Wikipedia gives two definitions of the Lie bracket $[X, Y]$ of vector fields $X$ and $Y$ on a manifold $M$, which it claims are equivalent:

Let $\Phi_t^Z$ be the flow associated with the vector field $Z$. At a point $x \in M$,

$[X, Y]_x := \left.\frac{\mathrm{d}}{\mathrm{d} t}\right|_{t=0} (\mathrm{d}\Phi^X_{-t}) Y_{\Phi^X_t(x)}$.

This is the definition that appears in the texts that I have seen. However, the article claims that the Lie bracket can also be defined as follows:

$[X, Y]_x := \left.\frac12\frac{\mathrm{d}^2}{\mathrm{dt}^2}\right|_{t=0} (\Phi^Y_{-t} \circ \Phi^X_{-t} \circ \Phi^Y_{t} \circ \Phi^X_{t})(x) = \left.\frac{\mathrm{d}}{\mathrm{d} t}\right|_{t=0} (\Phi^Y_{-\sqrt{t}} \circ \Phi^X_{-\sqrt{t}} \circ \Phi^Y_{\sqrt{t}} \circ \Phi^X_{\sqrt{t}})(x)$.

I find the latter definition quite intuitive in geometric terms. But I don't see how to show that the two are equivalent. Can someone point me in the right direction, or provide a reference?



This is an exercise in Warner Foundations of Differentiable Manifolds and Lie Groups. Let $M$ be a smooth manifold and let $X,Y \in \mathfrak{X}(M)$. Suppose $\phi_t$ and $\varphi_t$ are flows of $X$ and $Y$ respectively (for $t \in (−e, e)$ sufficiently small). Let: $$\beta(t)=\varphi_{-\sqrt[]{t}}\circ \phi_{-\sqrt[]{t}}\circ \varphi_{\sqrt[]{t}}\circ \phi_{\sqrt[]{t}}(m)$$ Show that for every $f$ we have: $$([X,Y]f)(m)=\lim_{t \to 0}\frac{f(\beta(t))-f(\beta(0))}{t}$$

The function is only defined for $t \ge 0$ so we have to compute the limit from the right. This is basically computing the limit interchanging $t$ with $t^2$ in the formula: $$([X,Y]f)(m)=\lim_{t \to 0^+}\frac{f(\beta(t^2))-f(\beta(0))}{t^2} $$ Define $F(t_1,t_2,t_3,t_4)= \varphi_{-t_1}\circ \phi_{-t_2}\circ \varphi_{t_3}\circ \phi_{t_4}(m)$ so we have: $$([X,Y]f)(m)=\lim_{t \to 0^+}\frac{f(F(t,t,t,t))-f(\beta(0))}{t^2} $$ We will use the Taylor rule to $g(t_1,t_2,t_3,t_4)=f \circ F$ as: $$g (t_1,t_2,t_3,t_4) =g(0,0,0,0)+\sum_i \frac{\partial g}{\partial t_i}|_{0}(t_i)+\frac{1}{2}\sum_{i,j}t_i \cdot t_j\int_0^1(1-s)\frac{\partial^2 g}{\partial t_i t_j}(s (t_1,t_2,t_3,t_4))ds $$ This means that we can write the limit as $$\lim_{t \to 0^+}\frac{\sum_i \frac{\partial g}{\partial t_i}|_{0}(t)+\frac{1}{2}\sum_{i,j}t ^2 \int_0^1(1-s)\frac{\partial^2 g}{\partial t_i t_j}(s(t,t,t,t))ds }{t^2} $$ We now observe using the fact that $\varphi_0=Id=\phi_0$ that $$\sum_i \frac{\partial}{\partial t_i}=-Y(f)(m)-X(f)(m)+Y(f)(m)+X(f)(m)=0$$ So we can write the limit as $$ \lim_{t \to 0^+}\frac{1}{2}\sum_{ij}\int_0^1(1-s)\frac{\partial^2 g}{\partial t_i t_j}(s(t,t,t,t))ds= \frac{1}{2}\sum_{ij}\int_0^1 \lim_{t \to 0^+}(1-s)\frac{\partial^2 g}{\partial t_i t_j}(s(t,t,t,t))ds=\frac{1}{2}\sum_{ij}\int_0^1 (1-s)\frac{\partial^2 g}{\partial t_i t_j}(0)ds$$ So we get $$\frac{1}{2}\sum_{ij}\frac{-1}{2}(1-s)^2\frac{\partial^2 g}{\partial t_i t_j}(0)\big|^{s=1}_{s=0}$$ And this will be equal to $$ \frac{1}{2}\sum_{ij}\frac{1}{2}\frac{\partial^2 g}{\partial t_i t_j}(0)$$ After calculating the derivatives will get the desired result $$ \frac{1}{2}\sum_{ij}\frac{1}{2}\frac{\partial^2 g}{\partial t_i t_j}(0)=XY(f)(m)-YX(f)(m) $$

  • $\begingroup$ Great, thanks for taking the time to give such a detailed answer $\endgroup$ – Simon Jul 30 '18 at 18:19
  • $\begingroup$ I am happy you found it useful $\endgroup$ – Elad Jul 31 '18 at 7:32

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.