Work in ZFC. Read transfinite recursion, initial ordinals, cofinality, Hartogs and infinite cardinal arithmetic first. No forcing, consistency, independence or large-cardinal existence result is assumed or proved.
Motivation
The first infinite cardinals describe only the beginning of an unbounded hierarchy. At a successor stage one can either take the next cardinal or take a power set. At a limit stage one takes a supremum. These choices produce the aleph and beth hierarchies, which need not agree without an additional hypothesis.
A limit stage also has a structural memory of its index: the cofinality of the resulting cardinal equals the cofinality of that limit index. This identifies many singular cardinals directly. It does not make every limit cardinal singular, and a fixed point of the aleph function need not be regular. We will prove these distinctions rather than infer them from suggestive notation.
Definitions
Definition 2.5.1
(Aleph and beth hierarchies)
Identify cardinals with initial ordinals and let be the next cardinal. Using Hartogs at successors, define
For each set ordinal, the earlier recursion theorem constructs the initial segment; uniqueness makes the segments agree, defining a class function. It does not create a set of all cardinals. The beth hierarchy is
Choice identifies the power-set cardinal with an initial ordinal. A class function on ordinals is normal here if strictly increasing and continuous at nonzero limit ordinals.
Definition 2.5.2
(Regular, singular and strong-limit cardinals)
An infinite cardinal is regular if , otherwise singular. It is a strong limit if for every cardinal . A limit cardinal is an infinite cardinal not the successor of another infinite cardinal; in statements about uncountable limit cardinals we exclude explicitly. A fixed point of a hierarchy is an ordinal with or , respectively.
CH means . GCH means for every infinite cardinal. These are statements to be analyzed, not conclusions of the definitions.
Theorem/proposition
Theorem 2.5.3
(The two hierarchies are normal and unbounded)
The aleph hierarchy enumerates every infinite cardinal exactly once. Both hierarchies are normal, dominate their ordinal indices, and satisfy . Every beth at a nonzero limit index is a strong-limit cardinal.
Theorem 2.5.4
(Successor regularity and limit cofinalities)
Every infinite successor cardinal is regular in ZFC. For every nonzero limit ordinal ,
The formula concerns limit indices; it is not a formula for successor indices, whose ordinal cofinality is one.
Theorem 2.5.5
(Fixed points exist without regularity)
Each hierarchy has fixed points above every prescribed ordinal. Its least fixed point is a singular cardinal of cofinality . Thus an aleph fixed point is not automatically an uncountable regular limit cardinal.
Proof sketch or proof idea
Proof
(Normality, enumeration and strong limits)
First, the supremum of a strictly increasing sequence of initial ordinals indexed by a nonzero limit is initial. If an ordinal smaller than that supremum were equinumerous with it, choose a stage cardinal larger than . Restricting the proposed bijection gives an injection of that initial ordinal into a smaller ordinal, impossible. Thus both recursions continue through limit stages as cardinal-valued functions. Hartogs gives strict increase for aleph successors, and Cantor gives it for beth successors. Limit continuity is their definition.
A strictly increasing continuous ordinal function dominates its index, by induction: the successor value exceeds the preceding one, and limits preserve the inequalities by suprema. Hence and . Given any infinite initial cardinal , some aleph exceeds it, for example the value at index . Let the first such index be . It cannot be zero or a limit, since all earlier values are at most and their supremum would still be at most . Thus , with . There is no intervening cardinal, so . Strict increase proves uniqueness.
Induction also gives . At a successor, is a cardinal, so is at least the next aleph. At a limit, take suprema. For at a nonzero limit, any cardinal is bounded by some , . Consequently
Here and power-set cardinality is monotone by extending binary functions. This proves the strong-limit claim.
Proof
(Successor regularity and the limit-cofinality formula)
For regularity of , suppose a cofinal sequence of ordinals had cardinal length . Then , and every has size at most . Choice selects the required injections simultaneously. The cofinal union is the whole ordinal , but its cardinal is at most
contradicting its definition. This is where both the family of encodings and the earlier infinite-square theorem enter. Regularity of follows separately because a finite set of natural numbers is bounded.
For the limit formula, let be either normal hierarchy and a nonzero limit. A cofinal map into remains cofinal after applying , proving . Conversely take a cofinal family in of minimal cardinal size. For each , choose the least with , which exists by continuity. The indices must be unbounded in : a bound would bound every by . Hence . The two inequalities prove equality.
Proof X-ray
Which property of a normal function each inequality uses
The upper bound uses only continuity at and monotonicity: if is cofinal in , any lies below some , and once . The lower bound runs the other way. Continuity again produces the index , and strict increase supplies the closing step , so a bounded set of indices cannot reach the top. The map takes least elements and needs no choice. Nothing in the argument mentions power sets or successor cardinals, which is why one proof serves both hierarchies.
Proof
(Iterating a normal function to a fixed point)
For fixed points of a normal , start at any ordinal and set . Index domination makes the sequence nondecreasing. If it becomes constant it has reached a fixed point. Otherwise it is strictly increasing, and is a nonzero limit. Continuity gives
Taking above a prescribed bound gives a fixed point above it. For the least fixed point, start at zero. Since , strict increase of makes every subsequent step strict. Any fixed point bounds every by induction, so their supremum is the least one. Its strictly increasing countable cofinal sequence gives cofinality : finite sets cannot be cofinal in a limit ordinal. It is a cardinal because it equals a hierarchy value, and it is uncountable, so this proves singularity.
Proof X-ray
Limit continuity is built into the hierarchy, not into power sets
At a limit index the definition takes a supremum; at a beth successor it takes a power set. Continuity of the class function therefore justifies replacing a cofinal set of indices by its hierarchy values. It does not justify commuting the power-set operation with an arbitrary increasing supremum of cardinals. The beth- calculation below gives a strict counterexample.
Worked examples
Worked example 2.5.6
(Two different limit cofinalities)
The ordinals and have countable cofinality, using and respectively. Thus and have cofinality and are singular. In contrast , since is a regular successor cardinal. This aleph is still singular: its value strictly exceeds . A singular cardinal need not have countable cofinality.
Worked example 2.5.7
(Beth- is a strong limit, but its power set is larger)
Let . It is strong limit and has cofinality . Shifting a cofinal sequence by one stage gives
Thus cardinal exponentiation need not preserve this supremum. The earlier König consequence moreover gives , although the base itself has countable cofinality. These are bounds and cofinalities, not an evaluation of the power-set cardinal.
Worked example 2.5.8
(What CH and GCH would identify)
CH identifies just with . GCH identifies with at every index: equality begins at zero, the successor step follows from the power-set assumption, and limit steps use identical suprema. Conversely, if all those values agree, the beth successor definition gives for each ; aleph enumeration then gives GCH for every infinite cardinal. This proves an equivalence of formulations, not either hypothesis. In particular it does not compute from the ZFC axioms.
Worked example 2.5.9
(A successor index does not inherit the limit formula)
For any ordinal , , because its last element alone is cofinal. Yet is a regular infinite successor cardinal, with cofinality equal to itself rather than one. The limit-index hypothesis therefore cannot be dropped. At a beth successor the earlier König bound gives , again not the index's cofinality. Nor does the notation make that beth a successor cardinal: taking a power set is not the same definition as taking the next cardinal. Without an additional hypothesis, the proved bound alone does not assert that the power-set cardinal is regular. None of these distinctions contradicts normality, whose continuity clause addresses only limit indices. Reading where the successor symbol appears is essential before applying a cofinality formula.
Common mistakes
Common mistake
(Successor ordinal versus successor cardinal)
A successor ordinal and a successor cardinal are different. An infinite successor cardinal is a limit ordinal, yet is regular in the stated choice background. Conversely a limit-index aleph has the index's cofinality, which need not equal its cardinal size.
Common mistake
(Strong limit is not regularity)
Strong limit and regular refer to different operations. Beth- is strong limit and singular, and the least aleph fixed point is singular too. No existence of an uncountable regular limit cardinal was proved or refuted by these examples.
Common mistake
(A class definition decides neither a set of all cardinals nor CH)
The functions are definable classes, each set-length initial segment obtained by recursion. Writing their definitions does not create a set of all alephs or beths. Finally, CH/GCH and other continuum-value claims require separate arguments; notation alone supplies none.
Counterexample mode
Regularity of successors does not pass to limit cardinals
Claim: since and every with are regular, every infinite cardinal is regular. Take . If , then is not an upper bound of all the , so for some . Hence the countable set is cofinal in , while no finite set is cofinal in a limit ordinal:
The successor proof breaks because no cardinal satisfies : every lies below some , so the cofinal stages are not uniformly bounded in size by . The correct statement keeps the successor hypothesis. At a nonzero limit index, regularity of would give , so .
Summary
Hartogs successors and transfinite suprema enumerate the alephs, while power sets define the beths. Both functions are normal, with beth limit stages strong limit. Successor cardinals are regular in ZFC; limit-index cofinality is inherited from the index. Countable normal-function iteration produces fixed points, including singular least fixed points, without deciding continuum hypotheses.
Exercises
Quick check
(In the successor-regularity proof, identify exactly where choice and the square theorem are used. Why does the argument not prove all limit cardinals regular?)
Solution · Answer
For a hypothetical short cofinal family below , each member separately has an injection into ; choice provides the entire indexed family of injections. The square theorem then bounds the union by . At a limit cardinal there need not be one smaller cardinal bounding the sizes of all cofinal stages, so the same estimate has no such uniform available. Beth- already supplies a singular limit counterexample.
Quick check
(Prove by a full transfinite induction, and explain why GCH changes each inequality into equality.)
Solution · Answer
Equality holds at zero. If the inequality holds at , Cantor gives the cardinal , hence it is at least . At a nonzero limit take suprema of the stage inequalities. Under GCH the successor step instead reads , and limit equalities follow by the definitions. This is a conditional proof throughout, not a proof that GCH holds.
Quick check
(Compute and , and determine whether these two cardinals are singular.)
Solution · Answer
They are and , respectively. The first uses the cofinal sequence in ; the second uses regularity of . Both hierarchy values exceed their respective cofinalities, since the first is uncountable and the second is at least . Both are therefore singular, with different cofinalities.
Quick check
(Prove every uncountable strong-limit cardinal equals for some nonzero limit ordinal .)
Solution · Answer
Since the beth function dominates indices, some value is at least ; let be the first such index. It is not zero, because . If , then and the strong-limit property gives , a contradiction. Hence is a nonzero limit. All earlier values are below , so their supremum is at most ; the chosen inequality makes it exactly . Thus .
Quick check
(If an uncountable regular limit cardinal exists, prove . Explain why the converse fails for the least aleph fixed point.)
Solution · Answer
Write by enumeration. Since it is an uncountable limit cardinal, is a nonzero limit ordinal. Regularity and the cofinality formula give , so and it is an aleph fixed point. The constructed least aleph fixed point has cofinality and is uncountable, hence is not regular. Fixedness is necessary under the hypothetical regular-limit assumptions, not sufficient and not an existence proof for those assumptions.