02.01.10 · analysis / topology

Fibre Homotopy Equivalence and Dold's Theorem

shipped3 tiersLean: none

Anchor (Master): Dold 1963 *Partitions of unity in the theory of fibrations* (Ann. of Math. 78); May 1999 *A Concise Course in Algebraic Topology* (University of Chicago Press) Ch. 7; tom Dieck 2008 *Algebraic Topology* (EMS) §5.5--5.6

Intuition Beginner

Two fibrations sit over the same base space $B$ . Each one is a way of stacking fibres above the points of $B$ . A fibrewise map between them is a continuous map of the total spaces that keeps everything over the point it started above: a point sitting over $b$ goes to a point still sitting over $b$ . It never slides sideways to a different part of the base.

A fibre homotopy equivalence asks for more. It is a fibrewise map that can be undone by another fibrewise map, where both round trips are homotopic to the identity through homotopies that are themselves fibrewise the whole time. So the two fibrations are the same not just as spaces, but as stacks of fibres over $B$ , in a way you can deform back and forth without ever leaving the column above each point.

Here is the surprise that this unit is built around. Suppose you have a fibrewise map and you only check that it is an ordinary homotopy equivalence of total spaces, forgetting the base entirely. Dold's theorem says that over a reasonable base this weaker fact is already enough: the map must have been a fibre homotopy equivalence all along.

In other words, for fibrations you do not have to verify the fibrewise inverse by hand. A plain homotopy equivalence upstairs upgrades itself, for free, into one that respects every fibre. That is a large saving, and it is why the theorem is used constantly.

Visual Beginner

Picture two bundles of vertical threads hanging over the same horizontal line $B$ . A fibrewise map slides each thread of the first bundle onto the thread of the second bundle sitting at the same spot on the line. It may stretch or fold a thread, but never carry it to a neighbour.

The dotted version of the same picture shows the inverse map and the deformations of the two round trips back to the identity. Every frame of those deformations keeps each thread over its own point. Dold's promise: if the solid map merely matches up the bundles as one big shape, the dotted picture can always be filled in.

Worked example Beginner

Take the base to be a single point. A fibration over a point is just a space $F$ , its fibre. A fibrewise map of two such fibrations is exactly an ordinary map $F \to F^{'}$ , because there is only one point to stay above. So over a point, fibre homotopy equivalence is the same as ordinary homotopy equivalence. Dold's theorem says nothing new here, and that is the point: the content lives in how it spreads this matching across a whole base.

Now grow the base to an interval $[0, 1]$ , which is contractible. A fibration over $[0, 1]$ looks like a stack of fibres, one over each point, possibly twisting as you move along. Dold's corollary says any such fibration is fibre homotopy equivalent to the plain product $[0, 1] \times F$ . The twisting can always be combed out fibrewise. This is the reason a fibration restricted to a small contractible patch of any base looks like a product up to fibre homotopy, which is the working picture most people carry in their head.

Check your understanding Beginner

Formal definition Intermediate+

Fix a base space $B$ . All maps in this section are over $B$ unless stated otherwise.

Definition (fibrewise map). Let $p : E \to B$ and $p^{'} : E^{'} \to B$ be maps. A continuous map $g : E \to E^{'}$ is fibrewise (a map over $B$ ) if $p^{'} \circ g = p$ . Equivalently $g$ carries the fibre $p^{- 1} (b)$ into $p^{' - 1} (b)$ for every $b \in B$ .

Definition (fibrewise homotopy). A homotopy $H : E \times [0, 1] \to E^{'}$ is fibrewise if each slice $H (-, t)$ is a fibrewise map, i.e. $p^{'} \circ H (e, t) = p (e)$ for all $e, t$ . We write $g ≃_{B} g^{'}$ when $g, g^{'}$ are joined by a fibrewise homotopy.

Definition (fibre homotopy equivalence). A fibrewise map $g : E \to E^{'}$ is a fibre homotopy equivalence if there is a fibrewise map $h : E^{'} \to E$ with $h \circ g ≃_{B} id_{E}$ and $g \circ h ≃_{B} id_{E^{'}}$ . The fibrations $p$ and $p^{'}$ are then fibre homotopy equivalent, written $E ≃_{B} E^{'}$ .

A fibre homotopy equivalence is in particular an ordinary homotopy equivalence of $E$ and $E^{'}$ , and it restricts on each fibre to a homotopy equivalence $p^{- 1} (b) ≃ p^{' - 1} (b)$ . The converse — recovering a fibrewise inverse from an ordinary one — is exactly the content of Dold's theorem below.

Standard situations.

Change of fibre. For a fibration $p : E \to B$ over a path-connected base, a path $γ$ from $b_{0}$ to $b_{1}$ in $B$ induces, by the homotopy lifting property, a homotopy equivalence $τ_{γ} : p^{- 1} (b_{0}) \to p^{- 1} (b_{1})$ depending only on the homotopy class of $γ$ . This is the change-of-fibre map.
Monodromy. Taking $γ$ a loop at $b_{0}$ gives a homomorphism $π_{1} (B, b_{0}) \to hAut (F)$ into the group of fibre homotopy classes of self-equivalences of the fibre $F$ — the $π_{1} (B)$ -action on the fibre.

Key theorem with proof Intermediate+

Theorem (Dold 1963). Let $p : E \to B$ and $p^{'} : E^{'} \to B$ be fibrations over a paracompact base $B$ , and let $g : E \to E^{'}$ be a fibrewise map. If $g$ is an ordinary homotopy equivalence of total spaces, then $g$ is a fibre homotopy equivalence.

Proof (sketch). The argument runs through an over- $B$ analogue of the Whitehead-type principle: a fibrewise map between fibrations that is "fibrewise a homotopy equivalence on each fibre" admits a fibrewise inverse, and one then has to promote the global hypothesis to that fibrewise one using paracompactness.

The local-to-global step is the heart of the matter. Cover $B$ by open sets $U_{α}$ over which both fibrations are fibrewise a product (such a cover exists because over a small enough — for instance contractible — neighbourhood every fibration is fibrewise a product). Over each $U_{α}$ the restricted map $g ∣_{p^{- 1} U_{α}}$ has a fibrewise inverse $h_{α}$ together with the two contracting fibrewise homotopies, because over $U_{α}$ the problem reduces to a homotopy equivalence of products. The inverses $h_{α}$ and their homotopies disagree on overlaps, but only up to fibrewise homotopy.

Choose a partition of unity ${ϕ_{α}}$ subordinate to the cover, available precisely because $B$ is paracompact. The functions $ϕ_{α}$ let one interpolate the local data: one glues the local inverses and homotopies along the "time" coordinate weighted by the $ϕ_{α}$ , using the fibrewise NDR (neighbourhood deformation retract) structure of the relevant pairs to make the patched homotopies continuous. The output is a single fibrewise inverse $h$ defined over all of $B$ together with global fibrewise homotopies $h g ≃_{B} id$ and $g h ≃_{B} id$ . That $g$ was an ordinary equivalence on total spaces is what guarantees the local pieces are equivalences to begin with; paracompactness is what lets them be assembled without sliding off the fibres. $□$

Bridge. Dold's theorem builds toward the change-of-fibre and classifying-space machinery: it is the foundational reason a fibration over a contractible base is fibrewise a product, which is exactly the local model that makes 02.01.07 (fibration) behave like a bundle up to homotopy. The over- $B$ Whitehead step is dual to the absolute statement of 03.12.20 (Whitehead theorem) — there a weak equivalence between CW complexes is upgraded to a homotopy equivalence; here a homotopy equivalence between total spaces is upgraded, fibre by fibre, to a fibrewise one — and this is exactly the same "promote a coarse equivalence to a structured one" pattern, with paracompactness playing the role CW structure plays in Whitehead's theorem. Putting these together, the central insight is that partitions of unity convert a global homotopy-theoretic fact into local-then-glued fibrewise data, a move that generalises from this theorem to the entire theory of numerable bundles and their classification. The bridge is that fibre homotopy type, not point-set type, is the correct invariant of a fibration, and Dold's theorem is what makes that invariant computable; it appears again in the proof that homotopic maps induce fibre homotopy equivalent pullbacks.

Exercises Intermediate+

Exercise 2 (medium, short-answer).

Deduce from Dold's theorem that any fibration $p : E \to B$ over a contractible paracompact base $B$ is fibre homotopy equivalent to the product fibration $pr_{1} : B \times F \to B$ , where $F$ is the fibre.

Hint

A contraction of $B$ gives, by homotopy lifting, a fibrewise comparison map; then check it is an ordinary equivalence of total spaces.

Answer

Let $H : B \times [0, 1] \to B$ contract $B$ to $b_{0}$ , so $H_{0} = id_{B}$ and $H_{1} \equiv b_{0}$ . Lifting $H$ along $p$ (homotopy lifting property) produces a fibrewise map $E \to B \times F$ over $B$ , where $F = p^{- 1} (b_{0})$ . Because $B$ is contractible, both $E$ and $B \times F$ deformation retract onto $F$ , so the comparison map is an ordinary homotopy equivalence of total spaces. Dold's theorem then promotes it to a fibre homotopy equivalence $E ≃_{B} B \times F$ . Rubric: full credit for the lifted comparison map, the total-space equivalence, and the appeal to Dold.

Exercise 3 (medium, short-answer).

Let $f_{0} ≃ f_{1} : B^{'} \to B$ be homotopic maps and $p : E \to B$ a fibration over a paracompact base $B^{'}$ . Show the pullbacks $f_{0}^{*} E$ and $f_{1}^{*} E$ are fibre homotopy equivalent over $B^{'}$ .

Hint

Pull back along the homotopy $B^{'} \times [0, 1] \to B$ and restrict to the two ends; use that $B^{'} \times [0, 1]$ retracts to $B^{'} \times {0}$ .

Answer

Let $K : B^{'} \times [0, 1] \to B$ be the homotopy. Then $K^{*} E \to B^{'} \times [0, 1]$ is a fibration, and its restrictions to the two ends are $f_{0}^{*} E$ and $f_{1}^{*} E$ . The inclusion $B^{'} \times {0} ↪ B^{'} \times [0, 1]$ is a fibrewise (over $B^{'}$ ) homotopy equivalence of bases, and lifting gives a fibrewise comparison between $f_{0}^{*} E$ and the restriction of $K^{*} E$ , which is an ordinary equivalence on total spaces. Dold's theorem upgrades it; doing this at both ends yields $f_{0}^{*} E ≃_{B^{'}} f_{1}^{*} E$ . Rubric: full credit for the pullback over the cylinder and the Dold upgrade.

Exercise 4 (hard, short-answer).

Give an example showing the paracompactness hypothesis in Dold's theorem cannot simply be dropped, by indicating where the partition-of-unity step fails over a non-paracompact base.

Hint

The interpolation of local inverses needs a subordinate partition of unity on the base; without one there is no continuous gluing parameter.

Answer

The proof produces local fibrewise inverses $h_{α}$ over a cover ${U_{α}}$ and glues them by weighting the contracting homotopies with a partition of unity ${ϕ_{α}}$ subordinate to the cover. On a non-paracompact base such a subordinate partition of unity need not exist, so there is no continuous family of weights with which to interpolate the $h_{α}$ across overlaps; the local equivalences cannot in general be assembled into a global fibrewise inverse. The standard counterexamples use long-line or large-ordinal bases where the local product structure holds patchwise but no global numeration exists. Rubric: full credit for locating the failure precisely at the partition-of-unity / interpolation step and tying it to the absence of a subordinate partition of unity.

Advanced results Master

Dold phrased the theorem in the language of numerable coverings — those admitting a subordinate partition of unity — rather than paracompactness, and worked with the broader class of numerable fibrations (maps satisfying the homotopy lifting property after restriction to a numerable cover). This is the natural setting: the entire argument is driven by partitions of unity, and numerability is exactly the hypothesis the gluing needs. Over a paracompact base every open cover is numerable, so the paracompact statement is a corollary.

Theorem (Dold, numerable form). Let $p, p^{'}$ be numerable fibrations over $B$ and $g : E \to E^{'}$ a fibrewise map. If $g$ is a homotopy equivalence of total spaces, then $g$ is a fibre homotopy equivalence. More generally, a fibrewise map that is fibrewise a homotopy equivalence over each member of a numerable cover is a fibre homotopy equivalence.

Corollary (fibrations over a contractible base are fibrewise a product). A numerable fibration over a contractible base is fibre homotopy equivalent to a product $B \times F$ .

Corollary (homotopy invariance of pullbacks). Homotopic maps $f_{0} ≃ f_{1} : B^{'} \to B$ induce fibre homotopy equivalent pullback fibrations $f_0^ E \simeq_{B'} f_1^* E $o v er a p a r a co m p a c t$ B' $. * T hi s i s t h ee n g in e b e hin d t h ec l a ss i f i c a t i o n o f f ib r a t i o n s b y h o m o t o p y c l a sseso f ma p s in t o a c l a ss i f y in g s p a ce : t h e a ss i g nm e n t$ [B', B] \to {\text{fibrations over } B'}/\simeq_{B'}$ is well defined precisely because of Dold's theorem.

Change of fibre and the fundamental-group action. For a fibration over a path-connected base, the change-of-fibre construction $τ_{γ} : F_{b_{0}} \to F_{b_{1}}$ depends only on the homotopy class of $γ$ — itself a small instance of Dold-type reasoning, since $τ_{γ}$ and $τ_{γ^{'}}$ for homotopic paths are joined by lifting the path homotopy. Restricting to loops gives the monodromy action $π_{1} (B, b_{0}) \to hAut (F)$ , equivalently the action of the loop space $Ω B$ on the homotopy fibre. This action is the local coefficient system that governs the $E_{2}$ -page of the Leray–Serre spectral sequence and detects when a fibration is simple.

Synthesis. The central insight is that the correct equivalence relation on fibrations is fibre homotopy equivalence, and Dold's theorem is the foundational reason this relation is testable by a single global condition rather than infinitely many fibrewise ones. This is exactly the over- $B$ shadow of the absolute Whitehead philosophy: where Whitehead's theorem says a map of CW complexes inducing isomorphisms on homotopy groups is an equivalence, Dold says a fibrewise map inducing an equivalence on total spaces is a fibrewise equivalence, and putting these together one sees both as instances of "a coarse equivalence between sufficiently structured objects is automatically a structured equivalence." The change-of-fibre action and the contractible-base product structure are dual faces of the same fact: the former measures how the fibre is permuted as one moves around the base, the latter says that over a base with no loops to move around, no permutation survives. This generalises from numerable fibrations to the abstract theory of fibrant objects in a model category, where Dold's theorem becomes the statement that a weak equivalence between fibrant objects over a base is an equivalence in the over-category, and the bridge to classifying-space theory is that homotopy invariance of pullbacks is what makes a representing object — a classifying space — exist at all.

Full proof set Master

Proposition (fibrewise homotopy is an equivalence relation refining ordinary homotopy). For fixed fibrations $p, p^{'}$ over $B$ , the relation $≃_{B}$ on fibrewise maps $E \to E^{'}$ is an equivalence relation, and $g ≃_{B} g^{'}$ implies $g ≃ g^{'}$ .

Proof. Reflexivity uses the constant homotopy $H (e, t) = g (e)$ , which is fibrewise because $p^{'} g (e) = p (e)$ for all $t$ . Symmetry reverses the time parameter, $\overset{ˉ}{H} (e, t) = H (e, 1 - t)$ , preserving the fibrewise condition slice by slice. Transitivity concatenates two fibrewise homotopies $H, H^{'}$ with $H_{1} = H_{0}^{'}$ by the usual reparametrised gluing $t \mapsto H (2 t)$ for $t \leq 1/2$ and $H^{'} (2 t - 1)$ otherwise; each slice still satisfies $p^{'} \circ (-) = p$ , so the concatenation is fibrewise and continuous by the pasting lemma. Finally a fibrewise homotopy is in particular a homotopy of the underlying maps, giving $g ≃ g^{'}$ . $□$

Proposition (fibre homotopy equivalence is detected fibrewise). If $g : E \to E^{'}$ is a fibre homotopy equivalence over $B$ , then for each $b \in B$ the restriction $g_{b} : p^{- 1} (b) \to p^{' - 1} (b)$ is an ordinary homotopy equivalence.

Proof. Let $h$ be a fibrewise inverse with fibrewise homotopies $H : h g ≃_{B} id_{E}$ and $H^{'} : g h ≃_{B} id_{E^{'}}$ . Because $h$ is fibrewise it restricts to $h_{b} : p^{' - 1} (b) \to p^{- 1} (b)$ , and because $H$ is fibrewise each slice $H (-, t)$ maps $p^{- 1} (b)$ into $p^{- 1} (b)$ ; hence $H$ restricts to a homotopy $h_{b} g_{b} ≃ id$ on the fibre $p^{- 1} (b)$ . The same restriction of $H^{'}$ gives $g_{b} h_{b} ≃ id$ on $p^{' - 1} (b)$ . Therefore $g_{b}$ is a homotopy equivalence with inverse $h_{b}$ . $□$

Theorem (contractible base, explicit form). Let $p : E \to B$ be a fibration over a contractible paracompact base $B$ with $H : B \times [0, 1] \to B$ contracting $B$ to $b_{0}$ and fibre $F = p^{- 1} (b_{0})$ . Then there is a fibre homotopy equivalence $E ≃_{B} B \times F$ .

Proof. Lift the homotopy $H \circ (p \times id) : E \times [0, 1] \to B$ starting from $id_{E}$ along $p$ to obtain $\tilde{H} : E \times [0, 1] \to E$ with $\tilde{H}_{0} = id_{E}$ and $p \tilde{H}_{1} \equiv b_{0}$ , so $\tilde{H}_{1}$ lands in $F$ . Define $g : E \to B \times F$ by $g (e) = (p (e), \tilde{H}_{1} (e))$ ; this is fibrewise since its first coordinate is $p (e)$ . On total spaces $g$ is a homotopy equivalence: $B \times F ≃ F ≃ E$ , the last equivalence because $\tilde{H}$ deformation retracts $E$ onto $F$ and $B$ is contractible. By Dold's theorem $g$ is a fibre homotopy equivalence. $□$

Connections Master

Fibration 02.01.07. Dold's theorem is the structural fact that makes the homotopy-theoretic study of fibrations possible: it is the reason a fibration is determined up to fibre homotopy by data that can be checked on total spaces and on a numerable cover. The contractible-base corollary is precisely the local product picture that the fibration unit treats informally as "looks like a product over small patches."
Whitehead theorem 03.12.20. Dold's theorem is the over- $B$ , fibrewise analogue of Whitehead's principle. Where Whitehead promotes a weak equivalence of CW complexes to a homotopy equivalence, Dold promotes an ordinary equivalence of total spaces to a fibrewise one; both are instances of upgrading a coarse equivalence between adequately structured objects to a structured equivalence.
Homotopy and homotopy group 03.12.01. The change-of-fibre construction and the $π_{1} (B)$ -action on the fibre that Dold's theorem organises are the source of the local coefficient systems in the long exact sequence and spectral sequence of a fibration; the monodromy homomorphism $π_{1} (B) \to hAut (F)$ lives in the homotopy-group framework set up there.

Historical & philosophical context Master

The decisive paper is Albrecht Dold's 1963 Partitions of unity in the theory of fibrations (Ann. of Math. 78) ^{[Dold 1963]}, which isolated numerability — the existence of a subordinate partition of unity — as the exact hypothesis under which the homotopy theory of fibrations becomes tractable. Before Dold, results comparing fibrations were typically proved over polyhedral or paracompact bases by ad hoc cell-by-cell or simplex-by-simplex arguments inherited from the Steenrod–Whitney bundle theory. Dold's move was to replace those combinatorial inductions by a single analytic device, the partition of unity, which interpolates locally defined homotopy data into a global fibrewise construction. The theorem that a fibrewise homotopy equivalence is detected on total spaces, and its companion that numerable fibrations are classified up to fibre homotopy by homotopy classes of maps, are the load-bearing consequences.

Philosophically, Dold's theorem encodes a recurring lesson of mid-century algebraic topology: the right notion of sameness for a geometric object carrying extra structure (here, the projection to a base) is the homotopy-theoretic one relative to that structure, and partitions of unity are the bridge between point-set local product structure and global homotopy-theoretic invariance. This perspective fed directly into the formulation of model categories and the over-category viewpoint, where Dold's theorem reappears as the statement that weak equivalences between fibrant objects are equivalences in the slice, and into the modern treatment in May's A Concise Course in Algebraic Topology (Ch. 7) ^{[May 1999]}, which presents the result as the cornerstone of the elementary theory of fibrations.

Bibliography Master

@article{Dold1963Partitions,
  author  = {Dold, Albrecht},
  title   = {Partitions of unity in the theory of fibrations},
  journal = {Ann. of Math. (2)},
  volume  = {78},
  year    = {1963},
  pages   = {223--255}
}

@book{MayConciseAlgebraicTopology,
  author    = {May, J. Peter},
  title     = {A Concise Course in Algebraic Topology},
  publisher = {University of Chicago Press},
  year      = {1999}
}

@book{tomDieckAlgebraicTopology,
  author    = {tom Dieck, Tammo},
  title     = {Algebraic Topology},
  publisher = {European Mathematical Society},
  year      = {2008}
}

@book{HatcherAlgebraicTopology,
  author    = {Hatcher, Allen},
  title     = {Algebraic Topology},
  publisher = {Cambridge University Press},
  year      = {2002}
}

Prerequisites

02.01.07
03.12.01
03.12.20

Tier anchors

beginner: May 1999 *A Concise Course in Algebraic Topology* (University of Chicago Press) Ch. 7 (informal account of fibrewise maps); 3Blue1Brown bundle / fibre intuition
intermediate: May 1999 *A Concise Course in Algebraic Topology* (University of Chicago Press) Ch. 7; tom Dieck 2008 *Algebraic Topology* (EMS) §5.5
master: Dold 1963 *Partitions of unity in the theory of fibrations* (Ann. of Math. 78); May 1999 *A Concise Course in Algebraic Topology* (University of Chicago Press) Ch. 7; tom Dieck 2008 *Algebraic Topology* (EMS) §5.5--5.6

References

Dold 1963 — Partitions of unity in the theory of fibrations · Ann. of Math. 78, 223--255 (originator paper for the theorem)
May — A Concise Course in Algebraic Topology · Ch. 7, fibrewise homotopy equivalence and Dold's theorem
tom Dieck — Algebraic Topology · §5.5--5.6, h-equivalences of fibrations over paracompact bases

Estimated time

beginner: 18m
intermediate: 45m
master: 80m