For all the doom and gloom expressed by many here, my impression is that mathematics is far closer to being able to do without the big publishers than many other fields. (Spend some time reading what the medics have to say on blogs like Bad Science if you want to see how bad it could be.)

For one thing, (almost) all mathematicians use latex, which has the potential to bring typesetting expenses right down. There are also some good examples of journals run essentially 'by the community'. It may not be actually open access, but Geometry & Topology is run not for profit, by mathematicians, and is of a very high standard.

One Site to rule them all,
One Site to find them,
One Site to bring them all
and in the darkness bind them

Very laudable. But very difficult to set up and get going - who's going to look at such a site before it has lots of information on it, and who's going to put the information on without the surety that lots of people are going to look at it? Parts of this idea is what we discussed on the rForum about a year ago.

(Oh, and +1 for the sarcasm: "It is even conceivable that the beneficial things that journal publication gives us (peer-review and some measure of the importance of a paper) could be incorporated into such a unified model.".)

Surely an ultimate goal would be for the arXiv, mathscinet, and electronic versions of everything published in journals to be all completely managed together in one nice cohesive whole. One place to see all the papers, all reviews of those papers, updated versions and errata, as well as the papers currently being worked on.

Laudable, but I can't help recalling a simile I once heard used by a statistician: trying to get mathematicians to do anything concertedly was supposed to be "like herding cats".

(I am also instinctively worried about quality control, but am aware that other people have different opinions on just what journals do and don't provide here.)

arXiv versus google scholar: I must admit that I don't use google scholar, I've yet to need to. MathSciNet, the arXiv, and just plain old google have done me fine so far. One reason that I hesitate about using Google Scholar is that I know nothing about how it indexes stuff. At least on the arXiv, the author of the paper has chosen how it should be indexed and what searches should find it. So I prefer the arXiv because I presume that the author of the paper knows who ought to be reading it. (It's worth pointing out that the arXiv now has full text searches as well)

arXiv versus MathSciNet: Of course, if I want a really good idea of what you were working on five years' ago, I'll use MathSciNet. If I want to know about current research, well, it's not so good.

arXiv versus journals: No contest. If journals actually served a useful purpose internal to academia then there might be an argument to answer here. But they don't, so there isn't.

arXiv versus archive: I find Andy's argument astonishing. Firstly, the idea that someone might actually be interested in what I wrote in 100 years' time is a strange one. Very few people read the actual originals except for historical interest. We've largely subsumed the maths of 100 years' ago and we have new books and articles summarising stuff. Secondly, although paper as a technology might have survived that long, for actual bits of paper the survival process is a lot more haphazard and random. We tend to rely on bits of paper that have been used for other purposes, stuffed down the back of a mattress for 50 years, and then not quite used as a fire-starter! If a journal actually has an archival policy, then great, but I've yet to hear of any journal that does that (which isn't to say that they don't, just that they don't make a big fuss about it). A far better strategy would be to have a central body with responsibility for this, in which case they could just print out stuff from the arXiv as well. Thirdly, I certainly don't publish in journals for archival purposes. It's never crossed my mind and I'd be astonished to hear that I was in a minority here. I publish in journals because I'm told to. Otherwise, I wouldn't bother.

That's not to say that the arxiv is perfect. It's not, far from it. But it's the best system we have by a long way.

It's easy to say what is wrong with it. It tries to do two things: it tries to be both sourceforge and freshmeat. Sourceforge (for those that don't know) is where programmers can host their programs while they develop them. You can track a program, keep downloading updates, and so forth. Freshmeat is for announcing and indexing programs. You can announce a particular release of a program, keep up to date with latest stable versions of programs that you're interested in, and so forth. Of course, there's overlap, but as with such things keeping distinct roles in distinct places helps with "mission creep".

Of the two, I think that the announcing and indexing part is the key one. I suspect that the arXivers do too, since (to counter BCnrd's argument), updates are announced separately to new papers, and only 5 updates are announced - after that, new versions don't get on to the list. I check the "What's new" page each day and bookmark what look like interesting papers to go back and read at leisure. If someone just puts something on their web page, how am I supposed to know it's there? Okay, I can do a google search but I can only do that if I know what to look for! This is particularly important for someone like me who works on a bit of a boundary. I'm a topologist, but I use quite a bit of functional analysis. Because I don't actually do research in functional analysis, I don't really know what goes on there, or who the major people are. But I want to know of any results that might be useful to me, so I need to keep an eye on what's being done. The arXiv is invaluable to me in that.

So please, please, please go back and put everything you've ever done on the arXiv. Of course you don't have to update every time you correct a spelling. You can even put a link in the 'abstract' field to your webpage saying, "For the most recent version, see my webpage". At least then someone will know to look for it!

Go on. Do it now. You weren't doing anything particularly important right now, were you, otherwise you wouldn't have been reading this forum.

I don't think I have any arguments against any of your points. As I wrote, I am reasonably convinced in theory about the virtues of posting to the arxiv; I simply haven't gotten around to doing this in practice.

One thing to remark is that posting to the arxiv is relatively less common in number theory than in other fields, I think (although it is becoming more common over time).

(By the way: I find it strange that "not being able to update the paper as often as I need" is an argument against the arXiv: I doubt the proponents of those observations update their printed papers! :) )

I also don't post on the arxiv (with the exception of joint papers posted by a coauthor), out of a mixture of laziness and a similar concern to Brian's about not wanting multiple versions to pile up, and I do post on my web-page. However, it has been pointed out to me that for people (mathematicians, but perhaps even more students) who are not in the thick of the action, and so who may not know whose web-sites to look at, the arxiv is a more universal (and so more democratic) place to post. I haven't found this argument compelling enough to change my behaviour (yet), but it has made me feel somewhat guilty about not posting on the arxiv.

mathematicians