What papers can you ask someone to cite for using your software?

Software authors often request that you cite a paper of theirs if you are using there software.

But what logic governs what kind of papers they can reasonably ask to cite? I see 5 cases:

1. 
   

   The paper is about the software explicitly. Eg Theano

   
   


2. 
   

   The software is a direct implementation of an algorithm described in the paper. Eg anything from here

   
   


3. 
   

   They created the software to use in research presented in the paper. It has not changed since.

   
   


4. 
   

   They originally created the software for use in the research paper, but since then it has undergone fairly large changes as it has continued to be developed. Eg DeepLearnToolbox

   
   


5. 
   

   The paper has nothing in common with the software, except sharing an author.

   
   

In which cases is it acceptable to request the paper be cited? 5 seems like an obvious no-no, but are the others in the clear?

This question is the flip-side of Should I cite Conference Papers to acknowledge usage of tools?