Ranking patterns of unfolding models of codimension one Hidehiko - - PowerPoint PPT Presentation

ranking patterns of unfolding models of codimension one
SMART_READER_LITE
LIVE PREVIEW

Ranking patterns of unfolding models of codimension one Hidehiko - - PowerPoint PPT Presentation

Recent developments on geometric and algebraic methods in Economics August 24, 2014 Hokkaido University Ranking patterns of unfolding models of codimension one Hidehiko Kamiya (Nagoya University) Joint work with H. Terao and A. Takemura This


slide-1
SLIDE 1

Recent developments on geometric and algebraic methods in Economics August 24, 2014 Hokkaido University

Ranking patterns of unfolding models of codimension one

Hidehiko Kamiya (Nagoya University) Joint work with H. Terao and A. Takemura

slide-2
SLIDE 2

This talk is based on

[1] H. Kamiya, A. Takemura, H. Terao, “Ranking patterns of unfolding models of codimension

  • ne,” Advances in Applied Mathematics, Vol.

47, Iss. 2 (2011), pp. 379–400. [2] H. Kamiya, A. Takemura, H. Terao, “Arrangements stable under the Coxeter groups,” In: Configuration Spaces: Geometry, Combinatorics and Topology (ed. A. Bjorner,

  • F. Cohen, C. De Concini, C. Procesi, M.

1

slide-3
SLIDE 3

Salvetti), CRM Series, Vol. 14 (2012) pp. 327-354, Scuola Normale Superiore Pisa. 2

slide-4
SLIDE 4

Table of Contents

  • 1. Introduction
  • 2. Unfolding as a braid slice
  • 3. All braid slices
  • 4. Unrealizable braid slices
  • 5. Number of ranking patterns
  • 6. Inequivalent ranking patterns
  • 7. Concluding remarks

3

slide-5
SLIDE 5

1 Introduction

  • The unfolding model

(Coombs (1950), Psychol. Rev. 57, 145–158), also known as the ideal point model.

  • A model for preference rankings.
  • Objects 1, 2, . . . , m.
  • An individual ranks these m objects.

4

slide-6
SLIDE 6

Unfolding model

  • The objects 1, 2, . . . , m are represented by

µ1, µ2, . . . , µm ∈ Rn.

  • The individual is represented by

y ∈ Rn (his/her ideal point).

  • Rn: the joint space.

5

slide-7
SLIDE 7

Preference

  • Preference: The nearer, the more preferred,

i.e., y prefers i to j ⇐ ⇒ ∥y − µi∥ < ∥y − µj∥.

  • y has ranking ( i1
  • best

i2

  • 2nd best

· · · im

  • worst object

) iff ∥y − µi1∥ < ∥y − µi2∥ < · · · < ∥y − µim∥. 6

slide-8
SLIDE 8

Coffee Brands

Sourness Richness

Mocha Colombia Brazil Kilimanjaro Costa Rica Blue Mountain Hawaii Kona Mandheling Guatemala 1 4 2 4

y

y likes Mocha best, Blue Mountain next,.... 7

slide-9
SLIDE 9

Admissible/inadmissible rankings

  • In general, we can think of m! rankings

among m objects.

  • Not all m! rankings are generated.

→ admissible/inadmissible rankings:            ∃y ∈ Rn s.t. ∥y − µi1∥ < · · · < ∥y − µim∥ → (i1i2 · · · im) : admissible; ∄y ∈ Rn s.t. ∥y − µi1∥ < · · · < ∥y − µim∥ → (i1i2 · · · im) : inadmissible. 8

slide-10
SLIDE 10

Admissible Rankings

4 3 1 2

1 1 1 2 2 2 3 3 3 4 4 4

Inadmissible rankings: (1243) (1423) (2143) (2413) (4123) (4213) (4321) (3421) (4231) (2431) (2341) (2314) (2134) (1234) (3241) (3124) (1324) (3142) (1432) (4132) (4312) (3412) (1342) (3214)

2, m=4

18 out of 4! = 24 rankings are admissible. (For simplicity, µ1, µ2, µ3, µ4 are written as 1, 2, 3, 4.)

9

slide-11
SLIDE 11

Ranking pattern

  • Define the ranking pattern

RPUF(µ1, . . . , µm) := {admissible rankings}

  • f the unfolding model with µ1, . . . , µm.
  • The number of admissible rankings, i.e.,

|RPUF(µ1, . . . , µm)|, was obtained in Good and Tideman (1977), Kamiya and Takemura (1997), Zaslavsky (2002). 10

slide-12
SLIDE 12

Different ranking patterns

  • A different m-tuple (µ1, . . . , µm),

→ a different RPUF(µ1, . . . , µm).

4 3 1 2

1 1 1 2 2 2 3 3 3 4 4 4

Inadmissible rankings: (1243) (1423) (2143) (2413) (4123) (4213) (4321) (3421) (4231) (2431) (2341) (2314) (2134) (1234) (3241) (3124) (1324) (3142) (1432) (4132) (4312) (3412) (1342) (3214)

4 3 1 2

Inadmissible rankings: (2143) (2413) (2431) (4123) (4213) (4231)

3 3 3 2 2 2 4 4 4 1 1 1

(4321) (3421) (1423) (1243) (2341) (2314) (2134) (1234) (3241) (3124) (1324) (3142) (1432) (4132) (4312) (3412) (1342) (3214)

2, m=4

11

slide-13
SLIDE 13

Problem

  • How many ranking patterns are possible by

changing µ1, . . . , µm?

  • When n ≥ m − 1, we have

RPUF(µ1, . . . , µm) = Pm (the set of all permutations of {1, . . . , m}).

  • We are interested in the case 1 ≤ n ≤ m − 2.
  • We are also interested in the case where we

ignore difference by a permutation of objects (“inequivalent” ranking patterns). 12

slide-14
SLIDE 14

Problem of ranking patterns

  • Thrall (1952), Michigan Math. J. 1, 81–88:

n = 1; upper bound; Young tableaux.

  • Kamiya, Orlik, Takemura and Terao (2006),
  • Ann. Comb. 10, 219–235:

n = 1; exact number; mid-hyperplane arrangement.

  • This talk:

n = m − 2; exact number; inequivalent ranking patterns. 13

slide-15
SLIDE 15

Table of Contents

  • 1. Introduction
  • 2. Unfolding as a braid slice
  • 3. All braid slices
  • 4. Unrealizable braid slices
  • 5. Number of ranking patterns
  • 6. Inequivalent ranking patterns
  • 7. Concluding remarks

14

slide-16
SLIDE 16

2 Unfolding as a braid slice

  • Given µ1, . . . , µm ∈ Rn (not necessarily

n = m − 2).

  • We will see:

RPUF(µ1, . . . , µm) can be obtained by slicing the “braid arrangement” Bm by an affine subspace K (a “braid slice”). 15

slide-17
SLIDE 17

Affine subspace K

  • Assume µ1, . . . , µm are normalized so that

∑m

i=1 µi = 0, (1/m) ∑m i=1 ∥µi∥2 = 1.

  • Let

W :=     µT

1

. . . µT

m

    , u := − 1 2     ∥µ1∥2 − 1 . . . ∥µm∥2 − 1     ,

and define K := u + col W . 16

slide-18
SLIDE 18

u colW K = u + colW V: x1+...+xm = 0

17

slide-19
SLIDE 19

K ⊂ V

  • Because of the normalization, we have

K ⊂ V := {(x1, . . . , xm)T ∈ Rm : x1 + · · · + xm = 0} ≃ Rm−1.

  • All of our discussions will be in V .

18

slide-20
SLIDE 20

Braid arrangement Bm in V

  • Next, consider the braid arrangement Bm in

V : Bm := {Hij : 1 ≤ i < j ≤ m}, Hij := {(x1, . . . , xm)T ∈ V : xi = xj} ⊂ V. 19

slide-21
SLIDE 21

x2 = x1 x2 = x3 x3 = x1 V: x1+x2+x3 =0

3

m=3:

20

slide-22
SLIDE 22

Chambers of Bm

  • Then, Bm cuts V into m! regions (called

chambers): Ci1···im = {(x1, . . . , xm)T ∈ V : xi1 > · · · > xim} ⊂ V, (i1 · · · im) ∈ Pm. 21

slide-23
SLIDE 23

C123 C213 C132 C231 C312 C321 x2 = x1 x2 = x3 x3 = x1 x1>x2>x3 V: x1+x2+x3 =0

3

m=3:

22

slide-24
SLIDE 24

Ci1···im intersecting K

  • We can show: for each (i1 · · · im) ∈ Pm,

(i1 · · · im) is admissible ⇐ ⇒ K ∩ Ci1···im ̸= ∅.

  • Let’s see a simple example.

23

slide-25
SLIDE 25

Example: braid slice (m = 3, n = 1)

  • m = 3, n = 1:

µ1= −3/ 14/3

1

√ µ2= 1/ 14/3 √ µ3= 2/ 14/3 √

  • The ranking pattern is

1 2 2 3 1 3

(123) (213) (321) (231)

µ1 µ3 µ2

1

RPUF(µ1, µ2, µ3) = {(123), (213), (231), (321)}.

24

slide-26
SLIDE 26

Example: braid slice (m = 3, n = 1)

  • K in this case is

K =   −1 2   µ2

1 − 1

µ2

2 − 1

µ2

3 − 1

  + t   µ1 µ2 µ3   : t ∈ R    =    1 28   −13 11 2   + t   −3 1 2   : t ∈ R    ⊂ V. 25

slide-27
SLIDE 27

C123 C213 C132 C231 C312 C321 x2 = x1 x2 = x3 x3 = x1 x1>x2>x3 V: x1+x2+x3 =0

3

m=3:

26

slide-28
SLIDE 28

C123 C213 C132 C231 C312 C321 x2 = x1 x2 = x3 x3 = x1 K V: x1+x2+x3 =0

3

m=3:

27

slide-29
SLIDE 29

C123 C213 C132 C231 C312 C321 V: x1+x2+x3 =0 x2 = x1 x2 = x3 x3 = x1

3

K

n=1, m=3:

1 2 2 3 1 3

(123) (213) (321) (231)

µ1 µ3 µ2

1

28

slide-30
SLIDE 30

Proof

  • The proof is quite simple:

∥y − µi1∥ < · · · < ∥y − µim∥, ∃y ∈ Rn ⇐ ⇒ µT

i1y − 1

2(∥µi1∥2 − 1) > · · · > µT

imy − 1

2(∥µim∥2 − 1), ∃y ∈ Rn ⇐ ⇒ W y + u ∈ Ci1···im, ∃y ∈ Rn ⇐ ⇒ K ∩ Ci1···im ̸= ∅. 29

slide-31
SLIDE 31

Unfolding model as a braid slice by K

  • Therefore,

Proposition: Unfolding as a braid slice ✓ ✏ Ranking patterns of unfolding models can be obtained as RPUF(µ1, . . . , µm) = {(i1 · · · im) : K ∩ Ci1···im ̸= ∅}. ✒ ✑ 30

slide-32
SLIDE 32

Unfolding model as a braid slice by K

  • In this way, RPUF(µ1, . . . , µm) can be
  • btained by “slicing” Bm by K.

C123 C213 C132 C231 C312 C321 V: x1+x2+x3 =0 x2 = x1 x2 = x3 x3 = x1

3

K

n=1, m=3:

1 2 2 3 1 3

(123) (213) (321) (231)

µ1 µ3 µ2

1

31

slide-33
SLIDE 33

Codimension one: n = m − 2

  • Hereafter, we will assume n = m − 2, and

that µ1, . . . , µm are “generic.”

  • Then, dim K = n = m − 2 = dim V − 1.
  • Hence, in this case, we will say the unfolding

model is of codimension one.

  • We can also see 0 /

∈ K.

  • Thus,

K is an affine hyperplane in V . 32

slide-34
SLIDE 34

Arbitrary affine hyperplane K˜

v

  • Let’s forget the unfolding model for a

moment.

  • In general, an arbitrary affine hyperplane in V

can be indexed by its normal vector ˜ v ∈ V : K˜

v := {x ∈ V : ˜

vT (x − ˜ v) = 0}, ˜ v ̸= 0. 33

slide-35
SLIDE 35

V

v ~ Kv

~

34

slide-36
SLIDE 36

Braid slice RP(˜ v) by K˜

v

  • Let’s consider slicing Bm by K˜

v:

RP(˜ v) := {(i1 · · · im) : K˜

v ∩ Ci1···im ̸= ∅}.

  • We call RP(˜

v) (the ranking pattern of) the braid slice by K˜

v.

35

slide-37
SLIDE 37

C123 C213 C132 C231 C312 C321 Kv

~

v ~ RP(v) = {(213), (231), (321), (312)} ~

36

slide-38
SLIDE 38

K = Kv of unfolding model

  • Getting back to the unfolding model, we can

write K = u + col W as K = Kv with v := u − projcol W (u) = { Im − W (W T W )−1W T } u. 37

slide-39
SLIDE 39

v u colW K = u + colW = Kv V: x1+...+xm = 0

38

slide-40
SLIDE 40

RPUF(µ1, . . . , µm) = RP(v(µ1, . . . , µm))

  • v is a function of µ1, . . . , µm:

v = v(µ1, . . . , µm).

  • We can write RPUF(µ1, . . . , µm) as

RPUF(µ1, . . . , µm) = {(i1 · · · im) : Kv(µ1,...,µm) ∩ Ci1···im ̸= ∅} = RP(v(µ1, . . . , µm)). 39

slide-41
SLIDE 41

Stability of RP(v(µ1, . . . , µm))

  • m-tuple (µ1, . . . , µm) changes just a little,

→ v = v(µ1, . . . , µm) and thus Kv change just a little, → the braid slice by Kv remains exactly the same.

  • (µ1, . . . , µm) changes beyond a certain

extent, → the braid slice changes. 40

slide-42
SLIDE 42

v D Kv

C213 C231 C312 C321

v’ D’ Kv’

C123 C213 C231 C321

41

slide-43
SLIDE 43

Stability of RP(˜ v), ˜ v ̸= 0

  • Let’s investigate the stability of the braid

slices RP(˜ v), ˜ v ̸= 0, in general. 42

slide-44
SLIDE 44

Table of Contents

  • 1. Introduction
  • 2. Unfolding as a braid slice
  • 3. All braid slices
  • 4. Unrealizable braid slices
  • 5. Number of ranking patterns
  • 6. Inequivalent ranking patterns
  • 7. Concluding remarks

43

slide-45
SLIDE 45

3 All braid slices

  • We will investigate the stability of the braid

slices RP(˜ v), ˜ v ̸= 0, and find all (different) braid slices. 44

slide-46
SLIDE 46

All-subset arrangement Am

  • First, define the all-subset arrangement Am

in V by Am := {HI : I ⊂ {1, . . . , m}, 1 ≤ |I| ≤ m − 1}, HI := {(x1, . . . , xm)T ∈ V : ∑

i∈I

xi = 0} ⊂ V. 45

slide-47
SLIDE 47

x2 = 0 ( x1+x3 = 0) H{2} = H{1,3} x1 = 0 ( x2+x3 = 0) H{1} = H{2,3} x3 = 0 ( x1+x2 = 0) H{3} = H{1,2} V: x1+x2+x3 =0

m=3:

3

46

slide-48
SLIDE 48

Chambers of Am

  • Let Ch(Am) := {chambers D of Am}.

D

47

slide-49
SLIDE 49

Generic braid slices

  • We only consider RP(˜

v) for “regular” ˜ v: ˜ v ∈ V \ ∪

H∈Am

H = ⊔

D∈Ch(Am)

D.

  • We will say the braid slices are “generic” in

this case. 48

slide-50
SLIDE 50

RPD := RP(˜ v), ∀˜ v ∈ D

  • For ˜

v’s in a chamber D ∈ Ch(Am), RP(˜ v) are the same: RP(˜ v) : the same for all ˜ v ∈ D. We can define RPD := RP(˜ v), ˜ v ∈ D.

  • For different chambers D and D′, RP(˜

v) changes: RPD ̸= RPD′ for D ̸= D′. 49

slide-51
SLIDE 51

~

v D

~

Kv

C213 C231 C312 C321

v’

~

D’

~

Kv’

C123 C213 C231 C321

50

slide-52
SLIDE 52

{chambers of Am} ↔ {generic braid slices}

  • Therefore,

Proposition: {D} ↔ {RPD} ✓ ✏ We have a bijection Ch(Am) → {RPD : D ∈ Ch(Am)}, D → RPD = RP(˜ v), ˜ v ∈ D. ✒ ✑ 51

slide-53
SLIDE 53

v(µ1, . . . , µm) ∈ V \ ∪

H∈Am H

  • Let’s get back to the unfolding model.
  • We can show

v(µ1, . . . , µm) ∈ V \ ∪

H∈Am

H = ⊔

D∈Ch(Am)

D.

D v(µ1,..., µm)

52

slide-54
SLIDE 54

RP(v(µ1, . . . , µm)): generic

  • Thus,

RPUF(µ1, . . . , µm) = RP(v(µ1, . . . , µm)) is a generic braid slice: RPUF(µ1, . . . , µm) = RPD for D ∋ v(µ1, . . . , µm). 53

slide-55
SLIDE 55

D

v(µ1,..., µm)

54

slide-56
SLIDE 56

D not attained by v( · , . . . , · )

  • However, not all D ∈ Ch(Am) can be

attained by v( · , . . . , · ).

  • Hence,

{r.p.’s of unfoldings} {generic braid slices}.

  • Not all generic braid slices can be realized by

the unfolding model. 55

slide-57
SLIDE 57

C123 C213 C132 C231 C312 C321 Kv

~

v ~ RP(v) = {(213), (231), (321), (312)} ~

56

slide-58
SLIDE 58

D not attained by v( · , . . . , · )

  • We have to identify Ds which can/cannot be

attained by v( · , . . . , · ). 57

slide-59
SLIDE 59

Table of Contents

  • 1. Introduction
  • 2. Unfolding as a braid slice
  • 3. All braid slices
  • 4. Unrealizable braid slices
  • 5. Number of ranking patterns
  • 6. Inequivalent ranking patterns
  • 7. Concluding remarks

58

slide-60
SLIDE 60

4 Unrealizable braid slices

  • We will identify D ∈ Ch(Am) which cannot

be attained by v( · , . . . , · ), and thereby identify unrealizable braid slices RPD. 59

slide-61
SLIDE 61

Image of v( · , . . . , · )

  • We can show im v := {v(µ1, . . . , µm) :

generic µ1, . . . , µm ∈ Rm−2} is given by im v = ⊔

D∈Ch(Am), D̸=−D1,...,−Dm

D where −D1, . . . , −Dm ∈ Ch(Am) are −Di := {v ∈ V : vi < 0, vj > 0 for all j ̸= i}. 60

slide-62
SLIDE 62

−D1 −D3 −D2 (x1 < 0, x2,x3 > 0) (x2 < 0, x1,x3 > 0) (x3 < 0, x1,x2 > 0)

x2 = 0 ( x1+x3 = 0) H{2} = H{1,3} x1 = 0 ( x2+x3 = 0) H{1} = H{2,3} x3 = 0 ( x1+x2 = 0) H{3} = H{1,2} V: x1+x2+x3 =0

m=3:

3

61

slide-63
SLIDE 63

Unrealizable braid slices

  • Therefore,

Theorem: Unrealizable braid slices ✓ ✏ RPD cannot be realized by the unfolding model if and only if D = −Di for some i = 1, . . . , m. ✒ ✑ 62

slide-64
SLIDE 64

Unrealizable braid slices

  • That is, exactly

RP−D1, . . . , RP−Dm are unrealizable braid slices. 63

slide-65
SLIDE 65

−D1 −D3 −D2 (x1 < 0, x2,x3 > 0) (x2 < 0, x1,x3 > 0) (x3 < 0, x1,x2 > 0)

x2 = 0 ( x1+x3 = 0) H{2} = H{1,3} x1 = 0 ( x2+x3 = 0) H{1} = H{2,3} x3 = 0 ( x1+x2 = 0) H{3} = H{1,2} V: x1+x2+x3 =0

m=3:

3

64

slide-66
SLIDE 66

C123 C213 C132 C231 C312 C321 Kv

~

v ~ RP(v) = {(213), (231), (321), (312)} ~

65

slide-67
SLIDE 67

Table of Contents

  • 1. Introduction
  • 2. Unfolding as a braid slice
  • 3. All braid slices
  • 4. Unrealizable braid slices
  • 5. Number of ranking patterns
  • 6. Inequivalent ranking patterns
  • 7. Concluding remarks

66

slide-68
SLIDE 68

5 Number of ranking patterns

  • So far, we have seen

{r.p.’s of unfoldings} {generic braid slices} ↕ {chambers of Am} and {generic braid slices} \ {r.p.’s of unfoldings} = {RP−D1, . . . , RP−Dm}. 67

slide-69
SLIDE 69

Main result 1 Theorem: Number of r.p.’s ✓ ✏ The number q(m) of ranking patterns of unfolding models of codimension one with m objects is q(m) = |{chambers of Am}| − m. ✒ ✑ 68

slide-70
SLIDE 70

q(m), m ≤ 8 Corollary: q(m) for m ≤ 8 ✓ ✏ For m ≤ 8, we have q(3) = 3, q(4) = 28, q(5) = 365, q(6) = 11286, q(7) = 1066037, q(8) = 347326344. ✒ ✑ 69

slide-71
SLIDE 71

Table of Contents

  • 1. Introduction
  • 2. Unfolding as a braid slice
  • 3. All braid slices
  • 4. Unrealizable braid slices
  • 5. Number of ranking patterns
  • 6. Inequivalent ranking patterns
  • 7. Concluding remarks

70

slide-72
SLIDE 72

6 Inequivalent ranking patterns

  • Let’s consider the case where we ignore

difference by a permutation of objects.

  • We consider two ranking patterns are

essentially the same if they are just the relabelings of the objects of each other.

  • What is the number of essentially different

ranking patterns of unfolding models of codimension one? 71

slide-73
SLIDE 73

Equivalence of ranking patterns

  • Let Sm be the symmetric group on m letters

(all bijections σ : {1, . . . , m} → {1, . . . , m}).

  • We will say RPD and RPD′ are equivalent:

RPD ∼ RPD′ if and only if RPD = σRPD′, ∃σ ∈ Sm, where σRPD′ := {(σ(i1) · · · σ(im)) : (i1 · · · im) ∈ RPD′}. 72

slide-74
SLIDE 74

Action of Sm on Ch(Am)

  • Consider the action of Sm on Ch(Am):

Sm × Ch(Am) ∋ (σ, D) − → σD ∈ Ch(Am), where σD := {σv : v ∈ D} with σv := (vσ−1(1), . . . , vσ−1(m))T for v = (v1, . . . , vm)T . 73

slide-75
SLIDE 75

RPD ∼ RPD′ ⇐ ⇒ O(D) = O(D′)

  • We can check

RPσD = σRPD.

  • Thus,

RPD ∼ RPD′ ⇐ ⇒ D = σD′, ∃σ ∈ Sm (i.e., O(D) = O(D′)), where O(D) := {σD : σ ∈ Sm} is the Sm-orbit containing D. 74

slide-76
SLIDE 76

RPD ∼ RPD′ ⇐ ⇒ O(D) = O(D′)

  • Therefore, the number of inequivalent braid

slices RPD, D ∈ Ch(Am), is equal to |{Sm-orbits of Ch(Am)}|. 75

slide-77
SLIDE 77

Orbit of excluded chambers

  • On the other hand, the excluded m chambers

−D1, . . . , −Dm constitute one orbit O(−Dm) = {−D1, . . . , −Dm}. 76

slide-78
SLIDE 78

Number of inequivalent ranking patterns

  • Therefore,

Proposition: # of inequivalent r.p.’s ✓ ✏ The number of inequivalent ranking pat- terns of unfolding models of codimension

  • ne is

|{Sm-orbits of Ch(Am)}| − 1. ✒ ✑ 77

slide-79
SLIDE 79

|{Sm-orbits of Ch(Am)}|

  • Our job is to count

|{Sm-orbits of Ch(Am)}|. 78

slide-80
SLIDE 80

{F ∈ Ch(Am ∪ Bm) : F ⊂ C1···m}

  • There is a one-to-one correspondence

{F ∈ Ch(Am ∪ Bm) : F ⊂ C1···m} ← → {Sm-orbits of Ch(Am)} given by F − → O(DF ), where DF := unique D ∈ Ch(Am) such that F ⊂ D for F ∈ Ch(Am ∪ Bm). 79

slide-81
SLIDE 81

F ↔ O(DF ), m = 3

x2 = x1 x2 = x3 x3= x1 x2 =0 (x1+x3 = 0) x1 =0 (x2+x3 = 0) x3 =0 (x1+x2 = 0)

3 3

m =3:

F F ’

F

F C123

x2 = x1 x2 = x3 x3 = x1 x2 =0 (x1+x3 = 0) x1 =0 (x2+x3 = 0) x3 =0 (x1+x2 = 0)

3 3

x3 =0 (

m =3:

DF (DF) | (DF)|=3

{Sm-orbits of Ch(Am)} = {O(DF ), O(DF ′)}. 80

slide-82
SLIDE 82

|{Sm-orbits of Ch(Am)}|

  • Since |Ch(Bm)| = |Sm| = m!, we obtain

|{Sm-orbits of Ch(Am)}| = |{F ∈ Ch(Am ∪ Bm) : F ⊂ C1···m}| = |Ch(Am ∪ Bm)| m! . 81

slide-83
SLIDE 83

Main result 2 Theorem: # of inequivalent r.p.’s ✓ ✏ The number qIE(m) of inequivalent rank- ing patterns of unfolding models of codi- mension one is qIE(m) = |Ch(Am ∪ Bm)| m! − 1. ✒ ✑ 82

slide-84
SLIDE 84

qIE(m), m ≤ 9 Corollary: qIE(m) for m ≤ 9 ✓ ✏ For m ≤ 9, we have qIE(3) = 1, qIE(4) = 3, qIE(5) = 11, qIE(6) = 55, qIE(7) = 575, qIE(8) = 16639, qIE(9) = 1681099. ✒ ✑ 83

slide-85
SLIDE 85

Illustration: m = 4

  • We illustrate our results when m = 4.

84

slide-86
SLIDE 86

F, F ′, ˜ F, ˜ F ′ ⊂ C1234

  • Chamber

C1234 : x1 > x2 > x3 > x4

  • f B4 contains exactly 4 chambers

F, F ′, ˜ F, ˜ F ′ ∈ Ch(A4 ∪ B4). 85

slide-87
SLIDE 87

F, F ′, ˜ F, ˜ F ′ ⊂ C1234

C1234: x1 > x2 > x3 > x4

x2>x4 x1+x4>0 x1>x2 x1>x4 x1>x3 x3>x4 x2>x3 x1+x3>0 x1+x2>0 x1>0 x4>0 x2>0 x3>0 F ’ F ~ F ’ F ~ x2>x4 x1>x2 x1>x4 x1>x3 x3>x4 x2>x3

C1234

F ~ ’ F F ’ F ~ , , , ⊂ C1234

4 4 4

(intersection with the unit sphere S2 in V ≃ R3)

86

slide-88
SLIDE 88

x2>x4 x1+x4>0 x1>x2 x1>x4 x1>x3 x3>x4 x2>x3 x1+x3>0 x1+x2>0 x1>0 x4>0 x2>0 x3>0 F ’ F ~ F ’ F ~

A4 ∪ B4 (intersection with the unit sphere S2 in V ≃ R3)

87

slide-89
SLIDE 89

DF , DF ′, D ˜

F , D ˜ F ′ ∈ Ch(A4)

  • The chambers of A4 containing F, F ′, ˜

F, ˜ F ′ are DF = −D4 : x1 > 0, x2 > 0, x3 > 0, DF ′ = D1 : x4 < 0, x3 < 0, x2 < 0, D ˜

F = E : x2 + x3 > 0, x1 + x3 > 0, x3 < 0,

D ˜

F ′ = E′ : x3 + x2 < 0, x4 + x2 < 0, x2 > 0.

88

slide-90
SLIDE 90

DF , DF ′, D ˜

F , D ˜ F ′ ∈ Ch(A4)

x1+x3>0 x3<0 x2+x3>0 x4+x2<0 x3>0 x1>0 x2>0 F

~ D =E

F

~

F D = -D4

F

x

4

< x

2

< x

3

<

F ’

F’

D = D1 D =E’

F ’

~

F ’

~

x2>0 x3+x2<0

89

slide-91
SLIDE 91

Cross section

  • These DF , DF ′, D ˜

F , D ˜ F ′ constitute a

complete set of representatives of the orbits:

{Sm-orbits of Ch(Am)} = {O(DF ), O(DF ′), O(D ˜

F ), O(D ˜ F ′)}.

90

slide-92
SLIDE 92

Four inequivalent braid slices

  • Four inequivalent braid slices:

RPDF = RP−D4, RPDF ′ = RPD1, RPD ˜

F = RPE, RPD ˜ F ′ = RPE′.

– RP−D4: not realizable. – RPD1, RPE, RPE′: realizable. 91

slide-93
SLIDE 93

Three inequivalent r.p.’s of unfoldings

  • There are exactly

three inequivalent ranking patterns RPD1, RPE, RPE′

  • f unfolding models.

92

slide-94
SLIDE 94

Unrealizable RP−D4

  • Unrealizable braid slice:

RP−D4 = P4 \ {(4123), (4132), (4213), (4231), (4312), (4321)}. (“object 4 is never ranked first”)

93

slide-95
SLIDE 95

Realizable RPD1, RPE, RPE′

  • Realizable braid slices:

RPD1 = P4 \ {(2341), (2431), (3241), (3421), (4231), (4321)}, (“object 1 is never ranked last”) RPE = P4 \ {(3412), (3421), (4312), (4321), (4132), (4231)}, RPE′ = P4 \ {(3412), (4312), (3421), (4321), (3241), (4231)}.

94

slide-96
SLIDE 96

95

slide-97
SLIDE 97

Realizing unfolding models

  • The unfolding models realizing

RPD1, RPE, RPE′ are illustrated in the following figures (µ1, µ2, µ3, µ4 are written as 1, 2, 3, 4 in the figures). 96

slide-98
SLIDE 98

RPD1

4 1 2 3

2 2 2 3 3 3 1 1 1 4 4 4

Inadmissible rankings: (2341) (2431) (3241) (3421) (4231) (4321) (4132) (1432) (4312) (3412) (3142) (3124) (3214) (2314) (1342) (1234) (2134) (1243) (2413) (4213) (4123) (1423) (2143) (1324)

97

slide-99
SLIDE 99

RPE

2 2 2 4 4 4 3 3 3 1 1 1

3 2 1 4

(1423) (1432) (1342) (3124) (3142) (3214) (3241) (2341) (2431) (2413) (4213) (4123) (1243) (1324) (1234) (2314) (2134) (2143) Inadmissible rankings: (3412) (3421) (4312) (4321) (4132) (4231)

98

slide-100
SLIDE 100

RPE′

4 1 2 3

Inadmissible rankings: (3412) (4312) (3421) (4321) (3241) (4231)

1 1 1 3 3 3 4 4 4 2 2 2

(4132) (1432) (2431) (2341) (3142) (3124) (3214) (2314) (1342) (1234) (2134) (1243) (2413) (4213) (4123) (1423) (2143) (1324)

99

slide-101
SLIDE 101

Table of Contents

  • 1. Introduction
  • 2. Unfolding as a braid slice
  • 3. All braid slices
  • 4. Unrealizable braid slices
  • 5. Number of ranking patterns
  • 6. Inequivalent ranking patterns
  • 7. Concluding remarks

100

slide-102
SLIDE 102

7 Concluding remarks

  • We found the number of ranking patterns of

unfolding models when n = m − 2.

  • Open problem: What is the number of

ranking patterns when 1 < n < m − 2? 101

slide-103
SLIDE 103

Acknowledgment: H. Kamiya was partially supported by JSPS KAKENHI Grant Numbers 22540134, 25400201. 102

slide-104
SLIDE 104

Thank you.

103

slide-105
SLIDE 105

Notes

  • ˜

F ′ is obtained from ˜ F by changing (x1, . . . , x4) → (−x4, . . . , −x1).

  • RPE′ can be obtained from RPE by

reversing the indices : i → 5 − i (i = 1, . . . , 4) and taking the opposite rankings : (i1i2i3i4) → (i4i3i2 104

slide-106
SLIDE 106
  • RPE and RPE′ are not equivalent.
  • The same is true for F and F ′ (E and E′).

105

slide-107
SLIDE 107
  • Orbit sizes are

(|O(−D4)| = 4) |O(D1)| = 4, |O(E)| = |O(E′)| = (4 3 ) × 3 = 12, so q(4) = 28 (mentioned ealier) can also be confirmed by q(4) = 4 + 2 × 12 = 28. 106

slide-108
SLIDE 108
  • Good and Tideman (1977), J. Combin.

Theory A 23, 34–45;

  • Kamiya and Takemura (1997), J. Multivariate
  • Anal. 61,1–28;
  • Zaslavsky (2002), Discrete Comput. Geom.

27, 303–351. 107