When to update an Arxiv posting vs. creating a new post?

Question

I have a paper on the Arxiv which has recently been referenced in someone else's published work. I have since used the new work (of someone else) to significantly extend my current Arxiv work. Since my previous Arxiv posting has not yet appeared anywhere, I am sending the old work along with the extensions to a journal in a single paper. So my question is this: should I update the Arxiv listing for the previous paper with the new extended version as a revision or should I add it as a new paper?

I guess my worry is this: let A be my paper, B be the other paper that references A and A' be my extended paper which builds on some ideas from B to significantly extend A. If I post A' as a revision to A, then there is a weird referencing problem where B refers to my Arxiv post without knowing that A' builds on B. However, if I post A' as a new paper, then it includes essentially all of A in it (though presented in a more clear manner) and so there seems to be some duplication.

Answer 1

I agree with Mangara that there's no problem with updating the previous paper if your new version subsumes the previous one. It's understood that results in preprints are sometimes improved before publication, and so this won't trouble anyone. Furthermore, there are several reasons why you should (and not just can) do this:

1. It's best not to leave incomplete versions of the paper lying around, since someone may run across them and never realize there's a better version elsewhere. For example, it's better if readers following the reference from B learn that there's a new version, rather than just seeing the old one and potentially assuming that's all there is.

2. Posting very similar papers can look bad, like you are trying to inflate your paper count with minor variations. That's not your intent, but someone browsing through your arXiv papers won't having any way of knowing. You might also trigger the arXiv flag for text overlap between your papers, in which case the similarity will be explicitly pointed out to everyone.

The comments field in your arXiv submission can help address the potential for confusion. Whenever you make major changes to an arXiv paper, it's a good idea to explain in the comments what has changed, so that someone who has already read a previous version knows whether it's worth another look. Highlighting the changes in this way will also help clarify things for readers following older references.

Answer 2

As long as the results of A are a subset of A', I see no problem with posting it as an update.

To avoid confusion due to the circular references, you could mention something like "Paper B builds on an earlier version of this work" when you first cite B.

Answer 3

I at one point took the route of adding a new paper, and had it come out badly, so I'd recommend replacing. In summary, we had paper A, a lengthy preprint which developed a lot of theory in a somewhat confused order. We then realized that a large fraction of the results could be proved in a more general context and in a much cleaner way, which became paper B. However, there remained a number of elegant combinatorial arguments which only worked in the original context, and became paper C.

I have had to tell a lot of people not to read preprint A, just B and C (which are now published). In retrospect, I think I should have made C a replacement for A. It seems that the argument for this is strongerin your case, where you are only replacing a paper by one paper, not two.

Answer 4

Ideally the someone-else should have included the version number of the arxiv posting. In that case, there is no possibility of confusion.