I posted my questions in a previous post MO, but it seems that a more refined version for question on the Euler-Lagrange equation is needed. So, I post my question again.

In standard symplectic geometry, we assume the "canonical coordinates" $(p_i,q_i)$ at least locally, under with the symplectic form $\omega$ is written as \begin{equation} \omega=\sum_i dp_i \wedge dq_i \ . \end{equation}

Then the Hamilton's equations are $\dot{p_i}= -\frac{\partial H}{\partial q_i}$ and $\dot{q_i}= \frac{\partial H}{\partial p_i}$ while the Euler-Lagrange equation is \begin{equation} \frac{d}{dt}\frac{\partial L}{\partial \dot{q}_i}=\frac{\partial L}{\partial q_i} \end{equation}

However, I am curious about how to write the Hamilton's equations and Euler-Lagrange equation in "arbitrary" coordinates $(p'_i,q'_i)$ where $\omega \neq \sum_i dp'_i \wedge dq'_i$.

In "Classical Dynamics : A Contemporary Approach" (1998) by Jose Saletan, the Euler Lagrange equation is expressed in a coordinate-free way by \begin{equation} \mathcal{L}_\Delta \theta_L = d L \end{equation} where $\Delta$ is the vector field generating time translation and $\theta_L$ is the canonical $1$-form corresponding to the Lagrangian $L$. However, he only defines these quantities in terms of canonical (local) coordinates as \begin{equation} \theta_L : =\sum_i \frac{\partial L}{\partial \dot{q}_i} dq_i \text{ while } \Delta:= \sum_i \dot{q_i} \frac{\partial}{\partial q_i} + \ddot{q_i} \frac{\partial}{\partial \dot{q_i}} \end{equation}

So, I don't think it "truly" coordinate-free...

Could anyone please provide the general formula for the Euler-Lagrange equation either

  1. In a "truly" coordinate-independent way (which is more desired for me)
  2. In "arbitrary" local coordinates $(p'_i,q'_i)$ in which $\omega \neq \sum_i dp'_i \wedge dq'_i$

I would deeply appreciate any help...

  • 1
    $\begingroup$ Hamilton's equations are a special case of Euler-Lagrange equations derived from the variational principle $S[p,q] = \int (\omega(p,\dot{q}) - H(p,q)) dt$. You can take the dynamical variables $z = (p,q)$, rewrite them in any set of coordinates you please and derive the EL equations with respect to them. This kind of question comes up from time to time. See some discussion and references here. $\endgroup$ Commented Jan 12 at 16:31
  • $\begingroup$ @IgorKhavkine Thank you. I will read your link very carefully. $\endgroup$
    – Isaac
    Commented Jan 12 at 16:38
  • $\begingroup$ @IgorKhavkine Just to check my understanding : in your formula above, $z=(p,q)$ is NOT necessarily the canonical coordinates under which $\omega=\sum_i dp_i \wedge dq_i$. Am I correct? $\endgroup$
    – Isaac
    Commented Jan 12 at 16:38

1 Answer 1


The answer is in this short paper by Erik Curiel. The Geometry of the Euler-Lagrange Equation

The key idea is to express the Euler-Lagrange equation in the tangent bundle, in terms of a structure tensor that the tangent bundle automatically comes equipped with.

Let $Q$ be the coordinate manifold. Pick coordinate chart $q_i$ of $Q$. Then $q_i, \dot{q}_i$ is a chart of the tangent bundle $TQ$.

Define a $(1,1)$ tensor $J$ on $TQ$ by

$$ J \begin{bmatrix} \vec{\dot{q}} \cr \vec{\ddot{q}} \end{bmatrix} = \begin{bmatrix} 0 \cr \vec{\dot{q}} \end{bmatrix} $$

You can check that this is well defined and doesn't depend on the coordinate chart.

Let $\xi$ be a vector field in $TQ$, which is to be the solution of the Euler-Lagrange equation. Then the equation can be expressed as:

$$ \mathscr{L}_\xi ({dL} \cdot J) = dL $$

  • $\begingroup$ Thank you. So, your $q_i$'s depend on time? How do you know that $q_i, \dot{q}_i$ are linearly independent so that they comprise a local chart for $TQ$? These issues have always confused me... $\endgroup$
    – Isaac
    Commented Jan 13 at 19:05
  • $\begingroup$ $q_i$ is a chart of $Q$ by definition, so $q_i$ does not directly depend on time. $q_i, \dot{q}_i$ is the associated chart of $TQ$, where a vector at $\vec{q}$ is expressed as $V = V^i \dot{q}_i $. The way time comes into it is though $\xi$. The time evolution of the system is the flow through $TQ$ generated by $\xi$. So $\mathcal{L}_\xi$ is $d/dt$ $\endgroup$ Commented Jan 14 at 14:57
  • $\begingroup$ ah, sorry that should be $V = \dot{q}^i \partial / \partial q_i $ $\endgroup$ Commented Jan 14 at 15:09
  • $\begingroup$ by the way, I just picked up Manuel de Leon’s Methods of Differential Geometry in Analytical Mechanics and skimmed some of the chapters last night, and it has some pretty good coverage of these concepts $\endgroup$ Commented Jan 14 at 18:01
  • 1
    $\begingroup$ @Isaac Section 7.1. Equation (7.2) is equivalent to the version I gave here, you can derive it by applying Cartan's magic formula to expand the Lie derivative, and then applying the 2nd order condition on for $\xi$ $\endgroup$ Commented Jan 16 at 19:00

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.