The limit point of the pentagram map
Max Glick Ohio State University
November 10, 2018 Three features of interest in a system
A
B = T (A) Geometric description
y1 1 −1 , , 3 2 , 4 (y1, y2, y3, y4) y2 y (1 + y ) y 1+y2 4 3 4 3 µ2
1 2 1 2
Integrability Cluster algebra structure The pentagram map
A
B = T (A)
T = the pentagram map Outline
1. Pentagram map background 2. Cluster (Y -Pattern) structure for the pentagram map. 3. The search for the limit point
b b b P b Prehistory [ ..., Clebsch 1800’s, ..., Schwartz 1992]
Theorem If A is a closed pentagon then there is a projective transformation of the plane taking A to T (A). Theorem If A is a closed hexagon then there is a projective transformation of the plane taking A to T 2(A). A A
2 T (A) T (A) Modern introduction [Schwartz 1992]
Theorem If A is a convex n-gon then the sequence of polygons T k (A) converges exponentially to a point P =(X , Y ) ∈ R2 as k → ∞.
b b b P b Coordinate system [Schwartz 2008] ◮ Easier to work with twisted n-gons, e.g.
b b b
A =
b b b Coordinate system [Schwartz 2008] ◮ Easier to work with twisted n-gons, e.g.
b b b
A =
b b b
◮ Cross ratio (projectively natural) coordinates
A 7→ (x1,..., x2n). Coordinate system [Schwartz 2008] ◮ Easier to work with twisted n-gons, e.g.
b b b
A =
b b b
◮ Cross ratio (projectively natural) coordinates
A 7→ (x1,..., x2n). ◮ Evolution equation
1 − xj−3xj−2 xj−1 , j even ′ 1 − xj+1xj+2 xj = 1 − xj+3xj+2 xj+1 , j odd 1 − xj−1xj−2 Integrability of the pentagram map
The pentagram map is:
Twisted polygons Closed polygons Liouville integrable [OST 2010] [OST 2013] Algebro-geometric integrable [Soloviev 2013] [Soloviev 2013] “Cluster integrable” [GSTV 2012]
where ◮ OST = Ovsienko-Schwartz-Tabachnikov ◮ GSTV = Gekhtman-Shapiro-Tabachnikov-Vainshtein Other pentagram map work
1. Connection to cluster algebras (Part 2 of this talk) 2. Continuous limit n → ∞ (Boussinesq equation) ◮ Ovsienko-Schwartz-Tabachnikov 2010 3. Incidence theorems ◮ Schwartz 2001, 2008 ◮ G. 2015 ◮ Yao 2014 (preprint) 4. Generalizations ◮ Gekhtman-Shapiro-Tabachnikov-Vainshtein 2012 ◮ Khesin-Soloviev 2013, 2015, 2016 ◮ Mari Beffa 2013, 2015 ◮ Schwartz 2013 ◮ G.-Pylyavskyy 2016 The y-parameters
The y-parameters of a twisted polygon A are the cross ratios
−1 y2k (A)= −(χ(Ak−1, B, C, Ak+1))
y2k+1(A)= −χ(D, Ak , Ak+1, E)
for B, C, D, E as below.
A +3 b k A +2 b k A +2 bb k bb b b Ak+1 E b C Ak+1 b B b b Ak Ak b bb
Ak−2 Ak−1 bb b b Ak−2 Ak−1 D Cluster structure for the pentagram map [G. 2011]
Theorem The y-parameters of a twisted n-gon A evolve under T as
(1 + yj−1)(1 + yj+1) yj − − , j even ′ (1 + y 1 )(1 + y 1 ) yj = j−3 j+3 −1 odd yj , j
which agrees with the Y-pattern mutation dynamics of the quiver Qn.
To be defined: ◮ Y -pattern mutations ◮ the quiver Qn Y -patterns [Fomin, Zelevinsky 2007]
1 1 1 −1 , , 3 2 , 4 (y1, y2, y3, y4) (y y2 y (1 + y ) y ) 1+y2
4 3 4 3 µ2
1 2 1 2
Y -patterns are a family of discrete dynamical systems encoded by directed graphs (quivers) satisfying many nice properties: ◮ subtraction free rational expressions ◮ a natural Poisson structure [Gekhtman, Shapiro, Vainshtein] ◮ frequent cancellation of factors under iteration ◮ (sometimes) combinatorial formulas for iterates Y -seeds and mutations
A Y -seed is a pair (y, Q) where y =(y1,..., ym) is a collection of rational functions and Q is a quiver, i.e. a directed graph on vertex set {1, 2,..., m} without oriented 2-cycles.
Given a Y -seed (y, Q) and some k ∈ {1,..., m}, the mutation ′ ′ µk (y, Q)=(y , Q ), where Y -seeds and mutations
A Y -seed is a pair (y, Q) where y =(y1,..., ym) is a collection of rational functions and Q is a quiver, i.e. a directed graph on vertex set {1, 2,..., m} without oriented 2-cycles.
Given a Y -seed (y, Q) and some k ∈ {1,..., m}, the mutation ′ ′ µk (y, Q)=(y , Q ), where ◮ The vector y′ is obtained from y via the following steps:
1. For each j → k in Q, multiply yj by 1+ yk . 1 2. For each k → j in Q, multiply yj by 1+ −1 . yk 3. Invert yk . Y -seeds and mutations
A Y -seed is a pair (y, Q) where y =(y1,..., ym) is a collection of rational functions and Q is a quiver, i.e. a directed graph on vertex set {1, 2,..., m} without oriented 2-cycles.
Given a Y -seed (y, Q) and some k ∈ {1,..., m}, the mutation ′ ′ µk (y, Q)=(y , Q ), where ◮ The vector y′ is obtained from y via the following steps:
1. For each j → k in Q, multiply yj by 1+ yk . 1 2. For each k → j in Q, multiply yj by 1+ −1 . yk 3. Invert yk . ◮ The quiver Q′ is obtained from Q via the following steps: 1. For every length 2 path i → k → j, add an arc from i to j. 2. Reverse the orientation of all arcs incident to k. 3. Remove all oriented 2-cycles. An example of a Y -seed mutation
1 1 1 −1 , , 3 2 , 4 (y1, y2, y3, y4) (y y2 y (1 + y ) y ) 1+y2
4 3 4 3 µ2
1 2 1 2 The pentagram quiver (n = 8)
4
6 2
5 3
7 1 8 16 9 15
11 13
10 14
12 Application: combinatorial formulas [G. 2011]
k k Fj−1,k Fj+1,k yj (T (A)) = yj+3i Fj−3,k Fj+3,k i=−k ! Y where
Fj,k = “Order ideal generating functions of Aztec diamond posets” Application: combinatorial formulas [G. 2011]
k k Fj−1,k Fj+1,k yj (T (A)) = yj+3i Fj−3,k Fj+3,k i=−k ! Y where
Fj,k = “Order ideal generating functions of Aztec diamond posets”
Example
y2 y4
y3
y0 y6
F3,2 =1+ y0 + y6 + y0y6 + y0y3y6(1 + y2 + y4 + y2y4) The limit point [Schwartz 1992]
Theorem If A is a convex n-gon then the sequence of polygons T k (A) converges exponentially to a point P =(X , Y ) ∈ R2 as k → ∞.
b b b P b
Question Are X and Y analytic functions of the coordinates x1, y1,..., xn, yn of the vertices of A? Example of limit point 3 A 2
1
1 2 3 Can estimate (X , Y ) ≈ (1.60947, 1.83760) Example of limit point 3 A 2
1
1 2 3 Can estimate (X , Y ) ≈ (1.60947, 1.83760) 482 − 49t 614 − 51t (X , Y )= , t2 − 27t + 178 t2 − 27t + 178 where t ≈−12.8777 Example of limit point 3 A 2
1
1 2 3 Can estimate (X , Y ) ≈ (1.60947, 1.83760) 482 − 49t 614 − 51t (X , Y )= , t2 − 27t + 178 t2 − 27t + 178 where t ≈−12.8777 t = largest magnitude root of λ3 − 7λ2 − 160λ + 1236 The main theorem
Question (Schwartz 1992)
Are X and Y analytic functions of the coordinates x1, y1,..., xn, yn of the vertices of A? Theorem (G. 2017) The answer to the question is “yes”, and even better, X and Y lie in a degree 3 extension of Q(x1, y1,..., xn, yn). Projective transformations
For φ˜ ∈ GL3(R), the linear map
φ˜ : R3 → R3
induces a projective transformation
φ : R2 → R2
φ11x + φ12y + φ13 φ21x + φ22y + φ23 (x, y) 7→ , φ31x + φ32y + φ33 φ31x + φ32y + φ33 of R2 ⊆ RP2.
◮ Write PGL3(R)= {projective transformations}. Schwartz’s question for pentagons and hexagons
◮ Let A be a convex pentagon. Then ∃φ ∈ PGL3(R) such that
T (A)= φ(A) Schwartz’s question for pentagons and hexagons
◮ Let A be a convex pentagon. Then ∃φ ∈ PGL3(R) such that
T (A)= φ(A) =⇒ T k (A)= φk (A) for all k Schwartz’s question for pentagons and hexagons
◮ Let A be a convex pentagon. Then ∃φ ∈ PGL3(R) such that
T (A)= φ(A) =⇒ T k (A)= φk (A) for all k =⇒ (X , Y ) a fixed point of φ Schwartz’s question for pentagons and hexagons
◮ Let A be a convex pentagon. Then ∃φ ∈ PGL3(R) such that
T (A)= φ(A) =⇒ T k (A)= φk (A) for all k =⇒ (X , Y ) a fixed point of φ X =⇒ Y an eigenvector of φ˜ 1 Schwartz’s question for pentagons and hexagons
◮ Let A be a convex pentagon. Then ∃φ ∈ PGL3(R) such that
T (A)= φ(A) =⇒ T k (A)= φk (A) for all k =⇒ (X , Y ) a fixed point of φ X =⇒ Y an eigenvector of φ˜ 1 ◮ It follows that X , Y are algebraic of degree 3 over Q(x1, y1,..., x5, y5). ◮ A similar argument works for hexagons. Recipe for (X , Y ) in general
T 3 Lift (xj , yj ) to vj =[xj yj 1] ∈ R . Define a linear map
L : R3 → R3 n |vj−1, u, vj+1| L(u)= − vj |vj−1, vj , vj+1| j=1 X where |·, ·, ·| denotes a 3-by-3 determinant. Recipe for (X , Y ) in general
T 3 Lift (xj , yj ) to vj =[xj yj 1] ∈ R . Define a linear map
L : R3 → R3 n |vj−1, u, vj+1| L(u)= − vj |vj−1, vj , vj+1| j=1 X where |·, ·, ·| denotes a 3-by-3 determinant. Proposition For any convex n-gon, X Y 1 is an eigenvector of L. Properties of LA
3 3 LA : R → R n |vj−1, u, vj+1| LA(u)= − vj |vj−1, vj , vj+1| j=1 X Proposition Let A be a convex n-gon, B = T (A).
1. LA is independent of the choice of lifts v1,..., vn
2. Q ∈ conv(B)=⇒ LA(Q) ∈ conv(A)
3. LA = LB Picture of (projectivization of) LA
Aj Aj−1 Aj+1
b Q
n 1 Area(△QAj−1Aj+1) LA(Q)= Aj Λ Area(△Aj−1Aj Aj+1) j=1 X (Λ = normalization constant) Underlying combinatorics
There are identifications
{polygons} {edge weightings of Γ} {vertex weightings of Q}
for which
T Postnikov transformations Y -pattern mutations
where
Γ=
Q = dual quiver Underlying combinatorics (continued)
Method 1 (G. 2011, Gekhtman et al. 2012, ...)
◮ edge weights = areas A1A2A3, A1A2A4, A1A3A4, A2A3A4, ...
◮ y-variables = cross ratios [A1, B1, B2, A3],....
A2 A1 A3 b b b b B1 B2 B3 A4 Underlying combinatorics (continued)
Method 2 (based on Kenyon)
◮ (complex) edge weights = displacements A2 − B0, A2 − B1, A2 − B2, A2 − B3, A3 − B1, ... ◮ y-variables = double ratios (not projectively natural)
A2 b A1 A3 b b B1 B2 A2 − B0 A2 − B3 b y = b A2 − B1 A2 − B2 B0 B3 A duality on polygons
A B
◮ sides of A parallel to sides of B A duality on polygons
A B
◮ sides of A parallel to sides of B ◮ sides of T (A) parallel to sides of T −1(B) A duality on polygons
A B
◮ sides of A parallel to sides of B ◮ sides of T (A) parallel to sides of T −1(B) ◮ sides of T 2(A) parallel to sides of T −2(B) A duality on polygons
A B
◮ sides of A parallel to sides of B ◮ sides of T (A) parallel to sides of T −1(B) ◮ sides of T 2(A) parallel to sides of T −2(B) ◮ ... Say A and B are dual about the line at infinity. A duality on polygons (continued)
Proposition A closed convex polygon has a (possibly twisted) dual about a given line or through a given point.
A B A duality on polygons (continued)
Proposition A closed convex polygon has a (possibly twisted) dual about a given line or through a given point.
A B
Theorem A closed convex polygon has a closed dual through its limit point.