A Recurrence Relation

Kangyeon Moon
April 1, 2024

Consider the recurrence relation


a1 = 1

n
1 X an+1−m . (1)
an+1 =1+ n≥1
4 m=1 m3


In this article, we show that an converges to 4/(4 − ζ(3)) using generating function
and Tauberian theory.

1 Generating Function
Since the sum in the recurrence relation is an additive convolution, it would be useful
to consider the power series generating function
∞
X
f (z) = an z n . (2)
n=1

By multiplying z n+1 to the second line of (1) and summing for all n ≥ 1, we obtain
the functional equation
∞ ∞ ∞ n
X
n+1
X
n+1 1 X n+1 X an+1−m
an+1 z = z + z 3
, (3)
n=1 n=1
4 n=1 m=1
m
∞ ∞
z2 1 X zm X
f (z) − z = + ak z k , (4)
1 − z 4 m=1 m3 k=1
z 1
f (z) = + f (z)Li3 (z). (5)
1−z 4

1
Note that we also used the initial condition a1 = 1 while passing to the (4) above.
By solving this for f (z), we get
z
f (z) = .
(1 − z)(1 − Li3 (z)/4)

All the above power series operations are formal, since we ignored convergence issues.
However,
∞
X zn
Li3 (z) =
n=1
n3
converges absolutely for |z| < 1, and |Li3 (z)/4| ≤ ζ(3)/4 < 1, by re-defining f by
the above
P equation, we get a holomorphic function on |z| < 1, and hence its power
n
series an z must converge absolutely there. Thus we can justify all the above
manipulations.

2 Getting Behavior
We consider alternative generating function
∞
X
g(z) = (1 − z)f (z) = z + (an − an−1 )z n . (6)
n=2

Define the sequence bn by

∞
X
g(z) = bn z n . (7)
n=1

Then we easily see that

n
X
b m = an . (8)
m=1
P
Therefore, finding the limit of an is equivalent to find the sum of the series bn ,
and this is intimately related to the behavior of g(z) near z = 1, since the formal
substitution z = 1 shows that
∞
X
g(1) = bn . (9)
n=1

2
 First, let’s investigate the behavior of Li3 (z). If |z| > 1, then the series diverges.
If |z| ≤ 1, the series converges absolutely and uniformly (use Weierstrass M-test).
Hence we see that Li3 (z) is continuous for |z| ≤ 1, with
∞
X 1
lim Li3 (z) = Li3 (1) = 3
= ζ(3). (10)
|z|≤1
n=1
n
z→1

Therefore, we see that

z
lim g(z) = lim (11)
|z|<1 |z|<1 1 − Li3 (z)/4
z→1 z→1
1
= (12)
1 − ζ(3)/4
4
= . (13)
4 − ζ(3)

Thus one may naively guess that g(1) = 4/(4 − ζ(3)).

3 Tauberian Theorem and the Conclusion

P
Historical Introduction. In 1826, Niels Henrik Abel has proved that a series bn
converges to b, then
∞
X
lim− bn rn → b. (14)
r→1
n=1
P
If the above convergence holds, we say that the series bn is ‘Abel summable’ to
b. Because of this history, we can a theorem ‘Abelian’ if it states that a kind of
summability implies the other.
In general, converse of the above assertion is false;
∞
X 1 1
lim− (−1)n rn = lim− = (15)
r→1
n=0
r→1 1+r 2

but
∞
X
(−1)n
n=0

3
is P
divergent. However, in 1897, Alfred Tauber has proved
P a partial converse, that
if bn is Abel summable to b and bn = o(1/n), then bn = b. Following Hardy
and Littlewood, we call a theorem ‘Tauberian’ if it provides a partial converse of
Abelian
P∞ theorem. Now we use a Tauberian theorems to conclude that limn→∞ an =
m=1 bm = g(1) = 4/(4 − ζ(3)).
P
Theorem 1. (Hardy-Littlewood) If Abel summable to b, and if bn ≥ −A/n
bn is P
for some constant A and for all n ∈ N, then bn = b.

For thePproof, refer to Corollary 5.9 of [1]. From the previous section, it is
clear that bn is Abel summable to 4/(4 − ζ(3)). So now we prove the inequality
bn ≥ −A/n by induction. Choose a constant A so that the inequality holds for b1
and b2 , as a base case. For n ≥ 2, suppose that bk ≥ −A/k for k = 1, 2, · · · , n. Then

bn+1 = an+2 − an+1 (16)

n+1 n
!
1 X an+2−m X an+1−m
= 3
− (17)
4 m=1
m m=1
m3
n
AX 1
≥− 3
(18)
4 m=1 m (n + 1 − m)

and it suffices to show that

n
X 1 4
≤ . (19)
m=1
m3 (n + 1 − m) n

Since m(n + 1 − m) ≥ n/2, we have

n n
X 1 2X 1 2ζ(2) 4
3
≤ 2
< < . (20)
m=1
m (n + 1 − m) n m=1 m n n

as ζ(2) = π 2 /6 < 2. This completes the proof.

References
[1] Hugh L. Montgomery & Robert C. Vaughan, 2006, Multiplicative Number Theory
I. Classical Theory, Cambridge Studies in Advanced Mathematics, Cambridge
University Press.