Laplaces Equation

Separation of variables two examples Laplaces Equation in Polar Coordinates

Derivation of the explicit form An example from electrostatics

A surprising application of Laplaces eqn

Image analysis This bit is NOT examined

Laplaces Equation
2 2 + 2 =0 2 y x
In the vector calculus course, this appears as 2

x where = =0 y

Note that the equation has no dependence on time, just on the spatial variables x,y. This means that Laplaces Equation describes steady state situations such as: steady state temperature distributions steady state stress distributions steady state potential distributions (it is also called the potential equation steady state flows, for example in a cylinder, around a corner,

Stress analysis example: Dirichlet conditions

Steady state stress analysis problem, which satisfies Laplaces equation; that is, a stretched elastic membrane on a rectangular former that has prescribed out-of-plane displacements along the boundaries x w(x,y) is the displacement in y w = w0 sin a z-direction b
y w=0 w=0 z x a x

To solve:
2 2

w=0

Boundary conditions w(0, y ) = 0, for 0 y b

w w + 2 =0 2 x y

w( x,0) = 0, w(a, y ) = 0, w( x, b) = w0 sin

for 0 x a for 0 y b

x,

for 0 x a

Solution by separation of variables

w( x, y ) = X ( x)Y ( y ) from which X Y + XY = 0 X Y and so + =0 X Y Y X as usual = =k Y X
where k is a constant that is either equal to, >, or < 0.

Case k=0
X ( x) = ( Ax + B ), Y ( y ) = (Cy + D)
w(0, y ) = 0 B = 0 or C = D = 0 if C = D = 0, then Y ( y ) 0, so w( x, y ) 0 Continue with B = 0 :w( x, y ) = Ax(Cy + D)

w( x,0) = 0 ADx = 0 Either A = 0 (so w 0) or D = 0 Continue with w( x, y ) = ACxy

w(a, y ) = 0 ACay = 0 A = 0 or C = 0 w( x, y ) 0
That is, the case k=0 is not possible

Case k>0
Suppose that k = 2 , so that w( x, y ) = ( A cosh x + B sinh x)(C cos y + D sin y )
Recall that cosh 0 = 1, sinh 0 = 0 w(0, y ) = 0 A(C cos y + D sin y ) = 0 C = D = 0 w( x, y ) 0 Continue with A = 0 w( x, y ) = B sinh x(C cos y + D sin y )

w( x,0) = 0 BC sinh x = 0 B = 0 w( x, y ) 0 Continue with C = 0 w( x, y ) = BD sinh x sin y

w(a, y ) = 0 BD sinh a sin y = 0 so either B = 0 or D = 0 w( x, y ) 0
Again, we find that the case k>0 is not possible

Final case k<0

Suppose that k = 2 w( x, y ) = ( A cos x + B sin x)(C cosh y + D sinh y )
w(0, y ) = 0 A(C cosh y + D sinh y ) = 0 as usual, C = D = 0 w 0 continue with A = 0 w( x, y ) = B sin x(C cosh y + D sinh y )

w( x,0) = 0 BC sin x = 0 B =0 w0 continue with C = 0 w( x, y ) = BD sin x sinh y

w(a, y ) = 0 BD sin a sinh y = 0 B = 0 or D = 0 w 0 sin a = 0 = n

wn ( x, y ) = BD sin n

x sinh n

Solution
Applying the first three w( x, y ) = boundary conditions, we have

nx ny K n sin sinh a a n =1

The final boundary condition is: w( x ,b ) = w0 sin x a

which gives:

nx nb w0 sin = K n sin sinh a n =1 a a

We can see from this that n must take only one value, namely 1, so that K 1 = and the final solution to the stress distribution is

w0 sinh

b
a

w( x, y ) =

w0 sinh

b
a

sin

x
a

sinh

y
a

Check out: http://mathworld.wolfram.com/HyperbolicSine.html

More general boundary condition

y b w=0 w=0

w = w0 f ( x )

a w=0 Then

nx nb w0 f ( x) = K n sin sinh a a n =1

and as usual we use orthogonality formulae/HLT to find the Kn

Types of boundary condition

1. The value ( x, y ) is specified at each point on the boundary: Dirichlet conditions 2. The derivative normal to the boundary n ( x, y ) is specified at each point of the boundary: Neumann conditions 3. A mixture of type 1 and 2 conditions is specified
Johann Dirichlet (1805-1859)
http://www-gap.dcs.stand.ac.uk/~history/Mathematicians/Dirichlet.html

Carl Gottfried Neumann (1832 -1925)

http://www-history.mcs.standrews.ac.uk/history/Mathematicians/Neumann_Carl.html

A mixed condition problem

A steady state heat transfer problem y b

w = w0 sin

2a

To solve:

w w + 2 =0 2 x y
2 2

w=0

w =0 x
a x

Boundary conditions

w=0

w(0, y ) = 0, w( x,0) = 0, w = 0, x x = a w( x, b) = w0 sin

for 0 y b for 0 x a for 0 y b

There is no flow of heat across this boundary; but it does not necessarily have a constant temperature along the edge

2a

x,

for 0 x a

Solution by separation of variables

w( x, y ) = X ( x)Y ( y ) from which X Y + XY = 0 X Y and so + =0 X Y Y X as usual = =k Y X
where k is a constant that is either equal to, >, or < 0.

Case k=0
X ( x) = ( Ax + B ), Y ( y ) = (Cy + D)
w(0, y ) = 0 B = 0 or C = D = 0
if C = D = 0, then Y ( y ) 0, so w( x, y ) 0 Continue with B = 0 :w( x, y ) = Ax(Cy + D)

w( x,0) = 0 ADx = 0 Either A = 0 (so w 0) or D = 0 Continue with w( x, y ) = ACxy

w = 0 ACy = 0 A = 0 or C = 0 w( x, y ) 0 x x = a
That is, the case k=0 is not possible

Case k>0
Suppose that k = 2 , so that w( x, y ) = ( A cosh x + B sinh x)(C cos y + D sin y )
Recall that cosh 0 = 1, sinh 0 = 0 w(0, y ) = 0 A(C cos y + D sin y ) = 0 C = D = 0 w( x, y ) 0 Continue with A = 0 w( x, y ) = B sinh x(C cos y + D sin y )

w( x,0) = 0 BC sinh x = 0 B = 0 w( x, y ) 0 Continue with C = 0 w( x, y ) = BD sinh x sin y

w = 0 BD cosh a sin y = 0 x x = a so either B = 0 or D = 0 w( x, y ) 0
Again, we find that the case k>0 is not possible

Final case k<0

Suppose that k = 2 w( x, y ) = ( A cos x + B sin x)(C cosh y + D sinh y )
w(0, y ) = 0 A(C cosh y + D sinh y ) = 0 as usual, C = D = 0 w 0 continue with A = 0 w( x, y ) = B sin x(C cosh y + D sinh y )

w( x,0) = 0 BC sin x = 0 B =0 w0 continue with C = 0 w( x, y ) = BD sin x sinh y

w = 0 BD cos a sinh y = 0 x x = a B = 0 or D = 0 w 0 cos a = 0 = (2n 1) (2n 1) (2n 1) x sinh y wn ( x, y ) = BD sin 2a 2a 2a

Solution
Applying the first three boundary conditions, we have
w( x, y ) = K n sin
n =1

(2n 1) (2n 1) x sinh y 2a 2a

The final boundary condition is:

w( x, b) = w0 sin

x
2a

(2n 1) (2n 1) = K n sin x sinh b which gives: w0 sin 2a n =1 2a 2a

We can see from this that n must take only one value, namely 1, so that K1 = and the final solution to the stress distribution is

w0

sinh

2a

w( x, y ) =

w0 sinh

b
a

sin

x
a

sinh

y
a

Check out: http://mathworld.wolfram.com/HyperbolicSine.html

PDEs in other coordinates

In the vector algebra course, we find that it is often easier to express problems in coordinates other than (x,y), for example in polar coordinates (r,) Recall that in practice, for example for finite element techniques, it is usual to use curvilinear coordinates but we wont go that far
We illustrate the solution of Laplaces Equation using polar coordinates*
*Kreysig, Section 11.11, page 636

A problem in electrostatics
Radius a r

V ( r , , z ) = U on the upper half

This is a cross section of a charged cylindrical rod.

Thin strip of insulating material 0V

I could simply TELL you that Laplaces Equation in cylindrical polars is:

2V 1 V 1 2V 2V V= 2 + + 2 + 2 =0 2 z r r r r
2

brief time out while I DERIVE this

2D Laplaces Equation in Polar Coordinates

r x y

x = r cos y = r sin
r = x2 + y2
y = tan x
1

2 2 u u 2 u= 2 + 2 =0 x y

where

x = x(r , ), y = y( r , )

u ( x, y ) = u ( r , ) u ( r , ) = 0
2

So, Laplaces Equation is

We next derive the explicit polar form of Laplaces Equation in 2D

Recall the chain rule:

u u r u = + x r x x

Use the product rule to differentiate again

2u u 2 r u r u 2 u + + = + 2 2 2 r x x r x x x x x

(*)

and the chain rule again to get these derivatives

u u r = + x r r r x

u r x

2u r 2u = 2 + r x r x

u u r + = x r x

u 2u r 2u = + 2 r x x x

The required partial derivatives

x = r cos
2 2 2

y = r sin

r= x +y
2

y = tan x
1

r r x r = x + y 2r = = 2x x r x

r y = Similarly, y r

in like manner .

2r y 2 2r x 2 = 3, 2 = 3 2 x r y r y x = 2, = 2 x r y r 2 2 xy 2 2 xy = 4 , 2 = 4 2 x r y r

Back to Laplaces Equation in polar coordinates

Plugging in the formula for the partials on the previous page to the formulae on the one before that we get:

2u 2u x 2 u y 2 2u 2 xy u 2 xy 2u y 2 = 2 2+ + + + 2 4 2 3 3 4 r r r r r r r r x
2 2 2 2 2 2 2 u u y u x u 2 xy u 2 xy u x Similarly, = 2 2 + + + 2 4 2 3 3 4 r r r r r r y r r

So Laplaces Equation in polars is

2u 2u 2u 1 u 1 2u =0 + 2 + 2 = 2+ 2 2 r r r r y x

u u + 2 =0 2 x y is equivalent to
2 2

u 1 u 1 u =0 + + 2 2 2 r r r r
2 2

Example of Laplace in Cylindrical Polar Coordinates (r, , z)

Consider a cylindrical capacitor Radius a r

V ( r , , z ) = U on the upper half

P(r,,z) z

r x
Thin strip of insulating material 0V

Laplaces Equation in cylindrical polars is:

Boundary conditions

V ( a , ) = U : 0 V ( a , ) = 0 : 2
In the polar system, note that the solution must repeat itself every = 2
V should remain finite at r=0

2V 1 V 1 2V 2V 2 + 2 + 2 =0 V= 2 + 2 r r r r z

There is no variation in V in the z-direction, so

This means we can treat it as a 2D problem Using separation of variables

V =0 z 2V 1 V 1 2V + 2 =0 + 2 2 r r r r

V = R(r ) ( )
1 1 R + R + 2 R = 0 r r

As before, this means

R + R r = R r2

R + R r = = k, a constant 2 R r

The case k=0

= 0 ( ) = a + b R R + = 0 R(r ) = (C ln r + D ) r and so V (r , ) = (a + b)(C ln r + D )
The solution has to be periodic in 2: a=0 The solution has to remain finite as r0: c=0

V (r , ) = bd = g, a constant

The case k<0

Suppose that k = m 2 + m 2 = 0 ( ) = ( Am cosh m + Bm sinh m ) R m 2 R + + 2 R = 0 R (r ) = Cm r m + Dm r m r r

The solution has to be periodic in , with period 2. This implies that A = B = 0 V ( r , ) 0

m m

The case k>0

Suppose that k = n 2 + n 2 = 0 ( ) = ( An cos n + Bn sin n ) R n 2 R + 2 R = 0 R(r ) = Cn r n + Dn r n r r V (r , ) = ( An cos n + Bn sin n ) Cn r n + Dn r n

Evidently, this is periodic with period 2 To remain finite as r0

Dn = 0

The solution
V (r , ) = g + r n ( An cos n + Bn sin n )
n

Notice that we have not yet applied the voltage boundary condition!! Now is the time to do so

U g + a n ( An cos n + Bn sin n ) = V (a, ) = n 0

2

< 2

Integrating V from 0 to 2:
2

V (a, )d = U
0

Left hand side:

n ( An cos n + Bn sin n )d = 2g g a + 0 n

and so g =

U 2

Solving for Am and Bm

So far, the solution is

U V (a, ) = + a n ( An cos n + Bn sin n ) 2 n

2

We apply the orthogonality relationships:

2

U V ( a , ) cos m d = 2 0

cos md +
0

a
0 n

( An cos n + Bn sin n ) cos md

U cos md = 0 + Am a m
0

0 = Am a m , and so Am = 0, for all m

2

U V a m d ( , ) sin = 2 0

sin md + a
0 0 n 2

( An cos n + Bn sin n ) sin md

U cos m U sin md = + Bm a m 2 m 0 0 2U = Bm a m , for odd m = (2n 1) m

V (r , ) =

U 2U + r ( 2 n 1) sin(2n 1) ( 2 n 1) 2 n (2n 1)a

U 2U V (r , ) = + 2

r ( 2 n 1 ) sin( 2 n 1 ) ( 2 n 1 ) n =1 ( 2 n 1 )a

Check for r = a, = /2:

U 2U V (r , ) = + 2

a ( 2 n 1 ) sin( 2 n 1 ) ( 2 n 1 ) 2 n =1 ( 2 n 1 )a

= =

U 2U 1 3 1 5 + sin + sin + sin + ...... 2 2 2 5 2 3 U 2U 1 1 1 + 1 + + ...... 2 3 5 7 =U 4

U r

U 2U = + 2

0V

An application in image analysis

We saw that the Gaussian is a solution to the heat/diffusion equation We have studied Laplaces equation The next few slides hint at the application of what we have done so far in image analysis This is aimed at engaging your interest in PDEs it is not examined

Laplaces Equation in image analysis

How do we compute the edges?

Image fragment

Remove the noise by smoothing Find places where the second derivative of the image is zero
Signal position, x Zero crossing atmax step. Butstep. Constant gradient + at the Note amplified noise noise doubly amplified

d2 intensity/ dx2 d intensity/ dx

Edge map

intensity

Gaussian smoothing

Blurring with a Gaussian filter is one way to tame noise

Zero crossings of a second derivative, isotropic operator, after Gaussian smoothing

I smooth = 0
2
An application of Laplaces Equation!

Limits of isotropic Gaussian blurring

A noisy image

Gaussian blurring

Gaussian is isotropic takes no account of orientation of image features so it gives crap edge features

 (G I ) = 0
2

As the blurring is increased, by increasing the standard deviation of the Gaussian, the structure of the image is quickly lost. Can we do better? Can we make blurring respect edges?

Anisotropic diffusion
t I = ( g ( x; t )I )
T |I |2 k2

g ( x; t ) = e

, for some constant k, or 1 g ( x; t ) = 2 2 1+ | I | / k

This is a non-linear version of Laplaces Equation, in which the blurring is small across an edge feature (low gradient) and large along an edge.

Anisotropic blurring of the noisy image Top right: Gaussian Bottom left and right: different anisotropic blurrings

Example of anisotropic diffusion brain MRI images, which are very noisy. Top: Gaussian blur Middle and bottom: anistropic blur

Anisotropic blurring of the house image retaining important structures at different degrees of non-linear blurring

Two final examples: Left original image Right anistropic blurring