We can solve some differential equation numerically using Runge-Kutta methods. i.e. for some function $f(x,y) = \frac{dy}{dx}$, then

$$ y_{n+1} = y_n + \sum_{i=1}^{s} c_i k_i $$

where e.g.

$$ k_2 = f(x_n+a_2 h, y_n + b_{21}k_1)$$

and $a_i$, $b_{ij}$ are some coefficeints e.g. Cash-Karp.

Now, I understand that there also exist symplectic Runge-Kutta methods. How would this method work? What would be the equivalent $y_{n+1}$ style equations for a symplectic Runge-Kutta? How are the coefficients chosen?


1 Answer 1


The following is sufficient and necessary for a Runge-Kutta method to be symplectic: $$ b_i a_{i,j} + b_j a_{j,i} - b_ib_j = 0 \, \quad 1 \leq i,j \leq s $$ where $s$ is the number of stages in your method. No explicit RK methods achieve this.

This condition is in addition to other consistency conditions too.

Also, don't forget the $dt$ in your method, and the convention is to use $b_i$ in the linear combination of stage derivatives $k_i$, $c_i$ multiplied by $h=dt$, and $a_{i,j}$ for the states.

source: http://www.sanzserna.org/pdf/56_physicaD.pdf

EDIT: As pointed out in the comments, this applies to symplectic Runge-Kutta methods, but generally methods may be symplectic but not Runge-Kutta.

  • $\begingroup$ This applies to the use of standard RK methods, like the midpoint method. A larger number of symplectic methods are constructed as partitioned methods, with different, but coupled stages for the position and impulse components of the system. $\endgroup$ Mar 14 at 17:02
  • $\begingroup$ Indeed. But those methods are not Runge-Kutta per se, so I am answering the OP. $\endgroup$
    – haavbj
    Mar 14 at 19:30

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.