Let $so(3)$ be the Lie Algebra of $SO(3)$ and $R\in SO(3); \Omega_1,\Omega_2 \in so(3)$ and $\Omega_n = \frac{d}{dt}R_n(t)$ at the point $t=0$. So $\Omega_n$ is the tangent vector of the curve $R_n(t)$ at $t=0$.

To show that the Lie bracket $[\Omega_1,\Omega_2]$ is an element of $so(3)$ my professor wrote:

$$ \tag{1}[\Omega_1,\Omega_2] = \left.\frac{d}{dt}R_1(t)\Omega_2 R_1(t)^{-1}\right|_{t=0} $$

I am trying to understand that, but I cannot see why this shows that the commutator is an element of $so(3)$.

I understand that for this tangential vector to be in $so(3)$, the product $R_1(t)\Omega_2 R_1(t)^{-1}$ has to be a parametrized curve in $SO(3)$.

But I do not see why it is a curve in $SO(3)$.

  • $\begingroup$ I want to be clear about what you are asking :) Do you want to know why the product $R_1(t)\Omega_2R_1(t)^{-1}$ "has to be a parametrized curve" living completely with $SO(3)$. Or, are you asking if it is a parametrized curve" living completely with $SO(3)$, then you can using it in the above formula to calculate $[\Omega_1, \Omega_2]$. $\endgroup$ – muaddib Aug 11 '15 at 22:16
  • $\begingroup$ I want to see the proof that the Lie bracket $[\Omega_1,\Omega_2]$ is an element of $so(3)$. In other words, this is the proof that the tangent space of $SO(3)$ at the unity matrix is a Lie algebra, if you use the commutator as the Lie bracket. My professor wrote the above formula $(1)$ as proof, but I don't understand it. $\endgroup$ – Bass Aug 12 '15 at 8:13
  • $\begingroup$ Edited question for clarification. $\endgroup$ – Bass Aug 12 '15 at 8:16

We want to show that $$ \tag{1}[\Omega_1,\Omega_2] = \left.\frac{d}{dt}R_1(t)\Omega_2 R_1(t)^{-1}\right|_{t=0} $$

We can start doing this by computing: \begin{eqnarray*} \left.\frac{d}{dt}\left(R_1(t)\Omega_2 R_1(t)^{-1}\right)\right|_{t=0} &=& \left.\left[\frac{d}{dt}R_1(t)(\Omega_2 R_1(t)^{-1})\right]\right|_{t=0} + \left.\left[(R_1(t)\Omega_2) \frac{d}{dt}R_1(t)^{-1}\right]\right|_{t=0} \\ &=& \left[\Omega_1\Omega_2 R_1(0)^{-1}\right] + \left.\left[R_1(t)\Omega_2 \frac{d}{dt}R_1(t)^{-1}\right]\right|_{t=0} \\ &=& \left[\Omega_1\Omega_2 R_1(0)^{-1}\right] + \left.\left[R_1(t)\Omega_2 (-R_1(t)^{-1})\left(\frac{d}{dt}R_1(t)\right)R_1(t)^{-1})\right]\right|_{t=0} \\ &=& \left[\Omega_1\Omega_2 R_1(0)^{-1}\right] - \left.\left[R_1(t)\Omega_2 R_1(t)^{-1}\left(\frac{d}{dt}R_1(t)\right)R_1(t)^{-1})\right]\right|_{t=0} \\ &=& \left[\Omega_1\Omega_2 R_1(0)^{-1}\right] - \left[R_1(0)\Omega_2 R_1(0)^{-1}\left.\left(\frac{d}{dt}R_1(t)\right|_{t=0}\right)R_1(0)^{-1})\right] \\ &=& \left[\Omega_1\Omega_2 R_1(0)^{-1}\right] - \left[R_1(0)\Omega_2 R_1(0)^{-1}\Omega_1R_1(0)^{-1})\right] \\ \end{eqnarray*}

See Derivatives of Inverse Matrix for details on the third line's derivation.

There isn't much further we can go without knowing more details about $R_1$. I actually think it might have been defined as $$R_i(t) = e^{\Omega_it}$$ But it's only necessary to require $R_i(0) = I$. Then the final line above gives $\Omega_1\Omega_2 - \Omega_2\Omega_1$.

Update to include additional requirements:

The curve is in $SO(3)$

This is not always true. A Lie group is a subgroup of $GL$ but a Lie algebra always contains the $0$ matrix. It maps to the identity of the Lie group. So if $\Omega_2 = 0$ in the above, then the curve $C(t) = 0$ does not lie in $SO(3)$.

Show $[\Omega_1, \Omega_2]$ is in $\mathfrak{so}(3)$

This is true independently of the above considerations. $\Omega_1, \Omega_2$ are elements of $\mathfrak{so}(3)$ which is closed under addition and multiplication (matrices of the form $X^T = -X$).

  • $\begingroup$ I'm sorry if I didn't express myself clearly. I see that formula (1) is correct, you showed it using the "product rule of differentiation". However, for $[\Omega_1,\Omega_2]$ to be an element of $so(3)$, this is not sufficient. You also need to show that $C(t) = R_1(t)\Omega_2 R_1(t)^{-1}$ is a curve in $SO(3)$ with $C(0)=\mathbb{1}$. $\endgroup$ – Bass Aug 12 '15 at 13:28
  • $\begingroup$ @BastianTreichler - Updated $\endgroup$ – muaddib Aug 12 '15 at 13:43
  • $\begingroup$ Thanks, but your last remark is not true. $\mathfrak{so}(3)$ is not closed under multiplication: $\begin{bmatrix}0 & 0 & 1\\0 & 0 & 0\\-1 & 0 & 0\end{bmatrix}\begin{bmatrix}0 & 1 & 0\\-1 & 0 & 0\\0 & 0 & 0\end{bmatrix} = \begin{bmatrix}0 & 0 & 0\\0 & 0 & 0\\0 & -1 & 0\end{bmatrix}$ The Lie bracket IS the multiplication in $\mathfrak{so}(3)$, and this does not seem trivial. $\endgroup$ – Bass Aug 12 '15 at 14:11
  • $\begingroup$ @BastianTreichler - Yes, by "multiplication", I meant the one for the lie-algebra, namely the bracket. What I am saying is $[\Omega_1, \Omega_2] = \Omega_1\Omega_2 - \Omega_2\Omega_1$ is part of the definition of the Lie Algebra. $\endgroup$ – muaddib Aug 12 '15 at 14:13
  • 1
    $\begingroup$ @BastianTreichler hmm, well once you obtain the formula from the above that $\Omega_1\Omega_2 - \Omega_2\Omega_1$ is the commutator, you can show it is an element of $\mathfrak{so}(3)$ as follows. Elements of $\mathfrak{so}(3)$ are those matrices that satisfy $X^T = -X$. Then $[X, Y]^T = (XY - YX)^T = Y^TX^T - X^TY^T = -[X^T, Y^T] = -[(-X), (-Y)] = -[X, Y]$. $\endgroup$ – muaddib Aug 12 '15 at 14:19

Formula (1) is a bit strange indeed, $R_1(t)\Omega_2 R_1(t)^{-1}$ is not necessarily in $SO(3)$ ($\Omega_2$ could have a null determinant for example).

One way to prove the result, is to consider that (for $t \to 0$):

$$ R_1(t) = I + t \Omega_1 + t^2 W_1 + o(t^2)$$ $$ R_2(t) = I + t \Omega_2 + t^2 W_2 + o(t^2)$$


$$ R_1(t)R_2(t) = I + t (\Omega_1 + \Omega_2) + t^2 (W_1 + W_2 + \Omega_1 \Omega_2) + o(t^2)$$ $$ R_2(t)R_1(t) = I + t (\Omega_1 + \Omega_2) + t^2 (W_1 + W_2 + \Omega_2 \Omega_1) + o(t^2)$$

The you can consider $L(t) = R_1(t)^{-1}R_2(t)^{-1}R_1(t)R_2(t)$ which is a curve in $SO(3)$. Since $R_2(t)R_1(t)L(t) = R_1(t)R_2(t)$, you can develop term by term each side and find that:

$$ L(t) = I + t^2([\Omega_1, \Omega_2]) + o(t^2)$$

If you re-parametrise L with $s = t^2$ you have $L(s) = I + s([\Omega_1, \Omega_2]) + o(s)$ Thus you have found a curve in $SO(3)$ whose tangent is $[\Omega_1, \Omega_2]$.

  • 1
    $\begingroup$ Thanks, but the last term $L(s)=I+s([\Omega_1,\Omega_2])+O(s)$ still contains that $O(s)$ term. What if that term was $sM+O(s^2)$ with some matrix $M$. Then the tangent of $L(s)$ is $M+[\Omega_1,\Omega_2]$, which is different from $[\Omega_1,\Omega_2]$. $\endgroup$ – Bass Aug 12 '15 at 10:05
  • $\begingroup$ Why is $R_2(t)^{-1}R_1(t)^{-1}L(t) = R_1(t)R_2(t)$ ? I think the inversions are wrong, but you didn't accept my edit. $\endgroup$ – cheesus Aug 12 '15 at 11:05
  • $\begingroup$ @BastianTreichler: Thanks for the remark. My $O(.)$ are indeed $o(.)$, this was a typo. I edited it. $\endgroup$ – Tantto Aug 12 '15 at 14:47
  • $\begingroup$ @cheeesus: I think the way wrote is correct. If $L(t) = R_1(t)^{-1}R_2(t)^{-1}R_1(t)R_2(t)$ then you need to multiply from the left by $R_1(t)$ and then by $R_2(t)$ to obtain the second expression. $\endgroup$ – Tantto Aug 12 '15 at 14:47
  • $\begingroup$ Exactly, but you are left-multiplying by $R_1(t)^{-1}$ and then by $R_2(t)^{-1}$. $\endgroup$ – cheesus Aug 12 '15 at 16:09

Found a solution, but it has nothing to do with $(1)$:

$$ \begin{align} [\Omega_1,\Omega_2] & = \Omega_1\Omega_2-\Omega_2\Omega_1 = (-\Omega_1^T)(-\Omega_2^T) - (-\Omega_2^T)(-\Omega_1^T) \\ & = \Omega_1^T\Omega_2^T-\Omega_2^T\Omega_1^T=(\Omega_2\Omega_1)^T-(\Omega_1\Omega_2)^T = (\Omega_2\Omega_1-\Omega_1\Omega_2)^T \\ & = -[\Omega_2,\Omega_1] \end{align} $$


Here is a possibility as to how the professor was arriving at the result that $[\Omega_1, \Omega_2]$ is an element of the Lie algebra. Suppose that we take the above formula as the definition of the Lie Bracket: $$ \tag{1}[\Omega_1,\Omega_2] = \left.\frac{d}{dt}R_1(t)\Omega_2 R_1(t)^{-1}\right|_{t=0} $$ There is a proposition that for any Lie group element $A$ and Lie algebra element $X$ (of its associated Lie algebra), that $AXA^{-1}$ is also in the Lie algebra. (Proof from Hall: $e^{t(AXA^{-1})} = Ae^{tX}A^{-1}$ and the latter is in the Lie group).

This means in the above, the curve $$C(t) = R_1(t)\Omega_2 R_1(t)^{-1}$$ lies entirely in the Lie algebra. Hence its derivative at $t = 0$ is also in the Lie algebra (because a Lie algebra is a vector space). So the misconception from the original question was that $C(t)$ lives in $\mathfrak{so}(3)$ instead of $SO(3)$.

  • $\begingroup$ Could you elaborate a bit more about the Proof from Hall part? If you read my question, this was exactly the thing I wanted to see. I can't yet see why $C(t)$ must be a curve in $SO(3)$. $\endgroup$ – Bass Aug 12 '15 at 15:59
  • $\begingroup$ What misconception do you mean? I never asked why $C(t)$ is in $so(3)$. My original question was: why is $C(t) \in SO(3)$ for an interval around $t=0$. $\endgroup$ – Bass Aug 12 '15 at 16:01
  • $\begingroup$ It is not a curve in $SO(3)$ it is a curve in $\mathfrak{so}(3)$. Regarding the proof, the RHS is an element of the group because it is the product of three group elements. Then by definition of the Lie Algebra, if $e^{t(AXA^{-1}}$ in the Lie group for all $t$, then $AXA^{-1}$ is in the Lie algebra. $\endgroup$ – muaddib Aug 12 '15 at 16:03
  • $\begingroup$ Yes, what I am saying is that is where you went wrong. $\endgroup$ – muaddib Aug 12 '15 at 16:03
  • $\begingroup$ Oh now I see... I always expected $C(t)$ to be in $SO(3)$ so that $(1)$ defines a tangential vector and thus a member of $so(3)$. Okay, so now that this is sorted out, do you have more information about that "Proof of Hall"? When I google it, I only find pages about Hall's theorem, which is something else. $\endgroup$ – Bass Aug 12 '15 at 16:06

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.