# A "known" Pythagorean identity in algebra?

Some will recognize this as similar to a question I asked before, but I want to ask it without the trigonometry.

Let $e_k$ be the $k$th-degree elementary symmetric polynomial in $x_1,x_2,x_3,\ldots$. If $k$ is more than the number of $x$s, then $e_k$ is the sum of no terms and is $0$. From one POV, the following Pythagorean identities are as elementary as anything not in the high-school curriculum: $$(e_0+e_2+e_4+\cdots)^2 - (e_1+e_3+e_5+\cdots)^2 = (1-x_1^2)(1-x_2^2)(1-x_3^2)\cdots$$ $$(e_0-e_2+e_4-\cdots)^2 + (e_1-e_3+e_5-\cdots)^2 = (1+x_1^2)(1+x_2^2)(1+x_3^2)\cdots$$ So are these "known" in the sense of being in refereed publications one could cite? And if not, are they "known" in the sense that some people see them mentioned or explicitly used from time to time?

(And if there should happen to be infinitely many $x$s, could this still be considered only algebra by thinking of these as a sort of "formal" series?)

• What's the complete homogeneous symmetric function analogue? I might recognize that more easily. I get the feeling this is standard or at least a routine application of something standard in symmetric function theory. The right hand sides look like the squares of Cauchy kernels. Feb 13 '13 at 0:42

These are both simple corollaries of $$\sum_{k\geq 0} t^ke_k(x_1,x_2,\dots)=\prod_{k\geq 0}(1+tx_k).$$
There is a typo in both your identities. They should read, $$(e_0+e_2+\cdots)^2-(e_1+e_3+\cdots)^2=(e_0+e_1+e_2+\cdots)(e_0-e_1+e_2-\cdots)$$ $$=\prod_{k\geq 0}(1+x_k)\prod_{k\geq 0}(1-x_k)=(1-x_1^2)(1-x_2^2)\cdots,$$ and $$(e_0-e_2+e_4-\cdots)^2+(e_1-e_3+e_5-\cdots)^2=(\mathfrak {Re}[\prod_{k\geq 0}(1+ix_k)])^2+(\mathfrak{Im}[\prod_{k\geq 0}(1+ix_k)])^2$$ $$=(1+x_1^2)(1+x_2^2)\cdots$$
• The odd thing is that when the $x$s are sines and the first identity generalizes $1-\sin^2\theta=\cos^2\theta$, or the $x$s are tangents and the second identity generalizes $1+\tan^2\theta=\sec^2\theta$, the validity of the identities in no way depends on the fact that the circle is parametrized by arc length rather than in some other way, but yet the way I found these identities was by deriving them from an identity one side of which was $\tan(\theta_1+\theta_2+\theta_3+\cdots)$. Feb 13 '13 at 4:01