constraints on hypercomputation greg michaelson 1 & paul cockshott 2 1 heriotwatt university, 2...
TRANSCRIPT
Constraints on Hypercomputation
Greg Michaelson1 &
Paul Cockshott2
1HeriotWatt University,2 University of Glasgow
Church-Turing Thesis
• effective calculability– A function is said to be ``effectively
calculable'' if its values can be found by some purely mechanical process ... (Turing 1939)
• Church-Turing Thesis– all formalisations of effective calculability are
equivalent– e.g. Turing Machines (TM), λ calculus,
recursive function theory
Hypercomputation
• are there computations that are not effectively calculable?
• Wegner & Eberbach (2004) assert that:– TM model is too weak to describe e.g. the
Internet, evolution or robotics – superTuring computations (sTC) are a
superset of TM computations – interaction machines, calculus & $-calculus
capture sTC
Challenging Church-Turing 1
• a successful challenge to the Church-Turing Thesis should show that:– all terms of some C-T system can be reduced
to terms of the new system, – there are terms of the new system which
cannot be reduced to terms of that C-T system
Challenging Church-Turing 2
• might demonstrate:1. some C-T semi-decidable problem is now
decidable
2. some C-T undecidable problem is now semi-decidable
3. some C-T undecidable problem is now decidable
4. characterisations of classes 1-3
5. canonical exemplars for classes 1-3
C-T & Physical Realism 1
• new system must encompass effective computation:– physically realisable in some concrete
machine
• potentially unbounded resources not problematic– e.g. unlimited TM tape
C-T & Physical Realism 2
• reject system if: – its material realisation conflicts with the laws
of physics;– it requires actualised infinities as steps in the
calculation process.
C-T & Physical Realism 2
• infinite computation?– accelerating TMs (Copeland 2002)
• relativistic limits to function of machine
• analogue computation over reals? (Copeland review 1999)– finite limits on accuracy with which a physical
system can approximate real numbers
Interaction Machines 1
• Wegner & Eberbach allege that:– all TM inputs must appear on the tape prior to
the start of computation;– interaction machines (IM) perform I/O to the
environment.
• IM canonical model is the Persistent Turing Machine(PTM) (Goldin 2004) – not limited to a pre-given finite input tape;– can handle potentially infinite input streams.
Interaction Machines 2
• Turing conceived of TMs as interacting open endedly with environment– e.g. Turing test formulation is based on
computer explicitily with same properties as TM (Turing 1950)
• TM interacting with tape is equivalent to TM interacting with environment e.g. via teletype– by construction – see paper
Interaction Machines 3
• IMs, PTMs & TMs are equivalent– by construction – see paper– PTM is a classic but non-terminating TM– PTM's, and thus Interaction Machines, are a
sub-class of TM programs
Calculus 1
calculus is not a model of computation in the same sense as the TM– TM is a specification of a buildable material
apparatus– calculi are rules for the manipulation of strings
of symbols– rules will not do any calculations unless there
is some material apparatus to interpret them
Calculus 2
• program can apply calculus re-write rules of the to character strings for terms calculus has no more power than underlying
von Neumann computer• language used to describe calculus
– channels, processes, evolution – implies physically separate but
communicating entities evolving in space/time• does the calculus imply a physically
realisable distributed computing apparatus?
Calculus 3
• cannot build a reliable parallel/ distributed mechanism to implement arbitrary calculus process composition – synchronisation implies instantaneous transmission of
information – i.e. faster than light communication if processes are
physically separated
• for processors in relative motion, unambiguous synchronisation shared by different moving processes is not possible– processors can not be physically mobile for 3 way
synchronisation (Einstein 1920)
Calculus 4
• Wegner & Eberbach require implied infinities of channels and processes– could only be realised by an actual infinity of
fixed link computers– finite resource but of unspecified size like a
TM tape – for any actual calculation a finite resource is
used, but the size of this is not specified in advance
Calculus 5
• Wegner & Eberbach interpret ‘as many times as is needed' as meaning an actual infinity of replication– deduce that the calculus could implement
infinite arrays of cellular automata (CA) – cite Garzon (1995) to the effect that they are
more powerful than TMs.
• CAs require a completed infinity of cells– cannot be an effective means of computation.
Conclusion 1
• Wegner & Eberbach do not demonstrate for IM or calculus: 1.some C-T semi-decidable problem which is
now decidable 2.some C-T undecidable problem which is now
semi-decidable 3.some C-T undecidable problem which is now
decidable 4.characterisations of classes 1-35.canonical exemplars for classes 1-3
Conclusion 2
• Wegner & Eberbach do not demonstrate physical realisability of IM or calculus
• longer paper submitted to Computer Journal (2005) includes: – fuller details of constructions– critique of $-calculus