Subcubic graph number

The subcubic graph numbers are the outputs of an extremely fast-growing al function. They were devised by Harvey Friedman.

A subcubic graph is a finite graph in which each vertex has a valence of at most three. (For the sake of this article, subcubic graphs are allowed to be s or s, and are not required to be connected.) Suppose we have a sequence of subcubic graphs G1, G2, ... such that each graph Gi has at most i + k vertices (for some integer k) and for no i < j is Gi into (i.e. is a graph minor of) Gj.

The Robertson-Seymour theorem proves that subcubic graphs are well-founded by homeomorphic embeddability, implying such a sequence cannot be infinite. So, for each value of k, there is a sequence with maximal length. We denote this maximal length using SCG(k).

Specific values
It is possible to show that SCG(0) = 6. The first graph is one vertex with a loop, the second is two vertices  connected by a single edge, and the next four graphs consist of 3, 2, 1, and 0 unconnected vertices. All maximal sequences will peak and decline this way. The following bounds have been shown by Googology Wiki user.


 * \(\text{SCG}(1) > f_{\varepsilon_22}(f_{\varepsilon_02}(f_{\varepsilon_0+1}(f_{\varepsilon_0}(f_{\omega^\omega+1}(f_{\omega^5+\omega^2+\omega}(\\f_{\omega^23+1}(f_{\omega^22+1}(f_{\omega^2+\omega3+1}(f_{\omega^2+1}(f_{\omega^2}(2^{3\times2^{95}})))))))))))\).
 * \(\text{SCG}(2) > f_{\vartheta(\Omega^\omega)}(f_{\varepsilon_22}(f_{\varepsilon_02}(f_{\varepsilon_0+1}(f_{\varepsilon_0}(f_{\omega^\omega+1}(\\f_{\omega^5+\omega^2+\omega}(f_{\omega^23+1}(f_{\omega^22+1}(f_{\omega^2+\omega3+1}(f_{\omega^2+1}(f_{\omega^2}(2^{3\times2^{95}}))))))))))))\)

These bounds use a non-standard choice of fundamental sequences for ordinals &mdash; by using a particular, highly complex bijection between ordinals and small graphs, which we will denote here by \(f\), we define \(\alpha[n]=\max\{\beta: \beta<\alpha\text{ and } f(\beta)\text{ is a graph with }\leq n\text{ vertices}\}\).

Since the graph indices start at one, it is also valid to say that SCG(-1) = 1, consisting only of the empty graph.

Friedman proved that SCG(13) is greater than the halting time of any such that it can be proven to halt in at most 20002 symbols in \(\Pi^1_1\)-\(\text{CA}_0\). Not much is known about it, although it is known that it's far larger than TREE(3). Since SCG(n) is a computable function, SCG(13) is still no match for Rayo's number.

The growth rate of SCG(n) is about \(\psi_{\Omega_1}(\Omega_\omega)\) in the fast-growing hierarchy. It is upper bounded by the Takeuti-Feferman-Buchholz ordinal.

Matrix interpretation
An alternate way of describing the SCG function is as follows. Define an incidence matrix as a matrix with entries in {0, 1, 2} where each column sums to exactly 2 and each row sums to at most 3. Given a nonnegative integer k, we construct a sequence of n incidence matrices M1, M2, ..., Mn such that each matrix Mi has at most i + k rows, and no matrix can be changed into an earlier one by repeated applications of any of the following processes:


 * Reordering rows or columns.
 * Deleting columns.
 * Deleting rows, then deleting any columns that do not sum to 2.
 * Adding two rows so their sum produces at least one 2, and deleting the column containing that 2.

SCG(k), then, is the largest possible value of n.

Simple subcubic graph numbers
If we require the subcubic graphs to be simple (i.e. not pseudographs or multigraphs), we get the simple subcubic graph numbers, denoted SSCG. Adam P. Goucher has shown that SSCG(2) << TREE(3) << SSCG(3). Moreover, he has shown that even TREEn(3) for even very large n (for example n=TREE(3)) does not compete at all with SSCG(3).

SCG(n) and SSCG(n) have comparable growth rates: Goucher proved that \(SSCG(4n+3) \geq SCG(n)\). Further, it can be proven that \(SSCG(3n+4) \geq 3 SCG(n)\).

Values and bounds

 * SSCG(0) = 2
 * SSCG(1) = 5
 * SSCG(2) \(\geq 3*2^{3*2^{95}}-8 \approx 10^{3.5775*10^{28}}\)