Motivation
A homeomorphism preserves every topological property, but it is often more structure than a loop calculation needs. The punctured plane has two-dimensional neighborhoods and the circle has one-dimensional neighborhoods, yet every point of the punctured plane can be moved radially to the circle without passing through the missing origin. This deformation suggests that the two spaces should have the same fundamental group even though they are not homeomorphic.
A second useful construction combines the circle calculation with products. A loop on a torus has two coordinate loops, one in each circle factor. If the equivalence relation and multiplication also operate coordinate by coordinate, the fundamental group should be a product of groups. Both ideas become precise once we distinguish homotopy of maps from homotopy of paths relative to endpoints.
The distinction about basepoints remains essential. A deformation may move the chosen point, so its induced maps can have different target groups. The correct comparison then includes transport along the path traced by that point. Ignoring this can make an otherwise correct geometric argument have ill-typed algebraic equations.
Definitions
Definition 5.2.1
(Homotopy, homotopy equivalence and contractibility)
A homotopy between continuous maps is a continuous map with and . Write . A homotopy is relative to if for and all . Homotopy of maps does not otherwise fix points.
A continuous map is a homotopy equivalence if there is a continuous with and . Such a is a homotopy inverse. A nonempty space is contractible if its identity is homotopic to a constant map. This is equivalent to being homotopy equivalent to a one-point space: the maps to and from that point give the two required composites.
Definition 5.2.2
(Retraction and strong deformation retraction)
For an inclusion , a retraction is a continuous with . It is a strong deformation retraction if, in addition, by a homotopy relative to . In particular, every point of stays fixed throughout. A mere retraction supplies no such homotopy.
For a path from to , write
This is the change-of-basepoint isomorphism proved in the preceding path chapter. Products of groups use component-wise multiplication. All products of spaces in this chapter carry the product topology.
Theorem/proposition
Theorem 5.2.3
(Homotopy and the moving basepoint)
If and , let . Then
Thus homotopic maps induce the same homomorphism when the homotopy fixes the basepoint. Without that restriction, the equation includes basepoint transport.
Theorem 5.2.4
(Homotopy equivalence preserves the fundamental group)
If is a homotopy equivalence and , then
is an isomorphism. In particular, a strong deformation retraction induces inverse isomorphisms and at every basepoint in the retract.
Theorem 5.2.5
(Fundamental group of a product)
For based spaces and , the projection maps induce an isomorphism
No path-connectedness hypothesis on the whole spaces is needed: all loops stay in their respective basepoint components.
Proof sketch or proof idea
Proof
(Boundary of the homotopy square)
For the moving-basepoint formula, let be a loop at and consider on the square. Traverse its boundary starting at along the bottom, right, top in reverse and left in reverse. Its image is the loop
up to harmless reparametrization. This loop is null-homotopic: contract the boundary traversal linearly to inside the convex square, then compose with . The contraction fixes the traversal's endpoints at that corner, so its image is a based null-homotopy. Canceling path classes gives , as claimed. This proof uses the square's entire interior, not an unsupported claim that every boundary loop is null in the boundary itself.
Proof X-ray
Keep the basepoint path in the equation
The square-boundary proof tracks four edges. The two vertical edges both trace , with opposite traversal directions. They are the reason conjugation appears. When is constant, these contributions cancel to constant paths and the familiar equality of induced maps is valid. When it moves, the two homomorphisms land in different based groups, so omitting transport is already a type error before any group calculation is attempted.
Proof
(Homotopy inverses give isomorphisms)
Now suppose is a homotopy inverse of and put , . Homotopy of to the identity, followed by the first theorem, shows that
is an isomorphism: its composition with a basepoint-change isomorphism is the identity. Hence is surjective. Apply the other homotopy to at . Its induced map, which factors as the map induced by after , is also an isomorphism. Consequently is injective. Therefore is an isomorphism, and is an isomorphism. This argument keeps the different basepoints in the two compositions distinct.
For a strong deformation retraction, is literally the identity and is homotopic to the identity fixing any . Functoriality and the first theorem give and , so the two homomorphisms are inverse. A contractible space is path-connected: for each , the contraction traces a path from to its constant endpoint. Its fundamental group is trivial by homotopy equivalence with a point. Thus every nonempty contractible space is simply connected.
Proof X-ray
Two composites, two halves of an isomorphism
The proof never inverts directly. The homotopy , through the moving-basepoint theorem, makes an isomorphism, which shows only that is surjective. The other homotopy, , is used at the basepoint ; there is followed by the map induced by at , not at , and that composite being an isomorphism makes injective. Only once is bijective does follow. A one-sided hypothesis, as for a bare retraction, delivers only one of these halves.
Proof
(Coordinatewise loops and coordinatewise homotopies)
For the product theorem, the displayed map is well-defined and a homomorphism by continuity of projections and functoriality. To prove surjectivity, pair any loops in and in : the function is continuous by the product universal property and has the required projections. To prove injectivity, suppose two loops have path-homotopic coordinate loops. If are the respective endpoint-fixing homotopies, then is continuous and fixes the pair of endpoints. It is a path homotopy of the original loops. In particular, both coordinate classes trivial imply the product loop class trivial. This proves the isomorphism without guessing from a picture of the torus.
Worked examples
Worked example 5.2.6
(The punctured plane retracts to its circle)
For , let . It is continuous and restricts to the identity on . Define
The coefficient is positive for every and , so the deformation never reaches the deleted origin. At time zero it is , at time one it is , and points of the circle remain fixed. Therefore this is a strong deformation retraction. At it yields . The positive-coefficient check is what turns a radial drawing into a valid proof.
Worked example 5.2.7
(A cylinder and an annulus)
The cylinder strongly deformation retracts to by . It therefore has fundamental group . The closed annulus similarly retracts to its inner circle by the radial formula above: the new radius is , still between and . Neither deformation crosses a point outside its declared space.
Worked example 5.2.8
(The torus has two independent winding coordinates)
For the product torus based at , the product theorem and circle calculation give . The loop
represents . Concatenation adds both coordinates. A loop is null-homotopic exactly when both windings vanish, because injectivity of the product isomorphism supplies a simultaneous coordinate homotopy. The two basic loops commute in the fundamental group; this follows from the product theorem, not from a claim that their parametrized concatenations are identical.
Worked example 5.2.9
(Retraction alone is weaker)
The constant map is a retraction onto one point. If it were a deformation retraction, the circle would be homotopy equivalent to a point, so its fundamental group would be trivial. The circle calculation gives instead. Thus a retraction need not be a deformation retraction. The existence of only proves that is injective; it does not prove the reverse composition is the identity.
Composition, extension over the disk and further deformations
Two general facts complete the basic toolkit. The first makes homotopy equivalence behave like an equivalence relation between spaces. The second translates the algebraic statement “ is trivial” for a loop into a geometric statement about extending a map.
Proposition 5.2.10
(Homotopy respects composition)
If and , then . Consequently homotopy equivalence of spaces is reflexive, symmetric and transitive.
Proof
(Two homotopies, one after the other)
Homotopy of maps is an equivalence relation on continuous maps , by the same formulas as for path homotopy: , , and the pasting of with along , which is continuous by finite closed pasting. Now let and . The map is a homotopy from to , and is a homotopy from to . Both are continuous as composites, and transitivity joins them.
For the consequence, the identity is a homotopy equivalence and the definition is symmetric. If has homotopy inverse and has homotopy inverse , the composition rule gives
and in the same way .
Write and regard as its boundary, with basepoint . Let , whose class generates .
Proposition 5.2.11
(Null-homotopic circle maps extend over the disk)
For a continuous map with , the following are equivalent:
- is homotopic to a constant map;
- extends to a continuous map ;
- is the trivial homomorphism.
Proof
(Collapse one end of a cylinder)
Two quotient maps do the topological work. The map , , is a continuous surjection from a compact space onto a Hausdorff space. It is therefore closed, hence a quotient map. Its only nontrivial fiber is . Similarly , , is a quotient map whose only nontrivial fibers are the pairs .
. Let satisfy and for a fixed point . Then is constant on every fiber of , so by the universal property of quotients from products and quotients there is a continuous with . For , .
. If is the inclusion, then and . The disk is convex, so is trivial and so is .
. The loop at represents , which is trivial. Choose a path homotopy from to the constant loop at . Because the endpoints stay at , for all , so is constant on the fibers of and factors as with continuous. Then and .
Applied to the identity of , whose induced map is the identity of , the proposition says the identity does not extend over the disk. This is the statement that is not a retract of , proved from the fundamental group in the unit on covering maps, lifting and the circle.
Worked example 5.2.12
(Punctured Euclidean spaces and a dimension argument)
For the formula is a strong deformation retraction of onto , by the same positive-coefficient check as in the plane. For the sphere is simply connected, as shown in the unit on van Kampen's theorem, so is simply connected.
This distinguishes from for . A homeomorphism would restrict to a homeomorphism , and translation identifies the latter with . The first space has fundamental group and the second is simply connected, so no such exists. The same removal of a point shows that is not homeomorphic to for : removing a point disconnects but leaves path-connected.
Worked example 5.2.13
(The Möbius band, its central circle and its boundary)
Define by
and let be its image. Since , the angle of the point around the vertical axis is recorded for . Within that half-plane the point is , which records . So is injective on , and the only identifications are : a strip glued to itself with a half-twist. As a compact subset of , is Hausdorff.
The central circle is the circle of radius in the horizontal plane. Put . This is well defined because . It is continuous: the map from the compact space onto the Hausdorff space is a closed continuous surjection, hence a quotient map, and the composite is continuous. At it is the identity, at it lands in , and it fixes throughout. Thus is a strong deformation retract of , and .
The boundary is a single circle, traced by the loop
based at ; the two formulas agree at because . The loop is injective on and onto , so it induces a continuous bijection from onto the Hausdorff space . That bijection is a homeomorphism because is compact, and therefore generates . With , a direct substitution gives , a loop winding twice around . Since is a homotopy equivalence, is an isomorphism. Hence the inclusion induces multiplication by between two copies of , the sign depending only on the chosen generators.
Consequently is not a retract of . A retraction would satisfy . If generates , then , so would be divisible by two in , where is a generator. The inner circle of an annulus, by contrast, is a strong deformation retract of it. Both spaces deformation retract onto a circle, but their boundary circles sit inside them in different ways.
How changing the transport path changes the isomorphism
The basepoint transport isomorphism depends on a path, not merely on its endpoints. Suppose and both run from to . Their transports have the same domain and codomain, so we can compare them. Put , a loop based at . For any loop at , substitution into the transport formulas gives
Thus changing the path changes transport by an inner automorphism of the target group. This is a precise form of the phrase “the fundamental group does not depend on the basepoint” within a path component: its isomorphism type does not, but a particular identification can depend on a connecting path. If the group is abelian, every inner automorphism is the identity, so all paths with those endpoints give the same transport. Without commutativity, the displayed conjugation must be retained. Even then some path changes can have no effect, namely when the difference loop represents a central element; noncommutativity does not force every two choices to give different maps.
Concept lensStructural
Transport as arrows of a groupoid
Collect the path classes between all pairs of points of , composed by concatenation. Every class has an inverse, so they form a groupoid, and is its group of arrows from to itself. Transport is conjugation by the arrow , and the computation above says that two arrows with the same ends differ by a loop. The moving-basepoint theorem says that a homotopy supplies, for each , an arrow from to that conjugates -images to -images.
A retraction supplies another useful distinction about what has been proved. The identity makes injective and surjective, even without a deformation. It does not make surjective. In the retraction from the circle onto a point, the inclusion-induced map has trivial domain and is injective, while the circle still has its nontrivial winding classes. The extra homotopy in a deformation retraction supplies the missing opposite identity. An algebraic left inverse and a two-sided inverse are different conclusions.
For the torus map of the exercises below, the image subgroup is generated by and and has index six. To see this directly, subtract an integer multiple of from any so the second coordinate becomes one of . Next subtract a multiple of so the first coordinate becomes zero or one. These six representatives are distinct modulo the image: a difference between two of them can have second coordinate divisible by three only when that difference is zero, and its remaining first coordinate can be even only when it too is zero. This does not contradict injectivity. An infinite group can be isomorphic to a proper finite-index subgroup of itself, as the simpler map on already shows. A kernel computation and an image-index computation measure different features of the same induced map.
These equations compare homotopy classes, not identical parametrized loops. Different bracketings of a concatenation can traverse the same pieces at different speeds. The endpoint-fixing reparametrization homotopies established in the path chapter are what make multiplication of classes associative. Keeping that distinction visible permits the algebraic cancellation above without silently equating different functions on the parameter interval.
Common mistakes
Common mistake
(Checking a proposed deformation retraction)
A homotopy equivalence is weaker than a homeomorphism. A strong deformation retraction need not preserve metric distances or dimension, but it must be continuous, remain inside the space at every time and fix the retract. Check all three conditions directly for a proposed formula.
Common mistake
(Simply connected does not mean contractible)
Contractibility implies simple connectedness; simple connectedness alone does not provide a contraction. The converse is false: the two-sphere is simply connected but not contractible, and, as the note below explains, the second half of that statement cannot be detected by the fundamental group. The product theorem also does not identify a torus with a wedge of circles: a wedge has a different gluing construction and requires a different fundamental-group calculation.
Summary
Homotopy of maps compares induced homomorphisms after the basepoint path is accounted for. Homotopy equivalences and deformation retractions preserve the fundamental group. Product spaces turn loop classes into pairs of loop classes, giving explicit computations for the punctured plane, cylinders, annuli and the product torus. A circle map is null-homotopic exactly when it extends over the disk, and the Möbius band deformation retracts onto a circle even though its boundary circle is not a retract.
Exercises
Quick check
(Prove that any nonempty convex subset of is contractible, and identify the point fixed by your contraction.)
Solution · Answer
Fix and set . Convexity keeps the image in ; the formula is continuous, starts at the identity and ends at the constant . Also for every , so the chosen basepoint remains fixed.
Quick check
(Find the fundamental group of at using either a deformation or the product theorem.)
Solution · Answer
The Euclidean factor is contractible by , so the product strongly retracts to , giving . Alternatively the product theorem gives , the same group.
Quick check
(On , define . Compute under and determine its kernel and image.)
Solution · Answer
On , the first coordinate becomes and the second . Thus , represented by the integer matrix with rows and . Its kernel is zero: gives , then gives . Its image consists exactly of with divisible by and even. These conditions are necessary by the formula and sufficient by choosing , .
Quick check
(Explain precisely why the radial map cannot be extended continuously to a retraction from onto .)
Solution · Answer
The formula is undefined at zero, and approaching zero along different radii gives different limiting unit vectors, so that formula has no continuous extension. More strongly, no different continuous retraction can repair it: a retraction would make the inclusion-induced map injective, which is impossible. The second argument excludes all possible extensions, not just this formula.
Quick check
(Let be the solid torus, based at , with boundary torus . Compute the homomorphism induced by the inclusion and deduce that is not a retract of .)
Solution · Answer
The inclusion is , where is the inclusion. Under the product isomorphisms it induces , that is, , because is convex and is trivial. Its kernel contains , the class of a meridian circle , which bounds a disk in . A retraction would give , forcing the inclusion-induced map to be injective. Hence no retraction exists.
Quick check
(Show that any two continuous maps are homotopic when is contractible. Show also that every continuous map from a contractible space to a path-connected space is homotopic to every constant map .)
Solution · Answer
Let , where is the constant map at . For any , is a homotopy from to the constant map at . Two maps homotopic to the same constant map are homotopic to each other. For the second claim, let the constant map at . Then joins to the constant map at . Path-connectedness of is used to move this constant: a path from to any gives the homotopy between the two constant maps. Hence is homotopic to every constant map .
Quick check
(Prove that a homotopy equivalence induces a bijection between the sets of path components of and of .)
Solution · Answer
A continuous map sends each path component into a single path component, so induces a function on path components, and . If , then for each the path joins to , so is the identity. The homotopy gives in the same way. Therefore is a bijection with inverse . Together with the isomorphism theorem above, this shows that a homotopy equivalence preserves both the number of path components and the fundamental group of each one.