- Mail actions: [ respond to this message ] [ mail a new topic ]
- Contemporary messages sorted: [ by date ] [ by thread ] [ by subject ] [ by author ] [ by messages with attachments ]

From: Volker Springel <volker_at_MPA-Garching.MPG.DE>

Date: Mon, 19 Nov 2007 09:51:40 +0100

Anna Artigas wrote:

*> Dear Gadget list,
*

*>
*

*> I am performing simulations with gas and dark matter with Gadget 2 and I have
*

*> noticed that the linear momentum doesn't conserve. In my simulations the
*

*> center of mass moves appreciably during the simulation.
*

*> Does anybody have the same problem? How could I avoid this?
*

*>
*

*> Thank you,
*

*> Anna
*

*>
*

Hi Anna,

The tree-algorithm used by gadget provides an approximation of the

gravitational force, and there is no guarantee that the errors in the

forces all add up to give exactly zero. In other words, one cannot

expect that the total momentum is preserved exactly. (There are however

other types of tree algorithms, most notably the 'fast multipole'

method, which have this property to machine precision, see for example

the code by Walter Dehnen, 2000, ApJ, 536, L39. But this method cannot

be readily used with individual timesteps, this is the main reason why

it is not used by gadget at the moment.)

Normally, errors due to non-conservation of momentum are pretty small

and inconsequential, especially when one avoids imposing a wrong center

(say fixed at the origin) for an isolated object whose core has drifted

away from the initial coordinate. However, the initial conditions and

code parameter settings can have a large impact on the size of this

effect. It's clear that one should make sure that the initial

center-of-mass velocity is zero if one simulates an isolated object,

otherwise a drift will obviously occur. The code parameters for force

accuracy and time integration are highly important however, too. The

better the force accuracy, the smaller this error will become. So

improving the force accuracy setting should lead to a substantial

reduction of any induced motion of the center of mass. Equally important

is the time integration accuracy. If you use a too large

MaxSizeTimestep, in particular, then the finite error in the total force

will be 'amplified' with the use of a too large timestep.

Volker

*>
*

*>
*

*> -----------------------------------------------------------
*

*>
*

*> If you wish to unsubscribe from this mailing, send mail to
*

*> minimalist_at_MPA-Garching.MPG.de with a subject of: unsubscribe gadget-list
*

*> A web-archive of this mailing list is available here:
*

*> http://www.mpa-garching.mpg.de/gadget/gadget-list
*

Received on 2007-11-19 09:51:30

Date: Mon, 19 Nov 2007 09:51:40 +0100

Anna Artigas wrote:

Hi Anna,

The tree-algorithm used by gadget provides an approximation of the

gravitational force, and there is no guarantee that the errors in the

forces all add up to give exactly zero. In other words, one cannot

expect that the total momentum is preserved exactly. (There are however

other types of tree algorithms, most notably the 'fast multipole'

method, which have this property to machine precision, see for example

the code by Walter Dehnen, 2000, ApJ, 536, L39. But this method cannot

be readily used with individual timesteps, this is the main reason why

it is not used by gadget at the moment.)

Normally, errors due to non-conservation of momentum are pretty small

and inconsequential, especially when one avoids imposing a wrong center

(say fixed at the origin) for an isolated object whose core has drifted

away from the initial coordinate. However, the initial conditions and

code parameter settings can have a large impact on the size of this

effect. It's clear that one should make sure that the initial

center-of-mass velocity is zero if one simulates an isolated object,

otherwise a drift will obviously occur. The code parameters for force

accuracy and time integration are highly important however, too. The

better the force accuracy, the smaller this error will become. So

improving the force accuracy setting should lead to a substantial

reduction of any induced motion of the center of mass. Equally important

is the time integration accuracy. If you use a too large

MaxSizeTimestep, in particular, then the finite error in the total force

will be 'amplified' with the use of a too large timestep.

Volker

Received on 2007-11-19 09:51:30

*
This archive was generated by hypermail 2.3.0
: 2023-01-10 10:01:30 CET
*