Problem when using Kernel Gradient Correction in 2D Mode

philst_germain

Hi,

I have noticed that simply turning on the kernel gradient correction for a run in 2D makes the fluid particles to blow up in all directions. Could there be a bug in the code that causes this?

Thanks,

Philippe St-Germain

Alex

The kernel gradient correction is implemented in the code to be formally correct from the point of view of Maths.

However, in the practice, it is not recommended to use it since it is very sensitive to the case you are working on.

I suggest to turn off it always.

elisewin

Hi,

first of all I would like to thank all the developers for their nice work and the new release of the Code.
I am interested in a feature that was actually available in the old release (v.2), the kernel gradient correction.
It is probably no longer supported...I would like anyway to ask you informations about that.

I can read in this thread that it has given instability issues and its usage was not suggested;
were these problems confined to 2D geometries or did someone experienced it also with 3D geometries?
I actually tried [with KGC on] the 3D dambreak [your benchmark] and it runs without problems.

"However, in the practice, it is not recommended to use it since it is very sensitive to the case you are working on."
is there some kind of "black list" of test cases which has given problems with this correction? Do they have common features? I don´t know, tc with moving boundaries or other features..

And what is the reason at the origin of this instability/at what "point" of the correction the problem occurred?

Any hint would be really appreciated, I thank you all in advance

Best regards

Mojtaba

Dear elisewin,
Hi,

For your information about use of KGC in SPH you could read this paper:
Title: "Completeness, conservation and error in SPH for fluids"
Written by Vaughan, 2008

Regards,
Mojtaba

elisewin

Hi Mojtaba,

thanks for your suggestion, I have a look at the paper.

Kind regards

Alex

Hello

as stated in the work of Bonet and Lock 1999, the kernel correction guarantees the conservation of the linear momentum while kernel gradient correction guarantees the preservation of the angular momentum

if you were testing an example like 2-D lid-driven cavity flow without gravity as shown in https://wiki.manchester.ac.uk/spheric/index.php/Test3 then you must be worried about the conservation of angular momentum

however in most of the applications we study with DualSPHysics there is no much need of being worried about that and in the end, applying the KGC makes the code more complex and difficult to maintain and to extend to the GPU implementation.

that is why we remove it. we included it first since it is a prove of being mathematically correct, but in the practice makes the code slower and even inestable for some engineering problems

you can play with testcases of DSPH v2.0 and see the difference in the simulation when using KGC or not and you will not find different accuracy, only the same or worse

so , it will depend on your application. if you are a matematician and you are studying empirical solutions with analytical problems such as the one mentioned above, go ahead.

Regards

Alex