[petsc-dev] PetscCitations: software or underlying math

[Barry Smith]

   I think that —citations should be only citations; that is previous original work that deserves to be cited in the new publication because,  

   1) the new publication builds on/depends on the previous work

   2) the previous work is of some “quality” 

   3) the previous work is not yet common knowledge and already well cited. (so, for example, Newton, additive Schwarz and GMRES would not get cites) NASPIN could get a cite

   4) is where the original work was done, not a textbook that reviews the work (so, for example, if you did cite GMRES you would cite the 86 paper, not Golub and Van Loan)

  the original work could be analysis, algorithms or software

  Of course there is subjectivity in deciding these. 

   Some examples, the new GMRES variants, including pipelined (but not FGMRES) would get cites. FAS, Broyden would not get cites

   Additional reading and background material should be presented but not via the —citations mechanism. For example if we resurrected my https://bitbucket.org/petsc/petsc/pull-request/50/feedback-on-html-formatting/commits then when it displayed —xxx_view it could show links to manual pages with all the appropriate background reading. 

    Another way to look at it is that —citations is JUST a way to get petsc users to do reasonable citations without them having to do all the background research they would need to do to pick the right things to cite. 


   Barry



On Oct 24, 2013, at 11:31 AM, Jed Brown <jedbrown at mcs.anl.gov> wrote:

> Matthew Knepley <knepley at gmail.com> writes:
>> I think we can differentiate between mathematical background and analysis,
>> e.g.
>> 
>>  Proving that GMRES converges with these matrices, etc.
>> 
>> and showing exactly how to structure an algorithm:
>> 
>>  Saad and Schultz, 96
> 
> *86*, but what about GCR?  After all, GMRES is an incremental
> modification of GCR.  It also contains mispredictions like:
> 
>  "In practical implementation it is usually more suitable to replace
>  the Gram-Schmidt algorithm of step 2 by the modified Gram-Schmidt
>  algorithm"
> 
> If someone uses LGMRES, would we produce a citation only to Baker et al,
> or also to Saad & Schultz?  What about the BiCG family, containing many
> more variants that are slight variations on existing methods?  Or
> identical methods that were published twice under different names?
>