Continuous Degenerations

Whenp is not covering at least two elements, the tropical linear formα_p has just one term and defines an empty tropical hyperplane since the maximum can never be achieved twice. Hence, letRdenote the set of allp ∈P˜ covering at least two elements and define a tropical hyperplane arrangementH (P,λ)inR^P with tropical hyperplanesH_p = H_αp for allp ∈R. By construction, the facets ofT (H (P,λ))are the linearity regions ofφ_t for t ∈ (0, 1]^P˜.

The reason this subdivision will help study the combinatorics of marked poset polytopes is the following: by Proposition7.5.2the combinatorics ofO_t(P,λ)can be determined by pulling points back toO(P,λ)and looking at the relation⊣_x. But forr ∈Randp ∈P we havep ⊣_x r if and only ifp ∈ tc(x)_r, so the information encoded in⊣_x is equivalent to knowing the minimal cell ofT (H (P,λ))containingx.

Using this tropical hyperplane arrangement, we can define a polyhedral subdivision of O(P,λ).

Definition 7.6.2. LetT (H (P,λ))be the polyhedral subdivision ofR^P associated to the marked poset(P,λ). Thetropical subdivisionT (P,λ)ofO(P,λ)is given by the intersection of faces ofO(P,λ)with the faces ofT (H (P,λ)):

T (P,λ)= {F ∩G|F ∈ F (O(P,λ)),G ∈ T (H (P,λ)) }. Fort ∈ [0, 1]^P˜ define the tropical subdivision ofO_t(P,λ)as

T_t(P,λ)= {φ_t(Q) |Q ∈ T (P,λ)}.

Note thatT_t(P,λ)is polyhedral subdivision ofO_t(P,λ)sinceφ_t is linear on eachG ∈ T (P,λ)by construction. In particular,T_t(P,λ)is a coarsening of the subdivisionS_t into products of simplices and simplicial cones.

7.7. Continuous Degenerations i) A continuous mapρ: Q0× [0, 1] → Rⁿ, such that eachρ_t = ρ(−,t)is an embedding,

ρ0is the identical embedding ofQ0 and the image ofρ1 isQ1.

ii) Finitely many continuous functions f¹,f²,. . .,f^k: Rⁿ× [0, 1] →Rsuch that for alliandt the maps f_tⁱ = fⁱ(−,t): Rⁿ →Rare affine linear forms and satisfy

ρ_t(Q0)= {x ∈Rⁿ

f_tⁱ(x) ≥0 for alli}.

Hence, the imagesρ_t(Q0)are all polyhedra and we writeQ_t forρ_t(Q0)and say(Q_t)_t∈[0,1]

is a continuous deformation when the accompanying mapsρand fⁱ are clear from the context.

Note that a continuous deformation of polyhedra as defined here consists of both a map moving the points aroundand a continuous description in terms of inequalities for allt ∈ [0, 1].

Definition 7.7.2. A continuous deformation(Q_t)_t∈[0,1]as in Definition7.7.1is called a continuous degenerationif for allx ∈Q0,t < 1 andi =1,. . .,k we have f_tⁱ(ρ_t(x))=0 if and only if f₀ⁱ(x)= 0.

From this definition we immediately obtain the following.

Proposition 7.7.3. If (Q_t)_t∈[0,1] is a continuous degeneration, the polyhedraQ_t fort < 1 are all combinatorially equivalent and ρ_t preserves faces and their incidence structure.

Proof. Let the data of the continuous degeneration be given as in Definition7.7.1. For y ∈Q_t denote byI_t(y)the set of alli ∈ [k]such that f_tⁱ(y)= 0. The set of allI_t(y)for y ∈Q_t ordered by reverse inclusion is isomorphic toF (Q_t) \ {œ}since relative interiors of faces ofQ_t correspond to regions of constantI_t.

Since for allx ∈Q0,t < 1 andi =1,. . .,kwe havef_tⁱ(ρ_t(x))=0 if and only iff₀ⁱ(x)=0, the setsI_t(ρ_t(x))are fixed fort < 1 and henceρ_t preserves the face structure.

We continue by illustrating the definition of continuous degenerations in an example before proceeding with the general theory.

Example 7.7.4. Fort ∈ [0, 1]letQ_t ⊆ R² be the polytope defined by the inequalities 0 ≤ x1 ≤ 2, 0 ≤x2as well as

x2 ≤ (1−t)x1+1, and x2 ≤ (1−t)(2−x1)+1.

Fort =0,t = ¹2

andt =1 we have illustrated the polytope in Figure7.3. Together with the mapρ_t: Q0 →R²given byρ_t(x)1 =x1 for allt and

ρ_t(x)2 = {x2

(1−t)x¹+1 x¹+1

forx1 ≤ 1, x2

(1−t)(2−x¹)+1

(2−x¹)+1 forx1 ≥ 1

x1

x2

1 2

1 2

(a)Q0

x1

x2

1 2

1 2

(b)Q¹

2

x1

x2

1 2

1 2

(c)Q1

Figure 7.3.: The polytopes in the continuous degeneration from Example7.7.4fort =0, t = ¹2 andt = 1.

we obtain a continuous degeneration. Starting from the pentagon in Figure7.3aatt =0 we see increasingly compressed pentagons with the two top edges becoming more flat-angled until ending up with the rectangle in Figure7.3catt =1. The mapρ_t just scales thex2 coordinates accordingly, preserving the face-structure fort <1. ♢ The key result on continuous degenerations that will allow conclusions on face struc-ture of degenerations is that during a continuous degeneration, relative interiors of faces always map into relative interiors of faces. In other words, continuous degenerations can not “fold” faces ofQ0 so they split into different faces ofQ1, but only “straighten” some adjacent faces ofQ⁰to become one face ofQ¹.

Proposition 7.7.5. Let(Q_t)_t∈[0,1]be a continuous degeneration of polyhedra. WheneverF is a face of Q0, there is a unique faceGof Q1 such that

ρ1(relintF) ⊆ relintG.

Proof. As in the previous proof, let I_t(y) denote the set of indicesi ∈ [k] such that f_tⁱ(y) = 0. Using these incidence sets we may rephrase the proposition as follows:

wheneverx,x^′∈Q0satisfyI0(x)= I0(x), they also satisfyI1(ρ1(x))=I1(ρ1(x^′)). LetF be the face ofQ0having bothxandx^′in its relative interior and assume there exists aj ∈I1(ρ1(x)) \I1(ρ1(x^′))for sake of contradiction. Hence, we havef₁^j(ρ1(x))=0 while f₁^j(ρ1(x^′)) > 0. Letd denote the dimension of F then relintF is a manifold of dimensiond. Sinceρ1 is an embedding, ρ1(relintF) is a manifold of dimensiond as well. Since the affine hull of ρ_t(relintF)is of dimensiond for allt < 1, we conclude that the affine hull ofρ1(relintF)is of dimension at mostd. To see this, take anyd+1 pointsy0,. . .,y_d inρ1(relintF). Their imagesρ_t(ρ⁻¹₁ (y0)),. . .,ρ_t(ρ₁⁻¹(y_d))inρ_t(relintF) are affinely dependent fort <1, so they have to be affinely dependent fort = 1 as well by the continuity ofρint.

But asρ¹(relintF)is a manifold of dimensiond, we conclude that its affine hull has dimension exactlyd andρ1(relintF)is an open subset of its affine hull. Given that both ρ1(x)andρ2(x^′)are points in ρ1(relintF), we conclude that there exists anε > 0 such

7.7. Continuous Degenerations that the point

z= ρ1(x)+ε(ρ1(x) −ρ1(x^′))

is still contained inρ1(relintF). In particular,z ∈Q1. However, since f₁^j is an affine linear form, we have

f₁^j(z)= (1+ε)f₁^j(ρ1(x)) −ε f₁^j(ρ1(x^′))< 0. This contradictsz ∈Q1, which finishes the proof.

The consequence of Proposition7.7.5is that continuous degenerations induce maps between face lattices.

Corollary 7.7.6. When (Q_t)_t∈[0,1] is a continuous degeneration of polyhedra, we have a surjective order-preserving map of face lattices

dg : F (Q0) −→ F (Q1) determined by the property

ρ1(relintF) ⊆relint dg(F).

for non-empty F anddg(œ) = œ. Furthermore, the map satisfies dim(dg(F)) ≥ dimF for all F ∈ F (Q0).

We will refer to the map in Corollary7.7.6as thedegeneration map. Before coming back to marked poset polyhedra, we finish with a result on the f-vectors of continuous degenerations.

Proposition 7.7.7. Let (Q_t)_t∈[0,1] be a continuous degeneration of polyhedra. We have f_i(Q1) ≤ f_i(Q0)for alli.

Proof. Let G be an i-dimensional face of Q¹. We claim that there is at least one i -dimensional face F ofQ0 such that dg(F) = G. Since every polyhedron is the disjoint union of the relative interiors of its faces andρ1is a bijection, we have

relintG = ⨆

F∈dg⁻¹(G)

ρ1(relintF).

Since relintG is a manifold of dimension dimG and each ρ¹(relintF)is a manifold of dimension dimF ≤ dimG, there has to be at least oneF ∈dg⁻¹(G)of the same dimension asG.

7.7.2. Continuous Degenerations in the Continuous Family

We are now ready to apply the concept of continuous degenerations to the continuous family of marked poset polyhedra. Let us first identify for which pairs of parameters u,u^′∈ [0, 1]^P˜ we expect to have a continuous degeneration fromO_u(P,λ)toO_u^′(P,λ)and then specify the deformation precisely.

Definition 7.7.8. Letu ∈ [0, 1]^P˜and letI ⊆P˜be the set of indicesp, such thatu_p ∈ {0, 1}. Anyu^′∈ [0, 1]^P˜ such thatu_p^′ =u_p forp ∈I is called adegenerationofu.

Proposition 7.7.9. Letu^′be a degeneration of u. The map ρ: O_u(P,λ) × [0, 1] −→ R^P,

(x,ξ) ↦−→θ_u,ξu^′+(1−ξ)u(x)

is a continuous degeneration with the accompanying affine linear forms given by the equa-tions and inequalities in Definition7.1.1fort =ξu^′+(1−ξ)u.

Proof. The map ρ together with the affine linear forms given by Definition 7.1.1is a continuous deformation by Theorem7.2.1. The fact thatρis a continuous degeneration follows from Proposition7.5.2.

Now the machinery of continuous degenerations immediately yields degeneration maps and results on the f-vectors of marked poset polyhedra.

Corollary 7.7.10. Letu,u^′ ∈ [0, 1]^P˜ such thatu^′is a degeneration of u. The continuous degeneration in Proposition7.7.9yields a degeneration map dg

u,u^′: O_u(P,λ) → O_u^′(P,λ) in the sense of Corollary7.7.6. In particular, the f-vectors satisfy

f_i(O_u^′(P,λ)) ≤ f_i(O_u(P,λ)) for alli.

Furthermore, given a degenerationu^′′ of u^′, the degeneration maps satisfy dgu,u^′′ = dg

u^′,u^′′◦dg

u,u^′.

Proof. After applying Proposition 7.7.5, Corollary 7.7.6 and Proposition 7.7.7 to the situation at hand, all that remains to be proven is the statement about compositions of degeneration maps. This is an immediate consequence ofθ_u,u^′′ =θ_u^′,u^′′◦θ_u,u^′.

Im Dokument The Polyhedral Geometry of Partially Ordered Sets (Seite 92-96)