Published March 2, 2023|4 minute read
This post motivates the complete zeta function ζA(z)=π−z/2Γ(z/2)ζ(z), explaining how it arises in a natural way by including contributions from all possible completions of Q. The argument is from Andreas Knauf’s excellent lecture notes; I just fill in some details.
By Ostrowski's theorem, the only possible absolute values on Q are the usual absolute value ∣⋅∣=∣⋅∣∞ and the p-adic absolute value ∣⋅∣p. The completion of Q with respect to these absolute values give the real numbers R=Q∞ and the p-adic rationals Qp, respectively.
One performs Fourier analysis over R by executing a change of basis to the eigenfunctions χu,∞:=exp(2πiux) of the translation group R/Z. Here the frequency u takes values in Q∞. An analogous basis exists for the p-adics; we may take χu,p:=exp(2πi[ux]p), where u∈Qp and [ux]p denotes the “fractional part.”
In either case, we may define the Fourier transform
Fvf(u):=∫Qvf(x)χu,v(x)dxv.It is natural to ask about the F-invariant functions. In the case of Q∞, the answer is the familiar Gaussian exp(−πx2). For Qp, a possible solution is the closed ball
γp(x):={1,0,∣x∣p≤1,∣x∣p>1of radius 1. Indeed, we may write out
Fpγp(u)=∫Qp1∣x∣p≤1exp(2πi[ux]p)dxp=∫Zpexp(2πi[ux]p)dxp.Here the measure dxp satisfies
dxp(a+pnZp)=p−nfor integers a and n≥0. If ∣u∣p≤1, then all ux∈Zp, so the integral reduces to
∫Zpdxp=dxp(Zp)=1=γp(u),as claimed. If ∣u∣p=pk>1, then the integral becomes
a=0∑pk∫a+pkZpexp(2πi[ux]p)dxp=a=0∑pkexp(2πi[ua]p)dxp(a+pkZp)=p−ka=0∑pkexp(2πia/pk)=0=γp(u),again as claimed, where the second equality involves reshuffling the values of [ua]p.
Now define the function
ζv(s):=∫Qvγv(x)∣x∣vsdxv∗,where we now use the multiplicatively-invariant normalized Haar measure
dx∞∗:=∣x∣∞dx∞anddxp∗:=p−1p∣x∣pdxp,since the main object of integration ∣x∣ps transforms multiplicatively (i.e. ∣x∣ps∣y∣ps=∣xy∣ps). For the case of Q∞, we straightforwardly obtain the result
∫Re−πx2∣x∣sdx∞∗=2∫0∞e−πx2xs−1dx=π−s/2Γ(s/2).For the case of
Qp, we obtain
∫Qp1∣x∣p≤1∣x∣psdxp∗=p−1p∫Zp∣x∣ps−1dxp=p−1pn=0∑∞∫∣x∣p=p−n∣x∣ps−1dxp=p−1pn=0∑∞(p1−s)ndxp(pnZp−pn+1Zp)=p−1pn=0∑∞(p1−s)npn+1p−1=1−p−s1,where we have evaluated
dxp(pnZp−pn+1Zp)=a=1∑p−1dxp(apn+pn+1Zp)=pn+1p−1.It is then natural to combine all the ζv to obtain an adelic zeta function
ζA(s):=v∏ζv(s)=π−s/2Γ(s/2)ζ(s),which obeys the simple functional equation ζA(s)=ζA(1−s).