What a covering space is, and why we want one
Guide 2 of this rung built the fundamental group pi_1(X, x_0) out of loops, and crowned the construction with the computation pi_1(S^1) = Z via the winding-number argument. Look back at how that proof actually worked: we did not stay on the circle, we climbed onto the real line R sitting above it through the map p(t) = (cos 2 pi t, sin 2 pi t). A loop on the circle became a path on R, and the integer it landed on was the winding number. That maneuver — lift the downstairs loops to upstairs paths where they can no longer secretly close up — is the whole engine of this guide, promoted from one example to a general theory.
A covering space of X is a space E with a continuous surjection p: E -> X such that every point x in X has an open neighborhood U whose preimage p^{-1}(U) is a disjoint union of open sets, each mapped homeomorphically onto U by p. The picturesque phrase is that U is evenly covered: upstairs it appears as a stack of identical 'pancakes', and p flattens each pancake perfectly onto U. The local model is U x (discrete set); a covering space looks locally like the base times a scattering of dust, even though globally it can be braided together in surprising ways. The map R -> S^1 above is the prototype: each little arc of the circle has preimage a disjoint family of intervals on R, one per integer.
Lifting: the one property that does all the work
Everything a covering space can do for topology flows from two lifting properties, and they deserve to be stated cleanly. Path lifting: given a path gamma in X starting at x_0, and any choice of starting point e_0 upstairs with p(e_0) = x_0, there is a UNIQUE path gamma-tilde in E starting at e_0 with p(gamma-tilde) = gamma. Homotopy lifting: a homotopy of paths downstairs lifts to a homotopy upstairs, again uniquely once the start is pinned. The proofs are pure point-set topology — chop the path into small pieces each landing in an evenly covered U, lift one pancake at a time, and uniqueness glues the pieces with no freedom of choice once e_0 is fixed.
Watch what these two facts immediately buy. A loop gamma at x_0 lifts to a path gamma-tilde starting at e_0; its endpoint gamma-tilde(1) lies in the fiber p^{-1}(x_0), and it need NOT equal e_0. That endpoint depends only on the homotopy class of gamma (by homotopy lifting), so we get a well-defined action of pi_1(X, x_0) on the fiber: the class [gamma] sends e_0 to gamma-tilde(1). This is the monodromy action, and it is the secret heart of the subject. For R -> S^1 it is precisely 'add the winding number': the loop that wraps once carries the integer 0 to the integer 1, which is why pi_1(S^1) = Z. The whole Galois dictionary below is just this action, organized.
There is also a one-line consequence worth banking: p* : pi_1(E, e_0) -> pi_1(X, x_0) is INJECTIVE. A loop upstairs whose image is null-homotopic downstairs lifts that null-homotopy back up (homotopy lifting), so the original loop was already null-homotopic upstairs. Hence pi_1(E) embeds as a subgroup of pi_1(X). The image p*(pi_1(E, e_0)) is the characteristic subgroup of the cover — it consists of exactly those downstairs loops that lift to LOOPS (closed paths) rather than open paths. Two covers are 'the same' precisely when they have conjugate characteristic subgroups. Hold onto this subgroup; it is the index card in the card catalog we are about to build.
The universal cover and the lifting criterion
Among all covers of X there is a maximal one, and it is the keystone. The universal cover X-tilde is a covering space that is itself simply connected: pi_1(X-tilde) = 1. For R -> S^1 the line R, being contractible, is exactly the universal cover of the circle. The name 'universal' is apt because X-tilde covers EVERY other connected cover of X — it sits at the very top of the tower, with all loops of X fully unwound and none surviving upstairs. It exists whenever X is reasonable: connected, locally path-connected, and semilocally simply connected (every point has a neighborhood whose loops already die in X). That last condition is mild but real — the Hawaiian earring, an infinite shrinking nest of circles, fails it and has no universal cover.
How do you actually build X-tilde? With loops themselves. Fix a basepoint x_0 and let X-tilde be the set of homotopy classes of PATHS in X starting at x_0 (homotopies fixing both endpoints); the projection sends a class [gamma] to its endpoint gamma(1). A point upstairs is literally 'a point of X together with a remembered way of getting there'. Two routes to the same destination are different points upstairs exactly when the loop formed by going out one way and back the other is nontrivial in pi_1 — which is why the universal cover unwinds the loops. The fiber over x_0 is then pi_1(X, x_0) itself, so the number of sheets of the universal cover equals the order of the fundamental group.
The Galois correspondence of covering spaces
Now the payoff. Fix a nice base X with universal cover X-tilde and group G = pi_1(X, x_0). There is a perfect dictionary — order-reversing and exhaustive — between connected covering spaces of X (up to isomorphism over X) and subgroups of G (up to conjugacy). Up top sits X-tilde, matched with the trivial subgroup {1}; at the bottom sits X itself, the one-sheeted cover, matched with all of G. A cover with characteristic subgroup H has degree equal to the index [G : H], its fiber is the coset space G/H, and the monodromy action is just left multiplication on cosets. Larger subgroups give smaller covers; this inverted matching is exactly the shape of the Galois correspondence between field extensions and subgroups of the Galois group.
COVERING SPACES of X SUBGROUPS of G = pi_1(X, x_0)
------------------- -----------------------------
universal cover X-tilde <----> trivial subgroup {1}
| |
intermediate cover E_H <----> subgroup H <= G
| |
the base itself X <----> whole group G
degree of the cover = [G : H] (index)
fiber over x_0 = G / H (left cosets)
monodromy = G acting on G/H by left multiplication
NORMAL (regular) cover <----> NORMAL subgroup H |> G
deck group Deck(E/X) = G / HThis is not a loose analogy dressed up for slogans — it is a literal equivalence of categories. The category of covering spaces of X is equivalent to the category of G-sets, with the universal cover playing the role of the regular representation G acting on itself. Honesty check: the correspondence needs X to be connected, locally path-connected, and semilocally simply connected, the same hypotheses that guaranteed X-tilde. Drop semilocal simple connectivity and the dictionary breaks at the top because there is no universal cover to anchor it. And the matching is up to CONJUGACY of subgroups, not equality — conjugate subgroups give isomorphic covers because they differ only by which point of the fiber you called the basepoint. Forgetting the conjugacy clause is the single most common error here.
Deck transformations: the symmetries of a cover
Every cover carries its own symmetry group, and it completes the Galois picture. A deck transformation of p: E -> X is a homeomorphism phi: E -> E that respects the projection, p(phi(e)) = p(e) — it permutes the sheets while leaving everything downstairs fixed. For R -> S^1 the deck transformations are exactly the integer translations t -> t + n, so Deck(R/S^1) = Z, which is pi_1(S^1) all over again. That coincidence is no accident: for the universal cover, the deck group IS the fundamental group. The group G acts on X-tilde by deck transformations, freely (no nontrivial deck map fixes a point) and properly, and the quotient X-tilde / G recovers X. So you can read pi_1(X) off geometrically as 'the symmetries of the universal cover'.
For a general cover E_H corresponding to subgroup H, the deck group is the quotient N(H)/H, where N(H) is the normalizer of H in G. The cleanest case is when H is NORMAL: then N(H) = G, the deck group is the full quotient G/H, and the cover is called regular (or normal, or Galois). A regular cover is the topological twin of a Galois field extension — its deck group acts transitively on each fiber, permuting the sheets as freely as the Galois group permutes the roots of a polynomial. When H is not normal the cover is still perfectly good, but its symmetry group N(H)/H is smaller than the degree [G : H]; the sheets cannot all be swapped into one another, the topological echo of a non-normal field extension.
- Start with a space X (nice: connected, locally path-connected, semilocally simply connected) and compute G = pi_1(X, x_0); for a wedge of two circles, say, G is the free group on two generators a, b.
- Pick a subgroup H <= G. Its index [G : H] is the degree of the cover you will get, and the coset space G/H is the fiber on which G acts by left multiplication (the monodromy).
- Realize the cover as X-tilde / H: take the universal cover (here the infinite 4-valent tree, the Cayley graph of the free group) and quotient by H acting through deck transformations; the result is a graph mapping down to the wedge.
- Read off the symmetry: the deck group is N(H)/H. If H is normal the cover is regular with deck group G/H acting transitively on the fiber; if not, fewer sheets can be exchanged — the topological face of a non-Galois extension.
What this machinery is good for
Covering spaces are not a curiosity; they are a working tool that turns hard topology into manageable group theory and back. The flagship application is a theorem of pure algebra proved entirely by topology: every subgroup of a free group is free (the Nielsen-Schreier theorem). The proof is almost a joke once you have the dictionary — a free group is pi_1 of a wedge of circles, that wedge is a graph, every covering space of a graph is again a graph, and pi_1 of any connected graph is free. So a subgroup, being pi_1 of a cover, is automatically free. The Schreier index formula for the rank of the subgroup falls out as the Euler-characteristic bookkeeping of the covering graph. This is geometric group theory in miniature: study a group by studying a space it acts on.
The same dictionary powers the next guide. Knowing the covering spaces of X is knowing the subgroups of pi_1(X), so the Seifert-van Kampen theorem — which computes pi_1 of a space glued from pieces — and covering theory are two views of one object. In differential geometry the universal cover is how you tame multiply-connected manifolds: a flat torus is R^2 modulo a lattice acting by deck transformations, and the lattice IS its fundamental group; a space form is the round sphere, Euclidean space, or hyperbolic space quotiented by a group of deck isometries. Even rigidity phenomena like Mostow rigidity are statements about how little freedom the deck-group action of pi_1 has on a hyperbolic universal cover.
One honest caveat to carry forward. The whole theory presumes the base is well-behaved; for wild spaces (the Hawaiian earring and its kin) there is no universal cover and the dictionary simply does not apply — do not reach for it there. And the analogy with Galois theory, beautiful as it is, is an analogy of structure, not a theorem you can cite to transfer results blindly: there is no Galois group of a polynomial hiding inside a topological space, only a fundamental group that happens to organize covers the same way Galois groups organize extensions. Used within its hypotheses, though, the correspondence is one of the most reliable bridges in mathematics — algebra and geometry reading each other's minds.