1
$\begingroup$

Suppose I have n-1 linearly independent vectors in n dimensional space, and I wish to find a vector perpendicular to all of these. How would I go about doing this? When working in three dimensions, I can just find the cross product of the two given vectors. In 2 dimensions, I would solve the algebraic expression for the dot product. These two methods seem entirely different and so I am not sure if they can somehow be generalised. I could generalise the approach for the 2D case by writing a series of n-1 simultaneous equations which are the dot products of an unknown vector with the given vectors, and u could write this in matrix form, but then I don't know what to do with this...

$\endgroup$
1
$\begingroup$

As you've noted, the problem can be reduced to finding the null space of a matrix. But perhaps to others it is not obvious why this is so.

Let $v_1,\ldots,v_{n-1},x\in\mathbb{R}^n$ be column vectors. Consider: $$ A = \begin{bmatrix}v_1^T\\\vdots\\v_{n-1}^T\end{bmatrix} \;\;\;\implies\;\;\; Ax=\vec{0} =\begin{bmatrix}v_1^T x\\\vdots\\v_{n-1}^T x\end{bmatrix} =\begin{bmatrix}v_1\cdot x\\\vdots\\v_{n-1}\cdot x\end{bmatrix} $$ So, finding the null-space of $ A $ is equivalent to enforcing $v_i\cdot x=0$, meaning $x$ is orthogonal to the rest of the set.

It may also be worth mentioning the Gram-Schmidt procedure, which is a little more general.

Notice that this gives you an easy way to construct an orthonormal vector $v_n$, which is orthogonal to a set of orthonormal vectors $v_1,\ldots,v_{n-1}$. Just choose a random vector $r\in\mathbb{R}^n$, normalize it, and take: $$ v_n = r - \sum_{k=1}^{n-1} \frac{v_k\cdot r}{v_k\cdot v_k}v_k $$ where, if $r$ or $v_n$ turns out to be linearly dependent on one of the $v_i$, just choose a different $r$. (The probability of such an event is extraordinarily low; theoretically, zero in fact.)

$\endgroup$

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.