So, I thought I would go back to math briefly; as great as it is to get so much energy and comments around a sexier issue like journals, I don’t want that to entirely take over the blog. Fortuitously, I’ve also got a math topic I want to cover; Catharina Stroppel and I have just replaced our paper on two-block Springer fibers on the arXiv, with significant enough changes that they deserve some mention. (see the previous post on this here).

Most notably, now our main theorem is a geometric construction of Khovanov’s algebra, not some weird signed variant on it, which is a *good thing* given how the general picture of this stuff is supposed to fit together, but a tiny bit disappointing for me, since I had grown a little fond of the guy with weird signs. He always struck me as a scrappy underdog.

Now, we had known for a while that there was something a bit off about the previous version of the paper. It isn’t precisely that the theorems in there were wrong, but rather than they were the wrong theorems; we now give a geometric construction of Khovanov’s algebra, but it’s a subtly different construction than in the previous version which didn’t give us Khovanov’s algebra at all. Remarkably, a slight change of perspective suddenly popped everything that had been subtly off back into place.

I feel a lot better about the whole business now, though it is slightly embarrassing to have given a number of talks on the subject when I had a less than complete understanding of what was going on. So, if you happened to see one of those talks, you may want to blank the last few minutes of it from your memory, and replace that with what’s now in the paper.

So, what changed in our construction? The secret is the orientations. Previously, we had defined a convolution algebra structure on the cohomology of the intersections of components as follows:

Let be the disjoint union of the components of the Springer fiber, and let be the Springer fiber itself. Thus the fiber products and are the disjoint unions of the double and triple intersections of all doubles and all triples of Springer components. Of course, we have projections

given by forgetting a factor (including triple intersections into double intersections in the obvious way).

One can define a product on by the formula

.

But there’s a subtlety here: it uses pushforward of cohomology classes, which uses Poincare duality, which uses an orientation of your manifold, so you have to choose an orientation of everything in sight.

Now, when you’ve been thinking too long about complex things, you start to think pushforward is a canonical operation, since complex things always have a natural choice of orientation. This is the *wrong* perspective to have on things, and it messed us up for a long time.

Instead, you have to remember that orientations are something you choose, and how you choose them changes your algebra (in particular, whether it’s associative), so all our problems before arose from the fact that if you give everything the complex orientation, the resulting algebra is *not associative.*

Instead, you have to choose your orientations carefully, something I’m still working on how to do in general. In this particular case (and in the hypertoric case, where one has to do something similar), there’s an easy combinatorial prescription, but it’s not clear how to generalize it. Once you do this, however, all your problems melt away, and you get the right answer.

This still leaves unresolved the question of what this algebra means….I’m hoping it’s something about the Fukaya category, or failing that, coherent sheaves, but those are questions whose details will have to wait for another day.

Now, I’m really confused about the Ext algebra in the category of coherent sheaves (I mean the Ext of the direct sum of the structures sheaves). I thought that you and Rina had decided that it was equal to your sign changed Khovanov algebra. On the other hand, it is definitely associative and now you say that the sign changed Khovanov algebra is not associative!

The answer is that I’m confused too. At least now we just have one logical guess, but it’s still unclear to me how one really nails down the Ext-product in the category of coherent sheaves. In the last section, we define a interesting t-structure on the coherent sheaves on the convolution variety from the affine Grassmannian, and show that the combinatorics of that category is rather like that of category O, but that’s just multiplicities, not equivalences.

Actually, the Ext algebra _is_ a sign changed Khovanov algebra (I have included this in the current version of http://arxiv.org/abs/0711.4409). It is associative. It does not quite coincide with Ben and Catharina’s variant, but the discrepancies are awkwardly insignificant.

BTW, I’m in Boston for this week, so we could meet and discuss this at some moment.