A situs is a notion of generalised topological space, and defined as a simplicial object of the category of filters on sets or, equivalently, the category of finitely additive measures taking values 0 and 1 only, i.e. a simplicial set equipped, for each , with a filter, equiv. such (not quite) a measure, on the set of -simplices such that under any face or degeneration map the preimage of a large set is large. We denote the category of situses by sዋ. Most of what we say below would also apply to the category of simplicial sets with finitely additive measures obtained by dropping the restriction that the measure takes only two values. With appropriate definitions, sዋ is a full subcategory of .
Intuitively, these filters (measures) are viewed as additional structure of topological nature on a simplicial set (“the situs structure on a simplicial set”) giving a precise meaning to the phrase “a simplex is sufficiently small”: by definition, we say “a property holds for all small enough -simplices” iff it holds “almost everywhere” according to the filter (measure), i.e. on a set in the filter, or, resp., on a set of full measure. We refer to sets in the filter, resp., sets of full measure, as neighbourhoods. In particular, the situs structure “on a set” (i.e. on the simplicial set represented by the set) allows one to talk about an n-tuple of points being “sufficiently near each other” for . This extends to simplicial language the standard intuition of topology available for pairs of points (n=2): given a topological structure on a set, the precise meaning of the phrase “a property holds for all points sufficiently close to a given point ” is that the property holds on a neighbourhood of ; intuitively, the pair with near is thought of as small (a small simplex) because approximates up to a small error.
Situses generalise metric and topological spaces, filters, and simplicial sets, and the concept is designed to be flexible enough to formulate categorically a number of standard basic elementary definitions in various fields, e.g. in analysis, limit, (uniform) continuity and convergence, equicontinuity of sequences of functions; in algebraic topology, being locally trivial and geometric realisation; in geometry, quasi-isomorphism; in model theory, stability and simplicity and several Shelah’s dividing lines, e.g. NIP, NOP, NSOP, , , , , of a theory.
No homotopy theory for situses has been developed, although the naive definition of an interval object (namely, the simplicial set represented by the linear order equipped with some situs structure based on the metric/topology) leads to a directed (not symmetric) notion of homotopy, and in sዋ there is a diagram chasing reformulation of local triviality (becoming trivial after a certain base-change/pullback).
We now give a number of examples demonstrating the expressive power of the category of situses.
A simplicial set can be equipped with discrete or indiscrete situs structure: %The (in)discrete situs structure on a simplicial set for each , equip with the discrete or indiscrete filter, respectively.
Let be a metric space. View as a simplicial set represented by the set of points of , and equip each with the filter generated by uniform neighbourhoods of the diagonal, i.e. subsets containing all tuples of small enough diameter. With this situs structure, a map is uniformly continuous iff it induces a morphism of situses. In fact this defines a fully faithful embedding of the category of metric spaces with uniformly continuous maps into the category of situses.
We can also consider a different situs structure capturing the notion of quasi-isomorphism in large scale geometry. Equip each with the filter such that a subset of is large iff it contains all -tuples such that the distance between distinct points is at least , for some . With this situs structure, for quasi-geodesic metric spaces, a map is an isomorphism of situses iff is an quasi-isometry.
Given a filter on a set , there is a coarsest situs structure on viewed as a simplicial set (i.e. the simplicial set represented by ) such that its filter on the set of -simplices is finer than . Dually, there is a finest situs structure on such that its filter on the set of -simplices is coarser than . We denote these situses by and , respectively.
In fact this gives two fully faithful embeddings of the category of filters into the category of situses
In a similar way one can define two fully faithful embeddings of the category of filters on preorders and continuous monotone maps. We denote these by and , respectively.
More generally, given an arbitrary simplicial set and a filter on the set of -simplices , there is a coarsest/finest situs structure on such that its filter on the set of -simplices is finer/coarser than . Taking and the filter always indiscrete gives two fully faithful embeddings of the category of simplicial sets into the category of situses.
We use this to define situses corresponding to uniform and topological spaces.
Take a set and view it as a simplicial set (represented by ). Recall that a uniform structure on is a filter on ; take the coarsest situs structure with this filter on the set of -simplices. This is the situs associated with the uniform structure on . Intuitively, we defined the situs structure “on a set” such that two points of points are “sufficiently close to each other” in the uniform structure iff the 1-simplex is “sufficiently small”.
In fact, it is easy to define uniform spaces in terms of situses. A filter on is a uniform structure iff it is symmetric (i.e. the endomorphism of , permuting the coordinates is continuous) and this construction produces a situs such that the filter on is the coarsest filter such that the two maps , are continuous. This can be used to characterise situses arising from uniform structures as those symmetric situses such that the filter of has this property. We say that a situs is symmetric iff it factors though the category of non-empty finite sets.
The situs associated to a topological structure on is defined in the same way starting from the filter of non-uniform neighbourhoods of the diagonal on defined as consisting of the subsets of form where is a not necessarily open neighbourhood of .
A trivial verification shows these constructions define fully faithful embeddings of the categories of topological and of uniform spaces into the category of situses, and in fact there are corresponding forgetful functors to these categories such that the following compositions are the identity:
A filter on a metric space is Cauchy iff is continuous.
A sequence of functions , of metric spaces is uniformly equicontinuous iff the map is continuous. This gives a precise meaning to the phrase “if , and , are sufficiently close to each other, so are and ”. A different choice of the situs strucuture gives a different precise meaning:
A sequence of functions , from a topological space to a metric space is equicontinuous iff the map is continuous.
An endomorphism of the category of finite linear orders gives rise to an endomorphism of the category of situses. Of particular interest is the shift endomorphism adding a new least element (decalage considers the endomorphism adding a new greatest element rather than least) on objects, , and on morphisms, goes to , , , . The object and morphism allows one to talk about local properties of , e.g. limits and local triviality.
For example, taking a limit of a filter on a topological or metric space corresponds to taking the factorization
Indeed, the underlying simplicial set of is connected and thus maps to a single connected component of (here we consider the equality of the underlying simplicial sets); continuity of the map , means exactly that the first coordinate is a limit point of on .
Let and denote the set, resp. the linear order , equipped with the filter of cofinite subsets.
A metric space is complete iff either of the following equivalent conditions holds:
i. .
ii. .
iii. .
Such a reformulation raises the question whether the notion of completeness may be defined with help of the archetypal counterexample: is a metric space complete iff
It also allows to define the completion of a metric space in terms of a weak factorisation system
A topological space is compact iff for each ultrafilter either of the following equivalent conditions holds:
i. .
ii. .
A topological space is compact iff
Here , , and are viewed as filters on the set . (needs verification)
A map of topological or metric spaces is locally trivial with fibre iff in sዋ becomes a direct product with (“globally trivial”) after base-change . That is,
is of form .
The notion of geometric realization involves topological spaces and simplicial sets, which both are situses. This allows one to interpret the Besser-Drinfeld-Grayson construction of geometric realisation in sዋ, as follows.
View the standard geometric simplex consisting of sequences in as the space of monotone maps with Skorokhod-type metric . Recall that Skorokhod metric is used in probality theory to express the intution that two random variables are close if one can be obtained from another by a small perturbation of both time and space (values). The category of situses allows us to view both linear orders as situses: the situs structure on “remembers” the metric, and the situs structure on “remembers” the equality , i.e. is the finest situs structure such that the filter on the set of 0-simplices is indiscrete.
Then one may define a situs structure on the inner hom
of the underlying simplicial sets motivated by the definition of Skorokhod metric.
This gives the construction of geometric realisation due to
Besser, Drinfeld, and Grayson. See details at section 3.2 of geometric realization.
Let us now describe how situses can be used to reformulate two notions of model theory: stability and simplicity of a first-order theory. An alternative and more general approach is given in the next section on the Shelah’s dividing lines which completely covers simplicity(NTP).
Consider a model in a language , and a linear order . For an -ary -formula , we say that a sequence of elements of is -indiscernible (with repetitions) iff for either all or none of the subsequences (of distinct elements) the formula holds in .
Equip with the filter generated by the sets of all -tuples which are -indiscernible with repetitions, where varies through all -formulas. The situs so obtained is called {the generalised pre-Stone space of in sዋ} because the forgetful functor takes it to the set of elements of equipped with the preimage of Stone topology?; by this we mean the topology on generated by sets of elements realising unary -formulas. There are many variants of this definition, notably instead of being -indiscernible one may require being a part of an {infinite} -indiscernible sequence, and instead of consider the set of -types or .
We shall reformulate the following characterisation of stable theories as a lifting property in sዋ.
A first-order theory is stable iff in a saturated enough model it holds that each -indiscernible sequence of -tuples is in fact a -indiscernible set, for each and each formula of the language of the theory.
For this can be reformulated as a lifting property in sዋ as follows.
Fix a linear order .
Let be the situs associated with the preorder with the indiscrete filter. Recall that this is the simplicial set represented by as a linear order, equipped with indiscrete filters. Let denote the situs associated with the preorder with the {filter of tails} generated by the subsets containing all elements large enough.
Let denote the situs associated with the filter of tails on the set of elements of .
An indiscernible sequence indexed by a linear order is an injective continuous map .
An indiscernible set indexed by is an injective continuous map .
An eventually indiscernible sequence indexed on a linear order is an injective continuous map .
Let be a model. The following are equivalent:
i. each -indiscernible sequence of elements is in fact a -indiscernible set.
ii. the situs is symmetric
iii. the following lifting property holds in sዋ:
iii. says that each continuous map factors as . If these maps are injective, remarks above say it is equivalent to i. If not injective, then in some end segment each element occurs infinitely often, and in that case being indiscernible with repetitions means being set indiscernible.
(Simon,2021) implies that a theory is stable iff the lifting property iii. holds for the situs associated with considered in the language with arbitrary parameters, for a saturated enough model of the theory.
The definition of simplicity is not as simple combinatorially. A more general approach to simplicity is discussed in the next section on Shelah’s dividing lines.
First let us introduce the situs associated with a model for this purpose; this situs structure is defined to talk about consistency of instances of a formula. Fix a formula . As usual, the situs is based on the simplicial set represented by the set of elements of . The filter on is generated by a single set of those tuples such that
. Let us denote this situs by and call it the -characteristic situs of model . Note that -characteristic situs captures the same structure as the characteristic sequence of a first order formula in (M.Malliaris. The characteristic sequence of a first-order formula. 2010), see also (M.Malliaris. “Edge distribution and density in the characteristic sequence).
Recall that a formula φ(x;y) has the finite cover property if for arbitrarily large there exist such that is n-consistent but (n+1)-inconsistent.
Recall that a simplicial object , e.g. a situs, is said to have finite dimension if for some for each is the pullback of all the simplicial maps , . Let denote the formula .
has no finite cover property
the characteristic sequence has finite support (in terminology of Malliaris, Def.2.6)
the situs has finite dimension, i.e. explicitly, for some for each the filter on is the coarsest filter such that all the simplicial maps are continuous
there is such that for each holds iff for any -element subset
there is such that for each
Items 1 and 2 are Remark 2.7 of Malliaris, item 4 and 5 are both item 2 and item 3 written explicitly.
The reformulation in terms of situses uses the definition of a simple first-order theory which says that each formula of the theory has NTP (“not the tree property”) see Tent-Ziegler, Def.7.2.1, or 3,\S9 NTP is defined as a lifting property with respect to a morphism involving the following combinatorial structures.
We recall the definition of NTP and a simple theory.
[Tent-Ziegler,7.2.1] 1. A formula has the tree property (with respect to k) if there is a tree of parameters such that:
a) For all , is -inconsistent.
b) For all is consistent.
Let be an infinitely branching tree of infinite depth, viewed as preorder, and equipped with the indiscrete filter.
We may take to be . Let denote the corresponding situs. Recall that by definition is the set of ordered (weakly increasing) -tuples of vertices of , and there is only one large subset, namely the whole set.
Note that to give a morphism of the underlying simplicial sets is the same as to give a a tree of parameters . This morphism is continious iff these parameters satisfy item b, in notation of the definition: indeed, continuity means that the preigame of the large (by definition) set of -consistent tuples is large, i.e. the whole set of ordered tuples, for each .
Let be the simplicial set represented by the set of vertices of , namely .
Let denote the simplicial set equipped with the -tautological filter on defined as follows: a subset is not small iff it either contains
1) some tuple in weakly increasing order, or
2) all the lexicographically ordered tuples required to be inconsistent by the tree property with respect to a subtree-counterexample to the tree property.
In more detail, a subset is large iff
1’) it contains the subset of tuples in weakly increasing order
2’) for each isomorphic copy of in there is a vertex and its immediate (in ) descendants such that . %A verification shows that this indeed defines a filter.
Note that by item (i) the map is continuous. Also note that no tuple of increasing elements is required to be -inconsistent by the tree property.
The following are equivalent:
i. the formula has NTP with respect in the model
ii. in sዋ there is no morphism such that for each tuple -tuple , for each , of immediate descendants of the same vertex,
iii. In sዋ the following lifting property holds:
ii. is exactly the definition of NTP for formula as stated in (Tent-Ziegler, Def.7.2.1), cf. 3,\S9. In iii., one only needs to check that the unique lifting is continuous, namely that the set of tuples such that is large. By the definition of the filter, this set is large iff there is an infinitely branching subtree of infinite depth satisfying ii. This implies that ii. and iii. are equivalent.
Finally, let us prove our -tautological filters are well-defined. We need only to show that the union of any two small sets is small. Assume it is not small. Label each vertex of the tree with the largest such that the first small subset contains above the vertex all tuples required to be inconsisent in some copy of . Above each vertex in there are at most finitely many vertices in labelled by the same or greater number. Removing them leaves not small. But then we get that vertices of are labelled by numbers strictly decreasing along any branch, hence is of finite depth. Now pick a vertex labelled . This means that below that vertex there is no infinite set of siblings that each lexicographically ordered tuple is in , hence among any infinite set of siblings by Ramsey theorem there is an infinite set of siblings not in , i.e. in . Hence, is not small.
One can similarly define -tautological situs of a tree , for , and see that is defined by a lifting property. The same argument gives lifting properties related to cdt, inp, and sct patterns in classification theory.
This raises the question whether holds in the category of situses. It seems the standard proof would go through if one defines the corresponding lifting properties carefully enough. In particular, to reflect the use of Ramsey theorem, it may be necessary to replace by the simplicial set of types where is the set of -types, and also do the same for and for quantifier-free types in an appropriate language.
In fact, it would seem that the standard proof of gives that is the pushout of and , just as diagram chasing considerations show would be sufficient for the corresponding relation between the lifting properties. Though, possibly one needs to modify the definitions of the filters appropriately modified to reflect the need to use Ramsey theorem and consider the tree properties with respect all the finite conjunctions of at the same time.
The notion of a -characteristic situs captures the same structure as the characteristic sequence of a first order formula in (M.Malliaris. The characteristic sequence of a first-order formula. 2010), see also (M.Malliaris. “Edge distribution and density in the characteristic sequence). Moreover, it appears that several properties of characteristic sequences can be defined as lifting properties in the category of situses.
We show that in the category of situses a number of the Shelah’s divining lines, namely , and are expressed as Quillen lifting properties of form
where is the terminal object, and is a situs associated with a model and a formula, and and are objects of combinatorial nature.
Recall the common pattern of definitions of . As usually stated, such a property wrt a formula and a model require that there is no combinatorial structure (“counterexample” or “witness”) formed by elements of the model satisfying certain “positive” requirement that certain collections of formulas are consistent, and “negative” requirement that certain collections of formulas are inconsistent. The combinatorics is coded by the maps of underlying simplicial sets and . The positive requirements are coded by continuity of the morphism . The failure of negative requirement are coded by continuity of the diagonal morphism .
The underlying simplicial sets of , and are representable, and are so chosen that both morphisms and both correspond to maps of sets and . The ‘s and ’s represent the indices of tuples mentioned in positive, resp. negative, requirements. The filters on are taken indiscrete; thus continuity of means that for each , each tuple satisfies a certain formula. Namely, if is the situs of -indiscernible sequences, the continuity says that each such sequence is -indiscernile; if is the situs of consisent finite -types, the continuity says that each finite type is consistent.
To code by continuity the failure of negative requirements, we need to define non-trivial filters on . The first (incorrect) attempt would be to say that a subset of is large (a neighbourhood) iff it contains a tuple representing one of the negative requirements; unfortunately, this is not a filter. But if it were, then continuity of would mean exactly that one of negative requirements fails, and hence the map does not represent a counterexample/witness. To fix this definition, define a subset of to be large (a neighbourhood) iff it contains a tuple representing a negative requirement with respect to each substructure of of the same shape as required by the property.
We call the filters on ‘s tautological because of the tautological argument showing equivalence of the continuity and the failure of the negative requirements.
Below we do the reformulations for non-order properties , , and , which involve the situs of -indiscernible sequences. These reformulations involve reformutading the standard definitions in terms of -indiscernible sequences.
The no-tree-properties were formulated above and they involve the situs of consistent -types.
Note our considerations allow to formulate a precise conjecture corresponding to the equality of On the Antichain Tree Property. Indeed, all classes occurring in the formula are defined by lifting properties wrt the same morphism, and the equalitity should represented a relationship of the “combinatorial” morphisms on the left. Though doing so should perhaps require a rather more careful reformulations taking care of formulas of arbitrary arity.
Recall a formula has NOP (no order property) iff there no sequence such that .
Because NOP cares about {\em consistency of instances of a formula on tuples of ordered variables}, equip with the filter of subsets containing
Then rewrite the standard definition of NOP in a form with two items representing positve and negative requirements.
A formula has {\em NOP} iff there is no sequence such that \begin{itemize} \item[a)] for , is false. %each subsequence is not -indiscernible with repeated elments \item[b)] for , holds. \end{itemize}
With these definitions, defines NOP.
For , a formula has SOP (%the -strong order property) iff \begin{itemize} \item[a)] the set is inconsistent.
\item[b)] there are such that for all
\end{itemize} A theory is NSOP if no formula has SOP.
Let , denote the simplicial set represented by the set with elements which we denote by . Equip with the filter of subsets containing {,} and for each equip with the coarsest filter such that all simplicial maps are continuous.
Then a map is continuous iff either or holds for the subsequence of of distinct elements. Hence, item a) holds iff . Then\ NSOP can be stated as:
We also note that in the category of generalised topological spaces NFCP (no finite cover property) means that is of infinite dimension. Recall that “\bulletM\bulletM_\bulletN\ge 0n\ge N
M_\bullet(n^\leqslant)M_\bullet(n^\leqslant)\to M_\bullet(m^\leqslant)M_\bullet(m_1^\leqslant)\to M_\bullet(m_2^\leqslant)m,m_1,m_2\le N\lt P_n\gt\varphi$ has finite support’’.
Both stability and NIP can be expressed in terms of indiscernible sequences. Hence, we modify the filters on to reflect that: the filter on is generated by sets of {\em -indiscernible sequences with repeations}, i.e.~sequences such that whenever , , all the are distinct, and all the are distinct. %any subsequence of distinct elements -indiscernible. (If you feel it is more natural, you may require each subsequence with distinct elements to be part of an {\em infinite} -indiscernible sequence.)
This construction of generalises to formulas of arbitrary arity, and, moreover, for arbitrary collections of formulas.
There is a forgetful functor taking a generalised topological space into a topological space (possibly empty). For a unary formula , it takes defined above into the (usual) Stone space of -types.
Let { \ } equip with the filter of cofinite subsets, and equip each , resp.~, with the coarsest filter such that all simplicial maps , resp.~, are continuous. Then {each (infinite) -indiscernible sequence is a -indiscernible set iff \ }
Let { ,} and equip with the filter generated by subsets %, for each formula {, a formula} One can check that the map forgetting the first coordinate, is continuous.
Each indiscernible sequence in is eventually indiscernible over any parameter iff {} where is equipped with the filter containing the empty set, and is the map . Note that the map picks an arbitrary element of .
In the notions of homotopy and limit are closely related: the same construction in describes both picking a homotopy contracting a topological space (via singular complexes) and taking a limit of a filter on a topological space, and applies to an arbitrary morphism in . One may perhaps hope that this construction can somehow be used to define a useful notion of a {\em contractible} (generalised topological space associated with a) {\em model}, %, or that the notion of a {\em limit of a combinatorial structure on a model}, say a tree as in NTP, is useful (i.e.~taking a limit of the morphism associated with a combinatorial structure on a model, such as those arising in the lifting properties describing NTP).
Below we sketch this construction; see [6,\S3] for more details. We warn the reader that our considerations here are unusually preliminary, and apologise for including them; our excuse is that we are likely unable to pursue them.
[(Limit in a generalised space)] Let be a morphism in . A morphism is said to be a {\em limit morphism} (or simply {\em a limit})
of iff factors as
where is the shift
and is the expected map “rgetting the first coordinate’’.
To recover the Bourbaki definition of a limit of a filter on a topological space , associate with the simplical set represented by
equip with , and equip each %is equipped with the finest filter such that the diagonal map is continuous. A verification shows that (possibly discontinuous) liftings correspond to points of (indeed, as simplicial sets, is connected, and is the disjoint union of connected component which are copies of parametrised by ), and the continuity requirement means precisely that they are limit points. % (see Fig.~2).
[Limit of the tree in NTP] Let , , and denote the objects corresponding to a tree and a model appearing in the lifting property for NTP of formula . A verification shows the following.
A limit of is a point such that the map is continuous where .
To see this, first note that, as simplicial sets (i.e.~if we ignore the filters), is the disjoint union of connected component which are copies of parametrised by elements of , that both , are connected. Thus to give a map of simplicial sets or is the same as to pick a point . The continuity requirement on means precisely that is consistent whenever lie on the same branch of . The continuity requirement on means precisely that is consistent whenever lie on the same branch of .
The case of is similar. \def\sing{\operatorname{sing}} Now let and denote topological spaces.
A homotopy contracting in to a point (i.e.~a map
from the cone of to ), gives rise to a map ar complexes lifting the map :
Take to defined by
Here and is the cone of -simplex .
A map
continuous in a neighbourhood of “e top of the cone’‘ point is the same as a map in
where denotes the interval equipped with the filter of neighbourhoods of point , and denote the generalised topological spaces corresponding to and .
Some of these constructions are sketched in the drafts below.
Topology and analysis:
Geometric realisation:
Stability and simplicity:
[TentZiegler] K.Tent, M.Ziegler. A Course in Model Theory. CUP. 2012.
Maryanthe Malliaris. The characteristic sequence of a first-order formula. J Symb Logic 75, 4 (2010) 1415-1440. (pdf)
Maryanthe Malliaris. “Edge distribution and density in the characteristic sequence,” Ann Pure Appl Logic 162, 1 (2010) 1-19. pdf
[Scow2012] Lynn Scow. Characterization of nip theories by ordered graph-indiscernibles. Annals of Pure and Applied Logic, 163(11):1624 – 1641, 2012. (pdf)
[Simon2021] Pierre Simon. A note on stability and NIP in one variable. (pdf)
[AhnKimLee2021] JinHoo Ahn, Joonhee Kim, and Junguk Lee. On the Antichain Tree Property. pdf
[3] Misha Gavrilovich. Remarks on Shelah’s classification theory and Quillen’s negation. (pdf)
Last revised on July 14, 2022 at 17:05:39. See the history of this page for a list of all contributions to it.