arXiv:1806.05307v1 [math.CO] 14 Jun 2018
POSITIVE GRASSMANNIAN AND POLYHEDRAL
SUBDIVISIONS
ALEXANDER POSTNIKOV
Abstract. The nonnegative Grassmannian is a cell complex with rich geometric, algebraic, and combinatorial structures. Its study involves interesting
combinatorial objects, such as positroids and plabic graphs. Remarkably, the
same combinatorial structures appeared in many other areas of mathematics
and physics, e.g., in the study of cluster algebras, scattering amplitudes, and
solitons. We discuss new ways to think about these structures. In particular,
we identify plabic graphs and more general Grassmannian graphs with polyhedral subdivisions induced by 2-dimensional projections of hypersimplices.
This implies a close relationship between the positive Grassmannian and the
theory of fiber polytopes and the generalized Baues problem. This suggests
natural extensions of objects related to the positive Grassmannian.
1. Introduction
The geometry of the Grassmannian Gr(k, n) is related to combinatorics of the
hypersimplex ∆kn . Gelfand, Goresky, MacPherson, Serganova [GGMS87] studied
the hypersimplex as the moment polytope for the torus action on the complex
Grassmannian. In this paper we highlight new links between geometry of the positive Grassmannian and combinatorics of the hypersimplex ∆kn .
[GGMS87] studied the matroid stratification of the Grassmannian Gr(k, n, C),
whose strata are the realization spaces of matroids. They correspond to matroid
polytopes living inside ∆kn . In general, matroid strata are not cells. In fact, according to Mnëv’s universality theorem [Mnë88], the matroid strata can be as complicated as any algebraic variety. Thus the matroid stratification of the Grassmannian
can have arbitrarily bad behavior.
There is, however, a semialgebraic subset of the real Grassmannian Gr(k, n, R),
called the nonnegative Grassmannian Gr≥0 (k, n), where the matroid stratification
exhibits a well behaved combinatorial and geometric structure. Its structure, which
is quite rich and nontrivial, can nevertheless be described in explicit terms. In some
way, the nonnegative Grassmannian is similar to a polytope.
The notion of a totally positive matrix, that is a matrix with all positive minors, originated in pioneering works of Gantmacher, Krein [GK35], and Schoenberg
[Sch30]. Since then such matrices appeared in many areas of pure and applied mathematics. Lusztig [Lus94, Lus98a, Lus98b] generalized the theory of total positivity
Date: February 27, 2018.
2010 Mathematics Subject Classification. Primary 05E; secondary 52B, 52C, 13F60, 81T.
Key words and phrases. Total positivity, positive Grassmannian, hypersimplex, matroids,
positroids, cyclic shifts, Grassmannian graphs, plabic graphs, polyhedral subdivisions, triangulations, zonotopal tilings, associahedron, fiber polytopes, Baues poset, generalized Baues problem, flips, cluster algebras, weakly separated collections, scattering amplitudes, amplituhedron,
membranes.
1
2
ALEXANDER POSTNIKOV
in the general context of Lie theory. He defined the positive part for a reductive
Lie group G and a generalized flag variety G/P . Rietsch [Rie98, Rie99] studied
its cellular decomposition. Lusztig’s theory of total positivity has close links with
his theory of canonical bases [Lus90, Lus92, Lus93] and Fomin-Zelevinsky’s cluster
algebras [FZ02a, FZ02b, FZ03, BFZ05, FZ07].
[Pos06] initiated a combinatorial approach to the study of the positive Grassmannian. The positive (resp., nonnegative) Grassmannian Gr>0 (k, n) (Gr≥0 (k, n))
was described as the subset of the Grassmannian Gr(k, n, R) where all Plücker coordinates are positive (resp., nonnegative). This “elementary” definition agrees
with Lusztig’s general notion [Lus98a] of the positive part of G/P in the case when
G/P = Gr(k, n).
The positroid cells, defined as the parts of matroid strata inside the nonnegative
Grassmannian, turned out to be indeed cells. (The term “positroid” is an abbreviation for “positive matroid.”) The positroid cells form a CW-complex. Conjecturally,
it is a regular CW-complex, and the closure of each positroid cell is homeomorphic
to a closed ball. This positroid stratification of Gr≥0 (k, n) is the common refinement of n cyclically shifted Schubert decompositions [Pos06]. Compare this with
the result [GGMS87] that the matroid stratification of Gr(k, n) is the common refinement of n! permuted Schubert decompositions. The cyclic shift plays a crucial
role in the study of the positive Grassmannian. Many objects associated with the
positive Grassmannian exhibit cyclic symmetry.
Positroid cells were identified in [Pos06] with many combinatorial objects, such as
decorated permutation, Grassmann necklaces, etc. Moreover, an explicit birational
subtraction-free parametrization of each cell was described in terms of plabic graphs,
that is, planar bicolored graphs, which are certain graphs embedded in a disk with
vertices colored in two colors.
Remarkably, the combinatorial structures that appeared in the study of the
positive Grassmannian also surfaced and played an important role in many different areas of mathematics and physics. Scott [Sco05, Sco06] and [OPS15] linked
these objects with cluster algebra structure on the Grassmannian and with LeclercZelevinsky’s quasi-commutting families of quantum minors and weakly separated
collections. Corteel and Williams [CW07] applied Le-diagrams (which correspond
to positroids) to the study of the partially asymmetric exclusion process (PASEP).
Knutson, Lam, and Speyer [KLS13] proved that the cohomology classes of the
positroid varieties (the complexifications of the positroid cells) are given by the
affine Stanley symmetric functions, which are dual to Lapointe-Lascoux-Morse kSchur functions. They also linked positroids with theory of juggling. Plabic graphs
appeared in works of Chakravarty, Kodama, and Williams [CK09, KoW11, KoW14]
as soliton solutions of the Kadomtsev-Petviashvili (KP) equation, which describes
nonlinear waves. Last but not least, plabic graphs appeared under the name of
on-shell diagrams in the work by Arkani-Hamed et al [ABCGPT16] on scattering
amplitudes in N = 4 supersymmetric Yang-Mills (SYM) theory. They play a role
somewhat similar to Feynman diagrams, however, unlike Feynman diagrams, they
represent on-shell processes and do not require introduction of virtual particles.
In this paper, we review some of the main constructions and results from [Pos06,
PSW09, OPS15] related to the positive Grassmannian. We extend these constructions in the language of Grassmannian graphs. The parametrization of a positroid
cell in Gr≥0 (k, n) given by a Grassmannian graph can be thought of as a way to
POSITIVE GRASSMANNIAN AND POLYHEDRAL SUBDIVISIONS
3
“glue” the positroid cell out of “little positive Grassmannians” associated with vertices of the graph. The idea to think about parametrizations of cells as gluings of
Grassmannians came originally from physics [ABCGPT16], where vertices of onshell diagrams (i.e., plabic graphs) were viewed as little Grassmannians Gr(1, 3)
and Gr(2, 3).
We link this construction of parametrizations of Gr>0 (k, n) given by Grassmannian graphs with the study of polyhedral subdivisions induced by 2-dimensional
cyclic projections π : ∆kn → Q of the hypersimplex. Reduced Grassmannian
graphs parametrizing the positive Grassmannian Gr>0 (k, n) turn out to be in bijection with π-induced polyhedral subdivisions. Thus gluing of Grassmannians
from smaller Grassmannians is equivalent to subdividing polytopes into smaller
polytopes. The study of π-induced subdivisions for projections of polytopes is the
subject of Billera-Sturmfels’ theory [BS92] of fiber polytopes and the generalized
Baues problem (GBP) posed by Billera, Kapranov, Sturmfels [BKS94]. We also
mention the result of Galashin [Gal16] where plabic graphs are identified with sections of zonotopal tilings, and the construction from the joint work [LP] with Lam
on polypositroids where plabic graphs are viewed as membranes, which are certain
2-dimensional surfaces in higher dimensional spaces.
The correspondence between parametrizations of the positive Grassmannian and
polyhedral subdivisions leads to natural generalizations and conjectures. We discuss
a possible extension of constructions of this paper to “higher positive Grassmannians” and amplituhedra of Arkani-Hamed and Trnka [AT14].
I thank Federico Ardila, Nima Arkani-Hamed, Arkady Berenstein, David Bernstein, Lou Billera, Jacob Bourjaily, Freddy Cachazo, Miriam Farber, Sergey Fomin,
Pavel Galashin, Israel Moiseevich Gelfand, Oleg Gleizer, Alexander Goncharov,
Darij Grinberg, Alberto Grünbaum, Xuhua He, Sam Hopkins, David Ingerman,
Tamás Kálmán, Mikhail Kapranov, Askold Khovanskii, Anatol Kirillov, Allen Knutson, Gleb Koshevoy, Thomas Lam, Joel Lewis, Gaku Liu, Ricky Liu, George
Lusztig, Thomas McConville, Karola Mészáros, Alejandro Morales, Gleb Nenashev, Suho Oh, Jim Propp, Pavlo Pylyavskyy, Vic Reiner, Vladimir Retakh, Konni
Rietsch, Tom Roby, Yuval Roichman, Paco Santos, Jeanne Scott, Boris Shapiro,
Michael Shapiro, David Speyer, Richard Stanley, Bernd Sturmfels, Dylan Thurston,
Jaroslav Trnka, Wuttisak Trongsiriwat, Vladimir Voevodsky, Lauren Williams,
Hwanchul Yoo, Andrei Zelevinsky, and Günter Ziegler for insightful conversations.
These people made a tremendous contribution to the study of the positive Grassmannian and related combinatorial, algebraic, geometric, topological, and physical
structures. Many themes we discuss here are from past and future projects with
various subsets of these people.
2. Grassmannian and matroids
be the set of k-element
Fix integers 0 ≤ k ≤ n. Let [n] := {1, . . . , n} and [n]
k
subsets of [n].
The Grassmannian Gr(k, n) = Gr(k, n, F) over a field F is the variety of kdimensional linear subspaces in Fn . More concretely, Gr(k, n) is the space of k × nmatrices of rank k modulo the left action of GL(k) = GL(k, F). Let [A] = GL(k) A
be the element of Gr(k, n) represented by matrix A.
4
ALEXANDER POSTNIKOV
Maximal minors ∆I (A) of such matrices A, where I ∈ [n]
k , form projective
coordinates on Gr(k, n), called the Plücker coordinates. For [A] ∈ Gr(k, n), let
[n]
M(A) := {I ∈
| ∆I (A) 6= 0}.
k
The sets of the form M(A) are a special kind of matroids, called F-realizable ma
troids. Matroid strata are the realization spaces of realizable matroids M ⊂ [n]
k :
SM := {[A] ∈ Gr(k, n) | M(A) = M}.
The matroid stratification is the disjoint decomposition
G
SM .
Gr(k, n) =
M realizable matroid
The Gale order “” (or the coordinatewise order) is the partial order on [n]
k given
by {i1 < · · · < ik } {j1 < · · · < jk }, if ir ≤ jr for r ∈ [k]. Each matroid M has a
unique minimal element Imin (M) with respect to the Gale order.
For I ∈ [n]
k , the Schubert cell ΩI ⊂ Gr(k, n) is given by
G
ΩI := {[A] ∈ Gr(k, n) | I = Imin (M(A)} =
SM .
M: I=Imin (M)
F
They form the Schubert decomposition Gr(k, n) = ΩI . Clearly, for a realizable
matroid M, we have SM ⊂ ΩI if and only if I = Imin (M).
The symmetric group Sn acts on Gr(k, n) by permutations w([v1 , . . . , vn ]) =
[vw(1) , . . . , vw(n) ] of columns of [A] = [v1 , . . . , vn ] ∈ Gr(k, n).
It is clear that, see [GGMS87], the matroid stratification of Gr(k, n) is the common refinement of the n! permuted Schubert decompositions. In other words, each
matroid stratum SM is an intersection of permuted Schubert cells:
\
w(ΩIw ).
SM =
w∈Sn
Indeed, if we know the minimal elements of a set M ⊂
orderings of [n], we know the set M itself.
[n]
k
with respect to all n!
3. Positive Grassmannian and positroids
Fix the field F = R. Let Gr(k, n) = Gr(k, n, R) be the real Grassmannian.
Definition 3.1. [Pos06, Definition 3.1] The positive Grassmannian Gr>0 (k, n)
(resp., nonnegative Grassmannian Gr≥0 (k, n)) is the semialgebraic set of elements
[A] ∈ Gr(k, n) represented by k × n matrices A with all positive maximal minors
∆I (A) > 0 (resp., all nonnegative maximal minors ∆I (A) ≥ 0).
This definition agrees with Lusztig’s general definition [Lus98a] of the positive
part of a generalized flag variety G/P in the case when G/P = Gr(k, n).
Definition 3.2. [Pos06, Definition 3.2] A positroid cell ΠM ⊂ Gr≥0 (k, n) is a
nonempty intersection of a matroid stratum with the nonnegative Grassmannian:
ΠM := SM ∩ Gr≥0 (k, n).
POSITIVE GRASSMANNIAN AND POLYHEDRAL SUBDIVISIONS
5
such that ΠM is nonempty. The
A positroid of rank k is a collection M ⊂ [n]
k
positroid stratification of the nonnegative Grassmannian is the disjoint decomposition of Gr≥0 (k, n) into the positroid cells:
G
ΠM .
Gr≥0 (k, n) =
M is a positroid
Clearly, positroids, or positive matroids, are a special kind of matroids. The positive Grassmannian Gr>0 (k, n) itself is the top positroid cell Π([n]) for the uniform
k
.
matroid M = [n]
k
The cyclic shift is the map c̃ : Gr(k, n) → Gr(k, n) acting on elements [A] =
[v1 , . . . , vn ] ∈ Gr(k, n) by
c̃ : [v1 , . . . , vn ] 7−→ [v2 , v3 , . . . , vn , (−1)k−1 v1 ].
The shift c̃ induces the action of the cyclic group Z/nZ on the Grassmannian
Gr(k, n), that preserves its positive part Gr>0 (k, n). Many of the objects associated
with the positive Grassmannian exhibit cyclic symmetry. This cyclic symmetry is
a crucial ingredient in the study of the positive Grassmannian.
Theorem 3.3. [Pos06, Theorem 3.7] The positroid stratification is the common
refinement of n cyclically shifted Schubert decompositions restricted to Gr≥0 (k, n).
In other words, each positroid cell ΠM is given by the intersection of the nonnegative
parts of n cyclically shifted Schubert cells:
ΠM =
n−1
\
c̃ i (ΩIi ∩ Gr≥0 (k, n)).
i=0
So the positroid cells require intersecting n cyclically shifted Schubert cells, which
is a smaller number than n! permuted Schubert cells needed for general matroid
strata. In fact, the positroid cells ΠM (unlike matroid strata) are indeed cells.
Theorem 3.4. [Pos06, Theorem 3.5], [PSW09, Theorem 5.4] The positroid cells
ΠM are homeomorphic to open balls. The cell decomposition of Gr≥0 (k, n) into the
positroid cells ΠM is a CW-complex.
Conjecture 3.5. [Pos06, Conjecture 3.6] The positroid stratification of the nonnegative Grassmannian Gr≥0 (k, n) is a regular CW-complex. In particular, the
closure ΠM of each positroid cell in Gr≥0 (k, n) is homeomorphic to a closed ball.
This conjecture was motivated by a similar conjecture of Fomin and Zelevinsky
on double Bruhat cells [FZ99]. Up to homotopy-equivalence this conjecture was
proved by Rietsch and Williams [RW10]. A major step towards this conjecture was
recently achieved by Galashin, Karp, and Lam, who proved it for the top cell.
Theorem 3.6. [GKL17, Theorem 1.1] The nonnegative Grassmannian Gr≥0 (k, n)
is homeomorphic to a closed ball of dimension k(n − k).
By Theorem 3.3, positroids M and positroid cells ΠM ⊂ Gr≥0 (k, n) correspond
to certain sequences (I0 , I1 , . . . , In−1 ). Let us describe this bijection explicitly.
Definition 3.7. [Pos06, Definition 16.1] A Grassmann necklace J = (J1 , J2 , . . . , Jn )
such that, for any i ∈ [n], either
of type (k, n) is a sequence of elements Ji ∈ [n]
k
Ji+1 = (Ji \ {i}) ∪ {j} or Ji+1 = Ji , where the indices i are taken (mod n).
6
ALEXANDER POSTNIKOV
The cyclic permutation c ∈ Sn is given by c : i 7→ i + 1 (mod n). The action
of the symmetric group Sn on [n] induces the Sn -action on [n]
and on subsets
k
[n]
of k . Recall that Imin (M) is the minimal element of a matroid M in the Gale
order. For a matroid M, let
J (M) := (J1 , . . . , Jn ), where
Ji+1 = ci (Imin (c−i (M))), for i = 0, . . . , n − 1.
Theorem 3.8. [Pos06, Theorem 17.1] The map M 7→ J (M) is a bijection between
positroids M of rank k on the ground set [n] and Grassmann necklaces of type (k, n).
The sequence (I0 , I1 , . . . , In−1 ) associated with M as in Theorem 3.3 is related
to the Grassmann necklace (J1 , . . . , Jn ) of M by Ii = c−i (Ji+1 ), for i = 0, . . . , n−1.
The following result shows how to reconstruct
a positroid M from its Grassmann
,
the
Schubert
matroid is MI := {J ∈ [n]
necklace, cf. Theorem 3.3. For I ∈ [n]
k |
k
I J}, where “” is the Gale order.
Theorem 3.9. [Oh11, Theorem 6] For a Grassmann necklace J = (J1 , . . . , Jn ),
the associated positroid M(J ) = M is given by
M=
n−1
\
ci (MIi ),
i=0
−i
where Ii = c (Ji+1 ).
Let us describe positroids in the language of convex geometry. The hypersimplex
[n]
∆kn := conv eI | I ∈
k
P
is the convex hull of the nk points eI = i∈I ei , for all I ∈ [n]
k . Here e1 , . . . , en
is the standard basis in Rn . For a subset M ⊂ [n]
k , let PM := conv{eI | I ∈ M}
be the convex hull of vertices of ∆kn associated with elements of M.
By [GGMS87], M is a matroid if and only if every edge of the polytope PM
with |I ∩ J| = k − 1. Here is an analogous
has the form [eI , eJ ], for I, J ∈ [n]
k
description of positroids, which is not hard to derive from Theorem 3.9.
is a positroid if and only if
Theorem 3.10. [LP] A nonempty subset M ⊂ [n]
k
(1) Every edge of PM has the form [eI , eJ ], for I, J ∈ [n]
k with |I ∩ J| = k − 1.
(2) Every facet of PM is given by xi + xi+1 + · · · + xj = aij for some cyclic
interval {i, i + 1, . . . , j} ⊂ [n] and aij ∈ Z.
Many of the results on the positive Grassmannian are based on an explicit birational parametrization [Pos06] of the positroid cells ΠM in terms of plabic graphs.
In the next section we describe a more general class of Grassmannian graphs that
includes plabic graphs.
4. Grassmannian graphs
Definition 4.1. A Grassmannian graph is a finite graph G = (V, E), with vertex
set V and edge set E, embedded into a disk (and considered up to homeomorphism)
with n boundary vertices b1 , . . . , bn ∈ V of degree 1 on the boundary of the disk
(in the clockwise order), and possibly some internal vertices v in the interior of the
POSITIVE GRASSMANNIAN AND POLYHEDRAL SUBDIVISIONS
7
disk equipped with integer parameters h(v) ∈ {0, 1, . . . , deg(v)}, called helicities of
vertices. Here deg(v) is the degree of vertex v. We say that an internal vertex v is
of type (h, d) if d = deg(v) and h = h(v).
The set of internal vertices of G is denoted by Vint = V \ {b1 , . . . , bn }, and the set
of internal edges, i.e., the edges which are not adjacent to the boundary vertices, is
denoted by Eint ⊂ E. The internal subgraph is Gint = (Vint , Eint ).
A perfect orientation of a Grassmannian graph G is a choice of directions for
all edges e ∈ E of the graph G such that, for each internal vertex v ∈ Vint with
helicity h(v), exactly h(v) of the edges adjacent to v are directed towards v and the
remaining deg(v) − h(v) of adjacent edges are directed away from v. A Grassmannian graph is called perfectly orientable if it has a perfect orientation.
The helicity of a Grassmannian graph G with n boundary vertices is the number
h(G) given by
X
(h(v) − deg(v)/2).
h(G) − n/2 =
v∈Vint
For a perfect orientation O of G, let I(O) be the set of indices i ∈ [n] such
that the boundary edge adjacent to bi is directed towards the interior of G in the
orientation O.
Lemma 4.2. For a perfectly orientable Grassmannian graph G and any perfect
orientation O of G, we have |I(O)| = h(G). In particular, in this case, h(G) ∈
{0, 1, . . . , n}.
Remark 4.3. This lemma expresses the Helicity Conservation Law. We leave it as
an exercise for the reader.
For a perfectly orientable Grassmannian graph G of helicity h(G) = k, let
[n]
.
M(G) = {I(O) | O is a perfect orientation of G} ⊂
k
Here is one result that links Grassmannian graphs with positroids.
Theorem 4.4. For a perfectly orientable Grassmannian graph G with h(G) = k,
the set M(G) is a positroid of rank k. All positroids have form M(G) for some G.
Definition 4.5. A strand α in a Grassmannian graph G is a directed walk along
edges of G that either starts and ends at some boundary vertices, or is a closed
walk in the internal subgraph Gint , satisfying the following Rules of the Road: For
each internal vertex v ∈ Vint with adjacent edges labelled a1 , . . . , ad in the clockwise
order, where d = deg(v), if α enters v through the edge ai , it leaves v through the
edge aj , where j = i + h(v) (mod d).
A Grassmannian graph G is reduced if
(1) There are no strands which are closed loops in the internal subgraph Gint .
(2) All strands in G are simple curves without self-intersections. The only
exception is that we allow strands bi → v → bi where v ∈ Vint is a boundary
leaf, that is a vertex of degree 1 connected with bi by an edge.
(3) Any two strands α 6= β cannot have a bad double crossing, that is, a pair
of vertices u 6= v such that both α and β pass through u and v and both
are directed from u to v. (We allow double crossings where α goes from u
to v and β goes from v to u.)
8
ALEXANDER POSTNIKOV
(4) The graph G has no vertices of degree 2.
The decorated strand permutation w = wG of a reduced Grassmannian graph G
is the permutation w : [n] → [n] with fixed points colored in colors 0 or 1 such that
(1) w(i) = j if the strand that starts at the boundary vertex bi ends at the
boundary vertex bj .
(2) For a boundary leaf v connected to bi , the decorated permutation w has
fixed point w(i) = i colored in color h(v) ∈ {0, 1}.
A complete reduced Grassmannian graph G of type (k, n), for 0 ≤ k ≤ n, is
a reduced Grassmannian graph whose decorated strand permutation is given by
w(i) = i + k (mod n). In addition, for k = 0 (resp., for k = n), we require that G
only has n boundary leaves of helicity 0 (resp., of helicity 1) and no other internal
vertices.
Theorem 4.6. cf. [Pos06, Corollaries 14.7 and 14.10] (1) For any permutation
w : [n] → [n] with fixed points colored in 0 or 1, there exists a reduced Grassmannian
graph G whose decorated strand permutation wG is w.
(2) Any reduced Grassmannian graph is perfectly orientable. Moreover, it has an
acyclic perfect orientation.
(3) A reduced Grassmannian graph G is complete of type (k, n) if and only if its
helicity equals h(G) = k and the number of internal faces (excluding n boundary
faces) equals
X
f (h(v), deg(v)).
f (k, n) −
v∈Vint
where f (k, n) = (k − 1)(n − k − 1). A reduced Grassmannian graph is complete if
and only if it is not a proper induced subgraph of a larger reduced Grassmannian
graph.
Figure 1. Two complete reduced Grassmannian graphs of type
(2, 5) with 2 internal faces (left) and 1 internal face (right). The
internal vertices of types (1, 3) and (1, 4) are colored in white, the
type (2, 3) vertices colored in black, and the type (2, 4) vertex is
“chessboard” colored.
Let us now describe a partial ordering and an equivalence relation on Grassmannian graphs.
Definition 4.7. For two Grassmannian graphs G and G′ , we say that G refines
G′ (and that G′ coarsens G), if G can be obtained from G′ by a sequence of the
following operations: Replace an internal vertex of type (h, d) by a complete reduced
Grassmannian graph of type (h, d).
The refinement order on Grassmannian graphs is the partial order G ≤ref G′ if
G refines G′ . We say that G′ covers G, if G′ covers G in the refinement order.
POSITIVE GRASSMANNIAN AND POLYHEDRAL SUBDIVISIONS
9
Two Grassmannian graphs G and G′ are refinement-equivalent if they are in
the same connected component of the refinement order ≤ref , that is, they can be
obtained from each other by a sequence of refinements and coarsenings.
Definition 4.8. A Grassmannian graph is called a plabic graph if it is a minimal
element in the refinement order.
The following is clear.
Lemma 4.9. A Grassmannian graph is a plabic graph if and only if each internal
vertex in the graph has type (1, 3), (2, 3), (0, 1), or (1, 1).
In drawings of plabic and Grassmannian graphs, we color vertices of types (1, d)
in white color, and vertices of types (d − 1, d) in black color.
Let us now describe almost minimal elements in the refinement order.
Definition 4.10. A Grassmannian graph G is called almost plabic if it covers a
plabic graph (a minimal element) in the refinement order.
For example, the two graphs shown on Figure 4 are almost plabic. The following
lemma is also straightforward from the definitions.
Lemma 4.11. Each almost plabic Grassmannian graph G has exactly one internal
vertex (special vertex) of type (1, 4), (2, 4), (3, 4), (0, 2), (1, 2), or (2, 2), and all
other internal vertices of types (1, 3), (2, 3), (0, 1), or (1, 1). An almost plabic
graph with a special vertex of type of type (1, 4), (2, 4), or (3, 4) covers exactly two
plabic graphs. An almost plabic graph with a special vertex of type (0, 2), (1, 2), or
(2, 2) covers exactly one plabic graph.
Note that a reduced Grassmannian graph cannot contain any vertices of degree
2. So each reduced almost plabic graph covers exactly two reduced plabic graphs.
Definition 4.12. Two plabic graphs are connected by a move of type (1, 4), (2, 4),
or (3, 4), if they are both covered by an almost plabic graph with a special vertex
of the corresponding type. Two plabic graphs G and G′ are move-equivalent if they
can be obtained from each other by a sequence of such moves.
Figure 2. Three types of moves of plabic graphs: (1,4)
contraction-uncontraction of white vertices, (2,4) square move,
(3,4) contraction-uncontraction of black vertices.
Let us say that vertices of types (1, 2), (0, d), (d, d) (except boundary leaves) are
extraneous. A reduced graph cannot have a vertex of this form.
Theorem 4.13. (1) For two reduced Grassmannian graphs G and G′ , the graphs
are refinement-equivalent if and only if they have the same decorated strand permutation wG = wG′ .
(2) cf. [Pos06, Theorem 13.4] For two reduced plabic graphs G and G′ , the
following are equivalent:
(a) The graphs are move-equivalent.
10
ALEXANDER POSTNIKOV
(b) The graphs are refinement-equivalent.
(c) The graphs have the same decorated strand permutation wG = wG′ .
(3) A Grassmannian graph is reduced if and only if it has no extraneous vertices
and is not refinement-equivalent to a graph with a pair of parallel edges (two edges
between the same vertices), or a loop-edge (an edge with both ends attached to the
same vertex).
(4) A plabic graph is reduced if and only if it has no extraneous vertices and is
not move-equivalent to a plabic graph with a pair of parallel edges or a loop-edge.
Remark 4.14. Plabic graphs are similar to wiring diagrams that represent decompositions of permutations into products of adjacent transpositions. In fact, plabic
graphs extend the notions of wiring diagrams and, more generally, double wiring diagrams of Fomin-Zelevinsky [FZ99], see [Pos06, Remark 14.8, Figure 18.1]. Moves
of plabic graphs are analogous to Coxeter moves of decompositions of permutations.
Reduced plabic graphs extend the notion of reduced decompositions of permutations.
Let us now summarize the results about the relationship between positroids,
Grassmannian and plabic graphs, decorated permutations, and Grassmann necklaces. For a decorated permutation w : [n] → [n] (a permutation with fixed points
colored 0 or 1), define J (w) := (J1 , . . . , Jn ), where
Ji = {j ∈ [n] | c−i+1 w−1 (j) > c−i+1 (j)} ∪ {j ∈ [n] | w(j) = j colored 1}.
The helicity of w is defined as h(w) := |J1 | = · · · = |Jn |. Conversely, for a Grassmann necklace J = (J1 , . . . , Jn ), let
j if Ji+1 = (Ji \ {i}) ∪ {j},
i (colored 0) if i 6∈ Ji = Ji+1 ,
w(J ) := w, where w(i) =
i (colored 1) if i ∈ Ji = Ji+1 .
Theorem 4.15. cf. [Pos06] The following sets are in one-to-one correspondence:
(1) Positroids M of rank k on n elements.
(2) Decorated permutation w of size n and helicity k.
(3) Grassmann necklaces J of type (k, n).
(4) Move-equivalence classes of reduced plabic graphs G with n boundary vertices and helicity h(G) = k.
(5) Refinement-equivalence classes of reduced Grassmannian graphs G′ with n
boundary vertices and helicity h(G′ ) = k.
The following maps (described above in the paper) give explicit bijection between
these sets and form a commutative diagram:
(1) Reduced Grassmannian/plabic graphs to positroids: G 7→ M(G).
(2) Reduced Grassmannian/plabic graphs to decorated permutations: G 7→ wG .
(3) Positroids to Grassmann necklaces: M 7→ J (M).
(4) Grassmann necklaces to positroids: J 7→ M(J ),
(5) Grassmann necklaces to decorated permutations: J 7→ w(J ).
(6) Decorated permutations to Grassmann necklaces: w 7→ J (w).
Proof of Theorems 4.4, 4.6, 4.13, 4.15. In case of plabic graphs, most of these results were proved in [Pos06]. The extension of results to Grassmannian graphs
follows from a few easy observations.
POSITIVE GRASSMANNIAN AND POLYHEDRAL SUBDIVISIONS
11
Let G and G′ be a pair of Grassmannian graphs such that G refines G′ . Any
perfect orientation of G induces a perfect orientation of G′ . Conversely, any perfect
orientation of G′ can be extended (not uniquely, in general) to a perfect orientation
of G. Thus G is perfectly orientable if and only if G′ is perfectly orientable, and
M(G) = M(G′ ) and h(G) = h(G′ ). Any strand of G corresponds to a strand
of G′ . The graph G is reduced if and only if G′ is reduced. If they are reduced,
then they have the same decorated strand permutation wG = wG′ . Finally, any
Grassmannian graph can be refined to a plabic graph. So the results for plabic
graphs imply the results for Grassmannian graphs.
5. Weakly separated collections and cluster algebras
are weakly separated
Definition 5.1. [Sco05], cf. [LZ98] Two subsets I, J ∈ [n]
k
if there is no a < b < c < d such that a, c ∈ I \ J and b, d ∈ J \ I, or vise versa. A
collection of subsets S ⊂ [n]
k is weakly separated if it is pairwise weakly separated.
This is a variation of Leclerc-Zelevinsky’s notion of weak separation [LZ98] given
by Scott [Sco05]. It appeared in their study of quasi-commuting quantum minors.
Definition 5.2. The face labelling of a reduced Grassmannian graph G is the
labelling of faces F of G by subsets IF ⊂ [n] given by the condition: For each
strand α that goes from bi to bj , we have j ∈ IF if and only if the face F lies to
the left of the strand α (with respect to the direction of the strand from bi to bj ).
Let us stay that two reduced plabic graphs are contraction-equivalent if they can
be transformed to each other by the moves of type (1, 4) and (3, 4) (contractionuncontraction moves) without using the move of type (2, 4) (square move).
Theorem 5.3. [OPS15] (1) Face labels of a reduced Grassmannian graph form a
weakly separated collection in [n]
k , where k = h(G) is the helicity of G.
is the collec(2) Every maximal by inclusion weakly separated collection in [n]
k
tion of face labels of a complete reduced plabic graph of type (k, n).
(3) This gives a bijection between maximal by inclusion weakly separated collec
tions in [n]
k , and contraction-equivalence classes of complete reduced plabic graphs
of type (k, n).
Remark 5.4. Weakly separated collections are related to the cluster algebra structure [FZ02a, FZ03, BFZ05, FZ07] on the Grassmannian studied by Scott [Sco06].
In general, the cluster algebra on Gr(k, n) has infinitely many clusters. (See [Sco06]
for a classification of finite cases.) There is, however, a nicely behaved finite set of
clusters, called the Plücker clusters, which are formed by subsets of the Plücker coordinates ∆I . According to [OPS15, Theorem 1.6], the Plücker clusters for Gr(k, n)
are exactly the sets {∆I }I∈S associated with maximal weakly-separated collections
S ⊂ [n]
k . They are in bijection with contraction-equivalence classes of type (k, n)
complete reduced plabic graphs, and are given by the k(n − k) + 1 face labels of
such graphs. Square moves of plabic graphs correspond to mutations of Plücker
clusters in the cluster algebra.
Theorem 5.3 implies an affirmative answer to the purity conjecture of Leclerc and
Zelevinsky [LZ98]. An independent solution of the purity conjecture was given by
Danilov, Karzanov, and Koshevoy [DKK10] in terms of generalized tilings. The relationship between the parametrization a positroid cell given by a plabic graph G (see
12
ALEXANDER POSTNIKOV
Section 7 below) and the Plücker cluster {∆}I∈S associated with the same graph
G induces a nontrivial transformation, called the twist map, which was explicitly
described by Muller and Speyer [MuS16]. Weakly separated collections appeared in
the study of arrangements of equal minors [FP16]. In [GP17] the notion of weakly
separated collections was extended in the general framework of oriented matroids
and zonotopal tilings.
6. Cyclically labelled Grassmannian
Let us reformulate the definition of the Grassmannian and its positive part in
a more invariant form, which makes its cyclic symmetry manifest. In the next
section, we will consider “little positive Grassmannians” associated with vertices v
of a Grassmannian graph G whose ground sets correspond to the edges adjacent to
v. There is no natural total ordering on such a set of edges, however there is the
natural cyclic (clockwise) ordering.
We say that a cyclic ordering of a finite set C is a choice of closed directed cycle
that visits each element of C exactly once. A total ordering of C is compatible with
a cyclic ordering if it corresponds to a directed path on C obtained by removing an
edge of the cycle. Clearly, there are |C| such total orderings.
Definition 6.1. Let C be a finite set of indices with a cyclic ordering of its elements,
and let k be an integer between 0 and |C|. The cyclically labelled Grassmannian
|C|
Gr(k, C) over R is defined as the subvariety of the projective space P( k )−1 with
projective Plücker coordinates (∆I ) labelled by unordered k-element subsets I ⊂ C
satisfying the Plücker relations written with respect to any total order “<” on C
compatible with the given cyclic ordering:
X
(−1)|{a∈A, a>i}|+|{b∈B, b<i}| ∆A\{i} ∆B∪{i} = 0,
i∈A\B
where A and B are any (k+1)-element and (k−1)-element subsets of C, respectively.
(More precisely, Gr(k, C) is the projective algebraic variety given by the radical of
the ideal generated by the above Plücker relations.)
The positive part Gr>0 (k, C) is the subset of Gr(k, C) where the Plücker coordinates can be simultaneously rescaled so that ∆I > 0, for all k-element subsets
I ⊂ C.
Remark 6.2. The Plücker relations (written as above) are invariant with respect
to cyclic shifts of the ordering “<”. Thus the definition of the cyclically labelled
Grassmannian Gr(k, C) is independent of a choice of the total order on C. For
example, for k = 2 and C = {1, 2, 3, 4}, Gr(2, C) is the subvariety of P6−1 given by
the Plücker relation:
∆{1,3} ∆{2,4} = ∆{1,2} ∆{3,4} + ∆{1,4} ∆{2,3} .
Observe the cyclic symmetry of this relation! The ordering of indices 2 < 3 < 4 < 1
gives exactly the same Gr(2, C) with the same positive part Gr>0 (2, C).
Remark 6.3. There is a subtle yet important difference between the cyclically labelled Grassmannian Gr(k, C) with the Plücker coorinates ∆I and the usual definition of the Grassmannian Gr(k, n), n = |C|, with the “usual Plücker coordinates”
defined as the minors D(i1 ,...,ik ) = det(Ai1 ,...,ik ) of submatrices Ai1 ,...,ik of a k × n
matrix A.
POSITIVE GRASSMANNIAN AND POLYHEDRAL SUBDIVISIONS
13
The D(i1 ,...,ik ) are labelled by ordered collections (i1 , . . . , ik ) of indices. They are
anti-symmetric with respect to permutations of the indices i1 , . . . , ik . On the other
hand, the ∆{i1 ,...,ik } are labelled by unordered subsets I = {i1 , . . . , ik }. So they are
symmetric with respect to permutations of the indices i1 , . . . , ik .
The “usual Plücker relations” for the D(i1 ,...,ik ) have the Sn -symmetry with
respect to all permutations of the ground set. On the other hand, the above Plücker
relations for the ∆{i1 ,...,ik } have only the Z/nZ-symmetry with respect to cyclic
shifts of the ground set.
Of course, if we fix a total order of the ground set, we can rearrange the indices
in D(i1 ,...,ik ) in the increasing order and identify D(i1 ,...,ik ) , for i1 < · · · < ik , with
∆{i1 ,...,ik } . This identifies the cyclically labelled Grassmannian Gr(k, C) with the
usual Grassmannian Gr(k, n). However, this isomorphism is not canonical because
it depends on a choice of the total ordering of the index set. For even k, the
isomorphism is not invariant under cyclic shifts of the index set.
7. Perfect orientation parametrization of positroid cells
Positroid cells were parametrized in [Pos06] in terms of boundary measurements
of perfect orientations of plabic graphs. Equivalent descriptions of this parametrization were given in terms of network flows by Talaska [Tal08] and in terms of perfect
matchings [PSW09, Lam15]. Another interpretation of this parametrization was
motivated by physics [ABCGPT16], where plabic graphs were viewed as on-shell
diagrams, whose vertices represent little Grassmannians Gr(1, 3) and Gr(2, 3) and
edges correspond to gluings, see also [Lam15, Section 14] for a more mathematical
description. Here we give a simple and invariant way to describe the parametrization in the general setting of Grassmannian graphs and their perfect orientations.
It easily specializes to all the other descriptions. Yet it clarifies the idea of gluings
of little Grassmannians.
Let G = (V, E) be a perfectly orientable Grassmannian graph with n boundary
vertices and helicity h(G) = k, and let Gint = (Vint , Eint ) be its internal subgraph.
Also let Ebnd = E \ Eint be the set of boundary edges of G.
Informally speaking, each internal vertex v ∈ Vint represents the “little Grassmannian” Gr(h, d), where d is the degree of vertex v and h is its helicity. We “glue”
these little Grassmannians along the internal edges e ∈ Eint of the graph G to form
a subvariety in the “big Grassmannian” Gr(k, n). Gluing along each edge kills one
parameter. Let us give a more rigorous description of this construction.
For an internal vertex v ∈ Vint , let E(v) ⊂ E be the set of all adjacent edges
to v (possibly including some boundary edges), which is cyclically ordered in the
clockwise order (as we go around v). Define the positive vertex-Grassmannian
Gr>0 (v) as the positive part of the cyclically labelled Grassmannian
Gr>0 (v) := Gr>0 (h(v), E(v)).
(v)
Let (∆J ) be the Plücker coordinates on Gr>0 (v), where J ranges over the set
E(v)
of all h(v)-element subsets in E(v).
h(v)
Let us define several positive tori (i.e., positive parts of complex tori). The
>0
:= (R>0 )Ebnd ≃ (R>0 )n . The internal positive
boundary positive torus is Tbnd
>0
>0
>0
>0
Eint
torus is Tint := (R>0 )
, and the total positive torus Ttot
:= Tbnd
× Tint
. The
boundary/internal/total positive torus is the group of R>0 -valued functions on
boundary/internal/all edges of G.
14
ALEXANDER POSTNIKOV
These tori act on the positive vertex-Grassmannians Gr>0 (v) by rescaling the
>0
Plücker coordinates. For (te )e∈E ∈ Ttot
,
!
Y
(v)
(v)
te ∆J ).
(te ) : (∆J ) 7−→ (
e∈J
>0
The boundary torus Tbnd
also acts of the “big Grassmannian” Gr(k, n) as usual
Q
>0
(t1 , . . . , tn ) : ∆I 7→ ( i∈I ti ) ∆I , for (t1 , . . . , tn ) ∈ Tbnd
.
Recall that, for a perfect orientation O of G, I(O) denotes the set of i ∈ [n] such
that the boundary edge adjacent to bi is directed towards the interior of G in O.
For an internal vertex v ∈ Vint , let J(v, O) ⊂ E(v) be the subset of edges adjacent
to v which are directed towards v in the orientation O.
We are now ready to describe the perfect orientation parametrization of the
positroid cells.
Theorem 7.1. Let G be a perfectly orientable Grassmannian graph. Let µG be the
map defined on the direct product of the positive vertex-Grassmannians Gr>0 (v)
and written in terms of the Plücker coordinates as
µG :
×
v∈V
[n]
Gr>0 (v) −→ P( k )−1
int
µG :
(v)
(∆J )J∈(
×
v∈V
int
E(v)
h(v)
,
) 7−→ (∆I )I∈([n]
k )
where ∆I is given by the sum over all perfect orientations O of the graph G such
that I(O) = I:
X Y
(v)
∆J(v,O) .
∆I =
I(O)=I v∈Vint
[n]
(1) The image of µG is exactly the positroid cell ΠM ⊂ Gr≥0 (k, n) ⊂ P( k )−1 ,
where M = M(G) is the positroid associated with G.
>0
>0
(2) The map µG is Tint
-invariant and Tbnd
-equivariant, that is, µG (t·x) = µG (x)
>0
>0
′
′
for t ∈ Tint , and µG (t · x) = t · µG (x) for t′ ∈ Tbnd
.
(3) The map µG induces the birational subtraction-free bijection µ̄G
!
µ̄G :
×
v∈V
>0
Gr>0 (v) /Tint
−→ ΠG
int
if and only if the Grassmannian graph G is a reduced.
Remark 7.2. The phrase “birational subtraction-free bijection” means that both
µ̄G and its inverse (µ̄G )−1 can be expressed in terms of the Plücker coordinates by
rational (or even polynomial) expressions written without using the “−” sign.
Proof. Part (2) is straightforward from the definitions. Let us first prove the
remaining claims in the case when G is a plabic graph. In fact, in this case
this construction gives exactly the boundary measurement parametrization of ΠG
from [Pos06, Section 5]. The Plücker coordinates for the boundary measurement
parametrization were given in [Pos06, Proposition 5.3] and expressed by Talaska
[Tal08, Theorem 1.1] in terms of network flows on the graph G. The construction
of the boundary measurement parametrization (and Talaska’s formula) depends on
a choice of a reference perfect orientation O0 . One observes that any other perfect
POSITIVE GRASSMANNIAN AND POLYHEDRAL SUBDIVISIONS
15
orientation O of the plabic graph G is obtained from O0 by reversing the edges along
a network flow, which gives a bijection between network flows and perfect orientations of G. This shows that the above expression for ∆I is equivalent to Talaska’s
formula, which proves the equivalence of the above perfect orientation parametrization and the boundary measurement parametrization from [Pos06]. Parts (1) and
(3) now follow from results of [Pos06].
For an arbitrary Grassmannian graph G′ , let G be a plabic graph that refines G′ .
We already know that each “little plabic graph” Gv , i.e., the subgraph of G that
refines a vertex v of G′ , parametrizes each positive vertex-Grassmannian Gr>0 (v)
by a birational subtraction-free bijection µ̄v := µ̄Gv . We also know the map µ̄G for
the plabic graph G parametrizes the cell ΠM(G) if G reduced, or maps surjectively
but not bijectively onto ΠM(G) if G is not reduced. Then the map µ̄G′ is given by
the composition µ̄G ◦ (×v∈Vint (µ̄v )−1 ) and the needed result follows.
The above construction can be thought of as a gluing of the “big Grassmannian”
out of “little Grassmannians.” This is similar to a tiling of a big geometric object
(polytope) by smaller pieces (smaller polytopes). As we will see, this construction
literally corresponds to certain subdivisions of polytopes.
8. Polyhedral subdivisions: Baues poset and fiber polytopes
In this section we discuss Billera-Sturmfels’ theory [BS92] of fiber polytopes, the
generalized Baues problem [BKS94], and flip-connectivity, see also [Rei99, RS00,
ARS99, Ath01, AS02] for more details.
8.1. The Baues poset of π-induced subdivisions. Let π : P → Q be an affine
projection from one convex polytope P to another convex polytope Q = π(P ).
Informally, a π-induced polyhedral subdivision is a collection of faces of the
polytope P that projects to a polyhedral subdivision of the polytope Q.
Here is a rigorous definition, see [BS92]. Let A be the multiset of projections
π(v) of vertices v of P . Each element π(v) of A is labelled by the vertex v. For
σ ⊂ A, let conv(σ) denotes the convex hull of σ. We say that σ ′ ⊂ σ is a face of σ
if σ ′ consists of all elements of σ that belong to a face of the polytope conv(σ).
A π-induced subdivision is a finite collection S of subsets σ ⊂ A, called cells,
such that
(1) Each σ ∈ S is the projection under π of the vertex set of a face of P .
(2) For each σ ∈ S, dim(conv(σ)) = dim(Q).
(3) For any σ1 , σ2 ∈ S, conv(σ1 ) ∩ conv(σ2 ) = conv(σ1 ∩ σ2 ).
(4) S
For any σ1 , σ2 ∈ S, σ1 ∩ σ2 is either empty or a face of both σ1 and σ2 .
(5) σ∈S conv(σ) = Q.
π
The Baues poset ω(P → Q) is the poset of all π-induced subdivisions partially
ordered by refinement, namely, S ≤ T means that, for every cell σ ∈ S, there exists
a cell τ ∈ T such that σ ⊂ τ . This poset has a unique maximal element 1̂, called
the trivial subdivision, that consists of a single cell σ = A. All other elements are
π
π
called proper subdivisions. Let ω̂(P → Q)) := ω(P → Q) − 1̂ be the poset of proper
π-induced subdivisions obtained by removing the maximal element 1̂. The minimal
elements of the Baues poset are called tight π-induced subdivisions.
Among all π-induced subdivisons, there is a subset of coherent subdivisions that
come from linear height functions h : P → R as follows. For each q ∈ Q, let
16
ALEXANDER POSTNIKOV
F̄q be the face of the fiber π −1 (q) ∩ P where the height function h reaches its
maximal value. The face F̄q lies in the relative interior of some face Fq of P . The
π
collection of faces {Fq }q∈Q projects to a π-induced subdivision of Q. Let ωcoh (P →
π
Q) ⊆ ω(P → Q) be the subposet of the Baues poset formed by the coherent πinduced subdivisions. This coherent part of the Baues poset is isomorphic to the
π
face lattice of the convex polytope Σ(P → Q), called the fiber polytope, defined as
the Minkowskii integral of fibers of π (the limit Minkowskii sums):
Z
π
Σ(P → Q) :=
(π −1 (q) ∩ P ) dq
q∈Q
π
In general, the whole Baues poset ω(P → Q) may not be polytopal.
8.2. The generalized Baues problem and flip-connectivity. For a finite poset
ω, the order complex ∆ω is the simplicial complex of all chains in ω. The “topology
of a poset ω” means the topology of the simplicial complex ∆ω. For example, if ω
is the face poset of a regular cell complex ∆, then ∆ω is the barycentric subdivision
of the cell complex ∆; and, in particular, ∆ω is homeomorphic to ∆.
π
Clearly, the subposet ω̂coh (P → Q) of proper coherent π-induced subdivisions
homotopy equivalent to a (dim(P )−dim(Q)−1)-sphere, because it is the face lattice
of a convex polytope of dimension dim(P ) − dim(Q), namely, the fiber polytope
π
Σ(P → Q).
The generalized Baues problem (GBP) posed by Billera, Kapranov, Sturmfels
[BKS94] asks whether the same is true about the poset of all proper π-induced
π
subdivisions. Is it true that ω̂(P → Q) is homotopy equivalent to a (dim(P ) −
dim(Q) − 1)-sphere? In general, the GBP is a hard question. Examples of Baues
posets with disconnected topology were constructed by Rambau and Ziegler [RZ96]
and more recently by Liu [Liu16]. There are, however, several general classes of
projections of polytopes, where the GBP has an affirmative answer, see the next
section.
Another related question is about connectivity by flips. For a projection π :
P → Q, the flip graph is the restriction of the Hasse diagram of the Baues poset
π
ω(P → Q) to elements of rank 0 (tight subdivisions) and rank 1 (subdivisions that
cover a tight subdivision). The elements of rank 1 in the flip graph are called
flips. The flip-connectivity problem asks whether the flip graph is connected. The
coherent part of the flip graph is obviously connected, because it is the 1-skeleton
π
of the fiber polytope Σ(P → Q).
The GBP and the flip-connectivity problem are related to each other, but, strictly
speaking, neither of them implies the other, see [Rei99, Section 3] for more details.
8.3. Triangulations and zonotopal tilings. There are two cases of the above
general setting that attracted a special attention in the literature.
The first case is when the polytope P is the (n − 1)-dimensional simplex ∆n−1 .
The multiset A of projections of vertices of the simplex can be an arbitrary multiset
of n points, and Q = conv(A) can be an arbitrary convex polytope. In this case,
π
the Baues poset ω(∆n−1 → Q) is the poset of all polyhedral subdivisions of Q (with
vertices at A); tight π-induced subdivisions are triangulations of Q; and the fiber
π
polytope Σ(∆n−1 → Q) is exactly is the secondary polytope of Gelfand-KapranovZelevinsky [GKZ94], which appeared in the study of discriminants.
POSITIVE GRASSMANNIAN AND POLYHEDRAL SUBDIVISIONS
17
In particular, for a projection of the simplex ∆n−1 to an n-gon Q, π-induced
subdivisions are exactly the subdivisions of the n-gon by noncrossing chords. All of
them are coherent. Tight subdivisions are triangulations of the n-gon. There are
2n−4
1
the Catalan number Cn−2 = n−1
n−2 of triangulations of the n-gon. The fiber
polytope (or the secondary polytope) in this case is the Stasheff associahedron.
Another special case is related to projections π : P → Q of the hypercube P =
n := [0, 1]n . The projections Q = π( n ) of the hypercube form a special class
of polytopes, called zonotopes. In this case, π-induced subdivisions are zonotopal
tilings of zonotopes Q. According to Bohne-Dress theorem [Boh92], zonotopal
tilings of Q are in bijection with 1-element extensions of the oriented matroid
associated with the zonotope Q.
For a projection of the n-hypercube n to a 1-dimensional line segment, the fiber
polytope is the permutohedron. For a projection π : n → Q of the n-hypercube
n to a 2n-gon Q, fine zonotopal tilings (i.e., tight π-induced subdivisions) are
known as rhomus tilings of the 2n-gon. They correspond to commutation classes
of reduced decompositions of the longest permutation w◦ ∈ Sn .
9. Cyclic polytopes and cyclic zonotopes
Fix two integers n and 0 ≤ d ≤ n − 1.
Definition 9.1. A cyclic projection is a linear map
π : Rn → Rd+1 ,
π : x 7→ M x
given by a (d + 1) × n matrix M = (u1 , . . . , un ) (the ui are the column vectors) with
all positive maximal (d + 1) × (d + 1) minors and such that f (u1 ) = · · · = f (un ) = 1
for some linear form f : Rd+1 → R. In other words, M represents a point of the
positive Grassmannian Gr>0 (d + 1, n) with columns ui rescaled so that they all lie
on the same affine hyperplane H1 = {y ∈ Rd+1 | f (y) = 1}.
The cyclic polytope is the image under a cyclic projection π of the standard
(n − 1)-dimensional simplex ∆n−1 := conv(e1 , . . . , en )
C(n, d) := π(∆n−1 ) ⊂ H1 .
The cyclic zonotope is the image of the standard n-hypercube n := [0, 1]n ⊂ Rn
Z(n, d + 1) := π( n ) ⊂ Rd+1 .
Remark that, for each n and d, there are many combinatorially (but not linearly)
isomorphic cyclic polytopes C(n, d) and cyclic zonotopes Z(n, d + 1) that depend
on a choice of the cyclic projection π. Clearly, C(n, d) = Z(n, d + 1) ∩ H1 .
Ziegler [Zie93] identified fine zonotopal tilings of the cyclic zonotope Z(n, d + 1),
π
i.e., the minimal elements of the Baues poset ω( n → Z(n, d+1)), with elements of
Manin-Shekhtman’s higher Bruhat order [MaS86], also studied by Voevodsky and
Kapranov [VK91]. According to results of Sturmfels and Ziegler [SZ93], Ziegler
[Zie93], Rambau [Ram97], and Rambau and Santos [RS00], the GBP and flipconnectivity have affirmative answers in these cases.
Theorem 9.2. (1) [SZ93] For π : n → Z(n, d + 1), the poset of proper zonotopal
tilings of the cyclic zonotope Z(n, d + 1) is homotopy equivalent to an (n − d − 2)dimensional sphere. The set of fine zonotopal tilings of Z(n, d + 1) is connected by
flips.
18
ALEXANDER POSTNIKOV
(2) [RS00] For π : ∆n−1 → C(n, d), the poset of proper subdivisions of the
cyclic polytope C(n, d) is homotopy equivalent to an (n − d − 2)-dimensional sphere.
[Ram97] The set of triangulations of the cyclic polytope C(n, d) is connected by
flips.
10. Cyclic projections of the hypersimplex
Fixn three integers
0 ≤ k ≤ n and 0 ≤ d ≤ n − 1. The hypersimplex ∆kn :=
o
[n]
is the k-th section of the n-hypercube n ⊂ Rn
conv eI | I ∈ k
∆kn = n ∩ {x1 + · · · + xn = k}.
n
Let π : R → R
d+1
be a cyclic projection as above. Define the polytope
Q(k, n, d) := π(∆kn ) = Z(n, d + 1) ∩ Hk ,
where Hk is the affine hyperplane Hk := {y ∈ Rd+1 | f (y) = k}. Clearly, for k = 1,
the polytope Q(1, n, d) is the cyclic polytope C(n, d).
Let ω(k, n, d) be the Baues poset of π-induced subdivisions for a cyclic projection
π : ∆kn → Q(k, n, d):
π
ω(k, n, d) := ω(∆kn → Q(k, n, d)).
π
π
Let ωcoh
(k, n, d) := ωcoh (∆kn → Q(k, n, d)) ⊆ ω(k, n, d) be its coherent part. Note
π
that the coherent part ωcoh
(k, n, d) depends on a choice of the cyclic projection
π, but the whole poset ω(k, n, d) is independent of any choices. The coherent
π
part ωcoh
(k, n, d) may not be equal ω(k, n, d). For example they are not equal for
(k, n, d) = (3, 6, 2).
The poset ω(k, n, d) is a generalization of the Baues poset of subdivisions of the
cyclic polytope C(n, d), and is related to the Baues poset of zonotopal tilings of
the cyclic zonotope Z(n, d + 1) in an obvious manner. For k = 1, ω(1, n, d) =
π
ω(∆n−1 → C(n, d)). For any k, there is the order preserving k-th section map
π
Sectionk : ω( n → Z(n, d + 1)) → ω(k, n, d)
that send a zonotopal tiling of Z(n, d + 1) to its section by the hyperplane Hk .
Let ωlift (k, n, d) ⊆ ω(k, n, d) be the image of the map Sectionk . We call the
elements of ωlift (k, n, d) the lifting π-induced subdivisions. They form the subset
of π-induced subdivisions from ω(k, n, d) that can be lifted to a zonotopal tiling of
the cyclic zonotope Z(n, d + 1). Clearly, we have
π
ωcoh
(k, n, d) ⊆ ωlift (k, n, d) ⊆ ω(k, n, d).
The equality of the sets of minimal elements of ω(k, n, d) and ωlift (k, n, d) was
proved in the case k = 1 by Rambau and Santos [RS00], who showed that all
triangulations of the cyclic polytope C(n, d) are lifting triangulations. For d = 2,
the equality follows from the result of Galashin [Gal16] (Theorem 11.7 below) about
plabic graphs, as we will explain in the next section.
Theorem 10.1. The minimal elements of the posets ωlift (k, n, d) and ω(k, n, d) are
the same in the following cases: (1) k = 1 and any n, d; (2) d = 2 and any k, n.
Flip-connectivity [SZ93, Zie93] of zonotopal tilings of Z(n, d + 1) easily implies
the following claim.
Lemma 10.2. The minimal elements of ωlift (k, n, d) are connected by flips.
POSITIVE GRASSMANNIAN AND POLYHEDRAL SUBDIVISIONS
19
Indeed, for any pair of fine zonotopal tilings T and T ′ of Z(n, d+1) connected by
a flip, their k-sections Sectionk (T ) and Sectionk (T ′ ) are either equal to each other
or connected by a flip.
The Baues posets of the form ω(k, n, d) are good candidates for a general class of
projections of polytopes where the GBP and flip-connectivity problem might have
affirmative answers.
Problem 10.3. Is the poset ω(k, n, d) − 1̂ homotopy equivalent of a sphere? Can
its minimal elements be connected by flips? Is it true that ωlift (k, n, d) = ω(k, n, d)?
Example 10.4. For d = 1, the Baues poset ω(k, n, 1) is already interesting. Its
minimal elements correspond to monotone paths on the hypersimplex ∆kn , which
are increasing paths that go along the edges of the hypersimplex ∆kn . Such paths
are the subject of the original (non-generalized) Baues problem [Bau80], which was
proved by Billera, Kapranov, Sturmfels [BKS94] (for any 1-dimensional projection
of a polytope). More specifically, monotone paths on ∆kn correspond to directed
with edges I → J
paths from [1, k] to [n − k + 1, n] in the directed graph on [n]
k
if J = (I \ {i}) ∪ {j} for i < j.
It is not hard to see that, for k = 1, n − 1, there are 2n−2 monotone paths, and
the posets ω(1, n, 1) and ω(n − 1, n, 1) are isomporphic to the Boolean lattice Bn−2 ,
i.e., the face poset of the hypercube n−2 . For n = 2, 3, 4, 5, the Baues poset
ω(2, n, 1) has 1, 2, 10, 62 minimal elements.
Monotone paths on ∆kn might have different lengths. The longest monotone
paths are in an easy bijection with standard Young tableaux of the rectangular shape
Qn−k−1 i!
k × (n − k). By the hook-length formula, their number is (k(n − k))! i=0
(k+i)! .
Note, however, ω(k, n, d) 6= ωlift (k, n, d) for (k, n, d) = (2, 5, 1). Indeed, Galashin
pointed out that the monotone path {1, 2} → {1, 3} → {1, 4} → {2, 4} → {3, 4}
cannot be lifted to a rhombus tiling of the the 2n-gon Z(n, 2), because it is not
weakly separated.
11. Grassmannian graphs as duals of polyhedral subdivisions induced
by projections of hypersimplices
Let us now discuss the connection between the positive Grassmannian and combinatorics of polyhedral subdivisions. In fact, the positive Grassmannian is directly
related to the setup of the previous section for d = 2.
Theorem 11.1. The poset of complete reduced Grassmannian graphs of type (k, n)
ordered by refinement is canonically isomorphic to the Baues poset ω(k, n, 2) of
π-induced subdivisions for a 2-dimensional cyclic projection π of the hypersimplex
∆kn . Under this isomorphism, plabic graphs correspond to tight π-induced subdivisions and moves of plabic graphs correspond to flips between tight π-induced
subdivisions.
Theorem 4.13(2) ([Pos06, Theorem 13.4]) immediately implies flip-connectivity.
Corollary 11.2. The minimal elements of Baues poset ω(k, n, 2) are connected by
flips.
Example 11.3. The Baues poset ω(1, n, 2) is the poset of subdivisions of an n-gon
by non-crossing chords, i.e., it is the Stasheff’s associahedron. Its minimal elements
2n−4
1
correspond to the Catalan number n−1
n−2 triangulations of the n-gon.
20
ALEXANDER POSTNIKOV
We can think of the Baues posets ω(k, n, 2) as some kind of “generalized associahedra.” In general, they are not polytopal. But they share some nice features
with the associahedron. It is well-known that every face of the associahedron is a
direct product of smaller associahedra. The same is true for all ω(k, n, 2).
Proposition 11.4. For any element S in ω(k, n, 2), the lower order interval {S ′ |
S ′ ≤ S} in the Baues poset ω(k, n, 2) is a direct product of Baues posets of the same
form ω(k ′ , n′ , 2).
Proof. This is easy to see in terms of complete reduced Grassmannian graphs G.
Indeed, for any G, all refinements of G′ are obtained by refining all vertices of G
independently from each other.
This property is related to the fact that every face of the hypersimplex ∆kn is a
smaller hypersimplex, as we discuss below.
Remark 11.5. Among all reduced Grassmannian/plabic graphs, there is a subset
of coherent (or regular) graphs, namely the ones that correspond to the coherent
π-induced subdivisions from ωcoh (k, n, 2). Each of these graphs can be explicitly
constructed in terms of a height function. This subclass depends on a choice of
the cyclic projection π. Regular plabic graphs are related to the study of soliton
solutions of Kadomtsev-Petviashvili (KP) equation, see [KoW11, KoW14]. We will
investigate the class of regular plabic graphs in [GPW].
Let us now give more details on the correspondence between Grassmannian
graphs and subdivisions. A cyclic projection π : ∆kn → Q(k, n, 2) is the linear
map given by a 3 × n matrix M = (u1 , . . . , un ) such that [u1 , . . . , un ] ∈ Gr>0 (3, n)
and u1 , . . . , un all lie on the same affine plane H1 ⊂ R3 . Without loss of generality,
assume that H1 = {(x, y, z) | z = 1}. The positivity condition means that the
points π(u1 ), . . . , π(un ) form a convex n-gon with vertices arranged in the counterclockwise order.
The polytope Q := Q(k, n, 2) = π(∆kn ) is the convex n-gon in the affine plane
Hk = {(x, y, k)} ⊂ R3 with the vertices π(e[1,k] ), π(e[2,k+1] ), . . . , π(e[n,k−1] ) (in the
counterclockwise order) corresponding to all consecutive cyclic intervals of size k in
[n].
Notice that each face γ of the hypersimplex ∆kn is itself a smaller hypersimplex
of the form
γI0 ,I1 := {(x1 , . . . , xn ) ∈ ∆kn | xi = 0 for i ∈ I0 , xj = 1 for j ∈ I1 }
where I0 and I1 are disjoint subsets of [n]. So γ ≃ ∆hm , where h = k − |I0 | and
m = n − |I0 | − |I1 |. The projection π maps the face γ to the m-gon π(γ) that carries
an additional parameter h.
Thus the π-induced subdivisions S are in bijective correspondence with the tilings
of the n-gon Q by smaller convex polygons such that:
(1) Each vertex has the form π(eI ) for I ∈ [n]
k .
(2) Each edge has the form [π(eI ), π(eJ )] for two k-element subsets I and J
such that |I ∩ J| = k − 1.
(3) Each face is an m-gon of the form π(γI0 ,I1 ), as above.
Let S ∗ be the planar dual of such a tiling S. The graph S ∗ has exactly n
boundary vertices bi corresponding to the sides [π(e[i,i+k−1] ), π(e[i+1,i+k]) ] of the
n-gon Q. The internal vertices v of S ∗ (corresponding to faces γ of S) are equipped
POSITIVE GRASSMANNIAN AND POLYHEDRAL SUBDIVISIONS
21
with the parameter h = h(v) ∈ {0, . . . , deg(v)}. Thus S ∗ has the structure of a
Grassmannian graph. Moreover, each face F of S ∗ (corresponding a vertex π(eI ) of
S) is labelled by a subset I ∈ [n]
k . We can now make the previous theorem more
precise.
Theorem 11.6. The map S 7→ S ∗ is an isomporphism between the Baues poset
ω(k, n, 2) and complete reduced Grassmannian graphs G of type (k, n). For each
face F of G = S ∗ corresponding to a vertex π(eI ) of S, the subset I ⊂ [n]
is
k
exactly the face label IF (see Definition 5.2).
Proof. Let us first show that tight π-induced subdivisions S are in bijection with
complete reduced plabic graphs of type (k, n). That means that, in addition to
the conditions (1), (2), (3) above, we require the tiling S of the n-gon G has all
triangular faces. So S is a triangulation of the n-gon Q of a special kind, which we
call a plabic triangulation.
Such plabic triangulations of the n-gon are closely related to plabic tilings from
[OPS15]. The only difference between plabic triangulations and plabic tilings is
that the latter correspond not to (3-valent) plabic graphs (as defined in the current
paper) but to bipartite plabic graphs. A bipartite plabic graph G is exactly a
Grassmannian graph such that each internal vertex either has type (1, d) (white
vertex) or type (d − 1, d) (black vertex), and every edge of G connects vertices of
different colors. Each reduced 3-valent plabic graph G′ can be easily converted into
a bipartite plabic graph G by contrating edges connecting vertices of the same color.
It was shown in [OPS15, Theorem 9.12] that the planar dual graph of any reduced
bipartite plabic graph G can be embedded inside an n-gon as a plabic tiling with
black and white regions and all vertices of the form π(eI ). If we now subdivide the
black and white regions of such plabic tiling by chords into triangles, we can get
back the plabic triangulation associated with a (3-valent) plabic graph G′ . This
shows that any complete reduced (3-valent) plabic graph is indeed the planar dual
of a tight π-induced subdivision.
On the other hand, for each plabic triangulation S we can construct the plabic
graph by taking its planar dual G = S ∗ as described above. It is easy to check from
the definitions that the decorated strand permutation w of G is exactly w(i) = i + k
(mod n). It remains to show that this plabic graph G is reduced. Suppose that
G is not reduced. Then by Theorem 4.13(5), after possibly applying a sequence of
moves (1, 4), (2, 4), and/or (3, 4), we get a plabic graph with a pair of parallel edges
or with a loop-edge. It is straightforward to check that applying the moves (1, 4),
(2, 4), (3, 4) corresponds to local transformations of the plabic triangulation S, and
transforms it into another plabic triangulations S ′ . However, it is clear that if a
plabic graph G contains parallel edges of a loop-edge, then the dual graph is not a
plabic triangulation. So we get a contradiction, which proves the result for plabic
graphs and tight subdivisions.
Now let G′ be any complete reduced Grassmannian graph of type (k, n), and let
G be its plabic refinement. We showed that we can embed the planar dual graph
G∗ as a plabic triangulation S into the n-gon. The union of triangles in S that
correspond to a single vertex v of G′ covers a region inside Q. We already know
that this region is a convex m-gon (because we already proved the correspondence
for plabic graphs). Thus, for each vertex of G′ , we get a convex polygon in Q
and all these polygons form π-induced subdivision. So we proved that the planar
22
ALEXANDER POSTNIKOV
dual of G′ can be embedded as a polyhedral subdivision of Q. The inverse map is
S ′ 7→ G′ = (S ′ )∗ .
Let us mention a related result of Galashin [Gal16].
Theorem 11.7. [Gal16] Complete reduced plabic graphs of type (k, n) are exactly
the dual graphs of sections of fine zonotopal tilings of the 3-dimensional cyclic zonotope Z(n, 3) by the hyperplane Hk .
In view of the discussion above, this result means that any tight π-induced
subdivision in ω(k, n, 2) can be lifted to a fine zonotopal tiling of the cyclic zonotope
Z(n, 3). In other words, the posets ω(k, n, 2) and ωlift (k, n, 2) have the same sets
of minimal elements. A natural question to ask: Is the same true for all (not
necessarily minimal) elements of ω(k, n, 2)?
12. Membranes and discrete Plateau’s problem
Membranes from the joint project with Lam [LP] provide another related interpretation of plabic graphs. Let Φ = {ei − ej | i 6= j} ∈ Rn , where e1 , . . . , en are the
standard coordinate vectors.
Definition 12.1. [LP] A loop L is a closed piecewise-linear curve in Rn formed by
line segments [a, b] such that a, b ∈ Zn and a − b ∈ Φ.
A membrane M with boundary loop L is an embedding of a 2-dimensional
disk into Rn such that L is the boundary of M , and M is made out of triangles
conv(a, b, c), where a, b, c ∈ Z and a − b, b − c, a − c ∈ Φ.
A minimal membrane M is a membrane that has minimal possible area (the
number of triangles) among all membranes with the same boundary loop L.
The problem about finding a minimal membrane M with a given boundary loop
L is a discrete version of Plateau’s problem about minimal surface. Informally
speaking, membranes correspond to (the duals of) plabic graphs, and minimal
membranes correspond to reduced plabic graphs. Here is a more careful claim.
Theorem 12.2. [LP] Let w ∈ Sn be a permutation without fixed points with helicity
h(w) = k. Let Lw be the closed loop inside the hypersimplex ∆kn formed by the line
segments [a1 , a2 ], [a2 , a3 ], . . . , [an−1 , an ], [an , a1 ] such that ai+1 − ai = ew(i) − ei , for
i = 1, . . . , n, with indices taken modulo n.
Then minimal membranes M with boundary loop Lw are in bijection with reduced
plabic graphs G with strand permutation w. Explicitly, the correspondence is given
as follows. Faces F of G with face labels I = IF correspond to vertices eI of the
membrane M . Vertices of G with 3 adjacent faces labeled by I1 , I2 , I3 correspond
to triangles conv(eI1 , eI2 , eI3 ) in M .
Moves of plabic graphs correspond to local area-preserving transformations of
membranes. Any two minimal membranes with the same boundary loop Lw can be
obtained from each other by these local transformations.
13. Higher positive Grassmannians and Amplituhedra
The relation between the positive Grassmannian Gr>0 (k, n) and the Baues poset
ω(k, n, 2) raises a natural question: What is the geometric counterpart of the Baues
poset ω(k, n, d) for any d? These “higher positive Grassmannians” should generalize
Gr> (k, n) in the same sense as Manin-Shekhtman’s higher Bruhat orders generalize
POSITIVE GRASSMANNIAN AND POLYHEDRAL SUBDIVISIONS
23
the weak Bruhat order. The first guess is that they might be related to amplituhedra.
Arkani-Hamed and Trnka [AT14], motivated by the study of scattering amplitudes in N = 4 supersymmetric Yang-Mills (SYM) theory, defined the amplituhedron An,k,m = An,k,m (Z) as the image of the nonnegative Grassmannian Gr≥ (k, n)
under the “linear projection”
Z̃ : Gr≥0 (k, n) → Gr(k, k + m),
[A] 7→ [A Z T ]
induced by a totally positive (k + m) × n matrix Z, for 0 ≤ m ≤ n − k. The case
m = 4 is of importance for physics.
In general, the amplituhedron An,k,m has quite mysterious geometric and combinatorial structure. Here are a few special cases where its structure was understood
better. For m = n − k, An,k,n−k is isomorphic to the nonnegative Grassmannian
Gr≥0 (k, n). For k = 1, An,1,m is (the projectivization of) the cyclic polytope
C(n, m). For m = 1, Karp and Williams [KaW16] showed that the structure of the
amplituhedron An,k,1 is equivalent to the complex of bounded regions of the cyclic
hyperplane arrangement. In general, the relationship between the amplituhedron
An,k,m and polyhedral subdivisions is yet to be clarified.
References
[ABCGPT16] N. Arkani-Hamed, J. Bourjaily, F. Cachazo, A. Goncharov, A. Postnikov,
J. Trnka. Grassmannian Geometry of Scattering Amplitudes. Cambridge University
Press, 2016.
[ARS99]
C. A. Athanasiadis, J. Rambau, F. Santos. The generalized Baues problem for
cyclic polytopes II. In “Geometric combinatorics (Kotor, 1998).”. Publ. Inst. Math.
(Beograd) (N.S.) 66(80) (1999), 3–15.
[AS02]
C. A. Athanasiadis, F. Santos. On the topology of the Baues poset of polyhedral
subdivisions. Topology 41 (2002), 423–433.
[Ath01]
C. A. Athanasiadis. Zonotopal subdivisions of cyclic zonotopes. Geometriae Dedicata
86 (2001), no. 1-3, 37–57.
[AT14]
N. Arkani-Hamed, J. Trnka. The amplituhedron. J. High Energy Phys., (10):33,
2014.
[Bau80]
H. J. Baues. Geometry of loop spaces, the cobar construction. Mem. Amer. Math.
Soc. 25 (1980) 1–171.
[BFZ05]
A. Berenstein, S. Fomin, A. Zelevinsky. Cluster algebras III. Upper bounds and
double Bruhat cells. Duke Math. J. 126 (2005), no. 1, 1–52.
[Boh92]
J. Bohne. Eine kombinatorische analyse zonotopaler raumaufteilungen. Univ., Diss.Bielefeld, 1992.
[BS92]
L.J. Billera, B. Sturmfels. Fiber polytopes. Ann. Math. 135 (1992) 527–549.
[BKS94]
L.J. Billera, M.M. Kapranov, B. Sturmfels. Cellular strings on polytopes. Proc.
Amer. Math. Soc. 122 (1994) 549–555.
[CK09]
S. Chakravarty, Y. Kodama. Soliton solutions of the KP equation and application
to shallow water waves. Studies in Applied Mathematics 123 (2009), 83–151.
[CW07]
S. Corteel, L. Williams. Tableaux combinatorics for the asymmetric exclusion process. Advances in Applied Math 39 (2007), no. 3, 293–310.
[DKK10] V. I. Danilov, A. V. Karzanov, G. A. Koshevoy. On maximal weakly separated
set-systems. J. of Algebraic Combinatorics 32 (2010), no. 4, 497–531.
[FP16]
M. Farber, A. Postnikov, Arrangements of equal minors in the positive Grassmannian, Adv. Math. 300 (2016), 788–834.
[FZ99]
S. Fomin, A. Zelevinsky. Double Bruhat cells and total positivity. Journal of the
American Mathematical Society 12 (1999), no. 2, 335–380.
[FZ02a]
S. Fomin, A. Zelevinsky. Cluster algebras. I. Foundations. J. Amer. Math. Soc. 15
(2002), no. 2, 497–529.
24
[FZ02b]
ALEXANDER POSTNIKOV
S. Fomin, A. Zelevinsky. The Laurent phenomenon. Adv. in Appl. Math. 28 (2002),
no. 2, 119–144.
[FZ03]
S. Fomin, A. Zelevinsky. Cluster algebras. II. Finite type classification. Invent. Math.
154 (2003), no. 1, 63–121.
[FZ07]
S. Fomin, A. Zelevinsky. Cluster algebras. IV. Coefficients. Compositio Mathematica
143 (2007), no. 1, 112-164.
[Gal16]
P. Galashin. Plabic graphs and zonotopal tilings. arXiv:1611.00492, 2016.
[GGMS87] I. M. Gelfand, R. M. Goresky, R. D. MacPherson, V. V. Serganova. Combinatorial geometries, convex polyhedra, and Schubert cells. Advances in Math. 63 (1987),
301–316.
[GK35]
F. R. Gantmacher, M. G. Krein. Sur les matrices oscillatoires. C. R. Acad. Sci.
Paris 201 (1935), 577–579.
[GKL17]
P. Galashin, S. N. Karp, T. Lam. The totally nonnegative Grassmannian is a ball.
arXiv:1707.02010, 2017.
[GKZ94]
I. M. Gelfand, M. M. Kapranov, A. V. Zelevinsky. Discriminants, Resultants, and
Multidimensional Determinants. Mathematics: Theory & Applications, Birkhäuser,
Boston, 1994.
[GP17]
P. Galashin, A. Postnikov. Purity and separation for oriented matroids.
arXiv:1708.01329, 2017.
[GPW]
P. Galashin, A. Postnikov, L. Williams. Regular plabic graphs. In preparation.
[KLS13]
A. Knutson, T. Lam, D. Speyer. Positroid varieties: juggling and geometry. Compositio Mathematica 149 (2013), no. 10, 1710–1752.
[KaW16] S. N. Karp, L. Williams. The m = 1 amplituhedron and cyclic hyperplane arrangements. arXiv:1608.08288, 2016.
[KoW11] Y. Kodama, L. Williams. KP solitons, total positivity, and cluster algebras. Proceedings of the National Academy of Sciences, (2011) 108 (22), 8984–8989.
[KoW14] Y. Kodama, L. Williams. KP solitons and total positivity on the Grassmannian.
Inventiones mathematicae 198 (2014), no. 3, 637–699.
[Liu16]
G. Liu. A counterexample to the extension space conjecture for realizable oriented
matroids. arXiv:1606.05033, 2016.
[Lam15]
T. Lam. Totally nonnegative Grassmannian and Grassmann polytopes,
arXiv:1506.00603, 2015.
[LP]
T. Lam, A. Postnikov. Polypositroids I. In preparation.
[Lus90]
G. Lusztig. Canonical bases arising from quantized enveloping algebras. J. Amer.
Math. Soc. 3 (1990), no. 2, 447–498.
[Lus92]
G. Lusztig. Canonical bases in tensor products. Proc. Nat. Acad. Sci. U.S.A. 89
(1992), no. 17, 8177–8179.
[Lus93]
G. Lusztig, Introduction to quantum groups, Progress in Mathematics, vol. 110,
Birkhäuser, Boston, 1993.
[Lus94]
G. Lusztig. Total positivity in reductive groups. Lie Theory and Geometry: In Honor
of B. Kostant, Progress in Mathematics, vol. 123, Birkhäuser, Boston, 1994, pp. 531–
568.
[Lus98a]
G. Lusztig. Total positivity in partial flag manifolds. Representation Theory, 2
(1998), 70–78.
[Lus98b]
G. Lusztig. Introduction to total positivity. Positivity in Lie theory: open problems,
133–145, de Gruyter Expositions in Mathematics, 26, de Gruyter, Berlin, 1998.
[LZ98]
B. Leclerc, A. Zelevinsky. Quasicommuting families of quantum Plücker coordinates. American Mathematical Society Translations, Ser. 2 181, 1998.
[Mnë88]
N. E. Mnëv. The universality theorems on the classification problem of configuration
varieties and convex polytope varieties. pp. 527–543, in “Topology and Geometry:
Rohlin Seminar,” O. Ya. Viro ed., Lecture Notes in Mathematics 1346, SpringerVerlag, Berlin, 1988.
[MaS86]
Yu. I. Manin, V. V. Shekhtman. Higher Bruhat orderings connected with the symmetric group. Funktsional. Anal. i Prilozhen., 20(2):74–75, 1986.
[MuS16]
G. Muller, D. E. Speyer, The twist for positroid varieties, arXiv:1606.08383, 2016.
[Oh11]
S. Oh. Positroids and Schubert matroids. J. Combin. Theory Ser. A 118 (2011), no. 8,
2426–2435.
POSITIVE GRASSMANNIAN AND POLYHEDRAL SUBDIVISIONS
[OPS15]
[Pos06]
[PSW09]
[Ram97]
[Rei99]
[Rie98]
[Rie99]
[RS00]
[RW10]
[RZ96]
[Sco05]
[Sco06]
[Sch30]
[SZ93]
[Tal08]
[VK91]
[Zie93]
25
S. Oh, A. Postnikov, D. E. Speyer. Weak separation and plabic graphs. Proc. Lond.
Math. Soc. (3) 110 (2015), no. 3, 721–754.
A. Postnikov. Total positivity, Grassmannians, and networks. arXiv:math/0609764,
2006.
A. Postnikov, D. Speyer, L. Williams. Matching polytopes, toric geometry, and
the non-negative part of the Grassmannian. Journal of Algebraic Combinatorics 30
(2009), no. 2, 173–191.
J. Rambau. Triangulations of cyclic polytopes and higher Bruhat orders, Mathematika
44 (1997), no. 1, 162–194.
V. Reiner. The generalized Baues problem. In New perspectives in algebraic combinatorics (Berkeley, CA, 1996–97), 293–336, Math. Sci. Res. Inst. Publ., volume 38,
Cambridge Univ. Press, Cambridge, 1999.
K. Rietsch. Total Positivity and Real Flag Varieties, Ph.D. Thesis, M.I.T. 1998.
K. Rietsch. An algebraic cell decomposition of the nonnegative part of a flag variety,
Journal of Algebra 213 (1999), no. 1, 144-154.
J. Rambau, F. Santos. The generalized Baues problem for cyclic polytopes I. European Journal of Combinatorics 21 (2000), no. 1, Pages 65–83.
R. Rietsch, L. Williams. Discrete Morse theory for totally non-negative flag varieties.
Advances Math. 223 (2010), no. 6, 1855–1884.
J. Rambau, G. M. Ziegler. Projections of polytopes and the generalized Baues conjecture. Discrete Comput. Geom. 16 (1996) 215–237.
J. S. Scott. Quasi-commuting families of quantum minors. J. Algebra 290 (2005),
no. 1, 204-220.
J. S. Scott. Grassmannians and cluster algebras. Proc. London Math. Soc. (3) 92
(2006), no. 2, 345-380.
I. J. Schoenberg. Uber variationsvermindende lineare Transformationen. Math. Z.
32 (1930), 321–328.
B. Sturmfels, G. M. Ziegler, Extension spaces of oriented matroids. Discrete Comput. Geom. 10 (1993), 23–45.
K. Talaska. A formula for Plücker coordinates associated with a planar network. Int.
Math. Res. Not. IMRN 2008, 19 pp.
V. A. Voevodskii, M. M. Kapranov. The free n-category generated by a cube,
oriented matroids and higher Bruhat orders. Funktsional. Anal. i Prilozhen., 25(1):62–
65, 1991.
G. M. Ziegler. Higher Bruhat orders and cyclic hyperplane arrangements. Topology,
32(2):259–279, 1993.
Department of Mathematics, Massachusetts Institute of Technology, 77 Massachusetts
Avenue, Cambridge MA 02139
E-mail address: apost@math.mit.edu