Where we are on the ladder
You arrive at the algebraic-topology rung carrying point-set topology in your pocket: a topological space, open sets, continuity, connectedness, compactness, and the notion of a homeomorphism — a continuous bijection with continuous inverse. Homeomorphism is the gold standard of 'same shape', but it is brutally rigid. It cannot tell you that a thickened figure-eight and a thin figure-eight are morally the same, and it gives you no algebra to compute with. This rung's whole project is to relax 'same shape' into something coarser, computable, and still astonishingly powerful. The relaxation begins with one idea: homotopy.
An honesty note before we start. These are graduate topics. We assume you are fluent in point-set topology, linear algebra, and the smooth-manifold survey from Volume I, and that abstract algebra — groups, homomorphisms, generators and relations — is comfortable, because by the next guide it becomes the language of the fundamental group. Volume I words like manifold, homeomorphism, and covering space are used here, not re-derived; when one appears we link back rather than re-teach. One more piece of honesty about order: homotopy theory, homology, and differential topology are partly independent towers, and 'homotopy before homology' is a common but not forced route — we take it because π_1 is the most concrete handle on the whole subject.
Homotopy of maps: a continuous slide through time
Start with two continuous maps f, g: X -> Y. We say f is homotopic to g if you can continuously slide f into g, with no jumps along the way. Formally, a homotopy is a continuous map H: X × [0,1] -> Y such that H(x, 0) = f(x) and H(x, 1) = g(x) for every x. Read the second slot as time: at time t = 0 you see f, at time t = 1 you see g, and in between H(·, t) is a continuous family of maps morphing one into the other. The single word that makes this work is continuous — continuous jointly in the point x and the time t, so the whole movie has no rips.
A tiny concrete example fixes the idea. Let Y = R^n, and let f, g be any two maps into it. The straight-line homotopy H(x, t) = (1 − t) f(x) + t g(x) interpolates linearly: each point f(x) marches in a straight segment to g(x) as t runs from 0 to 1. It is plainly continuous, so in R^n every pair of maps is homotopic — there is nothing in the way. That triviality is itself the lesson: homotopy detects obstructions in the target, holes you cannot slide across. R^n has no holes, so everything collapses. The interest begins the moment Y has a hole.
Homotopy is an equivalence relation on the set of maps X -> Y, and checking this is a healthy first exercise in the bookkeeping of the subject. Reflexivity: the constant-in-time homotopy H(x, t) = f(x) shows f is homotopic to itself. Symmetry: run time backwards, H'(x, t) = H(x, 1 − t), turning a slide from f to g into one from g to f. Transitivity: if H slides f to g on [0,1] and K slides g to h, concatenate them — do H at double speed on [0, 1/2], then K at double speed on [1/2, 1] — and continuity at the seam holds because both agree on g. The equivalence classes are called homotopy classes, written [f]; they are the atoms the rest of the theory is built from.
Homotopy equivalence: when two spaces count as the same
Now lift the idea from maps to spaces. A homeomorphism demanded maps f: X -> Y and g: Y -> X whose compositions are exactly the identities. Relax 'exactly' to 'up to homotopy'. We say X and Y are homotopy equivalent if there exist continuous f: X -> Y and g: Y -> X with g ∘ f homotopic to the identity on X and f ∘ g homotopic to the identity on Y. The maps f and g need not be bijections, need not even be injective or surjective — they only have to undo each other after you are allowed to slide. This is a strictly weaker, strictly more forgiving notion of sameness than homeomorphism, and it is the right one for everything in this rung.
Here is the example that should be burned into memory. The punctured plane R^2 minus the origin is homotopy equivalent to the circle S^1. Let f: S^1 -> R^2 ∖ {0} be the inclusion, and g: R^2 ∖ {0} -> S^1 the radial projection g(v) = v / |v|. Then g ∘ f is literally the identity on S^1, and f ∘ g is homotopic to the identity on the punctured plane via the radial slide H(v, t) = (1 − t) v + t (v / |v|), which never hits 0 because the origin was removed. So the two-dimensional punctured plane and the one-dimensional circle are the same to homotopy — they share the same hole, and homotopy sees only the hole, not the dimension. That single hole is what the fundamental group of the next guide will measure as the integer winding number.
Deformation retracts: shrinking a space onto a subspace
The punctured-plane example used a special, very visual kind of homotopy equivalence: we shrank the big space down onto a subspace sitting inside it, leaving that subspace pinned in place the whole time. That move deserves its own name. Given a subspace A ⊆ X, a deformation retraction is a homotopy H: X × [0,1] -> X with H(x, 0) = x for all x, H(x, 1) ∈ A for all x, and crucially H(a, t) = a for all a ∈ A and all t — the subspace A never moves. When such an H exists, A is a deformation retract of X, and X is homotopy equivalent to A as a free consequence. So deformation retraction is a concrete, hands-on way to prove a homotopy equivalence: do not hunt for two abstract maps, just slide the whole space down onto the part you care about.
Deformation retraction of X onto a subspace A:
H : X x [0,1] -> X , continuous, with
H(x, 0) = x (start at the identity)
H(x, 1) in A (end inside A)
H(a, t) = a (A held fixed for all t) <-- this is the 'strong' clause
Consequence: X is homotopy equivalent to A .
Example: R^n deformation-retracts to a point {0}
H(x, t) = (1 - t) x ==> R^n is contractible.A subtle but standard caution about that third clause. The version above, where A is held fixed for all time, is called a strong deformation retraction; some books drop the per-time fixing and ask only that the final map land in A and restrict to the identity there, calling that a deformation retraction. Both still yield a homotopy equivalence X ≃ A, so for the purposes of this rung the distinction rarely bites — but it bites occasionally in delicate arguments, and the words genuinely differ between Hatcher, Bredon, and tom Dieck. Pick one convention, write down which clause you are assuming, and warn your reader; mixing the strong and weak versions is a quiet source of false 'theorems'.
Contractible spaces, and what homotopy chooses to forget
The most collapsed case is worth a name. A space X is contractible if it is homotopy equivalent to a single point — equivalently, if the identity map of X is homotopic to a constant map. The straight-line homotopy H(x, t) = (1 − t) x shows R^n is contractible, and the same trick works for any convex set and indeed any star-shaped set: pick the center, slide everything radially into it. A contractible space has, from homotopy's point of view, no features at all — no holes of any dimension. So when the next guides compute that the fundamental group of a contractible space is trivial, that is not a deep fact but a definition unwinding: nothing to wind around.
Now the honest part: what does this coarsening throw away? A great deal. Homotopy equivalence cannot distinguish a point from a closed ball, nor a circle from an annulus, nor R^3 from R^3 minus a point's worth of nothing — it is blind to dimension, to local smoothness, to whether a space is a manifold at all. It keeps only the 'hole structure', the connectivity data that the homotopy and homology groups will quantify. This is a feature, not a bug: by deliberately forgetting the rigid metric and smooth structure, you trade detail for computability, and you gain invariants — the fundamental group, the higher homotopy groups, later homology — that are actually calculable and that no homeomorphism-level invariant could organize so cleanly.
Invariants, the road ahead, and one honest limit
The payoff of all this relaxing is a strategy. Algebraic topology assigns to each space an algebraic object — a group, a ring, a sequence of groups — in such a way that homotopy-equivalent spaces get isomorphic objects. Such an object is a homotopy invariant. If two spaces receive different invariants, they cannot be homotopy equivalent, hence cannot be homeomorphic either; the algebra rules out the topology. The first and most hands-on invariant is the fundamental group, built from homotopy classes of loops based at a point, which we construct next guide and compute as π_1(S^1) = Z. The crude but classical Euler characteristic is another homotopy invariant, recovered later from homology.
- Want to show two spaces are NOT homeomorphic? Compute a homotopy invariant (start with π_1) of each; if they differ, you are done — no homeomorphism can exist.
- Want to show two spaces ARE homotopy equivalent? Deformation-retract each onto a small core, then match the cores — do not chase abstract maps by hand.
- Want to show a space has no holes? Exhibit a contraction — a homotopy from the identity to a constant map — and every homotopy and homology group collapses at once.
Finally, one honest limit, because invariants are powerful but not omnipotent. Equal invariants do NOT prove homotopy equivalence — they only fail to disprove it. Two spaces can share the same fundamental group, the same homology, even the same Euler characteristic, and still differ; the invariants are coarse shadows, not fingerprints. The deep partial converse, the Whitehead theorem of guide 5, says that a map between nice spaces (CW complexes) inducing isomorphisms on all homotopy groups IS a homotopy equivalence — but note the hypotheses, which are easy to drop: you need an actual map, and you need all the homotopy groups, and you need CW complexes. Stating the slogan without the hypotheses is exactly the kind of false closure this ladder asks us to avoid.