S ep.

1 7

The L aplace/ P oisson Equations: Explicit Formulas In this lecture we study the properties of the Laplace equation u = 0, and the Poisson equation u = f, with Dirichlet boundary conditions u= g through explicit representations of solutions. We call a function u harmonic in a region if x Rd x Rd x u = 0 in . (1) ( 2) ( 3)

1 . Fundamental solution. Recall that to solve a linear constant-coecient PDE P( D) u = f in Rd , it suces to nd D such that P( D) = .

( 4)

Now we try to nd such for P( D) =

We notice that the equation = is invariant with respect to rotations/ reections, in other words, if ( x) is a solution, then ( O x) is also a solution where O is an orthogonal matrix. Therefore it is reasonable to suspect that the solution takes the form Simple computation yields 2 xi = x

2 x 2 1

2 . x 2 d

( x) = V( r) , r 1 = x2 + x i 2 1 x = V ( r) i ; r x i

r = x2 + 1
1 /2

+ x2 n xi r

1/2

( 5)

+ x2 n

0,

( 6) ( 7) ( 8)

which gives Summing up we have

2 x2 i = V ( r) 2 + V ( r) r x 2 i = V ( r) + n 1 V ( r) r

1 x2 i . r r3

when r > 0. Since = , from general theory of distributional solutions of elliptic PDEs we know that Therefore is C in Rn \ { 0} , which means Thus it is reasonable to solve sing supp G sing supp = { 0} . ( 9)

= 0 holds in the classical sense for x x

0. ( 1 0)

=0

0.

rst to narrow down the candidates. When depends only on r = | x | , we have V ( r) + This gives log( V ) = n 1 V ( r) = 0 r r > 0. a rn 1 (11) ( 1 2)

for some constant a. Integrating we obtain

for constants b and c. In particular, we can take special b and c to obtain

b ln r + c n = 2 V( r) = b n 2 + c n 3 r

1n r

V ( r) =

( 1 3)

Denition 1 . The distribution

where ( n) is the volume of the n- dimensional unit ball (or equivalently, n ( n) is the area of the n 1 dimensional unit sphere), is called the fundamental solution of the Laplace s equation. One can verify that tional derivatives, = holds in the sense of distributions. According to the denition of distribu( ) ( ) ( 1 ) (
2

1 ln | x | n=2 2 ( x) 1 1 ( 2 n) n ( n) | x | n 2 n = 3

( 1 4)

) = (

) =
Rn

( 1 5)

where the last equality comes from the fact that is locally integrable, we only need to show that
Rn for any C0 . To show this, we need to rst establish the so-called Green s formula:

= ( 0)

( 1 6)

Lemma 2. Let u , v be C 2 in , then u

vv

u dx =

u v v dS. n n

( 1 7)

where n is the outward unit normal vector of . Proof. Recall the Gauss theorem: Now we have u vv u dx = = =

f dx = [ (u (u u (n u

f n dS. u v] [ (v u) v u] dx

( 1 8)

v) vv

u) dx u) dS ( 1 9)

v) v ( n

u v v dS. n n

Now we show that = for n = 2, and leave Take any C0 R2 , there is R > 0 such at the origin and with radius R. Now take > to u = and v = . Keep in mind that has Since = 0 in , we have
R2

the n 3 case as an exercise. that supp BR where BR denotes the open ball centered 0 small. We set = B R \ B and apply the Green s formula two parts.
B

= lim

BR \ B

Here n out = for x BR while n in ( x) = Checking term by term, we have B = 0 for the same reason.

x R

+ lim n ou t 0 n ou t 0 B R A + B + C + D. = lim
x

n in n in ( 20)

for x B . = 0.

A = 0 since supp BR which means

n o u t

For C, we have
B

n in

C | log | ,

| log | sup |
x B

| ( 21 )

therefore C = 0. Finally, for D, we have for x B , Thus x = n in = x 1 2 1 x r r

r= | x | =

1 . 2

( 22) ( 23)

D = lim

1 = lim n in 2

= ( 0) .
B

In summary, u = f solves the Poisson equation u= f in the sense of distributions for any f E ( Rn ) .

( 24)

2. The Green s function for the Laplace equation. In practice, we are more interested in solving the Poisson equation on a domain with boundary instead of the full space. We explicitly dene where is the fundamental solution, to make the presentation clearer. The Dirichlet problem takes the form u= f In this case, we have Theorem 3. Suppose is a bounded domain in Rn , and that u C 1 ( ) C 2 ( ) . Then for any a there holds u u( a) = ( a , x) u( x) dx ( a , x) ( 27) ( x) u( x) ( a , x) dSx . n x n x The proof uses the Green s formula and can be found in F. Lin and Q. Han Elliptic Partial Dierential Equations. One further notices that for any ( a , x) satisfying x = 0 ( here x denotes the Laplacian with respect to the variable x) , we have u( a) = where

( a , x) = ( x a)

( 25)

in ,

u= g

on

( 26)

( a , x)

u( x) dx

u ( a , x) ( x) u( x) ( a , x) n x n x

dSx .

( 28) ( 29)

( a , x) = ( a , x) + ( a , x) . Using the equation and boundary conditions, we have u( a) =

( a , x) f( x) dx +

g( x)

( a , x) dS n x

u ( a , x) ( x) . n x

( 30)

To be able to nd u( a) , all we need to do is to nd a function G( a , x) such that 1 . G( a , x) = ( a , x) + ( a , x) where is harmonic w. r. t. x, For such G, we have 2. G( a , x) = 0 for x . u( a) =

G( a , x) f( x) dx +

g( x)

G ( a , x) dS. n x

( 31 )

where both terms on the RHS are known ( of course only when we can construct this G! ) . The conditions on G implies that solves =0 in , ( a , x) = ( a , x) on . ( 32)

This can be done by cleverly transforming and combining s when the domain has simple geometry. However when for general domains obviously the problem is as hard as the original Dirichlet problem for u. Example 4. ( Green s function for the ball BR ) For each a BR , we need to nd ( a , x) such that
x( a,

x) = 0

in BR ;

( a , x) = ( a , x)

on BR .

( 33)

For simplicity of presentation, we abuse notation and write ( a , x) as ( | x a | ) . The idea is to nd a point b = b( a) BR and a constant c, then set ( a , x) = ( c | x b( a) | ) . Note that for any b B R , ( c | x b | ) is harmonic in BR . Thus all we need to do is nding appropriate c and b so that From the explicit formula of we see that this can hold if and only if | x a | = c | x b | for all x BR .
a1

( | x a | ) = ( c | x b | ) .

( 34)

an

bn

Simplifying, we obtain

1 c2

x2 + 1

+ x2 = 2 n

a 1 c2 b 1 x 1 +

+ a n c2 b n x n + c2 b2 i

For this to hold for all x BR , we need for all x B R . As a consequence we have 1 whose solutions are c2 = | a| R2 R2 R2 b = a/ c 2 = a. | a| 2 = | a| R2 x a R | a| 2 | a| R2 x a R | a| 2 . a2 i
2

1 c2 R2 = 2

a 1 c2 b 1 x 1 +

+ a n c2 b n x n + c2 i = 1, , n a2 i

c2

a i = c2 b i R2 = c2

b2 i

Thus

and

( a , x) =

G( a , x) = ( | x a | )

Remark 5. One easily checks that G( a , x) = G( x , a) for the Green s function on BR , by writing | a| R2 | a| R x 2 a = R x | a| a = R | a| | a| | x| 2 a x + R2 R2
2 2 1/2

This turns out to remain true for other Green s functions. A formal argument to convince oneself about this is the following.

2 ( x1 a1 ) +

Let a =

b =

b1

Then | x a | = c | x b | becomes

+ ( x n a n ) 2 = c2 ( x 1 b 1 ) 2 +

+ ( x n bn) 2 b2 i a2 i a2 . i

( 35)

( 36)

( 37)

( 38) ( 39)

( 40) ( 41 )

( 42)

( 43)

( 44)

Let a be arbitrary. We would like to show G( a , x) = G( x , a) for all x . Writing G 1 ( x) G( a , x) , the goal becomes to show that G 1 ( x) = G( x , a) . Now G 1 ( x) solves G 1 = a , Thus formally we have G 1 ( x) =

G1 = 0

on ,

( 45) ( 46)

G( x , y) a ( y) = G( x , a) .

Obviously the above argument has several holes. For example G( x , y) C( ) so the action of a ( y) on it is not dened. However one can make it rigorous by computing ( G1
\ B ( x 1 ) B ( x 2 )

G2 G2

G 1 ) dx

( 47)

where G 1 ( x) = G( x 1 , x) and G 2 ( x) = G( x , x 2 ) , and then let Equations for details. 3. Harmonic functions. A function u C 2 ( ) is said to be harmonic if u( a) =

0. See L. C. Evans Partial Dierential

u = 0. Let g be its boundary value, we see that G ( a , x) dSx . n x ( 48)

g( x)

In particular, when = BR , we have the Poisson representation formula u( a) = 4. Well-posedness. Given the Dirichlet problem R2 | a | 2 n ( n) R g( x) dSx . | a x| n on , ( 49)

B R

u = f,

u= g

( 50)

as soon as we have the Green s function, we can write u( a) =

G( a , x) f( x) dx +

g( x)

G ( a , x) dSx . n x

( 51 )

There are several issues need to be settled ( they will be settled in the following few weeks) . Existence. The formula itself does not give us existence of the solution per se. We need to show that u = f indeed holds and u takes g as its boundary value. In light of the next section, we can set v = u ( x y) f( y) which satises the Laplace equation v = 0 with some boundary conditions. Thus the question becomes whether we can show the existence of solutions to the Laplace equation. When = BR , it suces to show that the Poisson integral representation indeed gives a solution to the Laplace equation. For general , it is not possible to nd a Green s function with explicit formula. It turns out somehow one can show the existence of solution to the Laplace equation u = 0 through solving it iteratively on balls inside the domain. This is the Perron s method. Uniqueness. It can be easily seen that if u 1 , u 2 solves the same Poisson s equation, their dierence u 1 u 2 satises the Laplace equation with zero boundary condition. Thus we only need to show that the zero function is the only solution in this situation. This is a consequence of the maximum principle. Regularity. In the following section we prove a trivial regularity result. In the next lecture we will prove much more rened regularity versions.

5. Regularity a glimpse. 2 Let ( x) be the fundamental solution of the Laplacian. Assuming f Cc ( Rn ) ( twice continuously differentiable with compact support) , then we have u = R n ( x y) f( y) C 2 ( Rn ) . First write u=
Rn

( x y) f( y) =

Rn

( y) f( x y) .
f( x x i

( 52) y) and ( y)
2 f( x x i x j

Recall that is locally integrable, thus by assumption both ( y) integrable. As a consequence one has u = x i ( y) f( x y) x i and

y) are ( 53)

2 u 2 = ( y) f( x y) x i x j x i x j
f( x x i

following from properties of the Lebesgue integrals. 1 Since both formly continuous, u C 2 .

y) and

2 f( x x i x j

y) are uni-

The above result is obviously not satisfactory as no improvement on regularity is obtained, while intuitively, since f is a linear combination of the double derivatives of u, one would expect the regularity of u u C 2 ( which is obviously to be better than that of f. For example, it is natural to guess that f C true in the one dimensional case) . Unfortunately this particular conjecture is wrong, but u turns out to be indeed twice more dierentiable than f if we use the right function spaces. We will discuss this issue more in the next lecture. Further readings. L. C. Evans, Partial Dierential Equations, 2. 2. D. Gilbarg, N. S. Trudinger, Elliptic Partial Dierential Equations of Second Order, Chapter 2, Classics in Mathematics, Springer. Fanghua Lin, Qing Han, Elliptic Partial Dierential Equations, 1 . 3, Courant Lecture Notes 1 , AMS.

Exercises.
E xercise 1 . P rove that the fundamental solution for dimensional at least 3: ( x) = solves the equation 1 1 ( 2 n) n ( n) | x | n 2 = in the sense of distributions. Remember that n ( n) is the area of the unit sphere in E xercise 2 . C onstruct the G reen s function for the half-space { x n > 0} . ( 0,

( 55) ( 5 6) Rn. results.

1 . O ne can also prove directly using nite dierences. For example, let e i be the unit vector in x i direction, that is e i = , 1 , , 0) with the only 1 in the i-th p osition. T hen

u( x + e i ) u( x) =
2 S ince f Cc , f ( x + ei y) f ( x y) h

( y)
Rn

f ( x + e i y) f ( x y) d y. h
f x i

( 5 4)

f (x x i

y) uniformly because

is uniformly continuous.