4

Henning once told me that,

[i]t follows from the Euler characteristic of the plane that the average face degree of a 3-regular planar graph with $F$ faces is $6-12/F$, which means that every 3-regular planar graph has at least one face with degree $5$ or lower.

I tried to understand and extend this and got the following:

Given a $k$-regular graph. Summing over the face degrees $f_n$ gives twice the number of edges $E$ and this is $k$ times the number of vertices $V$: $$ \sum f_n = 2E =kV \tag{1} $$ Further we have Euler's formula saying $$ V+F = E +\chi, $$ where $\chi$ is Euler's characteristic of the surface where the graph lives on. Again we insert $E=\frac k2V$ and get: $$ F=\left( \frac k2 -1 \right)V+\chi\\ V=\frac{F-\chi}{\frac k2 -1}. \tag{2} $$ Dividing $(1)$ by $F$ and inserting $(2)$ gives: $$ \frac{\sum f_n}{F}= \frac{k(F-\chi)}{(\frac k2 -1)F}=\frac {2k}{k -2} \left( 1-\frac{\chi}{F}\right) \tag{3} $$

or

$$ \sum f_n=\frac {2k}{k -2} \big( F-\chi\big). \tag{3$^\ast$} $$

Plug in $k=3$ and $\chi=2$ (the characteristic of the plane), we get back Henning's formula, but when e.g. $\chi=-2$, so the surface we draw could be a double torus, we get the average degree to be:

$$ 6 \left( 1+\frac{2}{F}\right) $$

How to find cubic graphs where the $\min(f_k)>6$ on surfaces with $\chi<0$?

EDIT The graph should not be a snark.

draks ...
  • 18,449
  • I would try to approach it from the other direction, taking the dual graph of a triangulation in which every vertex has valency more than 6. For instance, $K_8$ with 2 extra edges added across the non triangular faces. – jp26 Aug 04 '13 at 17:47
  • 1
    I have been trying all afternoon, but so far not quite there! Here are some links for other people to try too... Double Torus examples: http://users.wpi.edu/~bservat/blanusa08.pdf
    Embedding scheme on page 5 of http://www.math.caltech.edu/~2012-13/2term/ma006b/12Ma6bChp35.pdf
    – jp26 Aug 04 '13 at 20:30
  • 1
    Are you sure $\chi=1$ for the plane and $\chi=-2$ for a double torus uses the same convention for $\chi$? The Wikipedia article seems to use $\chi$ for what you call $1+\chi$. – hmakholm left over Monica Aug 06 '13 at 00:23
  • @HenningMakholm $+1$ for the component... – draks ... Aug 06 '13 at 05:25
  • Note to myself: http://www.personal.kent.edu/~rmuhamma/GraphTheory/MyGraphTheory/embedding.htm – draks ... Aug 18 '13 at 20:54

2 Answers2

2

For orientable surfaces, here's a representative element of a family of non-snarky cubic graphs on an $n$-torus with $4n-2$ vertices, $6n-3$ edges and a single $(12n-6)$-sided face.

drawing of graph goes here

If it is a problem that some pairs of vertices have more than one edge going between them, that can easily be fixed with some local rearrangements (which can be chosen to preserve the green-blue Hamiltonian cycle and thus non-snarkiness).


For non-orientable surfaces, I think it is easiest to start by appealing to the classification theorem and construe the surface as a sphere with $k\ge 2$ cross-caps on it. Now if you start with a planar graph and place a cross-cap straddling one of its edges (so following the edge from one end to another will make you arrive in the opposite orientation), the net effect is to fuse the two faces the edge used to separate.

Therefore, start with an arbitrary planar non-snarky cubic graph with $2k-2$ vertices and $3k-3$ edges. Select an arbitrary spanning tree (comprising $2k-3$ edges), and place cross-caps on the $k$ edges not in the spanning tree. This will fuse all of the original graph's faces, and we're left with a graph with a single $(6k-6)$-sided face.


In each of the above cases, if you want more faces, simply subdivide the single one you've already got with new edges. The face itself, before you glue its edges together to form the final closed surface, is just a plain old $(12n-6)$- or $(6k-6)$-gon, so you can design the subdivision as a plane drawing.

  • @draks: Answer updated to give a construction for non-orientable surfaces too. – hmakholm left over Monica Aug 06 '13 at 12:50
  • Henning, thanks again. Do you think it's possible to construct a non-hamiltonian example on a (non-)orientable surface, starting with let's say Tutte's graph? – draks ... Aug 06 '13 at 21:09
  • 1
    The non-orientable case in the answer will work for the Tutte graph as well as for any other. And the orientable case is a special case of a construction that can be applied to an arbitrary planar graph: Draw your starting graph on one side of a wooden board, and then drill a hole through the board inside each face. Now select a spanning tree and add an edge around the backside of the board for each edge not in the spanning tree (these correspond to the partially-visible red edges in the diagram). That again gives a single-faced graph, which is non-Hamiltonian if the starting graph is. – hmakholm left over Monica Aug 06 '13 at 23:40
0

$\hskip2.5in$enter image description here

This genus-$2$ regular map, shown to the right, has six octagonal faces, of which three meet at each of its $16$ vertices. It has $24$ edges, and a Euler characteristic of $-2$.

Since it's bipartite there is a $3$-edge-coloring, so it's not a snark...

draks ...
  • 18,449