next up previous
Next: Probability vs Descriptive Complexity Up: Measures and Probability Distributions Previous: Universal TM-Induced Measures

Universal CEM vs EOM with Random Input

Corollary 4.3 and Lemma 4.2 below imply that $mu^E$ and $mu_0$ are essentially the same thing: randomly selecting the inputs of a universal EOM yields output prefixes whose probabilities are determined by the universal CEM.

Corollary 4.3   Let $mu_0$ denote the universal CEM of Theorem 4.1. For $x in B^*$,

\begin{displaymath}
mu^E(x) = O(mu_0(x)).
\end{displaymath}

Lemma 4.2   For $x in B^*$,

\begin{displaymath}
mu_0(x)= O(mu^E(x)).
\end{displaymath}

Proof. In the enumeration of EOMs in the proof of Theorem 4.1, let $EOM_0$ be an EOM representing $mu_0$. We build an EOM $T$ such that $mu_T(x) = mu_0(x)$. The rest follows from the Invariance Theorem (compare Proposition 3.1).

$T$ applies $EOM_0$ to all $x in B^*$ in dovetail fashion, and simultaneously simply reads randomly selected input bits forever. At a given time, let string variable $z$ denote $T$'s input string read so far. Starting at the right end of the unit interval $[0,1)$, as the $Vbar{mu}_0(x)$ are being updated by the algorithm of Theorem 4.1, $T$ keeps updating a chain of finitely many, variable, disjoint, consecutive, adjacent, half-open intervals $VI(x)$ of size $Vbar{mu}_0(x)$ in alphabetic order on $x$, such that $VI(y)$ is to the right of $VI(x)$ if $y succ x$. After every variable update and each increase of $z$, $T$ replaces its output by the $x$ of the $VI(x)$ with $0.z in VI(x)$. Since neither $z$ nor the $VCmu_0(x)$ in the algorithm of Theorem 4.1 can decrease (that is, all interval boundaries can only shift left), $T$'s output cannot either, and therefore is indeed EOM-computable. Obviously the following holds:

\begin{displaymath}
Cmu P_T(x) =
CP_T(x) =
Cmu_0(x)
\end{displaymath}

and

\begin{displaymath}
mu P_T(x) = sum_{z in B^{sharp}} P_T(xz) = mu_0(x).
\end{displaymath}

Summary. The traditional universal c.e. measure [40,45,29,16,17,41,14,30] derives from universal MTMs with random input. What is the nature of our novel generalization here? We simply replace the MTMs by EOMs. As shown above, this leads to universal cumulatively enumerable measures. In general these are not c.e. any more, but they are ``just as computable'' in the limit as the c.e. ones -- we gain power and generality without leaving the constructive realm and without giving up the concept of universality. The even more dominant approximable measures, however, lack universality.


next up previous
Next: Probability vs Descriptive Complexity Up: Measures and Probability Distributions Previous: Universal TM-Induced Measures
Juergen Schmidhuber 2003-02-13