Where we are on the ladder
You arrive here already fluent in homotopy. You have met the fundamental group pi_1(X), you have classified covering spaces, and you know how a CW complex is glued together cell by cell. Homotopy theory is powerful but stubbornly noncommutative and hard to compute: pi_1 of a wedge of circles is a free group, and the higher homotopy groups of even the 2-sphere are a famous swamp. Homology is the trade you make to regain computability. It throws away the noncommutative information on purpose, replacing groups of loops-up-to-homotopy with abelian groups that you can pin down by linear algebra over the integers. The bargain is honest: you lose some refinement, you gain a machine that actually turns the crank.
An honesty note before we start. These are graduate topics; we assume comfort with point-set topology, abstract algebra (groups, rings, and especially modules over the integers), and the homotopy survey from the previous rung. Words like homeomorphism, homotopy, and CW complex are used here, not re-derived. Note also that there is no single forced order through algebraic topology: many courses do homotopy before homology, as we are, but homology can be built first and is in some ways simpler. We follow the homotopy-then-homology route only because it lets us contrast the two; nothing below secretly depends on pi_1.
The standard simplex and singular simplices
The whole construction starts from one rigid model shape in each dimension: the standard n-simplex Delta^n. Delta^0 is a point, Delta^1 is a line segment, Delta^2 is a filled triangle, Delta^3 is a solid tetrahedron, and in general Delta^n is the convex hull of the n+1 standard basis points e_0, e_1, ..., e_n in R^(n+1). It comes with an ordering of its vertices baked in, and that ordering is not decoration — it is what will let us attach signs later. A singular n-simplex in a space X is then nothing fancier than a continuous map sigma: Delta^n -> X. The word 'singular' is a warning label: sigma need not be injective, need not be smooth, may crush the triangle to a point or fold it on itself. We allow every continuous map, and that generosity is exactly why singular homology is defined for any topological space at all.
From these maps we build a place for them to live. The group of singular n-chains C_n(X) is the free abelian group on the set of all singular n-simplices: a chain is a finite formal integer combination like 3*sigma - 2*tau + sigma', where each summand is a continuous map of the triangle into X and the integers are bookkeeping multiplicities. Do not over-interpret the coefficients; a chain is not a subset of X but a formal ledger entry. The reason we take formal sums with integer coefficients, rather than mere sets of simplices, is that we want to add and subtract, and we want minus signs — and the minus signs are about to do all the work.
The boundary operator, and why d composed with d is zero
Now the key gadget, the boundary operator d_n: C_n(X) -> C_(n-1)(X). On a single simplex sigma it spits out the alternating sum of its faces. The i-th face of Delta^n is the (n-1)-simplex you get by deleting vertex e_i, and restricting sigma to that face gives a singular (n-1)-simplex written sigma composed with the i-th face inclusion. The boundary is the sum of all of these with alternating signs: d sigma = sum over i of (-1)^i times (the i-th face of sigma). For an edge sigma: Delta^1 -> X from a point a to a point b, this reads d sigma = b - a, the endpoint minus the startpoint — exactly the signed boundary you would draw. The alternating signs are precisely the contribution of the vertex ordering we insisted on keeping.
Everything hinges on one identity: d_(n-1) composed with d_n is the zero map, written tersely as d^2 = 0. Concretely, the boundary of a boundary always vanishes. Take a filled triangle sigma: Delta^2 -> X with vertices 0, 1, 2. Its boundary is the three edges [1,2] - [0,2] + [0,1], traversed with those signs so they chain head-to-tail around the loop. Apply d again: each edge contributes its two endpoints with signs, and when you collect terms every vertex appears exactly twice with opposite sign and cancels. The reason is structural, not lucky — deleting vertex i and then vertex j gives the same sub-simplex as deleting j then i, and the alternating signs are arranged so the two orders cancel in pairs. This single cancellation is the algebraic shadow of the geometric fact that a boundary has no boundary: the rim of a disk is a circle, and a circle has no endpoints.
d sigma = sum_{i=0}^{n} (-1)^i ( sigma restricted to the i-th face )
edge [a,b] : d = b - a
triangle [0,1,2]: d = [1,2] - [0,2] + [0,1]
d^2 = 0 (boundary of a boundary is always zero)
... --d_3--> C_2(X) --d_2--> C_1(X) --d_1--> C_0(X) --> 0
with d_n o d_{n+1} = 0 at every stageCycles, boundaries, and the homology groups
The identity d^2 = 0 organizes the chain groups into a chain complex: a sequence of abelian groups with boundary maps such that each composite vanishes. That single condition lets us name two subgroups inside each C_n(X). A cycle is a chain z with d z = 0; the group of n-cycles is Z_n = ker d_n. A boundary is a chain of the form d w for some (n+1)-chain w; the group of n-boundaries is B_n = image d_(n+1). Because d^2 = 0, every boundary is a cycle — applying d to d w gives zero — so B_n sits inside Z_n as a subgroup. The whole point of d^2 = 0 is exactly to guarantee this containment, which is what makes the next definition legal.
The n-th singular homology group is the quotient H_n(X) = Z_n / B_n: cycles modulo boundaries. The slogan is worth memorizing — cycles that are not boundaries detect holes. A cycle is a closed thing with no edge of its own; it becomes trivial in homology exactly when it is the boundary of something one dimension up, i.e. when it can be filled in. So a 1-cycle around the hole of an annulus is a genuine class because nothing inside the annulus fills it, while a 1-cycle bounding a disk inside the plane is zero in H_1 because the disk fills it. Two cycles are called homologous when their difference is a boundary; H_n(X) is the group of homology classes. For a path-connected space H_0(X) is always the integers Z, since 0-cycles are points and any two points are homologous via the path between them — H_0 just counts path-components.
Functoriality: maps become homomorphisms
Here is the feature that turns a definition into a tool. A continuous map f: X -> Y induces a homomorphism on chains: push a singular simplex sigma: Delta^n -> X forward to f composed with sigma: Delta^n -> Y. This chain map f_# commutes with the boundary operator — f_# of d sigma equals d of f_# sigma, since composing with f does not touch which faces you take — and a chain map that respects d automatically sends cycles to cycles and boundaries to boundaries. Therefore it descends to a homomorphism on homology, f_*: H_n(X) -> H_n(Y). Two clean bookkeeping facts come for free: the identity map induces the identity, and (g composed with f)_* equals g_* composed with f_*. In the language of categories, H_n is a functor from topological spaces to abelian groups. That is the entire reason homology is useful rather than merely defined.
Functoriality is what lets homology prove theorems by pure formal pressure. Suppose two spaces had isomorphic homology only as an afterthought; instead, because H_n is a functor, a homeomorphism X -> Y forces an isomorphism H_n(X) -> H_n(Y), so homology is a genuine topological invariant — spaces with different homology cannot be homeomorphic. Better still, homology is a homotopy invariant: homotopic maps induce the same homomorphism, the deep reason being that a homotopy between maps builds an explicit chain homotopy between their chain maps, an algebraic gadget whose existence forces f_* = g_*. Consequently a homotopy equivalence gives an isomorphism on every H_n. This is how the no-retraction theorem and Brouwer's fixed-point theorem fall out almost mechanically: a retraction of the disk onto its boundary circle would force an impossible factorization of the identity through H_1, and the functor refuses it.
Two faces of the same theory, and what comes next
Singular homology is wonderfully general — defined for any space, manifestly a functor, obviously homotopy-invariant — but it is hopeless to compute directly: C_1(S^1) is already a free abelian group on uncountably many maps. Its computational twin is simplicial homology, where you triangulate the space into finitely many simplices and the chain groups become finite-rank, so the homology is literally a kernel-mod-image computation with integer matrices, the same Smith-normal-form linear algebra you would do by hand. The miracle, proved with subdivision, is that for any triangulable space the two theories agree: simplicial and singular homology are canonically isomorphic. You get the conceptual cleanliness of singular and the finite computability of simplicial, and you are free to switch between them.
Let me set honest expectations about what this guide did and did not establish. We constructed the groups H_n(X), checked d^2 = 0, and proved functoriality and homotopy invariance in outline — but the two genuinely hard pillars are still owed. We have not proved that homotopic maps induce equal homomorphisms in full (that needs the prism operator that builds the chain homotopy), and we have not proved excision, the cut-and-paste principle that actually makes homology computable. Both are real work, deferred to guide 2 of this rung, where the long exact sequence of a pair, excision, and Mayer-Vietoris turn these definitions into a calculus you can run on the torus, the Klein bottle, and the spheres. Treat what is above as the load-bearing definitions, not the finished theory.
One last connection worth holding onto, because it foreshadows the rest of the geometry ladder. The pattern 'a sequence of groups, a squaring-to-zero differential, and homology as cycles modulo boundaries' is not unique to topology. You have already met its smooth cousin: differential forms with the exterior derivative d satisfy d^2 = 0, and their cycles-mod-boundaries is de Rham cohomology. The de Rham theorem says that cohomology agrees with the cohomology dual to the singular homology built here, over the reals. So the chain complex you just met is the topological skeleton, and the smooth forms put flesh on it — same algebra, two incarnations. The rank of H_n, by the way, is the n-th Betti number, the integer count of n-dimensional holes that we will read off explicitly in guide 3.