odd function of x satisfies
f (− x ) = − f ( x ). Prove the following symmetry theorems:

a) If f ( x ) is an even function of x, then f 0 ( x ) is an odd function of x.

b) If f ( x ) is an odd function of x, then f 0 ( x ) is an even function of x.

Solutions to the Problems

Lecture 10 | Iteration equations
View this lecture on YouTube

The governing equations for a steady flow field may be written as

∂2 ψ ∂2 ψ ∂2 ω ∂2 ω
     
Re ∂ψ ∂ω ∂ψ ∂ω
− + 2 = e2ξ ω, − + 2 = − .
∂ξ 2 ∂θ ∂ξ 2 ∂θ 2 ∂ξ ∂θ ∂θ ∂ξ

We define ψi,j = ψ(ξ i , θ j ) and ωi,j = ω (ξ i , θ j ) and use second-order centered difference
approximations for all the derivatives. Application of the SOR method (Numerical Methods
for Engineers, Lecture 66) results in the equations

k +1 k rψ  k 
ψi,j = (1 − rψ )ψi,j + ψi+1,j + ψik−1,j + ψi,j
k
+
k 2 2ξ i k
1 + ψi,j−1 + h e ωi,j ,
4  
k +1 k r ω k k k k Re k
ωi,j = (1 − rω )ωi,j + ωi+1,j + ωi−1,j + ωi,j+1 + ωi,j−1 + f ,
4 8 i,j

where
     
k
f i,j = ψik+1,j − ψik−1,j ωi,j
k
+1 − ω k
i,j−1 − ψ k
i,j+1 − ψ k
i,j−1 ω k
i +1,j − ω k
i −1,j .

Because these equations are nonlinear, iterations can easily become unstable. To maintain
stability, the relaxation parameters, rψ and rω , may need to be less than one. Numerical
experimentation may be necessary to obtain the best trade-off between computationally
stability and computational speed. In addition, solutions at higher Reynolds numbers are
best initialized by solutions at slightly lower Reynolds numbers.
The convergence of the iterations need to be monitored. We define

εkψ+1 = max ψi,j

k +1 k
− ψi,j , εkω+1 = max ωi,j
k +1 k
− ωi,j .
i,j i,j

Iterations can be stopped when the values of εkψ+1 and εkω+1 are less than some pre-defined
error tolerance, say 10−8 .

21
 WEEK II. STEADY FLOWS 22

Problems for Lecture 10

1. Derive the SOR finite difference equations for

a) the stream function equation,

∂2 ψ ∂2 ψ
 
− + 2 = e2ξ ω;
∂ξ 2 ∂θ

b) the scalar vorticity equation,

∂2 ω ∂2 ω
   
Re ∂ψ ∂ω ∂ψ ∂ω
− + = − .
∂ξ 2 ∂θ 2 2 ∂ξ ∂θ ∂θ ∂ξ

Solutions to the Problems

Lecture 11 | Free-stream boundary
conditions
View this lecture on YouTube

The dimensionless free-stream velocity field is given by 𝑢 = 𝑖, which in polar coordi-

nates is
ˆ
u = cos θr̂ − sin θ 𝜃.

Since rur = ∂ψ/∂θ and uθ = −∂ψ/∂r, the stream function satisfies the free-stream condi-
tions
∂ψ ∂ψ
= r cos θ, = sin θ.
∂θ ∂r
The solution of these equations can be seen to be

ψ(r, θ ) = r sin θ,

which also satisfies our boundary conditions ψ = 0 when θ = 0 and π. On our log-polar
coordinate grid, then, the free-stream boundary condition at ξ = ξ n is given by

ψn,j = eξ n sin θ j .

The free-stream flow field has zero vorticity, and the simplest, effective boundary condi-
tion on the scalar vorticity is
ωn,j = 0.

23
 WEEK II. STEADY FLOWS 24

Problems for Lecture 11

1. Instead of assuming the scalar vorticity to be zero in the free stream, that is,

ωn,j = 0,

another option is to take the normal derivative to be zero, that is,

∂ω
= 0.
∂ξ (n,j)

From second-order Taylor series expansions of ωn−1,j and ωn−2,j , derive the corresponding
boundary condition on ωn,j .

Solutions to the Problems

Lecture 12 | Cylinder boundary
conditions
View this lecture on YouTube

Boundary conditions on the cylinder are determined from the no-penetration condition,
ur = 0, and no-slip condition, uθ = 0. And since rur = ∂ψ/∂θ and uθ = −∂ψ/∂r, the
stream function is constant and its normal derivative is zero. Since we know ψ = 0 at
y = 0, the cylinder boundary conditions are

∂ψ
ψ1,j = 0, = 0.
∂ξ (1,j)

The normal derivative boundary condition on the stream function will become a boundary
condition on the vorticity. The scalar vorticity everywhere satisfies

∂2 ψ ∂2 ψ
 
ω = −e−2ξ + 2 ,
∂ξ 2 ∂θ

and on the cylinder ξ = 0 and ψ is independent of θ so that

∂2 ψ
ω1,j = − .
∂ξ 2 (1,j)

To obtain a boundary condition on the vorticity, we Taylor series expand the stream func-
tion one and two grid points away from the cylinder surface:

∂ψ 1 ∂2 ψ 1 ∂3 ψ
ψ2,j = ψ1,j + h + h2 + h3 + O( h4 ),
∂ξ (1,j) 2 ∂ξ 2 (1,j) 6 ∂ξ 3 (1,j)
∂ψ ∂2 ψ 4 ∂3 ψ
ψ3,j = ψ1,j + 2h + 2h2 + h3 + O( h4 ).
∂ξ (1,j) ∂ξ 2 (1,j) 3 ∂ξ 3 (1,j)

Since the stream function and its normal derivative are zero on the cylinder surface, these
Taylor series expansions reduce to

1 1 ∂3 ψ 4 ∂3 ψ
ψ2,j = − h2 ω1,j + h3 + O( h4 ), ψ3,j = −2h2 ω1,j + h3 + O( h4 ).
2 6 ∂ξ 3 (1,j) 3 ∂ξ 3 (1,j)

After multiplying the first equation by eight and subtracting the second equation, we
obtain
8ψ2,j − ψ3,j = −2h2 ω1,j + O(h4 );

and the boundary condition on the vorticity, accurate to second-order, is given by

1
ω1,j = (ψ3,j − 8ψ2,j ).
2h2

25
 WEEK II. STEADY FLOWS 26

Problems for Lecture 12

1. The cylinder boundary condition on the scalar vorticity may be rewritten in terms of
the stream function as
∂2 ψ 1
= (8ψ2,j − ψ3,j ).
∂ξ 2 (1,j) 2h2

This boundary condition was derived using

∂ψ
ψ1,j = 0, = 0.
∂ξ (1,j)

A stream function that is quadratic in ξ near the cylinder surface satisfies the above two
conditions. Use it to test the validity of the derived boundary condition.

Solutions to the Problems

Lecture 13 | Summary of the
boundary conditions
View this lecture on YouTube

We collect all the boundary conditions and summarize them here:

ξ = 0, 0≤θ≤π: ψ1,j = 0, ω1,j = (ψ3,j − 8ψ2,j )/2h2 ;

ξ = ξ max , 0≤θ≤π: ψn,j = eξ n sin θ j , ωn,j = 0;
θ = 0, 0 ≤ ξ ≤ ξ max : ψi,1 = 0, ωi,1 = 0;
θ = π, 0 ≤ ξ ≤ ξ max : ψi,m = 0, ωi,m = 0.

27
 WEEK II. STEADY FLOWS 28

Problems for Lecture 13

1. Consider a grid with n and m gridpoints in the ξ and θ directions, respectively. Suppose
the SOR iteration begins with the flow fields ψ = ω = 0 everywhere except on the free-
stream boundary, where
ψn,j = eξ n sin θ j .

How many SOR iterations does it take for ω to obtain its first nonzero value?

Solutions to the Problems

Lecture 14 | Matlab program
(steady) (Part A)
View this lecture on YouTube

Successful Matlab programs will (1) calculate the stream function and scalar vorticity for
the two-dimensional steady flow around a cylinder (simulation code), and; (2) visualize
the solution (graphics code). We will provide students with a skeleton of the simulation
code, and a finished graphics code. The solution for Re = 10 will be graded and students
can perform additional calculations at other Reynolds numbers, if they so desire.
We have written the simulation code as a function, with the stream function psi and
scalar vorticity omega as outputs, that is,

function [psi, omega] = flow_around_cylinder_steady

To make the calculation fast, we use only 101 grid points in both the ξ and θ direc-
tions (1012 = 10,201 total grid points). The iteration is to be stopped when the absolute
value of the change in omega or psi after one iteration is less than the error tolerance
delta=1.e-08 at every grid point.
The structure of the simulation code is as follows:
1. Define the grid.
2. Initialize the flow fields, including the free-stream boundary condition.
3. Set the relaxation parameters, error tolerance and any extra variables.
4. Perform the main SOR iteration loop:
(a) Loop over all the grid points of the psi equation;
(b) Set the omega boundary condition on the cylinder;
(c) Loop over all the grid points of the omega equation;
(d) Test for convergence of the fields, and continue the iteration or end the loop.
5. Plot the contours of the stream function.

29
 WEEK II. STEADY FLOWS 30

Problems for Lecture 14

1. Using the SOR method, compute the steady solution for the stream function and scalar
vorticity when Re = 10.

Solutions to the Problems

Lecture 15 | Matlab program
(steady) (Part B)
View this lecture on YouTube

The graphics code that we provide plots the stream function for Re = 10. The stream
function computation is on a grid in log-polar coordinates and the stream function needs
to be plotted in Cartesian coordinates. Apart from the graphics itself, we will need to
interpolate the psi values on the ξ-θ grid onto an x-y grid.
We make use of the Matlab functions meshgrid.m and interp2.m to interpolate the
stream function onto the Cartesian grid points xi_i and theta_i. The Cartesian grid
is created using meshgrid.m and is given by the two arrays X and Y. The interpolation
points in the ξ-θ space are given by

xi_i=0.5*log(X.^2+Y.^2);
theta_i=wrapTo2Pi(atan2(Y,X));

The graphics code is written as a function with stream function psi as input, that is,

function plot_Re10(psi)

The structure of the graphics code is as follows:

1. Define the ξ-θ grid.
2. Define the x-y grid.
3. Construct interpolation points.
4. Interpolate the values of psi onto the Cartesian grid.
5. Scale the interpolated values of psi for a better plot.
6. Set the color map.
7. Plot the color contours.
8. Set the contour line values.
9. Plot the contour lines.
10. Draw a black circle to represent the cylinder.
11. Beautify the plot.

31
 WEEK II. STEADY FLOWS 32

Problems for Lecture 15

1. Study the graphics code and try to understand it. You may need to Google some
unfamiliar Matlab functions.

Solutions to the Problems

 Week III

Unsteady Flows

In this week’s lectures, we will solve for a time-dependent stream function and scalar vorticity.
Our goal is to simulate the famous Kármán street, where periodic vortices are shed from the top
and bottom of the cylinder. Applying periodic boundary conditions in the polar angle θ, we use
a second-order finite difference method to discretize the equations in space. The stream-function
equation is a system of linear equations and can be solved by a direct method using the Matlab
backslash operator. To increase the speed of the computation, we make use of the LU decomposition.
The time integration of the vorticity equation is performed using the Matlab function ode23.m.

For the computational project, students will flesh out the skeleton of a code that time-advances the
vorticity field. Experiments have shown that the steady solution becomes unstable for Re < 46
and we ask students to perform a calculation at Re = 60. After building a program that passes the
Matlab Grader, students can then write code to create a movie file of the Kármán street.

33
Lecture 16 | Periodic boundary
conditions
View this lecture on YouTube

In the unsteady flow, vortices are shed above and below the cylinder. We will need to
solve for the flow over the entire range of the polar angle, 0 ≤ θ < 2π.
We implement periodic boundary conditions. Let m be the total number of grid points
in the θ angle. We define the first grid point to be one grid cell below zero, and the last
grid point to be 2π (see figure).

grid points
2, m
1, m-1

The polar angle is now discretized as

θ j = ( j − 2)h, j = 1, 2, . . . , m,

and since θm = 2π, we have h = 2π/(m − 2).

The first and last grid points are duplicates of the (m − 1) and second internal grid
points and are called ghost points. Periodic boundary conditions then requires for these
ghost points

ψi,1 = ψi,m−1 , ωi,1 = ωi,m−1 ; ψi,m = ψi,2 , ωi,m = ωi,2 .

34
 WEEK III. UNSTEADY FLOWS 35

Problems for Lecture 16

1. Our implementation of periodic boundary conditions in the polar angle uses two extra
ghost points: one point below the x-axis and one point on the x-axis. It is equally fine to
implement periodic boundary conditions with one point above the x-axis and one point
on the x-axis, as shown in the figure below.

grid points
2, m
1, m-1

Write down the two appropriate periodic boundary conditions for the first and last grid
points. What is the difference between this implementation and the one in lecture?

Solutions to the Problems

Lecture 17 | Finite difference
equations
View this lecture on YouTube

If we know the vorticity ω at time t, we can solve a Poisson equation for the stream
function,
∂2 ψ ∂2 ψ
 
− + 2 = e2ξ ω.
∂ξ 2 ∂θ
Then with both the stream function and vorticity known at time t, we can time integrate
the vorticity equation,

2e−2ξ ∂2 ω ∂2 ω
   
∂ω −2ξ ∂ψ ∂ω ∂ψ ∂ω
= + +e − .
∂t Re ∂ξ 2 ∂θ 2 ∂ξ ∂θ ∂θ ∂ξ

We discretize the spatial derivatives as before, using a second-order finite difference

scheme. The stream-function equation is

4ψi,j − ψi+1,j − ψi−1,j − ψi,j+1 − ψi,j−1 = h2 e2ξ i ωi,j ,

and the vorticity equation becomes

dωi,j 2e−2ξ i

= 2 ωi+1,j + ωi−1,j + ωi,j+1 + ωi,j−1 − 4ωi,j
dt h Re
e−2ξ i 




+ 2
ψi+1,j − ψi−1,j ωi,j+1 − ωi,j−1 − ψi,j+1 − ψi,j−1 ωi+1,j − ωi−1,j .
4h

In the next few lectures, we discuss how to solve these two coupled equations.

36
 WEEK III. UNSTEADY FLOWS 37

Problems for Lecture 17

1. Derive the finite difference equation for the scalar vorticity from the time-dependent
vorticity equation,

2e−2ξ ∂2 ω ∂2 ω
   
∂ω ∂ψ ∂ω ∂ψ ∂ω
= 2
+ 2 + e−2ξ − .
∂t Re ∂ξ ∂θ ∂ξ ∂θ ∂θ ∂ξ

Solutions to the Problems

Lecture 18 | Stream-function
computation
View this lecture on YouTube

The discrete stream-function equation is given by

4ψi,j − ψi+1,j − ψi−1,j − ψi,j+1 − ψi,j−1 = h2 e2ξ i ωi,j .

We define the right-hand side of this equation as

ω̃i,j = h2 e2ξ i ωi,j ,

and apply natural ordering [Numerical Methods for Engineers, Lecture 63] to map (i, j) →
k = i + ( j − 1)n. The stream-function equation transforms to

4ψk − ψk+1 − ψk−1 − ψk+n − ψk−n = ω̃k .

This is a matrix equation Aψ = b, where each value of k corresponds to a row of the

matrix A. The matrix A has dimension mn-by-mn and has five diagonal bands: a four on
the main diagonal, and a negative one on one and n above and below the main diagonal.
These diagonal bands extend only over the rows in which k indexes an interior point of
the domain.
When k indexes a boundary point, however, the discrete stream function equation no
longer applies and the corresponding rows of the matrix A and right-hand side b are
modified to enforce the boundary conditions. We discuss the boundary conditions in the
next lecture.

38
 WEEK III. UNSTEADY FLOWS 39

Problems for Lecture 18

1. Consider the calculation of ω̃k in Matlab, where

ω̃i,j = h2 e2ξ i ωi,j ,

the dimension of ω is n-by-m, and we have applied natural ordering to map (i, j) → k =
i + ( j − 1)n.
Map the mathematical variables above into the following Matlab variables: omega_tilde,
h, xi, omega, n, m. Write the Matlab code that computes the mn sized vector omega_tilde
given that all the other variables have previously been defined.

Solutions to the Problems

Lecture 19 | Stream-function
boundary conditions
View this lecture on YouTube

We want to solve the matrix equation Aψ = b and consider now the rows of A corre-
sponding to points on the boundaries of the computational domain. The values of k on
the boundaries are shown in the figure below, where we borrow the colon notation from
Matlab. We label the four sides: bottom (B), top (T), left (L) and right (R). On (L) where

ξ = 0, we have ψ1,j = 0; and on (R) where ξ = ξ n , we have the free-stream condition

ψn,j = eξ n sin θ j . These boundary conditions are implemented by replacing the kth row of
A by the corresponding kth row of the mn × mn identity matrix, and setting the right-hand
side row element either to zero or to the free-stream condition.
Periodic boundary conditions are implemented on (B) and (T). On (B), we implement
ψi,1 − ψi,m−1 = 0. That is, in rows k = 1:n of the matrix A, we place the difference
of two rows of the identity matrix: rows 1:n minus rows 1 + (m − 2)n:(m − 1)n. The
corresponding row elements in b are set to zero. On (T), we implement ψi,m − ψi,2 = 0.
That is, in rows k = 1 + (m − 1)n:mn of the matrix A, we place the difference of two
rows of the identity matrix: rows 1 + (m − 1)n:mn minus rows (1 + n):2n. And again the
corresponding row elements in b are set to zero.

40
 WEEK III. UNSTEADY FLOWS 41

Problems for Lecture 19

1. Construct the nine-by-nine matrix equation for the stream function on the three-by-
three grid shown below, where the k indexing is labeled. Note that only grid point five is
an internal point.

7 8 9

4 5 6

1 2 3

Solutions to the Problems

Lecture 20 | Vorticity computation
View this lecture on YouTube

The discrete scalar vorticity equation at interior grid points is given by

dωi,j 2e−2ξ i

= 2 ωi+1,j + ωi−1,j + ωi,j+1 + ωi,j−1 − 4ωi,j
dt h Re
e−2ξ i 




+ ψi+1,j − ψi−1,j ωi,j+1 − ωi,j−1 − ψi,j+1 − ψi,j−1 ωi+1,j − ωi−1,j
4h2

This equation is a system of (n − 2)(m − 2) first-order differential equations. The stream

function and scalar vorticity are assumed to be known at time t, and this system of equa-
tions is integrated to solve for the scalar vorticity at time t + ∆t.
Although we are solving only for the scalar vorticity at the interior points, the right-
hand side requires the values of the stream function and scalar vorticity at the boundaries.
Our matrix equation for the stream function already includes boundary conditions, and
the boundary conditions on the scalar vorticity are

ψ3,j − 8ψ2,j
ω1,j = , ωn,j = 0, ωi,1 = ωi,m−1 ωi,m = ωi,2 .
2h2

To integrate the system of first-order differential equations, a Matlab ode solver can
be used. We have timed the solvers ode45.m, ode23.m and ode113.m using the default
error tolerance RelTol = 1.e-3, and the most commonly used solver, ode45.m, appears
to require about 50% more derivative evaluations than the other two solvers. We have
therefore chosen to use ode23.m in our program.

42
 WEEK III. UNSTEADY FLOWS 43

Problems for Lecture 20

1. In an efficiently written Matlab program, often one line of code takes up most of the
computational time. In the unsteady flow problem, it will certainly be the call to ode23.m.
But within the function that defines the time derivative of the vorticity field, which line of
code do you think will require most of the computational time?

Solutions to the Problems

Lecture 21 | Matlab program
(unsteady) (Part A)
View this lecture on YouTube

To compute the Kármán vortex street, one can first compute a time series of the scalar
vorticity in the unsteady flow regime, and second, create images of the vorticity field and
add them to a movie file. For the assessment, we will only ask students to fill out the
skeleton of a code to time-advance the vorticity field. Students, on their own, can create
a time series of fields and produce a movie of the Kármán street, as explained in the next
Lecture.
The code template we give students uses 101 grid points in the ξ direction and 202 grid
points in the θ direction. Dimensionless time is to be advanced to only t = 0.5, and the
resulting vorticity field will be graded. We found that a near steady flow field develops
by t = 50, after which a growing instability can be seen visually near t = 900, with a fully
formed Kármán street at t = 1100.
The structure of the simulation code for the assessment is as follows:
1. Define the grid and time-stepping parameters.
2. Initialize the flow fields, including the free-stream boundary condition.
3. Construct the matrix A for the ψ equation and compute A = LU.
4. Compute and save time-independent factors.
5. Advance the vorticity field using ode23.m.
6. Compute the stream function and obtain the vorticity boundary values.
9. Plot the vorticity field.
The function called by ode23.m has the following structure:
1. Compute the stream function.
2. Set the boundary conditions on the vorticity field.
3. Compute the time-derivative of the vorticity.

44
 WEEK III. UNSTEADY FLOWS 45

Problems for Lecture 21

1. Using ode23.m, compute the unsteady solution for the scalar vorticity when Re = 60.

Solutions to the Problems

Lecture 22 | Matlab program
(unsteady) (Part B)
View this lecture on YouTube

After you have a code that computes a time-dependent vorticity field, you might want
to create a movie file of the Kármán vortex street. One could add movie frames on the fly
as you compute the fields. But since most of the computational time is spent computing
the fields, and often we need to tweak the graphics, a better solution is to write a time
series of fields into .mat files. These fields can then be loaded into another program that
creates the movie file.
To write the .mat files, we define a file name such as
output_file='.\fields\Re_60_';

where we choose to write the fields into a subdirectory of our current working directory.
We then loop over the call to ode23.m, advance the vorticity field over a fixed time interval,
and save each computed vorticity field to disk using the Matlab function call
save([output_file,num2str(k),'.mat'],'omega');

where k is the loop variable, for instance.

The separate graphics program then reads in the fields one-by-one, and plots the vor-
ticity using the Matlab function imagesc. To initialize the movie file, we use
writerObj = VideoWriter('movie.mp4','MPEG-4');
writerObj.Quality = 100;
writerObj.FrameRate=24;
open(writerObj);

To capture movie frames and save them into a movie file, we use the commands
frame=getframe(gcf);
writeVideo(writerObj,frame);

Finally, we conclude the program with the command

close(writerObj);

I will provide you a copy of my graphics code, but feel free to modify it to suit your
desired outcome and taste.

46
 WEEK III. UNSTEADY FLOWS 47

Problems for Lecture 22

1. Study the graphics code that creates the movie file and try to understand what it does.

Solutions to the Problems

 Problem solutions and
Matlab learner templates

48
 PROBLEM SOLUTIONS AND MATLAB LEARNER TEMPLATES 49

Solutions to the Problems for Lecture 1

1. The first component of the Navier-Stokes equation reduces to

d2 u
ν = 0.
dy2

The boundary conditions on the plates are u(0) = 0 and u(d) = U . We find u(y) = Uy/d,
so that
Uy
𝑢= 𝑖.
d

2. The first component of the Navier-Stokes equation reduces to

1 dp d2 u
− + ν 2 = 0.
ρ dx dy

Using dp/dx = − G leads to

d2 u G
2
=− ,
dy νρ
which can be solved using the no-slip boundary conditions u(0) = u(d) = 0. We find

Gd2  y   y
𝑢= 1− 𝑖.
2νρ d d

3. The first component of the Navier-Stokes equation becomes

∂2 u ∂2 u
 
1 dp
− +ν + 2 = 0.
ρ dx ∂y2 ∂z

Using dp/dx = − G leads to

∂2 u ∂2 u G
2
+ 2 =− .
∂y ∂z νρ
p
We use polar coordinates in the y-z plane. With r = y2 + z2 , and the Laplacian written
in polar coordinates, the differential equation for u = u(r ) then becomes
 
d du Gr
r =− ,
dr dr νρ

with no-slip boundary condition u( R) = 0. The first integration from 0 to r yields

du Gr2
r =− ;
dr 2νρ

and after division by r, the second integration from r to R yields

GR2
  r 2 
u (r ) = 1− ,
4νρ R
 PROBLEM SOLUTIONS AND MATLAB LEARNER TEMPLATES 50

so that
GR2 y2 + z2
 
𝑢= 1− 𝑖.
4νρ R2

Solutions to the Problems for Lecture 2

1.

a) We prove that ∇ × {(𝑢 · ∇)𝑢} = ∇ × (𝜔 × 𝑢) . To begin, we first prove the identity

1
𝑢 × (∇ × 𝑢) = ∇(𝑢 · 𝑢) − (𝑢 · ∇)𝑢.
2

Proof proceeds by considering the ith component of the left-hand side:

∂um
[𝑢 × (∇ × 𝑢)]i = eijk u j eklm (curl in component notation)
∂xl
∂um
= ekij eklm u j (eijk = ekij )
∂xl
∂um
= (δil δjm − δim δjl )u j (ekij eklm = δil δjm − δim δjl )
∂xl
∂u j ∂u
= uj − uj i (δij A j = Ai )
∂xi ∂x j
1 ∂ ∂u
= (u u ) − u j i
2 ∂xi j j ∂x j
 
1
= ∇(𝑢 · 𝑢) − [(𝑢 · ∇)𝑢]i . (back to vector notation)
2 i

Therefore,
1
(𝑢 · ∇)𝑢 = ∇(𝑢 · 𝑢) − 𝑢 × (∇ × 𝑢),
2
and
 
1
∇ × (𝑢 · ∇)𝑢 = ∇ × ∇(𝑢 · 𝑢) − 𝑢 × (∇ × 𝑢)
2
1
= ∇ × (∇(𝑢 · 𝑢)) − ∇ × (𝑢 × (∇ × 𝑢)) (distribution law for curl)
2
= −∇ × (𝑢 × (∇ × 𝑢)) (curl of a gradient equals zero)
= −∇ × (𝑢 × 𝜔 ) (∇ × 𝑢 = 𝜔)
= ∇ × (𝜔 × 𝑢). (𝑢 × 𝜔 = −𝜔 × 𝑢)

b) We prove that
∇ × (𝜔 × 𝑢) = (𝑢 · ∇)𝜔 − (𝜔 · ∇)𝑢,

using the vector identity

∇ × (𝑢 × 𝑣 ) = 𝑢(∇ · 𝑣 ) − 𝑣 (∇ · 𝑢) + (𝑣 · ∇)𝑢 − (𝑢 · ∇)𝑣.

 PROBLEM SOLUTIONS AND MATLAB LEARNER TEMPLATES 51

We have

∇ × (𝜔 × 𝑢) = 𝜔 (∇ · 𝑢) − 𝑢(∇ · 𝜔 ) + (𝑢 · ∇)𝜔 − (𝜔 · ∇)𝑢 (vector identity)

= (𝑢 · ∇)𝜔 − (𝜔 · ∇). (∇ · 𝑢 = 0, ∇ · 𝜔 = 0)

Solutions to the Problems for Lecture 3

1. The cylinder boundary in Cartesian coordinates is given by

x 2 + y2 = R2 ,

and in polar coordinates is given by

r = R.

The free stream velocity in Cartesian coordinates is given by

𝑢 f = U𝑖,

and in polar coordinates is given by

𝑢 f = U (cos θ r̂ − sin θ θ̂ ).

Solutions to the Problems for Lecture 4

1. We have

𝑖 𝑗 𝑘
𝜔×𝑢 = 0 0 ω = −vω𝑖 + uω𝑗,
u v 0

and

𝑖 𝑗 𝑘  
∂ ∂
∇ × (𝜔 × 𝑢) = ∂/∂x ∂/∂y 0 = (uω ) + (vω ) 𝑘
∂x ∂y
−vω uω 0
   
∂u ∂v ∂ω ∂ω
= + ω+ u +v 𝑘
∂x ∂y ∂x ∂y
 
∂ω ∂ω
= u +v 𝑘.
∂x ∂y

2.
𝑖 𝑗 𝑘  
∂v ∂u
𝜔 = ∇ × 𝑢 = ∂/∂x ∂/∂y 0 = − 𝑘.
∂x ∂y
u v 0
 PROBLEM SOLUTIONS AND MATLAB LEARNER TEMPLATES 52

The scalar vorticity is then defined as

∂v ∂u
ω= − .
∂x ∂y

3. In Cartesian coordinates, the two-dimensional scalar vorticity equation is written as

∂2 ω ∂2 ω
 
∂ω ∂ω ∂ω
+u +v =ν + .
∂t ∂x ∂y ∂x2 ∂y2

Solutions to the Problems for Lecture 5

1. The two-dimensional incompressibility condition in Cartesian coordinates is given by

∂u ∂v
+ = 0.
∂x ∂y

The appropriate definition of the stream function ψ = ψ( x, y) that satisfies this condition
is
∂ψ ∂ψ
u= , v=− .
∂y ∂x

2. The scalar vorticity in Cartesian coordinates is given by

∂v ∂u
ω= − .
∂x ∂y

Using the definition of the stream function, we have

∂2 ψ ∂2 ψ
 
ω=− + 2
∂x2 ∂y

3. Since ∇ · 𝑢 = 0 we can write 𝑢 = ∇ × 𝐴, where 𝐴 is called a vector potential. In two

dimensions, we have

𝑖 𝑗 𝑘  
∂A3 ∂A3 ∂A2 ∂A1
u𝑖 + v𝑗 = ∂/∂x ∂/∂y 0 = 𝑖− 𝑗+ − 𝑘.
∂y ∂x ∂x ∂y
A1 A2 A3

Therefore,
∂A3 ∂A3
u= , v=− ,
∂y ∂x
and the stream function can be identified as the third component of the vector potential
𝐴.
 PROBLEM SOLUTIONS AND MATLAB LEARNER TEMPLATES 53

Solutions to the Problems for Lecture 6

1. In Cartesian coordinates, d𝑟 = dx𝑖 + dy𝑗, and with the velocity field parallel to d𝑟, we
have
0 = 𝑢 × d𝑟 = (u𝑖 + v𝑗 ) × (dx𝑖 + dy𝑗 ) = (−vdx + udy)𝑘.

But since the stream function ψ is defined by

∂ψ ∂ψ
u= , v=− ,
∂y ∂x

we have determined that

 
∂ψ ∂ψ
0= dx + dy 𝑘 = dψ𝑘,
∂x ∂y

or that the stream function is constant along streamlines.

Solutions to the Problems for Lecture 7

1. The nondimensional vorticity equation can be written as

  
Re ∂ω 1 ∂ψ ∂ω ∂ψ ∂ω
∇2 ω = + − ;
2 ∂t r ∂θ ∂r ∂r ∂θ

and if we set Re = 0, we obtain the Laplace equation ∇2 ω = 0. Since ∇2 ψ = −ω, taking

the Laplacian of both sides results in the biharmonic equation, usually written as

∇4 ψ = 0.

The attempt to solve this equation for the flow around an infinite circular cylinder leads
to Stokes’ paradox.

Solutions to the Problems for Lecture 8

1. The time-dependent governing equations are given by

∂2 ψ ∂2 ψ
 
+ 2 = −e2ξ ω,
∂ξ 2 ∂θ
∂2 ω ∂2 ω
   
∂ω ∂ψ ∂ω ∂ψ ∂ω 2 −2ξ
+ e−2ξ − = e 2
+ 2 .
∂t ∂θ ∂ξ ∂ξ ∂θ Re ∂ξ ∂θ

In a steady flow, we have ∂ω/∂t = 0. The governing equations can then be written neatly
as
∂2 ψ ∂2 ψ ∂2 ω ∂2 ω
     
2ξ Re ∂ψ ∂ω ∂ψ ∂ω
+ 2 = −e ω, + = − .
∂ξ 2 ∂θ ∂ξ 2 ∂θ 2 2 ∂θ ∂ξ ∂ξ ∂θ
 PROBLEM SOLUTIONS AND MATLAB LEARNER TEMPLATES 54

Solutions to the Problems for Lecture 9

1.

a) Suppose f ( x ) is an even function of x so that f (− x ) = f ( x ). Then using the defini-

tion of the derivative,

f (− x + h) − f (− x )
f 0 (− x ) = lim
h →0 h
f ( x − h) − f ( x )
= lim
h →0 h
f ( x ) − f ( x − h)
= − lim
h →0 h
= − f 0 ( x ).

b) Suppose f ( x ) is an odd function of x so that f (− x ) = − f ( x ). Then using the

definition of the derivative

f (− x + h) − f (− x )
f 0 (− x ) = lim
h →0 h
− f ( x − h) + f ( x )
= lim
h →0 h
f ( x ) − f ( x − h)
= lim
h →0 h
= f 0 ( x ).

Solutions to the Problems for Lecture 10

1.

a) The stream function equation is given by

∂2 ψ ∂2 ψ
 
− + 2 = e2ξ ω.
∂ξ 2 ∂θ

The second-order finite difference approximation to the second partial derivatives

yields
ψi+1,j − 2ψi,j + ψi−1,j ψi,j+1 − 2ψi,j + ψi,j−1
 
− 2
+ = e2ξ i ωi,j
h h2
We multiply by h2 and collect terms to obtain

4ψi,j − ψi+1,j − ψi−1,j − ψi,j+1 − ψi,j−1 = h2 e2ξ i ωi,j

To construct the SOR method, we rewrite this equation as

1 
ψi,j = ψi,j + ψi+1,j + ψi−1,j + ψi,j+1 + ψi,j−1 − 4ψi,j + h2 e2ξ i ωi,j .
4
 PROBLEM SOLUTIONS AND MATLAB LEARNER TEMPLATES 55

We make the left-hand side the (k + 1)-th iteration and the right-hand side the k-th
iteration, and we adjust the correction to ψi,j by a relaxation factor rψ . The result
becomes

k +1 k rψ  k 
ψi,j = ψi,j + ψi+1,j + ψik−1,j + ψi,j
k
+
k k 2 2ξ i k
1 + ψi,j−1 − 4ψi,j + h e ωi,j
4
k rψ  k 
= (1 − rψ )ψi,j + ψi+1,j + ψik−1,j + ψi,j
k
+1 + ψ k
i,j−1 + h 2 2ξ i k
e ω i,j .
4

b) The scalar vorticity equation is given by

∂2 ω ∂2 ω
   
Re ∂ψ ∂ω ∂ψ ∂ω
− + = − .
∂ξ 2 ∂θ 2 2 ∂ξ ∂θ ∂θ ∂ξ

Second-order finite difference approximations for the first- and second-derivatives

yields

ωi+1,j − 2ωi,j + ωi−1,j ωi,j+1 − 2ωi,j + ωi,j−1

 
− +
h2 h2
ψi+1,j − ψi−1,j ωi,j+1 − ωi,j−1 ψi,j+1 − ψi,j−1 ωi+1,j − ωi−1,j
     
Re
= − .
2 2h 2h 2h 2h

Multiplying by h2 and collecting terms yields

4ωi,j − ωi+1,j − ωi−1,j − ωi,j+1 − ωi,j−1

Re 




= ψi+1,j − ψi−1,j ωi,j+1 − ωi,j−1 − ψi,j+1 − ψi,j−1 ωi+1,j − ωi−1,j
8

For notational clarity, we define

f i,j = ψi+1,j − ψi−1,j ωi,j+1 − ωi,j−1 − ψi,j+1 − ψi,j−1 ωi+1,j − ωi−1,j .

We then rewrite the equation as

 
1 Re
ωi,j = ωi,j + ωi+1,j + ωi−1,j + ωi,j+1 + ωi,j−1 − 4ωi,j + f .
4 8 i,j

Again, we make the left-hand side the (k + 1)-th iteration and the right-hand side
the k-th iteration, and we adjust the correction to ωi,j by a relaxation factor rω . The
result becomes
 
k +1 k rω Re k
ωi,j = ωi,j + ωik+1,j + ωik−1,j k
+ ωi,j +1 + k
+ ωi,j
f −1
k
− 4ωi,j
4 8 i,j
 
k rω k k k k Re k
= (1 − rω )ωi,j + ωi+1,j + ωi−1,j + ωi,j+1 + ωi,j−1 + f ,
4 8 i,j

with
     
k
f i,j = ψik+1,j − ψik−1,j ωi,j
k
+1 − ω k
i,j−1 − ψ k
i,j+1 − ψ k
i,j−1 ω k
i +1,j − ω k
i −1,j .
 PROBLEM SOLUTIONS AND MATLAB LEARNER TEMPLATES 56

Solutions to the Problems for Lecture 11

1. We Taylor series expand the scalar vorticity one and two grid points inside the free-
stream boundary. That is,

∂ω h2 ∂2 ω
ωn−1,j = ωn,j − h + + O( h3 )
∂ξ (n,j) 2 ∂ξ 2 (n,j)
∂ω 2
(2h) ∂2 ω
ωn−2,j = ωn,j − 2h + + O( h3 ).
∂ξ (n,j) 2 ∂ξ 2 (n,j)

We assume that the first derivative terms are zero and multiply the first equation by four
and subtract the second equation to find

4ωn−1,j − ωn−2,j = 3ωn,j + O(h3 ).

The boundary condition on ωn,j is therefore

4ωn−1,j − ωn−2,j
ωn,j = .
3

Solutions to the Problems for Lecture 12

1. Near the cylinder surface, we assume that

ψ(ξ ) = aξ 2 .

Then
∂2 ψ
= 2a, ψ2,j = ah2 , ψ3,j = 4ah2 ,
∂ξ 2 (1,j)

and
1 1 ∂2 ψ
2
(8ψ2,j − ψ3,j ) = 2 (8ah2 − 4ah2 ) = 2a = .
2h 2h ∂ξ 2 (1,j)

Solutions to the Problems for Lecture 13

1. The SOR iteration formulas are

k +1 k rψ  k 
ψi,j = (1 − rψ )ψi,j + ψi+1,j + ψik−1,j + ψi,j
k
+
k 2 2ξ i k
1 + ψi,j−1 + h e ωi,j ,
4  
k +1 k r ω k k k k Re k
ωi,j = (1 − rω )ωi,j + ωi+1,j + ωi−1,j + ωi,j+1 + ωi,j−1 + f ,
4 8 i,j

where
     
f ijk = ψik+1,j − ψik−1,j ωi,j
k
+1 − ω k
i,j−1 − ψ k
i,j+1 − ψ k
i,j−1 ω k
i +1,j − ω k
i −1,j .
 PROBLEM SOLUTIONS AND MATLAB LEARNER TEMPLATES 57

The iterations start with ψ and ω everywhere zero except for

ψn,j = eξ n sin θ j , for j = 1 to m.

Note that every term in the iteration equation for the scalar vorticity ω k+1 is proportional
to ω k . Therefore, ω k can only become nonzero through the boundary condition on the
cylinder:
ω1,j = (ψ3,j − 8ψ2,j )/2h2 ,

and ω will first become nonzero after ψ3,j becomes nonzero. In the iteration equation for
k +1
the stream function ψi,j , there is a term ψik+1,j so that the nonzero boundary condition at
i = n propagates downward in i with each iteration until it finally reaches i = 3. This will
take n − 3 iterations of the SOR equations.
Considerations such as this are useful when debugging code and monitoring how the
fields develop with each iteration.
 PROBLEM SOLUTIONS AND MATLAB LEARNER TEMPLATES 58

Solutions to the Problems for Lecture 14

1. Complete your solution in the Matlab Grader using the Learner Template:

function [psi, omega] = flow_around_cylinder_steady

Re=10;
%%%%% define the grid %%%%%
n=101; m=101; % number of grid points
N=n-1; M=m-1; % number of grid intervals
h=pi/M; % grid spacing based on theta variable
xi=(0:N)*h; theta=(0:M)*h; % xi and theta variables on the grid
%%%%% Initialize the flow fields %%%%%
psi=zeros(n,m);
omega=zeros(n,m);
psi(n,:)=... % Write the free-stream boundary condition here
%%%%% Set relax params, tol, extra variables %%%%%
r_psi=1.8; r_omega=0.9; % relaxation parameters
delta=1.e-08; % error tolerance
error=2*delta; % initialize error variable
%%%%% Add any additional variable definitions here %%%%%
...
...
%%%%% Main SOR Loop %%%%%
while (error > delta)
psi_old = psi; omega_old = omega;
for i=2:n-1
for j=2:m-1
psi(i,j)=... % Write the psi equation here
end
end
error_psi=max(abs(psi(:)-psi_old(:)));
omega(1,:)=... % Write the boundary condition here
for i=2:n-1
for j=2:m-1
omega(i,j)=... % Write the omega equation here
end
end
error_omega=max(abs(omega(:)-omega_old(:)));
error=max(error_psi, error_omega);
end
plot_Re10(psi);
 PROBLEM SOLUTIONS AND MATLAB LEARNER TEMPLATES 59

Solutions to the Problems for Lecture 15

1. Here is a copy of the graphics code.

function plot_Re10(psi)
Re=10;
%%%%% xi-theta grid %%%%%
n=size(psi,1); m=size(psi,2);
N=n-1; M=m-1;
h=pi/M; % grid spacing
xi=(0:N)*h; theta=(0:M)*h;
[XI, THETA] = meshgrid(xi,theta);
%%%%%% x-y grid %%%%%
nx=640; ny=480/2; % number of pixels in x and half of y
xmin=-1.5; xmax=2.5; ymax=(xmax-xmin)*ny/nx; ymin=-ymax;
x=linspace(xmin,xmax,nx+1); y=linspace(0,ymax,ny+1);
[X,Y]=meshgrid(x,y);
%%%%% construct interpolation points %%%%%
xi_i=0.5*log(X.^2+Y.^2);
theta_i=wrapTo2Pi(atan2(Y,X));
%%%%% interpolate %%%%%
psi_xy=interp2(XI,THETA,psi',xi_i,theta_i);
%%%%% set psi to zero inside cylinder %%%%%
psi_xy(xi_i<0)=0;
%%%%% scale contour levels %%%%%
%%%%% negative values have same range as positive values %%%%%
psi_min=min(psi_xy(:));
psi_max=max(psi_xy(:));
psi_xy(psi_xy<0)=psi_xy(psi_xy<0)/abs(psi_min);
psi_xy(psi_xy>0)=psi_xy(psi_xy>0)/abs(psi_max);
%%%%% set colormap for contours %%%%%
levels=linspace(-1,1,1000);
cmap=flipud(jet(length(levels)));
colormap(cmap);
%%%%% plot color contours %%%%%
imagesc(x,y,psi_xy); hold on;
imagesc(x,-y,-psi_xy); % negative values of y
%%%%% plot contour lines %%%%%
v=[-0.9:0.2:-0.1,0,0.0005,0.001,0.002:0.004:0.01,0.02:0.04:0.9];
contour(X,Y,psi_xy,v,'LineColor','k');
contour(X,-Y,-psi_xy,-v,'LineColor','k');
%%%%% draw black circle for cylinder %%%%%
t=linspace(0,2*pi, 1000);
a=cos(t); b=sin(t);
fill(a,b,[0 0 0]);
%%%%% beautify plot %%%%%
set(gca,'YDir','normal');
axis([xmin xmax ymin ymax]); set(gcf,'color','w'); axis equal; axis off;
text(xmin+0.75*(xmax-xmin),ymin+0.08*(ymax-ymin),...
['Re = ', num2str(Re,'%3.0f')],'FontSize',22,'Color','k');
 PROBLEM SOLUTIONS AND MATLAB LEARNER TEMPLATES 60

Solutions to the Problems for Lecture 16

1. Periodic boundary conditions requires for the end points

ψi,1 = ψi,m−1 , ωi,1 = ωi,m−1 ; ψi,m = ψi,2 , ωi,m = ωi,2 .

The boundary conditions are the same as the implementation in the lecture, but the values
of θ are now shifted by one grid point, starting at zero rather than −h and ending at 2π + h
rather than at 2π.

Solutions to the Problems for Lecture 17

1. The scalar vorticity equation is given by

2e−2ξ ∂2 ω ∂2 ω
   
∂ω ∂ψ ∂ω ∂ψ ∂ω
= 2
+ 2 + e−2ξ − .
∂t Re ∂ξ ∂θ ∂ξ ∂θ ∂θ ∂ξ

Second-order finite difference approximations for the first- and second-derivatives yields

dωi,j 2e−2ξ i ωi+1,j − 2ωi,j + ωi−1,j ωi,j+1 − 2ωi,j + ωi,j−1

   
= + +
dt Re h2 h2
ψi+1,j − ψi−1,j ωi,j+1 − ωi,j−1 ψi,j+1 − ψi,j−1 ωi+1,j − ωi−1,j
     
−2ξ i
e − .
2h 2h 2h 2h

Collecting terms yields

dωi,j 2e−2ξ i

= 2 ωi+1,j + ωi−1,j + ωi,j+1 + ωi,j−1 − 4ωi,j +
dt h Re
e−2ξ i 




ψi+1,j − ψi−1,j ωi,j+1 − ωi,j−1 − ψi,j+1 − ψi,j−1 ωi+1,j − ωi−1,j .
4h2

Solutions to the Problems for Lecture 18

1. One straightforward implementation that uses two for loops is

omega_tilde=zeros(n,m);
for j=1:m
for i=1:n
omega_tilde(i,j}=h^2*exp(2*xi(i))*omega(i,j);
end
end
omega_tilde = omega_tilde(:);

An implementation that doesn’t use for loops could use the Matlab repmat.m function
to duplicate xi across m columns, and may be written as
 PROBLEM SOLUTIONS AND MATLAB LEARNER TEMPLATES 61

omega_tilde=h^2*repmat(exp(2*xi(:)),1,m).*omega;
omega_tilde=omega_tilde(:);

Solutions to the Problems for Lecture 19

1. Only the fifth row satisfies the Poisson equation, being the only internal grid point.
Rows one to three and rows seven to nine obey periodic boundary conditions, row four
satisfies the cylinder boundary condition ψ4 = 0 and row six satisfies the free-stream
condition ψ6 = eξ 3 sin θ2 .

Φ1
    
1 0 0 −1 0 0 0 0 0 0
0 1 0 0 −1 0 0 0 0  Φ2   0
    

0  Φ3  
    
0
 0 1 0 0 −1 0 0     0 

0 0 0 1 0 0 0 0 0  Φ4   0
    

0 Φ5  = h e ω2,2 
    
0 −1 0 −1 4 −1 0 −1     2 2ξ 2
 
0 0 0 0 0 1 0 0 0 Φ6   eξ 3 sin θ2 
    

  Φ7  
    
0
 0 0 −1 0 0 1 0 0    0 

0 0 0 0 −1 0 0 1 0  Φ8   0
    

0 0 0 0 0 −1 0 0 1 Φ9 0

With only three grid points in θ, all the θ values are actually located at the same point
(equivalent to θ = 0)! Nevertheless, this result can be used to debug your construction of
the matrix for the stream-function equation.

Solutions to the Problems for Lecture 20

1. We have found that the line of code that solves the stream-function equation Aψ = b
using the backslash operator takes almost 90% of the computational time. Replacing A by
LU, where the LU decomposition need only be done once, saves substantial computational
time.
 PROBLEM SOLUTIONS AND MATLAB LEARNER TEMPLATES 62

Solutions to the Problems for Lecture 21

1. Complete your solution in the Matlab Grader using the Learner Template:

function omega = flow_around_cylinder_unsteady

Re=60;
%%%%% define the grid %%%%%
n=101; m=202; % number of grid points
N=n-1; M=m-2; % number of grid intervals: 2 ghost points, theta=-h,2*pi
h=2*pi/M; % grid spacing based on theta variable
xi=(0:N)*h; theta=(-1:M)*h; % xi and theta variables on the grid
%%%%% time-stepping parameters %%%%%
t_start=0; t_end=0.5; % vortex street starts at around t=900
tspan=[t_start t_end];
%%%%% Initialize vorticity field %%%%%
omega=zeros(n,m);
%%%%% Construct the matrix A for psi equation %%%%%

%%%%% Find the LU decomposition %%%%%

[L,U]=lu(A); clear A;
%%%%% Compute any time-independent constants %%%%%

%%%%% advance solution using ode23 %%%%%

options=odeset('RelTol', 1.e-03);
omega=omega(2:n-1,2:m-1); % strip boundary values for ode23
omega=omega(:); % make a column vector
[t,omega]=ode23...
(@(t,omega)omega_eq(omega,L,U, (additional arguments),...
tspan, omega, options);
%%%%% expand omega to include boundaries %%%%%
temp=zeros(n,m);
temp(2:n-1,2:m-1)=reshape(omega(end,:),n-2,m-2);
omega=temp; clear temp;
%%%%% compute stream function (needed for omega boundary values) %%%%%

%%%%% set omega boundary conditions %%%%%

%%%%% plot scalar vorticity field %%%%%

plot_Re60(omega);
end
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
function d_omega_dt=omega_eq(omega,L,U,%additional arguments)
%%%%% expand omega to include boundary points %%%%%
temp=zeros(n,m);
index1=2:n-1; index2=2:m-1;
temp(index1,index2)=reshape(omega,n-2,m-2);
omega=temp; clear temp;
%%%%% compute stream function %%%%%

%%%%% compute derivatives of omega %%%%%

end
 PROBLEM SOLUTIONS AND MATLAB LEARNER TEMPLATES 63

Here is a copy of the code plot_Re60.m.

function plot_Re60(omega,t_plot)
% Plot vorticity for Re = 60 from flow_past_circle_unsteady
Re=60;
% xi-theta grid
n=size(omega,1); m=size(omega,2);
N=n-1; M=m-2;
h=2*pi/M; %grid spacing
xi=(0:N)*h; theta=(-1:M)*h; %xi and theta variables on the grid
[XI, THETA] = meshgrid(xi,theta);

% x-y grid
nx=640; ny=480; %number of pixels in x and y
xmin=-1.5; xmax=10; ymax=((xmax-xmin)/2)*ny/nx; ymin=-ymax;
x=linspace(xmin,xmax,nx+1); y=linspace(ymin,ymax,ny+1);
[X,Y]=meshgrid(x,y);

%construct interpolation points

xi_i=0.5*log(X.^2+Y.^2);
theta_i=wrapTo2Pi(atan2(Y,X));

omega_xy=interp2(XI,THETA,omega',xi_i,theta_i);

%inside circle
omega_xy(xi_i<0)=0;

%set colormap for contours

levels=linspace(-1,1,1000);
v=[levels(1) levels(end)];
cmap=jet(length(levels));
cmap=flipud(cmap);
colormap(cmap);

%color contours
imagesc(x,y,omega_xy,v); hold on;

%draw black circle for cylinder

t=linspace(0,2*pi, 1000);
a=cos(t); b=sin(t);
fill(a,b,[0 0 0]);

%neaten plot
set(gca,'YDir','normal');
axis([xmin xmax ymin ymax]); set(gcf,'color','w'); axis equal; axis off;
text(xmin+0.75*(xmax-xmin),ymin+0.08*(ymax-ymin),...
['t = ', num2str(t_plot,'%3.1f')],'FontSize',22,'Color','k');
 PROBLEM SOLUTIONS AND MATLAB LEARNER TEMPLATES 64

Solutions to the Problems for Lecture 22

1. Here is a copy of the graphics code.

function plot_Re60_movie
Re=60;
%%%%% Initialize movie file %%%%%
writerObj = VideoWriter('movie.mp4','MPEG-4');
writerObj.Quality = 100; writerObj.FrameRate=24; open(writerObj);
input_file='.\fields\Re_60_';
%%%%% Loop over fields to construct plot and put in movie file %%%%%
nfields=1000;
for ij=1:nfields
load([input_file num2str(ij)],'omega');
%%%%% xi-theta grid %%%%%
n=size(omega,1); m=size(omega,2);
N=n-1; M=m-2;
h=2*pi/M; %grid spacing
xi=(0:N)*h; theta=(-1:M)*h;
[XI, THETA] = meshgrid(xi,theta);
%%%%%% x-y grid %%%%%
nx=640; ny=480; % number of pixels in x and y
xmin=-1.5; xmax=20; ymax=((xmax-xmin)/2)*ny/nx; ymin=-ymax;
x=linspace(xmin,xmax,nx+1); y=linspace(ymin,ymax,ny+1);
[X,Y]=meshgrid(x,y);
%%%%% construct interpolation points %%%%%
xi_i=0.5*log(X.^2+Y.^2); theta_i=wrapTo2Pi(atan2(Y,X));
%%%%% interpolate %%%%%
omega_xy=interp2(XI,THETA,omega',xi_i,theta_i);
%%%%% set omega to zero inside cylinder %%%%%
omega_xy(xi_i<0)=0;
%%%%% set colormap for contours %%%%%
levels=linspace(-1,1,1000);
v=[levels(1) levels(end)];
cmap=flipud(jet(length(levels))); colormap(cmap);
%%%%% plot color contours %%%%%
imagesc(x,y,omega_xy,v); hold on;
%%%%% draw black circle for cylinder %%%%%
t=linspace(0,2*pi, 1000);
a=cos(t); b=sin(t);
fill(a,b,[0 0 0]);
%%%%% beautify plot %%%%%
set(gca,'YDir','normal');
axis([xmin xmax ymin ymax]); set(gcf,'color','w'); axis equal; axis off;
text(xmin+0.75*(xmax-xmin),ymin+0.08*(ymax-ymin),...
['t = ', num2str(ij,'%3.0f')],'FontSize',22,'Color','k');
%%%%% add frame to movie file %%%%%
frame=getframe(gcf);
writeVideo(writerObj,frame); close;
fprintf('%g\n',ij)
end
close(writerObj);