29
$\begingroup$

I've seen the statement "The matrix product of two orthogonal matrices is another orthogonal matrix. " on Wolfram's website but haven't seen any proof online as to why this is true. By orthogonal matrix, I mean an $n \times n$ matrix with orthonormal columns. I was working on a problem to show whether $Q^3$ is an orthogonal matrix (where $Q$ is orthogonal matrix), but I think understanding this general case would probably solve that.

$\endgroup$
52
$\begingroup$

If $$Q^TQ = I$$ $$R^TR = I,$$ then $$(QR)^T(QR) = (R^TQ^T)(QR) = R^T(Q^TQ)R = R^TR = I.$$ Of course, this can be extended to $n$ many matrices inductively.

$\endgroup$
5
  • $\begingroup$ Very succinct answer, thank you. But you used commutative property in the proof, and in general matrices don't obey multiplicative commutativity. Could you please elaborate on that? $\endgroup$ – Kashmiri Nov 16 '20 at 5:25
  • $\begingroup$ @YasirSadiq hm? Nowhere did I use commutativity. I did use associativity and the fact that $(AB)^T = B^TA^T$ -- I added an extra expression to clarify. Both associativity and the transpose-of-product-is-reverse-product-of-transposes can be verified by writing out the elements of the matrices. $\endgroup$ – user217285 Nov 17 '20 at 5:01
  • $\begingroup$ @YasirSadiq You should learn what the commutative property is before littering all the answers with the same erroneous belief. $\endgroup$ – Ted Shifrin Nov 17 '20 at 5:07
  • $\begingroup$ @user217285Yes it's clear now ,Thank you so much. $\endgroup$ – Kashmiri Nov 17 '20 at 5:58
  • $\begingroup$ @Ted Shifrin I apologize shall I delete all my comments? $\endgroup$ – Kashmiri Nov 17 '20 at 5:58
33
$\begingroup$

As an alternative to the other fine answers, here's a more geometric viewpoint:

Orthogonal matrices correspond to linear transformations that preserve the length of vectors (isometries). And the composition of two isometries $F$ and $G$ is obviously also an isometry.

(Proof: For all vectors $x$, the vector $F(x)$ has the same length as $x$ since $F$ is an isometry, and $G(F(x))$ has the same length as $F(x)$ since $G$ is an isometry; hence $G(F(x))$ has the same length as $x$.)

$\endgroup$
2
  • 3
    $\begingroup$ Intuitively, I think of it as a "hyperdimensional rotation". Is this roughly an apt expression? $\endgroup$ – MackTuesday Sep 1 '15 at 17:35
  • 4
    $\begingroup$ @MackTuesday With one catch; It is not only rotation but also reflection that can preserve vector lengths. I prefer the intuitive notion presented by this answer: There is no way, using pure rotation and reflection, to change the length of a vector. If I take a stick and swing it around or look at it in a (flat) mirror, no matter what I do the stick remains the same size. $\endgroup$ – Iwillnotexist Idonotexist Sep 2 '15 at 1:08
9
$\begingroup$

Let the orthogonal matrices be known as $M$ and $N$. By the definition of orthogonal matrices, $M \cdot N$ must be orthogonal, as

$$(M \cdot N)^T \cdot (M\cdot N) = N^T \cdot M^T \cdot M \cdot N = N^T \cdot N = I $$ $$(M \cdot N) \cdot (M\cdot N)^T = M \cdot N \cdot N^T \cdot M^T = M \cdot M^T = I $$

$\endgroup$
2
  • 1
    $\begingroup$ Nice answer. But you used commutative property in the proof, and in general matrices don't obey multiplicative commutativity. Could you please elaborate on that $\endgroup$ – Kashmiri Nov 16 '20 at 5:51
  • $\begingroup$ @YasirSadiq This doesn't actually use commutativity. One property of taking the transpose of a product of matrices is that the order of those matrix factors is reversed, in addition to them individually being transposed. E.g. (A * B)^T = B^T * A^T. This answer just takes advantage of that property. $\endgroup$ – Starkle Nov 18 '20 at 0:59
8
$\begingroup$

Let $A$ and $B$ be two orthogonal matrices. You have $$AA^T = A^TA = I$$ and $$BB^T = B^TB =I.$$

So, we have $$(AB)^T(AB) = B^TA^TAB = I.$$

$\endgroup$
2
  • 1
    $\begingroup$ +1 but You used commutative property in the proof, and in general matrices don't obey multiplicative commutativity. Could you please elaborate on that $\endgroup$ – Kashmiri Nov 16 '20 at 5:51
  • $\begingroup$ @YasirSadiq (copied from my other comment) This doesn't actually use commutativity. One property of taking the transpose of a product of matrices is that the order of those matrix factors is reversed, in addition to them individually being transposed. E.g. (A * B)^T = B^T * A^T. This answer just takes advantage of that property. $\endgroup$ – Starkle Nov 18 '20 at 1:00

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.