Connectedness and compactness detect coarse features of a space. To detect holes one needs a finer invariant. The fundamental group records which loops in a space can be continuously deformed into one another, and it turns the intuitive statement "the circle has a hole" into a precise algebraic one. This chapter constructs the group and proves that it is functorial and, up to isomorphism, independent of the basepoint in a path-connected space. The unit on covering maps, lifting and the circle computes it for the circle.
Motivation
Two loops in the plane based at the same point can always be shrunk to the base point or deformed into one another by sliding straight-line segments. In the punctured plane a loop that goes once around the origin cannot be shrunk to a point without crossing the origin. To make this precise we need three things: a notion of deforming paths while keeping their endpoints fixed, a way of composing loops, and a check that the composition is well defined on deformation classes and has the properties of a group. The last point is subtle, because composition of paths is not associative on the nose: and traverse the three pieces at different speeds. Associativity holds only up to a reparametrization, which we spell out explicitly.
Definitions
Throughout .
Definition 5.1.1
(Path homotopy)
Let be paths with and . A path homotopy from to is a continuous map with
We write when such an exists.
Definition 5.1.2
(Concatenation, constant path and reverse path)
If define by for and for . This is continuous by the pasting lemma, since both formulas give at . The constant path at is . The reverse of is .
Definition 5.1.3
(Fundamental group and induced homomorphism)
A loop at is a path from to . The fundamental group is the set of path-homotopy classes of loops at with the operation .
For a continuous (so ) define by .
Concept lensGeometric
Loops as maps of a circle
Gluing the ends of turns a loop at into a map through , and a path homotopy of loops into a homotopy of such maps keeping at . So is the set of based homotopy classes of maps from the circle. In this picture is trivial exactly when the circle map extends continuously to the closed disk: a null-homotopy fills the circle radially.
Definition 5.1.4
(Change of basepoint and simple connectivity)
If is a path from to , define by , where products of three paths are bracketed in either way (the class does not depend on the bracketing, by the groupoid properties proved below).
A space is simply connected if it is path-connected and is trivial.
Theorem/proposition
Proposition 5.1.5
(Path homotopy is an equivalence relation)
Path homotopy is an equivalence relation on the set of paths from to .
Proposition 5.1.6
(Groupoid properties of concatenation)
Let be paths with and , and let , .
- If and then .
- (Reparametrization) If is continuous with and , then .
- (Associativity) ; more precisely for an explicit piecewise-linear .
- (Identities) .
- (Inverses) and .
In particular is a group, with identity and inverse .
Proposition 5.1.7
(Functoriality of induced homomorphisms)
The map is a well-defined group homomorphism. Moreover and . Hence a homeomorphism induces an isomorphism.
Proposition 5.1.8
(Change of basepoint is an isomorphism)
For a path from to the map is a group isomorphism with inverse . If is a loop at , then is the inner automorphism of .
Proposition 5.1.9
(Convex sets are simply connected)
Let be nonempty and convex. Any two paths in with the same endpoints are path-homotopic. In particular is trivial for every , so is simply connected.
Proof sketch or proof idea
All proofs are given in full. Throughout we use the pasting lemma: if (or ) is a union of two closed sets on each of which a function is continuous, and the two functions agree on the overlap, then the glued function is continuous.
Proof
(Reflexivity, symmetry and transitivity by pasting)
Reflexive: . Symmetric: if is a path homotopy from to then is one from to . Transitive: if goes from to and from to , set for and for . At both equal , so is continuous by the pasting lemma, and it has the correct boundary values.
Proof
(Explicit homotopies for the groupoid laws)
(1) Let and . Define for and for . At the two formulas give , so is continuous. It restricts to at and to at , and its sides are constant.
(2) Put . This is continuous, since as a convex combination of points of . At it is , at it is , and , because , .
(3) Define by
It is continuous (the values at are both , at both ), with and . Recall equals on , on , on , while equals on , on , on . For we get . For we get . For we get . So , and (2) gives .
(4) We have with for and for : indeed on the left side is the constant . Similarly with for and for . Both are continuous and fix and , so (2) applies.
(5) Define for and for . At both give , so is continuous. At : for , and for , , which is . At it is the constant . At and it is for all . So is a path homotopy from to . Applying this to in place of (note ) gives .
Statements (1) to (5) show that multiplication of classes of loops is well defined, associative, has the identity , and every has the inverse .
Proof X-ray
One reparametrization lemma does most of the work
The five parts rest on two constructions. Part (2) is the engine. Its homotopy interpolates in the parameter interval , which is convex, and never in . That is why the groupoid laws need no hypothesis on the space. The conditions and are used exactly once, to keep the two endpoints fixed for every .
Parts (3) and (4) are then reduced to checking an identity of functions: each product is written as the other path composed with an explicit fixing and . Part (5) cannot be handled this way. Writing forces on and on , so and the lemma does not apply; besides, the constant loop is not, in general, a reparametrization of . The homotopy used instead is : at time the loop runs along only up to the parameter and then returns. Finally, part (1) is what turns (3)–(5), which are statements about representatives, into the group axioms for classes.
Proof
(Composing homotopies with a map)
If is a path homotopy from to then is a path homotopy from to , so is well defined. Since (compare the defining formulas), . Functoriality is and . If is a homeomorphism, also induces a homomorphism and , similarly the other way.
Proof
(Conjugating by the basepoint path)
By the groupoid properties the class is well defined, and it is a loop at since ends at and starts at and ends at . Homomorphism: using ,
Inverse: and symmetrically in the other order. If is a loop then and .
Proof
(Straight-line interpolation)
Let be paths in with the same endpoints. Define . By convexity ; it is continuous; , , and , are independent of . Taking shows each loop is null-homotopic. Also, any two points of the nonempty convex set are joined by their straight segment, so the set is path-connected as required.
Worked examples
Worked example 5.1.10
(Associativity fails on the nose)
In let , , . Then , while . So the two products are different functions. By part (3) of the groupoid properties, they differ by the reparametrization : , and , as required.
Worked example 5.1.11
(A straight-line homotopy and a reparametrization)
In let and , both from to . Then is a path homotopy, as in the convex-set proposition. Also and in differ by the reparametrization , and the homotopy of part (2) of the groupoid properties is .
Worked example 5.1.12
(No canonical identification of fundamental groups)
If is a loop at , the change-of-basepoint proposition says that is conjugation by . Thus in a path-connected space the groups and are isomorphic, but the isomorphism depends on the choice of path up to conjugation, and is canonical only when the fundamental group is abelian (proved in the exercises below). One writes only when this ambiguity does not matter.
Worked example 5.1.13
(Basepoint matters without path-connectedness)
Let be the disjoint union of an isolated point and a circle. A loop at the isolated point has connected image containing , and is open and closed in , so the image is and the loop is constant. Hence is trivial. By the unit on lifting and the circle, however, (where ), which is not trivial. So the groups at different basepoints can differ when is not path-connected; the change-of-basepoint proposition needs a path joining the two basepoints.
Worked example 5.1.14
(Retractions give injective inclusion maps)
Let with inclusion , and suppose is continuous with (a retraction). Then on by functoriality, so is injective and is surjective. This is the template for several arguments in the next chapter.
Common mistakes
Common mistake
(Treating concatenation as associative on the nose)
The paths and are different functions in general; the first worked example exhibits a parameter where they differ. Associativity holds only after passing to path-homotopy classes. More generally, the elements of are equivalence classes, and an equality of classes asserts that a path homotopy exists, not that two functions agree.
Common mistake
(Letting the endpoints move during the homotopy)
If endpoints were allowed to move, every path would be homotopic to the constant path at via , and the relation would detect nothing about paths. The definition of uses homotopies relative to the endpoints, and each homotopy in the proofs above must be checked to satisfy and .
Common mistake
(Calling the change-of-basepoint isomorphism canonical)
In a path-connected space all the groups are isomorphic, but the isomorphism depends on the path ; two paths give isomorphisms differing by an inner automorphism. Writing without a basepoint is harmless only when that ambiguity does not affect the statement.
A related slip is to use straight-line interpolation outside a convex set. The convex-set proposition needs every segment to stay in ; for a non-convex set this formula can leave the set, as the last exercise shows for the punctured plane.
Summary
Path homotopy relative to endpoints is an equivalence relation compatible with concatenation. Concatenation is associative, has identities and inverses up to path homotopy, so classes of loops at a point form the group . A continuous map induces a homomorphism, functorially; homeomorphic spaces have isomorphic fundamental groups. A change of basepoint along a path induces an isomorphism, which for loops is conjugation. Nonempty convex subsets of are simply connected via straight-line homotopies.
Exercises
Quick check
(For loops at prove that if and only if .)
Solution · Answer
If then by parts (1) and (5) of the groupoid properties. Conversely if , then , using .
Quick check
(Suppose is simply connected. Prove that any two paths in with the same endpoints are path-homotopic.)
Solution
Let be paths from to . The path is a loop at , so by hypothesis. Then
by parts (4), (5), (3), (1) and (4) of the groupoid properties.
Quick check
(Suppose is path-connected and is abelian. Prove that the isomorphism does not depend on the choice of path from to .)
Solution
Let be paths from to and a loop at . Since ,
Set ; then and the displayed relation reads . The group is isomorphic to via , hence abelian, so and .
Quick check
(Let be a retract of with . Prove that is injective.)
Solution · Answer
Let be a retraction, so . Then by functoriality (with ). A homomorphism with a left inverse is injective: if then .
Quick check
(Let and be the upper and lower unit semicircles from to . Write an explicit path homotopy from to in , and show that the same formula does not give a path homotopy in the punctured plane .)
Solution · Answer
The plane is convex, so the convex-set proposition applies with . This is continuous, equals at and at , and , for all . In the punctured plane the formula does not define a map into : at , we get . This shows only that the straight-line formula fails, not that no path homotopy exists; in fact and are not path-homotopic in the punctured plane, because winds once around the origin, which the unit on lifting and the circle will show is not null-homotopic.