The Thesis That Wasn’t

Jennifer Galovich
October 19, 2010
The guy that started it all…
Major Percy A. MacMahon

1854 - 1929
• Best known for Combinatory Analysis (1915),
the first important work in English in the field
of enumerative combinatorics.
• “The indices of permutations and the
derivation therefrom of functions of a single
variable associated with the permutations of
any assemblage of objects”, Am J. Math 35,
1913.
 Background
A permutation of {1,2,…,n} is a bijection from
{1,2,…,n} to itself.
Two line (Math 241)notation:
1234
3142
Cycle (Math 331) notation:
(1 3 4 2)
One line (Math 322) notation:
3142 word w
 Indices on permutations

Descent Number, des(w):

A word has a descent in position i if wi+1 > wi.
Example: w = 3142 has descents in positions
1 and 3, so des(w) = 2.
3•14•2

Major index: The major index of w, or maj (w) is the sum of the descent
positions.
Example: maj(3142) = 1 + 3 = 4.

The “other” major index: Maj(w), is the sum of the descent positions weighted
by the amount of the descent.
Example: Maj (3142) = 1(3 – 1) + 3 (4 – 2) = 8
 More generally…
A word of type i2
...n in of some arrangement of
1i1 2consists
i1 ones, i2 twos, etc.

Example: w = 44133121 is a word of type 13213242.

Indices work the same way as they did before:

44•133•12•1

des(w) = 3 (in positions 2, 5 and 7)

maj(w) = 2 + 5 + 7= 14
Maj(w) = 2 (4 – 1) + 5 (3 – 1) + 7 (2 -1)= 23
 Big question:
How are the values of the statistics distributed?
N = 3: N = 4:
word maj word maj
123 0 1234 0
132 2 1243 3
213 1 1342 3
231 2 .
312 1 .
321 3 .
4321 6

Generating Function: Generating Function:

1 q0 + 2q1 +2q2 + 1q3= 1 q0 + 3q1 +5q2 + 6q3 +5q4 +3q5 +1q6=
(1)(1 +q)(1 + q + q2)= (1)(1 +q)(1 + q + q2)(1 + q + q2+q3)=
[1]q [2]q [3]q =[3]! [1]q [2]q [3]q [4]q =[4]!
MacMahon proved that the GF for maj on
permutations of {1,2,…,n} was [n]!.

He also showed that the GF for maj on words of

type 1i 2i ...ni
1 2 n

was
[i1  i2  ...  in ]!
[i1 ]![i2 ]!...[in ]!
 What about Maj?
For binary words of type 0i1j , maj = Maj, and we have
  [i  j ]!
  [i ]![ j ]!
 
 
For ternary words of type 0i1j2k, MacMahon finds a similar formula in
terms of [i], [j] and [k].
 
 
For words of type 0i1j2k3l, MacMahon says:
 
“The determination of this expression is reserved for a future occasion.”
 Distribution of Maj is ugly!!

Ugly, but asymptotically pretty – that is I was able to show that it

was asymptotically normal -- as N gets larger,
the distribution looks more and more like the Gaussian.
 Another look at the GF…
For n=2 , we have [2]q=(1+q)
 
And for n=3 we have
 
=(1+q+q2)(1+q2 )=1+q+2q2 +q3 +q4
 So…consider the general case:
 

For example, when n = 5 this is

(1+q+q2 +q3+q4 )(1 + q2 +q4 + q6 )(1 + q3 +q6)(1 +q4 )(1)

 
Prettier pictures
 Definition of “pretty”
A generating function is “pretty” if
• the coefficients are symmetric (same, read
left to right and right to left)
• the coefficients are unimodal (they are non-
decreasing for a while and then they are non-
increasing – that is, they go up and then they
go down
 Example
1+q2 +2q3+5q4+2q5+q6+q7
is both symmetric and unimodal.
The coefficients are
1,1,2,5,2,1,1

1+q2 +2q3+5q4+2q5+q7+q8
Is neither symmetric nor unimodal.
The coefficients are
1,1,2,5,2,0,1
So.. we have some “prettier” generating
functions. But they are not the ones we
started with. So

What are they

generating?
 A new index
Let w be a permutation. First, we count the number of
inversions in w, that is, the number of pairs that are out of
order.
There is an easy way to do this! For each entry, count the
number of entries to the right and smaller.

Example:
w =7413652
w =6301210
w is called the left to right inversion vector of w. If we add
up the entries of w we get the number of inversions.
For my new index, however, I am going to
weight the entries by position:
dot product!

G(w) = w • (1,2,…,n)
= (6,3,0,1,2,1,0) •(1,2,3,4,5,6,7)
= 6+6+0+4+10+6+0 =32
 Theorem
The generating function of the new index is:
 It is easy to see that this GF is symmetric, but
 
 
Is it unimodal?????
 

(the thesis that wasn’t)

Answer: Yes
Proof by moral certainty:
I have
  checked it for n up to 200 (so far)
Idea…make a picture!
More precisely, make a poset P(n) for for
for which the GF we have is the
rank generating function

N=4
The rank numbers are
1,1,2,3,3,4,3,3,2,1,1

The GF is
= 1 q0 + 1q1 +2q2 + 3q3
+3q4 +4q5 +3q6 +3q7 +2q8 +
1q9+1 q10
Another picture of the same poset 3210

2210

ETC.
 What happened here?
Make a diagram for each inversion vector as follows:

(b1,b2,…,bn-1,0) b1 1s,b2 2s,…,bn-1 n-1s

 
 
(2,2,1,0)
 

 
 
and partially order the diagrams by inclusion.
• From the definition of unimodality, we want to
show (somehow) that the number of elements
at any rank is less than or equal to the number
at the next higher rank, until you get to the
middle, when things reverse.
 Plan
• To show unimodality, look for an injection
from one rank to the next one down till you
get to the middle.
 Unimodality? Stuck.

What else could I discover about this

poset?
Maximal chains
 1

2
 Fact
The maximal chains in my poset are lattice words
(aka Yamanouchi words aka ballot sequences)
Read left to right, at any point, the number of 1s beats
the number of 2s beats the number of 3s, etc.

Example: 1 1 1 2 1 2 1 3 2 1
Or 1122131211

But not 1 2 2 1 1 2 3 1 1 1

BAD!!
 There is another poset with this property– and
there are some other interesting parallels

M(n) My poset P(n)

• elements of M(n) are
subsets of {1,2,…,n}, written
as n-tuples of 0s and 1s
{2,3} (0,1,1,0)
•  ranked by set sums
(0,1,1,0) is rank 2
• Maximal chains are lattice
words
• Known to be unimodal!!
• Elements are inversion
vectors
• ranked by multiset sums
(2,2,1,0) is rank 7 = (2)(1)+(2)
(2)+(1)(3)
• Maximal chains are lattice
words
• Hope for unimodality???
 Lots of directions to go…

Unimodality SYTs and Schensted

???????

Other statistics? Count maximal chains

MacMahon’s Maj Other staircase shapes

 Shameless self-promotion…

Math 322
Combinatorics and Graph
Theory
9:40 even days