The Dirichlet Test

Theorem (The Dirichlet Test) Let X be a metric space. If the functions fn : X → C,

Proof: The trick to this proof is the summation by parts formula, which we now derive.
n
X
sn (x) = fk (x)gk (x)
k=1
n
X
= F1 (x)g1 (x) + [Fk (x) − Fk−1 (x)]gk (x)
k=2
Xn n
X
= F1 (x)g1 (x) + Fk (x)gk (x) − Fk−1 (x)gk (x)
k=2 k=2
n
X n−1
X
= Fk (x)gk (x) − Fk (x)gk+1 (x)
k=1 k=1
Xn
 
= Fk (x) gk (x) − gk+1 (x) + gn+1 (x)Fn (x)
k=1

So if m > n, the difference between the mth and nth partial sums is
m
X  
sm (x) − sn (x) = Fk (x) gk (x) − gk+1 (x) + gm+1 (x)Fm (x) − gn+1 (x)Fn (x)
k=n+1
 
If M = sup |Fn (x)|  x ∈ X, n ∈ IN ,
m
X
   
sm (x) − sn (x) ≤ M gk (x) − gk+1 (x) + M gm+1 (x) + M gn+1 (x)
k=n+1
  (1)
= M gn+1 (x) − gm+1 (x) + M gm+1 (x) + M gn+1 (x)
= 2M gn+1 (x)
since gm+1 (x) ≥ 0, gn+1 (x) ≥ 0 and every gk (x) − gk+1 (x) ≥ 0. For each fixed x,
lim gn+1 (x) = 0. So (1) guarantees that {sn (x)} is a Cauchy sequence and hence con-
n→∞
verges. Call the limit s(x). Taking the limit of (1) as m → ∞ gives
 
s(x) − sn (x) ≤ 2M gn+1 (x)

c Joel Feldman. 2008. All rights reserved. February 3, 2008 The Dirichlet Test 1
Since gn+1 (x) converges uniformly to zero as n → ∞, we have that sn (x) converges uni-
formly to s(x) as n → ∞.

z n
P∞
P∞ 1 z
n
Example. We shall consider three different power series: n=0 R , n=0 n R and
P∞ 1 z
n
n=0 n2 R , for some fixed R > 0. For all three series, the radius of convergence is
exactly R since, for ℓ ∈ {0, 1, 2},
q  q ℓ
lim sup nℓ Rn = R lim sup n n1
n 1 1 1
= R1
n→∞ n→∞

So all three series converge for all complex numbers z with |z| < R and diverge for all
complex numbers with |z| > R. What if |z| = R?
z n
P∞

We’ll start with the series n=0 R . Then we can compute exactly the partial sum

z n+1


n  1 − R
if z 6= R

z m
X

Fn (z) = R
= 1 − z (2)
 R
m=0 
n+1 if z = R
As expected, if |z| < R this converges to 1−1 z as n → ∞. Also as expected, this diverges
R
 z
n+1 
 =  z n+1 → ∞. I claim that this also diverges whenever
 
for |z| > R, because  R R
z n+1


|z| = R. For z = R, it is obvious because n + 1 → ∞. For |z| = R with z 6= R, R
does not blow up as n → ∞, but it cannot converge either, because
 z
n+2 z n+1 
  z n+1  z   
 − 1 =  z − 1



R − R =  
R R R
P∞ z

n
is independent of n. So the geometric series n=0 R , which has radius of convergence
R, converges if and only if |z| < R.
P ∞ 1 z
n
The third series, n=0 n2 R , converges for all |z| ≤ R, by comparison with
P∞ 1
n=0 n2 . As the series Phas radius of convergence R, it converges if and only if |z| ≤ R.
∞ z n
The middle series n=0 n1 R


has a more interesting domain of convergence. Of
course the radius of convergence is exactly R, so the series converges for all complex
numbers z with |z| < R and diverges for all complex numbers with |z| > R. What if
z n
P∞ P∞
|z| = R? Well, if z = R, then the series is n=0 n1 R = n=0 n1 which diverges. So

that leaves |z| = R but with z 6= R. This is where the Dirichlet test comes in handy. Fix
any ε > 0 and set

X

X = z ∈ C  |z| = R, |z − R| ≥ ε
z

n
fn (z) = R
n
z m
X

Fn (z) = as in (2)
R |z − R| = ε
m=0
|z| = R
gn = n1

c Joel Feldman. 2008. All rights reserved. February 3, 2008 The Dirichlet Test 2
For z ∈ X
z n+1 

 z n+1
 1 − R 1 +


Fn (z) =  ≤
 
R 2R
 1− z 1 ≤
R

R
|R − z| ε
so that the hypotheses of the Dirichlet test are satisfied and the series converges uniformly
z n
P∞
on X. We conclude that n=0 n1 R


converges for |z| < R and for |z| = R, z 6= R and
diverges for |z| > R and for z = R.

c Joel Feldman. 2008. All rights reserved. February 3, 2008 The Dirichlet Test 3