Obviously, the set of points that maximizes our sum is equal to the one that minimizes with the signs of the numbers swapped (our sum is an odd function), so:
and
With :
(1)
(2)
(3)
Equating the partial derivatives to 0, we have:
or (4)
So I will assume that lambda is negative and that there are b numbers of xi's that are negative and (n-b) numbers of xi's that are positive. (5)
Applying (5) and (4) to (3), we get:
(6)
Applying (5) and...
Your result for n=3 is positive for my conjecture.For n=4, the solutions I obtained from the partial derivatives do not follow the imposed restrictions. For forcing such restrictions on the obtained system I arrived at xi=(2/n)^(1/2) or xi= - (1/2n)^(1/2); it's easy to see that this only adds up...
I applied the partial derivatives you suggested and arrived at x1=0, the problem is that for n=3, the optimal solution does not contain null terms. I apologize if this is too obvious, I'm just a humble medical student who likes numbers.
To solve the elegant asymmetry, I took ai = 1/n + xi. (n^2-n)^(-1/2), such a transformation led me to non-negative real numbers such that:
a1+a2+a3...+an=1
It is
a1^2+a2^2+a3^2...+an^2= (n-1)^(-1). The problem is to get the maximum value of a1^3+a2^3+a3^3...+an^3.