この文書はAlweさんの企画された Mathematical Logic Advent Calendar 2023 の15日目の記事です。まだ日本語を使わなくてすみません🙏。
I learned the following argument from Garrett Ervin.
An elementary chain a sequence $C=\{M_i:i<\alpha\}$ of models over a common first-order language $\mathcal{L}$, indexed by ordinals $i$ below $\alpha$, such that for all $i\le j<\alpha$, $M_i\preceq M_j$ is an elementary extension.
We say $C$ is increasing if for all $i$ such that $i+1<\alpha$, it holds that $M_i\subsetneqq M_{i+1}$ is a proper extension.
We say $C$ is continuous if for all limit ordinals $\beta<\alpha$, it holds that $M_\beta=\bigcup_{i<\beta}M_i$ is the union of its predecessors $\{M_i:i<\beta\}$.
Let $M$ be an arbitrary model of size $\aleph_0$ over a countable language. The following hold:
In the cases (2), (3), and (4), the elementary chain $\{M_i:i<\kappa^+\}$ is maximal: The next model $M_{\kappa^+}=\bigcup_{i<\kappa^+}M_i$ always has size $\kappa^+$, so it cannot be isomorphic to the earlier models $M_i$ of size $\kappa$.
It is currently an open question whether case (3) always holds in $\textsf{ZFC}$.
The following lemma reveals the combinatorial content of the main theorem.
An increasing continuous chain of models $C=\{M_i:i<\alpha\}$ is defined similarly as an increasing continuous elementary chain, except for all $i\le j<\alpha$, we no longer require $M_i\preceq M_j$ to be an elementary extension, but we instead assume $M_i\subseteq M_j$ is some extension (where the inclusion $M_i\subseteq M_j$ respects the function, relation, and constant symbols from the language).
Given an infinite cardinal $\kappa$ and an ordinal $\alpha$, the following are equivalent:
Let $M=(\mathbb{Q},{<})$. Then the increasing continuous chain $\{M_i=(I_i,{<}):i<\alpha\}$ of linear orders is what we wanted.
To prove the implication (2 ⇒ 1), we need a theorem by Ehrenfeucht–Mostowski.
Every complete first-order theory $T$ has an extension $T^*$ into an expanded first-order language $\mathcal{L}(T^*)\supseteq\mathcal{L}(T)$, such that $|T^*|=|T|$, and $T^*$ has built-in Skolem functions: For every formula $\phi(v,\bar{w})$ in the language of $T^*$, there is a function symbol $f_\phi(\bar{w})$ such that $T^*$ contains the sentence $\forall\bar{w}[(\exists v\,\phi(v,\bar{w}))\rightarrow\phi(f_\phi(\bar{w}),\bar{w})]$.
For every complete first-order theory $T$ with infinite models and every linear order $(I,{<})$, there is a model $\mathfrak{C}$ of $T$ with a sequence $\{a_i:i\in I\}$ of order indiscernibles: For every two increasing sequences $i_1<\ldots< i_n$ and $j_1<\ldots< j_n$ in $I$ and every formula $\phi(x_1,\ldots,x_n)$, we have $\mathfrak{C}\vDash\phi[a_{i_1},\ldots,a_{i_n}]\leftrightarrow\phi[a_{j_1},\ldots,a_{j_n}]$.
Given a complete first-order theory $T$ with built-in Skolem functions, a model $\mathfrak{C}\vDash T$ with order indiscernibles $\{a_i:i\in I\}$, and a suborder $I_0\subseteq I$, the Skolem hull of $\mathbf{I}_0=\{a_i:i\in I_0\}$ is the $\mathcal{L}(T)$-submodel $\text{EM}(\mathbf{I}_0)\subseteq\mathfrak{C}$ with universe $\{f_\phi^{\mathfrak{C}}(\bar{a}):f_\phi\in \mathcal{L}(T),\bar{a}\subseteq\mathbf{I}_0\}$.
Let $T$, $\mathfrak{C}\vDash T$, $\mathbf{I}=\{a_i:i\in I\}$, and $\text{EM}({\cdot})$ be as in the definition above. For any $\mathbf{I}_0,\mathbf{I}_1\subseteq\mathbf{I}$, the following hold:
Let $\text{ED}(M)=\{\phi(c_{m_1},\ldots,c_{m_n}):M\vDash\phi[m_1,\ldots,m_n]\}$ be the elementary/complete diagram of $M$, so that $\text{ED}(M)$ is a countable complete first-order theory extending the theory $\text{Th}(M)$ of $M$, and every model $N\vDash\text{ED}(M)$ admits an elementary embedding $j_{M,N}:M\to N$ (mapping $m\mapsto c_m^N$), such that if $j_{N,N'}:N\to N'$ is an elementary embedding between models of $\text{ED}(M)$, then $j_{N,N'}\circ j_{M,N}=j_{M,N'}$ (see [1] Lemma 2.3.3).
Let $T\supseteq\text{ED}(M)$ be a countable complete first-order extension (in an expanded language) with built-in Skolem functions, and let $\mathfrak{C}\vDash T$, $\mathbf{I}=\{a_i:i\in I\}$, and $\text{EM}({\cdot})$ be as in the Ehrenfeucht–Mostowski theorem, where $I=\bigcup_{i<\alpha}I_i$.
For every $i<\alpha$, let $\mathbf{I}_i=\{a_i:i\in I_i\}$, and let $M_i=\text{EM}(\mathbf{I}_i)$. We can check that $\{M_i:i<\alpha\}$ is the elementary chain that we wanted:
In light of the lemma from the previous section, the main theorem is equivalent to the following proposition:
In this section we will prove parts (1) and (2).
Note that in the notation above, we cannot have $\alpha>\kappa^+$, or else the increasing continuous property of the chain implies that $I_{\kappa^+}$ always has size $\kappa^+>\kappa$, so that $I_{\kappa^+}$ cannot be isomorphic to previous linear orders of size $\kappa$.
Assume $\alpha<\kappa^+$. Let $\beta$ be a limit ordinal such that $\max\{\kappa,\alpha\}\le\beta<\kappa^+$. For $i<\alpha$, we can define the linear order
$$I_i=\beta\cup\{j+\tfrac{1}{2}:j< i\},$$
such that $(I_i,{<})$ is a well-order, and $\beta$ being a limit ordinal implies that $(I_i,{<})\cong(\beta,{<})$ for all $i<\alpha$. It is easy to check that $\{I_i:i<\alpha\}$ is an increasing continuous chain of linear orders of size $\kappa$ which are all isomorphic to $\beta$.
The next parts (2), (3), and (4) all concern the case $\alpha=\kappa^+$, so it is convenient to define a term for this case.
We say that a linear order $(I,{<})$ of size $\kappa$ is robust if there exists an increasing continuous chain $\{(I_i,{<}):i<\kappa^+\}$ of linear orders which are all isomorphic to $(I,{<})$.
Do robust linear orders always exist in every infinite cardinality?
We will show that $(\mathbb{Q},{<})$ is robust of size $\aleph_0$.
For $i<\omega_1$, we define the linear order
$$(I_i,{<})=\left((1+i)\times\mathbb{Q},{<}_{\text{lex}}\right).$$
This $I_i$ is constructed by replacing every point in the linear order $1+i$ with a copy of $\mathbb{Q}$. Then every $I_i$ is a dense linear order without endpoints of size $\aleph_0$, so by Cantor's back-and-forth method (see [1] Theorem 2.4.1), we get an isomorphism $(I_i,{<})\cong(\mathbb{Q},{<})$.
It's easy to see that $\{I_i:i<\omega_1\}$ forms an increasing continuous chain, witnessing robustness of $(\mathbb{Q},{<})$.
In this section we'll see that the Proper Forcing Axiom ($\textsf{PFA}$) implies a robust linear order of size $\aleph_1$.
A set of reals $X\subseteq\mathbb{R}$ is called $\kappa$-dense if for every nonempty open interval $U\subseteq\mathbb{R}$, there is $|X\cap U|=\kappa$.
Assume $\textsf{PFA}$. Every two $\aleph_1$-dense sets of reals $X$ and $Y$ are isomorphic.
Let $X$ be an $\aleph_1$-dense set of reals. We'll show that $(X,{<})$ is robust.
$\textsf{PFA}$ implies $|\mathbb{R}|=\aleph_2$ ([3] Theorem 31.23), so we can pick $\aleph_2$ many distinct reals $\{x_i\in\mathbb{R}:i<\omega_2\}$ that are not in $X$. For $i<\omega_2$, define
$$I_i=X\cup\{x_j:j< i\}.$$
Then every $I_i$ is $\aleph_1$-dense since $X\subseteq I_i$ is $\aleph_1$-dense and $|I_i|=\aleph_1$, and so Baumgartner's theorem implies $(I_i,{<})\cong (X,{<})$.
It's easy to see that $\{I_i:i<\omega_2\}$ is increasing continuous, which witnesses that $(X,{<})$ is robust of size $\aleph_1$.
I was given this argument in a Mathematics Stack Exchange answer by username Holo when I first asked about robust linear orders.
In an unpublished result , Itay Neeman has shown the consistency of “every two $\aleph_2$-dense sets of reals are isomorphic”.
In this section, we will prove part (4) of the main proposition:
If $\kappa$ is an infinite cardinal such that $\kappa^{<\kappa}=\kappa$, then there exists a robust linear order of size $\kappa$.
The assumption that $\kappa^{<\kappa}=\kappa$ is equivalent to the assumptions that $\kappa$ is regular and $2^{<\kappa}=\kappa$.
The Continuum Hypothesis ($\textsf{CH}$) implies that $\kappa=\aleph_1$ satisfies this condition, and so part (3) of the main proposition follows. More generally, the Generalized Continuum Hypothesis ($\textsf{GCH}$) implies that every infinite cardinal $\kappa$ satisfies $2^{<\kappa}=\kappa$, which then implies there are robust linear orders of size any regular cardinal $\kappa$.
In the discussions below, we fix a regular cardinal $\kappa$ satisfying $2^{<\kappa}=\kappa$.
$(2^\kappa,{<}_\text{lex})$ is the lexicographic ordering on binary strings of length $\kappa$.
Let $\mathcal{I}\subseteq2^\kappa$ be the set of all $x\in2^\kappa$ which are not eventually constant. The linear order $(\mathcal{I},{<})$ is the suborder of $(2^\kappa,{<}_\text{lex})$ induced by $\mathcal{I}$.
Define $\mathcal{J}\subseteq\mathcal{I}$ to be the set of all elements of $\mathcal{I}$ of the form $\sigma^\frown(01)^\kappa=\sigma0101\ldots$ for some $\sigma\in 2^{<\kappa}$.
Note that $|\mathcal{I}|=2^\kappa$ and $|\mathcal{J}|=2^{<\kappa}=\kappa$.
Let $A,B\subseteq\mathcal{I}$ be two subsets of $\mathcal{I}$ such that $|A|<\kappa$, $|B|<\kappa$, and for all $a\in A$ and $b\in B$ there is $a< b$. Then there exists $c\in\mathcal{J}$ such that $a< c$ for all $a\in A$, and $c< b$ for all $b\in B$.
We can define a supremum $f=\sup(A)\in 2^\kappa$ recursively, such that for all $i<\kappa$, $f(i)\in 2$ is the smallest such that for all $a\in A$, there is $a\upharpoonright(i+1)\le_\text{lex}(f\upharpoonright i)^\frown f(i)$. It is not hard to show that for all $a\in A$, there is $a\le_\text{lex} f$, and for all $g<_\text{lex}f$ there exists some $a\in A$ such that $g<_\text{lex}a$.
We split into cases.
Define a linear order $(I,{<})$ of size $\kappa$ to be saturated if for every $A,B\subseteq I$ such that $|A|<\kappa$, $|B|<\kappa$, and $a< b$ for all $a\in A$ and $b\in B$, there exists $c\in I$ such that $a< c$ for all $a\in A$ and $c< b$ for all $b\in B$.
Every two saturated linear orders $(L,{<})$ and $(K,{<})$ of size $\kappa$ are isomorphic.
Let $L\sqcup K=\{x_i:i<\kappa\}$.
We can build a chain of partial isomorphisms $\{f_i:i<\kappa\}$ such that $\text{dom}(f_i)\subseteq L$ and $\text{img}(f_i)\subseteq K$ both have size $<\kappa$, $f_i$ is an isomorphism between $(\text{dom}(f_i),{<}_L)$ and $(\text{img}(f_i),{<}_K)$, and $x_i\in\text{dom}(f_{i+1})\cup\text{img}(f_{i+1})$.
Finally, we get an isomorphism $f=\bigcup_{i<\kappa}f_i$ from $(L,{<})$ to $(K,{<})$.
We'll show the linear order $(\mathcal{J},{<})$ is robust of size $\kappa$.
Since $|\mathcal{I}|=2^\kappa$, we can pick $\kappa^+$ many distinct elements $\{x_i\in\mathcal{I}:i<\kappa^+\}$ that are not in $\mathcal{J}$. For $i<\kappa^+$, define
$$I_i=\mathcal{J}\cup\{x_j:j< i\}.$$
By the saturation lemma part 1, every $I_i$ is saturated of size $\kappa$, because $\mathcal{J}\subseteq I_i$. By the saturation lemma part 2, $(I_i,{<})\cong(\mathcal{J},{<})$.
It's easy to see that $\{I_i:i<\kappa^+\}$ is increasing continuous, which shows that $(\mathcal{J},{<})$ is robust of size $\kappa$.
This argument is due to Garrett Ervin in a private conversation.
From Shelah [4] Theorem 8.4.7, there exist saturated linear orders of size $\kappa$ if and only if $\kappa^{<\kappa}=\kappa$. For a simple argument when $\kappa=\aleph_\omega$, see [1] §4.3.