What are Deal.II killer features when comparing against Comsol?

[Konstantin Ladutenko]

Any Comsol users switched to Deal.II to share their experience? Or using both Deal.II and Comsol on everyday tasks?

Some questions behind this: 

1) How I can motivate my colleagues to try Deal.II? 

2) What are possible arguments to invest time into Deal.II instead of buying one more Comsol license?

My thought on this (still I am not actually using any of them, so any comments are welcome):

1) License cost?

Comsol license is quite expensive. However, if you put the scientist salary on the board, it is not that straightforward.  With Comsol it is quite easy to become productive from day zero. To achieve the same technical level with Deal.II you will need few weeks or even months of everyday use... There are so many technical details under the hood hidden from Comsol user which you are forced to be aware of with Deal.II. And even after mastering them both it looks like using Comsol to create a new model (with new geometry and physics) is cheaper compared to deal.ii as soon as you already have a license. And this works even better when you scale up - for sure you need much less number of licenses than a stuff count in the lab..

2) Control. 

This is a real benefit from my point of view. You can put the C++ into version control system and get all the improvements with simple diff. Probably Comsol has some kind of "model review mode" to show all check boxes and list choices, however, I was not able to locate this feature fast.

3) Adaptive mesh with hanging nodes.

It seems to be a unique Deal.II feature to be able to do a local refinement or coarsening of the model. For sure it is very good, especially for large-scale simulations.

The side effect is that it is very natural to do hp-refinement. For time domain problems it is easy for Deal.II to follow with dense mesh alongside with places, where it is actually needed.

4) Large models for HPC

This is mostly related to engenering tasks. However, if you have a large problem it can be easier to solve it with Deal.II using HPC cluster.  Using 100-1000 cores or more to get the result can speed up the progress a lot, and such kind of clusters are quite affordable our days, it is quite easy to get a HPC access for academic research. It looks like it is not feasible with Comsol (at list I was not able to get data on more than a 20x parallel speedup with 0.8 efficiency for Comsol 4.2  in the internet). 

5) Comsol is an integrated environment.

So it has the drawing, meshing, setting the model, evaluation, visualisation in the same window, which is rather convenient. Actually this an argument aginst Deal.II, where you should mostly rely on external tools for most of the tasks. It is not a crucial problem, however, it complicates it a bit, e.g. to set up a working environment or to become really productive...

Any comments on this? More killer features of Deal.ii are welcome!

Best regards,

Konstantin Ladutenko

[Wolfgang Bangerth]

Konstantin,
I have never used Comsol, and so can't really answer your question. But I
think you may enjoy reading some of the answers we got on the user survey and
which are listed here:
https://docs.google.com/forms/d/1K2nhtfGuf15Cdw6E4_nbPk82zS9-XNfuRqOjFBWkp7E/viewanalytics

Best
Wolfgang

> --
> The deal.II project is located at http://www.dealii.org/
> For mailing list/forum options, see https://groups.google.com/d/forum/dealii?hl=en
> ---
> You received this message because you are subscribed to the Google Groups
> "deal.II User Group" group.
> To unsubscribe from this group and stop receiving emails from it, send an
> email to dealii+un...@googlegroups.com
> <mailto:dealii+un...@googlegroups.com>.
> For more options, visit https://groups.google.com/d/optout.


--
------------------------------------------------------------------------
Wolfgang Bangerth email: bang...@colostate.edu
www: http://www.math.colostate.edu/~bangerth/

[Konstantin Ladutenko]

Dear Wolfgang,

Thank you for your answer! I followed the link, the most interesting part, in the context of my question, are answers to "Why did you decide to use deal.II?" However, this survey is mostly among Deal.ii users, who already use it and like it. So I tried to estimate the total size of simulation software community by simply counting paper mentioning the product and indexed by Scopus in 2015 for names titled in Wiki page on FEM software. The result:

ansys	6413	53,02	%
comsol	2289	18,92	%
Ls-dyna	1211	10,01	%
OpenFOAM	868	7,18	%
abaqus fem	717	5,93	%
nastran	221	1,83	%
freefem++	183	1,51	%
deal.ii	95	0,79	%
code_saturne	67	0,55	%
libmesh	32	0,26	%

So deal.ii is the second in the list of generic FEM free codes (after FreeFem++) that looks to be quite good. However, most of the new knowledge generated with FEM is done using Ansys (which is widely used in industry) and Comsol (more dedicated to scientific research) are cleary dominates here. I have probably missed some titles, however the general situation should be clear. So, any more reasons (in addition I have mentioned above) to prefer Deal.ii?

Best regards,

Kostya

 

понедельник, 26 сентября 2016 г., 4:54:20 UTC+3 пользователь Wolfgang Bangerth написал: