You are on page 1of 2

U

which it operated. Of the early machines, the EDSAC,


SEAC, Pilot ACE, EDVAC, and UNIVAC I all had ultra-
sonic memories.
The principle is illustrated in Fig. 1. A train of pulses
Ultrasonic memories played an important role in the representing the number to be stored is modulated
early development of digital computers, but are now onto a carrier and applied to a piezoelectric crystal in
only of historical interest. The report on the EDVAC contact with a column of mercury. The ultrasonic
drafted by von Neumann in June 1945 on behalf of the pulses so generated travel along the column until they
group at the Moore School of Electrical Engineering, reach another crystal at the f a r end. This converts them
Philadelphia, clearly envisaged this type of memory, back into electric signals, which are amplified and
although it did not describe the physical principles on rectified. Theresultingpulses are applied to a gate

Modulator tank Mercury Amplifier and


demodulator

CLEAR
waveform
AL waveform
Write
wavefo
Digits in Digits out
Figure 1. Ultrasonic memory.
1812 UNDECIDABLE PROBLEMS

together with pulsesfrom a continuously running clock carries out a computational task. At that time, “com-
pulse generator. This gating operation serves the twin puter” meant “one who computes” (Le. a person who
purposes of regeneration and synchronization. The carries out a calculation), but the model applies equally
emerging pulses, which are exact replicas of the orig- well to modern electronic computers. Turing then
inal pulses, are reapplied to the modulator and con- proved rigorouslythat no Turing machine can solve the
tinue to circulate. The operations of reading, clearing, so-called HaltingProblem, the problem of determining
and writing can be performed by applying to the gates whether Turing machines eventually halt. Some do,
shown suitable waveforms accurately synchronized some don’t: the problem is, given a Turing machine
with the clock. A typical main memory consisted of a along with its input data, to determine whether or not
group of 32 tanks, as the columns were called, each it will ever finish its task and halt. Of course, one can
between 0.5 and 1.5 meters long and giving a delay of run the Turing machine step by step, and, if it halts,
between one-third and 1 ms. one would see that it had. The difficulty isthat if it does
not halt then this method will not reveal that fact; no
In the rnid-l950s, ultrasonic memories using a fine
matter howlong one continues the simulation, it is
nickel wire in the form of a coil as the propagation
conceivable that the machine would have halted if
medium appeared in some low-cost computers. The
processing had continued an additional few moments.
waves were excited by making use of the magneto-
strictive properties of the nickel. Turing’s proofis a formalization of the following
idea. Establish a pairing between all possible Turing
Bibliography machines (there are infinitely many) and all possible
1956. Wilkes, M. V. Automatic Digital Computers. New York: sets of input data (also infinitely many) in a systematic
John Wiley. way. Now suppose there were a Turing machine H that
1985. Wilkes, M. V. Memoirs of a Computer Pioneer. could solve the Halting Problem. Starting from H, one
Cambridge, MA: MIT Press. could build a “perverse” machine P that acts as fol-
Maurice V. Wilkes lows: P takes its input data x and determines (by using
H a s a subroutine) whether the particular machine M,
which is paired with x, would halt or run forever on
input x; and P then perversely either halts or enters
an infinite loop, whichever causes it to behave differ-
ently from M. Now P has succeeded in behaving dif-
ferently from every machine; specifically,it differs from
machine M on the particular input data x that is
One of this century’s major intellectual discoveries is matched with M in the pairing. But this is impossible,
that there aresome perfectly precise problems that can since P would have to differ evenfrom itself. So H can-
never be solved. This is a technical result, not a mys- not exist after all.
tical one. There is nothing ineffable about these prob-
Turing’s method of proof is very general. It applies
lems; indeed, one can give specific examples of them.
equally well to show that itis impossible to write a
Nor should this result be confused with the claim that
Pascal program, say, that can test Pascal programs
some problems (e.g. what is the meaning of life?) are
for infinite loops. If this were the end of the story,
too imprecise to be solved computationally. Actually,
Turing’s result would merely establish certain limits
these unsolvable problems are just as precise as the
on self-referentialapplication, suggesting, for example,
problem of computing a sum. Also, the result is not
that if one wants to test programs written in a partic-
based on the fact that some problems are just too large
ular language, then one must write the tests in a differ-
to solve in practical terms. Rather, these problems are
ent language. But at the same time that Turing was
unsolvable even under the assumption that wholly un-
describing his machines, Alonzo Church and various
reasonable amounts of time and space are available for
other logicians were proving that manymodels of
their solution. By way of contrast, it is a trivial matter
computation (including Turing machines) thatare
under such an assumption to write a program that
superficially very different are in fact all equivalent.
plays a perfect game of chess: just explore allof the
As a result of this work, theChurch-Turing Thesis has
possibilitiessystematically. (There are only a finite
become generally accepted: any method of perform-
number of possibilities, but that number is so astro-
ing a computation that might conceivablybe proposed
nomically large that this’ exhaustive approach will
in thefuture,nomatter how apparently powerful,
never be of practical use.)
will turn out to be performable by a Turing machine.
In the 1930s, the mathematician Alan Turing intro- This implies that the Halting Problem is unsolvable,
duced a conceptual automaton, now called a Turing not merely in the sense that no Turing machine can
machine, to model the process by which a computer solve it, but that more generally, no computational