Take the 2-minute tour ×
Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. It's 100% free, no registration required.

I'm trying to re-write $v (u x)$ where $v$ and $u$ are row vectors and $x$ is a column vector as some expression $M x$ (or $\bar{v}x$, etc.). The motivation is because I'm trying to compute the gradient (and $\nabla^{'}_x [ v (u x) ] \cdot x$ should yield a row vector).

Intuitively, since the expression $v (ux)$ can be re-written as

$[v_1 (ux), v_2(ux), \cdots, v_n(ux)]$

(where $v_i$ is a scalar), this quantity can be computed. However, doing this and substituting into a larger expression is pretty obnoxious and I'd like to do it using standard matrix operations (non element-wise) if possible.

I know this has something to do with tensors, but I don't understand how it all fits together. Any help is appreciated.

share|improve this question

1 Answer 1

up vote 1 down vote accepted

If by $Mx$ you mean $M$ is a matrix, then clearly you cannot write $v(ux)$ as $Mx$. Why? A matrix represents a linear operation that sends column vectors to column vectors. But your operation sends the column vector $x$ to a row vector. What you can do is write

$$ v(ux) = \left((v^T\otimes u)x\right)^T $$

where $v^T\otimes u$ is the tensor product of the column vector $v^T$ and the row vector $u$, and can be represented by the matrix $M$ with entries $M_{ij} = v_iu_j$. So you multiply $x$ by this matrix and then take the transpose.

Then taking the gradient is as simple as taking the gradient of $(Mx)^T$.

share|improve this answer

Your Answer

 
discard

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.