Pyscal interpolation option causing large number of failing simulations

[edubarrosTNO]

I noticed that when using the interpolate: true entry in relative_permeability, a lot more simulations fail already with the prior ensemble as set up for the Norne case (current version on flownet-testdata / master). I also observed that the simulations take much longer to run when interpolate: true compared to interpolate: false

[olelod]

Do we have a simple way of plotting/comparing the relative permeability curves produced by either method for some flow tubes? A number close to -1 will in practice use low values for all the relative permeability parameters, and a number close to 1 will use all high values. Maybe the curves are getting numerically difficult due to that? When we are not interpolating, the different values are drawn independently, so we are maybe not ending up in a very pessimistic or very optimistic scenarios?

[wouterjdb]

You could potentially test with running pyscal in normal mode and not "fast mode" - that should spotted (obvious) errors.

[olelod]

I don't think the functionality around interpolating between low/base/high has a fast mode implemented. I could not find a way to pass along the fast parameter, at least.

[wouterjdb]

I don't think the functionality around interpolating between low/base/high has a fast mode implemented. I could not find a way to pass along the fast parameter, at least.

Ok, then that doesn't help.

[olelod]

I have been checking a little and thinking a bit. So, as of now, the most pessimistic relperm curve is made up from all the low input values, but come to think of it, some low values are actually good for us.... This makes the initial ensemble look quite weird in the interpolation case:

Generating curves without interpolation gives an ensemble like this:

[olelod]

231 will hopefully help here

[olelod]

231 has been merged so hopefully you will see improved performance now @edubarrosTNO

[wouterjdb]

@edubarrosTNO will test today.