1
$\begingroup$

For a homogeneous degree $d$ polynomial $P$, the symmetric or Waring rank $W(P)$ is the minimum $r$ such that $P = \sum_{j=1}^r l_j^d$, where $l_j$s are linear forms. Now, is the Waring rank sub-multiplicative, i.e. for two homogeneous degree $d$ polynomials $P$ and $Q$, $W(P Q) \leq W(P) \, W(Q)$?

$\endgroup$
4
$\begingroup$

You asked:

Is $W(P \otimes Q) \leq W(P) W(Q)$ for two homogeneous polynomials $P$ and $Q$?

First, I think you have to be a little bit careful about the difference between tensor product and multiplication of polynomials. If you are just using the ordinary tensor product, then the resulting $P \otimes Q$ is no longer a symmetric tensor. For example, if $P = P(x,y) = xy$ and $Q = Q(z,w) = zw$, then as tensors, $P = x \otimes y + y \otimes x$, $Q = z \otimes w + w \otimes z$, and $$ P \otimes Q = x \otimes y \otimes z \otimes w + \dotsb + y \otimes x \otimes w \otimes z, $$ a sum of $4$ terms. But it is not a symmetric tensor because it is missing terms such as $x \otimes w \otimes y \otimes z$.

So perhaps a better question would be:

Is $W(PQ) \leq W(P) W(Q)$ for two homogeneous polynomials $P$ and $Q$?

Here the answer is no. For a simple example, let's say... $P=xy$, $Q=zw$. Then $W(P)=W(Q)=2$, but $W(PQ)=W(xyzw)=8$, see for example https://arxiv.org/abs/1104.3648.

And in fact one should not expect $W(PQ) \leq W(P)W(Q)$ anyway. If we write $P=\sum_{i=1}^r \ell_i^d$ and $Q=\sum_{j=1}^s m_j^e$ (where $r=W(P)$, $s=W(Q)$, $\ell_i$ and $m_j$ are linear forms, $d=\deg(P)$, and $e=\deg(Q)$), then we get $$ PQ = \sum \ell_i^d m_j^e, $$ a sum of $rs$ terms. Naively we would expand each $\ell_i^d m_j^e$ into a sum of powers. This needs $W(x^d y^e) = \max(d+1,e+1)$ terms for each term in the above summation. This proves that $$ W(PQ) \leq W(P) W(Q) \max(\deg(P)+1, \deg(Q)+1).\tag{$*$} $$

The inequality can be strict. For example let, once again, $P=xy$ and $Q = zw$, so $W(P)=W(Q)=2$, $\deg(P)=\deg(Q)=2$, and $W(PQ) = 8$, strictly less than the naive $W(P) W(Q) W(\ell^2 m^2) = 2 \cdot 2 \cdot 3 = 12$.

Also equality is possible. For example if $W(P)=W(Q)=1$, and $P=x^d$, $Q=y^e$ for independent variables $x,y$ (as opposed to $P=x^d$, $Q=x^e$ for the same variable), then equality occurs. Off the top of my head, that's the only example I can think of where equality occurs in $(*)$. Slightly more generally say $P=x^d$, $Q=y_1^{a_1} y_2^{a_2} \dotsm y_k^{a_k}$, $1 \leq a_1 \leq \dotsb \leq a_k$, $d \geq \deg(Q) = a_1+\dotsb+a_k$. Then $W(P)=d+1$ and $W(Q)=(a_2+1)\dotsm(a_k+1)$ (see https://arxiv.org/abs/1112.3474). So $(*)$ provides the bound $W(PQ) \leq 1 \cdot W(Q) \cdot (d+1)$. Since $PQ$ is a monomial we know that $W(PQ) = (a_2+1)\dotsm(a_k+1)(d+1)$, which matches the bound.

But if $1 \leq d < a_1 \leq a_2 \leq a_k$ then we are back to strict inequality: $W(PQ) = (a_1+1)\dotsm(a_k+1) = (a_1+1) W(Q)$, which is strictly less than $W(P)W(Q)\max(\deg(P)+1,\deg(Q)+1)$, since that would be $(a_1+\dotsb+a_k)W(Q)$. So $W(P)=1$ or $W(Q)=1$ is certainly not a sufficient condition for equality in $(*)$.

Is it a necessary condition?

Suppose $P$ and $Q$ are homogeneous polynomials such that $$W(PQ) = W(P) W(Q) \max(\deg(P)+1,\deg(Q)+1).$$ Does it follow that $W(P)=1$ or $W(Q)=1$?

My hunch is that the answer is probably negative, but I don't know. I haven't thought about it.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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