It is well known that for any nonzero integer $k$ the Mordell Equation $x^2 = y^3 + k$ has finitely many solutions $x$ and $y$ in $\mathbf Z$, but it has solutions modulo $n$ for all $n$. One proof of this involves using the Weil Bound to show that $x^2 = y^3 + k$ has solutions modulo $p$ for all $p > 3$, then manually checking for $2, 3$, and then using the lifting lemma. However, the first step doesn't seem very elementary. Does anybody know how to prove that it has solutions modulo $p$ for all primes $p$.


There is an elementary solution using Jacobi sums, which takes advantage of the equation's diagonal nature (each variable appearing separately to a power).

If $p \not\equiv 1 \bmod 3$ (this includes $p = 3$) then every element of $\mathbf Z/(p)$ is a cube in a unique way, so $x^2 \equiv y^3 + k \bmod p$ has $p$ solutions (for each $x$ there is exactly one $y$). If $k \equiv 0 \bmod p$ then the equation becomes $x^2 \equiv y^3 \bmod p$, which also has $p$ solutions. So now let's assume $p \equiv 1 \bmod 3$ and $k \not\equiv 0 \bmod p$.

As in Dylan's answer, let $N_p$ be the number of mod $p$ solutions. Since $p \equiv 1 \bmod 3$ there is a cubic character $\chi$ on $(\mathbf Z/(p))^\times$. In fact there are exactly two cubic characters, $\chi$ and its complex conjugate $\overline{\chi}$. Since $p$ is odd, there is a quadratic character $\psi$ on $(\mathbf Z/(p))^\times$ (it's the Legendre symbol, of course). To count solutions to $x^2 \equiv y^3 + k \bmod p$ we will count solutions $(a,b)$ to the simpler equation $a \equiv b + k \bmod p$ and then count how often $a$ is a square mod $p$ and $b$ is a cube mod $p$. The number of ways $a$ is a square mod $p$ is $1 + \psi(a)$, and the number of ways $b$ is a cube mod $p$ is $1 + \chi(b) + \chi(b)^2 = 1 + \chi(b) + \overline{\chi}(b)$. Therefore $$ N_p = \sum_{(a,b)} (1 + \psi(a))(1 + \chi(b) + \overline{\chi}(b)), $$ where we sum over all $(a,b)$ mod $p$ for which $a \equiv b + k \bmod p$ (either $a$ or $b$ determines the other). Expanding out the product, we get a sum of 6 terms over all the pairs $(a,b)$ where $a \equiv b + k \bmod p$: $$ N_p = \sum_{(a,b)} (1 + \psi(a) + \chi(b) + \psi(a)\chi(b) + \overline{\chi}(b) + \psi(a)\overline{\chi}(b)). $$ Split this up into 6 sums. The first sum is $p$ since the number of possible $(a,b)$ is $p$ (both $a$ and $b$ determine each other and either one is free to take on any value). The second, third, and fourth sums are $0$ since the sum of a nontrivial multiplicative character over $\mathbf Z/(p)$ is $0$. We're left with the sums of $\psi(a)\chi(b)$ and $\psi(a)\overline{\chi}(b)$, and at this point let's write $b$ directly in terms of $a$ (and $k$) so we can write the sums as running over all $a \bmod p$: $$ N_p = p + \sum_{a} \psi(a)\chi(a-k) + \sum_{a}\psi(a)\overline{\chi}(a-k). $$ Since $k \not\equiv 0 \bmod p$, we can make the change of variables $a \mapsto ka$ in both sums and pull out the character values at $k$ and $-k$: $$ N_p = p + \psi(k)\chi(-k)\sum_{a} \psi(a)\chi(1-a) + \psi(k)\overline{\chi}(-k)\sum_{a}\psi(a)\overline{\chi}(1-a). $$ Set $J = \sum_{a} \psi(a)\chi(1-a)$, which is a Jacobi sum, so $$N_p = p + \psi(k)\chi(-k)J + \psi(k)\overline{\chi}(-k)\overline{J} = p + 2{\rm Re}(\psi(k)\chi(-k)J).$$ For any complex number $z$, $|{\rm Re}(z)| \leq \sqrt{|z|}$, so $|N_p - p| \leq 2\sqrt{|J|}$. Since $\psi$ and $\chi$ are nontrivial characters of different order, it is standard that the modulus of the Jacobi sum is $\sqrt{p}$. Therefore $|N_p - p| \leq 2\sqrt{p}$.

For completeness, my answer at https://mathoverflow.net/questions/134352/the-modular-arithmetic-contradiction-trick-for-diophantine-equations?rq=1 discusses how to turn the count on mod $p$ solutions into the existence of a solution mod $p^r$ for all $r$.

The argument I made with quadratic and cubic multiplicative characters mod $p$ carries over to any finite field $\mathbf F_q$ in place of $\mathbf Z/(p)$: for any $k$ in $\mathbf F_q$, the equation $y^2 = x^3 + k$ has a solution $x$ and $y$ in $\mathbf F_q$.


The Hasse-Weil bound for elliptic curves (like $x^2=y^3+k$) tells you that the number of solutions mod $p$, which I'll call $N_p$, satisfies $|p+1 - (N_p+1)|\leq2 \sqrt{p}$.

The smallest $N_p$ can be is $N_p = p - 2\sqrt{p}$, which is definitely $>1$ for $p>5$. For the people who know elliptic curves, I have used $N_p + 1$ because I assume OP doesn't want the point at infinity.

  • $\begingroup$ But OP wanted a proof not using such high-powered results, right? $\endgroup$ – Gerry Myerson Jul 24 '14 at 12:13

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.