2
$\begingroup$

Let $\boldsymbol{c}_1, ..., \boldsymbol{c}_n$ be $n$ orthonormal, $m$-dimensional complex vectors, with $\boldsymbol{c}_i = (c_{i,1}, ..., c_{i,m})$. Consider the following polynomial in $x_1,..., x_m$: $$ (c_{11} x_1 + c_{12} x_2 + ... + c_{1m} x_m) (c_{21} x_1 + c_{22} x_2 + ... + c_{2m} x_m) \cdots (c_{n1} x_1 + c_{n2} x_2 + ... + c_{nm} x_m) = \prod_{i=1}^n \sum_{j=1}^m c_{ij} x_j. \tag{A} $$ Is there a way to control the resulting sums of terms containing all different $x_i$? With this I mean, given a target normalised complex vector $\boldsymbol v$ of length $\binom{m}{n}$, can I find a set of orthogonal $\boldsymbol{c}_i$ such that the $n$-linear terms of (A), renormalised, give $\boldsymbol v$?

As an example, in the $m=3, n=2$ case, (A) can be written as $$ (c_{11} c_{21} x_1^2 + c_{12} c_{22} x_2^2 + c_{13} c_{23} x_3^2) + (c_{11} c_{22} + c_{12} c_{21}) x_1 x_2 + (c_{11} c_{23} + c_{13} c_{21}) x_1 x_3 + (c_{12} c_{23} + c_{13} c_{22}) x_2 x_3, $$ so that the problem amounts to solving the following nonlinear system: \begin{align} c_{11} c_{22} + c_{12} c_{21} &= N v_1, \\ c_{11} c_{23} + c_{13} c_{21} &= N v_2, \\ c_{12} c_{23} + c_{13} c_{22} &= N v_3, \end{align} for some normalisation constant $N$ and $\boldsymbol c_1, \boldsymbol c_2$ constrained to be orthonormal.

In the general case this is clearly related to the problem of computing permanents of submatrices of a matrix $C$ having the $\boldsymbol c_i$ as row/columns.

Has this kind of thing been studied? Is there relevant literature I can look at?

$\endgroup$
  • $\begingroup$ surely not always: you have $(n-1)+(n-2)+\dots+(n-m+1)$ degrees of freedom, and want to control $\binom{n}m-1$ parameters $\endgroup$ – Fedor Petrov Jun 25 '17 at 9:17
2
$\begingroup$

Yes something close to your question has been studied before see these MO questions:

  1. Grassmann-Plücker relations for permanents
  2. which homogeneous polynomials split into linear factors?

Basically there is a rather complicated system of algebraic equations of degree $n+1$ called the Brill-Gordan equations for the coefficients of your polynomial which tell you if it is a product of linear factors. You would need more work to eliminate the coefficients of the non-multilinear terms in order to get a system of equation for the multilinear terms only. I don't know what the effect of imposing orthogonality would be.

| cite | improve this answer | |
$\endgroup$

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.