General recursive functions of natural numbers’). Von S.C. Kleene in Madison (Wis., U.S.A.). The substitution 1) (hay 25 Gn) = BCL (ys oe 0s ads + +s Xm (rs ++ +9 Be))s and the ordinary recursion with respect to one variable Q) (0, 2y, «5 Tn) = Y (Ly, --) Zn) PUY HU, yy 0005 Ga) = K(Ys 9 (Ys Bas + +s La)s as ++ Be)s where 6,4,,---,%m» Y»% are given functions of natural numbers, are examples of the definition of a function y by equations which provide a step by step process for computing the value y(k,,...,%,) for any given set &,,...,k, of natural numbers. It is known that there are other definitions of this sort, e. g. certain recursions with respect to two or more variables simultaneously, which cannot be reduced to a succession of substitutions and ordinary recursions®). Hence, a characterization of the notion of recursive definition in general, which would include all these cases, is desirable. A definition of general recursive function of natural numbers was suggested by Herbrand to Gédel, and was used by Gédel with an important modification in a series of lectures at Princeton in 1934. In this paper we offer several observations on general recursive functions, using essentially Gdédel’s form of the definition. The definition will be stated in §1. It consists in specifying the form of the equations and the nature of the steps admissible in the computation of the values, and in requiring that for each given set of arguments the computation yield a unique number as value. The ope- rations on symbols which occur in the computation have a similarity to ordinary recursive operations on numbers. This similarity will be utilized, by the Gédel method of representing formulas by numbers, to prove that every (general) recursive function is expressible in the form vleyle(z,----, Zn, y) = 0]) where y and @ are ordinary or ,,primitive“ 1) Presented to the American Mathematical Society, ‘September 1935. 2) W. Ackermann, Zum Hilbertschen Aufbau der reellen Zahlen, Math. An- nalen 99 (1928), §.118—133; Rézsa Péter, Konstruktion nichtrekursiver Funk- tionen, Math. Annalen 111 (1935), S. 42—60.728 S.C. Kleene. recursive functions and (x,,..., %) (Ey) [0 (%,,---) Zn, ¥) = O]%). Also, it is seen directly that, for any recursive function @ (2,,..., 2n,Y) £Y [0 (%,---> Bn» y) = 0} is a recursive function, provided (z,,-.., tn) (Ey) Le (13+. as y) = O} In §2, the problem is raised, which systems of equations define recursive functions under the general definition. The systems which do cannot be recursively enumerated, if by a recursive enumeration is under- stood one such that the numbers ordered by the Gédel method to the systems of equations in the enumeration are a recursive sequence (i. e. the successive values of a recursive function of one variable), since from any recursive sequence of such numbers we can obtain the recursive definition of a new function by the familiar process of diagonalizing and adding 1. For the same reason, a recursive process of deciding which systems define recursive functions is unattainable, if by a recursive process is meant one such that there is a recursive function of the corresponding numbers whose value is 0 or 1 according to the result obtained. Since the condition under which a recursive function of x variables is defined can be expressed in the form (2, ..., Z,) (Ey) [o(a,,--+, ns y) = O], we are afforded an approach (somewhat different than Gédel’s‘)) to the existence of undecidable number-theoretic proposi- tions in formal logics satisfying certain general conditions. Roughly speaking, every such formal logic must contain undecidable propositions of the form (2) (Ey) [o (2, y) = 0], where o (x, y) is a primitive recursive function, because otherwise the logic could be used to decide recursively which systems of equations define recursive functions, which we know in advance to be impossible. Every problem of the form, whether or not (x) (Ey) [o (a, y) = 0], where o (a, y) is a recursive function, is in- cluded in the problem, which systems of equations define recursive func- tions of one variable. Also, there are non-recursive functions definable using only one quantifier, thus: +(x) = 0 if (y)[o (a, y) = 0], +(x) = 1 otherwise, where o (x, y) is primitive recursive. 3) In the “functions” which we consider, the arguments are understood to range over the natural numbers (i. e. non-negative integers) and the values to be natural numbers. Also, for abbreviation, we use propositional functions of natural numbers, calling them ,,relations‘ (alternatively “classes”, when there is only one variable) and employing the following notations: (x) A (x) [for all natural numbers, A(z), (Hz)A(z) [there is a natural number z such that A(z), e2[A (z)] [the least natural number x such that A(z), or 0 if there is no such number}, — [not], V for], & [and], > [implies], = [is equivalent to]. 4) Kurt Godel, Uber formal unentscheidbare Satze der Principia Mathematica und verwandter Systeme I, Monatsh. far Math. u. Physik 38 (1931), S.173—198.General recursive functions. 729 ci The relation between primitive and genera! recursive functions. A recursive function (relation) in the sense of Gédel‘) (S. 179—180) will now be called a primitive recursive function (relation), By using S(z%)=2+1 (the successor function), (3) C(t) =0 (the constant function 0), U? (a, 2) = % (identity functions) as initial functions, the definition of primitive recursive function can be phrased thus: Definition 1. A function is primitive recursive if it can be defined from the functions (3) by (zero or more) successive applications of schemas (1) and (2) (m,n = 1, 2,...; 7 =1,..., )5). In the study of general recursive functions, we treat the defining equations formally, as sequences of symbols. For abbreviation, we may omit to distinguish between the functions and numbers, and the symbols or sets of symbols which stand for them. Now consider expressions consisting of finite sequences of the follow- ing symbols: 0 (the numeral 0), S (the suecessor function), ,, w,,... (numerical variables), 0), 0,,.-- (variables for functions of 74,7,,... at- guments, where 7,,7,,... is a sequence of positive integers in which each occurs infinitely many times, say 1, 1, 2,1, 2,3,...), (,),> , = (pa- rentheses, comma, equality symbol). We define term thus: 0, w,, w,,... are terms; if a,,a,,... are terms, S(a,), Qy(@,, ---s 1), 01 (4,3 ++ +» Gr) -+. are terms. By numeral is meant one of the expressions 0, S(0), 8(S(0)),.. If a and 6 are terms (and if o,,...,0, are functional variables*) such that a least one of o,,...,0, occurs in a or b, but no functional variables other than o,,..., 0, occur in a or b), a = 6 will be called an equation (in o,,...,0,). By a system of equations we mean a finite sequence of equations. st seal shall denote the result of pee On substituting 6; for a; (i = 1,...,) throughout A (A itself, if a,,..., a, 5) This form of the definition was introduced by Gédel to avoid the necessity of providing for omissions of arguments on the right in schemas (1) and (2). The operations in the construction of primitive recursive functions can be further restricted. See Rézsa Péter, Uber den Zusammenhang der verschiedenen Begriffe der rekursiven Funktionen, Math. Annalen 110 (1934), S. 612—632. 6) That is, if o,,...,0, stand for e, ,--» @,, for some set of distinct num- bers a,,-+» a, (then we use #, for r,,). Similarly, in R, below it is meant that 2yss+ t stand for wy... wy, for some set of distinct numbers A...» » Bye a730 S. C. Kleene. do not occur in A). E -.,,.... F shall denote that the expression F is derivable from the expressions E by (zero or more) applications of the operations R,,, R, Mp9 0+ We list the ry on expressions’): R,: to replace A by si ae) where &,,...,%, are the numerical rise An variables which occur in A, and k,,..., hy are numerals. R,: to pass from A and o(k,,...,k,) =k to the result of substituting k for a particular occurrence of o(k,,...,k,) in A, where k,,...,h,, k are numerals. R,: to pass from A and B=C to the result of substituting C for a particular occurrence of B in A. The Herbrand-Gédel definition of general recursive function of natural numbers can be formulated thus*): Definition 2a. Given functional variables o,,...,0,, let Ef denote the set of equations oj(k,,...,h,,) =k where k& is the “value” of Gj (ky, «++ ke) as presently defined. The functions o,,...,0, are defined recursively by the system of equations (EZ, :.. E,) if, for each i (i=1, ..., n), EB, is a system of equations in o,,...,0,, each of the form o;(a,,...,4,) = 6 where o; does not occur in a,,...,@,,, such that for each set of numerals k,,...,%,, there is exactly one numeral & (called the value of 0; (k,,--.k.,)) for which Ef, ..., Bly, By bis 0 (bys +++ be) = & A function o, is recursive if there is an (E,...E,) of this description. We understand a function g(a,,..., 2) to be recursive under this definition, if it is possible to define it by recursion equations of the type described, whether or not originally the function is so defined. More explicitly, a given function 9(z,,...,%m) is recursive under Def. 2a, if there exists an (H,...E,) as described in Def.2a in which o, may be regarded as representing y. 0, may be regarded as representing 9, if Sy == m and whenever k,, ..., k,, are the numerals S(...2, times ...S(0)...), + S(..-%m times ...S(0)...), resp., the “value of o, (hy, -.-5 be,)” under Def.2a is the numeral S(... 9 (,,.--s 2m) times... S(0)...). A similar remark applies to Def.2b below. 1) In these operations we do not require that 4 and B = C be equations and that c be a functional variable, since R,—R, as stated when applied to equations generate equations. Thereby, our proof of IV is simplified. 8) In what follows, the word “recursive” (when not qualified by the adjective “primitive”) will mean recursive under any one of the definitions 2a, 2b and 2c, except when the definition involved is mentioned explicitly (as is necossdiy in the course of establishing the theorems VI and IX on their equivalence).General recursive functions. 731 We now show that Def.2a is not more general than the following (which will later be proved equivalent to it): Definition 2b. The functions o,,..., 0, are defined recursively by E, if E is a system of equations in o,, ..., 0, such that for each i (i = 1, ..., n) and each set of numerals k,,...,%,, there is exactly ‘one numeral & (called the value of o, (k,,---,k.,)) for which E 'y,30,(k,,..., k,) = k. A function o, is recursive if there is an E of this description ®). For the system of equations (Z,...E,) of Def.2a can be proved to be a system E for Def. 2b thus: Clearly, (E, ... E,) is a system of equa- tions in o,,...,¢,, and for each i and set of numerals k,,..., fas (B, ... En) bis 0. (ky, «++, he,) = & where & is the value of o, (ky, .-., ks,) under Def. 2a. It remains to be shown that (Z, ...£,) Fy, 5 0; (k,,---, 4.) = for 1 a numeral only when 1 = &. Now each equation of (H,...E,) is verifiable (for each replacement of its numerical variables by numerals) by use of the values under Def. 2a, since, on examination, the supposition of the contrary is found to conflict with the hypothesis that for given i and numerals k,,...,%,, there is only one numeral & such that EX, ..., Et 1, Ey Frye ar (ys oes k,) = &%*). Moreover, R, and R, applied to verifiable equations yield verifiable equations. Hence, if (E,... E,) Fis 0,(k,,-.., 4.) = 1 where k,,...,%,,, 1 are numerals, the values of o,(k,,.-.,%,,) and J must be the same, i.e. 2 must be the value & of o;(k,,...,%,,) under Def. 2a. The set of operations R,,R, may be replaced in Def.2b by a set R; (¢ = 0,1, 2,...) of single-valued binary operations, defined over all pairs of equations as follows: Ws Ry, : to pass from A and B to S 'S(w) 4° R341: t0 pass from A and B to spi). R3.42: to pass from A and B = C to the result of replacing the occurrence of B in A beginning with the i+ 18 symbol by C, if there is such an occurrence; otherwise, to A itself. For, under the conditions of Def.2b, Ebi1,2, .. o:(41, +++ ky) sl (J a numeral) when / is the value of o;(k,,...,%,,) under Def. 2b and only then (as is easily shown). 8) A more general definition would not be obtained by allowing under Ry also the substitution of B for C, since B may be chosen to include b = a when- ever a = b is included. 9a) Similarly, the equations of the aystem # of Def. 2b are verifiable by use of the values under Def. 2b, if they are of the form o(a,,.-.,a,) = b.732 S. C. Kleene. We now assign numbers to symbols, expressions, finite sequences of expressions, etc., by the Gédel method [loc. cit.*) 8. 179—182], letting numbers correspond to symbols thus: Ogee Sr) Seer Op (ead yer 1S: $e? os Bitty (OR +6 Bits where p; denotes the i» prime number. Then if the numbers correspon- ding to N,,...,Nz are ”,,...,%, resp., the number corresponding to the sequence N,,..., N, is pi? ... pe*. Employing Gédel’s notations (in- cluding the use of italics to indicate the correspondent for numbers of a given notion relating to expressions) and his methods of exhibiting the primitive recursiveness of functions and relations®»), we adopt 1—10 of his list, modifying 6, and define further primitive recursive functions and relations, as follows: 6. nGla=eyly member. The significances ascribed to I(x), x+y, etc., refer only to the case m,,-.., %, > 0”), Me rtysezeSrkesyty. Ike >ye¢+yar-y,ifersyc~y=0. 2 [F]=e2kSe&e+y> a). 2 13. Rem (a,y) = «=~ ((F)) y- 14. Dy (0) = 1, Dy (+1) = ez [2 < 3 & {[1GI Dy () < 241 Dy (hy &2 = QL Olds th) + 132 GtDy HY V [1GI Dy (k) > 241 Dy (k) > 0 &z = 21 I Dy (&) Bl2 Gt Dy (H] + 1] V [2G1 Dy (k) = 0 & z = BU dy a+ i} 90) Also see Th. Skolem, Begrindung der elementaren Arithmetik durch die rekurrierende Denkweise ohne Anwendung scheinbarer Verdnderlichen mit unend- lichem Ausdehnungsbereich, Videnskapsselskapets Skrifter 1923. I. Mat.-naturv. KL, Nr. 6, 8. 1—38. 10) Note that 1(1) = 0 and z*1 = lex =z [I(z) > 0}.General recursive functions. 133, Dy (k) represents the & + 1* pair of numbers in the following order: 00; 01, 11, 10; 02, 12, 22, 21, 20;.... 15. v Occ x = (En) [0 I(p) + nbz = z}}]- St(z, n, a, y) is the result of substituting the expression y for the oceurrence of the expression a in the expression x beginning with the n+ 1 symbol, if there is such an occurrence; otherwise, x itself. - 20. Ri (i, 2, 9) = S(2, Pri +7), R(3)¢E (Pr (i+7))) Ry, & y) = S(u, Pri +7), B(\)). Baie y) = St (ai ep[p Sy & (Egg Syky = p+ RO)*a], eqla S y& (Ep)p S y&y = p+ RO) +4). Mathematische Annalen, 112. 48734 S. C. Kleene. Ry (1,2, y), By (i, 2, y), Re (i, 2, y) correspond to the operations R;;, Boi+1, Bes+a, resp. 21. R’(n, 2, y) is Bs ([F a a, y)4 RY (t=) 2,y) t RPS} a, y) & [fn [B@2= — Bs ([F $b on | Vv [n+2|3&2 = Ri ("3") y)] v [> 2/2 = ar([52}. «)]]]- R’(n, x,y) corresponds to the operation R,. 22. Z(0) = R(I), Z(n+ 1) = R(3)+E (Z(n)). Z(n) corresponds to the numeral S(... times ... S(0)). 23. Eval, (n, y, &,,---)%») = (Ex) {e Sy &y = R([Pr(n+7)}P)+ FE (Z(a,) «R(7)*...*R(1)+Z(z,))*R(5)*Z(2)} (for a fixed number 7). y corresponds to an expression of the form 0@,(2,,..., 2) = z, where x is a numeral. 24. Val(y) = ex{e < y&(Em)[m < y&y = m+Z(z)}}. If y corresponds to an expression of the form a = « where x is a numeral, then Val (y) = 2. Supposing the function p(n,z,y) given, we define a series of functions as follows: (0, 2, y) = x, y(n +12, ¥) = p(n, «, y). 4(0, 2) = 1 (2), A+ 1, z) = [+ 1)-2(k, 2). 7 (0,2) =z, Ret o21 tk+l2= H [Pr(n + 1} exp \o([zaeab {[L@lDy (Rem(n, n=0 4. (k, 8))] + 1] le (k, 2), [[267 Dy (Rem (n, 2 (k, 2)°))] +1]GIx(k, 2). B(n,2) = et[t < nbBan< s 4(4, 2) i=0 (0, 2) y(n,z)=[ XY Ali, 2] ~—n. i=o 6(z,m) = v(m, 2)Glt(u(m, 2), 2). Then if 2 or 7(0,2) is the Gédel number for the sequence S, of the A(0, z) numbers 2,, ..., % (2, +++ % > 0), 7(4 + 1, 2) is the Gédel numberGeneral recursive functions. 735, for the sequence S, +; of the A(k + 1, z) numbers y(n, 2, y), forn = 0,..., k and 2 and y ranging over S,, in a certain order. Since y(0, x, y) = 2, S, includes all numbers in S$; forO 0, u(n, 2) and y(n,z) as n= 0,1,2,... take successively the pairs of values 04(0,2),04(0, z) —1,...,01; 14 (1,2), 14 (1,2)—1,...,11;.... Hence 6(z, m) for m = 0,1,2... are the members of S,(k = 0,1,2,...). But these are (with repetitions) the numbers obtainable from z,, ..., 2 by zero or more applications of the operations (0, z,y), p(1, 2%, y),-... Since 6(z, m) was defined in a manner which shows that it can be obtained from g(n, x,y) and known primitive recursive functions by substitutions and primitive recursions, we have proved: I. Given a function y(n, x,y), there is a function 6(2, m), primitive recursive in p(n, 2, y)™), such that, whenever z = pi! ... pi! (2,, ... % > 0), then @(z, 0), 6(z,1),... t8 an enumeration (with repetitions) of the least class C(x) such that C(2,), .... C (2) and (n, 2, y) [C(2) & C(y) > C(p(n,z, y)))- We note here the following two theorems for later use: II. Given a class A(x), a relation x,yBz, and a number k which belongs to the least class C (x) such that (x) [A (x) > C(x)] and (a, y,z)[C(2) &Cly)& x, yBz+C (2), there is a function 4(m), primitive recursive in A(x) and 2,yBz, such that 4(0),n(1),... ts am enumeration (with repetitions) of C (a). n(m) is the function 6(R(k), m) when 6(z,m) is chosen as in I taking for p(n, 2, y) the function az < nthe l{nral(a [a] * «= [§)v(a(R)e-=a)flv [n+ eel(s vB ]e+=[5") Voie = fi If a member & of a class R(x) is given, the class is enumerated (allowing repetitions) by the function ey[y < m+k& {(R(m)&y =m) V (R(m) &y = 2)}], which is primitive recursive in the class. Similarly: ps 11) We call a function g primitive recursive in other functions y,, if p becomes primitive recursive under the supposition that y, are primitive recursive. wy) yd, 9) res 4 (8 3) and 4(% 3) are primitive recursive in y(z.9) and (z,3,n). Here we use z,9,3 as abbroviations for 2,5 2.4) 245 Ys oor Yo Fo oo Sp TOD and we shall continue to do so when convenient. i 18) If k= 0, replace “4(0,z) = T(z)” by “2(0, 2) = 1” in the definition of 4 (z, m). . . 48+736 S. C. Kleene. III. Given a relation R(x, y) and a number k such that (Ey) R(k, y), there is a function y(m), primitive recursive in R(z,y), such that 7 (0), 7 (1), ... ts an enumeration (allowing repetitions) of the class (Ey) R(z, y). y(m) = eyly S [1Glm] + & & {(R(1Glm, 261m) & y = 161m) V (RU Glm, Elm) & y = &}). By applying I, taking for gy(n,z,y) the function R'(n, 2, y) (21), we obtain a primitive recursive function: 25. H(z, m). If z corresponds to a system of equations Z, H(z, 0), H(z, 1),... is an enumeration (with repetitions) of the numbers corresponding to equations Y such that Zbi,1,0,... Y. Now let ¢ (x) be a recursive function in the sense of Def. 2a or Def. 2b. Then there is a system E of equations defining y recursively under Def. 2b; suppose that 9, stands for » in E. The system FE has a Gédel number e. Using 23 and 25, if R(x, y) = Eval,, (a, H(e, y),2), then, by Def. 2b, (z) (Ey) R(z, y). Furthermore, using 24, if p(y) = Val(H(e, y)), then (zt) = p(ey[R(z, y)]). We have now proved: IV. Every function recursive in the sense of Def. 2a (or Def.2b) is expressible in the form p(ey[R(z, y)]), where p(y) is a primitive recursive function and R(x, y) a primitive recursive relation and (x) (Ey) R(s, y). Thus the extension of general over primitive recursive functions consists only in that to substitutions and primitive recursions is added the operation of seeking indefinitely through the series of natural numbers for one satisfying a primitive recursive relation. By Gédel 8.180 IV, ey[R(x,y)] is primitive recursive in R(z,y) and any function z(z) which bounds y. Hence, in a certain sense, the length of the computation algorithm of a recursive function which is not also primitive recursive grows faster with the arguments than the value of any primitive recursive function). Given a relation R(z), the function 9(z) which is 0 or 1, according as R(x) holds or not, may be called the representing function of R(z). As with primitive recursions, we say that R(x) is recursive, if its representing function is recursive (under Def. 2a)"). 18) Besides the method, for demonstrating that a function is not primitive recursive (or not definable by given additional means, such as recursions with respect to n variables simultaneously), which consists in finding a lower bound for the values, we have the method, for demonstrating relationships of the opposite kind, which consists in finding an upper bound for the number of steps in the computation algorithm. 14) This is equivalent to saying that there is a recursive function 9'(z) such that R(x) ~ [o’(x) = 0}, since then 9 (x) = I= (1 = 0! (2)).General recursive functions. 737 Let R(z,y) be a recursive relation such that (x) (Ey) R(x,y). Then | the function 2 (x,y) = 4 © (z,7), where g(x, y) is the representing function i=0 of R(z,y), is recursive (under Def.2a); and the function u(x) = ey[R(x,y)] satisfies the following relations in terms of x (x,y): oOLy=y (4) o(S(2),3,y) = o(x(z,S(y)),3,8(y)), (8) = o(2(z, 0), z, 0). These equations (supplemented by the equations defining 2 (x, y) recursively under Def. 2a) form a system E defining yu (zx) recursively under Def. 2a. Hence: V. If R(z,y) ts a recursive relation, and (x)(Ey)R(z,y), then ey[R(z,y)] ts recursive (under Def. 2a). This shows that the converse of IV is true, and gives us as an operation of recursive definition the formation of ey[R(z,y)] from a recursive relation R (x, y) such that (x) (Ey) R(z,y)*). Also, the equivalence of Def. 2a and Def. 2b is now established: VI. The class of recursive functions under Def. 2b is identical with that under Def. 2a. For, as noted earlier, Def. 2b is not less general than Def. 2a, and now we have by IV and V that any function recursive under Def. 2b is expressible in the form y(u(z)) where p(y) and s(x) are recursive under Def. 2a. “ VII. Let R(z,y) be a@ relation such that for every x R(x,y) holds for infinitely many y’s, and let v(x,n) denote the n* y such that R(x, y) in order of magnitude. If R(x,y) is recursive, then v(z,n) is recursive. For »(z,n) satisfies the relations v(z,0) = ¢ y[R(z, y)] and »(z,S(n)) = &(z,r(z,n)) where &(z,z) = ey[R(z,y) &y > 2, from which its recursiveness follows by use of V. The converse of VII holds, since R (z,y) = (En)[n < y & v(z,n) = y], which is primitive recursive in » (x, 2). 18) By IV, the use of this operation repeatedly and with A(z, y) a general recursive relation gives no extension of the class of functions obtainable by a single application of it with R(x, y) primitive recursive. We had already as an operation of recursive definition the formation of ey(B(x,y)] from a recursive relation R(z,y) such that there is a recursive function z(z) for which R(z, y) > y = x (2) (by Godel, 8.180, IV). This and the present result correspond to different methods of expressing ey[R tosh] recursively in terms of 9 (z, y).738 8. C. Kleene. Thus, omitting the parameters x, an infinite class is recursively enumerable without repetitions in order of magnitude if and only if it is recursive. VIII. If the function €(ax) is recursive and takes infinitely many values, and n(n) denotes the n‘* in order of first occurrence in £ (0), f (1), .--; then n(n) is recursive. For y(n) = (»(n)) when »(n) is chosen by VII for RYy) = (le (Em) [0 (m) = J}. For, given any recursive function 6 (m) such that (m, x) (Ey) T,(6(m), z,y), then a fortiori (x) (Ey) T,(6(x),2,y), and by X there is a number / such that (2) (Ey) 7, (f,2,y) but (Em) [0 (m) = fl. XII. The class (x) (Ey) T,(z,2,y) of the numbers z which define functions p(x) recursively is not recursive. For if it were recursive, it would be enumerated by a recursive function, contradicting XI. Indeed, given any recursive class R(z) such that (z){R(z) > (z)(Ey) T, (2, 2, y)}, a number f such that (x) (Ey) 7, (f,2,y) but R(f) is obtained by X, when 0 (2) = ey [(R (2) &y = 2) V (R(e)&y =H), where & is any number such that (2) (Ey) T, (k, z, y). 17) Then 9 (x,,...,2n) is that one of the values x determined by H for which the Gdel number of pc, (21, ++» Zn) = z occurs earliest in the list H(e,0), H(¢,1),-..- 18) Since the means given for passing from definitions under Def. 2a (2b) to definitions under Def. 2c, and vice versa, ate effective, the problem which we now study (which numbers define functions recursively) is equivalent to the one first proposed (which systems of equations define functions recursively).740 S. C. Kleene. The definability of a non-recursive class by use of quantifiers applied to a recursive relation gives the existence of undecidable number-theoretic propositions in certain formal logics from the consideration (somewhat different from that employed by Gédel) that otherwise the logics could be used to construct recursive definitions of the class. XIII. Given a formal logic 8, swppose that the propositions (x) (Ey) T,(z,z,y) (z = 0,1,2,...) can be expressed in S by formulas A,, and that numbers can be assigned to the formulas of S, in such a fashion that (1) to distinct formulas are assigned distinct numbers, (2) the class A(x) of the numbers assigned to axioms, and the relation x,y Bz between numbers of being assigned to formulas in the relation of immediate consequence, are recursive, (3) z is a recursive function B(a,) of the number a, of A,, (4) the class C (n) of the numbers a, is recursive, (5) if A, is provable, then (x) (Ey) T, (z,2,y) is true. Then there are 2’s for which A, is not provable although (@) (Ey) Ty (2, 2) is true™), For suppose that there is a number & such that A, is provable. Then, by (2) and II, given a,, there is a recursive function H (m) which enumerates the numbers assigned to provable formulas of S; and, by (1), (3) and (4), the recursive function 6(y) = 6 (em [{C(H (y)) &m = H(y)} v {C(A(y)) &m = a,\]) enumerates the 2’s for which A, is provable. By (5), (2) (Ey) T, (6 (2), 2, y). Hence, by X, there is a number f such that (#q) [6(q) = f] (which implies that A, is not provable in S) and (x) (Ey) 7, (f, 2, y)*). 19) The relation of “immediate consequence” we suppose to be a given reiation between a formula and a pair of formulas, and the class of “provable formulas” to be the least class which contains the given class of “axioms” and has the property that Z is provable whenever X and Y are provable and Z is an immediate consequence of X and Y. If more details of the structure of § were suitably specified, condition (5) could be given a more metamathomatical appearance, such as the following {analogous to Gédel’s condition of «-Widerspruchsfreiheit, S. 187): for no relation F (z,y) and natural number & are all of the formulas F (k,0), F(k,1), ..-. (22) (y) FU) provable. On the further sssumption that for no relation F(z, y) and sequence of natural numbers ky,ky,... are all of the formulas F (0, ko), F(1,&), «++, () Hy) FO») provable, the conclusion could be given the form, that there are z's for which 4, is formally undecidable, i.e. for which neither A, nor A, is provable. (The conditions need to be assumed merely for certain relations F (2, y).) %) The undecidable proposition 4, can be effectively constructed for a given logic, whenever the number a,,, recursive definitions of 4 (x); z,y Bz; B(y) and O(n), and effective means of constructing A, from f, are given. Whenever the supposition in this proof, that there is a & such that A, is provable, is not realized, the theorem holds trivially.General recursive functions. 741 XIV. The function ¢y[T, (zx, x, y)] 1s non-recursive™). For, if o(z) is any recursive function, the function 7(z) = Val (H(z, @(2))) +1 is recursive, and (z) = Val (H(f,ey[T, (7,2,9)])) holds for any number / defining 7(z) recursively. Now if o(f) = ey[T,(,f.y)), two different values are obtained for 7(/). Hence o(f) + ey(Ti (hy) Thus ey[T,(x, x, y)] differs from each recursive function for some value of x. Note also that (Fy) T,(/,f,y). XV. The class (Ey) T, (x, x, y) is non-recursive. Thus non-recursive functions can be defined by the schema [ 0 if (Ey) R(z,y) (1 if Ey) Rey)’ where R(z,y) is primitive recursive. This follows from XIV, since, if (By) Riky) and 2(@) = (1+ 7(@)]-e+r(@)-k, then ey[R(z,y)] = [l+1(2)]-ey[R(A(z),y)], which is recursive if +(x) is recursive. To analyze the situation more fully, let S(x) be any recursive class such that (x) {S(x) > (Ey) T,(x,x,y)}, and o(z) the representing func- tion of S(x). If k is any number which defines a function recursively, then (Ey) 7,(k,k,y), and we set (2) = (1+ o(z)]-2-+o0(z)-k and @ (2) = [1+ a(2)]-ey[T, (u (2), 2(a), y)]. @ (x) is recursive, and as in the proof of XIV, there is an f such that 0 (f) + ey[T,(/,f,y)] and (Ey) 7, (4,f,y)- If S(/), then o(f) = 0 and o(f) = ey[T,(f,f,y)} Hence S(f). XVI. The class (Ey) T, (x, x, y) is not recursively enumerable**). For by III, the complementary class (Fy) T, (x, z, y) is enumerated by a recursive function y(m). Now if (Ey) T,(c, 2, y) is enumerated by ~(m) and we set £(m) = on {[m|2en = “(3D Vv |m + 1|2&n =+((23)}: we have (Ey) 2, (x, 2,y) =e m[&(m) = 2]|2, which would contradict XV if x(m) were recursive. XVII. Given a recursive relation R(x,y), there is a number e such that (x) (Ey) R(z,9) = (2) (Ey) T,(e,z,y). Given a recursive relation R(y), there is a number e such that (En) R(y) = (Ey) T,(e,e, y)*). t (x)= 21) We recall that ey [R(z,y)] = 0 when (Hy) R(z, y)- 2).The proof given here is non-constructive. The writer has a constructive proof that for certain recursive relations R(z,y) the class (Ey) R(z,y) is not recursively enumerable. From that proof, the existence in certain formal logics of undecidable propositions involving only one quantifier (which can be concluded non-constructively from present results) is obtainable in the same manner as XIII. 23) From the great generality of the problems, which e’s define recursively functions of one variable, and which ¢’s “determine recursively” the e'* value of a function of one variable, as displayed by this theorem, the result, that they are not “effectively” soluble, could have been anticipated.742 S. C. Kleene, General recursive functions. For, to every proposition of the form (z)(#y)R(z,y), there is an equivalent proposition of the form (x) (Hy) R(zx,y) obtained by utilizing the recursive enumerability of n-tuples of natural numbers, or introducing fictive variables*); and the Gédel number e of the system E of equations which defines ey[R(zx,y)] in the proof of V on the supposition that (x) (Ey) R(a,y) satisfies the present theorem. Similarly, (Ey) R(y) has an equivalent (Ey) R(y), and for e we may take the Gédel number of the equations defining ey[R(y) & z = 2] on the supposition that (Ey) R(y)*). My thanks are due to Prof. Paul Bernays for the suggestion of improvements in the presentation. 24) EL ge (ay tay 29) B (2, ta, ts) = (2) (By) [RB (1GLz, 2G1 x, 3GL2)& y =). 2) XV, XVI, and XVII are similar, respectively, to results obtained in a different connection by Prof. Alonzo Church (An unsolvable problem of elementary number theory, see Bull. Amer. Math. Soc. Abstract 41—5— 205), Dr. J. B. Rosser (unpublished), and the present writer (A theory of positive integers in formal logic, Part II, Amer. Jour. Math. 57 No. 2, pp. 230 ff.). (Eingegangen am 7. 7. 1935.)