4
$\begingroup$

A symmetric function is a function $f:\mathbb R^n\to \mathbb R$ such that $f(x_1,\ldots,x_n)=f(\sigma(x_1,\ldots,x_n))$ for every permutation $\sigma\in S_n.$

The most commonly encountered symmetric functions are polynomial functions.

Question 1: Is it true that, given a symmetric function $f:\mathbb R^n\to \mathbb R,$ there exist symmetric polynomials $p_1,\ldots,p_m:\mathbb R^n\to \mathbb R$ and functions $h:\mathbb R^m\to \mathbb R$ and $g_1,\ldots,g_n:\mathbb R\to \mathbb R$ such that $f$ can be factorized as $$f(x_1,\ldots,x_n)=h(p_1(g_1(x_1),\ldots, g_1(x_n)),p_2(g_2(x_1),\ldots, g_2(x_n)),\ldots,p_m(g_m(x_1),\ldots, g_m(x_n)))?$$

Question 2: If $f$ has some regularity, e.g., continuous, $C^\infty$ or analytic, can we find the above $g_i$ and $h$ to be also regular?

To illustrate the idea behind the question I furnish an example.

If $f:\mathbb R^2\to \mathbb R$ with $$f(x,y)=(x^2+y^2)\ln(\frac{xy}{x+y})+\sin(x)+\sin(y),$$ we have $m=4$, $p_1(x,y)=x^2+y^2,$ $p_2(x,y)=xy$, $p_3(x,y)=x+y$, $p_4(x,y)=x+y$, $g_1,g_2,g_3=\operatorname{id}_{\mathbb R}$, $g_4(x)=\sin(x)$ and $h(x,y,z,t)=x\ln(\frac{y}{t})+t$.

Notice that Question 1 seems to be true thanks to an argument similar to that in the answer of Qi Zhu to Can every symmetric function be written as some function of a sum?.

Question 3: does this argument works here?

$\endgroup$

2 Answers 2

2
$\begingroup$

$\newcommand\R{\mathbb R}$For any $(x_1,\dots,x_n)\in\R^n$,
$$P(x):=\prod _{i=1}^{n}(x-x_{i})=\sum _{k=0}^{n}(-1)^{k}e_k(x_1,\dots,x_n)x^{n-k},$$ where the $e_k$'s are the elementary symmetric polynomials in $x_1,\dots,x_n$. Therefore and because "the roots of a polynomial are determined by its coefficients", there is a function $g\colon\R^m\to[\R^n]$ such that $$[x_1,\dots,x_n]=g(e_0(x_1,\dots,x_n),\dots,e_n(x_1,\dots,x_n))$$ for all $(x_1,\dots,x_n)\in\R^n$, where $m:=n+1$, $$[x_1,\dots,x_n]:=\{(x_{\pi(1)},\dots,x_{\pi(n)})\colon\pi\in S_n\},$$ $$[\R^n]:=\{[x_1,\dots,x_n]\colon(x_1,\dots,x_n)\in\R^n\},$$ and $S_n$ is the set of all permutations of the set $\{1,\dots,n\}$.

On the other hand, if a function $f\colon\R^n\to\R$ is symmetric, then there clearly exists a function $u\colon[\R^n]\to\R$ such that $$f(x_1,\dots,x_n)=u([x_1,\dots,x_n])$$ for all $(x_1,\dots,x_n)\in\R^n$.

Thus, $$f(x_1,\dots,x_n)=h(e_0(x_1,\dots,x_n),\dots,e_n(x_1,\dots,x_n)) \tag{1}\label{1}$$ for all $(x_1,\dots,x_n)\in\R^n$, where $h:=u\circ g$. This answers your Question 1. $\quad\Box$


As an illustration of the previous consideration, here is the particular case of identity \eqref{1} for the symmetric function $f\colon\R^2\to\R$ given by the formula $f(x,y):=x\sin y+y\sin x$:
$$x\sin y+y\sin x \\ =\Big(-\frac{p(x,y)}2+\sqrt{\frac{p(x,y)^2}4-q(x,y)}\Big) \\ \times\sin\Big(-\frac{p(x,y)}2-\sqrt{\frac{p(x,y)^2}4-q(x,y)}\Big) \\ +\Big(-\frac{p(x,y)}2-\sqrt{\frac{p(x,y)^2}4-q(x,y)}\Big) \\ \times\sin\Big(-\frac{p(x,y)}2+\sqrt{\frac{p(x,y)^2}4-q(x,y)}\Big) \\ $$ for all $(x,y)\in\R^2$, where $p(x,y):=x+y$ and $q(x,y):=xy$ are elementary symmetric polynomials in $x,y$.


It is now an easy exercise to show that $h$ is continuous if $f$ is continuous. Hints: (i) note that the polynomial $P$ is monic and use a bound on the roots of a monic polynomial in terms of the bounds on its coefficients; (ii) using compactness and the continuity of the value of a polynomial with respect to the coefficients and the argument of the polynomial, show that the map $g$ is continuous; (iii) the map $f\mapsto u$ is clearly continuous in any reasonable topology; (iv) so, $h=u\circ g$ is continuous.

$\endgroup$
6
  • 1
    $\begingroup$ The definition of the function $g$ is missing some steps - first, I think you want the target to be $[\mathbb R]$ and not $\mathbb R$, and second, some kind of extension / interpolation is required since not every polynomial with real coefficients has real roots. $\endgroup$
    – Will Sawin
    Commented Feb 24, 2023 at 17:12
  • $\begingroup$ @WillSawin : Thank you for your comment. (i) I have fixed the codomain. (ii) I think it's OK here: even though not every polynomial with real coefficients has real roots, all the roots of the polynomial $\sum _{k=0}^{n}(-1)^{k}e_k(x_1,\dots,x_n)x^{n-k}=\prod _{i=1}^{n}(x-x_{i})$ for real $x_1,\dots,x_n$ are real. $\endgroup$ Commented Feb 24, 2023 at 17:32
  • 1
    $\begingroup$ (ii) Fair - since you're not claiming continuity, I guess it is obvious that a function $g$ exists, but it's worth pointing out (if only in a comment), that $g$ is not unique, whereas in the complex analogue it would be unique. $\endgroup$
    – Will Sawin
    Commented Feb 24, 2023 at 18:06
  • $\begingroup$ Should $[\mathbb R]^n$ be $[\mathbb R^n]$? \\ How would your formula for $x\sin(y) + y\sin(x)$ change if $\sin$ were replaced by $\cos$? You couldn't use exactly the same formula, I think, since the one you've written seems to rely on both $(x, y) \mapsto x$ and $(x, y) \mapsto \sin(y)$ being odd functions. $\endgroup$
    – LSpice
    Commented Feb 24, 2023 at 18:20
  • 2
    $\begingroup$ @LSpice : Thank you for your comment. (i) I have fixed the typo with $[\mathbb R^n]$. (ii) No, I did not use any parity considerations. Any expression of the form $v(x)w(y)+v(y)w(x)$ in place of $x\sin y+y\sin x$ will similarly do, by (1). $\endgroup$ Commented Feb 24, 2023 at 18:30
2
$\begingroup$

Q1 and Q2 both have positive answers. I believe these are established results in the literature. Have a look at the papers of Glaeser and Schwarz that I referenced in this answer.

$\endgroup$
1
  • $\begingroup$ Thanks for these references, I will have a look at them. $\endgroup$ Commented Feb 25, 2023 at 12:14

You must log in to answer this question.

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