Since I'm on vacation these days, I decided to finally submit my published research papers to arXiv. A previous attempt was about as pleasant as a visit to the dentist, but somewhat longer (and unsuccessful). This time around I tried to submit other papers, and things went more or less smoothly, as soon as I learned to follow some guidelines (for documents produced using LaTeX):
- Make sure that all .eps figures are of reasonable size: no files above 6MB and no more that 10MB for the whole submission. I downsized some very large files by first converting them to .pdf and cropping to remove white margins (using Acrobat) then opened the .pdf files in Photoshop and saved a copy in .eps format (without preview and using jpeg compression).
- Check that the figures are correctly invoked in the .tex file (the name should be case-sensitive, something that is not required on Windows systems). File names should not contain special characters (more details).
- Run LaTeX locally until the compilation is error-free.
- Create a .zip archive with the .tex file, the .eps figures and the .bbl file (the compiling process on arXiv does not include Bibtex) and, if necessary, with supplementary material in .ps format.
- Hope that the compilation works without errors, otherwise wade through about three screens of output, fix things and reload files. The supplementary files are simply appended to the final .pdf; this is an easy way to include additional material or tricky parts that fail to compile on the arXiv server. This is how I managed to add to one paper a page-wide table in landscape orientation, which had resisted all other methods.
The final result is here.
For Word documents the process should be much easier, since one can directly submit the .pdf version. I only tried this for one paper, currently on hold because it has line numbers in the margin. This restriction is not mentioned anywhere on the arXiv site (neither is the size limit, by the way.)
Overall, the entire procedure was easier than I thought. Still, the system is far from user-friendly (metadata retrieval using the DOI would be nice), the interface is firmly stuck in the 90s, some limitations seem a bit arbitrary and the help could be more detailed.