arXiv:1503.04564vl [math.LO] 16 Mar 2015
A CLASSIFICATION OF 2-CHAINS HAVING 1-SHELL
BOUNDARIES IN ROSY THEORIES
BYUNGHAN KIM, SUNYOUNG KIM, AND JUNGUK LEE
Abstract. We classify, in a non-trivial amenable collection of
functors, all 2-chains up to the relation of having the same 1-shell
boundary. In particular, we prove that in a rosy theory, every
1-shell of a Lascar strong type is the boundary of some 2-chain,
hence making the 1st homology group trivial.
We also show that, unlike in simple theories, in rosy theories
there is no upper bound on the minimal lengths of 2-chains whose
boundary is a 1-shell.
1. Introduction
In [5], [6], J. Goodrick, A. Kolesnikov and the first author developed
a homology theory for any amenable collection of functors in a very
general context. But the most interesting examples appear in model
theory. Namely, given any strong type p G S(A) in a rosy theory T,
we may assign a non-trivial amenable collection of functors preserving
thorn-independence and compute the corresponding homology groups.
By the general theory, if T has n-complete amalgamation (n > 2) over
A = acl(A) then the (n — l)-th homology group of p G S(A) consists
of ( n — l)-shells with the support n + 1 = {0,... ,n}. Hence, in any
simple T (where, due to 3-amalgamation, every 1-shcll is the boundnary
of some 2-simplex), the 1st homology group is trivial. But the question
remained whether the same would hold in rosy theories. In this paper,
we show that the answer is yes (as long as p is a Lascar type). A crucial
ingredient in our proof is the fact that a and b realize the same Lascar
type if and only if their Lascar distance is finite, i.e., dA{a,b) < oj. In
the proof, the number of 2-simplices involved in a 2-chain having the
1-shcll boundary is proportional to dA(o,,b). Therefore one may guess
that, there does not exist a uniform bound for the minimal lengths of 2-
chains with 1-shell boundaries for various Lascar types in rosy theories,
The first author was supported by NRF of Korea grants 2011-0021916 and
2013R1A1A2073702. The second and third authors were supported by NRF of Ko¬
rea grants 2012-030479 and 2012-044239. The third author was supported by Euro¬
pean Community’s Seventh Framework Programme [FP7/2007-2013] grant 238381.
1
2
BYUNGHAN KIM, SUNYOUNG KIM, AND JUNGUK LEE
contrary to the case of simple theories where the bound is 1, due to
3-amalgamation. A series of rosy examples in [2] where the Lascar
distances increase are candidates. However in order to confirm that
in each example that a candidate 2-chain has the expected minimal
length, we need to rule out all other possibilities. For this goal we start
to classify all the 2-chains having the same 1-shcll boundary in a very
general amenable context. The classification also has its own research
interests. We obtain some interesting and important results in regard
to the classification.
There are basically two operations on the class of 2-chains preserv¬
ing the length and boundary of a chain. The first one is called the
crossing (CR-)operation and the second one is called the renaming-of-
support (RS-)operation. Two 2-chains are said to be equivalent if one
is obtained from the other by finitely many applications of the two
operations.
In the remainder of this section, we recall the definitions of an
amenable family of functors and the corresponding homology groups
introduced in We thank Hyeung-Joon Kim and John Goodrick
for their valuable suggestions and comments.
Notation. Throughout the paper, s denotes an arbitrary finite set
of natural numbers. Given any subset X C V(s), we may view X
as a category where for any u,v G X , Mor(w,r>) consists of a single
morphism l U jV if u C v, and Mor(w,u) = 0 otherwise. If /: X —> C is
any functor into some category C then for any u,v G X with u C v,
we let /“ denote the morphism f(t u ,v) G Mor c(f(u),f(v)). We shall
call X C P(s) a primitive category if X is non-empty and downward
closed , i.e., for any a,uG ’P(s), if u C v and v G A" then u G X. (Note
that all primitive categories have the empty set 0 C cu as an object.)
Remark/Definition 1.1. Given any primitive categories X and Y,
define
X + Y := {t U k | t G X, k G Y}
which is clearly a primitive category itself containing X and Y as sub¬
categories. And, for any t G X, define
X t := {k G A | t n k = 0} and X\ t := {k G X t \ t U k G X}
both of which are clearly primitive subcategories of X. Observe:
(1) X\ t C X t C X
(2) X C X t + V(t)
( 3 ) x\ t = U{v(u\t) \tcue x}.
A CLASSIFICATION OF 2-CHAINS HAVING 1-SHELL BOUNDARIES IN ROSY THEORIES
Moreover, it is easy to check that the following are equivalent:
x = x t + v(t) <*=* x t = x\ t « x = U {V{u) | tcuex}.
If one of these equivalent conditions holds, we shall say that X splits
at t.
For any functor /: X —> C to some category C and for any t € X, the
localization of f att is the functor f\ t : X\ t —>■ C defined as follows: for
any u C v e X\ t , (/| t )“ = and f\ t (v) = f(t U v).
Definition 1.2. Let X C V(s) and Y C V(t) be any primitive cat¬
egories (where s,t are some finite sets of natural numbers). And let
/: X —> C and g: Y —» C be any functors to some category C.
(1) We say that / and g are isomorphic if there is an order-preserving
Injection r: s —> t such that Y = {t(u) \ u E X} and there is
a family of isomorphisms { h u : f(u) —> g(r(u)) \ u € A"} C
Mor(C) such that, for any u Cue X,
h o f u — o h
u v u J v — g T ( v ) u U u-
(2) We say that / and g are permutations of each other if there is a
Injection cr: s —» t (not necessarily order-preserving) such that
Y = {cr(u) | u G X} and, for any u C v e Y , g(v) = /(<t _1 (u))
and (<?)“ = fp-ij^y I n this case, we write g = f o cr -1 .
Note that, if / and g are permutations of each other via an order¬
preserving map cr: s —>■ t, then / and g are isomorphic.
Definition 1.3. Let A be a non-empty family of functors from various
primitive categories into some fixed category C. We say that A is
amenable if it satisfies the following properties:
(1) (Closed under isomorphism and permutation) If / e A then
any functor g which is isomorphic to / or is a permutation of /
also belongs to A.
(2) (Closed under restriction and union) For any functor /: X —» C
from some primitive category X into C,
/ e A <=> for every t € X , / ( V(t) € A.
(3) (Closed under localization) If /: X —> C is any functor in A
then for every t € X, f\ t : X\ t -A C is also in A.
(4) (Extensions of localizations are localizations of extensions) Let
/: X —y C be any functor in A which splits at some t e X.
Then whenever f\ t can be extended to some functor g: Z —>• C
in A where t fl 1J Z — 0, / can be extended to some functor
h : V(t) + Z —> C in A such that h\ t = g.
4
BYUNGHAN KIM, SUNYOUNG KIM, AND JUNGUK LEE
Definition 1.4. By a ( regular ) n-simplex in a category C , we mean a
functor /: V(s) —>■ C where s C u has the size n + 1. We call s the
support of f and denote it by supp(/).
Definition 1.5. Let A be an amenable family of functors into some
category C. Let B e Ob(C). If / is a functor in A such that /(0) = B ,
we shall say that / is over B. And we define:
S n (A; B ) {f & A \ f is a regular n-simplex over B }
C n (A ; B) := the free abelian group generated by A n (^l; B).
The elements of C n (^4; B) are called the n-chains over B in A. For
each i — 0,..., n, we define a group homomorphism
d l n : C n (A: B) —>• C' n _ 1 ( v 4; B)
by letting, for any n-simplex /: 'P(s) —> C in S n (A; B ) where s = {s 0 <
A },
^(/) :=/r^(a\W)
and then extending linearly to all n-chains in C n (A: B). Then we define
the boundary map
d n :C n (A;B)^C n ^(A;B)
by
9n(c) := 5] (-1 )%(c).
0<i<n
We shall often refer to d n (c) as the boundary of c. Next, we define:
Z n (A\ B) \= Ker d n
B n (A ; B) \= Im d n+1 .
The elements of Z n (A; B) and B n (A] B) are called n-cycles and n-
boundaries , respectively. It is straightforward to check
dn —i ° d n — 0.
Hence we may define
H n (A, B) := Z n (A] B)/B n (A] B)
called the n-th (simplicial) homology group of A over B.
Notation 1.6. (1) For c G Z n (A ; B), [c] denotes the coset of B n (A ; B)
containing c.
(2) When n is clear from context, we shall often omit n from d l n
and d n , writing simply as d l and d.
A CLASSIFICATION OF 2-CHAINS HAVING 1-SHELL BOUNDARIES IN ROSY THEORIES
(3) When we write an n-chain c G C n (A; B) as
k
C =E
i =1
we shall assume, unless stated otherwise, that n^’s are nonzero
integers and ff s are distinct n-simplices. (This form is called
the standard form of a chain.) For such an n- chain c, we define
the length of c and the support of c as |c| := Xu=i \ n i\ and
supp(c) := Uti{supp(/ i )}, respectively.
(4) For c,d 6 C n (A',B), we say that d is a subchain (or sub sum¬
mand) of c if they are in the standard forms
k
c = ^2 Uifi and d = E
i=l i£J
where J C {1 ,,k} and, for each i G •/, n, • m* > 0 and
|mj| < |rii|.
Remark/Definition 1.7. Let c be any n-chain and let d be a sub-
summand of c. For any n-chain d', we shall say that the n-chain
d := c — d + d'
is obtained by replacing the subsummand d in c by d!. Note that, if
\d'\ < \d\ then |c'| < |c|.
Remark/Definition 1.8. Given any bijection a: co —> oj (not neces¬
sarily order-preserving), we may induce an automorphism c*: C n (A: B ) —)■
C n (A: B ) for each n as follows: for any n-chain c = fT ll nifi G G n (^4, B ),
where each f % is an n-simplex with Sj := supp(/j) = {s ii0 < • • • <
we let cxj := a \ Si and ti := cxi(si) = {t^o < ■ ■ ■ < We dehne
<T*{c) ■= E 71 *! 0 ^ °
i
(see Dehnition ll.2lOL with \ai\ := sign(cr') (= ±1) where o[ G Sym(n +
1) such that for j < n, ^(sjj) = For example
= \&i\fi o af 1 .
Moreover, a* commutes with the boundary map, i.e., d o a* = cr* o d.
This can be verified inductively by first checking the case where cr is a
transposition.
Next we dehne the amalgamation properties. For n = {0,..., n — 1},
we let V~(n) := V{n) \ {n}. i.e., V~{n) is the set of all the proper
subsets of 7i.
6
BYUNGHAN KIM, SUNYOUNG KIM, AND JUNGUK LEE
Definition 1.9. Let A be an amenable family of functors into a cate¬
gory C.
(1) A has n-amalgamation (n > 1) if every functor /: V~{n) —> C
in A can be extended to some functor g: V(n) —» C in A.
(2) A has n-complete amalgamation (written n-CA) if it has k-
amalgamation for every 1 < k < n.
(3) A has strong 2-amalgamation if, whenever /: V(s) —>• C and
g: V(t) C are simplices in A which agree on V(s fl t), then
there exists some simplex h: V(sUt) —> C in A extending both
/ and g.
Remark 1.10. It is easy to verify that, for any amenable family A:
(1) strong 2-amalgamation =>■ 2-amalgamation.
(2) (1-amalgamation + strong 2-amalgamation) =>■ A has n-simplices
for every n > 0.
Definition 1.11. An amenable family of functors is called non-trivial
if it has 1-amalgamation and strong 2-amalgamation (in particular, it
has 2-CA).
Definition 1.12. An n-chain c G C* n (^l; B) is called an n-shell if it is
in the form
c = ± Y. (-^Vi
0<i<n+l
where s are n-simplices satisfying
d 1 fj = cp-'fi whenever 0<i<j<n + l.
We define E n (A ; B) := {cG Cn^A] B) \ c is an n-shell }.
It is straightforward to verify the following proposition.
Proposition 1.13. (1) E n (A', B) C Z n (A]B).
(2) For every f e S n (A ; B), d n (f ) G S„_i(^4; B).
(3) If c — ± (—1)*/« an U n-shell, then \ supp(c)| = n + 2.
0<i<n+l
Moreover, there exists a unique functor g : P~(supp(c)) —* C in
A extending all the fi’s. More precisely, if we let supp(c) =
{s 0 < • • • < s n+ i}, then g \ P(supp(c) \ {sj}) = ./( for each i.
(4) A has (n + 2)-amalgamation if and only if for any n-shell c,
there exists some (n + 1 )-simplex d such c = ±d(d).
Definition 1.14. An amenable family of functors has weak 3- amalgamation
if each 1-shell is the boundary of some 2-chain c with |c| < 3.
A CLASSIFICATION OF 2-CHAINS HAVING 1-SHELL BOUNDARIES IN ROSY THEORIES
The following result due to 0,0 illustrates the importance of the
notion of shell.
Fact 1.15. [5] © Let A be any non-trivial amenable family of functors.
If A has (n + l)-CA for some n > 1, then
H n (A ; B ) = {[c] | c G E n (A; B ), supp(c) = {0,..., n + 1} }.
In particular,
(1) Hi(A; B) = 0 E^A; B) c B^A] B)
(2) If A has weak 3 -amalgamation then Hi(A] B ) = 0.
In the remainder of the paper, A shall denote a non-trivial
amenable family of functors into a category C.
Now we begin to talk about the prototypical examples of an amenable
family of functors : complete types in rosy theories. In the sequel
we work with a large saturated model Ai = Al eq and its the¬
ory T which is rosy. Recall that a theory is called rosy if there is
a ternary independence relation As on the small sets of its model, sat¬
isfying the basic independence properties. (See [I], pf] for the precise
definition.) We take As here to be thorn-independence. Any simple or
o-minimal theory is known to be rosy. Moreover, if a simple theory T
has elimination of hyperimaginaries then non-forking independence is
equal to thorn-independence. So we assume that any simple T in
this paper has elimination of hyperimaginaries. (Of course this
is just for convenience as we can work in A l heq without the assump¬
tion.) In particular, we assume that 3-amalgamation holds over any
algebraically closed set in simple T.
We fix any algebraically closed small subset B C Ai and consider the
category Cb whose objects are all the small subsets of Ai containing
B, and whose morphisms are elementary maps over B (i.e., fixing B
pointwise). We also fix any p{x) G S(B) (where x could be an infinite
tuple). When / is any functor from a primitive category X into Cb
and u C v G A", we shall abbreviate /“(/(«)) as ff{u).
Definition 1.16. By a closed independent functor in p(x), we mean
a functor / from some primitive category X into Cb satisfying the
following:
(1) Whenever {z} C uj is an object in X, we can choose a realization
b |= p{x) such that, if we let C := /®. j(0) then /({«}) = acl (Cb)
and b As B C.
BYUNGHAN KIM, SUNYOUNG KIM, AND JUNGUK LEE
(2) Whenever u(^ 0) C u is an object in A", we have
/(«) — acl ((J/FfW)
\iGu
and {/i^({i})| i G u} is independent over /®(0).
We let A(p) be the family of all closed independent functors in p.
Fact 1.17. [6] A{p) is a non-trivial amenable family of functors.
Notation 1.18. We shall abbreviate S n (A(p); B), C n (A(p); B),... as
S n A(p), C n A(p ),.... We shall also abbreviate H n (A(p)] B) simply as
H n {p). Other than this, we use standard notation. For example a =a b
denotes tp (a/A) = tp (b/A); and a =\ b denotes Ltp(a/H) = Ltp(6/A),
i.e., the Lascar (strong) types of a, b over A are the same.
2. H\ (p) IN ROSY THEORIES
If a theory T is simple then due to 3-amalgamation and Fact 11.151
we know Hi(p) = 0. In this section we show the same holds for any
rosy T as long as p is a Lascar type.
Let /: A" —* C B be any functor in A(p) with /(0) — B. If u G X with
u = {i 0 < ■ ■ ■ < ik}, we shall write f{u) = [a 0 ,..., a*], where aj |= p,
f(u) = acl (B, a 0 ■ ■ ■ a*), and acl (ajB) = fu h \{ij}). Thus {a 0 ,..., a^.}
is independent over B.
Theorem 2.1. If B is a model, then A(p) has weak 3-amalgamation
over B (so H x (p) = 0).
Proof. Let f — a 12 — ao 2 + aoi be any 1 -shell in EiA(p) where each
Uij: V({i,j}) —> Cb is a 1-simplex. We want to find a 2-chain g with
length 3 such that dg = /. For this goal there is no harm in assuming
that a 0 i({l}) = [a] = ai 2 ({l}) and ai 2 ({2}) = [ b} = a 02 ({2}). Let
a 0 i({ 0 }) := [c] and ao 2 ({ 0 }) := [c'j, and let q be a coheir of p over
Babcc'. Choose any c" |= q. Then d' abed (see [3]) and cd' = B
dc". Now let g := a ±23 — 0023 + aoi 3 where a ^-3 are 2-simplices hav¬
ing support {i,j, 3} extending a^- such that ai 23 ({l, 2, 3}) = [a, b, c"],
a 0 23({0, 2, 3}) = [d,b, c"], a 0 i 3 ({0,1, 3}) = [c, a,d']. Hence we may as¬
sume <9 0 (a 023 ) = <9°(ai 23 ) and <9 0 (a 0 i 3 ) = d 1 (ai 23 )- But cd' = B dc"
implies that we may further assume ^(aois) = 9 1 (a 0 23)- Therefore
dg — f as desired. □
Remark 2.2. Of course the same proof shows that weak 3-amalgamation
(over a model) holds not only in A(p) but more generally inside A4
(with arbitrary vertices).
A CLASSIFICATION OF 2-CHAINS HAVING 1-SHELL BOUNDARIES IN ROSY THEORIES
Recall that, for any tuples a and b, we write ds{a , b ) < n iff there is
a sequence of tuples c 0 ,... ,c n with c 0 = a and c n = b , such that each
CjCj + i begins some B -indiscernible sequence. The smallest such n (if it
exists) is denoted by de(a, b) (called the Lascar distance between a and
b). Recall the fact that a — f b iff de(a, b) < u> in any rosy theory.
Lemma 2.3. Let I = (a 0 ,ai,...) be any B-indiscernible sequence.
Then for any cq there is c=b cq such that c^b ao^i and cao =b ca\.
Proof. Extend / to I' indiscernible over B having a sufficiently large
length. Then by the extension axiom there is c' =b Cq such that
c'-J-b/'. Moreover, by the pigeonhole principle, there are a^a,,- G /'
(i < j ) such that c'cq =b c'aj. Now, by R-indiscernibility, there is c
such that caocq =b c'ataj. Then c is the desired tuple. □
Theorem 2.4. Suppose thatp is a Lascar strong type. Then H\{p) = 0.
Proof. For notational simplicity we let B — 0. As in the proof of
Theorem 12.11 given any 1-shell / = cq 2 — ao 2 + aoi hi EiA(p) where
each aij: P({i,j}) —* Cb is a 1-simplex, we want to fold a 2-chain g
such that dg — f. Again there is no harm in assuming that «oi({l}) =
[a] = ai 2 ({l}) and ai 2 ({2}) = [b] = a 02 ({2}). Let a 0 i({0}) := [c]
and a 02 ({0}) := [c'j. By extension we can further assume {a, b, c, c'}
is independent. Now c,c' (= p and let d(c, c') = n. So there are
c — Cq, ... ,c n — d such that c,q + i begins an indiscernible sequence,
for i < n. We can further assume that ab^ cc i CiC n _i; so ab dL c Q ■ ■ ■ c n .
Then by Lemma [2731 there are e* |= p (i < n) such that CjCj+i
and e, : Cj = ejCj +1 (*). Again by extension we suppose ah ^' c , c . t+1 e«, so
that each of the {a, c*, , {a, c i+1 , ef\ is independent. Moreover each
{a, e n _i, b}, {e n _i, c n , b} is independent as well (**).
Now there is g 0 := gf — gf where gf,gf are 2-simplices with sup¬
port {0,1,3} such that ^({0,1, 3}) = [co,a, eo] and #,{({0,1,3}) =
[ci,a,e 0 ]; d°gf = d°gf] d l gf = d 1 gf (this is possible by (*)); and gf
extends aoi (i.e., d 2 gf = aoi). Hence dgo = aoi — d 2 gf.
By iteration we can End := gf — g~ (0 < i < n — 1) where gf , gf
are 2-simplices with support {0,1, 3} such that gf ({0,1, 3}) = [q, a, ef\
and #u({0,1,3}) = [c m ,a,ej; d°g + = d°g ~; d 1 g + = d 1 g~ (this again
is possible by (*)); and d 2 gf = d 2 gf_ 1 . Therefore we have
d(do + • • • + g n - 2 ) = aoi — d 2 g n _ 2 -
The rest of the proof is similar to that of Theorem 12.11 We put
9n-i '■= 9n—i — ® 023 T®i 23 where aj 23 is a 2-simplex with support { j , 2, 3}
extending a j2 such that a 023 ({0, 2, 3}) = [c n , 6 , e n _i], ai 23 ({l, 2, 3}) =
[a, b, e n _i] (see (**)). Also gf_ { is a 2 -simplex with #)}_ 1 ({ 0 , 1 , 3}) =
10
BYUNGHAN KIM, SUNYOUNG KIM, AND JUNGUK LEE
[Cn-i, a, e n _i] extending d 2 g n _ 2 - Moreover again by (*), we have d 1 gif_ l =
<9 1 Q'023- Thus it follows
dg n -1 = 9~g+_ i — ao2 + a 12 = 9 2 g n _ 2 — ao2 + «i 2 •
Therefore for g := g 0 + • • • + g n - 1, we have dg — f as desired. □
3. A CLASSIFICATION OF 2-CHAINS WITH A 1-SHELL BOUNDARY
In this section, we bring our attention back to a non-trivial amenable
family of functors A and classify 2-chains of A having 1-shell bound¬
aries. Basically we show that any 2-chain having a 1-shcll boundary is
equivalent to one of two types of 2-chains, called the NR-type and the
RN-type.
We start by introducing two operations on 2-chains called the cross¬
ing operation and the renaming-of-support operation, respectively. For
any distinct real numbers x and y, we shall abbreviate the open interval
(min{x, y}, max{x, y}) as [(x,y)] = [{y,x)j.
Definition 3.1. Let v e C^A, B) be a 2-chain and let w := 6101+6202
be a subsummand of v , where cq’s are 2-simplices with for i = 1,2,
6j = ±1, supp(oj) = {^i,^2,^j} being all distinct numbers) such
that 01 and 02 agree on the intersection of their domains, namely
V({£i, ^2})- Further assume that, if we let 7 := cq ( P^i, ^2}), then
7 does not appear in d(w), i.e., the two 7 terms in d(w) have opposite
signs and cancel each other.
Now by strong 2-amalgamation, there exists some 3-simplex /i ex¬
tending both Oj. For i = 1, 2, let fa := /j \ V({k\, k 2 , £i}) and
, fe2 A + 61 2 if e+2 = —1, and exactly one of &2, h belongs to [(ki, £2)]
6i pi + 62 P2 otherwise.
Then the operation of replacing the subsummand w in v by w' is
called the crossing operation (or simply CR-operation).
Example 3.2. Let /o, /1, / 2 , h be 2-simplices with supp(./)) = {0,1, 2, 3}\
{i}. Assume that fi and fj agree on their intersection, for every pair
i,j. Consider the 2-chain c = / 0 — /1 + /2. Then we can apply the
CR-operation to the subsummand / 0 — f\ to obtain a new 2-chain
c = (-/ 2 + h) + f'i or simply / 3 .
This example illustrates in particular that a CR-operation may not be
reversible, i.e., once we apply a CR-operation to a 2-chain, we may not
be able to recover the original 2-chain by applying more CR-operations
(unless we allow 2-chains to be written redundantly as / 3 — / 2 + f-i).
A CLASSIFICATION OF 2-CHAINS HAVING 1-SHELL BOUNDARIES IN ROSY THEORIES
Next, we define an operation on n-chains called the renaming-of-
support operation.
Definition 3.3. Let c be an n-chain in C n (A', B) and let d be a sub¬
summand of c. Let j G supp(d) such that j f supp(<9 n (d)). (In this
situation, we say that d has a vanishing support, namely j, in its bound¬
ary.) Choose any k £ supp(c) and any bijection a: u> —* u which sends
j i—>■ k but which fixes the rest of the elements in supp(c). Then
the operation of replacing the subsummand d in c by cr*(d) is called
the renaming-of-support operation (or simply RS-operation). (See Re¬
mark/Definition [L8] to recall the definition of cr*.)
Remark 3.4. When we apply the CR- and RS-operation to some sub¬
summand of an n-chain c, the resulting n-chain has the same boundary
as c (guaranteed by the fact that cr* commutes with the boundary map
d) and has a shorter or equal length as c (by Remark/Definition 11.71
and the clear fact that cr* preserves the lengths of n-chains).
Remark/Definition 3.5. A 2-chain c is called proper if its length
|c| does not change after any finitely many applications of CR/RS-
operations to its subsummands. It is clear that any 2-chain may be
reduced to a proper 2-chain after finitely many applications of the
two operations. Any CR-operation (also RS-operation) applied to any
proper 2-chain is in fact reversible. This allows us to define an equiva¬
lence relation ~ among proper 2-chains by: c ~ d <=>• c can be obtained
from c' by finitely many applications of the CR/RS-operations to its
subsummands. Note that c ~ c! implies d[c ) = d(c') and |c| = \d\.
We are now ready to introduce the notions of renameable type and
non-renameable type for 2-chains having 1-shell boundaries.
Definition 3.6. Let a be a 2-chain having a 1-shell boundary.
(1) We say a is of renameable type (or simply RN-type ) if some
subsummand of a has a vanishing support. Otherwise, a is
said to be of non-renameable type (or simply NR-type ).
(2) a is called minimal if it is proper, and for any proper a' equiv¬
alent to a, there does not exist any subsummand j3 of a' such
that d(/3) = 0.
Remark 3.7. Suppose that a is a 2-chain having a 1-shcll boundary.
(1) Note that a is of NR-type iff none of the CR or RS-operation
is applicable to a, i.e. nothing else is equivalent to a except a
itself. So an NR-type chain is minimal.
12
BYUNGHAN KIM, SUNYOUNG KIM, AND JUNGUK LEE
As was the case in Example 13.21 an RN-type a can sometimes
be transformed to an NR-type by CR-operations. But if a is
proper then its RN/NR-type is preserved under equivalence.
(2) We can always find some minimal 2-chain a' such that d(a) =
d{a!). Such an a! can be obtained from a by finitely many
applications of CR/RS-operations and deleting subsummands
having trivial boundary.
There is a 2-chain (3 with |/3| = 5 having a 1-shell boundary
such that any subsummand of (3 does not have the trivial bound¬
ary but f3' with \/3 '| =5 obtained from /3 by the CR.-operation
has a subsummand with the boundary 0 .
(3) If a is minimal then any a ' equivalent to a is minimal as well
(of course \a\ = \a'\ and d(a ) = d(a!) too).
Notation. Let / be any simplex. For any subset {jo, • • • ,jk} Q supp(/),
we shall abbreviate / \ V({jo ,... ,jk}) as p 0, '"^ k . Also, given a chain
c = ^2 i€l n ifi (in its standard form), and any subset {jo, ■ ■ ■ ,jk} Q
supp(c), we shall write to denote the subchain where
J:={i el | supp (ft) = {j 0 ,..., jjfc}}.
Example 3.8. Of course any 2-simplex is of NR-type. The following
is an NR-type 2-chain with length 5: Let a = cq + a 2 + a 3 — a 4 — a 5 be
a 2 -chain with 2 -simplices a* having supp(cq) = { 0 , 1 , 2 } such that;
Ui , do — cla , do — Uc are distinct,
• a 2 , a j = a 5 , a 3 = a 4 are distinct;
1 0,1 0,1 0,1 0,1 0,1
• and so are a 3 , aj = a 4 , a 2 = a 5 ’ .
Then a is of NR-type with a 1-shcll boundary a{
T o
0,1
3 '
Before stating our Erst main theorem of the classification, we intro¬
duce a notion called chain-walk which will be used in our proof.
Remark 3.9. Recall that if a is a 2-chain with a 1-shcll boundary,
then its length is always an odd positive number.
For the rest of this section, we fix a 1-shell boundary /i 2 — / 02+/01
with supp (f jk ) = {j < k}.
Definition 3.10. Let a be a 2-chain having the boundary /i 2 — /o 2 +/oi-
m
A subchain f3 = e A °f a (where e, = ±1 and bi is a 2 -simplex, for
i=0
each i) is called a chain-walk in a from f m to — f 02 if
( 1 ) there are non-zero numbers k 0 ,..., k m+ 1 (not necessarily dis¬
tinct) such that k 0 = 1 , k m+ 1 = 2 , and for i < m, supp(frj) =
{fci,A: i+ i,0};
A CLASSIFICATION OF 2-CHAINS HAVING 1-SHELL BOUNDARIES IN ROSY THEORIES
(2) (<9e 0 &o) 0,1 = foi% (<9e m fe m ) 0 ’ 2 = -/ 02 ; and
(3) for 0 < i < m,
(deA)° M+1 + (de i+1 b i+ i)°’ ki+1 = 0.
m
The sum eA with its order is called a representation of the chain-
i =0
walk (5. Unless said otherwise a chain-walk is written in the form of
a representation. Notice that a chain-walk may have more than one
representation. For example, a reordering of terms in f3 above may also
satisfy conditions (l)-(3). By a section of the chain-walk (3, we shall
mean a subchain of [3 in the form
m!
& :=£ eA for some 0 < j < m' < m
i=j
and the sequence (kj, kj + 1 ,..., k m >, k rn i + \) is called the walk sequence of
fd'. A chain-walk [d in a is called maximal (in a) if it has the maximal
possible length. We say a is centered at 0 if some (hence every) maximal
chain-walk in a from /01 to —/02 is, as a chain, equal to a.
We similarly define such notions as a chain-walk in a from —/02 to
fi 2 , a is centered at 2, and so on.
Remark 3.11. In the definition above, if 6 is a chain-walk in a from
foi to —fo 2 , then 0 G supp(&,;) for all i, but 0 ^ supp(<9/3 — /01 + / 02 ); and
the walk sequence of /d is a sequential arrangement of (supp(hj) \ {0})’s
without repetition of the overlapped support.
Note now that given any 2-chain a as in the definition above, since
there are only finitely many 2-simp lex terms in a, we can always find
a chain-walk say, from / 0 i to —fo 2 '- We start with any 2-simplex term
b in a such that d 2 b = foi and then keep finding a term in a (with the
coefficient) cancelling out adjacent 1-simplcx boundaries. This process
must stop with a term having —/02 as its boundary.
Even if 0 is in the support of every simplex term of a, it need not
be centered at 0: Let a = c 0 + C\ — c 2 such that dc 0 = g\ 2 — /02 + foi ;
da = fi 2 ~ 902 + , 9 or; and d(-c 2 ) = -912 + 9 o 2 - 9o 1 , where f tJ ± g iy
Then Co itself is a maximal chain-walk in a from f m to — f 02 - Note
that a = c 0 + Ci — c 2 is not a chain-walk from / 01 to — f 02 , whereas it
is a chain-walk from /j 2 to / 01 , i.e, a is centered at 1 .
Lemma 3.12. Let a be a 2-chain with the 1-shell boundary / 12 — / 02 +
m
foi. Let fd = e A a chain-walk in a, say from —/02 to /j 2 . Assume
i =0
m!
there is a section fd' — Jj) eA of fd such that for supp( 6 j) = { 2 , fcj, k i+ 1 },
14
BYUNGHAN KIM, SUNYOUNG KIM, AND JUNGUK LEE
either h, &W +1 for all i = j,, m'; or ki kj for all i — j +
1,..., ml + 1. Then by finitely many applications of CR-operations to
j3 r , we obtain a simplex c with supp(c) = {2, kj, k m f +1 } such that, for
j-1 m
some e = ± 1 , (3" := e A + ec + e A 5 ^// a chain-walk from
i= 0 %>m'
fo2 tO /i 2 .
Proof. When j = mf there is nothing to prove. Assume the lemma
holds when m'—j = n. Let us show that the lemma holds when m'—j =
n + 1. Assume ki k, m / + \ for all i = j,..., m!. Then we can apply the
CR-operation to + e m >b m ', and we get + e' m ,b' m ,
with supp(&A-i) = {2,k m i-i,k m i + i}, having the same boundary. Due
m! —2
to the induction hypothesis applied to e A + 4 '-iA-d we are
i=j
done. When k t ^ kj for all i = j + 1,..., m! + 1, we apply the CR-
operation to €jbj + €j + ibj + i, and similarly we are done. □
Remark/Definition 3.13. In Lemma [3. 121 we call f3", a reduct of [3.
The walk sequence of (3" is also called a reduct of the walk sequence
of (3. So given a chain-walk its reducts are also chain-walks, which are
obtained by the repeated applications of the CR-operation as described
in Lemma 13.121
Theorem 3.14. Let a be a minimal 2-chain with the boundary / 12 —
fo2 + foi-
(1) Assume a is of NR-type. Then |a| = 1 or |a| > 5. If |a| >
5 then any chain-walk in a from / 01 to — f 02 is of the form
2 n
l)*a* which is as a chain equal to a such that /i 2 = a^-
i= o
for some 1 < j < n — 1 .
(2) a is of RN-type iff a is equivalent to a 2-chain
2n—l
of = do + + ® 2 n.
2=1
(-n > 1 ) which is a chain-walk from / 0 i to — / 02 such that
d°a 2n = fi 2 , d 1 (a 2n ) = -fo 2 and supp(a 2n ) = { 0 , 1 , 2 }. (The
representation of a' is called standard.J
Proof. (1) As mentioned in Remark 13.111 a chain-walk [3 in a from
/oi to —/02 exists. Now since a is of NR-type, supp(o) = {0,1,2}. If
\f3 1 < | a | then it follows a—f3 has a vanishing support 0 , a contradiction.
Hence \a\ = \f3\ and a = (3. Suppose now that |cr| = 3. So the chain-
walk is a 0 — ai + a 2 = a and either d°a 0 = / i2 or d°a 2 = f\ 2 . Then
A CLASSIFICATION OF 2-CHAINS HAVING 1-SHELL BOUNDARIES IN ROSY THEORIES
either d°ai = d°a 2 or <9 0 «o = d°a\. In either case, the subchain of a
has a vanishing support 1 or 2, a contradiction. Hence \a\ = 1 or > 5.
When |a| > 5, all we need to show is that /i 2 ^ d°a 0 and f 12 ^ d°a 2n .
If fi 2 = d°a 0 then a — a 0 has a vanishing support 1 , a contradiction.
Hence f\ 2 ^ d°a 0 . Similarly, we can show f\ 2 ^ d°a 2n .
(2) It follows supp(<9(a' — a 2n )) = {0,1}, i.e., a' — a 2n has
a vanishing support, so a' is of RN-type. Since CR/RS-operations
preserve the minimality and the chain types, a is also an RN-type.
(=^) We prove this in a series of claims. Note that |a| > 3.
Claim 1. There is a 2 -chain on ~ a which is centered at 2 such that
| supp(ai)| > 3.
Proof of Claim 1. Let a 2 a if |supp(a)| > 3. Otherwise since
a is of RN-type, we can apply RS-operations to obtain some a 2 ~ a
with |supp(a 2 )| > 3. Now there is /3 := a maximal chain-
iei
walk in a 2 from — f 02 to f\ 2 . If fi — a 2 we put a\ := a 2 and we are
done. Otherwise let 7 := a 2 — (3, and then 7 has a vanishing support
2 in its boundary. By applying the RS-operation to 7 we find 7 ' with
2 f supp( 7 ') such that a 2 ~ a' 2 := /3 + 7 '.
Assume now inductively we can find a desired an ~ of 2 when | 7 / | = m.
Let ly'l = m + 1. Note that f\ 2 — /o 2 + foi ^ d(/3), since otherwise
< 9 ( 7 ') = 0 contradicting the minimality of a’ 2 . Hence there is io € I with
supp(6j 0 ) = {2,no,ni} such that / 0 i) with a coefficient, stays
in d((3). Therefore there must be a term 6j 0 bj 0 (e JO G {1, —1}) in 7 ' such
that (2 ^)supp( 6 JO ) = {no, ni, n 2 } and b'f°’ ni = b n C ni is cancelled out
in d(ei 0 bi 0 + 6j 0 bj 0 ). Now applying the CR-operation to e tQ b lQ + e ](j b jQ , we
gef e'i o b' io +e' jo b' jQ with supp( 6 'J = { 2 , n 0 , n 2 }, supp( 6 ' o ) = { 2 ,m,n 2 },
preserving the boundary. Then from f3, we obtain (3' by substituting
e' o b' l(: + e'j Q b'j o for e io b io . Notice that f3' is still a chain-walk from — f 02
to /i 2 while a 2 ~ a" := f3' + ( 7 ' — £j 0 bj 0 ). Hence by the induction
hypothesis there is a desired cd ~ a 2 . We have proved Claim 1.
Claim 2. There is a 2 -chain a 2 ~ cti that has a 1 -simplcx term c
(with the coefficient 1 ) such that supp(c) = { 0 , 1 , 2 }, and f\ 2 — /o 2 =
<9°(c) — ^(c).
Proof of Claim 2. For notational simplicity, let (0,1, 2, 3} C supp(a 1 ),
2 n
and write an = e * c u a chain-walk from — f 02 to f\ 2 . So for some
i=0
jo < 2 n, we have e jo = 1, supp(c io ) = {0,1,2}, and <9 2 (c io ) = f 01 .
io-l 2 n
Let /9 0 := £iCi, and /A := 6^. We shall find the desired c
i= 0 i=j o+l
16
BYUNGHAN KIM, SUNYOUNG KIM, AND JUNGUK LEE
Figure 1. A standard RN-type 2-chain
(and a 2 ) by applying the process in Lemma 13.121 and finding reducts
of chain-walks, starting from aq. Each time, the reduced chain-walk
together with the deleted terms is equivalent to aq.
Case 1) 3 ^ supp(/9i): So 3 G supp(/3 0 ). Now let Jo := (0,..., 3,..., 0)
be the walk sequence of /3o, and let I\ = ( 1 ,..., 1 ) be the walk sequence
of d\. So J 0 /i is the walk sequence of aq. Now J 0 = J$J\ such that J\
starts with 3 but all other components 7 ^ 3. Then due to Lemma [3.121
(applied to Jih), we can find 71 , a reduct of aq, whose walk sequence
is J 0 (3,1). Now J 0 = (0,...).
If 3 is not in J 0 then again by Lemma 13.121 we can further find a
reduct of 71 whose walk sequence is (0, 3,1), then again further reduce
it with the walk sequence ( 0 , 1 ), and we are done.
If 3 is in J 0 then in general, by finding a sequence of all 3’s in J 0
and applying Lemma 13.121 we can reduce Jo to a sequence of the form
Jq = (03, Aq3, Ay3,...; Ay) where each k j 7 ^ 3. If none of the /q’s is 0
then by applying Lemma 13.121 again to Jg(3, 1) we directly reduce it
to (0,1) and we are done. Otherwise, one of the Ay’s is 0, and we can
similarly reduce Jg to a sequence of the form (03, 03,...; Ay). Now the
reduced walk sequence is (03, 03,...; Ay; 3,1). If Ay 7 ^ 1 then it can
directly be reduced to (0,1) and we are done. If Ay = 1 then it can be
reduced to (0,1; 3,1) and further reduced to (0, 3,1) and to (0,1), so
we are done.
Case 2) 3 ^ supp(/3 0 ): Then 3 G supp(/3i) and the proof will be
similar to Case 1.
Case 3) 3 G supp(/d 0 ) 0 supp( / di): By an argument similar to that
in Case 1, the walk sequence of aq can be in general reduced to I =
(03, 03,...; k, 3)(0,1)(3, k 31,..., 31). Now by the argument in the
last part of the proof of Case 1, (03, 03,...; Ay 3)(0,1) can be reduced
to (0,1). Hence / can be reduced to (0,1)(3, k'\ 31,..., 31). Then by
A CLASSIFICATION OF 2-CHAINS HAVING 1-SHELL BOUNDARIES IN ROSY THEORIES
the same argument it can finally be reduced to ( 0 , 1 ), and we have
proved Claim 2.
Now lastly we simply take a chain-walk 7 from / 0 i to — / 02 in a 2
terminating with c (= the 1-simplex described in Claim 2). Then by
an argument similar to that in the proof of Claim 1, we repeatedly
apply the CR-operation to 7 (while keeping c unchanged), and obtain
a desired a' ~ ct 2 centered at 0 forming a chain-walk from / 0 1 to —/o 2 -
Then we take the reverse order of the representation of the chain-walk
a'. □
In an upcoming paper [7], it is shown that for any minimal 2-chain
whose boundary is a 1 -shell, there is an equivalent 2 -chain which has
the same boundary with support size three.
4. Examples
This section is devoted to exhibiting a certain family of examples
of 2 -chains of types in rosy theories whose boundaries are 1 -shells.
The existence of these examples implies that, in rosy theories, there is
no uniform bound for the minimal lengths of 2 -chains having 1 -shell
boundaries.
We recall the examples described in [2]. For a positive integer n,
consider a (saturated) structure M n = (|M n |; S, g n ), where \M n \ is a
circle; S' is a ternary relation such that S(a,b,c ) holds iff a,b,c are
distinct and b comes before c going around the circle clockwise starting
at a; and g n is a rotation (clockwise) by 27r/n-radians. When n is
obvious from context, g n is often written as g. The following Fact 14.11
14.21 are from [2].
Fact 4.1. (1) Th (M n ) has the unique 1-complete type p n (x ) over
0, which is isolated by the formula x = x.
(2) Th(M n ) is K ^-categorical and has quantifier-elimination.
(3) For any subset A C M n , acl(A) = dcl(kl) = \J 0<i<n g l n {A ) (in
the home-sort), where g l n = g n 0 • • • 0 g n -
i times
(4) Foreacha G M n withn > 1, and an integer i, S(g l (a),x, g l+1 (a))
isolates a complete type over a.
In what follows, we assume n > 1.
Fact 4.2. (1) There are a,b E M n such that d(a,b ) > n/2.
(2) For any a,b E M n , the following are equivalent:
(i) a,b begin some 0 -indiscernible sequence,
18
BYUNGHAN KIM, SUNYOUNG KIM, AND JUNGUK LEE
(ii) a and b have the same type over some elementary substruc¬
ture of M n ,
(Hi) a = b V S (a, b, g n (a )) V S(b, a, g n (b )) holds.
Thus the unique 1-complete type p n is also a Lascar type.
Theorem 4.3. (1) Th(M n ) has weak elimination of imaginaries.
(2) Th (M n ) is rosy having thorn U-rank 1 with a trivial pregeome¬
try.
Proof. (1) We claim that if a set D in ( M n ) k is definable over Ao and
Ai respectively where A , = acl(Hj) = dcl(Hj) (in the home-sort) then
it is definable over B := A 0 fl Ap. We sketch the proof of the claim by
freely using Fact 14.11 Let k = 1. Due to quantifier elimination, D is
some union of finitely many arcs on M n . Clearly each end-point of a
connected component of D is in dcl(H,;) and so in B as well. Hence D
is indeed H-definable. Now for induction, assume the claim holds for
k — 1 . We want to show it holds for k. Suppose that ..., Xk, of)
defines D where a* G H, : . Then, for each element b, the set Db de¬
fined by (pi(xi, ..., Xk-i, b, at) is definable over Bb , by the induction
hypothesis. But due to No-categoricity (so there are only finitely many
formulas over 0 up to equivalence), it easily follows that for each y,
tpi(x i,..., Xk-i, y, of) is definable over B , i.e. D is definable over B as
we wanted.
Now let E(x, y ) be an 0-definable equivalence relation on (M n ) fc . For
a G ( M n ) k , let a 1 denote a finite tuple of algebraic closure of a in the
home-sort. Let b be the maximal subtuplc of a' which is algebraic over
a/E. Thus there is a" = ac i (a/E) of such that b — a! C I a" as sets. Hence
due to the claim, a/E G dcl cq (6) and b G acl (a/E). We have proved
(1) '
(2) Due to (1), Th (M n ) is rosy having thorn U -rank 1 as pointed
out in [TJ. Notice that M n has the same pregeometry as the n-copies
of a half-closed interval, and so M n forms a trivial pregeometry with
its algebraic closures. □
Definition 4.4. Let a, b G M n be any elements with acl(a) ^ acl(6).
(1) We define the S-distance of b from a, denoted by Sd (a, b) as
follows: Sd (a, b) = k iff M n \= S(g k (a), b, g k+1 (a)). For integers
k < l, we write k < Sd (a, b) < l if M n |= S(g k (a),b, g l+1 (a)).
(2) We define the S-distance of b from a, denoted by Sd (a, b), as
similar manner as Sd (a, b), using the formula
S(x, y,z ) = (x z A S(x, y, z)) V (x — z A x ^ y).
A CLASSIFICATION OF 2-CHAINS HAVING 1-SHELL BOUNDARIES IN ROSY THEORIES
Remark 4.5. Let x,y,z G M n have mutually disjoint algebraically
closures. Then for any k, l, m G Z,
(1) Sd (y,x) = - Sd (x,y) - 1;
(2) (a) for l — k ^ —1,0 (mod n), if k < Sd (x,y) <1 — 1, and
Sd (y, z ) = m, then m + k < Sd (x, z) <m + l\
(b) for l — k = — 1 (mod n), if k < Sd (x,y) <1 — 1, and
Sd (y, z) = m, then g k+m (x) z.
(1) ' Sd(y,x) = -Sd{x,y)-1- U
(2) ' for k ^ l (mod n), if k < Sd (x, y) < l — 1, and Sd (y, z) = m,
then m + k < Sd (x, z) <m + l.
Lemma 4.6. Let k and l Q ,... ,l m be fixed integers and Lj := YH= o h-
Let a and d 0 ,, d m +1 (m + 1 < n) be elements in M n such that
(*)rn : Sd (a, do) = k, Sd (di, d i+1 ) = U, 0 < i < m.
Then
k L m < Sd (u, d m _|_i) < k ~\~ L m -)- 771 1.
Moreover, by choosing appropriate elements for a and do,..., d m+ i, the
quantity Sd (a,d m+ 1 ) can be made to be any integer in [k + L m , k +
L m + m + 1 ] (**) m .
Proof. We show this using induction on m. For m = 0, by Remark
I4.51( 2) / . it follows from (*) 0 that
k T Iq < Sd (cz, d \) < k T In T 1.
Moreover it is not hard to see (**) 0 holds.
Now assume the lemma holds for m — 1 with m + 1 < n. Let us show
the lemma for m. For i < m + 1, a, di G M n are given which satisfy
(*) m . Firstly, by the induction hypothesis for m — 1,
k + L m - 1 < Sd (a, d m ) < k + L m -i + m.
Since m + 1 < n,
k + L m -1 < Sd (a, d m ) < k + L m - 1 + m.
Then again by Remark 14. 5 If 2)'.
k + L m < Sd (a, dm+i) < k + L m + m + 1.
Secondly, we show the moreover part. Fix L m < j < L m + m + 1 and
a' G M n . If j = L m , then j — l rn — L m -i and due to the induction
hypothesis, there are d' Q ,...,d' m that satisfy (*) m _i and
Sd (a', d' m ) = k + j-lm ■
20
BYUNGHAN KIM, SUNYOUNG KIM, AND JUNGUK LEE
So, Sd {a',g lm {d' m )) = k + j, and M n \= S (g lm (d' m ), d' m+1 , g k+j+1 (a'))
for some d! m+1 G M n . Thus
Sd (d' m , d! m+1 ) = l m , Sd (a', d' m+l ) = k + j.
So, a' and d! i for i < m + 1 satisfy the required condition. Now for
j > L m , the proof is similar to the case j = L m except that we replace
j — l m by j — l m ~ 1 and take d' m+1 in M n such that
M n ^S(g k+ i(a'),d' m+l ,g , - + \d' m )).
□
Now, let A(pn) be the family of all the closed independent func¬
tors in p n . We follow the notation given at the beginning of Section
[2j given a closed independent functor / over 0 in p n with u = {i 0 <
■■• < i k } G dom(/), we write f(u) = [a 0 , • • •, a*,], where aj G M n ,
f(u ) = acl(a 0 ,.. ■, Ofe), and acl(aj) = fu h \{ij})- When we write
f(u) = [b 0 , ..., b k ], it of course means that [a 0 , • • •, a k \ = [bo, ■ ■ ■, b k ].
By Theorem 14.31 it is equivalent to saying a 0 ■ ■ ■ a k = b 0 ■ ■ -b k .
m
Remark 4.7. Let r = e di (L; 2-simplex) be a chain-walk (in p n )
i=0
from foi to — /02 such that D t = supp(tj) = {0, /q+i} with k 0 = 1,
k m+ 1 = 2. Then putting together the triangles t 0 (D 0 ), ..., t m (D m ) side
by side centered at 0, we can find elements a and d 0 ,. .., d m+ 1 in M n
such that for 0 < i < m,
t i D \— I [ a ,di, d i+ 1] if ki < k i+ 1
[a, d.j , i. dj] if fcj > /cj_)_i.
Combining the classification results in Section [3] and Lemma 14.61
we will show that there does not exist any finite upper bound for the
minimal lengths of 2-chains with 1-shell boundaries in the types p n .
Theorem 4.8. Let A be a non-trivial amenable collection and let s be
a 1-shell. Define B(s), and B(A ) as follows:
(1) B(s ) := min{ |r| : t is a (minimal) 2-chain and <9(r) = s }.
(If s is not the boundary of any 2-chain, define B(s) := —oo.)
(2) B{A) := max{R(s): s is a 1-shell of A }.
Let n > 1 and let s = s 12 — S 02 + S 01 be a 1-shell from A(p n ) with
supp(sjj) = {i,j}- Then there are a,b,c,d in M n and some integers
k \, k 2 , ko with 0 < ki < n such that,
• Sd (a, c) = ki, Sd (a, b ) = k 2 , and Sd (6, d) = k 3 ;
• s 0 i({0,1}) = [a,c],s 02 ({0,2}) = [a, b\, and si 2 ({1,2}) = [c’,b].
A CLASSIFICATION OF 2-CHAINS HAVING 1-SHELL BOUNDARIES IN ROSY THEORIES
Let 0 < fc 4 (< n ) = k 2 — (ky — £ 3 ) (mod n) and let
n s := min{ 2 (n — kf) — 1 , 2 fc 4 + 1 }.
Then
B(s ) = n s .
Moreover, taking k\ = 0, k 2 = 0, and k 3 = [|] ; we get n s > n — 1 and
B(A(p n )) > n — 1 . Therefore lim -B(7l(p n )) = cx).
n—>-oo
Proof. (1) H(s) > n ,5 : By Theorem 12.41 and Corollary 13.141 there is a
2m
chain-walk r = (T](—l)*tj from s 0 i to — S 02 and <9(r) = s. We want to
i =0
show |r| > n s . Suppose not, i.e., |r| = 2m+ 1 < n — 1. By Remark l4~7l
there are dfs (0 < i < 2 m + 1) in M n such that ac = ado, d 2 m +i = b;
and
• Sd (d 0 , di) = l 0 , Sd (d 2 m-i) d 2m ) = hm. for some integers If,
• t 0 ({ 0 ,fc 0 ,fci}) = [a,d 0 ,d 1 ], t 2 i _i({ 0 , k 2 j-i,k 2 j}) = [a,d 2j ,d 2j -i\,
and t 2 j({ 0 , k 2j , k 2j+1 }) = [a, d 2j , d 2j+1 ] for 1 < j < m.
Now dr = s implies d°t 2 j 0 = s 12 for some 0 < 2 j 0 < 2m; and for any
0 < ji 7 ^ jo < m there is 0 < j 2 7 ^ jo < m (indeed a bijection) such
that d°t 2n = d°t 2 j 2+1 . So
• Sd (d 2jo , d 2j 0 + i) = -k 3 - 1; and
• [d2j 1 ,d 2 j 1+ i\ = [d 2 j 2+ 2,d 2 j 2+ i\.
By Remark IdlT iy. Sd (d 2 j 1 , d 2 j 1 +i) = — Sd (d 2 j 2 , d 2 j 2 +i ) — 1. Therefore
2m
l 2 j 0 = —k 3 — 1 and l 2 j 2 +i — ~hj! — 1, so h = ~^3 — m ~ 1- Hence
3 =0
due to Lemma 14.61 and 2m + 1 < n — 1, we have k\ — k 3 — m — 1 <
Sd (a, b) < ki — k 3 + m. Thus
Sd (a, b) = k 2 , and k\ — k 3 — m — 1 < Sd (a, b) < k\ — k 3 + m.
We rewrite it as
Sd(g kl ~ k3 (a),b) = k 2 -(k 1 -k 3 y, and -m-1 < Sd (g kl - k3 (a), b) < m.
We can replace k 2 — (k\ — k 3 ) by k± and we have n — (m +1) < +1 or
m + 1 > k^ In either case, we have m > min{n — fc 4 — 1, fc 4 }. Therefore
2m +1 > 2 min{n — fc 4 — 1, fc 4 } +1 = min{2(n — k 4 ) — 1, 2A : 4 +1} = n s ,
a contradiction. We have proved B(s) > n s .
n s — 1
(2) B(s) < n s : We construct a chain-walk 7 = ^ with supp( 7 ) =
i=0
{0,1,2} and <9y = s as follows: Note that since n s is odd, m s : =
(n s — l )/2 is an integer. Also note that, if we let Aj := k\ — k 3 — m s — 1
and N 2 := k 3 — k 3 + m s , then k 2 = N t (mod n) (i — 1 or 2). Hence we
22
BYUNGHAN KIM, SUNYOUNG KIM, AND JUNGUK LEE
have Sd (a, b ) = N\ or Sd (a, b ) = JV 2 . Applying Lemma [4761 with fcj and
Zo, - - -, ^ 2 m s such that / 2 ?;+i = —Z 2 * — 1 for 0 < i < m s and l 2ms = —k 3 — 1 ,
2m s
we obtain = L 2rns — ~ m s — k 3 — l, and L 2ms + 2 m s +1 = m s — k 3 .
i o
Therefore if j is chosen to be such that j = N 3 — k\ or = N 2 — k 3 , and
by applying (**) 2ms in Lemma SH we can find d ' 0 ,..., d' 2ms+1 (= d' n J
such that
Sd (a, d' 0 ) = ki, Sd (d[, d' i+1 ) = ^ for 0 <i< 2 m s , and Sd (a, d' n ) = k 2 .
Then due to Fact 14.1( 4) and Remark 14.51 (1)', it follows that ad' 0 =
ac, ad' ns = ab, d! ns _ x d! na = c '6 and d , 2i d , 2i+1 = d' 2i+1 d' 2i+2 for 0 < i < m s .
n a —1
Hence clearly we have a desired 2-chain 7 — r i such that
4 = 0
^({ 0 , 1 , 2 })
[a, d' tl d' i+l ] if i = 0 (mod 2)
[a, d' i+v d'] if i = 1 (mod 2).
□
Corollary 4.9. For each n > 5, A(p n ) does not have weak 3-amalgamation.
References
[1] Hans Adler. Explanations of Independence. Ph. D. Thesis, Univ. of Freiburg
(2005).
[2] Enrique Casanovas, Daniel Lascar, Anand Pillay, and Martin Ziegler. Galois
groups of first order theories. Journal of Math. Logic , 1 (2001) 305-319.
[3] Reinhard Diestel. Graph Theory, 2nd edition. Springer, New York (2000).
[4] Clifton Ealy and Alf Onshuus. Characterizing rosy theories. Journal of Symbolic
Logic, 72 (2007) 919-940.
[5] John Goodrick, Byunghan Kim, and Alexei Kolesnikov. Homology groups of
types in model theory and the computation of H 2 {p). Journal of Symbolic Logic,
78 (2013) 1086-1114.
[6] John Goodrick, Byunghan Kim, and Alexei Kolesnikov. Amalgamation functors
and homology groups in model theory. To appear in Proceedings of ICM 201 f.
[7] SunYoung Kim and Junguk Lee. More on classification of 2-chains having 1-shell
boundaries in rosy theories. Preprint.
Department of Mathematics, Yonsei University, 50 Yonsei-Ro, Seodaemun-
Gu, Seoul 120-749, Korea
E-mail address: bkim@yonsei.ac.kr
E-mail address: sy831@yonsei . ac .kr
E-mail address: ljw@yonsei.ac.kr
