Unofficial archive — problems, solutions & results © IMC, reproduced with permission.

IMC / 2005 / Problems / Day 2, P11

IMC 2005 · Day 2 · P11

killer

Find all r>0r > 0 such that whenever f:R2Rf : \mathbb{R}^2 \to \mathbb{R} is a differentiable function such that gradf(0,0)=1|\operatorname{grad} f(0,0)| = 1 and gradf(u)gradf(v)uv|\operatorname{grad} f(u) - \operatorname{grad} f(v)| \le |u - v| for all u,vR2u, v \in \mathbb{R}^2, then the maximum of ff on the disk {uR2:ur}\{ u \in \mathbb{R}^2 : |u| \le r \} is attained at exactly one point. (gradf(u)=(1f(u),2f(u))\operatorname{grad} f(u) = (\partial_1 f(u), \partial_2 f(u)) is the gradient vector of ff at the point uu. For a vector u=(a,b)u = (a, b), u=a2+b2|u| = \sqrt{a^2 + b^2}.)

Solution (official)

To get an upper bound for rr, set f(x,y)=xx22+y22f(x, y) = x - \dfrac{x^2}{2} + \dfrac{y^2}{2}. This function satisfies the conditions, since gradf(x,y)=(1x,y)\operatorname{grad} f(x, y) = (1 - x, y), gradf(0,0)=(1,0)\operatorname{grad} f(0, 0) = (1, 0) and gradf(x1,y1)gradf(x2,y2)=(x2x1,y1y2)=(x1,y1)(x2,y2)|\operatorname{grad} f(x_1, y_1) - \operatorname{grad} f(x_2, y_2)| = |(x_2 - x_1, y_1 - y_2)| = |(x_1, y_1) - (x_2, y_2)|.

In the disk Dr={(x,y):x2+y2r2}D_r = \{ (x, y) : x^2 + y^2 \le r^2 \} f(x,y)=x2+y22(x12)2+14r22+14.f(x, y) = \frac{x^2 + y^2}{2} - \left( x - \frac{1}{2} \right)^2 + \frac{1}{4} \le \frac{r^2}{2} + \frac{1}{4}. If r>12r > \frac{1}{2} then the absolute maximum is r22+14\frac{r^2}{2} + \frac{1}{4}, attained at the points (12,±r214)\left( \frac{1}{2}, \pm\sqrt{r^2 - \frac{1}{4}} \right). Therefore, it is necessary that r12r \le \frac{1}{2} because if r>12r > \frac{1}{2} then the maximum is attained twice.

Suppose now that r1/2r \le 1/2 and that ff attains its maximum on DrD_r at u,vu, v, uvu \ne v. Since gradf(z)gradf(0)r|\operatorname{grad} f(z) - \operatorname{grad} f(0)| \le r, gradf(z)1r>0|\operatorname{grad} f(z)| \ge 1 - r > 0 for all zDrz \in D_r. Hence ff may attain its maximum only at the boundary of DrD_r, so we must have u=v=r|u| = |v| = r and gradf(u)=au\operatorname{grad} f(u) = au and gradf(v)=bv\operatorname{grad} f(v) = bv, where a,b0a, b \ge 0. Since au=gradf(u)au = \operatorname{grad} f(u) and bv=gradf(v)bv = \operatorname{grad} f(v) belong to the disk DD with centre gradf(0)\operatorname{grad} f(0) and radius rr, they do not belong to the interior of DrD_r. Hence gradf(u)gradf(v)=aubvuv|\operatorname{grad} f(u) - \operatorname{grad} f(v)| = |au - bv| \ge |u - v| and this inequality is strict since DDrD \cap D_r contains no more than one point. But this contradicts the assumption that gradf(u)gradf(v)uv|\operatorname{grad} f(u) - \operatorname{grad} f(v)| \le |u - v|. So all r12r \le \frac{1}{2} satisfies the condition.

How the field did

contestants scored
226
average (of 20)
1.11
solved (≥ 80%)
1.8%
near-0 (≤ 10%)
91.6%
discrimination
0.30

Score distribution (field cohort)

Computed on contestants with a meaningful total (field cohort); discrimination is the corrected item–total correlation.

Similar problems

IMC 2006 · Day 1 · P6killeravg 0.3/10 · solved 1% · near-0 95% · disc 0.18
IMC 2001 · Day 2 · P11killeravg 1.4/10 · solved 3% · near-0 63% · disc 0.30
IMC 2021 · Day 1 · P4killeravg 0.1/10 · solved 1% · near-0 98% · disc 0.21
IMC 2012 · Day 1 · P4killeravg 0.2/10 · solved 0% · near-0 95% · disc 0.25