Estermann’s evaluation of Gauss Sum

We show Esterman’s evaluation of the Gauss sum

\displaystyle g_p=\sum_{n=0}^{p-1} e^{2 \pi in^{2} / p}=\sum_{m=0}^{p-1}\left( \frac{m}{p}\right)e^{2 \pi i m / p}.

It’s easy to see by squaring that

\displaystyle g_p^2 = \left( \frac{-1}{p}\right) p

which implies that

\displaystyle g_{p}=\pm \sqrt{p} \text { if } p \equiv 1~(\bmod 4) \quad \text { and } g_{p}=\pm i \sqrt{p} \text { if } p \equiv 3 ~(\bmod 4)

What about the sign?

Both cases together can be written as

\displaystyle \frac{1}{2}(1-i)\left(1+i^{p}\right) g_{p}=\pm \sqrt{p}

To show that the sign is positive we will prove that the real part of the above expression {\frac{1}{2}(1-i)\left(1+i^{p}\right) g_{p}} is at least {-\sqrt p}.

Start with

\displaystyle g_p-1= \sum_{n=1}^{p-1} e^{2 \pi in^{2} / p}. = 2\sum_{n=1}^{(p-1)/2} e^{2 \pi in^{2} / p}.

Multiplying with {1+i^p,} we have

\displaystyle (1+i^p)\sum_{n=1}^{(p-1)/2} e^{2 \pi in^{2} / p}=\sum_{n=1}^{(p-1)/2} e^{2 \pi in^{2} / p}+ \sum_{n=1}^{(p-1)/2} e^{2 \pi in^{2} / p} e^{\frac{\pi i p}{2}}

\displaystyle \implies (1+i^p)\sum_{n=1}^{(p-1)/2} e^{2 \pi in^{2} / p}= \sum_{n=1}^{(p-1)/2} e^{2 \pi in^{2} / p} + \sum_{n=1}^{(p-1)/2} e^{2 \pi i (p/2-n)^{2} / p}

\displaystyle \implies (1+i^p)\sum_{n=1}^{(p-1)/2} e^{2 \pi in^{2} / p} = \sum_{n=1}^{p-1} e^{\pi in^{2} / 2p}.

So we have

\displaystyle \frac{1}{2}(1-i)\left(1+i^{p}\right) g_{p} = (1-i) \sum_{n=1}^{p-1} e^{\pi in^{2} / 2p}.

Now consider the real part of the sum {\displaystyle (1-i)\sum_{n=1}^{p-1} e^{\pi in^{2} / 2p}}

which equals

\displaystyle \sum_{n=1}^{p-1} \left(\cos \frac{\pi n^{2}}{2 p}+\sin \frac{\pi n^{2}}{2 p}\right).

First the contribution of the terms {n=1,2, \cdots [\sqrt p]} is at least {[\sqrt p]} since each of the term is

\displaystyle \left(\cos \frac{\pi n^{2}}{2 p}+\sin \frac{\pi n^{2}}{2 p}\right) \ge 1, \quad \text{for } n \le \sqrt p.

What about the rest of the terms?

We have

\displaystyle \left(a_{n}-a_{n-1}\right) b_{n}=2 i e^{\pi i n ^{2} /(2 p)},

where

\displaystyle a_{n}=e^{\pi i n(n+1) /(2 p)} \text{ and } b_{n}=\csc \left(\frac{\pi n }{2 p}\right).

Therefore the sum

\displaystyle 2i\sum_{[\sqrt p]+1}^{p-1} e^{\pi i n ^{2} /(2 p)} =\sum_{[\sqrt p]+1}^{p-1} (a_n-a_{n-1})b_n
\displaystyle =\sum_{[\sqrt p]+1}^{p-1} (b_n-b_{n-1})a_n - a_{[\sqrt p]}b_{[\sqrt p]+ 1} +a_{p-1}b_p

And we have the bound

\displaystyle \left| 2 \sum_{[\sqrt p]+1}^{p-1} e^{\pi i n ^{2} /(2 p)} \right| \le \sum_{[\sqrt p]+1}^{p-1} (b_n-b_{n-1}) +b_{[\sqrt p]+ 1} + b_p =2b_{[\sqrt p]+ 1} \le 2 \sqrt p

Hence the real part of \displaystyle (1-i) \sum_{[\sqrt p]+1}^{p-1} e^{\pi i n ^{2} /(2 p)} is bounded by {\sqrt 2 \sqrt p}

Also {\text{Re}\left(\frac{1}{2}(1-i)\left(1+i^{k}\right)\right) } is either {0} or {1}, so we get

\displaystyle \text{Re} \left(\frac{1}{2}(1-i)\left(1+i^{p}\right) g_{p}\right) \ge [\sqrt p] -\sqrt{2} \sqrt p >-\sqrt p .

But we know that

\displaystyle \text{Re} \left(\frac{1}{2}(1-i)\left(1+i^{p}\right) g_{p}\right) =\pm \sqrt p.

Hence we are done!

Posted in $.

Leave a comment