There’s a 2009 article by Arai on interated recursive Mahloness. The main idea of the paper is that stronger systems of reflection can be analysed in terms of iterated $\Pi_2$-reflection.
In section 5, Arai writes the following:
A $\Pi_3$-reflecting ordinal $K$ is understood to be $<\varepsilon_{K+1}$[-ply]-recursively Mahlo.The implication here is that in terms of strength, the existence of a $\Pi_3$-reflecting is equivalent to the existence of a $K$ that is $<\varepsilon_{K+1}$-ply-recursively Mahlo.
So a $\Pi_3$-reflecting is equivalent, in some sense, to a $K$ that is $F(\varepsilon_{K+1})$-ply-recursively Mahlo, were $F(\alpha)=\alpha$. Now the question is: what are the ‘Arai functions’ $F$ for higher reflecting ordinals?
For a $\Pi_2$-reflecting, the Arai function is (obviously) $1$. For a $\Pi_{n+1}$-reflecting, the arai function is $F(x)=x^{G(x)}$ where $G$ is the Arai function for $\Pi_n$-reflection.
The final cases are $X$-reflecting ordinals that are also $\Pi_n$-reflecting onto $Y$-reflecting ordinals. Suppose the Arai function for $X$-reflection is $x^{x^{\ldots^{G(x)}}}$ with $n-2$ $x$s and that for $Y$-reflection is $x^{x^{\ldots^{H(x)}}}$ with $n-2$ $x$s, then the Arai function for $X$-reflecting ordinals that are also $\Pi_n$-reflecting onto $Y$-reflecting ordinals is $x^{x^{\ldots^{H(x)+G(x)}}}$.
Collapsing functions
Fortunately, Arai has a solution for that. First, we agree to notate Arai functions by the value of $F(K)$, where $K$ is some suitable large ordinal. Essentially, given an iterated Mahlo hierarchy $M$ and Skolem hulls $C$, define $\psi^\alpha_\kappa(\beta)$ to be the least ordinal $\pi$ that is in $M^\alpha$ and such that $C(\beta,\pi)\cap\kappa=\pi$.
Since defining $C$ is trivial, we focus on defining $M$. We see that the definition must somehow unpack the Arai function $F$ knowing $F(K)$. However, there are ordinals not described by Arai functions, such as $\kappa$ which are $\kappa$-ply-recursively Mahlo, and are important to reach the full strength of reflection. So we instead notate Arai functions by the value of $F(\varepsilon_{K+1})$. To obtain the Arai function from the value, the solution is the standard diagonalization trick:
$M^\alpha$ is the set of $\kappa$ such that:
- $\kappa< K$
- $C(\alpha,\kappa)\cap K=\kappa$
- For all $\beta\in C(\alpha,\kappa)\cap\alpha$, $\kappa$ is $\Pi_2$-reflecting on $M^\beta$
For completeness, we also define the Skolem hulls $C(\alpha,\kappa)$ as the colsure of $\kappa\cup\{0,K,\varepsilon_{K+1}\}$ under:
- $\beta+\gamma$
- $\omega^\beta$
- $\psi^\beta_\kappa(\gamma)$, provided $\beta,\gamma<\alpha$
If we restrict our ordinals to be below $\varepsilon_{K+1}$, then voilà, we have rediscovered Rathjen’s $\Psi$, modulo “technical convenience” (Rathjen’s wording).
Degrees of Reflection
Those of you familiar with Taranovsky may have noticed a resemblance between the Arai function assignments above and Taranovsky’s Degrees of Reflection. Indeed, if the Arai function is $F$, then the Taranovsky degree is $\Omega^{F(\Omega)}$.
Thus, we can use this to analyse DoR. Note that both DoR and our $\Pi_n$-ref OCF both have additional structure beyond the Arai functions.
In our OCF, an Arai function is coded as $F(\varepsilon_{\Omega+1})$, and in Taranovsky’s, it is coded as $\Omega^{F(\Omega)}$. For the Arai functions corresponding to basic reflection, the former is much larger than the latter, so it seems plausible that DoR misses too much structure to reach $\Pi_n$-ref. Consider the original degrees of reflection. Then the degrees range over precisely the same range as Arai and Rathjen use to analyse $\Pi_3$-ref. Thus, the strength of raw DoR is probably $\Pi_3$-ref. However, using Taranovsky’s suggestion to use Veblen’s function with DoR makes $F(\varepsilon_{\Omega+1})\approx\Omega^{F(\Omega)}$ for most functions $F$ under consideration, so Taranovsky’s claims in this case are most likely correct, though the assignments may need to be lowered.
Degrees of Reflection ordinal assignments (conjectural)
Based on the previous section, we construct a function $C$ which, conjecturally, models the DoR notation system. We use Taranovskian notation.
Skolem hulls $H(a,b)$ are defined as the closure of $b\cup\{0,\Omega\}$ under $C$ with first argument below $a$ and $O$ provided the output is above $\Omega$.
Define hirarchies $M(a)$ as the set of $b<\Omega$ such that:
- $H(a,b)\cap\Omega=b$
- For all $c\in H(a,b)\cap a$, $b$ is $\Pi_2$-reflecting on $M(c)$
$C(a,b)$ is the least element of $M(a)$ that is greater than $b$.
Of course this assignment is suboptimal–it does not use limits, so the outputs of $C$ are above $\omega_1^{CK}$. However, modifying it to use limits and yield ordinals below $\omega_1^{CK}$ decreases the simplicity.