I must first preface that while this is indeed a question on an exercise, I believe this is advanced enough for MathOverflow.

Let $\kappa$ be a regular uncountable cardinal. Recall that the notion of a stationary subset makes sense for subsets of limit ordinals of uncountable cofinality. In Jech's Set Theory (Third Millenium Edition), he defined an ordering $<$ on stationary subsets $S,T \subseteq \kappa$ by (Definition 8.18, or (2.6) of the linked paper below): $$ S < T \iff S \cap \alpha \text{ is stationary for almost all } \alpha \in T $$ Note that we implicitly assume that almost all $\alpha \in T$ are limit ordinals of uncountable cofinality. The standard examples of such stationary sets are of the form: $$ E_\lambda^\kappa := \{\alpha < \kappa : \operatorname{cf}(\alpha) = \lambda\} $$ This was first introduced in his paper Stationary subsets of inaccessible cardinals. He proved in Lemma 8.19 (Theorem 2.4 of the paper) that $<$ is a well-founded relation, so it makes sense to define a rank function on this relation, $o(S)$, which he calls the order of the set.

He then proceeds to give two exercises on this matter:

  1. (Exercise 8.13) If $\lambda < \kappa$ is the $\alpha^\text{th}$ regular cardinal, then $o(E_\lambda^\kappa) = \alpha$.

  2. (Exercise 8.14) $o(\kappa) \geq \kappa$ if and only if $\kappa$ is weakly inaccessible; $o(\kappa) \geq \kappa + 1$ if and only if $\kappa$ is weakly Mahlo.

I have no idea how to solve either exercise. It may be helpful to note that in his paper, he defined the notion of a canonical stationary set (of order $\nu$), and mentioned without proof that the set $E_\lambda^\kappa$ is the canonical stationary set of order $\lambda$. However, I do not see why this is true.

  • $\begingroup$ Just a small remark, you don't need to "implicitly assume almost all ordinals in $T$ are limit ordinals", since it is true that almost all ordinals are limit ordinals. $\endgroup$
    – Asaf Karagila
    Apr 26, 2022 at 8:28
  • $\begingroup$ @AsafKaragila The implicit assumption includes "of uncountable cofinality", and that's definitely a constraint on $T$, since the ordinals of cofinality $\omega$ form a stationary set. $\endgroup$ Apr 26, 2022 at 14:03
  • $\begingroup$ In the first of the two exercises, it would be better not to use $\alpha$ for two different things. $\endgroup$ Apr 26, 2022 at 14:04

2 Answers 2


Here were my solutions from a few years ago, when I worked through Jech's 1984 paper on this topic.

Exercise 8.13: If $\lambda<\kappa$ is the $\alpha$th regular cardinal, then $o(E_{\lambda}^{\kappa})=\alpha$.

Proof. Now, $o(E_{\lambda}^{\kappa})\geq \alpha$ follows from the fact that $E_{\nu}^{\kappa}<E_{\mu}^{\kappa}$ when $\nu<\mu<\kappa$ are regular cardinals. (This later fact follows from $E_{\mu}^{\kappa}\subset {\rm Tr}(E_{\nu}^{\kappa})$, by the previous exercise.)

The other inequality is by induction, and we follow the proof given in Jech [1984: Stationary Subsets of Inaccessible Cardinals]. (We will show that any stationary set of order $\alpha$ must have an element of cofinality the $\alpha$th regular cardinal.) All of our stationary sets will be restricted to limit ordinals.

For each ordinal $\alpha<\kappa^{+}$, we say that a stationary set $E$ is canonical of order $\alpha$ if (1) $o(X)=\alpha$ for every stationary set $X\subset E$, and (2) $E$ meets every stationary set of order $\alpha$. Such a canonical set, if it exists, is clearly unique (modulo nonsingular sets). If it exists, we write $E_{\alpha}$ for such a set. (We will eventually show $E_{\alpha}=E_{\lambda}^{\kappa}$.)

Lemma 3.1 from Jech's paper: For a stationary set $S$, if $o(S)>\alpha$ and $E_{\alpha}$ exists, then $S\cap {\rm Tr}(E_{\alpha})\neq \emptyset$.

Proof of Lemma 3.1. The proof is by contradiction, letting $S$ be a minimal counterexample. Let $T<S$ be stationary of order $\nu$. Write $T=A\cup B\cup C$ where $A=T\cap E_{\alpha}$, $B=T\cap {\rm Tr}(E_{\alpha})$, and $C=T-(A\cup B)$. Since $A\subset E_{\alpha}$ we have ${\rm Tr}(A)\subset {\rm Tr}(E_{\alpha})$, so ${\rm Tr}(A)$ is disjoint from $S$. Next, $B\subset {\rm Tr}(E_{\alpha})$, hence ${\rm Tr}(B)\subset {\rm Tr}({\rm Tr}(E_{\alpha}))\subset {\rm Tr}(E_{\alpha})$ is also disjoint from $S$. From $T<S$ we have $S\subset {\rm Tr}(C)\mod I_{\rm NS}$. In particular, $C$ is stationary and $C<S$. Since taking the order reverses inclusion, we have $o(C)\geq o(T)=\alpha$. From heredity, we cannot have $o(C)=\alpha$, so the inequality is strict. But then we have a contradiction on the minimality of $S$. $\square_{\rm Lemma\ 3.1}$

Corollary 3.2 from Jech's paper: ${\rm Tr}(E_{\alpha})$ is the largest set (modulo nonstationary sets) of order $>\alpha$, if it is non-empty modulo $I_{\rm NS}$. In other words, for any stationary set $S$, if $o(S)>\alpha$ then $S>E_{\alpha}$.

Proof of Corollary 3.2. Assume not, so there is some stationary set $S$ with $o(S)>\alpha$ but $S$ is not contained in ${\rm Tr}(E_{\alpha})$ (modulo $I_{\rm NS}$). Let $T:=S-{\rm Tr}(E_{\alpha})$, which is a stationary set by the assumption. Further, $T$ is disjoint from ${\rm Tr}(E_{\alpha})$ and $o(T)\geq o(S)$ (from order reversal from $T\subset S$), a contradiction to the previous lemma. $\square_{\rm Corollary\ 3.2}$

Lemma 3.3 from Jech's paper: The set $E_{\alpha}$ exists if and only if there exists a largest stationary set $M_{\alpha}$ of order $\alpha$. Then \begin{equation}\tag{A}\label{Eq:3.3fromJech} M_{\alpha}=E_{\alpha}\cup {\rm Tr}(E_{\alpha}) \end{equation} and \begin{equation}\tag{B}\label{Eq:3.4fromJech} E_{\alpha}=M_{\alpha}-{\rm Tr}(M_{\alpha}). \end{equation} The union in (\ref{Eq:3.3fromJech}) is disjoint, and ${\rm Tr}(M_{\alpha})\subset M_{\alpha}$. (Both statements are taken modulo $I_{\rm NS}$.) Moreover, ${\rm Tr}(E_{\alpha})={\rm Tr}(M_{\alpha})$.

Proof of Lemma 3.3. $(\Rightarrow)$: Assume $E_{\alpha}$ exists. Note that $E_{\alpha}$ and ${\rm Tr}(E_{\alpha})$ are disjoint modulo $I_{\rm NS}$ since $o({\rm Tr}(E_{\alpha}))>\alpha$ and the heredity property on $E_{\alpha}$. Putting $M:=E_{\alpha}\cup {\rm Tr}(E_{\alpha})$, the order of $M$ is $\alpha$. [Prove by induction, using minimal counterexample, that $o(S\cup T)=\min\{o(S),o(T)\}$ for stationary sets. We call this the union property.] If $o(S)=\alpha$ write $S=A\cup B$ where $A=S\cap E_{\alpha}$ and $B=S-A$. If $B$ is nonstationary, then $S=A\subset M$ modulo $I_{\rm NS}$. If $B$ is stationary, by the second defining property on $E_{\alpha}$ and the union property we have $o(B)>\alpha$, and so $B\subset {\rm Tr}(E_{\alpha})\mod I_{\rm NS}$ by the previous corollary. Hence in this case we also have $S\subset M\mod I_{\rm NS}$; proving that $M$ is the largest set of order $\alpha$.

$(\Leftarrow)$: Assume $M_{\alpha}$ exists. By the union property [or by trivial considerations when ${\rm Tr}(M_{\alpha})$ is nonstationary] it follows that $M_{\alpha}\cup {\rm Tr}(M_{\alpha})$ has order $\alpha$, so ${\rm Tr}(M_{\alpha})\subset M_{\alpha}$ (modulo $I_{\rm NS}$) by maximality. Let $E=M_{\alpha}-{\rm Tr}(M_{\alpha})$. The order of $E$ is $\alpha$ since $M_{\alpha}=E\cup {\rm Tr}(M_{\alpha})$, by the union property, etc.

We first show that $E$ has order $\alpha$ hereditarily. Let $S\subset E$ be stationary. If $o(S)>\alpha$ then there is a stationary set $T<S$ of order $\alpha$. By maximality of $M_{\alpha}$ we have $T\subset M_{\alpha}\mod I_{\rm NS}$, so $S\subset {\rm Tr}(T)\subset {\rm Tr}(M)\mod I_{\rm NS}$ which contradicts the fact that $S$ does not intersect ${\rm Tr}(M_{\alpha})$.

Finally we show that $E$ meets every stationary set of order $\alpha$. Suppose, by way of contradiction, that $S$ is stationary of order $\alpha$ and $S\cap E=\emptyset$. We have $S\subset M_{\alpha}\mod I_{\rm NS}$, and since $S\cap E=\emptyset$ we have $S\subset {\rm Tr}(M_{\alpha}\mod I_{\rm NS}$, hence $M<S$, contradicting the order. This proves that $E$ is the canonical set of order $\alpha$.

Finally (working modulo $I_{\rm NS}$ everywhere), since $E\subset M$ we have ${\rm Tr}(E)\subset {\rm Tr}(M)$, while the reverse containment follows from the previous corollary.$\square_{\rm Lemma\ 3.3}$

Corollary 3.4 from Jech's paper: Assuming $M_{\alpha}$ exists, for any stationary set $S$, then $o(S)\geq \alpha$ if and only if $S\subset M_{\alpha}\mod I_{\rm NS}$.

Proof of Corollary 3.4. This is just a direct application of the definition of $M_{\alpha}$ and the union property (taking the union with $M_{\alpha}$). [I don't see how it is a corollary.] $\square_{\rm Corollary\ 3.4}$

Lemma 3.5 from Jech's paper: [Note, it is misnumbered there.] Let $\alpha$ be a limit ordinal and assume that $M_{\beta}$ exists for all $\beta<\alpha$. Then $M_{\alpha}$ exists if and only if ${\rm inf}_{\beta<\alpha}M_{\beta}$ exists, and is nonzero, in the Boolean algebra $P(\kappa)/I_{\rm NS}$.

Proof of Lemma 3.5. $(\Rightarrow)$: Assume $M_{\alpha}$ exists. We have $M_{\beta}\supset M_{\alpha}\mod I_{\rm NS}$ when $\beta<\alpha$ by the previous corollary, so $M_{\alpha}$ is a lower bound on the $M_{\beta}$. If $S$ is any stationary set which is a lower bound on each $M_{\beta}$ modulo $I_{\rm NS}$, then again by the previous corollary $o(S)\geq \beta$ for each $\beta<\alpha$, hence $o(S)\geq \alpha$, so $S\subset M_{\alpha}\mod I_{\rm NS}$ as desired.

$(\Leftarrow)$: If the inf exists and is nonzero, call it $M$. As it is contained in each $M_{\beta}$ (for $\beta<\alpha$) it has order at least $\alpha$. Given any stationary subset $S$ with $o(S)=\alpha$ then by the previous corollary $S\subset M_{\beta}\mod I_{\rm NS}$ for each $\beta<\alpha$, hence it lies inside the infimum. In particular, the infimum has exactly order $\alpha$ and is the largest set containing all such sets.$\square_{\rm Lemma\ 3.5}$

Lemma 3.6 from Jech's paper: If $M_{\alpha}$ exists and the height of $\kappa$ is at least $\alpha+2$ (i.e.\ there is a set of order $\alpha+1$), then $M_{\alpha+1}={\rm Tr}(M_{\alpha})$.

Proof of Lemma 3.6. We know ${\rm Tr}(S)\geq o(S)+1$ (when ${\rm Tr}(S)$ is stationary). Thus $o({\rm Tr}(M_{\alpha}))\geq \alpha+1$. By Corollary 3.4, ${\rm Tr}(M_{\alpha})\subset M_{\alpha+1}\mod I_{\rm NS}$. Conversely, $o(M_{\alpha+1})>\alpha$ so by Corollary 3.2 $M_{\alpha+1}\subset {\rm Tr}(E_{\alpha})={\rm Tr}(M_{\alpha})\mod I_{\rm NS}$. $\square_{\rm Lemma\ 3.6}$

As diagonal intersection gives inf for sets of cardinality $<\kappa^{+}$, the previous two lemmas complete an induction, so we see that we can find the maximal sets $M$ inductively for $\alpha<\kappa^{+}$ (when the height of $\kappa$ is bigger than $\alpha$) by: $M_{0}$ is the limit points below $\kappa$, $M_{\alpha+1}={\rm Tr}(M_{\alpha})$, and for limits $\alpha$ we have $M_{\alpha}=\triangle_{\beta<\alpha}M_{\beta}$.

In the first paragraph we showed that the height of $\kappa$ is at least the ordinality of the sequence of regular infinite cardinals $<\kappa$. It is easy to check (via induction) using these definitions that $E_{\alpha}=E_{\lambda}^{\kappa}$ as claimed.$\square_{\rm Exercise\ 8.13}$

Exercise 8.14: The height of $\kappa$ is at least $\kappa$ if and only if $\kappa$ is weakly inaccessibly; the height of $\kappa$ is at least $\kappa+1$ if and only if $\kappa$ is weakly Mahlo.

Proof. First suppose $\kappa$ is an uncountable, regular, successor cardinal. Say $\kappa$ is the $\alpha$th regular cardinal. We have $\alpha<\kappa$. If $\alpha$ is a successor ordinal, say $\alpha=\beta+1$, let $\lambda$ be the $\beta$th regular cardinal. Then $E_{\beta}=E_{\lambda}^{\kappa}$ and ${\rm Tr}(E_{\beta})=\emptyset$ by Exercise 8.12, so the height of $\kappa$ is $\alpha$ by Lemma 3.6 in the solution to the previous problem. If $\alpha$ is a limit ordinal, then since $\triangle_{\beta<\alpha}E_{\alpha}=\triangle_{\lambda<\kappa:\lambda\text{ regular}}E_{\lambda}^{\kappa}=\emptyset$, by Lemma 3.5 in the solution of the previous problem we have the height of $\kappa$ is $\alpha$.

Second suppose $\kappa$ is an uncountable, regular, limit cardinal. This means the sequence of limit cardinals $<\kappa$ is a cofinal sequence, so the height of $\kappa$ is at least $\kappa$, again by the previous problem.

Putting these together, we see $h(\kappa)\geq \kappa$ iff $\kappa$ is weakly inaccessible.

Now, fix $\kappa$ to be weakly inaccessible. The diagonal intersection of the sequence $E_{\lambda}^{\kappa}$ consists of regular cardinals $<\kappa$, so this is stationary if and only if $\kappa$ is weakly Mahlo. Hence by Lemma 3.5 of the previous solution, $h(\kappa)\geq \kappa+1$.

Conversely, if $h(\kappa)\geq \kappa+1$, the diagonal intersection of $E_{\lambda}^{\kappa}$ is stationary, hence there is a stationary sequence of regular cardinals $<\kappa$, so $\kappa$ is weakly Mahlo.$\square_{\rm Exercise\ 8.14}$

  • $\begingroup$ Pace, thank you very much for taking the time to write out the detailed answer. However, I've spent the past few days digesting the proof in Monk's notes (finished shortly after I saw you posted the answer), so I'll accept my own answer instead. $\endgroup$ May 3, 2022 at 6:42

After some googling I found the notes by J. D. Monk, which have answered the questions in the span of Theorem 2.68 to 2.91.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.