#### Weakly Regular Subdivisions

Discrete Comput Geom
Weakly Regular Subdivisions
Lionel Pournin 0
0 L. Pournin ( ) Mathematics Institute, Ecole Polytechnique Fédérale de Lausanne , 1015 Lausanne , Switzerland
It is shown that 2-dimensional subdivisions can be made regular by moving their vertices within parallel 1-dimensional spaces. As a consequence, any 2-dimensional subdivision is projected from the boundary complex of a 4-polytope. Let A be a d -dimensional point configuration, that is, a finite subset of Rd whose affine hull has full dimension. A polyhedral subdivision of A is a collection of subsets of A whose convex hulls form a polyhedral complex that covers conv(A). Some of the polyhedral subdivisions of A can be obtained by projecting the lower facets of a (d + 1)-dimensional polytope. Such polyhedral subdivisions are called regular and play an important role in many fields of engineering and mathematics. Recently, the author has generalized the notion of regularity using more general polytope projections [14]: A subdivision of a d -dimensional point configuration is k-regular if it is projected from the boundary complex of a polytope p ⊂ Rd+k . The usual notion of regularity then becomes 1-regularity. Now consider transforming a subdivision S of A by moving its vertices while preserving its topological structure. Such transformations are used in triangulationbased algorithms designed for various applications (see [6, 12, 13, 22]). Following
Polyhedral subdivisions; Triangulations; k-Regularity; Weak regularity; Steinitz's theorem; Flip-graph connectivity
1 Introduction
the terminology introduced by Lee [
8, 9
], a subdivision S is weakly regular if a
transformation of this kind results in a regular subdivision of some point configuration.
A formal definition of weak regularity is given in Sect. 2.
This paper sheds some light on the link between weakly regular subdivisions and
k-regularity. It is proven in Sect. 2 that if a subdivision can be transformed into a
regular subdivision by moving its vertices within parallel k-dimensional affine spaces,
then this subdivision is (k + 1)-regular. In Sect. 3, the case of 2-dimensional
subdivisions is addressed. A well known consequence of Steinitz’s theorem [
20
] is that
all the subdivisions of 2-dimensional point configurations are weakly regular [
1
]. It
is shown in Sect. 3 that all the subdivisions of any 2-dimensional point
configuration can be made regular by moving their vertices along parallel 1-dimensional affine
spaces. This implies that all the subdivisions of 2-dimensional point configurations
are 2-regular, which settles the 2-dimensional case of a conjecture formulated in [
14
].
In the last section, it is shown how to regularize a 2-dimensional subdivision by
explicitly performing the construction described in Sect. 3.
2 Weak Regularity and k-Regular Subdivisions
This section begins with a formal definition of k-regularity (see also [
14
]). A pair
(p, π ) is a polytope projection if p ⊂ Rn is a polytope and π : Rn → Rd is an affine
map whose restriction to the vertex set of p is injective. Let (p, π ) be a polytope
projection and denote by V the vertex set of p. Let φ be the inverse of the restriction
of π to V (i.e. φ is the unique map from π(V) onto V so that π ◦ φ is the identity).
According to the definition found in [
3, 4
], a polyhedral subdivision S is induced by
π from p if it is a subdivision of π(V) so that for all s ∈ S, the convex hull of φ (s) is a
face of p. The set of all polyhedral subdivisions induced by π from p will be denoted
by ω(p, π ) in the following of this paper. Note that ω(p, π ) can be given the structure
of a poset using the refinement relation (see [
3, 4
]). The following definition is found
in [
14
]:
Definition 1 A subdivision S is called k-regular if there exists a polytope projection
(p, π ) so that S ∈ ω(p, π ) and dim(p) − dim(π(p)) ≤ k.
In other words a polyhedral subdivision of a d-dimensional point configuration
is k-regular if it is projected from the boundary complex of a polytope p ∈ Rd+k .
According to Proposition 1 in [
14
], 1-regularity coincides with the usual notion of
regularity. Note that regularity is more commonly defined using height functions that
lift a d-dimensional point configuration to the vertex set of a polytope in Rd+1. In
the sequel, v(S) will denote the vertex set of a polyhedral subdivision S and δ(S) the
smallest integer k such that S is k-regular. A formal definition of weak regularity is
now given:
Definition 2 Let S be a subdivision. An injection ψ : v(S) → Rd regularizes S if
ψ (S) is a regular subdivision of ψ (v(S)).
The set of all injections that regularize a subdivision S will be called R(S) and S
will be called weakly regular if R(S) = ∅. In other words, a subdivision is weakly
regular if, through some motion of its vertices, it can be transformed into a regular
subdivision. For any ψ ∈ R(S), denote by ε(ψ ) the affine dimension of {a − ψ (a) :
a ∈ v(S)} and by ε(S) minimum of ε over R(S); by convention, ε(S) = +∞ if
R(S) = ∅.
Note that any subdivision is k-regular for some (large enough) integer k. Indeed,
any subdivision of a d -dimensional configuration A of n points is necessarily (n −
d − 1)-regular as any subset of A can be projected from the boundary complex of a
(n − 1)-dimensional simplex. This provides a bound on δ(S) which depends on the
size of v(S). The following theorem gives a better bound on δ(S) in the case S is
weakly regular. Note that this bound only depends on ε(S) and not on the cardinality
of v(S):
Theorem 1 For any polyhedral subdivision S, δ(S) ≤ ε(S) + 1.
Proof Let d be the affine dimension of v(S). Assume without loss of generality that
v(S) ⊂ Rd . Let ψ be a map that regularizes S and satisfies ε(ψ ) = ε(S). As ψ (S)
is regular, there exists a polytope projection (p, π ) so that p ⊂ Rd+1 and for all s ∈
ψ (S), the convex hull of φ (s) is a face of p, where φ is the inverse of the restriction
of π to the vertex set of p.
Call k = ε(S) + 1. In the remainder of the proof, Rd+1 is identified with a subspace
of Rd+k . Denote by κ the orthogonal projection from Rd+k onto Rd+1. Since ker(κ)
and the affine hull H of {a − ψ (a) : a ∈ v(S)} both have dimension k − 1, there exists
an isomorphism τ : ker(κ) → H . For all a ∈ v(S), define
φ (a) = φ ◦ ψ (a) + τ −1 a − ψ (a) .
Note that the restriction of κ to φ (v(S)) is an injection into the vertex set of p. As a
consequence, the convex hull of φ (v(S)) is a polytope p whose vertex set is exactly
φ (v(S)). Denote by ι is the orthogonal projection from Rd+k onto ker(κ) and by
π the affine map π ◦ κ + τ ◦ ι. According to this construction, the restriction of π
to φ (v(S)) is a bijection onto v(S) and φ is its inverse. The pair (p , π ) then is a
polytope projection.
Now let s be a face of S. There exists an affine map χ : Rd+1 → R so that conv(φ ◦
ψ (s)) is the face of p on which χ is maximal. It follows from the construction of p
that conv(φ (s)) is the subset of p that maximizes the affine map χ ◦ κ . This proves
that conv(φ (s)) is a face of p and that S is induced by π from p . As p ⊂ Rd+k
the result follows.
Let A be a d -dimensional point configuration and S a subdivision of A. If S is
weakly regular then necessarily, ε(S) ≤ d . The corollary below therefore immediately
follows from Theorem 1:
Corollary 1 Any weakly regular subdivision of a d -dimensional point configuration
is (d + 1)-regular.
The end of this section focuses on the notion of flip. Let T be a triangulation of
some point configuration A, that is, a subdivision of A whose elements are affinely
independent. Under given conditions, it is possible to transform T into another
triangulation of A performing a local operation in T called flip. This operation basically
consists in replacing the star of a simplex in T by another triangulation of this star. An
exact definition of flips is not required nor given here; the interested reader will find
formal definitions in [
14, 17–19
]. The graph whose vertices are the triangulations of
A and whose edges link two triangulations that can be transformed into one another
by a flip is called the flip-graph of A.
For practical applications, flip-graph connectivity, especially in the case of
3-dimensional point configurations, is an important issue as flips are used to build
regular triangulations from non-regular ones. It is shown in [
14
] that the subgraph
induced in the flip-graph of A by 2-regular triangulations is connected. Hence the
following result becomes a consequence of Theorem 1:
Corollary 2 Let T be a triangulation of some d -dimensional point configuration. If
ε(T ) ≤ 1 then T can be flipped to a regular triangulation.
Note that flip-graph connectivity for general 3- and 4-dimensional point
configurations remains an open question. For 2-dimensional point configurations, it was
shown by Lawson [
7
] that flip-graphs are always connected. The next section
focuses on dimension 2: it is proven that if S is a subdivision of a 2-dimensional point
configuration then ε(S) ≤ 1. This will have several consequences, among which an
alternative proof for the connectivity result in [
7
].
3 Two-Dimensional Subdivisions Are Weakly Regular
Let S be a subdivision of some 2-dimensional point configuration. The set of the
interior vertices of S will be denoted by vi (S) and the set of its interior edges by ei (S).
For any vertex a of v(S), denote by l(a, S) the set of all vertices of S adjacent to a. It
will be shown in this section that ε(S) ≤ 1. A characterization of regularity originally
formulated by Maxwell [
10, 11
] will be used to this end. This characterization links
the local convexity or concavity of a 3-dimensional polyhedral surface to the signs
of the constraints induced in its edges as they are projected into the plane. Several
results have been obtained since then on the topic (see in particular [
16, 21
]). In
this characterization, the sufficient condition for regularity does not require the set of
edges in the plane to be the 1-skeleton of a polyhedral complex: edges could a priori
cross and cells be non-convex. This is particularly convenient when displacing the
points of a polyhedral subdivision. The following result is a direct consequence of
Theorem 5.1 in [
16
]:
Theorem 2 Let S be a subdivision of some 2-dimensional point configuration and
ψ : v(S) → R2 an injection so that ψ (v(S) \ vi (S)) lies within the relative boundary
of its convex hull. Let τ : ei (S) → R be a map so that for all a ∈ vi (S),
τ {a, b} ψ (b) − ψ (a) = 0.
(1)
If τ is positive on all the interior edges of S then there exists a regular subdivision R
of ψ (v(S)) so that ψ (S) ⊂ R.
The lemma below will be useful in the proof of Theorem 3. It states sufficient
conditions under which a square matrix admits an inverse. Similar conditions for
matrix non-singularity can be found in [
2
].
Lemma 1 Let M be a square matrix whose columns sum to a non-zero, non-negative
vector. If αI − M is non-negative and (αI − M)k is positive for some α > 0 and some
k ∈ N then M admits an inverse.
Proof Call n the number of lines of matrix M and denote A = I − α−1M . For any
i ∈ N, denote by yi the vector obtained by summing the columns of Ai . According to
the hypotheses, A is a non-negative matrix and so are its successive powers. Hence yi
is non-negative for all i ∈ N. Moreover, as the columns of M sum to a non-negative
vector, then all the coefficients of y1 are in [
0, 1
].
Let i be a positive integer. By writing the coefficients of Ai+1 as scalar products
of the rows of Ai with the columns of A, one finds that yi+1 = Ai y1. In other words,
the coefficients of yi+1 are the scalar product of the rows of Ai with vector y1. As
both Ai and y1 are non-negative and as 1 is an upper bound on the coefficient of y1,
one then obtains yi+1 ≤ yi .
Using this last inequality recursively, one shows in particular that for all i ∈ N,
all the coefficients of yi are in [
0, 1
]. According to the hypotheses, Ak is a positive
matrix. Moreover, as the columns of M sum to a non-zero vector, then at least one
coefficient of y1 is strictly smaller than 1. One therefore deduces from the equality
yk+1 = Aky1 that the strict inequality yk+1 < yk holds, which proves that every row
of matrix Ak+1 sums to a number in [0, 1[.
For every vector x ∈ Rn, the inequality Ak+1x ∞ < x ∞ then holds. As a
consequence, all the eigenvalues of Ak+1 have their moduli in [0, 1[ and so do the
eigenvalues of A. This proves that the series i≥0 Ai converges, which is equivalent to
the non-singularity of I − A (see Lemma 2.1 in Chap. 6 of [
2
]). As matrix M is equal
to I − A up to a non-zero multiplicative coefficient, the lemma is proven.
The following theorem is the main result of this paper. Not only does it show
that all the subdivisions of 2-dimensional point configurations are weakly regular but
also that it is possible to regularize such subdivisions by moving their vertices along
parallel lines.
Theorem 3 If S is a subdivision of a 2-dimensional point configuration then S is
weakly regular and ε(S) ≤ 1.
Proof Let u and v be two orthogonal unit vectors in R2 such that all the vertices of
S have distinct coordinates along v. In a first step, it is shown that there is a map
τ : ei (S) → ]0, +∞[ so that for all a ∈ vi (S),
This will be achieved by inductively building a map σ : vi (S) →]0, +∞[ that satisfies
given properties and subsequently deducing τ from σ . For any vertex a ∈ v(S), call
v−(a, S) the set of all vertices b ∈ v(S) so that b · v ≤ a · v and denote:
l−(a, S) = b ∈ v−(a, S) : {a, b} ∈ ei (S) .
In other words, l−(a, S) is the set of the vertices b of S whose coordinates along v
are smaller than a · v and so that {a, b} is an interior edge of S. Call n+(a, S) the
number of the edges {a, b} interior to S that satisfy a · v < b · v. Now let a ∈ v(S).
It is proven by induction on the cardinality of v−(a, S) that there exists a map σ :
v−(a, S) →]0, +∞[ so that for all x ∈ v−(a, S) ∩ vi (S),
n+(x, S)σ (x) =
σ (y).
y∈l−(x,S)
If v−(a, S) only has one element then this element is necessarily a and one defines a
positive map on v−(a, S) by choosing an arbitrary positive number for σ (a). Notice
that, at this stage, (3) does not need to be satisfied as a ∈ vi (S). Now assume that
v−(a, S) contains at least two vertices. Call a the vertex in v−(a, S) that satisfies
v−(a, S) = v−(a , S) ∪ {a}. By induction there exists a map σ : v−(a , S) →]0, +∞[
so that for all x ∈ v−(a , S) ∩ vi (S), (3) is satisfied. This map can be extended to
v−(a, S) by choosing a positive number for σ (a) if a is in the boundary of S and,
1
σ (a) = n+(a, S)
y∈l−(a,S)
σ (y),
if a is interior to S. Doing so is possible as, by induction, σ already has been defined
on l−(a, S). Observe then that σ (a) > 0 and further that if a is interior to S, (3) is
necessarily satisfied for x = a, which finishes the induction.
This shows the existence of a positive map σ on v(S) so that for every interior
vertex x of S, (3) is satisfied. Now consider an interior edge of S, denote by a and b
its vertices with the convention that a · v < b · v, and call
σ (a)
τ {a, b} = (b − a) · v .
By (3), this defines a positive map τ on ei (S) so that (2) is satisfied for every interior
vertex a of S. In a second step, an injection ψ : v(S) → R2 will be found so that (1)
holds for all the interior vertices of S with map τ .
(2)
(3)
For every vertex a ∈ vi (S), call βa = b∈l(a,S) τ ({a, b})(b − a) · u and consider
the following linear equation system with variables (xa )a∈vi (S):
∀ a ∈ vi (S),
τ {a, b} xa −
τ {a, b} xb = βa .
(4)
As a matrix equation, this system reads MX = B, where X and B are the vectors
with coefficients xa and βa , respectively, and M is a square matrix whose diagonal
coefficients are the sums b∈l(a,S) τ ({a, b}) for a ∈ vi (S) and whose non-diagonal
coefficients are either equal to zero or to −τ ({a, b}) for some (a, b) ∈ vi (S)2. Hence,
the diagonal coefficients of M are positive and its non-diagonal coefficients are
nonpositive. Moreover, the sum of the coefficients of matrix M in the row corresponding
to a vertex a ∈ vi (S) is either positive (if l(a, S) ⊂ vi (S)) or equal to zero (if l(a, S) ⊂
vi (S)). As there necessarily exist vertices in vi (S) adjacent to some boundary vertex
of S, the columns of matrix M sum to a non-zero, non-negative vector. Call α the
largest diagonal coefficient of M . The matrix αI − M is non-negative and its
nonzero off diagonal coefficients exactly correspond to the edges in the 1-skeleton of S.
As the 1-skeleton of S is a connected graph, there exists an integer k ∈ N so that
(αI − M)k is positive. According to Lemma 1, M is therefore invertible and system
(4) admits a unique solution. Let (xa)a∈vi (S) be this solution. For all a ∈ vi (S), define
ψ (a) = a + xau and for all a ∈ v(S) \ vi (S), define ψ (a) = a. Using ψ instead of
(xa)a∈vi (S) in the system of (4), it is found that, for all a ∈ vi (S),
τ {a, b} ψ (b) − ψ (a) · u = 0.
(5)
Let a be a vertex of S. As a and ψ (a) have the same coordinate along v, one obtains
from (2) and (5) that (1) holds for vertex a. This shows that maps τ and ψ satisfy the
hypotheses of Theorem 2. According to this theorem, ψ (S) is a subset of some regular
subdivision R of ψ (v(S)). As R and ψ (S) are (topologically) 2-dimensional balls
with the same boundary, this inclusion implies the equality ψ (S) = R. This proves
that ψ ∈ R(S) and that S is weakly regular. Moreover, the affine hull of {a − ψ (a) :
a ∈ v(S)} is either equal to {0} or to the 1-dimensional space spanned by vector u.
As a consequence ε(S) ≤ 1.
Observe that the proof of Theorem 3 gives an explicit way to regularize a
subdivision S of any 2-dimensional point configuration: first, a positive map τ on the interior
edges of S is built by considering the vertices of S ordered along a given axis. At this
stage τ satisfies (2) for all a ∈ vi (S). Secondly, the system of (4) is solved, which
provides a map ψ that defines motions for the interior vertices of S so that ψ (S) is a
regular subdivision of ψ (v(S)). This construction will be explicitly carried out on an
example in Sect. 4.
Three corollaries of Theorem 3 are now given. Let S be a subdivision of some
2-dimensional point configuration. As a consequence of Theorems 1 and 3, S is
2-regular. In other words, it is projected from the boundary complex of some
4-dimensional polytope. This settles the 2-dimensional case of the conjecture
formulated in [
14
]:
Corollary 3 All the subdivisions of any 2-dimensional point configuration are
2-regular.
Observe that this corollary is a tight result as there exist 2-dimensional point
configurations with non-regular subdivisions (see Sect. 4). The two following corollaries
only give alternative proofs to already known results. It is a well known consequence
of Steinitz’s theorem [
20
] that every 3-connected planar graph is isomorphic to the
1-skeleton of a polyhedral subdivision of some 2-dimensional point configuration.
Indeed, if p is the polytope built following Steinitz’s theorem, such a subdivision can
be obtained by centrally projecting the boundary complex of p from a point placed
close to one of its 2-dimensional faces. The obtained subdivision would be regular if
a linear projection were used instead of a central one. This issue was settled in [
1
]
where it is proven that one can prescribe the faces of p that (linearly) project to the
boundary of the subdivision. Theorem 3 provides an alternative proof of this result:
Corollary 4 Every 3-connected planar graph is isomorphic to the 1-skeleton of some
regular subdivision of a 2-dimensional point configuration.
Finally, recall that the subgraph induced by 2-regular subdivisions in the flip-graph
of a point configuration is connected (Theorem 2 in [
14
]). Hence, the following result,
already obtained by Lawson in [
7
], gets an alternative proof from the combination of
Corollary 3 with [
14
]:
Corollary 5 The flip-graph of any 2-dimensional point configuration is connected.
4 Examples and Discussion
In this section, the construction used in the proof of Theorem 3 is explicitly applied to
the non-regular triangulation shown in the left of Fig. 1. Let d and e be two positive
numbers. The matrix below gives the coordinates in R2 of six points that collectively
form a point configuration A. Those coordinates are expressed in an orthogonal basis
(u, v) of R2. Each column of the matrix corresponds to a point whose coordinates
along u and v can be, respectively, read in the first and second row:
A =
−d/2
−√3d/2
(e − 1)/2
−√3(e + 1)/2
d
0
1 + e/2
√3e/2
−e√− 1/2
3/2
√−3dd//22 .
Each point in A will be referred to using the number of its column in the above
matrix, and subsets of A will be denoted by simply enumerating the points they
contain. Consider the following subsets of A:
124, 126, 134, 245, 256, 345, 356.
If d > 1 and e < (√d2 + 6d − 3 − d − 1)/2, then these seven subsets of A are the
2-dimensional faces of a non-regular triangulation Td,e of A. This triangulation is
depicted on the left of Fig. 1 for d = 3 and e = 1/3. One can check that Td,e is
actually non-regular by using the acyclicity criterion described in [
5
]: edges 14, 26,
and 35 form a cycle of overlapping faces with respect to the centroid of triangle 245.
The dotted lines drawn in the left of Fig. 1 help visualize this property. It is proven in
[
5
] that a subdivision with such a cycle is necessarily non-regular.
In the proof of Theorem 3, an orthonormal basis (u, v) is first chosen so that
the points in A have distinct coordinates along v. The basis chosen here to express
the coordinates of the points in A already fulfills this requirement. Next, a map σ :
A → ]0, +∞[ is built so that (3) is satisfied by every vertex x interior to Td,e. Here,
following the procedure found in this proof, and choosing √3/2 as the image under
σ of all the boundary vertices of Td,e, one obtains
σ (2) =
√3
6 ;
σ (4) =
7√3
6 ;
σ (5) =
.
A map τ : ei (Td,e) → ]0, +∞[ is then constructed from map σ and from the
coordinates along v of the vertices of Td,e. Here, one finds
τ (12) = 1/(d − e − 1);
τ (25) = 1/(6 + 3e);
τ (35) = 1;
τ (14) = 1/(d + e);
τ (26) = 1/(3 + 3e + 3d);
τ (45) = 7/(3 − 3e);
τ (24) = 1/(3 + 6e);
τ (34) = 1/e;
τ (56) = 11/(3d − 3).
The system (4) can then be written using those values for τ and solved. While this
can actually be done formally as a function of d and e, the expression of the solution
obtained is very complicated and will be omitted here. The exact value of this solution
for d = 3 and e = 1/3 is given instead: x2 = −214/1047, x4 = 176/349, and x5 =
1846/1047. By Theorem 3, those quantities are motions along vector u of vertices 2,
4, and 5, respectively, so that the resulting triangulation is regular. This triangulation
is shown on the right of Fig. 1, where the original triangulation is drawn with dashed
lines and the motion of the vertices takes place along the horizontal dotted lines. One
can see in particular that edges 14, 26, and 35 no longer form a cycle of overlapping
faces after the motion of the vertices.
A natural question is whether Theorem 3 carries over to higher dimensions. The
answer is given by Rudin’s triangulation [
15
]. This triangulation of a 3-dimensional
point configuration is non shellable. Non shellability is a topological property and as
a consequence, any triangulation obtained by moving the vertices of Rudin’s
triangulation while preserving its topology will remain non shellable. As regularity implies
shellability, Rudin’s triangulation cannot be weakly regular. Hence, above
dimension 2, subdivisions exist that are not weakly regular. It has been shown in [
14
],
though, that Rudin’s triangulation is 2-regular. While it is unlikely that Corollaries 3
and 5 carry over to higher dimensions (see [
14
]), a subdivision of some 3-dimensional
point configuration that is not 2-regular is yet to be found.
Finally, in the perspective of practical applications [
6, 12, 13, 22
], the key question
is whether a subdivision known to be weakly regular can be transformed by flips into
a regular subdivision. While this question remains open in dimension 3, the tools
described in this paper contribute to provide insights on possible answers.
Acknowledgement I am grateful to Tom Liebling for his kind and patient help with the writing of this
article, and for his friendship.
1. Barnette , D.W. : Projections of 3-polytopes . Isr. J. Math. 8 , 304 - 308 ( 1970 )
2. Berman , A. , Plemmons , R.J.: Nonnegative Matrices in the Mathematical Sciences. Classics in Applied Mathematics , vol. 9 . SIAM, Philadelphia ( 1994 )
3. Billera , L.J. , Sturmfels , B. : Fiber polytopes . Ann. Math. 135 , 527 - 549 ( 1992 )
4. Billera , L.J. , Kapranov , M.M. , Sturmfels , B. : Cellular strings on polytopes . Proc. Am. Math. Soc . 122 , 549 - 555 ( 1994 )
5. Edelsbrunner , H. : An acyclicity theorem for cell complexes in d dimension . Combinatorica 10 , 251 - 260 ( 1990 )
6. Ferrez , J.-A. , Liebling , Th.M.: Dynamic triangulations for efficient detection of collisions between spheres with applications in granular media simulations . Philos. Mag., B 82 , 905 - 929 ( 2002 )
7. Lawson , C.L. : Software for C1 interpolation . In: Rice, J . (ed.) Mathematical Software III , pp. 161 - 194 . Academic Press, New York ( 1977 )
8. Lee , C.W. : Regular triangulations of convex polytopes . In: Gritzmann, P. , Sturmfels , B . (eds.) Applied Geometry and Discrete Mathematics-The Victor Klee Festschrift. DIMACS Series in Discrete Mathematics and Theoretical Computer Science , vol. 4 , pp. 443 - 456 . American Mathematical Society, Providence ( 1991 )
9. Lee , C.W. : Subdivisions and triangulations of polytopes . In: Goodman, J.E. , O'Rourke , J . (eds.) Handbook of Discrete and Computational Geometry , pp. 383 - 406 . CRC Press, Boca Raton ( 2004 )
10. Maxwell , J.C. : On reciprocal figures and diagrams of forces . Philos. Mag. Ser. 7 27 , 250 - 261 ( 1864 )
11. Maxwell , J.C. : On reciprocal figures frames, and diagrams of forces . Trans. R. Soc. Edinb . 26 , 1 - 40 ( 1869 )
12. Pournin , L. , Liebling , Th.M.: Constrained paths in the flip-graph of regular triangulations . Comput. Geom . 37 , 134 - 140 ( 2007 )
13. Pournin , L. , Liebling , Th.M.: From spheres to spheropolyhedra: generalized Distinct Element Methodology and algorithm analysis . In: Cook, W. , Lovász , L. , Vygen , J . (eds.) Research Trends in Combinatorial Optimization, pp. 347 - 363 . Springer, Berlin ( 2009 )
14. Pournin , L. : A result on flip-graph connectivity . Adv. Geom . (in press)
15. Rudin , M.E. : An unshellable triangulation of a tetrahedron . Bull. Am. Math. Soc . 64 , 90 - 91 ( 1958 )
16. Rybnikov , K. : Stresses and liftings of cell-complexes . Discrete Comput. Geom . 21 , 481 - 517 ( 1999 )
17. Santos , F. : A point set whose space of triangulations is disconnected . J. Am. Math. Soc . 13 , 611 - 637 ( 2000 )
18. Santos , F. : On the refinements of a polyhedral subdivision . Collect . Math. 52 , 231 - 256 ( 2001 )
19. Santos , F. : Geometric bistellar flips: the setting, the context and a construction . In: Sanz-Sole, M. , Soria , J. , Varona , J.L. , Verdera , J . (eds.) Proceedings of the ICM 2006 , vol. III, pp. 931 - 962 . Eur. Math. Soc. , Zurich ( 2006 )
20. Steinitz , E.: Polyeder und Raumeinteilungen . Enzykl. Math. Wiss. 3 , 1 - 139 ( 1922 )
21. Whiteley , W.: Motions and stresses of projected polyhedra . Topol. Struct . 7 , 13 - 38 ( 1982 )
22. Xue , X.J. , Righetti , F. , Telley , H. , Liebling , Th.M. , Mocellin , A. : The Laguerre model for grain growth in three dimensions . Philos. Mag., B 75 , 567 - 585 ( 1997 )