A series of secants

Jordan Bell
November 3, 2014

Let ={τ:Imτ>0}. Define C: by

C(τ)=2n=-1eπinτ+q-πinτ=n=-secπnτ,τ.

We take as granted that C is holomorphic on .

First we calculate the Fourier transform of xsechπx.11 1 Elias M. Stein and Rami Shakarchi, Complex Analysis, p. 81, Example 3.

Lemma 1.

For ξR,

-e-2πiξxsechπx𝑑x=sechπξ.
Proof.

Let ξ and define

f(z)=e-2πizξcoshπz.

The poles of f are those z at which coshπz=0, thus z=ni+i2, n. Taking γR to be the contour going from -R to R, from R to R+2i, from R+2i to -R+2i, and from -R+2i to -R, the poles of f inside γR are i2 and 3i2. Because (coshπz)=πsinhπz, we work out

Resz=i/2f(z)=e-2πii2ξπsinhπi2=eπξπisinπ2=eπξπi

and

Resz=3i/2f(z)=e-2πi3i2ξπsinhπ3i2=e3πξπisin3π2=e3πξ-πi.

We bound the integrals on the vertical sides as follows. For z=-R+iy,

|coshπz|=|eπz+e-πz|2||eπz|-|e-πz||2=|e-Rπ-eRπ|2=eRπ-e-Rπ2,

and, for 0y2,

|e-2πizξ|=e2πyξ=e4πξ.

For z=R+iy,

|coshπz|=|eπz+e-πz|2||eπz|-|e-πz||2=|eRπ-e-Rπ|2=eRπ-e-Rπ2,

and, for 0y2,

|e-2πizξ|=e2πyξ=e4πξ.

Therefore

|-R-R+2if(z)𝑑z|-R-R+2i|f(z)|𝑑z2e4πξ2eRπ-e-Rπ=e4πξeRπ-e-Rπ

and likewise

|RR+2if(z)𝑑z|e4πξeRπ-e-Rπ.

As R, each of these tends to 0. Therefore,

-f(z)𝑑z++2i-+2if(z)𝑑z=2πi(eπξπi+e3πξ-πi)=-2e2πξ(eπξ-e-πξ),

i.e.,

-f(z)𝑑z=-+2i+2if(z)𝑑z-2e2πξ(eπξ-e-πξ).

For the top horizontal side,

-R+2iR+2if(z)𝑑z =-RRe-2πi(x+2i)ξcosh(πx+2πi)𝑑x
=-RRe-2πixξe4πξcosh(πx)cosh(2πi)+sinh(πx)sinh(2πi)𝑑x
=e4πξ-RRe-2πixξcoshπx𝑑x
=e4πξ-RRf(x)𝑑x.

Writing

I=-f(z)𝑑z,

this gives us

I=e4πξI-2e2πξ(eπξ-e-πξ),

and so

I=-2e2πξeπξ-e-πξ1-e4πξ=2eπξ-e-πξe2πξ-e-2πξ=2eπξ-e-πξ(eπξ-e-πξ)(eπξ+e-πξ)=sechπξ,

which is what we wanted to show. ∎

Corollary 2.

For t>0 and aR,

-e-2πiξxe-2πiaxsechπxt𝑑x=tsech(π(ξ+a)t),ξ.
Proof.
-e-2πixξe-2πiaxsechπxt𝑑x =-e-2πi(ξ+a)xsechπxt𝑑x
=t-e-2πi(ξ+a)txsechπx𝑑x
=tsech(π(ξ+a)t).

Theorem 3.

For all τH,

C(τ)=iτC(-1τ).
Proof.

For fL1(), we define f^: by

f^(ξ)=e-2πiξxf(x)𝑑x,ξ.

Following Stein and Shakarchi, for a>0, define 𝔉a to be the set of those functions f defined on some neighborhood of in such that f is holomorphic on the set {z:|Imz|<a} and for which there is some A>0 such that

|f(x+iy)|A1+x2,x,|y|<a,

and we set 𝔉=a>0𝔉a. The Poisson summation formula22 2 Elias M. Stein and Rami Shakarchi, Complex Analysis, p. 118, Theorem 2.4. states that for f𝔉,

nf(n)=nf^(n).

For z=x+iy with |y|<12,

|sechπzt| =2|eπ(x+iy)-e-π(x+iy)|
2||eπ(x+iy)|-|e-π(x+iy)||
=2|eπx-e-πx|
=sechπ|x|.

Let t>0. Because the zeros of coshπz are ni+i2, n, the function f(z)=sechπzt belongs to 𝔉t2. Corollary 2 with a=0 gives us

f^(ξ)=tsechπξt,

so applying the Poisson summation formula we get

nsechπnt=tnsechπnt,

or,

nsecπint=tnsecπint,

i.e.,

C(it)=tC(it).

For τ=it this reads

C(τ)=iτC(-1τ).

But τC(τ) and τiτC(-1τ) are holomorphic on , so by analytic continuation this identity is true for all τ. ∎

Theorem 4.
C(1-1τ)4τieπiτ2,Imτ+.
Proof.

Let t>0 and define f(z)=e-πizsechπzt, which we check belongs to 𝔉t2. Corollary 2 with a=12 tells us that for t>0,

f^(ξ)=-e-2πiξxe-πixsechπxt𝑑x=tsech(π(ξ+12)t),ξ.

Thus the Poisson summation formula gives, as (-1)n=e-iπn,

n(-1)nsechπnt=tnsech(π(n+12)t),

or

n(-1)nsecπint=tnsec(πi(n+12)t).

For τ=it this reads

n(-1)nsecπnτ=τinsec(π(n+12)τ).

Now,

sec(πn(1-1τ))=1cosπncos-πnτ-sinπnsin-πnτ=(-1)nsecπnτ,

so the above states that for τ=it, t>0,

C(1-1τ)=τinsec(π(n+12)τ). (1)

We assert that both sides of (1) are holomorphic on , and thus by analytic continuation that (1) is true for all τ.

Write τ=σ+it. For ν>0,

secπντ=2eiπντ+e-iπντ=2e-iπντ(e2πiντ+1)=2eiπντ(1+O(|e2πiντ|)),

or,

secπντ=2eiπντ+O(|e3πiντ|).

Now,

|e3πiτ2|=e-3πt2,

so,

secπντ=2eiπντ+O(e-3πt2).

For ν<0,

secπντ=sec(-πντ)=2e-iπντ+O(e-3πt2).

For ν=12,

secπντ=2eiπτ2+O(e-3πt2),

and for ν=-12,

secπντ=2eiπτ2+O(e-3πt2).

It follows that

nsec(π(n+12)τ)=2eiπτ2+2eiπτ2+O(e-3πt2)=4eiπτ2+O(e-3πt2).

Using this with (1) yields

C(1-1τ)=4τieiπτ2+O(|τ|e-3πt2),τ=σ+it,

proving the claim. ∎

Define θ: by

θ(τ)=neπin2τ,τ.

By proving that Cθ2 is a modular form of weight 0, it follows that it is constant, and one thus finds that C=θ2.33 3 Elias M. Stein and Rami Shakarchi, Complex Analysis, p. 304. One reason that θ is significant is that, for q=eiπτ,

θ(τ)2 =(n1qn12)(n2qn22)
=(n1,n2)×qn12+n22
=n=0r2(n)qn,

where r2(n) denotes the number of ways that n can be expressed as a sum of two squares. We can write C(τ) as

C(τ) =2n=-1qn+q-n
=2n=-qn1+q2n
=1+4n=1qn1+q2n
=1+4n=1qn1-q2n1-q4n
=1+4n=1(qn1-q4n-q3n1-q4n).

Therefore the identity θ(τ)2=C(τ) can be written as

n=0r2(n)qn=1+4n=1(qn1-q4n-q3n1-q4n).

We write

n=1qn1-q4n=n=1qnm=0(q4n)m=n=1m=0qn(4m+1)=k=1a(k)qk,

where a(k) denotes the number of divisors of k of the form 4m+1, and

n=1q3n1-q4n=n=1q3nm=0(q4n)m=n=1m=0qn(4m+3)=k=1b(k)qk,

where b(k) denotes the number of divisors of k of the form 4m+3. Thus for n1,

r2(n)=4(a(n)-b(n)).