Counting Cells Through a Connected Sum
Euler characteristic is useful in surface constructions because it can be computed from a finite cell decomposition: vertices contribute \(+1\), edges contribute \(-1\), and faces contribute \(+1\). A connected sum changes a surface in two places at once: one disk is removed from each surface, and the resulting boundary circles are identified. To find the net change, we must count both operations rather than treat the gluing as an informal joining of pictures.
Throughout, a surface means a compact connected surface without boundary. For two such surfaces \(M\) and \(N\), their connected sum \(M\#N\) is formed by removing the interior of a disk from each and identifying the two boundary circles by a homeomorphism. The identification is chosen so that the result is a surface. The Euler characteristic \(\chi(M)\) is the alternating sum \(V-E+F\) from a finite cell decomposition of \(M\), where \(V,E,F\) are its numbers of vertices, edges and two-dimensional cells. The invariance of this number under subdivision is established in the earlier material on Euler characteristic.
We may choose finite cell decompositions adapted to the disks being removed: each disk is a two-dimensional cell, and its boundary is a subcomplex. Subdivision lets us use the same number \(k\) of boundary edges on both circles, with \(k\) boundary vertices on each. We take \(k\geq 3\), which is always possible by subdividing the boundary. The cells outside the disk are retained after its interior is removed. This is the cell-counting setup for the proof; the final result does not depend on the particular subdivision.
The Connected-Sum Formula
Proof. Choose cell decompositions adapted to the removed disks, with \(k\) boundary vertices and \(k\) boundary edges on each disk. Write \(V_M,E_M,F_M\) for the cell counts on \(M\) before removing its disk, and \(V_N,E_N,F_N\) for the counts on \(N\). Removing each disk deletes the interior two-dimensional cell but leaves the boundary vertices and edges. Thus the two punctured surfaces have counts $$ (V_M,E_M,F_M-1) \quad\text{and}\quad (V_N,E_N,F_N-1). $$ When the boundary circles are identified, their \(k\) pairs of vertices become \(k\) vertices, reducing the combined vertex count by \(k\). Their \(k\) pairs of edges become \(k\) edges, reducing the combined edge count by \(k\). No two-dimensional cells are identified. Consequently, $$ \begin{aligned} V_{M\#N}&=V_M+V_N-k,\\ E_{M\#N}&=E_M+E_N-k,\\ F_{M\#N}&=(F_M-1)+(F_N-1). \end{aligned} $$ Substituting these counts into the definition gives $$ \begin{aligned} \chi(M\#N) &=(V_M+V_N-k)-(E_M+E_N-k)+(F_M-1)+(F_N-1)\\ &=(V_M-E_M+F_M)+(V_N-E_N+F_N)-2\\ &=\chi(M)+\chi(N)-2. \end{aligned} $$ The two boundary corrections, \(-k\) for vertices and \(-k\) for edges, cancel because vertices enter positively and edges negatively. The two removed disk cells account for the remaining \(-2\). \(\square\)
This proof is a direct cell count. It does not require the boundary identification to preserve an orientation: the count uses only the fact that the two boundary circles are identified cell by cell. Nor does it require the surfaces to be orientable.
Worked Checks on Familiar Surfaces
Worked Example: The Sphere Connected-Summed with the Torus
Compute \(\chi(S^2\#T^2)\), where \(S^2\) is the sphere and \(T^2\) is the torus. The sphere has a cell decomposition with one vertex, no edges and one face, so $$ \chi(S^2)=1-0+1=2. $$ The standard square cell structure for the torus has one vertex, two edges and one face, so $$ \chi(T^2)=1-2+1=0. $$ Apply the connected-sum formula: $$ \chi(S^2\#T^2)=2+0-2=0. $$ The arithmetic uses the sphere contribution \(2\), the torus contribution \(0\), and the subtraction \(2\) from removing one disk in each surface. As a check, the net result equals \(\chi(T^2)\); this is consistent with the familiar fact that taking a connected sum with a sphere does not change the surface's homeomorphism type.
Worked Example: The Torus Connected-Summed with Itself
For each torus, the one-vertex, two-edge, one-face structure gives $$ \chi(T^2)=1-2+1=0. $$ Therefore $$ \chi(T^2\#T^2)=0+0-2=-2. $$ We can also check the count by using cell structures adapted to the removed disks. Let the punctured copies have counts \((V_1,E_1,F_1)\) and \((V_2,E_2,F_2)\). The disk-removal count gives $$ V_1-E_1+F_1=-1,\qquad V_2-E_2+F_2=-1, $$ since each torus starts with Euler characteristic \(0\) and loses one face. After gluing, the boundary vertices and edges cancel in the alternating count, so the total is $$ (-1)+(-1)=-2. $$ This agrees with the formula computed from the two closed tori.
Worked Example: The Projective Plane Connected-Summed with the Klein Bottle
The standard polygon cell structure for the projective plane has one vertex, one edge and one face. Hence $$ \chi(\mathbb{R}P^2)=1-1+1=1. $$ The standard square cell structure for the Klein bottle has one vertex, two edges and one face. Hence $$ \chi(K)=1-2+1=0. $$ It follows that $$ \chi(\mathbb{R}P^2\#K)=1+0-2=-1. $$ The two removed disk faces account for the decrease from \(1+0=1\) to \(-1\). In particular, the formula applies even though both input surfaces are nonorientable; orientability is not a hypothesis of the cell count.
Why the Boundary Gluing Contributes Zero
The cancellation in the proof can also be understood by counting the cells in a circle. A circle subdivided into \(k\) edges has \(k\) vertices and \(k\) edges, so its alternating count is $$ k-k=0. $$ When the boundary circles are identified, one copy of each boundary vertex and edge is lost from the combined count. The contribution of this loss to the alternating sum is \(-k-(-k)=0\). Thus the gluing itself contributes no net change to Euler characteristic. The decrease comes entirely from the two disk interiors that were removed before the gluing.
This explanation concerns the boundary circle as a one-dimensional cell structure. It does not say the circles are absent from the connected sum: their points are present, but paired points are identified. Counting vertices and edges separately records that identification exactly.
A Common Counting Error
Common pitfall: subtracting \(1\) instead of \(2\). One might reason that the boundary gluing has Euler characteristic zero and conclude that the connected sum has the same Euler characteristic as the sum of the original values. That overlooks the disk removals. Each removed open disk contributes one deleted two-dimensional cell, so two cells are deleted altogether.
For example, using the torus cell count gives \(\chi(T^2)=1-2+1=0\). The incorrect calculation would be $$ \chi(T^2\#T^2)=0+0-1=-1, $$ which subtracts only one for the two disk removals. The correct calculation is $$ \chi(T^2\#T^2)=0+0-2=-2. $$ The correction can be checked before gluing: removing one disk from each torus changes each Euler characteristic from \(0\) to \(0-1=-1\). The two punctured surfaces therefore have combined alternating count \((-1)+(-1)=-2\). Identifying the boundary circles changes the vertex and edge counts by equal amounts, so it leaves this value at \(-2\), not \(-1\).
The formula is especially useful as a consistency check after a surface has been built by cutting and gluing. It predicts the Euler characteristic from the two input surfaces without requiring a complete cell decomposition of the final surface. It does not, by itself, identify the resulting surface: different surfaces can share an Euler characteristic, so additional information is needed for that task.
Check Your Understanding
In each question, state the Euler characteristic values you use and show the arithmetic in the connected-sum formula.
- Compute the Euler characteristic of the connected sum of two projective planes, using \(\chi(\mathbb{R}P^2)=1\).
- Compute \(\chi(S^2\#S^2)\), using the one-vertex, zero-edge, one-face cell structure for each sphere.
- Explain why identifying boundary circles with \(k\) edges and \(k\) vertices does not change the alternating cell count.
- A calculation gives \(\chi(M\#N)=\chi(M)+\chi(N)-1\). Which part of the construction has been undercounted?
- Does the cell-counting formula require either surface to be orientable? Justify your answer from the proof.