Minor issue related to [soln-plugin-residual] nstep - pyfr-1.4.0

General
1

Hi,

I have just found out a curious thing that when changing the nstep value to 1 (unity) in [soln-plugin-residual] closure, the code does not produce residuals at all in the output file, a *.csv file is created but stays empty.

Best regards,
Piotr Prusiński, M.Sc. Eng.

2

Hi Piotr,

I have just found out a curious thing that when changing the nstep value
to 1 (unity) in [soln-plugin-residual] closure, the code does not
produce residuals at all in the output file, a *.csv file is created but
stays empty.

This is due to the design of the residual plugin which estimates the
residual through:

  residual = (prev_soln - curr_soln) / delta_t

with the plugin itself having two phases: the first phase is run one
time step before we are due to save out the residual and its purpose is
to assign:

self._prev = curr

and then in the second phase (run on the next time step) we actually
compute an approximation to the residual itself.

The problem when nsteps = 1 is that in a single time step we have to
perform both phases. This confuses the code as it is currently written.

I can certainly update the plugin to support this. However, generally
speaking nsteps = 1 is extremely frequent in terms of residual output.
Usually, we would not run below nsteps = 50 given that with an explicit
code such as PyFR the time steps themselves are usually extremely small.

Regards, Freddie.

3

Ok, no question :slight_smile: Anyway, such an information could be introduced somewhere in the manual that the nstep is in fact not an integer but rather an “even integer” starting from 2 and then wider explanation like yours.

Best regards,
Piotr

W dniu czwartek, 28 kwietnia 2016 00:11:45 UTC+2 użytkownik Freddie Witherden napisał:

4

Hi Piotr,

Thanks for the feedback.

We can add a note indicating nsteps > 1. But I don’t think nsteps has to be even?

Cheers

Peter

5

Ok, now I got the point. The issue comes when the nsteps is equal to 1 and starts with the very first step as it cannot read a previous step. Now it’s clear :slight_smile: So yes the “nsteps > 1” will be a clear definition of what is allowed.

Thank you and best regards,
Piotr