[sega]

Hello World.

I made a strange experience some days ago.
I started a two-phase-flow simulation not knowing how long it will take so reach its final equilibrium state.

My first, second and third guesses of the simulation time were too short.
In all cases I restarted the calculation by putting the latest time into the startTime entry.

Due to insufficient disc space I had to recalculate the simulation on another system.
As a matter of fact I knew the correct simulation time and simulated straight from 0 to the final timestep.

I compared a characteristic value of the two simulations (magU in this case) and found that they differ over time!

Interrupted simulation:



Continuous simulation:



Strange to me are the tree peaks at the interrupted simulation plot.
They are not exactly at but following a short time after the three interruption-times.

What went wrong? Did I made something wrong?

[niklas]

did you write your data as ascii or binary?

[sega]

Ascii in both cases.

[fra76]

Use binary, and run in double precision.
Your restart will behave much better!
I guess this is what Niklas was thinking of...

Francesco

[niklas]

It is

restarting in ascii files is nasty business.
You lose alot of the accuracy, which you just have witnessed I guess.

I still remember the trouble I faced when I was implementing the spray and saved the data in ascii...

If a drop is close to a face it might actually end up in another cell when you do a restart, because of the round-off errors.

would be nice to see what you got when you use binary.