research/cantera
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

[1D] Fixed a case where the Newton solver could get stuck

Browse source 
Sometimes, while trying to solve the steady-state problem, the Newton solver
would get stuck in a loop where it couldn't find a suitable damping ratio even
after re-revaluating the Jacobian, causing it to get stuck in an loop where it
would keep re-evaluating the Jacobian at the same point (i.e. without having
made a successful damped step).

This change detects this condition and stops the Newton solver so that the 1D
solver can advance the solution by timestepping before trying to solve the
steady-state problem again.
This commit is contained in:
Ray Speth 2013-01-22 20:28:34 +00:00
parent 021f578ebd
commit 80f3c9d24d
1 changed files with 5 additions and 0 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

5
src/oneD/MultiNewton.cpp
View file

@ -312,6 +312,7 @@ int MultiNewton::solve(doublereal* x0, doublereal* x1,
bool frst = true;
doublereal rdt = r.rdt();
int j0 = jac.nEvals();
int nJacReeval = 0;
while (1 > 0) {
@ -370,6 +371,10 @@ int MultiNewton::solve(doublereal* x0, doublereal* x1,
else if (m < 0) {
if (jac.age() > 1) {
forceNewJac = true;
if (nJacReeval > 3) {
goto done;
}
nJacReeval++;
if (loglevel > 0)
writelog("\nRe-evaluating Jacobian, since no damping "
"coefficient\ncould be found with this Jacobian.\n");