Mathematical works of Vladimir A. Uspensky: A commentary
aa r X i v : . [ m a t h . HO ] A ug Mathematical works of Vladimir A. Uspensky:a commentary ∗ Alexander Shen † Abstract
Vladimir Andreevich Uspensky [1930–2018] was one of the Soviet pioneers of the theoryof computation and mathematical logic in general. This paper is the survey of his mathe-matical works and their influence. (His achievements in linguistics and his organizationalrole are outside the scope of this survey.)
Harmonic functions
The first paper of Uspensky [3] appeared when he was an undergraduate student. It suggestsan elementary approach to harmonic functions that is based on the definition of a harmonicfunction on R as a function that has the mean value property. The main tool in the followingobservation: for fixed two poinst A , B the oriented angle ACB is a harmonic function of C , and this function is a locally constant function on any circle that goes through A and B . Master thesis
Uspensky’s thesis advisor was Andrei Kolmogorov; the thesis [4] was written and defended in1952. It contains the description of the model of computation suggested by Kolmogorov and nowknown as
Kolmogorov–Uspensky machines . It is shown that this model is equivalent to partialrecursive functions (defined in terms of substitution, recursion and µ -operator). Moreover,this model is used to define relative computability with respect to some function f . For that,the graph of f is represented as an infinite graph (a complex ) that is available to the graphtransformation algorithm together with the input [definition (A) on p. 64]. This definition iscompared with other definition of relative computability. For that, Uspensky reformulates theTuring–Post definition [84, 86], see definition (T) on p. 63, and shows that definitions (T) and(A) are equivalent. Moreover, Uspensky proves the equivalence with one more definition givenin terms of the closure of the function f (and basic functions) under substitution, recursionand µ -operator.The thesis advisor (Kolmogorov) writes in his opinion:This paper analyses (in more details than it was done before) the very notion ofalgorithmic computability.(1) The author reproduces the only completely formal definition of algorithmic re-ducibility (p. 22) that he ascribes to Boris Trakhtenbrot: a function γ is reducible ∗ Vladimir Andreevich Uspensky was my teacher (and undegraduate and Ph.D advisor). Here I concentrateon his mathematical works; I hope to express my deep gratitude to him elsewere. † LIRMM, University of Montpellier, CNRS, Montpellier, France. Supported by RaCAF–ANR-15-CE40-0016. This is easy to see, because both direction angles CA and CB satisfy the mean value property and thesame is true for their difference.
1o another function δ if γ belongs to the recursive closure of δ . The author showsthat in fact such a reduction can be performed in some simple canonical way, usingsome fixed primitive recursive functions τ ( u ) and ω ( u ) and some primitive recursivefunctions h ( u, v, w ) and ϕ ( m ) that depend on γ and δ ; see theorem on p. 28. Thisis the main result of the paper from the purely mathematical viewpoint. Trakhtenbrot’s definition, on the other hand, needs some “justification” that showsthat it corresponds to the intuitive notion of reduction: there is a “mechanical” wayof obtaining γ ( x ) assuming that the values δ ( x ) are somehow made “accessible” foreach x . The general framework for such a justification were given by Post; thetranslation of the corresponding part of Post’s paper [86] is reproduces in the thesis. Then the author gives a completely formal definition of reduction that corresponds tothis idea (probably for the first time) and proves its equivalence to Trakhtenbrot’sdefinition. This is also a very significant achievement of the author.Also the paper contains a good survey of the definitions of algorithmic computabilityfor function y = γ ( x ) with numerical arguments and values. It is centered aroundthe definition suggest by myself. The thesis provides the motivation for this defini-tion and proves that it is equivalent to the previous ones. In a sense this equivalencecan be considered as a “justification” for the previous definitions since my definitionmakes especially clear the idea of an algorithmic computability; the algorithms thatare used are models of the real computation devices, only the amount of “memory”is assumed to be unbounded .To understand the value of this paper, one should recall the historical context (now almostforgotten). Let us make few historical comments.
Partial recursive functions
Ask an expert what is a partial recursive function. Most probably the answer would be: thisis a function that can be obtained from basic functions (projection function, zero constant andsuccessor function) using substitution, recursion and minimization ( µ -operator). This definitioncan be found in the classical text of Odifreddi [108, p. 127] and other sources (see, e.g., [100, 106]or Wikipedia article [113]).However, the original definition was different. The traces of this old definition can be foundin another classical textbook [70, Section 1.5] and in Wolfram MathWorld [114] site. Thisdefinition in equivalent (gives the same class of partial function), but it is different, and oneshould have this difference in mind while reading old papers.Let us try to clarify the history. Recursive definitions were well known for a long time(recall the Fibonacci sequence). They were systematically used to define arithmetic functionsin the paper of Skolem [72]. He realized that in this way one can define not only addition ormultiplication (by the recurrent formula like x + y ′ = ( x + y ) ′ or x · y ′ = x · y + x , where x ′ is thesuccessor of x ), but also many other functions that appear in the elementary number theory.After these definitions are given, one could prove basic facts of number theory by induction. Skolem did not explicitly consider the class of functions that can be defined recursively inthis way. However, in his 1925 talk Hilbert [73] says that “the elementary means that we haveat our disposal for forming functions are substitution (that is, replacement of an argument bya new variable or function) and recursion (according to the schema of the derivation of the The goal was to show that many mathematical results can be proven by a simple and robust way, justby using recursive definitions and inductive arguments. Now the corresponding theory is known as primitiverecursive arithmetic. n + 1 from that for n )”. He then considers the sequence of functions ϕ ( a, b ) = a + b, ϕ ( a, b ) = a · b, ϕ ( a, b ) = a b ,ϕ ( a, b ) = [ b th term in the sequence a, a a , a ( a a ) , a a ( aa ) . . . ]that can be defined in general by the recurrent formula ϕ ( a, b ) = a + b, ϕ n +1 ( a,
1) = a, ϕ n +1 ( a, b + 1) = ϕ n ( a, ϕ n +1 ( a, b )) , and mentions Ackermann’s result saying that the function ϕ n ( a, b ) “cannot be defined by sub-stitutions and ordinary, step-wise recursions” (this result was later published in [74]). Whenstating this negative result, Hilbert implicitly considers the class of function that can be de-fined by “substitutions and ordinary, stepwise recursions” (even though this class is not definedexplicitly and there is no name for the functions from this class.)Such a definition (and name) appeared in the classical work of G¨odel [75, p. 179]: a functionis called recursive ( rekursiv in German) if it can be obtained by a sequence of substitution andrecursion operation (we construct ϕ assuming that ψ and µ are already constructed): ϕ (0 , x , . . . , x n ) = ψ ( x , . . . , x n ) ϕ ( k + 1 , x , . . . , x n ) = µ ( k, ϕ ( k, x , . . . , x n ) , x , . . . , x n )(scheme (2) p. 179). G¨odel proves that these functions could be represented in a formal system,so for him this class of functions is more a tool than an object.G¨odel’s definition does not cover more general recursive definition (like the one used byAckermann). How can they be treated? Herbrand (in a letter to G¨odel and in [76]) suggestedthat one could consider systems of functional equations that related the functions we definewith the already defined ones. He writes in [76, p. 5, p. 624 of the English translation]:We may also introduce any number of functions f i ( x , . . . , x n i ) together with hy-potheses such that(a) The hypotheses contain no apparent variables;(b) Considered intuitionistically, they make the actual computation of f i ( x , . . . , x n i )of the f i ( x , . . . , x n i ) possible for every given set of integers, and it is possibleto prove intuitionistically that we obtain a well-determined result.The reference to intuitionism sounds a bit unclear; probably it means that it is not enoughto have a functional equation or a system of equations for which we can prove (using arbitrarilypowerful tools) that it has a unique solution. We require that the proof is constructive andprovides a method to compute the values of the functions starting from the equations. Indeed,later Kalmar [89] gave an example of a system of functional equations that uniquely defines anon-computable function.G¨odel returns to Herbrand’s suggestion (Herbrand died in mountains just after sending hispaper [76] to the editors) in his Princeton’s lectures. (The lecture notes circulated at that timeand later were reprinted, see [76].) As before, he considers ¡¡recursive functions¿¿ that can beobtained from basic functions by substitutions and “ordinary” recursions; however, in Section 9he mentions the recursive definitions of more general type. The functions defined in this wayare called “general recursive functions”. He says: This expression means: whey they are translated into ordinary language, considered as a property of integersand not as a mere symbol. [Herbrand’s footnote] φ denotes an unknown function, and ψ , . . . , ψ k are known functions, and if the ψ ’sand the φ are substituted in one another in the most general fashions and certainparts of the resulting expressions are equated, then if the resulting set of functionalequations has one and only one solution for φ , φ is a recursive function.”(and mentions Herbrand’s letter as a reference). Then he added some restrictions that clarifyHerbrand’s idea:We shall make two restrictions on Herbrand’s definition. The first is that the left-hand side of each of the given functional equations defining φ shall be of the form φ ( ψ i ( x , . . . , x n ) , ψ i ( x , . . . , x n ) , . . . , ψ il ( x , . . . , x n )) . The second (as stated below) is equivalent to the condition that all possible sets ofarguments ( n , . . . , n l ) of φ can be so arranged that the computation of the valueof φ for any given set of arguments ( n , . . . , n l ) by means of the given equationsrequires a knowledge of the values of φ only for sets of arguments which precede( n , . . . , n l ).G¨odel does not specify the ordering on the tuples (used as arguments), so the exact meaningof this definition is unclear. But later he specifies the derivation rules that allow to derive anequality from the other ones, and says:Now our second restriction on Herbrand’s definition of recursive function is that foreach set of natural numbers k , . . . , k l there should be one and only one m such that φ ( k , . . . , k l ) = m is a derived equation.In this way G¨odel gives a quite formal definition of some class of functions called “generalrecursive functions” (usually translated to Russian as «общерекурсивные функции» . However,as Kleene explains in [105], at the time of these lectures (1934) G¨odel was not sure that thisclass of functions is general enough: ¡¡However, G¨odel, according to a letter he wrote to MartinDavis on 15 February 1965, “was, at the time of [his 1934] lectures, not at all convinced that[this] concept of recursion comprises all possible recursions”¿¿ [105, p. 48]. Davis writes in [99,p. 40]:In the present article [Davis discussed [77]] G¨odel shows how an idea of Herbrand’scan be modified so as to give a general notion of recursive function h . . . i G¨odelindicates (cf. footnote 3) that he believed that the class of functions obtainable byrecursion of the most general kind were the same as those computable by a finiteprocedure. However, Dr. G¨odel has stated in a letter that he was, at the time of theselectures, not at all convinced that his concept of recursion comprised all possiblerecursions; and that in fact the equivalence between his definition and Kleene’s inMath. Ann. 112 [this is [80] in our list] is not quite trivial. So despite appearancesto the contrary, footnote 3 of these lectures is not a statement of Church’s thesis.Footnote 3 [99, p. 44] discusses the claim that every primitive recursive function (obtainedby substitutions and “ordinary recursions, see below) can be computed by a finitary process,and says that “The converse seems to be true, if, besides recursions according to the scheme(2) [primitive recursion], recursions of other forms (e.g., with respect to two variables simulta-neously) are admitted. This cannot be proved, since the notion of finite computation is notdefined, but it serves as a heuristic principle”.R´osza P´eter in [78] studies the “ordinary recursions” and proves, for example, that one mayuse several values of the function (for smaller arguments) in the recursive definition and still get4he same class of functions. She introduces the name “primitive Rekursion” for the “ordinary”recursions considered by her predecessors.Then Kleene in [80] (1936) introduces the name “primitive recursive functions” ( «прими-тивно рекурсивные функции» in Russian) for functions that can be obtained by substitutionsand primitive (=“ordinary”) recursion. At the same time, Kleene suggests to consider a biggerclass of functions. He calls the functions from this class “general recursive function” (the titleof his paper is
General recursive functions of natural numbers ). This class is defined followingHerbrand and G¨odel; Kleene considered different versions of derivation rules for equalities andshows that they lead to the same class of functions.Kleene also introduces “ ε -operator”. Namely, εx [ A ( x )] is defined as the minimal x suchthat A ( x ) or 0 if such an x does not exists. This operator is used in Theorem IV that saysthat every general recursive function can be represented as ψ ( εy [ R ( x, y )]) , for some primitiverecursive function ψ and some primitive recursive predicate R (this means that R can berepresented as r = 0 for some primitive recursive function r ), such that for every x thereexists y such that R ( x, y ). The next Theorem V says that the reverse statement is also true:every function that can be presented in this way is a general recursive function (in the sense ofHerbrand and G¨odel). There this representation can be considered as an equivalent definitionof the class of general recursive functions. Moreover, this definitions can be used to providesome numbering of all general recursive functions if we add an additional argument e to R ; notall values of e lead to total functions. One could say that it this way we get a numbering of afamily of partial functions, but in this paper Kleene does not considers this class (later called“partial recursive functions”).Church (also in 1936) publishes his paper [79] where he defines some other class of functionswith natural arguments and values in terms of some calculus (called λ -calculus) and claims thatthis class captures the intuitive idea of computability:The purpose of the present paper is to propose a definition of effective calculability which is thought to correspond satisfactorily to the somewhat intuitive notion.Here ( ) is Church’s footnote:As will appear, this definition of effective calculability can be stated in either of twoequivalent forms, (1) that a function of positive integers shall be called effectivelycalculable if it is λ -definable in the sense of § § λ -definability is due jointly to the present author and S.C. Kleene h . . . i The notion of recursiveness in the sense of § h . . . i The proposal to identify these notions with the intuitive notion ofeffective calculability is first made in the present paper. . .Church adds (a footnote in § λ -definability had previously been proposedby the author independently.It is clear from this footnote that for Church the suggestion to identify the intuitive notionof effective calculability with the formally defined class of functions (for which two equivalent Note that the clause in the definition of ε -operator that lets the value to be 0 when x does not exists, is notused in Theorem IV; so one can use the standard µ -operator instead. (For µ -operator the value is undefined if y does not exst.) Church’sthesis .Almost at the same time Turing publishes his paper [79] where he defines the model ofcomputation now called
Turing machines . Turing calls them a -machines (‘a’ for ‘automatic’).Turing also constructs the universal machine that can simulate any Turing machine whenequipped by a suitable problems. Turing uses this type of machines to define the notion ofa computable real number (the digits in the positional representation can be computer by amachine), and also gives his proof of the undecidability of the Entscheidungsproblem (thereis no algorithm that can tell whether a given first order formula is logically valid, i.e., truein all the interpretations of the language). Earlier similar results (for equivalent definitions ofcomputability) were proven by G¨odel and Kleene, as well as Church (see [99, p.109] for details).In an Appendix (added August 28, 1936) Turing sketches the proof of equivalence betweentwo definitions of computability of a sequence: in terms of a -machines and in terms of λ -calculus.Describing this result in the Introduction, he writes:In a recent paper Alonzo Church has introduced an idea of “effective calculabil-ity”, which is equivalent to my “computability”, but is very differently defined.Church also reaches similar conclusions about the Entscheidungsproblem. Theproof of equivalence between “computability” and “effective calculability” [i.e., λ -definability] is outlined in an appendix to the present paper.Independently of Turing (and almost simultaneously) Post publishes his paper [81], where heintroduces the notion of a “finite combinatory process” that is very similar to Turing machines.Some technical details are different; one should mention also that Post never speaks about amachine. He describes how a “problem solver or worker” follow “the set of directions” of afixed type. Then Post writes:The writer expects the present formulation to turn out to be logically equivalent torecursiveness in the sense of the G¨odel–Church development. Its purpose, however,is not only to present a system of a certain logical potency but also, in its restrictedfield, of psychological fidelity. In the latter sense wider and wider formulations arecontemplated. On the other hand, our aim will be to show that all such are logicallyequivalent to formulation 1 [the definition suggest by Post]. We offer this conclusionat the present moment as a working hypothesis . And to our mind such is Church’sidentification of effective calculability with recursiveness. h . . . i The success of theabove program would, for us, change this hypothesis not so much to a definition orto an axiom but to a natural law .In a footnote Post adds:Actually the work already done by Church and others carries this identificationconsiderably beyond the working hypothesis stage. But to mask this identificationunder a definition hides the fact that a fundamental discovery in the limitationsof the mathematizing power of Homo Sapiens has been made and blinds us to theneed of its continual verification. It is clear that in 1936 the puzzle (as we know it now) was almost completely assembled:there are several definitions of computability that are shown to be equivalent (the classes of Probably now this “fundamental discovery” has lost its value and even may be its meaning: when speakingabout the equivalence between the intuitive idea of algorithmic computability and a formal definition, we assumethat this intuitive idea was developed independently of any model of computation or programming language.But now it would be almost impossible to find anyone who learned the intuitive notion of algorithm beforehaving some programming experience. µ -operator though all the tools to prove the equivalence areready and this equivalence is mentioned explicitly by Kleene in 1943 [85, p. 53, Corollary].Second, more important difference is that all these papers consider only total functions(defined for all natural arguments). Partial functions appear only later, in Kleene’s paper [83](published in 1938) where the computable notation systems for ordinal are considered (andpartial computable functions are essential). Kleene describes the process of derivation in thesense of Herbrand and G¨odel and assumes that such a derivation exists only for one functionvalue (for given arguments). Then he writes:If we omit the requirement that the computation process always terminate, weobtain a more general class of functions, each function of which is defined over asubset (possibly null or total) of the n -tuples of natural numbers, and possesses theproperty of effectiveness when defined. These functions we call partial recursive.In this way the notion of a partial recursive function is introduced. Kleene considers substi-tutions and recursions (that can be naturally extended to partial functions), and then defines µ -operator for partial functions: µy [ R ( m, y ) = 0] = n for a partial function R if R ( m, n ) is defined and equals 0 while all previous values R ( m, , . . . , R ( m, n −
1) are defined and are not zeros. It is obvious that n with this property is unique; however, itmay not exist, and in this case the µ -operator defines a non-total function (that is undefinedon m ). Kleene notes that the class of partial recursive functions defined in the language of Her-brand and G¨odel is closed under all three operations (substitution, recursion and µ -operator).He notes also that for every n there exists a universal function Φ n ( z, x ) of n + 1 variables suchthat every partial recursive function of n variables x can be obtained from Φ n by fixing somevalue of the first argument z . This universal function Φ n can be represented asΦ n ( z, x ) = S ( z, µyT n ( z, x , y )) , where S is some primitive recursive function and T n is a primitive recursive predicate (sayingthat some primitive recursive function equals 0). Informally speaking, z is a natural numberthat encodes a system of functional equations (in Herbrand – G¨odel style) that defines somepartial recursive function of n variables, and y is an encoding of a derivation that, starting withthese equations, establishes the value of this partial recursive function on x . The predicate T n checks the correctness of this derivation, and the function S extracts the function value from it. This result is called “Kleene’s normal form theorem”; it implies that partial recursive functioncould be equivalently defined as functions that can be obtained by substitution, recursion, and µ -operator. One may also require additionally that the µ -operator is used only once (beingapplied to a primitive recursive functions). However, this way of defining partial recursivefunctions is not mentioned by Kleene.The same framework and terminology is used in a later paper of Kleene [85] (1943, wherehe consider the arithmetical hierarchy) and in his classical book of 1952 [65] that remained astandard reference for logic and computability theory for a long time. Let us mention again The traditional Russian translation of this name is частично рекурсивная функция . It sound even morestrange than общерекурсивная функция for general recursive functions; one could think that the function isnot completely recursive but only partially recursive. Kleene provides z as the first argument to the function S but this is not necessary. Relative (oracle) computability
One can define the notion of computability of a function relative to some other function (or set,if we identify sets with their characteristic functions). This definition was first considered inTuring’s Ph.D thesis (1939, see [84]); however, it was only a side remark and only reducibilityto some specific set was considered. Turing writes:Let us suppose that we supplied with some unspecified means of solving number-theoretic problems; a kind of oracle as it were. We will not go any further into thenature of this oracle than to say that it cannot be a machine. With the help of theoracle we could form a new kind of machine (call them o -machines), having as one ofits fundamental processes that of solving a given number-theoretic problem. Moredefinitely these machines are to behave in this way. The moves of the machine aredetermined as usual by a table except in the case of moves from a certain internalconfiguration o . If the machine is in the internal configuration o and if the sequenceof symbols marked with l is then the well formed formula A , then the machine goesinto the internal p or t according as it is or is not true that A is dual. The decisionas to which is the case is referred to the oracle.These machines may be described by tables of the same kind as used for the descrip-tion of a -machines, there being no entries, however, for the internal configuration o .The definition of Turing reducibility for the general case was given by Post in his famousarticle [86, Section 11] where he formulated Post’s problem (asking whether there exists arecursive enumerable non-recursive set X that it is not Turing-complete: not all recursivelyenumerable sets are reducible to X ). Formally speaking, Post considers the case when bothsets (the one being reduced and the other to which it is reduced) and recursively enumerable,but the definition is the same for the general case of arbitrary sets of natural numbers. ThePost’s definition follows the scheme sketched by Turing. Kleene in 1943 [85] suggests a differentapproach: we define general recursive functions using Herbrand – G¨odel derivations but extendthe list of “axioms” adding the full information about the values of some fixed total functions ψ , . . . , ψ k . The functions that are definable in this way are then called general recursivefunctions in ψ , . . . , ψ k :A function φ which can be defined from given functions ψ , . . . , ψ k by a series ofapplications of general recursive schemata we call general recursive in the givenfunctions; and in particular, a function φ definable ab initio by these means we call general recursive .However, Kleene does not develop this idea (which remains a side remark), and does not definerelative computability for the case of partial functions (only total functions are considered).In 1952 book Kleene extends the definitions to partial functions and proves that the resultingdefinition (in Herbrand – G¨odel style) is equivalent to the definition of relative computabilitygiven by Turing and Post [65, § • For the first time, the (now standard) definition of partial recursive functions in termsof substitutions, recursions, and µ -operator was stated explicitly (with a reference to an“idea of Boris Trakhtenbrot” [4, p. 22]). • It was shown (simultaneously with [65, §
69] and in much more clear way) that thisdefinition is equivalent to other definitions of (absolute and relative) computability. • For the first time, a “machine-independent” definition of relative computability was given.Here machine independence means that the definition does not use any model of compu-tation but only the class of computable functions. It was shown that this definition isequivalent to other definitions of relative computability. • Finally, it was the first paper that presents the model of computation suggested by Kol-mogorov (later it was published in a joint paper by Kolmogorov and Uspensky [16]), thedefinition of relative computability in terms of this model, and the proof of equivalenceof this definition to other definitions of relative computability.The third item in this list requires some clarifications. The Turing – Post definition ofrelative computability is a modification of the corresponding definition for (absolute) com-putability: we extend the class of Turing machines by allowing them to get “answers” from anoracle. Similarly, the Kleene’s definition of the relative computability modifies the definition ofthe computable (partial recursive) functions. So even if we have already agreed on the definitionof (absolute) computability, we still may not left this definition behind when defining relativecomputability. Instead, in the latter definition we need to return to the model of computationand make some modifications (that allow some kind of “oracle access”).On the other hand, Uspensky defines relative computability in terms of a dialog with anoracle, and this dialog should be computable in the sense that some (partial) functions thatdescribe this dialog should be computable. These function should describe the dialog in thefollowing sense: they specify the next question to the oracle (or output if no more questionsare needed) given the input and the list of previous questions and oracle answers.Now the “machine-independent” definitions of relative computability are quite standard.For example, one of them can be found in the classic textbook of Rogers [70, Section 9.2](without any references to previous work). One can also note that Uspensky’s definition has atechnical advantage: unlike the definition from [70] it can be naturally generalized to a partialoracles ψ , and the class of functions that are obtained in this way is equal to the closure ofthe partial recursive functions and ψ with respect to substitutions, recursions and µ -operator.However, Uspensky did not consider this generalization and always assumes that oracle is atotal function (though the proof could be easily adapted to the case of partial oracles). G¨odel’s incompleteness theory and theory of computability
The G¨odel incompleteness theorem and the class of recursive functions appeared not only atthe same time but also together like Siamese twins. The classical paper where G¨odel provedincompleteness of Principia Mathematica and related systems [75] also introduced the notionof a recursive function (a primitive recursive function in modern terminology, see above), andthis notion played an important technical role in the proof. Namely, several functions relatedto the encoding of formulas and proofs by natural numbers (their “G¨o numbers”) were defined Unfortunately this paper was not published, though both reports (by Kolmogorov, the thesis advisor, andby Petr Sergeevich Novikov, the reviewer) recommended its publication. So — alas — it hardly could play anyrole in the further developments. In 1950 he gave [87] a similar interpretation for theRosser’s version of incompleteness theorem: it corresponds to the existence of two inseparableenumerable sets. So all the crucial observations were made by Kleene before 1950. Still theexposition both in this 1950 paper and in the 1952 textbook [65] is intertwined with the languageof primitive recursive function (it is enough to say that the exposition in [87] starts by “Let T be the primitive recursive predicate so designated in a previous paper by the author”), and theembedding of the inseparable sets into a formal theory is not described explicitly.Shortly after than (but most probably, independently) Kolmogorov also realized the connec-tion between G¨odel’s incompleteness theorem and theory of algorithms. As Uspensky writes in[51, p. 323],At December 2, 1952 Kolmogorov explained me main ideas relating G¨odel’s incom-pleteness theorem for general calculi to the existence of [enumerable] sets that arenot recursive, and pairs of [enumerable] sets that can not be separated by a recursiveset. The explanation was quite concise (maybe, five minutes) but then he gave mea short written note entitled “G¨odel and recursive enumerability”, so I could readand copy it. The note was written just for himself, and it was not easy for me tounderstand both the note and his oral comments. Then it became more clear, andon May 8, 1953 Kolmogorov submitted my short paper “G¨odel’s theorem and thetheory of algorithms” to Soviet Math. Doklady. When Kolmogorov worked with hisstudents, he made them feel that they are the authors (and he became a coauthor ofhis students much more rarely than he deserved it) h . . . i a paper “On the definitionof an algorithm” was published in Успехи математических наук ; in this papermy role was essentially technical.Here Uspensky speaks about two papers [6, 16]. The second paper contains the detailed ex-position of a model of computation based on graph transformations that appeared already inUspensky’s master thesis [4] and is known as Kolmogorov – Uspensky machines (see above).The first paper [6] explains (without any reference to primitive recursive functions) that G¨odel’sincompleteness theory (formal arithmetic is incomplete and cannot be completed) is a corollaryof two facts: (1) there exist recursively inseparable enumerable sets; (2) this pair of inseparablesets can be embedded into the formal arithmetic (in modern language, can be m -reduced tothe pair (provable formulas, refutable formulas). Moreover, for every enumerable set of addi-tional axioms (that keeps the theory consistent) one can effectively point out a formula thatis is neither provable nor refutable in this extended system, and this fact is a corollary of theexistence of two effectively inseparable sets.Let me stress again that all these observations were made already by Kleene in [87]; it seemsthat Kolmogorov and Uspensky did not see that paper at the time. Uspensky’s paper [6] has Now people say “computably enumerable” instead of “recursively enumerable”. Since we do not considerother type of enumerations, we call these sets enumerable in the sequel.
10 reference to Kleene’s 1943 paper [85]; however, when speaking about inseparable enumerablesets, Uspensky does not refer to Kleene’s 1950 paper [87] where they were constructed andnotes only that they were constructed by Novikov (and provides a reference to Trakhtenbrot’spaper of 1953).Generally speaking, there are two complementary views on G¨odel’s theorem. The originalG¨odel’s argument is a version of the liar’s paradox. This self-referential paradox notes that thestatement “This statement is false” cannot be either true or false. If we consider instead thestatement “This statement is not provable” (which can be, unlike the previous one, formulatedin the language of arithmetic), we get a statement that is true and (therefore) not provable— or false and provable, but we assume that formal arithmetic is consistent. This reasoningdoes not rely on the theory of algorithms; however, to show that one can translate finitaryarguments into the language of formal arithmetic one can use primitive recursive functions asa technical tool (following G¨odel).On the other hand, G¨odel incompleteness theorem is a consequence of the existence of anenumerable undecidable set (or, in a more symmetric version, of the existence of two recursivelyinseparable enumerable sets). In this way self-referential nature of the argument is hidden. Butit is just moved to the proof of the existence of an undecidable enumerable set (or an inseparablepair). Indeed, this proof uses “diagonal argument” that goes back to Cantor, and this diagonalargument is of self-referential nature (the “diagonal” function appear when we apply a functionto its own number, or run a program on its own text).Much later Uspensky published a popular exposition of the proof of G¨odel’s theorem basedon the algorithms theory (together with the introduction to this theory) in [22]. The extendedversion of this paper was published as a brochure [25] (in the series “Popular lectures in math-ematics” published by Nauka publishing house in Moscow). This work is probably the mostaccessible (and correct) non-technical exposition of G¨odel’s incompleteness theorem in Russianliterature (at least if we consider its algorithmic side).In addition to that, these publications [22, 25] suggest a way to explain theory of algo-rithms that was quite unusual at the time (one may compare them to Rogers’ textbook [70]).Usually the exposition started with a detailed analysis of some specific model of computation.The choice of this model changed with time. Initially most expositions used partial recursivefunctions; then Turing machine became the preferred model. In Russia Markov and his schoolpreferred the so-called normal algorithms . The analysis of this model required a lot of efforts(and space). Only after that the readers can learn the basic facts like Post’s theorem (an enu-merable set with enumerable complement is decidable), etc. Of course, the impatient readercould skip the boring first part, but then all the considerations in the rest of the textbookbecame baseless.What can be done? Uspensky suggested the following approach used in [22] (and before inhis 1972/73 lectures, and may be even earlier). We consider the class of computable functionassuming that this class satisfies some properties (“axioms”). These properties include thefollowing ones: • some specific functions (e.g., the pair numbering functions) are computable; some specificconstructions (e.g., the conditional execution or loops) preserve computability; • The tracing axiom : for every algorithm A there exist a decidable set R whose elementsare called “traces”, and two computable functions α and ω . Informally, elements of R are traces of terminating runs of A on all possible inputs (that include all informationabout the computation); this set should be decidable since one can check that the traceis indeed a trace of A . The function α recovers the input from the trace; the function ω recovers the output. This is an informal explanation why this axiom is plausible; theformal requirement is only that A ( x ) = y if and only if there exists r ∈ R such that11 ( r ) = x and ω ( r ) = y . • The program axiom : there exists a decidable set P (whose elements are called “programs”and an algorithm U that can be used to apply an arbitrary program p ∈ P to arbitraryinput x (so the input of U is a pair h p, x i ). The axiom requires that every computablefunction f has some program p such that U ( p, x ) = f ( x ) for every x . The last equalitysign is understood as follows: either both sides are undefined or both sides are definedand equal.After we agree with these axioms, we can prove results about computability without goinginto the technical details. On the other hand, it is quite clear what is missing in this pictureto get a formally sound mathematical theory: • We need to choose some model of computation. • We need to be able to program (in this model) some constructions used in the proofs. Infact, they could be not so simple (recall the priority arguments, for example). • We need to prove the tracing axiom and the program axiom for this model.This looks like a good plan for the first introductory course in the theory of algorithmsthat postpones some things that could be postponed. The model of computation then could beintroduced later when proving the undecidability of specific mathematical problems or defin-ing the complexity classes. Still a psychological barrier remains: many people who are quitefluent in mathematics and can easily deal with complicated constructions still have a feelingof uncertainty when they touch the algorithms theory, but at least this barrier becomes moreexplicit. For the proof of incompleteness theorem we need one more axiom (that is not a consequenceof the previous ones): the arithmetization axiom saying that every computable function canbe expressed by an arithmetical formula. (Later this axiom can be proved for some specificcomputation model.)If we use this machine-independent approach to the computability theory, we are not allowedto refer to a model of computation when speaking about (say) program transformations ororacle computations. Instead, we should provide all necessary definitions using only the classof computable functions. As we have said, the definition of relative computability that hasthis form appeared (for the first time) in the master thesis of Uspensky. Then it was donefor enumeration reducibility. To deal with program transformations, Uspensky introduced thenotion of a “main numbering” (see the next section for the enumeration reducibility and mainnumberings).One can also note that this axiomatic approach to computability theory provide a formaljustification for the following standard observation: most results of the computability theorycan be “relativized”, i.e., remain true if we replace the class of computable functions by theclass of A -computable function for some oracle A . Here A can be a set or a total function.Indeed, one could check that all axioms (except, of course, the arithmetization axiom) for thisclass. After that we know that all theorems (derived from the axioms) are true for this class. Nowadays the situation is a bit different; one should take into account that most of the people have a lot ofprogramming experience when starting to learn computability. A modern version of Uspensky’s approach couldbe something like that: we start with a programming language that is familiar to the students, and add somelibrary functions: (a) an interpreter for this language, i.e., a function that gets two inputs, a program string p and some other string y , and simulates program p on input y ; this corresponds to the program axiom; (b)a step-by-step debugger that gets also the number n of steps that should be simulated (a combination of thetracing axiom and program axiom). One can even add a library function without arguments that returns theprogram text, this would make the fixed point theorem obvious. A -computable functions for all oracles A , is a consequenceof his axioms. It turned out that the (positive) answer is easy to get (after the question isstated), see [104]. Computable mappings of sets and enumeration reducibility
The notion of reducibility introduces by Turing and Post (and considered in the master thesisof Uspensky, see above) can be called “decision reducibillity”. If A is reducible to B , and B is decidable, then A is decidable. One may say that in this definition we “reduce the decisionproblem for A to the decision problem for B ”.In [9] Uspensky gives the definition of enumeration reducibility where we reduce the task“enumerate the set A ” to the task “enumerate the set B ”. This definition uses the notion of acomputable operation on sets (introduced in the same paper). Let us describe this notion.Let us consider the simple case when unary operation is applied to subsets of N and mapsthem also to subsets of N . Consider the set P ( N ) of all subsets of N as a topological space.Namely, for each finite set X ⊂ N consider the family O ( X ) of all subsets of N that aresupersets of X . The families O ( X ) and all their unions are considered as open in P ( N ).After the topology on P ( X ) is defined, we consider all mappings F : P ( N ) → P ( N ) that arecontinuous with respect to this topology. It is easy to check that all continuous F are monotone(if U ⊂ V , then F ( U ) ⊂ F ( V )), and the value F ( U ) is determined by the values F ( X ) forfinite subsets X ⊂ U (is the union of F ( X ) for all finite X ⊂ U . The values of F on finite sets X can be described by the set of pairs {h n, X i | n ∈ F ( X ) } (here n is a natural number, and X is a finite set of natural numbers.Uspensky gives the following definition: a continuous mapping F : P ( N ) → P ( N ) is a computable operation is the corresponding set of pairs (see above) is an enumerable sets. Notethat pairs h n, X i are finite objects, so the notion of an enumerable set of pairs makes sense.Now the enumeration reducibility is defined: a set A ⊂ N is enumeration reducible to a set B ⊂ N if there exists a computable operation F that maps B to A . Uspensky notes that Turingreducibility can be described in terms of enumeration reducibility: a total function ϕ is Turingreducible to a total function ψ (i.e., computable with oracle ψ ) if and only if the graph of ϕ isenumeration reducible to the graph of ψ . We can also characterize the Turing reducibility forsets in the same way; for that we consider the graphs of characteristic functions of those sets.He says also that one can characterize partial recursive operators in the sense of Kleene [65],but here the terminology is confusing (see the discussion below).Finally, in this paper ([9]) Uspensky notes that the definition of a computable operations interms of topology (discussed above) is equivalent to two “machine-dependent” definitions. Thecorresponding notions are called “Kolmogorov operations” and “Post operations” by Uspensky(though they do not appear explicitly in the works of Kolmogorov and Post).In another 1955 paper ([10], see also an exposition of its results with some extensions in [12])Uspensky introduces the notion of a numbering (following Kolmogorov’s talk given in 1954 atthe seminar on recursive arithmetic, Moscow State University mathematics department), intro-duces the notion of a “main numbering” ( «главная нумерация» in Russian) and related thecomputable operations on enumerable sets (as defined in [9]) with algorithmic transformationsof their numbers.Let us explain Uspensky’s contribution in more detail. Assume that we want to considercomputable transformations of programs for computable functions (or enumerable sets). Thenit is not enough to know which functions are computable (or which sets are enumerable). Weneed also to make some assumptions on the “programming methods” (or languages, «способыпрограммирования» in Russian that are used for establish the correspondence between pro-13rams and computable functions. Programs are usually strings (words), but one could identifystrings with natural numbers via some computable bijective numbering of strings. Then aprogramming language (method) for computable functions defines a universal function of twoarguments: U ( n, x ) is the output of the n th program on input x (we assume that inputs andoutputs are also natural numbers). A programming language for enumerable set defines a uni-versal set of pairs h n, x i such that x belongs to the n th enumerable sets. In a different (butequivalent) language one may say that a programming method for computable functions (resp.enumerable sets) is a natural numbering of the set of all computable functions (enumerablesets), i.e., a (total) mapping of N onto the set of all computable functions (enumerable sets): anumber n is mapped to a computable functions (enumerable set) that corresponds to the n thprogram.Not all programming methods (numbering) are equally good. A reasonable theory that de-scribes the algorithmic transformations of programs needs some additional assumptions. Theseassumptions essentially appeared in Kleene’s work under the name of “ s - m - n -theorem”, butappeared explicitly for the first time in [10] where Uspensky defines the notion of a main num-bering. This definition consists of two requirements. First, to be main, a numbering shouldbe computable. This means that the corresponding universal function is a computable partialfunction of two arguments (for the case of sets: the corresponding universal set of pairs isenumerable). Second, any other computable numbering should be reducible to the main num-bering. This means that for any other computable numbering of the same family there existsa computable translation functions that transforms a number in this other numbering into anumber of the same function (set) in the main numbering.Fix some main numbering for the family of enumerable sets. Then we may define computablemappings of this family into itself. Here computability of a mapping P means that there existan algorithm that, given a number of some enumerable set X , returns (some) number for the set P ( X ). In other words, we consider computable transformations of programs (or numbers) thatpreserves the equivalence relation: if two program p and p ′ are equivalent, i.e., are programs ofthe same set, then they are transformed into two equivalent programs. Uspensky proved [10,Section 6] that computable mappings of the family of enumerable sets are exactly computableoperations on the family of all sets, restricted to the subfamily of enumerable sets. He alsoproved a similar statement for a subfamily of function graphs: every computable mapping ofthe family of computable functions into itself is a restriction of a computable operation on thefamily on all function graphs.Let us describe the connections of this work of Uspensky to the other research of that time. Rice [88] considered completely recursively enumerable classes of enumerable set. A family X of enumerable set is called completely recursively enumerable if the set of all programs forall elements of X is enumerable. Rice formulated a conjecture [88, p.361]: every completelyrecursively enumerable family is the family of all supersets of finite sets from some enumerablefamily of finite sets. This conjecture becomes Theorem 5 in Uspensky paper [10, Theorem5] (1955) and is a crucial point in the proofs of his results about computable transformations. The definition of reducibility for numbering also was published in [9] with a reference to Kolmogorov’seminar talk, also probably for the first time. Unfortunately (see below the quote from Uspensky’s memoirs) all three publications of him [9, 10, 12] areshort notes in the
Soviet Math. Doklady [9, 10] and a resume of a talk in the Moscow Mathematical Society [12];they contain only the statements of the theorems and lemmas used in the proofs. The full proofs were publishedin Uspensky’s PhD thesis [11]. Formally speaking, this thesis was publicly available (it can be ordered andaccessed in few libraries in the USSR), but it hardly could influence the developments in the field. Probably theshort notes [9, 10, 12] were not read outside the USSR, too. Later Uspensky wrote a monograph [18] that becomehis “habilitation text” ( «докторская диссертация» ); this book was translated into French. Unfortunately,it included only the definition of main numberings, but not the results on computable transformations andmappings. § § «сводимость поперечислимости» used by Uspensky).The notion of a main numbering ( «главная нумерация» in Uspensky’s terminology) wasalso rediscovered by Rogers (see [94]) under the name of “G¨odel numbering”. Rogers startswith a “machine-dependent” definition: “A G¨odel numbering is a numbering equivalent to thestandard numbering” (p. 333); however, later he provides a machine-independent characteri-zation (as the maximal element with respect to reducibility — as in the Uspensky definition,though without references to Uspensky). Nowadays the names “admissible numbering” (see,e.g., Soare’s book [112]) and “acceptable numbering” (see, e.g., [108, Definition II.5.2]) are used;in both cases a “machine-dependent” definition is given.When comparing Uspensky’s work to the similar publications of others, one should havein mind that there are different (and often mixed) notions of reducibility for partial functions.Assume that f and g are two partial functions (with natural arguments and values). Considerthe following three definitions of “ f is reducible to g ” (= f is computable relative to g ); eachof them is strictly stronger than the previous ones:1. The graph of f is enumeration reducible to the graph of g .2. Consider (following Uspensky) the family U of all partial functions with natural argumentsand values, and consider the following topology in U : the basic open sets are sets ofall extensions of some finite partial functions. Call a continuous mapping F : U → U a computable operation if its restriction to finite functions has an enumerable graph,i.e., if the set of all pairs hh x, y i , u i , where x and y are natural numbers, u is a finitepartial function and [ F ( u )]( x ) = y , is enumerable. Then we require that there exists acomputable operation that maps g to f .3. We may extend Trakhtenbrot’s definition (see the discussion of Uspensky’s master thesisabove) to partial function and require that f belongs to the closure of the family of allpartial recursive functions with g added under substitution, recursion and µ -operation.(This requirement appears, for example, in [100].)15he third condition in this list can be equivalently reformulated in the oracle computationslanguage. This reformulation repeats the definition from Uspensky’s master thesis but allowspartial functions (that were not considered by Uspensky). Namely, an algorithm, given x , com-putes f ( x ); it is allowed to ask questions about g ( y ) for arbitrary y — but it should be donesequentially and as soon as it asks for g ( y ) that is undefined, the computation hangs withoutproviding any result (so f ( x ) remains undefined for the corresponding x ). The second require-ment also can be reformulated in terms of oracle computations if we allow asking questionsabout several values g ( y ) in parallel (the computations continues while waiting for the oracle’sanswers; it is required that the result of the computation does not depend on delays before theoracle answers are provided).To see why each requirement is stronger than the previous one, we may consider two exam-ples. The first example separates the first two requirements.Let f be an arbitrary total function with natural arguments and values. Let g be a partialfunction whose values are all zeros, and whose domain is the set of all numbers of pairs h n, f ( n ) i for all n . (We assume that some computable numbering of pairs is fixed.) Then the firstrequirement is true for these f and g while the second one is false unless f is computable itself(a computable mapping that maps g to f should map the zero function to f , since the zerofunction extends g ). This example is mentioned in the Uspensky’s footnote to the Russiantranslation of Rogers’ book [70, p.362] with a reference to D.G. Skordev; the original argumentof Rogers is much more complicated.The second example [108, Proposition II.3.20, with a reference to Sasso’s 1971 thesis] showsthat the third property is stronger than the second one. Let g be an arbitrary partial functionwith natural arguments that has only zero values. Construct another partial function f , alsowith zero values, in the following way: the value f ( n ) is defined (and equals 0) if and only if atleast of the one values g (2 n ) and g (2 n + 1) is defined. Then the second requirement is satisfiedfor sure: for input n we ask in parallel what are the values g (2 n ) and g (2 n + 1); as soon asone of the answers is given, we return 0. However, if we have to ask the oracle sequentially,this argument does not work: if we first ask for g (2 n ) and g (2 n ) is undefined, then f ( n ) isundefined even if g (2 n + 1) is defined. (Of course, this is only an explanation why the previousconstruction is no more valid; to show that indeed the third requirement may be false we needa simple diagonal argument.)The first requirement corresponds to the notion that is called “partial recursive operators”in Rogers’ book [70, § § ∗† from Kleene’s book [65, p. 332]. However, this Thesis (see the top of p. 332) does notuse the name “partial recursive functional” that does not appear on p. 332 at all. The subjectindex refers to page 326 for “partial recursive functional”, but this page does not mention sucha notion. It defines the notion of a partial function ϕ that is partial recursive relative to par-tial functions ψ , . . . , ψ k that corresponds to our first requirement (enumeration reducibility ofgraphs) and mentions some “scheme” F but does not say whether this scheme F should definea function for all possible ψ , . . . , ψ k or only for the specific functions. (The numberings of allfunctions that are computable with an oracle are considered only for the case when the oracleis total.) Still Myhill and Sheperdson clarify the situation and say that for their result theyneed partial recursive functionals that are defined (and produce functions) for all argumentsthat are functions, so essentially they consider the second requirement (as well as Uspensky inhis 1955 papers).Odifreddi in [108, Definition II.3.6] defines partial recursive functionals with reference toKleene [65]; however, he uses the third version of the definition (a composition of substitutions,recursions and µ -operators applied to partial recursive functions and input functions) — one16hat does not appear in [65]. He uses the names “effectively continuous functional” or “recursiveoperator” for the second requirement and the name “partial recursive operator” for the firstone. He uses topological notions in his definitions (as Uspensky did).Let us summarize the contribution of Uspensky’s papers [9, 10, 12]: • the historically first definition of enumeration reducibility; • the definition of a numbering and reducibility of numbering was published for the firsttime (with reference to Kolmogorov’s talk); • the analysis of the properties of numberings of computable functions and enumerable setsneeded for the results about program transformation; the definition of main numberings(later rediscovered by Rogers); • the proof of Rice’s conjecture about completely recursive enumerable classes of enumer-able sets (and similar results for functions, including the undecidability of all non-trivialproperties of computable functions); • the definition of a computable operation (in topological terms) and the proof that algo-rithmic transformations of programs for computable functions or enumerable sets can bedescribed as restrictions of computable operations on functions or sets.As we have said, these achievements were unavailable to the international community andthe corresponding results were independently obtained by other researchers (at the same timeor a bit later). Let us note, to avoid possible misunderstanding, that Uspensky does not consider algorithms that are defined on all programs of total functions and give the same resultsfor equivalent programs. The corresponding work of Kreisel, Lacombe and Shoenfild (1959,see [96]) later generalized by Tseitin [98] to constructive metric spaces, have no intersectionswith Uspensky’s work.In the following quote from Uspensky’s memoirs ([63, p. 905–907, 912]) he recalls his 1955results and the Third All-Union Mathematical Congress (1956) where these results were pre-sented:In the survey talk (June, 26) “On algorithmic reductions” I spoke about four kindsof reductions and relations between them. These four notions are the following:First, computability reduction where the task “compute f ” for some function f isreduced to the task “compute g ” for some other function g . Second, the decidabilityreduction : the task “construct a decision procedure for A ”, where A is some set, isreduced for the same task for some other set B . Third, the enumerability reduction:the task “enumerate A ” for some set A is reduced to the same problem for some otherset B . Finally, this is reduction of mass problems that reduces one mass problem toanother one h . . . i The notion of mass problems was introduced by Yury Medvedev,who was Kolmogorov’s student, who defined also the corresponding reductions. h . . . i Another talk of mine (July, 2) was named “The notion of a program and com-putable operators”, and a short communication (July, 3) “Computable operations,computable operators and effectively continuous functions” was closely related tothat talk.In the last communication I formulated (without proof, of course) the result whichnow I consider as my main mathematical achievement and still remember the cir-17umstances when it came to my min; it was called “Theorem 3” (see below). Thisresult was the core of my Ph.D. thesis that was defended in October 1955. I neverpublished the proof of this result, except for the thesis itself; this thesis is available(or at least was available) in the math department library. Why? Mostly due tomy laziness (shame on me). Another reason, may be less embarrassing, but stupid,was my desire to present this result in the most general form (but one cannot reachthe limits of generalization). h . . . i Theorem 3 . Let g be a function with natural arguments and values. Assumethat this function has the following property: if m and n are programs of the samecomputable s -ary function, then g ( m ) and g ( n ) are programs of the same unaryfunction. Then there exists a computable operator V such that for every function θ with program n the value V ( θ ) is a function with program g ( n ) . A philosophical comment : a semiotic interpretation of Theorem 3 goes as fol-lows: a “well-behaved” computable transformation of names is accompanied by acomputable transformation of named objects.
Constructivism and classical mathematics
The idea of a constructive interpretation of mathematical statements (and, more general, logicalconnective) goes back to Brouwer and his “intuitionistic” school; later it was developed in adifferent way by Andrei Markov, jr., and his students under the name of “constructivism”. Inparticular, the constructive interpretation of the statement “for every x there exists y suchthat. . . ” is that there exist a way to get this “existing” y for every value of x .Usually this constructive approach was combined with the change in the understandingof logical connectives (that makes the excluded middle law invalid). Still there is anotherpossibility that initially was not very popular: consider the “effective” versions of classicalnotions and results as a part of usual (“classical”, “non-constructive”) mathematics that usesstandard mathematical tools. Many people thought that if we are studying algorithms, thisshould be done in some “constructive” or “finitistic” way. Uspensky stressed that this isnot the only option and one can study constructive notions inside the classical universum ofmathematics.Here are two examples that he considered. The first is the notion of a computable real num-ber. There are different construction of real numbers (Dedekind cuts, fundamental sequences,common points of intervals of decreasing lengths, decimal expansions, etc.). For each of theconstructions one can consider its effective version. For example, we can consider Dedekindcuts such that there exists an algorithm that says for a rational number whether it belongs tothe left or right part. For a fundamental sequence x n of rational numbers one may require thissequence to be computable (given n , one can compute x n ), and also require the existence of acomputable modulus of convergence (an algorithm that, given rational ε >
0, computes some N such that | x k − x l | < ε for all k, l > N ). For an infinite decimal fraction one may requirethe computability of the function n ( n th digit), and so on.Each of these definitions leads to some subset of R that consists of the numbers that haveeffective representations in the corresponding sense. One can ask (still working in the frameworkof classical mathematics) whether these definitions lead to the same subset of to different ones. Theorem 3 was interesting for me from the semiotic viewpoint, even if I did not know the word “semiotics”at that time. I remember how I was walking along Moscow streets thinking about this question only. Theinsight came when I was at my mother-in-law apartment (on Big Spasoglinitschevskii lane in Moscow). Myson was not born yet, my wife and her mother went to their jobs in the morning, there was no phone in theapartment (and, of course, no mobiles!). Suddenly I’ve understood how it works. [Uspensky’s footnote]
18t is not difficult to see that they define the same subset (in different ways), and the elementsof this subset can be called computable real numbers (following Turing [82]).This example can be used to illustrate the difference with Markov-style constructivism. Forconstructivists there are no such things as “real numbers” in the usual sense, so they cannotconsider the set of computable real numbers as a subset of the set of all real numbers. Forthe a (computable) real number is a pair of algorithms: one, given n , computes x n , and theother computes the modulus of convergence. Note that not all definitions mentioned aboveare equally good. For example, the definition with decimal fractions has problems: we cannotdefine addition, i.e., there is no algorithm that transforms two constructive real numbers (i.e.,the algorithms for their representations) into their sum (i.e., the corresponding algorithm).However, as Uspensky notes, the same problem can be analyzed in the framework of classicalmathematics. For that, we consider numberings of computable reals that correspond to differentdefinitions. We may ask then whether these numberings are equivalent (whether one canalgorithmically transform the number of a computable real in one numbering into a numberof the same real in another numbering). And here the same problem with decimal fractionsreappears — and the other positional systems also have this problem. In [18] Uspensky providesnecessary and sufficient conditions for the reducibility of two numberings of computable reals(with different bases).Another example studied by Uspensky [19]: the effective versions of the notion of an infiniteset of natural numbers. We may say that a set X is infinite if for every natural n the set X contains at least n different elements. Or: X is infinite if it differs from any finite set X :for every finite F there exists some number that belongs to the symmetric difference F △ X .Both definitions lead to natural effective versions. In the first case we require that there is analgorithm that, given n , produces a list of n different elements of X . In the second case werequire that there is an algorithm that, given a finite set X , produces some element of F △ X .It is easy to see that these two effective definitions are equivalent (and we may even modify thesecond definition requiring only that the algorithm gives an element of X \ F for finite subsets F of X ). Using the terminology from Post’s paper [86] all these properties are equivalent tonon-immunity of X (i.e., to the existence of an enumerable infinite subset of X ).On the other hand, not all definitions of infinity lead to equivalent effective versions. Forexample, we may say that x is infinite if for every n there exists an initial segment [0 , N ]that contains at least n elements if X . The effective version of this definition would be: thereexists an algorithm that for every n computes some N with this property. This is a weakerproperty of “effective infiniteness”: as Uspensky noted in [15](answering the question of Kol-mogorov; A.V. Kuznetsov and Yu.T. Medvedev independently answered the same question),this requirement means that the set is not hyperimmune in the sense of Post [86].One may also note (though this has no relation to Uspensky’s work) that the basic definitionin algorithmic randomness, the definition of randomness given by Martin-L¨of in 1966 [101] isalso an effective version of the definition of a null set (a set of Lebesgue measure 0). Thisclassical definition says that a set X ⊂ [0 ,
1] is a null set if for every ε > X by intervals whose total measure does not exceed ε . For obvious reason wemay consider only rational values of ε and only interval with rational endpoints. Then both ε and the intervals are constructive objects, and one may consider the effective version of thedefinition and require that an algorithm gets ε > Algorithmic information theory
It is strange that Uspensky, being a student of Kolmogorov and his colleague at the MathematicsDepartment of the Moscow State University, was not involved in the research initiated byKolmogorov in 1960s when he introduced the notion of algorithmic complexity of finite objects(now known also as Kolmogorov complexity). I have asked him about that but it still remains akind a mystery for me. As Uspensky told me, he came into this field only when preparing (withAlexei L. Semenov) the talk for the Urgench conference [24, 26]. In this talk Uspensky andSemenov suggested a general scheme for defining different versions of complexity (or algorithmicentropy, as Uspensky preferred to name them) known at the time: plain, prefix, monotone,decision entropies, as well as conditional versions of entropy. Initially (see [107]) this approachused the notions of f -spaces and their continuous mappings. In a sense this can be consideredas an extension of the topological approach to computability suggested by Uspensky long ago.However, this was definitely an overkill, and Uspensky and Semenov [24, 26] suggested a muchmore simple version of this scheme that used only the “compatibility relation” on objectsand descriptions that is enough to cover most of the cases. Later this simplified scheme wasexplained in [39, 41]; a detailed exposition from the topological viewpoint (but without f -spaces) can be found in [59]The different notions of randomness are discussed also in a survey [35] and in a mono-graph [59]. In 2005 Uspensky gave a talk at the “Modern mathematics” school for undergrad-uates devoted to algorithmic randomness. A brochure based on this talk was published in2006 [49] and was reprinted as a part of a monograph [59].One of the questions asked by Uspensky, Semenov and An. Muchnik [44] remains open.They asked whether the Martin-L¨o randomness is equivalent to the absence of a computablestrategy in non-monotone games (“non-predictability”). See [50, 59] for more details. Popular science
There are different ideas about “popular science” (in French one says “vulgarization”, and itsounds embarrassing though partially correct). One may tell stories about life and fate of greatscientists. One can try to retell stories found in other popular science books adding more funnyjokes. All this may be a good thing, but Uspensky’s approach was different. During all his lifehe tried to explain faithfully the real scientific achievements. These explanation could be easilyaccessible or technically difficult (depending on the audience); still it was always a serious andhonest explanation of a material that can be explained with a clear indication of what remainswithout proof (or clarification). And he never was afraid of explaining basic and “well known”things: as Aristotle wrote in
Poetics , “subjects that are known are known only to a few”.While being a student, Uspensky (with a senior coauthor, Evgeny B. Dynkin) wrote abook [4] that was based on the materials of mathematical circles in Moscow. Uspensky firstwas a participant of these circles, and later one of the teachers there. The book covers severaltopics (graphs’ coloring, the basics of number theory and probability theory). These topics arepresented as a sequence of problems (as it was done in the circles’ meetings), and the solutionsof these problems are provided. This was not the first problem book based on the materialsof mathematical circles, but and important new idea was that these problems, taken together,form a coherent exposition of some mathematical theory. This book for a long time was veryhard to find (before it was reprinted in 2004 and before its appearance on the Internet).Several popular brochures written by Uspensky were based on his lectures for high school20tudents (in particular, for the participants of the mathematical olympiads) and appeared inthe series “Popular lectures on mathematics”. Some of there were not related to his own math-ematical specialty: he wrote a brochure about applications of mechanics to mathematics [17]and about Pascal’s triangle [20]. The latter includes also a philosophical discussion: what is acombinatorial problem and why do we fix the list of operations that are allowed in the answerfor such a problem (e.g., including factorials but not the notation for binomial coefficients).Two other brochures in this series written by Uspensky (“The Post machine” [23] and “TheG¨odel incompleteness theorem”) are covering topics from mathematical logic and algorithms’theory. The first is quite elementary and is based on the lessons given by Uspensky to elementaryschool students. The other one (as we have mentioned) is based on the article publishedin
Russian Mathematical Surveys and assumes significant mathematical culture (but still isaccessible to competent high school students). One more popular exposition [27] written byUspensky was devoted to the non-standard analysis where the tools from mathematical logicare used to proved a mathematically correct approach to infinitesimals. The extended versionof this brochure was published few years later [31].Like Josef Knecht (from Hesse’s
Das Glasperlenspiel ) Uspensky switched to more andmore basic things when becoming older. He started to preach mathematics among human-ities students (and researchers). This preaching started in 1960 when he developed andimplemented the mathematics curriculum for the Division of Theoretical and Applied Lin-guistics of the Philology Department of the Moscow State University. However, duringthe two last decades of his life he addressed to a much wider audience. Several of his lec-tures during the summer school on mathematics and linguistics (in Dubna, a town nearMoscow) were videotaped (thanks to Vitaly Arnold) and are available (see the referencesin ). Theygive some idea about Uspensky’s approach to teaching, but one could fully appreciate it onlyduring university courses (first of all, a non-obligatory ones, «спецкурсы» in Russian). Us-pensky always was preaching mathematics, not preaching “about mathematics”. He explainedsimple things, but seriously and with proofs. One of his last books [56] is even called “Verysimple examples of mathematical proofs” (probably not a good name from the advertisingviewpoint). The other book [45] was named “What is an axiomatic approach?”, and it alsocontains a lot of examples, including “school geometry” — not the part that is taught in highschool but the axiomatic part that is omitted. For example, this book explains how one canderive from the axioms that for every line there is a point that does not belong to this line.The materials from these two books were included in a collection of Uspensky’s papernamed “Mathematics’ Apology”[54], together with the some other (more general) essays aboutmathematics. And strangely his preaching was successful — at least if we interpret successin the same sense as for Saint Anthony of Padua’s preaching to the fish: in 2010 Uspenskygot the “Enlightenment” award established by Dmitry Borisovich Zimin, Russian engineer andphilanthropist, the founder and main sponsor of the
Dynasty foundation.In addition to his own books, Uspensky organized the translation and publication of manyclassical textbooks: he translated (following the suggestion of Kolmogorov) R. Peter’s bookon recursive functions [64], was the editor for the translations of monographs of Kleene [65],Rogers [70], Davis [71] (the latter translation probably was the first Russian-language bookabout non-standard analysis), Church’s logic textbook [67], the first volume of the “Elementsof mathematics” by Bourbaki [68], and Ashby’s book on cybernetics [66].21 eferences
Uspensky’s papers in the Internet: [1] Uspensky’s page at mathnet.ru : [2] Uspensky’s page at the Logic and Theory of Algorithms Division of the MathematicalDepartment of Moscow Lomonosov State University: http://lpcs.math.msu.su/~uspensky/ Publications [3] A geometric approach to proving main properties of harmonic functions[
Геометрический вывод основных свойств гармонических функций , in Russian]
Успехи математических наук , 1949, vol. IV, issue 2(30), p. 201–205, http://lpcs.math.msu.su/~uspensky/bib/Uspensky_1949_UMN_Geometr_vyvod.pdf , http://mi.mathnet.ru/umn8612 [4] A general definition of algorithmic computability and algorithmic reducibility.
Общееопределение алгоритмической вычислимости и алгоритмической сводимости , inRussian . Master thesis (advisor A. Kolmogorov) Moscow State Lomonosov University,Math. Department механико- математический факультет . A typescript. 90 pp. http://lpcs.math.msu.su/~uspensky/bib/Uspensky_1952_Diploma.pdfhttps://archive.org/details/uspensky-1952-master-thesis-and-reviews [5] E.B. Dynkin, V.A. Uspensky,
Mathematical Conversations: Multicolor Problems,Problems in the Theory of Numbers, and Random Walks . [
Е. Б. Дынкин,В. А. Успенский,
Математические беседы. Задачи о многоцветной раскраске.Задачи из теории чисел. Случайные блуждания , in Russian] (Mathematical circles’library, vol. 6. [
Библиотека математического кружка, вып. 6 ]). Moscow, Leningrad:State Publisher of Technical and Theoretical Books. [
Москва– Ленинград:Государственное издательство технико- теоретической литературы ], 1952. http://ilib.mccme.ru/djvu/bib-mat-kr/besedy.htm , . 2d ed.: Nauka[ Наука ,] 2004. English translation published by D.C. Heath, Boston in 1963 (in threebrochures) and later was published as one book: E.B. Dynkin, V.A. Uspenskii,
Mathematical Conversations: Multicolor Problems, Problems in the Theory of Numbers,and Random Walks , Dover books in mathematics, Dover Publications, 2006, ISBN0-486-45351-0.[6] On the notion of algorithmic reducibility. [
О понятии алгоритмической сводимости , inRussian] A summary of a talk given at the meeting of Moscow Mathematical Society,March 17, 1953.
Успехи математических наук , vol. VIII, issue 4(56), 1953,July–August, p. 176, see http://mi.mathnet.ru/umn8234 [7] G¨odel’ theorem and the theory of algorithms [
Теорема Гёделя и теория алгоритмов , inRussian]. A summary of a talk given at the meeting of Moscow Mathematical Society,March 24, 1953.
Успехи математических наук , vol. VIII, issue 4(56), 1953,July–August, p. 176–178, see http://mi.mathnet.ru/umn8234 [8] G¨odel’ theorem and the theory of algorithms [
Теорема Гёделя и теория алгоритмов , inRussian],
Доклады Академии наук СССР , vol. 91, issue 4, p. 737–740 (1953), https://istina.msu.ru/publications/article/92662634/ ,22 ttps://archive.org/details/uspensky-1953-dan-091-4-godel-algorithms .English translation: G¨odel’s theorem and the theory of algorithms,
AmericanMathematical Society Translations, Series 2, Advances in the Mathematical Sciences ,vol. 23 (1963), 103–107,
DOI10.1090/trans2/023/06 [9] On computable operations [
О вычислимых операциях , in Russian],
Доклады Академиинаук СССР , vol. 103, issue 5 (1955), p. 773–776, https://istina.msu.ru/publications/article/92662640/ , https://archive.org/details/uspensky-1955-dan-103-5-computable-operations [10] Systems of enumerable sets and their numberings [ Системы перечислимых множеств иих нумерации , in Russian],
Доклады Академии наук СССР , vol. 105, issue 6 (1955),p. 1155–1158, https://istina.msu.ru/publications/article/92662649/ , https://archive.org/details/uspensky-1955-dan-105-6-enumerable-sets-numerations .[11] On computable operations [ О вычислимых операциях , in Russian], Ph.D thesis,Moscow State Lomonosov University, Mathematics Department, October 1955.[12] Computable operations and the notion of a program. [
Вычислимые операции и понятиепрограммы , in Russian]. A summary of a talk given at the meeting of MoscowMathematical Society, February 28, 1956,
Успехи математических наук , vol. XI,issue 4(70), 1956, July–August, p. 172–176, http://mi.mathnet.ru/umn7861 [13] Third All-Union Mathematical Congress. Plenary talk “On algorithmic reducibility”.[ «Об алгоритмической сводимости» , in Russian], June 26, 1955. Talk “The notion of aprogram and computable operators” [ «Понятие программы и вычислимыеоператоры» , in Russian], July 2,1955. Short talk “Computable operations, computableoperators and constructively continuous function” [ «Вычислимые операции,вычислимые операторы и конструктивно-непрерывные функции» , in Russian], July 3,1955. The resumes of the talks is published: Proceedings of the Third All-UnionMathematical Congress [
Труды третьего всесоюзного Математического съезда , inRussian]. Moscow: Academy of Science Publications, 1956. Vol. 2, p. 66–69 (plenarytalk), vol. 1, p. 186 (talk), vol. 1, p. 185 (short talk).[14] On the uniform continuity theorem. [
К теореме о равномерной непрерывности , inRussian].
Успехи математических наук , vol. XII, issue 1(73), 1957, January–February,p. 100–142, http://mi.mathnet.ru/umn7524 [15] Some remarks on [recursively] enumerable sets [
Несколько замечаний о перечислимыхмножествах , in Russian].
Zeitschrift f¨ur mathematische Logik und Grundlagen derMathematik , Bd. 3, Heft 12, S. 157–170 (1957), https://onlinelibrary.wiley.com/toc/15213870/1957/3/12 , https://istina.msu.ru/publications/article/92666188/ , https://archive.org/details/uspensky-1957-zml-zamechanie-perechisl-mnozhestv .English translation: Some remarks on recursively enumerable sets, AmericanMathematical Society Translations, Series 2, Advances in the Mathematical Sciences ,vol. 23 (1963), 89–101,
DOI10.1090/trans2/023/05 [16] A.N. Kolmogorov, V.A.Uspensky, On the definition of an algorithm [
К определениюалгоритма , in Russian].
Успехи математических наук , vol. XIII, issue 4(82), 1958,July–August, p. 3–28, http://mi.mathnet.ru/umn7453 . English translation (ElliottMendelson): Kolmogorov A.N., Uspenskij V.A., On the definition of an algorithm,
American Mathematical Society Translations, Series 2, Advances in the MathematicalSciences , vol. 29 (1963), 217–245,
DOI10.1090/trans2/029/07
Some application of mechanics to mathematics [ Некоторые приложения механики кматематике , in Russian] (Popular mathematics lectures [
Популярные лекции поматематике ], issue 27). Moscow, State Physics and Mathematics Publishers[
Государственное издательство физико- математической литературы ], 1958. 48 pp., https://math.ru/lib/plm/27 [18] On the relations between different systems of constructive real numbers [
К вопросу осоотношении между различными системами конструктивных действительных чисел ,in Russian],
Communication of High Education Institutions [
Известия высших учебныхзаведений , mathematics, 1960, issue 2(15), p. 199–208, http://mi.mathnet.ru/ivm2028 [19]
Lectures on computable functions [ Лекции о вычислимых функциях ]. Moscow StatePhysics and Mathematics Publishers [
Государственное издательствофизико- математической литературы ], 1960. 492 pp., https://archive.org/details/uspenskij-1960ru . French translation: OuspenskiV.A., Le¸cons sur les fonctions calculables. Paris, Hermann, 1966. 412 p.[20]
Pascal’s Triangle [ Треугольник Паскаля , in Russian]. (Popular mathematics lectures[
Популярные лекции по математике ], issue 43). Moscow, Nauka Publishers, Physicsand Mathematics Division [
Наука, главная редакция физико- математическойлитературы ], 1966. 35 pp. Second extended edition: 1979. 48 pp. [21] Reductions of computable and potentially computable numberings [
О сводимостивычислимых и потенциально вычислимых нумераций , in Russian]
Математическиезаметки , vol. 6, issue 1 (1969), p. 3–9, http://mi.mathnet.ru/mz6891 / Englishtranslation: V.A. Uspenskii, Reduction of computable and potentially computablenumerations,
Mathematical Notes of the Academy of Sciences of the USSR , 1969, vol. 6,no. 1, 461–464,
DOI10.1007/BF01450246 , https://istina.msu.ru/publications/article/92645436/ [22] An elementary exposition of G¨odel’s incompleteness theorem [ Теорема Гёделя онеполноте в элементарном изложении , in Russian].
Успехи математических наук ,vol. XXIX, issue 1(175), p. 3–47 (1974, January–February), http://mi.mathnet.ru/umn4322 . English translation: (E. Lichfield): Uspenskii V.A.,An elementary exposition of G¨odel’s incompleteness theorem,
Russian MathematicalSurveys , vol. 29 (1974), no. 1, 63–106,
DOI10.1070/RM1974v029n01ABEH001280 , full textavailable at https://istina.msu.ru/publications/article/92645484/ [23]
Post’s machine [ Машина Поста , in Russian]. (Popular mathematics lectures[
Популярные лекции по математике ], issue 54). Moscow, Nauka Publishers, Physicsand Mathematics Division [
Наука, главная редакция физико- математическойлитературы ], 1979. 96 pp., [24] Uspensky V.A., Semenov A.L., What are the gains of the theory of algorithms: Basicdevelopments connected with the concept of algorithm and with its application inmathematics,
Algorithms in Modern Mathematics and Computer Science, Proceedings,Urgench, Uzbek SSR, September 16–22, 1979 . Edited by A.P. Ershov and D. Knuth,Lecture Notes in Computer Science, 122, Springer, 1981, p. 100–234, https://archive.org/details/uspensky-semenov-urgench [25]
G¨odel’s incompleteness theorem [ Теорема Гёделя о неполноте , in Russian]. (Popularmathematics lectures [
Популярные лекции по математике ], issue 57). Moscow, Nauka24ublishers, Physics and Mathematics Division [
Наука, главная редакцияфизико- математической литературы ], 1982, . English translation by N. Kolblitz waspublished in1987 (Mir Publisher) and later in
Theoretical Computer Science , (1994),239–319, [26] Uspensky V.A., Semenov A.L. Theory of algorithms: main discoveries and applications.[ Успенский В.А., Семёнов А.Л., Теория алгоритмов: основные открытия иприложения , in Russian]. Published in the proceedings volume:
Algorithms in modernmathematics and its applications. The proceedings of the international symposium,Urgench, Uzbek SSR, September 16–22, 1979. [ Алгоритмы в современной математикеи её приложениях. Материалы международного симпозиума, Ургенч, УзССР,16–22 сентября 1979 г. ] Edited by A.P. Ershov, D. Knuth, Part I, p. 99–342, https://archive.org/details/uspensky-semenov-urgench-rus [27]
Nonstandard, or non-Archimedian, analysis . [
Нестандартный, или неархимедов,анализ , in Russian]. Moscow,
Знание publishers, 1983. 61 pp. https://archive.org/details/uspensky-nonstandard-znanie [28] Uspensky V.A., Kanovei V.G., Luzin’s problems on constituents and their fate[
Проблемы Лузина о конституантах и их судьба , in Russian].
Вестник Московскогоуниверситета. Серия 1: Математика, механика , 1983, issue 6, p. 73–87, https://istina.msu.ru/publications/article/93856782/ . English translation:Uspenskii V.A., Kanovei V.G., Luzin’s problems on constituents and their fate,
MoscowUniversity Mathematics Bulletin , vol. 38, no. 6 (1983), 86–102. (Allerton Press, inc.)[29] Luzin’s contribution to the descriptive theory of sets and functions: concepts, problems,predictions [
Вклад Н. Н. Лузина в дескриптивную теорию множеств и функций:понятия, проблемы, предсказания , in Russian].
Успехи математических наук ,vol. 40, issue 3(243), p. 85–116, http://mi.mathnet.ru/umn2648 . English translation:Uspenskii V.A., Luzin’s contribution to the descriptive theory of sets and functions:concepts, problems, predictions,
Russian Mathematical Surveys , vol. 40, no. 3 (1985),97–134, available at https://istina.msu.ru/publications/article/92645592/ [30] A.L. Semenov, V.A. Uspensky, Mathematical logic in computer science and computerapplications [
Математическая логика в вычислительных науках и вычислительнойпрактике , in Russian]
Вестник Академии наук СССР , (7), 93–103 (1986)[31] What is nonstandard analysis? [ Что такое нестандартный анализ? , in Russian].Moscow, Nauka Publishers, Physics and Mathematics Division [
Наука, главнаяредакция физико- математической литературы ], 1987. 128 pp., https://archive.org/details/chto-takoe-nestandartny-analiz-djvu [32] V.A. Uspensky, A.L. Semenov, Theory of algorithms: main discoveries and applications.[
В. А. Успенский, А. Л. Семёнов,
Теория алгоритмов: основные открытия иприложения , in Russian], Moscow, Nauka Publishers, Physics and MathematicsDivision [
Наука, главная редакция физико- математической литературы ], 1987.(Programmer’s library, vol. 49 [
Серия «Библиотечка программиста», выпуск 49 ].)288 pp., https://archive.org/details/uspensky-semenov-1987-algoritmy . Englishtranslation: Vladimir Uspensky, Alexei Semenov,
Algorithms: Main Ideas andApplications , Kluwer Academic Publishers, 1993, https://doi.org/10.1007/978-94-015-8232-2
Алгоритмы ислучайность , in Russian],
Теория вероятностей и её применения , vol. XXXII, issue 3,1987 (July, August, September), p. 425–455, http://mi.mathnet.ru/tvp1437 . Englishtranslation: Kolmogorov A.N., Uspenskii V.A., Algorithms and randomness,
Theory ofProbability and Its Applications , SIAM Publishers, vol. 32, no. 3, 389–412. http://dx.doi.org/10.1137/1132060 , https://istina.msu.ru/publications/article/92647240/ [34] V.A. Uspensky, V.G. Kanovei, M.Ya. Suslin’s contribution to set- theoretic mathematics[ Вклад М. Я. Суслина в теоретико- множественную математику , in Russian].
ВестникМосковского университета. Серия 1: Математика, механика , 1988, issue 5, p. 8–12, https://istina.msu.ru/publications/article/93856823/ . English translation:Uspenskii V.A., Kanovei V.G., M.Ya. Suslin’s contribution to set- theoretic mathematics,
Moscow University Mathematics Bulletin , vol. 43, no. 5 (1988), 29–40. (Allerton Press,inc.)[35] V.A. Uspensky, A.L. Semenov, A. Shen, Can an individual sequence of zeros and ones berandom? [
Может ли (индивидуальная) последовательность нулей и единиц бытьслучайной? , in Russian].
Успехи математических наук , vol. 45, issue 1(271),p. 105–162 (1990, January–February), http://mi.mathnet.ru/umn4692 . Englishtranslation: V.A. Uspensky, A.L. Semenov, A. Shen, Can an individual sequence of zerosand ones be random?
Russian mathematical surveys , vol. 45, issue 1, 121–189 (1990), https://archive.org/details/uspensky-semenov-shen-1990 [36] V.A. Uspensky, V.E. Plisko, Diagnostic propositional formulas [
Диагностическиепропозициональные формулы , in Russian].
Вестник Московского университета.Серия 1: Математика, механика , 1991, issue 3, p. 7–12, available at https://istina.msu.ru/publications/article/92717778/ [37] V.A. Uspensky, N.K. Vereshchagin, V.E. Plisko,
An introductory course of mathematicallogic [ Вводный курс математической логики ], Moscow State Lomonosov UniversityPublishers, 1991. 2nd edition: Moscow,
Физматлит , 2004, https://archive.org/details/uspensky-vereshchagin-plisko [38]
Complexity and Entropy: An Introduction to the Theory of Kolmogorov Complexity . In:
Kolmogorov Complexity and Computational Complexity , Osamu Watanabe, editor.Springer, 1992, ISBN 3-540-55840-3, p. 85–102, https://archive.org/details/uspensky-1992-watanabe-book [39] Kolmogorov and mathematical logic,
The Journal of Symbolic Logic , volume 57, number2, June 1992, 385–412, https://doi.org/10.2307/2275276 , https://archive.org/details/uspensky-1992-jsl-kolmogorov-mathematical-logic [40] Vladimir A. Uspensky and Valery Ye. Plisko, Review: Raymond M. Smullyan, G¨odel’sincompleteness theorems, The Journal of Symbolic Logic , volume 60, issue 4 (1995),1320–1324, https://projecteuclid.org/euclid.jsl/1183744885 [41] V.A. Uspensky, A. Shen, Relations Between Varieties of Kolmogorov Complexities,
Mathematical Systems Theory , , 271–292 (1996), https://link.springer.com/article/10.1007/BF01201280 , lpcs.math.msu.su/~uspensky/bib/Uspensky_1996_MST_Shen_Relations_between_varieties_of_Kolmogorov_complexities.pdf Proceedings of the 21st International Symposium on Mathematical Foundations ofComputer Science 1996 (MFCS96), Cracow, Poland, September 2–6, 1996 (LectureNotes in Computer Science, v. 1113), 1996, p. 156–166, https://link.springer.com/chapter/10.1007/3-540-61550-4_145 , http://lpcs.math.msu.su/~uspensky/bib/Uspensky_1996_LNCS_Kolmogorov_Complexity_Recent_trents_Moscow.pdf [43] Mathematical logic in the former Soviet Union: brief history and current trends, Logicand Scientific Methods , M.L. Dalla Chiare et al., editors, Kluver Academic Publishers, , 457–483.[44] Andrei A. Muchnik, Alexei L. Semenov, Vladimir A. Uspensky, Mathematicalmetaphysics of randomness,
Theoretical Computer Science , , 263–317 (1998), (fulltext available), http://lpcs.math.msu.su/~uspensky/bib/Uspensky_1998_TCS_Muchnik_Semenov_Math_metaphysics_randomness.pdf [45] What is an axiomatic method? [ Что такое аксиоматический метод?
Izhevsk,Udmurt university [
Ижевск: издательский дом «Удмуртский университет» ], 2000.100 pp. ISBN 5-7029-0337-4.[46] Why Kolmogorov complexity? In: E.Goles and C.Martinez (eds.),
Complex systems (Series: Nonlinear Phenomena and Complex Systems, Vol. 6), Kluwer AcademicPublishers, 2001, p. 201–260. ISBN 0-7923-6830-4, https://link.springer.com/chapter/10.1007/978-94-010-0920-1_5 [47]
Non-Mathematical Works, with an Appendix: Semiotic Letters of A.N. Kolmogorov to theauthor and his friends , [
Труды по нематематике с приложением семиотическихпосланий А. Н. Колмогорова к автору и его друзьям . , in Russian] Moscow, ОГИ , 2002.1409 pp., ISBN 5-94282-086-4, [48] B. Durand, V. Kanovei, V.A. Uspensky, N.K. Vereshchagin. Do stronger definitions ofrandomness exist?
Theoretical Computer Science , v. 290, No. 3, p. 1987–1996 (2001),available as [49] V.G. Kanovei, V.A. Uspenksy, On the equivalence of the two versions of the ContinuumHypothesis, [
Об эквивалентности двух форм континуум- гипотезы , in Russian].
Вестник Московского университета. Серия 1: Математика, механика , 2005,issue 3, p. 62–64, https://istina.msu.ru/publications/article/100287822/ [50] Four algorithmic faces of randomness, [
Четыре алгоритмических лица случайности , inRussian].
Математическое просвещение , ser. 3, issue 10, Moscow, MCCME, 2006,p. 71–108, http://mi.mathnet.ru/mp188 . Included as an appendix in the book [59].[51] Kolmogorov as I remember him [
Колмогоров, каким я его помню , in Russian]. In:
Kolmogorov in the memories of his students [ Колмогоров в воспоминаниях учеников , inRussian], edited by A.N. Shiryaev and N.G. Khimchenko, Moscow, MCCME, 2006,272–371.[52] V.G. Kanovei, V.A. Uspensky, On the uniqueness of nonstandard extensions [
Оединственности нестандартных расширений , in Russian].
Вестник Московского ниверситета. Серия 1: Математика, механика , 2006, issue 5, p. 3–10, available at https://istina.msu.ru/publications/article/100285758/ [53] V.G. Kanovei, T. Linton, V.A. Uspensky, A game approach to the Lebesgue measure,[ Игровой подход к мере Лебега , in Russian]
Математический сборник , vol. 199,issue 11 (2008), 21–44, http://mi.mathnet.ru/msb3948
English translation:V.G.Kanovei, Tom Linton and Vladimir A. Uspensky, Lebesgue measure and gambling,
Sbornik: Mathematics , volume 199, no. 11, p. 1597, http://dx.doi.org/10.1070/SM2008v199n11ABEH003974 [54]
Mathematics’ apology [ Апология математики , in Russian]. Saint-Petersbourg,
Амфора ,2009. 554 pp.[55] On the history of Goldbach’s problem [
К истории проблемы Гольдбаха , in Russian]. In:
Историко- математические исследования. Вторая серия . РАН, Институтестествознания и техники им. С. И. Вавилова . Vol. 13(48). Moscow,
Янус-К , 2009,ISBN 978-5-8037-0449-2, p. 273–283.[56]
Basic examples of mathematical proofs [ Простейшие примеры математическихдоказательств , in Russian]. (¡¡
Математическое просвещение ¿¿ series, issue 34).Moscow, MCCME, 2009. 56 pp. ISBN 978-5-94057-492-7. [57] V.A. Uspensky, V.V. Vyugin, The emergence of algorithmic information theory in Russia[
Становление алгоритмической теории информации в России , in Russian].
Информационные процессы , vol. 10, issue 2, p. 145–158.[58] G¨odel’s theorem and four ways to it, [
Теорема Гёделя и четыре дороги, ведущие к ней ,in Russian].
Математическое просвещение , ser. 3, issue 15, Moscow, MCCME, 2011,p. 35–75, http://mi.mathnet.ru/mp309 [59] N.K. Vereshchagin, V.A. Uspensky, A. Shen,
Kolmogorov complexity and algorithmicrandomness [ Колмогоровская сложность и алгоритмическая случайность , inRussian]. Moscow, MCCME, 2013. 575 pp. (English version: A. Shen, V. Uspensky,N. Vereshchagin,
Kolmogorov complexity and algorithmic randomness , AmericanMathematical Society, 2017.)[60] [V.A. Uspensky, M.S. Gelfand], Mathematics is a part of humanities [
Математика— этогуманитарная наука (интервью с В. А. Успенским ведёт М. С. Гельфанд) ]. In the book:
Математические прогулки. Сборник интервью , Moscow,
Паулсен , 2017. ISBN978-5-98797-057-7, p. 198–207. [The English translation of the book:
MathematicalWalks. A Collection of Interviews , ISBN: 978-5-98797-167-3, Moscow, Paulsen, 2017.][61] Vladimir Uspenskiy and Alexander Shen, Algorithms and Geometric Constructions,
Computability in Europe, 2018 , Lecture Notes in Computer Science, v. 10936, Springer,p. 410–420 (2018), see also arXiv:1805:12579 [62]
Non-Mathematical Works. Second extended and corrected edition. In five parts. Part 5.Memories and observations . [
Труды по не математике. Второе издание, исправленноеи дополненное. В пяти книгах. Книга пятая. Воспоминания и наблюдения. , inRussian]. Moscow, Объединённое гуманитарное издательство. Фонд«Математические этюды» . 2018. 1118 pp.[63] Third Mathematical Congress. In: [62, p. 897–905]. Commentaries. Ibidem, p. 905–912.28 ublications translated or edited by Uspensky [64] R. Peter,
Recursive functions [ Р. Петер,
Рекурсивные функции , in Russian], translatedfrom German by V.A. Uspensky, edited by A.N. Kolmogorov, with a preface written byA.N. Kolmogorov. Moscow,
Издательство иностранной литературы , 1954. (Originalbook:
Rekursive Funktionen , von R´osza P´eter, Budapest, 1951.)[65] S. Kleene, Introduction to metamathematics [
Стефен К. Клини,
Введение вметаматематику , in Russian], translated from English by A.S. Esenin-Volpin, editedby V.A. Uspensky. Moscow,
Издательство иностранной литературы , 1957. (Originalbook: Stephen Cole Kleene,
Introduction to metamathematics , D. van Nostrand company,New York, Toronto, 1952.)[66] W. Ross Ahsby,
An introduction to cybernetics [ У. Росс Эшби,
Введение вкибернетику , in Russian]. Translated from English by D.G. Lakhuti. Edited byV.A. Uspensky. With a preface written by A.N. Kolmogorov. Moscow,
Издательствоиностранной литературы , 1959. 428 pp. (Original book:
An introduction to cybernetics ,by W. Ross Ashby, London, Chapman&Hall Ltd., 1956.)[67] A. Church, Introduction to mathematical logic, volume I. [
А. Чёрч,
Введение вматематическую логику, I , in Russian]. Translated from English by V.S. Chernyavsky.Edited by V.A. Uspensky. Moscow,
Издательство иностранной литературы , 1960.(Original book:
Introduction to mathematical logic by Alonzo Church. Volume I.Princeton University Press, 1956.)[68] N. Bourbaki,
Elements of Mathematics. Part 1. Fundamental structures of analysis. Book1. Set theory [ Н. Бурбаки,
Начала математики. Первая часть. Основныеструктуры анализа. Книга первая. Теория множеств , in Russian]. Translated fromFrench by G.N. Povarov and Yu.A. Shikhanovich. Edited by V.A. Uspensky. With apreface written by V.A. Uspensky. Moscow,
Мир , 1965. (Original book: ´El´ements demath´ematique par N. Bourbaki, XVII, XX, XXII, I. Premiere partie. Les structuresfondamentales de l’analyse. Livre I. Th´eorie des ensembles. Hermann, 1956–1960.)[69]
Mathematics in the modern world [ Математика в современном мире , in Russian], acollection of translations of papers from Scientific American special issue
Mathematics inthe modern world , Scientific American, 1964. Translated from English by N.G. Rychkova.Edited by V.A. Uspensky. With a preface written by V.A. Uspensky. Moscow,
Мир , 1967.[70] H. Rogers, Theory of recursive functions and effective computability [
Х. Роджерс,
Теория рекурсивных функций и эффективная вычислимость , in Russian], translatedfrom English by V.A. Dushsky, M.I. Kanovich, E.Yu. Nogina. Edited by V.A. Uspensky.Moscow,
Мир , 1972. (Original book: Hartley Rogers, Jr.,
Theory of recursive functionsand effective computability , McGraw-Hill Book Company, 1967.A preliminary version withthe same name. Volume I. Mimeographed. Technology Store, Cambridge, Mass., 1957.)[71] M. Davis,
Applied nonstandard analysis [ М. Дэвис,
Прикладной нестандартныйанализ , in Russian]. Translated from English by S.F. Soprunov. Edited byV.A. Uspensky. With a preface written by V.A. Uspensky. Moscow,
Мир , 1980. (Originalbook: Martin Davis,
Applied nonstandard analysis , Wiley&Sons, 1977.)29 ther references [72] Th. Skolem,
Begr¨undung der elementaren Arithmetik durch dir rekurrierende Denkweiseohne Anwendung scheinbarer Ver¨anderlichen mit unendlichem Ausdehnungsbereich (Videnskapsselskapets Scrifter, I. Mat.-naturv. Klasse, 1923, No. 6), Kristiania, 1923.(English translation: The foundations of elementary arithmetic established by means ofthe recursive mode of thought without the use of apparent variables ranging over infinitedomains. In [102, p. 302–333].)[73] David Hilbert, ¨Uber das Unendliche,
Mathematische Annalen , Bd. 95, S. 161–190 (1926).(English translation: On the Infinite, [102, p. 367–392].)[74] Wilhelm Ackermann in G¨ottingen, Zum Hilbertschen Aufbau der reellen Zahlen,
Mathematische Annalen , Bd. 99, 118–133 (1928). (English translation: On Hilbert’sconstruction of the real numbers, [102, p. 493–507].)[75] Kurt G¨odel in Wien, ¨Uber formal unentscheidbare S¨atze der Principia Mathematica undverwandter Systeme I,
Monatshefte f¨ur Mathematik und Physik , , 173–198 (1931).(English translation: On formally undecidable propositions of Principia Mathematica andrelated systems I, [102, p. 596–616] or [99, p. 4–38].)[76] J. Herbrand `a Paris, Sur la non-contradiction de l’Arithm´etique,
Journal f¨ur die reineund angewandte Mathematik , Bd. 166, S. 1–8 (1932), (English translation: On the consistency of arithmetic [102, p. 618–628].)[77] Kurt G¨odel,
On undecidable propositions of formal mathematical systems , Lecture notes,Institute for Advanced Study (Princeton), Spring 1934. Reprinted in [99, p. 39-74][78] R´osza P´eter, ¨Uber den Zusammenhang der verschiedenen Begriffe der rekursivenFunktionen,
Mathematische Annalen , Bd. 110, n. 1, S. 612–632 (1935), https://doi.org/10.1007/BF01448046 , https://link.springer.com/article/10.1007%2FBF01448046 [79] Alonzo Church, An unsolvable problem of elementary number theory, American Journalof Mathematics , vol. 58, no. 2 (April 1936), 345–363. Reprinted in [99, p. 88–107][80] S.C. Kleene, General recursive functions of natural numbers,
Mathematische Annalen ,Bd. 112, S. 727–742 (1936), https://eudml.org/doc/159849 . Reprinted in [99,p. 236–253].[81] Emil L. Post. Finite combinatory processes. Formulation I.
The Journal of SymbolicLogic , vol. 1 (1936), p. 103–105.[82] A.M. Turing, On computable numbers, with an application to the Entscheidungsproblem,
Proceedings of the London Mathematical Society , ser. 2, vol. 42 (1936–7), p. 230–265;Correction in the next volume of the same journal: vol. 43 (1937), p. 544–546. Reprintedin [99, p. 116–154].[83] S.C. Kleene, On notations for ordinal numbers,
Journal for Symbolic Logic , , 150–155(1938), [84] A.M. Turing, Systems of logic based on ordinals, Proc. London Math. Soc. (2), vol. 45(1939), pp. 161–228, https://doi.org/10.1112/plms/s2-45.1.161 . (Reprinted [99,p. 154–222]. Turing’s Princeton Ph.D. thesis (with the same name) availablesas )3085] S.C. Kleene, Recursive predicates and quantifiers,
Transactions of the AmericanMathematical Society , , number 1, 41–73 (1943), https://doi.org/10.1090/S0002-9947-1943-0007371-8 . Reprinted in [99, p.254–287].[86] Emil L. Post, Recursively enumerable sets of positive integers and their decisionproblems, Bulletin of the American Mathematical Society , , 284–316 (1944). https://projecteuclid.org/download/pdf_1/euclid.bams/1183505800 . Reprintedin [99, p.304–337].[87] S.C. Kleene, A symmetric form of G¨odel’s theorem (Presented to the AmericanMathematical Society, October 29, 1949. Communicated by Prof. L.E.J. Brouwer at themeeting of April 29, 1950). Koninklijke Nederlandse Akademie van Wetenschappen,Volume 53, deel 6 (1950), 800–802, [88] H.G. Rice, Classes of recursively enumerable sets and their decision problems, Transactions of the American Mathematical Society , vol. 74 (1953), p. 358–366.[89] L´aszlo Kalm´ar in Szeged, Ungarn, ¨Uber ein Problem, betreffend die definition desBegriffes der allgemein-rekursiven Funktion,
Zeitschrift f¨ur mathematische Logik undGrundlagen der Mathematik , Bd. 1, S. 93–95 (1955), https://onlinelibrary.wiley.com/doi/abs/10.1002/malq.19550010204 [90] John Myhill, A fixed point theorem in recursion theory, abstract, Eighteenth Meeting ofthe Association of Symbolic Logic,
The Journal of Symbolic Logic , volume 20, no.2 (June1955), p. 205.[91] J. Myhill in Berkeley, California (USA), J.C. Sheperdson in Bristol, England, Effectiveoperations in partial recursive functions,
Zeitschrift f¨ur mathematische Logik undGrundlagen der Mathematik , Bd. 1, S. 310–317 (1955), https://onlinelibrary.wiley.com/doi/abs/10.1002/malq.19550010407 [92] H.G. Rice, On completely recursive enumerable classes and their key arrays,
The Journalof Symbolic Logic , volume 21, number 3, Sept. 1956, p. 304–308. (Received September 14,1955) [93] Martin Davis,
Computability and Unsolvability , McGraw-Hill Book Company, 1958.[94] Hartley Rogers, Jr. G¨odel numberings of partial recursive functions,
Journal of SymbolicLogic , Volume 23, Number 3, Sept. 1958, p. 331–341 (Received July 7, 1958), [95] Richard M. Friedberg, Hartley Rogers jr., Reducibility and Completeness for Sets ofIntegers,
Zeitschrift f¨ur mathematische Logik und Grundlagen der Mathematik , Bd. 5,S. 117–125 (1959), https://doi.org/10.1002/malq.19590050703 [96] Kreisel, G., Lacombe, D., Shoenfield, J.R.,
Partial recursive functionals and effectiveoperations , in
Constructivity in Mathematics , A. Heyting, editior, North Holland, 1959,p. 195–207.[97] John Myhill, Note on degrees of partial functions,
Proceedings of the AmericanMathematical Society , (1961), p. 519–521, https://doi.org/10.1090/S0002-9939-1961-0125794-X Алгорифмическиеоператоры в конструктивных метрических пространствах , in Russian]. Proceedings ofMoscow Steklov Institute, LVII, Problems in constructive mathematics, 2. (Constructiveanalysis) [
Труды математического института имени В. А. Стеклова, LVII,Проблемы конструктивного направления в математике, 2 (Конструктивныйматематический анализ) ], a collection of papers. Edited by N.A. Shanin. Moscow,Leningrad, Academy of Science Publishers,
Издательство Академии наук СССР,Москва, Ленинград , 1962, p. 295–361.[99]
Basic Papers On Undecidable Propositions, Unsolvable Problems and ComputableFunctions , a collection compiled by Martin Davis, Raven Press, Hewlett, New York, 1965.[100] A.I. Maltsev,
Algorithms and recursive functions [ Алгоритмы и рекурсивныефункции , in Russian], Moscow,
Наука , 1965. (2nd edition, 1986)[101] Per Martin-L¨of, The definition of random sequences,
Information and Control , volume9, issue 6, December 1966, p. 602–619, https://doi.org/10.1016/S0019-9958(66)80018-9 [102]
From Frege to G¨odel. A Source Book in Mathematical Logic, 1879–1931 , a collection ofpapers compiled by Jean van Heijenoort, Harvard University Press, Cambridge,Massachusetts, 1967.[103] John Case, Enumeration reducibility and partial degrees,
Annals of mathematical logic ,vol. 2, no. 4 (1971), 419–439. (Received 9 September 1969), [104] A. Shen, Axiomatic approach to the theory of algorithms and relativized computability,[
Аксиоматический подход к теории алгоритмов и относительная вычислимость , inRussian],
Вестник Московского университета. Серия 1: Математика, механика. ,1980, issue 2, p. 27–29. (English translation made by the author is availableas https://hal-lirmm.ccsd.cnrs.fr/lirmm-01923123 .)[105] Stephen C. Kleene, The theory of recursive functions, approaching its centennial.(Elementarrekursiontheorie vom h¨oheren Standpunkte aus.)
Bulletin of the AmericanMathematical Society , volume 4, number 1, July 1981, 43–61.[106] Nigel Cutland,
An introduction to recursive function theory , Cambridge University Press,1980. Russian translation by Albert A. Muchnik, edited by S.Yu. Maslov:
Н. Катленд,
Вычислимость. Введение в теорию рекурсивных функций.
М.:Мир, 1983. [107] A. Shen, Algorithmic versions of the notion of entropy [
Алгоритмические вариантыпонятия энтропии , in Russian],
Доклады Академии наук , 1984, vol. 276, issue 3,p. 563–566. (English translation: Soviet Math. Doklady, (3), 1984, 569–573.)[108] Piergiorgio Odifreddi, Classical Recursion Theory. The Theory of Functions and Sets ofNatural Numbers (Studies in logic and the foundations of mathematics, volume 125),Elsevier, 1989. xix+668 pages.[109] Robert I. Soare, Computability and Recursion,
Bulletin of Symbolic Logic , (3), 284–321(1996). See also [110] V.V. Vyugin, Ergodic theorems for individual random sequences, Theoretical ComputerScience , volume 207, issue 2, November 6, 1998, p. 343–361, https://doi.org/10.1016/S0304-3975(98)00072-3 .32111] Mariya I. Soskova, The Turing Universe in the Context of Enumeration Reducibility. In:Bonizzoni P., Brattka V., L¨owe B. (eds),
The Nature of Computation. Logic, Algorithms,Applications. CiE 2013 . Lecture Notes in Computer Science, vol 7921. Springer, Berlin,Heidelberg, p. 371–382, https://doi.org/10.1007/978-3-642-39053-1_44 [112] Robert I. Soare,
Turing Computability. Theory and Applications . Springer, 2016, ISBN978-3-642-31932-7, https://doi.org/10.1007/978-3-642-31933-4 [113] Wikipedia pages:
Primitive recursive function , https://en.wikipedia.org/wiki/Primitive_recursive_function and µ -recursivefunction , https://en.wikipedia.org/wiki/%CE%9C-recursive_function . (Version ofNovember 5, 2018)[114] Szudzik, Matthew. Recursive Function . From MathWorld — A Wolfram Web Resource,created by Eric W. Weisstein. http://mathworld.wolfram.com/RecursiveFunction.html . (Accessed November 5,2018) 33 r X i v : . [ m a t h . HO ] A ug Математические работы Владимира АндреевичаУспенского: комментарии ∗ Александр Шень † Аннотация
Мой учитель Владимир Андреевич Успенский (1930–2018) был одним из пионе-ров теории вычислений и в целом математической логики в СССР. В этом обзорепредпринимается попытка описать его математические работы и их роль в разви-тии теории алгоритмов и математической логики в СССР. (Его организационнаядеятельность и достижения в лингвистике выходят за рамки этого обзора.)
Гармонические функции
Первая (студенческая) работа Успенского [3] предлагает элементарное изложение ос-новных свойств гармонических функций (с минимальным использованием сведений изанализа— определением считается утверждение теоремы о среднем). Изложение осно-вано на таком наблюдении: угол, под которым виден фиксированный отрезок из пере-менной точки, является гармонической функцией этой точки в том смысле, что длянего верна теорема о среднем, и функция эта является ступенчатой на окружности,проходящей через концы отрезка. Дипломная работа
В дипломной работе Успенского [4] излагается модель вычислений, предложеннаяА. Н. Колмогоровым и известная теперь как «машины Колмогорова–Успенского». До-казывается, что эта модель эквивалентна частично рекурсивным функциям, опреде-лённым с помощью подстановки, рекурсии и минимизации (то есть даёт тот же классвычислимых функций). Кроме того, в рамках этой модели вводится понятие вычисли-мости относительно некоторой функции f : график функции f представляется в видебесконечного графа (комплекса), доступного алгоритму вместе со входом [определение(A) на с. 64]. Это определение относительной вычислимости сравнивается с други-ми. Для этого Успенский переформулирует определение Тьюринга– Поста (машиныс оракулом [84, 86]) в терминах вычислимого протокола взаимодействия с оракулом[определение (T) на с. 63], и доказывает, что полученное определение эквивалентно ∗ Владимир Андреевич Успенский был моим учителем (и научным руководителем на старших кур-сах и в аспирантуре, как это формально называется на мехмате МГУ). В этом обзоре говоритсяпрежде всего о его математических работах; надеюсь выразить свою благодарность за всё остальноев отдельном тексте. † LIRMM, University of Montpellier, CNRS, Montpellier, France. Грант RaCAF–ANR-15-CE40-0016 Это следует из того, что среднее направление из фиксированной точки на переменную точкуокружности совпадает с направлением на её центр.
Б. А. Трахтенброту: функция γ сводится к функции δ , если γ принадлежит рекурсивному замыканию δ . Автор показывает, что в действительности такая сводимость может бытьвсегда осуществлена очень простым каноническим образом при помощи разнавсегда заданных примитивно-рекурсивных функций τ ( u ) и ω ( u ) и зави-сящих от пары γ, δ примитивно рекурсивных функций h ( u, v, w ) и ϕ ( m ) .См. об этом теорему на стр. 28. Это основной новый с чисто математи-ческой точки зрения результат работы. Определение сводимости по Трахтенброту нуждается в известном «оправ-дании» его соответствия интуитивной идее сводимости в смысле существо-вания «механического» способа получения при любом x значения γ ( x ) впредположении , что получение значений δ ( x ) сделано каким-то способом«доступным» для любого x . Общие контуры возможной формализации этойидеи были намечены Поустом [так Колмогоров пишет фамилию Поста (EmilPost)]. В дипломной работе полностью воспроизведён перевод соответству-ющего места статьи Поуста [имеется в виду статья [86]]. Автор дипломнойработы, по-видимому впервые, даёт соответствующее этой идее опреде-ление сводимости с полной отчётливостью и показывает его эквивалент-ность определению Трахтенброта. Это тоже весьма существенное достиже-ние автора дипломной работы.(2) Кроме того в работе содержится хороший обзор различных предлагав-шихся ранее определений алгоритмической вычислимости числовой функ-ции y = γ ( x ) . В центре изложения помещено определение, предложенноемною, интерес которого на мой взгляд убедительно аргументирован авто-ром дипломной работы. Доказана равносильность этого определения преж-де предлагавшимся. В известном смысле слова этот результат можнорассматривать как «обоснование» прежних определений, так как в моёмопределении становится особенно ясной основная идея алгоритмическойвычислимости, которая отличается от вычислимости обыкновенным ре-альным счётным механизмом только неограниченным объёмом «запоми-нающего устройства» механизма.
Чтобы оценить содержание работы, важно представлять себе исторический кон-текст. Сейчас этот контекст почти забыт, и о нём надо сказать несколько слов.
Частично рекурсивные функции
Если спросить, что такое частично рекурсивная функция (partial recursive function),большинство современных специалистов ответят, что это функция, которая может быть2олучена из базовых функций (проекция, нулевая функция и функция прибавленияединицы) с помощью операций подстановки, рекурсии и минимизации ( µ -оператора).Это определение можно найти в классической книге Одифредди [108, с. 127], в другихучебниках [100, 106] и в википедии [113].Но раньше определение было другим, и следы этого старого определения сохрани-лись в другом классическом учебнике [70, раздел 1.5] и в cправочном ресурсе WolframMathWorld [114]. Хотя это определение и эквивалентно приведённому выше (задаёт тотже класс функций), но разницу между ними важно иметь в виду при чтении старыхработ.История вопроса здесь такова. Рекуррентные определения были известны давно(достаточно вспомнить о Фибоначчи), но их систематическое использование для по-строения арифметики появилось в работе Сколема 1923 года [72]. Он понял, что такимобразом можно определить не только простые функции, скажем, сложение и умно-жение (чтобы прибавить следующее за y число к x , надо прибавить y к x и взятьследующее число; чтобы умножить x на следующее за y число, надо умножить x на y и прибавить x ), но и много других функций, встречающихся в элементарной теории чи-сел. После этого базовые результаты этой теории можно доказать по индукции, исходяиз рекурсивных определений. Сколем не рассматривал явно класса всех функций, которые можно получить спомощью такого рода рекурсивных определений. Но уже в докладе 1925 года Гиль-берт [73] говорит об определениях функций «с помощью подстановок и рекурсий» иразличает «обычные, пошаговые рекурсии», где значение функции на каком-то числеопределяется через её значение на предыдущем числе, и более сложные схемы. В каче-стве примера такой более сложной схемы он приводит последовательность функций ϕ ( a, b ) = a + b, ϕ ( a, b ) = a · b, ϕ ( a, b ) = a b ,ϕ ( a, b ) = [ b -й член в последовательности a, a a , a ( a a ) , a a ( aa ) . . . ]и так далее, которую можно задать рекурсивно равенствами ϕ ( a, b ) = a + b, ϕ n +1 ( a,
1) = a, ϕ n +1 ( a, b + 1) = ϕ n ( a, ϕ n +1 ( a, b )) , и упоминает результат Аккермана о том, что функцию ϕ n ( a, b ) как функцию от трёхпеременных n, a, b нельзя задать «обычными» рекурсиями (этот результат опубликованпозже в [74]). Упоминание этого отрицательного результата Гильбертом означает, что уГильберта было уже понятие о классе функций, которые можно получить «обычными»рекурсиями (и подстановками), хотя не было специального названия для функций этогокласса и не было явно дано его определение.Такое название и такое определение появились в знаменитой работе Гёделя [75,с. 179]: функция называется рекурсивной (rekursiv по-немецки), если она получаетсяпоследовательным применением нескольких операций подстановки и операции рекур-сии такого вида (построение функции ϕ , если ψ и µ уже построены ранее): ϕ (0 , x , . . . , x n ) = ψ ( x , . . . , x n ) ϕ ( k + 1 , x , . . . , x n ) = µ ( k, ϕ ( k, x , . . . , x n ) , x , . . . , x n ) Мотивацией этой работы было —показать, что многие математические результаты можно обосно-вать простым и надёжным способом, индуктивно доказывая равенства между рекурсивно определён-ными функциями. Теперь соответствующую теорию называют примитивно рекурсивной арифмети-кой . f i ( x , . . . , x n i ) вместе с утверждениями о них, если:(a) эти утверждения не содержат связанных переменных;(б) рассматриваемые с интуиционистской точки зрения, то есть как утвер-ждения о натуральных числах, а не просто как символы, эти утвержде-ния позволяют вычислить значение f i ( x , . . . , x n i ) для любого наборачисловых аргументов, и можно интуиционистски доказать, что резуль-тат однозначно определён.Смысл этой оговорки про «интуиционистскую точку зрения», видимо, в том, что насне устроит само по себе функциональное уравнение, про которое из каких-то общихсоображений можно доказать, что его решение существует и единственно; нужно, что-бы это доказательство было в каком-то смысле конструктивно и давало способ вы-числения значений интересующих нас функций, исходя из задающих их равенств. (Идействительно, впоследствии Кальмар [89] привёл пример системы функциональныхуравнений, однозначно задающей невычислимую функцию.)Гёдель возвращается к предложению Эрбрана (погибшего в горах сразу после от-правки в редакцию статьи [76]) в своих лекциях в Принстоне (записки которых былиразмножены ещё тогда, а позднее перепечатаны, см. [77]). Он по-прежнему называет«рекурсивными» функции, которые получаются из базовых с помощью подстановок и«обычных» рекурсий, но в разделе 9 говорит о рекурсивных определениях более обще-го вида и задаваемых ими функциях, называя их “general recursive functions”. Гёдельвоспроизводит предложение Эрбрана так:One may attempt to define this notion [general recursive function] as follows:if φ denotes an unknown function, and ψ , . . . , ψ k are known functions, and ifthe ψ ’s and the φ are substituted in one another in the most general fashionsand certain parts of the resulting expressions are equated, then if the resultingset of functional equations has one and only one solution for φ , φ is a recursivefunction.”(В примечании к этому определению Гёдель ссылается на письмо Эрбрана.) Далее ондобавляет ограничения, уточняющие замысел Эрбрана:We shall make two restrictions on Herbrand’s definition. The first is that theleft-hand side of each of the given functional equations defining φ shall be of theform φ ( ψ i ( x , . . . , x n ) , ψ i ( x , . . . , x n ) , . . . , ψ il ( x , . . . , x n )) . The second (as stated below) is equivalent to the condition that all possible setsof arguments ( n , . . . , n l ) of φ can be so arranged that the computation of thevalue of φ for any given set of arguments ( n , . . . , n l ) by means of the given4quations requires a knowledge of the values of φ only for sets of argumentswhich precede ( n , . . . , n l ) .Гёдель не уточняет порядок на наборах аргументов, так что точный смысл этого неочень ясен. Но дальше описываются конкретные правила вывода одних равенств издругих и говорится:Now our second restriction on Herbrand’s definition of recursive function is thatfor each set of natural numbers k , . . . , k l there should be one and only one m such that φ ( k , . . . , k l ) = m is a derived equation.Тем самым даётся вполне точное определение некоторого класса функций, названных“general recursive functions”; по-русски обычно переводят этот термин (немного зага-дочно) как «общерекурсивные функции». Однако, как пишет Клини в [105], в моментчтения лекций (1934) сам Гёдель ещё не был уверен в том, что этот класс функцийдостаточно широк: «However, G¨odel, according to a letter he wrote to Martin Davis on15 February 1965, “was, at the time of [his 1934] lectures, not at all convinced that [this]concept of recursion comprises all possible recursions”» [105, p. 48]. В [99, p. 40] про этоговорится так:In the present article [речь идёт о [77]] G¨odel shows how an idea of Herbrand’scan be modified so as to give a general notion of recursive function h . . . i G¨odelindicates (cf. footnote 3) that he believed that the class of functions obtainableby recursion of the most general kind were the same as those computable bya finite procedure. However, Dr. G¨odel has stated in a letter that he was, atthe time of these lectures, not at all convinced that his concept of recursioncomprised all possible recursions; and that in fact the equivalence between hisdefinition and Kleene’s in Math. Ann. 112 [речь идёт о [80]] is not quite trivial.So despite appearances to the contrary, footnote 3 of these lectures is not astatement of Church’s thesis.Footnote 3 [99, p. 44] относится к утверждению о том, что всякая примитивно рекурсив-ная функция может быть вычислена с помощью конечной процедуры, и говорит “Theconverse seems to be true, if, besides recursions according to the scheme (2) [примитивнаярекурсия], recursions of other forms (e.g., with respect to two variables simultaneously) areadmitted. This cannot be proved, since the notion of finite computation is not defined, butit serves as a heuristic principle”.Роза Петер в [78] изучает возможности «обычных рекурсий» (например, доказывает,что разрешение использовать несколько значений функции в меньших точках сводитсяк схеме с одним предшественником) и вводит термин “primitive Rekursion” для этихсамых «обычных рекурсий».Следуя ей, Клини в 1936 году [80] вводит термин “primitive recursive functions” (при-митивно рекурсивные функции) для тех функций, которые получаются с помощьюподстановок и примитивных рекурсий и которые Гёдель в [75] называл просто «рекур-сивными». Одновременно Клини предлагает рассмотреть более общий класс функций,элементы которого он называет “general recursive functions” (его статья так и называ-ется,
General recursive functions of natural numbers ). Этот класс определяется в духеЭрбрана и Гёделя, при этом рассматриваются разные правила вывода одних равенствиз других, которые, однако (как доказывает Клини), задают один и тот же класс функ-ций. 5лини также вводит ε -оператор εx [ A ( x )] как наименьшее число, удовлетворяющееусловию A ( x ) , если таковое существует; в противном случае берётся нуль. Этот опе-ратор фигурирует в теореме IV, которая утверждает, что всякая общерекурсивнаяфункция может быть представлена в виде ψ ( εy [ R ( x, y )]) , где ψ —некоторая примитивнорекурсивная функция, а R — примитивно рекурсивный предикат (это означает, что пре-дикат R записывается как равенство нулю некоторой примитивно рекурсивной функ-ции), причём для всякого x существует y , при котором R ( x, y ) . Теорема V утверждает,что верно и обратное: всякая функция, представимая в указанном виде, рекурсивна всмысле определений в духе Эрбрана и Гёделя. Тем самым такое представление можетрассматривать как эквивалентное определение понятия рекурсивной функции. Крометого, из этого можно извлечь некоторый способ нумерации всех рекурсивных функ-ций, введя дополнительный параметр e в примитивно рекурсивный предикат R (хотяне при всех e получается всюду определённая функция; можно было бы сказать, чтонумеруются частичные функции, но пока Клини их не рассматривает).В том же (1936) году Чёрч публикует статью [79], в которой приводит другоеопределение некоторого класса числовых функций (в терминах так называемого λ -исчисления) как формализацию интуитивной идеи вычислимости:The purpose of the present paper is to propose a definition of effective calculability which is thought to correspond satisfactorily to the somewhat intuitive notion.В подстрочном примечании ( ) Чёрч пишет:As will appear, this definition of effective calculability can be stated in eitherof two equivalent forms, (1) that a function of positive integers shall be calledeffectively calculable if it is λ -definable in the sense of §2 below, (2) that afunction of positive integers shall be called effectively calculable if it is recursivein the sense of §4 below. The notion of λ -definability is due jointly to the presentauthor and S.C. Kleene h . . . i The notion of recursiveness in the sense of §4 isdue jointly to Jacques Herbrand and Kurt G¨odel h . . . i The proposal to identifythese notions with the intuitive notion of effective calculability is first made inthe present paper. . .и добавляет (примечание в §7):The question of the relationship between effective calculability and recursiveness(which it is here proposed to answer by identifying the two notions) was raisedby G¨odel in conversation with the author. The corresponding question of therelationship between effective calculability and λ -definability had previouslybeen proposed by the author independently.Видно, что Чёрч считает важным делом отождествление интуитивного понятия вы-числимости с принадлежностью к точно определённому классу функций (для которогоесть два эквивалентных определения). Это отождествление и назвали потом тезисомЧёрча .Почти в то же время Тьюринг публикует свою работу [82], в которой он опреде-ляет машины, называемые теперь машинами Тьюринга (сам Тьюринг употребляеттермин a -machine, от слова ‘automatic’), и строит универсальную машину (которая Та часть ß определения ε -оператора, где результат полагается равным нулю, когда искомого y несуществует, при этом роли не играет. Таким образом, здесь можно заменить ε -оператор на стандарт-ный µ -оператор, в котором значение считается неопределённым в случае отсутствия искомого y . λ -определимости. Описывая этот результат во введении к работе,он пишет:In a recent paper Alonzo Church has introduced an idea of “effective calculability”,which is equivalent to my “computability”, but is very differently defined. Churchalso reaches similar conclusions about the Entscheidungsproblem. The proof ofequivalence between “computability” and “effective calculability” [имеется ввиду λ -определимость] is outlined in an appendix to the present paper.Независимо от Тьюринга и почти одновременно с ним Пост публикует работу [81],где описывает своё определение «финитного комбинаторного процесса», отличающеесяот машин Тьюринга лишь техническими деталями, а также тем, что он не говорито машине, а описывает, как “problem solver or worker” следует указаниям (the set ofdirections) определённого вида. Далее Пост замечает:The writer expects the present formulation to turn out to be logically equivalentto recursiveness in the sensе of the G¨odel–Church development. Its purpose,however, is not only to present a system of a certain logical potency but also, inits restricted field, of psychological fidelity. In the latter sense wider and widerformulations are contemplated. On the other hand, our aim will be to showthat all such are logically equivalent to formulation 1 [предложенный Постомвариант определения]. We offer this conclusion at the present moment as a working hypothesis . And to our mind such is Church’s identification of effectivecalculability with recursiveness. h . . . i The success of the above program would,for us, change this hypothesis not so much to a definition or to an axiom but toa natural law .И добавляет в примечании:Actually the work already done by Church and others carries this identificationconsiderably beyond the working hypothesis stage. But to mask this identificationunder a definition hides the fact that a fundamental discovery in the limitationsof the mathematizing power of Homo Sapiens has been made and blinds us tothe need of its continual verification. Видно, что к 1936 году уже сложилась почти что современная картина: есть несколь-ко эквивалентных (задающих один и тот же класс функций) определений вычислимо-сти, есть понимание, что они отражают интуитивную идею алгоритма и вряд ли что-тоупущено (и даже есть некоторые интуитивные объяснения, почему это так). Сейчас, пожалуй, это фундаментальное открытие (fundamental discovery) уже практически утра-тило смысл: чтобы говорить о соответствии интуитивной идеи вычислимости (идеи алгоритма в нефор-мальном смысле этого слова) и формально определённого класса алгоритмов, нужно, чтобы эта инту-итивная идея была сформирована независимо от программистского опыта —а где теперь найти людей,которые бы познакомились с идеей алгоритма, не имея уже программистского опыта?
7о есть два отличия от современной картины: одно скорее терминологическое, адругое более принципиальное. Терминологическое состоит в том, что ни в одной изработ, говорящих о рекурсивных функциях, они не определяются как функции, полу-чаемых с помощью подстановки, рекурсии и µ -оператора, хотя все ингредиенты для до-казательства эквивалентности этого определения другим есть и сама эквивалентностьявно упоминается Клини в 1943 году [85, p. 53, Corollary].Во вторых, во всех этих работах говорится о всюду определённых функциях (опре-делённых на всех натуральных числах). Частичные функции появляются чуть позже,в другой работе Клини [83], где строятся вычислимые системы обозначений для орди-налов (тут без частичных функций уже не обойтись). Описав процесс вывода утвер-ждения о значении функции из равенств в духе Эрбрана и Гёделя и предположив, чтоон даёт не более одного ответа для искомого значения, он пишет:If we omit the requirement that the computation process always terminate,we obtain a more general class of functions, each function of which is definedover a subset (possibly null or total) of the n -tuples of natural numbers, andpossesses the property of effectiveness when defined. These functions we callpartial recursive.Таким образом впервые появляется понятие частично рекурсивной функции (partialrecursive function). Клини рассматривает (естественным образом обобщаемые на ча-стичные функции) операции подстановки и рекурсии, а также определяет действие µ -оператора для случая частичной функции: µy [ R ( m, y ) = 0] = n для частичной функции R , если R ( m, n ) определено и равно нулю, а все предыдущиезначения R ( m, , . . . , R ( m, n − определены и не равны нулю. (Очевидно, что такое n единственно, если существует; если же нет, то определяемая с помощью µ -операторафункция не определена на m .) Как отмечает Клини, все три операции (подстановка, ре-курсия и µ -оператор) не выводят из класса частично рекурсивных функций (определён-ных по Эрбрану и Гёделю). Он отмечает также, что для класса частично рекурсивныхфункций от любого числа ( n ) переменных существует универсальная функция Φ n ( z, x ) от n + 1 переменной. (Универсальность означает, что фиксацией первого аргумента z из Φ n можно получить любую частично рекурсивную функцию от n переменных.) Этауниверсальная функция может быть представлена в виде Φ n ( z, x ) = S ( z, µyT n ( z, x , y )) , где S — некоторая примитивно рекурсивная функция, а T n — некоторый примитивно ре-курсивный предикат (задаваемый условием обращения в нуль некоторой примитивнорекурсивной функции). В этом представлении, неформально говоря, z кодирует (в ви-де натурального числа) систему равенств, задающих частично рекурсивную (в смыслеЭрбрана и Гёделя) функцию от n переменных, а y является протоколом вывода изэтой системы равенств утверждения о значении функции на входе x . Предикат T n проверяет корректность этого вывода, а функция S извлекает из этого вывода установ-ленное значение функции. Из этого результата (который называют «теоремой Клини Русский термин тут, пожалуй, ещё более странный, чем слово «общерекурсивные» для всюдуопределённых вычислимых функций: создаётся впечатление, что функция лишь отчасти рекурсивна.Английский термин лучше, потому что в нём слово ‘partial’ относится к слову ‘function’. В формулировке Клини есть ещё первый аргумент z у функции S , но его можно было бы иопустить. µ -оператора (и даже допол-нительно потребовать, чтобы µ -оператор применялся только один раз к примитивнорекурсивной функции). Но такой вариант определения Клини по-прежнему не упоми-нает.Примерно эта же система понятий, терминология и способ изложения используютсяв более поздней статье Клини [85], посвящённой в основном арифметической иерархии,и в его же учебнике 1952 года [65], ставшем на многие десятилетия классическим. От-метим, помимо сказанного выше, ещё одну непривычную для нас вещь: формулировка«тезиса Чёрча» (отождествление интуитивной идеи вычислимой функции с точно опре-делённым классом функций) относится только к всюду определённым функциям. Относительная вычислимость (вычислимость с оракулом)
Можно определить относительную вычислимость одной функции относительно другой(или относительно некоторого множества, которое можно отождествить с его харак-теристической функцией). Впервые обсуждаемое определение сводимости предложилТьюринг в своей диссертации (1939, см. [84]), но это было там побочной темой и рас-сматривалось лишь в некотором частном случае (сводимость к некоторому конкретно-му множеству). Он пишет:Let us suppose that we supplied with some unspecified means of solving number-theoretic problems; a kind of oracle as it were. We will not go any further into thenature of this oracle than to say that it cannot be a machine. With the help ofthe oracle we could form a new kind of machine (call them o -machines), havingas one of its fundamental processes that of solving a given number-theoreticproblem. More definitely these machines are to behave in this way. The moves ofthe machine are determined as usual by a table except in the case of moves froma certain internal configuration o . If the machine is in the internal configuration o and if the sequence of symbols marked with l is then the well formed formula A , then the machine goes into the internal p or t according as it is or is not truethat A is dual. The decision as to which is the case is referred to the oracle.These machines may be described by tables of the same kind as used for thedescription of a -machines, there being no entries, however, for the internalconfiguration o .Общее определение сводимости по Тьюрингу предложил Пост в своей знаменитой ста-тье [86, раздел 11]— той самой, в которой он сформулировал проблему Поста о суще-ствовании перечислимого неразрешимого множества, не являющегося полным (к кото-рому сводятся не все перечислимые множества). Формально говоря, Пост в своём опре-делении ограничивается сводимостью одного перечислимого множества к другому, нореально требование перечислимости в его тексте не используется. Определение Постаследует схеме Тьюринга и использует машины с оракулом. Клини в статье 1943 года [85]предлагает другой вариант определения: к определению общерекурсивных функций спомощью выводов в исчислении равенств можно добавить равенства, выражающие таб-лицу значений для произвольных (всюду определённых) функций ψ , . . . , ψ k , и назватьте функции, которые можно определить таким способом, общерекурсивными относи-тельно ψ , . . . , ψ k :A function φ which can be defined from given functions ψ , . . . , ψ k by a series ofapplications of general recursive schemata we call general recursive in the given9unctions; and in particular, a function φ definable ab initio by these means wecall general recursive .Но это описание дальше не развивается и не используется, оставаясь лишь пояснениемк даваемому дальше определению, и для частично рекурсивных функций (в отличиеот общерекурсивных) относительная вычислимость не рассматривается. В книге 1952года Клини говорит и о частичных функциях и их вычислимости и доказывает, чтотакое определение относительной вычислимости (с выводами в исчислении равенствпо Эрбрану и Гёделю) эквивалентно определению Тьюринга– Поста [65, §69]. При этоморакул должен быть всюду определённой функцией (или набором таких функций). Но,в отличие от статьи Поста, где рассматривались лишь перечислимые множества в каче-стве оракулов, эта функция может быть произвольной всюду определённой функцией.Обзор различных определений относительной вычислимости можно найти в [109,Section 4.3, “History of Relative Computability” ].Теперь мы можем указать, в чём была новизна работы Успенского: • впервые было явно указано (со ссылкой на Трахтенброта— видимо, на неопубли-кованное сообщение) простое определение частичной рекурсивности, абсолютнойи относительной, с помощью операций подстановки, рекурсии и минимизации ( µ -оператора); • была доказана (одновременно с книгой Клини [65] и гораздо более отчётливо)эквивалентность этого определения с другими определениями вычислимости (аб-солютной и относительной); • было дано (впервые) определение относительной вычислимости, не связанное нис какой конкретной вычислительной моделью, а использующее лишь класс вы-числимых функций, и доказана эквивалентность этого определения другим опре-делениям относительной вычислимости; • наконец, в дипломной работе Успенского была впервые изложена конструкциямашин Колмогорова (позже опубликованная в совместной статье Колмогорова иУспенского [16]), дано определение относительной вычислимости в терминах этоймодели и доказана эквивалентность этого определения другим.Третий пункт этого перечня требует некоторых пояснений. Определение Тьюрин-га– Поста для относительной вычислимости является модификацией соответствующе-го определения для абсолютной вычислимости: мы расширяем класс машин Тьюринга,дополнительно разрешая получать ответы от оракула. Аналогичным образом опреде-ление Клини для относительной вычислимости модифицирует определение частичнорекурсивной функции. Таким образом, даже если мы договорились, какие функциимы считаем (абсолютно) вычислимыми, нам нельзя ещё забыть про конкретную мо-дель вычислений и не возвращаться к ней: определяя относительную вычислимость,надо снова вспомнить о модели и её модифицировать. В отличие от этой ситуации,в определении из работы Успенского относительная вычислимость определяется с по-мощью алгоритмов диалога с оракулом, то есть требуется вычислимость некоторых(частичных) функций, задающих этот диалог (функций, указывающих следующий во-прос к оракулу при известном входе, предыдущих вопросах и ответах на них). Таким К сожалению, дипломная работа не была опубликована, несмотря на рекомендации в отзывах, такчто вряд ли повлияла на дальнейшее развитие событий, но тем не менее.
Теорема Гёделя и теория алгоритмов
Теорема Гёделя и теория вычислимых функций появились не только одновременно, нои вместе, как сиамские близнецы. В классической статье Гёделя, где доказан его ре-зультат о неполноте формальных теорий, одновременно было введено понятие рекур-сивной функции (то, что теперь называется «примитивно рекурсивными» функциями,см. выше), и это понятие было важным техническим средством в доказательстве. Аименно, различные функции, связанные с кодированием формул и выводов натураль-ными числами (их «гёделевыми номерами», как раньше говорили), были определенырекурсивно, и это определение использовалось для погружения рассуждений о выводахв формальную систему.С другой стороны, определение общерекурсивных функций было дано в терминахформальной системы (исчисления равенств), восходящей к Эрбрану и Гёделю.Можно, пожалуй, сказать, что одним из первых достижений и в области теориивычислений, и в области теории доказательств, было разделение этих «сиамских близ-нецов», и было это не таким простым делом, как сейчас кажется. Сначала Тьюринг иПост предложили модель вычислений (машины Тьюринга– Поста), позволившую опре-делить вычислимость безо всякого упоминания формальных теорий и выводов в них.Общая природа теоремы Гёделя и её связи с теорией алгоритмов были осознаны в1940-е годы, видимо, в первую очередь Клини и Колмогоровым.В статье Клини 1943 года [85] была указано, что теорема Гёделя по существу озна-чает неперечислимость множества истинных формул, а в статье 1950 года [87] анало-гичная интерпретация была дана для теоремы Гёделя в форме Россера и указано, чтоона соответствует построению пары эффективно неотделимых перечислимых множе-ства. Но хотя по существу все необходимые наблюдения были уже сделаны, по формеизложение Клини и в этой статье 1950 года, и в классическом учебнике 1952 года [65]остаётся ещё тесно связанным с языком теории примитивно рекурсивных функций(достаточно сказать, что изложение в [87] начинается словами “Let T be the primitiverecursive predicate so designated in a previous paper by the author”), и процедура погруже-ния неотделимых множеств в формальную теорию скорее подразумевается, чем явноописана.Примерно в то же время, и, вероятно, независимо к пониманию соотношения междутеоремой Гёделя и теорией алгоритмов пришёл Колмогоров. Как рассказывает Успен-ский в [51, с. 323],2 декабря 1952 г. Колмогоров изложил мне весьма кратко, в течение пятиминут,— но зато дал списать с заготовленной им бумажки, озаглавленной11Гёдель и рекурсивная перечислимость»,— основополагающие идеи о связитеоремы Гёделя о неполноте аксиоматических систем (для самых общих ис-числений) с существованием множеств, не являющихся рекурсивными, ипар множеств, не являющихся рекурсивно отделимыми. Бумажка была на-писана им «для себя», и разобраться в ней, а тем более в его сопутствующихустных комментариях, мне было тогда непросто. Потом всё как-то выстрои-лось, и 8 мая 1953 г. Колмогоров представил в «Доклады АН СССР» моюзаметку «Теорема Гёделя и теория алгоритмов», написанную на основе егоидей. Высокое искусство Колмогорова как учителя состояло в умении со-здать у ученика впечатление, что именно он, ученик, полноценный авторстатьи. Колмогоров во много раз реже, чем имел на это все права, выступалв роли соавтора своих учеников h . . . i В 1958 г. в «Успехах математическихнаук» под двумя нашими фамилиями вышла статья «К определению алго-ритма», в которой мне принадлежит, по существу, лишь черновая работа.(Статьи, о которых идёт речь: [6, 16]. Вторая из них содержит изложение вычислитель-ной модели с преобразованием графов, которая фигурировала уже в дипломной работеУспенского.)В статье Успенского 1953 года [6] было отчётливо указано, уже без всякого упомина-ния о примитивно рекурсивных функциях, что теорема Гёделя о том, что достаточнобогатая формальная система (скажем, формальная арифметика) неполна и не можетбыть пополнена, следует из того, что существуют перечислимые неотделимые множе-ства и что эта пара множеств погружается в формальную систему— как сейчас сказалибы, сводится к паре (доказуемые формулы, опровержимые формулы). А эффективнаянепополнимость (тот факт, что по расширению формальной системы дополнительнымиаксиомами можно алгоритмически указать формулу, которая остаётся недоказуемой инеопровержимой) следует из существования эффективно неотделимых перечислимыхмножеств. Но, повторим ещё раз, всё это по существу уже было в работе Клини [87], окоторой Колмогоров и Успенский, судя по всему, тогда не знали. Успенский ссылаетсяна работу Клини 1943 года [85], говоря о рекурсивных функциях, но говоря о существо-вании перечислимых неотделимых множеств, не ссылается на [87], где они построены, аговорит лишь, что они были построены Новиковым (не указывая никакой публикацииНовикова, а лишь давая ссылку на работу Трахтенброта 1953 года).Говоря о ситуации в целом, можно сказать, что есть два взаимно дополнительныхвзгляда на теорему Гёделя. С одной стороны, она является реализацией парадокса лже-ца (в одном из вариантов он говорит, что утверждение «Это утверждение ложно» неможет быть ни ложным, ни истинным): если вместо этого сделать утверждение «Этоутверждение недоказуемо», то оно будет истинным (и потому неопровержимым), нонедоказуемым. Это объяснение не ссылается на теорию алгоритмов, хотя для обосно-вания возможности записать неформальные рассуждения в формальной арифметикеможно, следуя Гёделю, использовать примитивно рекурсивные функции как техниче-ское средство. С другой стороны, теорема Гёделя является следствием существованиянеразрешимых перечислимых множеств (или, в симметричном варианте, существова-ния неотделимых перечислимых множеств), и в таком изложении никакой «самоприме-нимости» не заметно. Но, конечно, она никуда не делась, переместившись в конструк-цию неразрешимого перечислимого множества (или неотделимых множеств). Эта кон-струкция следует идее «диагонального аргумента» Кантора, которая, в свою очередь,является формой проявления самоприменимости («диагональ» состоит из результатовприменения вычислимой функции к своему номеру, или алгоритма к его собственномутексту). 12последствии Успенский опубликовал подробное изложение доказательства теоре-мы Гёделя с помощью средств теории алгоритмов (а также изложение начал этой тео-рии) сначала в виде статьи [22], а затем (в расширенном виде) брошюры [25] в серии«Популярные лекции по математике». Это изложение до сих пор остаётся, пожалуй,наиболее доступным и корректным изложением теоремы Гёделя для неспециалистов(по крайней мере если говорить о её алгоритмическом аспекте).Помимо этого, в [22, 25] была намечена (совсем не очевидная в то время, достаточ-но сравнить с тем же учебником Роджерса [70]) схема изложения теории алгоритмов.Традиционно (в том числе и в книге самого Успенского [18]) изложение теории алго-ритмов начиналось с подробного разбора какой-то вычислительной модели (сначала вэтом качестве были популярны рекурсивные функции, потом машины Тьюринга; Мар-ков использовал для этого нормальные алгорифмы), и это занимало достаточно многоместа и времени. Лишь после этого оставшиеся слушатели (читатели) знакомились спростейшими фактами вроде теоремы Поста (перечислимое множество с перечисли-мым дополнением разрешимо), и т.п. Конечно, можно было пропустить первую часть,с построением конкретной вычислительной модели, и начинать прямо со второй, рас-суждая как в анекдоте о беспроволочном телеграфе («представьте себе длинную кошку,которую в одном городе дёргают за хвост, а в другом она мяукает— это проволочныйтелеграф,— а теперь то же самое, но без кошки»), тогда рассуждения становились про-стыми и наглядными, но беспочвенными.Выход, предложенный Успенским в [22] (а до этого использованный в его лекциях1972/3 года, но, вероятно, и в предыдущие годы), состоял в следующем: мы рассуждаемо классе вычислимых функций, приняв на веру (в качестве «аксиом») некоторые свой-ства этого класса, отчётливо сформулированные, но оставленные без доказательства.Помимо вычислимости конкретных функций (а также сохранения вычислимости приконкретных построениях одних функций из других), Успенский выделяет два такихсвойства, называя их «аксиомой протокола» и «аксиомой программы».
Аксиома протокола утверждает, что для всякого алгоритма A существует разреши-мое множество R , элементы которого называются «протоколами», и две вычислимыефункции α и ω . Неформально говоря, элементы R являются протоколами (програм-мисты сказали бы: «логами») вычисления алгоритма A , то есть записями всех после-довательных шагов его работы на некотором входе, в тех случаях, когда эта работазавершается и даёт какой-то результат. Функция α выделяет из протокола исходноеданное (вход), а ω —результат работы (выход). Формально же требуется выполнениетакого свойства: выход алгоритма A на входе x равен y тогда и только тогда, когдасуществует r ∈ R , для которого α ( r ) = x и ω ( r ) = y . Аксиома программы утверждает, что есть некоторое разрешимое множество P , эле-менты которого называются «программами», и алгоритм U применения произвольнойпрограммы p ∈ P к произвольному входу x (таким образом, входом алгоритма U явля-ется пара h p, x i ). При этом любая вычислимая функция f задаётся некоторой програм-мой p в том смысле, что U ( p, x ) = f ( x ) для всех x . Последнее равенство понимаетсятак: обе его части одновременно определены или одновременно не определены, и равныв том случае, когда определены.Приняв эти аксиомы, можно развивать теорию алгоритмов, не вдаваясь в техни-ческие детали модели вычислений. Вместе с тем остаётся совершенно понятно, чегонедостаёт в этих рассуждениях: мы должны указать конкретную модель вычисления,научиться в ней программировать те конструкции, которые использованы в доказа-тельствах (и которые не так уж и просты, достаточно вспомнить, скажем, метод при-оритета), а также проверить выполнение аксиом протокола и программы для этой мо-13ели вычислений. Конечно, и после этого некоторый психологический барьер остаётся(многим людям, которые легко ориентируются в достаточно сложных математическихконструкциях, теория вычислимости всё же кажется чем-то странным), но по крайнеймере он становится более явным и отчётливо видным. Для доказательства теоремы Гёделя нужна и третья аксиома, не следующая из этихдвух— что всякую вычислимую функцию можно выразить арифметической формулой(и тут снова приходится обращаться к конкретной модели вычислений).Заметим, что при таком «машинно-независимом» изложении теории вычислимостимы не имеем права вновь возвращаться к вычислительной модели, рассуждая, скажем,о преобразованиях программ или о вычислениях с оракулом, а должны давать все необ-ходимые определения, ссылаясь только на понятие вычислимой функции. Как мы ужеговорили, такое определение для относительной вычислимости было (впервые) даноУспенским в дипломной работе, затем это было сделано для сводимости по перечис-лимости, а также для «способов программирования» (формализацией которых сталовведённое Успенским понятие главной нумерации). Об этих двух последних достиже-ниях Успенского мы говорим в следующем разделе.Ещё можно отметить, что эта аксиоматизация теории алгоритмов позволяет строгообосновать известное наблюдение о том, что большинство результатов теории алго-ритмов «релятивизуются», то есть сохраняют силу, если вычислимые функции заме-нить функциями, вычислимыми с некоторым фиксированным оракулом (в качествекоторого можно взять множество или всюду определённую функцию). В самом деле,достаточно проверить, что для этого релятивизованного класса (функций, вычисли-мых с данным оракулом) выполнены все аксиомы теории алгоритмов (кроме свойстваарифметичности, естественно), и потому выполнены и все теоремы, выводимые из этихаксиом. Успенский поставил вопрос, полностью ли объясняет это наблюдение возмож-ность релятивизации, то есть верно ли, что всякое утверждение, выполненное для всехклассов вычислимых с некоторым оракулом функций, является следствием указанныхим аксиом. Оказалось, что когда этот вопрос поставлен, получить на него (положи-тельный) ответ уже несложно [104].
Вычислимые отображения множеств исводимость по перечислимости
Понятие сводимости, введённое Тьюрингом и Постом и рассмотренное в дипломнойработе Успенского (см. выше), можно назвать «сводимостью по разрешимости»: сво-димость множества A к множеству B гарантирует, что если B разрешимо, то и A разрешимо. Можно сказать, что в этом определении мы «сводим задачу разрешениямножества A к задаче разрешения множества B ». В работе [9] Успенский предлагаетопределение сводимости по перечислимости , в котором речь идёт о сведении задачиперечисления множества A к задаче перечисления другого множества B . Это опре- Сейчас ситуация с методической точки зрения изменилась, прежде всего потому, что большинствоприступающих к изучению теории вычислимых функций уже имеют программистский опыт. Возмож-но, современная реализация педагогических идей Успенского состояла бы в том, что мы предлагаемслушателям представить себе знакомый им язык программирования, к которому добавлены библио-течные функции интерпретатора этого же языка (аргументами которого являются строка, понимаемаякак программа, и вход; это соответствует аксиоме программы), а также пошагового отладчика (кото-рый комбинирует аксиому программы с аксиомой протокола; на вход ему подаются текст программы,вход и число шагов работы отлаживаемой программы). Если к этому добавить ещё и функцию (безаргументов), выдающую текст текущей исполняемой программы, это облегчит и доказательство тео-ремы о неподвижной точке, сделав её самоочевидной. вычислимой операции надмножествами.В простейшем случае (одноместная операция, аргументами и значениями которойявляются подмножества натурального ряда) вычислимые операции определяются так.Введём на множестве P ( N ) всех подмножеств натурального ряда топологию. Для каж-дого конечного множества X ⊂ N рассмотрим семейство O ( X ) всех его надмножеств,и будем считать открытыми в P ( N ) все такие семейства и все их объединения. Теперьрассмотрим все непрерывные в смысле этой топологии отображения F : P ( N ) → P ( N ) .Легко проверить, что все такие отображения монотонны (если U ⊂ V , то F ( U ) ⊂ F ( V ) ),и значение F на любом множестве U определяется значениями F на конечных под-множествах U (надо объединить F ( X ) для всех конечных X ⊂ U ). Значения F наконечных множествах можно описать множеством пар {h n, X i | n ∈ F ( X ) } (здесь n — натуральное число, а X — конечное множество натуральных чисел). Непрерывноеотображение F : P ( N ) → P ( N ) Успенский называет вычислимой операцией , если соот-ветствующее ему множество пар перечислимо (заметим, что эти пары являются ко-нечными объектами, так что можно говорить о перечислимости их множества). Послеэтого определяется сводимость по перечислимости: множество A сводится по перечис-лимости к множеству B , если существует вычислимая операция F , переводящая B к A . Отмечается, что сводимость по Тьюрингу можно определить в терминах своди-мости по перечислимости: всюду определённая функция ϕ вычислима по Тьюрингу соракулом для всюду определённой функции ψ тогда и только тогда, когда график ϕ сводится по перечислимости к графику ψ . Отсюда же получается и критерий своди-мости множеств (переходом к их характеристическим функциям). Указывается, чтов этих же терминах можно получить определить и понятие частично рекурсивногооператора в смысле Клини ([65], см. обсуждение ниже).Наконец, в этой же работе указывается эквивалентность предложенного определе-ния вычислимой операции с двумя «машинно- зависимыми» определениями. Соответ-ствующие понятия Успенский называет «операциями Колмогорова» и «операциямиПоста» (хотя в явном виде они в публикациях Колмогорова и Поста не встречаются).В другой работе 1955 года ([10], см. также изложение результатов этой работы снекоторыми добавлениями в [12]) вводится (со ссылкой на доклад Колмогорова насеминаре по рекурсивной арифметике в 1954 году) понятие нумерации, определяетсяпонятие главной нумерации системы перечислимых множеств и устанавливается связьвычислимых операций на перечислимых множествах в смысле [9] с алгоритмическимипреобразованиями номеров.Более подробно. Пусть мы хотим говорить об алгоритмических преобразованиях программ вычислимых функций (или перечислимых множеств). Тогда нам мало знать,какие функции вычислимы (какие множества перечислимы), но нужно ещё и оговоритькласс «способов программирования», используемых для записи их программ. Програм-мы являются обычно словами в некотором алфавите, но их можно отождествить с на-туральными числами (при какой-то естественной нумерации слов). Тогда способ про-граммирования вычислимых функций превращается в универсальную функцию двухаргументов: U ( n, x ) есть результат применения программы с номером n в входу x , кото-рый мы тоже считаем натуральным числом. Способ программирования перечислимыхмножеств тогда становится универсальным множеством пар h n, x i , для которых x при-надлежит перечислимому множеству, программа которого имеет номер n . На другом(но эквивалентном) языке можно сказать, что способ программирования вычислимыхфункций (перечислимых множеств) представляет собой натуральную нумерацию мно-жества вычислимы функций (перечислимых множеств), то есть отображение всего на-15урального ряда на множество вычислимых функций (соответственно перечислимыхмножеств): числу n ставится в соответствие вычислимая функция (перечислимое мно-жество), программа которого имеет номер n .Если не накладывать на способы программирования (нумерации) никаких ограниче-ний, то оказывается, что не все они одинаково хороши. Разумная теория, описывающаяалгоритмические преобразования программ, требует некоторых дополнительных огра-ничений. По существу они встречались в классических работах Клини ( s - m - n -теорема),но явно впервые были сформулированы в [10]. А именно, требуется, чтобы нумера-ция была главной . Определение главной нумерации включает в себя два требования.Во-первых, нумерация должна быть вычислимой: это значит, что соответствующаяуниверсальная функция является вычислимой (частичной) функцией двух аргументов(вариант для множеств: соответствующее универсальное множество пар должно бытьперечислимым множеством пар). Во- вторых, к этой нумерации должна сводиться лю-бая другая вычислимая нумерация. Это требование означает, что для любой другойвычислимой нумерации того же множества существует (всюду определённая) вычис-лимая функция, преобразующая номера в этой второй нумерации в номера в первой(главной).Используя понятие главной нумерации для перечислимых множеств, можно рас-смотреть вычислимые отображения семейства перечислимых множеств в себя.
Вычис-лимость означает, что есть алгоритм, который по (любой) программе перечислимогомножества даёт (какую-то) программу его образа. Другими словами, мы рассматри-ваем всюду определённые преобразования программ, которых сохраняют эквивалент-ность (эквивалентные программы, то есть программы, задающие одно и то же множе-ство, преобразуются в эквивалентные программы). Успенский доказывает [10, раздел6], что вычислимые отображения семейства перечислимых множеств в себя в точностипредставляют собой ограничения вычислимых операций (на семействе всех множеств)на класс перечислимых множеств. Аналогичное утверждение делается и для подсе-мейства униформных множеств пар (графиков функций): вычислимые отображениясемейства вычислимых функций в себя представляют собой ограничения вычислимыхопераций, отображающих семейство униформных множеств в себя.Опишем связи этих работ Успенского с работами других авторов того же перио-да. Райс [88] рассмотрел вполне перечислимые классы перечислимых множеств, тоесть такие классы перечислимых множеств, для которых множество всех программвсех множеств этого класса перечислимо, и сформулировал гипотезу о том, что всякийтакой класс состоит из всех надмножеств множеств из некоторого перечислимого се-мейства конечных множеств. Эта гипотеза приводится в [10] как теорема 5 и являетсяключевым шагом в доказательстве упомянутых результатов о вычислимых отображе-ниях. Она также доказана в статье самого Райса 1956 года [92], где говорится, что Определение сводимости нумераций также опубликовано в [10] со ссылкой на доклад Колмогорова(видимо, впервые). К сожалению (см. ниже отрывок из воспоминаний Успенского), все три публикации Успенско-го [9, 10, 12] представляют собой краткие заметки в Докладах Академии наук СССР (первые две) ирезюме доклада в Московском математическом обществе (третья), где приводятся только формули-ровки теорем и лемм, используемых в их доказательствах. Сами доказательства были опубликованыв кандидатской диссертации Успенского [11], которая хотя формально и была доступна (её можнобыло заказать и получить в нескольких библиотеках в СССР), но вряд ли повлияла на дальнейшееразвитие области. Да и статьи с кратким изложением [9, 10, 12], видимо, остались неизвестными внеСССР. Позже книга Успенского [18] (учебник по теории вычислимых функций, который стал доктор-ской диссертацией Успенского) была переведена на французский язык—к сожалению, в неё вошлолишь определение главной нумерации, но не результаты о вычислимых операциях и отображениях. f и g —две частичные функции (с натуральными аргументами и значениями).Сводимость f к g (вычислимость f относительно g ) можно понимать в трёх смыслах(каждый следующий сильнее предыдущего).1. График f сводится по перечислимости к графику g .2. Рассмотрим, следуя Успенскому, семейство U всех частичных функций из N в N с топологией, в которой базовыми открытыми множествами являются семей-ства всех продолжений некоторой конечной функции. Непрерывное отображение F : U → U мы будем называть вычислимой операцией, если его ограничение на17онечные функции имеет перечислимый график, то есть если множество всех пар hh x, y i , u i , где x и y —натуральные числа, а u —конечная частичная функция из N в N , и при этом [ F ( u )]( x ) = y , перечислимо. Теперь вычислимость f относительно g можно понимать как существование вычислимой операции, переводящей g в f .3. Можно распространить определение Трахтенброта (см. обсуждение дипломнойработы выше) на частичные функции и говорить, что частичная функция f вы-числима относительно частичной функции g , если f принадлежит частично ре-курсивному замыканию множества частично рекурсивных функций, к которомудобавлена функция g . (Это определение, например, приводится в [100].)Третье определение имеет эквивалентную переформулировку в терминах машиныс оракулом. Эта переформулировка по существу повторяет определение из дипломнойработы Успенского, но для частичных функций. А именно, значение f ( x ) вычисляет-ся алгоритмом, который получает на вход x и может запрашивать значения функции g в произвольных точках, но как только одно из запрошенных значений g окажетсянеопределённым, вычисление «зависает» (прерывается без результата) и f ( x ) остаётсянеопределённым. Второй вариант определения можно тоже переформулировать в тер-минах машин, дополнительно разрешив параллельные запросы нескольких значенийфункции g ; каждый из этих запросов не останавливает вычисление, которое продолжа-ется и получает информацию о запрошенных значениях, если они определены, черезкакое-то время. При этом требуется, чтобы результат вычисления не зависел от того,через какое время поступят запрошенные значения.Эти три определения различаются: каждое следующее сильнее предыдущего (болееограничительно). Разницу между этими определениями можно пояснить двумя приме-рами. Первый пример, разделяющий первый и второй варианты определения, таков.Пусть f —произвольная всюду определённая функция, а g —функция, принимающаятолько нулевые значения, и область определения функции g состоит из всех номеровпар h n, f ( n ) i (для какой-то вычислимой нумерации пар). Тогда f сводится к g в смыслепервого определения, но не обязательно сводится в смысле второго. (Это рассуждениеприведено в примечании Успенского на с. 362 русского перевода книги Роджерса [70]со ссылкой на Д. Г. Скордева; приведённое Роджерсом рассуждение существенно слож-нее.)Второй пример [108, Proposition II.3.20, ссылка на диссертацию Sasso 1971 года] по-казывает разницу между вторым и третьим определениями. Пусть g —произвольнаячастичная функция натурального аргумента, принимающая только нулевые значения.Построим другую частичную функцию f , которая тоже принимает только нулевыезначения, при этом f ( n ) определено (и равно нулю) тогда и только тогда, когда хо-тя бы одно из значений g (2 n ) и g (2 n + 1) определено. Тогда функция f вычислимаотносительно g в смысле второго определения, но не обязательно вычислима в смыс-ле третьего. (В терминах машин: если разрешено параллельно запрашивать g (2 n ) и g (2 n + 1) , ожидая, пока один из этих запросов будет удовлетворён, то вычислить f легко, но если можно лишь запрашивать их последовательно, то ничего не выйдет,потому что неудовлетворённый первый запрос помешает перейти ко второму. Конеч-но, это лишь неформальное пояснение, для доказательства различия нужно строитьсоответствующий пример функции g диагональным методом, и это легко сделать.)Первое определение соответствует тому, что в книге Роджерса [70, §9.8] названо «ча-стичнорекурсивными операторами» (partial recursive operators). Второе соответствуеттому, что названо там же «рекурсивными операторами» (recursive operators).18айхилл и Шепердсон [91] говорят о “partial recursive functionals”, ссылаясь наThesis I ∗† на с. 332 книги Клини [65], но этот тезис (начало страницы 332) не используеттермин “partial recursive functional” и вообще этот термин на с. 332 не встречается. Пред-метный указатель этой книги [65] отсылает к с. 326 по слову “partial recursive functional”,но и эта страница не содержит соответствующего упоминания. Правда, на этой страни-це даётся определение частичной рекурсивности частичной функции ϕ относительночастичных функций ψ , . . . , ψ k , соответствующее сводимости графиков по перечисли-мости (первый вариант из приведённых выше трёх). и говорится о «схеме» (scheme) F , но какие требования предъявляются к этой схеме (должна ли она давать функциютолько в применении к функциям ψ , . . . , ψ k , или к любым k функциям), из текста неясно. (А нумерации класса вычислимых с данным оракулом функций рассматриваютсятолько для случая, когда оракул представляет собой множество или всюду определён-ную функцию.) Но Майхилл и Шепердсон уточняют, что в их результате речь идётот частично рекурсивных функционалах, определённых (и дающих функции) для всехфункций в качестве аргументов, что эквивалентно второму определению, как и должнобыть.Одифредди [108, Definition II.3.6] определяет частично рекурсивные функционалысо ссылкой на Клини [65], но следует третьему варианту определения (композицияопераций подстановки, минимизации и рекурсии, применённых к частично рекурсив-ным функциям и аргументам), который в [65] не встречается. Понятие, соответствую-щее второму варианту определения, он называет “effectively continuous functional” или“recursive operator”, а первому— “partial recursive operator”.Возвращаясь к работам Успенского [9, 10, 12], можно отметить следующие дости-жения: • впервые было дано определение сводимости по перечислимости; • впервые было опубликовано (восходящее к Колмогорову) определение нумерациии сводимости нумераций; • были проанализированы свойства нумераций вычислимых функций и перечис-лимых множеств, необходимые для рассуждений о номерах программ, введенопонятие главной нумерации (впоследствии переоткрытое Роджерсом); • была доказана (до того, как это сделал сам Райс) гипотеза Райса об описаниивполне перечислимых классов перечислимых множеств; получены также анало-гичные результаты для вычислимых функций (вместо перечислимых множеств);в частности, доказана невозможность распознавания нетривиальных свойств вы-числимых функций по их номерам в главной нумерации; • было дано определение вычислимой операции в терминах топологического подхо-да (вычислимость как некоторый специальный случай непрерывности) и доказа-но (одновременно с Майхиллом и Шепердсоном), что ограничения вычислимыхоперацией на вычислимые функции можно эквивалентно описать как алгорит-мические преобразования программ, а также доказана аналогичная теорема дляопераций над перечислимыми множествами.Во избежание недоразумений отметим, что Успенский не рассматривает алгорит-мы, определённые на всех программах всюду определённых функций и дающие одина-ковые значения на эквивалентных программах: эти работы Крайзеля, Лакомба, Шён-филда [96] (позднее обобщённые Цейтиным [98] на конструктивные метрические про-странства) никак не пересекаются с работами Успенского.19риведём отрывок из воспоминаний Успенского [63, с. 905–907, 912], где он пишето своих результатах 1955 года и об их представлении на Третьем всесоюзном матема-тическом съезде (1956):В сделанном 26 июня обзорном докладе «Об алгоритмической сводимости»я рассказывал о четырёх видах сводимости и связях между ними. Это сво-димость по вычислимости , состоящая в сведении вычисления одной функ-ции к вычислению другой. Это сводимость по разрешимости , состоящаяв сведении задачи построения разрешающего алгоритма для одного множе-ства к задаче построения разрешающего алгоритма для другого множества.Это сводимость по перечислимости , состоящая в сведении перечисленияодного множества к перечислению другого. Это сводимость проблем , состо-ящая в сведении решения одной проблемы к решению другой. h . . . i [Гово-ря о понятии сводимости проблем, Успенский указывает на его источник:]В 1955 году интересную разновидность проблем ввёл ученик КолмогороваЮ. Т. Медведев; он же определил понятие сводимости для таких проблем. h . . . i Наименование моего доклада 2 июля было «Понятие программы и вычис-лимые операторы», а сообщения 3 июля— «Вычислимые операции, вычисли-мые операторы и конструктивно- непрерывные функции». Доклад и сообще-ние были тесно связаны тематически.В сообщении 3 июля был изложен (разумеется, без доказательства) резуль-тат, который я считаю своим главным математическим результатом; я пом-ню обстоятельства его получения — теорема 3 (см. ниже). Он составил ос-нову моей кандидатской диссертации, защищённой в октябре 1955 года. До-казательства этого результата, кроме текста диссертации, хранящейся— илитолько уже хранившейся— в библиотеке мехмата, я так никогда и не опубли-ковал. Основная причина, как ни стыдно в этом признаться, банальная лень.Дополнительная причина, не столь постыдная, но глупая, это преследовав-шее [в тексте: преследующее] меня, пока я не повзрослел, желание изложитьвсё в максимально общем виде, но достичь предела в обобщении нереально. h . . . i Теорема 3 . Пусть функция g с натуральными аргументами и значениямиобладает следующим свойством. Если m и n служат программами однойи той же вычислимой функции от s аргументов, то g ( m ) и g ( n ) так-же служат программами одной и той же функции от одного аргумента.Тогда существует вычислимый оператор V со следующим свойством. Длявсякой функции θ с программой n значением V ( θ ) оператора V на функции θ является функция с программой g ( n ) . Философский комментарий . Семиотический смысл теоремы 3 таков: «хо-рошее» вычислимое преобразование имён сопровождается вычислимым пре-образованием соответствующих объектов. Теорема 3 интересовала меня с семиотической точки зрения, h . . . i хотя слова ‘семиотика’ я тогда,скорее всего, ещё не знал. Помню, как я бродил по улицам и думал только об этом. Озарение пришло,когда в течение некоторого времени я проводил дневные часы в квартире моей тёщи в Большом Спасо-глинищевском переулке. Сын ещё не родился, жена и тёща уходили на работу, телефона в квартире небыло, мобильные телефоны ещё не были изобретены. Вот в этой обстановке меня внезапно осенило.»[Примечание В. А. Успенского] онструктивность в классической математике Идея о том, что можно понимать математические утверждения конструктивно, былаизвестна давно («интуиционизм» Брауэра и его последователей и позже «конструкти-визм» Маркова и его учеников). В частности, утверждения вида «для всех x суще-ствует такой y , что. . . » при конструктивном понимании предполагают, что существуетнекоторый способ получения этого самого «существующего» y по любому данному зна-чению x .Обычно вместе с этим предлагали изменить и саму логику, понимая конструктивно,в частности, связку «или» и не пользуясь законом исключённого третьего. Другое (напервый взгляд, напрашивающееся) предложение, а именно, рассматривать «эффектив-ные» аналоги классических определений и результатов внутри обычной («неконструк-тивной») математики, удивительным образом сначала было менее популярно, и частосчиталось, что если уж мы рассуждаем об алгоритмах, то это почему-то обязывает насрассуждать «конструктивно», «финитно» и т.п. В отличие от этой традиции, Успенскийсистематически пропагандировал «классический» подход к алгоритмическим поняти-ям. Вот два примера из его работ.Есть разные конструкции действительных чисел (сечения Дедекинда, фундамен-тальные последовательности, вложенные отрезки, бесконечные десятичные дроби).Для каждой из них можно рассмотреть её эффективный вариант. Скажем, для сече-ний Дедекинда можно требовать существования алгоритма, который по рационально-му числу говорит, в каком из двух множеств сечения оно лежит. Для фундаменталь-ной последовательности рациональных чисел естественно требовать существованияалгоритма, вычисляющего её члены, а также «регулятора сходимости»— алгоритма,находящего по (рациональному) ε > то место, начиная с которого члены последова-тельности отличаются менее чем на ε . Для бесконечной десятичной дроби естественнотребовать существования алгоритма, который по n указывает n - ю цифру дроби, и такдалее.В каждом из этих случаев возникает некоторое подмножество множества действи-тельных чисел (соответствующее тем числам, для которых имеются такие эффектив-ные представления). Можно поставить, оставаясь в рамках «классической математи-ки», вопрос о том, дают ли перечисленные варианты определений одно и то же подмно-жество или разные. Нетрудно убедиться, что одно и то же, и его элементы называют вычислимыми действительными числами .Здесь хорошо видна разница с конструктивистским подходом (скажем, в смыслешколы А. А. Маркова). Для конструктивистов никаких «обычных действительных чи-сел» не существует, и множество конструктивных действительных чисел не являетсяподмножеством никакого большего множества. Вместо этого конструктивным действи-тельным числом называется пара алгоритмов (один вычисляет члены последовательно-сти, другой является регулятором сходимости), и рассуждать о таких парах предлага-ется в рамках конструктивной логики. При этом не все варианты определения действи-тельных чисел равнозначны с точки зрения их конструктивизации. Скажем, подход сдесятичными дробями неудачный— потому, например, что для так определённых кон-структивных действительных чисел нельзя конструктивно определить сложение (какпреобразование, которое по алгоритмам для двух дробей давало бы алгоритм для ихсуммы).Но этот же дефект, замечает Успенский, можно проанализировать и в рамках клас-сической математики. Будем интересоваться не только тем, один и тот же класс дей-ствительных чисел возникает в рамках разных определений или разные, но также и21олее тонким вопросом: эквивалентны ли нумерации множества вычислимых действи-тельных чисел, которые получаются из этих определений (можно ли по номеру вы-числимого действительного числа в одной нумерации алгоритмически получить номертого же числа в другой). И тут возникает та же самая проблема с десятичными дробями(и с дробями в любых системах счисления). Подробно этот вопрос разобран в [18], гдеданы необходимые и достаточные условия, при которых переход от одного основанияк другому (в позиционной записи) эффективен.Другой пример, разобранный Успенским в [19]— эффективизация определения бес-конечного множества. Можно определить бесконечность множества X так: для всякого n в множестве X есть не менее n элементов. Или так: для всякого конечного множе-ства F есть число, которое отличает F от X , то есть принадлежит симметрическойразности F △ X . Оба эти определения можно эффективизировать, потребовав суще-ствования соответствующих алгоритмов. В первом случае речь идёт об алгоритме, ко-торый по n указывает список из n элементов множества X ; во втором случае речь идётоб алгоритме, который применим к любому конечному множеству F и даёт какой-тоэлемент разности F △ X . Можно проверить, что эти свойства (существование того идругого алгоритма) эквивалентны (и даже можно без нарушения эквивалентности тре-бовать, чтобы по F давался элемент разности X \ F ). В терминологии Поста [86] обаэти свойства равносильны тому, что множество X не является иммунным (содержитбесконечное перечислимое подмножество).Не все естественные определения бесконечности приводят к эквивалентным эффек-тивизациям. Скажем, можно сказать, что множество X бесконечно, если для всякого n можно указать начальный отрезок [0 , N ] натурального ряда, содержащий по край-ней мере n элементов множества X . Эффективный вариант этого определения требует,чтобы существовал алгоритм, указывающий N по n . Это более слабое определение эф-фективной бесконечности, которое, как доказал Успенский в [15], равносильно тому, чтомножество не является гипериммунным в смысле Поста [86]. (Параллельно и независи-мо, отвечая на вопрос Колмогорова, это же доказали А. В. Кузнецов и Ю. Т. Медведев.)Интересно отметить, хотя это и не имеет отношения к работам Успенского, чтоважнейшее достижение алгоритмической теории случайности, а именно, определениеслучайности по Мартин- Лёфу, данное им в 1966 году [101], тоже можно рассматриватькак естественную эффективизацию классического (во всех смыслах этого слова) опре-деления нулевого множества (множества нулевой меры в смысле Лебега). В этом клас-сическом определении (скажем, для подмножеств отрезка) говорится, что множество X ⊂ [0 , является нулевым, если для всякого ε > существует покрытие множества X интервалами с суммой длин не больше ε . По очевидным причинам можно ограничить-ся рациональными значениями ε и интервалами с рациональными концами. Тогда онибудут конструктивными объектами и можно эффективизировать определение, потребо-вав, чтобы существовал алгоритм, который, получив на вход ε , перечисляет интервалыпокрытия с требуемыми свойствами. Это и предложил Мартин- Лёф.Можно добавить, что многие вопросы и результаты алгоритмической теории случай-ности можно интерпретировать как вопросы об эффективизации классических понятийи теорем. Скажем, критерий Соловея случайности по Мартин- Лёфу, как заметил Алек-сандр Буфетов, является эффективным вариантом классической леммы Бореля–Кан-телли. При этом интересно, что стандартное её доказательство (хвосты сходящегосяряда могут быть сколь угодно малы) не эффективизируется, и приходится использо-вать другое (тоже естественное и несложное, см. подробности в [59]). Другой поучи-тельный пример того же рода— обнаруженное В. В. Вьюгиным (учеником Успенского)доказательство эффективного варианта эргодической теоремы [110].22 лгоритмическая теория информации Странным образом Успенский, будучи учеником Колмогорова и работая рядом с нимна мехмате МГУ, был в 1960-е годы в стороне от исследований Колмогорова, связан-ных с определением понятия сложности конечного объекта. По его словам, он впервыевплотную занялся этой областью, готовя доклад (с А. Л. Семёновым) на конференции вУргенче [24, 26]. В этом докладе была предложен подход к классификации различныхвидов сложности (или, как предпочитал говорить Успенский, «энтропии») для конеч-ных объектов, определённых к тому времени (простая, префиксная, условная, моно-тонная энтропии, а также энтропия разрешения). Изначально (см. [107]) этот подходбыл предложен в терминах f -пространств и операций над ними, что можно рассматри-вать как развитие подхода Успенского к определению вычислимых отображений какчастного случая непрерывных. Однако для целей классификации различных видовколмогоровской сложности можно было обойтись и без f -пространств, и Успенскийс Семёновым в [24, 26] предложили более простой вариант определения (в терминахотношения согласованности на описаниях и объектах), достаточный для целей класси-фикации. Впоследствии это упрощённое описание было изложено в [38, 40]; подробноеизложение с топологической точки зрения (но без f - пространств) можно найти в [59].Алгоритмической теории информации (точнее, различным определениям понятияслучайности) посвящён также обзор [35] и монография [59]. Популярная лекция (длястудентов- младшекурсников), посвящённая различным определениям случайности,была прочитана Успенским на летней школе «Современная математика» в Дубне в2005 году, и материалы этой лекции были изданы [50] и вошли в качестве приложенияв монографию [59].До сих пор остаётся открытым вопрос, поставленный Успенским, Семёновым и Муч-ником в [43] о том, совпадают ли понятия случайности по Мартин- Лёфу и «непредска-зуемости» (отсутствия выигрышной вычислимой стратегии в немонотонных играх, см.подробнее в [50, 59]). Популяризация
Есть разные представления о том, что такое «популяризация науки» (по-французскиэто называют vulgarisation, что для русского уха звучит обидно, хотя и не совсемнезаслуженно). Можно рассказывать байки о трудной судьбе или особенностях лич-ной жизни выдающихся учёных. Можно пересказывать недавно прочитанное в другойпопулярной книге, добавляя «оживляж». Видимо, это полезное дело— но Успенскийвсю жизнь занимался другим, пытаясь честно объяснить «суть дела». При этом слож-ность темы, естественно, зависела от того, на кого рассчитано объяснение, но всегдаэто было настоящее объяснение того, что объяснить можно, с отчётливым указанием,что именно оставлено без доказательства или уточнения. При этом он не боялся объ-яснить очевидное, помня, что известное известно немногим («Что мы знаем о лисе?Ничего, и то не все»— писал сосед Колмогорова по даче, знаменитый детский писательБорис Заходер.)Ещё будучи студентом, Успенский (вместе со старшим соавтором, Евгением Бори-совичем Дынкиным) написал книгу «Математические беседы» [5] по материалам мате-матических кружков, где он сначала был участником, а потом руководителем. В нейнесколько тем (раскраска графов, начала теории чисел и теории вероятностей) пред-ставлены в виде последовательности задач, как это и делалось на кружке, и приведенырешения этих задач. И до этого издавались книжки с задачами математических круж-ков, но тут идея была в том, что эти задачи в целом образуют изложение некоторой23атематической теории. Книга Дынкина и Успенского была библиографической редко-стью, пока не была переиздана уже сравнительно недавно, в 2004 году (и не появиласьв интернете).Успенский читал лекции для школьников (в частности, для участников матема-тических олимпиад) и написал несколько популярных брошюр в серии «Популярныелекции по математике», никак не связанных с его собственными научными интересами(математической логикой и теорией алгоритмов): про применение механики в матема-тике [17] и про треугольник Паскаля [20]. Впрочем, последняя брошюра затрагиваети логический вопрос: что означает решить комбинаторную задачу и почему нужнофиксировать список разрешённых операций (скажем, включив в него факториалы, ноисключив обозначения для биномиальных коэффициентов).Две другие брошюры в этой серии («Машина Поста» [23] и «Теорема Гёделя онеполноте» [25]) уже посвящены темам из математической логики и теории алгорит-мов. Первая из них совсем элементарна и основана на занятиях с младшеклассниками,вторая, наоборот, написана на основе статьи в «Успехах математических наук» [22] ипредполагает некоторую математическую культуру, но вполне может быть прочитана ипонята продвинутыми старшеклассниками. Популярному изложению ещё одной темы,так называемого «нестандартного анализа», в котором методы математической логи-ки используются для математически корректного рассмотрения бесконечно малых ибесконечно больших величин, посвящена брошюра [27]; её расширенный вариант былопубликован затем издательством «Наука» [31].Как герой «Игры в бисер» Гессе, становясь старше, Успенский объяснял всё бо-лее и более базовые вещи, занявшись проповедью математики среди «гуманитари-ев». Впрочем, началось это уже давно, в 1960-е годы, когда он разрабатывал и осу-ществлял курс математики для отделения теоретической и прикладной лингвисти-ки филологического факультета МГУ, но в последние два десятилетия своей жизнион обращался к гораздо более широкой аудитории. Несколько его лекций на лет-них школах по математике и лингвистике в Дубне, к счастью, сохранились как ви-деозаписи (прежде всего благодаря Виталию Арнольду), и по ним (см. ссылки в ) мож-но составить представление об Успенском как лекторе— хотя, конечно, в полной мереоценить его можно было только на университетских лекциях, особенно спецкурсах.При этом проповедь Успенского была именно проповедью математики, а не «о мате-матике». Он рассказывал простые вещи, но всерьёз, с определениями, примерами идоказательствами. Одна из его последних книжек [56] так и называется: «Простейшиепримеры математических доказательств». Другая книжка [44] называется «Что такоеаксиоматический метод?»— и там тоже подробно разобрано множество примеров (вчастности, из школьной геометрии, точнее, из той части школьной геометрии, котораяв школах пропускается). Например, объясняется, как вывести из аксиом, что для вся-кой прямой найдётся точка, на ней не лежащая. Материалы из этих двух книг вошлив сборник «Апология математики» [54] (вместе с другими статьями, уже более общегохарактера). И удивительным образом проповедь Успенского имела успех (по крайнеймере в том же смысле, что у Антония Падуанского): Успенскому была присужденапремия «Просветитель» (учреждённая Дмитрием Борисовичем Зиминым и фондом«Династия») за 2010 год в области естественных и точных наук.Помимо собственных книг, большой заслугой Успенского (то, что называют по- ан-глийски community service) является организация издания многих классических книгпо математической логике и теории алгоритмов на русском языке: он переводил (поинициативе Колмогорова) книгу Р. Петер о рекурсивных функциях [64], редактиро-24ал перевод классических монографий Клини [65] и Роджерса [70], книги Дэвиса онестандартном анализе [71] (видимо, первого изложения идей нестандартного анализа,появившегося по-русски), фундаментального учебника Чёрча по логике [67], первоготома «Начал математики» Бурбаки [68], а также книги Эшби о кибернетике [66].25 писок литературы
Сайты с работами В. А. Успенского: [1] Страница В. А. Успенского на сайте mathnet.ru : [2] Страница В. А. Успенского на сайте кафедры математической логики и теории ал-горитмов мехмата МГУ: http://lpcs.math.msu.su/~uspensky/ Работы В. А. Успенского [3] Геометрический вывод основных свойств гармонических функций.
Успе-хи математических наук , 1949, том IV, выпуск 2(30), с. 201–205, http://lpcs.math.msu.su/~uspensky/bib/Uspensky_1949_UMN_Geometr_vyvod.pdf , http://mi.mathnet.ru/umn8612 [4] Общее определение алгоритмической вычислимости и алгоритмиче-ской сводимости . Дипломная работа. Научный руководитель — ака-демик А. Н. Колмогоров. Московский государственный университет им.М. В. Ломоносова, механико- математический факультет. Машинопись. 90 с. http://lpcs.math.msu.su/~uspensky/bib/Uspensky_1952_Diploma.pdf [5] Е. Б. Дынкин, В. А. Успенский,
Математические беседы. Задачи о мно-гоцветной раскраске. Задачи из теории чисел. Случайные блуж-дания . (Библиотека математического кружка, вып. 6). Москва– Ле-нинград: Государственное издательство технико- теоретической лите-ратуры, 1952. http://ilib.mccme.ru/djvu/bib-mat-kr/besedy.htm , . 2- е издание:Наука, 2004. Английский перевод был выпущен в виде трёх брошюр издатель-ством D.C. Heath, Boston в 1963 году, и позже в виде единой книги: E.B. Dynkin,V.A. Uspenskii, Mathematical Conversations: Multicolor Problems, Problems inthe Theory of Numbers, and Random Walks , Dover books in mathematics, DoverPublications, 2006, ISBN 0-486-45351-0.[6] О понятии алгоритмической сводимости. Резюме доклада в Московском матема-тическом обществе 17 марта 1953 года,
Успехи математических наук , т. VIII,вып. 4(56), 1953, июль–август, с. 176, см. http://mi.mathnet.ru/umn8234 [7] Теорема Гёделя и теория алгоритмов. Резюме доклада в Московском матема-тическом обществе 24 марта 1953 года,
Успехи математических наук , т. VIII,вып. 4(56), 1953, июль–август, с. 176–178, см. http://mi.mathnet.ru/umn8234 [8] Теорема Гёделя и теория алгоритмов,
Доклады Академии наук СССР , т. 91, номер64 с. 737–740 (1953), https://istina.msu.ru/publications/article/92662634/ (есть полный текст). Английский перевод: G¨odel’s theorem and the theory ofalgorithms,
American Mathematical Society Translations, Series 2, Advances in theMathematical Sciences , vol. 23 (1963), 103–107,
DOI10.1090/trans2/023/06 [9] О вычислимых операциях,
Доклады Академии наук СССР , том 103, номер 5 (1955),с. 773–776, https://istina.msu.ru/publications/article/92662640/ (есть пол-ный текст) 2610] Системы перечислимых множеств и их нумерации,
Доклады Ака-демии наук СССР , том 105, номер 6 (1955), с. 1155–1158, https://istina.msu.ru/publications/article/92662649/ (есть полный текст)[11]
О вычислимых операциях , диссертация на соискание учёной степени кандидатафизико- математических наук, Московский государственный университет имениМ. В. Ломоносова, механико- математический факультет, октябрь 1955 г.[12] Вычислимые операции и понятие программы. Резюме доклада в Московском мате-матическом обществе 28 февраля 1956 года,
Успехи математических наук , т. XI,вып. 4(70), 1956, июль–август, с. 172–176, http://mi.mathnet.ru/umn7861 [13] Третий всесоюзный математический съезд. Обзорный доклад «Об алгоритмиче-ской сводимости» 26 июня 1955 года. Доклад «Понятие программы и вычислимыеоператоры» 2 июля 1955 года. Сообщение «Вычислимые операции, вычислимыеоператоры и конструктивно-непрерывные функции» 3 июля 1955 года. Краткоесодержание выступлений опубликовано:
Труды третьего всесоюзного Математи-ческого съезда . М.: издательство АН СССР, 1956. Т. 2, с. 66–69 (обзорный доклад),т. 1, с. 186 (доклад), т. 1, с. 185 (сообщение).[14] К теореме о равномерной непрерывности.
Успехи математических наук , т. XII,вып. 1(73), 1957, январь–февраль, с. 100–142, http://mi.mathnet.ru/umn7524 [15] Несколько замечаний о перечислимых множествах.
Zeitschrift f¨urmathematische Logik und Grundlagen der Mathematik , Bd. 3, S. 157–170 (1957), https://istina.msu.ru/publications/article/92666188/ (есть полный текст)Английский перевод: Some remarks on recursively enumerable sets,
AmericanMathematical Society Translations, Series 2, Advances in the Mathematical Sciences ,vol. 23 (1963), 89–101,
DOI10.1090/trans2/023/05 [16] А. Н. Колмогоров, В. А. Успенский. К определению алгоритма.
Успехи ма-тематических наук , т. XIII, вып. 4(82), 1958, июль— август, с. 3–28, http://mi.mathnet.ru/umn7453 . Английский перевод (Elliott Mendelson):Kolmogorov A.N., Uspenskij V.A., On the definition of an algorithm,
AmericanMathematical Society Translations, Series 2, Advances in the Mathematical Sciences ,vol. 29 (1963), 217–245,
DOI10.1090/trans2/029/07 [17]
Некоторые приложения механики к математике (Популярные лекции по мате-матике, выпуск 27). М.: Государственное издательство физико- математической ли-тературы, 1958. 48 с., [18] К вопросу о соотношении между различными системами конструктивных действи-тельных чисел,
Известия высших учебных заведений , математика, 1960, номер2(15), с. 199–208, http://mi.mathnet.ru/ivm2028 [19]
Лекции о вычислимых функциях . Москва, государственное издательство физи-ко- математической литературы, 1960. 492 с. Французский перевод: Ouspenski V.A.,Le¸cons sur les fonctions calculables. Paris, Hermann, 1966. 412 p.[20]
Треугольник Паскаля (Популярные лекции по математике, выпуск 43.) М.:Наука,главная редакция физико- математической литературы, 1966. 35 с. Второе издание(дополненное), 1979. 48с.
Математи-ческие заметки , т. 6, номер 1 (1969), с. 3–9, http://mi.mathnet.ru/mz6891 /Английский перевод: V.A. Uspenskii, Reduction of computable andpotentially computable numerations,
Mathematical Notes of the Academy ofSciences of the USSR , 1969, vol. 6, no. 1, 461–464,
DOI10.1007/BF01450246 , https://istina.msu.ru/publications/article/92645436/ (есть полный текст)[22] Теорема Гёделя о неполноте в элементарном изложении. Успехи мате-матических наук , т. XXIX, вып. 1(175), с. 3–47 (1974, январь–февраль), http://mi.mathnet.ru/umn4322 . Английский перевод (E. Lichfield): Uspenskii V.A.,An elementary exposition of G¨odel’s incompleteness theorem,
Russian MathematicalSurveys , vol. 29 (1974), no. 1, 63–106,
DOI10.1070/RM1974v029n01ABEH001280 , https://istina.msu.ru/publications/article/92645484/ (есть полный текст)[23] Машина Поста . (Популярные лекции по математике, выпуск 54). М.:Наука,главная редакция физико-математической литературы, 1979. 96 с., [24] Uspensky V.A., Semenov A.L., What are the gains of the theory of algorithms: Basicdevelopments connected with the concept of algorithm and with its application inmathematics,
Algorithms in Modern Mathematics and Computer Science, Proceedings,Urgench, Uzbek SSR, September 16–22, 1979 . Edited by A.P. Ershov and D. Knuth,Lecture Notes in Computer Science, 122, Springer, 1981, p. 100–234.[25]
Теорема Гёделя о неполноте . (Популярные лекции по математике, выпуск57) М.:Наука, главная редакция физико-математической литературы, 1982, . Английский перевод опубликован в1987 (Mir Publisher, перевёл N. Koblitz) и затем в
Theoretical Computer Science , (1994), 239–319.[26] Успенский В.А., Семёнов А.Л., Теория алгоритмов: основные открытия и прило-жения. В сборнике Алгоритмы в современной математике и её приложениях.Материалы международного симпозиума, Ургенч, УзССР, 16–22 сентября 1979г.
А.П. Ершов, Д. Кнут, редакторы. Часть I, с. 99–342.[27]
Нестандартный, или неархимедов, анализ . М.:Знание, 1983. 61 c.[28] Успенский В.А., Кановей В.Г., Проблемы Лузина о конституантах и их судьба.
Вестник Московского университета. Серия 1: Математика, механика , 1983, но-мер 6, с. 73–87, https://istina.msu.ru/publications/article/93856782/ . Ан-глийский перевод: Uspenskii V.A., Kanovei V.G., Luzin’s problems on constituentsand their fate,
Moscow University Mathematics Bulletin , vol. 38, no. 6 (1983), 86–102.(Allerton Press, inc.)[29] Вклад Н. Н. Лузина в дескриптивную теорию множеств и функций: понятия,проблемы, предсказания.
Успехи математических наук , т. 40, вып. 3(243),с. 85–116, http://mi.mathnet.ru/umn2648 . Английский перевод: Uspenskii V.A.,Luzin’s contribution to the descriptive theory of sets and functions: concepts,problems, predictions,
Russian Mathematical Surveys , vol. 40, no. 3 (1985), 97–134, https://istina.msu.ru/publications/article/92645592/ (есть полный текст)[30] А. Л. Семенов, В. А. Успенский, Математическая логика в вычислительных наукахи вычислительной практике,
Вестник Академии наук СССР , (7), 93–103 (1986)2831] Что такое нестандартный анализ?
М.:Наука, главная редакция физико- матема-тической литературы, 1987. 128 c.[32] В. А. Успенский, А. Л. Семёнов,
Теория алгоритмов: основные открытия и прило-жения , М.: Наука, главная редакция физико-математической литературы, 1987.(Серия «Библиотечка программиста», выпуск 49.) 288 с. Английский перевод:Vladimir Uspensky, Alexei Semenov,
Algorithms: Main Ideas and Applications , KluwerAcademic Publishers, 1993, https://doi.org/10.1007/978-94-015-8232-2 [33] Колмогоров А. Н., Успенский В. А., Алгоритмы и случайность,
Теория вероят-ностей и её применения , том XXXII, выпуск 3, 1987 год, июль, август, сентябрь,с. 425–455, http://mi.mathnet.ru/tvp1437 . Английский перевод: Kolmogorov A.N.,Uspenskii V.A., Algorithms and randomness,
Theory of Probability and Its Applications ,SIAM Publishers, vol. 32, no. 3, 389–412. http://dx.doi.org/10.1137/1132060 , https://istina.msu.ru/publications/article/92647240/ (есть текст)[34] Успенский В.А., Кановей В.Г., Вклад М. Я. Суслина в теорети-ко- множественную математику. Вестник Московского университе-та. Серия 1: Математика, механика , 1988, номер 5, с. 8–12, https://istina.msu.ru/publications/article/93856823/ . Английский пере-вод: Uspenskii V.A., Kanovei V.G., M.Ya. Suslin’s contribution to set- theoreticmathematics,
Moscow University Mathematics Bulletin , vol. 43, no. 5 (1988), 29–40.(Allerton Press, inc.)[35] В. А. Успенский, А. Л. Семёнов, А. Х. Шень, Может ли (индивидуальная) последо-вательность нулей и единиц быть случайной?
Успехи математических наук , т. 45,вып. 1(271), с. 105–162 (1990, январь–февраль), http://mi.mathnet.ru/umn4692 [36] Успенский В.А., Плиско В.Е., Диагностические пропозициональные формулы,
Вестник Московского университета. Серия 1: Математика, механика , 1991, но-мер 3, с. 7–12, https://istina.msu.ru/publications/article/92717778/ [37]
Complexity and Entropy: An Introduction to the Theory of Kolmogorov Complexity ,в книге:
Kolmogorov Complexity and Computational Complexity , Osamu Watanabe,editor. Springer, 1992, ISBN 3-540-55840-3, p. 85–102.[38] Kolmogorov and mathematical logic,
The Journal of Symbolic Logic , volume 57, number2, June 1992, 385–412.[39] Vladimir A. Uspensky and Valery Ye. Plisko, Review: Raymond M. Smullyan, G¨odel’sincompleteness theorems,
The Journal of Symbolic Logic , volume 60, issue 4 (1995),1320–1324, https://projecteuclid.org/euclid.jsl/1183744885 [40] V.A. Uspensky, A. Shen, Relation Between Varieties ofKolmogorov Complexities,
Mathematical Systems Theory , , 271–292 (1996), https://link.springer.com/article/10.1007/BF01201280 , lpcs.math.msu.su/~uspensky/bib/Uspensky_1996_MST_Shen_Relations_between_varieties_of_Kolmogorov_complexities.pdf [41] Kolmogorov complexity: recent research in Moscow. In W.Penczek, A.Szalas(eds.), Proceedings of the 21st International Symposium on MathematicalFoundations of Computer Science 1996 (MFCS96), Crakow, Poland, September –6, 1996 (Lecture Notes in Computer Science, v. 1113), 1996, p. 156–166, https://link.springer.com/chapter/10.1007/3-540-61550-4_145 , http://lpcs.math.msu.su/~uspensky/bib/Uspensky_1996_LNCS_Kolmogorov_Complexity_Recent_trents_Moscow.pdf [42] Mathematical logic in the former Soviet Union: brief history and current trends, Logicand Scientific Methods , M.L. Dalla Chiare et al., editors, Kluver Academic Publishers, , 457–483.[43] Andrei A. Muchnik, Alexei L. Semenov, Vladimir A. Uspensky, Mathematicalmetaphysics of randomness,
Theoretical Computer Science , , 263–317 (1998), (до-ступен полный текст), http://lpcs.math.msu.su/~uspensky/bib/Uspensky_1998_TCS_Muchnik_Semenov_Math_metaphysics_randomness.pdf [44] Что такое аксиоматический метод?
Ижевск: издательский дом «Удмуртскийуниверситет», 2000. 100 с. ISBN 5-7029-0337-4.[45] Why Kolmogorov complexity? In: E.Goles and C.Martinez (eds.),
Complexsystems (Series: Nonlinear Phenomena and Complex Systems, Vol. 6),Kluwer Academic Publishers, 2001, p. 201–260. ISBN 0-7923-6830-4, https://link.springer.com/chapter/10.1007/978-94-010-0920-1_5 [46]
Труды по нематематике с приложением семиотических посланийА. Н. Колмогорова к автору и его друзьям.
М.:ОГИ, 2002. 1409 с., ISBN 5-94282-086-4, [47] B. Durand, V. Kanovei, V.A. Uspensky, N.K. Vereshchagin. Do stronger definitions ofrandomness exist?
Theoretical Computer Science , v. 290, No. 3, p. 1987–1996 (2001), (до-ступен полный pdf-файл)[48] Успенский В.А., Верещагин Н.К., Плиско В.Е.,
Вводный курс математическойлогики , М.: Физматлит, 2004.[49] Кановей В.Г., Успенский В.А., Об эквивалентности двух форм континуум- гипоте-зы,
Вестник Московского университета. Серия 1: Математика, механика , 2005,номер 3, с. 62–64, https://istina.msu.ru/publications/article/100287822/ (есть текст как фото)[50] Четыре алгоритмических лица случайности,
Математическое просвещение , серия3, выпуск 10, М.:МЦНМО, 2006, с. 71–108, http://mi.mathnet.ru/mp188 . Вошло вкачестве приложения в книгу [59].[51] Колмогоров, каким я его помню.
Колмогоров в воспоминаниях учеников , ре-дактор- составитель А. Н. Ширяев, текст подготовлен Н. Г. Химченко, М:МЦНМО,2006, 272–371.[52] Кановей В.Г., Успенский В.А., О единственности нестандартных расширений,
Вестник Московского университета. Серия 1: Математика, механика , 2006, но-мер 5, с. 3–10, https://istina.msu.ru/publications/article/100285758/ (естьскан текста) 3053] В. Г. Кановей, Т. Линтон, В. А. Успенский, Игровой подход к мере Лебега,
Матема-тический сборник , т. 199, номер 11 (2008), 21–44, http://mi.mathnet.ru/msb3948
Английский перевод: V.G.Kanovei, Tom Linton and Vladimir A. Uspensky, Lebesguemeasure and gambling,
Sbornik: Mathematics , volume 199, no. 11, p. 1597, http://dx.doi.org/10.1070/SM2008v199n11ABEH003974 [54]
Апология математики . Санкт-Петербург: Амфора, 2009. 554 с.[55] К истории проблемы Гольдбаха, В кн.:
Историко-математические исследова-ния. Вторая серия . РАН, Институт естествознания и техники им. С. И. Вавилова.Вып. 13(48). М.: Янус-К, 2009, ISBN 978-5-8037-0449-2, с. 273–283.[56]
Простейшие примеры математических доказательств (Библиотека «Математи-ческое просвещение», вып. 34). М.:МЦНМО, 2009. 56 с. ISBN 978-5-94057-492-7. [57] В. А. Успенский, В. В. Вьюгин, Становление алгоритмической теории информациив России,
Информационные процессы , том 10, номер 2, с. 145–158.[58] Теорема Гёделя и четыре дороги, ведущие к ней,
Математическое просвещение ,серия 3, выпуск 15, М.:МЦНМО, 2011, с. 35–75, http://mi.mathnet.ru/mp309 [59] Н. К. Верещагин, В. А. Успенский, А. Шень,
Колмогоровская сложность и ал-горитмическая случайность , М.:МЦНМО, 2013. 575 с. (Английский перевод:A. Shen, V. Uspensky, N. Vereshchagin,
Kolmogorov complexity and algorithmicrandomness , American Mathematical Society, 2017.)[60] [В. А. Успенский, М. С. Гельфанд] Математика— это гуманитарная наука (интер-вью с В. А. Успенским ведёт М. С. Гельфанд), в книге:
Математические прогулки.Сборник интервью , М.:издательство Паулсен, 2017. ISBN 978-5-98797-057-7, с. 198–207.[61] Vladimir Uspenskiy and Alexander Shen, Algorithms and Geometric Constructions,
Computability in Europe, 2018 , Lecture Notes in Computer Science, v. 10936, Springer,p. 410–420 (2018), см. также arXiv:1805:12579 [62]
Труды по не математике. Второе издание, исправленное и дополненное. В пятикнигах. Книга пятая. Воспоминания и наблюдения. М.:Объединённое гуманитар-ное издательство. Фонд «Математические этюды». 2018. 1118 с.[63] Третий математический съезд. В кн.: [62, с. 897–905]. Примечания. Там же, с. 905–912.
Публикации, в подготовке которых участвовал В. А. Успенский [64] Р. Петер,
Рекурсивные функции , перевод с немецкого В. А. Успенского под редакци-ей и с предисловием А. Н. Колмогорова. М.:Издательство иностранной литературы,1954. (Оригинал:
Rekursive Funktionen , von R´osza P´eter, Budapest, 1951.)[65] Стефен К. Клини,
Введение в метаматематику , перевод с английскогоА. С. Есенина-Вольпина под редакцией В. А. Успенского. М.:Издательство ино-странной литературы, 1957. (Оригинал: Stephen Cole Kleene,
Introduction tometamathematics , D. van Nostrand company, New York, Toronto, 1952.)3166] У. Росс Эшби,
Введение в кибернетику , перевод с английского Д. Г. Лахути подредакцией В. А. Успенского с предисловием А. Н. Колмогорова. М.:Издательствоиностранной литературы, 1959. 428 с. (Оригинал:
An introduction to cybernetics , byW. Ross Ashby, London, Chapman&Hall Ltd., 1956.)[67] А. Чёрч,
Введение в математическую логику, I , перевод с английскогоВ. С. Чернявского под редакцией В. А. Успенского, М.:Издательство иностраннойлитературы, 1960. (Оригинал:
Introduction to mathematical logic by Alonzo Church.Volume I. Princeton University Press, 1956.)[68] Н. Бурбаки,
Начала математики. Первая часть. Основные структуры анали-за. Книга первая. Теория множеств . Перевод с французского Г. Н. Поварова иЮ. А. Шихановича под редакцией и с предисловием В. А. Успенского, М.:Мир, 1965.(Оригинал: ´El´ements de math´ematique par N. Bourbaki, XVII, XX, XXII, I. Premierepartie. Les structures fondamentales de l’analyse. Livre I. Th´eorie des ensembles.Hermann, 1956–1960.)[69]
Математика в современном мире , сборник переводов из специального выпуска
Mathematics in the modern world , Scientific American, 1964. Перевод с английскогоН. Г. Рычковой. Под редакцией и с предисловием В. А. Успенского. М.:Мир, 1967.[70] Х. Роджерс,
Теория рекурсивных функций и эффективная вычислимость , пере-вод с английского В. А. Душского, М. И. Кановича, Е. Ю. Ногиной под редакциейВ. А. Успенского. М.:Мир, 1972. (Оригинал: Hartley Rogers, Jr.,
Theory of recursivefunctions and effective computability , McGraw-Hill Book Company, 1967. Предвари-тельная версия с тем же названием. Volume I. Mimeographed. Technology Store,Cambridge, Mass., 1957.)[71] М. Дэвис,
Прикладной нестандартный анализ , перевод с английскогоС. Ф. Сопрунова под редакцией и с предисловием В. А. Успенского, М.:Мир,1980. (Оригинал: Martin Davis,
Applied nonstandard analysis , Wiley&Sons, 1977.)
Другие цитируемые работы [72] Th. Skolem,
Begr¨undung der elementaren Arithmetik durch dir rekurrierende Denkweiseohne Anwendung scheinbarer Ver¨anderlichen mit unendlichem Ausdehnungsbereich (Videnskapsselskapets Scrifter, I. Mat.-naturv. Klasse, 1923, No. 6), Kristiania, 1923.(Английский перевод: The foundations of elementary arithmetic established by meansof the recursive mode of thought without the use of apparent variables ranging overinfinite domains, в сборнике [102, p. 302–333].)[73] David Hilbert, ¨Uber das Unendliche,
Mathematische Annalen , Bd. 95, S. 161–190 (1926).(Aнглийский перевод: On the Infinite, [102, p. 367–392].)[74] Wilhelm Ackermann in G¨ottingen, Zum Hilbertschen Aufbau der reellen Zahlen,
Mathematische Annalen , Bd. 99, 118–133 (1928). (Aнглийский перевод: On Hilbert’sconstruction of the real numbers, [102, p. 493–507].)[75] Kurt G¨odel in Wien, ¨Uber formal unentscheidbare S¨atze der Principia Mathematicaund verwandter Systeme I,
Monatshefte f¨ur Mathematik und Physik , , 173–198 (1931).(Английский перевод: On formally undecidable propositions of Principia Mathematica and related systems I, [102, p. 596–616] или [99, p. 4–38].)3276] J. Herbrand `a Paris, Sur la non-contradiction de l’Arithm´etique,
Journalf¨ur die reine und angewandte Mathematik , Bd. 166, S. 1–8 (1932), (Английский перевод: On the consistency of arithmetic [102, p. 618–628].)[77] Kurt G¨odel,
On undecidable propositions of formal mathematical systems , записки лек-ций в Institute for Advanced Study (Принстон), весна 1934 года. Воспроизведенов [99, p. 39-74][78] R´osza P´eter, ¨Uber den Zusammenhang der verschiedenen Begriffeder rekursiven Funktionen,
Mathematische Annalen , Bd. 110,n. 1, S. 612–632 (1935), https://doi.org/10.1007/BF01448046 , https://link.springer.com/article/10.1007%2FBF01448046 [79] Alonzo Church, An unsolvable problem of elementary number theory, American Journalof Mathematics , vol. 58, no. 2 (April 1936), 345–363. Перепечатано в [99, p. 88–107][80] S.C. Kleene, General recursive functions of natural numbers,
Mathematische Annalen ,Bd. 112, S. 727–742 (1936), https://eudml.org/doc/159849 . Перепечатано в [99,p. 236–253].[81] Emil L. Post. Finite combinatory processes. Formulation I.
The Journal of SymbolicLogic , vol. 1 (1936), p. 103–105.[82] A.M. Turing, On computable numbers, with an application to theEntscheidungsproblem,
Proceedings of the London Mathematical Society , ser. 2,vol. 42 (1936–7), p. 230–265; исправления в следующем томе того же журнала:vol. 43 (1937), p. 544–546. Перепечатано в [99, p. 116–154].[83] S.C. Kleene, On notations for ordinal numbers,
Journal for Symbolic Logic , , 150–155(1938), [84] A.M. Turing, Systems of logic based on ordinals, Proc. London Math. Soc. (2), vol. 45(1939), pp. 161–228, https://doi.org/10.1112/plms/s2-45.1.161 . (Перепечатанов [99, p. 154–222]. Есть текст одноимённой диссертации Тьюринга в Принстоне, )[85] S.C. Kleene, Recursive predicates and quantifiers,
Transactions ofthe American Mathematical Society , , number 1, 41–73 (1943), https://doi.org/10.1090/S0002-9947-1943-0007371-8 . Перепечатано в [99,p.254–287].[86] Emil L. Post, Recursively enumerable sets of positive integers and their decisionproblems, Bulletin of the American Mathematical Society , , 284–316 (1944). https://projecteuclid.org/download/pdf_1/euclid.bams/1183505800 . Перепе-чатано в [99, p.304–337].[87] S.C. Kleene, A symmetric form of G¨odel’s theorem (Presented to theAmerican Mathematical Society, October 29, 1949. Communicated byProf. L.E.J. Brouwer at the meeting of April 29, 1950). KoninklijkeNederlandse Akademie van Wetenschappen, Volume 53, deel 6 (1950), 800–802, Transactions of the American Mathematical Society , vol. 74 (1953), p. 358–366.[89] L´aszlo Kalm´ar in Szeged, Ungarn, ¨Uber ein Problem, betreffend diedefinition des Begriffes der allgemein-rekursiven Funktion,
Zeitschrift f¨urmathematische Logik und Grundlagen der Mathematik , Bd. 1, S. 93–95 (1955), https://onlinelibrary.wiley.com/doi/abs/10.1002/malq.19550010204 [90] John Myhill, A fixed point theorem in recursion theory, abstract, Eighteenth Meetingof the Association of Symbolic Logic,
The Journal of Symbolic Logic , volume 20, no.2(June 1955), p. 205.[91] J. Myhill in Berkeley, California (USA), J.C. Sheperdson in Bristol,England, Effective operations in partial recursive functions,
Zeitschrift f¨urmathematische Logik und Grundlagen der Mathematik , Bd. 1, S. 310–317 (1955), https://onlinelibrary.wiley.com/doi/abs/10.1002/malq.19550010407 [92] H.G. Rice, On completely recursive enumerable classes and their key arrays,
TheJournal of Symbolic Logic , volume 21, number 3, Sept. 1956, p. 304–308. (ReceivedSeptember 14, 1955) [93] Martin Davis,
Computability and Unsolvability , McGraw-Hill Book Company, 1958.[94] Hartley Rogers, Jr. G¨odel numberings of partial recursive functions,
Journal ofSymbolic Logic , Volume 23, Number 3, Sept. 1958, p. 331–341 (Received July 7, 1958), [95] Richard M. Friedberg, Hartley Rogers jr., Reducibility and Completeness for Sets ofIntegers,
Zeitschrift f¨ur mathematische Logik und Grundlagen der Mathematik , Bd. 5,S. 117–125 (1959), https://doi.org/10.1002/malq.19590050703 [96] Kreisel, G., Lacombe, D., Shoenfield, J.R.,
Partial recursive functionals and effectiveoperations , in
Constructivity in Mathematics , A. Heyting, editior, North Holland, 1959,p. 195–207.[97] John Myhill, Note on degrees of partial functions,
Proceedingsof the American Mathematical Society , (1961), p. 519–521, https://doi.org/10.1090/S0002-9939-1961-0125794-X [98] Г. С. Цейтин, Алгорифмические операторы в конструктивных метрических про-странствах, Труды математического института имени В. А. Стеклова, LVII,Проблемы конструктивного направления в математике, 2 (Конструктивный ма-тематический анализ) , сборник работ под редакцией Н. А. Шанина, ИздательствоАкадемии наук СССР, Москва, Ленинград, 1962, с. 295–361.[99]
Basic Papers On Undecidable Propositions, Unsolvable Problems and ComputableFunctions , сборник работ, составитель Martin Davis, Raven Press, Hewlett, New York,1965.[100] А. И. Мальцев,
Алгоритмы и рекурсивные функции , М.:Наука, 1965. (2-е издание,1986) 34101] Per Martin-L¨of, The definition of random sequences,
Informationand Control , volume 9, issue 6, December 1966, p. 602–619, https://doi.org/10.1016/S0019-9958(66)80018-9 [102]
From Frege to G¨odel. A Source Book in Mathematical Logic, 1879–1931 , сбор-ник работ, составитель Jean van Heijenoort, Harvard University Press, Cambridge,Massachusetts, 1967.[103] John Case, Enumeration reducibility and partial degrees,
Annals ofmathematical logic , vol. 2, no. 4 (1971), 419–439. (Received 9 September 1969), [104] A. Шень, Аксиоматический подход к теории алгоритмов и относитель-ная вычислимость,
Вестник Московского университета. Серия 1: Мате-матика, механика. https://hal-lirmm.ccsd.cnrs.fr/lirmm-01923123 .)[105] Stephen C. Kleene, The theory of recursive functions, approaching its centennial.(Elementarrekursiontheorie vom h¨oheren Standpunkte aus.)
Bulletin of the AmericanMathematical Society , volume 4, number 1, July 1981, 43–61.[106] Nigel Cutland,
An introduction to recursive function theory , Cambridge UniversityPress, 1980. Русский перевод Ал. А. Мучника под редакцией С. Ю. Маслова:Н. Катленд,
Вычислимость. Введение в теорию рекурсивных функций.
М.:Мир,1983.[107] А. Х. Шень, Алгоритмические варианты понятия энтропии,
Доклады Академиинаук , 1984, том 276, номер 3, с. 563–566. (Английский перевод: Soviet Math. Doklady, (3), 1984, 569–573.)[108] Piergiorgio Odifreddi, Classical Recursion Theory. The Theory of Functions and Setsof Natural Numbers (Studies in logic and the foundations of mathematics, volume 125),Elsevier, 1989. xix+668 pages.[109] Robert I. Soare, Computability and Recursion,
Bulletinof Symbolic Logic , (3), 284–321 (1996). См. также [110] V.V. Vyugin, Ergodic theorems for individual random sequences, TheoreticalComputer Science , volume 207, issue 2, November 6, 1998, p. 343–361, https://doi.org/10.1016/S0304-3975(98)00072-3 .[111] Mariya I. Soskova, The Turing Universe in the Context of EnumerationReducibility. In: Bonizzoni P., Brattka V., L¨owe B. (eds),
The Natureof Computation. Logic, Algorithms, Applications. CiE 2013 . Lecture Notesin Computer Science, vol 7921. Springer, Berlin, Heidelberg, p. 371–382, https://doi.org/10.1007/978-3-642-39053-1_44 [112] Robert I. Soare,
Turing Computability. Theory and Applications . Springer, 2016, ISBN978-3-642-31932-7, https://doi.org/10.1007/978-3-642-31933-4
Primitive recursive function , https://en.wikipedia.org/wiki/Primitive_recursive_function и µ -recursivefunction , https://en.wikipedia.org/wiki/%CE%9C-recursive_function . Версия5 ноября 2018 года.[114] Szudzik, Matthew. Recursive Function . From MathWorld —A Wolfram Web Resource, created by Eric W. Weisstein. http://mathworld.wolfram.com/RecursiveFunction.htmlhttp://mathworld.wolfram.com/RecursiveFunction.html