The Diophantine Problem for Polynomial Rings and Fields of Rational Functions

Author(s): J. Denef
Source: Transactions of the American Mathematical Society, Vol. 242 (Aug., 1978), pp. 391-399
Published by: American Mathematical Society
Stable URL: http://www.jstor.org/stable/1997746 .
Accessed: 02/12/2013 14:16

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp

.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.

American Mathematical Society is collaborating with JSTOR to digitize, preserve and extend access to
Transactions of the American Mathematical Society.

http://www.jstor.org

This content downloaded from 134.58.253.30 on Mon, 2 Dec 2013 14:16:46 PM

All use subject to JSTOR Terms and Conditions
TRANSACTIONS OF THE
AMERICAN MATHEMATICAL SOCIETY
Volume 242, August 1978

THE DIOPHANTINEPROBLEMFOR POLYNOMIALRINGS

AND FIELDS OF RATIONALFUNCTIONS'
BY
J. DENEF2

We provethat the diophantineproblemfor a ringof polynomials

ABSTRACr.
over an integral domain of characteristiczero or for a field of rational
functionsover a formallyreal field is unsolvable.

1. Introduction.3During the last thirty years much work has been done to
provethat the elementarytheoryof variousringsis undecidable;see J. Ax [1],
Yu. Ersov [9], [10], A. Malcev [14], Yu. Penzin [15], J. Robinson [17]{-20],
R. M. Robinson [21], [22] and A. Tarski[23].
After M. Davis, Yu. Matijasevic,H. Putnamand J. Robinson (see, e.g., [4],
[6]) proved that the existentialtheoryof Z is undecidable,it is naturalto ask
whetherthe existentialtheoryof variousotherringsis undecidabletoo.
Let R be a commutativering with unity and let R' be a subringof R. We
say that the diophantineproblemfor R with coefficientsin R' is unsolvable
(solvable) if there exists no (an) algorithm to decide whether or not a
polynomial equation (in several variables) with coefficients in R' has a
solutionin R.
In [7] we proved that the diophantineproblem for the ring of algebraic
integersin any quadraticextensionof Q is unsolvable,and recentlywe have
extendedthis to some more algebraicintegerrings.For some very interesting
relatedresults,see L. Lipshitz[13].
The main resultsof this paper are:
THEOREMA. Let R be an integraldomainof characteristiczero; then the
diophantineproblemfor R [T] with coefficientsin Z[T] is unsolvable.(R [T]
denotesthe ring of polynomialsover R, in one variableT.)
THEOREMB. Let K be a formallyrealfield, i.e. - I iv not the sum of squares

Receivedby the editorsJune 20, 19 /

AMS (MOS) subjectclassifications(1970).Primary02G05, 10N05, 10B99.
Keywordsandphrases.Hilbert'stenthproblem,unsolvableproblems,diophantineequations.
'Dedicatedto ProfessorL. P. Bouckaerton the occasionof his seventiethbirthday.
2Thisworkhas been supportedby the "NationaalFonds voor Wetenschappelijk Onderzoek".
It was done at HarvardUniversity,whose generoushospitalityI greatlyappreciate.I am also
gratefulto M. Boffa and R. M. Robinsonfor simplifyingthe proofof Lemma2.1.
3Weuse the followingnotations:N is the set of naturalnumbers;Z is the ringof integers;Q is
the field of rationals;R is the field of real numbers;and C is the field of complexnumbers.
? American Mathematical Society 1978

391

This content downloaded from 134.58.253.30 on Mon, 2 Dec 2013 14:16:46 PM

All use subject to JSTOR Terms and Conditions
392 J. DENEF

in K. Then the diophantineproblemfor K(T) with coefficientsin Z[T] is

unsolvable.(K( T) denotesthefield of rationalfunctionsover K, in one variable
T.)
We prove TheoremA in ?2 and TheoremB in ?3.
It is obvious that the diophantineproblemfor R [T] with coefficientsin Z
is solvableif and only if the diophantineproblemfor R with coefficientsin Z
is solvable. And the same holds for K(T). (An algebraicclosed field, a real
closed field, the ring of p-adic integers and the ring of formal power series
over a decidable field of characteristiczero are examples of rings whose
diophantineproblemwith coefficientsin Z is solvable.)
R. M. Robinson [21]provedfor any integraldomainR that the elementary
theory of R [7T]is undecidable.M. Davis and H. Putnam[5] proved that the
diophantineproblem for Z[T] with coefficients in Z[T] is unsolvable. But,
after that the diophantineproblemfor Z was proved unsolvable,it becomes
trivial that the diophantine problem for Z[T] with coefficients in Z is
unsolvable.
A. Malcev [14] and A. Tarski [23] proved that the elementarytheory of
K(T) is undecidablewhen K is a real closed field. A simplerproof of this
result has been given by J. Robinson [20]. Later R. M. Robinson [22] exten-
ded this resultto any formallyreal field K. Yu. Ersov [9] and Yu. Penzin [15]
proved the undecidabilityof the elementarytheory of K(T) when K is a
finite field.
If K is a formallyreal field then so is K(T); thus TheoremB is also truefor
fields of rationalfunctionsin severalvariables.
It is interestingto compare our work with a result of J. Becker and L.
Lipshitz[2]: The diophantineproblemfor C[[T1,T2]](i.e. the ring of formal
power seriesover C, in the variablesT, and T2)with coefficientsin Z[Tp T2]
is solvable, althoughthe elementarytheory of C[[T1,T2]] is undecidable(see
Er'sov[10]).
Let R be a commutative ring with unity and let D (xx,..., x) be a
realtion in R. We say that D (xl, . .. , x,,) is diophantineover R if there exists a
polynomial P(xI .... a, X yl, *Ymn) over R such that for all xl, . . . , xn in
y.
R:
D (xl, .. **,xn)
<43yj1,. *,YmEER: P(xl,.. *Xwy
* ... *Ym)
* =?
We have the same definition for subsets of R by regardingthem as 1-ary
relations.Let R' be a subringof R and supposeP can be chosen such that its
coefficientslay in R', then we say that D (xl,... , xn) is diophantineover R
with coefficientsin R'.
If R is an integraldomain and if DI and D2 are diophantineover R [T]
with coefficients in Z[T], then also DI V D2 and Di A D2 are diophantine

This content downloaded from 134.58.253.30 on Mon, 2 Dec 2013 14:16:46 PM

All use subject to JSTOR Terms and Conditions
 THE DIOPHANTINE PROBLEM 393

over R [T] with coefficientsin Z[T]; indeed,

P,=OVP2=0O-->P1P2=O and P1= OAP2=O +
p2_ TP2 O.
Moreover,the same holds for K(T).
In this paper we prove also:
PROPOSITION1. Let R be an integral domain of characteristic zero. Suppose
there exists a subset S of R which contains Z and which is diophantine over
R [T]; then Z is diophantine over R [T]. In particular, this is true when R
contains Q.
PROPOSITION 2. Let K be a formally real field. Suppose there exists a subset S
of K which contains Z and which is diophantine over K(T); then Z is
diophantine over K(T). In particular, this is true when K contains the real
closure of Q.
In [8] we have provedthat a relationis diophantineover Z[T] if and only if
it is recursivelyenumerable.M. Boffa noticed that for a nondenumerable
languagethe situationcan be very different:
COROLLARY(M. BOFFA). Every subset D of N is diophantine over
R[T].
PROOF. Let r be the real number r = ? an/IOn ", where an = 0 for
n E D and an = I for n X D. Then we have
n EDn EN 3p, m E N: (m=lnO<m- <10)
But Z is diophantine over R[T] by Proposition 1, and every recursively
enumerablerelationin Z is diophantineover Z (see, e.g., [4], [6]).Thus, using
elementaryalgebra,we see that D is diophantineover R[T]. Q.E.D.
2. Polynomialrings. Let R be any integral domain of characteristiczero.
We considerthe Pell equation
X2_ (T2_ 1)y2 = I (1)
over R [ T]. Let U be an elementin the algebraicclosureof R [ T] satisfying
U2= T2_ 1. (2)
Define two sequences Xn, Yn,n = 0, 1, 2,..., of polynomials in Z[T], by
setting

Xn + UYn = (T + U)n. (3)

We prove that Lemma2.2 of M. Davis and H. Putnam[5] remainstrue when
Z is replacedby R:
LEMMA
2.1. The solutions of (1) in R [ T] are given precisely by
X = ?XXn, Y=?Yn, n = 0, 1, 2.....

This content downloaded from 134.58.253.30 on Mon, 2 Dec 2013 14:16:46 PM

All use subject to JSTOR Terms and Conditions
394 J. DENEF

PROOF.(1) is equivalentto
(X - UY)(X + UY) - (4)
From (3) and (2) follows
Xn - UYn= (T -U)n= (T + U)-n.
Hence the X", Ynare solutionsof (1).
Conversely,suppose X and Y in R [T] satisfy (1). Let us parametrisethe
curve(2) by

T= t2+1 U- 2t
t2 -it2-i
The rational functions X + UY and X - UY in t have poles only at t = +1.
Moreover(4) impliesthey have zeroesonly at t = + 1. Hence

X+ UY=c( t +1 )= C(T + U), c E R, m E Z.

Thus also X - UY = c(T U)m.But substituting this in (4) gives c2= 1,

-
whichprovesthe lemmaby (3). Q.E.D.
Throughoutthis section we write V, W to denote that the polynomialsV
0
and W in R [ T] take the same value at T = 1. Notice that the relationZ O
is diophantineover R [T] with coefficientsin Z[T], indeed
Z -0 *>3X E=R [T]: Z =(T- I)X.
The followinglemmawas used by M. Davis and H. Putnam[5, Lemma2.3]
too:
2.2. We have Yn - n,for n = 0, 1, 29.
LEMMA
PROOF.From (3) and (2) follows

2n-
Yn= Ti T - ) T
i=1
i odd

Substitutenow T = 1. Q.E.D.
Let us define the 1-aryrelationImt(Y) in R [T] by
I
Imt(Y)<>YYE R[T]A3X E R[T]:X2 -(T2 )y2= 1.
LEMMA 2.3. Wehave:
(i) TherelationImt(Y) is diophantine overR [T] withcoefficientsin Z[T].
(ii) If Y satisfiesImt(Y), thenthereexists an integerm such that Y m.
(iii) For every integerm thereexists a polynomialY satisfyingImt(Y) and
Y m.

PROOF.This follows at once from Lemmas2.1 and 2.2.

This content downloaded from 134.58.253.30 on Mon, 2 Dec 2013 14:16:46 PM

All use subject to JSTOR Terms and Conditions
 THEDIOPHANTINE
PROBLEM 395

PROOF OF THEOREM A. There exists an algorithm to find for any

polynomialP (zl, . . , z) over Z, a polynomial P*(ZI, ... , Zm) over Z[T]
such that
3zI .. ZnE Z:P (zl Zn) =?

*Zl.3.Z . z ZmEERI[T]: P*(ZI,* * Zm) = ? (S)

Indeedby Lemma2.3 we have
3ZInc zinEt
Za Pa(zrediphntzn)= 0 []whc3Zo,fiinsi
Zn RI,T]: Z
(Imt(Zi ) A ... A Imt(Zn) A P (Zi, ***, Zn) - ?).
Since Imt and - are diophantine over R [T] with coefficients in Z[ T], we
easily obtain a polynomial P* satisfying (5). Hence if the diophantine
problem for R [T] with coefficients in Z[T] would be solvable, then the
diophantineproblemfor Z wouldbe solvable. Q.E.D.
PROOFOF PROPOSITION 1. If S satisfies the conditions of the proposition,
then
z EZE-*3Z E R[T]: (Imt(Z)AZ-..z A ES).
Moreover,if R contains Q, then we define S by:
x E S-x E R[T] A(x =OV3Y E R[T]:xy = 1). Q.E.D.
3. Fieldsof rationalfunctions.Let F be a field. A projectivecurveE, given
by the affine equation cy2 = x3 + ax + b, is called an ellipticcurvedefined
over F if it is nonsingularand if a, b and c are in F. One defines (see, e.g.,
Cassels [3, ?7], Fulton [11, Chapter5, ?6] or Lang [12, Chapter 1, ??3, 4]) a
commutative group law " + " on the set E (F) of points on the elliptic curve E
which are rational over F. The neutral element of this group is the unique
point Q at infinity on E. We shall denote by (v, w) the point with affine
coordinates x = v, y = w.
EveryellipticcurveE definedover Q whosej invariant(j = 2833a3/(4a3+
27b2)) is not integralhas no complex multiplication,i.e. the only C-rational
maps from E into itself which fix 0 are the maps PH-4 m -P = P + P
+ * * - + P (m times), m E Z. (See, e.g., Lang [12, Chapter 1, ?5 and
Chapter5, ?2, Theorem4].)
From now on we fix an elliptic curve Eo defined over Q, without complex
multiplicationand with equation
Y2 = X3 + ax + b.(1)
To Eo we associatethe ellipticcurve
(T3 +aT+b)Y2=X3+aX+b, (2)
definedover Q(T), which we denote from now on by E. Obviouslythe point
P1 with coordinates(T, 1) lies on E(Q(T)).

This content downloaded from 134.58.253.30 on Mon, 2 Dec 2013 14:16:46 PM

All use subject to JSTOR Terms and Conditions
396 J. DENEF

Let K be any field of characteristiczero; then we have

LEMMA3.1. Thepoint PI is of infinite order and generates the group E (K(T))
modulopoints of order two.
PROOF.We identify T with the rational function (x, y)F-*x on Eo and we
denote the rationalfunction (x, y)v*y on Eo by U. The functionfield F of Eo
over K is thus F = K(T, U), where U2 = T3 + aT + b. Let 4I be the
F-rationalmap
4I: E-* Eo: (X, Y)f- (X, UY).
Notice that 4, is a group homomorphismsince it is rational and 4,(Q) = 0.
We denote the group of K-rationalmaps from Eo into Eo by RatK(E0,EO).
Let 42 be the map
Eo (F) -> RatK(EO,Eo)
42:

whichsends the point (V, W) on EO(F)to the K-rationalmap

{2 ( V, W): Eo ( V(x, y), W(x, y)).
Eo: (x, y)-
Obviously42 is a homomorphism.Considerthe grouphomomorphism
P = P2o {l: E (K (T)) -> RatK(E0, E0).
For all points (X, Y) on E(K(T)) we have
T o4(X,Y)=X, (3)
Uo fr(X,Y)=UY. (4)
Hence ' is injective.Since Eo has no complexmultiplication,we have
RatK(E0, EO) {tam.m E Z} E Eo(K), (5)

whereamis the map Ph-*m. P, and where we identify a point on Eo with the
constantmap from Eo onto this point. Notice that 4(P,) = a,, and {(m PI)
= am. Thus P1 is of infinite order. Moreover, if (X, Y) E E(K(T)) and
+(X, Y) E EO(K),then X E K by (3) and (2) yields Y = 0. This means that
(X, Y) is a point of ordertwo on E(K(T)). The lemma follows now from (5).
Q.E.D.
We denote, for any nonzero integerm, the affine coordinatesof m PI by
(Xm,Y,). Notice that Xmand Y, are in Q(T). For any V and W in K(T) we
write V W to denote that V - W (consideredas a rationalfunction on the
projectiveline over K) takes the value zero at infinity.
LEMMA3.2. Using the above notation we have Xm/ TYm m for all nonzero
integers m.

PROOF.Notice that T/ U is a local parameter on Eo at 0, hence

This content downloaded from 134.58.253.30 on Mon, 2 Dec 2013 14:16:46 PM

All use subject to JSTOR Terms and Conditions
 THE DIOPHANTINE PROBLEM 397

{ (T/U)o 0 } )

(See, e.g., Lang [12, Appendix 1, ?3].) On the other hand, from (3) and (4)
follows
Xm T0b (Xm, Ym) T oam = ( T T
TYm (UWo?(Xm, Ym))T/U (Uoam)T/U U Mr/ U-
Q.E.D.

Let us define the 1-aryrelationImt(Z) in K(T) by

Imt(Z)*-*Z E K(T)

A {Z = OV 3X, Y E K(T): ((X, Y) E 2 *E(K(T)) A 2TYZ = X)}.

LEMMA 3.3. (i) The relation Imt(Z) is diophantineover K(T) with coefficients
in Z[T].
(ii) If Z satisfies Imt(Z), then there exists an integer m such that Z m.
(iii) For every integer m, there exists an element Z in Q(T) satisfying Imt(Z)
and Z m.
PROOF.This follows at once from Lemmas3.1 and 3.2.
We considerthe relationCom(y) definedby
Com(y)-y E K(T)A3x E K(T):y2= X3- 4.

The followinglemma was used by R. M. Robinson [22, ?4] too:

LEMMA 3.4. (i) The relation Com(y) is diophantine over K(T) with
coefficients in Z.
(ii) If y satisfies Com(y), theny lies in K.
(iii) For every rational number z, there exists a rational numbery satisfying
Com(y) and y > z.
(iv) If K contains the real closure of Q, then every integery satisfies Com(y).
PROOF.(i) and (iv) are obvious.
(ii) Since y2 = X3 - 4 is a curve of genus 1, it admits no rational para-
metrization.
(iii) It is known (see, e.g., R. M. Robinson [22, ?4]) that the group of
rational points on the elliptic curveY2 = X- 4 is infinite. So the rational
points are everywheredense on the curve in the real plane. Indeed since the
curve is connected in the real plane, its group of real points is a topological
groupisomorphicto the circle group.But every infinite subgroupof the circle
groupis everywheredense. Q.E.D.
We define the 1-aryrelationZ 0 in K(T) by

This content downloaded from 134.58.253.30 on Mon, 2 Dec 2013 14:16:46 PM

All use subject to JSTOR Terms and Conditions
398 J. DENEF

Z?0 <->Z E K(T)A3X1,X2,X3,X4,X5,y E (T):

(Com (y) A (6)
(Y - T)Z2 + 1-X2 + X22+ X32+ 2 + X52 (7)
LEMMA3.5. (i) TherelationZ: 0 is diophantineover K(T) withcoefficients
in Z[T].
(ii) If thefield K is formallyreal and if Z 4- 0, thenZ 0
(iii) If Z E Q(T) and Z O, 0 thenZ A- 0.
PROOF.(i) is obvious.
(ii) Suppose there exist XI,... ,X5, y in K(T) satisfying (6) and (7).
Suppose we have not Z - 0, then deg Z > 0 (where deg Z denotes the
degree of the rational function Z). From (6) and Lemma 3.4(ii) follows
y E K. Hence deg ((y - T)Z2 + 1) is positive and odd. But deg (X2 + X2
+ * * * + X2) is even. Indeed there is no cancellationof the coefficients of
largestdegree,since a sum of squaresin a formallyreal field vanishesonly if
each termis zero. So we are in contradictionwith (7), hence Z- 0.
(iii) Let Z E Q(T) and Z 0, then TZ2 0 and thereis a naturalnumber
z such that
j(TZ2 )(r)l <2 when r E R andlrl > z.
By Lemma 3.4(iii) there exists a rational numbery satisfying Com(y) and
y > z > 0. Thus
((y - T)Z2 + 1)(r) > 0 forallrinR.
But a theoremof Y. Pourchet[16] states that every positive definite rational
functionover Q can be writtenas a sum of five squaresin Q(T). Hence there
exist Xl, .. ., X5 in K(T) satisfying (7), whence Z - O. Q.E.D.
PROOFOF THEOREMB. Thereexists an algorithmto find for any polynomial
P (zl .. , z) over Z, a polynomial P*(ZI, . .. , Z,,,)over Z[T] such that
3z1 ,..., zln E Z: P(z1, . ** Zn)=0
* 3Zi . . . , Zm E K(T): P*(ZI, *, Zm) = 0.
Indeed,by Lemmas3.3 and 3.5 we have
3zI, z,z E- Z: P(zl,.v zn,)= 0<-3Z,, Z, E K(T):
@zn-
(Imt(Zi A ... A Imt(Zn) A P (4l ..* * Zn), 0).
Proceednow as in the proof of TheoremA. Q.E.D.
PROOF OF PROPOSITION2. If S satisfies the conditions of the proposition,
then
z E Z -3Z E K(T): (Imt(Z) A Z-z -OA z E S).

This content downloaded from 134.58.253.30 on Mon, 2 Dec 2013 14:16:46 PM

All use subject to JSTOR Terms and Conditions
 THE DIOPHANTINE PROBLEM 399

Moreover,the last assertionof the propositionfollows from Lemma 3.4(iv).

Q.E.D.
REFERENCES
seriesfields, Proc. Amer. Math. Soc. 16 (1965), 846.
of powver
1. J. Ax, On the undecidability
2. J. Beckerand L. Lipshitz,Renarks on the elementarytheoriesof formal and convergent
powerseries,Fund. Math. (to appear).
3. J. W. S. Cassels,Diophantineequationswith specialreferenceto ellipticcurves,J. London
Math.Soc. 41(1966), 193-291.
4. M. Davis, Hilbert'stenthproblemis unsolvable,Amer.Math. Monthly80 (1973),233-269.
5. M. Davis and H. Putnam,Diophantine sets overpolynomialrings,IllinoisJ. Math.7 (1963),
251-256.
6. M. Davis, Yu. Matijasevicand J. Robinson, Diophantineequations:Positiveaspectsof a
negativesolution,Proc. Sympos.Pure Math., vol. 28, Amer. Math. Soc., Providence,R.I., 1976,
pp. 323-378.
7. J. Denef, Hilbert'stenthproblemfor quadraticrings, Proc. Amer. Math. Soc. 48 (1975),
214-220.
8. _ _, Diophantine sets over Z[T], Proc. Amer. Math. Soc. 69 (1978), 148-150.
of certainfields, Dokl. Akad.Nauk SSSR161(1965),349-352.
9. Yu. Ersov,Undecidability
10. _ _, New examples of undecidabletheories, Algebra i. Logika 5 (1966), 3747.
11.W. Fulton,Algebraiccurves,Benjamin,New York, 1969.
12.S. Lang,Ellipticfunctions,Addison-Wesley,London,1973.
13. L. Lipshitz,Undecidableproblemsfor additionand divisibilityin algebraicnumberrings.II,
Proc.Amer.Math.Soc. 64 (1977),122-128.
v 14. A. I. Malcev,On the undecidabilityof the elementarytheoriesof certainfields, SibirskMat.
Z. 1 (1960),71-77; ibid 2 (1961),639; Englishtransl.,Amer. Math. Soc. Transl.(2) 48 (1965).
15. Yu. Penzin, Undecidability of fields of rationalfunctionsover fields of characteristic2,
Algebrai. Logika12 (1973),205-210;244.
16. Y. Pourchet,Sur la representationen sommede carresdespolyno^mes a une indetermine sur
un corpsde nombresalgebriques, Acta Arith.19 (1971),89-104.
17. J. Robinson,Definabilityand decisionproblemsin arithmetic,J. SymbolicLogic 14 (1949),
98-114.
18. __ of algebraicringsandfields,Proc.Amer. Math. Soc. 10 (1959),
, The undecidability
950-957.
19. _ _, On the decisionproblemfor algebraicrings,Studiesin Math. Anal. and Related
Topics,Stanford,1962,pp. 297-304.
20. _ _, Thedecisionproblemfor fields, Sympos.on the Theoryof Models,North-Holland,
Amsterdam,1965,pp. 299-311.
21. R. M. Robinson,Undecidable rings,Trans.Amer.Math.Soc. 70 (1951),137-159.
22. _ _, The undecidability of pure transcendental extensionsof realfields, Z. Math. Logik
GrundlagenMath.10 (1964),275-282.
23. A. Tarski, The elementaryundecidability of pure transcendentalextensionsof real closed
fields, Notices Amer.Math. Soc. 10 (1963),A-355.
UNIVERSITY OF LEUVEN, DEPARTMENT OF MATMATICS, CELESTUINENLAAN200B, 3030
HEVERLEE, BELGIUM

Currentaddress: Department of Mathematics,Princeton University, Fine Hall, Box 37,

Princeton,New Jersey08540

This content downloaded from 134.58.253.30 on Mon, 2 Dec 2013 14:16:46 PM

All use subject to JSTOR Terms and Conditions