The Seifert-van Kampen Theorem: Gluing Fundamental Groups

The Precise Statement

Let X be a topological space with an open cover X = U ∪ V, where U, V, and U∩V are all path-connected and nonempty. Choose a basepoint x₀ ∈ U∩V. The inclusions induce homomorphisms i∗ : π₁(U∩V) → π₁(U) and j∗ : π₁(U∩V) → π₁(V). The theorem asserts that the natural map

• π₁(U,x₀) ∗ π₁(V,x₀) ⟶ π₁(X,x₀)

is surjective, and its kernel is the normal subgroup generated by all elements i∗(ω) j∗(ω)⁻¹ for ω ∈ π₁(U∩V). Equivalently, π₁(X) is the amalgamated free product π₁(U) ∗_{π₁(U∩V)} π₁(V). In the language of category theory, π₁ carries the pushout square of spaces (U∩V → U, V → X) to a pushout square of groups. The openness of U and V is what makes the compactness argument in the proof go through.

The Picture: Cutting and Regluing Loops

Think of a loop γ in X based at x₀. Because {U,V} is an open cover of the compact interval [0,1], you can chop γ into finitely many arcs, each of which lies entirely in U or entirely in V. Insert a little detour back to x₀ at each cut point (running along a path inside the overlap), and γ becomes a concatenation of honest loops, alternating between U and V. That is exactly the statement that π₁(U) and π₁(V) generate π₁(X).

The only ambiguity is the choice of detour path for a loop living in the overlap: read it in U you get i∗(ω), read it in V you get j∗(ω). These two must be declared equal — that is precisely the amalgamation relation. So the picture is: the two pieces contribute their generators freely, and the overlap glues them together by identifying its loops' two shadows.

The Key Idea of the Proof

Two things must be shown: surjectivity (every loop factors through the pieces) and that the amalgamation relations are the only ones. Both rest on the same engine — the Lebesgue number lemma applied to the compact domains [0,1] and [0,1]×[0,1].

• Generation. Pull back the open cover {U,V} along a loop γ : [0,1] → X. The preimages cover the compact interval, so a partition 0 = t₀ < t₁ < … < t_n = 1 exists with each subarg γ([t_{k−1},t_k]) inside one member. Path-connectedness of U∩V lets you close each subarc into a loop.
• Relations. Given a homotopy F : [0,1]² → X between words, subdivide the square into a grid so fine that each cell maps into U or V (Lebesgue number again). Reading the homotopy cell-by-cell rewrites one word into the other using only moves inside a single piece plus the overlap identifications — proving no extra relations are needed.

The formal packaging verifies the universal property of the pushout directly.

Worked Example: The Wedge of Two Circles

Let X = S¹ ∨ S¹, a figure-eight with wedge point x₀. To apply the theorem cleanly we thicken slightly: let U be one circle plus a little open arc of the other (so U deformation-retracts to a circle, π₁(U) ≅ ℤ = ⟨a⟩), and V symmetrically the other circle (π₁(V) ≅ ℤ = ⟨b⟩). Their intersection U∩V is a small contractible neighborhood of x₀, so π₁(U∩V) is trivial.

With trivial overlap group, the amalgamation relations i∗(ω)j∗(ω)⁻¹ are vacuous, and the amalgamated free product collapses to the plain free product:

• π₁(S¹ ∨ S¹) ≅ ℤ ∗ ℤ = F₂, the free group on two generators a, b.

This is striking: the two loops do not commute — ab ≠ ba — even though each individually generates an abelian ℤ. The non-abelian fundamental group detects that you cannot slide one loop past the other through the wedge point. More generally, a wedge of n circles has π₁ ≅ F_n.

Why the Hypotheses Matter

Every hypothesis earns its keep, and each fails informatively when dropped.

• Path-connected overlap. If U∩V is disconnected, the plain theorem fails. Take S¹ = U ∪ V with U, V two overlapping arcs meeting in two components. Both pieces are contractible, so the naive pushout of trivial groups would predict π₁ = 1 — but π₁(S¹) = ℤ. The second overlap component is exactly what generates that ℤ. The fix is Ronald Brown's fundamental groupoid version, which tracks basepoints in every component.
• Open cover. Openness underwrites the Lebesgue-number subdivision. With arbitrary closed pieces the compactness argument can break (though CW-pairs and cofibrations behave well, which is why the theorem is usually invoked for cell attachments).
• Path-connected pieces. Needed so that a single basepoint suffices to see all of π₁(U) and π₁(V).

The theorem is the π₁ analogue of the Mayer–Vietoris sequence in homology; the disconnected-overlap subtlety mirrors the reduced-homology bookkeeping there.

Applications and Significance

Seifert-van Kampen is the workhorse for computing fundamental groups, turning topology into finite group presentations.

• CW complexes. Attaching a 2-cell along a loop α adds the relation α = 1 (the new cell's boundary bounds a disk). Iterating gives every group as a fundamental group: pick generators (wedge of circles) and relators (glue 2-cells) — so π₁ realizes arbitrary presentations.
• Surfaces. The genus-g orientable surface is a 4g-gon with edges identified; one 2-cell attached to a wedge of 2g circles yields ⟨a₁,b₁,…,a_g,b_g | ∏[a_i,b_i]⟩.
• Knot theory. Wirtinger presentations of knot-group complements come straight from van Kampen applied to a knot diagram.
• 3-manifolds & graphs of groups. Gluing pieces along tori/annuli produces amalgamated products and HNN extensions, the foundation of Bass–Serre theory.

It is, alongside covering space theory, one of the two pillars that make π₁ genuinely computable rather than merely definable.

How the theorem specializes: the group of the overlap controls how the two pieces are glued.

Space X = U ∪ V	π₁(U∩V)	Resulting π₁(X)	Why
Wedge of two circles S¹∨S¹	trivial (overlap contractible)	ℤ ∗ ℤ (free group F₂)	amalgamation over trivial group = free product
Sphere S², U,V = two disks	ℤ (equator ≃ S¹)	trivial group	the ℤ from the overlap is killed in each contractible disk
Torus T² = S¹×S¹	ℤ (annular overlap)	ℤ² = ⟨a,b | aba⁻¹b⁻¹⟩	overlap relation forces a,b to commute
Klein bottle	ℤ (annular overlap)	⟨a,b | abab⁻¹⟩	the gluing reverses orientation, giving a twisted relation
Genus-g surface Σ_g	free group (from a wedge)	⟨a_i,b_i | ∏[a_i,b_i]⟩	one relator = boundary of the glued 2-cell