Skip to main content
2 of 2
Added a `real' counterexample.
KP Hart
  • 11.4k
  • 38
  • 48

Here is an easy counterexample under the Continuum Hypothesis, where the union of the topologies generates a non-normal topology. Take the upper half plane $\mathbb{H}$ (including $x$-axis) and enumerate $\mathbb{R}$ as $\langle x_\alpha:\alpha<\omega_1\rangle$. Now let topology $\mathcal{T}_\alpha$ be generated by the normal topology plus the tangent disk neighbourhoods at all points $(x_\beta,0)$ for $\beta\le\alpha$. Each topology $\mathcal{T}_\alpha$ is still separable and metrizable (regular with a countable base) but the union generates the Tangent Disk topology, which is not (even) normal.

This example can be made `real' as follows: take an injective sequence $\langle x_\alpha:\alpha<\omega_1\rangle$ of real numbers and consider the subset $X$ of $\mathbb{H}$ consisting of all points above the $x$-axis and the points $(x_\alpha,0)$ on the x-axis. Perform exactly the same construction as above; the intermediate topologies are again separable and metrizable and their union generates the Tangent Disk topology on $X$. In this topology the set $\{(x_\alpha,0):\alpha<\omega_1\}$ is closed and discrete but there is no pairwise disjoint family $\{U_\alpha:\alpha<\omega_1\}$ of open sets such that $(x_\alpha,0)\in U_\alpha$ for all $\alpha$. So the space is not collectionwise Hausdorff and hence not paracompact.

KP Hart
  • 11.4k
  • 38
  • 48