8
$\begingroup$

Let $R$ be a reduced root system, $W$ the associated Weyl group, and $w_0 \in W$ the longest element of $W$. In general $w_0$ admits more than one reduced decomposition into a product of reflections, a number which we denote by $d_R$. Where can one find a list of values of $d_R$ for low-dimensional root systems?

For example are the explicit values of $d_R$ known for the exceptional root systems?

$\endgroup$
  • 1
    $\begingroup$ For Types A and B(=C) there are product formulas for these numbers: see the famous paper math.mit.edu/~rstan/pubs/pubfiles/56.pdf and math.stackexchange.com/questions/2271510/…. I'm pretty sure for Type D there is not a product formula (as Stanley mentions there is a big prime, 193, in the factorization of the number of reduced words of the longest word in Type $D_4$). As for exceptionals I don't know of a list but this is in principle something a computer can do. $\endgroup$ – Sam Hopkins Aug 28 at 16:36
  • $\begingroup$ In the linked stackexchange webpage the B-series formula in the answer gives, for low values of $n$, a product with negative limits. Is this an error or is this an example of some convention I am not familar with? For example, what is the value for $B_2$? $\endgroup$ – Bas Winkelman Aug 28 at 17:07
  • $\begingroup$ Maybe the formula Zach wrote is not quite right. For $B_2$ the answer should be 2. It is the same as the number of linear extensions of the root poset (poset of positive roots whereby $\alpha \leq \beta$ if $\beta-\alpha$ is a nonnegative sum of simple roots). This poset is the same as the shifted trapezoid shape $(2n-1,2n-3,...,1)$ poset. This number also happens to be the same as the number of SYTs of $n\times n$ square shape. $\endgroup$ – Sam Hopkins Aug 28 at 17:14
  • $\begingroup$ You can see Corollary 5.2 of the paper of Stanley linked above for another way of writing the product formula for the Type B # of reduced decompositions (he lists it as a conjecture but it has been proven). $\endgroup$ – Sam Hopkins Aug 28 at 17:16
  • 1
    $\begingroup$ The basic thing that's going on here is that there's an Edelman-Greene style bijection in Types A and B (and also the non-Weyl types $I_2(m)$ and $H_3$- sometimes these types are called the 'coincidental types'). The other types don't have such a bijection. $\endgroup$ – Sam Hopkins Aug 28 at 17:18
5
$\begingroup$

This is easy to do in SageMath. E.g. the following code

G = WeylGroup("F4")
w = G.long_element_hardcoded()
print(w)
rw = w.reduced_words() 
len(rw)

outputs 2144892. If you want to look at some of these reduced words just examine the list rw. To create a list for classical types of different rank do

res = {}
for n in range(2,5):
    G = WeylGroup(["A", n])
    w = G.long_element_hardcoded()
    print("Calculating rank ", n)
    res[n] = len(w.reduced_words())
| cite | improve this answer | |
$\endgroup$
  • 2
    $\begingroup$ Note 2144892 = 2^2 x 3 x 47 x 3803, again suggesting lack of product formula. $\endgroup$ – Sam Hopkins Aug 28 at 19:34
  • $\begingroup$ I created findstat.org/StatisticsDatabase/St001585. If you have enough power to do E6 that would be great! $\endgroup$ – Martin Rubey Aug 28 at 19:44
  • $\begingroup$ @MartinRubey I will try some computer in Prague's Charles University to which I have remote access. But I am pessimistic. This algorithm actually generates and stores all the reduced words so it eats all available memory rather quickly. $\endgroup$ – Vít Tuček Aug 28 at 19:54
  • 1
    $\begingroup$ Counting reduced words is the same as counting directed paths between two vertices in a certain directed graph (the weak order graph), so it's possible you could write code by hand that does this faster than what's in Sage. $\endgroup$ – Sam Hopkins Aug 28 at 19:59
  • 1
    $\begingroup$ @SamHopkins, indeed, this is much faster! I added corresponding code to the findstat entry. $D_5$ is now easy. Unfortunately, FindStat has a limitation on the size of statistic values, so anything beyond $A_6$ is too large. $\endgroup$ – Martin Rubey Aug 28 at 20:20
5
$\begingroup$

Around the time Fomin and I wrote this paper, Tao Kai Lam applied the technique to type $D_n$. It emerged that it was "natural" to weight a reduced decomposition $\rho$ by $2^{d(\rho)}$, where $d(\rho)$ is the number of simple reflections in $\rho$ that correspond to the $n-2$ "nonbranch nodes" in the Coxeter diagram for $D_n$. Using this weighting, there is a nice product formula for the number of weighted reduced decompositions of the longest element, which I unfortunately have forgotten. I hope someone can redo this work.

| cite | improve this answer | |
$\endgroup$
  • $\begingroup$ Is this weighted sum related to linear extensions in some way? $\endgroup$ – Sam Hopkins Aug 28 at 19:09
  • 1
    $\begingroup$ @SamHopkins: I have a hazy recollection that it is a power of 2 times the number of SYT of the shifted shape $(2n-2,2n-4,\dots,4,2)$ (for which there is a nice product formula). $\endgroup$ – Richard Stanley Aug 28 at 19:41
  • $\begingroup$ That would make a lot of sense! In conservancy.umn.edu/bitstream/handle/11299/159973/… Williams calls the poset you're talking about the "flattened root poset" of $D_n$. $\endgroup$ – Sam Hopkins Aug 28 at 19:44
  • $\begingroup$ Actually, the result you mention might follow from this paper of Billey and Haiman: math.berkeley.edu/~mhaiman/ftp/schubert/schubert.pdf (see also the discussion on pg. 46 of the thesis of Williams linked above) $\endgroup$ – Sam Hopkins Aug 28 at 19:46

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.