In several classes where I have offered "Hints for Beginners" in research, I have stressed the benefits of beginning with what I lovingly call "Paper P". When you anchor your first research on one paper (or a few), you get a number of benefits:
To excise a phraise from an essay by Don Davis, "current work is close to (but not exactly!) a sufficient statistic for what has come before." Davis encouages us to take
advantage of this, and my interpretation is that we should take our (well selected) paper P as a legitmate summary (approximate sufficient statistic) of the record of the past.
I usually do not stress the point, but if you do not begin with "Paper P" you have to face some potentially uncomfortable risks:
I have a few papers that I really like but which have had very little follow-up.
Sometimes you have simply followed your own curiosity without any identification of an audience. My paper with Dan Rudolph "Sizes of Order Statistical Events of Stationary Processes" was an invention of our own curiosity.
I thought the problem was marvelous, and I still think so. Nevertheless, it was a "green field" project, and, so far, nothing has been built around it. If you like order statistics and information theory, you can be the first to follow up. I may even do a sequel myself. Thirty years is surely a sufficient mellowing period.
Sometimes a paper just falls into the "cracks". For example. I really like my paper with David Boyd "Monotone subsequences in the sequence of the fractional parts of multiples of an irrational." It also appears in a famous journal (where Abel published his papers), but it has had very few descendents. Apparently, it fell into a crack between number theory and combinatorics.
It's hard to say why a paper falls into a crack, but there may be pointers. The related paper with Ellis "Fast Sorting of Weyl Sequences by Comparisons" has not been a home run, but it has contributed to developmens in computer science by Devroye.
Work in mathematics garners few citations compared with some other fields. To the generic scientist or engineer the citation rate in mathematics must seem prudish. On the other hand, the citation customs in clinical medicine must seem down right promiscuous. In clinical medicine the citing authors may have read the abstracts of the papers they cite, but in many cases I doubt that they have read much more.
In mathematics, one must be reconciled to having 10 citations for a "successful" paper and perhaps not more than 100 or so for one's best. For a paper to go past 100 is really a matter of the ``Giant component" and may no longer be a measure of impact --- but of "ripple impact."
One always hopes to put most of ones effort into projects that have good prospects for follow-up. Having an anchoring "Paper P" is no guarantee, but I think it helps more than one might imagine.