"Resolution" of time in Mumax3
18 views
Skip to first unread message
Elizabeth Ramirez
Feb 12, 2026, 11:11:57 AMFeb 12
to mumax2
Hello Mumax community,
When defining time parameters for how long a simulation will run, up to how many decimal points after the nanosecond time frame is accepted. As an example:
Say I have been running a simulation for 5.00 ns. But I am curious how the simulation behaves at more specific points such as 5.08 ns, or even up 5.0008185...
What is the limit to this?
Thank you!
Say I have been running a simulation for 5.00 ns. But I am curious how the simulation behaves at more specific points such as 5.08 ns, or even up 5.0008185...
What is the limit to this?
Thank you!
Josh Lauzier
Feb 15, 2026, 6:11:44 AMFeb 15
to mumax2
Hi,
The main practical limitation in normal usage will be your timestep. That is the smallest step the simulation takes. The timestep can be either fixed, or adaptive if you're using one of the adaptive solvers. The exact timestep that is appropriate will vary depending on your problem (in particular, it is usually going to be dictated by how strong exchange is). The adaptive solvers will choose a timestep based on a target error per step, and this error threshold can be set (you can also set a minimum step size, or a maximum). Typically these timesteps will be somewhere between 1e-15 to 1e-11. If you try to save data that is finer than a single timestep and doesn't align with a step, I believe mumax just rounds you to the closest one.
As you shrink the timestep, eventually there will be a floor to the accuracy based on (single) numerical precision, where reducing the timestep will no longer reduce the error and you will hit a noise floor. The exact point will depend on which solver you are using. You can see a chart for this for the various solvers in the mumax paper, "the design and verification of mumax3" , in Fig 10 and Fig 11.
Keep in mind that smaller timesteps will lead to the simulation computation taking longer, so in general it is good to take as large as timesteps as possible while maintaining reasonable error. For reasonable parameters, not much will change from step to step, so basically what you will see as you shrink will be as if you interpolated between the previous and next step.
Cheers,
Josh L.
The main practical limitation in normal usage will be your timestep. That is the smallest step the simulation takes. The timestep can be either fixed, or adaptive if you're using one of the adaptive solvers. The exact timestep that is appropriate will vary depending on your problem (in particular, it is usually going to be dictated by how strong exchange is). The adaptive solvers will choose a timestep based on a target error per step, and this error threshold can be set (you can also set a minimum step size, or a maximum). Typically these timesteps will be somewhere between 1e-15 to 1e-11. If you try to save data that is finer than a single timestep and doesn't align with a step, I believe mumax just rounds you to the closest one.
As you shrink the timestep, eventually there will be a floor to the accuracy based on (single) numerical precision, where reducing the timestep will no longer reduce the error and you will hit a noise floor. The exact point will depend on which solver you are using. You can see a chart for this for the various solvers in the mumax paper, "the design and verification of mumax3" , in Fig 10 and Fig 11.
Keep in mind that smaller timesteps will lead to the simulation computation taking longer, so in general it is good to take as large as timesteps as possible while maintaining reasonable error. For reasonable parameters, not much will change from step to step, so basically what you will see as you shrink will be as if you interpolated between the previous and next step.
Cheers,
Josh L.
Reply all
Reply to author
Forward
0 new messages