hysteresis loops
32 views
Skip to first unread message
KL K
Nov 5, 2025, 4:29:56 AMNov 5
to mumax2
Hi,everyone.Currently, I can use mumax3 to simulate hysteresis loops under high-frequency (such as GHz and MHz) alternating magnetic fields. However, when the frequency is changed to kHz, the calculation becomes very slow. After checking the literature, I don't seem to find any simulations for frequencies below kHz either. Is it feasible to use mumax3 to simulate hysteresis loops under low-frequency (below kHz) alternating magnetic fields?
Josh Lauzier
Dec 11, 2025, 4:40:21 AM (yesterday) Dec 11
to mumax2
Hi,
I think doing this would be very challenging. The problem is the size of the time step. The solver will be limited in how big of a time step it can take while maintaining reasonable error. I don't recall the exact formula, but the time step is related to the cellsize, and the cellsize is going to be limited by material parameters like the exchange length (in particular, it's the exchange interaction that causes such hard limits). You generally need the cellsize to be half the exchange length or so, in order to properly resolve features like domain walls. For typical materials, you end up with timesteps that are something like ~1e-14 to 1e-11 or so. There's just not a whole lot you can do with such a small step size, which fundamentally means you will have to take many many steps.
Depending on what exactly you're interested in, you could try using relax() or minimize(), and the field is fixed at each value and then manually stepped. These methods are energy minimization techniques. Essentially, this would be treating the field as changing slow enough relative to the magnetic system that you can treat it as fixed, and that the system has long enough time to relax to the ground state. Code wise, it would basically just look like a normal hysteresis loop. If there are important dynamics on this timescale though, this probably would not be suitable.
If that isn't an option, I don't think there is much else you can do unless you can make some other simplifying approximations or something.
Best,
Josh L.
I think doing this would be very challenging. The problem is the size of the time step. The solver will be limited in how big of a time step it can take while maintaining reasonable error. I don't recall the exact formula, but the time step is related to the cellsize, and the cellsize is going to be limited by material parameters like the exchange length (in particular, it's the exchange interaction that causes such hard limits). You generally need the cellsize to be half the exchange length or so, in order to properly resolve features like domain walls. For typical materials, you end up with timesteps that are something like ~1e-14 to 1e-11 or so. There's just not a whole lot you can do with such a small step size, which fundamentally means you will have to take many many steps.
Depending on what exactly you're interested in, you could try using relax() or minimize(), and the field is fixed at each value and then manually stepped. These methods are energy minimization techniques. Essentially, this would be treating the field as changing slow enough relative to the magnetic system that you can treat it as fixed, and that the system has long enough time to relax to the ground state. Code wise, it would basically just look like a normal hysteresis loop. If there are important dynamics on this timescale though, this probably would not be suitable.
If that isn't an option, I don't think there is much else you can do unless you can make some other simplifying approximations or something.
Best,
Josh L.
Reply all
Reply to author
Forward
0 new messages