There is a cofinal branch. We give a proof that does not require distinct limit-level nodes to have different predecessor chains.

The set $S=\{\alpha <\lambda :\mathrm{cf}(\alpha )=\kappa \}$ is stationary. Indeed a strictly increasing continuous $\kappa$-sequence in any club set has supremum in that club set, below $\lambda$, of cofinality $\kappa$. At each $\alpha \in S$, there are fewer than $\kappa$ pairs of nodes on $T_{\alpha }$. For every pair whose predecessor chains below $\alpha$ differ, choose a height where they differ. Since $\mathrm{cf}(\alpha )=\kappa$, all these heights are bounded by some ${\beta }_{\alpha }<\alpha$. Consequently nodes in $T_{\alpha }$ having the same predecessor at ${\beta }_{\alpha }$ have identical predecessor chains below $\alpha$.

The Fodor lemma states that a regressive function on a stationary subset of a regular uncountable cardinal is constant on a stationary subset. Applying it here, ${\beta }_{\alpha }$ has a constant value $\beta$ on a stationary subset $S^{\prime }$. Choose $t_{\alpha }\in T_{\alpha }$ for each $\alpha \in S^{\prime }$. The level $T_{\beta }$ has fewer than $\kappa <\lambda$ nodes. Partitioning $S^{\prime }$ according to the predecessor of $t_{\alpha }$ at height $\beta$, one fiber $S^{\prime \prime }$ is stationary, since the union of fewer than $\lambda$ nonstationary sets is nonstationary. Let its common predecessor be $t$.

For $\alpha <{\alpha }^{\prime }$ in $S^{\prime \prime }$, the predecessor of $t_{{\alpha }^{\prime }}$ at level $\alpha$ and $t_{\alpha }$ have the same predecessor $t$ at level $\beta$. Their chains below $\alpha$ therefore agree. For each $\nu <\lambda$, choose $\alpha \in S^{\prime \prime }$ above $\nu$ and let $b_{\nu }$ be the predecessor of $t_{\alpha }$ at level $\nu$. The preceding comparison makes $b_{\nu }$ independent of that choice. The nodes $b_{\nu }$ form a chain through every level:This proves the uniformly narrow regular-height tree branch theorem. The uniform bound below the smaller regular $\kappa$ is stronger than merely bounding each level below $\lambda$.

Write ${◊}_S^{\prime }$ for the countable-family assertion in the PDF; its prime is not a superscript $c$. A single stationary diamond at a regular cardinal sequence gives such families by taking singletons, so ${◊}_S\Rightarrow {◊}_S^{\prime }$.

Conversely enumerate each countable family as $E_{\alpha }(n)$, $n<\omega$, padding finite families and allowing the empty set as a default. Fix a bijection $\pi :\lambda \times \omega \to \lambda$. The previous part applied to $B(\xi )=\{\pi (\xi ,n):n<\omega \}$ gives a club set $D$ on which $\pi [\alpha \times \omega ]\subseteq \alpha$. Define, for each $n$, a single candidate sequenceSuppose no candidate sequence witnesses ${◊}_S$. For each $n$ choose $X_n\subseteq \lambda$ and a club set $C_n$ such that $D_{\alpha }^n\ne X_n\cap \alpha$ for every $\alpha \in S\cap C_n$. Code all the counterexamples intoThe countable-family hypothesis guesses $X$ on a stationary set. Choose a guessing $\alpha$ in the club set $D\cap {\bigcap }_n{C}_n$, and choose $n$ with $E_{\alpha }(n)=X\cap \alpha$. For every $\xi <\alpha$, closure under $\pi$ givesThus $D_{\alpha }^n=X_n\cap \alpha$, contradicting $\alpha \in C_n$. At least one candidate is a diamond sequence, provingThis is the countable-family diamond equivalence, for every regular uncountable $\lambda$ and stationary $S\subseteq \lambda$.

First the increasing sequence is cofinal in $\kappa$. Write $\mu ={\sup }_{\xi <\delta }{\kappa }_{\xi }$. Strict increase gives $|\delta |\le \mu$ when the sequence is infinite, so its cardinal sum is $\max (|\delta |,\mu )=\mu$. The given sum therefore implies $\mu =\kappa$ and $|\delta |\le \kappa$.

For any function $f:\lambda \to \kappa$, choose an index ${\xi }_t$ with $f(t)<{\kappa }_{{\xi }_t}$ for every $t<\lambda$. There are at most $\lambda <\mathrm{cf}(\delta )$ such indices, so they are bounded below $\delta$. Enlarging the bound if needed gives an index $\xi$ with $\mathrm{ran}(f)\subseteq {\kappa }_{\xi }$. ThusConversely each summand is at most ${\kappa }^{\lambda }$, and $|\delta |\le \kappa \le {\kappa }^{\lambda }$ because $\lambda \ne 0$. Infinite cardinal multiplication consequently givesCombining the inequalities proves ${\kappa }^{\lambda }={\sum }_{\xi <\delta }{\kappa }_{\xi }^{\lambda }$.

Put $\theta =\mathrm{cf}(\delta )$. There is a strictly increasing cofinal map $a:\theta \to \delta$. If $\mathrm{cf}(\theta )=\nu <\theta$, choose a cofinal map $b:\nu \to \theta$. Then $a\circ b$ is cofinal in $\delta$, contradicting the definition of $\theta$ as its least cofinal order type. Since always $\mathrm{cf}(\theta )\le \theta$, we obtainFor a nonzero limit ordinal this is an infinite regular cardinal. If the convention includes zero among limit ordinals, its cofinality is zero and the identity is immediate there.

The structure $H_{\kappa }$ of hereditarily small sets is transitive. If $\kappa \le \omega$, all its sets are hereditarily finite, so it cannot satisfy Infinity. Therefore the assumed model has $\kappa >\omega$.

Let $\mu <\kappa$ be a cardinal. The ordinal $\mu$ belongs to $H_{\kappa }$, and every actual subset of $\mu$ also belongs to $H_{\kappa }$: its transitive closure has size at most $\max (\mu ,\omega )<\kappa$. The Power set axiom inside $H_{\kappa }$ therefore produces the actual $\mathcal{P}(\mu )$, since all subsets relevant to the internal definition are present. This power set itself belongs to $H_{\kappa }$, soThus $\kappa$ is a strong limit. Its regularity was assumed, and we have proved uncountability. Hence $\kappa$ is strongly inaccessible.