Talk:Energy drift

Should this article make reference to the flying ice cube problem in MD?

I'm not an MD person but this artefact seems thematically related to the flying ice cube problem in MD. Both are artefacts of numerical integration - one appears to violate conservation of energy and the other equipartition of energy. Would it be helpful to place each in the correspoding 'see also' section of the other page or would this just confuse matters? Eutactic (talk) 06:55, 15 September 2009 (UTC)

I wouldn't put it in the 'see also' but I might mention it in the text. Updating now... 7daysahead (talk) 15:50, 4 January 2012 (UTC)

Factor of three vs. two?

I'm not sure if I understand why the time step should be three-fold smaller than the shortest period, rather than two-fold, which I get by analogy with parametric resonance. Does anyone have any insights to help me understand this intuitively? Willow 16:30, 23 February 2007 (UTC)

Is the energy actually growing?

I'm not quite convinced if the energy should be growing. I know from my experience that useually it falls down and drift is something like -0.00sth.

See this article: http://scitation.aip.org/getpdf/servlet/GetPDFServlet?filetype=pdf&id=JCPSA6000126000004046101000001&idtype=cvips&prog=normal On 2nd page there is a nice plot of drifts. Greetings kassiel —Preceding unsigned comment added by 83.10.60.55 (talk) 23:46, 22 May 2008 (UTC)

"Energy drift is substantial for numerical integration schemes that are not symplectic"

I get what symplectic methods claim to do, but in practice they don't. The above claim simply isn't true. For context: Forward Euler method is a "1st-order symplectic method", Verlet method is "2nd-order symplectic method", etc. If you could take it to infinite order, then symplectic methods would actually conserve energy, but then forward Euler would also conserve energy if we take dt->0. 74.140.199.156 (talk) 06:50, 20 May 2021 (UTC)

This isn't quite right; standard Euler integration is not symplectic, only the slightly modified "semi-implicit" Euler is a first-order symplectic integrator. Beyond this nitpick, there is nothing incorrect about the quoted article excerpt. Energy drift tends to be more substantial for non-symplectic integrators, and that's a fact of practice, not theory. Having a conserved shadow Hamiltonian doesn't eliminate energy drift, but it strongly curtails it, since there is a truly conserved energy-like quantity, it just isn't the energy of the original unperturbed Hamiltonian. 128.220.159.211 (talk) 15:56, 9 September 2024 (UTC)