(I'm not sure how to tag this; I'm tagging it math.CO because that's where it arose, but math.FA or something might be appropriate as well. Feel free to edit or comment on what this should be.)

I have a polynomial in a bunch of variables $P(x_1, x_2, \ldots, x_n, y_1, y_2, \ldots, y_m)$. I know that if we evaluate the polynomial at

$x_1 = x_2 = \ldots = x_n = 1, y_1 = y_2 = \ldots = y_m = 0$

then the result is greater than the evaluation at

$x_1 = x_2 = \ldots = x_n = 0, y_1 = y_2 = \ldots = y_m = 1$.

Then are there necessarily functions

$f_1, \ldots, f_n$, with $f_i \in C^1[0,1], f_i(0) = 0, f_i(1) = 1$


$g_1, \ldots, g_m$, with $g_j \in C^1[0,1], g_j(0) = 1, g_j(1) = 0$

such that

$Q(t) = P(f_1(t), f_2(t), \ldots, f_n(t), g_1(t), \ldots, g_m(t))$

is monotonically decreasing on $[0,1]$?

  • $\begingroup$ I'm embarrassed at how trivial it is to see the necessary and sufficient condition -- I just got bogged down in the complexity of the application, I think. $\endgroup$ May 9, 2011 at 14:08

1 Answer 1


The answer is no. Is suffices to construct a polynomial such that these points are local minima. An example when $n=1$ is $$x^2(y-1)^2(\epsilon+y^2(x-1)^2)$$ with $0<\epsilon$ small.

  • $\begingroup$ I have a sneaking suspicion that I asked the question wrong and this situation doesn't apply; nevertheless, I think I have a better intuition for the problem now. $\endgroup$ May 9, 2011 at 13:59

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.