Sie sind auf Seite 1von 9

7th November, 2011.

By Samsung, SUM Sung Fung.

MATH4052 - Partial Differential Equations


Tutorial Notes #08
Samsung, SUM Sung Fung <delamath@ust.hk>

 Analysis on Diffusion Equation

The solution to the diffusion equation:

ut ku xx = f ( x, t ) x , t > 0
u ( x, 0) = ( x) Initial Conditions(IC)
ut kuxx = f ( x, t ) x ( a, b) , t > 0
OR u ( x, 0) = ( x) Initial Condition(IC)
+ Boundary Conditions(BCs) at x = a and x = b
is very hard to find by the tricks we have used for wave equations.
So what we are going to do is to convince you that we get the solution from
some secret magic.

Before telling you the solution, we shall prove that our solution (if exists) will
be the only possible solution, so that you cant find anything to challenge our
magic.

This leads to our consideration on the well-posedness criterion:

1. Existence: There is some solution u ( x, t ) satisfying the


diffusion PDE, as well as the IC and the BCs.
2. Uniqueness: If there are u1 ( x, t ) and u2 ( x, t ) both satisfying
all the conditions, then u1 = u2
3. Stability: The solution u ( x, t ) depends continuously on the
given IC and BCs, for example:
If v1 and v2 satisfy essentially the same diffusion equation except a
slight difference that u1 ( x,0) = 1 ( x) and u2 ( x, 0) = 2 ( x) , then

1 2 implies u1 u2

Remarks: There are many choices for measuring the closeness of two functions.

MATH4052 Tutorial Notes #08 P.1 of 9


7th November, 2011.
By Samsung, SUM Sung Fung.

 Maximum Principle for Diffusion Equation

One of the most powerful properties of the diffusion equation is the so-called
Maximum Principle, and (the weaker version) is stated as follows:

If u ( x, t ) is a function defining on the rectangular domain [ a, b] [0, T ] ,


which satisfies a homogeneous diffusion equation, then any (global)
maximum of u will be found either at t = 0 , x = a or x = b , i.e.
max {u ( x, t )} = max {u ( x, t )}
x[ a ,b ] x = a ,b
t[0,T ] t =0

The possible regions for maximum values are pointed out in the figure below:
t t

T T

x
a b
Furthermore, by applying Maximum Principle to the function
v ( x, t ) = u ( x, t ) , which is also a solution to a homogeneous diffusion equation,
and make use of the equality:
max {v} = max {u} = min {u}

one can derive the Minimum Principle for u .

 Uniqueness of Solution to General Diffusion Problem


If there are u1 ( x, t ) and u2 ( x, t ) both satisfying the same system of PDE:
ut ku xx = f ( x, t ) x ( a, b) , t > 0
u ( x, 0) = ( x) Initial Condition(IC)
u (a, t ) = g (t )
And Dirichlet Boundary Conditions(DBC): t > 0
u (b, t ) = h(t )

ux (a, t ) = g (t )
OR Neumann Boundary Conditions(NBC): t > 0
ux (b, t ) = h(t )
then u1 = u2 .

MATH4052 Tutorial Notes #08 P.2 of 9


7th November, 2011.
By Samsung, SUM Sung Fung.
Standard procedure:
Consider the function w( x, t ) u1 ( x, t ) u2 ( x, t ) and show that w = 0 .
It is easy to observe that w satisfies the following system of equations:
wt kwxx = f f = 0 x (a, b), t > 0

w( x, 0) = = 0
w(a, t ) = g g = 0
with DBC: t > 0 1
w(b, t ) = h h = 0

wx (a, t ) = g g = 0
OR NBC: t > 0
wx (b, t ) = h h = 0
Now we have two different methods to prove w = 0 , both are worth mentioning.
Method 1: Maximum Principle
Suppose DBC is specified, since w( x, t ) is a solution to a homogeneous
diffusion equation, the Maximum Principle and the Minimum Principle hold,
that is:

min {w( y , )} w( x, t ) max {w( y , )} for all x [ a, b ] and t 0


y = a ,b y = a ,b
=0 =0

But min {w( y , )} = max {w( y , )} = 0 , this forces w = 0 .


y = a ,b y = a ,b
=0 =0

Suppose NBC is specified, we can consider wx ( x, t ) as a solution to a


homogeneous diffusion equation:

Differentiating both wt kwxx = 0 and w( x, 0) = 0 with respect to x , we get

( wx )t k ( wx ) xx = 0 and wx ( x, 0) = 0

Also with NBC, wx (a, t ) = wx (b, t ) = 0 ; these force wx ( x, t ) = 0 by the


Maximum and Minimum Principles.

But wx ( x, t ) = 0 wxx ( x, t ) = 0 , put this back to wt kwxx = 0 , we get wt = 0

So w is unchanged with time t , i.e. w( x, t ) = w( x, 0) = 0

Method 2: Energy Method

Multiplying w to the equation wt kwxx = 0 , and making use of the identities:


1 2
w = wwt and ( wwx ) x = wwxx + wx2
2 t
We have:
1
0 = w ( wt kwxx ) = w2 + kwx2 k ( wwx ) x
2 t

MATH4052 Tutorial Notes #08 P.3 of 9


7th November, 2011.
By Samsung, SUM Sung Fung.

Integrating over the interval a < x < b , we get


b 1 b b
0 = w2 dx + kwx2 dx k ( wwx ) x dx

a 2
t a

a
 
b
kwwx
a

Here the last term vanishes by either DBC (w x = a ,b


=0 ) or NBC (w
x x = a ,b )
=0 ,

d b1 2 b 1 b
w dx = w2 dx = k ( wx ) dx 0
2

dt a 2
a 2
t a

b
Therefore, a
w2 dx is decreasing, we have

b b
[ w( x, t )] dx [ w( x, 0) ] dx = 0 for all t > 02
2 2
a a

And this forces w = 0 by vanishing theorem.

 Stability of Solution to General Diffusion Problem


The solution to a general diffusion problem depends continuously on the initial
condition, given that the differential equation and the boundary conditions are
fixed when the IC varies. That is,

Suppose u1 ( x, t ) and u2 ( x, t ) both satisfy ut ku xx = f ( x, t ) and the same


Boundary Conditions at x=a and x = b , but u1 ( x,0) = 1 ( x) and
u2 ( x, 0) = 2 ( x) , then
1 2 implies u1 u2

It turns out that we can use similar approaches as in the proof of uniqueness,
by considering the function w = u1 u2 , which now satisfies most conditions
in 1 EXCEPT we have to change the IC to w( x, 0) = 1 ( x) 2 ( x)

Method 1: Maximum Principle


Suppose the DBC is specified, we can use Maximum Principle to show that
Max.
Prin.
w( x, t ) max {w( y , )} max { w( y , ) } = max { 1 2 }
y = a ,b y = a ,b
=0 =0

Similarly by Minimum Principle, we have


Min.
Prin.
w( x, t ) min {w( y , )} min { w( y , ) } = max { w( y, ) } = max { 1 2 }
y = a ,b y = a ,b y = a ,b
=0 =0 =0

Combining the two inequalities, we have

max { 1 2 } w( x, t ) max { 1 2 }
max u1 u2 = max w max 1 2
x[ a ,b ] x[ a ,b ] x[ a ,b ]

MATH4052 Tutorial Notes #08 P.4 of 9


7th November, 2011.
By Samsung, SUM Sung Fung.

So for max 1 2 0 , we have max u1 u2 0


x[ a ,b ] x[ a ,b ]

This is called the uniform stability.

Method 2: Energy Method


From inequality 2 in the uniqueness proof, we have
b b b b
[u u2 ] dx = w2 dx [ w( x, 0) ] dx = [1 2 ] dx
2 2 2
a 1 a a a

b b
[ [u
2 2
So for 1 2 ] dx 0 , we have 1 u2 ] dx 0
a a

This is called the L2 -stability, generally weaker than the uniform one.

 Diffusion on the Whole Line

The solution to the system of diffusion equation with source:


ut kuxx = f ( x, t ) x , t > 0

u ( x,0) = ( x)
is given by
t
u ( x, t ) = ( ) S ( x , t )d + f ( y, s ) S ( x y, t s )dyds


  0

u homo ( x ,t ) u part ( x ,t )

y2
1
where S ( y, s ) = e 4 ks
is the Diffusion Kernel.
4 ks
Since we have proved the uniqueness of the solution and the stability of the
system, now what we need to show is the given u ( x, t ) fits the system:

(i) u ( x, t ) satisfies the PDE for x , t > 0


(ii) u ( x, 0) = ( x)

But the problem is that when t = 0 , S ( y , 0) is ill-defined, so we cannot


verify (ii) directly. Instead, based on the stability of the system, we can
consider a weaker form of satisfaction:
(ii)* lim u ( x, t ) = ( x )
t 0+

The property (i) can be verified by direct differentiations, and property (ii)*
can be verified once we grab some special characteristics of S ( y , t ) .

MATH4052 Tutorial Notes #08 P.5 of 9


7th November, 2011.
By Samsung, SUM Sung Fung.

 Special Properties of the Diffusion kernel

The diffusion kernel S ( y , t ) has the following properties:

a) St ( y, t ) kS yy ( y, t ) = 0

1
b) S ( y , t ) is Gaussian, centered at y = 0 , with peak value
4 kt

c)
S ( y, t )dy = 1 t > 0

d) For any fixed > 0 , max S ( y , t ) 0 as t 0 +


y <

Remarks: Property d is called uniform convergence, it guarantees exchanging of limit


and improper integrals!

Example: Solve the following IVP on diffusion equation:


ut uxx = 0 x , t > 0
0 x 2
u ( x, 0) = f ( x) =
1 x > 2
Solution:

u ( x, t ) = f ( y ) S ( x y, t )dy

2
= f ( y ) S ( x y, t )dy + f ( y ) S ( x y, t )dy
2
2
= 0 S ( x y, t )dy + 1 S ( x y, t )dy
2

= S ( x y, t )dy
2
( x y )2
1
=
4 kt
2
e 4 kt
dy

yx
Let p = , then
4kt
1 p2
4 kt
u ( x, t ) = 2 x e dp

2 x
1 1

p2 2
= e dp 4 kt e p dp
0
0

1 1 2 x
= Erf
2 2 4kt

MATH4052 Tutorial Notes #08 P.6 of 9


7th November, 2011.
By Samsung, SUM Sung Fung.

Appendix
Proof of Maximum Principle

Lets first denote the rectangular domain drawn in P.2 as = [ a , b ] [0, T ] and its
lower boundary as = {( x, t ) : x = a OR x = b OR t = 0} .

Suppose u ( x, t ) is a function defined on , and we have ut kuxx = 0 on ,


we want to show that the maximum will NOT be located at the place that the PDE
ut kuxx = 0 holds. In other words, the maximum can only exist on , where the
PDE does not hold. Denote M = max {u ( x, t )} as the maximum value on the lower

boundary, we shall show that u ( x, t ) M

Let > 0 and let v( x, t ) = u ( x, t ) + x 2 , our goal is to show that v( x, t ) M + b 2


on the whole domain , which in turns shows u ( x, t ) M + (b 2 x 2 ) . Since this is
true for any > 0 , we can conclude that u ( x, t ) M on the whole domain.

From the definition of v( x, t ) and M , we have v( x, t ) M + b 2 on . And v


satisfies vt kvxx = ut k (u + x 2 ) xx = ut ku xx 2 k = 2 k < 0 , which is the
diffusion inequality (note that it is a strict inequality).

Suppose v( x, t ) attends a maximum somewhere at the interior of , by calculus on


maxima, vt = 0 and u xx 0 , then vt kvxx 0 , violating the diffusion inequality.
Thus v( x, t ) attends NO maximum at the interior of .

Now suppose v( x, t ) attends a maximum along the top boundary t = T , say at point
( x0 , T ) , then vx ( x0 , T ) = 0 and vxx ( x0 , T ) 0 . Moreover, since from time t = T t
to t = T , v( x0 , t ) is increasing. So we have
v( x0 , T ) v( x0 , T t )
vt ( x0 , T ) = lim 0
t 0 t
The diffusion inequality is again violated, which means there is no allowable
maximum at t = T .

By Extreme Value Theorem (in Tutorial Notes#01), v( x, t ) must have a maximum


within the compact domain , so this maximum must be at , therefore
v( x, t ) vmax M + b2 .

at

This proves the (weaker version of) maximum principle.

MATH4052 Tutorial Notes #08 P.7 of 9


7th November, 2011.
By Samsung, SUM Sung Fung.

Verifying Property (ii)*: lim u ( x, t ) = ( x )


t 0+

Lets try to show that u ( x, t ) ( x) 0 as t 0 + .


Suppose ( x) is continuous and bounded in . By continuity of , for any given
> 0 and a fixed x , we can find > 0 such that ( y ) ( x) < whenever
y (x , x + ) .

Lets consider

u ( x, t ) ( x ) =
S ( x y, t ) ( y )dy ( x)

=

S ( x y, t ) [ ( y ) ( x) ] dy ( using


S ( x y, t )dy = 1)

S ( x y, t ) ( y ) ( x) dy

x +
= S ( x y, t ) ( y ) ( x) dy + S ( x y, t ) ( y ) ( x) dy
x x +
x
+ S ( x y, t ) ( y ) ( x) dy

For the first term:


x + x +
x
S ( x y, t ) ( y ) ( x) dy < S ( x y, t )dy S ( x y, t )dy =
 x
<
For the second and third terms, since is a bounded function on , there is some
fixed positive number M such that ( z ) M for all z , in particular,
( y ) ( x) ( y) + ( x) 2M . So,

lim S ( x y, t ) ( y ) ( x) dy 2M lim S ( x y, t )dy
t 0+ x + t 0+ x +


= 2M lim S ( z , t )dz (Let z = y x)
t 0+

= 2M lim S ( z , t )dz (By Uniform Convergence)
t 0+

= 2M 0dz = 0

x x
lim S ( x y, t ) ( y ) ( x) dy 2 M lim S ( x y, t )dy
t 0+ t 0+


= 2M lim S ( z , t )dz (Let z = y x)
t 0+


= 2M lim S ( z , t )dz (By Uniform Convergence)
t 0+


= 2M 0dz = 0

That means u ( x, t ) ( x) <

MATH4052 Tutorial Notes #08 P.8 of 9


7th November, 2011.
By Samsung, SUM Sung Fung.

Properties of the Error Function

2 x 2


The Error function Erf ( x ) e p dp has the following properties:
0

a) Oddness: Erf ( x ) = Erf ( x )

b) Limit at infinity: lim Erf ( x ) = 1


x +
d 2 x2
c) Derivative: Erf ( x ) = e
dx
1 2
d) Antiderivative: Erf ( x ) dx = x Erf ( x ) +
e x + C

Properties of Diffusion Kernel

a) Direct verification
b) Trivial fact
y2
1 y
c)
S ( y , t ) dy =
4 kt

e 4 kt
dy (put p =
4kt
)

1
p2


= e dp = 1

Where we have to use the polar coordinates trick:


2 2
I = e x dx = e y dy

2 2
=
e x e y dxdy

(
x2 + y 2 ) dxdy
=
e
2 2
= 0 0
e r rdrd
2
= 2 e r rdr
0


= e r ( 2
) 0
=

d) For any fixed > 0 , since S ( y , t ) is a Gaussian distribution,


2
1
max S ( y, t ) = S ( , t ) = e 4 kt
y < 4 kt
2
+ 1
As t 0 , max S ( y, t ) = e 4 kt 0 by l'Hpital's Rule.
y < 4 kt

MATH4052 Tutorial Notes #08 P.9 of 9

Das könnte Ihnen auch gefallen