Let $g:[0,\infty] \to [0,\infty]$ be a smooth strictly increasing function satisfying $g(0)=0$ and $g^{(k)}(0)=0$ for every natural $k$.

Is $\sqrt g$ is infinitely (right) differentiable at $x=0$?

I know that $\sqrt g \in C^1$ at zero*, and that in complete generality, one cannot expect for $\sqrt g$ to be even $C^2$. However, in the counter-example given in the linked question, $g$ was not monotonic.

Does this additional assumption of (strict) monotonicity save us? I tried to look at the literature, but did not find a treatment of this particular case.

*The proof that $\sqrt g \in C^1$ goes by rewriting $g(x)=x^2h(x)$ where $h \ge 0$ is smooth (this is possible since $g(0)=g'(0)=0$).


As pointed out by Igor Rivin, it seems that theorem 2.2 (on page 639) here (pdf) does the job. It states that any square root of $f$ "precised up to order $m$" is of class $C^m$. (The definition of a "square root precised up to order $m$" is Definition 1.1 on page 636).

This certainly settles the issue. However, I think it would still be nice to find a simpler approach, since here we assume much more-the strict monotonicity is a much stronger assumption than those assumed in the paper.


If we assume that $g''>0$ in a neighbourhood of zero (which implies that $g'>0$), then $\sqrt g \in C^2$. (details below).

I think that there is a chance for smoothness under the additional assumption that $g^{(k)}>0$ in a neighbourhood of zero for every $k$, but I am not sure. The calculations become quite messy even when trying to establish $\sqrt g \in C^3$.

A proof $\sqrt g \in C^2$ when $g',g''>0$ near zero: (We use these assumptions when applying L'Hôpital's rule).

$$\sqrt{g}'' = \frac{g''}{2\sqrt{g}} - \frac{(g')^2}{4g^{3/2}}.$$

Thus it is enough to prove that $(g'')^2/g\to 0$ and $(g')^4/g^3\to 0$.

$$ \lim_{x\to 0^+} \frac{(g'')^2}{g} = \lim_{x\to 0^+} 2\frac{g''g^{(3)}}{g'} = \lim_{x\to 0^+} 2\frac{g''g^{(4)}+(g^{(3)})^2}{g''} = 0, $$ where in the last equality we applied $\frac{(h')^2}{h}\to 0$ above for $h=g''$.

$$ \lim_{x\to 0^+} \frac{(g')^4}{g^3} = \lim_{x\to 0^+} \frac{4(g')^2g''}{3g^2} = \lim_{x\to 0^+} \frac{8(g'')^2 + 4g' g^{(3)}}{6g} = \lim_{x\to 0^+} \frac{2g' g^{(3)}}{3g} = \lim_{x\to 0^+} \left(\frac{2g^{(4)}}{3} + \frac{2g''g^{(3)}}{3g'}\right)=\lim_{x\to 0^+} \frac{2g''g^{(3)}}{3g'} = \lim_{x\to 0^+} \frac{2g^{(4)}}{3}+\frac{2(g^{(3)})^2}{3g''} = 0,$$

where in the first row we used the first calculation, and in the second we again applied $\frac{(h')^2}{h}\to 0$ to $h=g''$.

  • 1
    $\begingroup$ Just a comment: if all derivatives of $g$ are nonegative in a common neighborhood of zero, then $g$ can be expanded in a power series, by Bernstein theorem, and would be equal to 0. However, the question changes if the neighborhhod is allowed to depend on the derivative. $\endgroup$ May 10, 2020 at 7:56

2 Answers 2


The answer is yes, by the results of

Bony, Jean-Michel; Colombini, Ferruccio; Pernazza, Ludovico, On square roots of class (C^m) of nonnegative functions of one variable, Ann. Sc. Norm. Super. Pisa, Cl. Sci. (5) 9, No. 3, 635-644 (2010). ZBL1207.26004.

Here is the math review:

enter image description here

Clearly the condition is fulfilled in the OP (for any $m$).


Extend the domain of the function $g$ to $\mathbb R$ by letting $g(x):=0$ for real $x<0$. The resulting function, which we shall still denote by $g$, is $C^\infty$ on $\mathbb R$.

Theorem 3.5 on page 144 implies that a nonnegative $C^4$ function $f$ on $\mathbb R$ has a $C^2$ square root if for any minimum $x_0$ of $f$ we have $f(x_0)=0$.

This latter condition obviously holds for our function $g$ in place of $f$ -- because $g$ is strictly increasing on $[0,\infty)$ and thus has no minima in $(0,\infty)$, and $g=0$ on $(-\infty,0]$.

Therefore, we can conclude that $\sqrt g$ is $C^2$ on $\mathbb R$, even without assuming that $g''>0$ in a neighborhood of zero.

Yet, this conclusion falls short of your main goal, to show that $\sqrt g$ is $C^\infty$. Looking at the proof of the mentioned Theorem 3.5, this task may be too big for a usual MO answer and may require a full-blown paper.


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.