Undecidability in number theory

Bjorn Poonen

Does the equation x3 +y 3 +z 3 = 29 have a solution in integers? Yes: (3, 1, 1), for instance. How about x3 +y 3 +z 3 = 30? Again yes, although this was not known until 1999: the smallest solution2 is (283059965, 2218888517, 2220422932). And how about x3 + y 3 + z 3 = 33? This is an unsolved problem. Of course, number theory does not end with the study of cubic equations in three variables: one might ask also about x1729 y 1093 z 196884 163xyzt262537412640768000 = 561. D. Hilbert, in the list of 23 problems he published after a famous lecture in 1900, asked his audience to nd a method that would answer all such questions. More precisely, Hilberts tenth problem (hereafter denoted H10) asks for an algorithm that takes as input a multivariable polynomial f (x1 , . . . , xn ) with integer coecients and outputs YES or NO according to whether there exist integers a1 , a2 , . . . , an such that f (a1 , . . . , an ) = 0. In 1970, Yu. Matiyasevich, building on earlier work of M. Davis, H. Putnam, and J. Robinson, showed that no such algorithm exists. The purpose of this article is to discuss some of the concepts in the proof, a few by-products of the proof, and current research on related problems that are still open, such as the analogue for rational number solutions. H10 and the DPRM theorem The notion of algorithm. To make sense of the negative answer to H10, we need a precise notion of algorithm. In 1900 such a notion had not

yet been developed. But in the 1930s, several rigorous models of computation were proposed and were shown to be equivalent; one of these was the Turing machine. The equivalence made believable the Church-Turing thesis, which is the assertion that every purely mechanical procedure can be carried out by a Turing machine.3 Because of this, algorithm is taken to mean Turing machine. An informal description of a Turing machine may be more enlightening than a mathematically precise denition. A Turing machine is equivalent to a nite-length program running on a physical computer, except that the computer has unlimited time and memory and is not subject to physical errors (such as data loss from power outages). The memory is sometimes modelled as an innite tape, initialized to the binary representation of the nonnegative integer input. The computer reads and writes 0s and 1s from and to the memory tape during its operation, and may or may not print characters on a separate output tape, following the rules of its program. It might run forever, or it might halt when some condition specied by the program is satised. Turing machines may accept any objects as input if we x an encoding of these objects as nonnegative integers. For example, a polynomial with integer coecients could be represented by the concatenation of the ASCII codes of the characters in a TEX string for the polynomial. The exact encoding does not matter as long as a Turing machine can convert between the proposed encodings. Diophantine, listable, and computable sets. Davis, Putnam, Robinson, and Matiyasevich deduced the negative answer to H10 from a

1 This survey article has been accepted for publication in the Notices of the Amer. Math. Soc., and is expected to appear in the March 2008 issue. 2Discovered by E. Pine, K. Yarbrough, W. Tarrant, and M. Beck following an approach suggested by N. Elkies. 3Quantum computers might seem at rst not to t this framework. But they can be simulated by classical Turing machines in exponential time, and H10 asks for any algorithm without being fussy about its running time. When one ignores running time, quantum computers are no more powerful than classical ones.

stronger theorem having many more implications. To explain it, we need a few denitions. Denition 1. A set A Z is diophantine if there exists a polynomial p(t, x) Z[t, x1 , . . . , xn ] such that A = {a Z : (x Zn ) p(a, x) = 0}.

One should think of p as dening a family of polynomial equations, depending on a parameter t; then A is the set of values of the parameter for which the resulting equation in the remaining variables x1 , . . . , xn has a solution. Equivalently, if B is the set of solutions to p(t, x) = 0 in Z1+n , then A is the projection of B onto the rst co- Theorem 6 (Turing 1936). The halting problem ordinate. The denition can be extended in an is undecidable; that is, no Turing machine can obvious way to subsets of Zm for m > 1. solve it. Example 2. The subset N := {0, 1, 2, . . . } of Z is diophantine since for a Z, we have

The halting problem. The negative answer to H10 was proved by relating it to undecidability results in logic and computability theory from the 1930s. These undecidability results were proved using diagonalization arguments reminiscent of G. Cantors famous proof of the uncountability of R. One such result concerns the halting problem, which asks for an algorithm that takes as input a computer program p and an integer x, and outputs YES or NO, according to whether program p run on input x eventually halts (instead of entering an innite loop, say).

Sketch of proof. Fix an encoding of programs as nonnegative integers; identify programs with 2 2 a N (x1 , . . . , x4 Z) x1 + + x4 = a. their integer codes. Suppose that there were an algorithm for deciding when program p halts on Denition 3. A set A Z is listable (or recur- input x. Using this we could build a new program sively enumerable) if there is an algorithm that H such that for any x, prints A, i.e., a Turing machine such that A is H halts on input x the set of integers it prints out when left running program x does not halt on input x. forever. Example 4. The set of integers expressible as a sum of three cubes is listable. (Print out x3 + y 3 + z 3 for all |x|, |y|, |z| 10; then print out x3 + y 3 + z 3 for |x|, |y|, |z| 100; and so on.) A similar argument shows that any diophantine subset of Z is listable. Denition 5. A set A Z is computable (or recursive) if there is an algorithm for deciding membership in A, i.e., an algorithm that takes as input an integer a and outputs YES or NO according to whether a A. Any computable set is listable, since given an algorithm for deciding membership in A, one can apply it successively to 0, 1, 1, 2, 2, . . . and print each number for which the membership test returns YES. But it is not obvious that every listable set is computable. An algorithm that prints A does not immediately let one test whether 33 is in A, say: if after running the algorithm for a while the number 33 is not printed, it may be hard to decide whether it will appear later on. In fact, the next section shows that there exists a listable set that is not computable. Taking x = H, we nd a contradiction: H halts on input H if and only if H does not halt on input H. Corollary 7. There exists a listable set that is not computable. Proof. Let A be the set of numbers 2p 3x such that program p halts on input x. By Theorem 6, A cannot be computable. On the other hand, here is a program that prints A: loop over N = 1, 2, . . . , and during iteration N , for each p, x N , run program p on input x for N steps, and print 2p 3x if the program halts within these N steps. The DPRM theorem. We are now ready to state the following remarkable theorem.4 DPRM theorem (Davis, Putnam, Robinson, Matiyasevich 1970). A subset of Z is listable if and only if it is diophantine. To prove their theorem, these four authors essentially built a computer out of diophantine equations! They showed that diophantine equations are rich enough to simulate any computer in the sense that given a computer program, one can

4Historically, the notions of diophantine, listable, and computable and the DPRM theorem were stated for subsets

of N instead of Z. This makes little dierence, however: reductions in both directions are possible because of Example 2 and the equality Z = N (N).

construct a polynomial equation that has an integer solution if and only if the program halts. The proof of the DPRM theorem looks curiously like the construction of a complicated computer program, with high-level routines built out of more elementary ones, except that instead of routines one has diophantine equations everywhere. An improved version of the original proof may be found in Chapters 15 of [Mat93]. A brief history of the DPRM theorem. The DPRM theorem was conjectured in 1949 by Davis, who also carried out the rst reductions towards its proof. In 1961, Davis, Putnam, and Robinson proved its analogue for exponential diox 2 phantine equations over N (such as 2x3y z+x = 2 5x + yz). This meant that it remained to show that exponentiation was diophantine, i.e., that {(a, b, c) N3 : c = ab } was a diophantine set. Earlier, in 1952, Robinson had proved that the diophantineness of exponentiation would follow from the existence of a 2-variable diophantine relation of exponential growth. Finally, in 1970, Matiyasevich used properties of Fibonacci numbers Fn to prove that the relation m = F2n was diophantine; this gave what Robinson needed, and completed the proof of the DPRM theorem. For more history, see the references at the end of this article, including the lm [Csi08] and the website [H10web]. Negative answer to H10. The DPRM theorem easily implies a negative answer to H10, as we now explain. The undecidability of the halting problem gave us a listable set that is not computable. By the DPRM theorem, having this is the same as having a diophantine set that is not computable. By denition, this means that we have a polynomial p(t, x) such that there is no algorithm for deciding for which values a Z the equation p(a, x) = 0 has a solution in integers x1 , . . . , xn . Thus there cannot be an algorithm for deciding the existence of integer solutions to all polynomial equations. Remark. H10 was not the rst problem outside logic and computability theory to be proved undecidable. In 1947 A. A. Markov and E. Post independently found a nitely presented semigroup for which the word problem is undecidable, and in 1955 P. S. Novikov did the same for a nitely presented group. (The word problem for a nitely presented semigroup G with nite set of generators A is the problem of deciding, given two nite sequences of elements of A, whether the product

of the rst sequence equals the product of the second sequence in G.) The word problem for groups had been motivated by topology, and it was not long afterward that fundamental problems in topology itself were found to be undecidable: for instance, Markov in 1958 proved that the problem of deciding whether two nite simplicial complexes are homeomorphic is undecidable. Other fun consequences of DPRM Undecidability for polynomials of xed degree in a xed number of variables. The proof of the previous section shows that there is a pair (n, d) of positive integers such that there is no algorithm for deciding the existence of integer solutions to n-variable polynomial equations of total degree d. In the 1960s, before the DPRM theorem was proved, the fact that it would imply that equations of bounded degree in a bounded number of variables suce to represent all diophantine sets was considered by some as evidence that the theorem could not be true! After 1970, several authors, including Yu. Matiyasevich, J. Robinson, and Z. W. Sun, proved undecidability results for explicit small values of n and d. For instance, it is now known that there is no algorithm for deciding the existence of integer solutions to polynomial equations in 11 variables. In the positive direction, it is known only that there is an algorithm for polynomials in one variable! It is likely that the problem is decidable also for polynomials in two variables, but so far the elaborate machinery developed by arithmetic geometers is too weak to prove even this. As for degree, a trick discovered by T. Skolem in the 1920s shows that any polynomial equation in integers is equivalent to one of degree at most 4 (and the equivalence is constructive): for instance, y 2 = x5 + 7 is solvable if and only if (u x2 )2 + (v u2 )2 + (y 2 xv 7)2 = 0 is. Thus there is no algorithm for equations of degree 4. In the positive direction, there is an algorithm for equations of degree at most 2 in any number of variables. The situation for degree 3 is still unknown. Number of solutions. Theorem 8 (Davis 1972). Let A be a nonempty proper subset of N {0 }. There is no algorithm

that takes as input f (x) Z[x1 , . . . , xn ] and out- of RH as in [DMR76, p. 335] or [Mat93, 6.4]); puts YES or NO according to whether the cardi- then one can use the DPRM theorem to simulate nality of {a Zn : f (a) = 0} belongs to A. the program with a polynomial equation. M. Baker half-jokingly observed that one The proof, which is very short, shows that an almight try to prove RH by showing that the equagorithm for any A as above could be used to give tion has no solutions modulo 17, say! As one an algorithm for H10. might expect, however, things are not so easy: Simple equations whose smallest solution the equation produced by the DPRM theorem will have solutions modulo any xed positive inis huge. teger. Theorem 9. There is a polynomial p(t, x) such H10 over other rings that for any function F : Z N that is computable and dened on all of Z, there exists a Z Even before 1970, researchers began asking such that p(a, x) = 0 has a solution x Zn but Hilberts question for rings other than Z. no solution with max |xi | < F (a). Denition 11. Let R be a commutative ring. Proof. Use the same p as in the proof of the negThen Hilberts tenth problem over R (H10 over ative answer to H10. If there were a computable R) asks for an algorithm that takes as input bound on the size of the smallest solution when a f (x) R[x1 , . . . , xn ] and outputs YES or NO acsolution existed, then one could decide for which cording to whether there exists a Rn such that a Z the equation p(a, x) = 0 was solvable simf (a) = 0. ply by searching up to that bound. This contraTechnically, to make sense of this, we need to dicts the choice of p. x an encoding of elements of R suitable for inPrime-producing polynomials. Before the put into a Turing machine. In cases where this DPRM theorem was proved, Putnam observed is not possible (e.g., if R is uncountable), then that it would imply the following theorem. it is understood that we restrict the possible inTheorem 10. There exists a polynomial puts by requiring that the coecients of f belong F (x1 , . . . , xn ) Z[x1 , . . . , xn ] such that the posi- to some large countable subring R0 of R. For tive integers in its range (as a function Nn Z) instance, if R = C, we might take R0 to be the subeld of algebraic numbers. are exactly the prime numbers. The question of whether H10 over R has a Proof. The natural number version of the DPRM positive answer now depends on the ring R (and theorem gives a polynomial p(t, x) such that for possibly also R0 ). The remainder of this article a N, the equation p(a, x) = 0 is solvable in will focus on rings R that are of interest to numnatural numbers if and only if a is prime. Dene ber theorists. For more information about these F (t, x) := t(1 p(t, x)2 ). It can be positive only problems, see [DLPVG00, Shl07]. when p(t, x) = 0, and in this case, t is prime and F (t, x) = t. Conversely, every prime arises this H10 over rings of algebraic integers. The ring of Gaussian integers, Z[i] := {a + bi : a, b way. Z}, shares many properties with Z, so one might A reasonably simple prime-producing polyno- expect a negative answer for H10 over Z[i]. More mial in 26 variables was constructed in a paper by generally, inside any number eld k (i.e., nite J. P. Jones, D. Sato, H. Wada, and D. Wiens: see extension of Q), one has the ring of integers Ok , [Mat93, p. 55]. Later Matiyasevich constructed a dened as the set of k satisfying f () = 0 for 10-variable example. some monic f (x) Z[x]. Riemann hypothesis. The DPRM theorem gives an explicit polynomial equation that has a solution in integers if and only if the Riemann hypothesis (RH) is false. Indeed, one can write a computer program that searches for a counterexample to RH (e.g., by applying the argument principle and numerical integration to rectangles with corners in Q[i] lying in the strip 1/2 < Re s < 1, or by testing an equivalent formulation Conjecture 12. For any number eld k, H10 over Ok has a negative answer. Through work of J. Denef, L. Lipshitz, T. Pheidas, A. Shlapentokh, and the author spanning about 30 years, the following is known: Theorem 13. For a number eld k, H10 over Ok has a negative answer if any of the following hold:

(i) k is totally real (i.e., every homomorphism k C has image contained in R). (ii) k is a quadratic extension of a totally real number eld. (iii) k has exactly one conjugate pair of nonreal embeddings. (iv) There exists an elliptic curve E over Q such that E(Q) and E(k) have the same positive rank. To make sense of (iv), recall the Mordell-Weil theorem, which states that for any elliptic curve E over a number eld k, the abelian group E(k) of points on E with coordinates in k is nitely generated. Condition (iv) is probably satised for every number eld k, but this seems extremely dicult to prove. The reason that the proof of the negative answer for Z cannot be adapted directly to arbitrary Ok is that it uses the fact that the integer solutions to Pells equation x2 dy 2 = 1 for a xed nonsquare d N form an abelian group of rank 1. It is only for number elds like those in (i)(iii) of Theorem 13 that something close enough to this holds over Ok . In contrast with Conjecture 12, if Z is the ring of all algebraic integers, i.e., { C : f () = 0 for some monic f (x) Z[x]}, then H10 over Z has a positive answer, as shown by R. Rumely. H10 over Q. H10 over Q is equivalent to one of the big open problems in arithmetic geometry, namely whether there is a general algorithm for deciding whether a variety X over Q has a rational point5.

Z. Unfortunately, this goes the wrong way: the subproblem might still be decidable even though the whole problem is not.6 One way to get a reduction in the useful direction would be to show that Z is diophantine over Q, i.e., that there is a polynomial p(t, x) Q[t, x1 , . . . , xn ] such that Z equals the set of a Q such that p(a, x) = 0 has a solution x Qn . Indeed, we could use this to embed H10 over Z as a subproblem of H10 over Q: given a polynomial equation to be solved in integers, we could consider the same equation over Q together with auxiliary equations that force the rational variables to take integer values (this is where we need Z to be diophantine over Q). Actually, something a little weaker would sufce for the desired reduction. It would suce to have a diophantine model of the ring Z over Q, i.e., a diophantine set S Qn that looks like Z in the sense that it is equipped with a bijection : Z S such that the graphs of + and (subsets of Z3 ) correspond under to diophantine subsets of S 3 Q3n . Even more generally, it would suce to have a diophantine interpretation of Z over Q: this is like a diophantine model, except that Z is identied not with a diophantine subset of some Qn , but with a diophantine subset modulo a diophantine equivalence relation.

Remark. It has been suggested that one might try to build a diophantine model of Z over Q using an elliptic curve E with E(Q) Z. Such elliptic curves are easy to nd, and under the bijection Z E(Q) the graph of + on Z corresponds to a Reductions. Might one deduce a negative answer diophantine subset; unfortunately it is not clear to H10 over Q from the negative answer to H10 whether the same is true for the graph of . over Z? Given a polynomial equation over Q, Mazurs conjecture. B. Mazur has proposed a one can construct an equivalent system of poly- conjecture that, if true, would rule out some of nomials over Z by replacing each rational variable these approaches towards a negative answer to by a ratio of two new integer variables, clearing H10 over Q. If X is a variety over Q, then the denominators, and adding auxiliary equations to set X(R) of real points on X inherits a topology force the denominator variables to take nonzero from the topology of Rn . values in any solution (such auxiliary equations exist since the subset Z {0} of Z is diophan- Conjecture 14 (Mazur 1992). For any variety tine). Since a system of polynomial equations X over Q, the topological closure of X(Q) in f1 = = fn = 0 over Z is equivalent to a sin- X(R) has at most nitely many connected com2 2 gle polynomial equation f1 + + fn = 0 over ponents. Z, the previous sentence shows that H10 over Q A deep theorem of G. Faltings can be used to can be embedded as a subproblem of H10 over prove Mazurs conjecture for a curve X. But our
5Readers unfamiliar with the notion of variety will lose little generality, for our purposes, in thinking of X as a

system of polynomial equations, and a rational point as a simultaneous solution in rational numbers. 6On the other hand, if H10 over Z had had a positive answer, it would have implied a positive answer to H10 over Q. It has been argued that this, together with the fact that Hilbert asked his question for Z instead of Q, suggests that Hilbert expected a positive answer to his tenth problem.

almost complete lack of understanding of rational points on higher-dimensional varieties makes it dicult to gather much evidence for or against the conjecture in general. See [Maz94] for further discussion. Mazurs conjecture, together with some elementary topology, implies that for any set S Qn that is diophantine over Q, the closure of S in Rn has at most nitely many connected components. In particular, it implies that Z is not diophantine over Q. (This was Mazurs reason for introducing his conjecture.) A more complicated argument of G. Cornelissen and K. Zahidi involving the DPRM theorem shows that Mazurs conjecture implies also that there is no diophantine model of Z over Q. On the other hand, it is not known whether Mazurs conjecture rules out also a diophantine interpretation of Z over Q. Subrings of Q. Given that we have a negative answer for Z and do not know the answer for Q, we might ask about rings in between. Every such ring is Z[S 1 ] for some subset S of the set P of all primes: Z[S 1 ] consists of the rational numbers whose denominators are divisible only by primes in S. How large can we make S and still prove a negative answer for H10 over Z[S 1 ]? If S is nite, work of Robinson on diophantine denitions of valuation rings in Q implies that Z is diophantine over Z[S 1 ], so the negative answer for Z implies a negative answer for Z[S 1 ]. If S is innite, we may measure its size by dening the natural density of S as #{p S : p X} , lim X #{p P : p X} if the limit exists. In 2003 the author proved Theorem 15. There exists a computable set S P of density 1 such that (i) There exists a curve E such that E(Z[S 1 ]) is an innite discrete subset of E(R). (So the analogue of Mazurs conjecture for Z[S 1 ] is false.) (ii) There is a diophantine model of Z over Z[S 1 ]. (iii) H10 over Z[S 1 ] has a negative answer. The proof takes E to be an elliptic curve of rank 1 (minus its point at innity), and shows that by choosing S carefully, we can control the subset E(Z[S 1 ]) of E(Q) suciently well to obtain a discrete set that looks enough like Z to serve as a diophantine model.

Unfortunately, the complement of S in P, while sparse, is still innite, so Theorem 15 implies nothing about H10 over Q.

First-order sentences In terms of logic, H10 asks for an algorithm to decide the truth of positive existential sentences (x1 x2 xn ) f (x1 , . . . , xn ) = 0 in the language of rings, where the variables run over integers. More generally, one can ask for an algorithm to decide the truth of arbitrary rstorder sentences, in which any number of quantiers and boolean operations are permitted: a typical such sentence is (x)(y)(z)(w) (xz +3 = y 2 ) (z = x+w). Long before DPRM, the work of K. Gdel, o A. Church, and A. Turing in the 1930s made it clear that there was no algorithm for solving the harder problem of deciding the truth of rst-order sentences over Z. First-order sentences over Q. Though it is not known whether Z is diophantine over Q, we have Theorem 16 (Robinson 1949). One can characterize Z as the set of t Q such that a particular rst-order formula of the form (x)(y)(z)(w) p(t, x, y, z, w) = 0 is true, when the variables range over rational numbers. Combining this with the non-existence of an algorithm for rst-order sentences over Z, Robinson obtained Corollary 17. There is no algorithm to decide the truth of a rst-order sentence over Q. How complicated must a class of rst-order sentences be, in order that we are able to prove that no algorithm can decide the truth of all sentences in the class? Using quaternion algebras, the author in 2007 improved Robinsons result by dening Z in Q by a formula with 2 universal quantiers followed by 7 existential quantiers:

Theorem 18. The set Z equals the set of t Q For C the positive answers are a consequence of such that 19th century elimination theory. For R they come from A. Tarskis elimination theory for semialge(a, b)(x1 , x2 , x3 , x4 , y2 , y3 , y4 ) braic sets, subsets of Rn dened by polynomial (a + x2 + x2 + x2 + x2 )(b + x2 + x2 + x2 + x2 ) 1 2 3 4 1 2 3 4 equations and polynomial inequalities. For nite h` 2 2 2 2 2 elds Fq , the answers are trivially positive! By a x1 ax2 bx3 + abx4 1 # p-adic eld, we mean a nite extension of the eld 2309 Y` 2 2 2 2 Qp of p-adic numbers; A. Macintyre developed an + (n t 2x1 )2 4ay2 4by3 + 4aby4 4 elimination theory for these, though the positive n=0 answers were given before this, in work of J. Ax, =0 Yu. Ershov, S. Kochen, and A. Nerode. It is suris true, when the variables range over rational prising that the answers for the closely analogous numbers. eld Fq ((t)) of formal Laurent series over a nite Corollary 19. There is no algorithm for decid- eld are not known. We have already mentioned Rumelys positive ing, given an algebraic family of morphisms of varieties, whether there exists one that is surjec- answer for H10 over Z; this was extended to rstorder sentences by L. van den Dries. The negative on rational points. tive answers for rst-order sentences over a numCornelissen and Zahidi obtained an even bet- ber eld k and its ring of integers O are due to k ter result conditional on the truth of a plausible Robinson. conjecture about elliptic curves. By global function eld we mean the eld Fq (t) If we could eliminate the two universal quan- of rational functions with coecients in a nite tiers in Theorem 18, we would have a negative eld, or a nite extension of F (t). Such elds are q answer to H10 over Q. But we cannot see how to studied both because they are closely tied to algeeliminate even one of them. braic geometry and because they are analogous to Status of knowledge. The table below summa- number elds in many ways. The breakthrough giving the negative answer to H10 for Fq (t) for rizes what is known regarding the questions odd q was due to T. Pheidas. The extension to all Is there an algorithm for H10 over R? global function elds (and even nite extensions Is there an algorithm to decide the truth of F (t , . . . , t ) for n 2) was completed by q 1 n of arbitrary rst-order sentences over R? C. Videla, A. Shlapentokh, and K. Eisentrger. a over various rings R, listed roughly in order of The proofs use the Frobenius endomorphism in increasing arithmetic complexity7: an essential way, however, and hence cannot be adapted to number elds. Ring H10 1st order The negative answer to H10 over C(t1 , . . . , tn ) C YES YES for n 2 is due to K. H. Kim and F. W. Roush; R YES YES this result should be better known among algeFq YES YES braic geometers than it is since it implies that p-adic elds YES YES there is no algorithm for the general problem Fq ((t)) ? ? of deciding whether a rational map of varieties X Pn over C for xed n 2 admits a raZ YES YES tional section. The analogue with Pn replaced number eld ? NO by an arbitrary xed variety Y of dimension at Q ? NO least 2 was proved by K. Eisentrger using work a global function eld NO NO of L. Moret-Bailly. Although the answers for C(t) Fq (t) NO NO are unknown, the answers for R(t) are negative, C(t) ? ? as shown by J. Denef. C(t1 , . . . , tn ), n 2 NO NO Our list of results is by no means complete: R(t) NO NO for instance, we have said nothing about rings of Ok ? NO holomorphic or meromorphic functions, function Z NO NO
7There is no formal denition of arithmetic complexity, but for elds k we can look at the size of the absolute

Galois group Gal(ks /k), where ks is a separable closure of k. Domains may be considered more complex than their fraction elds, since they have extra structure coming from the divisibility relation.

elds over an algebraically closed eld of posi- [DLPVG00] Jan Denef, Leonard Lipshitz, Thanases Pheidas, and Jan Van Geel (eds.), Hilberts tenth tive characteristic, etc. There remain many open problem: relations with arithmetic and alproblems for anyone who is interested. Acknowledgements I have borrowed extensively from many excellent earlier expositions of the subject; some of these are listed below. I thank M. Davis, E. Frenkel, Yu. Matiyasevich, and A. Shlapentokh for many comments. Short list of references
[Csi08] George Csicsery, Julia Robinson and Hilberts tenth problem, 2008. Film in progress, Zala Films, www.zalafilms.com. Martin Davis, Yuri Matiyasevich, and Julia Robinson, Hilberts tenth problem: Diophantine equations: positive aspects of a negative solution, Mathematical developments arising from Hilbert problems (Proc. Sympos. Pure Math., Vol. XXVIII, Northern Illinois Univ., De Kalb, Ill., 1974), Amer. Math. Soc., Providence, R. I., 1976, pp. 323378. (loose erratum). MR 0432534 (55 #5522)

[H10web]

[Mat93]

[DMR76]

[Maz94]

[Shl07]

gebraic geometry, Contemporary Mathematics, vol. 270, American Mathematical Society, Providence, RI, 2000. Papers from the workshop held at Ghent University, Ghent, November 25, 1999. MR 1802007 (2001g:00018) Hilberts tenth problem page. Website created by Maxim Vsemirnov under the supervision of Yuri Matiyasevich, http://logic.pdmi.ras.ru/Hilbert10. Yuri V. Matiyasevich, Hilberts tenth problem, Foundations of Computing Series, MIT Press, Cambridge, MA, 1993. Translated from the 1993 Russian original by the author; With a foreword by Martin Davis. MR 1244324 (94m:03002b) B. Mazur, Questions of decidability and undecidability in number theory, J. Symbolic Logic 59 (1994), no. 2, 353371. MR 1276620 (96c:03091) Alexandra Shlapentokh, Hilberts tenth problem. Diophantine classes and extensions to global elds, New Mathematical Monographs, vol. 7, Cambridge University Press, Cambridge, 2007. MR 2297245