Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Here is a prototype of the problem I have in mind: Let $P:\mathbb{R}^2\rightarrow\mathbb{R}$ be a strictly convex, nonnegative polynomial such that $P(0,0)=0$. Let $\alpha\geq 0$, and consider the following version of its distribution function (in the sense of, say, harmonic analysis)

$$\lambda_P(\alpha)=|\{(x,y)\in\mathbb{R}^2: P(x,y)<\alpha\}|;$$

here by $|\cdot|$ I mean Lebesgue measure on the plane.

Question. Is $\lambda=\lambda_P$ a smooth ($C^\infty$) function of $\alpha$, for $\alpha>0$?

Thank you.

share|cite|improve this question
Don't you mean $|P(x,y)|<\alpha$? Otherwise $\lambda_P(\alpha)$ is going to be infinite for $\alpha>1$ and $P$ nonconstant. – Alex Becker Jul 5 '12 at 0:53
@Alex: Corrected, thanks. I guess one could equivalently ask about the function $$\lambda_P(\alpha)=|\{(x,y)\in K: |P(x,y)|>\alpha\}|,$$ where $K$ is some compact and convex subset of the plane... – user17240 Jul 5 '12 at 1:04
Since you assume strict convexity, why take the modulus? It would be more natural to consider the set where $P<\alpha$. Also, are you interested in $C^1$ or higher degrees of smoothness? – user31373 Jul 5 '12 at 1:27
@Alex: What polynomial are you taking? $P(x,y)=x^2+y^2$? Then $\lambda_P(\alpha)=\pi (1-\alpha)$ if $\alpha\leq 1$, right? In any case, the point is that I'm only interested in smoothness of $\lambda$ in the interior of its support (hence the requirement $\alpha>0$ in the original question). – user17240 Jul 5 '12 at 1:30
@user17240 Take $P=(x^2+y^2)^{10}-1$: the set $|P|<\alpha$ is the annulus with radii $(1\pm \alpha)^{1/20}$ and the area $\pi(1+ \alpha)^{1/10}-\pi(1- \alpha)^{1/10}$ (the second term disappears when $\alpha>1$). This is not smooth at $\alpha=1$. – user31373 Jul 5 '12 at 1:33
up vote 2 down vote accepted

I think it is indeed $C^\infty$ smooth. Let $r,\theta$ be polar coordinates. For each fixed $\theta$ the function $p_\theta(r)=P(r,\theta)$ is $C^\infty$ smooth on $(0,\infty)$ and has strictly positive derivative. Therefore, the inverse $R_\theta : (0,\infty)\to(0,\infty)$ is also $C^\infty$ smooth. We have $\lambda_P(\alpha)=\frac{1}{2}\int_0^{2\pi}R_\theta^2(\alpha)\,d\theta$, and this can be differentiated with respect to $\alpha$ as many times as we wish.

share|cite|improve this answer

Let's modify the problem slightly, considering instead a $1$-variable strictly convex monic polynomial $P(x)=x^n+a_{n-1}x^{n-1}+\cdots+a_0$ and $\lambda_P(\alpha)=|\{x\in\mathbb R: P(x)<\alpha\}|$. Since $P$ is strictly convex, any line intersects it at most twice including the line $y=\alpha$. In fact, it is easy to see that $P$ has even degree and thus $y=\alpha$ intersects $P$ at exactly two points for $\alpha>M$ the global minimum. Let $m$ be the unique value such that $P(m)=M$. Then we have two functions $f_1(\alpha)< m<f_2(\alpha)$ which give the upper and lower roots of $P(x)-\alpha$. Note that $\gamma_P(\alpha)=f_2(\alpha)-f_1(\alpha)$ so if $f_1,f_2$ are smooth then $\lambda_P$ is. Furthermore, since the coefficients $(a_{n-1},\ldots,a_1,a_0-\alpha)$ of $P(x)-\alpha$ are smooth functions of the roots $(f_1,f_2,r_3,\ldots,r_n)$, it follows from the inverse function theorem that the roots are locally given by smooth functions (I think we can get $C^\infty$) of the coefficients wherever the Jacobian is noninvertible. By the implicit function theorem, if the coefficients vary smoothly with several variables we get that $f_1$ and $f_2$ are locally smooth functions of these variables where the Jacobian is noninvertible.

Generalizing the original problem, let $P(x_1,\ldots,x_k)$ be a strictly convex polynomial in arbitrarily many variables. Let $\chi_P(x_1,\ldots,x_k)$ be the indicator function for the set $\{(x_1,\ldots,x_k):P(x_1,\ldots,x_k)<\alpha\}$. By Fubini's theorem, $$\lambda_P(\alpha)=\int_{\mathbb R^n} \chi_P(x_1,\ldots,x_k)dx_1\cdots dx_k=\int_{\mathbb R}\cdots \int_{\mathbb R} \chi_P(x_1,\ldots,x_k)dx_1\cdots dx_k$$ and note that $\chi_P(x_1,\ldots,x_k)$ is the indicator function of a set $\{x_1: f(x_1)<\alpha\}$ where $f$ is a polynomial with coefficients varying smoothly in the other varaibles. It follows that the roots $f_1,f_2$ are smooth functions of $\alpha,x_2,\ldots,x_k$ except where the relevant Jacobian is noninvertible, which by strict convexity corresponds to points at which $f_1,f_2$ fail to exist and so $f_1,f_2$ are smooth wherever they are defined. Since $$\lambda_P(\alpha)=\int_{\mathbb R}\cdots \int_{\mathbb R} (f_2(\alpha,x_2,\ldots,x_n)-f_1(\alpha,x_2,\ldots,x_n)) dx_2\cdots dx_k$$ it follows by differentiating under the integrals that $\lambda_P$ is smooth.

share|cite|improve this answer

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

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