Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

For example, consider the specific question: Given $a_{11},a_{12},a_{21},a_{22}$ does that uniquely determine

$A=\begin{bmatrix} a_{11}&a_{12}&a_{13} \\ a_{21}&a_{22}&a_{23} \\ a_{31}&a_{32}&a_{33} \end{bmatrix}$

where $A\in SO(3)$.

share|cite|improve this question
Both $\begin{pmatrix}1&0&0\\0&1&0\\0&0&\pm1\end{pmatrix}$ are orthogonal, so the matrix is not determined uniquely. – Martin Sleziak Jun 9 '12 at 4:41
@MartinSleziak With $-1$ in the lower right corner, the determinant is $-1$ and hence we do not have an element of $SO$. – anon Jun 9 '12 at 4:44
I see. I was thinking $O(3)$ instead of $SO(3)$. The title only mentioned orthogonal matrices - I should have read the body of the question more carefully. – Martin Sleziak Jun 9 '12 at 4:46
I changed the headline accordingly. I assumed this is okay with the OP. – Hauke Strasdat Jun 9 '12 at 10:34
@MartinSleziak, unless the $(n-1) \times (n-1)$ block is already orthogonal, we can negate both the final row and the final column to get a second element of $SO_n.$ In this process, $a_{nn}$ remains the same. – Will Jagy Jun 9 '12 at 16:05
up vote 4 down vote accepted

Hint 1: The sum of the squares of each column is $1$.

Hint 2: The third column is $\pm$ the cross product of the first two.

Hint 1 is immediately applicable to $n\times n$ matrices.

There is an $n$-dimensional analog of the cross product that extends Hint 2.

share|cite|improve this answer
For the last point: given $n-1$ vectors $v_1,\cdots,v_{n-1}$, we can pick a $w$ from the orthogonal complement of the generated subspace, ie $w\in\langle v_1,\cdots,v_{n-1}\rangle^\perp$, such that $\|w\|$ is equal to the magnitude of the Grammian determinant of the $v_i$'s. – anon Jun 9 '12 at 4:38
@anon: if the vectors $v_1,\dots,v_{n-1}$ are orthonormal, then yes, let $v_n$ be column $n$ of the cofactor matrix (which does not depend on column $n$). – robjohn Jun 9 '12 at 4:52
@anon: oh, duh! You were making a statement, not asking a question. :-) – robjohn Jun 9 '12 at 5:04
Ok I see that if $a_{11}^2+a_{12}^2 \neq 1$ then there are two choices for $a_{13}$. Each such value uniquely determines $a_{23}$ according to column 1 and column 2 being orthogonal. Assuming this value of $a_{23}$ is consistent with $a_{12}^2+a_{22}^2+a_{32}= 1$, then there is a 3rd column that ensures $\det A = 1$ and its negative which gives $\det A = -1$ – user782220 Jun 9 '12 at 5:35

Note that if $A \in SO_n(\mathbb R)$ with

$$ A \; = \; \left( \begin{array}{rr} E & F \\ G & H \end{array} \right) , $$
and submatrices $F,G$ rectangular and $E,H$ both perfectly square, then $$ \det E = \det H. $$ I'm just saying. In your case, if you have the upper left square, it turns out that $$ a_{33} = \; a_{11} a_{22} \; - \; a_{12} a_{21}. $$ Put another way, if you are given the upper left block and complete it, another choice for $A \in SO_n(\mathbb R)$ is to negate column $n$ and row $n,$ which means that $a_{nn}$ is negated twice and actually stays put.

Proof: $$ \left( \begin{array}{cc} E & F \\ 0 & I \end{array} \right) \left( \begin{array}{cc} E^t & G^t \\ F^t & H^t \end{array} \right) = \left( \begin{array}{cc} I & 0 \\ F^t & H^t \end{array} \right). $$ Note that this is more general as both $E,H$ are allowed to have size larger than one, if $n > 3$ anyway.

share|cite|improve this answer

Your Answer


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.