Ackermann-Péter Function is not Primitive Recursive
Theorem
The Ackermann-Péter function is not primitive recursive.
Proof
Let $A : \N^2 \to \N$ denote the Ackermann-Péter function.
It suffices to show that, for every primitive recursive $f : \N^k \to \N$, there exists some $t_f \in \N$ such that:
- $\forall x_1, \dotsc, x_k \in \N: \map f {x_1, \dotsc, x_k} < \map A {t_f, \map \max {x_1, \dotsc, x_k}}$
For, if $A$ were primitive recursive, we would have:
- $\map A {t_A, t_A} < \map A {t_A, \map \max {t_A, t_A}} = \map A {t_A, t_A}$
which is a contradiction.
We will proceed by induction on the definition of $f$.
Basic Primitive Recursive Functions
If $f$ is the zero function, we have:
- $t_{\Zero} = 0$
as:
\(\ds \map \Zero x\) | \(=\) | \(\ds 0\) | Definition of Zero Function | |||||||||||
\(\ds \) | \(<\) | \(\ds x + 1\) | ||||||||||||
\(\ds \) | \(=\) | \(\ds \map A {0, x}\) | Definition of Ackermann-Péter Function |
If $f$ is a successor function, we have:
- $t_{\Succ} = 1$
as:
\(\ds \map \Succ x\) | \(=\) | \(\ds x + 1\) | Definition of Successor Function | |||||||||||
\(\ds \) | \(<\) | \(\ds x + 2\) | ||||||||||||
\(\ds \) | \(=\) | \(\ds \map A {1, x}\) | Ackermann-Péter Function at (1,y) |
If $f$ is a projection function, we have:
- $t_{\pr_j^k} = 0$
as:
\(\ds \map {\pr_j^k} {x_1, \dotsc, x_k}\) | \(=\) | \(\ds x_j\) | Definition of Projection Function | |||||||||||
\(\ds \) | \(\le\) | \(\ds \map \max {x_1, \dotsc, x_k}\) | Max Operation Yields Supremum of Operands | |||||||||||
\(\ds \) | \(<\) | \(\ds \map \max {x_1, \dotsc, x_k} + 1\) | ||||||||||||
\(\ds \) | \(=\) | \(\ds \map A {0, \map \max {x_1, \dotsc, x_k} }\) | Definition of Ackermann-Péter Function |
$\Box$
Substitution
Suppose $f : \N^k \to \N$ is obtained by substitution from:
- $g_1, \dotsc, g_m : \N^k \to \N$
- $h : \N^m \to \N$
as:
- $\map f {x_1, \dotsc, x_k} = \map h {\map {g_1} {x_1, \dotsc, x_k}, \dotsc, \map {g_m} {x_1, \dotsc, x_k}}$
Let $t_g = \map \max {t_{g_1}, \dotsc, t_{g_m}}$
Then, we have:
- $t_f = t_h + t_g + 2$
For every $1 \le j \le m$:
\(\ds \map {g_j} {x_1, \dotsc, x_k}\) | \(<\) | \(\ds \map A {t_{g_j}, \map \max {x_1, \dotsc, x_k} }\) | Inductive hypothesis on $t_{g_j}$ | |||||||||||
\(\ds \) | \(\le\) | \(\ds \map A {t_g, \map \max {x_1, \dotsc, x_k} }\) | Ackermann-Péter Function is Strictly Increasing on First Argument |
Thus:
\(\ds \map f {x_1, \dotsc, x_k}\) | \(=\) | \(\ds \map h {\map {g_1} {x_1, \dotsc, x_k}, \dotsc, \map {g_m} {x_1, \dotsc, x_k} }\) | ||||||||||||
\(\ds \) | \(<\) | \(\ds \map A {t_h, \map \max {\map {g_1} {x_1, \dotsc, x_k}, \dotsc, \map {g_m} {x_1, \dotsc, x_k} } }\) | Inductive hypothesis on $t_h$ | |||||||||||
\(\ds \) | \(=\) | \(\ds \map A {t_h, \map {g_j} {x_1, \dotsc, x_k} }\) | where $\map {g_j} {x_1, \dotsc, x_k} = \map \max {\map {g_1} {x_1, \dotsc, x_k}, \dotsc, \map {g_m} {x_1, \dotsc, x_k} }$ | |||||||||||
\(\ds \) | \(<\) | \(\ds \map A {t_h, \map A {t_g, \map \max {x_1, \dotsc, x_k} } }\) | Ackermann-Péter Function is Strictly Increasing on Second Argument | |||||||||||
\(\ds \) | \(<\) | \(\ds \map A {t_h + t_g + 2, \map \max {x_1, \dotsc, x_k} }\) | Strict Upper Bound for Composition of Ackermann-Péter Functions | |||||||||||
\(\ds \) | \(=\) | \(\ds \map A {t_f, \map \max {x_1, \dotsc, x_k} }\) |
$\Box$
Primitive Recursion
Suppose $f : \N^{k + 1} \to \N$ is obtained by primitive recursion from:
- $g : \N^k \to \N$
- $h : \N^{k + 2} \to \N$
as:
- $\map f {x_1, \dotsc, x_k, z} = \begin{cases}
\map g {x_1, \dotsc, x_k} & : z = 0 \\ \map h {x_1, \dotsc, x_k, z - 1, \map f {x_1, \dotsc, x_k, z - 1}} & : z > 0 \end{cases}$
Then, we have:
- $t_f = \map \max {t_g, t_h} + 5$
It suffices to show that, for all $x_1, \dotsc, x_k, z \in \N$:
- $\map f {x_1, \dotsc, x_k, z} < \map A {\map \max {t_g, t_h} + 1, \map \max {x_1, \dotsc, x_k} + z}$
For:
\(\ds \map f {x_1, \dotsc, x_k, z}\) | \(<\) | \(\ds \map A {\map \max {t_g, t_h} + 1, \map \max {x_1, \dotsc, x_k} + z}\) | To be shown | |||||||||||
\(\ds \) | \(\le\) | \(\ds \map A {\map \max {t_g, t_h} + 1, 2 \map \max {x_1, \dotsc, x_k, z} }\) | Ackermann-Péter Function is Strictly Increasing on Second Argument | |||||||||||
\(\ds \) | \(<\) | \(\ds \map A {\map \max {t_g, t_h} + 1, 2 \map \max {x_1, \dotsc, x_k, z} + 3}\) | Ackermann-Péter Function is Strictly Increasing on Second Argument | |||||||||||
\(\ds \) | \(=\) | \(\ds \map A {\map \max {t_g, t_h} + 1, \map A {2, \map \max {x_1, \dotsc, x_k, z} } }\) | Ackermann-Péter Function at (2,y) | |||||||||||
\(\ds \) | \(<\) | \(\ds \map A {\map \max {t_g, t_h + 5, \map \max {x_1, \dotsc, x_k, z} } }\) | Strict Upper Bound for Composition of Ackermann-Péter Functions | |||||||||||
\(\ds \) | \(=\) | \(\ds \map A {t_f, \map \max {x_1, \dotsc, x_k, z} }\) |
In order to prove that statement, we will perform induction on $z$.
Basis for the Induction
First, suppose $z = 0$.
Then:
\(\ds \map f {x_1, \dotsc, x_k, 0}\) | \(=\) | \(\ds \map g {x_1, \dotsc, x_k}\) | ||||||||||||
\(\ds \) | \(<\) | \(\ds \map A {t_g, \map \max {x_1, \dotsc, x_k} }\) | ||||||||||||
\(\ds \) | \(<\) | \(\ds \map A {\map \max {t_g, t_h} + 1, \map \max {x_1, \dotsc, x_k} + 0}\) | Ackermann-Péter Function is Strictly Increasing on First Argument |
satisfying the basis for the induction.
$\Box$
Induction Hypothesis
The induction hypothesis is:
- $\map f {x_1, \dotsc, x_k, z} < \map A {\map \max {t_g, t_h} + 1, \map \max {x_1, \dotsc, x_k} + z}$
Assuming that, we want to show:
- $\map f {x_1, \dotsc, x_k, z + 1} < \map A {\map \max {t_g, t_h} + 1, \map \max {x_1, \dotsc, x_k} + z + 1}$
Induction Step
First, we will show that:
- $\map \max {x_1, \dotsc, x_k, z, \map f {x_1, \dotsc, x_k, z} } < \map A {\map \max {t_g, t_h} + 1, \map \max {x_1, \dotsc, x_k} + z}$
If the greatest among $\set {x_1, \dotsc, x_k, z, \map f {x_1, \dotsc, x_k, z} }$ is some $x_i$, then:
\(\ds x_i\) | \(\le\) | \(\ds \map \max {x_1, \dotsc, x_k}\) | Max Operation Yields Supremum of Operands | |||||||||||
\(\ds \) | \(\le\) | \(\ds \map \max {x_1, \dotsc, x_k} + z\) | ||||||||||||
\(\ds \) | \(<\) | \(\ds \map A {\map \max {t_g, t_h} + 1, \map \max {x_1, \dotsc, x_k} + z}\) | Ackermann-Péter Function is Greater than Second Argument |
If it is $z$, then:
\(\ds z\) | \(\le\) | \(\ds \map \max {x_1, \dotsc, x_k} + z\) | ||||||||||||
\(\ds \) | \(<\) | \(\ds \map A {\map \max {t_g, t_h} + 1, \map \max {x_1, \dotsc, x_k} + z}\) | Ackermann-Péter Function is Greater than Second Argument |
If it is $\map f {x_1, \dotsc, x_k, z}$, the statement holds by the induction hypothesis.
Thus, by Proof by Cases, the statement:
- $\paren 1 \quad \map \max {x_1, \dotsc, x_k, z, \map f {x_1, \dotsc, x_k, z} } < \map A {\map \max {t_g, t_h} + 1, \map \max {x_1, \dotsc, x_k} + z}$
is true.
Now:
\(\ds \map f {x_1, \dotsc, x_k, z + 1}\) | \(=\) | \(\ds \map h {x_1, \dotsc, x_k, z, \map f {x_1, \dotsc, x_k, z} }\) | ||||||||||||
\(\ds \) | \(<\) | \(\ds \map A {t_h, \map \max {x_1, \dotsc, x_k, z, \map f {x_1, \dotsc, x_k, z} } }\) | ||||||||||||
\(\ds \) | \(<\) | \(\ds \map A {t_h, \map A {\map \max {t_g, t_h} + 1, \map \max {x_1, \dotsc, x_k} + z} }\) | Ackermann-Péter Function is Strictly Increasing on Second Argument and $\paren 1$ | |||||||||||
\(\ds \) | \(\le\) | \(\ds \map A {\map \max {t_g, t_h}, \map A {\map \max {t_g, t_h} + 1, \map \max {x_1, \dotsc, x_k} + z} }\) | Ackermann-Péter Function is Strictly Increasing on First Argument | |||||||||||
\(\ds \) | \(=\) | \(\ds \map A {\map \max {t_g, t_h} + 1, \map \max {x_1, \dotsc, x_k} + z + 1}\) | Definition of Ackermann-Péter Function |
Thus, the induction step is satisfied.
$\Box$
By the Principle of Mathematical Induction, it holds that:
- $\map f {x_1, \dotsc, x_k, z} < \map A {\map \max {t_g, t_h} + 1, \map \max {x_1, \dotsc, x_k} + z}$
for all $z \in \N$.
So, the case is concluded by the remarks above.
$\Box$
By definition, every primitive recursive function $f$ must be obtainable from just those operations, and so necessarily satisfy the required property.
It follows from the remarks at the start of the proof that the Ackermann-Péter function is not primitive recursive.
$\blacksquare$