From: Axel Kohlmeyer (akohlmey_at_gmail.com)
Date: Mon Aug 11 2014 - 06:40:43 CDT
On Mon, Aug 11, 2014 at 12:16 AM, Kenno Vanommeslaeghe
<kvanomme_at_rx.umaryland.edu> wrote:
> On 08/10/2014 11:29 PM, Hadi wrote:
[...]
> (*) "slow" can only be defined relative to something else that's faster.
> "Analytically" integrating rigid body rotations is computationally
> surprisingly expensive (the trigonometric functions play a role there), and
> the high-performance codes I'm familiar with deliberately avoid doing so in
> favor of a constraint algorithm.
it is not only the arithmetic intensity that makes integrating the
rotational degrees of freedom slow, but rather that this integration
requires the use of a much shorter time step. please note, that time
integration in euler angles is rarely done these days, because of the
issue of their discontinuity is making the code extremely complex.
usually codes use quaternions for time integration, as those are
continuous. however, quaternions also have a well known problem of
requiring different numerical accuracy depending on the orientation of
the individual rigid bodies, and that can lead to quite significant
errors as well, since we use floating point math with limited
numerical accuracy.
-- Dr. Axel Kohlmeyer akohlmey_at_gmail.com http://goo.gl/1wk0 College of Science & Technology, Temple University, Philadelphia PA, USA International Centre for Theoretical Physics, Trieste. Italy.
This archive was generated by hypermail 2.1.6 : Wed Dec 31 2014 - 23:22:42 CST