\documentclass[12pt,reqno]{article}
\usepackage[usenames]{color}
\usepackage{amssymb}
\usepackage{graphicx}
\usepackage{amscd}
\usepackage{epsfig}
\usepackage[colorlinks=true,
linkcolor=webgreen,
filecolor=webbrown,
citecolor=webgreen]{hyperref}
\definecolor{webgreen}{rgb}{0,.5,0}
\definecolor{webbrown}{rgb}{.6,0,0}
\usepackage{color}
\usepackage{fullpage}
\usepackage{float}
\usepackage{psfig}
\usepackage{graphics,amsmath,amssymb}
\usepackage{amsthm}
\usepackage{amsfonts}
\usepackage{latexsym}
\usepackage{epsf}
\setlength{\textwidth}{6.5in}
\setlength{\oddsidemargin}{.1in}
\setlength{\evensidemargin}{.1in}
\setlength{\topmargin}{-.5in}
\setlength{\textheight}{8.9in}
\newtheorem{theorem}{Theorem}
\newtheorem{corollary}[theorem]{Corollary}
\newtheorem{lemma}[theorem]{Lemma}
\newtheorem{proposition}[theorem]{Proposition}
\newtheorem{conjecture}[theorem]{Conjecture}
\newtheorem{problem}[theorem]{Problem}
\newtheorem{definition}[theorem]{Definition}
\newcommand{\seqnum}[1]{\href{http://www.research.att.com/cgi-bin/access.cgi/as/~njas/sequences/eisA.cgi?Anum=#1}{\underline{#1}}}
\begin{document}
\begin{center}
\epsfxsize=4in
\leavevmode\epsffile{logo129.eps}
\end{center}
\begin{center}
\vskip 1cm{\LARGE\bf
A Parametrization of Equilateral Triangles \\
\vskip .1in Having Integer Coordinates}
\vskip 1cm
\large
Eugen J. Ionascu\\
Department of Mathematics\\
Columbus State University\\
Columbus, GA 31907 \\
USA\\
and Honorific Member of the \\ Romanian Institute of Mathematics
``Simion Stoilow"\\
\href{mailto:chamberl@math.grinnell.edu}{\tt ionascu\_eugen@colstate.edu}\\
\end{center}
\vskip .2in
\begin{abstract}
We study the existence of equilateral triangles of given side
lengths and with integer coordinates in dimension three. We show
that such a triangle exists if and only if their side lengths are of
the form $\sqrt{2(m^2-mn+n^2)}$ for some integers $m,n$. We also
show a similar characterization for the sides of a regular
tetrahedron in $\mathbb Z^3$: such a tetrahedron exists if and only
if the sides are of the form $k\sqrt{2}$, for some $k\in\mathbb N$.
The classification of all the equilateral triangles in $\mathbb Z^3$
contained in a given plane is studied and the beginning analysis for
small side lengths is included. A more general parametrization is
proven under special assumptions. Some related questions about the
exceptional situation are formulated in the end.
\end{abstract}
\section{Introduction}
It is known that there is no equilateral triangle whose vertices
have integer coordinates in the plane. One can easily see this by
calculating the area of such a triangle of side length $l$ using the
formula $\text{Area}=\frac{l^2\sqrt{3}}{4}$ and by using Pick's
theorem for the area of a polygon with vertices of integer
coordinates: $\text{Area}=\frac{\sharp b}{2}+\sharp i-1$ where
$\sharp b$ is the number of points of integer coordinates on the
boundary of the polygon and $\sharp i$ is the number of such points
in the interior of the polygon. Since Pick's theorem implies that
this area is a rational number of square units, the formula
$\text{Area}=\frac{l^2\sqrt{3}}{4}$ says that this is a rational
multiple of $\sqrt{3}$ since $l^2$ must be a positive integer by the
Pythagorean theorem. This contradiction implies that no such
triangle exists.
The analog of this fact in three dimensions is not true since one
can form a regular tetrahedron by taking as vertices the points
$O(0,0,0)$, $A(1,1,0)$, $B(1,0,1)$ and $C(0,1,1)$. It turns out that
the sides of such regular tetrahedra have to be of the form
$k\sqrt{2}$, $k\in \mathbb N$. Moreover and as a curiosity, one can
use the facts derived in this note to show that there are only three
regular tetrahedra in $\mathbb Z^3$ having the origin as one of
their vertices and of side lengths $9\sqrt{2}$, where the counting
has been done up to symmetries of the cube. In the figure below,
that we generated with Maple, we show three regular tetrahedra that
together with all their cube symmetries fill out the class just
described.
\begin{center}
\epsfig{file=fig3.eps,height=3in,width=3in}
\vspace{.1in}\\
{\it Figure 1: Regular tetrahedra of side lengths $9\sqrt{2}$}
\vspace{.1in}
\begin{table}
\centerline{
\begin{tabular}{|c|c|c|c|}
\hline \hline
$O(0,0,0)$& $A_1(9,9,0)$ & $B_1(9,0,9)$& $C_1(0,9,9)$ \\
\hline
$O(0,0,0)$& $A_2(-9,9,0)$& $B_2(-4,5,-11)$& $C_2(3,12,-3)$ \\
\hline
$O(0,0,0)$& $A_3(12,3,-3)$& $B_3(7,-8,-7)$& $C_3(3,3,-12)$ \\
\hline
\end{tabular}}
\vspace{0.1in} \caption{Coordinates of the three regular tetrahedra
in Figure~1}
\end{table}
\end{center}
\vspace{0.2in}
Along these lines, we mention the following related result of
Schoenberg, \cite{s}, who proved that a regular $n$-simplex exists
in $\mathbb Z^n$ in the following cases and no others:\par
(i) $n$ is even and $n+1$ is a square;\par
(ii) $n\equiv 3$ (mod 4);\par
(iii) $n\equiv 1$ (mod 4) and $n+1$ is the sum of two squares.
\par \noindent In conjunction to the results of this note, Schoenberg's characterization opens more questions such as:
what are the corresponding parameterizations in all these cases
when regular $n$-simplexes do exist? Or, what is the similar
characterization for the regular $n$-simplex side lengths?
Equilateral triangles with vertices of integer coordinates in the
three-dimensional space are numerous as one could imagine from the
situation just described. One less obvious example is the triangle
$CDO$ with $C(31,19,76)$ and $D(44,71,11)$ having side lengths equal
to $13\sqrt{42}$. Generating all such triangles is a natural problem
and we may start with one such triangle and then apply the group of
affine transformations
$T_{\alpha,O,\overset{\rightarrow}{y}}(\overset{\rightarrow}{x})=
\alpha O(\overset{\rightarrow}{x})+\overset{\rightarrow}{y}$ where
$O$ is an orthogonal matrix with rational coefficients, $\alpha \in
\mathbb Z$, and $\overset{\rightarrow}{y}$ a vector in $\mathbb
Z^3$. Naturally, we may obtain triangles that have certain length
sides but not all desired triangles can be obtained this way.
Indeed, such a transformation multiplies the side lengths with the
factor $\alpha$ and so for instance the triangle $OAB$ with side
lengths $\sqrt{2}$ cannot be transformed this way into the triangle
$CDO$. We are interested in parametrizations that encompasses all
equilateral triangles in $\mathbb Z^3$ that are contained in the
same plane. This requires further restrictions on the type of
transformations $T_{\alpha,O,\overset{\rightarrow}{y}}$.
In fact, in the next section we show that the side lengths which
appear from such equilateral triangles are of the form
$\sqrt{2N(\zeta)}$ where $\zeta$ is an Eisenstein-Jacobi integer and
$N(\zeta)$ is its norm. The Eisenstein-Jacobi integers are defined
as $\mathbb Z[\omega]$ where $\omega$ is a primitive cubic root of
unity, i.e., the complex numbers of the form $\zeta=m-n\omega$ with
$m,n\in \mathbb Z$ with their norm given by $N(\zeta)=m^2-mn+n^2$.
What makes the existence of such triangles work in space that does
not work in two dimensions? We show in Proposition~\ref{planeprop}
that the plane containing such a triangle must have a normal vector
$\overset{\rightarrow}{n}=a\overset{\rightarrow}{i}+b\overset{\rightarrow}{j}+c\overset{\rightarrow}{k}$
where $a,b,c$ are integers that satisfy the Diophantine equation
\begin{equation}\label{eq1}
a^2+b^2+c^2=3d^2, \ d\in \mathbb Z . \end{equation}
This equation has no non-trivial solutions when $abc=0$, according
to Gauss' characterization for the numbers that can be written as
sums of two perfect squares.
Our study of the existence of such triangles started with an
American Mathematics Competition problem in the beginning of 2005.
The problem was stated as follows:
\begin{problem}\label{originalpr} Determine the number of equilateral triangles whose
vertices have coordinates in the set $\{0,1,2\}$.
\end{problem}
It turns out that the answer to this question is 80. Let us
introduce the notation ${\cal E}{\cal T}(n)$ for the number of
equilateral triangles whose vertices have coordinates in the set
$\{0,1,2,\ldots,n\}$ for $n\in \mathbb N$. Some of the values of
${\cal E}{\cal T}(n)$ are tabulated in Table~2. \par
\vspace{0.1in}
\begin{table}
\centerline{
\begin{tabular}{|c|c|c|c|c|c|c|c|c|c|c|}
\hline
$n$ & $1$ & $2$ & $3$& $4$& $5$& $6$& $7$& $8$ & $9$ & $10$ \\ \hline
${\cal E}{\cal T}(n)$& $8$ & $80$ & $368$& $1264$ & $3448$& $7792$& $16176$& $30696$ & $54216$& $90104$ \\
\hline
\end{tabular}}
\vspace{0.1in} \caption{Sequence A 102698}
\end{table}
\vspace{0.1in} This sequence was entered in the on-line
Encyclopedia of Integer Sequences \cite{OL} by Joshua Zucker on
February 4th, 2005. The first $34$ terms in this sequence were
calculated by Hugo Pfoertner using a program in Fortran (private
communication). Our hope is that the results obtained here may be
used in designing a program that could calculate ${\cal E}{\cal
T}(n)$ for significantly more values of $n$.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\section{Planes containing equilateral triangles in $\mathbb Z^3$}%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
Let us denote the side lengths of an equilateral triangle $\triangle
OPQ$ by $l$. We are going to discard translations, so we may assume
that one of the vertices of such a triangle is $O(0,0,0)$. If the
other two points, $P$ and $Q$, have coordinates $(x,y,z)$ and
$(u,v,w)$ respectively, then as we have seen before the area of
$\triangle OPQ$ is given by
\begin{equation}\label{area}
\text{Area}=\frac{l^2\sqrt{3}}{4}=\frac{1}{2}|\overset{\rightarrow}{OP}
\times \overset{\rightarrow}{OQ}|=\left|
\begin{array}{ccc}
\overset{\rightarrow}{i} & \overset{\rightarrow}{j} & \overset{\rightarrow}{k} \\
x & y & z \\
u & v & w \\
\end{array}
\right|
.\end{equation}
This implies the following simple proposition but essential in our
discussion:
\begin{proposition}\label{planeprop}
Assume the triangle $\triangle OPQ$ is equilateral and its vertices
have integer coordinates with $O$ the origin and
$l=||\overset{\rightarrow}{OP}||$. Then the points $P$ and $Q$ are
contained in a plane of equation $a\alpha+b\beta+c\gamma =0$, where
$a,b,c$, and $d$ are integers which satisfy (\ref{eq1}) and
$l^2=2d$.
\end{proposition}
\begin{proof} \ Assume the coordinates of $P$ and $Q$ are denoted as
before. Let us observe that
$l^2=||\overset{\rightarrow}{OP}-\overset{\rightarrow}{OQ}||^2=(x-u)^2+(y-v)^2+(z-w)^2=2l^2-2(xu+yv+zw)$
which implies $xu+yv+zw=\frac{l^2}{2}=d\in \mathbb Z$. Then using
the fact that $\overset{\rightarrow}{OP}$ and
$\overset{\rightarrow}{OQ}$ are contained in the plane orthogonal on
the vector $\overset{\rightarrow}{OP} \times
\overset{\rightarrow}{OQ}=a\overset{\rightarrow}{i}+b\overset{\rightarrow}{j}+c\overset{\rightarrow}{k}$
with $a=yw-vz$, $b=zu-xw$ and $c=xv-yu$ the statement follows from
(\ref{area}).\end{proof}
The equation (\ref{eq1}) has infinitely many integer solutions
besides the obvious ones $a=\pm d$, $b=\pm d$, $c=\pm d$. For
instance we can take $a=-19$, $b=11$, $c=5$ and $d=13$ and the
triangle $OCD$ given in the Introduction has $C$ and $D$ in the
plane $\{(\alpha,\beta,\gamma)\in \mathbb R^3|\
-19\alpha+11\beta+5\gamma =0\}$. For the purpose of computing ${\cal
E}{\cal T}(n)$ one has to consider all the planes of form below
although we are going to concentrate only on those that contain the
origin.
\begin{definition}\label{planesig}
Let us consider the set $\cal P$ of all planes $\{
(\alpha,\beta,\gamma)\in \mathbb R^3|\ a\alpha+b\beta+c\gamma =e\}
$, such that $a^2+b^2+c^2=3d^2$ for some $a,b,c,d,e\in \mathbb Z$
and $\gcd(a,b,c)=1$.
\end{definition}
So, if one starts with a plane $\pi$ in $\cal P$, picks two points
of integer coordinates that belong to $\pi$, which amounts to
solving a simple linear Diophantine equation, the natural question
is weather or not there exists a third point of integer coordinates,
contained in $\pi$, that completes the picture to an equilateral
triangle (see Figure~\ref{figure2}, where $O$ and $P$ are the chosen
points and the third point is denoted here by $Q_{+}$ or $Q_{-}$
since there are two possible such candidates). In order for the
third point to exist one needs to take the first two points in a
certain way. But if one requires only that the new point have
rational coordinates it turns out that this is always possible and
the next theorem gives a way to find the coordinates of the third
point in terms of the given data.
\begin{theorem}\label{construction}
Assume that $P(u,v,w)$ ($u,v,w\in\mathbb Q$) is an arbitrary point
of a plane $\pi\in \cal P$ of normal vector $(a,b,c)$ and passing
through the origin $O$. Then the coordinates of a point $Q(x,y,z)$
situated in $\pi$ and such that the triangle $\triangle OPQ$ is
equilateral are all rational numbers given by:
\begin{equation}\label{eqfortp}
\begin{cases}
\displaystyle x=\frac{u}{2}\pm \frac{cv-bw}{2d}\\
\displaystyle y=\frac{v}{2}\pm \frac{aw -cu}{2d} \\
\displaystyle z =\frac{w}{2}\pm \frac{bu-av}{2d}.
\end{cases}
\end{equation}
\end{theorem}
\begin{center}\label{figure2}
\epsfig{file=plane.eps,height=3in,width=3in}
\vspace{.1in}\\
{\it Figure 2: Plane of normal (a,b,c)} \vspace{.1in}
\end{center}
\vspace{0.2in} \begin{proof} \ From the geometric interpretation of
the problem we see that a point $Q\in \pi$ such that $\triangle OQP$
becomes equilateral is at the intersection of plane $\pi$ and two
spheres of radius $OP$ and centers $O$ and $P$. There are exactly
two points with this property lying on a segment perpendicular to
$OP$ and passing through its midpoint. We want to show that one
point is given by taking the plus sign in all equalities in
(\ref{eqfortp}) and the other point corresponds to the minus sign in
all equalities in (\ref{eqfortp}). We are going to set
$\overset{\rightarrow}{n}=\displaystyle \frac{1}{d\sqrt{3}}(a,b,c)$
which is one of the two unit vectors normal to the plane $\pi$ and
let
$\overset{\rightarrow}{r}=\overset{\longrightarrow}{OP}=(u,v,w)$.
Then the cross product $\overset{\rightarrow}{r}\times
\overset{\rightarrow}{n}$ is given by
$$\overset{\rightarrow}{r}\times \overset{\rightarrow}{n}=\displaystyle \frac{1}{d\sqrt{3}}(cv-bw,aw -cu,bu-av).$$
So we observe that the solution $(x,y,z)$ is, in fact, if written in
vector notation,
$\overset{\longrightarrow}{OQ}_{\pm}=\frac{1}{2}\overset{\rightarrow}{r}\pm
\frac{\sqrt{3}}{2}\overset{\rightarrow}{r}\times
\overset{\rightarrow}{n}$. It is easy now to check that
$\overset{\rightarrow}{r}$ and $\overset{\longrightarrow}{OQ}_{\pm}$
have the same norm and make a $60^{\circ}$ angle in
between.\end{proof}
\section{Solutions of the Diophantine equation $a^2+b^2+c^2=3d^2$}
We have enough evidence to believe that for each $(a,b,c)$
satisfying (\ref{eq1}) for some $d\in \mathbb Z$, there are
infinitely many equilateral triangles in $\mathbb Z^3$ that belong
to a plane of normal vector $(a,b,c)$ and the purpose of this paper
is to determine a way to generate all these triangles. But how many
planes do we have in $\cal P$? Let us observe that if $d$ is even,
then not all of $a$, $b$, $c$ can be odd integers, so at least one
of them must be even. Then the sum of the other two is a number
divisible by $4$ which is possible only if they are also even.
Therefore we may reduce all numbers by a factor of two in this case.
So, if we assume without loss of generality that $\gcd(a,b,c)=1$,
then such solutions of (\ref{eq1}) must have $d$ an odd integer and
then this forces that $a$, $b$ and $c$ must be all odd integers too.
If in addition, disregarding the signs, we have $a,b,c\in\mathbb N$
and $a\le b\le c$, then such a solution will be referred to as a
{\it primitive} solution of (\ref{eq1}). One can find lots of
solutions of (\ref{eq1}) in the following way.
\begin{proposition}\label{paramofplane} (i) The following formulae give a three integer
parameter solution of (\ref{eq1}):
\begin{equation}\label{peqabc}
\begin{cases}
a=-x_1^2+x_2^2+x_3^2-2x_1x_2-2x_1x_3\\
b=x_1^2-x_2^2+x_3^2-2x_xx_1-2x_2x_3\\
c=x_1^2+x_2^2-x_3^2-2x_3x_1-2x_3x_2\\
d=x_1^2+x_2^2+x_3^2
\end{cases},\ \ x_1,x_2,x_3\in \mathbb Z.
\end{equation}
\par
(ii) Every nontrivial primitive solution of (\ref{eq1}) is of the
form (\ref{peqabc}) with $x_1,x_2,x_3 \in \mathbb
Z[\frac{1}{\sqrt{k}}]$ with $k=(3d-a-b-c)/2\in \mathbb N$.
\end{proposition}
\noindent \begin{proof}For the first part of this proposition, one
can check that $a$, $b$, $c$ and $d$ given by (\ref{peqabc}) satisfy
(\ref{eq1}). For the second, we assume that $(a,b,c,d)$ is a
primitive solution of (\ref{eq1}) not equal to $(1,1,1,1)$. Then we
can introduce $x_1=(d-a)t$, $x_2=(d-b)t$, $x_3=(d-c)t$ where $t\in
\mathbb R$ such that $x_1^2+x_2^2+x_3^2=d$. This gives
$$\begin{array}{c}
\displaystyle t^2=\frac{d}{(d-a)^2+(d-b)^2+(d-c)^2}=\frac{d}{6d^2-2d(a+b+c)}=\\
\\
\displaystyle
\frac{1}{2(3d-a-b-c)}=\frac{1}{2}\frac{t}{x_1+x_2+x_3},
\end{array}$$
\noindent and so $t=\frac{1}{2s}$ with
$s=x_1+x_2+x_3=\sqrt{(3d-a-b-c)/2}$. Every primitive solution of
(\ref{eq1}) must have $a$, $b$, $c$, and $d$ all odd numbers as we
have observed. This makes $k=(3d-a-b-c)/2$ an integer. Since $3d>
a+b+c$ is equivalent to
$3(a^2+b^2+c^2)-(a+b+c)^2=(a-b)^2+(b-c)^2+(a-c)^2> 0$, it follows
that $k\in \mathbb N$. Then, for instance, $x_1=\frac{d-a}{2}
\frac{1}{\sqrt{k}}\in \mathbb Z[\frac{1}{\sqrt{k}}]$ because $d-a$
is even. Therefore, in general, every nontrivial primitive solution
of (\ref{eq1}) is of the form (\ref{peqabc}) with $x_1,x_2,x_3 \in
\mathbb Z[\frac{1}{\sqrt{k}}]$.\end{proof}
\vspace{0.1in} \noindent {\bf Example:} The following example shows
that (\ref{peqabc}) does not cover all solutions of (\ref{eq1}).
Suppose $a=5$, $b=11$, $c=19$ and $d=13$. As we have seen in the
proof of Proposition~\ref{paramofplane}, $x_1$, $x_2$, $x_3$ in
(\ref{peqabc}) are uniquely determined: $x_1=\frac{4}{\sqrt{2}}$,
$x_2=\frac{1}{\sqrt{2}}$ and $x_3=-\frac{3}{\sqrt{2}}$.
\vspace{0.1in} \noindent Although formulae (\ref{peqabc}) provide
infinitely many solutions of (\ref{eq1}), the following proposition
brings additional information about its integer solutions.
\begin{proposition}\label{sofabcfedodd} The equation (\ref{eq1}) has non-trivial
solutions for every odd integer $d\ge 3$.
\end{proposition}
\noindent \begin{proof} If $d=2p+1$ for some $p\in \mathbb Z$,
$p\ge 1$, then $3d^2=3[4p(p+1)+1]= 8l+3$ which shows that
$3q^2\equiv 3$ (mod 8). From Gauss's Theorem about the number of
representations of a number as a sum of three squares
(\cite[Thm.\ 2, p.\ 51]{gr}), we see that the number of representations of
$3d^2$ as a sum of three squares is at least 24. Here, the change of
signs is counted so each solution actually generates eight solutions
by just changing the signs. Also, the six permutations of $a$, $b$
and $c$ get into the counting process. So, there must be at least
one solution which is nontrivial since the solution
$3d^2=d^2+d^2+d^2$ generates only eight solutions by using all
possible change of signs.~\end{proof}
{\bf Remark:} One can generate an infinite family of solutions of
(\ref{eq1}) by reducing it two separate equations, say for example
$146=3d^2-c^2$ and $ a^2+b^2=146$. The second equation admits as
solution, for instance, $a=11$ and $b=5$. The first equation has a
particular solution $d=7$ and $c=1$. Then one can use the recurrence
formulae to obtain infinitely many solutions of $146=3d^2-c^2$:
$$d_{n+1}=2d_n+c_n,\ \ c_{n+1}=3d_n+2c_n\ for\ n\in \mathbb N$$ and
$d_1=7$, $c_1=1$. A simple calculation shows that
$3d_{n+1}^2-c_{n+1}^2=3d_n^2-c_n^2$ so, by induction, $(d_n, c_n)$
is a solution of the equation $3d^2-c^2=146$ for all $n\in \mathbb
N$. It is easy to see that $q_n$ and $c_n$ are increasing sequences
so this procedure generates infinitely many solutions of
(\ref{eq1}).
The primitive solutions of (\ref{eq1}) for small values of $d$ are
included in the Table~3.
\vspace{0.1in}
\begin{table}\label{psoabc}
\centerline{
\begin{tabular}{||c|c||c|c||}
\hline \hline
d& (a,b,c) & d & (a,b,c )\\
\hline \hline
$1$& $\{(1,1,1)\}$ & $9$& $\{(1,11,11),(5,7,13)\}$\\
\hline $3$& $\{(1,1,5)\}$ & $11$& $\{(1,1,19),(5,7,17),(5,13,13)\}$\\
\hline $5$& $\{(1,5,7)\}$& $13$& $\{(5,11,19),(7,13,17)\}$\\
\hline $7$& $\{(1,5,11)\}$ & $15$& $\{(1,7,25),(5,11,25),(5,17,19)\}$\\
\hline \hline
\end{tabular}}\vspace{0.1in}
\caption{Primitive solutions of (\ref{eq1})} \vspace{0.1in}
\end{table}
As a curiosity the number of primitive representations as in
(\ref{eq1}) corresponding to $d=2007$ is 333.
\section{The first six parametrizations}
The simplest solution of (\ref{eq1}) is $a=b=c=d=1$. We are going to
introduce some more notation here before we give the parametrization
for this case.
\begin{definition}\label{etinplabc} For every $(a,b,c)$, a primitive
solution of (\ref{eq1}), denote by ${\cal T} _{a,b,c}$ the set of
all equilateral triangles with integer coordinates having the origin
as one of the vertices and the other two lie in the plane
$\{(\alpha,\beta,\gamma)\in \mathbb R^3 |a\alpha
+b\beta+c\gamma=0\}$.
\end{definition}
It is clear now that in light of Proposition~\ref{planeprop}, every
equilateral triangle having integer coordinates after a translation,
interchange of coordinates, or maybe a change of signs of some of
the coordinates, belongs to one of the classes ${\cal T}_{a,b,c}$.
Let us introduce also the notation $\cal T$ for all the equilateral
triangles in $\mathbb Z^3$. If we have different values for $a$, $b$
and $c$, how many different planes can one obtain by permuting $a$,
$b$ and $c$ in between and changing their signs? That will be $6$
permutations and essentially $4$ change of signs (note that
$ax+by+cz=0$ is the same plane as $(-a)x+(-b)y+(-c)z=0$) which gives
a total of $24$ such transformations. We are going to denote the
group of symmetries of the space determined by these transformations
and leave the origin fixed, by ${\cal S}_{cube}$ (it is actually the
group of symmetries of the cube). Hence we have
\vspace{0.1in}
\begin{equation}\label{partition}
{\cal T} =\underset{\begin{array}{c} s\in {\cal S}_{cube},s(O)=O,\\
a^2+b^2+c^2=3d^2\\ 01$.
\section{Characterization of side lengths}
We will begin with two preliminary results. The first we just need
to recall it since it is a known fact that can be found in number
theory books mostly as an exercise or as an implicit corollary of
more general theorems about quadratic forms or Euler's $6k+1$
theorem (see \cite{r}, pp. 568 and \cite{g}, pp. 56).
\begin{proposition}\label{ejtypet} An integer $t$ can be written as $m^2-mn+n^2$ for some
$m,n\in \mathbb Z$ if and only if in the prime factorization of $t$,
$2$ and the primes of the form $6k-1$ appear to an even exponent.
\end{proposition}
%Euler 6k+1 theorem says that every prime of the form 6k+1 can be written
%as x^2+3y^2 with x,y positive integers.
The next lemma is probably also known in algebraic number theory but
we do not have straight reference for it so we are going to include
a proof of it.
\begin{lemma}\label{thebeauty} An integer $t$ which can be written as $t=3x^2-y^2$ with
$x,y \in \Bbb Z$ is the sum of two squares if and only if $t$ is of
the form $t=2(m^2-mn+n^2)$ for some integers $m$ and $n$.
\end{lemma}
\begin{proof} . \ For necessity, by Proposition~\ref{ejtypet}, we have to
show that $t$ is even and $t/2$ does not contain in its prime factor
decomposition any of the primes $2$ or those of the form $6k-1$
except to an even power. First, let us show that $t$ must be even
and the exponent of $2$ in its prime factorization is odd. Since
$t=3x^2-y^2=a^2+b^2$ implies $3x^2=a^2+b^2+y^2$ we have observed
that either all $x$, $y$, $a$, and $b$ are even or all odd.
If $x$, $y$, $a$ and $b$ are all even we can factor out a $2$ from
all these numbers and reduce the problem to $t/4$ instead of $t$.
Applying this arguments several times one can see that
$t=2^{2l+1}t'$ with $t'$ odd and $l\in \Bbb Z$. Without loss of
generality we may assume that $l=0$. In this case $t$ contains only
one power of $2$ in its prime decomposition and so $x$, $y$, $a$ and
$b$ must be all odd.
Let us then suppose that $t/2$ is divisible by a prime $p=6k-1$ for
some $k\in \mathbb N$. We need to show that the exponent of $p$ in
the prime factorization of $t$ is even. Since $p$ divides
$t=3x^2-y^2$ we get that $3x^2\equiv y^2$ (mod $p$). If $p$ divides
$x$, then $p$ divides $y$ and so $p^2$ divides $t$ which reduces the
problem to $t/p^2$. Applying this argument several times we arrive
to a point when $p$ does not divide $x/p^i$. So, discarding an even
number of $p$'s from $t$, we may assume that $i=0$. This implies
that $x$ has an inverse modulo $p$ and then $z^2\equiv 3$ (mod $p$),
where $z=x^{-1}y$. Using the Legendre symbol this says that
$(\frac{3}{p})=1$. By the Law of Quadratic Reciprocity
(\cite[Thm.\ 11.7]{r}) we see that
$(\frac{p}{3})=(-1)^{\frac{(p-1)}{2}\frac{3-1}{2}}=(-1)^{3k-1}$. But
the equation in $w$, $w^2\equiv p$ (mod $3$), is equivalent to
$w^2\equiv -1$ (mod $3$) which obviously has no solution in $w$.
This implies $(\frac{p}{3})=-1$ and so $k$ has to be even. Therefore
$p=12k'-1=4j+3$ for some $j\in \Bbb Z$. But by hypothesis, $t$ is a
sum of two squares and so, from Euler's characterization of those
numbers, $p$ must have an even exponent in the prime decomposition
of $t$.
For sufficiency, let us assume that $t=3x^2-y^2=2(m^2-mn+n^2)$ for
some $x,y,m,n\in \Bbb Z$. Using Euler's characterization we have to
show that if $p=4k+3$ is a prime dividing $t$ then the exponent in
its prime decomposition is even. If $p=3$, then $3$ divides $y$
which implies $x^2-3{y'}^2=2({m'}^2-m'n'+{n'}^2)$. This is true
because of Proposition~\ref{ejtypet} which one has to use in both
directions.
If ${m'}^2-m'n'+{n'}^2$ is not divisible by $3$ then
${m'}^2-m'n'+{n'}^2\equiv 1$ (mod 3) since all the prime factors of
the form $6k-1$ and $2$ appear to even exponents. This implies
$x^2-3{y'}^2\equiv 2({m'}^2-m'n'+{n'}^2)$ (mod $3$) or $x^2\equiv 2$
(mod $3$) which is a contradiction. So, ${m'}^2-m'n'+{n'}^2$ must
contain another factor of $3$ and so the problem could be then
reduced to $t/9$ instead of $t$. Hence $3$ must have an even
exponent in the prime decomposition of $t$.
Let us assume that $k=3j-1$ with $j\in \Bbb Z$. Then
$p=12j-1=6(2j)-1$ and so these primes must appear to an even power
in the decomposition of $m^2-mn+n^2$. The case $k=3j+1$ ($p=12k+7$)
is not possible because that will contradict the Law of Quadratic
Reciprocity: $(\frac{3}{p})(\frac{p}{3})=1\not
=(-1)^{\frac{(p-1)}{2}\frac{3-1}{2}}$. \end{proof}
\begin{theorem}\label{main}
An equilateral triangle of side lengths $l$ and having integer
coordinates in $\Bbb R^3$ exists, if and only if
$l=\sqrt{2(m^2-mn+n^2)}$ for some integers $m$ and $n$ (not both
zero) .
\end{theorem}
\noindent \begin{proof} \ The sufficiency part of the theorem is
given by the triangles in $\cal T_{1,1,1}$ (Theorem~\ref{firstpar}).
For necessity let us start with an arbitrary equilateral triangle
having integer coordinates and non zero side lengths $l$. Without
loss of generality we may assume that one of its vertices is the
origin. Denote the triangle as before $\triangle OPQ$, with
$P(u,v,w)$ and $Q(x,y,z)$. As we have shown in
Proposition~\ref{planeprop} we know that $au+bv+wc=0$ and
$ax+by+cz=0$ for some $a,b,c$ satisfying $a^2+b^2+c^2=3d^2$ and
$\gcd(a,b,c)=1$. We noticed too that all $a,b,c$ have to be odd
integers and so, in particular, they are all non-zero numbers.
Then $$l^2=u^2+v^2+w^2=\left(\displaystyle
\frac{bv+cw}{a}\right)^2+v^2+w^2=\displaystyle
\frac{(a^2+b^2)v^2+2bcvw+(a^2+c^2)w^2}{a^2}.$$
Completing the square we have
\[
\begin{array}{c}
a^2l^2=(a^2+b^2)v^2+2bcvw+(a^2+c^2)w^2= (3d^2-c^2)\left
(v+\displaystyle \frac{bcw}{3 d^2-c^2}\right )^2+(3d^2-b^2)w^2-\\ \\
\displaystyle \frac{b^2c^2w^2}{3d^2-c^2}= (3d^2-c^2)\left
(v+\displaystyle \frac{bcw}{3d^2-c^2}\right
)^2+3\frac{d^2a^2w^2}{3d^2-c^2},
\end{array}\]
or
$$a^2(3d^2-c^2)l^2=[(3d^2-c^2)v+bcw]^2+3d^2a^2w^2.$$
This calculation shows that $a^2(3d^2-c^2)l^2=m'^2-m'n'+n'^2$ where
$m'=(3d^2-c^2)v+bcw+daw$ and $n'=2daw$. By Lemma~\ref{thebeauty} we
can write $3d^2-c^2=a^2+b^2=2(m''^2-m''n''+n''^2)$ for some
$m'',n''\in\mathbb Z$. Hence
$$l^2=2\frac{1}{(2a)^2}\frac{m'^2-m'n'+n'^2}{m''^2-m''n''+n''^2}.$$ Because
$l^2\in \mathbb Z$ and Proposition~\ref{ejtypet} we see that
$l^2=2(m^2-mn+n^2)$. \end{proof}
We include here a similar result and the last of this section which
is only based on Proposition~\ref{planeprop}.
\begin{proposition}\label{tetrahedra}
A regular tetrahedra of side lengths $l$ and having integer
coordinates in $\Bbb R^3$ exists, if and only if $l=m\sqrt{2}$ for
some $m\in \mathbb N$.
\end{proposition}
\noindent \begin{proof} For sufficiency, we can take the tetrahedra
$OPQR$ with $P(m,0,m)$, $Q(m,m,0)$ and $R(0,m,m)$.
For necessity, without loss of generality we assume the tetrahedra
$OPQR$ is regular and has all its coordinates integers. As before,
we assume $P(u,v,w)$ and $Q(x,y,z)$.
\begin{center}\label{figure3}
\epsfig{file=regtetrahedra.eps,height=3in,width=3in}
\vspace{.1in}\\
{\it Figure 3: Regular tetrahedra} \vspace{.1in}
\end{center}
Let $E$ be the center of the face $\triangle OPQ$. Then from
Proposition~\ref{planeprop} we know that
$\frac{\overset{\rightarrow}{ER}}{|\overset{\rightarrow}{ER}| }=
\frac{(a,b,c)}{\sqrt{a^2+b^2+c^2}}=\frac{1}{\sqrt{3}d}(a,b,c)$ for
some $a,b,c,d\in\mathbb Z$, $l^2=2d$. The coordinates of $E$ are
$(\frac{u+x}{3},\frac{y+v}{3},\frac{z+w}{3})$.
From the Pythagorean theorem one can find easily that $RE=l
\sqrt{\frac{2}{3}}$. Since
$\overset{\rightarrow}{OR}=\overset{\rightarrow}{OE}+\overset{\rightarrow}{ER}$,
the coordinates of $R$ must be given by
$$\left(\frac{u+x}{3}\pm
l\sqrt{\frac{2}{3}}\frac{1}{\sqrt{3}d}a,\frac{y+v}{3}\pm
l\sqrt{\frac{2}{3}} \frac{1}{\sqrt{3}d}b,\frac{z+w}{3}\pm
l\sqrt{\frac{2}{3}}\frac{1}{\sqrt{3}d}c\right)$$ or
$$\left(\frac{u+x}{3}\pm \frac{2\sqrt{2}}{3l}a,\frac{y+v}{3}\pm \frac{2\sqrt{2}}{3l}b,
\frac{z+w}{3}\pm \frac{2\sqrt{2}}{3l}c\right).$$ Since these
coordinates are assumed to be integers we see that
$l=m\sqrt{2}$.\end{proof}
Another natural question that one may ask at this point is weather
or not every family ${\cal T}_{a,b,c}$ contains triangles which are
faces of regular tetrahedra with integer coordinates. We believe
that every triangle in ${\cal T}_{a,b,c}$ is a face of such a
tetrahedra as long as its sides, in light of Theorem~\ref{main}, are
of the form $l=\sqrt{2N(m,n)}$ with $N(m,n)$ a perfect square. We
leave this conjecture for further study. To find values of $m$,
$n\in\mathbb Z$ such that $N(m,n)=m^2-mn+n^2$ is a perfect square,
of course one can accomplish this in the trivial way, by taking
$m=0$ or $n=0$ but there are also infinitely many non-trivial
solutions as one can see from Proposition~\ref{ejtypet}.
\section{A more general parametrization}
Our construction depends on a particular solution, $(r,s)\in \mathbb
Z^2$, of the equation:
\begin{equation}\label{rs}
2(a^2+b^2)=s^2+3r^2.
\end{equation}
As before let us assume that $a$, $b$, $c$ and $q$ are integers satisfying $a^2+b^2+c^2=3d^2$ with $d$ an
odd positive integer and $\gcd(a,b,c)=1$. By Lemma~\ref{thebeauty} we
see that $3d^2-c^2=a^2+b^2=2(f^2-fg+g^2)$ for some $f,g\in \mathbb
Z$ and so $2(a^2+b^2)=(2f-g)^2+3g^2$ which says that equation
(\ref{rs}) has always an integer solution.
\begin{theorem}\label{generalpar} Let $a$, $b$, $c$, $d$ be odd positive
integers such that $a^2+b^2+c^2=3d^2$, $a\le b\le c$ and
$\gcd(d,c)=1$. Then ${\cal T}_{a,b,c}=\{ \triangle OPQ|\
m,n\in\mathbb Z\}$ where the points $P(u,v,w)$ and $Q(x,y,z)$ are
given by
\begin{equation}\label{paramone}
\begin{cases}
u=m_um-n_un,\\
v=m_vm-n_vn,\\
w=m_wm-n_wn, \\
\end{cases}
\ \ \ \text{and} \ \ \
\begin{cases}
x=m_xm-n_xn,\\
y=m_ym-n_yn,\\
z=m_zm-n_zn,
\end{cases}\ \
\end{equation}
with
\begin{equation}\label{paramtwo}
\begin{array}{l}
\begin{cases}
m_x=-\frac{1}{2}[db(3r+s)+ac(r-s)]/q,\ \ \ n_x=-(rac+dbs)/q\\
m_y=\frac{1}{2}[da(3r+s)-bc(r-s)]/q,\ \ \ \ \ n_y=(das-bcr)/q\\
m_z=(r-s)/2,\ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ n_z=r\\
\end{cases}\\
\ \ \ \text{and}\ \ \\
\begin{cases}
m_u=-(rac+dbs)/q,\ \ \ n_u=-\frac{1}{2}[db(s-3r)+ac(r+s)]/q\\
m_v=(das-rbc)/q,\ \ \ \ \ n_v=\frac{1}{2}[da(s-3r)-bc(r+s)]/q\\
m_w=r,\ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ n_w=(r+s)/2\\
\end{cases}
\end{array}
\end{equation}
where $q=a^2+b^2$ and $(r,s)$ is a suitable solution of (\ref{rs}).
\end{theorem}
In order to prove Proposition~\ref{generalpar} we need the following
lemma.
\begin{lemma}\label{existence}
Suppose $A$ and $B$ are integers in such a way $A^2+3B^2$ is
divisible by $q$ where $2q$ can be written as $s'^2+3r'^2$ with
$r',s'\in \mathbb Z$. Then there exist a representation
$2q=s^2+3r^2$, $r,s\in \mathbb Z$, such that
$$Ar+Bs \equiv 0\ \ ({\rm mod}\ 2q),$$
and
$$As-3Br \equiv 0\ \ ({\rm mod} \ 2q).$$
\end{lemma}
\begin{proof} Let us observe that every number of the form $A^2+3B^2$
is an Eisenstein-Jacobi integer since $A^2+3B^2=(A+B)^2-(A+B)(2B)+(2B)^2$.
Conversely if $n$ is even then
$m^2-mn+n^2=(m-n/2)^2+3(n/2)^2$ and since $m^2-mn+n^2=(n-m)^2-(n-m)n+n^2$ we see that every
Eisenstein-Jacobi integer is of the form $A^2+3B^2$ for some $A,B\in
\mathbb Z$. Using Proposition~\ref{ejtypet} we can write
$$A^2+3B^2=2^{2\alpha}(\underset{t\in T}{\prod}
p_t)^2\underset{j\in J} {\prod} p_j,\ \ \alpha\in \mathbb N, $$
and
$$2q=2^{2\beta}(\underset{t\in I'}{\prod} p_t)^2\underset{j\in J'}
{\prod }p_j,\ \ 1\le \beta \le \alpha, \ T' \subset T,\ J'\subset
J,$$
\noindent with $p_t$ prime of the form $6k-1$ for $t\in T$ and $p_j$
prime of the form $6k+1$ or equal to $3$ for all $j\in J$. From
Euler's $6k+1$ theorem, for each $j\in J$ we can write
$p_j=(m_j+n_j\sqrt{3}i)(m_j-n_j\sqrt{3}i)$ and we make the choice of
$m_j$ and $n_j$ in $\mathbb Z$ in such a way
$A+B\sqrt{3}i=s\underset{j\in J}{\prod }(m_j+n_j\sqrt{3}i)$ using
the prime factorization in $\mathbb Z[\sqrt{3}i]$ of $A+B\sqrt{3}i$
and $h=2^{\alpha}\underset{t\in T}{\prod} p_t$.
Then we take $r$ and $s$ such that $s+r\sqrt{3}i=h'\underset{j\in
J'}{\prod }(m_j-n_j\sqrt{3}i)$ with $h'=2^{\beta}\underset{t\in
T'}{\prod} p_t$. Notice that
$2q=(s+r\sqrt{3}i)(s-r\sqrt{3}i)=s^2+3r^2$ and
$(A+B\sqrt{3}i)(s+r\sqrt{3}i)=2q(A'+B'\sqrt{3}i)$. Identifying the
coefficients we get $As+3Br=2qA'$ and $Ar+Bs=2qB'$ and the
conclusion of our lemma follows from this.\end{proof}
To return to the proof of Theorem~\ref{generalpar} we begin with the
next proposition.
\begin{proposition} \label{integer} For some particular solution $(r,s)\in \mathbb Z^2$ of
(\ref{rs}) all of the coefficients $m_u$, $m_v$, $m_w$, $n_u$, $n_v$, $n_w$, $m_x$,
$m_y$, $m_z$, $n_x$,$n_y$, $n_z$ in (\ref{paramtwo}) are integers. \end{proposition}
\begin{proof} One can check that
\begin{equation}\label{equationssbparf}
\begin{cases}
m_x^2+m_y^2+m_z^2=n_x^2+n_y^2+n_z^2=m_u^2+m_v^2+m_w^2=n_u^2+n_v^2+n_w^2=2d^2, \\
m_xn_x+m_yn_y+m_zn_z=m_un_u+m_vn_v+m_wn_w=d^2,\\
am_x+bm_y+cm_z=am_u+bm_v+cm_w=an_x+bn_y+cn_z=an_u+bn_v+cn_w=0
\end{cases}
\end{equation}
These identities insures that the points $P(u,v,z)$ and $Q(x,y,z)$
are in the plane of normal vector $(a,b,c)$ and containing the
origin, the $\triangle OPQ$ is equilateral for every values of $m$,
$n$ and its side lengths are $l=d\sqrt{2(m^2-mn+n^2)}$. These
calculations are tedious and so we are not going to include them
here. The only ingredients that are used in establishing all these
identities are the two relations between $a$, $b$, $c$, $d$, $r$ and
$s$.
From (\ref{rs}) we see that $r$ and $s$ have to be of the same
parity. Then, it is clear that $m_z$, $n_z$, $m_w$, and $n_w$ are
all integers. Because the equalities in (\ref{equationssbparf}) are
satisfied it suffices to show that $m_x$, $n_x$, $m_u$, and $n_u$
are integers for some choice of $(r,s)$ solution of (\ref{rs}). To
show that $n_x$ is an integer we need to show that $q$ divides
$N=rac+dbs$.
Let us observe that $c^2\equiv 3d^2$ and $a^2\equiv -b^2$ (mod
$q$). Multiplying together these two congruences we obtain
$(ac)^2+3(db)^2\equiv 0 $ (mod $q$). Using Lemma~\ref{existence} we
see that $N$ is divisible by $2q$ for some choice of $r$ and $s$ as
in (\ref{rs}). So, $n_x\in \mathbb Z$.
Next we want to show that $m_x$ is an integer. First let us observe
that if $M=3dbr-acs$ we can apply the second part of
Lemma~\ref{existence} to conclude that $2q$ divides $M$ also. Hence
$2q$ divides $M+N=db(3r+s)+ac(r-s)$ and so, $m_x$ is an integer.
Because $m_x+n_u=n_x$ and $m_u=n_x$ it follows that $n_u$ and $m_u$
are also integers. \end{proof}
{\bf Remark:} Let us observe that the parametric formulae
(\ref{paramone}) and (\ref{paramtwo}) exist under no extra
assumption on $a$, $b$ and $c$. The question is whether or not every
triangle in ${\cal T}_{a,b,c}$ is given by these formulae. So, with
these preparations we can return to prove that this is indeed the
case under the assumption of Theorem~\ref{generalpar}.
\begin{proof} We start with a triangle in ${\cal T}_{a,b,c}$ say
$\triangle OPQ$ with the notation as before. We know that
$P(u_0,v_0,w_0)$ and $Q(x_0,y_0,z_0)$ belong to the plane of
equation $a\alpha+b\beta+c\gamma=0$ and by
Theorem~\ref{construction} we see that the coordinates of $P$ and
$Q$ should satisfy (\ref{eqfortp}). Hence, using the same notation,
$cv_0-bw_0$, $aw_0-cu_0$ and $bu_0-aw_0$ are divisible by $d$. A
relatively simple calculation shows that $m_vn_w-m_wn_v=ad$,
$m_wn_u-m_un_w=bd$ and $m_un_v-m_vn_u=cd$. We would like to solve
the following system in $m$ and $n$:
\begin{equation}\label{fsystem}
\begin{cases}
u_0=m_um-n_un,\\
v_0=m_vm-n_vn,\\
w_0=m_wm-n_wn. \\
\end{cases}
\end{equation}
\noindent The equalities (\ref{equationssbparf}) and the fact that
$(u_0,v_0,w_0)$ is in the plane $a\alpha+b\beta+c\gamma=0$ insures
that (\ref{fsystem}) has a unique real solution in $m$ and $n$. We
want to show that this solution is in fact an integer solution. The
value of $n$ can be solved from each pair of these equations to get
$$n=\frac{v_0m_w-w_0m_v}{ad}=\frac{w_0m_u-u_0m_w}{bd}=\frac{u_0m_v-v_0m_u}{cd}.$$
Since $\gcd(a,b,c)=1$ we can find integers $a'$, $b'$, $c'$ such that
$aa'+bb'+cc'=1$. Hence the above sequence of equalities gives
$$n=
\frac{a'(v_0m_w-w_0m_v)+b'(w_0m_u-u_0m_w)+c'(u_0m_v-v_0m_u)}{d}$$
So, it suffices to show that $d$ divides $v_0m_w-w_0m_v$,
$w_0m_u-u_0m_w$ and $u_0m_v-v_0m_u$.
Next, let us calculate for instance $v_0m_w-w_0m_v$ in more detail:
$$\begin{array}{l}
\displaystyle v_0m_w-w_0m_v=v_0r-\frac{das-rbc}{q}w_0=\frac{v_0qr-(das-rbc)w_0}{q}=\\ \\
\displaystyle \frac{-dasw_0+v_0(3d^2-c^2)r+rbcw_0}{q}=\frac{c(bw_0-v_0c)r+3rv_0d^2-dasw_0}{q}.
\end{array}$$
From Theorem~\ref{planeprop} we see that $bw_0-v_0c=\pm
d(2x_0-u_0)$. Hence,
$$\displaystyle v_0m_w-w_0m_v=\frac{d[\pm c(2x_0-u_0)r+3rv_0d-asw_0]}{q}.$$
This shows that $d$ divides $v_0m_w-w_0m_v$ provided that
$\gcd(q,d)=1$. This is true since $\gcd(d,c)=1$ implies $\gcd(q,d)=1$.
Similar calculations show that $d$ divides $w_0m_u-u_0m_w$ and
$u_0m_v-v_0m_u$. Hence $n$ must be an integer. Similarly one shows
that $m$ is an integer. By replacing $P$ with $Q$ if necessary all
the equalities in (\ref{paramone}) have to hold true by
Theorem~\ref{planeprop}.\end{proof}
\section{Acknowledgements and further investigations}
It turns out that the Diophantine equation $a^2+b^2+c^2=3d^2$ has
plenty of solutions which satisfy $\gcd(a,b,c)=1$, $\gcd(a,d)>1$,
$\gcd(b,d)>1$ and $\gcd(c,d)>1$. Let us refer to these as {\it
degenerate} solutions. We have searched for an example of such a
degenerate solution but did not find one between all odd $d\le
4095$. Thanks to Professor Florian Luca who pointed us in the right
direction, we have found the following first concrete degenerate
solution: $a=(17)(41)(79)$, $b=(23)(31)(3361)$,
$c=(5)(13)(71)(241)(541)(2017)$ and $d=(3)(13)(41)(241)(3361)$. It
will be interesting to find the smallest $d$ for which a
corresponding degenerate solution exists. One needs to investigate
whether or not the proof of Theorem~\ref{generalpar} can be adapted
to include the case of degenerate solutions.
We would like to thank the referee of this paper who has helped us
understand more about the existence of such solutions by providing a
sketch for the proof of the following facts:
\begin{equation}
\#\{(a,b,c,d):1\le a\le b\le c\le x,
a^2+b^2+c^2=3d^2,\gcd(a,b,c)=1\}\ll x^2(\ln x)^2,
\end{equation}
\noindent and
\begin{equation}
\begin{array}{l}
\#\{(a,b,c,d):1\le a\le b\le c\le x,
a^2+b^2+c^2=3d^2,\gcd(a,b,c)=1,\\ \\
\min(\gcd(a,d),\gcd(b,d),\gcd(c,d))>1\}\gg \frac{x^2}{(\ln x)^\alpha},
\end{array}
\end{equation}
\noindent for some $\alpha>0$ and large enough $x$.
Also, determining the density of the degenerate solutions in the
set of all solutions becomes an interesting nontrivial problem in
analytic number theory.
\begin{thebibliography}{9}
\bibitem{g} R.~Guy, {\it Unsolved Problems in Number Theory}, Springer-Verlag, 2004.
\bibitem{gr} E.~Grosswald, {\it Representations of Integers as Sums of Squares}, Springer Verlag, 1985.
\bibitem{s} I.~J.~Schoenberg, {\it Regular simplices and
quadratic forms}, {\it J. London Math. Soc.} {\bf 12} (1937), 48--55.
\bibitem{r} K.~Rosen, {\it Elementary Number Theory}, Fifth Edition,
Addison-Wesley, 2004.
\bibitem{OL}
Neil J.~A. Sloane, {\em The On-Line Encyclopedia of
Integer Sequences}, 2007, \newblock published
electronically at \href{http://www.research.att.com/~njas/sequences/}{\tt http://www.research.att.com/$\sim$njas/sequences/}.
\end{thebibliography}
\bigskip
\hrule
\bigskip
\noindent 2000 {\it Mathematics Subject Classification}:
Primary 11A67; Secondary 11D09, 11D04, 11R99, 11B99, 51N20.\\
\noindent {\it Keywords}: Diophantine equation, equilateral
triangle, quadratic reciprocity.
\bigskip
\hrule
\bigskip
\noindent (Concerned with sequence \seqnum{A102698}.)
\bigskip
\hrule
\bigskip
\vspace*{+.1in}
\noindent
Received August 28 2006;
revised version received June 16 2007.
Published in {\it Journal of Integer Sequences}, June 18 2007.
\bigskip
\hrule
\bigskip
\noindent
Return to
\htmladdnormallink{Journal of Integer Sequences home page}{http://www.cs.uwaterloo.ca/journals/JIS/}.
\vskip .1in
\end{document}