Positive Grassmannian, lectures by A.

Postnikov

March 14, 2012

Contents
1 Lecture 1, 2/8/2012 2
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Plücker coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Matroid Stratification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

2 Lecture 2, 2/10/2012 4
2.1 Mnëv’s Universality Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.2 Schubert decomposition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.3 Classical definition of Schubert cells . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.4 Plücker embedding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

3 Lecture 3, 2/15/2012 7
3.1 The Grassmannian Gr(n, k) over finite fields Fq . . . . . . . . . . . . . . . . . . . . 7
3.2 More on the Plücker embedding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
3.3 Matroids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

4 Lecture 4, 2/17/2012 10
4.1 Two definitions of matroids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
4.2 Matroid polytopes and a third definition of matroids . . . . . . . . . . . . . . . . . . 11

5 Lecture 5, 2/20/2012 12
5.1 Moment Map . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
5.2 Normalized Volumes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

6 Lecture 6, 2/24/2012 17
6.1 Graphical Matroids MG . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
6.2 Generalization of Matroid polytopes . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
6.3 Graphical Zonotopes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
6.4 Chromatic polynomial of G . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

7 Lecture 7, 2/29/2012 20
7.1 Schubert Calculus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
7.2 Cohomology of Gr(k, n, C) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
7.3 Note on proofs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
7.4 Honeycomb version of the Littlewood Richardson rule . . . . . . . . . . . . . . . . . 23

1
8 Lecture 8, 3/2/2012 23
8.1 Symmetric functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
8.2 Gleizer-Postnikov web diagrams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

9 Lecture 9, 3/7/2012 25
9.1 Wiring diagrams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
9.2 String cone . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

10 Lecture 10 3/09/2012 30
10.1 Symmetries of Littlewood-Richardson coefficients . . . . . . . . . . . . . . . . . . . . 30
10.2 Total positivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

11 Lecture 10 3/14/12 33
11.1 Bruhat order . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
11.2 Fomin-Zelevinsky double wiring diagrams and double Bruhat cells . . . . . . . . . . 35
11.3 Totally nonnegative Grassmannian . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
11.4 Relation to classical total positivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

1 Lecture 1, 2/8/2012
1.1 Introduction
Fix 0 ≤ k ≤ n, a field F (C, R, Fq ), the Grassmannian Gr(k, n, F) be the manifold of k-
dimensional linear subspaces in Fn (it has nice geometry it is projective algebraic variety/smooth,. . . )

Example 1.1. k = 1, Gr(1, n) = Pn−1 the projective space:

Pn−1 = {(x1 , . . . , xn ) 6= (0, . . . , 0)}\(x1 , . . . , xn ) ∼ (λx1 , . . . , λxn )

= {(x1 : x2 : · · · : xn )}.

Take a k × n matrix A with rank k then

Gr(k, n) = {k × n matrices of rank k}\ row operations = GL(k) \ Mat* (k, n).

where GL(k) is the group of k × k invertible matrices over F and Mat* (k, n) is the set of k × n
matrices over F with rank k.
The row picture is considering the row space of A (this is the original definition of Gr(k, n)).
The column picture is to take the columns of A: δ1 , . . . , δn ∈ Fk . Note that dim Gr(k, n) = k(n−k).
y

v3
v4 v2
v1
x
v5
Example 1.2. k = 2: v6 ∈ Gr(2, n).

2
 z

2 3

1 8
7
5
x 4

k=3 y projective picture: ∈ Gr(3, n).

1.2 Plücker coordinates

Let [n] = {1, 2, . . . , n} and [n]
k = {I ⊂ [n] | #I = k}. For a k × n matrix A and I ⊂ [n]
k let
∆I (A) = det(k × k submatrix in column set I) (i.e. ∆I (A) is a maximal minor of A). Since A has
rank k, at least on ∆I (A) 6= 0.
n

k
A

|{z} | {z }
I


For B ∈ GL(k), ∆I (B · A) = det(B) · ∆I (A). Thus the nk minors ∆I (A) for A in Gr(k, n)
form projective coordinates on Gr(k, n).
The Plücker embedding is the map Gr(k, n) 7→ P(k )−1 , A 7→ (∆I1 (A) : ∆I2 (A) : · · · ).
n

Example 1.3. For Gr(2, 4), its dimension is 2 · 4 − 2 · 2 = 4. Gr(2, 4) → P5 , A 7→ (∆12 : ∆13 , ∆14 :
∆23 : ∆24 : ∆34 ). Moreover, the maximal minors satisfy the following relation called a Plücker
relation

∆13 ∆24 = ∆12 ∆34 + ∆14 ∆23 . 1 2 3 4 1 2 3 4 1 2 3 4

`
The Grassmannian has the following decomposition: Gr(k, n) = λ⊂k×(n−k) Ωλ where λ is a
Young diagram contained in the k × (n − k) rectangle, and

ΩI := {A ∈ Gr(k, n) | ∆I (A) is the lexicographically minimal nonzero Plücker coordinate}.

This is the Schubert decomposition of Gr(k, n).

Example 1.4.  
0 1 2 1 3
A= ∈ Ω{2,4} = Ω .
0 1 2 2 3

Identify I ∈ [n]
k with the Young diagram λ ⊂ k × (n − k). The set I gives the labels of the
vertical steps. Explicitly, λ = (λ1 ≥ λ2 ≥ · · · ) is identified with the set I = {i1 < i2 < · · · < ik }
where λj = n − k + j − ij .

3
Example 1.5. k = 3, n = 9, λ ∼ I = {2, 5, 7}
1
2
5 4 3

7 6

9 8

Schubert calculus is based on the following result.

Theorem 1.6. Ωλ ∼
= F|λ| where |λ| is the number of boxes of λ.

Theorem 1.7. If F = C then H ∗ (Gr(k, n, C)) has a linear basis [Ωλ ].

Example 1.8. Gr(1, 3) = {(x1 : x2 : x3 )} = {(1, x2 , x3 )} ∪ {(0, 1, x3 )} ∪ {(0, 0, 1)}

| {z } | {z } | {z }
Ω Ω Ω∅

1.3 Matroid Stratification

[Gelfand-Serganova stratification]


For M ⊂ [n]k a strata is SM = {A ∈ Gr(k, n) | ∆I 6= 0 ⇔ I ∈ M}.
There is the following important axiom:
Exchange Axiom: For all I, J ∈ M and for all i ∈ I there exists a j ∈ J such that (I\{i}) ∪ {j} ∈
M.
In fact the nonempty sets M that satisfy the exchange axiom are called matroids, the sets
I ∈ M are called bases of M and k is the rank of the M.

Lemma 1.9. If M is not a matroid then SM = ∅.

The best way to prove this Lemma is using the Plücker relations. The converse is not true.
A matroid M is a realizable matroid (over F) if and only if SM 6= ∅. In general it is a hard
question to characterize realizable matroids.

2 Lecture 2, 2/10/2012
[n]

From last time we saw Gr(k, n) = GL(k) \ Mat* (k, n), Plücker coordinates ∆I (A) for I ∈ k . We
saw two stratifications: the Schubert and matroid stratifications.
a a
Gr(k, n) = Ωλ = SM ,
λ M

where the cells Ωλ have a simple structure and the matroid strata SM have complicated structure.

4
2.1 Mnëv’s Universality Theorem
The strata SM can be as complicated as any algebraic variety (even for k = 3).

Example 2.1. For k = 3 and n = 6, consider the point of Gr(3, 6) given by the projective picture:
1 2 3

4 5

It is in Ω1,2,4 = Ω and M = [6]
3 \ {{1, 2, 3}, {1, 4, 6}, {3, 5, 6}}.

Any system of algebraic equations can be realized by a configuration like One of the main topics
of this course will be to give a third decomposition of Gr(k, n) called the Positroid decomposition
that sits between the Schubert decomposition and the matroid decomposition.

2.2 Schubert decomposition

` [n]

Recall the Schubert decomposition Gr(k, n) = λ⊂k×(n−k) Ωλ where Ωλ = ΩI where I ∈ k .

Definition 2.2. There are several ways to define a Schubert cell:

1. ∆I consists of the elements of Gr(k, n) such that ∆I is the lexicographic minimum non-zero
Plücker coordinate.
2. A point in Gr(k, n) is in correspondence with a non-degenerate k × k matrix A modulo row
operations. By Gaussian elimination such a matrix A has a canonical form: the reduced
row-echelon form of the matrix. So ∆I consists of the elements of Gr(k, n) with pivot set I.

Example 2.3.  
0 1 ∗ ∗ 0 ∗ 0 0 ∗
0 0 0 0 1 ∗ 0 0 ∗
A=
0
,
0 0 0 0 0 1 0 ∗
0 0 0 0 0 0 0 1 ∗
Let I = {2, 5, 7, 8} be the set of the indices of the pivot columns (the pivots are underlined).

Clearly ΓI ∼
= F#∗ and also clear that ∆I , where I is the set of pivots, is the lexicographically
minimum nonzero Plücker coordinates.


Definition 2.4. The Gale order is a partial order on [n] k given by I = {i1 < i2 < · · · < ik } and
d J = {j1 < j2 < · · · < jk } then I  J if and only if i1 ≤ j1 , i2 ≤ j2 , . . . , ik ≤ jk .

Proposition 2.5. For A ∈ ΩI , ∆J (A) = 0 unless I  J.

Proof. We use the reduced row echelon form.

Corollary 2.6. The set M := {J ∈ [n]k | ∆J (A) 6= 0} has a unique minimal element with respect
to the Gale order.

Thus, we get another equivalent definition of a Schubert cell:

5
 3. ∆I consists of the elements of Gr(k, n) such that ∆I is the Gale order minimum non-zero
Plücker coordinate.
Theorem 2.7 (F = C or R). ΩI ⊃ ΩJ if and only if I  J.
Proof. ⇒: For J such that I 6 J then by Proposition 2.5 ∆J (A) = 0 and so A ∈ ΩI . In the closure
it still holds that ∆I (A) = 0 for A ∈ ΩI . (Looking at the row echelon form, pivots can only jump
to the right).
⇒: this is left as an exercise.

From row echelon form, remove the k pivot columns to obtain a k × (n − k) matrix.
Example 2.8. Continuing with the matrix from Example 2.3, if we remove the pivot columns we
obtain:
   
0 1 ∗ ∗ 0 ∗ 0 0 ∗ 0 ∗ ∗ ∗ ∗
0 0 0 0 1 ∗ 0 0 ∗ 0 0 0 ∗ ∗
  7→   7→ mirror image of λ = .
0 0 0 0 0 0 1 0 ∗ 0 0 0 0 ∗
0 0 0 0 0 0 0 1 ∗ 0 0 0 0 ∗

Also note that λ corresponds to I = {2, 5, 7, 8} (see Example 1.5).

Theorem 2.9. Ωλ ∼
= F|λ| .

2.3 Classical definition of Schubert cells

We think of Gr(k, n) as the space of k-dimensional subspaces V ⊂ Fn . Fix the complete flag of
subspaces:
{0} ⊂ V1 ⊂ V2 ⊂ · · · ⊂ Vn = Fn ,
where Vi =< en , en−1 , . . . , en+1−i > and ei = (0, . . . , 0, 1, 0, . . . , 0) is the ith elementary vector.
Pick a sequence of integers d = (d0 , d1 , . . . , dn ).
4. Ωd = {V ∈ Gr(k, n) | di = dim(V ∩ Vi ) for i = 0, . . . , n}.
The conditions on di ∈ Z≥0 are:
(
d0 = 0, di+1 = di or di+1 ,
dn = k.
Proof that Definition 4 of the Schubert cell is equivalent to Definition 3. We already have Ωλ ↔
ΩI . We show Ωλ ↔ Ωd and then show ΩI ↔ Ωd .
Given λ ⊂ k × (n − k) we obtain d by doing the following: we start at the SW corner of the
Young diagram and follow the border of the diagram until the NE corner. We label the horizontal
and vertical steps, where we add 1 if we go north. See Example 2.11(a) for an illustration of this.


Given I ⊂ [n] k we obtain d by di = #(I ∩ {n − i + 1, n − i + 2, . . . , n}). Note that this is
precisely the dimension of the rowspace(A) ∩ Vi or equivalently the number of pivots in positions
n, n − 1, . . . , n − i + 1.

Exercise 2.10. Check that all these correspondences agree with each other.
Example 2.11. For k = 4 and n = 9 let λ = 3211, from the labels of the Young diagram we get
d = (0, 1, 2, 2, 3, 3, 3, 4, 4) (see Figure 1). And since di = #(I ∩ {n − i + 1, n − i + 2, . . . , n}) we get
that I = {2, 5, 7, 8}.

6
 4 B
3 3 4
2 3
2
1
0 0
A

Figure 1: Illustration of correspondence between λ and d.

2.4 Plücker embedding

Recall the definition of the Plücker map ϕ : Gr(k, n) → PN −1 where N = n
k by A 7→ (∆I1 : ∆I2 :
· · · : ∆IN ).
Lemma 2.12. The map ϕ is an embedding.
Proof. We know that at least one Plücker coordinate is nonzero, without loss of generality say
∆12...k (A) 6= 0. Then
A= A e = [Ik | xij ],
where (xij ) is a k × (n − k) submatrix. From the minors we can reconstruct (xij ). That is
e := ∆I (A)/∆12...k (A) so that ∆12...k (A)
let ∆I (A) e = 1. Then by the cofactor expansion xij =
e Thus a point in the image of ϕ determines the matrix A in Gr(n, k).
±∆12···i−1 k+j i+1 ... k (A).

3 Lecture 3, 2/15/2012
3.1 The Grassmannian Gr(n, k) over finite fields Fq
Say F = Fq where q = pr is a power of a prime p. What can we say in this case about Gr(k, n, Fq )?
The first way is to view it as GL(k,Fq ) \ Mat* (k, n, Fq ). Recall that:

# Mat* (k, n, Fq ) = (q n − 1)(q n − q)(q n − q 2 ) · · · (q n − q n−k+1 ),

#GL(k, Fq ) = (q k − 1)(q k − q) · · · (q k − q k−1 ).
Then
1 − q n 1 − q n−1 1 − q n−k+1
#Gr(n, k, Fq ) = · ··· .
1−q 1−q
k k−1 1−q
n
We can write this more compactly using q-numbers: [n]q = 1 + q + q 2 + · · · + q n−1 = 1−q
1−q ,
 
n [n]q !
[n]q ! = [1]q [2]q · · · [n]q and = [k]q ![n−k] . Then
k q q!

 
n
Theorem 3.1. #Gr(n, k, Fq ) = .
k q
`
The second way is to use the Schubert decomposition Gr(k, n, Fq ) = λ⊆k×(n−k) Ωλ which
implies X
#Gr(k, n, Fq ) = q |λ| .
λ⊆k×(n−k)

Thus

7
  
n P
Corollary 3.2. = λ⊆k×(n−k) q |λ| .
k q

Example 3.3.

(1 − q 4 )(1 − q 3 )
Gr(2, 4, Fq ) =
(1 − q 2 )(1 − q)
= 1 + q + 2q 2 + q 3 + q 4
∅

Problem 3.4. Given a diagram D ⊆ k × (n − k), fix complement D to be 0. One obtains XD ⊆

Gr(k, n, Fq ). For e.g. if D is a skew Young diagram. Find #XD /Fq . Note that Ricky Liu studied
the combinatorics of related problems but not the # of points in Fq .

3.2 More on the Plücker embedding

Recall, the Plücker embedding ϕ : Gr(k, n) 7→ P(k )−1 , A 7→ (∆I1 (A) : ∆I2 (A) : · · · ). The signed
n

Plücker coordinates are

∆i1 ,i2 ,...,ik = ±∆{i1 ,i2 ,...,ik } ,
where the sign is positive if i1 < i2 < . . . < ik and the sign changes whenever we switch two indices.
Also ∆i−1,...,ik = 0 if the indices repeat. In terms of these signed coordinates the Plücker relations
are: for any i1 , . . . , ik , j1 , . . . , jk ∈ [n] and r = 1, . . . , k:
X
∆i1 ,...,ik ,j1 ,...,jk = ∆i01 ,...,i0k ∆j10 ,...,jk0 , (3.5)

where we sum over all indices i1 , . . . , ik and j10 , . . . , jk0 obtained from i1 , . . . , ik and j1 , . . . , jk by
switching is1 , is2 , . . . , isr (s1 < s2 < . . . < sr ) with j1 , j2 , . . . , jr .

Example 3.6. For n = 4, k = 3 and r = 1 we have (∆32 = −∆23 )

∆12 · ∆34 = ∆32 · ∆14 + ∆13 · ∆24 .

Theorem 3.7 (Sylvester’s Lemma).

1. The image of Gr(k, n) in P(k )−1 is the locus of common zeros of the ideal Ikn generated by the
n

Plücker relations (3.5)1 in the polynomial ring C[∆I ] (where we treat ∆I as formal variables).
2. Ikn is the ideal in C[∆I ] of all polynomials vanishing on the image of Gr(k, n).
3. Ikn is a prime ideal.

[? ? ? check Fulton’s book for issue with r? ? ? ?]

We give the proof of the first part of this result. The full proof can be found in Fulton’s book.
1. k-vectors d1 , . . . , dk , w1 , . . . , wk . Let |v1 . . . vk | := det(v1 , . . . , vk ) then
X
|v1 . . . vk | · |w1 . . . wk | = |v10 . . . vk0 | · |w10 . . . wk0 |
1
To generate the ideal one only needs the relations (3.5) for r = 1.

8
where the sum in the right hand side is obtained by switching r vectors from v1 , . . . , vk with r
vectors from w1 , . . . , wk . Let f be the difference between the left-hand side and the right-hand side
(we want to show that f = 0). Note that (i) f is a multilinear function of v1 , . . . , vk , wk and (ii) f
is an alternating function of v1 , . . . , vk , wk .
Claim: If vi = vi+1 or vk = wk then f = 0.
First we do the case vk = wk where r < k by induction on k.
X
f = |v1 . . . vk | · |w1 . . . vk | − |v10 . . . vk | · |w10 . . . vk |

Assume that vk = [0 . . . 01]T = en and expand the determinants in the expression above with
respect to the last column. We obtain the equation for f involving k − 1 vectors. By induction we
get f = 0.
Second, if vi = vi+1 the cancellation is easy to see as the following example shows.
Example 3.8. For k = 3 and r = 1 we get

f = |v1 v1 v3 | · |w1 w2 w3 | − (|w1 v1 v3 | · |v1 w2 w3 | + |v1 w1 w3 | · |v1 w2 w3 |)

= 0 − (|w1 v1 v3 | · |v1 w2 w3 | − |w1 v1 w3 | · |v1 w2 w3 |) = 0.

Part 1. We want to show that the image of ϕGr(k, n) in P(k )−1 is the zero locus of Ikn . Let
n

{∆I }I∈([n]) be any point in P(k )−1 satisfying the Plücker relations. We need to find a k × n matrix
n

k
A such that ∆I = ∆I (A).
Suppose ∆12...k 6= 0. We can rescale the coordinates such that ∆12...k = 1. Let A be the matrix
e I = ∆I (A). We want tot show that ∆
[Ik | xij ] where xij = ∆1 2...i−1 j+k i+1 ...k . Let ∆ e I = ∆I for all
[n]

I∈ k .
We know the following:
1. ∆1 2... k = ∆e 1 2 ... k = 1,
2. For all I such that |I ∩ {1, 2, . . . , k}| = k − 1 we have ∆ e I = xij for some j but by construction
xij = ∆I ,
3. Both ∆ e I and ∆I satisfy the Plücker relations.


We claim that form the three observations above it follows that ∆I = ∆ e I for all I ∈ [n] . We
k
show this for r = 1 by induction on p = |I ∩ {1, 2, . . . , k}|. The base case is p = k, k − 1.
We use the Plücker relations to expand ∆1 2 ... k · ∆i1 ... ik :

X
k
∆1 2 ... k · ∆i1 ... ik = ∆1 2 ... s−1 i s+1 ...k · ∆s i2 ... ik ,
s=1
X
= e 1 2 ... s−1 i s+1 ...k · ∆
∆ e s i ... i ,
2 k

e 1 2 ... k · ∆
=∆ e i ... i ,
1 k

e s i ... i y induction on p.
where ∆s i2 ... ik = ∆ 2 k
At the end since ∆1 2 ... k = ∆ e 1 2 ... k we obtain ∆i ... i = ∆
e i ... i as desired.
1 k 1 k

3.3 Matroids


Recall the notion of matroid M ⊂ [n] k . The elements of the matroid satisfy the Exchange axiom:
For all I, J ∈ M and for all i ∈ I there exists a j ∈ J such that (I\{i}) ∪ {j} ∈ M.

9
 [n]

Theorem 3.9. Pick A ∈ Gr(k, n) and let M = {I ∈ k | ∆I (A) 6= 0}. Then M is a matroid.

Proof. Two sets I, J and in M if and only if ∆J · ∆I 6= 0. But by the Plücker relations:
X
∆J · ∆I = ±∆(J\j)∪i · ∆(I\i)∪j .
j∈J

Thus there exists a j ∈ J such that ∆(J\j)∪i · ∆(I\i)∪j 6= 0 which implies the stronger condition that
both (J\j) ∪ i and (I\i) ∪ j are in M.

We call the stronger condition implied by the proof the strong exchange axiom.

Exercise 3.10. Is the strong exchange axiom equivalent to the exchange axiom?

4 Lecture 4, 2/17/2012
Last time we saw the bf Stronger Exchange axiom: for all i1 , . . . , ir ∈ I there exists j1 , . . . , jr ∈ J
such that (I\{i1 , . . . , ir }) ∪ {j1 , . . . , jr } ∈ M, (J\{j1 , . . . , jr }) ∪ {i1 , . . . , ir } ∈ M.


Example 4.1 (non-realizable matroid). The Fano plane in [7] 3 which is illustrated in Figure 2.

2 4
7

6
3 5

Figure 2: The Fano plane.

Exercise 4.2. Check that the Fano matroid is not realizable over R.

4.1 Two definitions of matroids

Recall the Gale order: {i1 , . . . , ik }  {j1 , . . . , jk } if and only if i1 ≤ j1 , . . . , ik ≤ jk . A generalization
of this order is obtained by picking a permutation w = w1 · · · wn in Sn , order the set [n] by
w1 < w2 < . . . < wn and define a permuted Gale order w accordingly.


Definition 4.3. Let M ⊆ [n] k .
1. Exchange Axiom: for all I, J ∈ M and for all i ∈ I there exists a j ∈ J such that
(I\{i}) ∪ {j} ∈ M.
2. M is a matroid if for all w in Sn , M has a unique minimal element in ⊆w (permuted Gale
order ⊆w ).

The second definition above is connected to
the Grassmannian since in Lecture 2 we had seen
that if we fix A ∈ Gr(n, k) then M := {I ∈ [n]
k | ∆I (A) 6= 0} has a unique minimal element with
respect to the Gale order.

10
 Another related result from Lecture 2 was the Schubert and matroid stratification of Gr(k, n):
a a
Gr(k, n) = Ωλ = SM ,
λ⊆k×(n−k) M

where Ωλ depends on an ordered basis but SM depends on an unordered` basis. Thus we really
have n!-Schubert decompositions: for w ∈ Sn we have Gr(k, n) = λ⊆k×(n−k) Ωw λ where Ωλ has
w

the same definition as Γλ but with respect to the order ew1 < ew2 < . . . < ewn . (We can also think
of Ωw
λ as w (Γλ ) where Sn acts on C by permuting coordinates according to w.
−1 n

Theorem 4.4 (Gelfand-Goresky-MacPherson-Sergenova). Matroid stratification is the common

refinement of n! permuted Schubert decompositions.

Proof. [? ? ? pending ? ? ?]

Example 4.5. For k = 3 and n = 5 consider the matroid M from Figure 3. The minimal basis
with respect to the Gale order is {1, 2, 4}. For w = 34512 the minimal basis with respect to the w
permuted Gale order is {3, 4, 1}. So SM ∈ Ω1,2,4 ∩ Ω34512
3,4,1 ∩ · · · .

1 2 3

5
4

Figure 3: Matroid M. The minimal basis with respect to the Gale order is {1, 2, 4}. For w = 34512
the minimal basis with respect to the w permuted Gale order is {3, 4, 1}.

4.2 Matroid polytopes and a third definition of matroids

We denote by e1 , . . . , en the coordinate vectors in Rn . Given I = {i1 , . . . , ik } ∈ [n] we denote by
[n]

k
eI the vector ei1 + ei2 + · · · + eik . Then for any M ⊆ k we obtain the following convex polytope

PM = conv(eI | I ∈ M) ⊂ Rn ,

where conv means the convex hull. Note that PM ⊂ {x1 + x2 + · · · + xn = k} so dim PM ≤ n − 1.

Definition 4.6 (Gelfand-Serganova-M-S). PM is a matroid polytope if every edge of PM is

parallel to ej − ei , i.e. edges are of the form [eI , eJ ] where J = (I\{i}) ∪ {j}.

This gives us a third definition of a matroid.

[n]

Definition 4.7. 3. A matroid is a subset M ⊆ k such that PM is a matroid polytope.

Theorem 4.8. The three definitions of a matroid: 1. by the exchange axiom, 2. by the permuted
Gale order, and 3. by matroid polytopes are equivalent.

Exercise 4.9. Prove this theorem.

11

[n]

Example 4.10. If M = [n] k is the uniform matroid then PM = conv(eI | I ∈ k ). This polytope

is called the hypersimplex ∆kn . The hypersimplex has the following property: All eI for I ∈ [n]
k
are vertices of ∆kn and these are all the lattice points of ∆kn .

Question 4.11. What are the vertices of any PM ?

The answer is eI where I ∈ M (basis of the matroid).

Example 4.12. For k = 1, ∆1n = conv(e1 , e2 , . . . , en ) is the usual (n − 1)-dimensional simplex.

See Figure 4 for an illustration of ∆13 .

∆1,4 e2
e1
e3
e4

Figure 4: An illustration of ∆13 .

Example 4.13. For k = 2 and n = 4 the hypersimplex ∆24 has six vertices e12 = (1, 1, 0, 0), e13 =
(1, 0, 1, 0), ldots (see Figure 5(a)). The matroid polytopes are subpolytopes of ∆24 without new
edges. In Figure 5(b) there are three subpolytopes associated with M1 = {12, 13, 14, 23, 34}, M2 =
{12, 14, 23} and M3 = {12, 23, 34, 13} respectively. M1 and M2 are matroids but M3 is not (take
I = {3, 4}, J = {1, 2} and i = 3 then the Exchange Axiom fails).

Exercise 4.14. Consider Gr(n, 2n, Fq ). [? ? ? pending ? ? ?]

5 Lecture 5, 2/20/2012
Last time we talked about matroid polytopes.
[n]

If M = k (uniform matroid) then the matroid polytope is the hypersimplex ∆kn .

e13 e13 e13

∆2,4 M1 M2 M3
e14 e34 e14 e34 e14 e34
e12 e23 e12 e23 e12 e23 e12 e23

e24
(a) (b)

Figure 5: (a) the hypersimplex ∆24 , (b) three subpolytopes of ∆24 corresponding to M1 =
{12, 13, 14, 23, 34}, M2 = {12, 14, 23} and M3 = {12, 23, 34, 13} the first two are matroid poly-
topes and the third one is not (it has a bad edge [e12 , e34 ], or in terms of M3 take I = {3, 4},
J = {1, 2} and i = 3 then the Exchange Axiom fails).

12
5.1 Moment Map
First we will talk about the moment map which is important in symplectic and toric geometry.
Let T = (C∗ )n , it acts on Cn by (t1 , . . . , tn ) · (x1 , . . . , xn ) 7→ (t1 x1 , t2 x2 , . . . , tn xn ). This induces
a T -action on Gr(k, n, C).
Recall Gr(k, n) = GLk \M at(k, n). We claim that T acts by right multiplication byQ diag(t1 , . . . , tn )
(we rescale column i by ti ). In terms of Plücker coordinates: (t1 , . . . , tn ) · {∆I } 7→ { P i∈I ti ∆I }.
|∆ |2
We define the moment map: µ : Gr(k, n, C) → Rn , A 7→ (y1 , . . . , yn ) where yi = PI3i I2 .
|∆ I|
I

Example 5.1. k = 2 and n = 4, Gr(2, 4) → R4 , A 7→ (y1 , y2 , y3 , y4 ) where for instance y1 =

|∆12 |2 + |∆13 |2 + |∆14 |2
P .
I |∆I |
2

Theorem 5.2 (Atiyah-Guillemin-Sternberg). (1) Image of µ is a convex polytope.

(2) Moreover, pick any point A ∈ SM ⊂ Gr(k, n), then µ(T · A) is a convex polytope (A is fixed,
T · A is a set of matrices).

Exercise proof this case of convexity theorem.

Claim:


(1) µ(Gr(k, n)) = [n]
k .

(2) µ(T · A) = PM is a matroid polytope.

Idea proof of Claim.

(1) Clearly 0 ≤ yi ≤ 1, also y1 + · · · + yn = k. This means that µ(Gr(k, n)) ⊆ ∆kn . (Recall that
∆kn = {(y1 , . . . , yn ) | 0 ≤ yi ≤ 1, y1 + · · · + yn = k}.)
Pick AI to be the 0-1 matrix whose k × k submatrix indexed by I is the identity matrix, and
the other columns are 0. This is a fixed point of T in Gr(k, n). Actually this is the form of all the

[n]

fixed points. Thus there are nk such fixed points P (one for each set I ∈ k ).
Now ∆J (AI ) = δI,J . Then µ(AI ) = eI = i∈I ei and this is a vertex of ∆kn . From the convexity
theorem, µ is a convex polytope, this together with the fact that µ(AI ) = eI forces µ(Gr(k, n)) to
be ∆kn . 

Remark 5.3. To prove the convexity result we have to show that if

Plücker µ
A → {∆I } → y = (y1 , . . . , yn ),
Plücker
Y Plücker
A · diag(t1 , . . . , tn ) → { ti ∆I } → y 0 = (y10 , . . . , yn0 ),
i∈I

then y and y 0 are connected by a line where every point in the line corresponds to an image of
µ.

5.2 Normalized Volumes

P ⊂ Rn , polytope with integer vertices in Zn . Let d = dim(P ) and let V be a d-dimensional
affine subspace containing P (it may or may not contain the origin). Let L = V ∩ Zn = `0 + <
`1 , . . . , `d >Z , where < `1 , . . . , `d >Z is the set of all linear Z-combinations of {`1 , . . . , `d }.

13
 V `1
`0 `2

Define Vol(·): volume form of V such that Vol(`0 + Π(`1 , . . . , `d )) = 1 where Π(`1 , . . . , `d ) is the
parallelepiped spanned by {`1 , . . . , `d }.
Also we define normalized volume Vol f := d! Vol. Claim VolP
f ∈ Z and Vol f of the standard
coordinate simplex is 1.
f
Example 5.4. For the lines A and B, Vol(A) f
= 2 but Vol(B) = 1.

A
B

The following result justifies why normalized volumes are interesting.

Theorem 5.5. The degree of a torus orbit T · A is the normalized volume of µ(T · A) = PM .

Example 5.6. For Gr(2, 4):

α1 t1 t2 + α2 t1 t3 + · · · + α6 t3 t4 = 0
β1 t1 t2 + β2 t1 t3 + · · · + β6 t3 t4 = 0
γ1 t1 t2 + γ2 t1 t3 + · · · + γ6 t3 t4 = 0.

f 24 ). We can calculate this volume by showing that

The number of solutions of this system is Vol(∆
f 24 ) = 4.
we can divide ∆24 into four simplices of normalized volume 1. Thus Vol(∆

∆24

This motivates calculating the normalized volume of the hypersimplex. In general the nor-
malized volume of ∆kn is given by the Eulerian number Ak−1,n−1 , where Ak,n = #{w ∈ Sn |
des(w) = k} where des(w) is the number of descents of w (#{i | wi > wi+1 }).

14
 f kn ) = Ak−1,n−1 .
Theorem 5.7. Vol(∆

Example 5.8.
f kn )
permutations Ak−1,n−1 Vol(∆
123 f 14 ) = 1
A0,3 = 1 Vol(∆
213, 132 A1,3 = 4 f 24 ) = 4
Vol(∆
312, 231
321 A23 = 1 f 34 ) = 1
Vol(∆

Euler knew the following:

1
1 + x + x2 + x3 + · · · =
1−x
x
x + 2x + 3x + 4x + · · · =
2 3 4
(1 − x)2
x + x2
12 x + 22 x2 + 32 x3 + 42 x4 + · · · =
(1 − x)3
x + 4x2 + x3
13 x + 23 x2 + 33 x3 + · · · = .
(1 − x)4

And in general:

Proposition 5.9.
X
∞ Pn
k=0 Ak,n x
k+1
r x =
n r
,
(1 − x)n+1
r=1
P
or equivalently, Ak−1,n−1 = [xk ](1 − x)n r≥1 r
n−1 xr .

We think of the hypersimplex in two ways: as a section of the n-hypercube:

∆kn = [0, 1]n ∩ {x1 + · · · + xn = k},

or equivalently as a slice of the (n − 1)-hypercube:

∆kn = [0, 1]n−1 ∩ {k − 1 ≤ x1 + · · · + xn−1 ≤ k}.

Example 5.10. We divide [0, 1]3 into ∆14 , ∆24 and ∆34 .
∆24 ∆34

∆24

f = 2 but Vol(B)
VolA f = 1.

15
Proof. First let us take the first interpretation.
Take the positive octant (R≥0 )n . Define k∆ = (R≥0 )n ∩ {x1 + · · · + xn = k}, this is a k-dilated
f
(n − 1)-simplex. So Vol(k∆) = k n−1 .
By inclusion-exclusion we can decompose the n-hypercube as
X X
[0, 1]n = (R>0 )n − (ei + (R≥0 )n ) + (ei + ej + (R≥0 )n ) − · · ·
i i<j

This is really an identity for characteristic functions of subsets (modulo set of measure 0)

1
− +

+ −
0 1

Then the volume of ∆kn (using the first interpretation)

X X
f kn ) = Vol(k∆)
Vol(∆ f − f
Vol((k − 1)∆) + f
Vol((k − 2)∆) − · · ·
i i<j
   
n n
= k n−1 − n(k − 1)n−1 + (k − 2)n−1 − · · · + (−1)k .
2 k−1
P
From the expression Ak−1,n−1 = [xk ](1 − x)n r≥1 rn−1 xr from Proposition 5.9 one can show this
expression above gives Ak−1,n−1 .

Stanley’s proof. We use the interpretation of ∆kn = [0, 1]n−1 ∩ {k − 1 ≤ x1 + · · · + xn−1 ≤ k}. For
w ∈ Sn−1 , there is a simple triangulation of [0, 1]n−1 into ∆w = {0 < xw1 < xw2 < . . . < xwn < 1}
of (n − 1)! simplices.
This is not exactly the triangulation we need, it is not compatible with the slicing of the
hypercube:

x2

∆21
x1 + x2 = 2
∆12
x1

x1 + x2 = 1
x1 + x2 = 0

Instead define S : [0, 1]n−1 → [0, 1]n−1 to be the following piecewise-linear volume preserving
map. (x1 , . . . , xn−1 ) 7→ (y1 , . . . , yn−1 ) where xi = {y1 + · · · + yi } = y1 + · · · + yi − by1 + · · · + yi c
(where {x} is the fractional part of x). The forward map is:
(
xi − xi−1 if xi ≥ xi−1
yi =
xi − xi−1 + 1 if xi < xi−1 ,
where we assume x0 = 0. The maps S is piecewise linear map and volume preserving.

16
 We had a trivial triangulation ∆w−1 := {0 < xw−1 (1) < · · · < xw−1 (n−1) < 1}. We get (n − 1)!
simplices. Then xi−1 < xi ⇔ wi−1 > wi ⇔ (i − 1) is a descent of w e = 0w1 w2 . . . wn−1 . Thus if
Des(w)
e is the st of descents of w
e then
(
xi − xi−1 if i − 1 6∈ Des(w),
e
yi =
xi − xi−1 + 1 if i − 1 ∈ Des(w)e

Then y1 + y2 + · · · yn−1 = xn−1 − x0 + des(w), and so des(w) ≤ y1 + y2 + · · · + yn−1 ≤ des(w) + 1.

So S(∆w−1 ) is in the kth slice of the (n − 1)-cube where k = des(w) + 1. And S(∆w−1 ) is a
triangulation of the (n − 1)-cube with exactly Ak−1,n−1 simplices in the kth slice.

6 Lecture 6, 2/24/2012
We digress a bit from the Grassmannian to talk about Matroids.
f M?
Problem 6.1. For any matroid M, what is the integer VolP
The dual matroid if M∗ := {[n]\I | I ∈ M}. Clearly, PM∗ ∼
= PM by the map (x1 , . . . , xn ) 7→
(1 − x1 , . . . , 1 − xn ).

6.1 Graphical Matroids MG

If G is a graph with labelled edges {1, 2, . . . , n}, the bases of MG are the set of edges corresponding
spanning trees of G. We denote by G∗ the dual graph of G (when G is planar).
1 1 1 1
4 2 4 2 4 2 4 2

1 1 3 3 3 3
4 2 4 2

3 3
G

G∗

Figure 6: A graph G, its dual G∗ and the spanning trees of G and of G∗ .

Exercise 6.2. 1. Check exchange axiom on spanning trees of G.

2. If G is a planar graph, MG∗ = (MG )∗ .

Example 6.3. If G is an n-cycle then MG ∼
= ∆n−1 is an (n − 1)-simplex.
Example 6.4. If G is an (r + 1)-cycle glued at a vertex with a (s + 1)-cycle then PMG = ∆r × ∆s .


f M ) = r+s
And Vol(P G r

(s + 1)-cycle

(r + 1)-cycle

17
Proposition 6.5. In full dimensions Vol(A × B) = Vol(A) · Vol(B), if A is r-dimensional and B


f
is s-dimensional then Vol(A × B) = r+s f f
r Vol(A)Vol(B).

f M ) =? or give families of graphs G with nice formulas for Vol(P

Problem 6.6. What is Vol(P f M )=
G G
?.

6.2 Generalization of Matroid polytopes

A generalized permutahedron is a polytope P ⊂ Rn such that any edge of P is parallel to
ei − ej for some i 6= j. (such a permutahedron has dimension n − 1)

Example 6.7. n = 3, the lengths of the sides satisfy the following hexagon equation: a + f = c + d,
e + f = b + c, and a + b = c + d.
a

f
b

e
c
d

Example 6.8. The usual permutahedron Pn with n! vertices (w1 , . . . , wn ) for all w ∈ Sn .
(3, 2, 1) (3, 2, 1)

(3, 1, 2) (2, 3, 1) (3, 1, 2) (2, 3, 1)

P3 P3

(2, 1, 3) (1, 3, 2) (2, 1, 3) 12 3 (1, 3, 2)

(1, 2, 3) (1, 2, 3)

Figure 7: The permutahedron P3 , and tiling of P3 illustrating why Vol(Pn ) is nn−2 , the number of
Cayley trees with n vertices.

Exercise 6.9. Check that Pn is a generalized permutahedron.

f n ) = (n − 1)!nn−2 . This is Cayley’s formula
Exercise 6.10. Vol(Pn ) = nn−2 or equivalently Vol(P
for the number of trees with n labelled vertices.

case k = 3.

18
 + =

+ =

+ =

Figure 8: Examples of Minkowski sums.

6.3 Graphical Zonotopes

The Minkowski sum A + B = {x + y | x ∈ A, y ∈ B}.
A zonotope is a Minkowski sum of line segments. In the plane you get n-gons whose opposite
sides have same length and are parallel. From the examples above, in the plane only the square is
a zonotope. P ∼
P If G is a graph on n vertices labelled 1, 2, . . . , n, the graphical zonotope is ZG = (i,j)∈E(G) [ei , ej ] =
(i,j)∈E(G) [0, ej − ei ]. In the last equation you pick an orientation of edges, however the zonotope
does not depend on the orientation.

Proposition 6.11. For the complete graph, the zonotope is the permutahedron ZKn = Pn .
P
A Newton polytope of f ∈ C[x1 , x2 , . . . , xn ], then we can write f = ca1 ,...,an xa11 · · · xann .
Then New(f ) = conv({(a1 , . . . , an ) | ca1 ,...,an 6= 0})

Example 6.12. New(x2 y 3 + x + 27y 2 ) = conv((2, 3), (1, 0), (0, 2)).

Figure 9: The newton polytope of x2 y 3 + x + 27y 2 .

One of the most important features of the Newton polytope is the following property than says
that we can view New(·) as a generalized logarithm.

Proposition 6.13. New(f · g) = New(f ) + New(g).

Proof. The non-trivial point of this proof is that vertices are not cancelled.

Proposition 6.14. ZG is a generalized permutahedron.

19
using Newton polytopes. Recall the Vandermonde determinant
 
1 1 ... 1
 x1 x2 . . . xn 
  X Y
 x2 x2 2 . . . x2n  1 −1 w2 −1 n −1
 1 = (−1)sgn(w) xw x2 · · · xw = (xj − xi ).
 .. .. .. ..  w∈S 1 n
 . . . .  n i<j

x1n−1 xn−1
2 · · · x n−1
n

If we take the Newton polytope of both P sides and use Property [], New(LHS) = conv((w1 −
1, . . . , wn − 1)) = Pn and New(RHS) = [ei , ej ] = ZKn .

Theorem 6.15. Vol(ZG ) = #{spanning trees in G}.

Proof. We prove it by induction on the number of edges of G. Let t(G) = #{spanning trees in G},
this numbers satisfy the following deletion-contraction relation

t(G) = t(G\e) + t(G/e).

Where G\e is graph G with edge e deleted and G/e is the graph G with edge e contracted. One
can then show that ZG also satisfies the same relation.

6.4 Chromatic polynomial of G

χG (t) = #{proper t-colorings of vertices of G}, a proper coloring is one where the vertices of any
edge of G have different colors.

Theorem 6.16. χG (t) is a polynomial in t.

Proof. χG (t) satisfies a deletion-contraction relation χG = χG\e − χG/e , and show χG (t) = tn if G
consists of just n vertices with no edges.

Problem 6.17 (open). Is there a polytope such that some statistic gives TG (x, y). Is there a
polytopal realization of the Tutte polynomial.

7 Lecture 7, 2/29/2012
7.1 Schubert Calculus
We start with basic facts about (co) homology: Let X be a topological space and Hi (X) is the ith
homology of X which is some vector space over C. Its dual H i (X) := (Hi (x))∗ is the cohomology
of X. These are both topological invariant of X.
The Betti number is βi (X) = dim H i (X). If H ∗ (X) = H 0 ⊕ H 1 ⊕ · · · , this space has a
multiplicative structure (cup-product).
Suppose that X is a nonsingular complex algebraic variety and dimC X = N then the homology
and cohomology only live in even dimension:

H∗ (X) = H0 ⊕ H2 ⊕ · · · ⊕ H2N
H ∗ (X) = H 0 ⊕ H 2 ⊕ · · · ⊕ H 2N .

The fundamental class [X] is the canonical generator of H2N .

20
 Y0 Y0
Y Y

transversal not transversal

Figure 10: Example of transversal and non-transversal intersections. In the first example the
intersection of the tangent spaces at the points where the varieties meet is a point. In the second
example, the tangent spaces are the same.

n−k n−k

λ λ
k k
λ ∨ µ

(a) (b)

Figure 11: (a) Example of a partition λ and its complement λ∨ , and (b) example of partitions
obtained in Pieri rule.

We also consider Poincaré duality that says H i (X) ∼= H2N −i (X), or equivalently H i (X) ∼=
(H 2N −i (X))∗ . If Y ⊂ X is an algebraic subvariety with dim Y = m then [Y ] ∈ H (X) ∼
=
` C 2m
H 2N −2m (X) (it has codimension 2m). If X = i∈I Yi then we say that X has a cell decomposition
(CW-complex), where Yi ∼ = Cmi and Yi is an algebraic subvariety and Yi \Yi is a union of smaller
dimensional Y s.
Claim 1: Cohomology classes of [Yi ] are in H 2N −2mi (X) so they form a linear basis of H ∗ (X). In
particular H 0 (X) is spanned by [X]. And H 2N (X) is spanned by [point].
Claim 2: If Y and Y 0 are algebraic subvarieties of X and Y ∩ Y 0 = Z1 ∪ · · · ∪ Zr where
(i) codim Y + codim Y 0 = codim Zi for all i (proper intersection)
(ii) For every generic point z ∈ Zi , Tz Zi = Tz Y ∩ Tz Y 0 where Tz is the tangent space (transversal
intersection)
Then X
[Y ] · [Y 0 ] = [Zi ].

7.2 Cohomology of Gr(k, n, C)

a
Gr(k, n) = Ωλ ,
λ⊆k×(n−k)

where Ωλ ∼
= C|λ| is a Schubert cell. Let Xλ := Ωλ∨ where λ∨ = (n − k − λk , · · · , n − k − λ1 ) is the
complement of λ in k × (n − k).
Denote by σλ = [Xλ ] ∈ H 2|λ| (Gr(k, n, C)), these are the Schubert classes. The Schubert classes
do not depend on the choice of basis, just on the partition.

Theorem 7.1. The Schubert classes σλ for λ ⊆ k × (n − k) form a linear basis of H ∗ (Gr(k, n, C)).

21
Example 7.2. σ∅ = [Gr(k, n)] and σk×(n−k) = [point].
Remarks 7.3 (Special feature of this basis). This basis is self-dual with respect to Poincaré
duality. This means:
(i) B = {σλ | |λ| = i} basis of H 2i (Gr(k, n)),
(ii) B∗ = {σ|µ| | |µ| = k(n − k) − i} basis of H 2k(n−k)−2i (Gr(k, n)). B and B ∗ are dual basis (the
dual map is σλ 7→ σλ∨ ).
Let c ∈ C, for σ ∈ H 2k(n−k) (Gr(k, n)) where σ = c · [point] then < σ >:= c.
Theorem 7.4 (Duality Theorem). For partitions λ, µ such that |λ| + |µ| = k(n − k) then <
σλ · σµ >= δλ,µ∨ σk×(n−k) . Where the product of Schubert classes is in the cup product.
Theorem 7.5 (Pieri Formula). Let σr = σ ...
(k boxes) then
X
σλ · σr = σµ ,
µ

where the sum is over µ such that µ/λ is a horizontal r-strip.

In
Pterms of coordinates the partitions are interlaced n−k ≥ µ1 ≥ λ1 ≤ µ2 ≥ λ2 ≥ · · · ≥ µk ≥ λk
and (µi − λi ) = r.
Example 7.6.

7.3 Note on proofs

Consider σλ · σµ = [Xλ ] · [Xµ ]. We work with Xλ ∩ X eµ where Xλ corresponds to standard Schu-
eµ corresponds to the opposite
bert decomposition (ordering basis with permutation 12 · · · n) and X
ordering of the coordinates (permutation n n − 1 · · · 21). We do this choice of basis to obtain a
transversal intersection and then use Claim 2.
If |λ| + |µ| = k(n − k) and λ = µ∨ then we get a point. Otherwise you can show that Xλ ∩ X eµ .
Pieri formula uniquely defines the multiplicative structure of the Schubert cells.
X
σλ · σµ = cνλµ σν ,
ν,|ν|=|λ|+|µ|

where cνλµ are the Littlewood Richardson coefficients. By the duality theorem cνλµ =< σλ ·σµ ·σν ∨ >,
∨ ee
eµ ∩ X
i.e. cλµν := cνλµ = #{Xλ ∩ X ν }. Then cλµν ∈ Z≥0 and these coefficients have S3 -symmetry.

n−k n−k

λ 0
Xλ = k eµ =
X k
µ
0

n−k n−k

0 0
eµ =
Xλ ∩ X k eλ∨ =
Xλ ∩ X k
0 0

22
 (a, ∗, ∗) (∗, b, ∗) (∗, ∗, c)

k=3 (λ3 , ∗, ∗) (∗, µ1 , ∗)

(λ2 , ∗, ∗) (∗, µ2 , ∗)

(λ1 , ∗, ∗) (∗, µ3 , ∗)

(∗, ∗, −ν1 ) (∗, ∗, −ν2 ) (∗, ∗, −ν3 )

7.4 Honeycomb version of the Littlewood Richardson rule

This version was done by Knutson-Tao, it is a reformulation of a rule by Bernstein-Zelevinsky.
We work in the plane R2 = {(x, y, z) ∈ R3 | x + y + z = 0}. In this plane there are three types
of lines (a, ∗, ∗), (∗, b, ∗) and (∗, ∗, c) where a, b, c ∈ Z.

Theorem 7.7. cνλµ = #{integer honeycomb with fixed boundary rays}.

If we know the rational lengths `i of the internal edges we can reconstruct the honeycomb.
We can rescale the honeycomb such that `i ∈ Z≥0 and also λ1 + λ2 = λ1 − λ2 and the lengths
on a hexagon should satisfy the hexagon condition

8 Lecture 8, 3/2/2012
Recall from last time that H ∗ (Gr(k, n, C)) has a linear basis of Schubert classes Ωλ . In this lecture
we will mention the relation between H ∗ (Gr(k, n, C)) and symmetric functions.

8.1 Symmetric functions

Let Λ be the ring of symmetric functions. We build this ring in the following way: let Λk =
C[x1 , . . . , xP
n ] , the symmetric polynomials with k variables and let Λ = ←
Sk lim Λk .
−
• er = P1≤i1 <i2 <···<ir xi1 xi2 · · · xir (elementary symmetric functions)
• hr = 1≤j1 ≤j2 ≤···≤jr xj1 · · · xjr (complete symmetric functions)

Theorem 8.1 (Fundamental theorem of symmetric functions). Λ = C[e1 , e2 , . . .] = C[h1 , h2 , . . .].

Another well known fact about Λ is it has a linear basis of Schür functions sλ where λ is a
partition λ = .

Definition 8.2. We give two equivalent definitions of the Schür functions.

23
 • Given a partition λ = (λ1 , . . . , λk ) let α = (α1 , . . . , αk ) = (λ1 + k − 1, λ2 + k − 2, . . . , λk + 0).
 α1 
 x α1 α1 
 1α x2α · · · xk 
 x 2 x 2 ··· 
 1 2 
sλ (x1 , . . . , xk ) =  . . . .  / det(xk−i
j )
 . . .
. . . .
. 
 α 
 x k x αk · · · x αk 
1 2 k

then sP
λ = limk→∞ sλ (x1 , . . . , xk ).
• sλ = T ∈SSY T (λ) xT where SSY T (λ) is the set of semistandard Young tableaux

T = 1 1 1 2 2 2, xT = x31 x52 x33 x34 x35 x6 x7 .

2 2 3 3 4 7
3 4 4 6
5 5 5

In particular er = s1r and hr = s(r) .

From the second definition of sλ it is easy to see the following rule.

Theorem 8.3 (Pieri formula for sλ ).

X
hr · sλ = sµ .
µ

where µ are partitions such that µ/λ is a horizontal r-strip.

P
Equivalently er · sλ = µ sµ where µ/λ is a vertical r-strip.
P
Lemma 8.4. Suppose that we have an associative bilinear operation ∗ on Λ such that sr ∗sλ = µ sµ
where µ/λ is a horizontal r-strip, then (Λ, ∗) ∼
= (Λ, ·).
Proof. By the fundamental theorem we know that (Λ, ·) = C[h1 , h2 , . . .], since sr = hr , the Pieri-
type formula essentially says that (Λ, ∗) has the same product of by the algebraically independent
generators hr as (Λ, ·).

Definition 8.5. Let Λk,n = Λ/Ik,n where Ik,n :=< sλ | λ 6⊂ k × (n − k) >.

Exercise 8.6. Show that In,k =< ei , hj | i > k, j > n − k >. Show that sλ 6⊂ k × (n − k) form a
linear basis of In,k .

Theorem 8.7. H ∗ (Gr(k, n)) ∼

= Λk,n .
P
Proof. Define the map σλ 7→ sλ , sλ · sµ = ν cλµ sν .
ν

8.2 Gleizer-Postnikov web diagrams

The GP-diagrams involve four directions instead of the three directions of the Knutson-Tao puzzles.
In the latter λ, µ and ν have k parts, and in the former λ and µ have k parts and ν has k + ` parts.
Note that by convention s(λ1 ,...,λk ,0)) = s(λ1 ,...,λk ) and sλ = 0 if some λi < 0.
√
The horizontal edge at a height c from the x-axis is labelled c = 2h/ 3. The edges going from
NW-SE and from NE-SW are labelled according to their x-intercepts. See Figure 13.

24
 Knutson-Tao puzzles GP-web diagrams

k k 60o `
120o
λ µ
µ
λ
cνλµ 120o cνλµ 120o
k 120o
ν
120o
k k
` 60o
ν

Figure 12: Schematic comparison of Knutson-Tao puzzles and Gleizer-Postnikov web diagrams.

2h
a c= √
3 b

0 a b

Figure 13: How to label the edges of a GP-web diagram.

P
Example 8.8 (k = ` = 1). sr · ss = c≥r−s ss+c,r−c . We have a conservation law: a flow of r + s
is coming in and coming out. See Figure 14.

Theorem 8.9. The Littlewood-Richardson coefficient cνλµ is the number of web diagrams of type λ,
µ and ν.
P
Proof. Define ∗ product on Λ by sλ ∗ sµ = ν #{web diagrams type λ, µ, ν}sν . Next we prove the
Pieri rule for ∗-product.
For ` = 1 we get interlacing ν1 ≥ λ1 ≥ λ2 ≥ · · · ≥ νk ≥ λk ≥ νk+1 (see Figure 15). By
conservation law |λ| + µ1 = |ν|. This is equivalent to saying that ν/λ is a horizontal µ1 -strip.
Conversely, given ν such that ν/λ is a horizontal µ1 -strip we can build a unique web diagram. The
non-trivial part showing it is associative.

Example 8.10. Let’s verify that c631

21,52 = 2. See Figure 16

Problem 8.11. (Open) The GP-web diagrams with six directions are infinite. However with five
directions the diagrams are finite. In this case, what is the analogue of the Littlewood-Richardson
coefficients.

9 Lecture 9, 3/7/2012
A permutation w is a bijection between [n] → [n]. We multiply permutations from right to left. A
simple transposition is a permutation si = (i, i + 1). We can write any w as a product of simple

25
 r s

r−c s+c

s r−s
c≥r−s

Figure 14: Example of GP-web diagrams for k = ` = 1.

λ1 µ1
λ2
...
λk
ν1
ν2

...
νk+1 νk

Figure 15: Illustration of the Pieri-rule for GP-web diagrams

transpositions. A reduced decomposition is an expression of w as a product si1 si2 · · · si` of simple

transpositions of minimum possible length `. The reduced decompositions of w are related by
certain moves

si sj = sj si |i − j| ≥ 2, (2-move)
si si+1 si = si+1 si si+1 (3-move)

9.1 Wiring diagrams

 
1 2 3 4 5
w= , w = s4 s2 s1 s2 s3 .
3 2 5 1 4
We do not allow triple intersections and intersections occur at different heights.
We want to transform the GP-web diagrams into wiring diagrams.
The number of such plane partitions are the Littlewood-Richardson rule.

Exercise 9.1. Show that the classical Littlewood-Richardson rule corresponds to this rule in terms
of plane partitions.

If V is a vector space with basis e0 , e1 , e2 , . . . a Scattering matrix (R-matrix)

(
es+c ⊗ er−c if c ≥ r − s,
R(c) : V ⊗ V → V ⊗ V, er ⊗ es =
0 otherwise.

26
 2

4 1 1
0 1 3 6 0 1 3 6
0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 1 2 3 4 5 6

Figure 16: Example showing the two GP web diagrams when λ = 21, µ = 52 and ν = 6310.

2-move 3-move w = 32514

Figure 17: Illustration of 2 and 3-moves and of the wiring diagram of w = 32514 = s4 s2 s1 s2 s3 .

where e−i = 0 for i > 0. This corresponds to picture. (ei are levels of excitement of particles and
R(c) describes how they interact)
Given a partition λ = (λ1 , . . . , λk ) we map sλ 7→ eλk ⊗ eλk−1 ⊗ · · · ⊗ eλ1 ∈ V ⊗k .
We define an operator Mk,` : V ⊗k ⊗ V ⊗` → V ⊗(k+`) . Then Rij (c) is the operator on V ⊗m that
acts as R(c) on the ith and jth copies of V . Clearly Rij (c) commutes with Rîĵ (c) if #{i, j, î, ĵ} = 4
(they act on four different copies of V ).
Definition 9.2. X Y Y
Mk,` = Rij (cij ),
(cij i=1,...,k j=k+`,k+`−1,...,k+1

where the sum is over (cij ) such that cij ≥ ci0 j 0 ≥ 0 whenever i0 ≤ i < j ≤ j 0 .
P
For example M23 = R15 (c15 )R14 (c14 )R13 (c13 )R25 (c25 )R24 (c24 )R23 (c23 ) (you can choose any
linear extension of the poset - such number is the number of Young Tableaux on the rectangle)
Theorem 9.3 (LR-rule: R-matrix version). Given λ = (λ1 , . . . , λk ) and µ = (µ1 , . . . , µ` ) then
X
Mk,` (eλ ⊗ eµ ) = cνλ,µ eν .
ν=(ν1 ,...,νk+` )

The Pieri-rule is easy, the hard part is to show associativity: if W = V ⊗k ⊗ V ⊗` ⊗ V ⊗m

WB
| BB Idk ⊗M`,m
Mk,` ⊗Idm || BB
||| BB
|} | B!
WB W
BB ||
BB ||
B |
Mk+`,m BB! |
}|| Mk,`+m
W
Proposition 9.4.
Mk+`,m ◦ (Mk,` ⊗ Idm ) = Mk,`+m ⊗ (Idk ⊗ M`,m ).

27
 λ1 µ3

λ2 µ2
µ3 µ1
λ1 µ2
λ2 µ1 c23 c23
r s
r s
c13 c24 c13 c24
c
c14 c25 c14 c25

r−c s+c ν5 c15 c15

r−c s+c ν4 ν1
ν3 ν2
ν ν5
ν4 ν1
ν3 ν2
locally GP-web diagram wiring diagram plane partition

Figure 18: Illustration of how to transform a GP-web digram into a wiring diagram.

R-matrices satisfy the Yang-Baxter Equation which depends on two parameters α, β:

R23 (β)R13 (α + β)R12 (α) = R12 (α)R13 (α + β)R23 (β).

These diagrams satisfy a genrealized Yang-Baxter equation that depends on three parameters.
Proposition 9.5 (Generalized Yang-Baxter equation).

R23 (c23 )R13 (c13 )R12 (c12 ) = R12 (c012 )R13 (c013 )R23 (c023 ),

where 

c12 = min(c12 , c13 − c12 )
0

c013 = c12 + c23


 0
c23 = min(c23 , c13 − c12 ).
when c13 = α + β it reduces to the classical Yang-Baxter equation.

1 2 3 1 2 3 1 2 3 1 2 3

α α c12 c012
α+β α+β c13 c013
β β c23 c023

1 2 3 1 2 3 1 2 3 1 2 3

Yang-Baxter equation Generalized Yang-Baxter equation

Figure 19: Illustration of the Yang Baxter equation and the generalized Yang Baxter equation.

Exercise 9.6. Prove this generalized Yang-Baxter equation.

28
 This Generalized Yang-Baxter takes the wiring diagram with cij to an expression with c0ij . The
important point is to show that the inequalities on cij get translated to the inequalities on c0ij .
To finish this prove we need to generalize this transformation of inequalities to arbitrary wiring
diagrams.

9.2 String cone

[? ? ? This part needs polishing ? ? ?]
Let D be any wiring diagram for some reduced decomposition w = si1 · · · si` .

Example 9.7. We start from a wiring diagram of w = 4213 = s2 s3 s2 s1 and obtain a bibpartite
graph (see Figure 20).

1 2 3 4

c12 c12

c c13 −c12
c12 c13

c −c c c14 −c13
c13
c14

c23 −c23
c23
c23
−c14 c14

1 2 3 4 L1 L2 L3 L4
w = 4213

Figure 20: Illustration of how to obtain a bipartite graph from a wiring diagram of w = 4213.

We switch directions of some edges such that strands L1 , L2 , . . . , Li are directed down and
Li+1 , . . . , Ln are directed up.
GD,i look at directed path P from Li+1 to Li , each path gives an inequality: sum of weights of
edges in graph ≥ 0.

Example 9.8. Continuing from Example 9.7. For i = 2, there are two paths P : L3 → L2 with
the convention that cij = −cji . See Figure 21.

1 2 3 4

c12 c12
i=2
c12

c13 −c12
c12 c13
c13
−c12
c12 c13
c13
c14 −c13 c14
P2 −c13
P1 c14

c23 −c23
c23
c23
−c14 c14
−c23
c23
c23
−c14 c14

1 2 3 4 L1 L2 L3 L4 L1 L2 L3 L4
w = 4213

Figure 21: Paths from L3 to L2 in the bipartite graph obtained from a wiring diagram of w = 4213.

From the first path P1 , we obtain the inequality c41 + (c13 − c13 + c13 ) − c23 ≥ 0 which simplifies
to c41 + c13 + c32 ≥ 0. Form the second path P2 , we obtain the inequality c41 + (c13 − c13 ) + (c12 −
c12 + c12 ) + (c23 − c23 ) ≥ 0 which simplifies to c41 + c12 ≥ 0.

29
Claim This cone is what we really need. Every time we apply a 3-move and transform parameters
by the Generalized Yang-Baxter equation then the cone for one diagram transforms to the cone of
the diagram we obtain ... transform as needed. (piecewise linear continuous map)
If we have a wiring diagram for the associativity these inequalities become very simple (we get
the plane partition inequalities)

10 Lecture 10 3/09/2012
How about showing ∗-product is commutative. We know that associativity with Peiri rule shows
that ∗-product of Schur functions is equivalent to normal product of Schur functions (which are
commutative). Surprisingly, there is no direct way to see commutativity from the ∗-product picture.

10.1 Symmetries of Littlewood-Richardson coefficients

cνλµ := cλµν ∨ has S3 symmetry. It is not clear how to see this symmetry from the KT honeycomb.
0
Also cνλµ = cνλ0 µ0 where λ0 is the conjugate partition. There is a reformulation of KT honeycombs
in terms of puzzles that makes this symmetry explicit.
Also cνλµ = cνµλ and this is related to the Schützenberger involution. An interesting open question
is to understand this symmetry in the ∗-product setting (using Yang-Baxter equation...).

10.2 Total positivity

An m × n matrix is called totally positive or TP (totally nonnegative or TNN respectively)
if every minor is > 0 (≥ 0 respectively).
Example 10.1.  
a b
, a, b, c, d ≥ 0, ad − bc > 0.
c d

The following Lemma relates total positivity with combinatorics.

Lemma 10.2 (Lindström Lemma). Let G be a finite acyclic directed graph with P weights xeQ
on edges
and selected vertices: A1 , . . . , An , B1 , . . . , Bn . Define M = (Mij ) where Mij = p:Ai →Bj e∈P xe .
Then
X Y
n Y
det(M ) = (−1)sgn(w)
xe ,
(P1 ,...,Pn ),Pi :Ai →Bw(i) i=1 e∈Pi

where (P1 , . . . , Pn ) are families of non-crossing paths connection the As and Bs. Where non-crossing
means that no pair of paths Pi and Pj have a common vertex.
Remark 10.3. If we sum over all paths without the restriction that they are non-crossing we just
get a restatement of the definition of the determinant.
Example 10.4.
x

y z

 
x + ytz yt
t , M= , det(M ) = (x + ytz)t − yt · tz = xt.
tz t

30
Proof.

X Y X Y
n Y
det(M ) = (−1)sgn(w) Mi,w(i) = (−1)sgn(w) xe ,
w∈S i (P1 ,...,Pn ),Pi :Ai →Bw(i) i=1 e∈Pi

for any family of paths connecting As with Bs. Next we use the Involution principle. We build a
sign-reversing involution ϕ on families (P1 , . . . , Pn ) with a crossing. Find the min. Pi that intersects
a path and find first point c on Pi that intersects a path. On that path find the minimal Pj that
passes through c.
Pi c Pi0 c
Ai Bwi Ai Bwi
ϕ

Pj Pj0
Aj Bwj Aj Bwj

Figure 22: Illustration of sign reversing involution ϕ in the proof of Lindström Lemma.

We claim that this map is an involution, it preserves the weight of the path but reverses sign.

A special case of this Lemma is related to positivity.

Corollary 10.5. If G is a plane graph (embedded in a disk, see Figure 23) A1 , . . . , Am are on the
left-hand side of the boundary of the disk and B1 , . . . , Bn are on the right-hand side (ordered from
top to bottom). Assume edge weights xe > 0, then the matrix M = (Mij ) is totally nonnegative.

A1 B1
A2
B2
.. ..
. .
Am Bn

Figure 23: Plane graph G. If xe > 0, then the matrix M of Lindström’s Lemma is TNN.

Proof. Lindström Lemma implies that any minor is given by a nonnegative expression. That is, if
graph is planar all signs associated to noncrossing paths are positive.

31
 x

y z

 
x + ytz yt
t , M= , x, y, t, z > 0 det(M ) = (x + ytz)t − yt · tz = xt > 0.
tz t

Thus M is totally positive. Moreover, all 2 × 2 totally positive matrix can be written uniquely in
this way.
Claim: Any TNN matrix has this form (but not in a unique way).
The following is based on the work of Bernstein-Fomin-Zelevinsky related to previous results
by Lusztig. Assume that m = n and that M ∈ GLn . Recall the LUD-decomposition M = LU D
where L is lower triangular with ones on the diagonal, D is a diagonal matrix, and U is upper
triangular with ones on the diagonal.
It is well known that M is TNN iff L, U, D are TNN. So our first goal is to understand TNN
upper triangular matrices. Let Un be upper-triangular unipotent subgroup of GLn . The strata of
the TNN part of Un correspond to permutations w ∈ Sn .
 
1 x
Example 10.6. n = 2, where x ≥ 0. There are two possibilities: x = 0 in which case we
0 1
 
1 x
get the identity. If x > 0 we get the matrix . Given w, we write its wiring diagram (now
0 1
drawn from left to right). See Figure 24 for an example of this correspondence for S2 (in this
case we are not using the fact that the graph we obtain from the wiring diagram is bipartite, this
property will be important later).
1
A1 B1
 
12 = ι 1 0
M=
A2 B2 0 1
1

1 1
A1 B1  
1 x
21 = s1 x M=
0 1
A2 B2
1 1
w wiring diagram bipartite planar strata of U2
graph

Figure 24: How permutations in S2 correspond to strata of the TNN part of U2 .

For w ∈ Sn pick any reduced decomposition w = si1 · · · si` . Next we decompose M into a
product of certain elementary matrices. We illustrate this with an example.

32
1 1 A1 B1 A1 B1
x t x t
2 2 A2 B2 A2 B2
y y
3 3 A3 B3 A3 B3
z z
4 4 A4 B4 A4 B4

J1 (x) J2 (y) J3 (z) J1 (t)

Figure 25: Example of Bruhat cell.

 
1 2 3 4
Example 10.7. For w = = s1 s3 s2 s1 , we have
4 2 1 3
    
1 x 1 1 1 t
 1  1 y  1  1 
M = J1 (x)J2 (y)J3 (z)J1 (t) = 





 
1 1 1 z  1 
1 1 1 1
The matrices Ji (x) are called elementary Jacobi matrices.
Lemma 10.8. If si1 · · · si` is a reduced decomposition of w, the set of matrices {Ji1 (x), . . . , Ji` (x` ) |
x1 , . . . , x` > 0} depend only on the permutation w.
Proof. Seeing that the set of matrices does not change under a 2-move is easy. For 3-moves (see
Figure 26) we do the followng:

x z y0
y x0 z0

s1 s2 s1 s2 s1 s2

Figure 26: 3-move.

   
1 x + z xy 1 y0 y0z0
M = J1 (x)J2 (y)J1 (z) = 0 1 y  = 0 1 x0 + z 0 
0 0 1 0 0 1
Thus we obtain the system of equations
y0 =x+z y0 =x+z
xy
x0 +z 0 =y ⇒ z0 = x+z
xy yz
y0z0 = xy x00 = y − x+z = x+z .

Note that the solutions are subtraction-free. This means that for positive x, y, z we obtain a unique
positive solution x0 , y 0 , z 0 .

11 Lecture 10 3/14/12
Question 11.1. What is the number of potential nonzero minors of an upper
triangular matrices
(including the “empty” minor)? Why is it the Catalan numbers Cn = n+1
1 2n
n .

33
Example 11.2. For n = 1 there are 2 minors, for n = 2 there are five minors, for n = 3 there are
14 such minors.
Exercise 11.3. Answer the question above and explain what is the refinement of Catalan numbers
you get when you restrict to the size of the minor.
Pick a reduced decomposition of w = si1 · · · si` . Then the Bruhat cell is Bw = {Ji1 (x1 )Ji2 (x2 ) · · · Ji` (x` ) |
x1 , . . . , x` > 0} where Ji (x) is the n × n upper triangular matrix with ones on the diagonal, the
(i, i + 1) entry is x and the other entries are 0.
 
1
 1 x 
 
Ji (x) =  .. 
 . 
1
See Figure 25 for an Example of a Bruhat cell. We also showed:
Lemma 11.4. Bw depends only on w (not on its reduced decomposition).

11.1 Bruhat order

Definition 11.5. The (strong) Bruhat order on Sn is the partial order on permutations with
covering relations u l w such that
1. u = w(i, j)
2. `(u) + 1 = `(w).
Equivalently, u ≤ w if any (or some for another equivalent definition) reduced decomposition for
w = si1 · · · si` contains a subword sj1 · · · sjr which is a reduced decomposition for u.
Note that it is not true that you can pick a reduced decomposition of u and add reflections to
obtain a reduced decomposition of w. [? ? ? give an example of this ? ? ?] See Figure 27 for the
Hasse diagram of the Bruhat order in S3 .
s1 s2 s1

s1 s2 s2 s1

s1 s2

Figure 27: The strong Bruhat order in S3 .

Theorem 11.6 (Bernstein-Fomin-Zelevinsky).

`
1. The TNN of Un decompose as w∈Sn Bw ,
2. Bw ∼
= R`>0 where ` = `(w), and the isomorphism is Ji1 (x1 ) · · · Ji` (x` ) 7→ (x1 , x2 , . . . , x` ).
3. The closure B u ⊆ B v if and only if u ≤ w in the strong Bruhat order.
Remark 11.7. To start seeing why part 2. of the theorem above is related to the Bruhat order if
xi = 0 then Jij (0) = I is the identity matrix. This is analogous to considering subwords (one has
to check that in the closure we get xi ≥ 0).
[? ? ? say something about nonzero minors ? ? ?]
How about the whole of GLn ?

34
 1 a b
 
0 1 c
s1 s2 s1 0 0 1
1 x xy 1 x 0
   
0 1 y 0 1 y
s1 s2 s2 s1 0 0 1 0 0 1

1 x 0 1 0 0
   

s1 s2 0 1 0 0 1 y
0 0 1 0 0 1

ι
1 0 0
 
0 1 0
Example 11.8. 0 0 1

Figure 28: Illustration of correspondence between w in the Bruhat order of S3 and TNN upper-
triangular matrices. The zero minors are in a blue square.

11.2 Fomin-Zelevinsky double wiring diagrams and double Bruhat cells

We take two permutations u and w and shuffle two reduced decompositons of these permutations.
Example 11.9. u−1 = s1 s2 and w−1 = s2 s3 s1 .
1 1
2 2
3 3
4 4

Figure 29: Example of double wiring diagram.

We can convert these double wiring diagrams into trivalent graphs as shown in Figure 30.
We have two types of Jacobi matrices Ji (x) (x on the (i, i + 1) entry) as before and Ji (x) which
is the identity matrix and x on the (i + 1, i) entry. We get an analogous decomposition as in the
case of single Bruhat cells.
1 1 1 1
2 2 2 2
3 3 3 3
4 4 4 4

Figure 30: Going from double wiring diagram to trivalent graph.

Example 11.10. Continuing from Example 11.9, M = J1 (x1 )J2 (x2 )J2 (x3 )J3 (x4 )J1 (x5 ).
Definition 11.11. For u, w ∈ Sn with reduced decompositions u = si1 si2 · · · si`(u) and w =
sj1 sj2 · · · sj`(w) the double Bruhat cell is

Bu,w = {Ji1 (x1 )Jj1 (t1 )Ji1 (x2 )Jj1 · · · | x1 , x2 , . . . , x`(u) , t1 , t2 , . . . , t`(w) > 0}

35
Theorem 11.12 (Lusztig, Fomin-Zelevinsky).
`
1. The TNN of GLn decompose as u,w∈Sn Bu,w ,
2. Bu,w ∼
`(u)+`(w)+n
= R>0 , and the isomorphism is Ji1 (x1 ) · · · Ji` (x` ) 7→ (x1 , x2 , . . . , x`(u) , t1 , t2 , . . . , t`(w) ).
3. The closure B u,w ⊆ B v,z if and only if u ≤ v and w ≤ z in the strong Bruhat order.

11.3 Totally nonnegative Grassmannian

Definition 11.13. Gr≥0 (k, n, R) are the elements in Gr(k, n, R) such that all Plücker coordinates
∆I ≥ 0 (we use only maximal minors).

In this setting the matroid strata are

SM>0
= {A ∈ Gr(k, n) | ∆I (A) > 0 for I ∈ M, ∆J (A) = 0 for J 6∈ M}.
` >0
We also have that Gr≥0 (k, n) = SM .
Recall that for Gr(k, n) the matroid stratification can get complicated (recall Lecture ??, Mnëv’s
Universality Theorem,. . . ) but the TNN Grassmannian has a “nice” stratification.

11.4 Relation to classical total positivity

Let B be a k × (n − k) matrix. From this matrix we obtain a k × n matrix A such that there is a
one-to-one correspondence between minors of B (of all sizes) and maximal minors of A (and with
the same sign).
 
  1 ±b k1 ±b k2 · · · ±b k n−k
b11 · · · b1 n−k  .. 
 . ··· 
 .. ..   
B= . .  , A =  1 b b · · · b 
 31 32 3 n−k 
bk1 · · · bk n−k  1 −b21 −b22 · · · −b2 n−k 
1 b11 b12 · · · b1 n−k

So the classical total positivity embeds on the TNN Grassmannian. Moreover, there is a symmetry
feature in the latter (take first column and place it at the end and change sign by (−1)k−1 . This
operation does not change the TNN Grassmannian. There is no such operation in the classical
setting.

Thanks: to Darij Grinberg, Nan Li and Tom Roby for comments, proofreading and board photos!

36