Trying to outline an actual proof why @boboquack's result is minimal; please comment on how to finish this:
Let the numbers involved be $(a,b,c)$, WLOG in increasing order, and the squares involved be $(X,Y,Z)$ WLOG in increasing order. (Clearly X,Y,Z are distinct, otherwise at least two of the a,b,c would also be identical).
Then we must have $a+b=X, a+c=Y, b+c=Z$ for $a_i \geq 1$
Now we're trying to minimize the sum S of the numbers, which gives us a constraint on the sum of squares: $S = a+b+c = (X+Y+Z)/2$
Write $x^2=X, y^2=Y, z^2=Z$ and in fact $X=x^2, Y=(x+u)^2, Z=(x+u+v)^2$ for some positive integers $u,v \geq 1$. (We suspect our solution has u=v=1 and minimal x)
Then our constraint on S becomes:
$2S = 2(a+b+c) = (X+Y+Z) = x^2 + (x+u)^2 + (x+u+v)^2$
$= 3x^2 + 2(2u+v)x + u^2+(u+v)^2$
Now consider how we partition $X=a+b$ into two positive integer parts (a,b) in increasing order. The number of ways of doing that is $\lfloor X/2 \rfloor$, and the difference (b-a) is at most (X-2). Clearly we must have $x \geq 3$, i.e. $X \geq 9$
Similarly to @boboquack we use that to get a minimum value on X, and hence x. Then u,v. This then gives us the minimal $S_{min}$. We are told $S_{min} \leq (6^2+7^2+9^2)/2 = 83$
Another unused line of attack: Note the residues of the x,y,z modulo 3 must be (0,1,2) in some order. And the residues mod-3 of the squares $x^2=X, y^2=Y, z^2=Z$ must be $(0, 1, 2^2 \equiv 1)$ in some order. Similarly show by contradiction that one of the x,y,z is divisible by 4.
0
included? $\endgroup$