The fundamental group is easy to define and hard to compute. The circle is the first nontrivial case, and the computation rests on a single idea: the real line winds around the circle like a helix, and loops in the circle can be unwound uniquely into paths in the line. This chapter develops that idea for a general covering map and then applies it to prove .
Motivation
A loop in going once around the origin appears to be impossible to shrink, but a proof must exclude every conceivable continuous deformation. A good strategy is to compare with a simpler space that maps onto the circle. The map , , wraps the line around the circle infinitely often. A loop in lifts to a path in , and the endpoint of the lift is an integer that counts how many times the loop goes around. What is needed is a guarantee that the lift exists, is unique once its starting point is fixed, and does not change under path homotopy. The three statements are the path lifting property, the uniqueness of lifts, and the homotopy lifting property. They hold for covering maps, and the hypothesis that the map is a covering, rather than merely a local homeomorphism, is essential (see the local-homeomorphism example below).
Definitions
Definition 6.1.1
(Evenly covered set and covering map)
Let be a continuous surjection. An open set is evenly covered by if is a disjoint union of open sets (the sheets over ) such that for each the restriction is a homeomorphism. The map is a covering map if every point of has an evenly covered open neighborhood. The space is the covering space, and is the fiber over .
Definition 6.1.2
(Lift)
If is continuous, a lift of is a continuous with .
Definition 6.1.3
(Lebesgue number)
If is an open cover of a metric space , a number is a Lebesgue number of the cover if every subset of of diameter less than lies in a single member of .
Definition 6.1.4
(The exponential covering)
, . For let be the loop based at .
Definition 6.1.5
(Degree of a loop)
For a loop at in , let be the lift with (path lifting). Its endpoint lies in and is called the degree of .
Concept lensAnalytic
Degree as a winding integral
For a continuously differentiable loop at , the degree is an integral. Locally , and each sheet inverse is a differentiable branch of , so is continuously differentiable; differentiating gives . Hence
the winding number about from complex analysis. The lifting definition needs no differentiability at all.
Theorem/proposition
Theorem 6.1.6
(Lebesgue number lemma)
Lebesgue number lemma. Let be a compact metric space and an open cover. Then has a Lebesgue number. This applies to and with the Euclidean metric.
Theorem 6.1.7
(Path lifting)
Path lifting. Let be a covering map, a path, and with . Then there is exactly one lift of with .
Theorem 6.1.8
(Homotopy lifting)
Homotopy lifting. Let be a covering map, continuous, and with . Then there is exactly one lift with . If is a path homotopy (that is, and are independent of ), then so is .
Corollary (endpoints of lifts). Let be a covering map and let be path-homotopic paths in starting at . If are the lifts starting at , then and .
Theorem 6.1.9
(Fundamental group of the circle)
Fundamental group of the circle. The map , , is a well-defined group isomorphism. The group is generated by , and .
Theorem 6.1.10
(Injectivity for coverings)
Injectivity for coverings. If is a covering and , then is injective.
Theorem 6.1.11
(No retraction onto the circle)
No retraction onto the circle. There is no continuous map from the closed unit disk with for all .
Proof sketch or proof idea
All proofs are given in full.
Proof
(The Lebesgue number lemma from distances to complements)
If is empty, any positive number works. Otherwise, by compactness choose a finite subcover . If some , any works. Otherwise define , the distance to the closed nonempty set ; each is continuous, and for because is closed. Let , continuous and positive everywhere (each point lies in some ). By the extreme value theorem on the compact space (the compactness chapter), attains a minimum . If has diameter and , there is with , so the ball lies in and contains .
Counterexample mode
Without compactness a cover can have no Lebesgue number
Claim: every open cover of a metric space has a Lebesgue number. Take with the Euclidean metric and the open cover , ; every lies in once . Given any , let and . Then , but for every , because the point belongs to and is less than . So no works and the claim is false. In the proof above, the step that fails is the passage to a finite subcover: without it, is not defined, and the infimum of the distances to the complements is . The Lebesgue number lemma needs compact, which is why it is applied only to and .
Proof
(Path lifting by subdivision and connectedness)
Existence. Cover by evenly covered open sets . Then is an open cover of . Let be a Lebesgue number and choose with ; set . Each lies in an evenly covered . Define and extend inductively. Suppose is defined and continuous on with . Since , the point lies in , hence in one sheet . Define on . At this agrees with the earlier definition because and . By the pasting lemma is continuous on , and clearly there.
Uniqueness. Let be lifts with the same starting point and . Then . is open: if , let be an evenly covered neighborhood of and the sheet over containing . By continuity there is a neighborhood of in with . For , and is injective, so and . is closed: if , then both lie over . Let be as before. They cannot lie in the same sheet (because is injective), so they lie in distinct, disjoint sheets . By continuity there is a neighborhood of with and , so . Since is connected (the connectedness chapter), .
Proof
(Homotopy lifting square by square)
Existence. Cover by evenly covered sets; is an open cover of the compact metric space with a Lebesgue number . Choose with and set . The squares () each map under into some evenly covered . Order the squares by increasing and, for fixed , by increasing . Define successively on .
For : the point lies over , hence in one sheet ; set on . Suppose is defined and continuous on the union of the earlier squares, with . The intersection of with the union of earlier squares is the union of the left and bottom edges of (only the bottom edge if , only the left edge if ): the square meets in the left edge, in the bottom edge, and other earlier squares meet in at most a corner already on these edges. The set is connected (an "L" shape, a segment, or a point), so the connected set lies in and hence in a single sheet , the sheets being disjoint and open. Define on . On this agrees with the existing , since there and . By the pasting lemma (finitely many closed sets) is continuous. This completes the induction, giving a lift on all of .
Uniqueness. Let be lifts agreeing at . For let . Then and are lifts of the path starting at , so they are equal by path lifting; evaluating at gives .
Path homotopies. Suppose for all . The paths and are both lifts of the constant path at starting at , hence equal by path lifting. In the same way, and the constant path at are lifts of with the same starting point, so is constant.
Proof X-ray
Local inverses, global compatibility
Each evenly covered neighborhood supplies an inverse only on one sheet. Path lifting starts by choosing the sheet containing the prescribed initial point, then uses each newly fixed endpoint to choose the next sheet. Compactness provides finitely many intervals, and the pasting lemma turns these local formulas into one continuous path.
For a square homotopy, the overlap with earlier squares is connected. Its lifted image must stay in one sheet, so a single inverse formula agrees along the whole overlap. That connected-overlap check is the additional compatibility step; agreement at one corner alone would not directly prove agreement on an entire edge. Finally, fixed endpoints lift to constant endpoint paths by uniqueness. These are exactly the ingredients that make the integer endpoint of a circle lift invariant under based homotopy.
Proof
(The endpoint corollary from the lifted square)
Let be a path homotopy from to and its lift with . The path lifts and starts at , so it equals by path lifting. The edge is constant by homotopy lifting, so lifts and starts at , hence equals . The edge is constant, so , and is a path homotopy from to .
Proof
(The fundamental group of the circle through the degree)
First, is a covering map. Given , the set is open and , a disjoint union of open intervals. On the interval with index , is a bijection onto whose inverse is , where is the principal argument, continuous on minus the closed negative real axis. So each interval is a sheet and is a covering.
Well defined. By the endpoint corollary, path-homotopic loops at have lifts from with the same endpoint, so depends only on .
Homomorphism. Let be loops at with lifts from , ending at and . The path is a lift of (since ) starting at , ending at . So is defined, is a lift of starting at (check on each half), and ends at . Hence .
Surjective. , so its lift from is and .
Injective. Suppose , so is a loop at in . Since is convex, (convex sets are simply connected), say by . Then is a path homotopy from to the constant loop at . So is trivial. Hence is an isomorphism, and .
Proof X-ray
Where each property of the degree comes from
Each property of has a different source. Well-definedness is the endpoint corollary, so it rests entirely on homotopy lifting. Additivity uses a symmetry of : translation by an integer satisfies , so the lift of can be restarted at the endpoint of . Surjectivity is an explicit lift. Injectivity is the only step that uses the topology of itself: a degree-zero loop lifts to a loop in , convexity contracts it, and carries the contraction down. If the covering space had a loop that cannot be contracted, exactly this step would fail.
Proof
(Injectivity for coverings by lifting the null-homotopy)
Let be a loop at with trivial; thus . The lift of starting at is (by uniqueness, since is a lift), and the lift of the constant path starting at is the constant path . By the endpoint corollary, .
Proof
(The no-retraction proposition from the fundamental group)
Suppose exists and let be the inclusion; then . By the unit on paths and fundamental groups (retractions give injective inclusion maps, using functoriality), is injective. But is trivial because is convex (convex sets are simply connected), while by the theorem on the fundamental group of the circle. A homomorphism from to the trivial group is not injective.
Worked examples
Worked example 6.1.12
(Lifting a loop that winds twice)
Lifting a loop that winds twice. Let . The path is continuous, satisfies and , so it is the lift from by uniqueness. Its endpoint is . Hence and . By the endpoint corollary and the computation of the fundamental group of the circle, is path-homotopic to and not to or . (Explicitly, the straight-line homotopy projects to a path homotopy from to .)
Worked example 6.1.13
(The circle is not a retract of the disk)
The circle is not a retract of the disk, and the punctured plane has a nontrivial loop. The no-retraction proposition was proved above. Similarly the map , , is a retraction onto , so is injective (retractions give injective inclusion maps). Therefore the loop , regarded as a loop in the punctured plane, is not path-homotopic to a constant loop. This settles the claim made in the unit on paths and fundamental groups about the semicircles. With and one computes exactly (on each half of the parameter interval both sides equal ). If in the punctured plane, the argument of the first exercise in that unit (its first half works verbatim for paths with common endpoints) would make trivial there, which we have just excluded. So no path homotopy from to exists, and the failure of the straight-line formula was not a defect of that formula.
Worked example 6.1.14
(A local homeomorphism that is not a covering)
A local homeomorphism that is not a covering. Let , . It is continuous, surjective and a local homeomorphism (it is the restriction of to an open subset of , and is a local homeomorphism by the sheets above). It is not a covering. Suppose were an evenly covered neighborhood of . Then every fiber , , would have the same cardinality as the index set of the sheets, since each sheet contains exactly one point of the fiber. But has two points, while for small the fiber over is with three points. This contradiction shows no neighborhood of is evenly covered. Consequently path lifting cannot be assumed for local homeomorphisms. Take the path , , which starts at and ends at . A lift with would satisfy , so is a continuous integer-valued function on which vanishes at ; hence . But then . So has no lift to starting at .
Worked example 6.1.15
(Degree of maps of the circle)
Degree of maps of the circle. For let , , with . Then , so : under the identification , is multiplication by . In particular complex conjugation acts by , reversing orientation, and (constant) acts by . A homomorphism is determined by its value on , so every based continuous self-map of induces multiplication by some integer, the degree of the map. Maps of different degrees are not homotopic through based self-maps: composing such a homotopy with would equate loops of different degrees.
Common mistakes
Common mistake
(Treating the lift of a loop as a loop)
The lift of starting at is , which ends at . Lifts of loops are in general only paths, whose two endpoints may be different points of the fiber; their endpoints carry the information.
Common mistake
(Forgetting to fix the starting point)
Lifts are unique only once is prescribed. Different starting points in the fiber give different lifts (translated by integers in the exponential covering).
Common mistake
(Replacing “covering” by “local homeomorphism”)
The local-homeomorphism example shows that a surjective local homeomorphism need not have path lifts with a prescribed starting point. The even covering condition is essential.
- Using the heuristic "a circle has one hole" as a proof. The computation of requires the lifting argument; intuition does not exclude exotic deformations.
- Assuming the exponential covering is a homeomorphism on its whole domain. It is a homeomorphism only on each sheet; on it is not injective.
Summary
A covering map has evenly covered neighborhoods, and for such maps paths lift uniquely once a starting point is chosen, and homotopies lift uniquely too, with path homotopies lifting to path homotopies. For the exponential map the endpoint of the lift of a loop at is an integer that is homotopy invariant, additive under concatenation, and takes every integer value. This gives , generated by the loop that goes around once. Consequences include the injectivity of for coverings, the non-existence of a retraction of the disk onto its boundary, and the notion of degree for self-maps of the circle.
Exercises
Quick check
(Find the degree of the loop at and describe a path homotopy to a standard loop .)
Solution · Answer
The path is continuous with and , so it is the lift from . Its endpoint is , so . The homotopy is a path homotopy in from to the identity path, since both start at and end at (check: , ). Then is a path homotopy from to .
Quick check
(Prove that for a covering map and , the fiber is a discrete subspace of .)
Solution · Answer
Let , an evenly covered neighborhood, . For each pick the sheet . Since is injective, . Thus is open in the subspace , for every , which means the subspace topology is discrete.
Quick check
(Use the no-retraction proposition to prove the Brouwer fixed point theorem for the closed disk: every continuous has a fixed point.)
Solution
Suppose has no fixed point. For put . Let , where is the larger root of , namely
The discriminant is nonnegative since and , and because the square root is at least . Since is continuous and nonzero, and are continuous, and by construction. If then , so and . Thus is a retraction, contradicting the no-retraction proposition.
Quick check
(Compute on for .)
Solution · Answer
The map satisfies . So , and is multiplication by on .
Quick check
(Let be a covering and suppose is simply connected and path-connected. Show that for a loop at with lift from , the loop is null-homotopic if and only if is a loop.)
Solution · Answer
Let be a loop at and its lift from . If is null-homotopic, the endpoint corollary (with and the constant path) gives , because the lift of the constant path is constant; thus is a loop. Conversely, if is a loop at , then since is simply connected, in (it is a loop whose class in the trivial group is trivial). Applying gives .