Ramanujan’s sum

Jordan Bell
April 7, 2014

1 Definition

Let q and l be positive integers. Define

cq⁢(l)=∑gcd⁡(h,q)=11≤j≤qe-2⁢π⁢i⁢h⁢l/q=∑gcd⁡(h,q)=11≤h≤qe2⁢π⁢i⁢h⁢l/q=∑gcd⁡(h,q)=11≤h≤qcos⁡2⁢π⁢h⁢lq.

cq⁢(l) is called Ramanujan’s sum.

2 Fourier transform on ℤ/q and the principal Dirichlet character modulo q

For F:ℤ/q→ℂ, the Fourier transform F^:ℤ/q→ℂ of F is defined by

F^⁢(k)=1q⁢∑j∈ℤ/qF⁢(j)⁢e-2⁢π⁢i⁢j⁢k/q,k∈ℤ/q.

Define χ:ℤ/q→ℂ by χ⁢(j)=1 if gcd⁡(j,q)=1 and χ⁢(j)=0 if gcd⁡(j,q)>1. χ is called the principal Dirichlet character modulo q. The Fourier transform of χ is

χ^⁢(k)=1q⁢∑j∈ℤ/qχ⁢(j)⁢e-2⁢π⁢i⁢j⁢k/q=1q⁢∑gcd⁡(j,q)=11≤j≤qe-2⁢π⁢i⁢j⁢k/q.

Therefore we can write Ramanujan’s sum cq⁢(l) as cq⁢(l)=q⋅χ^⁢(l), thus cq=q⋅χ^.

The above gives us an expression for cq⁢(l) as a multiple of the Fourier transform of the principal Dirichlet character modulo q. cq:ℤ/q→ℂ, and we can write the Fourier transform of cq as

cq^⁢(k) = 1q⁢∑j∈ℤ/qcq⁢(j)⁢e-2⁢π⁢i⁢j⁢k/q
= ∑j∈ℤ/qχ^⁢(j)⁢e-2⁢π⁢i⁢j⁢k/q.

3 Dirichlet series

Here I am following Titchmarsh in §1.5 of his The theory of the Riemann zeta-function, second ed. Let μ be the Möbius function. The Möbius inversion formula states that if

g⁢(q)=∑d|qf⁢(d)

then

f⁢(q)=∑d|qμ⁢(qd)⁢g⁢(d).

(∑d|q is a sum over the positive divisors of q.)

Define

ηq⁢(k)=∑j∈ℤ/qe-2⁢π⁢i⁢j⁢k/q.

We have (this is not supposed to be obvious)

ηq⁢(k)=∑d|qcd⁢(k).

Therefore by the Möbius inversion formula we have

cq⁢(k)=∑d|qμ⁢(qd)⁢ηd⁢(k).

(Hence |cq⁢(k)|≤∑d|kd=σ1⁢(k), where σa⁢(k)=∑d|kda.)

If q|k then ηq⁢(k)=q, and if q|̸k then ηq⁢(k)=0. (To show the second statement: multiply the sum by e-2⁢π⁢i⁢k/q, and check that this product is equal to the original sum. Since we multplied the sum by a number that is not 1, the sum must be equal to 0.) Thus we can express the Möbius function using Ramanujan’s sum as μ⁢(q)=cq⁢(1).

Because ηd⁢(k)=d if k|d and ηd⁢(k)=0 if k|̸d, we have

cq⁢(k)=∑d|q,d|kμ⁢(qd)⁢d=∑d⁢r=q,d|kμ⁢(r)⁢d.

So

cq⁢(k)qs=∑d⁢r=q,d|k1qs⁢μ⁢(r)⁢d=∑d⁢r=q,d|k1ds⁢rs⁢μ⁢(r)⁢d=∑d⁢r=q,d|k1rs⁢μ⁢(r)⁢d1-s.

Therefore

∑q=1∞cq⁢(k)qs=∑q=1∞∑d⁢r=q,d|k1rs⁢μ⁢(r)⁢d1-s=∑r=1∞∑d|k1rs⁢μ⁢(r)⁢d1-s=∑r=1∞1rs⁢μ⁢(r)⁢∑d|kd1-s.

Then

∑q=1∞cq⁢(k)qs=σ1-s⁢(k)⁢∑r=1∞1rs⁢μ⁢(r)=σ1-s⁢(k)⁢1ζ⁢(s);

here we used that

1ζ⁢(s)=∑n=1∞μ⁢(n)ns.

On the other hand, if rather than sum over q we sum over k, then we obtain

∑k=1∞cq⁢(k)ks = ∑k=1∞1ks⁢∑d|q,d|kμ⁢(qd)⁢d
= ∑d|q∑m=1∞1(m⁢d)s⋅μ⁢(qd)⁢d
= ∑d|q∑m=1∞1ms⋅1ds⋅μ⁢(qd)⁢d
= ∑m=1∞1ms⁢∑d|q1ds⁢μ⁢(qd)⁢d
= ζ⁢(s)⋅∑d|qμ⁢(qd)⁢d1-s.