Truth and Provability

RAYMOND M. SMULLYAN

s is well known, in the rst quarter of the twentieth century, there were two mathematical systems, Principia Mathematica, and Zermelo-Fraenkel Set Theory, which were so powerful that it was generally assumed that all mathematical questions could be decided that is, either proved or disprovedin each of the systems. del startled the However, in 1933, the logician Kurt Go mathematical world by proving that this was not the case that in each of these systems, as well as in a variety of related systems, there must be sentences that, though true, could not be proved within the systems. This celebrated result is dels Incompleteness Theorem (more speciknown as Go dels First Incompleteness Theorem). Closely cally as Go related to this result is a theorem of the logician Alfred Tarski, roughly to the effect that in these systems truth of sentences of the system is not denable in the system. Tarskis theorem dels theorem, which in many ways is provides a proof of Go dels original proof. simpler than Go The purpose of this article is to provide the general reader, even those readers with no familiarity with the symbolism of mathematical logic, with the essential ideas behind the proofs del and Tarski theorems. We do this by constructof the Go ing a very simple system (an abstraction of part of Reference [1]), which, despite its simplicity, has enough power for the del arguments to go through. First we address Tarski and Go dels. Finally I present some Tarskis theorem, and then Go related results of my own. All proofs are presented as problems to be solved by the reader. Solutions are provided at the end of the article.

that X belongs to the set named by H, and we accordingly dene the sentence to be true if X really does belong to the set named by H. We call two sentences equivalent if they are either both true or both false (not true). In what follows, it is to be understood that for any predicate H, there is no expression X such that HX is a predicate. Two symbols N and R play a special role, according to the following two rules: RULE RFor any predicate H, the expression RH is also a predicate, and for any predicate K, the sentence RHK is equivalent to HKK. (I use the letter R to suggest the word repeat, because the expression XX is known as the repeat of X.) RULE NFor any predicate H, the expression NH is also a predicate, and for any sentence X, the sentence NHX is true if and only if HX is not true. (I use the letter N to suggest the word not.) From just these two rules follows something surprising. (At least I hope you will be surprised!)

T H E O R E M T (after Tarski). The set of true sentences is

not nameable. It will be helpful in the proof of theorem T to rst prove another result, which is of considerable interest in its own right. A xed point of a predicate H means a sentence X such that X is equivalent to HX. (Fixed points are closely related to self-reference. A xed point X of a predicate H can be thought of as a sentence that asserts its own membership in the set named by Hand this indeed is true if and only if it does belong to the set named by H.) From just Rule R alone follows:

R E T A R S K I . We consider a collection of elements called

symbols, and expression means any nite sequence of these symbols. For any expressions X and Y, by XY is meant the expression X followed by the expression Yfor example, if X is the expression ad and Y is the expression mgr, then XY is the expression admgr. Certain expressions are called predicates, and each predicate is the name of a set of expressions. A set of expressions is called nameable if some predicate names it. By a sentence, we shall mean any expression of the form HX, where H is a predicate and X is any expression whatsoever. We interpret the sentence to mean

T H E O R E M F (Fixed Point Theorem). Every predicate has

a xed point.

P R O B L E M 1. Prove Theorem F by displaying a xed point

of H, using just the symbols H and R. Now let as see how this is related to Theorem T. We will call a predicate H a truth predicate if it names the set of all true sentences. We are to show that no predicate is a truth

2012 Springer Science+Business Media, LLC DOI 10.1007/s00283-012-9328-6

predicate. If H is a truth predicate, then for every sentence X, the sentence HX is true if and only if X is true, which means that every sentence X is a xed point of H! Thus, to show that H is not a truth predicate, it sufces to show that there is at least one sentence that is not a xed point of H. Such a sentence might aptly be called a witness that H is not a truth predicate. Such a witness can in fact be explicitly displayed using the symbols H, N, and R.

P R O B L E M 2. Prove Theorem T by displaying such a witness, using those three symbols.

Note: All the results of this and the next section have applications to standard mathematical systems of the type del, such as Peano Arithmetic investigated by Tarski and Go or Zermelo-Fraenkel set theory. In Peano Arithmetic, what I have called predicates are formulas with one free variable. They are to be thought of as names, not of sets of expressions, but of sets of natural numbers. To each such predicate H, and each natural number n, a sentence is assigned that I will denote as H(n), which is interpreted to mean that n belongs to the set named by H. Actually, the sentence H(n) is the result of substituting the name of n for all free occurrences of the free variable of H. In this system, each expression X is assigned a natural del number of X. The previous number now called the Go results all hold for Peano Arithmetic if, for any predicate H and expression X, we interpret HX to be H(x), where x is the del number of X. As for Rule N, instead of the symbol N, Go which we took for the negation symbol, is the usual symbol *. For any sentence X, the sentence *X of Peano Arithmetic is indeed true if and only if X is not true, and thus Rule N holds. As for Rule R, the symbol R is not one of the symbols of Peano Arithmetic, but for any formula H there is a formula that I have dubbed RH, such that Rule R holds in the del devised an extremely ingenious following sense: Go argument, now known as the diagonal argument, which

rst involved proving the following: For any predicate K with del number k, the sentence K(k) (sometimes called the Go del number, which I will diagonalization of K) has a Go del cleverly showed that for any predicate H, denote as k*. Go there was a predicate H0 (which I denoted RH) such that for del number k, the sentence H0 (k) is any predicate K, with Go equivalent to H(k*), which in the language of this article is that for any predicate H and any predicate K, the sentence RHK is equivalent to HKK, which is Rule R. dels clever diagonal argument is I might remark that Go closely related to, and may well have been inspired by, Cantors famous diagonal argument proving that for any set S, its power set has higher cardinality. Because Rules N and R hold for Peano Arithmetic, so does Tarskis theorem, which for Peano Arithmetic says that there is no predicate H such that for all natural numbers n, the del number of sentence H(n) is true if and only if n is the Go a true sentence.
D E L . We now consider a mathematical sysENTER GO

tem that proves various sentences of the previous section. The system is wholly accurate in the sense that all provable sentences are true, as previously denedthat is, If HX is provable, then X really does belong to the set named by the predicate H, and if NHX is provable, then X does not belong to the set named by H. Unlike the set of true sentences, which we have seen is not nameable, the set of provable sentences IS nameableits name is the symbol P. Therefore, the set of provable sentences and the set of true sentences do not coincide, and because every provable sentence is true, there must be a true sentence X that is not provable in the system. Since its negation NX is not true, then NX is also not provable in the system, and so we have: del). There is a sentence X that is TH E O R E M G (after Go undecidable in the system.

.........................................................................................................................................................
AUTHOR

RAYMOND M. SMULLYAN has lived four different livesas a mathematician, musician, magician, and author

of essays and puzzle books. Born in 1919 in Far Rockaway, New York, he was equally interested in music and science as a child. In high school, he fell in love with mathematics, and he was torn between becoming a mathematician or a concert pianist. He began teaching piano at Roosevelt College in Chicago, but tendonitis in his right arm forced him off the concert stage. He then turned his attention to mathematics, learning mostly on his own, but also taking a few advanced courses at the University of Chicago while supporting himself as a professional magician. Strangely, before he had a college degree, or even a high school diploma, he received an appointment as a mathematics instructor at Dartmouth College on the basis of some papers he had written on mathematical logic. After his time at Dartmouth, the University of Chicago awarded him a Bachelor of Arts degree, based partly on courses he had never taken but had successfully taught. He received his Ph.D. in mathematics from Princeton University in 1959, and subsequently taught at Princeton, New York University, Graduate School, Lehman College, and Graduate Center, and, last, as a distinguished professor at Indiana University. He has published more than 40 research papers on mathematical logic and 26 books on various subjects, six of them on mathematical logic. Two more books will be published this year. Indiana University Bloomington, IN 47405 United States e-mail: [email protected]
THE MATHEMATICAL INTELLIGENCER

Moreover one can explicitly display such a true but not provable sentence, using the symbols P, N, and R.

P R O B L E M 3. Display such a sentence.

Note: The symbol P that I used does not occur in Peano del created a predicate Bew (known Arithmetic. Instead, Go as the provability predicate, and derived from the German word Beweis = proof), such that for any number n, the del number of sentence Bew(n) is true if and only if n is the Go a provable sentence. D O U B L I N G U P . The following are side issues, but I hope these are not without interest.

P R O B L E M 4. We have seen that a xed point of NH is RNHRNH. There is another one, also, of length 6. Can you nd it?
The following is a useful variant of Theorem F:

T H E O R E M F1 For any predicate H and any predicate K,

there is a sentence X that is equivalent to HKX.

P R O B L E M 5. Prove Theorem F1.

Next is my Double Fixed Point Theorem:

T H E O R E M FF. For any predicates H and K, there are sentences X and Y such that X is equivalent to HY, and Y is equivalent to KX. P R O B L E M 6. Prove Theorem FF.
Now let us go back to our construction of a sentence that is true but is not provable in the system.

P R O B L E M 7. Using the same three symbols, P,N,R, it is

possible to construct two sentences X and Y such that one of the two is true but is not provable in the system, but there is no way to tell which one it is! (Hint: Construct them such that X is true if and only if Y is provable, and Y is true if and only if X is not provable.) There are in fact two solutions in which neither X nor Y are of length more than 8. Can you nd them? I must tell you an amusing incident: The previous problem, together with proofs I have presented in one of my puzzle books, that either Tweedledee exists or Tweedledum exists, but there is no way to tell which led the logician Professor Melvin Fitting, a former student of mine, to introduce me at a lecture I presented once by saying: I now introduce Professor Raymond Smullyan, who will prove to you that either he doesnt exist, or that you dont exist; but you wont know which!

K, we see that RHRH is equivalent to HRHRH. Thus RHRH is a xed point of H. 2. In all that follows, it will be convenient to use the abbreviation iff for if and only if.1 Now, no xed point of NH can possibly be a xed point of H, because for any expression X, NHX is obviously not equivalent to HX. As seen in the solution of Problem 1, for any predicate H, the sentence RHRH is a xed point of H. Because NH is also a predicate, it follows that RNHRNH is a xed point of NH, and is thus not a xed point of H, hence is a witness that H is not a truth predicate. 3. Any xed point X of NP is such a sentence, because X is then true iff NPX is true, iff PX is not true, iff X is not provable. Thus X is true iff X is not provable, which means that either X is true and not provable, or is not true but provable. The latter alternative cannot hold, because it is given that all provable sentences are true. Thus X is true but not provable in the system. We know from Problem 2, taking P for H, that RNPRNP is a xed point of NP, hence it is true but not provable. (Incidentally, another solution is NRPNRP, as you can verify.) 4. A xed point of NH, other than RNHRNH, is NRHNRH, which can be seen as follows: To begin, it is obvious that if X and Y are equivalent sentences, then so are their negations NX and NY. Now, for any predicate H and any predicate Y, the sentence RHY is equivalent to HYY, and so we take NRH for Y, and we see that the sentences RHNRH and HNRHNRH are equivalent, hence so are their negations NRHNRH and NHNRHNRH. Thus NRHNRH is a xed point of NH. 5. RHKRH is equivalent to HKRHKRH, so X is equivalent to HKX, for X the sentence RHKRH. 6. One solution is to use Problem 5: RHKRH is equivalent to HKRHKRH. We take X to be RHKRH and Y to be KRHKRH, which is KX. Thus X is equivalent to KY, and Y is not only equivalent to HX, but IS KX! There is also another solution: We have seen that RHKRH is equivalent to HKRHKRH. Interchanging H with K, it is equally true that RKHRK is equivalent to KHRKHRK. We now take Y to be RKHRK and X to be HRKHRK, which is HY. Thus X is equivalent to HY, because X is HY, and Y is equivalent to KX. Let us record these two solutions: 1X = RHKRH, Y = KRHKRH. 2X = HRKHRK, Y = RKHRK. 7. Let us rst note the following: Suppose X and Y are sentences such that: (1) X is true iff Y is provable. (2) Y is true iff X is not provable. Then one of the two sentences, X, Y, must be true but not provable (but there is no way to tell which one it is). Here is why: The sentence Y must be true, because if it were false, then by (2), X would be provable, hence X is then true, and so by

Solutions
1. We know that RHK is equivalent to HKK for any predicate K, and so it is true for the predicate RH, and so, taking RH for
1

This abbreviation is credited to the late mathematician Paul Halmos.

2012 Springer Science+Business Media, LLC

(1), Y would be provable, thus Y would be false and provable, contrary to the given condition that only true sentences are provable. Thus Y must be true. Therefore by (2), X is not provable. Now, X is either true or not true. If X is true, then it is true but not provable. If X is not true, then by (1), Y is not provable, and so Y is then true but not provable. In summary, If X is true, then X is true but not provable. If X is not true, then it is Y that is true but not provable. There is no way to tell which case holds. We thus seek sentences X and Y such that X is equivalent to PY, and Y is equivalent to NPX. We use the double xed

point theoremin the two solutions (Problem 6) we take P for H and NP for K, and we have: 1X = RPNPRNP, Y = NPRPNPRNP (= NPX). 2X = PRNPPRNP (= PY), Y = RNPPRNP.

REFERENCE

[1] Raymond Smullyan, The Lady or the Tiger, Chapter 15, Alfred A. Knopf, New York, 1982. del incompleteness [2] George Boolos, A new proof of the Go theorem, Notices of the A.M.S. 36 (1989), no. 4, 388390.

THE MATHEMATICAL INTELLIGENCER