Suppose $f_1,f_2,\ldots $ is a sequence of convex functions that converges to a continuous convex $f$. Let $x_1^*,x_2^*$ be their respective (not necessarily unique) minima, and let y be a minima of $f$ (once again need not be unique). Can we prove that there exists a version of $x_1^*,x_2^*,\ldots$ such that $x_n^*\rightarrow y$ ?


No; here's a counterexample: let $f = 0$ and consider the minimizer $y = 0.$ Then you can construct convex functions which converge to $0$ pointwise but whose minima are always moving away from $y =0,$ e.g. $f_n(x) = (x - n)^2/n^n.$


No. Let $f_n=x^2/n$ for $n$ odd and $(x-1)^2/n$ for $n$ even. Then $x_n^*$ is an alternating sequence of $1$s and $0$s, which does not converge to anything. But $f_n$ converges pointwise to $f=0$.

We can modify this to make the convergence uniform, by using an absolute value instead of a square, or to make $f$ nonconstant, by adding $\max(|x-1/2|,1)$ to $f_n$.

If $f$ has a unique minimum, the statement is true. Let $a$ be the $\lim\inf$ of $x_n^*$ and $b$ be the $\lim\sup$. Let $y$ be the unique minimum of $f$. Assume $a< y$. Clearly $f( (a+y)/2) > f(y)$. By convexity, $f(a)> f((a+y)/2)$. So for $n$ sufficiently large, $f_n(a)> f_n((a+y)/2)> f_n(y)$. But this implies that the minimum of $f_n$ is greater than $(a+y)/2$. So the $\lim \inf$ of $x_n^*$ is greater than or equal to $(a+y)/2$, which is greater than $a$. This is a contradiction so $a\geq y$. By symmetry $b\leq y$. Hence because $a \leq b$, $a=b=y$ and $y$ is the limit.

  • $\begingroup$ Thanks Will. The last paragraph is unclear to me. To start did you mean if f has a unique minima? $\endgroup$ – gmravi2003 May 11 '12 at 22:06
  • $\begingroup$ Yes, I did. What else is confusing? $\endgroup$ – Will Sawin May 12 '12 at 4:46
  • $\begingroup$ It seems that there should be $(x-1)^2/n$ instead of $(x_1)^2/n$. $\endgroup$ – Alex Ravsky Jan 19 '15 at 5:27
  • $\begingroup$ Why $x_n^*\leq a$ occurs infinitely many times? $\endgroup$ – Alex Ravsky Jan 19 '15 at 5:40
  • $\begingroup$ @AlexRavsky fixed. $\endgroup$ – Will Sawin Jan 19 '15 at 14:23

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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