Harmonic Functions are Real Analytic1

On this very short note we prove that harmonic functions are real analytic functions.
Lemma 1 is used on the inductive step of the proof of the main theorem. We prove Lemma 1 using the mean
value property of harmonic functions, Greens theorem and the maximum principle.
Lemma 1 (estimate on first derivative) Suppose w is harmonic in Rn , > 0, y and B(y; ) .
Then for any i {1, . . . , n},
n
|wxi (y)| kwkL(B(y;)) .

Proof: Since w is harmonic, wxi is also harmonic. Therefore we can use the mean value property on wxi .
Using also Greens theorems,
   
 1 Z   1 Z 
|wxi (y)| =  w dx = w dx 
   
x i
 (n)n B(y;) i
  (n)n B(y;)
 


1 n
n(n)n1 kwkL (B(y;)) = kwkL (B(y;)) .
(n)n
Using the maximum principle on w,
n n
|wxi (y)| kwkL (B(y;)) kwkL (B(y;)) .



Corollary 1 (Liouvilles Theorem) Any harmonic and bounded function on Rn is constant.

Proof: Since is harmonic on Rn we can let goes to infinity on Lemma 1 and conclude that its derivative is
zero at any point.

The idea of the proof of Theorem 1 is to estimate a derivative of order k using balls with radius r j = j/(k+1),
for j = 1, . . . , k + 1. We proceed by induction on k.
Theorem 1 (estimates on higher derivatives) Suppose u is harmonic in R n , r > 0, z and
B(z; r) . Then for any multi-index of order k,
 k
kn
|D u(z)| kukL(B(z;r)) . (1)
r
Proof: We will prove by induction on k.
For k = 0, |u(z)| kukL(B(z;r)) .
Suppose (1) is true for || = k. Then given any multi-index of order k + 1, there exists a multi-index
of order k such that D u = (D u)xi for some i {1, . . . , n}. Since u is harmonic, D u is harmonic also. From
Lemma 1, for w = D u, y = z and = r/(k + 1), which we can apply since B(z; r/(k + 1)) B(z; r) , we
obtain
(k + 1)n
|D u(z)| kD ukL(B(z;r/(k+1))) . (2)
r
If x B(z; r/(k + 1)), then B(x; kr/(k + 1)) B(z; r) since r/(k + 1) + kr/(k + 1) = r. Therefore,
using the induction hypotheses (1) on for z = x, r = kr/(k + 1) (we can apply it since B(x; kr/(k + 1)) )
 k  k
kn(k + 1) n(k + 1)
|D u(x)| kukL(B(x;kr/(k+1))) kukL(B(z;r)) .
kr r
Using (2),
 k  k+1
(k + 1)n n(k + 1) n(k + 1)
|D u(z)| kukL(B(z;r)) kukL(B(z;r)) .
r r r
This verify (1) for || = k + 1.

Now we need an easy estimate on k k for our main Theorem 2. This can be done using Stirlings formula (see
Federer), but our proof below is more elementary.
1 Verso 03.april.2008

1
Lemma 2 (estimate on k k ) For any k N,

k k < k!ek1 .

Proof: One can prove by induction (see (J. R. Felicio; Frmula de Stirling em tempos de Maple; Revista de
Matemtica Universitria, 17, (1994)) that
k1
Y
k k = k! (1 + 1/j)j .
j=1

Since (1 + 1/j)j < e (the sequence is monotone using Newtons binomial formula) for every j, k k < k!ek1 .


Theorem 2 (harmonic functions are analytic) If u is harmonic in Rn then u is real analytic in .

Proof: Fix any x0 . We will show that its Taylor series is convergent in a neighborhood of x0 .
Let r = 21 dist(x0 , ). For each k, the remainder Rk (x) of the Taylor series of u is
X (x x0 )
Rk (x) = D u() , (3)
!
||=k

where = x0 + t(x x0 ), for some 0 t 1, t depending on x. Let us assume that |x x0 | < < r for some
> 0 that we are going to choose later. Since | x0 | = |t||x x0 | |x x0 | < < r, we see that B(x0 ; r).
Since B(; r) B(x0 ; 2r) , we can apply Theorem 1 for z = and obtain
 k  k
kn kn
|D u()| kukL(B(;r)) kukL(B(x0 ;2r)) . (4)
r r

Using Lemma 2 (estimate on k k ) and (4), defining M = kukL(B(x0 ;2r)) , we obtain

 k
kn M  en k
|D u()| M k!. (5)
r e r

It is easy to check that |(x x0 ) | |x x0 ||| . Since || = k and |x x0 | < ,

|(x x0 ) | k . (6)
k!
= nk ), we obtain
P
Putting together equations (3), (5) and (6), and using the multinomial theorem ( ||=k !

k
M en2

M  en k X k!
|Rk (x)| = .
e r ! e r
||=k

en2
Therefore Rk converges to zero uniformly for x B(x0 ; ) when k goes to infinity if r < 1. So the Taylor
series is convergent for any x B(x0 ; ) if < enr 2 < r.