Cofinality, Cardinal Exponentiation, and the Singular Cardinals Hypothesis

Intuition Beginner

Some infinite sizes can be reached "from below" by a short climb, and some cannot. Picture a tall tower whose height is one of the infinite sizes from the previous unit. Now ask: how few steps do you need, each step landing on a shorter sub-height, so that the heights of your steps creep all the way up to the top with nothing left uncovered? That smallest number of steps is the tower's cofinality — its true "reaching length."

For the very first infinite size, the one matching the counting numbers, you genuinely need an endless run of steps: no finite stack of shorter heights ever catches up. So its reaching length is that same first infinite size. A size whose reaching length equals itself is called regular — it is sturdy, not sneakily approachable by a shorter climb.

But there is a stranger kind of size. Imagine stacking the sizes one infinite size, then the next, then the next, forever, and asking for the size that sits just above this whole endless list. You reach that ceiling with only a counting-number's worth of steps, even though the ceiling itself is a far larger infinity. A size you can sneak up on with a shorter climb than the size itself is called singular. The gap between "how big you are" and "how short a climb reaches you" is the secret that makes the arithmetic of these sizes hard, and it is what this unit is about.

Visual Beginner

A size is regular when the shortest climb that reaches it is as long as the size itself, and singular when a shorter climb already gets there. The picture below contrasts the two.

   REGULAR size                      SINGULAR size
   (first infinity)                  (the size just above
                                      first, second, third, ... infinities)

      top ===                           top ===
       |                               /  /  /   <- only a counting-number
       |   needs a climb              /  /  /        of steps reaches the top
       |   as long as the           /  /  /
       |   size itself             1st  2nd  3rd  ... infinities as the steps
      ...                          (a short ladder reaches a tall tower)

The next table records the reaching length (cofinality) of a few sizes. Write ${\aleph }_0$ for the first infinite size, ${\aleph }_1$ for the next, and ${\aleph }_{\omega }$ for the size sitting just above the whole list ${\aleph }_0,{\aleph }_1,{\aleph }_2,\dots$.

size	shortest climb that reaches it	regular or singular
${\aleph }_0$	${\aleph }_0$ (endless run of finite steps)	regular
${\aleph }_1$	${\aleph }_1$	regular
${\aleph }_{17}$	${\aleph }_{17}$	regular
${\aleph }_{\omega }$	${\aleph }_0$ (the list of earlier infinities)	singular

The key idea: a size is singular exactly when it is the top of a shorter list of smaller sizes. The first place this happens among the infinite sizes is ${\aleph }_{\omega }$, reached by a mere counting-number of steps.

Worked example Beginner

Let us see by hand why ${\aleph }_{\omega }$ is reached by an endless-but-counting-number-long climb, that is, why its reaching length is ${\aleph }_0$.

Step 1. List the infinite sizes in order: the first is ${\aleph }_0$, the second ${\aleph }_1$, the third ${\aleph }_2$, and so on, one for each counting number.

Step 2. The size ${\aleph }_{\omega }$ is defined to be the smallest size bigger than every entry on that list. So the list ${\aleph }_0,{\aleph }_1,{\aleph }_2,\dots$ sits entirely below ${\aleph }_{\omega }$, and nothing is squeezed strictly between the whole list and ${\aleph }_{\omega }$.

Step 3. Use the list itself as your climb: step onto ${\aleph }_0$, then ${\aleph }_1$, then ${\aleph }_2$, taking one step per counting number. The heights of your steps grow without bound below ${\aleph }_{\omega }$ and leave no room underneath, so they "reach" ${\aleph }_{\omega }$.

Step 4. Count the steps: there is exactly one step per counting number, so the climb has length ${\aleph }_0$, the first infinite size.

Step 5. Could a shorter climb work? A finite stack of sizes below ${\aleph }_{\omega }$ has a single largest member, say ${\aleph }_5$, and stops there — it never passes ${\aleph }_6$, so it cannot reach the top. So no finite climb works, and the counting-number climb is the shortest.

What this tells us: ${\aleph }_{\omega }$ has reaching length ${\aleph }_0$, far smaller than ${\aleph }_{\omega }$ itself. That mismatch is exactly what "singular" means, and ${\aleph }_{\omega }$ is the first singular infinite size.

Check your understanding Beginner

Formal definition Intermediate+

All of the following takes place inside ZFC; ordinals and transfinite recursion are as in 42.03.02, and cardinals, the aleph function, the absorption laws $\kappa \oplus \lambda =\kappa \otimes \lambda =\max (\kappa ,\lambda )$, cardinal exponentiation ${\kappa }^{\lambda }$, and König's inequality are as in 42.03.03. The Axiom of Choice is assumed throughout.

Let $\alpha$ be a limit ordinal. A set $X\subseteq \alpha$ is cofinal (or unbounded) in $\alpha$ if $\sup X=\alpha$, that is, for every $\beta <\alpha$ there is $\gamma \in X$ with $\beta <\gamma$. The cofinality of $\alpha$, written $\mathrm{cf}(\alpha )$, is the least order type of a cofinal subset of $\alpha$ ^{[Kunen Ch. I §13]}. Equivalently, $\mathrm{cf}(\alpha )$ is the least cardinal $\mu$ such that $\alpha$ is the supremum of a $\mu$-indexed increasing sequence $⟨{\alpha }_{\xi }:\xi <\mu ⟩$ with each ${\alpha }_{\xi }<\alpha$. One sets $\mathrm{cf}(0)=0$ and $\mathrm{cf}(\beta +1)=1$ for successors; cofinality is interesting only at limits.

An infinite cardinal $\kappa$ is regular if $\mathrm{cf}(\kappa )=\kappa$ and singular if $\mathrm{cf}(\kappa )<\kappa$. Thus ${\aleph }_0$ is regular (a cofinal subset of $\omega$ must be infinite), every successor cardinal ${\aleph }_{\alpha +1}$ is regular (proved below, using Choice), and ${\aleph }_{\omega }$ is singular with $\mathrm{cf}({\aleph }_{\omega })=\omega ={\aleph }_0$. More generally, for a limit ordinal $\lambda$ one has $\mathrm{cf}({\aleph }_{\lambda })=\mathrm{cf}(\lambda )$, so ${\aleph }_{\lambda }$ is singular precisely when $\mathrm{cf}(\lambda )<\lambda$.

The gimel function is $\gimel (\kappa )={\kappa }^{\mathrm{cf}(\kappa )}$, defined on infinite cardinals. König's theorem (next section) gives $\kappa <\gimel (\kappa )$, so $\gimel (\kappa )\ge {\kappa }^{+}$ always; gimel is the single function to which all cardinal exponentiation reduces. A cardinal $\kappa$ is a strong limit if $2^{\mu }<\kappa$ for every $\mu <\kappa$. An uncountable cardinal that is both regular and a limit cardinal is (weakly) inaccessible, and a regular strong-limit uncountable cardinal is (strongly) inaccessible; these are the regular cardinals not reachable by the ordinary cardinal operations from below, and their existence is not provable in ZFC (cross-ref 42.03.09).

Counterexamples to common slips Intermediate+

• "Every limit cardinal is singular." The first uncountable limit cardinal ${\aleph }_{\omega }$ is singular, but an inaccessible cardinal is a regular limit cardinal. Limit-cardinal-hood concerns the aleph index being a limit ordinal; regularity concerns cofinality, a separate question.

• "$\mathrm{cf}(\kappa )$ can be singular." No: $\mathrm{cf}(\mathrm{cf}(\alpha ))=\mathrm{cf}(\alpha )$, so $\mathrm{cf}(\kappa )$ is always a regular cardinal. A shortest cofinal climb cannot itself be shortened.

• "$2^{{\aleph }_0}$ could equal ${\aleph }_{\omega }$." König forbids it: $\mathrm{cf}(2^{\kappa })>\kappa$, and $\mathrm{cf}({\aleph }_{\omega })={\aleph }_0$, so $2^{{\aleph }_0}\ne {\aleph }_{\omega }$. The continuum can be ${\aleph }_1,{\aleph }_2,{\aleph }_{\omega +1},\dots$ but never any cardinal of countable cofinality.

• "Easton's theorem decides the value of $2^{{\aleph }_{\omega }}$." Easton's freedom applies only to regular cardinals. The singular case is governed by SCH and pcf theory, where ZFC theorems (Silver, Shelah) constrain the value sharply — the singular continuum function is not free.

Key theorem with proof Intermediate+

The organising fact is König's theorem: it converts the equality "the absorption laws collapse $\oplus$ and $\otimes$" into the one genuine inequality that survives, and that inequality is precisely a statement about cofinality.

Theorem (König's theorem and the cofinality bound). For families of cardinals with ${\kappa }_i<{\lambda }_i$ for all $i\in I$, $$ \sum_{i \in I} \kappa_i < \prod_{i \in I} \lambda_i. $$ Consequently $\kappa <{\kappa }^{\mathrm{cf}(\kappa )}$ for every infinite cardinal $\kappa$, and therefore $\mathrm{cf}(2^{\kappa })>\kappa$ ^{[Kunen Ch. I §13]}.

Proof. The inequality $\le$ is routine, so the content is strictness; it suffices to show no $G:{⨆}_i{\kappa }_i\to {\prod }_i{\lambda }_i$ is surjective. For each $i$, the set of $i$-th coordinates $\{G(x)(i):x\text{ in the }i\text{-th block}\}$ has size $\le {\kappa }_i<{\lambda }_i$, so there is some $f(i)\in {\lambda }_i$ omitted by every $G(x)$ with $x$ in the $i$-th block (the choice of $f(i)$ across $i\in I$ uses the Axiom of Choice). The function $f=⟨f(i){⟩}_i\in {\prod }_i{\lambda }_i$ then satisfies $G(x)(i)\ne f(i)$ whenever $x$ lies in the $i$-th block, so $f\notin \mathrm{ran}(G)$ and $G$ is not surjective.

For the corollary, let $\mu =\mathrm{cf}(\kappa )$ and write $\kappa ={\sup }_{i<\mu }{\kappa }_i$ as an increasing union with each ${\kappa }_i<\kappa$. Then $\kappa ={\sum }_{i<\mu }{\kappa }_i$ (the sum is $\ge \kappa$ by the supremum and $\le \mu \otimes \kappa =\kappa$ by absorption). Apply the inequality with ${\lambda }_i=\kappa$ for all $i<\mu$: since ${\kappa }_i<\kappa ={\lambda }_i$, $$ \kappa = \sum_{i < \mu} \kappa_i < \prod_{i < \mu} \kappa = \kappa^\mu = \kappa^{\operatorname{cf}(\kappa)}. $$ Finally set $\theta =2^{\kappa }$ and suppose $\mathrm{cf}(\theta )\le \kappa$. Writing $\theta ={\sum }_{i<\mathrm{cf}(\theta )}{\theta }_i$ with ${\theta }_i<\theta$ and applying the inequality with ${\lambda }_i=\theta$ gives $\theta <{\theta }^{\mathrm{cf}(\theta )}\le {\theta }^{\kappa }=(2^{\kappa }{)}^{\kappa }=2^{\kappa \otimes \kappa }=2^{\kappa }=\theta$, a contradiction. Hence $\mathrm{cf}(2^{\kappa })>\kappa$. $\square$

Bridge. König's theorem is the foundational reason cardinal exponentiation does not collapse the way addition and multiplication do: the absorption laws of 42.03.03 kill $\oplus$ and $\otimes$, and this is exactly the residue that survives — a strict inequality reading off the cofinality of the exponential. It builds toward the gimel calculus below, where $\gimel (\kappa )={\kappa }^{\mathrm{cf}(\kappa )}$ is the very quantity König bounds from below, and the central insight is that every hard question about ${\kappa }^{\lambda }$ is a question about cofinality: the case split that reduces exponentiation to gimel turns on whether the exponent $\lambda$ lies below or above $\mathrm{cf}(\kappa )$. The cofinality bound $\mathrm{cf}(2^{\kappa })>\kappa$ generalises Cantor's $2^{\kappa }>\kappa$ from 42.03.03 — Cantor's diagonal run once, König's diagonal run along an entire cofinal family — and is dual to the absorption collapse: putting these together, what addition forgets, exponentiation remembers exactly as a cofinality constraint. This appears again in Easton's theorem below, which shows monotonicity and this König bound are the only constraints on the regular continuum function, and in the Singular Cardinals Hypothesis, where the cofinality of a singular $\kappa$ is the hinge on which $\gimel (\kappa )$ turns.

Exercises Intermediate+

Advanced results Master

Exponentiation, freed of the additive collapse, reduces by cases to two data: the continuum function $\kappa \mapsto 2^{\kappa }$ on regular cardinals and the gimel function $\gimel (\kappa )={\kappa }^{\mathrm{cf}(\kappa )}$ at singulars. The first is free; the second is constrained by ZFC theorems whose discovery reshaped the subject.

Theorem 1 (the gimel reduction). For infinite cardinals $\kappa ,\lambda$, the value ${\kappa }^{\lambda }$ is computed recursively from the continuum function on regulars and the gimel function at singulars ^{[Jech Ch. 5]}. The Hausdorff formula $({\aleph }_{\alpha +1}{)}^{{\aleph }_{\beta }}={\aleph }_{\alpha +1}\cdot ({\aleph }_{\alpha }{)}^{{\aleph }_{\beta }}$ peels successors, and the Bukovský-Hechler analysis treats the remaining cases by comparing $\lambda$ with $\mathrm{cf}(\kappa )$: for $\lambda <\mathrm{cf}(\kappa )$ functions $\lambda \to \kappa$ are bounded so ${\kappa }^{\lambda }={\sum }_{\mu <\kappa }{\mu }^{\lambda }$; for $\mathrm{cf}(\kappa )\le \lambda <\kappa$ one gets ${\kappa }^{\lambda }=\gimel (\kappa )$ when the lower exponentials stabilise, and otherwise ${\kappa }^{\lambda }=({\sup }_{\mu <\kappa }{\mu }^{\lambda }{)}^{\mathrm{cf}(\kappa )}$; for $\kappa \le \lambda$, ${\kappa }^{\lambda }=2^{\lambda }$. Every exponential is thus a finite expression in $2^{(\cdot )}$ on regulars and $\gimel$ at singulars.

Theorem 2 (Easton's theorem; the regular continuum function is free). Let $E$ be any class function from regular cardinals to cardinals satisfying (i) monotonicity $\kappa \le \lambda \Rightarrow E(\kappa )\le E(\lambda )$ and (ii) the König bound $\mathrm{cf}(E(\kappa ))>\kappa$. Then there is a class forcing extension of any ground model of GCH in which $2^{\kappa }=E(\kappa )$ for every regular $\kappa$ ^{[Kunen Ch. I §13]}. Hence on regular cardinals, monotonicity and König are the complete list of ZFC constraints on $2^{(\cdot )}$; the full forcing construction is developed in 42.03.08. Easton's theorem says nothing about singular cardinals, and the contrast is the entire point of the singular-cardinals problem.

Theorem 3 (the Singular Cardinals Hypothesis). SCH is the assertion that for every singular cardinal $\kappa$, if $2^{\mathrm{cf}(\kappa )}<\kappa$ then ${\kappa }^{\mathrm{cf}(\kappa )}={\kappa }^{+}$, equivalently $\gimel (\kappa )={\kappa }^{+}$ whenever $\kappa$ is singular above the continuum ^{[Jech Ch. 5]}. SCH follows from GCH and, more sharply, holds above any cardinal where the continuum function below has stabilised; under SCH all of cardinal exponentiation is computed from the continuum function on regulars alone. The negation of SCH — a singular strong-limit $\kappa$ with $2^{\kappa }>{\kappa }^{+}$ — is consistent only relative to large cardinals: Magidor first forced $\neg$SCH at ${\aleph }_{\omega }$ from a supercompact cardinal, and Gitik calibrated the exact consistency strength to a large cardinal in the region of a measurable of high Mitchell order, so the failure of SCH requires large cardinals (cross-ref 42.03.09).

Theorem 4 (Silver's theorem). SCH cannot first fail at a singular cardinal of uncountable cofinality ^{[Jech Ch. 8]}. Precisely, if $\kappa$ is singular with $\mathrm{cf}(\kappa )>\omega$ and $2^{\mu }={\mu }^{+}$ for stationarily many $\mu <\kappa$, then $2^{\kappa }={\kappa }^{+}$; and the same for $\gimel$. The proof runs a generic-ultrafilter / elementary-submodel argument on the stationary set of approximating cardinals, transferring the GCH instances below $\kappa$ up to $\kappa$ itself. Silver's theorem is a pure ZFC theorem with no large-cardinal hypothesis, and it explains why all known failures of SCH are engineered at cardinals of cofinality $\omega$, paradigmatically ${\aleph }_{\omega }$.

Theorem 5 (Shelah's pcf bound). If ${\aleph }_{\omega }$ is a strong limit cardinal then $2^{{\aleph }_{\omega }}<{\aleph }_{{\omega }_4}$ ^{[Shelah 1994]}. This absolute ZFC bound emerges from pcf theory — the study of the possible cofinalities $\mathrm{pcf}(A)$ of ultraproducts $\prod A/D$ of a set $A$ of regular cardinals — and stands in sharp contrast to Easton's complete freedom on the regular continuum function: the singular continuum function at ${\aleph }_{\omega }$ is bounded by ZFC alone, even though each individual value $2^{{\aleph }_n}$ is Easton-free. Inaccessible cardinals enter as the regular limit cardinals that pcf theory must treat as boundary cases, and the gap between ${\aleph }_{{\omega }_1}$ (Silver) and ${\aleph }_{{\omega }_4}$ (Shelah) marks the current frontier.

Synthesis. The foundational reason cardinal arithmetic splits into a free part and a constrained part is cofinality: the absorption laws of 42.03.03 discard all additive and multiplicative content, König's theorem leaves exactly the cofinality inequality $\mathrm{cf}(2^{\kappa })>\kappa$, and this is exactly the seam along which the regular and singular cases diverge. The central insight is that on regular cardinals monotonicity and König exhaust the ZFC constraints — Easton's theorem makes $2^{(\cdot )}$ on regulars a free class function — whereas at singulars the gimel function $\gimel (\kappa )={\kappa }^{\mathrm{cf}(\kappa )}$ is pinned by genuine theorems: Silver's, that SCH cannot first fail at uncountable cofinality, and Shelah's pcf bound $2^{{\aleph }_{\omega }}<{\aleph }_{{\omega }_4}$. These two regimes are dual: forcing (Easton, and the large-cardinal forcings breaking SCH, 42.03.08, 42.03.09) generates freedom by adding subsets, while pcf theory and Silver's reflection generate constraint by transferring information up a cofinal sequence — the bridge is that both read off the same cofinality datum that König first isolated. Putting these together, the entire modern theory of cardinal exponentiation is the theory of the gimel function, and the gimel function generalises the continuum problem of 42.03.03 from $2^{{\aleph }_0}$ to every singular cardinal at once, with the independence of the regular values 42.03.08 on one side and the ZFC theorems of pcf on the other.

Full proof set Master

Proposition 1 (cofinality is a regular cardinal). For every limit ordinal $\alpha$, $\mathrm{cf}(\alpha )$ is a regular cardinal, and $\mathrm{cf}(\mathrm{cf}(\alpha ))=\mathrm{cf}(\alpha )$.

Proof. First $\mathrm{cf}(\alpha )$ is a cardinal: a cofinal subset of $\alpha$ of order type $\mu =\mathrm{cf}(\alpha )$ exists, and if $|\mu |<\mu$ a cofinal subset of order type $|\mu |<\mu$ could be extracted by re-indexing along a bijection $|\mu |\to \mu$, contradicting minimality. For regularity, let $\mu =\mathrm{cf}(\alpha )$ with increasing cofinal $g:\mu \to \alpha$, and let $\nu =\mathrm{cf}(\mu )$ with increasing cofinal $h:\nu \to \mu$. The composite $g\circ h:\nu \to \alpha$ is increasing and cofinal: for $\beta <\alpha$ pick $\xi <\mu$ with $g(\xi )>\beta$, then $\eta <\nu$ with $h(\eta )>\xi$, so $g(h(\eta ))>\beta$. Thus $\mathrm{cf}(\alpha )\le \nu =\mathrm{cf}(\mu )\le \mu =\mathrm{cf}(\alpha )$, forcing $\mathrm{cf}(\mu )=\mu$. $\square$

Proposition 2 ($\mathrm{cf}({\aleph }_{\lambda })=\mathrm{cf}(\lambda )$ for limit $\lambda$). If $\lambda$ is a limit ordinal then $\mathrm{cf}({\aleph }_{\lambda })=\mathrm{cf}(\lambda )$; in particular ${\aleph }_{\lambda }$ is singular iff $\mathrm{cf}(\lambda )<\lambda$, and $\mathrm{cf}({\aleph }_{\omega })=\omega$.

Proof. The map $\xi \mapsto {\aleph }_{\xi }$ is a strictly increasing continuous map $\lambda \to {\aleph }_{\lambda }$ with ${\aleph }_{\lambda }={\sup }_{\xi <\lambda }{\aleph }_{\xi }$. If $X\subseteq \lambda$ is cofinal of order type $\mathrm{cf}(\lambda )$, then $\{{\aleph }_{\xi }:\xi \in X\}$ is cofinal in ${\aleph }_{\lambda }$ (continuity: any ${\aleph }_{\eta }<{\aleph }_{\lambda }$ has $\eta <\lambda$, so some $\xi \in X$ exceeds $\eta$ and ${\aleph }_{\xi }>{\aleph }_{\eta }$), of the same order type, so $\mathrm{cf}({\aleph }_{\lambda })\le \mathrm{cf}(\lambda )$. Conversely a cofinal $Y\subseteq {\aleph }_{\lambda }$ of order type $\mathrm{cf}({\aleph }_{\lambda })$ projects to $\{\eta :{\aleph }_{\eta }\le \sup (y\cap \text{level})\}$; concretely, for $y\in Y$ let $\rho (y)$ be the least $\xi$ with $y<{\aleph }_{\xi }$, then $\{\rho (y):y\in Y\}$ is cofinal in $\lambda$ of order type $\le \mathrm{cf}({\aleph }_{\lambda })$, giving $\mathrm{cf}(\lambda )\le \mathrm{cf}({\aleph }_{\lambda })$. Hence equality. With $\lambda =\omega$, $\mathrm{cf}(\omega )=\omega$ gives $\mathrm{cf}({\aleph }_{\omega })=\omega <{\aleph }_{\omega }$. $\square$

Proposition 3 (regularity of successor cardinals). Under AC, every successor cardinal ${\aleph }_{\alpha +1}$ is regular.

Proof. Suppose $\mu =\mathrm{cf}({\aleph }_{\alpha +1})\le {\aleph }_{\alpha }$, witnessed by an increasing cofinal $⟨{\gamma }_i:i<\mu ⟩$ in ${\aleph }_{\alpha +1}$ with $|{\gamma }_i|\le {\aleph }_{\alpha }$. Then ${\aleph }_{\alpha +1}={\bigcup }_{i<\mu }{\gamma }_i$ as a set of ordinals, so by the absorption law for cardinal sums of 42.03.03, ${\aleph }_{\alpha +1}=|{\bigcup }_{i<\mu }{\gamma }_i|\le {\sum }_{i<\mu }|{\gamma }_i|\le \mu \otimes {\aleph }_{\alpha }=\max (\mu ,{\aleph }_{\alpha })={\aleph }_{\alpha }$, where Choice supplies the surjections ${\aleph }_{\alpha }\twoheadrightarrow {\gamma }_i$ assembled into a surjection $\mu \times {\aleph }_{\alpha }\twoheadrightarrow {\aleph }_{\alpha +1}$. This contradicts ${\aleph }_{\alpha }<{\aleph }_{\alpha +1}$, so $\mu ={\aleph }_{\alpha +1}$. $\square$

Proposition 4 (König's inequality and the cofinality bound). If ${\kappa }_i<{\lambda }_i$ for all $i\in I$ then ${\sum }_i{\kappa }_i<{\prod }_i{\lambda }_i$; consequently $\kappa <{\kappa }^{\mathrm{cf}(\kappa )}$ and $\mathrm{cf}(2^{\kappa })>\kappa$.

Proof. For strictness it suffices that no $G:{⨆}_i{\kappa }_i\to {\prod }_i{\lambda }_i$ is surjective. For each $i$ the $i$-th coordinates of $G$ restricted to the $i$-th block form a set of size $\le {\kappa }_i<{\lambda }_i$, so some $f(i)\in {\lambda }_i$ is omitted (Choice over $i\in I$). The diagonal $f=⟨f(i){⟩}_i$ differs from every $G(x)$ in coordinate $i$ when $x$ is in the $i$-th block, so $f\notin \mathrm{ran}(G)$. For the corollary, $\kappa ={\sum }_{i<\mathrm{cf}(\kappa )}{\kappa }_i$ with ${\kappa }_i<\kappa$ and ${\lambda }_i=\kappa$ gives $\kappa <{\prod }_{i<\mathrm{cf}(\kappa )}\kappa ={\kappa }^{\mathrm{cf}(\kappa )}$. If $\mathrm{cf}(2^{\kappa })\le \kappa$ then $2^{\kappa }<(2^{\kappa }{)}^{\mathrm{cf}(2^{\kappa })}\le (2^{\kappa }{)}^{\kappa }=2^{\kappa \otimes \kappa }=2^{\kappa }$, impossible; so $\mathrm{cf}(2^{\kappa })>\kappa$. $\square$

Proposition 5 (Hausdorff recursion). For all $\alpha ,\beta$, $({\aleph }_{\alpha +1}{)}^{{\aleph }_{\beta }}={\aleph }_{\alpha +1}\otimes ({\aleph }_{\alpha }{)}^{{\aleph }_{\beta }}$.

Proof. If ${\aleph }_{\beta }\ge {\aleph }_{\alpha +1}$ then $({\aleph }_{\alpha +1}{)}^{{\aleph }_{\beta }}=2^{{\aleph }_{\beta }}=({\aleph }_{\alpha }{)}^{{\aleph }_{\beta }}$ and ${\aleph }_{\alpha +1}\otimes 2^{{\aleph }_{\beta }}=2^{{\aleph }_{\beta }}$, so equality holds. Suppose ${\aleph }_{\beta }<{\aleph }_{\alpha +1}$. By Proposition 3, ${\aleph }_{\alpha +1}$ is regular, so every $f:{\aleph }_{\beta }\to {\aleph }_{\alpha +1}$ has $\sup \mathrm{ran}(f)<{\aleph }_{\alpha +1}$ (${\aleph }_{\beta }<\mathrm{cf}({\aleph }_{\alpha +1})$). Thus ${}^{{\aleph }_{\beta }}{\aleph }_{\alpha +1}={\bigcup }_{\gamma <{\aleph }_{\alpha +1}}{}^{{\aleph }_{\beta }}\gamma$, a union of ${\aleph }_{\alpha +1}$ sets each of size $|\gamma {|}^{{\aleph }_{\beta }}\le ({\aleph }_{\alpha }{)}^{{\aleph }_{\beta }}$, so $({\aleph }_{\alpha +1}{)}^{{\aleph }_{\beta }}\le {\aleph }_{\alpha +1}\otimes ({\aleph }_{\alpha }{)}^{{\aleph }_{\beta }}$. The reverse holds by monotonicity, since ${\aleph }_{\alpha +1}\le ({\aleph }_{\alpha +1}{)}^{{\aleph }_{\beta }}$ and $({\aleph }_{\alpha }{)}^{{\aleph }_{\beta }}\le ({\aleph }_{\alpha +1}{)}^{{\aleph }_{\beta }}$ with the product absorbing into the larger factor. $\square$

Proposition 6 (Silver's theorem, the uncountable-cofinality case). Let $\kappa$ be singular with $\mathrm{cf}(\kappa )>\omega$. If $\{\mu <\kappa :2^{\mu }={\mu }^{+}\}$ contains a closed unbounded set (indeed is stationary) in $\kappa$, then $2^{\kappa }={\kappa }^{+}$.

Proof (sketch at the level of the proof set). Fix an increasing continuous cofinal sequence $⟨{\kappa }_i:i<\mathrm{cf}(\kappa )⟩$ with limit $\kappa$, and let $S=\{i:2^{{\kappa }_i}={\kappa }_i^{+}\}$ be stationary in $\mathrm{cf}(\kappa )$. To each $X\subseteq \kappa$ associate the sequence $⟨X\cap {\kappa }_i:i<\mathrm{cf}(\kappa )⟩$. Using a uniform ultrafilter $D$ on $\mathrm{cf}(\kappa )$ extending the closed-unbounded filter, two subsets $X,Y$ with $\{i:X\cap {\kappa }_i=Y\cap {\kappa }_i\}\in D$ are identified; the number of $D$-classes is bounded by ${\prod }_i{2}^{{\kappa }_i}/D$. On the stationary $S$ the values $2^{{\kappa }_i}={\kappa }_i^{+}$ are successors, and Fodor's pressing-down lemma (available because $\mathrm{cf}(\kappa )>\omega$ makes the closed-unbounded filter $\sigma$-complete and normal) bounds the ultraproduct ${\prod }_{i\in S}{\kappa }_i^{+}/D$ by ${\kappa }^{+}$. Reconstructing $X$ from its $D$-class up to a set of size $\le \kappa$ yields $2^{\kappa }\le {\kappa }^{+}\otimes \kappa ={\kappa }^{+}$, and ${\kappa }^{+}\le 2^{\kappa }$ is Cantor. The hypothesis $\mathrm{cf}(\kappa )>\omega$ is essential: Fodor's lemma fails for $\mathrm{cf}=\omega$, which is exactly why $\neg$SCH is engineered at ${\aleph }_{\omega }$. $\square$

Connections Master

• Cardinals and the arithmetic of the infinite 42.03.03 is the direct prerequisite and the source of every tool used here: the absorption laws $\oplus =\otimes =\max$, the aleph and beth functions, cardinal exponentiation ${\kappa }^{\lambda }$, and König's inequality in its sum-product form are proved there, and this unit is the cofinality refinement that 42.03.03 explicitly defers. That unit owns "exponentiation carries all the content"; this unit owns "the content is governed by cofinality," splitting it into the Easton-free regular case and the pcf-constrained singular case.

• Forcing and the independence of CH 42.03.08 is the co-produced companion that supplies the forcing machinery behind Easton's theorem stated here: Cohen forcing adds subsets of a regular $\kappa$ to make $2^{\kappa }$ any prescribed value obeying monotonicity and the König bound, and the Easton product realises an entire class function on the regular cardinals at once. This unit states Easton's theorem and the freedom of the regular continuum function; 42.03.08 builds the generic extensions that prove it, and the Prikry-type forcings that break SCH at ${\aleph }_{\omega }$ from large cardinals live at the boundary between the two units.

• Large cardinals and inaccessibility 42.03.09 is the co-produced unit where the inaccessible cardinals introduced here as regular (strong) limit cardinals become first-class objects, and where the consistency strength of $\neg$SCH is measured: Silver's theorem and Shelah's pcf bound are ZFC theorems, but their complement — actually forcing a singular strong-limit $\kappa$ with $2^{\kappa }>{\kappa }^{+}$ — provably requires large cardinals (Magidor from supercompactness, Gitik's exact calibration). The regular-limit cofinality analysis of this unit is exactly the entry point to that hierarchy.

Historical & philosophical context Master

Cofinality enters set theory with Felix Hausdorff, whose 1908 Grundzüge einer Theorie der geordneten Mengen (Math. Ann. 65) introduced regular and singular cardinals, the cofinality of an order type, and the recursion $({\aleph }_{\alpha +1}{)}^{{\aleph }_{\beta }}={\aleph }_{\alpha +1}\cdot {\aleph }_{\alpha }^{{\aleph }_{\beta }}$ that still bears his name; Hausdorff also formulated the generalized continuum hypothesis in the aleph form. Julius König's 1905 inequality $\sum {\kappa }_i<\prod {\lambda }_i$ ^{[Kunen Ch. I §13]}, which gives $\mathrm{cf}(2^{\kappa })>\kappa$, arose from a flawed attempt to refute the continuum hypothesis and survives as the one substantive ZFC constraint on the continuum function beyond Cantor's and monotonicity.

The freedom of the regular continuum function was settled by William Easton, whose 1970 Annals of Mathematical Logic paper used class forcing with an Easton-support product to realise any monotone, König-respecting class function on the regular cardinals as $\kappa \mapsto 2^{\kappa }$. The singular case proved far deeper. Jack Silver proved in 1974 that the (generalized) continuum hypothesis, and SCH, cannot first fail at a singular of uncountable cofinality ^{[Jech Ch. 8]}, using generic ultrafilters and the stationary-set machinery; Fred Galvin and András Hajnal sharpened the bounds on $2^{{\aleph }_{\omega }}$ shortly after. Menachem Magidor showed in 1977 that $\neg$SCH at ${\aleph }_{\omega }$ is consistent relative to a supercompact cardinal, and Moti Gitik calibrated its exact consistency strength, so the failure of SCH genuinely requires large cardinals. Saharon Shelah's pcf theory, presented in his 1994 Cardinal Arithmetic ^{[Shelah 1994]}, produced the absolute bound $2^{{\aleph }_{\omega }}<{\aleph }_{{\omega }_4}$ for strong-limit ${\aleph }_{\omega }$, a ZFC theorem about an exponential that Easton's theorem shows is otherwise free below the singular level. The standard modern developments are Kunen's ^{[Kunen Ch. I §13]} and Jech's ^{[Jech Ch. 5]}, from which the present treatment of cofinality, gimel, and the singular-cardinals problem is drawn.

Bibliography Master

@book{kunen2011set,
  author    = {Kunen, Kenneth},
  title     = {Set Theory},
  series    = {Studies in Logic},
  volume    = {34},
  publisher = {College Publications},
  year      = {2011}
}

@book{jech2003set,
  author    = {Jech, Thomas},
  title     = {Set Theory},
  edition   = {3},
  series    = {Springer Monographs in Mathematics},
  publisher = {Springer},
  year      = {2003}
}

@book{shelah1994cardinal,
  author    = {Shelah, Saharon},
  title     = {Cardinal Arithmetic},
  series    = {Oxford Logic Guides},
  volume    = {29},
  publisher = {Oxford University Press},
  year      = {1994}
}

@article{koenig1905koenig,
  author  = {K{\"o}nig, Julius},
  title   = {Zum Kontinuum-Problem},
  journal = {Mathematische Annalen},
  volume  = {60},
  year    = {1905},
  pages   = {177--180}
}

@article{hausdorff1908grundzuege,
  author  = {Hausdorff, Felix},
  title   = {Grundz{\"u}ge einer Theorie der geordneten Mengen},
  journal = {Mathematische Annalen},
  volume  = {65},
  year    = {1908},
  pages   = {435--505}
}

@article{easton1970powers,
  author  = {Easton, William B.},
  title   = {Powers of regular cardinals},
  journal = {Annals of Mathematical Logic},
  volume  = {1},
  year    = {1970},
  pages   = {139--178}
}

@article{silver1975singular,
  author    = {Silver, Jack},
  title     = {On the singular cardinals problem},
  journal   = {Proceedings of the International Congress of Mathematicians (Vancouver 1974)},
  volume    = {1},
  year      = {1975},
  pages     = {265--268}
}

@article{magidor1977singular,
  author  = {Magidor, Menachem},
  title   = {On the singular cardinals problem I},
  journal = {Israel Journal of Mathematics},
  volume  = {28},
  year    = {1977},
  pages   = {1--31}
}