Why Polygon Gluing Diagrams Need More Than a Visual Check
A polygon gluing diagram records more than which edges are paired. The direction of each pairing determines which endpoints become identified, and those identifications determine the vertices and the neighborhoods in the quotient. A mistaken arrow can therefore change the space being described, even when the drawing still looks plausible.
We use a polygonal disk whose boundary edges are paired in pairs by homeomorphisms. The quotient identifies points on paired edges according to those maps. An edge label records a paired edge, and the exponent \(+1\) or \(-1\) indicates whether that occurrence is traversed with or against the chosen direction for the label. Thus \(a\) and \(a^{-1}\) are not interchangeable: they prescribe different endpoint correspondences.
The phrase “smallest equivalence relation” matters. Endpoint identifications must be followed transitively: if one pairing identifies \(v_0\) with \(v_1\), and another identifies \(v_1\) with \(v_2\), then all three corners represent one quotient vertex. Counting the corners in the original polygon is not the same as counting quotient vertices.
Three Checks That Should Be Kept Separate
For a closed surface built from a polygon with every edge paired exactly once with another edge, check the construction in this order:
- Pairing check: each boundary edge occurs in exactly one pair, and the arrow specifies the endpoint map.
- Vertex and local-neighborhood check: compute the vertex equivalence classes, then inspect the link at each quotient vertex.
- Orientation and Euler-characteristic checks: test whether orientations fit across every paired edge, and calculate \(V-E+F\) from the quotient cells.
These checks answer different questions. A correct Euler characteristic does not establish that the quotient is a surface, and the Euler characteristic alone does not determine orientability or identify a surface.
Proof. In a sufficiently small neighborhood of a polygon corner, the quotient looks like a short radial segment from the vertex in each direction represented by that corner. Taking all corners in the vertex class and applying the endpoint identifications gives a neighborhood shaped like the cone on the link: each point of the link is joined to the cone point representing the quotient vertex. If the link is a circle, its cone is a disk, so a neighborhood of the vertex is an open disk after restricting to a sufficiently small part of that cone. Conversely, if the link is disconnected, deleting the vertex from every sufficiently small conical neighborhood leaves more than one component, whereas a punctured open disk is connected. If the link is not a circle, the conical neighborhood is not a disk neighborhood: at a link endpoint the cone has a boundary, and at a branching point it has more than two local directions. Thus a disk neighborhood occurs exactly when the link is a single circle. \(\square\)
In a complete edge pairing, link endpoints are joined in pairs, so every vertex of the link graph has degree two (with a loop counted twice). Each connected component is therefore a circle. The practical issue is to check that the link of a given quotient vertex is one circle, rather than several circles. Several circles mean that the quotient vertex does not have a disk neighborhood.
Proof. Orient the interior of the polygon. Its boundary traversal induces a direction on each occurrence of each edge. For the polygon orientation to extend across a glued edge to a consistent orientation of the surface, the two induced boundary directions must be identified in opposite directions: the two polygon sides lie on opposite sides of the seam. When the two occurrences have opposite exponents, the edge-pairing map identifies their boundary traversal directions oppositely, as required. If this holds for every pair, the polygon orientation extends consistently across every seam, giving an orientation of the surface. If a pair has equal exponents, its two boundary traversal directions are identified in the same direction, so the polygon orientation cannot extend across that seam. Reversing the orientation of the polygon reverses both directions and does not change whether the two exponents agree. Since the surface has only this one face, any orientation of the surface would restrict to one of the two orientations of its interior. Neither can extend across a same-exponent pair. Thus no orientation exists in that case. \(\square\)
Worked Examples: Trace the Identifications
Worked Example: The Paired Square for the Torus
Take a square with boundary word \(aba^{-1}b^{-1}\), read cyclically in the direction around its boundary. Let its corners, in order, be \(v_0,v_1,v_2,v_3\), with \(v_4=v_0\). Its four sides run from \(v_0\) to \(v_1\), from \(v_1\) to \(v_2\), from \(v_2\) to \(v_3\), and from \(v_3\) to \(v_0\).
Vertex count. The \(a\)-pair has opposite exponents. The first occurrence starts at \(v_0\), while the \(a^{-1}\) occurrence ends at \(v_3\), so \(v_0\sim v_3\). The first occurrence ends at \(v_1\), while the \(a^{-1}\) occurrence starts at \(v_2\), so \(v_1\sim v_2\). The \(b\)-pair also has opposite exponents: \(v_1\sim v_0\) and \(v_2\sim v_3\). Together these relations put all four corners in one equivalence class. Hence \(V=1\). There are two paired edges, so \(E=2\), and one face, so \(F=1\).
Local check. Call the corner intervals \(c_0,c_1,c_2,c_3\) in cyclic order. The endpoint identifications join them in the cycle \(c_0,c_3,c_2,c_1,c_0\). This is one circle, so the vertex has a disk neighborhood.
Orientation and Euler checks. Both labels occur once with each exponent, so the orientation test succeeds. The calculation is \(\chi=V-E+F=1-2+1=0\). Thus the checks agree with the torus model.
Check: the counts use quotient cells, not the four original corners: \(1-2+1=0\). Counting four distinct corners would instead give \(4-2+1=3\), but those corners are not distinct quotient vertices.
Worked Example: The Projective-Plane Word \(aa\)
Take a polygonal disk with two boundary sides, both labeled \(a\), and pair the sides in the direction indicated by the word \(aa\). The two sides are traversed with the same exponent.
Vertex count. Write the corners in boundary order as \(v_0,v_1,v_2\). The first \(a\)-side runs from \(v_0\) to \(v_1\); the second runs from \(v_1\) to \(v_2\). The pairing identifies start with start and end with end. Thus \(v_0\sim v_1\) and \(v_1\sim v_2\). Transitivity gives one quotient vertex, so \(V=1\). There is one paired edge, \(E=1\), and one face, \(F=1\). Therefore \(\chi=1-1+1=1\).
Local and orientation checks. The two corner intervals join end to end under the endpoint identifications, forming one circle; the quotient vertex therefore has a disk neighborhood. Both occurrences have exponent \(+1\), so the orientation test fails. This is the standard polygon construction of the projective plane: it is a surface, but it is nonorientable.
Check: the single vertex is consistent with the explicit chain \(v_0\sim v_1\sim v_2\), and the Euler count recomputes as \(1-1+1=1\).
Worked Example: The Klein-Bottle Word \(aba^{-1}b\)
Use a square with boundary word \(aba^{-1}b\), and name its successive corners \(v_0,v_1,v_2,v_3\), with \(v_4=v_0\).
Vertex count. For the \(a\)-pair, the first occurrence runs from \(v_0\) to \(v_1\), while the \(a^{-1}\) occurrence runs from \(v_2\) to \(v_3\). Opposite exponents identify \(v_0\sim v_3\) and \(v_1\sim v_2\). The two \(b\)-occurrences both have exponent \(+1\), so start is identified with start and end with end: \(v_1\sim v_3\) and \(v_2\sim v_0\). These relations join all four corners into one class. Therefore \(V=1\); there are \(E=2\) paired edges and \(F=1\) face.
Local check. The four corner intervals join in the cycle \(c_0,c_3,c_1,c_2,c_0\). The link is one circle, so the quotient vertex has a disk neighborhood.
Orientation and Euler checks. The \(b\)-pair has equal exponents, so the orientation test fails. The Euler characteristic is \(\chi=1-2+1=0\). The quotient is the Klein bottle: it is a closed surface, but it is nonorientable. This example also shows why \(\chi=0\) cannot by itself distinguish the torus from the Klein bottle.
Check: both edge labels occur twice, so \(E=2\); with the one vertex and one face, \(1-2+1=0\).
Common Mistakes and Their Corrections
Mistake 1: Treating arrows as decoration. In the Klein-bottle word \(aba^{-1}b\), changing the last \(b\) to \(b^{-1}\) changes which endpoints are identified. The actual word has two positive \(b\)-occurrences, so their starts are paired and their ends are paired. Replacing the last occurrence with \(b^{-1}\) instead pairs the start of one occurrence with the end of the other. That is a different quotient; it is not a harmless relabeling of the same diagram.
Mistake 2: Counting polygon corners before taking the quotient. A wrong calculation for the torus square is \(V=4\), \(E=2\), \(F=1\), giving \(\chi=4-2+1=3\). The correction is to use the endpoint relations: \(v_0\sim v_3\), \(v_1\sim v_2\), \(v_1\sim v_0\), and \(v_2\sim v_3\), so all corners are equivalent. Thus \(V=1\) and \(\chi=1-2+1=0\).
Mistake 3: Assuming edge pairings automatically give surface neighborhoods. Pairing every edge ensures that an interior point of a paired edge has two polygon sides around it. At a quotient vertex, however, all corner sectors must fit together into one circle. A link consisting of two circles would describe two disk-like neighborhoods pinched at their central vertex, not one disk neighborhood. Count the corner intervals and trace every endpoint identification; do not infer the answer from the drawing alone.
Mistake 4: Inferring orientability from Euler characteristic. The torus and Klein bottle examples both have \(\chi=0\), but their edge directions give different orientation checks. Use the direction data to test orientability; use \(V-E+F\) only as the Euler-characteristic calculation. Neither calculation replaces the other.
A reliable audit records the endpoint relations explicitly, closes those relations under transitivity, checks each vertex link, and only then reports orientation and Euler characteristic. Keeping the checks separate makes it easier to locate an error: an incorrect vertex count points to endpoint bookkeeping, a failed link test points to local structure, and a failed orientation test points to incompatible directions across a seam.
Check Your Understanding
For each question, write down the endpoint relations before deciding what the diagram represents.
- In the word \(aba^{-1}b^{-1}\), which pairs of polygon corners are identified by the \(a\)-pair, and which by the \(b\)-pair?
- Why does a link with two circle components fail to give a disk neighborhood at the quotient vertex?
- For the word \(aba^{-1}b\), which label pair prevents the one-face orientation from extending consistently?
- A square diagram has four original corners and two paired edges. What additional information is needed before calculating its Euler characteristic?
- Why can two closed surfaces with the same Euler characteristic still differ in orientability?