Invariance of the Euler characteristic

Theorem

Invariance of the Euler characteristic

T-121Home MU-401Threads space · structure

Statement

Let \(X\) be a compact topological space admitting a finite CW-complex (in particular, a finite simplicial complex) structure \(K\), and for each \(i \geq 0\) let \(c_i(K)\) denote the number of \(i\)-cells of \(K\). Define \(\chi(K) = \sum_{i=0}^{n} (-1)^i c_i(K)\), where \(n = \dim K\). If \(K_1\) and \(K_2\) are any two finite CW structures on spaces \(X_1\) and \(X_2\) with \(X_1 \cong X_2\) (homeomorphic), or more generally with \(X_1 \simeq X_2\) (homotopy equivalent), then \(\chi(K_1) = \chi(K_2)\). In particular, for a triangulated surface with \(V\) vertices, \(E\) edges and \(F\) faces, the quantity \(V - E + F\) depends only on the topological type of the surface, not on the chosen triangulation.

Why it matters

The Euler characteristic is the oldest and most robust link between the combinatorics of a decomposition (how many vertices, edges, faces you drew) and the topology of the underlying space (a number that cannot see the drawing at all). Its invariance is what licenses every "count \(V-E+F\) on this particular picture and read off a topological fact" argument, from Euler's formula for convex polyhedra to the classification of surfaces and the Gauss–Bonnet theorem.

More structurally, this theorem is the first serious payoff of homology theory: it converts a statement that looks combinatorial (cell counts) into a statement that is manifestly an algebraic invariant of the space (alternating sum of Betti numbers), and thereby explains why the combinatorial count could not have depended on the triangulation in the first place.

Hypotheses

\(X\) is compact and admits a finite CW / simplicial structure.Drop finiteness and the sum \(\sum(-1)^i c_i\) need not converge: \(\mathbb{R}\) with its usual (infinite) CW structure by unit intervals has infinitely many 0-cells and 1-cells, so no well-defined alternating sum exists without extra conventions (e.g. compactly supported Euler characteristic, a different invariant). The cell/simplex counts \(c_i(K)\) are all finite.If a single stage of the decomposition has infinitely many cells (e.g. an infinite discrete set of 0-cells), \(\chi\) is undefined by this formula; one must pass to compactly-supported or \(\ell^2\) versions, which are genuinely different invariants. \(K_1, K_2\) triangulate spaces that are homeomorphic (or homotopy equivalent).Without this, there is nothing to compare: a triangulated disk (\(\chi=1\)) and a triangulated circle (\(\chi=0\)) simply have different Euler characteristics, correctly, because the spaces are not homotopy equivalent. The theorem asserts equality only when the underlying spaces agree up to homotopy. Homology groups \(H_i(X;\mathbb{Q})\) are finite-dimensional for all \(i\), and zero for \(i \gt n\).This is automatic once \(K\) is a finite complex (Level A step below), but is the precise algebraic fact that makes the alternating sum in the Euler–Poincaré formula finite; for spaces with infinite-rank homology in some fixed degree the formula breaks down.

Proof

C_i(K) := \text{the } \mathbb{Q}\text{-vector space freely generated by the } i\text{-simplices (or } i\text{-cells) of } K, \qquad c_i(K) = \dim_{\mathbb{Q}} C_i(K).

Definition of the simplicial (cellular) chain complex with rational coefficients; each \(C_i(K)\) is finite-dimensional because \(K\) is a finite complex. A

\partial_i : C_i(K) \to C_{i-1}(K), \qquad \partial_i \partial_{i+1} = 0 \text{ for all } i.

Standard simplicial (or cellular) boundary map, extended linearly from the boundary of each simplex/cell; \(\partial^2=0\) is the defining identity of a chain complex, proved by direct computation on generators (each codimension-2 face is counted twice with opposite sign). A

Z_i := \ker \partial_i \subseteq C_i(K), \qquad B_i := \operatorname{im} \partial_{i+1} \subseteq C_i(K), \qquad B_i \subseteq Z_i \text{ by Step 2}.

\(Z_i\) (cycles) and \(B_i\) (boundaries) are well-defined subspaces because \(\partial_i\) is linear and \(\partial_i\partial_{i+1}=0\). A

H_i(K;\mathbb{Q}) := Z_i / B_i, \qquad b_i(K) := \dim_{\mathbb{Q}} H_i(K;\mathbb{Q}) \ (\text{the } i\text{-th Betti number}).

Definition of simplicial homology with rational coefficients; \(b_i(K)\) is finite since \(Z_i \subseteq C_i(K)\) is finite-dimensional. A

\dim C_i = \dim Z_i + \dim(\operatorname{im}\partial_i) = \dim Z_i + \dim B_{i-1}.

Rank–nullity theorem applied to the linear map \(\partial_i : C_i \to C_{i-1}\): \(\dim C_i = \dim\ker\partial_i + \dim\operatorname{im}\partial_i\), and \(\operatorname{im}\partial_i = B_{i-1}\) by definition (Step 3). C

\dim Z_i = b_i(K) + \dim B_i \qquad \text{(from } H_i = Z_i/B_i \text{)}.

Rank–nullity again, this time for the quotient map \(Z_i \to Z_i/B_i = H_i(K;\mathbb{Q})\): \(\dim Z_i = \dim B_i + \dim(Z_i/B_i)\). B

c_i(K) = \dim C_i = b_i(K) + \dim B_i + \dim B_{i-1}.

Substitute Step 6 into Step 5. This is the key algebraic identity: cell count in degree \(i\) splits into "genuine homology" plus two boundary-rank correction terms, one shared with degree \(i+1\) and one with degree \(i-1\). B

\chi(K) = \sum_{i=0}^n (-1)^i c_i(K) = \sum_{i=0}^n (-1)^i \Big( b_i(K) + \dim B_i + \dim B_{i-1} \Big) = \sum_{i=0}^n (-1)^i b_i(K),

Substitute Step 7 into the definition of \(\chi(K)\) and observe the sum telescopes: each \(\dim B_i\) appears once as \((-1)^i \dim B_i\) (from degree \(i\)) and once as \((-1)^{i+1}\dim B_i\) (from degree \(i+1\)), cancelling exactly; the boundary terms \(B_{-1}=0\) and \(B_n=0\) (no \((n{+}1)\)-cells) vanish at the ends. This is the Euler–Poincaré formula. B

H_i(K;\mathbb{Q}) \cong H_i(|K|;\mathbb{Q}) \quad \text{(simplicial homology of } K \cong \text{ singular homology of the underlying space } |K|).

Theorem (simplicial–singular comparison): for any simplicial complex, the natural chain map from simplicial to singular chains induces an isomorphism on homology in every degree. This is a substantial named theorem in its own right (proved via barycentric subdivision and the excision/Mayer–Vietoris machinery, or via acyclic models); it is cited here rather than reproved. C

X_1 \cong X_2 \ \text{(or } X_1 \simeq X_2 \text{)} \implies H_i(X_1;\mathbb{Q}) \cong H_i(X_2;\mathbb{Q}) \ \text{for all } i.

Singular homology \(H_i(-;\mathbb{Q})\) is a functor from topological spaces (or the homotopy category) to \(\mathbb{Q}\)-vector spaces, and functors send isomorphisms to isomorphisms; homotopy invariance of singular homology (a standard theorem: homotopic maps induce equal maps on \(H_*\), proved via the prism/subdivision operator constructing a chain homotopy) upgrades this to homotopy equivalences, not just homeomorphisms. C

\chi(K_1) \overset{\text{Step 8}}{=} \sum_i (-1)^i b_i(K_1) \overset{\text{Step 9}}{=} \sum_i (-1)^i b_i(X_1) \overset{\text{Step 10}}{=} \sum_i (-1)^i b_i(X_2) \overset{\text{Step 9}}{=} \sum_i (-1)^i b_i(K_2) \overset{\text{Step 8}}{=} \chi(K_2).

Chain the four preceding identities: \(\chi\) of a triangulation equals the alternating sum of Betti numbers of the underlying space (Steps 8–9), and that alternating sum is a homotopy invariant of the space alone (Step 10), so it cannot distinguish between two triangulations of homeomorphic/homotopy-equivalent spaces. B

Result

\chi(K) = \sum_{i=0}^{n}(-1)^i c_i(K) = \sum_{i=0}^{n}(-1)^i \dim_{\mathbb{Q}} H_i(X;\mathbb{Q}), \qquad \text{independent of the triangulation } K \text{ of } X.

Reading. No matter how you chop a (compact, triangulable) space up into vertices, edges, faces, and higher cells, the alternating sum \(V - E + F - \cdots\) always comes out the same number, because it secretly measures something the chopping cannot change: the ranks of the space's homology groups.

Scope. Applies to any compact space admitting a finite CW or simplicial structure — polyhedra, triangulated surfaces and manifolds, finite graphs, finite simplicial complexes generally. It does not by itself apply to non-compact or infinite-complex spaces (\(\chi\) may be undefined there) and it only asserts equality between triangulations of the same (or homotopy-equivalent) space, not between arbitrary spaces.

Corollaries & converses

\(\chi\) is a homotopy invariant, hence a fortiori a homeomorphism invariant: homotopy-equivalent finite complexes have equal Euler characteristic (this is exactly Step 10–11).
For a connected graph (1-dimensional complex), \(\chi = V - E = 1 - b_1\), so \(b_1\) (the number of independent cycles) is recoverable from a single vertex/edge count.
Euler's polyhedral formula \(V-E+F=2\) for convex polyhedra is the special case \(X = S^2\), since every convex polyhedron's boundary is a triangulable 2-sphere.
Converse fails: equal Euler characteristic does not imply homotopy equivalence. The circle \(S^1\) and any wedge of a sphere and... more simply, \(S^1 \times S^1\) minus a point has \(\chi=-1\)... the cleanest counterexample: the Möbius band and the annulus both have \(\chi = 0\) (both homotopy equivalent to \(S^1\)) yet are non-homeomorphic as manifolds-with-boundary in the orientable sense — but for a sharper converse failure, note \(\mathbb{RP}^2 \# \mathbb{RP}^2\) (Klein bottle, \(\chi=0\)) and the torus (\(\chi=0\)) have equal Euler characteristic but are not homeomorphic (one is orientable, the other is not). So \(\chi\) is an invariant, not a complete one.

Fails without

Compactness / finiteness dropped: give \(\mathbb{R}\) the CW structure with 0-cells at every integer and 1-cells \([n,n+1]\). Both \(c_0\) and \(c_1\) are (countably) infinite, so \(\sum(-1)^i c_i\) is not an absolutely convergent, well-defined number by this formula — the theorem's conclusion (a single well-defined invariant number) simply has no object to refer to.
Same-space hypothesis dropped: compare a triangulated disk \(D^2\) (\(V-E+F=1\), e.g. one triangle: \(V=3,E=3,F=1\)) with a triangulated circle \(S^1\) (e.g. a triangle boundary: \(V=3,E=3\), \(\chi=0\)). These are two "reasonable" one-dimension-apart cell structures with different \(\chi\), which is entirely consistent with the theorem — but if one mistakenly expected \(\chi\) to be a universal constant independent of the space itself (rather than of the triangulation of a fixed space), this pair shows that expectation is false.
Homotopy type changed by a "small" modification: take \(S^2\) (\(\chi=2\)) and remove a single point to get \(\mathbb{R}^2\) (\(\chi = 1\), since \(\mathbb{R}^2 \simeq \text{point}\)). A seemingly negligible change to the space (one point) changes \(\chi\), confirming \(\chi\) tracks homotopy type exactly, not "how similar the picture looks."

Common errors

Computing \(V-E+F\) on a non-triangulation (a "cell complex" where two faces share more than one edge, or an edge that isn't attached at two distinct vertices) and expecting the invariance theorem to still apply; the identity in Step 2 (\(\partial^2=0\)) and the simplicial approximation comparison (Step 9) genuinely require a valid CW/simplicial structure.
Concluding from \(\chi(X)=\chi(Y)\) that \(X \simeq Y\) — the theorem is one-directional (Step 10 shows homotopy equivalence \(\Rightarrow\) equal \(\chi\), not the converse); see Klein bottle vs. torus above.
Forgetting orientation/boundary is irrelevant to \(\chi\): assuming a surface with boundary must have \(\chi=0\) like a closed orientable surface of genus 1; in fact a disk has \(\chi=1\), a Möbius band has \(\chi=0\), a pair of pants has \(\chi=-1\), etc. — boundary and orientability are separate data from \(\chi\).
Sign errors in the alternating sum for complexes of dimension \(\geq 3\): forgetting that 3-cells are subtracted-then-added pattern continues as \(-,+,-,+,\ldots\), not resetting.
Treating the Betti number identity \(\chi = \sum(-1)^i b_i\) as the definition of \(\chi\) rather than a theorem about it, then being unable to explain why the combinatorial count \(V-E+F\) should match it without redoing this proof.

Discussion

Euler's original 1750s observation for convex polyhedra (\(V-E+F=2\)) predates any concept of homology by over a century; it was Poincaré who, in developing what became homology theory, recognised that Euler's combinatorial invariant was the shadow of an algebraic one. The identity proved in Step 8 is literally called the Euler–Poincaré formula in his honour, and it is the historical bridge between 18th-century polyhedral combinatorics and 20th-century algebraic topology.

The proof structure here — reduce a geometric/combinatorial invariant to an algebraic one via a chain complex, then invoke a comparison theorem (simplicial \(\cong\) singular homology) to transport topological invariance from the algebraic side back to the geometric side — is the template for essentially all of homological algebra's applications in topology: intersection numbers, the Lefschetz fixed-point theorem, and de Rham's theorem for smooth manifolds all follow the same three-step choreography (build a chain/cochain complex from geometric data, prove an algebraic telescoping identity, invoke a comparison theorem).

One subtlety glossed over at Level A: the argument as given uses \(\mathbb{Q}\)-coefficients throughout, which sidesteps torsion phenomena in integral homology entirely. Working over \(\mathbb{Z}\), one must instead use the structure theorem for finitely generated abelian groups, writing \(H_i(K;\mathbb{Z}) \cong \mathbb{Z}^{b_i} \oplus T_i\) with \(T_i\) finite torsion; the same telescoping argument (Steps 5–8) goes through verbatim on the free ranks \(b_i\), because rank is additive on short exact sequences of finitely generated abelian groups exactly as dimension is additive for vector spaces, while the torsion contributes nothing to \(\chi\) at all — a genuinely pleasant fact requiring the universal coefficient theorem to see cleanly.

A common misconception worth flagging explicitly: students sometimes think the Euler characteristic is defined as \(\sum(-1)^ib_i\) and that "invariance under triangulation" is therefore automatic or vacuous. It is not vacuous: the combinatorial definition \(\sum(-1)^ic_i(K)\) is prior, elementary, and a priori triangulation-dependent; the entire content of this theorem is the non-obvious fact (Steps 5–11) that this elementary combinatorial count always agrees with the homotopy-invariant Betti-number sum, however the triangulation is chosen.

Worked examples

\text{Tetrahedral triangulation of } S^2: \quad V=4,\ E=6,\ F=4.

Standard boundary complex of a tetrahedron: 4 vertices, each pair joined by an edge (\(\binom{4}{2}=6\)), 4 triangular faces. A

\chi = V - E + F = 4 - 6 + 4 = 2.

Direct substitution into the combinatorial definition of \(\chi\). A

\text{Octahedral triangulation of } S^2: \quad V=6,\ E=12,\ F=8, \qquad \chi = 6-12+8 = 2.

A different, finer triangulation of the same space \(S^2\) (octahedron boundary). By the theorem (Step 11) this must agree with the tetrahedral count, and it does: both equal \(\chi(S^2) = 2 = b_0 - b_1 + b_2 = 1 - 0 + 1\), matching the known homology \(H_0(S^2;\mathbb{Q})=\mathbb{Q}\), \(H_1(S^2;\mathbb{Q})=0\), \(H_2(S^2;\mathbb{Q})=\mathbb{Q}\). B

\chi(S^2) = 2, \text{ confirmed by two independent triangulations, as guaranteed by the theorem.}

Reading. Any polyhedron topologically a sphere — however many vertices, edges and faces it is built from — satisfies \(V-E+F=2\); this is Euler's original formula, recovered as a corollary.

\text{Minimal CW structure on the torus } T^2: \quad 1 \text{ vertex}, \ 2 \text{ edges } (a,b), \ 1 \text{ face (attached via } aba^{-1}b^{-1}).

Standard "square with opposite sides identified" CW model of the torus: all four corners of the square become the single vertex, the two pairs of identified sides become the two edges \(a,b\), and the interior of the square is the single 2-cell. A

\chi(T^2) = 1 - 2 + 1 = 0.

Direct substitution into \(\chi = c_0 - c_1 + c_2\) for this CW structure. A

\text{A genuine simplicial triangulation of } T^2 \text{ (e.g. the standard minimal one): } V=7,\ E=21,\ F=14.

A concrete simplicial complex triangulating the torus (7 vertices is the minimum possible for a simplicial, as opposed to CW, triangulation of \(T^2\)); each vertex has degree 6 by symmetry, giving \(E = 7\cdot 6/2=21\), and each face is a triangle with \(3F = 2E \Rightarrow F = 14\). B

\chi = 7 - 21 + 14 = 0,

matching Step 2 exactly, as the theorem requires since both structures triangulate the same space \(T^2\). A

\chi(T^2) = 0, \text{ from a 4-cell CW structure and a 42-cell simplicial triangulation alike.}

Reading. The torus's Euler characteristic \(0\) reflects its homology \(b_0=1,\ b_1=2,\ b_2=1\) (\(1-2+1=0\)), and this number cannot be changed by triangulating more finely — a direct illustration of the theorem with two triangulations of very different sizes.

Problems

A single filled triangle (2-simplex together with its faces) has \(V=3,E=3,F=1\). Compute \(\chi\) and interpret it in terms of the space's homotopy type.
Solution
\(\chi = 3-3+1=1\). The filled triangle is homeomorphic to a disk \(D^2\), which is contractible (\(\simeq\) a point), so \(b_0=1\) and \(b_i=0\) for \(i\geq 1\), giving \(\sum(-1)^ib_i = 1\), matching \(\chi=1\) exactly as the theorem predicts.
Using the connected-sum formula for closed surfaces \(\chi(X \# Y) = \chi(X) + \chi(Y) - \chi(S^2)\) (removing a disk from each, \(\chi(\text{disk})=1\), and gluing along the resulting circle boundaries, \(\chi(S^1)=0\), by inclusion–exclusion \(\chi(X\#Y)=\chi(X)+\chi(Y)-\chi(S^1)\), so more precisely \(\chi(X\#Y) = \chi(X)+\chi(Y)-2\)), compute \(\chi(T^2 \# T^2)\), the genus-2 surface.
Solution
\(\chi(T^2)=0\) for each torus (Worked Example 2). By inclusion–exclusion on the connected sum (cut a disk, \(\chi\) drops by \(\chi(D^2)-\chi(S^1) = 1-0=1\), from each summand, then glue): \(\chi(T^2\#T^2) = \chi(T^2)+\chi(T^2)-\chi(S^2) = 0+0-2=-2\). This matches the general genus-\(g\) surface formula \(\chi = 2-2g\) with \(g=2\): \(2-4=-2\). Because connected sum only depends on the homeomorphism types being combined, and \(\chi\) is a homeomorphism invariant by the theorem, this value is independent of which particular triangulation of the genus-2 surface one draws.
Explain, using the theorem, why \(\chi\) of the product CW complex \(X \times Y\) satisfies \(\chi(X\times Y) = \chi(X)\cdot\chi(Y)\), and use it to compute \(\chi(T^2 \times S^1)\).
Solution
If \(K\) has cells indexed by \(\{\sigma_i\}\) and \(L\) by \(\{\tau_j\}\), the product CW structure on \(K\times L\) has cells \(\sigma_i \times \tau_j\) of dimension \(\dim\sigma_i + \dim\tau_j\), so \(c_k(K\times L) = \sum_{p+q=k} c_p(K)c_q(L)\). Then \(\chi(K\times L) = \sum_k(-1)^k c_k(K\times L) = \sum_{p,q}(-1)^{p+q}c_p(K)c_q(L) = \big(\sum_p(-1)^pc_p(K)\big)\big(\sum_q(-1)^qc_q(L)\big) = \chi(K)\chi(L)\), a purely combinatorial identity that (by the theorem) computes a topological invariant of \(X\times Y\) regardless of which product cell structure is used. For \(T^2\times S^1\): \(\chi(T^2)=0\), \(\chi(S^1)=0\) (1 vertex, 1 edge: \(1-1=0\)), so \(\chi(T^2\times S^1)=0\cdot 0=0\).
The complete graph \(K_5\) (5 vertices, all \(\binom{5}{2}=10\) edges present) is a 1-dimensional CW/simplicial complex. Compute \(\chi(K_5)\) and use \(\chi = b_0 - b_1\) for a connected graph to find its first Betti number (the number of independent cycles / rank of \(H_1\)).
Solution
\(\chi(K_5) = V - E = 5 - 10 = -5\). Since \(K_5\) is connected, \(b_0=1\). From \(\chi = b_0-b_1\): \(-5 = 1-b_1 \Rightarrow b_1 = 6\). Sanity check via spanning tree: a spanning tree on 5 vertices has \(4\) edges; the remaining \(10-4=6\) edges each close off one independent cycle, giving \(b_1=6\) directly — matching the value predicted purely from the invariant \(\chi\).
Take a filled square (a disk) with the obvious minimal CW structure \(V=4,E=4,F=1\) (\(\chi = 4-4+1=1\)). Now subdivide it by adding one interior vertex connected by 4 new edges to the corners, splitting the single face into 4 triangular faces. Recompute \(\chi\) for the subdivided structure and confirm invariance.
Solution
New counts: \(V' = 4+1=5\) (four original corners plus the new interior vertex); \(E' = 4 \text{ (original boundary edges)} + 4 \text{ (new spokes)} = 8\); \(F' = 4\) (four triangles). Then \(\chi' = 5-8+4=1\), identical to the original \(\chi=1\), exactly as the theorem guarantees since both structures triangulate the same underlying disk (which is contractible, \(b_0=1\), \(b_i=0\) for \(i\geq1\), so \(\sum(-1)^ib_i=1\) independent of which of the two triangulations one used to compute it).