Posting preprints to arXiv

[CauchyProcess]

How close to complete does a particular research topic have to be for it to be appropriate to post a preprint on arXiv?

 

I'm in the following situation. I've developed a new computational algorithm for solving certain problems that is completely parallelizable, but do not yet have a proof of convergence. I have strong numerical evidence of convergence, and very good theoretical justifications to believe that the algorithm should always converge, but writing down an actual proof is highly nontrivial - most likely to the point that me and my advisor would need to collaborate with a specialist (likely a pure math professor) in order to work everything out. My advisor is asking me to write up the algorithm, together with practical results on how to initialize it, tune it, etc., and post it on arXiv so that people can start looking at it and/or using it, but personally, I'm hesitant - it just seems early, premature.

 

What is the norm regarding preprints? Is it appropriate to write up a short document describing the algorithm and post it on arXiv even without a proof that said algorithm works? Is arXiv the correct place, or is this better suited for something like a tech report?

Edited August 13, 2015 by CauchyProcess

[TakeruK]

I am going to tentatively give an answer based on my field (we also put preprints on arxiv) but from my past experience on the math/stats board, it seems like the way it works in my field tend to be different from the way it works in maths/stats! But I'll say how we use arxiv just to start the discussion.

 

In my field, people use arxiv for a variety of reasons. The most common use is to wait until the paper is already accepted by the journal and then put it up on arxiv so people can read it right away instead of waiting for the 1-3 months it takes between acceptance and actual online (or even worse, print) publication. This also allows people without journal subscriptions to access your work for free (journals in my field all grant authors to put copies of their published articles on arxiv, as long as it's the version the author made themselves, not the fancy PDF created by the journal).

 

However, another use is to put the submitted paper on the arxiv and specifically ask for comments so that you get "real" peer review alongside the journal's peer review process. An alternative version of this is to put a version on the arxiv prior to journal submission so that you get community review, incorporate those thoughts and then submit it to a journal. Of course, if you do this, you run the risk of someone trying to do the same thing as you scooping you! But in astronomy, sometimes the science is only possible if you have a certain telescope data set and as long as you are the only ones with access to that data, you are safe.

 

Putting this information on the arxiv earlier also means that you can establish priority earlier. However, in my field, you will really need to follow up that arxiv posting with an accepted paper soon, because most journals will discourage their authors from citing un-peer-reviewed work. So an early version on the arxiv is a good way to get people's attention and cause them to look out for the peer-reviewed paper coming soon. It may be a good idea to time these types of postings prior to, during, or right after a major conference.

 

And most rarely, some people will post articles on the arxiv that may be useful to the community overall but not really worth a peer-reviewed paper. For example, an astronomer wrote a very nice PDF describing distance estimation methods as a review article that is on the arxiv and many people refer to it and use it but it is not a peer reviewed article. This still brings many benefits to the author of this arxiv post though, as many people appreciate their work and recognize them as an expert. 

 

To answer your specific question about your specific case, I think that if you have developed something useful to other people, if you are sure that it is correct, then in the norms of my field, it would be better for you to share now rather than waiting. arxiv could be a good venue for this--at least there is an easy way to cite your work and arxiv is indexed. I'd go with your advisor's advice. In my field, you would eventually want a peer-reviewed article that people can cite so that you get credit for your cool tool, but arxiv citations aren't that bad either. 

[compscian]

I also work on "computational maths", and I faced a situation very similar to yours about a year back. We also had an algorithm which seemed to work very well on a problem studied in many areas from mathematical biology to operations research. However, we had no proof of correctness or proof of convergence (we had the intuition and numerical results but not the proof).

 

We ended up posting it on arxiv and also sent it to a conference where it was accepted. After another 4-5 months of work, we finally had the proof and also tested it on some nice large scale applications. We ended up writing another paper which we put on arxiv and sent to a journal where it got accepted.

 

Bottom line, if you feel that your work is at least worthy of a conference publication, put it up on arxiv. You may end up getting two papers like me. However, if you feel that your method will not be accepted or well received without concrete proofs (this ranges from field to field - in my case, numerical simulations were acceptable evidence), then you might want to hold off for sometime and upload a complete paper. It is also worth mentioning that you can always update your arxiv paper after you complete the proof. The only point is the date of contribution will not be backtracked (which is totally obvious and fair).