If a paper you are linking to has an associated Digital Object Identifier (DOI), please use the http://dx.doi.org/ address to link to it instead of the publisher's address. It is intended to be persistent, even if the journal changes publishers, or the publisher changes address formats (e.g. Wiley and Taylor and Francis).

Publishers usually either give an explicit DOI link; if not, one merely appends the DOI indicated in the page to http://dx.doi.org/.

Take a look at a similar discussion on TP.SE. DOI isn't always a good idea alone because the documents it links to are not usually free to read. My opinion was that, when possible, a DOI link and a link to a free repository (in that case arXiv) should be provided if available.

• The key, of course, is "when possible". :) People should feel free to provide a free source in addition to the one from the publisher. Dec 4 '11 at 16:03
• Provided of course that the free source is legal! Self-published on a uni website would be fine, but a scanned knock-off of a book available from some server would not be. That point aside, I think that it's very very important to link to free sources if possible because not everyone here is in a big company or uni. Anyway, I wouldn't trust anything, even DOI, to be as persistent as IDF claims. Dec 4 '11 at 16:11
• DOI is really an identification number. The DOI link is something separate. I presume it redirects to the publisher's website because that is considered the canonical source, but you can still use a DOI number to look up alternative versions of the documents which may be freely available. To me the appeal of DOI is that it's the closest thing to a cross-disciplinary "standard" identifier. (arXiv eprint IDs may be more common but that's only for subjects which are hosted on arXiv. Then again the arXiv abstract page does contain the DOI link...) Dec 6 '11 at 9:16
• @DavidZaslavsky: You're confusing the ideal with the reality. A DOI always links back to the publisher website. I have something like 1000 articles in my library, and all those with DOI (which is not a number as many other characters are allowed) link to a single source (via [dx.doi](dx.doi.org or CrossRef usually). I've never encountered any other behaviour when resolving a DOI. Of course the DOI is different to a resolved DOI, but the two are synonymous, even on CrossRef. Dec 6 '11 at 10:42
• Well, I was using "number" in a general sense ;-) The point is that the string "10.1103/PhysRevD.60.114023" (to pick an arbitrary example) can be used for more than just constructing the link dx.doi.org/10.1103/PhysRevD.60.114023. That particular link goes to the publisher's website, but you can also, say, search Google for "10.1103/PhysRevD.60.114023" and the arXiv page is the fourth result. The point is that having the DOI does allow you to look up other sources for the paper, including free ones. Dec 6 '11 at 18:30
• @David: You're right of course. This does rely on free sources listing the DOI though. In the case of arXiv, the authors add it, and sometimes forget. You're a little more likely to succeed simply by googling the title. Dec 7 '11 at 0:12