raspstephan / l96-visualization Goto Github PK

Dear Stephan,

Thanks a lot for making the code publicly available! It has been very helpful to me!

Is it possible that np.roll(X, shift_param) shifts in the opposite direction of what is intended?

Example:

print(initX[0:5])
> [7.22218496 2.99854488 1.42502222 3.22644313 3.42971768]
print(np.roll(initX[0:5], -1))
> [2.99854488 1.42502222 3.22644313 3.42971768 7.22218496]

I think this thus gives $x_{k + 1}$ not $x_{k -1}$ for the multiplier in the advection term from https://towardsdatascience.com/lorenz-96-is-too-easy-machine-learning-research-needs-a-more-realistic-toy-model-6add938f6cc0 for example.

Should thus the shift parameters be negated and changed to the following?

    def _rhs(self, X):
        """Compute the right hand side of the ODE."""
        dXdt = (
                - np.roll(X, 1) * (np.roll(X, 2) - np.roll(X, -1)) -
                X + self.F
        )
        return dXdt

I am unsure if this actually changes much as the ordering of the k X variables/X states is arbitrary.

Thanks in advance,
Kim