How to control the total loading strain

[mikejwg]

When loading at a constant strain rate using the following settings:

maxstep = 200000
timestepIntegrator = forceBsubcycle # "trapezoid" is commented out
deltaTT = 1.656930e-12
maxDT = 1.000000e-07
nextDT = 1.988316e-12
rTol = 10

Fast Multipole Method controls

fmEnabled = 0
fmMPOrder = 2
fmTaylorOrder = 5
fmCorrectionTbl = "inputs/fm-ctab.Cu.m2.t5.dat"

Loading conditions

loadType = 1
eRate = 1.0e03
edotdir = [
1
1
0
]
How can the system be loaded to achieve a specific total strain (for example, 0.005)? With the above code, it is difficult to control the final total strain due to varying convergence properties.
In addition, for high initial dislocation densities, such as 1e14 m^-2, convergence can be particularly slow. Are there any methods to accelerate the convergence process?

[nrbertin]

There is actually no parameter available to control the maximum strain in ParaDiS. However, this option is available in our new OpenDiS framework (e.g. see the max_strain option in here).

The computational cost of a simulation scales with the number of dislocation segments. To keep the cost manageable, at high dislocation densities one can use smaller simulation boxes to keep the same number of dislocation segments as in a simulation with lower dislocation density. In all cases, using the subcycling integrator and running on GPU can significantly improve the simulation performance. Both these features are available in OpenDiS.

(Note: although you specified to use the subcycling integrator timestepIntegrator = forceBsubcycle in your input, LLNL ParaDiS version does not implement it, so your simulation will default to using the much less efficient trapezoid integrator.)

[mikejwg]

There is actually no parameter available to control the maximum strain in ParaDiS. However, this option is available in our new OpenDiS framework (e.g. see the max_strain option in here).

The computational cost of a simulation scales with the number of dislocation segments. To keep the cost manageable, at high dislocation densities one can use smaller simulation boxes to keep the same number of dislocation segments as in a simulation with lower dislocation density. In all cases, using the subcycling integrator and running on GPU can significantly improve the simulation performance. Both these features are available in OpenDiS.

(Note: although you specified to use the subcycling integrator timestepIntegrator = forceBsubcycle in your input, LLNL ParaDiS version does not implement it, so your simulation will default to using the much less efficient trapezoid integrator.)

thanks very much for your help.