Opened 8 years ago

Last modified 5 years ago

#99 new defect - convergence

crash with non-conservation

Reported by: Gary J. Ferland Owned by: nobody
Priority: minor Milestone:
Component: ionization convergence Version: trunk
Keywords: Cc:

Description

reported by Tim Kinnear on yahoo groups on 2009 june 11,

c08 and r3090 gave this error: PROBLEM DISASTER PressureTotal?: the chemical species are not conserved. The sum of the densities of the atoms and ions for Carbon is 1.725e+08 which is greater than the gas-phase abundance of the element, 3.162e+07.

Ryan reports that newmole at this time did far better, "It fails after more than 1000 zones in the fifth iteration of 20 molecular convergence failures."

Attachments (1)

kinnear.in (67.7 KB) - added by Gary J. Ferland 8 years ago.
input for test

Download all attachments as: .zip

Change History (3)

Changed 8 years ago by Gary J. Ferland

Attachment: kinnear.in added

input for test

comment:1 Changed 5 years ago by Ryan Porter

Component: pressure convergenceionization convergence
Version: trunk

I can get this to pass now at r6340, but there are still a number of ionization convergence failures (OTSRatChng), and it takes It seems like the cause is when one iteration stops because NZONE was reached and all previous iterations stopped because the outer radius was reached. Metals are scaled up by 3e10 so the first iteration (which will begin with a guess based on Hydrogen) is really, really bad. Being more open in the rt initialization routines about what element dominates the optical depth may help this sim. All of this happened with "set nend 3000". The 5% molecular abundance change criterion looks to be the main reason so many zones are required. Changing component to "ionization convergence".

comment:2 Changed 5 years ago by Ryan Porter

Priority: majorminor

Bumping the number of zones still further, the sim finishes with only three convergence failures (two sub-1% temperature, one ~20% OTS). Code does not appear to be converged after seven iterations (the reporter used "iterate to convergence 7").

But I think I've at least documented enough to demote this to minor.

Note: See TracTickets for help on using tickets.