I am working through a book titled "An introduction to mechanics and symmetry" by Marsden and Ratiu. I have written up a brief summary trying to solidify my understanding of the general principles.

Two questions:

i) Is my understanding of the ideas sound? I realise I should probably know if I understand the material once I understand it but I think it takes some time to internalize this stuff.

ii) How does one actually prove that the flows of the vector field $X_L$ correspond to the Euler Lagrange equations of $L$.


$\newcommand{\rn}{\mathbb{R}} \newcommand{\fn}{\mathbb{F}} \newcommand{\into}{\rightarrow}$ Let $T^*M$ be the cotangent bundle of $M$. The Liouville 1-form $\Theta$ on $T^*M$ is defined as follows. Let $(x, \omega) \in T^*M_x$ and then let $W$ be in $TT^*M_{(x, \omega)}$. Then $\Theta_{(x, \omega)} (W) = \omega(d \pi^* (W))$ where $\pi^* : T^*M \into M$ is the projection map. With this, we can define a 2-form $\Omega$ on $T^*M$ by taking the exterior derivative, $\Omega = -d\Theta$.

For a function $H : T^*M \into \rn$, the Hamiltonian vector field $X_H$ on the cotangent bundle is the vector field satisfying the property that $dH = \iota_{X_H} \Omega$. Flows under this vector field represent solutions to the classical Hamiltonian equations under a suitable Hamiltonian function $H$.

To switch between Hamiltonian formalism and the Lagrangian formalism, we use the Legendre transform. Given a map $L : TM \into \rn$, define $\fn L : TM \into T^*M$ by $\fn L(v)(w) = \frac{d}{ds}\big|_{s=0} L(v + sw)$. We have the corresponding forms on $TM$ given by $\Theta_L = (\fn L)^* \Theta$ and $\Omega_L = (\fn L)^* \Omega$. $\Theta_L$ is called the Lagrangian 1-form and $\Omega_L$ the Lagrangian 2-form.

Then let $X_L$ be a vector field on $TM$ such that $\iota_X \Omega_L = dE$ where the energy $E$ is defined by $E(x,v) = \mathbb{F}L(v)\dot(v) - L(x,v) = \Theta_L(X)(v) - L(v)$

Then the integral curves (I think) of $X_L$ are solutions to the Euler Lagrange equations. Suppose a curve $v(t) \in TM$ satisfies $v^{(1)}(t) = X(x, v)$. Then does $v^i(t) = x^{i(1)}(t)$ and $L_x(x, v) - \frac{d}{dt} L_v (x, v) = 0$?

Thanks for your time.


2 Answers 2


You start with a manifold $M$. A Lagrangian function is just an element $L\in C^\infty(T M)$. Call the pair $(M,L)$ a Lagrangian system. A motion in the Lagrangian system is just a curve $\gamma:\mathbb{R}\to M$ whose lift, $(\gamma(t),\gamma^\prime(t))\in T_{\gamma(t)}M$, satisfies the Euler-Lagrange equations.

A Hamiltonian system is a triple $(X,\omega, H)$ where $(X,\omega)$ is a symplectic manifold and $H\in C^\infty(X)$ is called the Hamiltonian function. The integral curves of $X_H$, the Hamiltonian vector field, are characterized by solving Hamilton's equations. Note that you can always turn $T^\ast M$ into a symplectic manifold in the way you described above.

Given the Lagrangian system as above, we get the associated Legendre transform $\Phi_L: T M\to T^\ast M$. Under certain conditions on $L$, we get that $\Phi_L$ is a diffemorphism. The Legendre transform relates the two formulations via the following theorem:

A curve $\gamma:\mathbb{R}\to M$ is a motion in $(M,L)$ (i.e. satisfies the Euler-Lagrange equations) if and only if $\Phi_L(\gamma(t),\gamma^\prime(t))$ is an integral curve for $X_H$ (i.e. satisfies the Hamilton equations)

Here $H=L\circ \Phi_L^{-1}$. That is we are considering the Lagrangian system $(M,L)$ and the Hamiltonian system $(T^\ast M,\omega, H=L\circ\Phi_L^{-1})$ with $H\in C^\infty (T^\ast M)$ and $X_H\in \Gamma(T(T^\ast M))$.

  • 2
    $\begingroup$ Thank you for your incredibly clear answer :) $\endgroup$
    – muzzlator
    Jul 7, 2014 at 6:08
  • 1
    $\begingroup$ I think it should be that $H = E (\phi^{-1}_L)$ where $E:TM \rightarrow \mathbb{R}$ is the function $E(v) = (\phi_L(v))(v)-L(v)$. See below for the correct explanation. $\endgroup$
    – Sina
    Jan 14, 2015 at 17:37

Simply because you pull back the equation $\omega(X,\cdot) = dH$ to an equation on $T^*(TM)$ using the Legendre transformation. Indeed denote $\phi_L: TM \rightarrow T^*M$ the Legendre transformation and denote the Hamiltonian as $H(p) = p(\phi_L^{-1}(p)) - L(\phi^{-1}_L(p))$ (this is how you define an Hamiltonian from the Lagrangian in the classical setting). Then if you define the quantity "energy" as you defined above that is $E(v)= (\phi_L(v))(v) - L(v)$ it is easy to see that $E(v) = H(\phi_L(v))$

Now applying the pull back to the equation you get:

$$\phi^*_L (\omega(X_H,\cdot) = d(\phi_L^*H)$$

$$ (\phi^*_L\omega)(D\phi^{-1}_L X_H, \cdot) = dE$$

$$ \Omega_L(X_L,\cdot) = dE$$

Just one remark on how you can also pass to Lagrangian mechanics from Hamiltonian in a "natural way". As you have remarked you can build the canonical 1-form on $T^*M$ as $\Theta_{x,p}(W) = p\circ d\pi^*(W)$. Now if $\pi_*: TM \rightarrow M$ is the canonical projection, given a Lagrangian one can define a 1-form on $TM$ with almost the same procedure as follows: $(\Omega_L)_{x,v}(U) = \phi_L(v) ( d\pi^*(U))$ for $U$ in $T(TM)$. If you want the differential of this 1-form to be symplectic, you need the non-degeneracy condition $det(\frac{\partial^2L}{\partial v_i \partial v_j}) \neq 0$. This is the same symplectic form $\Omega_L$ above. You just use an isomorphism between $TM$ and $T^*M$ to transport the symplectic structure on $T^*M$ to $TM$. Infact in the cases where $L(v) = g(v,v) - V(\pi_*(v))$ then Lagrangian transformation is simply the isomorphism between $TM$ and $T^*M$ given by the metric (the musical isomorphism).


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.