Initial Residuals and Convergence in transient

[Traiwit]

Check these boxes if you have followed the posting rules.

• Q&A General is the most appropriate section for my question
• I have consulted the posting Guidelines on the Discussions front page
• I have searched the Discussions forum and my question has not been asked before
• I have searched the MOOSE website and the documentation does not answer my question
• I have formatted my post following the posting guidelines (screenshots as a last resort, triple back quotes around pasted text)

Question

Hi everyone,

I'm running into a weird issue with my transient analysis in MOOSE. In each analysis, the first frame (like F001) starts off with a massive initial residual around 1e+13 and converges pretty smoothly. But then, the next frame (F002) starts much lower, around 1e+08, and just won't converge.

Even weirder, if I kick off a brand new simulation using the end results of F001 as the starting point via IC, the first step (now F002) starts high again at 1e+13 and converges no problem. But the next one (F003)? Back to the same old problem. This keeps happening.

So, here's what I'm wondering:

1. Why does the first frame always start with a high residual when it’s the first in an analysis but not when it's a subsequent frame?
2. Any idea why this is happening or how to keep the residuals high like in the first frame for all steps?

Would really appreciate any thoughts or tips on this.

Thank you!

[GiudGiud]

Hello

are you using automatic_scaling by any chance?

[GiudGiud]

then you must be accessing out of bounds.
Is this in parallel?

maybe there is another ay to grab solution from a node?

yes actually, accessing the solution vector is not the sanctioned way.
You should use either the coupleable interface's routines (see coupleable.h), or variable.getNodalValue() from the mooseVariableFE.h

[Traiwit]

yes it's in parallel

if i use getNodalValue(), how do i update the value like the one below:

solution.set(dof_index, _target_pwp);

[GiudGiud]

you can try this routine on the variable

  virtual void setNodalValue(const OutputType & value, unsigned int idx = 0) override;

Are you developing in opt mode?
You should hit an assert on an OOB if you are developing in devel (highly recommended) or debug (when things go south)
To change the compilation method, you can use METHOD=devel make -j6
or export METHOD=devel

[Traiwit]

Noted, Thanks @GiudGiud!

hmm I didnt know about developing in opt mode at all, dont even know what it does.

btw, is there any example object that I can have a look for setNodalValue?

it's a bit late for me now, I will have a look at them first thing tomorrow.

[GiudGiud]

in opt mode a lot of simple checks are removed for performance purposes.
in debug a lot of non-trivial poorly-scaling checks are added, degrading performance a lot
devel is a healthy middle ground, I rarely use the other modes

you can see setNodalValue being used in AuxKernel.C and in InitalConditonTempl.C