About the "Update scaled absolute tolerance" check box.
This feature of the Time-dependent solver has been on by default in the last few versions of Comsol.
In some models in the Model Library, the instruction is to uncheck it, but it's given without comment. For example, in these models:
• Capacity Fade in a Lithium-Ion Battery
• Peristaltic Pump
• Nonlinear Acoustics: Modeling of the 1D Westervelt Equation
I can't find a single theme in them (but I haven't looked in-depth).
From the documentation, when the check box is checked, the behaviour is described essentially as an interpolation in time, between the absolute tolerance value as entered into the "Absolute tolerance" field, and the same thing multiplied by the maximum value of the dependent variable taken over the entire field, where it is smaller than 1. (Plus, the L2 norm of it is added, which will always be smaller than the L-infinity norm, so that must be for the case where L-infinity norm is greater than 1.)
So it seems reasonable (and, in fact, better) to adaptively adjust the absolute tolerance to be relative to the max value or the L2 norm of the dep. variable, like that.
My questions are:
• When should this *not* be used?
• In particular for the above sample models, why has this feature been disabled?
• Is there a way to access the value of the absolute tolerance as it's being adapted throughout the time-dependent simulation?
This feature of the Time-dependent solver has been on by default in the last few versions of Comsol.
In some models in the Model Library, the instruction is to uncheck it, but it's given without comment. For example, in these models:
• Capacity Fade in a Lithium-Ion Battery
• Peristaltic Pump
• Nonlinear Acoustics: Modeling of the 1D Westervelt Equation
I can't find a single theme in them (but I haven't looked in-depth).
From the documentation, when the check box is checked, the behaviour is described essentially as an interpolation in time, between the absolute tolerance value as entered into the "Absolute tolerance" field, and the same thing multiplied by the maximum value of the dependent variable taken over the entire field, where it is smaller than 1. (Plus, the L2 norm of it is added, which will always be smaller than the L-infinity norm, so that must be for the case where L-infinity norm is greater than 1.)
So it seems reasonable (and, in fact, better) to adaptively adjust the absolute tolerance to be relative to the max value or the L2 norm of the dep. variable, like that.
My questions are:
• When should this *not* be used?
• In particular for the above sample models, why has this feature been disabled?
• Is there a way to access the value of the absolute tolerance as it's being adapted throughout the time-dependent simulation?
