Monday, April 02, 2007

This week in cond-mat

Much as it pains me to admit this, I agree with Lubos Motl about something: Neither of us like the new numerical identifier system launched by Paul Ginsparg and company at the arxiv. Lubos nails both of my complaints. While the old system actually conveyed information (the subject area of the paper and how many papers in that category had been submitted that month), the new system manages to be both cryptic and uninformative. Frankly I don't care how many total papers have been submitted from all categories in the arxiv, and I'm not sure why anyone would. Somehow this reminds me of the apparent desire of the Powers that Be to switch NSF proposal submissions from FastLane, which works extremely well and is easy to use, to, which is completely arcane and annoying. Prof. Ginsparg, if you see this, please consider switching back to some incrementally changed form of the old system.

Meanwhile, here's one paper in the old numbering format, for old time's sake, that I thought looked interesting. Perhaps a theorist could take a look at this and tell me if it's as clever and neat as it seems to be.

cond-mat/0703768 - Ostlund, The strong coupling Kondo lattice model as a Fermi gas
The Kondo lattice is a model developed in an attempt to understand the heavy fermion compounds. In this model, there are itinerant conduction electrons, and a lattice of localized unpaired moments (f-shell electrons) representing the ion cores of the rare earth constituents of the heavy fermion material. Under the right conditions, the ground state of these materials is a Fermi liquid, meaning that there are distinct, gapless, electron-like (spin 1/2, charge -e) quasiparticles, but they have an effective mass hundreds of times higher than the free electron mass. The idea is that the conduction electrons have formed fully screened Kondo singlets with the rare earth f-electrons. The true ground state of the Kondo problem is a Fermi liquid, and in this limit the ground state of the Kondo lattice is also a Fermi liquid, though the antiferromagnetic screening of the ion cores leads to the high effective mass. Note that this is rather special - in semiconductors, the effective mass is a single-particle effect that comes from the lattice potential; in these systems, the effective mass is the result of many-body correlations. In this paper, the author explicitly writes down a canonical transformation (read: clever change of variables) that directly maps the Kondo lattice Hamiltonian into that of a weakly interacting Fermi gas. It looks clever to me, but I can't judge it in the context of other theoretical treatments of these strongly correlated systems.


gilroy0 said...

I'm not active in any field, so this is purely hypothetical to me. But I read the explanation on the site as to why they changed and -- while I'm not sure it would have convinced me -- it seems far from random or ill-thought-out. I can see why you'd like to have the archive reference convey classification information, but it seems that they deliberately suppressed that so that things can be re-classified when need be. In a world that will be increasingly mediated by electronic agents, decoupling meta-data from filenames is not a dumb thing.

An analogy: There's some information conveyed by the fact that your blog is hosted at IP (whatever it is). If your address was, I could (conceivably) look at that and say, "Ah, IP block 72. That's routed along such-and-such backbone and is coming from such-and-such place." But of course, relocating your blog would become essentially impossible if everyone had hard-coded the IP address. As it stands, you can change hosts, change machines, etc., and the magic of the DNS system makes it so that everyone can find you.

So maybe the real issue is: It's goofy to refer to the arXiv files directly anyway. There should be (maybe there is -- I told you I'm not current) some sort of DNS-like system, so that people could register their papers and the system would convert it into machine-relevant low-level data. This sort of goes on anyway, in that people refer to journal articles via a human-level shorthand (Laughlin 1998 or whatever) and not by full bibliographic reference -- at least, not every time. (Then the cited list acts like a local DNS to translate that shorthand into "legitimate" universal locator data.)

Put another way (he says grandiosely), the problem isn't the step that the arXIv took. The problem is, they didn't go far enough.

Just an unfounded opinion from an ex-patriate of the land of physics research.


Douglas Natelson said...

Bernie - I think you've got it exactly backwards. Sure, for arxiv internal bookkeeping it makes sense to have a numbering system that is very low level; similarly to your DNS idea, the arxiv interface itself does allow me to search for just condensed matter papers, for example. However, since people do need to actually refer to these papers in bibliographies, doesn't it make sense to use a reasonable high level descriptor when the end users have to refer to them? In your analogy, doesn't it make more sense to refer to my blog in "conversation" as rather than by some IP address? It seems like Ginsparg et al. have gone from a name-based high level descriptor to a low-level descriptor. In so doing, they've also broken some of their functionality. The "previous/next" links on the abstract pages used to take you forward and back within the subject category. Now they take you forward and back in the uninformative arxiv index. I don't want to go from a condensed matter paper forward to an astro theory paper, thank you very much....