Why Euler Characteristic Survives Subdivision

Why Subdivision Does Not Change the Alternating Count

A polygonal decomposition records a space by counting its vertices, edges, and faces. Refining the decomposition changes those counts, sometimes substantially. Yet the alternating count “vertices minus edges plus faces” remains unchanged. The reason is local: the elementary operations used to subdivide add cells in combinations whose alternating contributions cancel.

Throughout, consider a finite polygonal decomposition of a compact surface, or of a polygonal disk. Its vertices are its \(0\)-cells, its edges are its \(1\)-cells, and its faces are its \(2\)-cells. If their numbers are \(V\), \(E\), and \(F\), respectively, its Euler characteristic is \(\chi=V-E+F\). A subdivision replaces cells by smaller polygonal cells without changing the underlying space. We count cells, not their corners before gluing: if vertices or edges are identified in the decomposition, each resulting cell is counted once.

To prove invariance, we will track the change in each count. We first examine three elementary moves. For a face cut, the arc endpoints must be existing vertices of the face boundary. If an endpoint instead lies in the interior of a boundary edge, that edge must be subdivided first; this distinction affects the vertex and edge counts.

The Elementary Moves

Proof. The original edge is replaced by two edges meeting at the new vertex. Thus \(V\) increases by \(1\), \(E\) increases by \(1\), and no face is created or removed, so \(F\) does not change. Consequently \(\Delta\chi=\Delta V-\Delta E+\Delta F=1-1+0=0\). \(\square\)

Proof. The two endpoints are already vertices, so no vertex is added. The arc becomes one new edge, and the original face is replaced by two faces. Hence \(\Delta V=0\), \(\Delta E=1\), and \(\Delta F=1\), giving \(\Delta\chi=0-1+1=0\). The asserted division of a polygonal disk by a simple arc is the standard polygonal Jordan separation fact; its general proof is not needed here. \(\square\)

Proof. The interior point adds one vertex, and the \(k\) arcs add \(k\) edges. Replacing one face by \(k\) faces increases the face count by \(k-1\). Therefore \(\Delta\chi=1-k+(k-1)=0\). \(\square\)

From Elementary Moves to a General Face Subdivision

The preceding counts explain the usual local operations. To apply them to an arbitrary finite polygonal subdivision, we also need to see why the subdivision can be assembled using such operations. Work first inside one original polygonal face, viewed as a disk. Include every point where the subdivision divides its boundary as a boundary vertex, splitting the boundary edges there first. The target subdivision then has a finite polygonal graph inside the disk: its vertices and edges, together with the subdivided boundary.

We use two elementary facts about this graph. Its vertices and edges form a connected graph: otherwise a component disconnected from the boundary would enclose a region and leave a complementary region that is not a disk, contrary to the requirement that the subdivision faces be polygonal disks. Also, adding a tree edge from a new vertex to an existing vertex does not split a region, whereas adding a remaining edge whose interior avoids the graph splits one region into two. For the second fact, the edge together with a path already in the graph forms a polygonal closed curve; the polygonal Jordan separation fact gives the split. These facts let us order the construction as follows.

Why the construction gives the target subdivision. All target vertices are included by the end of Step 2, and all target edges by the end of Step 3. The regions left by the complete target graph are exactly the polygonal faces of the subdivision. Every step preserves the alternating count, so the total refinement of the original face preserves it as well.

Proof. Refine one original face at a time. Subdividing the boundary edges of the face preserves \(\chi\) by Proposition 1. The spanning-tree construction adds each interior vertex together with one edge, giving \((\Delta V,\Delta E,\Delta F)=(1,1,0)\) at that step. Each remaining edge splits one region, giving \((0,1,1)\). In each case the change \(\Delta V-\Delta E+\Delta F\) is zero. Thus the subdivision within that face preserves the alternating count. Applying this argument to every face gives total change zero: the counts are additive across the local changes, including when boundary cells are shared, since each shared cell’s net change is counted just once in the resulting decomposition. Therefore the original and subdivided decompositions have equal Euler characteristic. \(\square\)

Endpoints in Edge Interiors: A Common Pitfall

A frequent counting error is to treat any point of a face boundary as if it were already a vertex. It is not: a point in the interior of a boundary edge must first be made a vertex by subdividing that edge. Only after this preliminary step can the face-cut count \((0,1,1)\) be applied.

What the Invariance Does—and Does Not—Say

The individual cell counts are not invariants of the underlying space: the diagonal cut of a square changed \(E\) and \(F\), and the center-and-spokes subdivision changed all three counts. What survives these refinements is their alternating combination. This is why two different polygonal descriptions can be compared through \(\chi\) even when their numbers of cells differ.

The argument concerns actual subdivisions: the underlying space stays fixed, and the new cells refine the old ones. It does not say that every change to a polygon diagram preserves the alternating count. For example, removing an edge without replacing it by a subdivision operation changes the cell structure in a way not covered by the proof. Nor does the argument identify two spaces just because their Euler characteristics agree; it establishes only that the value computed from a decomposition is unaffected by subdividing that decomposition.