[t.teschner]

Can anyone suggest good books on unstructured, incompressible solvers? I pretty much digested Versteeg/Malalasekra's book as well as Ferziger/Peric's book. I am looking for something more focused/dedicated on unstructured, incompressible flow solver development. (I also went through the book of Löhner for the grid generation / data structure aspects).

The lecture script of Murthy (https://engineering.purdue.edu/ME608/webpage/main.pdf) is already quite good and sort of what I am looking for, but it focuses only on SIMPLE methods (I am more interested in the Artificial Compressibility method). Or if you have any link to an example code (something easy to understand like a channel or lid driven cavity code, AC method preferably), that would also help (any papers where the authors have published their code would help as well).

[sbaffini]

Dear Tom,

what you've already found are actually the most relevant sources. While focused on compressible flows, i would nonetheless also consider:

http://store.elsevier.com/Computatio...9780080999951/

whose freely accessible resources also contain free source codes. Note also that the AC method can be seen as a special case of preconditioned density based solvers (the main difference, i guess, being that the latter are solved coupled, while probably AC still relies on segregation).

I am not aware of any source specifically focusing on the AC method, especially for unstructured grids. This is mostly because of the contrast between the flexibility of unstructured grids and the limitations of AC.

What kind of difficulties are you facing which cannot be answered by the sources you already mentioned?

Edit: I just had an illumination, as this:

Drikakis, Rider: High-Resolution Methods for Incompressible and Low-Speed Flows

actually has a section on AC. Still, i don't think it's explicitly focused on unstructured grids.

[t.teschner]

ah yeah, i forgot blazek, i guess i am looking for something like the blazek book but more for incompressible flows. i just wanted to see if i had missed anything, there is nothing particular that i am looking for, i was just looking more for a reference book dealing only with unstructured approaches for incompressible flows. the book of drikakis and rider is more an overview about different methods but not what i am looking for (i am good with the AC method, i am more looking for the implementation aspects)

[FMDenaro]

As I can remember, the literature is focused on pressure methods, I don't know how much you can find about AC.

Just a simple search on google confirm that: https://www.google.it/search?q=incom...JsHA8gfU1r6wBw

Try to see also on the CTR site (Annual briefs and Summer schools)

[sbaffini]

Well, just to point to additional sources you might not know, this book just came out last year:

http://www.springer.com/us/book/9783319168739

and it seems pretty focused on implementation (even if, somehow, biased toward OpenFOAM). But, honestly, i don't think it adds anything relevant to the sources you already mentioned. It's just another point of view of the same matter.

[sbaffini]

Originally Posted by t.teschner

the book of drikakis and rider is more an overview about different methods but not what i am looking for (i am good with the AC method, i am more looking for the implementation aspects)

If you can connect the dots between the AC theory in Drikakis, the preconditioned density based method in Blazek and the actual blazek codes... well, i guess you don't need much more.

[FMDenaro]

Originally Posted by t.teschner

ah yeah, i forgot blazek, i guess i am looking for something like the blazek book but more for incompressible flows. i just wanted to see if i had missed anything, there is nothing particular that i am looking for, i was just looking more for a reference book dealing only with unstructured approaches for incompressible flows. the book of drikakis and rider is more an overview about different methods but not what i am looking for (i am good with the AC method, i am more looking for the implementation aspects)

why do you like AC the most?

[t.teschner]

Originally Posted by sbaffini

If you can connect the dots between the AC theory in Drikakis, the preconditioned density based method in Blazek and the actual blazek codes... well, i guess you don't need much more.

yeah thats what i figured, i just wanted to check if there is anything that i missed (literature wise).

i am not particularly interested in the AC method by itself, but we have an improved method that we use (both are hyperbolic which have some pretty neat advantages). i am looking at porting my structured version into an usntructured one so i am just collecting literature at the moment and wanted to see if i had missed anything, so thanks for all your suggestions

[sbaffini]

I can only think of two aspects which are more difficult, intrinsically complex, and require more coding for unstructured methods than structured ones.

One is the parallel aspect. Specifically, knowing which processor owns which cell etc. without storing large amounts of data per processor (scaling with number of processors or cells), is not trivial as for structured methods.

The second ones are some specific algorithms (e.g., find the grid cell containing a given point, grid-to-grid interpolations, etc.), mostly related to computational geometry. Handling robustly all the possible cases is the issue here (possibly without writing routines with thousands lines of code).

In my experience, these topics are not covered exhaustively in any reference and open source codes also typically lack some features you might need. Here you will typically need some thinking before starting the implementation.

The same also, somehow, applies for the overall code organization as people with the structured approach in mind tend to produce bad unstructured code. I found useful the following approach: store less, access less, use more, compute more. Obviously, the underlying architecture is going to play a role here. For example, a common approach is to store variables separately: U(1:ncells), V(1:ncells), etc. But, in some cases (e.g., coupled solvers), an approach like VARS(1:nvars,1:ncells) and GRAD(1:ndim,1:nvars,1:ncells) will produce much less cache misses and vectorizes naturally (or quite so). Thinking Object Oriented here (which is not typically promoted by the structured approach) will also help the code organization (such a VARS, for example, can be naturally generalized to different solvers without rewriting everything else).

Besides these topics, my experience is that unstructured methods can be implemented just as described in books and are far more simple than structured ones.

Also, they are much better from the software engineering point of view. Not only because you need much less code, but also and especially because the code that disappears is that related to special cases (i=1, i=N, etc.) or to stuff repeated 3/6 times (i.e., face loops).

So, i guess, my suggestion is just go for it.