Opened 8 years ago

Closed 5 years ago

#148 closed defect - convergence (fixed)

grid command fails with non-convergence

Reported by: Gary J. Ferland Owned by: nobody
Priority: major Milestone: C13 branch
Component: chemical network Version: trunk
Keywords: Cc:

Description

Donghui has found certain parameter values where grids fail. He is running PDR models and varying the density and the ISM radiation field. (For the CO comparison project).

most density/ G0 pairs are ok. the ones that do fail in the grid work ok by themselves. Sounds like an initialization problem.

in the grid the first point that fails is density 5.4 and G0 4.7.

the grid does not fail if he varies the density but keeps G0 fixed. The grid does fail if the density is fixed but G0 varied.

this is incomplete - will be expanded as he does more testing

Attachments (2)

grid.in (1.1 KB) - added by Gary J. Ferland 8 years ago.
two point grid, second fails
one.in (1.1 KB) - added by Gary J. Ferland 8 years ago.
the second grid point, is OK. T(En-Den) is 1 dex smaller than in grid!

Download all attachments as: .zip

Change History (7)

Changed 8 years ago by Gary J. Ferland

Attachment: grid.in added

two point grid, second fails

Changed 8 years ago by Gary J. Ferland

Attachment: one.in added

the second grid point, is OK. T(En-Den) is 1 dex smaller than in grid!

comment:1 Changed 8 years ago by Gary J. Ferland

Donghui isolated two simple models. The attached grid.in will run quickly, the second model has temperature convergence failures. these would continue until aborting if the sim were allowed to go deeper.

one.in is only the second model in the grid - it runs ok if you run it by itself. Note that the energy density temperature in the header is different by a large amount:

grid.in gives T(En-Den):3.029E+01 one.in (same draine) gives T(En-Den):2.829E+00

one successfully fine one 1 temperature of 84K. in the grid run the temperature failure is around 2295K.

the radiation field is being set up quite differently in the grid point.

comment:2 Changed 8 years ago by Gary J. Ferland

from Ryan:

\It's a log/linear issue. The parameters supplied in the grid command are automatically interpreted as logs. The parameter supplied on the standalone "table draine command" is interpreted according to the value and the presence or absence of "LOG."

thanks - I see this now. Donghui and I are just sitting down to look at this.

comparing the grid vs single model, the actual draine command was identical in the input script. here is how they were interpreted:

single model:

  • c they want for the model is actually twice the value below *
  • table draine 4.8 *

grid models:

  • c Quan: use grid command to vary incident intensity *
  • TABLE DRAINE LOG 4.700000 *

is this behavior documented in hazy? section 16.1 of hazy 1 does mention grid using logs unless linear is included on the grid command. but the draine command definition in section 6.11.6 does not even say that the log option exists. I just added this.

this is a trap. I certainly fell into it. is the rule that all commands modulated by a grid have arguments that are logs unless the linear option is specified on the grid command?

comment:3 Changed 8 years ago by Gary J. Ferland

from Ryan:

The more fundamental problem looks like oscillations in OH charge transfer.

comment:4 Changed 8 years ago by Gary J. Ferland

this is a trap. I certainly fell into it. is the rule that all commands modulated by a grid have arguments that are logs unless the linear option is specified on the grid command?

Peter's response: For most, yes, but with 4 (rather obscure) exceptions. This is all documented in section 16.1 of Hazy 1. The first example even shows exactly what you describe here.

comment:5 Changed 5 years ago by Ryan Porter

Resolution: fixed
Status: newclosed

Works fine at r6340. Closing.

Note: See TracTickets for help on using tickets.