Mod 2 Cohomology of Combinatorial Grassmannians
Laura Anderson* James F. Davis*
Oriented matroids have long been of use in various areas of combinatorics
[BLS+ 93]. Gelfand and MacPherson [GM92 ] initiated the use of oriented ma-
troids in manifold and bundle theory, using them to formulate a combinatorial
formula for the rational Pontrjagin classes of a differentiable manifold. MacPh*
*er-
son [Mac93 ] abstracted this into a manifold theory (combinatorial differential
(CD) manifolds) and a bundle theory (which we call combinatorial vector bun-
dles or matroid bundles). In this paper we explore the relationship between
combinatorial vector bundles and real vector bundles. As a consequence of our
results we get theorems relating the topology of the combinatorial Grassmanni-
ans to that of their real analogs.
The theory of oriented matroids gives a combinatorial abstraction of linear
algebra; a k-dimensional subspace of Rn determines a rank k oriented matroid
with elements {1; 2; : :;:n}. Such oriented matroids can be given a partial ord*
*er
by using the notion of weak maps, which geometrically corresponds to moving
the k-plane into more special position with respect to the standard basis of Rn.
The poset MacP (k; n) of rank k oriented matroids with n elements was defined
by MacPherson in [Mac93 ] and is often called the MacPhersonian. The limit of
the finite MacPhersonians gives an infinite poset MacP (k; 1), and its geometric
realization k MacP (k; 1)k is the classifying space for rank k matroid bundles.
Our main results are the Combinatorialization Theorem, which associates
a matroid bundle to a vector bundle, the Spherical Quasifibration Theorem,
which associates a spherical quasifibration to a matroid bundle, and the Com-
parison Theorem, which shows that the composition of these two associations is
essentially the forgetful functor. More precisely, if B is a regular cell compl*
*ex,
and if Vk(B), Mk(B), and Qk(B) denote the isomorphism classes of rank k
vector bundles, matroid bundles, and spherical quasifibrations respectively, we
construct maps
Vk(B) ! Mk(B) ! Qk(B)
whose composite is the forgetful map given by deletion of the zero section. The
first map is intimately related to the construction of a continuous map
": G(k; R1 ) ! k MacP (k; 1)k
and the Comparison Theorem leads to the following theorem.
____________________________*
Partially supported by grants from the National Science Foundation.
1
Theorem A. The map
"* : H*(k MacP (k; 1)k; Z2) ! H*(G(k; R1 ); Z2)
is a split surjection.
The mod 2 cohomology of G(k; R1 ) is well-known: it is a polynomial algebra
on the Stiefel-Whitney classes w1; w2; : :;:wk. The above theorem gives Stiefel-
Whitney characteristic classes for matroid bundles.
More generally, one can consider any rank n oriented matroid Mn as a
combinatorial analog to Rn. There is an associated combinatorial Grassmannian
(k; Mn ) which is a partially ordered set of rank k "subspaces" of Mn . The
MacPhersonian arises in this way as MacP (k; n) = (k; Mc), where Mc is the
unique rank n oriented matroid with elements {1; 2; : :;:n}. If Mn is realizabl*
*e,
any realization induces a simplicial map
": G(k; Rn) ! k(k; Mn )k
from a triangulation of the real Grassmannian of k-planes in Rn to the geometric
realization of the combinatorial Grassmannian, constructed by the same method
as for the special case of the MacPhersonian.
Theorem B. The map
"* : H*(k(k; Mn )k; Z2) ! H*(G(k; Rn); Z2)
is a split surjection.
There is a natural combinatorial analog to an orientation of a real vec-
tor space leading to the definition of an oriented combinatorial Grassmannian
"(k; Mn ) analogous to the Grassmannian "G(k; Rn) of oriented k-planes in Rn.
For any realizable oriented matroid Mn , there is a combinatorialization map "
from "G(k; Rn) to k"(k; Mn )k.
Theorem C. The map
"*: H*(k"(k; Mn )k; Z) ! H*(G"(k; Rn); Z)
has the Euler class in its image.
These results suggest substantial power for a combinatorial approach to char-
acteristic classes via matroid bundles.
The Comparison Theorem also leads to results on homotopy groups of the
combinatorial Grassmannian. We show that the second homotopy group of the
MacPhersonian is the same as that of the corresponding Grassmannian. (Simi-
lar results for the zero and first homotopy groups of MacP (k; n) were previous*
*ly
known.). In addition, we get results on the homotopy groups of general combi-
natorial Grassmannians (which need not even be connected ([MRG93 ])).
Theorem D. Let Mn be a realized rank k oriented matroid on n elements. Let
p be a point in the image of " : G(k; Rn) ! k(k; Mn )k.
2
1.ssk(k(k; Mn )k; p) has Z as a subgroup when k is even and n 2k.
2.ssi(k(k; Mn )k; p) has Z2 as a subquotient when i 1; 2 (mod 8), n-k i,
and k i.
3.ss4m (k(k; Mn )k; p) has Zam as a subquotient when m > 0, n - k 4m,
and k 4m. Here am is the denominator of Bm =4m expressed as a
fraction in lowest terms, and Bm is the m-th Bernoulli number.
Together with a previously known stability result ([And98 ]), this implies a
similar statement about the infinite MacPhersonian. These are the first results
giving nontrivial ssi(k(k; M)k; p) for general realizable M for any i.
Our results have potential interest in combinatorics and topology. A major
focus of work on oriented matroids has been construction of oriented matroids
behaving very differently from those in the image of ", but our partial compu-
tations of cohomology and homotopy groups should be viewed as an attempt to
tame the beast. As far as geometric topology is concerned, a matroid bundle
over a finite cell complex is a purely finite gadget, and we have shown that
matroid bundles have characteristic class information. MacPherson has conjec-
tured that in fact all characteristic classes of a vector bundle should be carr*
*ied
by the associated matroid bundle. Also, the study of matroid bundles is a
necessary first step in the study of CD manifolds.
The paper is organized as follows. We review the theory of oriented ma-
troids and develop the foundations of matroid bundles. We show that any
vector bundle whose base space is a regular cell complex defines an isomor-
phism class of matroid bundles, thus passing from topology to combinatorics.
We then construct a combinatorial "sphere bundle" (actually, a spherical quasi-
fibration) associated to any matroid bundle, and thus pass from combinatorics
back to topology. This construction is described in Section 2.3. This Com-
binatorialization Theorem depends on a deep result from combinatorics (the
Topological Realization Theorem) and a deep result from topology (Quillen's
Theorem B). From this construction we derive Stiefel-Whitney classes and Euler
classes for matroid bundles. The Comparison Theorem (see Section 5) is proven
by constructing a map of spherical quasifibrations from the canonical sphere
bundle over the real Grassmannian to the "sphere bundle" over the combina-
torial Grassmannian, which implies that the Stiefel-Whitney and Euler classes
we defined for matroid bundles map under "* to the analogous classes for real
vector bundles. This completes the proof of the Theorems A, B, and C. Theo-
rem D (Section 6) is a consequence of the Combinatorialization Theorem, the
Spherical Quasifibration Theorem, and the Comparison Theorem and homotopy
theoretic results on the image of the J-homomorphism.
We also show that these combinatorial characteristic classes have interpre-
tations analogous to those of their classical counterparts, as obstructions to
the existence of combinatorial "orientations" (Theorem 4.8) and combinatorial
"independent sets of vector fields" (Section 7).
We outline below the organization of the paper.
3
1 Preliminaries
1.1 Oriented matroids
1.2 Partially ordered sets
1.3 Combinatorial Grassmannians
2 Matroid bundles
2.1 Matroid bundles and their morphisms
2.2 Combinatorializing vector bundles: the Combinatorialization Theorem
2.3 Combinatorial sphere and disk bundles
3 Combinatorial bundles are quasifibrations
3.1 Quasifibrations
3.2 The Spherical Quasifibration Theorem
4 Stiefel-Whitney classes and Euler classes of matroid bundles
5 Vector bundles vs. matroid bundles: the Comparison Theorem
6 Homotopy groups of the combinatorial Grassmannian
7 Vector fields and characteristic classes
8 Some open questions
Appendix A Topological maps from combinatorial ones
Appendix B Babson's criterion
1 Preliminaries
1.1 Oriented Matroids
For introductions to oriented matroids, see [BLS+ 93] and [Mac93 ]. There are
several equivalent axiomatizations of oriented matroids; throughout this paper
we will use the covector axioms [BLS+ 93, p.159].
Definition 1.1.An oriented matroid M is a finite set E(M) and a subset
V*(M) {-; 0; +}E(M)
satisfying the following axioms:
1.0 2 V*(M).
2.If X 2 V*(M), then -X 2 V*(M).
3.(Composition) If X; Y 2 V*(M); then the function
X O Y : E(M) ! {-; 0; +}
ae
e 7! X(e)Y ifX(e)(6=e0)otherwise
is in V*(M).
4.(Elimination) If X(e) = + and Y (e) = -, then there is a Z 2 V*(M) such
that Z(e) = 0, and for all f 2 E(M) for which X(f) and Y (f) are not of
opposite signs, Z(f) = X O Y (f).
4
E(M) is called the set of elements of M. V*(M) is the set of covectors of
M.
The motivating example: consider n linear forms {OE1; : :;:OEn} on a finite
dimensional real vector space V . To any p 2 V we associate a sign vector
X(p) = (signOE1(p); : :;:signOEn(p)) 2 {-; 0; +}n. The set {X(p) : p 2 V } is
the set of covectors of an oriented matroid M with elements {1; 2; : :;:n}. The
set {OE1; : :;:OEn} is called a realization of M. Any M arising in this way is
called realizable. Note that any of the OEican be multiplied by a positive scal*
*ar
without changing M. Thus we can represent a realizable oriented matroid by an
arrangement of hyperplanes {OE-1i(0)} and a distinguished side OE-1i(R+ ) for e*
*ach
hyperplane. (If OEi= 0 then the corresponding "hyperplane" is the degenerate
hyperplane V and the "distinguished side" is ;.) By considering a form as
the inner product with a vector, we arrive at yet another way of viewing an
realizable oriented matroid: Take a finite collection E = {v1; : :;:vn} of vect*
*ors
in a finite dimensional real inner product space V , then the functions given by
{i 7! sign(v . vi) : v 2 V } are the covectors of an oriented matroid.
Definition 1.2.Let M be an oriented matroid with elements E. A subset I
of E is independent in M if for every e 2 I, there is a covector X so that
X(e) 6= 0, but X(I\{e}) = 0. The rank of M is the maximal order of a set of
independent elements of M.
If the oriented matroid arises from a set of vectors in a real inner product
space, then the rank of the oriented matroid equals the dimension of the span
of E.
Definition 1.3.[BLS+ 93, pp. 133-134] Let A E(M) where M is an oriented
matroid. Define two oriented matroids whose elements are E(M)\A.
1.The covectors of the deletion M\A are
{X|E(M)\A : X 2 V*(M)}:
2.The covectors of the contraction M=A are
{X|E(M)\A : X 2 V*(M) so thatX(a) = 0 for alla 2 A}:
For a realizable oriented matroid, the deletion is realized by forgetting the
linear forms in A, while the contraction is realized by restricting the forms in
E(M)\A to the intersection of the zero sets of the forms in A.
Oriented matroids are connected to topology and geometry by the Topolog-
ical Representation Theorem of Folkman and Lawrence. This gives a way to
associate a PL sphere to an oriented matroid. We will state a weak version of
the theorem here. First we take a definition from [BLS+ 93, x5.1].
Definition 1.4.A pseudosphere S in Sk-1 is the image of the equator Sk-2
under a homeomorphism h : Sk-1 ! Sk-1. An oriented pseudosphere S is a
5
pseudosphere together with a labeling S+ and S- of the connected components
of Sk-1\S. Here S+ and S- are called the (open) sides of S. A signed
arrangement of pseudospheres is a finite multiset A = (Se)e2E, where for
each e, Se is an oriented pseudosphere of Sk-1, provided the following three
conditions hold:
1.SA = \e2ASe is homeomorphic to a sphere, for all subsets A of E.
2.If SA 6 Se, for A E and e 2 E, then SA \Se is an oriented pseudosphere
in SA with sides SA \ S+eand SA \ S-e.
3.The intersection of a collection of closed sides is either a sphere or a b*
*all.
A is essential if \e2ESe = ;.
For instance, if {OE1; : :;:OEn} is a realization of M in Rk, then the arran*
*ge-
ment of equators {OE-1i(0) \ Sk-1}i2{1;2;:::;n}is a signed arrangement of pseud*
*o-
spheres. This arrangement is essential if and only if the rank of M is k.
Any essential signed arrangement of pseudospheres chops Sk-1 into a regular
cell complex. The Topological Realization Theorem shows that the cells of this
complex give the nonzero covectors of an oriented matroid and that essentially
all oriented matroids arise in this way.
Let A = (Se)e2E be a essential signed arrangement of psuedospheres in
Sk-1. Define a function
oe : Sk-1 ! {+; 0; -}E
by oe(x)(e) = +; 0; or - depending on whether x 2 S+e; Se; or S-e. Then
{oe-1(X) : X 2 oe(Sk-1)}
gives a regular cell decomposition of Sk-1 and L(A) = oe(Sk-1) [ {0} is the set
of covectors of an oriented matroid on E (see [BLS+ 93, x5.1]).
A loop in an oriented matroid is an element e so that X(e) = 0 for all
covectors X. An oriented matroid is loopfree if it has no loops.
Topological Representation Theorem. (cf. [FL78 ], [BLS+ 93])
1.If M is a loopfree, rank k oriented matroid on E, there is an essential
signed arrangement A = (Se)e2E of pseudospheres in Sk-1 with L(A) =
V*(M).
2.If A = (Se)e2E is an essential signed arrangement of pseudospheres in
Sk-1, then L(A) is the set of covectors of a loopfree, rank k oriented
matroid on E.
6
1.2 Partially ordered sets
Oriented matroids and geometric topology are related via partially ordered sets,
or posets. There are several partial orders associated to oriented matroids, wh*
*ere
moving up in the partial order corresponds to some notion of moving into more
general position. The transition from posets to geometric topology is through
a functor called geometric realization.
Definition 1.5.We define three partial orders:
1.On {-; 0; +}: The only strict inequalities are + > 0 and - > 0.
2.On {-; 0; +}E : Define X Y if X(e) Y (e) for all e 2 E.
3.On the set of oriented matroids on a set E: Define M1 M2 if for every
X 2 V*(M2) there is some Y 2 V*(M1) such that Y X.
This last relation is sometimes described by saying M2 is a weak map
image of M1, or that M2 is a specialization of M1. If {OE1; : :;:OEn} is a
realization of a rank k oriented matroid M1 and {1; : :;:n} is a realization of
M2, then there is a weak map M1 M2 if and only if sign(OEi1^ . .^.OEik)
sign(i1^ . .^.ik) for every {ii; : :;:ik} {1; : :;:n}.
An (abstract) simplicial complex K is a collection of non-empty finite
sets, closed under proper inclusion. The elements of K are called simplices.
An i-simplex is an simplex with i + 1 elements, and Ki K is the set of i-
simplices. Note that a simplicial complex is a poset, with partial order given *
*by
inclusion. A regular cell complex B is a CW complex so that every cell e has
a characteristic map Dn ! _ewhich is a homeomorphism. The face lattice F(B)
is the poset of closed cells of B, ordered by inclusion. A simplicial complex K
determines a regular cell complex, which we denote by kKk. A chain in a poset
P is a non-empty, finite, totally ordered subset of P . The order complex P
of a poset P is the simplicial complex whose simplices are the chains in P .
Thus there are functors
Posets-! Simplicial Complexesk-k-!Regular Cell ComplexesF-!Posets
For a poset P , we write kP k for kP k, and call it the geometric real-
ization of P . If P opis the poset given by reversing the inequalities, then
kP k ~=kP opk. For a regular cell complex B, the simplicial complex given by
F(B) is called the barycentric subdivision of B and there is a homeomor-
phism B ~=kF(B)k under which every closed cell oe of B maps homeomorphi-
cally to the subcomplex kF(B)oe k. Here if p is an element of a poset P , then
Pp = {q 2 P : q p}. For a simplicial complex K, there is a canonical
homeomorphism kKk ~=kF(K)k.
Order Homotopy Lemma. If f; g : P ! Q are poset maps so that for all
p 2 P , f(p) g(p) then kfk is homotopic to kgk.
7
The proof is easy. Simply let (1 > 0) be the poset with elements 1 and
0, and the only strict inequality being 1 > 0. Then apply k k to the poset
map h : P x (1 > 0) ! Q defined by h(p; 1) = f(p) and h(p; 0) = g(p). A
consequence of the lemma is that if P has a maximal or minimal element, then
kP k is contractible.
1.3 Combinatorial Grassmannians
The MacPhersonian MacP (k; n) is the poset of rank k oriented matroids on
the set {1; 2; : :;:n}, with M1 M2 if there is a weak map M1 M2. There
is an obvious embedding MacP (k; n) ,! MacP (k; n + 1), by adding n + 1 as a
loop to each oriented matroid. We identify MacP (k; n) with its image under
this embedding and define MacP (k; 1) to be the direct limit over n of the
MacP (k; n). For any rank n real vector space W with a fixed basis {w1; : :;:wn}
(and therefore a fixed inner product) there is a canonical function
: G(k; W ) ! MacP (k; n)
given by intersecting each V 2 G(k; W ) with the hyperplanes {w?1; : :;:w?n}
and considering the corresponding oriented matroid (V ). Equivalently one
projects the basis of W onto V and thus obtains an oriented matroid. This
function is definitely not continuous, but the point inverses give an interest*
*ing
decomposition of the Grassmannian. (See [BLS+ 93, Section 2.4] for more about
this decomposition.)
For future reference, we record a generalization of the MacPhersonian.
Definition 1.6.There is a strong map M1 ! M2 if V*(M2) V*(M1).
For instance, if {OE1; : :;:OEn} is a realization of M in V and W is a subsp*
*ace
of V , then {OE1|W ; : :;:OEn|W } is the realization of a strong map image of M.
If M is an oriented matroid, the combinatorial Grassmannian (k; M)
is the poset of all rank k strong map images of M, with partial order given by
weak maps. If M is the coordinate rank n oriented matroid, i.e., the unique rank
n oriented matroid with elements {1; 2; : :;:n}, then (k; M) is the MacPher-
sonian MacP (k; n).
Let M be a realizable rank n oriented matroid, and let {v1; : :;:vm } Rn
realize M. Then, as above, there is a function : G(k; Rn) ! (k; M) given
by intersecting each vector space V 2 G(k; Rn) with the oriented hyperplanes
{v?1; : :;:v?m} and taking the corresponding oriented matroid (V ).
2 Matroid bundles
We will define a combinatorial vector bundle over a regular cell complex B, to *
*be
an assignment of an oriented matroid M(e) to every cell e of B so that if f is a
face of e then M(e) weak maps to M(f). To see intuitively how a combinatorial
vector bundle is derived from a vector bundle over a finite regular cell complex
8
B, note that if we fix a metric for the bundle and a finite set S of sections, *
*then
for every element b of the base space, {s(b)}s2S determines an oriented matroid
Mb with elements S. If we choose such an S so that for every b, {s(b)}s2S
spans the fiber over b and if the function b 7! Mb is constant on the interior *
*of
each cell of B, then these oriented matroids determine a combinatorial vector
bundle structure on B. In this case we say S is tame with respect to B. We will
show that tame sections exist (perhaps after a subdivision of the base space)
for vector bundles over finite-dimensional regular cell complexes.
2.1 Matroid bundles and their morphisms
The following is a generalization of the definition in [Mac93 ].
Definition 2.1.A rank k matroid bundle = (B; M) is a poset B and a
poset map M : B ! MacP (k; 1).
Definition 2.2.The universal rank k matroid bundle is
flk = (MacP (k; 1); Id)
Definition 2.3.A rank k combinatorial vector bundle = (B; M) is a
piecewise-linear (PL) cell complex B and a poset map M from the set of cells
of a PL subdivision of B, ordered by inclusion, to MacP (k; 1). In other words
a combinatorial vector bundle (B; M) is a matroid bundle (F(B0); M) where
F(B0) is the poset of cells of a PL subdivision B0 of B.
Note every regular cell complex can be given the structure of a PL space via
a barycentric subdivision.
A matroid bundle (B; M) gives a combinatorial vector bundle (kBk; M0),
where M0(b0 < b1 < . .<.bm ) = M(bm ). A combinatorial vector bundle
= (B; M : F(B0) ! MacP (k; 1)) induces a combinatorial vector bundle
0 = (B; M0 : F(B00) ! MacP (k; 1)) for any PL subdivision B00of B0, by
sending any cell oe of B00to M(ffi(oe)), where ffi(oe) is the smallest cell of *
*B0
containing oe. Two combinatorial vector bundles over a PL cell complex B
are equivalent if they are equivalent under the equivalence relation generated
by PL subdivision. Two matroid bundles over a poset B are equivalent if
the associated combinatorial vector bundles are equivalent. Clearly there is a
bijective correspondence between equivalence classes of matroid bundles over
a poset and combinatorial vector bundles over its geometric realization, and
henceforth we blur the distinction.
Definition 2.4.If (B1; M1) and (B2; M2) are two matroid bundles, a mor-
phism from (B1; M1) to (B2; M2) is a triple (f; [Cf; Mf]), where f is a PL
map from kB1k to kB2k, and [Cf; Mf] is an equivalence class of combinatorial
vector bundles over the mapping cylinder of f, where Mf restricts to structures
equivalent to (Bi; Mi) at either end. Such a morphism is a morphism cover-
ing f. A morphism covering the identity on kBk is called a B-isomorphism.
Clearly equivalent bundles over B are B-isomorphic.
9
Definition 2.5.Let B1 be a poset, = (B2; M) a matroid bundle, and f :
kB1k ! kB2k a PL map. Then f is simplicial with respect to some PL subdi-
visions B01and B02of kB1k and kB2k. The composite map F(B01) ! F(B02) !
MacP (k; 1) defines a combinatorial vector bundle over kB01k, and any subdi-
vision of B01gives an equivalent bundle. Thus f defines an equivalence class
of combinatorial vector bundles over kB1k. We call any of the corresponding
matroid bundles a pullback of by f and denote it f*().
Definition 2.6.If (Bi; Mi), i 2 {1; 2; 3} are rank k matroid bundles and (f :
B1 ! B2; [Cf; Mf]) and (g : B2 ! B3; [Cg; Mg]) are morphisms between them,
then consider the space Cf [B2 Cg obtained from the disjoint union of Cf and
Cg by identifying each b 2 B2 Cf with b x 0 2 Cg. The matroid bundle
structures on Cf and Cg define a matroid bundle structure on this space. The
pullback of the PL map c : CgOf! Cf [B2 Cg defined by
ae
c[b; i] = [b;[2i]f(b); 2(ifii- 1=21_
if i 2)]1=2
and c[b] = [b] for all b 2 B3, defines an equivalence class [CgOf; MgOf] of mat*
*roid
bundles which we call the composition of the two original morphisms.
Note that two bundles are B-isomorphic if and only if there is a combinatori*
*al
vector bundle over kBk x I restricting on the ends to bundles equivalent to the
original ones.
The following familiar properties of bundles are easily verified.
Proposition 2.7. 1. Let B1 and B2 be posets. There is a morphism 1 ! 2
covering a PL-map f : kB1k ! kB2k if and only if f*2 is B1-isomorphic
to 1.
2.If OE; : A ! B are poset maps so that kOEk and k k are homotopic, and
if is a matroid bundle over B, then OE* is A-isomorphic to *.
3.Every rank k matroid bundle is a pullback of the universal rank k bundle.
4.If is a rank k matroid bundle and there are two morphisms ! flk
covering f and g respectively, then f and g are homotopic.
Recall that [X; Y ] is the set of homotopy classes of maps from X to Y .
Corollary 2.8. For a regular cell complex B, let Mk(B) be the set of B-isomorph*
*ism
classes of rank k combinatorial vector bundles over B. Then
Mk(B) ! [B; k MacP (k; 1)k]
[]7! [kM()k]
is a bijection, natural in B. The inverse is defined by applying the simplicial
approximation theorem to [f] to obtain a subdivision B0 of B, a poset map
f0 : B0! Fk MacP (k; 1)k, and thus a matroid bundle (B0; M) where M(oe) is
the maximal vertex of f0(oe).
10
Thus the classifying space for rank k combinatorial vector bundles is k MacP*
* (k; 1)k
with universal element flk.
Matroid bundles arise in combinatorics in a variety of ways: see [And99a ]
for examples. In addition, any real vector bundle yields a combinatorial vector
bundle, as described in the following section.
2.2 Combinatorializing vector bundles: the Combinatori-
alization Theorem
For any real rank k vector bundle = (p : E ! B)over a paracompact base
space there is a bundle map
E ---"c-!E(k; R1 )
?? ?
y ?y
B ---c-! G(k; R1 )
to the canonical bundle over the Grassmannian of k-planes in R1 , and c is
determined up to homotopy (cf. [MS74 ] Ch. 5). If B is the underlying space
of a regular cell complex, we call the map c tame if O c is constant on the
interior of each cell. Such a tame classifying map gives a combinatorial vector
bundle c() = (F(B); M) by defining M(oe) = (c(intoe)). Here M is a poset
map since if oe is a face of o , then
_______
c(intoe) \ c(into)6= ;
so (c(intoe)) (c(into)) by [BLS+ 93, 2.4.6]. A subdivision of the cell com-
plex leads to an equivalent combinatorial vector bundle.
Generalizing the notation a bit, for any vector space V , let G(k; V ) be the
Grassmannian of k-planes in V . A classifying map for a vector bundle =
(p : E ! B) is a map c : B ! G(k; V ) covered by a bundle map from
to the canonical bundle. For any finite set F , let MacP (k; F ) be the poset of
rank k oriented matroids with elements F . For any set A, let MacP (k; A) be
the direct limit of MacP (k; F ), taken over all finite subsets F of A. If V is*
* a
vector space with a finite basis A, define : G(k; V ) ! MacP (k; A) as we did
in Section 1.3, while if the basis A is infinite, define so that it restricts *
*to
: G(k; SpanF ) ! MacP (k; F ) for all finite subsets F of A. If V is a vector
space, a tame classifying map is a classifying map c : B ! G(k; V ) so that
O c is constant on the interior of each cell.
Theorem 2.9 (Combinatorialization Theorem). Let = (p : E ! B) be
a rank k real vector bundle, where B is the underlying space of a regular cell
complex.
1.For i = 0; 1, let ci : B ! G(k; Vi) be a tame classifying map for . Then
there is a tame classifying map h : B x I ! G(k; V0 V1) for x I,
restricting to ci on B x {i}.
11
2.If B is finite dimensional, any classifying map c : B ! G(k; V ) is homo-
topic to a classifying map which is tame with respect to some simplicial
subdivision of the barycentric subdivision of B.
Proof. Note that for a vector bundle = (p : E ! B), specifying a classifying
map c : B ! G(k; V ) together with a covering map "c: E ! E(k; V ) is equiv-
alent to specifying a map for ^c: E ! V which is linear and injective on each
fiber. (Here c(b) = ^c(p-1b).) If V has a basis A, then c is tame if and only if
the function
B ! subsets of {+; -; 0}A
b7! {a 7! sign(^c(e) . a)}e2p-1b
is constant on the interior of cells.
We prove something slightly more general than (1). Suppose that V has a
basis A and that c0; c1 : B ! G(k; V ) are two tame classifying maps for with
covering maps bc0; bc1. Suppose also that for every e 2 E and for every a 2 A,
bc0(e) . a and bc1(e) . a do not have opposite signs. Then
hbt(e) = (1 - t)cb0(e) + tcb1(e) 0 t 1
defines a classifying map h : B x I ! G(k; V ) for the bundle E x I ! B x I
which is tame with respect to the product cell structure on B x I and hence a
tame homotopy between c0 and c1. Applying this to V = V0 V1 gives (1).
For part (2), it suffices to prove it for the universal case. We will find a
triangulation of G(k; Rn) so that the identity map is tame, i.e., : G(k; Rn) !
MacP (k; n) is constant on the interior of simplices. Then given a classifying
map c : B ! G(k; V ) where B has dimension r, by the cellular approximation
theorem and the Schubert cell decomposition of the Grassmannian, there is a
homotopic map c0 : B ! G(k; V 0) where V 0 V is a vector space spanned
by k + r elements of the basis. Finally, apply the Simplicial Approximation
Theorem to map from the barycentric subdivision of B to the tame triangulation
of G(k; V ).
Thus the following lemma applied to the coordinate oriented matroid (where
(k; M) = MacP (k; n)) completes the proof of the combinatorialization theo-
rem.
Lemma 2.10. (cf. [Mac93 ]) Let M be a realizable rank n oriented matroid with
a fixed realization. Then there is a semi-algebraic triangulation T of G(k; Rn)
and a simplicial map with respect to its barycentric subdivision
" : G(k; Rn) ! k(k; M)k
such that for every vertex v in the barycentric subdivision, one has "(v) = (v).
Furthermore, the homotopy class of " is independent of the choice of semi-
algebraic triangulation.
12
Proof. This is an application of Appendix A, theorems on existence and unique-
ness of semi-algebraic triangulations, and the fact that (cf. 2.4.6 in [BLS+ 93*
*])
: G(k; Rn) ! k(k; M)k is upper semi-continuous.
A key tool is [Hir75, Semi-algebraic triangulation theorem] which proves that
for any finite partition {Ui}i2I of a bounded, semi-algebraic set S into semi-
algebraic sets there exists a semi-algebraic triangulation of S such that each *
*Ui
is a union of the interiors of simplices. Furthermore, by [Hir75, 2.4], for any*
* two
such semi-algebraic triangulations, there is a semi-algebraic triangulation whi*
*ch
is a common refinement. Thus S is a P L space.
In the case at hand {-1(N)}N2(k;M) is a semi-algebraic partition of G(k; Rn).
In the language of Corollary A.4, the corresponding triangulation refines the
stratification given by the upper semi-continuous map , so the result_fol-
lows. |__|
|___|
Corollary 2.11. Let B be a finite dimensional regular cell complex. Let Vk(B)
be the set of B-isomorphism classes of rank k vector bundles over B. There is
a "combinatorialization map"
C : Vk(B) ! Mk(B);
natural in B, defined by sending a vector bundle to the combinatorial vector
bundle given by a tame classifying map.
Proof of Corollary.Let be a k-dimensional vector bundle over B.
o Existence: The combinatorialization theorem shows there is a tame clas-
sifying map
c : B ! G(k; V ):
Define C[] to be the corresponding combinatorial vector bundle [c()].
o Uniqueness: Suppose
ci: B ! G(k; Vi) i = 0; 1
are two tame classifying maps. Applying the first part of the combinato-
rialization theorem gives a tame classifying map
h : B x I ! G(k; V0 V1);
for xI restricting to c0 and c1 at either end. The resulting combinatorial
vector bundle over B xI gives an B-isomorphism between c0() and c1().
o Naturality: Clear.
|___|
13
We wish to extend the map Vk(B) ! Mk(B) to infinite dimensional com-
plexes. The problem, as shown in [AD ], is that G(k; R1 ) has no tame triangu-
lation, i.e. no triangulation where is constant on simplices. But we do have
the following theorem.
Theorem 2.12. There is a map " : G(k; R1 ) ! k MacP (k; 1)k which re-
stricts to a map G(k; Rn) ! k MacP (k; n)k given by Lemma 2.10 for all n. The
homotopy class of " is well-defined.
Proof. This follows from Appendix A and results on existence of semi-algebraic_
triangulations, by the same argument as the proof of Lemma 2.10. |__|
Corollary 2.13. Let B be a regular cell complex. There is a "combinatorial-
ization map"
C : Vk(B) ! Mk(B);
natural in B, which for finite-dimensional B coincides with the map given by
sending a vector bundle to the combinatorial vector bundle given by a tame
classifying map.
Proof. By replacing B by kF(B)k, we may assume that B is the geometric
realization of a simplicial complex. Let c : B ! G(k; R1 ) be a classifying map
for . Apply the simplicial approximation theorem to "O c to find a subdivision
B0 of B and a map c0: F(B0) ! MacP (k; 1), where kc0k is homotopic to "O_c.
Then set C[] = [c0]. |__|
2.3 Combinatorial sphere and disk bundles
Gelfand and MacPherson, in their combinatorial formula for the Pontrjagin
classes of a differentiable manifold, constructed a "combinatorial sphere bundl*
*e"
associated to a matroid bundle:
Definition 2.14. For a matroid bundle = (B; M), define posets
E() = {(oe; X) : oe 2 B; X 2 V*(M(oe)) }
E0() = {(oe; X) : oe 2 B; X 2 V*(M(oe))\{0} }
with (oe; X) (oe0; X0) if oe oe0 and X X0.
The projection map ss0 : E0() ! B and ss : E() ! B are the combinato-
rial sphere bundle and combinatorial disk bundle associated to .
Example 2.15. Warning: The geometric realization of the combinatorial sphere
bundle may not be a topological sphere bundle! The Topological Representation
Theorem promises that the realization of each fiber of ss0 over a vertex is a PL
sphere. But in general, the realization of ss0 is not a topological sphere bund*
*le,
as we can see from the example in Figure 1.
This figure shows a weak map M1 M0 of rank 2 (realizable) oriented
matroids with elements {a; b; c}; in the second oriented matroid the element b
14
M1 = a M0 = a
b
0 = b
c c
Figure 1: A sphere bundle need not be a sphere bundle.
is the degenerate hyperplane 0? . For each oriented matroid, the nonzero cov-
ectors are given by the cell decomposition of the unit circle. We can define a
rank 2 matroid bundle over the poset (1 > 0) by sending 1 to M1 and 0 to
M0. The total space of the associated sphere bundle will contain a 3-simplex
{(1; a- b+ c+ ); (1; a- b0c+ ); (0; a- b0c+ ); (0; a0b0c+ )}. Hence the geomet*
*ric real-
ization of the combinatorial sphere bundle is not a topological circle bundle
over the 1-dimensional cell k1 > 0k.
It seems unlikely that the geometric realization of ss0 always gives a fibra*
*tion.
In Section 3 we will show that it is the next best thing, a spherical quasifibr*
*ation.
3 Combinatorial bundles are quasifibrations
In this section we review the notion (due to Dold-Thom [DT56 ]) of a quasifi-
bration and show that the combinatorial sphere bundle associated to a matroid
bundle is a spherical quasifibration. A key tool is a criterion for the geometr*
*ic
realization of a poset map to be a quasifibration. This criterion was formulated
in the Ph.D. thesis [Bab93 ] of Eric Babson. We give a proof of Babson's crite-
rion in Appendix B. It is an application of Quillen's work on the foundations
of algebraic K-theory.
3.1 Quasifibrations
Definition 3.1.A map p : E ! B is a fibration if it has the homotopy lifting
property [Whi78 ]. A map p : E ! B is a quasifibration if
p* : ssi(E; p-1b; e) ! ssi(B; {b}; b)
is an isomorphism for all i 0, for all b 2 B, and for all e 2 p-1b.
Definition 3.2.A spherical (quasi)-fibration of rank k is a (quasi)-fibration
p0 : E0 ! B so that for all b 2 B, p-10b has the weak homotopy type of Sk-1 (i.*
*e.,
there is a map Sk-1 ! p-10b inducing an isomorphism on homotopy groups).
15
A fibration is a quasifibration. An example of a quasifibration which is not
a fibration is given by collapsing a closed subinterval of an interval to a poi*
*nt.
A (quasi)-fibration has a long exact sequence in homotopy.
A construction of Bourbaki [Whi78 , xI.7] shows that every continuous map
f : E ! B has the homotopy type of a fibration, i.e. there is a fibration
ssf : Pf ! B and a homotopy equivalence h : E ! Pf so that f = ssf O h. Here
Pf = {(e; ff) 2 E x BI : f(e) = ff(0)}
ssf(e; ff) = ff(1) and h(e) = (e; conste). For b 2 B, f-1 b is the fiber above b
and ss-1fb is the homotopy fiber above p. The homotopy long exact sequence
of a fibration and the five lemma give the following alternative (and perhaps
better) definition of a quasifibration.
Proposition 3.3. A map p : E ! B is a quasifibration if and only if for all
b 2 B
p-1(b) ! ssp-1(b)
is a weak homotopy equivalence.
Definition 3.4.A morphism of (quasi)-fibrations p and p0 is a map f :
B ! B0 and a (quasi)-fibration Ef ! Cf over the mapping cylinder of f which
restricts on the ends to f and f0. Such a morphism is called a morphism cov-
ering f. A morphism covering the identity on B is called a B-isomorphism.
A CW-(quasi)-fibration, respectively a B-CW-isomorphism, is a (quasi)-
fibration, respectively B-isomorphism, in which the domain of each (quasi)-
fibration has the homotopy type of a CW-complex.
Let p0: E0 ! B and p : E ! B be two maps. A map g : E0 ! E is fiber-
preserving if pOg = p0. A fiber-preserving homotopy equivalence (f.p.h.e)
is a homotopy equivalence g : E0 ! E which is fiber-preserving. There is also
the notion of a fiber-preserving weak homotopy equivalence (f.p.w.h.e).
Two maps g; g0 : E0 ! E are fiberwise homotopic if there is a homotopy
G : E0x I ! E between them so that for all t, G(-; t) is fiber-preserving. A
fiber-preserving map g : E0 ! E is a fiber homotopy equivalence (f.h.e) if
there is a fiber-preserving map h : E ! E0 so that g O h and h O g are both
fiberwise homotopic to the identity. One says that p and p0have the same fiber
homotopy type. Of course a fiber homotopy equivalence is a fiber-preserving
homotopy equivalence.
Two quasifibrations p : E ! B and p0 : E0 ! B are B-isomorphic if
and only if they are equivalent under the equivalence relation generated by
f.p.w.h.e. Indeed if g : E0! E is a f.p.w.h.e., then the natural map Cg ! B x I
shows that p and p0 are B-isomorphic quasifibrations. Conversely given a B-
isomorphism p00: E00! B x I, the inclusion map gives a f.p.w.h.e. from p (or
p0) to prB O p00: E00! B.
Two fibrations p : E ! B and p0: E0 ! B are B-isomorphic if and only if
there is a fiber homotopy equivalence between them, which occurs if and only if
there is a fiber preserving homotopy equivalence between them. All of this is an
16
elementary, if somewhat confusing, exercise in the homotopy lifting property.
For a reference that B-isomorphism implies f.h.e., see [Whi78 , Theorem 7.25]
and for a reference that f.p.h.e. implies h.e., see [Dol63, Theorem 6.1].
The following theorem shows that in terms of homotopy theory, there is really
not much difference between fibrations and quasifibrations. This theorem, in
slightly different language, is due to Stasheff [Sta63].
Theorem 3.5. For a CW-complex B, let Q(B) (respectively F (B)) be the set of
B-CW-isomorphism classes of CW-quasifibrations (respectively CW-fibrations)
over B. There is a bijection,
Q(B) ! F (B);
given by converting a quasifibration p into a fibration ssp. The inverse is the
forgetful map, given by considering a fibration as a quasifibration.
Proof. This conversion process has two nice properties. The first is that it
sends a fiber-preserving homotopy equivalence to a fiber-preserving homotopy
equivalence. The second is that given a map p : E ! B where E and B have
the homotopy type of a CW-complex, then Pp also has the homotopy type of a
CW-complex [Mil59].
We need to see that the map Q(B) ! F (B) is well-defined. If p00: E00!
B x I is a B-isomorphism between quasifibrations p : E ! B and p0: E0! B,
then as above, there is a a f.p.w.h.e. from p (or p0) to prBO p00, which is a f*
*.p.h.e
by the CW-assumption. Now this conversion process takes a f.p.h.e to a f.p.h.e,
and hence the corresponding fibrations are B-CW-isomorphic.
If one first converts and then forgets, one obtains a quasifibration which
is f.p.h.e. to the original one, and hence equivalent. Conversely, if one has
a fibration and converts it, the result is a fibration equivalent to_the_origin*
*al
one. |__|
Remark 3.6. By the pullback, F (-) is a contravariant functor from topological
spaces to sets. However, since the pullback of a quasifibration need not be
a quasifibration, it is not clear that Q(-) is a functor. However, using the
equivalence in the above theorem, Q(-) does give a functor from the category
of spaces having the homotopy type of CW-complexes to sets. Furthermore, if
the pullback of a quasifibration p : E ! B under a map f : B0! B happens to
be a quasifibration, then the pullback f*E ! B0represents the correct induced
element of Q(B0).
3.2 The spherical quasifibration theorem
Spherical Quasifibration Theorem. For any matroid bundle = (B; M),
the geometric realizations of the combinatorial sphere and disk bundles
kss0k: kE0()k ! kBk
kssk: kE()k ! kBk
are quasifibrations.
17
We use the following criterion for the geometric realization of a poset map
to be a quasifibration.
Babson's Criterion. If f : P ! Q is a poset map satisfying both of the con-
ditions below, then kfk is a quasifibration.
1.kf-1 q \ Pp k is contractible whenever p 2 P , q 2 Q, and q f(p).
2.kf-1 q \ Pp k is contractible whenever p 2 P , q 2 Q, and q f(p).
This criterion was formulated in the Ph.D. thesis [Bab93 ] of Eric Babson,
and is an application of Quillen's work on the foundations of algebraic K-theor*
*y.
We give a proof in Appendix B. In this section we verify that the combinatorial
bundles satisfy Babson's criterion.
Lemma 3.7. If M M0 and e is a nonzero element of M0 then M=e M0=e.
Proof. Let X0 2 V*(M0=e) = {Z0 2 V*(M0) : Z0(e) = 0}. Since e is nonzero
in M0, there are covectors Z01and Z02of M0 so that Z01(e) = + and Z02(e) = -.
Since M M0, there are covectors X1 and X2 of M so that X1 X0O Z01and
X2 X0O Z02. Since X1(e) = + and X2(e) = -, we can apply the elimination
axiom in the definition of an oriented matroid. What results is a covector_X of
M=e so that X X0. |__|
Lemma 3.8. If M M0, rankM = rankM0, and X is a nonzero covector of
M, then there is a nonzero covector X0 of M0 so that X X0.
Proof. We induct on rank(M) and on the number of elements of M. When
rank(M) = 1, the existence of X0 is easy.
If rank(M) > 1, it suffices to consider the_case when X is not_maximal,_since
if X is maximal,_there is a nonzero covector X of M so that X > X and we
replace X by X . So assume that there is some nonzero element e of M such
that X(e) = 0. We have two cases:
o If e is zero in M0, then M\e M0\e. Then by induction on the number
of elements we get X02 V*(M0\e) = V*(M0) such that X X0.
o If e is nonzero in M0, then by Lemma 3.7 we have M=e M0=e. Since X 2
V*(M=e), induction on rank gives a nonzero X0 2 V*(M0=e) V*(M0)
such that X X0.
|___|
Remark 3.9. A consequence of this lemma is that if M M0 and rankM =
rank M0, there is a poset map : V*(M) ! V*(M0) which maps nonzero
covectors to nonzero covectors and so that X (X) for all X. Indeed, (X)
is defined to be the composition of all nonzero covectors of M0 which are less
than or equal to X. This map lends credence to the intuition that a weak
map corresponds to moving into special position. This map and variations are
explored further in [And ].
18
Lemma 3.10. Let K be a simplicial decomposition of a compact PL manifold
with boundary, let K0 = {oe 2 K : koek \ k@Kk = ;}, and assume that @K is
full, i.e., any simplex whose faces are all contained in @K is itself contained*
* in
@K. Then kKk ' kK0k.
Proof. Enumerate the simplices oe1; oe2; : :;:oek of @K so that the dimension is
monotone decreasing. Let
Ki= K\(Koe1 [ . .[.Koei-1):
Note K1 = K and Kk+1 = K0.
We show kK0k can be obtained from kKk by a sequence of elementary
collapses and expansions. Define an elementary collapse to be inward if it
collapses out a pair of simplices ! and ! [ {x} with {x} 62 @K. (We will use
the term inward to apply to collapses of any complex, not just K.) We show by
induction on dimension and induction on i that kKi+1k can be obtained from
kKik by a sequence of elementary collapses and expansions. Both initial cases
hold because @K is full.
We will use ~ to denote equivalent via a sequence of elementary collapses
inwards and elementary expansions.
k linkKoeik~ k linkoeik (induction on i and K ~ Ki)
i K
~ k linkoeik (induction on dimension and k linkoeik is a PL-ball)
K0 K
~ * (contractible by the last step and ~ * since K0 \ @K = ;)
We leave for the reader to verify that such a sequence of collapses inward
and expansions from k linkKioeik to a point gives a sequence of collapses inward
and expansions from k starKioeik = oei* k linkKioeik to k linkKioeik, and hence
from kKik to kKi+1k. __
|__|
Proof of Spherical Quasifibration Theorem.We apply Babson's Criterion, first
with P = E0() and Q = B, then with P = E() and Q = B. In the first
case, let (M; X) 2 E0() and M0 2 B. Then X 2 V*(M)\{0} and ss-1(M0) ~=
V*(M0)\{0}.
If M M0, then ss-1(M0) \ E0()(M;X) is isomorphic to the poset of all
covectors X0 of M0 such that X X0. This is the poset of all covectors in M0
corresponding to cells of
" ___ " ___ "
BXM0= ( S+e) \ ( S-e) \ ( Se)
e2X-1(+) e2X-1(-) e2X-1(0)
given as a subcomplex of the pseudosphere picture of M0. By the Topological
Representation Theorem, this is either empty, a P L-sphere, or a P L-ball. It
can't be a sphere since X 6= 0 and it is non-empty by Lemma 3.8. Thus the
first condition of Babson's Criterion is fulfilled.
19
If M0 M, then ss-1(M0) \ E0()(M;X) is isomorphic to the poset of all
covectors X0 of M0 such that X0 X. This is the poset of all covectors in M0
corresponding to cells in the interior of the cell complex
0 " ___+ " ___-
BMX = ( Se ) \ ( Se )
e2X-1(+) e2X-1(-)
given as a subcomplex of the pseudosphere picture of M0. Now BM0Xmust be
empty or a P L-ball, but is in fact a P L-ball of full rank since this is a non*
*-empty
intersection (containing cells corresponding to covectors X0 X > 0). By the
previous lemma, the realization of the poset of cells in the interior of this b*
*all
is contractible. Thus the second condition of Babson's Criterion is fulfilled, *
*and
so the realization of the combinatorial sphere bundle is a quasifibration.
In the case of the disk bundle, the first condition to check is trivial, sin*
*ce
the poset in question will have a unique minimal element. The second condition
follows immediately from the proof of the second condition for sphere_bundles.
|__|
Corollary 3.11. Let B be a regular cell complex and Qk(B) be the set of B-
isomorphism classes of rank k spherical quasifibrations over B. The geometric
realization of the combinatorial sphere bundle gives a well-defined map
kE0k : Mk(B) ! Qk(B)
natural in B.
We now have two maps Vk(B) ! Qk(B), the map above and the map given
by deleting the zero section of a vector bundle. In Section 5 we will show they
coincide.
4 Stiefel-Whitney classes and Euler classes of
matroid bundles
Recall the axioms for Stiefel-Whitney classes [MS74 , x4]:
1.For any vector bundle = (p : E ! B) there are classes wi() 2
Hi(B; Z2), with w0() = 1 and wi() = 0 when i is larger than the fiber
dimension.
2.If f : B0! B is covered by a bundle map 0! , then wi(0) = f*wi().
P n
3.wn(0 1) = i=0wi(0) [ wn-i(1):
4.The first Stiefel-Whitney class of the canonical line bundle over RP 1 is
non-trivial.
20
The construction of the Stiefel-Whitney classes of a vector bundle = (p : E*
* ! B)
with fiber Rk given in [MS74 , x8] is
wi() = OE-1SqiOE(1) 2 Hi(B; Z2)
where Sqiis the i-th Steenrod square and
OE : H*(B; Z2) ! H*+k(E; E0; Z2)
is the Thom isomorphism. We next review the construction of Stiefel-Whitney
classes and Euler classes for spherical (quasi)-fibrations.
Thom Isomorphism Theorem. Let p0 : E0 ! B be a rank k spherical quasi-
fibration. Let p : E ! B be a quasifibration with contractible fiber and a fibe*
*r-
preserving embedding E0 ! E.
1.There is a class U 2 Hk(E; E0; Z2), so that for all b 2 B, inc*U 2
Hk(p-1b; p-10b; Z2) ~=Z2 is non-zero. Furthermore
OE : Hi(B; Z2)! Hi+k(E; E0; Z2)
ff7! p*ff [ U
is an isomorphism for all i.
2.If there is a class U 2 Hk(E; E0), so that for all b 2 B, inc*U 2
Hk(p-1b; p-10b) ~=Z is a generator, then
OE : Hi(B)! Hi+k(E; E0)
ff7! p*ff [ U
is an isomorphism for all i.
Proof. For any quasifibration f : X ! Y and point y 2 Y , there is a Serre
spectral sequence
Ei;j2= Hi(Y ; Hj(f-1 y)) =) Hi+jX
given by the Serre spectral sequence of the associated fibration ssf. If f were*
* a
fibration to begin with, there are, a priori, two different Serre spectral sequ*
*ences,
since f can be considered as a quasifibration or as a fibration. They coincide,
since if f is a fibration then f and ssf have the same fiber homotopy type.
The collapsing of the relative Serre spectral sequence
Ei;j2= Hi(B; Hj(p-1b; p-10b; Z2) =) Hi+j(E; E0; Z2)
gives the Thom isomorphism, and the Thom class U is the image of 1 under the
Thom isomorphism.
With integer coefficients, the same argument applies except that the E2-term
might have twisted coefficients. However the existence of an integral Thom clas*
*s__
in (2) guarantees that the coefficients are untwisted (look at E0;k2). *
*|__|
21
Definition 4.1.U is called the Thom class and OE is called the Thom iso-
morphism. In case 2 above, the spherical (quasi)-fibration is called orientable
and a choice of Thom class U 2 Hk(E; E0) is called an orientation.
Definition 4.2.Suppose = (p0 : E0 ! B) is either a vector bundle with
the 0-section deleted, a combinatorial sphere bundle, or a spherical (quasi)-
fibration. In the three cases respectively, let p : E ! B be the vector bundle,
the combinatorial disk bundle, or the obvious map p : E ! B from the mapping
cylinder E of p0. Then the i-th Stiefel-Whitney class of is
wi() = OE-1SqiOE(1) 2 Hi(B; Z2):
If p0 is oriented with Thom class U 2 Hk(E; E0), the Euler class
e() 2 Hk(B)
is the image of the Thom class under
Hk(E; E0) ! HkE ~=HkB:
We next wish to show that Stiefel-Whitney classes and Euler classes satisfy
the axioms and the usual properties, but first we had better make clear what is
meant by Whitney sum.
Definition 4.3.If 1 = (B; M1 : B ! MacP (i; E1)) and 2 = (B; M2 : B !
MacP (j; E2)) are two matroid bundles, then the Whitney sum 1 2 is the
matroid bundle (B; M1 M2 : B ! MacP (i + j; E1 q E2)) sending each b to
the direct sum M1(b) M2(b). If 1 = (p1 : E1 ! B) and 2 = (p2 : E2 ! B)
are two spherical (quasi)-fibrations then the Whitney sum is
1 2 = (p1 *B p2 : E1 *B E2 ! B);
where
E1 *B E2 = {[e0; e1; t] 2 E1 * E2 : p0(e0) = p1(e1) whenever t 6= 0; 1}
is the fiberwise join.
It is not difficult to show that the geometric realization of the combinator*
*ial
sphere bundle of a Whitney sum of matroid bundles is the Whitney sum of the
resulting spherical quasifibrations.
Proposition 4.4. The four axioms for Stiefel-Whitney classes are satisfied for
vector bundles, matroid bundles, and for spherical (quasi)-fibrations.
Proof. It suffices to prove the axioms hold for spherical fibrations. Axiom 1
holds since Sq0 = Id and Sqi is zero on Hk for i > k. Axiom 2 is clear by
construction.
The Whitney sum formula (Axiom 3) holds since the Thom class OE(1) for the
fiberwise join is the external product of the Thom classes of the two summands
and there is a sum formula for the Steenrod squares.
As for Axiom 4, one may compute w1 by restricting the canonical line bundle
to the circle. Here the bundle is the M"obius strip, which has non-trivial_w1 by
direct computation. |__|
22
Remark 4.5. While the axioms characterize the Stiefel-Whitney classes of vec-
tor bundles (due to the splitting principle), there is no assertion that the ax*
*ioms
give a characterization for the other categories of bundles.
We next show that w1() = 0 if and only if is orientable.
The oriented MacPhersonian O MacP (k; n) is defined in [And98 ]. The ele-
ments of the poset O MacP (k; n) are all chirotopes of elements of MacP (k; n).
In [And98 ] it is shown that kO MacP (k; n)k is the universal double cover of
k MacP (k; n)k. One can also define O MacP (k; 1) and show that its geometric
realization is the double cover of k MacP (k; 1)k.
Definition 4.6.An orientation of a matroid bundle = (B; M) is a poset
lifting
O MacP (k; 1)
% #
B M! MacP (k; 1):
Proposition 4.7. Any topological lifting of kMk : kBk! k MacP (k; 1)k to
kO MacP (k; 1)k is the geometric realization of an orientation of M : B! MacP *
*(k; 1).
Proof. Any topological lifting is simplicial, and any simplicial lifting to a p*
*oset
covering space is the realization of a poset lifting. (This is clear from looki*
*ng_at
the lifting on individual simplices.) |__|
Thus an orientation of a matroid bundle is equivalent to an orientation of
the geometric realization of the associated combinatorial sphere bundle.
Theorem 4.8. Let be a vector bundle, a matroid bundle, or a spherical
(quasi)-fibration. Then is orientable if and only if w1() = 0.
Proof. Suppose first that = (B; M) is a matroid bundle. From the double
cover result, H1(k MacP (k; 1)k) ~=Z2, and is generated by w1(flk) (where flk
is the universal bundle) since the first Stiefel-Whitney class is a non-trivial
characteristic class.
Note that the map kO MacP (k; 1)k ! k MacP (k; 1)k is an S0 bundle, and
hence has a classifying map into RP 1. Thus we have maps
kO MacP (k; 1)k ! S1
# #
kBk kMk! k MacP (k; 1)k c! RP 1
Let fi : kBk ! RP 1 be the composition of the lower two maps, ! be the
generator of H1(RP 1; Z2). Then by covering space theory fi has a lifting if
and only if fi*! = 0. One can see directly that the combinatorial vector bundle
corresponding to the M"obius strip mapping to the circle is non-orientable, and
thus c*! 6= 0. Hence c*! = w1(flk), and the result follows.
In the other cases, it suffices to consider a spherical fibration. One could
either use the classifying spaces BSG(k) and BG(k) for (oriented) spherical
23
fibrations and proceed as above, or use the fact the Sq1 is the mod 2 Bockstein
to show that w1() = 0 if and only if the coefficients in the spectral sequence
used in the Thom isomorphism theorem are untwisted. __
|__|
Finally, we note that proof of the Whitney sum formula for matroid bundles
also shows:
Proposition 4.9. Let 1 and 2 be matroid bundles with orientations. Then
e(1 2) = e(1) [ e(2):
In particular, the Euler class is an unstable characteristic class. Indeed, *
*if ffl
is a trivial (M is constant) bundle of rank greater than zero, then e( ffl) =
e() [ 0 = 0.
5 Vector bundles vs. matroid bundles: the Com-
parison Theorem
Comparison Theorem. Let B be a regular cell complex. The composite of
the natural transformations
Vk(B) C-!Mk(B) kE0k---!Qk(B)
coincides with the forgetful map given by deleting the zero section of a vector
bundle.
Thus the Stiefel-Whitney classes of the combinatorialization of a real vector
bundle coincide with those of the original bundle. In particular, as a corollary
we have Theorem A. In addition, since for every realized rank n oriented matroid
M the map G(k; Rn) ! k MacP (k; 1)k factors as
G(k; Rn) ! k(k; M)k ! k MacP (k; 1)k;
and since G(k; Rn) ! G(k; R1 ) gives a split surjection on mod 2 cohomology,
we have Theorem B.
Remark 5.1. The Comparison Theorem could also be stated universally by
saying that there are maps
BO(k) ! k MacP (k; 1)k ! BG(k)
covered by maps of spherical quasifibrations on the universal sphere bundles.
Let M be a rank n oriented matroid realized by a collection {OE1; : :;:OEm }*
* of
linear forms on Rn. Let S(k; Rn) be the sphere bundle of the canonical bundle
over G(k; Rn). An element of S(k; Rn) is a pair (V; p) where V is a k-plane
in Rn and p 2 V has unit length. Let E0(k; M) be the combinatorial sphere
24
bundle of the canonical bundle over (k; M). Define (V; p) = ((V ); X(p)) 2
E0(k; M), where : G(k; Rn) ! (k; M) is the function defined in Section 1.3,
X(p) = (signOE1(p); : :;:signOEm (p)) is a sign vector. Thus we have a commuta-
tive diagram
S(k; Rn)----! E0(k; M)
?? ?
yp ?yss
G(k; Rn)----! (k; M)
with and upper semi-continuous.
Lemma 5.2. Let M be a realized rank n oriented matroid. Then there is a
homotopy commutative diagram of continuous maps
S(k; Rn)---"-! kE0(k; M)k
?? ?
yp ?ykssk
G(k; Rn)---"-! k(k; M)k
so that there is a V 2 G(k; Rn) so that " gives a homotopy equivalence from the
fiber above V to the fiber above "(V ). Furthermore " is the map specified by
Lemma 2.10.
Proof. We again use Appendix A and the semi-algebraic triangulation theorem
of [Hir75]. As in the proof of Lemma 2.10, there is a semi-algebraic triangulat*
*ion
TG : kLk ! G(k; Rn) refining the stratification of G(k; Rn) and a map " :
G(k; Rn) ! k(k; M)k so that " O TG is simplicial. Now consider the semi-
algebraic stratification of S(k; Rn) given by the intersections of the preimages
of elements under and the preimages of simplices under TG-1O p. By [Hir75],
there is a triangulation TS : kKk ! S(k; Rn) refining this stratification, and *
*so
by Corollary A.4 there is a map " : S(k; Rn) ! kE0(k; M)k so that "O TS is
simplicial.
To see that these maps make the above diagram commute up to homotopy,
let s 2 S(k; Rn). Then s lies in a simplex S(k; Rn) of TS and there is
a simplex of TG so that p() G(k; Rn). By construction "() is
contained in the simplex spanned by the totally ordered set () and "() is
contained in the simplex spanned by the totally ordered set (). Then "(p(s))
and kssk("(s)) both lie in the closed simplex spanned by (), so there is a
straight-line homotopy from kssk O "to " O p.
Finally, note that for every vertex V 2 G(k; Rn) of TG , the topological
realization theorem gives a homotopy equivalence from the fiber over V to_the
fiber over "(V ) = (V ). |__|
25
Lemma 5.3. There is a homotopy commutative diagram
S(k; R1 )---"-! kE0(k; 1)k
?? ?
y p ?ykssk
G(k; R1 )---"-! k MacP (k; 1)k
so that there is a V 2 G(k; R1 ) so that " gives a homotopy equivalence from
the fiber above V to the fiber above "(V ). Furthermore " is in the homotopy
class of maps specified by Theorem 2.12.
Proof. The proof uses Theorem A.5 and the techniques of the proof of_the_
previous lemma. |__|
Proof of the Comparison Theorem.Let = (p : E ! B) be a vector bundle.
Convert the map kssk to a fibration ssksskand consider the diagram
E0() ----! S(k; R1 )---"-! kE0(k; 1)k --h--! Pkssk
?? ? ? ?
y ?yp ?ykssk ?ysskssk
B ---c-! G(k; R1 )---"-! k MacP (k; 1)k--Id--!k MacP (k; 1)k
By the homotopy lifting property, there is a map "0' hO" so that ssksskO"0 =
"O p. Furthermore, h gives a homotopy equivalence on fibers (since kssk is a
quasifibration) and " gives a homotopy equivalence on a fiber, so "0 gives a
homotopy equivalence on fibers.
Recall C[] is defined by applying the Simplicial Approximation Theorem to
find a subdivision B0 of B and a map c0: F(B0) ! MacP (k; R1 ), where kc0k is
homotopic to " O c. Then C[] = [c0].
We then have the following equations in Qk(B):
kE0k O C[] = kE0k[c0]
= [kE0kc0*(flk)]
= [kc0*E0(flk)k]
= [kc0k*kE0(flk)k]
= [c*"*kE0(flk)k]
= [c*"*Pkssk]
= [c*S(k; R1 )]
= [E0()]
|___|
26
6 Homotopy groups of the combinatorial Grass-
mannian
One can use the classical J-homomorphism to obtain limited information about
ssik MacP (k; n)k, or more generally about the homotopy groups of k(k; Mn )k
for realizable Mn . The idea is use vector bundles over spheres to construct
elements and use homotopy groups of spheres and characteristic classes to detect
them.
A few remarks will give a context for these results. The duality theorem
for oriented matroids gives k MacP (k; n)k ~=k MacP (n - k; n)k. It is easy to
show k MacP (k; n)k is connected, but there exist examples of Mn such that
k(n - 1; Mn )k is disconnected ([MRG93 ]). In [And98 ], it was shown that
ss1k MacP (k; n)k ~=ss1G(k; Rn), and stability results for large n were establi*
*shed.
There are also some results on homotopy type of combinatorial Grassmannians
for small values of k or n and for oriented matroids with few elements (cf. [MZ*
*93 ],
[Bab93 ], [SZ93 ]).
See [Whi78 ] for the definition of the J-homomorphism Ji;k: ssiO(k) !
ssi+kSk. The limit as k ! 1 is denoted
Ji: ssiO ! ssSi:
Stability results for the domain and range of J show
ImJi;k! Im Ji;k+1 is an epimorphism if k i + 1 and
ImJi;k! Im Ji is an isomorphism if k > i:+ 1
A group H is a subquotient of a group G if H is isomorphic to a subgroup
of a quotient group of G.
Theorem 6.1. Let Mn be a realized rank n oriented matroid. Let p be a point
in the image of " : G(k; Rn) ! k(k; Mn )k.
1.Im Ji-1;kis a subquotient of ssi(k(k; Mn )k; p) when n - k i.
2.Im Ji-1 is a subquotient of ssi(k(k; Mn )k; p) when n - k i and k > i.
Proof. Let G(k) denote the monoid of self-homotopy equivalences of Sk-1, given
the compact-open topology. Its classifying space BG(k) classifies rank k spher-
ical (quasi)-fibrations [Sta63]. G0(k + 1) denotes the monoid of self-homotopy
equivalences of Sk which fix a base point *.
The result follows from commutativity up to homotopy of the diagram
G(k; Rn) ---"-! k(k; Mn )k
?? ?
yfi ?yffi
G(k; R1 )---fl-! BG(k) ---ffl-!BG0(k + 1);
the surjectivity of ssi(fi) when n - k i, and the identification of ssi(ffl O *
*fl)
with Ji-1;k. The map fi is given by inclusion, the map fl by classifying the
27
canonical bundle minus its zero section, and the map ffl is B applied to the
injection of monoids G(k) ! G0(k) given by suspension. The map ", the map
ffi, and the homotopy ffi O " ' fl O fi, are given by our three main theorems:
the Combinatorialization Theorem, Spherical Quasifibration Theorem, and the
Comparison Theorem (see also Lemma 5.2).
The surjectivity of ssi(fi) when n - k i follows either from the Cellular A*
*p-
proximation Theorem applied to the Schubert cell decomposition, or by viewing
G(k; n) as O(n)=(O(k)xO(n-k)) and using the homotopy long exact sequence
of a fibration.
Now ssiG(k; R1 ) ~=ssi-1O(k) since there is a fibration O(k) ! V (k; R1 ) !
G(k; R1 ) and the Stiefel manifold is contractible. Also ssiBG0(k+1) ~=ssiG0(k+
1), which is in turn ssi+kSk by the adjoint property of smash and mapping spaces
in the category of based CW complexes. Thus we have identified the domain
and range of ssi(ffl O fl) with that of Ji-1;k, and the identification of the_t*
*wo maps
consists of tracing through these identifications. |__|
Corollary 6.2. Let Mn be a realized rank n oriented matroid. Let p be a point
in the image of " : G(k; Rn) ! k(k; Mn )k.
1.
8
><0 if k = 1
ss2(k MacP (k; n)k) ~=ss2(G(k; Rn)) ~=> Z if k = 2 and n - k 2
: Z2 if k 3 and n - k 3.
2.ssk(k(k; Mn )k; p) has Z as a subgroup when k is even and n 2k.
3.ssi(k(k; Mn )k; p) has Z2 as a subquotient when i 1; 2 (mod 8), n-k i,
and k i.
4.ss4m (k(k; Mn )k; p) has Zam as a subquotient when m > 0, n - k 4m,
and k 4m. Here am is the denominator of Bm =4m expressed as a
fraction in lowest terms, and Bm is the m-th Bernoulli number.
Proof. 1. By [And98 ], the combinatorialization map ss2(G(k; Rn)) ! ss2(k MacP *
*(k; n)k)
is surjective, so (1) follows from (2) and (3). An alternate route to (1) is to*
* use
characteristic classes to detect elements of ss2k MacP (k; n)k, by applying the
Euler class and second Stiefel-Whitney class of the combinatorialization of the
complex Hopf bundle over the 2-sphere.
2. There is a characteristic class version and a homotopy theoretic version *
*of
the proof. The characteristic class proof is as follows. Consider the oriented *
*com-
binatorial Grassmannian "(k; Mn ), defined analogously to O MacP (k; n). The
forgetful map k"(k; Mn )k ! k(k; Mn )k is a double cover, so ssk(k(k; Mn )k; p)*
* ~=
ssk(k"(k; Mn )k; p0) for each lifting p0 of p. The tangent bundle of the k-sphe*
*re
combinatorializes to a map Sk ! "G(k; Rn) -"!k"(k; Mn )k. The evaluation of
the Euler class of the tangent bundle of a manifold on the fundamental class of
that manifold is the Euler characteristic of the manifold (cf. 11.12 in [MS74 ]*
*). In
28
particular, the Euler class of the tangent bundle of an even-dimensional sphere
represents twice the generator of Hk(Sk) ~=Z. The Euler class applied to ori-
ented vector bundles over k-spheres can be viewed as a homomorphism from
ssk(BSG(k)) to Hk(Sk), where BSG(k) classifies rank k oriented spherical fi-
brations. Thus the combinatorialization of this tangent bundle generates an
infinite subgroup of ssk(k"(k; Mn )k; p0)).
The homotopy theoretic version is to consider the Hopf invariant
H : ImJk-1;k! Z:
The domain of Jk-1;kis ssk-1(O(k)), which classifies k-bundles over Sk. Now
H O Jk-1;kis the Euler class, so 2Z Im H, using the tangent bundle of the
k-sphere again. (For the reader's edification, we note the Hopf invariant is on*
*to
if and only if k = 2; 4 or 8, as can be seen by Bott periodicity or by Adams's
work on Hopf invariant one.)
3.,4. This follows from Theorem 6.1 and the deep homotopy theoretic com-
putation of Im J due to Adams [Ada65 ]. The result is that Im Ji-1 is Z2 for_
i 1; 2 (mod 8), Zam for i = 4m, and is zero otherwise. |__|
An exposition of Bernoulli numbers and topology is given in [MS74 , Ap-
pendix B]. The first few values of am are :
__a1_|a2__|_a3__|a4__|a5__|__a6___|a7_|_a8___
24 | 240 |504 |480 | 264 |65520 |24 | 16320
By the stability results of [And98 ], the results of our corollary also appl*
*y to
the MacPhersonian MacP (k; 1).
7 Vector fields and characteristic classes
The classical motivation for characteristic classes was as obstructions to the
existence of linearly independent vector fields of a manifold, or more generall*
*y,
independent sections of a vector bundle. This section gives a combinatorial
analog.
Definition 7.1.The combinatorial Stiefel manifold Vl(k; n) is the poset of
all M 2 MacP (k; n + l) satisfying both of the conditions below
1.rank(M\{n + 1; : :;:n + l}) = k
2.{n + 1; : :;:n + l} is independent in M.
Note that deleting {n + 1; : :;:n + l} gives a surjective map Vl(k; n) !
MacP (k; n) if l k.
Definition 7.2.If (B; ) is a matroid bundle, an independent set of l vector
fields is a lifting
Vl(k; n)
% #
B ! MacP (k; n):
29
Lemma 7.3. If a matroid bundle = (B; M : B ! MacP (k; n) admits a set
: B ! Vl(k; n) of l independent vector fields, then:
1.The map
Q : B ! MacP (k - l; n)
oe 7! (oe)={n + 1; : :;:n + l}
is a matroid bundle.
2.If ffll is the trivial rank l bundle over B sending each cell to the rank l
oriented matroid with elements {n+1; : :;:n+l}, then the matroid bundles
and Q ffll are kBk-isomorphic.
Q is called the quotient bundle of .
Proof. (1) This follows immediately from Lemma 3.7
(2) The proof is by induction on l, and relies on the Order Homotopy Lemma.
If l = 1, first note that (oe) (oe) for all oe, so ' . (Here ' means kBk-
isomorphic. All homotopies occur in k MacP (k; n + l)k; there is no need to go *
*to
1.) Thus, it suffices to show that (oe) Q(oe) ffl1(oe) for all oe. Any covect*
*or
X x Y 2 V*(Q(oe) ffl1(oe)) is built from covectors X 2 V*(Q(oe)) V*((oe))
and Y : {n + 1} ! {+; -; 0}. Since {n + 1} is independent in (oe), there is a
Ye 2 V*((oe)), so that eY Y . Thus X O eYis a covector of (oe) which is greater
than or equal to X x Y .
For l > 1, let 0 be the set of l - 1 independent vector fields obtained
from by deleting n + l from each oriented matroid (oe), and let Q0 be the
resulting quotient bundle. Note that the vector field l on Q0given by n + l is
non-vanishing and Q is the quotient bundle of Q0by l. Hence we have
' Q0 ffll-1 by the induction hypothesis
' Q ffl1 ffll-1 by the l = 1 case.
|___|
Theorem 7.4. If a rank k matroid bundle admits an independent set of l
vector fields, then wk-l+1() = 0.
Proof of Theorem 7.4.If is a set of l independent vector fields in and Q is
the resulting quotient bundle, then by the above lemma, ' Q ffll. Thus by
the Whitney sum formula, w() = w(Q)w(ffll) where w = 1+w1+w2+. . .is the
total Stiefel-Whitney class. But w(ffll) = 1 since ffllis trivial, and wk-l+1(Q*
*)_= 0
since Q is a rank k - l bundle. |__|
Similarly, we have
Theorem 7.5. If a rank k matroid bundle admits a non-zero cross section
(i.e. an independent set of 1 vector field) then the Euler class e() vanishes.
30
8 Some open questions
There are open questions everywhere you spit; we list a few.
1.Is a CD manifold a Poincare complex? Does a CD manifold satisfy
Poincare duality?
2.Give a definition of isomorphism of CD-manifolds; show that a diffeo-
morphism class of smooth manifolds determines an isomorphism class of
CD-manifolds.
Discussion: Problems (1) and (2) are manifold theoretic analogues of the
bundle theoretic results of this paper. Macpherson [Mac93 ] defined a CD
manifold and showed how a smooth manifold with a smooth triangulation
determines a CD manifold. He asked whether a CD-manifold was a topo-
logical manifold; this was shown in a special case in [And99b ].
3.Are there exotic mod 2 characteristic classes?
4.Are there Pontrjagin classes?
Discussion: The authors together with Eric Babson have outlined a con-
struction of rational Pontrjagin classes. If these classes were integral c*
*o-
homology classes, that would imply the existence of exotic CD 7-spheres.
5.Is MacP (1; 1) an infinite loop space?
6.Compute the homotopy groups of MacP (1; 1).
Discussion: Solving questions 5 and 6 would show that combinatorial vec-
tor bundles give a generalized cohomology theory and compute the coeffi-
cient groups (isomorphism classes of matroid bundles over spheres.)
A Topological maps from combinatorial ones
Section 1.3 described a natural map : G(k; Rn) ! (k; M) for any rank n
oriented matroid M with a fixed realization in Rn, given by intersecting the
hyperplanes of the realization with V 2 G(k; Rn) and taking the corresponding
oriented matroid. This appendix describes how we use this map to make a
simplicial map from a triangulation of G(k; Rn) to (k; M), unique up to
homotopy. We also construct a topological map G(k; R1 ) ! k MacP (k; 1)k
which is in some sense the limit of the simplicial maps obtained when (k; M) =
MacP (k; n). As discussed in [AD ], this limit map is not PL.
Actually, we will work more generally, considering maps from spaces to
posets satisfying certain properties.
Definition A.1. A triangulation of a topological space X is a homeomor-
phism T : kKk ! X where K is a simplicial complex. We will abuse language
slightly and refer to the image T (koek) of a simplex koek under a triangulation
T as a simplex of T . If X resp. Y are spaces equipped with triangulations S
31
resp. T then a map f : X ! Y is simplicial if T -1O f O S is. A subdivision
of a simplicial complex K is a homeomorphism S : kK0k ! kKk where K0 is a
simplicial complex and for every simplex oe02 K0, there is a simplex oe 2 K so
that S(koe0k) koek and S is linear on koe0k. The triangulation T 0= T O S is a
subdivision of the triangulation T . An example of such is the barycentric
subdivision T : kKk ! X. Two triangulations of a space X are equivalent
if they have a common subdivision. A PL space is a space equipped with a
fixed equivalence class of triangulations, called the PL triangulations. A map
f : X ! Y between PL spaces is a PL map if there are PL triangulations so
that f is simplicial.
Definition A.2. Let : G ! M be a function from a space to a poset. The
partition {-1(m) : m 2 M} of G is the stratification of G induced by
. The map is upper semi-continuous if every g 2 G has a neighborhood
U so that (U) M(g) . Thus the closure of a stratum -1(m) maps to
the lower order ideal Mm of M. A triangulation T : kKk ! G refines the
stratification if the interior of every simplex maps under to a single element
of the poset M.
Lemma A.3. Let : G ! M be upper semi-continuous and let T : kKk ! G
be a triangulation refining the stratification. Then for any simplex ff of the
barycentric subdivision, (T kffk) is totally ordered.
Proof. For a simplex oe of K, let denote both the barycenter of the simplex
koek kKk and the corresponding vertex of kKk. Let ff = {oen > . .>.oe0} 2
K be a chain of simplices of K. Note that
kffk (intkoenk) [ k{oen-1 > . .>.oe0}k
hence inductively,
(T kffk) = {(T ); : :;:(T )}:
Since the triangulation refines the stratification and is upper semi-continuou*
*s,_
(T ) (T ) for all i. |__|
Corollary A.4. Let : G ! M be upper semi-continuous and let T : kKk ! G
be a triangulation refining the stratification.
1.There is a map "T : G ! kMk, simplicial with respect to the barycentric
subdivision of T , which agrees with on the vertices of T .
2.If T 0is a subdivision of T , then "T ' "T0.
Proof. Part (1) follows from the last lemma by defining "T on vertices and
extending by linearity with respect to kKk. Part (2) follows from a straight-li*
*ne_
homotopy t"T (a) + (1 - t)"T0(a) across the simplices of kMk. |__|
We need an infinite version of the corollary.
32
Theorem A.5. Let G1 G2 . .b.e a sequence of PL spaces, M1 M2 . . .
a sequence of posets, and 1 : G1 ! M1; 2 : G2 ! M2; : :a:sequence of upper
semi-continuous maps so that i|Gi-1 = i-1. Let T1; T2; : :b:e a sequence of
PL triangulations of G1; G2; : :r:efining the stratifications given by 1; 2; : *
*:,:
so that the restriction of the i-th triangulation is a subdivision of the (i - *
*1)-st
triangulation. Let M be the union of the Mi's and G be the direct limit of the
Gi's.
1.There is a continuous map " : G ! kMk which, for every i, restricts to
a map "i: Gi ! kMik which is simplicial with respect to the barycentric
subdivision of Ti, and so that "i agrees with i on the vertices of Ti in
Gi\Gi-1.
2.Let T10; T20; : :b:e another sequence of PL triangulations satisfying the *
*same
hypotheses as T1; T2; : :.:Let "0: G ! kMk be the map given by part (1)
using the Ti0's. Then " ' "0 and the homotopy restricts to a homotopy
"i ' "0ifor all i.
Proof. We will inductively define maps Si: Gi! Mi so that
1.Si|Gi-1= Si-1;
2.For a vertex v 2 Gi\Gi-1 of Ti,
Si(v) = {(v)};
3.For a point p 2 Giin the interior of a simplex of Tiwhich is spanned by
vertices {v0; : :;:vn} Gi,
Si(p) = [jSi(vj):
Assume inductively that Si-1 has been defined satisfying (1), (2), and (3),
and also assume inductively that Si-1 satisfies property
4. max Si-1(p) = (p) forp 2 Gi-1.
We then use property (1) to define Si on Gi-1 and properties (2) and (3) to
define Si on Gi\Gi-1. We need to verify three things: first that property (3)
continues to hold for Si and p 2 Gi-1, second that Si(p) is totally ordered for
p 2 Gi\Gi-1, and third that property (4) holds. We leave the verification of the
first part to the reader.
We now show that Si(p) is totally ordered for p 2 Gi\Gi-1. Suppose p is in
the interior of a simplex of Ti with vertices
{v0; : :;:vn} [ {w0; : :;:wm }
with the v's in Gi-1 and the w's in Gi\Gi-1. Then by choosing a point q in
the interior of the simplex spanned by the v's and by the (omitted) proof that
property (3) holds for Si and q, we see [jSi(vj) is totally ordered. The proof
33
of Lemma A.3 shows that [jSi(wj) = {(w0); : :;:(wm )} is totally ordered.
Because Gi-1 is a subcomplex of the triangulation Ti, because we have taken
a barycentric subdivision, because is upper semi-continuous, and because
Ti refines the stratification, (wj) (vk) for all j and k. Hence Si(p) is
totally ordered. Finally property (4) holds since we have taken a barycentric
subdivision.
Next we use Si to define "iinductively. For p 2 Gi-1, let "i(p) = "i-1(p).
For a vertex p 2 Gi\Gi-1 of Ti, define "i(p) = (p). Then for p in the
interior of a simplex spanned by {v0; : :;:vk}, define "i(p) by linearity, noti*
*ng
inductively that i(p) is in the closed simplex spanned by Si(p). This completes
the proof of part (1) of the Theorem.
For part (2), it suffices to consider the case when each Ti0subdivides Ti.
Then note that S0i(p) Si(p), so we can use the straight-line homotopy across
the simplex spanned by Si(p). __
|__|
B Babson's criterion
This appendix gives the criterion of Babson for the realization of a poset map
to be a quasifibration. We first state two results of Quillen [Qui73, page 98],
specialized from general categories to posets.
Quillen's Theorem A. Let f : P ! Q be a poset map. If for all q 2 Q,
kf-1 (Qq )k is contractible, then kfk : kP k ! kQk is a homotopy equivalence.
A commutative square of spaces
A ---ff-!B
? ?
fi?y ?yfl
C ---ffi-!D
is homotopy cartesian if for all c 2 C, the induced map on homotopy fibers
ss-1fi(c) ! ss-1fl(ffic)
is a homotopy equivalence.
Quillen's Theorem B. Let f : P ! Q be a poset map such that for every
inequality q q0 in Q, the inclusion map kf-1 (Qq )k ! kf-1 (Qq0 )k is a
homotopy equivalence. Then for any q 2 Q,
kf-1 (Qq )k ----! kP k
?? ?
y ?y
kQq k ----! kQk
is homotopy cartesian.
34
Since the geometric realization of a poset can be identified with the real-
ization of the poset obtained by reversing the inequalities, one may reverse the
inequalities in Quillen's Theorem A and B. Similar remarks apply to the various
lemmas below.
In Babson's thesis, both of the previous two results of Quillen were recast.
Lemma B.1. If f : P ! Q is a poset map satisfying both of the conditions
below, then kfk is a homotopy equivalence.
1.kf-1 qk is contractible for all q 2 Q.
2.kf-1 q \ Pp k is contractible whenever p 2 P , q 2 Q, and q f(p).
Proof. By Theorem A applied to f and condition 1, we only need show that the
realization of the inclusion i : f-1 q ! f-1 (Qq ) is a homotopy equivalence for
all q 2 Q. But this follows from condition 2 and by applying Theorem A to i,
noting that
i-1(f-1 (Qq )p ) = f-1 q \ Pp
|___|
Recall that a chain in a poset P is a non-empty, totally ordered, finite sub*
*set
of P , and P denotes the poset of chains in P , where the partial order is given
by inclusion. Note kP k is the barycentric subdivision of kP k. If f : P ! Q
is a poset map, then there is a poset map f : P ! Q sending a chain c to
the chain f(d). Note that for a chain d of Q, the symbols f-1 d and (f)-1d
have different meanings.
Lemma B.2. If f : P ! Q is a poset map satisfying the conditions below, then
kfk is a quasifibration.
1.kf-1 q \ Pp k is contractible whenever p 2 P , q 2 Q, and q f(p).
2.kf-1 q \ Pp k is contractible whenever p 2 P , q 2 Q, and q f(p).
Proof. We wish to apply Quillen's Theorem B to the induced map of posets
f : P ! Q. For a chain d 2 Q, there is a retraction
r : (f)-1(Qd )! (f)-1d
c7! c \ f-1 d:
By the order homotopy lemma, krk is actually a deformation retraction since
r(c) c for every c 2 (f)-1(Qd ).
For an inequality d d0in Q, consider the following commutative diagram:
(f)-1d ----! (f)-1(Qd )
? ?
ff?y ?y
(f)-1d0 ---r- (f)-1(Qd0 )
where ff(c) = c \ f-1 d0, r is the retraction and the other two arrows are incl*
*u-
sions. To verify the hypothesis of Quillen's Theorem B applied to f, it suffices
to prove the following lemma.
35
Lemma B.3. Let f : P ! Q be a poset map satisfying the conditions of Lemma
B.2. Given an inequality d d0 in Q, the geometric realization of the map
ff(d d0) : (f)-1d! (f)-1d0
c 7! c \ f-1 d0
is a homotopy equivalence.
Proof. To prove this it suffices to check only those inequalities in Q given by
deleting the smallest or largest element of a chain. Let d = (qr > . .>.q1 > q0)
be a chain in Q. Let d0 = (qr > . .>.q2 > q1) and d00= (qr-1 > . .>.
q1 > q0). We will use Lemma B.1 to prove that kff(d d0)k and kff(d d00)k
are homotopy equivalences. Note ff(d d0)-1c0 is isomorphic to (f-1 q0 \
Pminc0 ) by the isomorphism c 7! c \ f-1 q0, and the geometric realization of
(f-1 q0 \ Pminc0 ) is contractible by condition (1).
Now if c0 c \ f-1 d0 where c and c0 are chains which map to d and d0
respectively,
ff(d d0)-1c0\ ((f)-1d)c
has a maximum element, namely (c \ f-1 q0) [ c0, so its geometric realization is
contractible. Thus by Lemma B.1, kff(d d0)k is a homotopy equivalence. The
proof that kF (d d00)k is a homotopy equivalence is similar and uses_property_
(3) of Lemma B.2. |__|
Now we return to the proof of Lemma B.2. Note that for d 2 Q, the
geometric realization of the diagram
(f)-1d ----! (f)-1(Qd )
?? ?
y ?y
d ----! Qd
is homotopy cartesian. Thus for all vertices and barycenters of kQk the inclusi*
*on
of the fiber of kfk : kP k ! kQk in the homotopy fiber is a homotopy equivalenc*
*e.
Since this is true for the barycenters and since kfk is a simplicial map, this *
*is __
true for all points in the interior of a simplex, so kfk is a quasifibration. *
* |__|
References
[AD] L. Anderson and J. F. Davis. Schubert stratifications and triangula-
tions of the infinite Grassmannian. To appear.
[Ada65] J. F. Adams. On the groups J(X). IV. Topology, 5:21-71, 1965.
[And] L. Anderson. Representing weak maps of oriented matroids. To
appear in European Journal of Combinatorics.
36
[And98] L. Anderson. Homotopy groups of the combinatorial Grassmannian.
Discrete Comput. Geom., 20:549-560, 1998.
[And99a] L. Anderson. Matroid bundles. In New Perspectives in Algebraic
Combinatorics, MSRI book series. Cambridge University Press, 1999.
[And99b] L. Anderson. Topology of combinatorial differential manifolds. Topol-
ogy, 38(1):197-221, 1999.
[Bab93] E. Babson. A combinatorial flag space. PhD thesis, MIT, 1993.
[BLS+ 93]A. Bj"orner, M. Las Vergnas, B. Sturmfels, N. White, and G. M.
Ziegler. Oriented matroids, volume 46 of Encyclopedia of Mathematics
and its Applications. Cambridge University Press, 1993.
[Dol63] A. Dold. Partitions of unity in the theory of fibrations. Ann. of Math.
(2), 78:223-255, 1963.
[DT56] A. Dold and R. Thom. Une generalisation de la notion d'espace fibre.
Application aux produits symetriques infinis. C. R. Acad. Sci. Paris,
242:1680-1682, 1956.
[FL78] J. Folkman and J. Lawrence. Oriented matroids. J. Combin. Theory
Ser. B, 25:199-236, 1978.
[GM92] I. M. Gelfand and R. D. MacPherson. A combinatorial formula for
the Pontrjagin classes. Bull. Amer. Math. Soc. (N.S.), 26:304-309,
1992.
[Hir75] H. Hironaka. Triangulations of algebraic sets. In Algebraic geometry
(Proc. Sympos. Pure Math., Vol. 29, Humboldt State Univ., Arcata,
Calif., 1974), pages 165-185. Amer. Math. Soc., 1975.
[Mac93] R. D. MacPherson. Combinatorial differential manifolds. In Topolog-
ical methods in modern mathematics (Stony Brook, NY, 1991), pages
203-221. Publish or Perish, 1993.
[Mil59] J. Milnor. On spaces having the homotopy type of a CW-complex.
Trans. Amer. Math. Soc., 90:272-280, 1959.
[MRG93] N. Mn"ev and J. Richter-Gebert. Two constructions of oriented ma-
troids with disconnected extension space. Disc. and Comp. Geometry,
10(3):271-286, 1993.
[MS74] J. W. Milnor and J. D. Stasheff. Characteristic classes. Number 76
in Annals of Mathematics Studies. Princeton University Press, 1974.
[MZ93] N. Mn"ev and G. Ziegler. Combinatorial models for the finite-
dimensional Grassmannians. Discrete Comput. Geom., 10(3):241-
250, 1993.
37
[Qui73] D. Quillen. Higher algebraic K-theory, I: Higher K-theories. In Proc.
Conf., Battelle Memorial Inst., Seattle, Wash., 1972, number 341 in
Lecture Notes in Mathematics, pages 85-147. Springer-Verlag, 1973.
[Sta63] J. Stasheff. A classification theorem for fibre spaces. Topology, 2:23*
*9-
246, 1963.
[SZ93] B. Sturmfels and G. Ziegler. Extension spaces of oriented matroids.
Discrete Comput. Geom., 10(1):23-45, 1993.
[Whi78] G. W. Whitehead. Elements of Homotopy Theory. Springer-Verlag,
1978.
Department of Mathematics
Texas A&M University
College Station, TX 77843
laura.anderson@math.tamu.edu
http://www.math.tamu.edu/~ laura.anderson/
Department of Mathematics
Indiana University
Bloomington, IN 47405
jfdavis@indiana.edu
http://www.indiana.edu/~ jfdavis/
38