Fighting Fish: enumerative properties. Enrica Duchi. Veronica Guerrini & Simone Rinaldi DIISM, Università di Siena.

Transcription

1 Fighting Fish: enumerative properties Enrica Duchi IRIF, Université Paris Diderot Veronica Guerrini & Simone Rinaldi DIISM, Università di Siena Gilles Schaeffer LIX, CNRS and École Polytechnique Séminaire Philippe Flajolet, 27

2 Summary of the talk Fighting fish, a new combinatorial model of discrete branching surfaces Exact counting formulas for fighting fish Decompositions for fighting fish Fighting fish VS classical combinatorial structures a bijective challenge...

3 Fighting fish, definition Cells 45 o tilted unit square (of thin paper or cloth) upper left edge lower left edge Build surface by gluing cells along edges in a coherent way: upper left with lower right or lower left with upper right. upper right edge lower right edge glued edges free edges = These objects do not necessarily fit in the plane so my pictures are projections of the actual surfaces: Apparently overlapping cells are in fact independant.

4 Fighting fish, definition Directed cell aggregation. Restrict to only three legal ways to add cells: by lower right gluing, upper right gluing, or simultaneous lower and upper right gluings from adjacent free edges. lower right gluing simultaneous right gluing upper right gluing

5 Fighting fish, definition Lemma. Single cell + aggregations a simply connected surface Remark. Such surfaces can be recovered from their boundary walk.

6 Fighting fish, definition Fighting fish A fighting fish is a surface that can be obtained from a single cell by a sequence of directed cell agregations We are interested only in the resulting surface, not in the aggregation order (but type of aggregation matters) 2 = 2 = but 8 free edges free edges

7 Small fighting fish n = 2 n = 3 n = 4 n = 5

8 Fighting fish versus polyominoes Polyomino = edge-connected set of cells of the planar square lattice

9 Fighting fish versus polyominoes Polyomino = edge-connected set of cells of the planar square lattice Directed polyominoes: there is a cell, the head, from which all cells can be reached by a left-to-right path.

10 Fighting fish versus polyominoes Polyomino = edge-connected set of cells of the planar square lattice Directed polyominoes: there is a cell, the head, from which all cells can be reached by a left-to-right path. Proposition. A fighting fish is a directed polyomino iff its projection in the plane is injective. fighting fish do not all fit in the plane, ie they are not all polyominoes.

11 Fighting fish versus polyominoes Polyomino = edge-connected set of cells of the planar square lattice Directed polyominoes: there is a cell, the head, from which all cells can be reached by a left-to-right path. Proposition. A fighting fish is a directed polyomino iff its projection in the plane is injective. fighting fish do not all fit in the plane, ie they are not all polyominoes.

12 Fighting fish versus polyominoes Polyomino = edge-connected set of cells of the planar square lattice Directed polyominoes: there is a cell, the head, from which all cells can be reached by a left-to-right path. Proposition. A fighting fish is a directed polyomino iff its projection in the plane is injective. fighting fish do not all fit in the plane, ie they are not all polyominoes. Conversely there are directed polyominoes that are not fighting fish:

13 Fighting fish versus polyominoes Polyomino = edge-connected set of cells of the planar square lattice Directed polyominoes: there is a cell, the head, from which all cells can be reached by a left-to-right path. Proposition. A fighting fish is a directed polyomino iff its projection in the plane is injective. fighting fish do not all fit in the plane, ie they are not all polyominoes. Conversely there are directed polyominoes that are not fighting fish: Fighting fish are a generalization of directed polyominoes without holes.

14 Fighting fish versus polyominoes Polyomino = edge-connected set of cells of the planar square lattice Directed polyominoes: there is a cell, the head, from which all cells can be reached by a left-to-right path. Proposition. A fighting fish is a directed polyomino iff its projection in the plane is injective. fighting fish do not all fit in the plane, ie they are not all polyominoes. Conversely there are directed polyominoes that are not fighting fish: Fighting fish are a generalization of directed polyominoes without holes.

15 Parameters of fighting fish head tails fin Area = # cells branch points Size = semi-perimeter = #{upper free edges} = #{upper left free edges} + #{upper right free edges} The fin length = #{ lower free edges from head to first tail }

16 Parameters of fighting fish head tails fin Area = # cells branch points Size = semi-perimeter = #{upper free edges} = #{upper left free edges} + #{upper right free edges} The fin length = #{ lower free edges from head to first tail } Fighting fish with exactly tail

17 Parameters of fighting fish head tails fin Area = # cells branch points Size = semi-perimeter = #{upper free edges} = #{upper left free edges} + #{upper right free edges} The fin length = #{ lower free edges from head to first tail } Fighting fish with exactly tail = parallelogram polyominoes aka staircase polygons in this case, fin length = semi-perimeter

18 Enumerative results

19 Enumerative results Theorem (folklore) { } parallelogram polyominos # = with semi-perimeter n + { } parallelogram polyominos with # i top left and j top right edges fighting fish with tail 2n + = i+j ( ) 2n n ( i+j i )( i+j j )

20 Enumerative results Theorem (folklore) { } parallelogram polyominos # = with semi-perimeter n + { } parallelogram polyominos with # i top left and j top right edges fighting fish with tail 2n + = i+j ( ) 2n n ( i+j i )( i+j j ) { # Theorem (D., Guerrini, Rinaldi, Schaeffer, 26) { } ( ) fighting fish 2 3n # = with semi-perimeter n + (n + )(2n + ) n } fighting fish with = i top left and j top right edges (2i+j )(2j+i ) ( 2i+j i )( 2j+i j )

21 Fighting fish as random branching surfaces Let F n be a fighting fish taken uniformly at random among all fighting fish of size n.

22 Fighting fish as random branching surfaces Let F n be a fighting fish taken uniformly at random among all fighting fish of size n. Theorem (D., Guerrini, Rinaldi, Schaeffer, J. Physics A, 26) The expected area of F n is of order n 5/4

23 Fighting fish as random branching surfaces Let F n be a fighting fish taken uniformly at random among all fighting fish of size n. Theorem (D., Guerrini, Rinaldi, Schaeffer, J. Physics A, 26) The expected area of F n is of order n 5/4 Compare to the known expected area n 3/2 of random parallelogram polyominoes of size n

24 Fighting fish as random branching surfaces Let F n be a fighting fish taken uniformly at random among all fighting fish of size n. Theorem (D., Guerrini, Rinaldi, Schaeffer, J. Physics A, 26) The expected area of F n is of order n 5/4 Compare to the known expected area n 3/2 of random parallelogram polyominoes of size n Uniform random fighting fish of size n gives a new model of random branching surfaces with original features.

25 Fish tails We start by giving the definition of a slightly more general class: Fighting fish tails.

26 Fish tails We start by giving the definition of a slightly more general class: Fighting fish tails. A cell is made up by two scales Left scale Right scale

27 Fish tails We start by giving the definition of a slightly more general class: Fighting fish tails. A cell is made up by two scales Left scale A fish tail is a surface that can be obtained from a strip of right scales by a sequence of directed cell agregations. Right scale

28 Fish tails We start by giving the definition of a slightly more general class: Fighting fish tails. A cell is made up by two scales Left scale A fish tail is a surface that can be obtained from a strip of right scales by a sequence of directed cell agregations. Right scale

29 Fish tails We start by giving the definition of a slightly more general class: Fighting fish tails. A cell is made up by two scales Left scale A fish tail is a surface that can be obtained from a strip of right scales by a sequence of directed cell agregations. area= the number of left and right scales in the fish tail height= the number of right scales in the strip size= the number of upper left and right free edges Right scale

30 A recursive decomposition

31 A recursive decomposition The empty fish is the unique fish tail with height. T T T T l + l l l l l l l k T 2 k T 2 k T 2 k T 2 Operation u Operation h Operation h Operation d

32 A recursive decomposition The empty fish is the unique fish tail with height. T T T T l + l l l l l l l k T 2 k T 2 k T 2 k T 2 Operation u Operation h Operation h Operation d Every fish tail can be obtained in a unique way using operations u, h, h, d.

33 A recursive definition

34 h A recursive definition

35 h A recursive definition

36 A recursive definition h h

37 A recursive definition h h

38 A recursive definition h h d

39 A recursive definition h h d

40 A recursive definition h h h d

41 A recursive definition h h u h d h u

42 A recursive definition h h u h d h u

43 Fish tails vs fighting fish Fish tails with height and n free upper edges are in one-to-one correspondence with fighting fish with n + free upper edges.

44 Fish tails vs fighting fish Fish tails with height and n free upper edges are in one-to-one correspondence with fighting fish with n + free upper edges. n = n = 2 n = 3 n = 4

45 Fish tails vs fighting fish Fish tails with height and n free upper edges are in one-to-one correspondence with fighting fish with n + free upper edges. n = 2 n = 3 n = 4 n = 5

46 The generating function Let FT be the set of fish tails without the empty fish tail. Then T (v, q, x, t) = T FT vh(t ) q a(t ) x c(t ) t n(t ) denote the generating function of fish tails, where h(t ) is the height of T a(t ) is the area of T c(t ) is the number of tails of T n(t ) is the semi-perimeter of T

47 The generating function Let FT be the set of fish tails without the empty fish tail. Then T (v, q, x, t) = T FT vh(t ) q a(t ) x c(t ) t n(t ) denote the generating function of fish tails, where h(t ) is the height of T a(t ) is the area of T c(t ) is the number of tails of T n(t ) is the semi-perimeter of T Let us denote T (v, q) T (v, q, x, t) f(q) = [v]t (v, q) We are going to write the functional equation associated with the previous construction.

48 The functional equation for fish tails T (v, q, x, t) = T FT vh(t ) q a(t ) x c(t ) t n(t )

49 The functional equation for fish tails T (v, q, x, t) = T FT vh(t ) q a(t ) x c(t ) t n(t ) T l + l T n(t ) = n(t ) + n(t 2 ) + h(t ) = h(t ) + h(t 2 ) + T 2 c(t ) = c(t ) + c(t 2 ) if l := h(t ) Operation u c(t ) = c(t 2 ) + if l = a(t ) = a(t ) + a(t 2 ) + 2l +

50 The functional equation for fish tails T (v, q, x, t) = T FT vh(t ) q a(t ) x c(t ) t n(t ) T l + l T n(t ) = n(t ) + n(t 2 ) + h(t ) = h(t ) + h(t 2 ) + T 2 c(t ) = c(t ) + c(t 2 ) if l := h(t ) Operation u c(t ) = c(t 2 ) + if l = a(t ) = a(t ) + a(t 2 ) + 2l + Operation u gives the term tvqt (vq 2, q, x, t)(t (v, q, x, t) + ) + tvqx(t (v, q, x, t) + ) = tvq(t (vq 2, q) + x)(t (v, q) + )

51 The functional equation for fish tails T (v, q, x, t) = T FT vh(t ) q a(t ) x c(t ) t n(t ) T T T T n(t ) = n(t ) + n(t 2 ) + l l l l h(t ) = h(t ) + h(t 2 ) k T 2 k T 2 Operation h Operation h c(t ) = c(t ) + c(t 2 ) (l ) a(t ) = a(t ) + a(t 2 ) + 2l Operation h and h give the term 2t(T (vq 2, q)(t (v, q) + )

52 The functional equation for fish tails T (v, q, x, t) = T FT vh(t ) q a(t ) x c(t ) t n(t ) T T n(t ) = n(t ) + n(t 2 ) + l l h(t ) = h(t ) + h(t 2 ) (l > ) k Operation d T 2 c(t ) = c(t ) + c(t 2 ) a(t ) = a(t ) + a(t 2 ) + 2l Operation u gives the term t vq (T (vq2, q) vq 2 f(q))(t (v, q) + )

53 Enumeration wrt the perimeter and number of tails The functional equation T (v, q) = tvq(t (vq 2, q) + x)(t (v, q) + ) + 2tT (vq 2, q)(t (v, q) + ) + t vq (T (vq2, q) vq 2 f(q))(t (v, q) + )

54 Enumeration wrt the perimeter and number of tails The functional equation T (v, q) = tvq(t (vq 2, q) + x)(t (v, q) + ) + 2tT (vq 2, q)(t (v, q) + ) + t vq (T (vq2, q) vq 2 f(q))(t (v, q) + ) Letting q = the master equation reduces to T (v) = tv(t (v) + x)(t (v) + ) + 2tT (v)(t (v) + ) + t v (T (v) vf)(t (v) + ) where T (v) T (v, ) and f f()

55 Enumeration wrt the perimeter and number of tails The functional equation T (v, q) = tvq(t (vq 2, q) + x)(t (v, q) + ) + 2tT (vq 2, q)(t (v, q) + ) + t vq (T (vq2, q) vq 2 f(q))(t (v, q) + ) Letting q = the master equation reduces to T (v) = tv(t (v) + x)(t (v) + ) + 2tT (v)(t (v) + ) + t v (T (v) vf)(t (v) + ) where T (v) T (v, ) and f f() This equation is now a polynomial equation with one catalytic variable and it admits an explicitly computable algebraic solution. (Bousquet-Mélou and Jehanne, J. Combin. Theory Ser.B, 26)

56 Enumeration wrt semi-perimeter and number of tails T (v) = tv(t (v) + x)(t (v) + ) + 2tT (v)(t (v) + ) + t v (T (v) vf)(t (v) + ) We apply the Bousquet-Mélou Jehanne trick:

57 Enumeration wrt semi-perimeter and number of tails T (v) = tv(t (v) + x)(t (v) + ) + 2tT (v)(t (v) + ) + t v (T (v) vf)(t (v) + ) We apply the Bousquet-Mélou Jehanne trick: Upon deriving with respect to v we obtain the following equation ( tv(t (v) + x) 2tT (v) t v (T (v) vf) (tv + 2t + t v = (T (v) + )(t(t (v) + x) t v 2 (T (v) vf) t v f) )(T (v) + )) dt (v) dv =

58 Enumeration wrt semi-perimeter and number of tails T (v) = tv(t (v) + x)(t (v) + ) + 2tT (v)(t (v) + ) + t v (T (v) vf)(t (v) + ) We apply the Bousquet-Mélou Jehanne trick: Upon deriving with respect to v we obtain the following equation ( tv(t (v) + x) 2tT (v) t v (T (v) vf) (tv + 2t + t v = (T (v) + )(t(t (v) + x) t v 2 (T (v) vf) t v f) )(T (v) + )) dt (v) dv = There is a power series V V (t), such that setting v = V cancels the left hand side: ( tv (T (V ) + x) 2tT (V ) t V (T (V ) V f) (tv + 2t + t V we then also have and the main equation gives a third equation. )(T (V ) + )) = (T (V ) + )(t(t (V ) + x) t V 2 (T (V ) V f) t V f) =

59 Enumeration wrt semi-perimeter and number of tails Simplifying the previous system of equations we obtain V = t( + V + xv 2 V )2 f = xv x 2 V 3 ( V ) 2 T (V ) = xv 2 V 2

60 Enumeration wrt semi-perimeter and number of tails Simplifying the previous system of equations we obtain V = t( + V + xv 2 V )2 f = xv x 2 V 3 ( V ) 2 T (V ) = xv 2 V 2 By setting x = we obtain equations for fighting fish according to the semi-perimeter only.

61 Enumeration wrt semi-perimeter and number of tails Simplifying the previous system of equations we obtain V = t( + V + xv 2 V )2 f = xv x 2 V 3 ( V ) 2 T (V ) = xv 2 V 2 By setting x = we obtain equations for fighting fish according to the semi-perimeter only. V = t ( V ) 2 f = V V 3 ( V ) 2

62 Enumeration wrt semi-perimeter and number of tails Simplifying the previous system of equations we obtain V = t( + V + xv 2 V )2 f = xv x 2 V 3 ( V ) 2 T (V ) = xv 2 V 2 By setting x = we obtain equations for fighting fish according to the semi-perimeter only. V = t ( V ) 2 f = V V 3 ( V ) 2 Lagrange inversion formula The number of fighting fish with size n + [t n ]f = 2 (n+)(2n+)( 3n n )

63 Enumeration wrt semi-perimeter and number of tails. The same approach can be applied to (re)derive the number of fighting fish of size n + with one tail.

64 Enumeration wrt semi-perimeter and number of tails. The same approach can be applied to (re)derive the number of fighting fish of size n + with one tail. [x]f = [x ]V = V f = xv x 2 V 3 ( V ) 2 where V = t( + V ) 2 V = t( + V + xv 2 V )2 [xt n ]f = 2n n+( n Parallelogram polyominoes of size n + )

65 Enumeration wrt semi-perimeter and number of tails. The same approach can be applied to (re)derive the number of fighting fish of size n + with one tail. [x]f = [x ]V = V f = xv x 2 V 3 ( V ) 2 where V = t( + V ) 2 V = t( + V + xv 2 V )2 [xt n ]f = Our decomposition generalizes a Temperly like decomposition for parallelogram polyominoes. l + l T l l T l l T 2n n+( n Parallelogram polyominoes of size n + l l T T 2 T 2 T 2 T 2 ) Operation u Operation h Operation h Operation d

66 Enumeration wrt semi-perimeter and number of tails. The same approach can be applied to (re)derive the number of fighting fish of size n + with one tail. [x]f = [x ]V = V f = xv x 2 V 3 ( V ) 2 where V = t( + V ) 2 V = t( + V + xv 2 V )2 [xt n ]f = 2n n+( n Parallelogram polyominoes of size n + ) l + l T l l T l l T l l T where T is a fighting fish with tail Operation u Operation h Operation h Operation d

67 Enumeration wrt semi-perimeter and number of tails. The same approach can be applied to (re)derive the number of fighting fish of size n + with one tail. [x]f = [x ]V = V f = xv x 2 V 3 ( V ) 2 where V = t( + V ) 2 V = t( + V + xv 2 V )2 [xt n ]f = 2n n+( n Parallelogram polyominoes of size n + ) l + l l l l l l l Operation u Operation h Operation h Operation d

68 Enumeration wrt semi-perimeter and number of tails. The same approach can be applied to (re)derive the number of fighting fish of size n + with one tail. [x]f = [x ]V = V f = xv x 2 V 3 ( V ) 2 where V = t( + V ) 2 V = t( + V + xv 2 V )2 [xt n ]f = 2n n+( n Parallelogram polyominoes of size n + ) More generally generating function for fighting fish with size n + and c tails is rational in the Catalan generating function. However explicit expressions are not particularly simple

69 Enumeration wrt semi-perimeter and number of tails. The same approach can be applied to (re)derive the number of fighting fish of size n + with one tail. [x]f = [x ]V = V f = xv x 2 V 3 ( V ) 2 where V = t( + V ) 2 V = t( + V + xv 2 V )2 [xt n ]f = 2n n+( n Parallelogram polyominoes of size n + ) More generally generating function for fighting fish with size n + and c tails is rational in the Catalan generating function. However explicit expressions are not particularly simple The number of fighting fish with size n + and a marked tail is n ( 3n 2 ) n df dx x= + Lagrange inversion formula

70 Enumeration wrt semi-perimeter and number of tails. The same approach can be applied to (re)derive the number of fighting fish of size n + with one tail. [x]f = [x ]V = V f = xv x 2 V 3 ( V ) 2 where V = t( + V ) 2 V = t( + V + xv 2 V )2 [xt n ]f = 2n n+( n Parallelogram polyominoes of size n + ) More generally generating function for fighting fish with size n + and c tails is rational in the Catalan generating function. However explicit expressions are not particularly simple The number of fighting fish with size n + and a marked tail is n ( 3n 2 ) n df dx x= + Lagrange inversion formula The average number of tails of fighting fish of size n + is [x n ] df dx x= [x n ]f = (n+)(2n+) 3(3n )

71 Enumeration of fighting fish wrt the area We are going to count fighting fish weighted by their area. Let A A(x, t) be the total area generating function. Then A(x, t) = F a(f )tn(f ) = (qf(q)) q q= = f + (f(q)) q q=

72 Enumeration of fighting fish wrt the area We are going to count fighting fish weighted by their area. Let A A(x, t) be the total area generating function. Then A(x, t) = F a(f )tn(f ) = (qf(q)) q Let us consider the master equation q= = f + (f(q)) q q= T (v, q) = tvq(t (vq 2, q) + x)(t (v, q) + ) + 2tT (vq 2, q)(t (v, q) + ) + t vq (T (vq2, q) vq 2 f(q))(t (v, q) + )

73 Enumeration of fighting fish wrt the area We are going to count fighting fish weighted by their area. Let A A(x, t) be the total area generating function. Then A(x, t) = F a(f )tn(f ) = (qf(q)) q Let us consider the master equation q= = f + (f(q)) q q= T (v, q) = tvq(t (vq 2, q) + x)(t (v, q) + ) + 2tT (vq 2, q)(t (v, q) + ) + t vq (T (vq2, q) vq 2 f(q))(t (v, q) + ) By deriving with respect to q and by setting q = we obtain ( tv(t (v, ) + x) 2tT (v, ) t v (T (v, ) vf) (tv + 2t + t T )(T (v) + )) (v, ) v q = (T ( (v, ) + ) ) (tv + 2t + t T ) 2v v v (v, ) + tv(t (v, ) + x) t f (T (v, ) vf) 2tf t v q ()

74 Enumeration of fighting fish wrt the area We are going to count fighting fish weighted by their area. Let A A(x, t) be the total area generating function. Then A(x, t) = F a(f )tn(f ) = (qf(q)) q Let us consider the master equation q= = f + (f(q)) q q= T (v, q) = tvq(t (vq 2, q) + x)(t (v, q) + ) + 2tT (vq 2, q)(t (v, q) + ) + t vq (T (vq2, q) vq 2 f(q))(t (v, q) + ) By setting v = V we have (tv + 2t + t V ) 2V T v (V, ) + tv (T (V, ) + x) t V (T (V, ) V f) 2tf t f q () To obtain T v (V, ) we apply again the kernel method.

75 The area generating function The generating function A A(x, t) for the total area of fighting fish with size n + satisfies V ( V ) 2 A 2 + 2( V ) 2 ( V 2 + xv 2 )A 4xV ( V 2 + xv 2 ) =

76 The area generating function The generating function A A(x, t) for the total area of fighting fish with size n + satisfies V ( V ) 2 A 2 + 2( V ) 2 ( V 2 + xv 2 )A 4xV ( V 2 + xv 2 ) = Extracting the coefficient of x, [x]a = A, yields 2( V ) 2 ( V 2 )A 4V ( V 2 ) = where V = [x ]V is a Catalan generating function satisfying V = t( + V ) 2 we obtain the generating function for the total area of parallelogram polyominoes A = 2V ( V ) 2 = 2t 4t The simplification to a rational function of t is a well-known feature of parallelogram polyominoes.

77 The area generating function The generating function A A(x, t) for the total area of fighting fish with size n + satisfies V ( V ) 2 A 2 + 2( V ) 2 ( V 2 + xv 2 )A 4xV ( V 2 + xv 2 ) = Extracting the coefficient of x, [x]a = A, yields 2( V ) 2 ( V 2 )A 4V ( V 2 ) = where V = [x ]V is a Catalan generating function satisfying V = t( + V ) 2 we obtain the generating function for the total area of parallelogram polyominoes A = 2V ( V ) 2 = 2t 4t The simplification to a rational function of t is a well-known feature of parallelogram polyominoes. The average area for parallelogram polyominoes of size n is 4 n /C n where C n are the Catalan numbers. The average area for parallelogram polyominoes of size n scales like n 3 2

78 The area generating function The generating function A A(x, t) for the total area of fighting fish with size n + satisfies V ( V ) 2 A 2 + 2( V ) 2 ( V 2 + xv 2 )A 4xV ( V 2 + xv 2 ) = Extracting the coefficient of x, [x]a = A, yields 2( V ) 2 ( V 2 )A 4V ( V 2 ) = where V = [x ]V is a Catalan generating function satisfying V = t( + V ) 2 we obtain the generating function for the total area of parallelogram polyominoes A = 2V ( V ) 2 = 2t 4t The simplification to a rational function of t is a well-known feature of parallelogram polyominoes. The average area for parallelogram polyominoes of size n is 4 n /C n where C n are the Catalan numbers. The average area for parallelogram polyominoes of size n scales like n 3 2 The generating function of the total area of fighting fish with c tails and size n + is a rational function of the Catalan generating function V

79 The area generatin function The average area A n of fighting fish with size n + grows like n 5 4

80 The area generatin function The average area A n of fighting fish with size n + grows like n 5 4 Indeed, let us reconsider the equation for V, f, and A, V = t ( V ) 2 f = V V 3 ( V ) 2 A = V (+V )( 3V ) V ( V ) and get asymptotics by singularity analysis

81 The area generatin function The average area A n of fighting fish with size n + grows like n 5 4 Indeed, let us reconsider the equation for V, f, and A, V = t ( V ) 2 f = V V 3 ( V ) 2 A = V (+V )( 3V ) V ( V ) We obtain: [t n ]f cte n 5 2 t n c n Then the average area is and get asymptotics by singularity analysis [t n ]A cte n 4 5 t n n c [t n ]A [t n ]f n cte n 5 4

82 A refinement of the main formula T (v, q, x, t) = T FT al(t ) b r(t ) v h(t ) q a(t ) x c(t ) t n(t ) denote the generating function of fish tails, where l(t ) is the number of upper-left free edges of T r(t ) is the number of upper-right edges of T T T T T l + l l l l l l l k T 2 k T 2 k T 2 k T 2 Operation u Operation h Operation h Operation d T (v, q) = tvqb(t (vq 2, q) + x)(t (v, q) + ) + at (vq 2, q)(t (v, q) + )+ { # +bt (vq 2, q)(t (v, q) + ) + t vq a(t (vq2, q) vq 2 f(q))(t (v, q) + ) } ) = fighting fish with i top left and j top right edges (2i+j )(2j+i ) ( 2i+j i )( 2j+i j

83 n algebraic decomposition for parallelogram polyominoes. Since the gf function f of fighting fish is algebraic we would like to find an algebraic decomposition.

84 n algebraic decomposition for parallelogram polyominoes. Since the gf function f of fighting fish is algebraic we would like to find an algebraic decomposition. We try a new decomposition starting from a grammar-like decomposition on parallelogram polyominoes.

85 n algebraic decomposition for parallelogram polyominoes. Since the gf function f of fighting fish is algebraic we would like to find an algebraic decomposition. We try a new decomposition starting from a grammar-like decomposition on parallelogram polyominoes. The wasp-waist decomposition for parallelogram polyominoes

86 n algebraic decomposition for parallelogram polyominoes. Since the gf function f of fighting fish is algebraic we would like to find an algebraic decomposition. We try a new decomposition starting from a grammar-like decomposition on parallelogram polyominoes. The wasp-waist decomposition for parallelogram polyominoes Let P = P t P be the GF of parallelogram polyominoes according to the size, then P = t + 2tP + tp 2

87 A new decomposition Extend the wasp-waist decomposition of parallelogram polyominoes: remove one cell at the bottom of each diagonal, from left to right along the fin, until this creates a cut +

88 A new decomposition Extend the wasp-waist decomposition of parallelogram polyominoes: remove one cell at the bottom of each diagonal, from left to right along the fin, until this creates a cut + Two more cases must be considered for fighting fish...

89 A glipse of the proof Let F (u) = f t f u fin(f) x tail(f) be the GF of fighting fish according to the size, fin length and number of extra tails. Then F (u) = tu( + F (u)) 2 F () F (u) + xtuf (u) with f = F (). u

90 A glipse of the proof Let F (u) = f t f u fin(f) x tail(f) be the GF of fighting fish according to the size, fin length and number of extra tails. Then F (u) = tu( + F (u)) 2 F () F (u) + xtuf (u) with f = F (). u Case x =. Fish with one tail, ie parallelogram polyominoes: we have the usual algebraic equation for the GF of Catalan numbers.

91 A glipse of the proof Let F (u) = f t f u fin(f) x tail(f) be the GF of fighting fish according to the size, fin length and number of extra tails. Then F (u) = tu( + F (u)) 2 F () F (u) + xtuf (u) with f = F (). u Case x =. Fish with one tail, ie parallelogram polyominoes: we have the usual algebraic equation for the GF of Catalan numbers. But in the general case we have again a polynomial equation with one catalytic variable... The question to find a direct algebraic decomposition of fighting fish remain.

92 Bijections and parameter equidistributions?

93 Sloane s Online Encyclopedia of Integer Sequences { # fighting fish with semi-perimeter n + } = 2 (n + )(2n + ) ( ) 3n n, 2, 6, 9, 48, This integer sequence was already in Sloane s OEIS! The number of fighting fish of size n + (with i left and j down top edges) is equal to the number of: Two-stack sortable permutations of {,..., n} (i ascending and j descending runs) (West, Zeilberger, Bona, 9 s) Rooted non separable planar maps with n edges (i + vertices, j + faces) (Tutte, Mullin and Schellenberg, 6 s) Left ternary trees with n noeuds (i + even, j odd vertices) (Del Lungo, Del Ristoro, Penaud, 999)

94 Left ternary trees and further equidistributions Natural embedding of a ternary tree: root vertex has label vertex with label i left child i +, central child i, right child i. 2 -

95 Left ternary trees and further equidistributions Natural embedding of a ternary tree: root vertex has label vertex with label i left child i +, central child i, right child i. Left ternary tree = ternary tree without negative labels

96 Left ternary trees and further equidistributions Natural embedding of a ternary tree: root vertex has label vertex with label i left child i +, central child i, right child i. Left ternary tree = ternary tree without negative labels. Core = binary subtree of the root after pruning all right edges

97 Left ternary trees and further equidistributions Natural embedding of a ternary tree: root vertex has label vertex with label i left child i +, central child i, right child i. Left ternary tree = ternary tree without negative labels. Core = binary subtree of the root after pruning all right edges Theorem (DGRS 26): The number of fighting fish with size n + and fin length k equals the number of left ternary trees with n nodes and core size k

98 Left ternary trees and further equidistributions Natural embedding of a ternary tree: root vertex has label vertex with label i left child i +, central child i, right child i. Left ternary tree = ternary tree without negative labels. Core = binary subtree of the root after pruning all right edges Theorem (DGRS 26): The number of fighting fish with size n + and fin length k equals the number of left ternary trees with n nodes and core size k. 2 Proof? We computed the gf of fighting fish wrt size and fin length. Compute the gf of left ternary trees wrt size and core size

99 Left ternary trees and further equidistributions Core = binary subtree of the root after pruning all right edges

100 Left ternary trees and further equidistributions 2 2 Easy for ternary trees: - Core = binary subtree of the root after pruning all right edges = +

101 Left ternary trees and further equidistributions Core = binary subtree of the root after pruning all right edges = + Easy for ternary trees: Proposition The GF τ(u) of ternary trees wrt size and core size is τ(u) = + tuτ(u) 2 τ() generalizing τ() = + tτ() 3

102 Left ternary trees and further equidistributions Core = binary subtree of the root after pruning all right edges = + Easy for ternary trees: Proposition The GF τ(u) of ternary trees wrt size and core size is τ(u) = + tuτ(u) 2 τ() generalizing τ() = + tτ() 3 But: No known simple decomposition of left ternary trees. Known decompositions are complex and do not preserve core.

103 Left ternary trees and further equidistributions Core = binary subtree of the root after pruning all right edges = + Easy for ternary trees: Proposition The GF τ(u) of ternary trees wrt size and core size is τ(u) = + tuτ(u) 2 τ() generalizing τ() = + tτ() 3 But: No known simple decomposition of left ternary trees. Known decompositions are complex and do not preserve core. Idea: Solve a more general problem...

104 Left ternary trees and further equidistributions Core = binary subtree of the root after pruning all right edges = + Easy for ternary trees: Proposition The GF τ(u) of ternary trees wrt size and core size is τ(u) = + tuτ(u) 2 τ() generalizing τ() = + tτ() 3 But: No known simple decomposition of left ternary trees. Known decompositions are complex and do not preserve core. Idea: Solve a more general problem... Proposition The GFs wrt size and core size of left ternary trees with root label i satisfy τ (u) = + tuτ (u)τ (u) = + i i i+ i i- τ i (u) = + tuτ i+ (u)τ i (u)τ i () for i >

105 Left ternary trees and further equidistributions Core = binary subtree of the root after pruning all right edges The solution of the previous infinite system of equation is known for u = :

106 Left ternary trees and further equidistributions Core = binary subtree of the root after pruning all right edges Theorem (Di Francesco 5, Kuba ) The size GF of left ternary trees with root label i is τ i = τ ( Xi+5 ) ( X i+4 ) ( X i+2 ) ( X i+3 ) The solution of the previous infinite system of equation is known for u = : where { τ = + tτ 3 X = ( + X + X 2 ) τ τ.

107 Left ternary trees and further equidistributions Theorem (Di Francesco 5, Kuba ) The size GF of left ternary trees with root label i is τ i = τ ( Xi+5 ) ( X i+4 ) ( X i+2 ) ( X i+3 ) Core = binary subtree of the root after pruning all right edges The solution of the previous infinite system of equation is known for u = : where { τ = + tτ 3 X = ( + X + X 2 ) τ τ Proof by guessing the formula and checking it satisfies the recurrence..

108 Left ternary trees and further equidistributions Theorem (Di Francesco 5, Kuba ) The size GF of left ternary trees with root label i is τ i = τ ( Xi+5 ) ( X i+4 ) ( X i+2 ) ( X i+3 ) Core = binary subtree of the root after pruning all right edges The solution of the previous infinite system of equation is known for u = : where { τ = + tτ 3 X = ( + X + X 2 ) τ τ Proof by guessing the formula and checking it satisfies the recurrence. Case i = of this thm gives formula for left ternary trees of size n.

109 Left ternary trees and further equidistributions Theorem (Di Francesco 5, Kuba ) The size GF of left ternary trees with root label i is τ i = τ ( Xi+5 ) ( X i+4 ) τ i (u) = τ(u) H i(u) H i (u) ( X i+2 ) ( X i+3 ) Proof by guessing the formula and checking it satisfies the recurrence. X i+2 X i+3 Core = binary subtree of the root after pruning all right edges The solution of the previous infinite system of equation is known for u = : where where { τ = + tτ 3 X = ( + X + X 2 ) τ τ Case i = of this thm gives formula for left ternary trees of size n We extend the theorem and its proof by guessing the bivariate formula Theorem (DGRS6) The bivariate size and core size GF of left ternary trees with label i is { τ(u) = + tuττ(u) 2 H j (u) = ( X j+ )XT (u) ( + X)( X j+2 )..

110 Left ternary trees and further equidistributions Core = binary subtree of the root after pruning all right edges The solution of the previous infinite system of equation is known for u = : Theorem (DGRS 26): The number of fighting fish with size n + and fin length k equals the number of left ternary trees with n nodes and core size k. Conjecture (DGRS 26): The previous computation can be refined to prove joined equidistribution of: fin length core size number of tails number of right branches number of left/right free edges number of even/odd labels

111 Bijections? fighting fish 2SS-permutations left ternary trees ns planar maps

112 Bijections? fighting fish 2SS-permutations left ternary trees ns planar maps Tutte recursive decomposition + GF

113 Bijections? fighting fish Zeilberger recursive decomposition + GF 2SS-permutations left ternary trees ns planar maps Tutte recursive decomposition + GF

114 Bijections? fighting fish Zeilberger recursive decomposition + GF 2SS-permutations left ternary trees Goulden-West (isomorphic recursive decompositions) ns planar maps Tutte recursive decomposition + GF

115 Bijections? fighting fish 2SS-permutations left ternary trees Zeilberger recursive decomposition + GF Goulden-West (isomorphic recursive decompositions) ns planar maps Del Lungo et al (isomorphic recursive decompositions) Tutte recursive decomposition + GF

116 Bijections? fighting fish direct enumeration Schaeffer 2SS-permutations left ternary trees Zeilberger recursive decomposition + GF Goulden-West (isomorphic recursive decompositions) ns planar maps Del Lungo et al (isomorphic recursive decompositions) Schaeffer (direct bijection) Tutte recursive decomposition + GF

117 Bijections? recursive decomposition + GF today s talk fighting fish direct enumeration Schaeffer 2SS-permutations left ternary trees Zeilberger recursive decomposition + GF Goulden-West (isomorphic recursive decompositions) ns planar maps Del Lungo et al (isomorphic recursive decompositions) Schaeffer (direct bijection) Tutte recursive decomposition + GF

118 Bijections? recursive decomposition + GF today s talk fighting fish direct bijection? direct enumeration Schaeffer 2SS-permutations left ternary trees Zeilberger recursive decomposition + GF Goulden-West (isomorphic recursive decompositions) ns planar maps Del Lungo et al (isomorphic recursive decompositions) Schaeffer (direct bijection) Tutte recursive decomposition + GF

119 Bijections? recursive decomposition + GF today s talk direct enumeration? fighting fish direct bijection? direct enumeration Schaeffer 2SS-permutations left ternary trees Zeilberger recursive decomposition + GF Goulden-West (isomorphic recursive decompositions) ns planar maps Del Lungo et al (isomorphic recursive decompositions) Schaeffer (direct bijection) Tutte recursive decomposition + GF

120 Bijections? recursive decomposition + GF today s talk direct enumeration? Fang (27) (isomorphic recursive decompositions) Zeilberger recursive decomposition + GF 2SS-permutations Goulden-West (isomorphic recursive decompositions) fighting fish ns planar maps Tutte recursive decomposition + GF direct bijection? left ternary trees Del Lungo et al (isomorphic recursive decompositions) Schaeffer (direct bijection) direct enumeration Schaeffer

121 Bijections? recursive decomposition + GF today s talk direct enumeration? Fang (27) (isomorphic recursive decompositions) Zeilberger recursive decomposition + GF 2SS-permutations Goulden-West (isomorphic recursive decompositions) fighting fish ns planar maps Tutte recursive decomposition + GF direct bijection? left ternary trees Del Lungo et al (isomorphic recursive decompositions) Schaeffer (direct bijection) direct enumeration Schaeffer

122 THANK YOU

123 THANK YOU

124

125 The area generatin function The average area A n of fighting fish with size n + grows like n 5 4

126 The area generatin function The average area A n of fighting fish with size n + grows like n 5 4 Indeed, let us reconsider the equation for V, f, and A, V = t ( V ) 2 f = V V 3 ( V ) 2 A = V (+V )( 3V ) V ( V ) and get asymptotics by singularity analysis

127 The area generatin function The average area A n of fighting fish with size n + grows like n 5 4 Indeed, let us reconsider the equation for V, f, and A, V = t ( V ) 2 and get asymptotics by singularity analysis f = V V 3 ( V ) 2 A = V (+V )( 3V ) V ( V ) V = generating function of a simple family of trees. Square root singularity expansion near dominant singularity V = 3 cte ( t t c ) + O( t t c ) with t c = 4 27

128 The area generatin function The average area A n of fighting fish with size n + grows like n 5 4 Indeed, let us reconsider the equation for V, f, and A, V = t ( V ) 2 and get asymptotics by singularity analysis f = V V 3 ( V ) 2 A = V (+V )( 3V ) V ( V ) V = generating function of a simple family of trees. Square root singularity expansion near dominant singularity V = 3 cte ( t t c ) + O( t t c ) with t c = 4 27 f = ( t t c ) ( t t c ) O(( t t c ) 2 )

129 The area generatin function The average area A n of fighting fish with size n + grows like n 5 4 Indeed, let us reconsider the equation for V, f, and A, V = t ( V ) 2 f = V V 3 ( V ) 2 and get asymptotics by singularity analysis V = generating function of a simple family of trees. A = V (+V )( 3V ) V ( V ) Square root singularity expansion near dominant singularity V = 3 cte ( t t c ) + O( t t c ) with t c = 4 27 f = ( t t c ) ( t t c ) O(( t t c ) 2 ) A = 3 cte t + O( t ) = 3 cte ( t ) 4 + O(( t ) 2 ) t c t c t c t c

130 The area generating function We have the singular expansions: f = ( t t c ) ( t t c ) O(( t t c ) 2 ) A = 3 cte ( t t c ) 4 + O(( t t c ) 2 ) From transfert theorems: g ( t t c ) α we obtain: [t n ]A cte n 4 5 t n n c [t n ]g n α Γ( α) t n c Then the average area is [t n ]f cte n 5 2 t n c n [t n ]A [t n ]f n cte n 5 4