Landsberg-Schaar relation

The Landsberg–Schaar relation is a reciprocity formula for quadratic exponential sums. It transforms a finite sum with quadratic phase qn^2/p into another finite quadratic sum in which the roles of p and q are essentially reversed. The transformation includes the striking phase factor \exp\left(\frac{\pi i}{4}\right) . This factor is not incidental: it is the same analytic phase that appears in the evaluation of quadratic Gauss sums and, ultimately, in quadratic reciprocity.

The relation is as follows.

Theorem. For all positive integers p and q ,

\displaystyle \frac{1}{\sqrt p}\sum_{n=0}^{p-1}\exp\left(\frac{2\pi i qn^2}{p}\right)=\frac{\exp\left(\frac{\pi i}{4}\right)}{\sqrt{2q}}\sum_{m=0}^{2q-1}\exp\left(-\frac{\pi i pm^2}{2q}\right).

No coprimality assumption is required. The formula is already a reciprocity law: a quadratic sum of length p is transformed into a quadratic sum of length 2q .

The proof has one central idea. Rather than evaluating either finite sum directly, we embed the left-hand sum into a Gaussian-damped infinite theta series. This theta series can then be analyzed in two ways. First, its periodic coefficients recover the finite sum on the left-hand side. Second, Poisson summation transforms it into a dual theta series whose periodic coefficients recover the finite sum on the right-hand side. The phase \exp\left(\frac{\pi i}{4}\right) arises from the Fourier transform of a complex Gaussian.

Gaussian Averaging

We begin with a standard consequence of Poisson summation. For h>0 and \alpha\in\mathbb R ,

\displaystyle \sum_{k\in\mathbb Z}e^{-\pi h(k+\alpha)^2}=\frac{1}{\sqrt h}\sum_{\ell\in\mathbb Z}e^{-\pi \ell^2/h}e^{2\pi i\ell\alpha}.

As h\downarrow0 , every term with \ell\ne0 is exponentially small. Hence

\displaystyle \sum_{k\in\mathbb Z}e^{-\pi h(k+\alpha)^2}=\frac{1}{\sqrt h}+o(h^{-1/2}).

In other words, a very broad Gaussian samples many integers, so its discrete sum is asymptotic to its integral. We will apply this observation after grouping integers into residue classes.

Theta Series

For \varepsilon>0 , define the Theta series \displaystyle \Theta_\varepsilon:=\sum_{n\in\mathbb Z}\exp\left(\frac{2\pi i qn^2}{p}\right)e^{-\pi\varepsilon n^2/p}.

The Gaussian factor makes the sum absolutely convergent. As \varepsilon\to0^+ , the damping weakens, and the leading term of \Theta_\varepsilon reveals a finite quadratic sum.

The phase \exp\left(\frac{2\pi i qn^2}{p}\right) is periodic modulo p , since replacing n by n+p changes qn^2/p by an integer. Writing n=r+kp , with 0\le r\le p-1 , gives

\displaystyle \Theta_\varepsilon=\sum_{r=0}^{p-1}\exp\left(\frac{2\pi i qr^2}{p}\right)\sum_{k\in\mathbb Z}e^{-\pi p\varepsilon(k+r/p)^2}.

For each fixed r , the inner sum has the form above, with h=p\varepsilon . Therefore,

\displaystyle \sum_{k\in\mathbb Z}e^{-\pi p\varepsilon(k+r/p)^2}=\frac{1}{\sqrt{p\varepsilon}}+o(\varepsilon^{-1/2}),

uniformly in r . Multiplying by \sqrt\varepsilon , we obtain

\displaystyle \lim_{\varepsilon\downarrow0}\sqrt\varepsilon\Theta_\varepsilon=\frac{1}{\sqrt p}\sum_{r=0}^{p-1}\exp\left(\frac{2\pi i qr^2}{p}\right).

This is the first interpretation of the theta sum: the finite quadratic sum on the left-hand side of the Landsberg–Schaar relation is the coefficient of the dominant \varepsilon^{-1/2} growth.

Poisson Summation

We now analyze the same theta sum by Fourier methods. The definition of \Theta_\varepsilon can be rewritten as

\displaystyle \Theta_\varepsilon=\sum_{n\in\mathbb Z}\exp\left(-\pi\frac{\varepsilon-2iq}{p}n^2\right).

For a complex number A with {\text{Re}}(A)>0 , the Fourier transform of a Gaussian is

\displaystyle \int_{\mathbb R}e^{-\pi A x^2}e^{-2\pi i\xi x}dx=A^{-1/2}e^{-\pi\xi^2/A}.

The square root is chosen by continuous extension from positive real values of A . This branch choice is responsible for the phase \exp\left(\frac{\pi i}{4}\right) .

Applying Poisson summation with \displaystyle A=\frac{\varepsilon-2iq}{p}, we obtain

\displaystyle \Theta_\varepsilon=\sqrt{\frac{p}{\varepsilon-2iq}}\sum_{m\in\mathbb Z}\exp\left(-\pi\frac{pm^2}{\varepsilon-2iq}\right).

This is the decisive step. Poisson summation has replaced the original quadratic phase qn^2/p by a dual quadratic phase.

Since \displaystyle \frac{1}{\varepsilon-2iq}=\frac{\varepsilon}{\varepsilon^2+4q^2}+i\frac{2q}{\varepsilon^2+4q^2}, the m -th term in the transformed sum is

\displaystyle \exp\left(-\pi\frac{p\varepsilon}{\varepsilon^2+4q^2}m^2\right)\exp\left(-\pi i\frac{2pq}{\varepsilon^2+4q^2}m^2\right).

As \varepsilon\downarrow0 ,

\displaystyle \frac{p\varepsilon}{\varepsilon^2+4q^2}=\frac{p\varepsilon}{4q^2}+O(\varepsilon^3),\qquad \frac{2pq}{\varepsilon^2+4q^2}=\frac{p}{2q}+O(\varepsilon^2).

A comparison on the effective range |m|\ll\varepsilon^{-1/2} , together with Gaussian tail estimates, yields

\displaystyle \sum_{m\in\mathbb Z}\exp\left(-\pi\frac{pm^2}{\varepsilon-2iq}\right)=\sum_{m\in\mathbb Z}e^{-\pi p\varepsilon m^2/(4q^2)}\exp\left(-\frac{\pi i pm^2}{2q}\right)+o(\varepsilon^{-1/2}).

The oscillatory factor \displaystyle \exp\left(-\frac{\pi i pm^2}{2q}\right) is periodic modulo 2q . Indeed, replacing m by m+2q changes pm^2/(4q) by an integer. Thus we may group the sum by residue classes modulo 2q . Writing m=r+2qk , we have

\displaystyle e^{-\pi p\varepsilon m^2/(4q^2)}=e^{-\pi p\varepsilon(k+r/(2q))^2}.

Applying the Gaussian averaging principle again gives

\displaystyle \sum_{m\in\mathbb Z}\exp\left(-\pi\frac{pm^2}{\varepsilon-2iq}\right)=\frac{1}{\sqrt{p\varepsilon}}\sum_{r=0}^{2q-1}\exp\left(-\frac{\pi i pr^2}{2q}\right)+o(\varepsilon^{-1/2}).

Finally, \displaystyle \sqrt{\frac{p}{\varepsilon-2iq}}\longrightarrow\exp\left(\frac{\pi i}{4}\right)\sqrt{\frac{p}{2q}}\quad(\varepsilon\downarrow0).

The phase \exp\left(\frac{\pi i}{4}\right) appears because -2iq has argument -\pi/2 , so its inverse square root has argument \pi/4 .

Combining the transformed theta-sum formula with these asymptotics, we obtain

\displaystyle \lim_{\varepsilon\downarrow0}\sqrt\varepsilon\Theta_\varepsilon=\frac{\exp\left(\frac{\pi i}{4}\right)}{\sqrt{2q}}\sum_{r=0}^{2q-1}\exp\left(-\frac{\pi i pr^2}{2q}\right).

The two preceding limits compute the same quantity. Equating them gives

\displaystyle \frac{1}{\sqrt p}\sum_{n=0}^{p-1}\exp\left(\frac{2\pi i qn^2}{p}\right)=\frac{\exp\left(\frac{\pi i}{4}\right)}{\sqrt{2q}}\sum_{m=0}^{2q-1}\exp\left(-\frac{\pi i pm^2}{2q}\right).

This is the Landsberg–Schaar relation.

The proof is important for more than the final identity. It begins with a finite quadratic phase modulo p , but it does not evaluate that phase by finite algebra alone. Instead, it embeds the finite sum into a theta function on the real line. Poisson summation then exchanges the integer lattice with its Fourier-dual lattice.

Under this transformation, the quadratic coefficient \displaystyle \frac{q}{p} is replaced by the dual coefficient \displaystyle -\frac{p}{4q}, while the Fourier transform contributes the phase \exp\left(\frac{\pi i}{4}\right) . Thus the Landsberg–Schaar relation forms a bridge between finite arithmetic and real Fourier analysis. The finite sums arise because the quadratic phases are periodic, while the phase \exp\left(\frac{\pi i}{4}\right) comes from the square root in the Fourier transform of a complex Gaussian.

Quadratic Reciprocity: The Landsberg–Schaar relation is not itself the quadratic reciprocity law, but it has the same underlying structure. Suppose that p and q are distinct odd primes, and define the quadratic Gauss sum

\displaystyle G(a;\ell):=\sum_{x\bmod\ell}\exp\left(\frac{2\pi i ax^2}{\ell}\right),\qquad G_\ell:=G(1;\ell).

For \ell\nmid a , one has the standard change-of-coefficient identity

\displaystyle G(a;\ell)=\left(\frac{a}{\ell}\right)G_\ell,

where \left(\frac{a}{\ell}\right) is the Legendre symbol.

The left-hand side of the Landsberg–Schaar relation therefore becomes

\displaystyle \frac{1}{\sqrt p}G(q;p)=\left(\frac{q}{p}\right)\frac{G_p}{\sqrt p}.

On the right, splitting the sum over m\bmod 2q into even and odd residue classes gives

\displaystyle \sum_{m\bmod 2q}\exp\left(-\frac{\pi i pm^2}{2q}\right)=\bigl(1-i^{pq}\bigr)G(-p;q).

Using the change-of-coefficient identity once more, \displaystyle G(-p;q)=\left(\frac{-p}{q}\right)G_q.

Thus the Landsberg–Schaar relation compares the two Legendre symbols \left(\frac{q}{p}\right) and \left(\frac{p}{q}\right), with the remaining sign supplied by the Gaussian phase and the congruence classes of p and q modulo 4 . After inserting the classical evaluation of quadratic Gauss sums, this comparison becomes

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

This is quadratic reciprocity. The sign in quadratic reciprocity is therefore not an arbitrary parity correction. It is the finite-arithmetic shadow of the phase \exp\left(\frac{\pi i}{4}\right) that appears in the Fourier transform of a complex Gaussian. In this sense, the Landsberg–Schaar relation is a Fourier-analytic reciprocity law from which the classical quadratic reciprocity law naturally emerges.

Leave a comment