Math 557 Sep 26

Midterm 1 Review

Take-home Problem 1


Prove unique readability for the set of \(\mathcal{L}\)-formulas.

Solution 1. First prove readibility, i.e. the statement that for any \(\mathcal{L}\)-formula \(\varphi\), exactly one of the following cases holds:

  1. There exist terms \(s,t\) such that \(\varphi \equiv s=t\).
  2. There exists a relation symbol \(R\) and, if \(n\) is the arity of \(R\), terms \(t_1, \dots, t_n\) sucht that \(\varphi \equiv Rt_1\dots t_n\).
  3. There exists a formula \(\psi\) such that \(\varphi \equiv \neg \psi\).
  4. There exist formulas \(\psi, \theta\) such that \(\varphi \equiv (\psi \land \theta)\).
  5. There exists a variable \(x\) and a formula \(\psi\) such that \(\varphi \equiv \exists x \psi\).

To see this, let \(F\) be the subset of \(\mathcal{L}^*\) such that every \(\theta \in F\) satisfies exactly one of (1)-(5). Then \(F\) contains all atomic formulas and is closed under \(\neg, \land, \exists x\). Since the set of \(\mathcal{L}\)-formulas is the smallest such set, it follows that every \(\mathcal{L}\)-formula is contained in \(F\), and therefore has the desired property.

Second we argue that a proper initial segment of a formula cannot be a formula. We proceed by induction on the length \(l\) of \(\varphi\). For \(l=1\) there is no formula of that length. Now suppose \(\varphi\) is a formula of length \(l+1\), and \(\beta\) is a proper initial segment. We may assume \(\beta\) is not empty. We consider theses (1)-(5) from readability.

Case \(\varphi \equiv s=t\)
Either \(\beta \equiv s = t'\) with \(t \subset t'\), but this would contradict that no proper initial segment of a term is a term. Or \(\beta \equiv s=\): An easy induction over the length of a formula shows that no formula ends with \(=\). Or \(\beta \subseteq s\). Another easy induction shows that no formula can be a term or an initial segment of a term.
Case \(\varphi \equiv Rt_1\dots t_n\)
If \(\beta\) were a formula, it has to be of the form \(Rs_1\dots s_n\). Comparing terms, we would see that one of the \(s_i\) is a proper initial segment of some \(t_j\), which is impossible.
Case \(\varphi \equiv \neg \theta\)
Then \(\beta \equiv \neg \beta'\) with \(\beta'\) a proper initial segment of \(\theta\). By inductive hypothesis, \(\beta'\) is not a formula, hence \(\neg \beta'\) is not a formula either.
Case \(\varphi \equiv (\psi \land \theta)\)
Then \(\beta\) starts with \((\) but does not end with \()\). Another induction shows that for any formula, the number of \((\)’s always equals the number of \()\)’s.
Case \(\varphi \equiv \exists x \varphi\)
Neither ‘\(\exists\)’ nor ‘\(\exists x\)’ are formulas. If \(\beta\) is longer than that, it is of the form \(\exists x \beta'\). By inductive hypothesis, \(\beta'\) is not a formula, and hence \(\beta\) is not a formula (by readability). varphi

Finally, we prove unique readability: The choices in (1)-(5) are unique.

Case (1)
Suppose \(\varphi \equiv s=t \equiv s'=t'\). Since no proper initial segment of a term is a term, it must hold that \(s\equiv s'\) and \(t \equiv t'\).
Case (2)
Suppose \(\varphi \equiv Rt_1\dots t_n \equiv S s_1\dots s_k\). Then \(R \equiv S\) and \(n=k\). Comparing terms inductively, using again the fact that no proper initial segment of a term is a term, we get \(s_i \equiv t_i\) for all \(i\).
Case (3)
Suppose \(\varphi \equiv \neg \psi \equiv \neg \theta\). Immediately, we infer \(\psi \equiv \theta\).
Case (4)
Suppose \(\varphi \equiv (\psi_1 \land \theta_1) \equiv (\psi_2\land \theta_2)\). Assume \(\psi_1 \not\equiv \psi_2\). Then \(\psi_1\) is a proper initial segment of \(\psi_2\) or vice versa (since both of them are part of \(\varphi\)). Either case is impossible due to the fact that no proper initial segment of a formula is a formula. Hence \(\psi_1 \equiv \psi_2\) and thus also \(\theta_1 \equiv \theta_2\).
Case (5)
Suppose \(\varphi \equiv \exists x \psi \equiv \exists y \theta\). By comparing entries, we get \(x \equiv y\) and thus $, as desired.
Take-home Problem 2


Let \(\mathcal{L}\) be any finite language and let \(\mathcal{M}\) be a finite \(\mathcal{L}\)-structure. Show that there is an \(\mathcal{L}\)-sentence \(\varphi\) such that \[ \mathcal{N} \models \varphi \; \iff \; \mathcal{N} \cong \mathcal{M}. \]

Solution 2. Suppose \(\mathcal{L} = \{c_1, \cdots, c_k, f_1^{(a_1)}, \dots f_l^{(a_l)}, R_1^{(b_1)}, \dots, R_j^{(b_j)} \}\) where the \((a_i), (b_m)\) denote the arities of the respective symbols.

Since \(\mathcal{M}\) is finite, we may assume \(M = \{1, \dots, n\}\) for some \(n \in \mathbb{N}\).

The basic idea is to collect all “elementary facts” about \(\mathcal{M}\) in a single formula (think of a group multiplication table, just for all functions and relations).

Define

\[\begin{aligned} \varphi \equiv \; \exists x_1, \dots, x_n \biggl ( & \bigwedge_{i \neq m} x_i \neq x_l \; \land \; \forall y \bigvee_{i \leq n} y = x_i \\ & \land \bigwedge_{i \leq k} c_i = x_{c^{\mathcal{M}}_i} \\ & \land \bigwedge_{i\leq l} \bigwedge_{\pi \in \{1,\dots, n\}^{a_i}} f_i x_{\pi(1)} \dots x_{\pi(a_i)} = x_{f^{\mathcal{M}}_i(\pi(1), \dots, \pi(a_i))} \\ & \land \bigwedge_{i\leq j} \bigwedge_{\pi \in \{1,\dots, n\}^{b_i}} \delta_\pi R_i x_{\pi(1)} \dots x_{\pi(b_i)} \biggr ) \end{aligned}\]

where \(\delta_\pi\) is empty if \(R(\pi(1), \dots, \pi(b_i))\) holds in \(\mathcal{M}\), and \(\neg\) if not.

Assme \(\mathcal{N} \models \varphi\). Due to the first line of the equation, \(N\) has exactly \(n\) elements, and let \(r_i \in N\) be the witness to \(\exists x_i\). Define a mapping \(\tau: M \to N\) by letting \(\tau(i) = r_i\). We claim that \(\tau\) is an isomorphism.

By definition we have \(\tau(c_i^{\mathcal{M}}) = r_{c_i^{\mathcal{M}}}\). Moreover, by the second line of the formula, \(r_{c^{\mathcal{M}}_i}\) is the unique element of \(N\) that makes the formula \(c_i = x_{c_i^{\mathcal{M}}}\) true. It follows that \(\tau(c_i^{\mathcal{M}}) = c_i^{\mathcal{N}}\) for all \(1 \leq i \leq k\).

Using a similar argument with the third line of the formula, we obtain \[ \tau(f_i^{\mathcal{M}}(s_1, \dots, s_{a_i})) = r_{f_i^{\mathcal{M}}(s_1, \dots, s_{a_i})} = f^{\mathcal{N}}(r_{s_1}, \dots, r_{s_{a_i}}) = f^{\mathcal{N}}(\tau(s_1), \dots, \tau(s_{a_i}) \] The argument for \(R^{\mathcal{M}}(s_1, \dots, s_{b_i}) \iff R^{`\mathcal{N}}(\tau(s_1), \dots, \tau(s_{b_i}))\) is similar.

On the other hand, if \(\mathcal{M} \cong \mathcal{N}\) via \(\tau\), then \(\mathcal{N} \models \psi[\tau(1), \dots, \tau(n)]\), where \(\psi\) is such that \(\varphi \equiv \exists x_1, \dots, x_n \: \psi\), due to the fact that \(\tau\) is an isomorphism, and thus \(\mathcal{N} \models \varphi\).

Take-home Problem 3


Give an example of a language \(\mathcal{L}\) and an \(\mathcal{L}\)-sentence \(\psi\) such that

  • there is at least one \(\mathcal{L}\)-structure \(\cal A\) such that \(\cal A \models \psi\),
  • for all \(\cal L\)-structures \(\cal A\), if \(\cal A \models \psi\), then the universe \(A\) of \(\cal A\) is infinite.

Solution 3. Let \(\mathcal{L} = \{<\}\), where \(<\) is a binary relation symbol. Define

\[\begin{aligned} \psi \equiv \; & \forall x \; x \nless x \\ & \land \forall x,y \; x < y \lor x=y \lor y < x \\ & \land \forall x,y,z \; (x < y \land y < z) \to x < z \\ & \land \forall x \exists y \; x < y \end{aligned}\]

The formula says that \(<\) is a linear order with no maximal element.

Clearly, \((\mathbb{Z}, <) \models \psi\).

Now suppose \(\mathcal{M} \models \psi\). Due to the last line of \(\psi\), there exists a function \(f: M \to M\) such that \(x < f(x)\) for all \(x \in M\). We claim that for any \(x\) and for any \(n \neq m\), \[ f^{(n)}(x) \neq f^{(m)}(x) \] This follows from antireflexivity (line one) and transitivity (line three).

Therefore, the set \[ \{x, f^{(1)}(x), f^{(2)}(x), \dots \} \] is an infinite subset of \(N\).

Take-home problem 4


Show that

\[\begin{aligned} \{\varphi \to \psi \} & \vdash \exists x \varphi \: \to \: \exists x \psi \\ \{\varphi \to \psi \} & \vdash \forall x \varphi \: \to \: \forall x \psi \\ \end{aligned}\]

Solution 4. We give derivations below (with brief justifications). We collect simple substeps into a single one.

For \(\{\varphi \to \psi \} \vdash \exists x \varphi \: \to \: \exists x \psi\;\):

Formula Justification
\(\varphi \to \psi\) given
\(\psi \to \exists x \psi\) (Q2)
\(\varphi \to \exists x \psi\) tautology
\(\neg \exists x \psi \to \neg \varphi\) tautotolgy
\(\forall x ( \neg \exists x \psi \to \neg \varphi)\) \(\forall\)-intro
\(\neg \exists x \psi \to \forall x \varphi\) (Q1), \(x\) not free in \(\neg \exists x \psi\)
\(\neg \forall x \neg \varphi \to \exists x \psi\) tautology
\(\exists x \varphi \to \exists x \psi\) (Q3)


For \(\{\varphi \to \psi \} \vdash \forall x \varphi \: \to \: \forall x \psi \;\):

Formula Justification
\(\varphi \to \psi\) given
\(\forall x \varphi \to \varphi\) Example 2.6.1 (d)
\(\forall x \varphi \to \psi\) Tautology
\(\forall x (\forall x \varphi \to \psi)\) \(\forall\)-intro
\(\forall x \varphi \to \forall x \psi\) (Q1), \(x\) not free in \(\forall x \varphi\)
Take-home Problem 6


Use the compactness theorem to show (without using the Axiom of Choice) that every set can be linearly ordered.

Try to strengthen this to:

Every partial order can be extended to a linear order.

Solution 5. We first argue that any finite partial order \((F,<)\) can be extended to a linear order.

We proceed by induction on the cardinality of \(F\).

For \(|F|=0\) there is nothing to prove.

Now assume \(|F|=n+1\). Pick an arbitrary \(a \in F\) and consider \(F \setminus \{a\}\). By inductive hypothesis, \(F \setminus \{a\}\) can be extended to a linear order \(a_1 < a_2 < \dots < a_n\).

If there does not exist \(i\) such that \(a_i <_F a\), we can extend to a linear order on \(F\) by putting \(a < a_i\) for all \(i\). Otherwise let \(k\) be maximal such that \(a_k < a\). Then \[ a_1 < \dots < a_k < a < a_{k+1} < \dots < a_n \] defines a linear order that extends \((F,<)\).

Now let \((P,<)\) be an arbitrary partial order. We extend the language \(\mathcal{L}_<\) of orders to \(\mathcal{L}_P\), where we add a new constant symcol \(c_p\) for every \(p \in P\).

Let \(T\) be the theory of linear orders (see first three lines of sentence \(\varphi\) in Problem 3) together with \[ \{ c_p < c_q : p <_P q \} \] In other words, if \(p\) is less than \(q\) according to \(P\), we add a corresponding axiom to our theory.

Any model of \(T\) is a liner order that extends \(P\).

Moreover, any finite subset \(T_0 \subseteq T\) induces a partial order on a finite subset of \(P\) (induced by those \(p\) for which \(c_p\) is part of a formula in \(T_0\)). By our argument in the first part, this finite partial order extends to a linear order, thereby giving a model of \(T_0\).

By compactness, \(T\) has a model \(\mathcal{M} = (M,<)\). By choice of \(T\), \((M,<)\) is a linear order. Furthermore, the mapping \[ c_p \mapsto c_p^{\mathcal{M}} \] is one-to-one since \(\mathcal{M} \models \forall x \ x \nless x\).

We can “pull back” the order on \(\mathcal{M}\) to \(P\) by letting \[ p <' q \iff c_p^{\mathcal{M}} <^{\mathcal{M}} c_q^{\mathcal{M}} \] By definition of \(T\), \(<'\) is a linear order that extends \(<_P\).