I'm having trouble understanding an assignment regarding the Maurer-Cartan form on orthogonal matrices:

Let $\text{O}(n)\subset\text{GL}(n,\mathbb{R})$ be the matrix Lie group of orthogonal matrices. Given $A\in\text{O}(n)$ let $A_i$ denote the $i$-th column vector. Writing $A=(A_1,\ldots A_n)$, we see that the elements $A\in\text{O}(n)$ are precisely the orthonormal bases $\{A_1,\ldots A_n\}$ of $\mathbb{R}^n$. Let $\pi_i:\text{SO}(n)\to\mathbb{R}^n$ be the map given by $\pi_i(A)=A_i$. Define 1-forms $\omega^i_j\in\Omega^1(\text{O}(n))$ as follows: given $v\in T_A\text{O}(n)$ $$ d\pi_i(v)=:\omega^j_i(v)A_j $$

The assignment then splits to 4 parts which eventually lead to one concluding that $\omega=(\omega^i_j)$ is in fact the Maurer-Cartan form on $\text{O}(n)$. I can go into more detail with this if the added context helps to understand my issue.

This is a homework assigment, so I would prefer no solutions. I'm just having trouble understanding the definition of $\omega=(\omega^i_j)$. A couple questions:

  • Shouldn't the domain of $\pi_i$ be $\text{O}(n)$ instead of $\text{SO}(n)$? Is this a typo?
  • What is $\omega^j_i(v)$? it seems apparent to me that $\omega^j_i(v)$ is just a scalar, but I know that, regarding manifolds, the "$j$'' could imply that one sum's along the terms $\omega^j_i(v)A_j$, with the Einstein convention.
  • Lastly, if $\omega^j_i(v)$ were just a scalar, I believe that $j$ must equivalent to $i$, since it's evident to me that $d\pi_i(v)$ simply projects vectors in $\text{O}(n)$ onto their $i$th column. But then $\omega^j_i$ would not be defined for $j\neq i$.

It would be fantastic if I could get some help clarifying, and I would be glad to give more detail on the question if needed. I feel like half the work in differential geometry is understanding the notation.

Also, are there any texts which cover the Maurer-Cartan form thoroughly? Our texts (Lee - Introduction to Smooth Manifolds/D.Carmo - Differential Forms and Applications/Spivak - A Comprehensive Introduction to Differential Geometry) don't seem to cover it and I would really like a reference text.

Thanks in advance!


Yes, the domain should be $O(n)$.

Each $\omega_i^j$ is a one-form, so it sends a single argument to a number. The argument is a vector in some space (in this case, $T_A(O(n))$).

The formula given does include an implied summation over $j$, I believe. Otherwise it'd be bound in the left side and un-bound in the right, and that makes no sense. Also, your conjecture about $j$ is part right: it's an index that over the same values as $i$.

Alas, I have no recommendation for a good book that discusses M-C forms. Frankly, I've never really understood them myself. Fortunately, I've never needed to. :)

Answer to followup question: You ask "isn't $\omega_i^j = 0$ for $j \ne i$?", and I don't know. But I figure I can work it out with an example. Let's look at $SO(2)$, which consists of all matrices $$ U(t) = \begin{bmatrix} \cos t & -\sin t \\ \sin t & \cos t \end{bmatrix} $$

Since $t \mapsto U(t)$ defines a curve in this 1-dimensional group, we can take its derivative at $t = 0$ to find a nice example tangent vector at $A = U(0) = I$. I get $$ U'(0) = \begin{bmatrix} 0 & -1 \\ 1 & 0 \end{bmatrix}. $$

(A slight generalization shows that the tangent space to $SO(n)$ at the identity consists of all skew-symmetric matrices, by the way.)

So now $A = I$ and $v$ is the matrix $U'(0)$ above. What about $\pi_1$ and $\pi_2$? They send a matrix to its first and second columns, respectively. The derivative of $\pi_1$ will therefore send $v$ to its first column, $\begin{bmatrix} 0\\1\end{bmatrix}$. And we need to express this in terms of the first and second column of the identity. I get this: \begin{align} d\pi_1(v) &=\begin{bmatrix} 0\\1\end{bmatrix}\\ &=0 \cdot \begin{bmatrix} 1\\0\end{bmatrix} + 1 \cdot \begin{bmatrix} 0\\1\end{bmatrix} \end{align} from which I conclude that $\omega_1^2 = 1$. A similar computation shows that $\omega_2^1 = -1$, but that $\omega_i^i = 0$ for $i = 1, 2$.

So your conjecture -- that $\omega_i^j$ is zero for $j \ne i$ -- turns out to be exactly wrong. :) And in fact I'll bet it's not too tough to show that the $\omega_i^j$ form a skew-symmetric matrix in general.

  • $\begingroup$ Thank you! That definitely clears up a lot! One more question: If $$ d\pi_i(v)=\omega^1_i(v)A_1+\cdots +\omega^2_i(v)A_n, $$ then wouldn't all $\omega^j_i(v)=0$ for all $j\neq i$? But then we can't have that, because if $\omega^j_i(v)\neq 0$, then $\omega^j_i(v)$ wouldn't take values in the skew-symmetric matrices. Since you said you're not very good with M-C forms, I understand if you're not sure. I just thought I would ask. $\endgroup$ – Blake Apr 4 '15 at 22:09
  • 1
    $\begingroup$ See additional answer. Thanks for asking. I should have worked this out for myself 35 years ago, but never did until now. :( $\endgroup$ – John Hughes Apr 5 '15 at 13:56
  • $\begingroup$ Thank you! I was definitely having a mental block about the former question, but you helped break through it quite nicely. I sincerely appreciate you taking the time out of your day to help me. $\endgroup$ – Blake Apr 5 '15 at 14:45

This is an example of a very general construction for matrix groups. Each group element $g$ is a matrix that depends on on some parameters (the co-ordinates on the group manifold $G$) . Let $dg$ be the matrix whose entries are the exterior derivative of the entries with respect to the parameters, then the left-invariant Muarer-Cartan form is the object $g^{-1}dg$. It is a matrix-valued one form on the group manifold.

Although constructed from elements of the group, the matrices $g^{-1}dg$ lie in the Lie algebra (for example, in the orthogonal group $g$ obeys $g^T=g^{-1}$ while $g^{-1}dg$ evaluated on any vector in $T(G)$ is a skew-symmetric matrix) so we can write $g^{-1}dg = \omega^a \lambda_a$ where $\lambda_a$ form a basis for the Lie algebra. The scalar-valued one forms $\omega_a$ are also refered to as Maurer-Cartan forms and are dual to the Lie algebra: If $L_a$ is the left-invariant vector field on the group manifold corresponding to the basis element $\lambda_a$, then $\omega^a(L_b)=\delta^a_b$ at all points in $G$.


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.