In metric spaces you learned to measure "how far apart" two points are, and then discovered something quietly radical: almost every theorem of analysis — limits, continuity, convergence — never actually uses the number d(x, y). It uses the open sets the metric builds. Continuity turned out to mean "the preimage of every open set is open." Convergence meant "the tail eventually enters every open ball around the limit." The metric was scaffolding; the open sets were the building.

So here is the daring move that opens all of modern topology. Throw the ruler away. Keep only the answer to the single question "which subsets count as open?" — and promote that answer from a consequence of a metric to the definition of the space itself. What survives is a structure vastly more general than a metric space, yet rich enough to still carry continuity, connectedness, compactness and convergence. That structure is a topological space, and it is the language in which nearly all of geometry, analysis and even parts of logic are now written.

The catch, of course, is that you cannot let any collection of subsets be "the open sets." Which collections deserve the name? The genius of the definition is that just three axioms — distilled from properties the metric open sets always had — are exactly enough. Everything else in topology is downstream of them.

What did open sets in a metric space actually do?

Recall the metric definition: U \subseteq X is open if every point of U owns a little ball B(x, r) \subseteq U sitting entirely inside — an open set is "all-interior, no skin." From that one definition, three facts fall out, and they hold in every metric space:

• The whole space and the empty set are open. Every point of X trivially has a ball inside X; and \varnothing has no points to check, so it is open by default.
• Any union of open sets is open — no matter how many. A point of the union lies in some member, inherits that member's ball, and the ball lands in the union too.
• The intersection of two open sets is open. If a point has a ball of radius r_1 inside U and a ball of radius r_2 inside V, the smaller ball of radius \min(r_1, r_2) sits inside both, hence inside U \cap V.

That is the entire genetic code of "openness." The definition of a topology is nothing but these three facts, cut loose from the metric that first proved them and hardened into law.

The three axioms of a topology

Let X be any set. A topology on X is a collection \mathcal{T} of subsets of X — a family living inside the power set \mathcal{P}(X) — subject to three rules:

The pair (X, \mathcal{T}) is a topological space, and the members of \mathcal{T} are — by decree, with no ruler in sight — called the open sets. In (T2) the index set I may be anything: finite, countable, or wildly uncountable. In (T3) it must be finite — and the asymmetry between those two axioms is the single most important thing on this page.

Why "finite"? The intersection that leaks

Axiom (T3) restricts intersections to finitely many sets, and the restriction is not bureaucratic caution — infinite intersections of open sets genuinely need not be open. Here is the canonical witness, on the real line \mathbb{R} with its usual open sets.

For each positive integer n, the interval

is open — a perfectly good open interval around 0. Now intersect all of them:

Why exactly \{0\}? The point 0 lies in every U_n, so it survives. But any other number x \ne 0 has some n large enough that \tfrac{1}{n} < |x| (that is the Archimedean property of the reals), and then x \notin U_n — so x is thrown out of the intersection. Only 0 remains.

And \{0\} is not open in the usual topology: no open interval around 0, however tiny, fits inside a set with a single point. Each individual U_n was open; their infinite intersection collapsed to a single non-open point. That single example is why the axiom says finite — and it is worth memorising, because it is the trap every beginner falls into once.

The picture behind the axioms

Before the gallery of examples, hold onto the mental image that makes "open" feel inevitable. An open set in the plane is a region with no skin: wherever you stand inside it, you have a little breathing room — a small disc around you that is still completely inside. Points on the very edge would have no breathing room (any disc around them pokes out), so an open set simply does not include its edge. Press Play to reveal the region, a point, and its private breathing-room ball.

This is exactly the metric-space definition of open — and it is the intuition the three axioms encode without ever mentioning a disc. Unions keep the breathing room (a point in the union keeps the ball it had in one piece). A finite intersection keeps it too (take the smallest of finitely many balls). An infinite intersection can shrink the available radius to zero — which is precisely how \bigcap_n(-\tfrac1n,\tfrac1n) lost its openness.

A gallery of topologies

The abstraction earns its keep the moment you see how many different, useful topologies fit the three axioms — including some that no metric could ever produce.

• The standard (Euclidean) topology on \mathbb{R}: \mathcal{T} is all sets that are unions of open intervals (a, b). This is the topology the absolute-value metric generates, and the one silently in force whenever anyone says "open subset of the reals." Its higher cousins, the Euclidean topologies on \mathbb{R}^n, are the home ground of calculus.
• The discrete topology: \mathcal{T} = \mathcal{P}(X) — every subset is open. It satisfies all three axioms trivially (the power set is closed under all unions and intersections). Here singletons \{x\} are open, so points are perfectly isolated. It is exactly the topology the discrete metric produces, and the finest topology any set can carry.
• The indiscrete (trivial) topology: \mathcal{T} = \{\varnothing, X\} — the bare minimum (T1) forces, and nothing more. Only two open sets exist. Points cannot be separated at all; if X has two or more points, no metric generates this topology. It is the coarsest topology any set can carry.
• The cofinite topology: declare a set open iff it is \varnothing or its complement is finite. Check the axioms and you meet a lovely reason for "finite": a union of cofinite sets has complement equal to the intersection of the (finite) complements, still finite; and a finite intersection of cofinite sets has complement equal to the union of finitely many finite sets, still finite — but an infinite intersection could make the complement infinite and break openness. On an infinite X this topology comes from no metric, yet it is central to algebraic geometry (it is the Zariski topology on a line).
• The Sierpiński space: on the two-point set X = \{0, 1\}, take \mathcal{T} = \{\varnothing, \{1\}, \{0,1\}\}. Three open sets; note \{1\} is open but \{0\} is not. This tiny, lopsided space is the smallest one that is not indiscrete or discrete, and it is a surprisingly deep test-object — it is, in a precise sense, the space that "detects openness" for every other topological space.

Worked example: is this collection a topology?

Deciding whether a given family \mathcal{T} is a topology is a three-item checklist: (T1) both \varnothing and X present, (T2) closed under unions, (T3) closed under finite intersections. Let us drill it on X = \{a, b, c\}.

Candidate 1. \mathcal{T} = \big\{\varnothing,\; \{a\},\; \{b\},\; \{a, b, c\}\big\}. (T1) holds — \varnothing and X are both there. But test (T2) on \{a\} \cup \{b\} = \{a, b\}: that set is not in the family. Not a topology — the union axiom fails.

Candidate 2. \mathcal{T} = \big\{\varnothing,\; \{a\},\; \{a, b\},\; \{a, b, c\}\big\}. Run the checklist. (T1): \varnothing, X present — good. (T2): the only non-trivial unions are \{a\} \cup \{a,b\} = \{a,b\} and unions with X or \varnothing, all already in the family — good. (T3): \{a\} \cap \{a,b\} = \{a\}, present; every other pairwise intersection is \varnothing, \{a\}, \{a,b\}, or a set itself — all present. Every box ticks: this is a topology (a "chain" topology, in fact).

Candidate 3. \mathcal{T} = \big\{\varnothing,\; \{a, b\},\; \{b, c\},\; \{a, b, c\}\big\}. (T1) fine. (T2): \{a,b\} \cup \{b,c\} = \{a,b,c\} = X, present — unions are fine. But (T3): \{a,b\} \cap \{b,c\} = \{b\}, and \{b\} is not in the family. Not a topology — the finite-intersection axiom fails. (Notice how (T2) and (T3) probe opposite operations; a family can pass one and flunk the other.)

The moral: to reject a candidate you need only one broken axiom and one explicit witness set; to accept one you must confirm all three, which on a finite space means grinding through the finitely many unions and pairwise intersections.

Coarser, finer, and the fact that a set has many topologies

A single set almost never carries just one topology — it carries a whole lattice of them, and they can be compared. Given two topologies on the same X, we say \mathcal{T}_1 is coarser than \mathcal{T}_2 (equivalently \mathcal{T}_2 is finer) when

that is, every set open in the coarser one is also open in the finer one — the finer topology simply has more open sets. Think of it as resolution: a finer topology sees more distinctions, can separate more points, makes more functions continuous out of the space and fewer continuous into it.

These comparisons have universal endpoints. For any set X:

The indiscrete topology is the coarsest possible (nested inside every topology); the discrete topology is the finest possible (it contains every topology). Any topology you build on X lives somewhere in that sandwich. On a three-point set there are exactly 9 distinct topologies (up to relabelling); on larger sets the count explodes — the number of topologies on an n-point set is a famously hard combinatorial quantity with no closed form. One set, a teeming zoo of geometries.