We introduce the class of treeless theories. These theories are defined in terms of a certain kind of indiscernible collapse which informally corresponds to the inability of the theory to code trees. This approach carves out a natural model-theoretic setting that contains both the stable theories and the binary theories. We build on the study of generically stable partial types begun in [Reference SimonSim20] to develop a theory of independence, called GS-independence, which allows us to establish the rudiments of a structure theory for this class. Although the genesis of this approach comes from theories without the independence property, we show that treelessness has strong consequences for the largely orthogonal setting of theories in the SOP
$_{n}$
hierarchy.
We begin, in Section 1, with a study of generically stable global partial types, as defined in [Reference SimonSim20]. We show that, in an arbitrary theory, every complete type over a set of parameters A extends to a unique maximal global partial type which is generically stable over A. This is then used to define
$\mathrm {GS}$
-independence: a is said to be
$\mathrm {GS}$
-independent from b over A if b satisfies
$\pi |_{Aa}$
, where
$\pi $
is the maximal global partial type which is generically stable over A and extends
$\text {tp}(b/A)$
. In Section 2, we study the properties of this independence relation in general and find that it satisfies many of the basic properties of independence relations.
In order to define treeless theories, we introduce in Section 3, a new kind of indiscernible tree, which we call a treetop indiscernible. The index structure in a treetop indiscernible is, in essence, the same as that of a strongly indiscernible tree, together with a predicate identifying the leaves of the tree. We show that finite trees (in a language with symbols for the tree partial order, the lexicographic order and the binary meet function) together with a predicate for the leaves form a Ramsey class and hence structures with this age give rise to a sensible notion of generalized indiscernible. In the tree
$\omega ^{\leq \omega }$
, the set
$\omega ^{\omega }$
of leaves carries the structure of a dense linear order (under
$<_{lex}$
) but also carries considerably more structure induced by the tree structure. The treeless theories are defined in Section 3 to be those theories in which, in any treetop indiscernible, this additional structure on the leaves is irrelevant, that is, the sequence of tuples indexed by the leaves ordered lexicographically is an indiscernible sequence. In Section 4, we connect treelessness to the above-mentioned work on
$\mathrm {GS}$
-independence, showing that, in treeless theories, GS-independence is symmetric and satisfies base monotonicity.
In Section 5, we prove that all stable theories are treeless and then in the remaining sections, we explore the consequences treelessness has for the SOP
$_{n}$
hierarchy. In Section 6, we prove that NSOP
$_1$
treeless theories are simple. We obtain this result as a rapid consequence of the fact that GS-independence and Kim-independence coincide over models in NSOP
$_{1}$
theories, but we also give an alternative argument for the corollary that binary NSOP
$_{1}$
theories are simple, using only tools from the theory of Kim-independence, which may be of independent interest. In Section 7, we show that every treeless NSOP
$_{3}$
theory with indiscernible triviality is NSOP
$_{2}$
. These hypotheses are met by any binary NSOP
$_{3}$
theory and therefore, modulo Mutchnik’s recent result [Mut22] that NSOP
$_{1}$
= NSOP
$_{2}$
, our results establish that every binary NSOP
$_{3}$
theory is simple. This means, for example, that the known classification for simple binary homogeneous structures due to [Reference KoponenKop18] applies directly to the a priori much broader class of homogeneous binary NSOP
$_{3}$
structures.
1 Generically stable partial types
In the following two subsections, we recall definitions and basic properties of generically stable partial types from [Reference SimonSim20]. The main result of the section is Corollary 1.9, which entails that every complete type over a set A has a unique maximal extension to a global partial type which is generically stable over A. This will serve as the basis of a notion of independence introduced in Section 2.
1.1 ind-definable partial types
We will work in a monster model
$\mathbb {M}$
of a fixed complete theory T. A partial type
$\pi (x)$
(over
$\mathbb {M}$
) is a consistent set of formulas with parameters in
$\mathbb {M}$
closed under finite conjunctions and logical consequences, that is:
-
○
$\phi (x), \psi (x)\in \pi \Longrightarrow \phi (x)\wedge \psi (x) \in \pi $ ;
-
○
$\phi (x)\in \pi \wedge \mathbb {M} \vDash \phi (x) \to \psi (x) \Longrightarrow \psi (x)\in \pi $ .
Given a set A of parameters,
$\pi |_A$
or
$\pi | A$
denotes the partial type obtained by taking the subset of
$\pi $
composed of formulas with parameters in A. Note that, because we require
$\pi $
to be closed under logical consequence, if
$a\vDash \pi |_A$
, then
$\pi \cup \mathrm {tp}(a/A)$
is consistent.
A partial type
$\pi $
is A-invariant if it is invariant under automorphisms of
$\mathbb {M}$
fixing A pointwise.
Definition 1.1. We say that a partial type
$\pi $
is ind-definable over A if for every
$\phi (x;y)$
, the set
${\{b : \phi (x;b)\in \pi \}}$
is ind-definable over A (i.e., is a union of A-definable sets).
As noted in [Reference SimonSim20, Section 2], one can represent an A-ind-definable partial type as a collection of pairs

where
$\phi _i(x;y)\in L$
,
$d\phi _i(y)\in L(A)$
such that
$\pi (x)$
is equal to
$\bigcup _i \{\phi _i(x;b) : b\in d\phi _i(\mathbb {M})\}$
(the same formula
$\phi (x;y)$
can appear infinitely often as
$\phi _i(x;y)$
). And, conversely, given a family of pairs
$(\phi _i(x;y),d\phi _i(y))$
, if the partial type
$\pi (x)$
generated by
$\bigcup _i \{\phi _i(x;b) : b\in d\phi _i(\mathbb {M})\}$
is consistent, then it is ind-definable. Observe that the partial types
$(\phi (x;y),d\phi (y))$
and
$(d\phi (y)\to \phi (x;y); y=y)$
are the same.
Fact 1.2 [Reference SimonSim20, Lemma 2.2].
Let
$\pi (x)$
be a partial A-invariant type. Then
$\pi $
is ind-definable over A if and only if the set
$X=\{(a,\overline {b}) : \overline {b}\in \mathbb {M}^{\omega }, a\vDash \pi |A\bar b\}$
is type-definable over A.
Let
$\pi (x)$
and
$\eta (y)$
be two A-invariant partial types, where
$\pi $
is ind-definable over A. Then there is an A-invariant partial type
$(\pi \otimes \eta )(x,y)$
such that
$(a,b)\vDash \pi \otimes \eta $
if and only if
$b\vDash \eta $
and
$a\vDash \pi |\mathbb {M} b$
. Indeed,
$(\pi \otimes \eta )(x,y)$
is generated by
$\eta (y)$
along with pairs
$(d\phi (y,z)\to \phi (x;y,z), z=z)$
(with
$\phi \in L$
and
$d\phi \in L(A)$
), where the partial type
$(\phi (x;y,z),d\phi (y,z))$
is equal to
$\pi (x)$
. If in addition
$\eta $
is ind-definable over A, then so is
$\pi \otimes \eta $
. As usual, we define inductively
$\pi ^{(1)}(x_0) = \pi (x_0)$
and

Also, set

All those types are ind-definable over A.
Instead of a partial type
$\pi $
, one could also consider the dual ideal
$I_\pi $
of
$\pi $
defined as the ideal of formulas
$\phi (x)$
such that
$\neg \phi (x)\in \pi $
. Then an
$I_\pi $
-wide type (namely a type not containing a formula in
$I_\pi $
) is precisely a type over some A containing
$\pi |A$
.
1.2 Generic stability
Definition 1.3. Let
$\pi (x)$
be a partial type. We say that
$\pi $
is generically stable over A if
$\pi $
is ind-definable over A and the following holds:
(GS) if
$(a_k:k<\omega )$
is such that
$a_k \vDash \pi |Aa_{<k}$
and
$\phi (x;b)\in \pi $
, then for all but finitely many values of k, we have
$\vDash \phi (a_k;b)$
.
If
$\pi $
is a global partial type generated by
$\pi _{0}$
and ind-definable over A, to show that
$\pi $
is generically stable, it suffices to check that
$\pi $
satisfies the condition (GS) for the formulas in
$\pi _{0}$
.
Definition 1.4. We say that a partial type
$\pi (x)$
over
$\mathbb {M}$
is finitely satisfiable in A if any formula in it has a realization in A (recall that we assume
$\pi $
to be closed under conjunctions).
The following facts record some basic properties of generically stable partial types:
Fact 1.5 [Reference SimonSim20, Lemma 2.4].
Let
$\pi $
be a partial type ind-definable over A. Let
$a\vDash \pi |A$
and b such that
$\mathrm {tp}(b/Aa)$
is finitely satisfiable in A. Then
$a\vDash \pi |Ab$
.
Fact 1.6 [Reference SimonSim20, Proposition 2.6].
Let
$\pi $
be a partial type generically stable over A. Then:
(FS)
$\pi $
is finitely satisfiable in every model containing A;
(NF) let
$\phi (x;b)\in \pi $
, and take
$a\vDash \pi |A$
such that
$\vDash \neg \phi (a;b)$
. Then both
$\mathrm {tp}(b/Aa)$
and
$\mathrm {tp}(a/Ab)$
fork over A.
Fact 1.7 [Reference SimonSim20, Lemma 2.9].
Let
$\pi (x)$
be generically stable over A, and let
$\pi _0(x)\subseteq \pi (x)$
be a partial ind-definable type, ind-definable over some
$A_0\subseteq A$
(i.e., the parameters in the ind-definitions of
$\pi $
come from
$A_{0}$
). Then there is
$\pi _*(x)\subseteq \pi (x)$
containing
$\pi _0(x)$
which is generically stable and ind-defined over some
$A_*\subseteq A$
of size
$\leq |A_0|+|T|$
.
The following lemma is new but is a strengthening of [Reference SimonSim20, Lemma 2.11]:
Lemma 1.8. Let
$\pi (x), \lambda (x)$
be two partial types ind-definable over A. Assume that
$\lambda $
is generically stable over A and that
$\pi (x)|_A \cup \lambda (x)|_A$
is consistent. Then
$\pi (x)\cup \lambda (x)$
is generically stable over A.
Proof. We show by induction on
$n<\omega $
that there is
$\bar a=(a_i:i<n)$
such that
$\bar a\vDash \pi ^{(n)}(x)|_A$
and
$\bar a^* \vDash \lambda ^{(n)}(x)|_A$
, where
$\bar a^* = (a_{n-1},a_{n-2},\ldots ,a_0)$
. For
$n=1$
, this is the hypothesis. Assume we know it for n, witnessed by
$\bar a=(a_i:i<n)$
. Since
$\pi (x)|_A \cup \lambda (x)|_A$
is consistent so is
$\pi (x)|_A \cup \lambda (x)$
. Let
$\bar b=(b_i:i<\kappa )$
be a long Morley sequence in that partial type. Since we assume our partial types are closed under logical consequence, the fact that
$\overline {a} \vDash \pi ^{(n)}|_{A}$
implies that
$\pi ^{(n)} \cup \text {tp}(\overline {a}/A)$
is consistent. Thus, composing by an automorphism over A, we may assume that
$\bar a\vDash \pi ^{(n)}|_{A\bar b}$
. By generic stability of
$\lambda $
, there is
$i<\kappa $
such that
$b_i\vDash \lambda |_{A\bar a}$
. It follows that
$(b_i)^{\frown }\bar a \vDash \pi ^{(n+1)}|_A$
and that
${\overline {a}^{*}}^{\frown } (b_i) \vDash \lambda ^{(n+1)}|_A$
. This finishes the induction.
This being done, we can construct, by Fact 1.2 and compactness, a sequence
$\bar d=(d_i:i<\omega )$
which is a Morley sequence of
$\pi $
over A such that the sequence in the reverse order is a Morley sequence of
$\lambda $
over A. We can further assume that
$\bar d\vDash \pi ^{(\omega )}|_{\mathbb {M}}$
. The set of formulas over
$\mathbb {M}$
that are true on almost all elements of
$\bar d$
contains
$\lambda (x)$
, and therefore
$\pi (x)\cup \lambda (x)$
is consistent.
Finally, we conclude that
$\pi (x) \cup \lambda (x)$
is generically stable over A. Let
$\mu (x)$
be the partial type generated by
$\pi (x) \cup \lambda (x)$
. It is clear that
$\mu (x)$
is ind-definable over A using Fact 1.2 and the fact that
$\{(a,\overline {b}) : \overline {b} \in \mathbb {M}^{\omega }, a \vDash \mu (x)|_{A\overline {b}}\}$
is equal to the intersection
$\{(a,\overline {b}) : \overline {b} \in \mathbb {M}^{\omega }, a \vDash \pi (x)|_{A\overline {b}}\} \cap \{(a,\overline {b}) : \overline {b} \in \mathbb {M}^{\omega }, a \vDash \lambda (x)|_{A\overline {b}}\}$
. If
$\varphi (x;b) \in \mu (x)$
, then there are
$\psi _{0}(x;c) \in \pi (x)$
and
$\psi _{1}(x;d) \in \lambda (x)$
such that
$\psi _{0}(x;c) \wedge \psi _{1}(x;d) \vdash \varphi (x;b)$
. Taking
$I = (a_{i} : i < \omega ) \vDash \mu ^{(\omega )}|_{A}$
, since I is Morley over A in both
$\pi $
and
$\lambda $
, we know that both
$\{i : \vDash \psi _{0}(a_{i};c)\}$
and
$\{i : \vDash \psi _{1}(a_{i};d)\}$
are cofinite so
$\{i : \vDash \varphi (a_{i};b)\}$
is cofinite as well. This shows
$\mu $
is generically stable over A.
Corollary 1.9. Let
$p(x)\in S(A)$
. There is a unique maximal global partial type
$\pi _{p}$
generically stable over A consistent with p – that is, if
$\pi $
is a global generically stable partial type consistent with p, then
$\pi \subseteq \pi _{p}$
. It follows, in particular, that
$\pi _{p}$
extends p.
Proof. By Lemma 1.8, if
$\pi (x)$
and
$\lambda (x)$
are two generically stable partial types consistent with p, ind-definable over A, then
$\pi (x)\cup \lambda (x)$
is consistent and even generically stable over A. Hence, we can define
$\pi _p(x)$
as the union of all generically stable partial types consistent with p and ind-definable over A. Then
$\pi _p(x)$
is consistent and is the maximal A-invariant generically stable partial type consistent with p. As p itself is generically stable over A, it follows that
$\pi _{p}$
extends p.
Lemma 1.10. Suppose
$p(x)$
is a complete type over A and
$E(x,y)$
is an equivalence relation which is
$\bigvee $
-definable over A and has unboundedly many classes represented by realizations of p. If
$\pi \supseteq p$
is the maximal generically stable partial type over A extending p, then
$\pi \vdash \neg E(x,c)$
for all
$c \in \mathbb {M}$
.
Proof. Let
$\pi _{0}(x)$
be the global partial type defined by closing the set of formulas

under conjunction and logical consequence. Then
$\pi _{0}$
is a consistent partial type by our assumption that
$E(x,y)$
is
$\bigvee $
-definable and has unboundedly many classes among realizations of p. We have
$\pi _{0}$
is ind-definable over A since, writing
$E(x,y) = \bigvee \psi _{i}(x,y)$
, we can ind-define
$\pi _{0}$
(on the generating formulas) via the schema
$(\varphi (x), y=y)_{\varphi (x) \in p}$
and
$(\neg \psi _{i}(x,y), y = y)_{i}$
. If
$(a_{i} : i < \omega )$
is a sequence with
$a_{i} \vDash \pi _{0}|_{Aa_{<i}}$
, then we have
$\neg E(a_{i},a_{j})$
for all
$i \neq j$
. Therefore, if
$c \in \mathbb {M}$
, then c can be E-equivalent to at most one
$a_{i}$
. Therefore, if
$\chi (x,c) \in S$
, then we have
$\vDash \chi (a_{j},c)$
for all but at most one j. Since S generates
$\pi _{0}$
, this shows
$\pi _{0}$
is a generically stable partial type over A and is therefore contained in the maximal one extending p by Corollary 1.9.
The following proposition is essentially [Reference SimonSim20, Remark 6.13]:
Proposition 1.11. Let
$\pi (x,y)$
be generically stable over A. Then the partial type
$\eta (x) = (\exists y)\pi (x,y)$
(which is also the restriction of
$\pi $
to the x variable) is generically stable over A.
Proof. Note that for any set
$B\supseteq A$
,
$\eta | B = (\exists y)(\pi (x,y)|_B)$
.
Since
$\pi (x,y)$
is A-invariant,
$\eta (x)$
is also A-invariant. We first show that
$\eta $
is ind-definable using Fact 1.2. Fix a variable
$\bar z$
, and let
$X(x,y,\bar z)$
be the set of triples
$\{(a,b,\bar c) :(a,b) \vDash \pi |A\bar c\}$
. For any tuples a and
$\bar c$
, we have
$a\vDash \eta |A\bar c$
if and only if there is b such that
$(a,b,\bar c)\in X$
. As X is type-definable by Fact 1.2, this whole condition is type-definable. By one more application of Fact 1.2,
$\eta $
is ind-definable.
We next show (GS). Assume for a contradiction that for some
$\phi (x;c)\in \eta $
, the set
$\eta ^{(\omega )}(x_k:k<\omega ) \cup \{\neg \phi (x_k;c) : k<\omega \}$
is consistent. Let
$(a_k)_{k<\omega }$
realize it. Note that if we replace
$(a_k:k<\omega )$
by a sequence
$(a^{\prime }_k:k<\omega )$
which has the same type over A, then we can find
$c'\equiv _A c$
such that
$\neg \phi (a^{\prime }_k;c')$
holds for all k. By invariance of
$\eta $
, we have
$\phi (x;c')\in \eta $
, so
$(a^{\prime }_k : k < \omega )$
also witnesses a failure of (GS).
Choose
$b_0$
such that
$(a_{0},b_{0}) \models \pi |_{A}$
, which is possible since
$a_0\vDash \eta |A$
. We build by induction on k tuples
$(b_k:k<\omega )$
such that
$\mathrm {tp}(a_k,b_k/A)=\mathrm {tp}(a_{0},b_{0}/A)$
and
$(a_k,b_k)\vDash \pi | Aa_{<k}b_{<k}$
. Assume we have found
$b_k$
. As
$a_{k+1}\vDash \eta |Aa_{\leq k}$
, there is an automorphism
$\sigma $
fixing
$Aa_{\leq k}$
such that
$\sigma (a_{k+1})\vDash \eta | Aa_{\leq k}b_{\leq k}$
. By the remark above, we may replace the sequence
$a_{>k}$
by
$\sigma (a_{>k})$
since this does not alter the type of the full sequence
$(a_i)_{i<\omega }$
. Hence, we may assume that actually
$a_{k+1}\vDash \eta |Aa_{\leq k}b_{\leq k}$
and then we find
$b_{k+1}$
as required.
We now have a sequence
$(a_kb_k:k<\omega )$
such that
$(a_k,b_k)_{k<\omega }\vDash \pi ^{(\omega )}(x_k: k<\omega )$
and c such that
$\phi (x;c)\in \pi $
and
$\neg \phi (a_k;c)$
holds for all k. Since the condition
$(a_k,b_k)_{k<\omega }\vDash \pi ^{(\omega )}(x_k :k<\omega )|_{A}$
is type definable by Fact 1.2, we can apply Ramsey and compactness and assume that the sequence
$(a_kb_k:k<\omega )$
is indiscernible over
$Ac$
. Using (GS) for the type
$\pi $
, we conclude that for every k,
$(a_k,b_k)\vDash \pi |Ac$
. But by the definition of
$\eta $
, this means that
$a_k\vDash \eta |Ac$
. Contradiction.
The following corollary is [Reference SimonSim20, Proposition 2.13]. It follows immediately from Lemma 1.8 and Proposition 1.11.
Corollary 1.12. Let
$\alpha (y)$
be a partial type, generically stable over A. Fix some
$a,b\in \mathbb {M}$
,
$b\vDash \alpha (y)|_A$
, and let
$\rho (x,y)\subseteq \mathrm {tp}(a,b/A)$
. Then the partial type
$\pi (x):= (\exists y) (\alpha (y) \wedge \rho (x,y))$
is generically stable over A.
2 GS-independence
We write if for every partial type
$\pi (x)$
generically stable over A, if
$b\vDash \pi |_A$
, then
$b\vDash \pi |_{Aa}$
. Note that this is equivalent to saying that
$b \vDash \pi _*|_{Aa}$
, where
$\pi _*$
is the maximal A-invariant generically stable partial type extending
$\mathrm {tp}(b/A)$
. If p is a partial type, we say that
$p \mathrm {GS}$
-forks over A if there is some B such that there is no
$a \vDash p$
with
.
Lemma 2.1. If or
, then
.
Proof. Immediate by Fact 1.6.
Theorem 2.2. The relation satisfies:
-
1. (invariance) If
and
$\sigma \in \mathrm {Aut}(\mathbb {M})$ , then
.
-
2. (normality) If
, then
.
-
3. (monotonicity) If
,
$A'\subseteq A$ ,
$B'\subseteq B$ , then
.
-
4. (left and right existence) For all A and B,
and
.
-
5. (right and left extension) If
and
$B'\supseteq B$ , then there is
$A' \equiv _{BC} A$ such that
. Similarly, if
$A' \supseteq A$ , then there is
$B'\equiv _{AC} B$ such that
.
-
6. (finite character) We have
if and only if for all finite
$A_0\subseteq A$ and
$B_0\subseteq B$ , we have
.
-
7. (left transitivity) If
$C \subseteq B \subseteq A$ ,
and
, then
.
-
8. (local character on a club) For every finite tuple a and for every set of parameters B, there is a club
$\mathcal {C} \subseteq [B]^{\leq |T|}$ such that
and
for all
$C \in \mathcal {C}$ .
-
9. (antireflexivity) We have
if and only if
$a\in \mathrm {acl}(C)$ .
-
10. (algebraicity)Footnote 1 If
, then
and
.
Proof. Invariance is clear from the definition. The implication from to
is also clear from the definition, and the statement of normality follows from this by extension. Monotonicity follows from the fact that adding dummy variables to a generically stable partial type preserves generic stability.
Existence (on both sides) follows directly from Lemma 2.1 since clearly and
.
To prove right extension, assume that and let
$B'= B\cup B"$
. Let
$\pi (x\hat {~}x")$
be the unique maximal global partial type consistent with
$\mathrm {tp}(BB"/C)$
which is generically stable over C. We first show that
$\mathrm {tp}(B/CA)\cup \pi |_{CA}$
is consistent. By Proposition 1.11, the partial type
$(\exists x")\pi (x\hat {~}x")$
is generically stable over C. It is therefore consistent with
$\mathrm {tp}(B/CA)$
, and the result follows. To conclude, let
$B_{*}' = (B_{*}, B_{*}") \vDash \mathrm {tp}(B/CA)\cup \pi |_{CA}$
. By an automorphism over
$CA$
, we may assume
$B_{*} = B$
. Then
by Corollary 1.9 and
$BB_{*}" \equiv _{C} BB"$
. Pick
$A'$
such that
$A'BB" \equiv _{C} ABB_{*}"$
. Then, in particular, we have
$A' \equiv _{CB} A$
and, by invariance,
as desired.
Left extension follows by definition: If
$B \vDash \pi |_{AC}$
for
$\pi $
generically stable over C, then
$\pi |_{CA'} \cup \text {tp}(B/AC)$
is consistent, so let
$B'$
realize it.
Finite character on the left follows from the definition. To see finite character on the right, assume that we have . Then there is a generically stable partial type
$\pi (x)$
extending
$\mathrm {tp}(B/C)$
and a formula
$\phi (x)\in \pi |_{AC}$
such that
$B\vDash \neg \phi $
. The formula
$\phi $
only involves a finite subset
$B_0\subseteq B$
. Write
$B = B_0\cup B'$
and correspondingly split the variable
$x = x_0\hat {~}x'$
. By Proposition 1.11, the partial type
$\pi _0(x_0) := (\exists x')\pi (x_0 \hat {~} x')$
is generically stable over C. Then the formula
$\phi (x)$
is a consequence of
$\pi _0$
and we see that
$B_0$
does not satisfy
$\pi _0 |_{AC}$
. Hence,
.
Next, we consider left transitivity. We will assume and
. Let
$\pi \supseteq \text {tp}(d/C)$
denote the maximal global partial type that is generically stable over C, and let
$\tilde {\pi } \supseteq \text {tp}(d/Cb)$
denote the maximal global partial type that is generically stable over
$Cb$
. We want to show
$d \vDash \pi |_{Cab}$
so pick
$\phi (x;a,b) \in \pi $
, and we will show that
$\vDash \phi (d;a,b)$
. By our assumption that
, we know that
$\text {tp}(d/Cb) \cup \pi $
is consistent. Since
$\mathrm {tp}(d/Cb)$
is a complete type and
$\pi $
is generically stable over C, we clearly have that
$\text {tp}(d/Cb) \cup \pi $
is generically stable over
$Cb$
(one can also see this using Lemma 1.8), hence contained in
$\tilde {\pi }$
. Thus,
$\phi (x;a,b) \in \tilde {\pi }$
and the fact that
entails that
$\vDash \phi (d;a,b)$
as desired.
We now prove local character on a club. By Lemma 2.1, if , then
and
. In particular, this happens if
$\text {tp}(B/aC)$
is finitely satisfiable in C. Therefore, it suffices to show that the set
$\mathcal {C}$
defined by

is a club of
$[B]^{\leq |T|}$
. The set
$\mathcal {C}$
is clearly closed under unions of chains of length
$\leq |T|$
, so we show it is unbounded. Pick any
$X \in [B]^{\leq |T|}$
. Inductively, we will build a sequence of sets
$(C_{i})_{i < \omega }$
such that, for all
$i < \omega $
, we have the following:
-
○
$X \subseteq C_{i} \subseteq C_{i+1} \subseteq B$ .
-
○
$|C_{i}| \leq |T|$ .
-
○ If
$\varphi (x;y) \in L(C_{i})$ and there is some
$b \in B$ with
$\vDash \varphi (b;a)$ , then there is some
$b' \in C_{i+1}$ with
$\vDash \varphi (b';a)$ .
There is no problem in carrying out the induction: We begin with
$C_{0} = X$
, and since
$|C_{i}| \leq |T|$
, there are only
$|T|$
many formulas
$\varphi (x;a)$
realized by some tuple in B and we form
$C_{i+1}$
by adding to
$C_{i}$
one tuple from B for each such formula. Then we put
$C = \bigcup _{i} C_{i}$
. By construction,
$\text {tp}(B/Ca)$
is finitely satisfiable in C and hence
$X \subseteq C \in \mathcal {C}$
.
For antireflexivity, note that the partial type generated by
$\{x\neq b : b\in \mathbb {M}\}$
is a generically stable partial type, consistent with
$\mathrm {tp}(a/C)$
if
$a\notin \mathrm {acl}(C)$
. Therefore,
implies that
$a\in \mathrm {acl}(C)$
. For the other direction, suppose
$a \in \mathrm {acl}(C)$
and let A be the finite set of realizations of
$\mathrm {tp}(a/C)$
. By extension, there is
$A' \equiv _{C} A$
such that
, but, as a set, we must have
$A = A'$
so
follows by monotonicity.
Algebraicity: Suppose that . The fact that
follows by right extension and invariance. Similarly,
follows from left extension and invariance.
Remark 2.3. The form of local character in (5) was first isolated for Kim-independence in NSOP
$_{1}$
theories in [Reference Kaplan, Ramsey and ShelahKRS19]. It, of course, implies the usual formulation of local character but is a more suitable analogue of the local character of non-forking independence in simple theories for contexts without base monotonicity. Additionally, the proof of local character plus Fact 1.6 imply local character on the left since finite satisfiability implies non-forking. That is, the proof establishes that, for every finite tuple a and set B, there is a club
$\mathcal {C} \subseteq [B]^{\leq |T|}$
such that
for all
$C \in \mathcal {C}$
.
Consider the following property:
-
(P) If
$\pi (x)$ is generically stable, then so is
$\pi ^{(\omega )}(x_0, x_1,\ldots )$ .
Proposition 2.4. Assume that (P) holds, then satisfies symmetry: For any
$A,a,b$
, we have

Proof. Assume that , but
. Let
$\pi (x)$
be generically stable over A, consistent with
$\mathrm {tp}(a/A)$
, but not
$\mathrm {tp}(a/Ab)$
. Let
$\phi (x,y)\in \mathrm {tp}(a,b/A)$
be such that
$\neg \phi (x,b)\in \pi |Ab$
. Let
$n<\omega $
be maximal such that there is
$(a_1,\ldots ,a_n)\vDash \pi ^{(n)}|_{A}$
with
$\bigwedge _{i\leq n} \phi (a_i,b)$
. (Note that such an n exists by generic stability and ind-definability.) Consider the partial type

This type is generically stable by property (P) and Corollary 1.12, and it is consistent with
$\mathrm {tp}(b/A)$
by definition. As
, it is consistent with
$\mathrm {tp}(b/Aa)$
. But this means that we can find
$a_1,\ldots ,a_n\vDash \pi ^{(n)}|_{Aa}$
with
$\bigwedge _{i\leq n} \phi (a_i,b)$
. But then
$(a_0:=a,a_1,\ldots ,a_n)\vDash \pi ^{(n+1)}|_{A}$
and
$\bigwedge _{i< n} \phi (a_i,b)$
holds. This contradicts the maximality of n.
Remark 2.5. In [Reference SimonSim20, Example 2.12], there is an example which shows that property P does not hold in general for generically stable partial types.
Question 2.6. Is symmetric in general? Does it always satisfy transitivty on the right?
3 Treeless theories
In this section, we define the treeless theories. We begin by showing that treetop indiscernibles, defined in the first subsection, have the modeling property. Then we define treelessness in terms of a form of indiscernible collapse from the structure on the leaves of the treetop indiscernible to an indiscernible sequence.
3.1 Generalized indiscernibles and Ramsey classes
In this subsection, we will define generalized indiscernibles and introduce a new kind of indiscernible tree, which allow us later on to define the treeless theories.
Definition 3.1. Suppose I is an
$L'$
-structure, where
$L'$
is some language.
-
1. We say
$(a_{i} : i \in I)$ is a set of I-indexed indiscernibles if whenever
$(s_{0}, \ldots , s_{n-1})$ ,
$(t_{0}, \ldots , t_{n-1})$ are tuples from I with
$$ \begin{align*}\text{qftp}_{L'}(s_{0}, \ldots, s_{n-1}) = \text{qftp}_{L'}(t_{0}, \ldots, t_{n-1}), \end{align*} $$
$$ \begin{align*}\text{tp}(a_{s_{0}},\ldots, a_{s_{n-1}}) = \text{tp}(a_{t_{0}},\ldots, a_{t_{n-1}}). \end{align*} $$
-
2. We define the (generalized) EM-type of
$(a_{i})_{i \in I}$ , written
$\mathrm {EM}_{L'}(a_{i} : i \in I)$ , to be the partial type
$\Gamma (x_{i} : i \in I)$ such that
$\varphi (x_{i_{0}}, \ldots , x_{i_{n-1}}) \in \Gamma $ if and only if
$\vDash \varphi (a_{j_{0}}, \ldots , a_{j_{n-1}})$ for all tuples
$(j_{0}, \ldots , j_{n-1})$ from I with
$(j_{0}, \ldots , j_{n-1}) \vDash \mathrm {qftp}_{L'}(i_{0}, \ldots , i_{n-1})$ . If
$(b_{i} : i \in I) \vDash \mathrm {EM}_{L'}(a_{i} : i \in I)$ , we say
$(b_{i} : i \in I)$ is locally based on
$(a_{i} : i \in I)$ .
-
3. We say that I-indexed indiscernibles have the modeling property if, given any
$(a_{i} : i \in I)$ from
$\mathbb {M}$ , there is an I-indexed indiscernible (b i : i ∈ I) in
$\mathbb {M}$ locally based on
$(a_{i} : i \in I)$ .
Remark 3.2. When I-indexed indiscernibles have the modeling property and J is an
$L'$
-structure with
$\mathrm {Age}(I) = \mathrm {Age}(J)$
, we additionally have that, given
$(a_{i})_{i \in I}$
, there is a J-indexed indiscernible
$(b_{i})_{i \in J}$
locally based on
$(a_{i})_{i \in I}$
. This follows easily by compactness, and we will often use the modeling property in this form.
For the remainder of the paper, except for the familiar case of indiscernible sequences, we will only ever consider I-indexed indiscernibles in the case where I is a tree, though there are important differences between the notions of indiscernibility one obtains based on different choices of language for the tree I. The language
$L_{0}$
is the language consisting of two binary relations
$\unlhd $
and
$\leq _{lex}$
and a binary function
$\wedge $
. The tree
$\omega ^{<\omega }$
, for example, may be naturally viewed as an
$L_{0}$
-structure, where
$\unlhd $
is interpreted the tree partial order,
$\leq _{lex}$
as the lexicographic order and
$\wedge $
as the binary meet function. If I is an
$L_{0}$
-structure with
$\mathrm {Age}(I) = \mathrm {Age}(\omega ^{<\omega })$
, then we refer to I-indexed indiscernibles as strongly indiscernible trees.
If
$\alpha $
is an ordinal, we define a language
$L_{s,\alpha }$
which consists of
$L_{0}$
, together with unary predicates
$P_{\beta }$
for every
$\beta < \alpha $
. The tree
$\omega ^{<\alpha }$
can be viewed as an
$L_{s,\alpha }$
-structure by giving the symbols of
$L_{0}$
their natural interpretation and interpreting each predicate
$P_{\beta }$
as
$\omega ^{\beta }$
, that is, as the set of nodes at level
$\beta $
in the tree. If
$(a_{i})_{i \in I}$
is an I-indexed indiscernible for some
$L_{s,\alpha }$
-structure I with
$\mathrm {Age}(I) = \mathrm {Age}(\omega ^{<\alpha })$
for some
$\alpha $
, then we refer to
$(a_{i})_{i \in I}$
as an s-indiscernible tree.
Fact 3.3 [Reference Kim, Kim and ScowKKS14, Theorem 4.3] [Reference Takeuchi and TsuboiTT12, Theorem 16].
Let denote I
s
be the L
s, ω
-structure (ω <ω
, ⊴, <
lex
,
$\wedge$
, (P
α
)
α<ω
) with all symbols being given their intended interpretations and each P
α
naming the elements of the tree at level α, and let
$I_{0}$
denote its reduct to
$L_{0} = \{ \unlhd , \leq _{lex}, \wedge \}$
. Then both
$I_{0}$
-indexed indiscernibles (strongly indiscernible trees) and I
s
-indexed indiscernibles (s-indiscernible trees) have the modeling property.
Remark 3.4. Trees of height greater than
$\omega $
may also be considered as s-indiscernible trees, though this requires adding additional predicates to the language on the index model: We say, for example, that
$(a_{\eta })_{\eta \in \omega ^{<\beta }}$
is an s-indiscernible tree if it is an
$\omega ^{<\beta }$
-indexed indiscernible where
$\omega ^{<\beta }$
is considered as a structure in the language
$L_{s,\beta }$
which contains predicates
$(P_{\alpha })_{\alpha < \beta }$
for all
$\beta $
levels of the tree. As the language on the index model of an s-indiscernible tree is typically clear from context, we will not specify it explicitly.
We will use the phrase Fraïssé class to denote a uniformly locally finite class of finite structures satisfying the hereditary property, the joint embedding property and the amalgamation property. Given any L-structures
$A,B$
, we write
$\mathrm {Emb}_{L}(A,B)$
to denote the set of embeddings from A to B. We omit the L subscript when it is understood from context.
Recall that a Fraïssé class
$\mathcal {K}$
has the Ramsey property if, given any
$A \subseteq B$
and
$r \in \omega $
, there is some
$C \in \mathcal {K}$
such that, if
$\chi : \mathrm {Emb}(A,C) \to r$
, there is some
$\alpha \in \mathrm {Emb}(B,C)$
such that
$\chi |_{\alpha \circ \mathrm {Emb}(A,B)}$
is constant, where

A Fraïssé class satisfying the Ramsey property is called a Ramsey class.
There is a tight connection between Ramsey classes and generalized indiscernibles with the modeling property, established by the following theorem of Scow:
Fact 3.5 [Reference ScowSco12, Theorem 3.12].
Suppose I is an infinite, locally finite structure expanding a linear order in the language
$L'$
, such that quantifier-free types are isolated by quantifier-free formulas. Then I-indexed indiscernibles have the modeling property if and only if
$\mathrm {Age}(I)$
is a Ramsey class.
The language
$L_{0,P} = \{\unlhd , \wedge , <_{lex}, P\}$
, where P is a unary predicate. The class
$\mathbb {K}_{0,P}$
consists of all finite
$\wedge $
-trees A in which every element of
$P^{A}$
is a leaf – that is, each
$A \in \mathbb {K}_{0,P}$
satisfies the axiom

Note that if
$\omega ^{\leq \omega }$
is viewed as an
$L_{0,P}$
structure in which
$\wedge , \unlhd $
and
$<_{lex}$
receives their natural interpretations and P is interpreted as
$\omega ^{\omega }$
, then
$\mathrm {Age}(\omega ^{\leq \omega }) = \mathbb {K}_{0,P}$
.
Definition 3.6. We define a treetop indiscernible to be any I-indexed indiscernible where I is an
$L_{0,P}$
-structure with
$\mathrm {Age}(I) = \mathbb {K}_{0,P}$
.
We aim to show that treetop indiscernibles have the modeling property or, equivalently, that
$\mathbb {K}_{0,P}$
is a Ramsey class. In the arguments below, it will be useful to introduce the following notation: If I is an
$L_{0,P}$
-structure with
$\mathrm {Age}(I) = \mathbb {K}_{0,P}$
, we will write
$I_{+}$
for
$P(I)$
, and we will write
$I_{-}$
for
$I \setminus P(I)$
. In other words,
$I_{+}$
names the leaves of the tree I and
$I_{-}$
names the nonleaves.
Recall that the tree
$\omega ^{\leq \omega }$
may be viewed as an index model for s-indiscernible trees, in which case this tree is viewed as a structure in the language
$L_{s,\omega +1} = \{\wedge , \unlhd , \leq _{lex}, (P_{\alpha })_{\alpha \leq \omega }\}$
, where
$P_{\alpha }$
is interpreted as the
$\alpha $
th level of the tree. We may regard the
$L_{0,P}$
-structure on
$\omega ^{\leq \omega }$
as a reduct of its
$L_{s,\omega +1}$
-structure, identifying P with
$P_{\omega }$
.
Lemma 3.7. Suppose
$\overline {\eta }, \overline {\nu }$
are
$\wedge $
-closed tuples from
$\omega ^{\leq \omega }$
, and we write

such that
$\overline {\eta }_{-},\overline {\nu }_{-}$
are tuples from
$\omega ^{<\omega }$
and
$\overline {\eta }_{+}, \overline {\nu }_{+}$
are from
$\omega ^{\omega }$
. Then if
$\overline {\eta }_{-} = \overline {\nu }_{-}$
and
$\mathrm {qftp}_{L_{0,P}}(\overline {\eta }) = \mathrm {qftp}_{L_{0,P}}(\overline {\nu })$
, then we have
$\mathrm {qftp}_{L_{s,\omega +1}}(\overline {\eta }) = \mathrm {qftp}_{L_{s,\omega +1}}(\overline {\nu })$
.
Proof. Since
$\overline {\eta }$
and
$\overline {\nu }$
are
$\wedge $
-closed and
$\mathrm {qftp}_{L_{0,P}}(\overline {\eta }) = \mathrm {qftp}_{L_{0,P}}(\overline {\nu })$
, it is enough to show that the map
$\overline {\eta } \mapsto \overline {\nu }$
preserves every predicate of the form
$P_{i}$
for
$i \leq \omega $
. But this mapping takes
$\overline {\eta }_{-}$
to
$\overline {\nu }_{-}$
so preserves
$P_{i}$
for every
$i < \omega $
. The mapping also takes
$\overline {\eta }_{+}$
to
$\overline {\nu }_{+}$
so preserves
$P_{\omega }$
as well.
We will argue that
$\mathrm {Age}_{L_{0,P}}(\omega ^{\leq \omega })$
is a Ramsey class. In order to do this, it suffices, by Fact 3.5, to show the following:
Lemma 3.8. Given any
$(a_{\eta })_{\eta \in \omega ^{\leq \omega }}$
, there is some
$(b_{\eta })_{\eta \in \omega ^{\leq \omega }}$
which is treetop indiscernible and locally based on
$(a_{\eta })_{\eta \in \omega ^{\leq \omega }}$
.
Proof. Let
$(a^{\prime }_{\eta })_{\eta \in \omega ^{\leq \omega }}$
be an s-indiscernible tree locally based on
$(a_{\eta })_{\eta \in \omega ^{\leq \omega }}$
.
Claim 3.9. It suffices to find
$(b_{\eta })_{\eta \in \omega ^{\leq \omega }}$
which is treetop indiscernible and locally based on
$(a^{\prime }_{\eta })_{\eta \in \omega ^{\leq \omega }}$
.
Proof of claim.
Suppose
$(b_{\eta })_{\eta \in \omega ^{\leq \omega }}$
is treetop indiscernible and locally based on
$(a^{\prime }_{\eta })_{\eta \in \omega ^{\leq \omega }}$
. Suppose further that
$\overline {\eta }$
is a tuple from
$\omega ^{\leq \omega }$
and
$\vDash \varphi (b_{\overline {\eta }})$
. By the local basedness of
$(b_{\eta })_{\eta \in \omega ^{\leq \omega }}$
as a treetop indiscernible, there is
$\overline {\nu }$
in
$\omega ^{\leq \omega }$
with
$\mathrm {qftp}_{L_{0,P}}(\overline {\eta }) = \mathrm {qftp}_{L_{0,P}}(\overline {\nu })$
and
$\vDash \varphi (a^{\prime }_{\overline {\nu }})$
. Then as
$(a^{\prime }_{\eta })_{\eta \in \omega ^{\leq \omega }}$
is locally based on
$(a_{\eta })_{\eta \in \omega ^{\leq \omega }}$
as an s-indiscernible tree, there is
$\overline {\xi }$
in
$\omega ^{\leq \omega }$
such that
$\mathrm {qftp}_{L_{s,\omega +1}}(\overline {\nu }) = \mathrm {qftp}_{L_{s,\omega +1}}(\overline {\xi })$
and
$\vDash \varphi (a_{\overline {\xi }})$
. It follows then that
$\mathrm {qftp}_{L_{0,P}}(\overline {\eta }) = \mathrm {qftp}_{L_{0,P}}(\overline {\xi })$
. This shows
$(b_{\eta })_{\eta \in \omega ^{\leq \omega }}$
is locally based on
$(a_{\eta })_{\eta \in \omega ^{\leq \omega }}$
.
So now let
$\mathrm {EM}_{L_{0,P}}((a^{\prime }_{\eta })_{\eta \in \omega ^{\leq \omega }})$
denote the partial type in the variables
$(x_{\eta })_{\eta \in \omega ^{\leq \omega }}$
consisting of the following set of formulas:

Let
$\Gamma $
denote the partial type consisting of
$\mathrm {EM}_{L_{0,P}}((a^{\prime }_{\eta })_{\eta \in \omega ^{\leq \omega }})$
, and the collection of formulas asserting that
$(x_{\eta })_{\eta \in \omega ^{\leq \omega }}$
is treetop indiscernible. By Claim 3.9, it suffices to show
$\Gamma $
is consistent. A finite subset of
$\Gamma $
will be contained in

for some finite
$\Delta $
, a finite tuple
$\overline {\xi }$
from
$\omega ^{\leq \omega }$
and
$\wedge $
-closed tuples
$\overline {\eta }_{i},\overline {\nu }_{i}$
with
$\overline {\nu }_{i} \vDash \mathrm {qftp}_{L_{0,P}}(\overline {\eta }_{i})$
for all
$i < k$
. Let C be a finite
$L_{0,P}$
-substructure of
$\omega ^{\leq \omega }$
containing
$\overline {\xi }$
and
$\overline {\eta }_{i}, \overline {\nu }_{i}$
for all
$i < k$
and so
$C_{-}$
is the
$L_{0}$
-substructure of
$\omega ^{<\omega }$
consisting of the elements of
$C \setminus P(C)$
.
For each
$i < k$
, let
$q_{i} = \mathrm {qftp}_{L_{0,P}}(\overline {\eta }_{i})$
and define a coloring
$c_{i} : q_{i}(\omega ^{\leq \omega }) \to S^{l(\overline {\eta }_{i})}_{\Delta }(\emptyset )$
by

for all
$\overline {\zeta } \in q_{i}(\omega ^{\leq \omega })$
. Note that, since
$\Delta $
is finite, we know
$S^{l(\overline {\eta }_{i})}_{\Delta }(\emptyset )$
is finite.
Let, for each
$i < k$
,
$\overline {\eta }_{-,i}$
be the subtuple of
$\overline {\eta }_{i}$
consisting of those elements not in
$\omega ^{\omega }$
and likewise for
$\overline {\nu }_{-,i}$
. Let
$q_{-,i} = \mathrm {qftp}_{L_{0}}(\overline {\eta }_{-,i}) = \mathrm {qftp}_{L_{0}}(\overline {\nu }_{-,i})$
. Then we define a coloring
$c_{-,i} : q_{-,i}(\omega ^{<\omega }) \to S^{l(\overline {\eta }_{i})}_{\Delta }(\emptyset )$
by setting, for each
$\overline {\mu } \in q_{-,i}(\omega ^{<\omega })$
,

for any
$\overline {\zeta } \in q_{i}(\omega ^{\leq \omega })$
with
$\overline {\zeta }_{-} = \overline {\mu }$
. By Lemma 3.7 and the s-indiscernibility of
$(a^{\prime }_{\eta })_{\eta \in \omega ^{\leq \omega }}$
,
$c_{-,i}$
is well-defined. As
$\mathrm {Age}_{L_{0}}(\omega ^{<\omega })$
is a Ramsey class, by Fact 3.3, there is some
$C_{-}' \cong C_{-}$
, an
$L_{0}$
-substructure of
$\omega ^{<\omega }$
, such that
$c_{-,i}|_{q_{-,i}(C^{\prime }_{-})}$
is constant for all
$i < k$
. Choose any
$C' \supseteq C^{\prime }_{-}$
, with
$C'$
a substructure of
$\omega ^{\leq \omega }$
and
$C'$
isomorphic to C as an
$L_{0,P}$
-structure. Then, unravelling definitions, we have that
$c_{i}|_{q_{i}(C')}$
is constant for all
$i < k$
. Letting
$\overline {\xi }'$
,
$\overline {\eta }^{\prime }_{i}$
and
$\overline {\nu }^{\prime }_{i}$
denote the corresponding tuples in
$C'$
, we have that
$a^{\prime }_{\overline {\xi '}}$
,
$(a^{\prime }_{\overline {\eta }_{i}})_{i < k}$
and
$(a^{\prime }_{\overline {\nu }_{i}})_{i < k}$
realize the desired finite subset of
$\Gamma $
. This concludes the proof.
Corollary 3.10.
$\mathbb {K}_{0,P}$
is a Ramsey class.
As
$\mathbb {K}_{0,P}$
is a Ramsey class, it is, in particular, a Fraïssé class, by [Reference BodirskyBod15, Theorem 2.13]. We denote the Fraïssé limit of
$\mathbb {K}_{0,P}$
by
$\mathcal {T}$
. This structure will play an important role in the definition of treeless theories in the subsection below.
3.2 Treeless theories
Given an
$L_{0,P}$
-structure I with
$\mathrm {Age}(I) = \mathrm {Age}(\omega ^{\leq \omega })$
and
$\eta \in I$
, let
$C(\eta ) = \{\nu \in P(I) : \eta \unlhd \nu \}$
, that is, the leaves of I that are in the cone above
$\eta $
.
Definition 3.11. Say that T is treeless if whenever
$(a_{\eta })_{\eta \in \mathcal {T}}$
is treetop indiscernible and
$\xi \in \mathcal {T}$
, then
$(a_{\eta })_{\eta \in C(\xi )}$
is an indiscernible sequence over
$a_{\xi }$
(i.e., is order-indiscernible over
$a_{\xi }$
with respect to
$<_{lex}$
).
Proposition 3.12. The following are equivalent:
-
1. T is treeless.
-
2. If
$\mathcal {S}$ is any
$L_{0,P}$ -structure with
$\mathrm {Age}(\mathcal {S}_{\xi }) = \mathbb {K}_{0,P}$ for all
$\xi \in \mathcal {S}_{-}$ , where
$S_{\xi } = \{\eta \in \mathcal {S} : \xi \unlhd \eta \}$ , and
$(a_{\eta } : \eta \in \mathcal {S})$ is treetop indiscernible, then for any
$\eta \in \mathcal {S}$ ,
$(a_{\eta } : \eta \in C(\eta ))$ is order indiscernible over
$a_{\eta }$ .
-
3. If
$(a_{\eta } : \eta \in \omega ^{\leq \omega })$ is treetop indiscernible, then
$(a_{\eta } : \eta \in \omega ^{\omega })$ is order indiscernible over
$a_{\emptyset }$ .
Proof. The implication
$(2)\implies (1)$
is trivial and
$(1) \implies (3)$
is easy, using that
$\mathrm {Age}(\omega ^{\leq \omega }) = \mathrm {Age}(\mathcal {T})$
, so we show
$(3)\implies (2)$
. Assume (3), and suppose
$\mathcal {S}$
is an
$L_{0,P}$
-structure with
$\mathrm {Age}(\mathcal {S})=\mathbb {K}_{0,P}$
,
$(a_{\eta } : \eta \in \mathcal {S})$
is a treetop indiscernible and
$\xi \in \mathcal {S}_{-}$
. We must show
$(a_{\eta } : \eta \in C(\xi ))$
is order-indiscernible over
$a_{\xi }$
. By assumption,
$S_{\xi }$
satisfies
$\mathrm {Age}(S_{\xi }) \supseteq \mathrm {Age}(\omega ^{\leq \omega })$
. Consequently, for each finite tuple
$\overline {\eta }$
from
$\omega ^{\leq \omega }$
, there is some
$\overline {\nu }$
in
$\mathcal {S}_{\xi }$
such that
$\mathrm {qftp}_{L_{0,P}}(\overline {\eta }) = \mathrm {qftp}_{L_{0,P}}(\overline {\nu })$
. We define the type
$p_{\overline {\eta }}(x_{\overline {\eta }})$
to be
$\text {tp}(a_{\overline {\nu }})$
for some (equivalently, all) such
$\overline {\nu }$
. Then, by compactness,
$\Gamma (x_{\eta } : \eta \in \omega ^{\leq \omega }) = \bigcup _{\overline {\eta }} p_{\overline {\eta }}$
is consistent, where
$\overline {\eta }$
ranges over all finite tuples of
$\omega ^{\leq \omega }$
. Moreover, letting
$(b_{\eta } : \eta \in \omega ^{\leq \omega })$
be a realization, we have that
$(b_{\eta } : \eta \in \omega ^{\leq \omega })$
is treetop indiscernible. By assumption, then,
$(b_{\eta } : \eta \in \omega ^{\omega })$
is order indiscernible over
$b_{\emptyset }$
. By construction, this entails that
$(a_{\eta } : \eta \in C(\xi ))$
is order indiscernible over
$a_{\xi }$
. As the case of
$\xi \in \mathcal {S}_{+}$
is trivial, this completes the proof.
If T is NIP, the definition of treeless can be weakened to omit the condition that the leaves are order indiscernible over the root:
Proposition 3.13. Assume T is NIP. Suppose that for all treetop indiscernibles
$(a_{\eta })_{\eta \in \omega ^{\leq \omega }}$
, the sequence
$(a_{\eta })_{\eta \in \omega ^{\omega }}$
is an indiscernible sequence. Then T is treeless.
Proof. Suppose
$(a_{\eta })_{\eta \in \omega ^{\leq \omega }}$
is treetop indiscernible. We must show that
$(a_{\eta })_{\eta \in \omega ^{\omega }}$
is indiscernible over
$a_{\emptyset }$
. By compactness, we may stretch the given treetop indiscernible to
$(a_{\eta })_{\eta \in \kappa ^{\leq \omega }}$
with
$\kappa = |T|^{+}$
. Since T is NIP, by [Reference SimonSim15, Proposition 2.8], there is an end segment
$J \subseteq \kappa ^{\omega }$
such that
$(a_{\eta })_{\eta \in J}$
is
$a_{\emptyset }$
-indiscernible. By treetop indiscernibility, it follows that
$(a_{\eta })_{\eta \in \kappa ^{\omega }}$
is
$a_{\emptyset }$
-indiscernible as well. Therefore, T is treeless.
Question 3.14. Is Proposition 3.13 true without the assumption that T is NIP? Note that weakened notion of treeless, in which the leaves indexed by
$\omega ^{\omega }$
in a treetop indiscernible
$(a_{\eta })_{\eta \in \omega ^{\leq \omega }}$
are only required to be an indiscernible sequence (not necessarily indiscernible over
$a_{\emptyset }$
) suffices for many of the observations.
The following related question was suggested to us by Artem Chernikov:
Question 3.15. To check treelessness, does it suffice to consider triples of leaves? More precisely, if whenever
$(a_{\eta })_{\eta \in \mathcal {T}}$
is a treetop indiscernible and, for all
$\eta _{0} <_{lex} \eta _{1} <_{lex} \eta _{2}$
and
$\nu _{0} <_{lex} \nu _{1} <_{lex} \nu _{2}$
from
$\mathcal {T}_{-}$
, we have
$(a_{\eta _{0}},a_{\eta _{1}}, a_{\eta _{2}}) \equiv _{a_{\emptyset }} (a_{\nu _{0}}, a_{\nu _{1}}, a_{\nu _{2}})$
, does it follow that T is treeless?
Example 3.16. Any structure homogeneous in a binary language. Any theory of a pure linear order is (distal and) treeless since it eliminates quantifiers in a binary language [Reference SimonSim15, Lemma A.1].
Example 3.17. The theory of any nontrivial ordered abelian group is not treeless. To see this, let G be any nontrivial ordered abelian group. We may assume G is
$\aleph _{0}$
-saturated, and hence we can fix some
$g> 0$
in G which is n-divisible for all n (take g to be in the intersection of
$n\cdot G$
for all
$n<\omega $
). Fix
$2 \leq n,m < \omega $
. Then for each
$\eta \in n^{\leq m}$
, as g is k-divisible for all k, we can define

Consider some
$\eta _{0} <_{lex} \eta _{1} <_{lex} \eta _{2} <_{lex} \eta _{3}$
in
$n^{m}$
with

and

(and thus
$(\eta _{0} \wedge \eta _{2}) = (\eta _{0} \wedge \eta _{3}) = (\eta _{1} \wedge \eta _{2}) = (\eta _{1} \wedge \eta _{3})$
). Then we have

and

Hence, by compactness and Corollary 3.10, we can find a treetop indiscernible
$(b_{\eta })_{\eta \in \omega ^{\leq \omega }}$
in a model of
$\mathrm {Th}(G)$
satisfying the same pair of inequalities, which shows that
$(b_{\eta })_{\eta \in \omega ^{\omega }}$
is not an indiscernible sequence, hence
$\mathrm {Th}(G)$
is not treeless.

Remark 3.18. Even if T is treeless, it may be the case that
$(a_{\eta })_{\eta \in \omega ^{\leq \omega }}$
is s-indiscernible and
$(a_{\eta })_{\eta \in \omega ^{\omega }}$
is not an indiscernible sequence (this
$(a_{\eta })_{\eta \in \omega ^{\leq \omega }}$
will be necessarily not treetop indiscernible). For example, let T be the model companion of the theory in the language
$L = \{R_{n} : n < \omega \}$
that says that the binary relation
$R_{n}$
is a graph for each n. So in T, each
$R_n$
defines a random graph and these graphs interact totally independently. We may choose vertices
$(a_{\eta })_{\eta \in \omega ^{\leq \omega }}$
so that, for leaves
$\eta , \nu \in \omega ^{\omega }$
,
$\vDash R_{n}(a_{\eta }, a_{\nu })$
holds if and only if the length of
$\eta \wedge \nu $
is n. This is preserved when passing to an s-indiscernible tree locally based on the
$(a_{\eta })_{\eta \in \omega ^{\leq \omega }}$
, so we can assume
$(a_{\eta })_{\eta \in \omega ^{\leq \omega }}$
is s-indiscernible. Clearly,
$(a_{\eta })_{\eta \in \omega ^{\omega }}$
is not an indiscernible sequence. However, T eliminates quantifiers and the language L is binary, so T is treeless.
Proposition 3.19. Suppose the theory
$T'$
is interpretable in the treeless theory T. Then
$T'$
is treeless.
Proof. Suppose
$T'$
is interpretable in T and E is a T-definable equivalence relation such that if
$M \vDash T$
, then
$M^{n}/E$
is the domain of a model of
$T'$
whose relations are definable in T. Let
$\mathbb {M}' = \mathbb {M}^{n}/E$
and let
$\pi : \mathbb {M}^{n} \to \mathbb {M}'$
denote the interpretation map. Suppose
$(a_{\eta })_{\eta \in \omega ^{\leq \omega }}$
is a treetop indiscernible in
$\mathbb {M}'$
. Then for each
$\eta \in \omega ^{\leq \omega }$
, we can choose some
$\tilde {a}_{\eta } \in \pi ^{-1}(a_{\eta })$
. We can then take
$(b_{\eta })_{\eta \in \omega ^{\leq \omega }}$
which is treetop indiscernible and locally based on
$(\tilde {a}_{\eta })_{\eta \in \omega ^{\leq \omega }}$
in
$\mathbb {M}$
. As T is treeless,
$(b_{\eta })_{\eta \in \omega ^{\omega }}$
is an indiscernible sequence over
$b_{\emptyset }$
. In particular,
$(\pi (b_{\eta }))_{\eta \in \omega ^{\omega }}$
is an indiscernible sequence over
$\pi (b_{\emptyset })$
. But since
$(a_{\eta })_{\eta \in \omega ^{\leq \omega }}$
was taken to be treetop indiscernible in
$\mathbb {M}'$
, we have, by local basedness, that
$(a_{\eta })_{\eta \in \omega ^{\leq \omega }} \equiv (\pi (b_{\eta }))_{\eta \in \omega ^{\leq \omega }}$
, hence
$(a_{\eta })_{\eta \in \omega ^{\omega }}$
is an indiscernible sequence over
$a_{\emptyset }$
, which shows
$T'$
is treeless.
Recall the following:
Definition 3.20. Suppose
$k \geq 1$
. We say that a formula
$\varphi (x;y_{0}, \ldots , y_{k-1})$
has the k-independence property (k-IP) if there is some array
$(a_{i,j} : i < k, j < \omega )$
such that, for all
$X \subseteq \omega ^{k}$
, there is some
$b_{X}$
such that

We say that a theory T has the k-independence property if some formula does modulo T. A theory without k-IP is called k-dependent.
Note that if a theory is k-dependent, then it is
$k'$
-dependent for all
$k' \geq k$
. The independence property is the same as
$1$
-IP. The k-dependence hierarchy was introduced by Shelah in [She07]. See also [Reference Chernikov, Palacin and TakeuchiCPT19] for further details on these classes of theories.
Proposition 3.21. If T is treeless, then T is
$2$
-dependent. In particular, T is k-dependent for all
$k \geq 2$
.
Proof. We prove the contrapositive. Suppose T has
$2$
-IP witnessed by the formula
$\varphi (x;y,z)$
. Then, by compactness, there is a sequence
$(b_{\eta }, c_{\eta } : \eta \in \omega ^{\omega })$
such that, for all
$X \subseteq \omega ^{\omega } \times \omega ^{\omega } $
, there is some
$a_{X}$
such that

Now, for each
$\eta \in \omega ^{\omega }$
, let

Choose, for each
$\eta \in \omega ^{\omega }$
some