Lagrange’s Four-Square Theorem — a special case of the Fermat-Cauchy Polygonal Number Theorem (FCPNT) — states that every natural number can be written as the sum of the squares of at most four integers.

I’m trying to find a new proof of a conditional version of part of that theorem:

Theorem: If $n$ is a natural number such that $2n+1$ can be written as the sum of the squares of four integers, then $2n+1$ can be written as the sum of squares of four integers which sum to $1$. In other words, one can always find integers $t,u,v,w$ such that \begin{align} 2n+1 &= t^2+u^2+v^2+w^2, \\ 1 &= t+u+v+w. \end{align}

The non-conditional version of this has been proven many times in the past (cf. Pollock, Cauchy, Bradley, etc.); but as far as I know, it has always been proven by first assuming the Fermat-Gauss theorem [that every natural number is the sum of at most three triangular numbers].

I’m hoping to find an elementary proof — existing or new — of this conditional result that doesn’t rely on any other part of the FCPNT. Any references or hints would be appreciated.

  • 1
    $\begingroup$ How do you know that the proposition is true when it is not proved? Is it a conjecture? Do you have any computational arguments in support of your proposition? $\endgroup$ – SARTHAK GUPTA Jun 20 at 4:37
  • $\begingroup$ @SARTHAKGUPTA: It has been proven, many times. So, yes, I know it’s true. The problem is, it is always proved by assuming the Fermat-Gauss theorem, and I’m trying to prove it “from scratch”. $\endgroup$ – Kieren MacMillan Jun 20 at 13:28
  • $\begingroup$ @SARTHAKGUPTA: See, for example, the last page of Bradley’s article jstor.org/stable/3619945 $\endgroup$ – Kieren MacMillan Jun 20 at 13:48

The problem is equivalent to determine whether the Fourier coefficient of $$ \Theta(q)=\sum_{ \begin{array}{cc} n_1,n_2,n_3,n_4\in\textbf{Z}\\ n_1+n_2+n_3+n_4=1 \end{array} }q^{n_1^2+n_2^2+n_3^2+n_4^2} $$ is always greater or equal to 1.

We can write equivalent $n_4=1-n_1-n_2-n_3$. Hence $$ \sum^{4}_{j=1}n_j^2= $$ $$ =1+2n_1+n_2^2+2n_2+2n_1n_2+2n_2^2+2n_3+2n_1n_3+2n_2n_3+2n_3^2= $$ $$ =(n_1+n_2+n_3)^2+2(n_1+n_2+n_3)+n_1^2+n_2^2+n_3^2+1 $$ Hence $$ \Theta(q)=\sum_{ \begin{array}{cc} n_1,n_2,n_3,t\in\textbf{Z}\\ n_1+n_2+n_3=t \end{array} }q^{(t+1)^2+n_1^2+n_2^2+n_3^2}= $$ $$ =\sum^{\infty}_{t=-\infty}q^{(t+1)^2}\sum_{ \begin{array}{cc} n_1,n_2,n_3\in\textbf{Z}\\ n_1+n_2+n_3=t \end{array} }q^{n_1^2+n_2^2+n_3^2}= $$ $$ =\sum^{\infty}_{t=-\infty}q^{(t+1)^2}\sum^{\infty}_{s=0}q^s\sum_{ \begin{array}{cc} n_1,n_2,n_3\in\textbf{Z}\\ n_1+n_2+n_3=t\\ n_1^2+n_2^2+n_3^2=s\\ \end{array} }1. $$ But the equations $$ n_1+n_2+n_3=t\textrm{, }n_1^2+n_2^2+n_3^2=s $$ have solutions $$ n_1=\frac{1}{2}\left(-n_3+t-\sqrt{-2n_3^2+2s+2n_3t-t^2}\right) $$ $$ n_2=\frac{1}{2}\left(-n_3+t+\sqrt{-2n_3^2+2s+2n_3t-t^2}\right) $$ and $$ n_1=\frac{1}{2}\left(-n_3+t+\sqrt{-2n_3^2+2s+2n_3t-t^2}\right) $$ $$ n_2=\frac{1}{2}\left(-n_3+t-\sqrt{-2n_3^2+2s+2n_3t-t^2}\right) $$ Hence if we set $$ A(t,s)=\sum_{ \begin{array}{cc} n_1+n_2+n_3=t\\ n_1^2+n_2^2+n_3^2=s\\ \end{array} }1, $$ then, if the discriminant of $n_1$ and $n_2$: $-2n_3^2+2s+2n_3t-t^2$ is zero possess one double root $(n_1=n_2)$. Otherwise if $-2n_3^2+2s+2n_3t-t^2=m^2$, $m\in\textbf{Z}^{*}$ two different. Hence if $s'=\left[\sqrt{s}\right]$ is the floor of the square root of $s\geq 0$, we get $$ A(t,s)=\sum_{ \begin{array}{cc} -s'\leq n_3\leq s'\\ -3n_3^2+2s+2n_3t-t^2=0 \end{array} }1+ $$ $$ +2\sum_{ \begin{array}{cc} -s'\leq n_3\leq s'\\ -3n_3^2+2s+2n_3t-t^2\neq 0 \end{array} }X_{\textbf{Z}}\left(\frac{1}{2}\left(-n_3+t+\sqrt{-3n_3^2+2s+2n_3t-t^2}\right)\right), $$ where $X_{\textbf{Z}}(n)$ is the characteristic function on integers.

By this way we get if $r_0(t,2s)$ is the number of representations of $2s$ in the form $$ 3x^2-2xt+t^2 $$ and $r_1(t,2s)$ the number of representations of $2s$ in the form $$ 3x^2-2xt+m^2+t^2, $$ then $r(t,2s)=r_0(t,2s)+2r_1(t,2s)=A(t,s)$ is the Fourier coefficient of $$ \Theta(q)=\sum^{\infty}_{t=-\infty}\sum^{\infty}_{s=0}A(t,s)q^{(t+1)^2+s}\textrm{, }|q|<1 $$ If $s-t=p$, then $$ 2s=3x^2-2xt+m^2+t^2\Leftrightarrow 3x^2-2tx+t^2-2t+m^2=2p.\tag 1 $$ We want to show that the above last equation have always integer solutions for every even non negative integer $p\geq0$ and for $p$ odd positive none. This will enable us to conclude, that the odd Fourier coefficients of $\Theta(q)$ are always occur and are non zero iff $p$ is even. For $p$ odd we have no representations. For this we define: $$ \phi(q)=\sum^{\infty}_{n,t=-\infty}q^{3n^2-2tn+t^2-2t} $$ and assume the transformation $n\rightarrow an+bt$, $t\rightarrow c n+d t$, where $a=-1$, $b=1$, $c=1$, $d=-2$. Then $ad-bc=1$ and $$ \phi(q)=2\sum^{\infty}_{n,t=-\infty}q^{n^2-2n+8 t^2+4t}. $$ Hence $$ \phi(q)=2q^{-1}\left(\sum^{\infty}_{n=-\infty}q^{n^2-2n+1}\right)\left(\sum^{\infty}_{t=-\infty}q^{8t^2+4t}\right)= $$ $$ =2q^{-1}\theta_3(q)\sum^{\infty}_{n=-\infty}q^{8n^2+4n}, $$ where $\theta_3(q)=\sum^{\infty}_{n=-\infty}q^{n^2}$, $|q|<1$. Hence (1) have analog:
$$ \phi(q)\theta_3(q)=\sum_{n,k,m\in\textbf{Z}}q^{3n^2-2nk+k^2-k+m^2}=2q^{-1}\theta_3(q)^2\sum^{\infty}_{n=-\infty}q^{8n^2+4n}= $$ $$ =2q^{-1}\theta_3(q)^2\phi_e(q^2), $$ where $\phi_e(q)=\sum^{\infty}_{n=-\infty}q^{4n^2+2n}$. But $$ \theta_2(q)=q^{1/4}\sum^{\infty}_{n=-\infty}q^{n^2+n}=q^{1/4}\sum^{\infty}_{n=-\infty}q^{4n^2+2n}+\sum^{\infty}_{n=-\infty}q^{(2n+1+1/2)^2} $$ and $$ \sum^{\infty}_{n=-\infty}q^{(2n+1+1/2)^2}=\sum^{\infty}_{n=-\infty}q^{(2n+3/2-2)^2}= $$ $$ \sum^{\infty}_{n=-\infty}q^{(2n-1/2)^2}=q^{1/4}\sum^{\infty}_{n=-\infty}q^{4n^2-2n}. $$ Hence $$ \theta_2(q)=2q^{1/4}\sum^{\infty}_{n=-\infty}q^{4n^2+2n} $$ and therefore $$ \phi(q)\theta_3(q)=q^{-3/2}\theta_3(q)^2\theta_2(q^2).\tag 2 $$

Hence is equivalent to show that $$ x^2+y^2+2z^2+2z=2p+1,\tag 3 $$ have always solutions when $p$ is even non negative integer and none when $p$ positive odd. But indeed: $2z^2+2z\equiv 0(4)$ and when $p$ is odd we have no solutions since no integer of the form $4n+3$ have representation as a sum of two squares. Hence we left with case $p$ even.

The function $f(z)=\theta_3(q)^2\theta_2(q^2)$, $q=e(z)=e^{2\pi i z}$, $Im(z)>0$ is the associated theta function of (3) and is a modular form of weight $3/2$ in $\Gamma(8)$, i.e. it holds $$ f\left(\frac{az+b}{cz+d}\right)=\epsilon_{c,d}(cz+d)^{3/2}f(z), $$ when $Im(z)>0$ and $a,b,c,d$ integers with $ad-bc=1$, $a,d\equiv 1(8)$, $c,b\equiv 0(8)$. The function $\epsilon_{c,d}$ is such that $\epsilon_{c,d}^4=1$. But I don't have much experience to go further.


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.