Hereditarily normal spaces

In the field of general topology, mathematicians seek to classify abstract spaces based on their intrinsic properties, much like biologists classify organisms. Some of these properties, termed hereditary, are reliably passed down to any subspace, offering a sense of stability and predictability. However, not all intuitive properties behave this way. One of the most fundamental separation properties, normality, surprisingly fails this test of inheritance, creating a crucial knowledge gap and motivating a deeper investigation into the structure of space.

This article delves into this fascinating subtlety within topology. It explores the concepts of normal and hereditarily normal spaces, uncovering the elegant reasons behind their differences and connections. Across the following chapters, you will gain a clear understanding of this important topological distinction. The "Principles and Mechanisms" chapter will dissect why normality is not hereditary, introduce the stronger condition of complete normality, and establish their equivalence. Subsequently, the "Applications and Interdisciplinary Connections" chapter will reveal where these robustly normal spaces are found, from the familiar metric spaces of geometry and analysis to the more abstract realms of ordered spaces, situating hereditary normality within the grand map of mathematical ideas.

In our journey through the world of topology, we act like explorers mapping out a vast, unseen landscape. We classify different "spaces" by their properties, much like a biologist classifies species. Some properties are simple and robust. For example, if a space is ​​Hausdorff​​ (meaning any two distinct points can be put in their own separate open "bubbles"), any piece you carve out of that space—any ​​subspace​​—will also be Hausdorff. Such properties, which are passed down to all subspaces, are called ​​hereditary​​. It's a bit like having a family trait that every single descendant inherits without exception.

But not all traits are so simple. This brings us to a wonderfully subtle and important property called ​​normality​​.

A topological space is called ​​normal​​ if it satisfies two conditions: first, individual points are closed sets (a $T_1$ space), and second, any two disjoint "islands" that are closed sets can be separated by disjoint open "moats." That is, for any two disjoint closed sets $C_1$ and $C_2$, you can always find two disjoint open sets $U_1$ and $U_2$ such that $C_1$ is entirely inside $U_1$ and $C_2$ is entirely inside $U_2$.

This seems like a very reasonable and stable property. It suggests a certain "roominess" in the space, enough to build a buffer zone between any two closed, non-overlapping regions. So, we must ask the natural question: is normality hereditary? If we take a normal space and look at one of its subspaces, must that subspace also be normal?

The answer, rather shockingly, is no. This is one of those delightful moments in mathematics where our intuition, honed on simpler spaces, is shown to be beautifully incomplete. Normality is like a complex genetic trait that can, under the right circumstances, fail to be passed on.

To see how this inheritance can fail, we need to meet a famous character in the menagerie of topological spaces: the ​​Tychonoff plank​​. Constructing it is an adventure in itself. We start with two special building blocks.

First, imagine the set of whole numbers $0, 1, 2, \dots$ and add a single point at the very "end" of the line, which we'll call $\omega$. This gives us the set $[0, \omega]$. The second block is far more exotic. It's the set of all countable ordinal numbers, which you can think of as a way of counting that goes far, far beyond the regular integers. At the end of this incredibly long line of countable numbers, we place the first number that is uncountable, a mind-bogglingly distant point we call $\Omega$. This gives us the set $[0, \Omega]$. A key feature of this set is that if you take any countable collection of points from $[0, \Omega)$ (all points except the very last one, $\Omega$), you can always find another point in $[0, \Omega)$ that is larger than all of them. You can't reach $\Omega$ by taking a countable number of steps.

Now, let's form a rectangle by taking the product of these two lines: $X = [0, \Omega] \times [0, \omega]$. This space $X$ is a "compact Hausdorff" space, a type of space that is known to be very well-behaved. And because of this, it is guaranteed to be ​​normal​​.

The mischief begins when we take this perfectly normal rectangular space and pluck out a single point: the "top-right corner," $(\Omega, \omega)$. The remaining space, $Y = X \setminus \{(\Omega, \omega)\}$, is the Tychonoff plank. This subspace $Y$ is no longer normal.

Let's see why. In this new space $Y$, consider two sets: the "right edge" (without its top point), $A = \{(\Omega, n) \mid n \omega\}$, and the "top edge" (without its right point), $B = \{(x, \omega) \mid x \Omega\}$. In the space $Y$, these two sets are closed and they are disjoint. If $Y$ were normal, we should be able to find two disjoint open sets, $U$ and $V$, that contain $A$ and $B$, respectively.

But here is where the strange nature of our building blocks foils us. Any open set $U$ that contains the entire right edge $A$ must, for each point $(\Omega, n) \in A$, also contain some points to its left. However, because there are only countably many points in $A$, we can find a single point $\beta \Omega$ that acts as a boundary for all this "spillage" from $U$. Now, consider a point on the top edge $B$, say $(\gamma, \omega)$, where $\gamma$ is chosen to be to the right of our boundary $\beta$. Any open set $V$ containing this point must contain some points directly "below" it. Inevitably, the open set $U$ spilling to the left and the open set $V$ spilling downwards will overlap. No matter how you try to draw the open sets, they will always intersect. The two disjoint closed sets $A$ and $B$ cannot be separated.

So, we have found a subspace of a normal space that is itself not normal. The property of normality is not hereditary.

This discovery is not a failure, but an invitation. It tells us that "normal" isn't the whole story. We need a stronger condition for spaces that do pass normality down to all their descendants. We can define such a property directly: a space is ​​hereditarily normal​​ if every single one of its subspaces is normal.

While this definition is clear, it's not very practical. How could we possibly check every single subspace? There are often uncountably many! What we need is an equivalent condition that we can check on the original space itself. This leads to a beautiful and powerful idea. We must refine what it means for sets to be "apart."

Let's define two sets, $A$ and $B$, to be ​​separated​​ if not only are they disjoint, but the closure of $A$ doesn't touch $B$, and the closure of $B$ doesn't touch $A$. Formally, $\overline{A} \cap B = \emptyset$ and $A \cap \overline{B} = \emptyset$. This is a stronger condition than just being disjoint. Disjoint sets can still be "infinitesimally close," but separated sets cannot.

Now for the remarkable insight: a space is hereditarily normal if and only if any two separated sets can be contained in disjoint open sets. This new, equivalent property is often called ​​complete normality​​ (or $T_5$). The terms "hereditarily normal" and "completely normal" describe the exact same class of spaces. We've traded a definition requiring an infinite number of checks for a single, more subtle condition on the parent space. This is a common and powerful theme in mathematics: finding an intrinsic characterization for an extrinsically defined property. And yes, by its very nature, complete normality is a hereditary property; any piece of a completely normal space is also completely normal.

It's crucial to appreciate the subtlety here. A space where only its closed subspaces are normal is simply a normal space and nothing more. The real power comes from demanding that all subspaces, open, closed, or neither, inherit normality.

So, which spaces have this desirable property of being completely normal? Is there an easy way to spot them? One important class of such spaces are the ​​perfectly normal​​ spaces.

A space is perfectly normal if it is normal and has an additional property: every closed set is a ​​$G_{\delta}$-set​​. A $G_{\delta}$-set is any set that can be written as a countable intersection of open sets. Think of zooming in on a target: a point on the real line can be described as the intersection of the open intervals $(-1, 1)$, then $(-1/2, 1/2)$, then $(-1/4, 1/4)$, and so on.

It turns out that every perfectly normal space is also completely (and thus hereditarily) normal. The proof of this fact is a beautiful extension of a famous result called Urysohn's Lemma, where one constructs a continuous function to distinguish between sets. This shows a deep link between the topological structure of a space (like sets being $G_{\delta}$) and the functions it can support.

The best part is that many familiar spaces are perfectly normal. Every ​​metric space​​—any space where you can define a notion of distance, like our everyday Euclidean space $\mathbb{R}^n$—is perfectly normal. This explains why our intuition often fails us! We live and think in a metric space, which is perfectly normal, hereditarily normal, and all-around very well-behaved. The strange behavior of the Tychonoff plank only appears when we venture into the more exotic, non-metric parts of the topological universe.

We have found a strong, stable, hereditary property: complete normality. It's possessed by many important spaces, like all metric spaces. But even this property has its limits. We must ask one more question: if we take the product of two completely normal spaces, is the result also completely normal?

Consider the ​​Sorgenfrey line​​, $\mathbb{R}_l$. This is the real number line but with a peculiar topology where the basic open sets are half-open intervals like $[a, b)$. This space is a bit strange, but it turns out to be perfectly normal, and thus completely normal.

Now, what about the product of two Sorgenfrey lines, the ​​Sorgenfrey plane​​ $\mathbb{R}_l \times \mathbb{R}_l$? This space, built from two perfectly well-behaved components, is famously not even normal, let alone completely normal. The product operation, which seems so simple, can destroy even strong properties like normality.

This final example leaves us with a profound lesson. The world of topological spaces is more intricate and surprising than we might first imagine. Properties like normality can be fragile, failing to be inherited by subspaces or preserved by products. In response, mathematicians define stronger properties like hereditary normality, uncovering deep equivalences and connections along the way. Each counterexample is not a roadblock but a signpost, pointing us toward a richer and more nuanced understanding of the fundamental nature of space.

Now that we have grappled with the precise definitions of normal and hereditarily normal spaces, you might be wondering, "What is all this for?" Are these just abstract categories invented by mathematicians for their own amusement, or do they describe something fundamental about the spaces we actually use and encounter? The answer, perhaps surprisingly, is the latter. The property of hereditary normality is not some exotic beast found only in the far corners of the topological zoo; it is a feature of many of the most foundational and well-behaved spaces in all of mathematics. This chapter is a journey to discover where this property lives, why it matters, and how it connects to the grand, intricate map of topological ideas.

Let us begin with the most familiar space of all: the real number line, $\mathbb{R}$. We walk on it, we measure with it, we graph functions on it. It feels intuitive and predictable. Does it possess this high degree of "normality"? The answer is a resounding yes. The reason is profound yet simple: the real line is a metric space. We can define the distance between any two points $x$ and $y$ as $|x-y|$.

This single fact—the existence of a distance function—unleashes a cascade of beautiful consequences. Any subspace of a metric space is, naturally, also a metric space. If you take a chunk of the real line, you can still measure distances within it. Furthermore, it is a cornerstone theorem of topology that every metric space is normal. Combining these facts gives us a powerful chain of reasoning: since every subspace of $\mathbb{R}$ is a metric space, and every metric space is normal, it follows that every subspace of $\mathbb{R}$ is normal. And that is precisely the definition of a hereditarily normal (or completely normal) space.

This is not just true for $\mathbb{R}$. It holds for the Euclidean plane $\mathbb{R}^2$, three-dimensional space $\mathbb{R}^3$, and any finite-dimensional Euclidean space $\mathbb{R}^n$. It even holds for more abstract infinite-dimensional spaces used in functional analysis, as long as they have a metric. The property of hereditary normality is, in a sense, baked into the very fabric of spaces where we can measure distance.

This "niceness" is also a topological invariant, meaning it is preserved under homeomorphisms. Imagine taking two separate intervals, say $[0, 1]$ and $[2, 3]$, and gluing the point $1$ to the point $2$. The resulting shape might seem a bit contrived, but it is topologically identical—homeomorphic—to the simple, single interval $[0, 2]$. Since $[0, 2]$ is a metric space, it is hereditarily normal, and therefore our "glued" space must be as well. The underlying structure shines through the superficial construction.

Is a metric the only way to guarantee such well-behaved separation properties? What if we dig deeper? The real line has another fundamental structure besides its metric: it is linearly ordered. Every two numbers can be compared; one is always less than the other. Could this order be the true source of its topological grace?

This question leads us to a remarkable and sweeping generalization. Consider any set with a linear order, and bestow upon it the natural "order topology" generated by open intervals. Such a space is called a ​​Linearly Ordered Topological Space (LOTS)​​. It turns out that every single LOTS is hereditarily normal. This is a theorem of immense power. It tells us that the simple, intuitive act of arranging elements in a line is sufficient to impose a very strong and orderly topological structure.

This principle allows us to venture far beyond the familiar. Consider the ​​long line​​, a famous topological object constructed by placing a copy of the interval $[0, 1)$ after every countable ordinal number. It is "long" in a way that is hard to visualize—uncountably long—and it is not metrizable. Yet, because its topology comes from a linear order, we know immediately and without further effort that it must be hereditarily normal. The power of a general theorem is that it allows us to understand the properties of even the most bizarre-looking spaces, as long as we can identify the fundamental principle that built them.

Having seen where hereditary normality thrives, a good scientist—or mathematician—must ask: where does it fail? By exploring the boundaries, we sharpen our understanding of the concept itself.

The definitions in topology are not arbitrary; they are load-bearing pillars. The definition of a completely normal space rests on the foundation of it being a $T_1$ space, where individual points are closed sets. What happens if we remove that pillar? Consider the set of integers $\mathbb{Z}$ with a strange topology where the only non-trivial open sets are "right-infinite rays" of the form $\{n, n+1, n+2, \ldots\}$. In this space, you cannot find an open set containing the number 3 that does not also contain the number 4. It is impossible to topologically isolate points from those larger than them. The space is not $T_1$, and as a direct consequence, it cannot be completely normal. The entire hierarchy of separation axioms crumbles if the foundation is weak.

Let's look at a more subtle case. The ​​Sorgenfrey line​​, $\mathbb{R}_l$, is the real numbers with a topology generated by half-open intervals $[a, b)$. It is a strange world: it is separable, but not second-countable; it is first-countable, but not metrizable. It is not a LOTS. It fails many of the "niceness" tests that the standard real line passes. And yet, it is completely normal. This is a crucial lesson: hereditary normality is a more resilient property than metrizability. It can persist in spaces that are, in other respects, quite peculiar.

Even in the friendliest of spaces, our intuition can be led astray. Let's return to the standard real line, $\mathbb{R}$. Consider the sets $A = (-\infty, 0)$ and $B = (0, \infty)$. These two sets are "separated"—the closure of $A$, which is $(-\infty, 0]$, does not touch $B$, and the closure of $B$, which is $[0, \infty)$, does not touch $A$. Since $\mathbb{R}$ is completely normal, we can find disjoint open sets containing $A$ and $B$. But now look at their closures, $\overline{A} = (-\infty, 0]$ and $\overline{B} = [0, \infty)$. Are they separated? No! They share the point $0$. They cannot be contained in disjoint open sets. This might seem like a paradox, but it is a vital clarification. The ability to separate two sets $A$ and $B$ does not, in general, grant you the ability to separate their boundaries. Nature is subtle, and definitions must be followed with exacting precision.

We have seen that hereditary normality is a strong property, but how does it relate to the other great concepts of topology, like metrizability or paracompactness? Can we use it as a stepping stone to reach even more desirable conclusions? This is one of the central games in general topology: understanding the intricate web of implications between different properties.

Let's ask a natural question: If a space is hereditarily normal and has some other nice properties, must it be metrizable? We saw that the Sorgenfrey line gives a powerful "No." The Sorgenfrey line is hereditarily normal, first-countable, and even a Baire space, but it is not metrizable. Adding these extra conditions is not enough.

What about paracompactness, an important generalization of compactness? If a space is hereditarily normal and even hereditarily separable (meaning every subspace has a countable dense subset), must it be paracompact? Again, the answer is no. While the standard counterexamples are too complex to construct here, it is a known fact that such non-paracompact spaces exist. This tells us that hereditary normality, for all its strength, does not automatically imply these other key properties. The map of topology is not a simple linear hierarchy; it is a rich, multidimensional landscape.

But the story does not end with negative results. Sometimes, adding just the right ingredient creates a beautiful chain reaction. Consider a hereditarily normal space that has a ​​point-countable base​​ (a basis for the topology where any given point belongs to only a countable number of basis elements). This condition, which might seem technical, is transformative. It is a deep result in topology that a hereditarily normal space with a point-countable base must be metrizable! The argument is a stunning cascade of theorems: these properties together imply the space is a "collectionwise normal Moore space," which in turn is known to be metrizable. And since every metric space is also "perfectly normal" (where every closed set is a countable intersection of open sets), we get that property for free as well.

This final example encapsulates the spirit of the field. We start with an abstract property, test it on familiar ground, explore its limits with strange new spaces, and finally discover its hidden connections to the other great ideas on the map. The journey from definition to application reveals that hereditary normality is not just a category, but a key player in the grand, unified story of geometric structure.