TURING'S HALTING PROBLEM ALGORITHMIC INFORMATION THEORY AND SHANNON'S THEORY

James Constant

math@coolissues.com

Shannon's Information Theory Finds that Proofs of Turing's Halting Problem and Algorithmic Information Theory are Invalid But Likely to Reformulate by Setting Decisional Thresholds

Algorithmic Information Theory (AIT)

Algorithmic Information Theory (AIT) teaches how a piece of information can be reproduced by an algorithm which indicates the "complexity" or "randomness" of the piece of information. The theory relates to Kurt Godel's incompleteness theorem and Alan Turing's Halting Problem (THP) from which Gregory Chaitin proves that there is no algorithm for testing whether any algorithm is the shortest way to compress a given piece of information, and therefore no true measure of the complexity of any sort of formal numeric entity.1. AIT asserts two definitions, its presumed axioms, and a corollary:

AIT Complexity Axiom: The amount of information in a given sequence of characters (string) is defined as the length of the shortest least complex program which will produce that string as output. One way to think about this is to regard the AIT information content as the lower limit to the compressibility of the string. This quantity is known variously as Kolmogorov complexity, algorithmic complexity and program-length complexity. AIT holds that a compressed representation of a string and the string itself are the same amounts of information. This differs from standard theory which tells us that compression degrades the information in a string.

AIT Randomness Axiom: A string is said to be random when its own length is equal to its program-length complexity - i.e. the string is as short as the shortest program that would produce it as output. This differs from the usual definition of randomness, which is a statistical one, specifying that all substrings (of up to a certain length) must appear equally often in the string.

AIT Randomness Corollary: One interesting feature of AIT's definition of randomness is that it is in general undecidable whether a string is random - because the shortest program that produces a given string as output is uncomputable, except when N is finite. These facts tend to go against there being many practical applications of AIT.

The terms Algorithmic Information Theory and program-length complexity were coined by Chaitin to describe features of his own work on the foundations of mathematics.

Turing's Halting Problem (THP)

Turing showed that no algorithm, no mathematical theory, can ever tell us which computer program will halt and which will not. His proof was made by the method of reductio ad absurdum variously known as "proof by double negation" or "proof by contradiction". A proof by contradiction is a weak proof and is rejected by some. It proves a theorem by assuming that the conjecture is false and finding that it is a contradiction. When a contradiction is found the conjecture is then "proved" to be true. Marilyn vos Savant tells us that

"Proof by double negation is fraught with dangers. For example, lets say the human animal can see only in black and white . . . We want to prove that an object is white, so we manage to prove that it isn't black. By double negation, the object is proved to be white. But considering all the colors of the rainbow, this shouldn't be a valid proof. Maybe the object is red. But how would we know it? Reductio ad absurdum is an "indirect proof". Its opposite is known as a 'direct proof', which satisfies the most stringent requirements"2

Clearly, the increased strength of a proof by contradiction rests upon the decreased absence of other colors. In the context of THP, considering all colors of the rainbow of information, from a complete or no correspondence between a compressed representation of a string and the string itself, THP shouldn't be a valid proof. Maybe the compressed representation of a string is only a fraction of the information of the string itself. But how would we know it? The answer follows.

Shannon's Information Theory

In Claude Shannon's standard information theory, the maximum information in a string of N bits or bytes is equal to the capacity of the computer H=C=logN. If the string is compressed becoming a string of n<N bits or bytes, the maximum information in the compressed string is Hc=CC=logn which reduces the size of the computer and information since Cc<C and since Hc<H. Generally, compression results in lost information since the information in the compressed string is logn=klogN or n=Nk where 0<k<1. If we are not interested in reducing the size of the computer Cc but only interested in preserving the information H in computer C, we can set Mlogn=logN or nM=N M integer.

The main consequence of invalid THP for AIT is that Chaitin's proof, which rests upon the validity of THP, is also invalid. We can now say that Shannon's information algorithm Hc=logn is a way to compress a piece of information H=logN to any shortness desired provided we either take (n=Nk k fraction) or compensate for (n=NM M integer) the loss of information. In this respect a computer does not halt but continues providing a degraded "all colors" output, contrasting THP's and AIT's presumed "black and white" output.

Suppose our string consists of a single page in a book with N=10,000 ASCII bytes and suppose we apply a 100:1 compression program to it so that n=100. The information at the output of the computer is logn<logN. If we accept the degradation of information at level logn=klogN or n=Nk, the result is we are left reading a page with 1/2 the letters missing. If N=10,000 and n=1,000 we are left reading a page with 1/4 the letters missing, a better result. The point is that we can always accept a degraded string at the computer output by setting a decisional threshold. Or, if capacity C is available, we can preserve the undegraded string by setting N=nM M integer, by a serial recycling the information M times through a single program or by a parallel arrangement of M identical programs.

The Future of THP and AIT

Since the compressed representation of a string is only a fraction of the information of the string itself, THP and AIT can survive by reformulating their results as decisional thresholds rather than as presumed absolutes. THP may now be restated as "Conventional Information Theory tells us which computer program will degrade by a fraction k and which will not." AIT's Complexity Axiom may be restated as "The amount of information in a given sequence of characters (string) is defined as the length of the shortest least complex program which will produce a degraded copy of that string as output" and AIT's Redundancy Axiom may be restated as "A string is said to be random when its own length is equal than its program-length complexity". If such or similar changes are made, AIT's Randomness Corollary may be restated as "A string is said to be random when its own length is equal to its program length complexity - i.e., the string is as short as the shortest program that produces it undegraded at its output". Underlines emphasize changes in the original axioms. If only capacity Cc<C is available, we must accept the decrease in capacity and complexity by a corresponding degradation of information by a fraction k. Or, if capacity C is available, we can preserve the undegraded string by setting N=nM M integer. In either case, setting decisional thresholds makes it possible to decide the "complexity" and "randomness" of program-lengths, opening AIT to practical applications.

As stated, AIT can determine program-length complexity where it is below or equal to N. But, program-length complexity can only be guessed when N is infinite. As example, in the case of pi there exist short programs which will go on producing digits of pi forever. It is hard to see that a degraded program-length of pi represents an undegraded pi, a presumed infinite string of digits. The problem with N infinite is that we have no way of knowing the information content of logN. In the pi example we do not know if its short program length preserves a fraction or the entire content of pi. However, since mathematicians have found no repeatability in pi, i.e., M infinite, we can only guess the latter.

While AIT brings some new insights into the ideas of "complexity" and "randomness" it is, nevertheless, a part of standard information theory to which it must be reconciled.


1 Gregory Chaitin, "The Limits of Reason", Scientific American, March 2006 pages 74-81

2 Marilyn vos Savant "The World's Most Famous Math Problem", St. Martin's Press, New York 1993 pages 34-35

Copyright © 2006 by James Constant

By same author: http://www.coolissues.com/mathematics/sameauthor.htm