Lagrange's four square theorem states that each $m\in\mathbb N=\{0,1,2,\ldots\}$ can be written as a sum of four squares.

Recently, I found that the diophantine equation $x^2+7y^2=2^n$ has certain surprising connnection with the four-square theorem. Namely, I have the following conjecture.

Conjecture. Let $a,b\in\mathbb N$ with $a\in\{2,4,6,\ldots\}$ or $2\mid b$. Then $m=2^a(2b+1)$ can be written as $x^2+y^2+z^2+w^2$ $(x,y,z,w\in\mathbb N)$ with $x^2+7y^2=2^n$ for some $n\in\mathbb N$.

I have verified this for all $m=1,\ldots,4\times 10^8$. For example, $$4\times3+1=2^2+0^2+0^2+3^2\ \ \text{with}\ 2^2+7\times0^2=2^2,$$ and $$4\times3+1=2^2+2^2+2^2+1^2\ \ \text{with}\ 2^2+7\times2^2=2^5.$$

QUESTION. What nontrivial things can we say about the diophantine equation $x^2+7y^2=2^n$? Why this equation is connected with the four-square theorem in the conjecture manner? Any ideas to solve the above conjecture?

Your comments are welcome!

  • 1
    $\begingroup$ $4\times 3+1=2^a(2b+1)?$ $\endgroup$ – Toni Mhax Oct 15 at 5:48
  • $\begingroup$ I don't understand the relation with sum of four squares, and your comments in the answer are confusing. If you can deterministically express $n$ as sum of 4 squares, this will be new result (probabilistic solution is possible). $\endgroup$ – joro Oct 15 at 7:55
  • 1
    $\begingroup$ @Toni $4\times3+1=2^0(2b+1)$ with $b=6$ even. $\endgroup$ – Zhi-Wei Sun Oct 15 at 9:27
  • 4
    $\begingroup$ This is a typical conjecture of OP's: one that has an obvious probabilistic justification, but with little hope of saying a second sentence beyond that. The values allowed for $x^2+y^2$ form a set with about $(\log n)^2$ elements (note that $x$ and $y$ need not be coprime), and one wants $n-x^2 -y^2$ to be a sum of two squares, which happens with probability about $1/\sqrt{\log n}$. So one fails with probability $(1-c/\sqrt{\log n})^{d (\log n)^2} = \exp(-C (\log n)^{\frac 32})$. Since this is very small, "Borel--Cantelli" would indicate that there are at most finitely many exceptions. $\endgroup$ – Lucia Oct 15 at 16:10
  • 1
    $\begingroup$ In my 2017 JNT paper I proved that any positive integer can be written as $x^2+(2xy)^2 +(xz)^2+w^2$ with $y,z,w\in\mathbb N$ and $x$ a power of two. $\endgroup$ – Zhi-Wei Sun Oct 15 at 18:18

The Diophantine equation $x^2 + 7y^2 = 2^n$ is "trivial" in the sense that its solutions can be described in a very simple way (see boxed formula near the end for solutions in odd numbers when $n \geq 3$), so I am suspicious that there is anything nontrivial connecting it to the four-square theorem.

Let's take $n \geq 1$. Then $n \geq 2$ since $x^2 + 7y^2 = 2$ has no integral solutions. Reducing both sides of $x^2 + 7y^2 = 2^n$ modulo 2, we get $x + y \equiv 0 \bmod 2$, so a necessary condition on a solution is that $x \equiv y \bmod 2$. That congruence condition is nicely compatible with the fact that the ring of integers of the number field $K := \mathbf Q(\sqrt{-7})$ is $$ \mathcal O_K := \mathbf Z\left[\frac{1 + \sqrt{-7}}{2}\right] = \left\{\frac{x + y\sqrt{-7}}{2} : x, y \in \mathbf Z, x \equiv y \bmod 2\right\}. $$ This ring is a UFD since it is a PID: $K$ is an imaginary quadratic field with class number $1$, or you can look at the recent arXiv post by Paul Pollack and Noah Snyder at https://arxiv.org/abs/2010.05033 for a proof that $\mathcal O_K$ is a UFD that doesn't rely on algebraic number theory. A factorization of $2$ into primes in $\mathcal O_K$ is $$ 2 = \frac{1+\sqrt{-7}}{2} \frac{1 - \sqrt{-7}}{2} = \pi\overline{\pi}, $$ where $\pi = (1 + \sqrt{-7})/2$ and $\overline{\pi} = (1 - \sqrt{-7})/2$ are both prime (their norms are both 2).

Since $n \geq 2$, the Diophantine equation $x^2 + 2y^2 = 2^n$ can be rewritten as ${\rm N}((x + y\sqrt{-7})/2) = 2^{n-2}$, where ${\rm N}$ is the norm function on $\mathcal O_K$. A prime element of $\mathcal O_K$ has norm equal to the power of a prime number, so the prime factors of $(x+y\sqrt{-7})/2$ must be prime factors of $2$. The only prime factors of 2 in $\mathcal O_K$ up to sign (the units of $\mathcal O_K$ are $\pm 1$) are $\pi$ and $\overline{\pi}$, so $$ {\rm N}\left(\frac{x + y\sqrt{-7}}{2}\right) = 2^{n-2} \Longleftrightarrow \frac{x+y\sqrt{-7}}{2} = \pm \pi^k\overline{\pi}^\ell $$ where $k$ and $\ell$ are nonnegative integers such that $k + \ell = n-2$.

If $k$ and $\ell$ are both positive then $\pi^k\overline{\pi}^\ell$ is divisible in $\mathcal O_K$ by $\pi\overline{\pi} = 2$, which makes $x$ and $y$ both even: $\pi^{k-1}\overline{\pi}^{\ell - 1} = (a + b\sqrt{-7})/2$ for some integers $a$ and $b$, so $$ \frac{x+y\sqrt{-7}}{2} = \pm 2\frac{a+b\sqrt{-7}}{2} \Longrightarrow x = \pm 2a, y = \pm 2b. $$ Solutions of $x^2 +7y^2 = 2^n$ with even $x$ and $y$ and $n \geq 2$ can be divided through by $4$ in every term of the equation, so the "interesting" integral solutions of $x^2 + 7y^2 = 2^n$ are the ones where $x$ or $y$ is odd, which means in fact that both $x$ and $y$ are odd, since $x \equiv y \bmod 2$. That forces $k = 0$ or $\ell = 0$: $$ x^2 + 7y^2 = 2^n \ {\sf with } \ x, y \ {\sf odd} \Longleftrightarrow \frac{x+y\sqrt{-7}}{2} = \pm \pi^{n-2} \ {\sf or } \ \pm \overline{\pi}^{n-2}. $$ Changing signs on $x$ and $y$ is an easy adjustment, so for $n \geq 2$ the only integral solution of $x^2 + 7y^2 = 2^n$ in odd $x$ and $y$, up to sign, is the coefficients in
$$ \boxed{\frac{x+y\sqrt{-7}}{2} = \pi^{n-2} = \left(\frac{1+\sqrt{-7}}{2}\right)^{n-2}.} $$

When $n = 2$, $(x + y\sqrt{-7})/2 = \pi^0 = 1$, so $x = 2$ and $y = 0$, which are not odd (this is just saying $x^2 + 7y^2 = 4$ only when $(x,y) = (\pm 2,0)$). For $n \geq 3$, $x$ and $y$ are odd (a proof is below). For example, the only solution of $x^2 + 7y^2 = 2^9$ in odd numbers comes from coefficients of $$ \frac{x+y\sqrt{-7}}{2} = \pi^7 = \frac{-13 + 7\sqrt{-7}}{2}, $$ so $(x,y) = (13,7)$ up to sign. The unique solution of $x^2 + 7y^2 = 2^n$ in positive odd numbers for $n = 3, 4, 5, 6, 7, 8$ are $(1,1)$, $(3,1)$, $(5,1)$, $(1,3)$, $(11,1)$, and $(9,5)$.

Let's show the coefficients $x$ and $y$ in the boxed formula above are odd when $n \geq 3$: setting $((1+\sqrt{-7})/2)^m = (a_m + b_m\sqrt{-7})/2$, when $m \geq 1$ the integers $a_m$ and $b_m$ are odd. Since $a_m^2 + 7b_m^2 = 2^{m+2} \equiv 0 \bmod 2$, we have $a_m \equiv b_m \bmod 2$, so it suffices to show $a_m$ is odd. By considering the conjugate formula $((1-\sqrt{-7})/2)^m = (a_m - b_m\sqrt{-7})/2$, $$ a_m = \left(\frac{1+\sqrt{-7}}{2}\right)^m + \left(\frac{1-\sqrt{-7}}{2}\right)^m. $$ To check how divisible the integer $a_m$ is by $2$, view the formula for it in the $2$-adic integers, which contains two square roots of $-7$: $$ 1 + 2 + 2^3 + 2^6 + \ldots, \ \ \ 1 + 2^2 + 2^4 + 2^5 \cdots $$ Embed $\mathbf Q(\sqrt{-7})$ into $\mathbf Q_2$ by letting $\sqrt{-7}$ be the first $2$-adic expansion above. Then the other expansion is $-\sqrt{-7}$, so in $\mathbf Z_2$ $$ \frac{1 + \sqrt{-7}}{2} = 2 + 2^2 + 2^5 + \cdots, \ \ \ \frac{1 - \sqrt{-7}}{2} = 1 + 2 + 2^3 + 2^4 + \cdots, \ \ \ $$ Thus $(1+\sqrt{-7})/2 \equiv 0 \bmod 2\mathbf Z_2$ and $(1 - \sqrt{-7})/2 \equiv 1 \bmod 2\mathbf Z_2$, so for $m \geq 1$ we have $((1+\sqrt{-7})/2)^m \equiv 0 \bmod 2\mathbf Z_2$ (not if $m$ is zero!) and $((1 - \sqrt{-7})/2)^m \equiv 1 \bmod 2\mathbf Z_2$. Therefore $a_m \equiv 0 + 1 \equiv 1 \bmod 2\mathbf Z_2$ for $m \geq 1$, so $a_m$ is odd.

| cite | improve this answer | |
  • $\begingroup$ Thanks. But this does not imply the conjecture. $\endgroup$ – Zhi-Wei Sun Oct 15 at 4:16
  • 14
    $\begingroup$ So do you want me to delete my answer because of that? I am just showing that solutions in odd numbers to $x^2 + 7y^2 = 2^n$ (for $n \geq 3$) can be very easily described. Solutions in even numbers (for varying $n$) must come from multiplying an odd solution by powers of 2, so there is nothing deep about the integral solutions to this Diophantine equation. $\endgroup$ – KConrad Oct 15 at 4:18
  • 5
    $\begingroup$ @Zhi-WeiSun: this looks to me like a pretty exhaustive answer to your actual question "What nontrivial things can we say about the diophantine equation $x^2+7y^2=2^n$. $\endgroup$ – Alex B. Oct 15 at 11:05
  • 6
    $\begingroup$ @Alex My feeling is that the OP uses this site as a place to advertise his "conjectures", many of which are just experimental results without much thought. See all his other posts. $\endgroup$ – WhatsUp Oct 15 at 23:27
  • 1
    $\begingroup$ @WhatsUp yes, and on NMBRTHRY for years and years and years listserv.nodak.edu/cgi-bin/wa.exe?A0=NMBRTHRY archives $\endgroup$ – Will Jagy Oct 16 at 14:37

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.