Building spaces one cell at a time
Four guides into this rung you can compute the fundamental group of almost anything you can present: loops in the circle gave pi_1(S^1) = Z, covering spaces organized subgroups, and Seifert-van Kampen glued presentations together. But notice what every one of those tools secretly assumed: that you had a space made of recognizable pieces you could cut and reassemble. This last guide makes that assumption into a method. A CW complex is a space built inductively by attaching cells — points (0-cells), arcs (1-cells), disks (2-cells), and so on — each n-cell glued in along a map from its boundary sphere S^(n-1) into the part you have already built.
Hold a concrete picture. To build the torus as a CW complex, take one 0-cell, two 1-cells (a loop a and a loop b, both starting and ending at that single point), and one 2-cell — a square — whose boundary you glue down along the word a b a^(-1) b^(-1). That is the entire torus: one vertex, two edges, one face. The Klein bottle is built the same way but with boundary word a b a b^(-1), and the sphere S^2 needs only one 0-cell and one 2-cell whose whole boundary collapses to that point. The data of a CW complex is exactly this: a list of cells in each dimension and an attaching map for each, telling you how its rim sits on the lower skeleton.
Why CW complexes are the right category
It would be fair to ask why we bother with a special class of spaces at all — why not work with every topological space? The honest answer is that homotopy theory only behaves on CW complexes, and two theorems show why. The first is the cellular approximation theorem: any continuous map between CW complexes is homotopic to a cellular one, a map that sends the n-skeleton into the n-skeleton. So when you study maps up to homotopy — which is all homotopy theory ever does — you may always assume your map respects the cell structure, reducing a continuous problem to a combinatorial one. This is the workhorse behind nearly every computation in the rung.
The second is the Whitehead theorem, and it draws the exact boundary of how much homotopy groups can see. It says: a map f: X -> Y between CW complexes that induces isomorphisms on all homotopy groups pi_n for every n is a homotopy equivalence. This is the precise sense in which the family of all pi_n is a complete invariant of CW homotopy type. But read the hypothesis honestly, because the slogan 'isomorphism on homotopy groups means equivalent' is a famous trap. It requires a single map inducing those isomorphisms — two spaces can have abstractly isomorphic homotopy groups in every degree and still fail to be homotopy equivalent if no map realizes the isomorphisms. And it requires CW complexes: drop that and the theorem is simply false.
Higher homotopy groups: holes a loop cannot feel
The fundamental group asked: how many ways can a loop — a map of S^1 — sit in X up to deformation? There is nothing sacred about the number 1. The n-th higher homotopy group pi_n(X, x_0) is the set of homotopy classes of based maps from the n-sphere S^n into X, with maps required to send a chosen basepoint of S^n to x_0. The group operation generalizes concatenation of loops: squash two maps onto two hemispheres of a single S^n. For n = 1 this recovers pi_1 exactly. The reason to climb is that pi_1 is blind to higher holes — the sphere S^2 has trivial pi_1, since every loop on a sphere contracts, yet S^2 plainly has a two-dimensional hole, and that hole is exactly what pi_2(S^2) = Z detects.
Two structural facts make the higher groups feel different from pi_1, and both reward a moment's thought. First, for every n at least 2 the group pi_n is abelian — the extra room in S^n lets you slide one map past another, an argument (the Eckmann-Hilton trick) that fails for loops where there is no room to commute. Second, pi_n is far harder to compute than pi_1, and the single sharpest illustration is honest and humbling: the homotopy groups of spheres are not fully known. We have pi_2(S^2) = Z and pi_3(S^2) = Z (the Hopf fibration, below), but pi_n(S^k) for general n > k is a famously hard, only partially understood object — there is no closed formula, and computing it is the subject of decades of research.
Fibrations and the long exact sequence
If higher homotopy groups are this hard, how does anyone ever compute one? The single most important tool is the fibration. A fibration p: E -> B is a map with the homotopy lifting property: any homotopy of maps into the base B that starts at a map already lifted to the total space E can be lifted, as a whole, back up to E. The intuition is that E is the base B with a fiber F = p^(-1)(b) hanging over every point, glued together so that paths downstairs can always be followed upstairs. A covering space is the special case where the fiber is discrete; a fibration lets the fiber be any space, which is exactly the generalization the higher groups demand.
The payoff is the long exact sequence of a fibration, the computational engine this guide has been building toward. From the fiber F, total space E, and base B it threads together all their homotopy groups into one infinite exact sequence, where 'exact' means the image of each map equals the kernel of the next — so knowing two of the three groups in any local stretch pins down the third. This is the higher-homotopy analogue of how a covering map gave you pi_1 of the base from the total space in Guide 3; now it relates pi_n of fiber, total space, and base in every degree at once.
... -> pi_n(F) -> pi_n(E) -> pi_n(B) -> pi_(n-1)(F) -> ... -> pi_0(F) Hopf fibration: S^1 -> S^3 -> S^2 (F = S^1, E = S^3, B = S^2) ... -> pi_3(S^1) -> pi_3(S^3) -> pi_3(S^2) -> pi_2(S^1) -> ... ... -> 0 -> Z -> pi_3(S^2) -> 0 -> ... exactness forces pi_3(S^2) = Z
Reading the sequence: the Hopf fibration
Let the Hopf fibration earn its starring role, because it is the cleanest worked example in the whole subject. The 3-sphere S^3 can be seen as the unit sphere in C^2; the map sending a point to the complex line through it lands in CP^1, which is just S^2. The fiber over each point is a circle S^1 — the unit-modulus complex numbers acting on that line. So S^3 is woven from circles, one over every point of S^2, and no two of those circles are unlinked: this is a genuinely nontrivial bundle, not a product, and that nontriviality is the entire content of pi_3(S^2) = Z.
- Write down the long exact sequence around degree 3 for the fibration S^1 -> S^3 -> S^2, with fiber F = S^1, total space E = S^3, base B = S^2.
- Plug in what you already know about the circle: pi_n(S^1) = 0 for every n at least 2, because the universal cover of S^1 is the contractible line R (recall Guide 3). So both pi_3(S^1) and pi_2(S^1) vanish.
- Read the local stretch 0 -> pi_3(S^3) -> pi_3(S^2) -> 0. Exactness says the middle map is both injective and surjective — an isomorphism.
- Use the only sphere fact you need at the top: pi_n(S^n) = Z (a degree, the wrapping number), so pi_3(S^3) = Z. Conclude pi_3(S^2) = Z — a nonzero higher homotopy group of the 2-sphere, with the Hopf map as its generator.
Stand back and see the shape of the whole rung in this one calculation. You used the cell-by-cell discipline of CW complexes to know spheres are clean objects; you used the higher homotopy group pi_n to ask a question pi_1 could not; and you used a fibration and its long exact sequence to convert that question into two lines of bookkeeping. Be honest about the limits as you leave: this worked because the circle's higher homotopy vanished and the spheres lined up perfectly — for most spaces the sequence has nonzero groups on both sides and you are left with an extension problem, not an answer. The general homotopy groups of spheres remain unknown, and that open horizon is exactly where this rung hands you off to the homology and cohomology that come next.