Let $k\ge 2$, and let $f(x):[0,1]\to [0,1]$ have $k$ continuous derivatives. Define the $C^k$ norm of $f$ as— $$||f||_{C^k} = \max_{0\le i\le k} \max_{0\le x\le 1} |f^{(i)}(x)|.$$

Now, given a sequence of $n$ items $(p_n, v_n)$ and given that $f(p_n) = v_n$ for each $n$, what is a tight upper bound (or, if practical, even the exact value) on $\inf ||f||_{C^k}$ over all functions mapping $[0, 1]$ to $[0, 1]$ with $k$ continuous derivatives?

Obviously, $\inf ||f||_{C^k}$ is at least $\max |f|$ and at least $|\frac{v_{n+1}-v_n}{p_{n+1}-p_n}|$, but is greater in general.

Specific Cases

I would like you to answer this question for the case where:

  • $k$ is 3, 4, 5, or 6.
  • $f(0) = f(1) = 0$.
  • Either:
    • $f(1/2)$ is known.
    • $f(1/4)$, $f(1/2)$, and $f(3/4)$ are known.

If your answer can be easily adapted to solve a more general case, that would be nice to have.


Suppose $f(x)=2x(1-x)$. For $f$ and its first three derivatives:

  • $\max |f| = 1/2.$
  • $\max |f\prime| = 2.$
  • $\max |f^{(2)}| = 4.$
  • $\max |f^{(3)}| = 0.$

Thus, $||f||_{C^3} = 4$. Now suppose there is a $C^3$ function $g$ that is unknown except that it maps [0, 1] to [0, 1] and that $g(0) = 0$, $g(1/2) = 1/2$, and $g(1) = 0$. Then obviously the $C^3$ norm must be at least 1/2. But is 4 the lowest $C^k$ norm that $g$ can have? Or can it be even lower?

This illustrates the problem of finding a tight upper bound (or an exact value) on $\inf ||f||_{C^k}$ given a set of points that interpolate $f$. I have not been able to find existing results on this matter, not even on the subject of "derivative bounds" or "bounds on derivatives". A related result is Bernstein's inequality used to find derivative bounds on polynomials (but not necessarily the tightest possible), but I don't see how or whether it can help in the problem at hand, which involves a broader class of functions than polynomials.


The motivation is to give evidence on a conjecture on a scheme to build polynomials that converge from above and below to a function and meet "consistency" requirements on their coefficients.

See also the following:


1 Answer 1


Fefferman has studied this problem, reasonably recently. I think this reference gives an overview https://www.ams.org/journals/bull/2009-46-02/S0273-0979-08-01240-8/S0273-0979-08-01240-8.pdf

  • $\begingroup$ Now that appears rather non-trivial. If you can, you should expand your answer to explain the special case when $f(0)=f(1)=0$, and $f(1/2) \in (0, 1)$, and $k$ is 3 or 4, since that is currently the case that interests me the most. $\endgroup$
    – Peter O.
    Aug 8, 2022 at 20:08
  • $\begingroup$ If you have a really specific question of that form then maybe ask it separately or edit the main question to just ask about the specific case...? $\endgroup$
    – T_M
    Aug 8, 2022 at 21:38
  • $\begingroup$ Edited to indicate the specific cases. $\endgroup$
    – Peter O.
    Aug 8, 2022 at 22:06

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .