Skip to article frontmatterSkip to article content

Inaccessible cardinals

The cardinality of VαV_\alpha grows rather fast relative to α{}\alpha. For example,

Vω+α=α|V_{\omega+\alpha}| = \beth_\alpha

where the beth function α\beth_\alpha is defined as

0=0α+1=2αλ=sup{α ⁣:α<λ} λ limit\begin{gather*} \beth_0 = \aleph_0 \\ \beth_{\alpha+1} = 2^{\beth_\alpha} \\ \beth_{\lambda} = \sup \{ \beth_\alpha \colon \alpha < \lambda\} \quad \text{ $\lambda$ limit} \end{gather*}

The existence of inaccessible cardinals ensures that the von-Neumann hierarchy is “long enough” for α{}\alpha to eventually “catch up” with the cardinality of VαV_\alpha.

Recall the enumeration of all cardinals by means of the \aleph-sequence:

0=ω,α+1=α+,λ=sup{ξ ⁣:ξ<λ}   for limit λ.\aleph_0 = \omega, \quad \aleph_{\alpha+1} = \aleph_\alpha^+, \quad \aleph_\lambda = \sup \{ \aleph_\xi \colon \xi<\lambda\} \; \text{ for limit } \lambda.

Here κ+\kappa^+ is the least cardinal >κ> \kappa. Some cardinals are limits of short sequences of cardinals -- for example,

ω=limnn\aleph_\omega = \lim_n \aleph_n

is uncountable, but a limit of a countable sequence of smaller cardinals. Generally, cardinals who are a limit of a sequence of cardinals of length smaller than their cardinality are called singular. Non-singular cardinals are called regular:

reg(κ):    α<κ  f  (f:ακ    supξ<αf(ξ)<κ).\Op{reg}(\kappa): \iff \forall \alpha < \kappa \; \forall f \;( f: \alpha \to \kappa \; \to \; \sup_{\xi < \alpha} f(\xi) < \kappa).

In other words, a regular cardinal κ{}\kappa cannot be reached by less then κ{}\kappa-many steps. The first example of a regular cardinal is 0\aleph_0.

On the other hand,

ω,ω+ω,ω,ω,\aleph_\omega, \aleph_{\omega+\omega}, \aleph_{\aleph_{\omega}}, \aleph_{\aleph_{\aleph_\omega}}, \ldots

are singular and this suggests the question:

Are there regular cardinals of the form λ\aleph_\lambda with λ\lambda limit?

This is captured by the notion of inaccessibility.

Under the Generalized Continuum Hypothesis,

(GCH)α    2α=α+(\mathsf{GCH}) \quad \forall \alpha \;\; 2^{\aleph_\alpha} = \aleph_{\alpha}^+

weakly and strongly inaccessible cardinals coincide.

If κ>ω\kappa > \omega is inaccessible, then κ=κ\kappa = \aleph_\kappa. Moreover, we have

Proof

It suffices to show that Vα<κ|V_\alpha| < \kappa for all α<κ\alpha < \kappa. This follows by a straightforward induction, using the fact that κ{}\kappa is strongly inaccessible.

This in turn implies we can bound the cardinality of elements of VκV_\kappa.

Proof

(\Rightarrow) xVκx \in V_\kappa implies x<Vκ|x| < |V_\kappa|. Apply Proposition 1.

(\Leftarrow) Since xVκx \subseteq V_\kappa, each yxy \in x has rank <κ< \kappa. Since x<κ|x| < \kappa, by regularity of κ{}\kappa,

rank(x)=sup{rank(y)+1 ⁣:yx}<κ\Op{rank}(x) = \sup\{\Op{rank}(y)+1 \colon y \in x\} < \kappa

which implies xVκx \in V_\kappa.

We have already seen that for limit α>ω\alpha > \omega, VαV_\alpha is a model of all ZFC\ZFC axioms except Replacement.

Proof

We verify that VκV_\kappa satisfies the axiom of Replacement. Suppose xVκx \in V_\kappa and f:xVκf:x \to V_\kappa is a function. Then f[x]Vκf[x] \subseteq V_\kappa, and by Proposition 2, f[x]x<κ|f[x]| \leq |x| < \kappa. Applying the other direction of Proposition 2 to f[x]f[x], we obtain f[x]Vκf[x] \in V_\kappa, as desired.

Suppose an inaccessible cardinal exists, and let κ{}\kappa be the least inaccessible.
It is not hard to verify that

VκZFC+“there does not exist an inaccessible cardinal”.V_\kappa \models \ZFC + \text{``there does not exist an inaccessible cardinal''}.

(You verify that being a inaccessible cardinal is absolute for VκV_\kappa.) Therefore, the existence of an inaccessible cardinal is not provable from ZFC\ZFC. This fact also follows from Gödel’s second incompleteness theorem.

Measurability

We have seen that (assuming the Axiom of Choice) there subsets of R\Real that are not Lebesgue measurable. Inspecting the proof, we see that we only use the following properties of Lebesgue measure:

  • σ{}\sigma-additivity,
  • translation invariance (λ(A)=λ(A+r)\lambda(A) = \lambda(A+r)),
  • λ(A)>0\lambda(A) > 0 for some AA.

For spaces without an additive structures, instead of translation invariance, we can consider a non-triviality condition:

m({x})=0 for all xm(\{x\})=0 \quad \text{ for all $x$}

The generalized measure problem asks whether there exists a set MM together with a measure function

m:P(M)[0,),\begin{gather*} m: \mathcal{P}(M) \to [0,\infty), \end{gather*}

so that the following conditions are met:

  • (M1) \quad m(M)=1m(M) =1
  • (M2) \quad xM  m({x})=0\forall x \in M \; m(\{x\})=0
  • (M3) \quad if (Ai)i<ω(A_i)_{i < \omega} is a countable sequence of disjoint sets M\subseteq M, then
m(i<ωAi)=i<ωm(Ai)m\left(\bigcup_{i<\omega} A_i\right ) = \sum_{i<\omega} m(A_i)

The structure of the set MM does not play any role here, so we can replace it by a cardinal κ{}\kappa outright. One can also consider strengthening σ{}\sigma-additivity to κ{}\kappa-additivity:

If γ<κ\gamma < \kappa and (Aξ)ξ<γ(A_\xi)_{\xi< \gamma} is a sequence of disjoint subsets of κ{}\kappa, then

m(ξ<γAξ)=ξ<γm(Aξ).m(\bigcup_{\xi<\gamma} A_\xi) = \sum_{\xi<\gamma} m(A_\xi).

A transfinite sum ξ<γ\sum_{\xi<\gamma} is given as the supremum of all sums over finite subsequences:

ξ<γm(Aξ)=sup{ξFm(Aξ) ⁣:Fγ finite}.\sum_{\xi<\gamma} m(A_\xi) = \sup \left \{ \sum_{\xi \in F} m(A_\xi) \colon F \subseteq \gamma \text{ finite}\right \}.

Hence, ω1\omega_1-additive is the same as σ{}\sigma-additive.

Proof

Suppose mm is a measure on κ{}\kappa that is not κ{}\kappa-additive. Then, for some γ<κ\gamma < \kappa, there exists a sequence (Aξ)ξ<γ(A_\xi)_{ \xi< \gamma} of disjoint subsets of κ{}\kappa so that

m(ξ<γAξ)ξ<γm(Aξ).m(\bigcup_{\xi<\gamma} A_\xi) \ne \sum_{\xi<\gamma} m(A_\xi).

Since a measure is always σ{}\sigma-additive, γ>ω\gamma > \omega has to hold, and there can be at most countably many AξA_\xi with m(Aξ)>0m(A_\xi)>0.

We can drop those AξA_\xi, and by the σ{}\sigma-additivity of mm for the remaining ξ\xi it has to hold that m(Aξ)=0m(A_\xi)=0 while m(ξ<γAξ)=r>0m \left(\bigcup_{\xi<\gamma} A_\xi \right) = r >0.

By putting

m(X)=m(ξXAξ)r\overline{m}(X) = \frac{m(\bigcup_{\xi \in X} A_\xi)}{r}

we obtain a measure on γ<κ\gamma < \kappa, contradicting the minimality of κ{}\kappa.

Measurable cardinals

If mm is a measure on κ{}\kappa, the associated ideal

Im={xκ ⁣:m(x)=0}\mathcal{I}_m = \{x\subseteq \kappa \colon m(x) = 0 \}

is a σ{}\sigma-ideal, or, complementing the notion of ω1\omega_1-additivity, a ω1\omega_1-complete ideal.

The corresponding filter

Fm={xκ ⁣:m(x)=1}\mathcal{F}_m = \{x\subseteq \kappa \colon m(x) = 1\}

is then ω1\omega_1-complete, too.

A measure mm is two-valued if it only assumes the values 0 and 1. In this case the corresponding filter Fm\mathcal{F}_m is an ultrafilter (and Im\mathcal{I}_m is a prime ideal).

Conversely, if UU is ω1\omega_1-complete, non-principal ultrafilter on κ{}\kappa, we can define a two-valued measure m:P(κ){0,1}m: \mathcal{P}(\kappa) \to \{0,1\} on κ{}\kappa by letting

m(x)={1if xU,0otherwise.m(x) = \begin{cases} 1 & \text{if } x \in U, \\ 0 & \text{otherwise}. \end{cases}

In the following, we will see that measurability implies inaccessibility.

Proof

Since UU is non-principal, no singleton set {x}\{x\} can be in UU (for this would imply κ{x}U\kappa\setminus \{x\} \notin U and therefore no subset of it would be in UU either, contradicting the non-principality of UU).

If XUX \in U and X<κ|X| < \kappa, then XX is the union of <κ< \kappa many singletons. Since ¬U\neg U is a κ{}\kappa-complete prime ideal, this implies X¬UX \in \neg U, contradiction.

Proof

If κ{}\kappa were singular, it would be the union of <κ<\kappa-many sets of cardinality <κ<\kappa. Applying Lemma 1 leads to a contradiction.

Proof

By Proposition 3, any measurable cardinal is regular. Assume for a contradiction there exists γ<κ\gamma < \kappa with 2γκ2^\gamma \geq \kappa. As 2γκ2^\gamma \geq \kappa, there exists a set SS of functions f:γ{0,1}f: \gamma \to \{0,1\} with S=κ|S| = \kappa. Let UU be a κ{}\kappa-complete, non-principal ultrafilter on SS.

For α<γ,i{0,1}\alpha < \gamma, i \in \{0,1\}, let

Xα,i={fS ⁣:f(α)=i}X_{\alpha,i} = \{ f \in S \colon f(\alpha) = i\}

and let g(α)=ig(\alpha) = i if and only if Xα,iUX_{\alpha,i} \in U. Since UU is an ultrafilter, gg is well-defined on γ\gamma.

Since γ<κ\gamma < \kappa and UU is κ{}\kappa-complete,

X=α<γXα,g(α)X = \bigcap_{\alpha < \gamma} X_{\alpha, g(\alpha)}

is in UU. But X1|X| \leq 1, since the only function possibly in XX is gg. This contradicts Lemma 1.

Thus, if κ{}\kappa is real-valued measurable but not measurable, then the continuum 202^{\aleph_0} has to be very large.

Partition properties

Another concept of largeness is related to the existence of large homogeneous sets for partitions.

For given set SS and nNn \in \Nat, let

[S]n:={XS ⁣:X=n}[S]^n := \{ X \subseteq S \colon \: |X| = n \}

be the set of all nn-element subsets of SS. For cardinals κ,λ\kappa, \lambda, we define

κ(λ)kn\kappa \to (\lambda)^n_k

to mean that any partition F:[S]n{1,,k}F: [S]^n \to \{1, \dots, k\} with S=κ|S| = \kappa has an FF-homogeneous subset of cardinality λ\lambda, that is, a set HH, H=λ|H| = \lambda, such that

F[H]n constant.F|_{[H]^n} \equiv \text{ constant}.

Ramsey’s theorem (1929/39) says that for any n,kNn,k \in \Nat,

0(0)kn.\aleph_0 \to (\aleph_0)^n_k.

Do there exist uncountable cardinals with similar properties?

A cardinal κ{}\kappa is weakly compact if it is uncountable and κ(κ)22\kappa \to (\kappa)^2_2 holds.

Measurable cardinals have even stronger homogeneity properties. Let [S]<ω[S]^{<\omega} be the set of all finite subsets of SS. If F:[S]<ωIF: [S]^{<\omega} \to I is a partition, then HSH \subseteq S is FF-homogenenous if

F[H]n constantF|_{[H]^n} \equiv \text{ constant}

for all nNn \in \Nat.

In general, any cardinal that satisfies the statement of the theorem is called Ramsey.

To prove Theorem 4, we introduce normal ultrafilters.

Let us assume as a convention that a filter on a cardinal κ{}\kappa always contains the end-segments {ξ ⁣:αξ<κ}\{\xi \colon \alpha \leq \xi < \kappa\}.

Proof (Rowbottom’s Theorem)

Let UU be a normal filter over κ{}\kappa. We show that for every nn, for any g:[κ]nγg: [\kappa]^n \to \gamma with γ<κ\gamma < \kappa, there is a set HnUH_n \in U such that g[Hn]nconstg \Rest{[H_n]^n} \equiv \text{const}. The intersection of the HnH_n is again in UU and satisfies the statement of the the theorem.

We proceed by induction. The case n=1n=1 follows from the κ{}\kappa-completeness of UU. Now assume g:[κ]n+1γg:[\kappa]^{n+1} \to \gamma, with γ<κ\gamma < \kappa.

For each S[κ]nS \in [\kappa]^n, define gS:κγg_S : \kappa \to \gamma by

gS(α)={g(S{α}) if maxS<α0otherwiseg_S(\alpha) = \begin{cases} g(S \cup \{\alpha\}) & \text{ if } \max S < \alpha \\ 0 & \text{otherwise} \end{cases}

By κ{}\kappa-completeness of UU, gSg_S is constant on a set YSUY_S \in U, say

gSYSδS<γ.g_S\Rest{Y_S} \equiv \delta_S < \gamma.

We now define a function h:[κ]nγh: [\kappa]^n \to \gamma by letting

h(S)=δS.h(S) = \delta_S.

By induction hypothesis, hh is constant on a set ZκZ \subseteq \kappa in UU (and hence of size κ{}\kappa), say h[Z]nδ<κh\Rest{[Z]^n} \equiv \delta < \kappa.

For each α<κ\alpha < \kappa, let

Yα={YS ⁣:maxSα}Y_\alpha = \bigcap \{Y_S \colon \max S \leq \alpha\}

By κ{}\kappa-completeness, YαUY_\alpha \in U, and by normality

H=ZΔα<κYαUH = Z \cap \Delta_{\alpha < \kappa} Y_\alpha \in U

By Lemma 1, HH has cardinality κ{}\kappa.

We claim that gg is constant on [H]n+1[H]^{n+1}: Let T[H]n+1T \in [H]^{n+1}. Write TT as S{α}S \cup \{\alpha\} with maxS<α\max S < \alpha. Then

αHαΔγ<κYγαβ<αYβαYmaxSαYSgS(α)=δS\begin{align*} \alpha \in H & \Rightarrow & \alpha \in \Delta_{\gamma < \kappa} Y_\gamma \\ & \Rightarrow & \alpha \in \bigcap_{\beta < \alpha} Y_\beta \\ & \Rightarrow & \alpha \in Y_{\max S} \\ & \Rightarrow & \alpha \in Y_S \\ & \Rightarrow & g_S(\alpha) = \delta_S \end{align*}

On the other hand, SHS \subseteq H implies SZS \subseteq Z and hence by definition of ZZ, h(S)=δS=δh(S) = \delta_S = \delta, so g(T)=gS(α)=δS=δg(T) = g_S(\alpha) = \delta_S = \delta.