(London Mathematical Society Lecture Note Series) Claire C. Ralph, Santiago R. Simanca-Arithmetic Differential Operators Over The P-Adic Integers. 396-Cambridge University Press (2012)

L O ND O N MATH EMAT I C A L SO C I E T Y L E C T U R E N O T E SER I ES
Managing Editor: Professor M. Reid, Mathematics Institute,

University of Warwick, Coventry CV4 7AL, United Kingdom
The titles below are available from booksellers, or from Cambridge University Press
at http://www.cambridge.org/mathematics
287 Topics on Riemann surfaces and Fuchsian groups, E. BUJALANCE, A.F. COSTA & E. MARTNEZ (eds)
288 Surveys in combinatorics, 2001, J.W.P. HIRSCHFELD (ed)
289 Aspects of Sobolev-type inequalities, L. SALOFF-COSTE
290 Quantum groups and Lie theory, A. PRESSLEY (ed)
291 Tits buildings and the model theory of groups, K. TENT (ed)
292 A quantum groups primer, S. MAJID
293 Second order partial differential equations in Hilbert spaces, G. DA PRATO & J. ZABCZYK
294 Introduction to operator space theory, G. PISIER
295 Geometry and integrability, L. MASON & Y. NUTKU (eds)
296 Lectures on invariant theory, I. DOLGACHEV
297 The homotopy category of simply connected 4-manifolds, H.-J. BAUES
298 Higher operads, higher categories, T. LEINSTER (ed)
299 Kleinian groups and hyperbolic 3-manifolds, Y. KOMORI, V. MARKOVIC & C. SERIES (eds)
300 Introduction to Mbius differential geometry, U. HERTRICH-JEROMIN
301 Stable modules and the D(2)-problem, F.E.A. JOHNSON
302 Discrete and continuous nonlinear Schrdinger systems, M.J. ABLOWITZ, B. PRINARI & A.D. TRUBATCH
303 Number theory and algebraic geometry, M. REID & A. SKOROBOGATOV (eds)
304 Groups St Andrews 2001 in Oxford I, C.M. CAMPBELL, E.F. ROBERTSON & G.C. SMITH (eds)
305 Groups St Andrews 2001 in Oxford II, C.M. CAMPBELL, E.F. ROBERTSON & G.C. SMITH (eds)
306 Geometric mechanics and symmetry, J. MONTALDI & T. RATIU (eds)
307 Surveys in combinatorics 2003, C.D. WENSLEY (ed.)
308 Topology, geometry and quantum eld theory, U.L. TILLMANN (ed)
309 Corings and comodules, T. BRZEZINSKI & R. WISBAUER
310 Topics in dynamics and ergodic theory, S. BEZUGLYI & S. KOLYADA (eds)
311 Groups: topological, combinatorial and arithmetic aspects, T.W. MLLER (ed)
312 Foundations of computational mathematics, Minneapolis 2002, F. CUCKER et al (eds)
313 Transcendental aspects of algebraic cycles, S. MLLER-STACH & C. PETERS (eds)
314 Spectral generalizations of line graphs, D. CVETKOVIC, P. ROWLINSON & S. SIMIC
315 Structured ring spectra, A. BAKER & B. RICHTER (eds)
316 Linear logic in computer science, T. EHRHARD, P. RUET, J.-Y. GIRARD & P. SCOTT (eds)
317 Advances in elliptic curve cryptography, I.F. BLAKE, G. SEROUSSI & N.P. SMART (eds)
318 Perturbation of the boundary in boundary-value problems of partial differential equations, D. HENRY
319 Double afne Hecke algebras, I. CHEREDNIK
320 L-functions and Galois representations, D. BURNS, K. BUZZARD & J. NEKOVR (eds)
321 Surveys in modern mathematics, V. PRASOLOV & Y. ILYASHENKO (eds)
322 Recent perspectives in random matrix theory and number theory, F. MEZZADRI & N.C. SNAITH (eds)
323 Poisson geometry, deformation quantisation and group representations, S. GUTT et al (eds)
324 Singularities and computer algebra, C. LOSSEN & G. PFISTER (eds)
325 Lectures on the Ricci ow, P. TOPPING
326 Modular representations of nite groups of Lie type, J.E. HUMPHREYS
327 Surveys in combinatorics 2005, B.S. WEBB (ed)
328 Fundamentals of hyperbolic manifolds, R. CANARY, D. EPSTEIN & A. MARDEN (eds)
329 Spaces of Kleinian groups, Y. MINSKY, M. SAKUMA & C. SERIES (eds)
330 Noncommutative localization in algebra and topology, A. RANICKI (ed)
331 Foundations of computational mathematics, Santander 2005, L.M PARDO, A. PINKUS, E. SLI & M.J. TODD (eds)
332 Handbook of tilting theory, L. ANGELERI HGEL, D. HAPPEL & H. KRAUSE (eds)
333 Synthetic differential geometry (2nd Edition), A. KOCK
334 The NavierStokes equations, N. RILEY & P. DRAZIN
335 Lectures on the combinatorics of free probability, A. NICA & R. SPEICHER
336 Integral closure of ideals, rings, and modules, I. SWANSON & C. HUNEKE
337 Methods in Banach space theory, J.M.F. CASTILLO & W.B. JOHNSON (eds)
338 Surveys in geometry and number theory, N. YOUNG (ed)
339 Groups St Andrews 2005 I, C.M. CAMPBELL, M.R. QUICK, E.F. ROBERTSON & G.C. SMITH (eds)
340 Groups St Andrews 2005 II, C.M. CAMPBELL, M.R. QUICK, E.F. ROBERTSON & G.C. SMITH (eds)
341 Ranks of elliptic curves and random matrix theory, J.B. CONREY, D.W. FARMER, F. MEZZADRI &
N.C. SNAITH (eds)
342 Elliptic cohomology, H.R. MILLER & D.C. RAVENEL (eds)
343 Algebraic cycles and motives I, J. NAGEL & C. PETERS (eds)
344 Algebraic cycles and motives II, J. NAGEL & C. PETERS (eds)
345 Algebraic and analytic geometry, A. NEEMAN
346 Surveys in combinatorics 2007, A. HILTON & J. TALBOT (eds)
347 Surveys in contemporary mathematics, N. YOUNG & Y. CHOI (eds)
348 Transcendental dynamics and complex analysis, P.J. RIPPON & G.M. STALLARD (eds)
349 Model theory with applications to algebra and analysis I, Z. CHATZIDAKIS, D. MACPHERSON, A. PILLAY &
A. WILKIE (eds)
350 Model theory with applications to algebra and analysis II, Z. CHATZIDAKIS, D. MACPHERSON, A. PILLAY &
A. WILKIE (eds)
351 Finite von Neumann algebras and masas, A.M. SINCLAIR & R.R. SMITH
352 Number theory and polynomials, J. MCKEE & C. SMYTH (eds)
353 Trends in stochastic analysis, J. BLATH, P. MRTERS & M. SCHEUTZOW (eds)
354 Groups and analysis, K. TENT (ed)
355 Non-equilibrium statistical mechanics and turbulence, J. CARDY, G. FALKOVICH & K. GAWEDZKI
356 Elliptic curves and big Galois representations, D. DELBOURGO
357 Algebraic theory of differential equations, M.A.H. MACCALLUM & A.V. MIKHAILOV (eds)
358 Geometric and cohomological methods in group theory, M.R. BRIDSON, P.H. KROPHOLLER & I.J. LEARY (eds)
359 Moduli spaces and vector bundles, L. BRAMBILA-PAZ, S.B. BRADLOW, O. GARCA-PRADA &
S. RAMANAN (eds)
360 Zariski geometries, B. ZILBER
361 Words: Notes on verbal width in groups, D. SEGAL
362 Differential tensor algebras and their module categories, R. BAUTISTA, L. SALMERN & R. ZUAZUA
363 Foundations of computational mathematics, Hong Kong 2008, F. CUCKER, A. PINKUS & M.J. TODD (eds)
364 Partial differential equations and uid mechanics, J.C. ROBINSON & J.L. RODRIGO (eds)
365 Surveys in combinatorics 2009, S. HUCZYNSKA, J.D. MITCHELL & C.M. RONEY-DOUGAL (eds)
366 Highly oscillatory problems, B. ENGQUIST, A. FOKAS, E. HAIRER & A. ISERLES (eds)
367 Random matrices: High dimensional phenomena, G. BLOWER
368 Geometry of Riemann surfaces, F.P. GARDINER, G. GONZLEZ-DIEZ & C. KOUROUNIOTIS (eds)
369 Epidemics and rumours in complex networks, M. DRAIEF & L. MASSOULI
370 Theory of p-adic distributions, S. ALBEVERIO, A.YU. KHRENNIKOV & V.M. SHELKOVICH
371 Conformal fractals, F. PRZYTYCKI & M. URBANSKI
372 Moonshine: The rst quarter century and beyond, J. LEPOWSKY, J. MCKAY & M.P. TUITE (eds)
373 Smoothness, regularity and complete intersection, J. MAJADAS & A. G. RODICIO
374 Geometric analysis of hyperbolic differential equations: An introduction, S. ALINHAC
375 Triangulated categories, T. HOLM, P. JRGENSEN & R. ROUQUIER (eds)
376 Permutation patterns, S. LINTON, N. RUKUC & V. VATTER (eds)
377 An introduction to Galois cohomology and its applications, G. BERHUY
378 Probability and mathematical genetics, N. H. BINGHAM & C. M. GOLDIE (eds)
379 Finite and algorithmic model theory, J. ESPARZA, C. MICHAUX & C. STEINHORN (eds)
380 Real and complex singularities, M. MANOEL, M.C. ROMERO FUSTER & C.T.C. WALL (eds)
381 Symmetries and integrability of difference equations, D. LEVI, P. OLVER, Z. THOMOVA & P. WINTERNITZ (eds)
382 Forcing with random variables and proof complexity, J. KRAJCEK
383 Motivic integration and its interactions with model theory and non-Archimedean geometry I, R. CLUCKERS,
J. NICAISE & J. SEBAG (eds)
384 Motivic integration and its interactions with model theory and non-Archimedean geometry II, R. CLUCKERS,
J. NICAISE & J. SEBAG (eds)
385 Entropy of hidden Markov processes and connections to dynamical systems, B. MARCUS, K. PETERSEN &
T. WEISSMAN (eds)
386 Independence-friendly logic, A.L. MANN, G. SANDU & M. SEVENSTER
387 Groups St Andrews 2009 in Bath I, C.M. CAMPBELL et al (eds)
388 Groups St Andrews 2009 in Bath II, C.M. CAMPBELL et al (eds)
389 Random elds on the sphere, D. MARINUCCI & G. PECCATI
390 Localization in periodic potentials, D.E. PELINOVSKY
391 Fusion systems in algebra and topology M. ASCHBACHER, R. KESSAR & B. OLIVER
392 Surveys in combinatorics 2011, R. CHAPMAN (ed)
393 Non-abelian fundamental groups and Iwasawa theory, J. COATES et al (eds)
394 Variational Problems in Differential Geometry, R. BIELAWSKI, K. HOUSTON & M. SPEIGHT (eds)
395 How groups grow, A. MANN
396 Arithmetic Differential Operators over the p-adic Integers, C.C. RALPH & S.R. SIMANCA
397 Hyperbolic geometry and applications in quantum Chaos and cosmology, J. BOLTE & F. STEINER (eds)
LO N D O N MAT HE MAT ICAL SOCIE TY LECTU RE
NOT E SE RIE S : 396
Arithmetic Differential Operators

over the p-adic Integers
C L A I R E C. R A L P H
Cornell University, USA
S A N T I A G O R. S I M A N C A
Universit de Nantes, France
c am b r i dge u n i ve r s i t y pr e ss
Cambridge, New York, Melbourne, Madrid, Cape Town,
Singapore, So Paulo, Delhi, Tokyo, Mexico City
Cambridge University Press
The Edinburgh Building, Cambridge CB2 8RU, UK
Published in the United States of America by Cambridge University Press, New York
www.cambridge.org
Information on this title: www.cambridge.org/9781107674141
C. Ralph and S. Simanca, 2012
This publication is in copyright. Subject to statutory exception

and to the provisions of relevant collective licensing agreements,
no reproduction of any part may take place without the written
permission of Cambridge University Press.
First published 2012
Printed in the United Kingdom at the University Press, Cambridge
A catalogue record for this publication is available from the British Library
ISBN 978-1-107-67414-1 Paperback
Cambridge University Press has no responsibility for the persistence or

accuracy of URLs for external or third-party internet websites referred to in
this publication, and does not guarantee that any content on such websites is,
or will remain, accurate or appropriate.
Contents
1 Introduction page 1
2 The p-adic numbers Qp 8
2.1 A pragmatic realization of Qp 11
2.2 The p-adic integers Zp and their eld of fractions 13
2.3 The topology of Qp 15
2.4 Analytic and algebraic properties of Qp 17
2.5 (p 1)-roots of unity in Qp 20
3 Some classical analysis on Qp 23
3.1 The ArtinHasse exponential function 30
3.2 The completion of the algebraic closure of Qp 33
3.3 Zeta functions 38
4 Analytic functions on Zp 48
4.1 Strassmanns theorem 53
5 Arithmetic dierential operators on Zp 55
5.1 Multiple primes I 61
6 A general view of arithmetic dierential operators 65
6.1 Basic algebraic concepts 67
6.2 General p -functions: arithmetic jet spaces 73
6.3 The analogue of a p -linear operators for group
schemes 82
6.4 Multiple primes II 86
7 Analyticity of arithmetic dierential operators 103
8 Characteristic functions of discs in Zp : p-adic
coordinates 107
8.1 Characteristic functions of discs of radii 1/p 107
8.2 Characteristic functions of discs of radii 1/pn 116
v
vi Contents
9 Characteristic functions of discs in Zp : harmonic co-

ordinates 119
9.1 A matrix associated to pm 120
9.2 Analytic functions and arithmetic dierential
operators 122
10 Some dierences between -operators over Zp and
ur
Z 127
p
References 135
Index 138
1
Introduction
Our purpose in this monograph is to provide a concise and complete

introduction to the study of arithmetic dierential operators over the
p-adic integers Zp . These are the analogues of the usual dierential op-
erators over say, the ring C[x], but where the role of the variable x is
replaced by a prime p, and the roles of a function f (x) and its deriva-
tive df /dx are now played by an integer a Z and its Fermat quotient
p a = (a ap )/p.
In making our presentation of these type of operators, we nd no bet-
ter way than discussing the p-adic numbers in detail also, and some of
the classical dierential analysis on the eld of p-adic numbers, empha-
sizing the aspects that give rise to the philosophy behind the arithmetic
dierential operators. The reader is urged to contrast these ideas at will,
while keeping in mind that our study is neither exhaustive nor intended
to be so, and most of the time we shall content ourselves by explain-
ing the dierential aspect of an arithmetic operator by way of analogy,
rather than appealing to the language of jet spaces. But even then, the
importance of these operators will be justied by their signicant ap-
pearance in number theoretic considerations. One of our goals will be to
illustrate how dierent these operators are when the ground eld where
they are dened is rather coarse, as are the p-adic integers Zp that we
use.
In order to put our work in proper perspective, it is convenient to
introduce some basic facts rst, and recall a bit of history. Given a
prime p, we may dene the p-adic norm p over the eld of rational
numbers Q. The completion of the rationals in the metric that this norm
induces is the eld Qp of p-adic numbers, and this eld carries a non-
Archimedean p-adic norm extending the original p-adic norm on Q. This
is the description of Qp as given by K. Hensel circa 1897 (see, for instance,
1
2 Introduction
[28]). Two decades later, A. Ostrovski [39] proved that any nontrivial
norm on Q is equivalent to either the Euclidean norm or to a p-adic norm
for some prime p. In this way, there arose the philosophical principle that
treats the real numbers and all of the p-adic numbers on equal footing.
In the twentieth century, the p-adic numbers had a rich history. We
briey mention some major results.
The idea that studying a question about the eld Q can be answered
by putting together the answers to the same question over the elds R
and Qp for all ps was born with the HasseMinkowskis theorem. This
states that a quadratic form over Q has a nontrivial zero in Qn if, and
only if, it has a nontrivial zero in Rn and a nontrivial zero in Qnp for
each prime p. This theorem was proven by Hasse in his thesis around
1921 [27], the problem having been proposed to him by Hensel who had
proven the n = 2 case a few years earlier. Such a principle fails for cubics.
The development above came after several interesting results that pre-
ceded the introduction of the p-adic numbers. The local-to-global princi-
ple embodied in the HasseMinkowski theorem had a precedent in Rie-
mannian geometric, since as recently as 1855, Bonnet had proved that
if the curvature of a compact surface was bounded below by a positive
constant, then its diameter was bounded above by a quantity depending
only on the said constant. Strictly on the arithmetic side of things, in
the seventeenth century J. Bernoulli dened the Bernoulli numbers Bk ,

the coecients in the expansion et /(et 1) = k Bk tk /k!, used them
m
to compute closed-form expressions for the sums j=0 j n , and devel-
oped several identities that these numbers satisfy. A century later, the
Bernoulli numbers were used by Euler to show heuristically that if is

the Riemman zeta function, then (1 k) = n=1 1/n1k = Bk /k for
any integer k 2. In the mid nineteenth century, Riemman proved that

(s) = n=1 1/ns is a meromorphic function on the complex plane C,
giving Eulers argument complete s sense. Further, he used the Gamma
s2
function to dene (s) = 2 (s) and proved the functional equa-
tion (s) = (1 s). The intimate relationship between the Bernoulli
numbers and the values of (s) at negative integers led to the idea that
these numbers have profound arithmetical properties, a fact discovered
by Kummer in his work on Fermats last theorem circa 1847. The ideal
class group of Q(N ), N a primitive N -th root of unity, is the quotient
of the fractional ideals of Q(N ) by the set of principal ideals, and it
turns out to be a group of nite order hN with respect to ideal multipli-
cation. A prime p is said to be regular if p hp , and irregular otherwise.
Kummer proved that p is regular if, and only if, p does not divide the
Introduction 3
numerator of B2 , B3 , . . . , Bp3 and that Fermat last theorem holds for

all regular primes. He also proved that, if m n 0 mod p 1, then
Bm /m Bn /n mod p, the congruences that are nowadays named after
him. They led to the proof that that there are innitely many irregular
primes. Since heuristically it can be proven that there is a large per-
centile of regular primes, Kummers ideas had remarkable implications
in the study of Fermats last theorem. Thus, algebraic number theory
and the theory of L-functions were born and replaced the elementary
methods used before him in the analysis of this problem.
C. Chevalley dened the adèle ring and idèle group [20], and used them
to reformulate class eld theory [21] around 1932. For convenience, if we
denote by the Euclidean norm in R, which we think of as Q , the
eld of p-adic numbers corresponding to p = , we take the Cartesian

product Q p Qp , and dene the adèle ring AQ to be

AQ = (a , a2 , a3 , a5 , . . .) Q Qp : ||ap ||p 1 for almost all ps .
p
Its ring structure is obtained by dening addition and multiplication

component-wise; it contains an isomorphic image of Q via the mapping
aQ
Q q (q, q, . . .) AQ .

For a Q p Qp , we dene ||a||p = ||ap ||p . Then a AQ if, and only
if, ||a||p 1 for all but nitely many ps. The subset IQ of AQ consisting
of all as such ||a||p = 0 for all ps, and ||a||p = 1 for all but nitely many of
them, is the idèle multiplicative group. It contains an isomorphic image
of Q by restriction of the mapping aQ above. If F is an extension of
Q, the norms on Q can be extended to norms on F , and we naturally
dene IF also. There is a norm homomorphism IF IQ , and its image
N (IF /IQ ) is a group. The Galois group of F/Q is naturally isomorphic
to IQ /Q N (IF /IQ ). Chevalley proved this fact using the local theory,
avoiding the use of tools from analytic number theory. He generalized it
also for number elds, elds that are extensions of Q of innite degree.
In his thesis, J. Tate used real harmonic analysis on the adèles to
prove functional equations for the Riemann zeta function. T. Kubota
and H.W. Leopoldt [32] introduced a p-adic version of the Riemann zeta
function, and used it to interpret Kummers congruences for Bernoulli
numbers mentioned above, which date back to 1851. Y.I. Manin and
B. Mazur [38] interpreted the result of Kubota and Leopoldt in terms
of a p-adic Mellin transform, and found p-adic interpretations of L-
4 Introduction
functions of elliptic curves. The p-adic integers Zp were known to appear

as Galois groups of some innite cyclotomic extensions. K. Iwasawa con-
sidered the completed group algebras of these Galois groups, which act
on class groups and make them modules over the completed groups.
These modules have some invariants. Iwasawa conjectured that these
invariants could be read o from classical Dirichlet L-functions after a
p-adic interpolation, using the p-adic Mellin transform. This conjecture
was proved by B. Mazur and A. Wiles [33]. Triggered by the work of
Tate, B. Dwork studied p-adic dierential equations, and gave a p-adic
proof of the rationality of Weils zeta function [23], taking then a major
step in the settling of all of the Weil conjectures about this function
[48], work that was completed by P. Deligne [22]. J.-P. Serre and N.
Katz studied several other p-adic functions of arithmetic interest, and
A. Grothendieck studied p-adic cohomology and crystalline cohomology.
The list of problems in the eld is outstanding, and the list of contrib-
utors to their understanding and resolution is important. We have not
come even close to exhausting either one. But we can now retake the
main theme of our work in this introduction with a better perspective
in mind.
In the course of modern mathematical history, analogies between func-
tions and numbers have played an important role in the development of
number theory. The fundamental theorems of algebra and arithmetic
can be seen as counterparts to each other, with the integers 1, 0 and 1
playing the role of the constant polynomials in C[x]. This point of view is
once again motivational to the philosophy of arithmetic dierential oper-
ators, the idea at the level of the integers Z being to nd an appropriate
substitute p : Z Z for the derivative operator
d
x = : C[x] C[x] .
dx
Indeed, given a number x, lets us think of it as a function, and
consider the expression x xp , one that makes frequent appearances in
number theoretic considerations. For Fp , the nite eld of p elements, the
identity x xp = 0 holds for all elements. In the more general situation,
we can restrict our attention to numbers such that x xp 0 mod p.
We think of x as a function of p, and interpret the dierence x xp as
the variation of x as its argument changes to p. We then use the Fermat
quotient
x xp
p x = ( )
p
Introduction 5
to dene the notion of arithmetic derivative of x in the direction of p.

This is the notion that we shall be studying here, most of the time
restricting our attention to xs that are taken from the ring of p-adic
integers Zp . At this point, though, this quotient is just a heuristic state-
ment.
The theory of arithmetic dierential operators that ensues from the
idea outlined above was proposed by A. Buium [6, 8], with p serving in
the role of the arithmetic analogue of the operator x on the polynomial
ring C[x]. At the purely arithmetic level, it serves also as a substitute
for Dworks operator
d
: Fp [x] Fp [x]
dx
in his theory of p-adic dierential equations over the dierential eld
Fp [x], Fp the algebraic closure of the eld Fp with p-elements. In Dworks
theory [25], the xs are still being viewed as an argument to the func-
tions rather than as functions themselves. But the arithmetic dieren-
tial operator p exhibits an additional fundamental dierence with the
Dworks operator that is worth pointing out now: p is highly nonlin-
ear, with additivity holding only modulo a lower order term measured
by a polynomial with integer coecients, and a Leibniz rule that holds
but only highly intertwined with the p-th power homomorphism, and
modulo terms that are p-adically smaller.
In fact, more can be said at this point. If we were to develop a dier-
ential theory with operators of the type

du dr u
u P u, , . . . , r
dx dx
where P (x0 , . . . , xr ) is a polynomial function, we would obtain the
RittKolchin theory of ordinary dierential equations with respect
d d
to dx , cf. [41, 30, 19]. This would lead to the notion of the dx -character
of an algebraic group, which should be viewed as the analogue of a linear
ordinary dierential operator on an algebraic group (cf. to the Kolchin
logarithmic derivative of algebraic groups dened over Z ur , [30, 19], and
p
the Manin maps of Abelian varieties dened over Z ur
p [[q]] [38, 12], Zp
ur
the unramied completion of the ring of p-adic integers).
If instead we were to develop a theory with operators that are the
p-adic limits of P (u, p u, . . . , pr u), P (x0 , . . . , xr ) a polynomial, we would
then obtain the arithmetic analogue of the ordinary dierential equations
of Buium, as found in [8, 9, 6]. In particular, we would then arrive at
6 Introduction
the notion of a p -character of a group scheme, the arithmetic version

of a linear ordinary dierential equation on a group scheme.
In this monograph, we apply and study Buiums idea over the rather
coarse ring of p-adic integers Zp . We think of the elements in this ring
as functions over a space of dimension zero that vary innitesimally ac-
cording to the heuristic equation ( ) at the prime p. The ensuing notion
of derivative is the one alluded to in the title, and on which we shall elab-
orate extensively in what follows. We will pause at some point to dene
these arithmetic operators with the generality given in Buiums work.
This will benet the interested reader while allowing us to contrast the
behaviour of these operators when dened over Zp or Z ur . Ultimately,
p
it is the fact that we can cast these operators as global functions on
a suitable arithmetic jet space, for any smooth scheme of nite type,
which allows for their interpretation as dierential operators of sorts,
the way the usual dierential operators on a manifold are sections of its
jet bundles.
Given such a notion of arithmetic derivative, we then may dene in the
obvious manner an arithmetic dierential operator of order n, where n is
an arbitrary positive integer. Over the ring of p-adic integers Zp , we have
also the classical notion of an analytic function. We shall show that all
arithmetic dierential operators turn out to be analytic functions. Quite
remarkably in fact, characteristic functions of p-adic discs are shown
to be equal to arithmetic operators of an order that depends upon the
radius of the disc, generalizing a result that we rst prove via an explicit
construction, namely that the characteristic function of a disc of radius
1/2 over the 2-adic integers is an arithmetic dierential operator of order
one. The extended result for a general prime is a bit surprising, point
upon which we will elaborate in due course.
We organize our work as follows: in Chapter 2, we summarize the
construction of the p-adic numbers and the p-adic integers, describe its
topology as a metric space, its analytic and algebraic properties, and
study the (p 1)-roots of unity in it. In Chapter 3 we study some results
from classical analysis on Qp , including Mahlers theorem that estab-
lishes a bijection between the sets of restricted sequences and that of con-
tinuous functions on Zp , we present basic properties of the
ArtinHasse function, and study the analytic completion of the alge-
braic closure of Qp , the p-adic alter ego of the complex numbers that
result when we complete Q in the Euclidean metric instead. In Chapter
4 we introduce the set of analytic functions as a required preliminary to
our discussion later on. The arithmetic dierential operators make their
Introduction 7
rst appearance in Chapter 5, where we tie them to their associated ho-

momorphisms. This in turn allows us to prove that equation ( ) denes
the only arithmetic dierential operator over Zp since this ring carries
just one automorphism. Using it as a building block, we dene an arith-
metic dierential operator of any order. We discuss also the basic rings
that must be used in the theory when we have multiple primes, essen-
tially to indicate the additional diculties that arise then. In Chapter 6
we pause to dene arithmetic operators in general, developing succinctly
the theory of arithmetic jet spaces of Buium. In order to make things
easier for analysts not accustomed to algebraic concepts, we present a
list of the concepts from commutative algebra and schemes that are
needed in the development of the general theory. In the case of group
schemes, we discuss the characters that have been alluded to earlier, the
analogs in the theory of the linear dierential operators. And we outline
the theory for multiple primes also, in a succinct manner. In Chapter 7
we prove that all arithmetic dierential operators over Zp are analytic
functions. In Chapter 8 we study characteristic functions of p-adic discs
from the point of view of the theory of arithmetic dierential operators,
and prove that they are indeed dierential operators of an order depend-
ing upon the radii of the discs. The prime p = 2 manifests itself in a
rather special manner here, as we are able to prove by way of a direct
argument that the characteristic function of a discs of radius 1/2 over
the 2-adic integers is an arithmetic dierential operator of order one.
This work is carried out in standard coordinates, and leads to some for-
mal power series representations of the characteristic functions when the
prime in question is singular, a concept that we dene then. In Chapter
9 we work with harmonic coordinates, and improve signicantly upon
the result in the previous chapter, showing that all analytic functions on
Zp are arithmetic dierential operators, with the order being equal to
the level of analiticity. This last concept had made its rst appearance
earlier, in the context of Chapter 4. Finally, in Chapter 10, we exhibit
some fundamental dierences in the behavior of arithmetic dierential
operators that manifest when we work over the ring Z ur
p instead of Zp .
In particular, we indicate how to show that as soon as we adjoin one
unramied root of unity to Zp , the counterpart of the theorem above on
the characteristic function of discs no longer holds.
2
The p-adic numbers Qp
The eld Qp of p-adic numbers is the completion of the eld Q of rational

numbers with respect to the p-adic norm. In this chapter, we explain
their construction from various points of view, all, of course, equivalent
to each other.
Let p Z be a prime that we x hereafter. For a Z, we let ordp a
be the exponent of p in the prime factorization of a, that is to say, the
integer l such that a = pl r, where r Z is not divisible by p. This
notion is extended to a rational number q = a/b by setting ordp q =
ordp a ordp b, and the resulting function is multiplicative, that is to
say, it has the property that ordp q1 q2 = ordp q1 + ordp q2 . We then
dene the p-adic norm function on Q by
1
||q||p = . (2.1)
pordp q
We shall denote by dp the metric on Q that this norm induces.

In the resulting norm on Q, a rational q has ||q||p 1 if, and only if,
the denominator b of its reduced rational form a/b is not divisible by p.
Integers are closer to each other in the metric dp on Q the higher the
power of p that divides their dierence. So, for instance,
1
d5 (2, 1) = ||1 2||5 = 1, while d5 (2, 127) = ||2 127||5 = .
53
The p-adic norm satises a condition stronger than the triangle in-
equality. Indeed, if q = a/b and r = c/d, since the biggest power of p
that divides ad+bc is at least the minimum of the biggest power dividing
8
The p-adic numbers Qp 9
ad and the biggest power dividing bc, we have that

ad + bc
ordp (q + r) = ordp
bd
min {ordp ad, ordp bc} ordp b ordp d
= min {ordp a + ordp d, ordp b + ordp c} ordp b ordp d
= min {ordp a ordp b, ordp c ordp d}
= min {ordp q, ordp r} ,
and therefore,
1
||q + r||p = max {pordp q , pordp r } = max {||q||p , ||r||p } .
pordp (q+r)
(2.2)
The triangle inequality now follows readily. This stronger inequality
(2.2), referred to as the non-Archimedean property of p , produces
some geometric results that contrast a bit with those from our more
traditional point of view in Euclidean geometry. Triangles, for instance,
are all isosceles.
For let us assume that we have a triangle with vertices at 0, q and r,
respectively. We then know that ||q r||p max {||q||p , ||r||p }. If ||q||p <
||r||p , the non-Archimedean property of the norm implies that
||q r||p ||r||p .
Since ||r||p = ||q (q r)||p max {||q||p , ||q r||p }, it follows that
||r||p ||q r||p
also. Thus, ||r||p = ||q r||p and so, in the geometry generated by p ,
all triangles are isosceles, with the two largest sides equal to each other
in length. Sometimes we shall refer to this as the isosceles triangle
property of p .
Now we describe briey the general process that denes Qp as the
metric completion of Q in the distance dened by the p-adic norm. This
yields Qp as the unique complete eld, up to isometric isomorphism,
that contains a p -isometric dense copy of the eld Q.
We say that a sequence {qn } of rational numbers is Cauchy with re-
spect to the norm p , if for any real number > 0 there exists N such
that ||qn qm ||p < for all n, m N . We say that the sequence {qn } is
null if for any > 0 there exists N such that ||qn ||p < for all n N .
Given rational Cauchy sequences {qn } and {rn }, we dene their ad-
dition and multiplication by
{qn } + {rn } = {qn + rn } , {qn }{rn } = {qn rn } .
10 The p-adic numbers Qp
Let R be the set of all rational Cauchy sequences, and let M be the
subset of all null sequences. The operations above provide R with a ring
structure, and M becomes an ideal in R. In fact, M is a maximal ideal.
For if {qn } R is not null, there exists an > 0 and an integer N such
that ||qn ||p > for any n > N , and we may set

0 for nN,
rn = 1
for n>N.
qn
This is a Cauchy sequence also, and we have
{qn }{rn } = {0, . . . , 0, 1, 1, . . .} = {1, 1, . . .} {1, . . . , 1, 0, 0, . . .} .
If I were an ideal with M I properly, then I would contain a non-null

sequence {qn }. Since the sequence {1, . . . , 1, 0, 0, . . .} is null, the argu-
ment above with rn would imply that the constant sequence {1, 1, . . .}
must be contained in I, and so I would have to be equal to R. Thus, M
is a maximal ideal. The quotient eld R/M is, by denition, the eld of
p-adic numbers Qp .
An additional detail exhibits a fundamental dierence between this
construction of Qp and the analogous construction of R as the completion
of Q in the Euclidean norm. Given a p-adic number = {qn } + M , if the
Cauchy sequence {qn } is null, we set ||||p = 0. Otherwise, there exist
a positive real number and an integer N such that ||qn ||p > for any
n > N . We may choose N suciently large so that ||qn qm ||p < for
n, m > N also. Then, by the isosceles triangle property, we have that
||qn ||p is constant for all n > N , and so we may dene
||||p = lim ||qn ||p ,

n
extending in this manner the p-adic norm on Q to a p-adic norm on Qp .

Using this extension, we may also extend the notion of p-adic order.
In the construction of R as the completion of Q in the Euclidean norm,
this said norm admits an extension to a norm on R as well. However, in
spite of the fact that their constructions derive from exactly the same
procedure, this and the p-adic norm above exhibit a substantial dier-
ence. For unlike the case of the Euclidean norm on R, the extended
p -norm on Qp still ranges over the set {0} {pn }nZ , the same range
this norm function has over Q, whereas, for R, the range of the Euclidean
norm is R itself.
2.1 A pragmatic realization of Qp 11
We have the mapping

Q Qp = R/M
q {q} + M ,
where {q} stands for the constant sequence all of whose terms are equal
to q. It denes an isometric isomorphism onto its image, which by abuse
of notation we denote by Q also. It is dense in Qp . For if = {qn }+M
Qp , and > 0 is arbitrary, there exists some N such that ||qn qm ||p <
for n, m > N . Then for any n > N , the constant rational sequence
= {qn , qn , qn , . . .} is such that || ||p = limm ||qm qn ||p .
That Qp is complete follows by construction. If n is a Cauchy se-
quence in Qp , we may nd a sequence q1 , q2 , . . . in Q such that
1
||n qn ||p < , n = 1, 2, . . . .
n
Since
||qn qm ||p ||qn n ||p + ||n m ||p + ||m qm ||p ,
the said rational sequence is Cauchy also, and its limit in Qp exists. We
set limn qn = , so = {qn } + M . Since
|| n ||p || qn ||p + ||qn n ||p ,
we see that the original sequence n has limit in Qp .
2.1 A pragmatic realization of Qp

It is convenient to have a pragmatic description of the elements of Qp
that allows us to carry calculations with some ease. We show one of these
next, describing a p-adic number in a way analogous to the description
of a real number by, say, its decimal expansion.
Let Qp . Since its p-adic norm is some integer power of p, we may
write as
n
= pn u , where ||||p = ||p||p , ||u||p = 1 . (2.3)
The unit u is the limit of some rational sequence {rj }. Let N be an
integer such that ||u rj ||p < 1 for all j N . By the isosceles triangle
property, we have that ||u||p = ||rj ||p = 1 for these js. Thus, rj is a
rational unit, and
u + M = rj + M .
We set qn = rN .
Since qn is a rational unit, qn = an /bn , where an and bn are relatively
prime to p. In particular, there exist integers v, w such that
vbn + wp = 1 ,
and v is the inverse of bn mod p. Therefore,

an an (1 vbn )
qn an v = an v = ,
bn bn
and the right side has p-adic norm less than 1.
We denote by P the set of p-adic numbers of p-adic norm less than 1.
If we set cn = an v Z, we then have that
u + P = qn + P = cn + P
and
||u cn ||p < 1 , || cn pn ||p < ||pn ||p .
Thus,
= upn = cn pn + (u cn )pn = cn pn + 1 ,
n
where 1 = (u cn )pn is a p-adic number such that ||1 ||p < ||p||p . We
m
must then have ||1 ||p = ||p||p for some m > n, and we can iterate the
procedure started in (2.3) with the role of now played by 1 . Since
n+j
||p||p 0 as j , we conclude the following.
Proposition 2.1 Any p-adic number can be written in the form

= cj pj , (2.4)
j=n
n
where each cj is an integer, and ||||p = ||p||p . The representation is
unique if each coecient cj is chosen in the range 0 cj p 1.
The expansion
cn cn+1
= + n1 + + c0 + c1 p + c2 p2 +
pn p
is usually abbreviated as
= cn cn+1 . . . c0 .c1 c2 . . . .
2.2 The p-adic integers Zp and their eld of fractions 13
Example 2.2 We nd the expansion of 1/5 in Q3 . This rational has

3-adic norm 1, and the only integer c0 in the range 0 c0 < 3 such that
the 3-adic norm of 1/5 c0 is less than 1 is c0 = 2. We have

1 1 9 1
=2+ 2 = 2 = 2 32 . (2.5)
5 5 5 5
Hence, the coecient c1 is zero, c2 = 1 because

1 1 2
= 1 + 1 = 1 3,
5 5 5
and we have
1 2
= 2 + 1 32 33 .
5 5
Now we see that
2 4 1
= 2 3 = 2 + 1 3 33 ,
5 5 5
and therefore
1 1
= 2 + 1 32 + 2 33 + 1 34 36 .
5 5
Thus, c3 = 2, c4 = 1, c5 = 0, and comparing with (2.5), we conclude
that c6 = c2 , and that the 3-adic expansion of 1/5 must be periodic,
with period 0121. In other words, we have
1
= 2.01210121 . . .
5
in Q3 .
Using the fact that

1 plk
1 + pk + p2k + + p(l1)k =
1 pk
1
converges p-adically to 1p k , and ideas similar to the one above, we
can show that Qp is an element of Q if, and only if, its canonical
expansion (2.4) is periodic.
2.2 The p-adic integers Zp and their eld of fractions

The set Zp of p-adic integers is dened by
Zp = { Qp : ||||p 1} ,
the set of p-adic numbers whose canonical expansion (2.4) begins at

n = 0 or above. By the non-Archimedean property (2.2) of the p-adic
norm, Zp is a subring of Qp . From this point on, the elements of Z will
be referred to as rational integers, reserving the term integer for the
elements of Zp instead.
We shall say that two p-adic numbers , are congruent mod pn if
|| ||p pn , and write mod pn . This is equivalent to saying
that ( )/pn Zp , that is to say, the expansion of contains
a nonzero coecient no sooner than in the pn -th position. Clearly, this
notion extends the congruence mod pn over Z.
We denote by pn Zp the ideal generated by pn in Zp . The congruence
mod pn simply says that pn Zp . The set P of p-adic
numbers of norm less than 1 used in the previous section is just pZp , a
maximal ideal in Zp .
n1
If = j=0 cj pj is an element of Zp , we let an = j=0 cj pj . This
denes a sequence of rational integers, and

an = pn cj pjn = pn
j=n
where Zp . The said sequence {an } Z satises the congruences
an mod pn , n = 1, 2, . . . (2.6)
and
an+1 an mod pn ,
for each n 1. Any p-adic integer satises an innite set of congruences

of this form, and these congruences characterize the p-adic integer in
question.
Example 2.3 Let us consider the sequence of rational integers given

by

n1
an = 3 10j .
j=0
We have that an+1 an = 3 10n , and so an+1 an mod 5n . Thus, the

said sequence yields a 5-adic number as n . This limit is in fact the
number 1/3. For we have that 3an = 10n 1, and so ||3an + 1||5 0.
Of course, we could have used the prime 2 instead, and show that the
said sequence converges to 1/3 in Q2 also.
2.3 The topology of Qp 15
The discussion above can be reversed [42]. For if An = Z/pn Z is the

ring of classes of integers mod pn , we obtain an obvious and natural
surjective homomorphism
n : An An1 ,
with kernel pn1 An , and the sequence
An An1 A2 A1 (2.7)
forms a projective system indexed by the whole numbers. Then we
may dene Zp to be the projective limit of the system (An , n ) above,
Zp = lim
(An , n ), where by denition, an element of this limit is a
sequence {an } with an An such that n (an ) = an1 . Addition and
multiplication are dened component-wise. If each An is given the dis-

crete topology, Zp is a subring of the space An endowed with the
product topology, which is compact. Thus, Zp inherits a topology that
makes it into a compact space, being a closed subspace of the compact
ambient space where it is embedded.
If n : Zp An is the n-th component function, we have the short
exact sequence
pn
0 Zp Zp
n
An 0 ,
which permits to realize the identication of Zp /pn Zp with An = Z/pn Z.
The congruences (2.6) exhibit explicitly this identication.
The eld Qp is the eld of fractions of Zp , Qp = Zp [p1 ].
2.3 The topology of Qp

As a metric space, the topology of Qp is given by the basis of open sets
consisting of discs of the form
a + pn Zp = D(a, pn ) = {x Qp : ||x a||p 1/pn }
for any center a Qp and radius pn .
Lemma 2.4 If b D(a, pn ), then D(a, pn ) = D(b, pn ).
In other words, every point of the disc D(a, pn ) is a center.
Proof. If x D(a, pn ), the non-Archimedean property of p implies
that ||x b||p = ||x a + a b||p max {||x a||p , ||a b||p } pn , and
so D(a, pn ) D(b, pn ). The opposite inclusion follows by reversing
the roles of a and b in this argument.
It then follows that if two discs have a nontrivial intersection, one of

them must be contained in the other. We thus see that all of the basis
elements of the topology of Qp are also closed sets; for given any a Qp ,
we have that Qp \ (a + pn Zp ) = aQp \(a+pn Zp ) (
a + pn Zp ) is a union of
basis elements.
Given a subset X of Qp , we shall sometimes let DX (a, pn ) stand for
the relative disc D(a, pn ) X.
The ring of integers Zp inherits the subspace topology, with its basis of
open sets given by {(a + pn Zp ) Zp }aQp . Since the rational integers are
dense in Zp , and every point of a disc is a center, the countable family
{(a + pn Zp )}nZ
aZ,0a<pn constitutes a basis for the topology of Zp also.
If n is any nonnegative rational integer, the collection
n
1
{DZp (a, pn )}pa=0
forms a covering of Zp by discs that are pairwise disjoints. Below we
shall use this fact on several occasions.
We have observed the compactness of Zp earlier, when we derived
the p-adic integers as a projective limit. Since that argument makes
use of Tychonos theorem, we pause to provide a simpler proof of the
compactness of Zp proceeding directly, from the denition.
Proposition 2.5 The space Zp is compact. The space Qp is locally
compact.
Proof. We need to prove that any covering of Zp has a nite subcov-
ering. Since open sets are unions of discs, it suces to show that an
arbitrary covering of Zp by discs {D(ai , pni )}iI
ai ,ni Z,0ai <pni , where I
is some index set, has a nite subcovering. We prove that there exists a
nite subset I of I such that {D(ai , pni )}iI covers Zp also.
If at least one of the ni s is less or equal than zero, the radius of the
corresponding disc would be greater or equal than one, and by Lemma
2.4, that single disc would suce to cover Zp . Thus, we only need to
consider the case where ni > 0 for each i I. Under this assumption,
= inf{ni : i I}. Then, there exists at least one disc in the given
let n
covering of radius pn .
For each rational integer a, let na be an integer such that a is contained
in a disc of the given covering of radius 1/pna , and dene
N = max{na : 0 a pn 1} .
We know that
Zp = pj=01 DZp (j, pN ) ,
N
and by Lemma 2.4 and the remark immediately after its proof, we con-
clude that for each of the discs D(j, pN ) there exists an index ij I
such that D(j, pN ) D(aij , pnij ). If I = {i0 , . . . , ipN 1 }, we then
have that
Zp iID(ai , pni ) ,
proving the desired result.

That Qp is locally compact follows since the linear mapping x
a + pn x is a homeomorphism between Zp and the disc D(a, pn ), respec-
tively.
Remark 2.6 Compactness of Zp can be proven in yet another way,

by showing that Zp is sequentially compact, which on metrizable spaces
is equivalent. Indeed, given any sequence {n } in Zp , we can construct
a subsequence all of whose elements have the same rst digit in their
p-adic expansions (2.4), and by iteration of this argument k times, nd a
subsequence {nk j } j=1 such that the rst k digits in the p-adic expansions
of all the nk j s coincide. We then diagonalize to dene a subsequence
{nj j } of {n } that converges by choosing the rst element of the rst
subsequence, the second of the second, and so on.
Lemma 2.7 The set of p-adic numbers Qp is totally disconnected.
Proof. Let S be any subset with more than one point. If x and y are
two elements of S, then we have that ||x y||p > 0, and if D = {z
S : ||z x||p < ||x y||p }, then the pair of open sets D, S \ D forms a
separation of S.
2.4 Analytic and algebraic properties of Qp

Let p and p be two distinct primes. Since the sequence {pn } converges in
Qp but not in Qp, as topological elds, Qp and Qp are not isomorphic.
We show here that, in fact, they are not even isomorphic as elds. This
shall follow by a simple analysis of solutions to quadratic equations.
Let us consider the polynomial f (x) = x2 a, where a Z is not
divisible by p2 . If p divides a then f cannot have a root in Qp . For if

= a is in Qp , since ||a||p = || a||p || a||p , we must have that || a||p =
1
||p||p2 , which is not possible because p ranges over integer powers of
||p||p .
If p does not divide a and we can nd a p-adic number such that
||f ()||p < 1, by the congruences (2.6) for , we can nd a1 Z such

that
a21 a mod p , (2.8)
which says that a is a quadratic residue mod p. If p is an odd prime, this
equation has two solutions in the range 1 a1 p 1. Proceeding by
induction, we nd two sequences of rational integers {an } that converge

to p-adic numbers a such that a an mod pn , and where the starting
element a1 is equal to each one of the said solutions of the congruence
(2.8), respectively. Thus, f has two distinct roots in this case.
If a is a quadratic nonresidue mod p, then f cannot have a root Qp .
For otherwise, there would exist c Z such that
+ pZp = c + pZp ,
and so,
a + pZp = 2 + pZp = c2 + pZp ,
which implies that
c2 a mod p .
Thus, a would be a quadratic residue, contrary to the original assump-
tion.
The argument given above is truly built on the p-adic version of New-
tons iteration algorithm. We provide some of the details of this in order
to explain the extra care that must be exercised in our argument in the
case when the prime p is 2.
Indeed, if f is any monic polynomial in Zp [x], and we have a1 Z
such that ||f (a1 )||p < 1, then the sequence given by

an+1 = an (f (an ))1 f (an ) (2.9)
is expected to converge to a root of f under suitable
hypothesis
on the

p-adic norm of f (a1 ). Indeed, if we assume that f (a1 ) = 1, by the
p
Taylor expansion

f (x + h) = f (x) + hf (x) + h2 ef (x, h) ,
we see that

2

f (a1 ) f (a1 ) f (a1 )
f (a2 ) = f (a1 ) f (a1 ) + ef a1 , ,
f (a1 ) f (a1 ) f (a1 )

where the error ef (a1 , f (a1 )/f (a1 )) has p-adic norm bounded above
2
by 1. Then, ||f (a2 )||p ||f (a1 )||p , and ||a2 a1 ||p ||f (a1 )||p . Proceeding
by induction, we see that
n1
2n
||an+1 an ||p ||f (a1 )||2p , ||f (an+1 )||p ||f (a1 )||p .
Thus,
a = a1 + (a2 a1 ) + (a3 a2 ) + = lim an
n
converges p-adically, and the limit is a p-adic integer root of f (x).

This argument must be modied slightly when dealing with the prime
p = 2 and the quadratic polynomial f (x) = x2 a. For in this case,

f (a1 ) = 2a1 will have relatively small norm, which aects the conver-
Butit does not prevent it. For if f (a1 )
gence rate of the iteration scheme.

has norm less than 1, and 0 = f (a1 ) < 1, as long as we assume that
p

f (a1 )/f (a1 )2 has p-adic norm less than 1, the iteration scheme above
produces a sequence that converges p-adically to a root of f .
By applying this iteration scheme to our quadratic polynomial for all
possible choices of the prime p, we obtain the following result.
Theorem 2.8 Let a be any rational integer not divisible by p2 . Con-

sider the equation x2 a = 0. Then:
a). If p | a, the equation has no solution in Qp .

b). If p = 2 and p a, the equation has two solution in Qp if a is a
quadratic residue mod p, and no solution if it is not.
c). If p = 2 and 2 a, the equation has two solutions in Qp if a is
congruent to 1 mod 8, and no solution if it is not.
We now reconsider the distinct primes p and p of the beginning, with

say, p < p. If these primes are both odd, and p is a quadratic residue
mod p, the equation x2 p = 0 has two solutions in Qp, but none in Qp .
If, on the other hand, p is a quadratic nonresidue mod p, we may choose
a quadratic nonresidue n mod p with n < p. Then np is a quadratic
residue mod p, and the equation x2 np = 0 has two solutions in Qp
but none in Qp .
The case where p is an odd prime and p = 2 is treated similarly. If 2
is a quadratic residue mod p, the equation x2 2 = 0 has two solutions
in Qp but none in Q2 . If 2 is a quadratic nonresidue mod p, let n be an
odd quadratic nonresidue mod p. Then the equation x2 2n = 0 has
solutions in Qp but none in Q2 .
We thus conclude the following.
Theorem 2.9 If p and p are distinct primes, the elds Qp and Qp are
not isomorphic.
Remark 2.10 The Newtons iteration algorithm on Zp discussed above

is commonly known as Hansels lemma. A convenient formulation that
can be used to nd roots of polynomial equations in several variables goes
as follows: Let F (x) = (f1 (x), . . . , fn (x)) with fj (x) = fj (x1 , . . . , xn )
Zp [x1 , . . . , xn ], j = 1, . . . , n. Let a1 Znp be such that F (a1 ) 0 mod p
and such that the Jacobian matrix F (a1 ) = F (x)
|x=a1 is invertible. If
||F (a1 )||p 1 and ||F (a1 )||p (F (a1 )) F (a1 )p < 1, then there exists

1 1
a unique a Znp such that F (a) = 0 and a a1 mod p.

The proof follows along the lines outlined above in the case of a single
variable, where starting from the approximate solution a1 , the elements
in the sequence (2.9) brings us closer to the actual root at every step.
As an example, let us consider the quadratic polynomial f (x, y, z) =
3x2 + y 2 + 4z 2 + 2(xy + xz + yz), and let p be the prime 7. For a1 =
(0, 1, 1) Z37 we have that f (a1 ) = 7 0 mod 7 and f (a1 ) =
(4, 4, 10). In order to analyze the roots of f by way of the stated
result, we consider the polynomial function F : Z37 Z37 given by
F (x, y, z) = (f (x, y, z), y 6, z 6). Then the sequence (2.9) yields a2 =
a1 + (F (a1 ))1 F (a1 ) = (91/4, 6, 6), and we have that F (a2 ) = (3 72
137/24 , 0, 0). Now we set a3 = a2 (F (a2 ))1 F (a2 ) = (6, 937/600, 6, 6),
and we have that F (a3 ) = (74 1372 /(26 354 ), 0, 0). Iterating this proce-
dure, we obtain a Cauchy sequence {aj } in Z37 such that ||F (aj )||7 1/7j ,
and the continuity of polynomial functions will imply that the limit a
in Z37 is a root of F , and so a root of its rst coordinate function f . By
construction, a a1 mod 7.
2.5 (p 1)-roots of unity in Qp

The eld Qp contains all (p 1)-roots of unity. Indeed, the equation
xp1 1 = 0
has exactly p1 roots in Qp , the maximum number of possible solutions,

as we now see.
By Eulers theorem, given any integer a in the range from 1 to p 1,
n+1 n
we have that a(p ) = ap (p1) 1 mod pn+1 . Thus, the series
n 2
a = lim ap = a + (ap a) + (ap ap ) +
2.5 (p 1)-roots of unity in Qp 21
converges. For using the non-Archimedean property of p , it suces

to show that
n+1 n n n
ap ap = ap (ap (p1)
1) ,
goes to zero as n goes to , which is so as the last factor on the right
of this expression is divisible by pn+1 . We also have that
n
ap1 = lim ap (p1)
= 1,
and so a is a (p 1)-root of unity in Qp .
If two of these roots were to coincide,
a + pZp = a + pZp ,
we would have that a a mod p, and so a would have to be equal to a.
p1
Thus, the collection of p-adic numbers {a }a=1 exhausts the set of all
(p 1)-roots of unity.
We summarize our discussion in the form of a theorem. We let Z p
denote the subgroup of p-adic units in Zp , and let Fp be a canonical
realization of a nite eld with p-elements. We denote by Fp the multi-
plicative group of nonzero elements of Fp .
Theorem 2.11 Consider 1 + pZp as a multiplicative subgroup of Zp ,
and let (Zp ) denote the multiplicative subgroup of roots of unity, {x
Zp : xp1 = 1}. Then we have that Z p = (Zp ) (1 + pZp ), and (Zp )
is the unique subgroup of Z
p isomorphic to F
p.
We have the eld identications Zp /pZp = Z/pZ

= Fp .
Let us now recall that for a nite eld Fq with q elements and char-
acteristic p, the Frobenius mapping is given by

Fq Fq
x xp .
It denes the unique automorphism of Fq that xes the subeld Fp , and
is a generator of Aut(Fq ). Since xp = x for any x Fp , the Frobenius
mapping reduces to the identity when q = p. This fact implies that the
identity is the unique lift to Zp of the p-th power Frobenius isomor-
phism on Fp Zp , and in turn, we may now derive the following simple
consequence, which will be of crucial importance to us later on, when
we in fact introduce the arithmetic dierential operators on Zp . The
reader may nd our discussion a bit odd at this point, since after all,
the identity is the unique ring automorphism on Zp . This oddity will be
claried later, and the reader should merely keep in mind now that the
study arithmetic dierential operators can be done with more generality

[8, 6, 15] over rings larger than Zp , and where the lift of the p-th power
Frobenius isomorphism is not the identity.
Theorem 2.12 For all x Zp , we have that x xp pZp .
Proof. We have that x = x + pu for elements u and x in Zp , x

p
a p 1 root of unity. Since 0 mod p for 1 j < p, we then
j
p p
have that x = x + pZp = x + pZp = x + pZp , and the desired result
follows.
3
Some classical analysis on Qp

The convergence of a p-adic series n=0 an is somewhat easy to ana-
lyze. The convergence of the series implies that limn an = 0, but the
remarkable fact is that this condition alone implies the converse. For if
n
sn = j=0 aj is the sequence of partial sums, by the non-Archimedean
property of the norm we have that
||sn sm ||p = ||an + + am+1 ||p max ||aj ||p 0 ,

m+1jn
showing that {sn } is a Cauchy sequence, and so its limit exists in Qp .

1/j
Using this same argument, it follows that if r = 1/ lim sup ||aj ||p , the
power series

F (x) = an xn Qp [[x]] (3.1)
n=0
converges if ||x||p < r, diverges if ||x||p > r, and the case ||x||p = r is
decided by simply looking at lim ||aj ||p rj , which implies convergence or
divergence contingent upon it being zero or not.
Let us assume that the series F (x) in (3.1) converges, and so we have

that limn an xn = 0. Let F (x) denote the formal power series of
term-by-term derivatives,

F (x) = nan xn1 .
n=0

Since nan xn1 p an xn1 p 0, the series F (x) converges also.

Furthermore, F (x) is the limit of the usual quotient of increments
23
24 Some classical analysis on Qp
(F (x + y) F (x))/y. For, by the binomial expansion, we have that

n

F (x + y) F (x) n
= an xnk y k1 ,
y k
n=1 k=1
and if ||y||p < ||x||p , we have

an n xnk y k1 ||an || ||x||n1 0 .
k p p
p
Hence, the series above converges uniformly in y so long as y is p-adically

smaller than x, and the innite summation and limit as y 0 can be
intertwined. We obtain
n

F (x + y) F (x) n
F (x) = lim = an lim xnk y k1
y0 y y0 k
n=1 k=1

= nan xn1 = F (x) ,
n=1
as desired.
By iteration we conclude that a convergent power series has convergent
power series derivatives of all order.
In a sense to be made precise below, the polynomial functions x

x
play the role in p-adic analysis that the monomials xn do in real
n
analysis. We lead to the discussion of this assertion, a theorem of Mahler
[36], by rst showing the following result (see [2]).
Theorem 3.1 Let y Qp be a p-adic number such that ordp y 0.

Then the binomial series

y y
(1 + x) = xn
n
n=0
converges for all x with ordp x > 1/(p 1).
Notice that, for all p other than 2, 1/(p 1) Z, and so the lower
bound we give above for the p-adic order of x appears to be of a rather
peculiar nature. The reason for doing so is simple. The binomial series
function can be dened also on the completion of the algebraic closure
of Qp , a rather large eld to which the p-adic norm and order admit
extensions (see 3.2 below). The function there dened will converge
precisely under the condition stated in the theorem.
Some classical analysis on Qp 25
Proof. We have

y y(y 1) (y n + 1)
= .
n n!

y
Since the p-adic order of y is nonnegative, the p-adic order of is
n
bounded below by that of 1/n!, and we conclude that

y xn
ordp xn ordp .
n n!
Let n = a0 + a1 p + + ak pk be the base p expansion of n. We dene

sn = a0 + + ak , and set s0 = 0. The p-adic order of n is the smallest
integer l such that al = 0, and we see easily that sn1 = l(p 1) + sn 1,
so ordp n = (sn1 sn + 1)/(p 1). By the multiplicative property of
ordp , it follows that
n sn
ordp n! = . (3.2)
p1
Hence,
xn
ordp = n ordp x ordp n!
n!
n sn
= n ordp x (3.3)
p 1
sn
= n ordp x p11
+ ,
p1
which grows with no bounds under the stated hypothesis, and so the
n-th term of the series dening (1 + x)y converges p-adically to zero.
The result follows.
We may show that the function of y that we obtained above is con-
tinuous. For all y in Zp , the argument

given shows that if x has p-adic

y
order greater than 1/(p1), we have xn ||xn /n!||p 0, and
n p
the series dening (1 + x)y converges uniformly as a function of y. But
uniform convergence of continuous functions yield a continuous limit. So
the function (1 + x)y is continuous as function of y Zp . We address
this type of issues in further detail now.
We restrict our attention to the space C(Zp , Qp ) of Qp -valued con-
tinuous function on Zp . Since Zp is compact, any f C(Zp , Qp ) is
uniformly continuous. If we were to consider C(D, Qp ) instead, where
D = D(a, pn ) is the disc in Qp of radius pn centered at a, a similar
result to the one below would still hold since D(a, pn ) is homeomorphic
to Zp .
Since the nonnegative rational integers form a dense subset of Zp , a
function f C(Zp , Qp ) is completely determined by its values on the
set {0, 1, 2, . . .}.
Given any f C(Zp , Qp ), let us consider the sequence

n

n
an = (1) k
f (n k) Qp , (3.4)
k
k=0
and the associated power series

x
F (x) = an Qp [[x]] . (3.5)
n
n=0

m
For any nonnegative rational integer m, = 0 if n > m. Then
n

m

m
ml

m k k+l m
F (m) = an = f (l) (1) .
n l k+l
n=0 l=0 k=0
Since

k+l m m ml
= ,
l k+l l k
we obtain

mm ml k ml
F (m) = l=0 f (l) k=0 (1)
l k

m m
= l=0 f (l)(1 1)ml
l
= f (m) .
Thus, the power series (3.5) interpolates the values of the function f over
the dense subset of nonnegative rational integers. Therefore, if we were to
prove that this series converges and denes an element of C(Zp , Qp ), the
said element will have to be f (x) itself. We proceed to prove exactly that
in a slightly oblique but succinct manner. The end result is a theorem
due to K. Mahler [36].

x
Let us notice that the polynomial functions x map rational
n
integers to rational integers, and so dene elements of C(Zp , Zp ). In
what follows, we shall say that a sequence {an } n=0 Qp is restricted if
an 0 p-adically.
Theorem 3.2 (Mahler [36].) Let S = {{an }

n=0 Qp , an 0} be the
set of restricted sequences. The mapping
S C(Zp , Qp )

x (3.6)
{an } n=0 an
n
is a bijection.
Given a restricted series {an }, the mapping (3.6) in this statement is

given by the associated series (3.5). We show rst that this mapping is
well dened.
The elementary combinatorial identity

n

x xy y
=
n k nk
k=0
holds for p-adic integers x, y also. This can be shown by using Theorem
3.1 to determine the coecient of tn in the series expansion of the iden-
tity (1 + t)x = (1 + t)xy (1 + t)y . Given an arbitrary integer x in Zp , let
y be a nonnegative rational integer such that

x y

n! 1 .
p
We have

y

n k 1 ,
p
as the argument to the function p on the left side is just the standard
binomial coecient. On the other hand,

xy (x y)(x y 1) (x y k + 1) xy
= = uk ,
k k! k!
where uk Zp , and therefore,

x y
1 .
k
p
By the non-Archimedean property of the p-adic norm, we conclude then

that

n

x x y y
1 ,
n = k n k
p k=0 p
estimate that implies that the series in the statement of the theorem de-
nes a continuous function, and so the map (3.6) is well-dened. Indeed,
the series converges because the said estimate implies that

an x ||an || 0 .
n p p
On the other hand, the sequence of partial sums are polynomials, hence
continuous, and the limit, being uniform, must be also a continuous
function.
The injectivity of (3.6) can now be argued with ease. Indeed, let us
consider a restricted sequence {an } such that its associated power se-
ries F (x) in (3.5) is the function identically zero. By the combinatorial
identity

x+1 x x
= +
n n n1
that holds for any p-adic integer x also, we have that

x
0 = F (x + 1) F (x) = an .
n1
n=1
We now observe that the value of the continuous function F (x+1)F (x)
at x = n Z0 is equal to an+1 , and so proceeding by induction, we
conclude that an = 0 for all n 1. So F (x) reduces to the zero term a0 ,
and this must be zero also since, by assumption, so is F .
In order to nish the proof of Theorem 3.2, we are left with the task
of showing the surjectivity of (3.6). So let us take f C(Zp , Qp ). By
compactness of Zp , f is bounded and so, after multiplication by a suit-
able power of p, we obtain a function that ranges in Zp . Thus, without
loss of generality, we may assume that f is an element of C(Zp , Zp ). We
view the latter as a Banach space with the supremum norm.
By completeness of C(Zp , Zp ), the desired surjectivity follows if we
show that for any N 0 there exist {an } Zp and fN C(Zp , Zp )
such that

x
f (x) = an + pN fN (x) .
n
n
We argue this by induction on N , starting with f0 = f , cf. [34], pp.

99-100.
Notice that
it suces
to prove that fN is, modulo p, a linear combi-
x
nation of the s. By the uniform continuity of fN , there exists M
n
such that
1 1
||fN (x) fN (y)||p for all x, y Zp such that ||x y||p .
p pM
Hence, the map

fN
Zp Zp Fp = Zp /pZp
factors through
Z/pM Z Fp .
It is rather elementary to see that for any k in the range 0 k < pM ,

the map

x
k
Zp Zp Fp
factors through
Z/pM Z Fp
also. Since Fp is discrete and nite, the desired task will be accomplished
if we show that the set of tuples {(
a0 , . . . , a i Fp }, and the set
pM 1 ) : a
of mappings Maps(Z/pM Z, Fp ) are in 1-to-1 correspondence with each
other via the function
pM 1

x
pM 1 )
a0 , . . . , a
( a
n .
n
n=0
The said sets have the same number of elements. On the other hand,
we can prove the injectivity of the map above by an argument similar
to that used to prove the injectivity of (3.6) itself. This completes the
proof of the Theorem.
Example 3.3 Consider the continuous function Z5 y (1 5)y

given by Theorem 3.1. We have that

y
(1 5) =y
(5) n
,
n
n=0
where the series on the right converges in Z5 .

3.1 The ArtinHasse exponential function

Using (3.2) we derived identity (3.3). It follows that if ordp x > 1/(p1),
ordp (xn /n!) , and the exponential series

xn
ex =
n=0
n!
converges for these xs. If otherwise x is such that ordp x 1/(p 1), we
may take n = pk . We then have that sn = 1, and ordp (xn /n!) ,
so the series above does not converge p-adically in this case. Thus, the
1
exponential function has p-adic radius of convergence equal to p p1 .
Bigger and bigger denominators in the coecients of the series help its
convergence when we use the classical Euclidean norm, but the opposite
is true p-adically when the denominators in question are divisible by
larger and larger powers of p. This makes the p-adic radius of convergence
of the exponential function relatively small compared to its radius of
convergence over Q .
Let (n) be the Mobius function, dened by

(1)k if n is the product of k distinct primes,
(n) =
0 otherwise .
We recall that by the prime factorization theorem, we have that

(d) = 0 for n > 1 . (3.7)
d|n
Lemma 3.4 We have the identity

(n)
ex = (1 xn ) n .
n=1
Proof. Taking log on the right side, we obtain that

(n) (n) xnm xj
log (1 xn ) = = (d) = x ,
n=1
n n=1
n m=1 m j=1
j
d|j
where the last equality follows by (3.7).

The identity in Lemma 3.4 shows that the failure for the series dening
ex to converge p-adically arises from the terms where n is divisible by p
and square-free. This leads to the following natural denition:
Denition 3.5 Given a prime p, the ArtinHasse exponential function

3.1 The ArtinHasse exponential function 31
is dened by

(n)
Ep (x) = (1 xn ) n .
n=1
pn
It follows that Ep (x) 1 + xZp [[x]], and therefore, Ep (x) converges

for ||x||p < 1.
As in the proof of Lemma 3.4, we may apply the log function to the
innite product dening Ep (x). If we then use the identity

1 if n is a power of p,
(d) = (3.8)
0 otherwise ,
d|n
pd
we see that

pj
x
Ep (x) = exp j
. (3.9)
j=0
p
The expression above may be used as an alternative denition for

Ep (x), in which case this function will be the exponential of a power
series in Q[[x]]. In so doing, it becomes harder to conclude that Ep (x)
is in eect an element of Zp [[x]], as the coecients of the series whose
exponential yields Ep (x) have norms that grow without bound in Qp .
We address this issue next, and derive the p-adic integrality property
of the coecients of the said series as a consequence of a beautiful and
remarkable result in p-adic analysis due to Dwork, which we take the
opportunity to state and prove. In spirit, this result is fundamental for
the theory of arithmetic dierential equations. It studies the interplay
between the p-th power of the function of a number and the function of
the p-th power of the number.
We begin this line of reasoning by rst showing the relationship be-
tween Ep (xp ) and Ep (x)p .
Lemma 3.6 We have the identity
Ep (xp ) = Ep (x)p exp (px) .

xp
j
Proof. Let us set L(x) = j=0 pj . Then we have that
Ep (x) = exp (L(x)) ,

and the desired identity amounts to showing that L(xp ) = pL(x) px,
which is clear.
By Lemma 3.6, we see that

Ep (xp ) np
n
n np
n1
= exp (px) = (1) x = 1 + px (1) xn .
Ep (x)p n=0
n! n=1
n!
Now by (3.2), we see that

n1
p n sn (n 1)(p 2) + sn 1
ordp = n1 = ,
n! p1 p1
which is nonnegative. Hence, exp (px) 1 + pxZp [[x]], and we have
that
Ep (xp )
1 + pxZp [[x]] . (3.10)
Ep (x)p
In other words, modulo pZp [[x]], the rational series Ep (x) in (3.9) com-
mutes with the p-th power map. This is amenable for the application of
the following result.

Lemma 3.7 (Dworks lemma) Let f (x) = ai xi 1+xQp [[x]]. Then
f (x) 1 + xZp [[x]] if, and only if, f (xp )/(f (x))p 1 + pxZp [[x]].
This can be stated in a more general context (cf. Lemma 2.4 in [15]),
but that will not be used in here.
Proof. If f (x) 1 + xZp [[x]] and since (a + b)p ap + bp mod p, by
Theorem 2.12 we have that f (x)p = f (xp )+pg(x) where g(x) xZp [[x]].
Then f (xp )/(f (x))p = 1 p(g(x)/f (x)p ), which belongs to 1 + pxZp [[x]]
because f (x)p 1 + pxZp [[x]] has an inverse in the multiplicative group
1 + pxZp [[x]].
Conversely, let us assume that f (xp ) = (f (x))p g(x) with g(x) =

bi xi 1 + pxZp [[x]]. Since by hypothesis the leading coecient is
1 for both, f and g, we have that

p

pi i i
1+ ai x = 1 + ai x 1+ bi x . (3.11)
i=1 i=1 i=1
with the bi s are all rational integers congruent to zero mod p. We use
induction to show that ai Zp for all i.
Let us assume that ai is a rational integer for all i < n. If p n, the
coecient of xn in the right side of (3.11) is given by an expression of
the form
pan + R(a1 , . . . , an1 , b1 , b2 , . . . , bn )
where R(a, b) is a p-adic integer that depends nonlinearly on a and lin-

early on b. Since b is congruent to zero mod p, we have that R(a, b) pZp .
By comparison with the left side of (3.11), this coecient must be equal
to zero, and so we have
1
an = R(a1 , . . . , an1 , b1 , b2 , . . . , bn ) Zp ,
p
as desired. If on the other hand, p | n, the coecient of xn in the right
side of (3.11) is given by an expression of the form
pan + apn|p + R(a1 , . . . , an1 , b1 , b2 , . . . , bn ) ,
where R(a, b) is a p-adic integer in pZp for the same reasons as before.
In this case, by comparison with the left side, we have that
an|p = pan + apn|p + R(a1 , . . . , an1 , b1 , b2 , . . . , bn ) .
By Theorem 2.12, we conclude that

1
an = an|p apn|p R(a1 , . . . , an1 , b1 , b2 , . . . , bn ) Zp ,
p
as desired.
Dworks lemma applies to the function dened by (3.9), with (3.10)
showing that the required hypothesis for its application holds. We con-
clude that Ep (x) 1 + xZp [[x]]. This is a bit of miracle if we look at
the right side of (3.9), which realizes Ep (x) as the exponential of a series
whose coecients diverge p-adically.
One of the results that the alluded miracle for (3.9) encodes is Wilsons
theorem. Indeed, the coecient of xp in this series is given by
1 1 1 + (p 1)!
+ = ,
p! p p!
and since all the coecients of the series are in Zp , among them this
particular one, we conclude that (p 1)! 1 mod p.
Since Ep (x) Zp [[x]], the series converges on any disc of radius less
than 1. This is precisely the disc of convergence of Ep (x).
3.2 The completion of the algebraic closure of Qp

By Theorem 2.8, we may conclude readily that Qp is not an algebraically
closed eld. Let Qp be its algebraic closure, the union of all nite eld
extensions of Qp . We see rst how to extend the p-adic norm and p-adic
order to Qp .
Let K be a nite eld extension of degree n of a normed eld F , which
is assumed to be locally compact. We then have that K is a vector space
of dimension n over F , and in such case, we see that norms on K must
be equivalent to each other. If K is provided with a eld norm, the norm
of the power of an element is the said power of the norm of the element.
Using this property, we can see easily that all norms on K extending the
norm on F must be, in fact, equal to each other. Thus, there exists at
most one eld norm on K that extends the norm on F .
Let be an algebraic number of degree n over a eld F . We denote
by
i (x) = xn + a1 xn1 + + an1 x + an
the unique monic irreducible polynomial with coecients in F that is

annihilated by . For the nite eld extension K = F (), we dene
NK/F () = (1)n an .
n
This number can be cast also as i=1 i where the i s are all the conju-
gates of 1 = over F , or as the determinant of the automorphism of K
given by multiplication by . In particular, when K = Qp (), we have
that NQp ()/Qp () = (1)n an , and this is a p-adic number. This discus-
sion provides a way of nding out the proper denition of an extension
of the p-adic norm to Qp .
For if K is an extension of F with a norm that extends the norm
on F , given Aut(K), the uniqueness of the norm extension implies

that x = (x) = x. If K = Qp () is a Galois extension, we can
nd automorphisms of K taking any conjugate of to itself, which
implies that all of these conjugates must have the same norm. Since
NQp ()/Qp () is a p-adic number, we must have that

n
NQ = NQ ()/Q () =
() p i = i = n ,
p ()/Qp p p
i=1
and so the extended norm of should be dened by

1 1 1
= NQp ()/Qp ()pn = ||i (0)||pn = ||an ||pn .
The expression above can be rephrased intrinsically. For if K is any -

1
nite eld extension of Qp that contains , then = NK/Qp ()p[K:Qp ] ,
where [K : F ] denotes the degree of the eld extension K of F .
Denition 3.8 Let be an algebraic number over Qp contained in a

eld extension K of Qp . We dene ||||p by
1
||||p = NK/Qp ()p[K:Qp ] , (3.12)
where NK/Qp () is the value at x = 0 of the unique monic irreducible

polynomial i (x) of , and [K : Qp ] is the degree of K over Qp .
Expression 3.12 is clearly multiplicative in , it is only zero when

= 0, and it yields the p-adic norm of in the case when Qp . Thus,
this function is an extension of the p-adic norm function on Qp . This
extension also satises the non-Archimedean property (2.2), fact whose
proof we leave to the interested reader. Thus, expression 3.12 denes a
non-Archimedean norm on Qp that is an extension of the p-adic norm
on Qp .
For the same reasons as before, the set {x Qp : ||x||p 1} forms a
ring, the ring of integers of Qp .
For any in a nite eld extension K of Qp of degree n, we dene
1
ordp = logp ||||p = logp NK/Qp ()p ,
n
where the logarithm is computed in base p. This clearly yields an ex-
tension to K of the ordp function on Qp . The image of K under this
function is an additive subgroup of n1 Z, and therefore, it must be of the
form 1e Z for some positive integer e dividing n. This integer is the in-
dex of ramication of K over Qp . If e = 1, the eld K is said to be an
unramied extension of Qp .
Example 3.9 We can revisit the result in Theorem 2.8 by considering
the quadratic equation f (x) = x2 p. Any root Qp has p-adic norm
1
||||p = 1/p 2 , and so neither of them is in Qp . With more generality,
if m/n Q with (m, n) = 1, then the polynomial f (x) = xn pm is
m
irreducible over Qp , and each of its roots in Qp has p-adic norm 1/p n .
Example 3.10 The primitive p-roots of unity in Qp are roots of the

irreducible polynomial ip (x) = xp1 + xp2 + + x + 1, and so have
degree p 1 over Qp . If p is a root of this polynomial, then ||p ||p = 1,
and any other root is the form pk for 1 k p 1.
We have that xip (x + 1) = (x + 1)p 1, and the polynomial ip (x + 1)
is irreducible over Qp , and have degree p 1. Since
the roots of ip (x + 1)
have the form pk 1, it follows that pk 1p = pp11
. We have that
K = Qp (p ) is a ramied eld extension of Qp of degree p 1. Its index

of ramication is p 1 also.
d1
In fact, for any d 1, the polynomial ip,d (x) = ip (xp ) is irreducible
over Qp , and its roots are the pd -roots of unity in Qp . If pd is a primitive
root, any other is of the form pkd for 1 k pd 1, p k. We have that

pd = 1 and pd 1 = 1
d1 . Thus, the only p-root of unity
p p p(p1)p
in Qp is = 1 when p is an odd prime, or = 1 if p = 2.
Let us now consider the situation where the order of the root of unity
is not divisible by p. We recall rstly the result in 2.5, where we proved
that all (p 1)-roots of unity lie in Qp . If we consider any primitive
(pr 1)-root of unity pr 1 , r 1, it has degree r over Qp and irreducible

polynomial 0lr1 (x pl r 1 ). Thus, ||pr 1 ||p = ||pr 1 1||p = 1.
Now, if d is any positive rational integer not divisible by p, we let md
be the smallest positive integer m such that pm 1 mod d. Since the
group of roots of the polynomial xn 1 in Qp is cyclic (as is every nite
multiplicative group in a eld), if pmd 1 is any primitive (pmd 1)-root
of unity, we obtain a d-root of unity by taking
md 1
kp m
= pmd 1
d
,
where k is any integer coprime to (pmd 1)/md . We have that ||||p = 1,
|| 1||p = 1, and Qp if, and only if, md = 1.
The ramied notion developed for nite eld extensions of Qp applies
to all elements in Qp . We say that is ramied if ||||p is a fractional
power p, and dene its ramication degree to be the smallest integer
e() such that e() is unramied. Unramied elements of Qp are those
for which e() = 1. The example shows that the roots of unity of order
coprime to p are unramied elements of Qp .
The union of all nite unramied extensions of Qp is denoted by Qur p ,
and it is referred to as the maximal unramied extension of Qp .
Proposition 3.11 The maximal unramied extension Qur p of Qp is
obtained by adjoining to Qp all the n-roots of unity with (n, p) = 1. The
set of unramied integers Zurp = {x Qp : ||x||p 1} is a (local) ring
ur
with a unique maximal ideal pZp , and the residue eld Zur
ur ur
p /pZp is the
algebraic closure Fp of the eld Fp with p-elements.
Proof. The nite unramied extensions of Qp are in 1-to-1 corre-
spondence with the nite extensions of Fp . Since the splitting eld of
xp x = x(xp 1 1) is the unique extension of Fp of degree k, it fol-
k k
lows that Qp has a unique unramied extension of degree k for every k,

the splitting eld of the said polynomial. Thus, a degree k unramied

extension of Qp is obtained by adjoining the (pk 1)-roots of unity to
Qp , and Qurp , which corresponds to the algebraic closure of Fp , is ob-
tained by adjoining all the (pk 1)-roots of unity for all k. If k is a
multiple of p, the order m = pk 1 of the root is not divisible by p. On
the other hand, for any m coprime to p we have that p(m) 1 mod m,
and as we saw in Example 3.10, any m-root of unity is unramied and
is obtained as a suitable power of a primitive (p(m) 1)-root of unity.
We therefore conclude that Qur p is obtained by adjoining to Qp all the
m-roots of unity with (m, p) = 1.
Since the p-adic norm is non-Archimedean, the unramied integers
Zur
p = {x Qp : ||x||p 1} form a ring. Since every element of Zp
ur ur
can be written as pk u for some u with ||u||p = 1 and some nonnegative

integer k, we see that pZur
p is a maximal ideal, and in fact, the only one
such. By the discussion above (cf. the arguments in Example 3.10), the
residue eld Zurp /pZp is the algebraic closure Fp of the eld Fp with
ur
p-elements.
As a metric space, Qp is not complete. Its completion p turns out to

be algebraically closed.
Let us think of Qp as the analogue of R when in the completion of Q
we use the p-adic norm p instead of the Euclidean norm. Then p is
to be though as the analogue of C. There is though an interesting p-adic
dierence: C is obtained from R as a eld extension of degree two that is
algebraically closed and metrically complete, while the algebraic closure
Qp of Qp requires an additional metric completion to ultimately yield
p . Neither Qp nor p are locally compact: their unit spheres centered
at the origin have plenty of sequences without convergent subsequences,
for instance, sequences of distinct roots of unity of order coprime to p.
We denote by Z ur p , the completion of the unramied extension
p
p of Zp . The theory of arithmetic dierential operators we will discuss
Zur
in 5 over Zp has a much richer avour over Z ur . Cf. [8, 6, 15]. The reader
p
will come to appreciate this fact in our outline of the most general version
of these operators in 6.2.
Remark 3.12 Any (pk 1)-root of unity, k 2, lies in the boundary

ur \ Zp .
of Z p
Remark 3.13 We revisit the statement in Theorem 3.1. The binomial


y
series n0 xn converges to (1 + x)y for all x p such that
n
ordp x > 1/(p 1), and the resulting function is continuous as a function
of y.
3.3 Zeta functions

Let F be a eld. We denote by AnF the n-dimensional ane space over
F, AnF = {(x1 , . . . , xn )| xi F , 1 i n}. the set of n-tuples of
elements in F. Then the projective n-space PnF over F is the set of one-
dimensional subspaces of An+1 F . A point p PnF is usually written as
a homogeneous vector [X0 : . . . : Xn ], by which is meant the F-line
spanned by (X0 , . . . , Xn ) An+1
F \ {0}.
The zero locus of a nite family {fi }iI , fi F[X1 , . . . , Xn ] denes
an ane algebraic variety over F:
V{fi }iI = {(X1 , . . . , Xn ) AnF : fi (X1 , . . . , Xn ) = 0} .
The ideal I(V ) of V consists of the set of all F-polynomials that vanish
on points of V . In fact, any ideal I F[X1 , . . . , Xn ] denes a corre-
sponding locus of points V = {X AnF : f (X) = 0 for all f I}, and
the Hilberts basis theorem asserts that I has a nite basis f1 , . . . , fk
so that this locus is in eect an ane algebraic variety. However, the
corresponding ideal I(V ) could be larger than I.
A polynomial f F[X0 , . . . , Xn ] does not in general descend to a
function on PnF . However, if f is a homogeneous polynomial of degree d,
the notion of zeroes of f in PnF makes sense because we have the rela-
tion f (X0 , . . . , Xn ) = d f (X0 , . . . , Xn ). A projective algebraic variety
V PnF over F is dened to be the zero locus of a collection of homoge-
neous polynomials in F[X0 , . . . , Xn ], and its ideal I(X) consists of the
set of polynomials that vanish on X.
For instance, the rational normal curve Cd PnF of degree d is dened
to be the image of the map P1F PdF given by
[X0 : X1 ] [X0d : X0d1 X1 : . . . : X0 X1d1 : X1d ] = [Z0 : . . . : Zd ] .
It is the common zero locus of the polynomials pij = Zi Zj Zi1 Zj+1
for 1 i j d 1, and has associated with it the ideal I(Cd ) :=
{f F[Z0 , . . . , Zn ] | f 0 on Cd }. This ideal is, in fact, generated by
the family of polynomials pij .
If E is a eld extension of F and V = V{fi }iI is a projective algebraic
variety over F dened by homogeneous polynomials fi F[X], then V is

dened also over the extension E as the coecients of the fi s whose zeros
dene V over F are also all elements of E. Thus, we can talk about the set
V (E) of E-points of V , V (E) = {(X0 , . . . , Xn ) PnE : fi (X0 , . . . , Xn ) =
0, i I}. This makes of the F-variety V a functor from the category of
eld extensions of F and their morphisms to a suitable category of sets
and morphisms, with the functor mapping an extension E of F to the
set V (E) of E-points of the variety. Let us observe in passing that using
restrictions when possible, we may also carry out this idea in the opposite
direction, and nd the points of a variety that lie on a subring of F when
the variety in question is dened by polynomials whose coecients are
elements of the subring (see [31] for an elementary expansion on this
point that relates to the roots of unity).
Let us assume now that F is a nite eld Fq of characteristic p. A
simple combinatorial argument shows that number of bases of the vector
space Fnq over Fq is given by
n(n1)
(q n 1)(q n q) (q n q n1 ) = q 2 (q 1)n [n]q ! ,
where
[n]q = 1 + q + q 2 + + qn1 ,
and where the factorial is dened by
[n]q ! = [1]q [2]q [n]q .
Proceeding similarly, we see that the number of linearly independent

k-element subsets of Fnq is equal to
k(k1)
(q n 1)(q n q) (q n q k1 ) = q 2 (q 1)k [n]q !/[n k]q !,
and so if k n, the number of subspaces of Fnq of dimension k is

n [n]q !
= .
k q
[n k]q ![k]q
In particular, we see that the number of points in PnFq is equal to q n +

q n1 + + q + 1.
Given a variety V dened over Fq , we denote by Nj the number of
Fqj -points of V . We may encode the combinatorial information given by
all of the numbers Nj , j = 1, 2, . . . in the Weil zeta function of V that
is dened as

N
(V /Fq , t) = exp
j j
t 1 + tQ[[t]] , (3.13)
j
j=1
Example 3.14 We consider the case where V = PnFq . Then the number
Nj of Fqj -points of this variety is q jn + q j(n1) + + q j + 1, and we
have that
1
n
1
(Pn /Fq , t) = = .
(1 t) (1 q t) j=0 1 q j t
n
This is a rational function with coecients in Z that satises the identity

n(n+1)
Z(Pn /Fq , 1/q n t) = q 2 tn+1 Z(Pn /Fq , t). Let us think of the integer
n as a topological invariant of V . Then this last relationship illustrates
the close interconnection between topological invariants of V and the
geometry of the algebraic variety over a nite eld.
If we consider the case where V = AnFq instead, then Nj = q jn , and
we have that
1
(An /Fq , t) = .
1 qn t
Remark 3.15 We may also dene the zeta function of nonnegative

elliptic pseudodierential operators. Their connection with the number
theoretic considerations introduced above is made by the spectral analy-
sis of linear operators on Banach spaces. We illustrate quickly this issue
via the case of the Neumann operator.
Let us consider any manifold M that is the the boundary of a closed
connected Riemannian manifold with boundary M . The Riemannian
structure allows us to dene a vector eld that is normal to the boundary,
the normal vector eld. Given a distribution u0 on M , let u be the unique
solution to the Dirichlet problem on M that has boundary value u0 . The
Neumann operator N acts on u0 , and maps it to the restriction to the
boundary of the normal derivative of u.
The Neumann operator N is a nonnegative elliptic pseudodierential
operator. As a function on the cotangent bundle T M , its principal
symbol is (N )(y, ) = i|| [46]. The kernel of N is the set of constant
functions.
The manifold M itself is Riemannian with Riemannian structure in-
duced by the Riemannian structure on M . Let be the Laplacian of

the metric. Then = N modulo terms of order zero or less, and in
this sense, N is a scalar Dirac operator. We consider the self-adjoint
extension of N to L2 (M ), which we shall denote by N also.
Let T be any nonnegative self-adjoint operator on a Hilbert space with
discrete spectrum. Let 0 < 1 2 be the sequence of nonzero
eigenvalues of T counted with multiplicity. The zeta function of T is
dened by

1
T (s) = .
s
n=1 n
This function is considered with domain the region of the complex plane
where the series converges.
The determinant of T is dened in terms of its zeta function by the

identity det T = eT (0) .
Let N = NS1r be the Neumann operator on the circle of radius r in
C centered at the origin (see [45] for the key role this operator plays in
the analysis of the Zaremba problem on planar domains). Then we have
that

1
N (s) = 2rs s
= 2r s (s) .
n=1
n
where (s) is the Riemann zeta function. Indeed, we just need to observe
that
1
ein/r
2r
is an eigenfunction of N of eigenvalue |n|/r, and the stated assertion
follows easily from that.

Using this relationship and the fact that (0) = 1/2 and (0) =

(1/2) log 2, it follows N (0) = log (2r). Thus, the determinant of
N over the r-circle is given by

det NS1r = eN (0) = 2r ,
the length of S1r .
It would be remarkable to produce an interpretation of the zeta func-

tion of nonnegative elliptic pseudodierential operators in terms of the
arithmetic dierential operators, which are the subject of our work here.
At present, we are far from being able to do so.
We call a variety V dened over F a hypersurface if V is dened as
the zero locus of a single polynomial in F[x].
Lemma 3.16 Let V be a variety over the nite eld Fq . Then its zeta
function (V /Fq , t) in (3.13) is the product of the zeta functions of a -
nite number of hypersurfaces, or their inverses, and in fact, (V /Fq , t)
1 + tZ[[t]].
Proof. By the Hilberts basis theorem, the ideal I(V ) of V has a nite
basis {f1 , . . . , fk }, and we have V = V{f1 ,...,fk } . Given a nite number of
polynomials fi1 , . . . , fil , we denote by Hfi1 ,...,fil the hypersurface dened
by the product fi1 fil , and by Nj,Hfi ,...,fi its number of Fqj -points.
1 l
By the inclusion-exclusion combinatorial principle, we have

Nj = (1)l+1 Nj,Hfi ,...,fi ,
1 l
i1 <<il
and so
l+1
(V /Fq , t) = (Hfi1 ,...,fil /Fq , t)(1) ,
i1 <<il
which proves the rst assertion.

Let us now consider a hypersurface Hf . Let E be an extension of Fq ,
and consider an E-point p. We let r0 be the smallest positive integer
such that all of the components of p are in Fqr0 . Each component of p
has r0 conjugates, and by varying all of these components among their
conjugates, we form a set of r0 points p1 , . . . , pr0 . These points must
be distinct. They are Fqr -points of Hf if, and only if, r0 divides r, so
they contribute r0 to each of the numbers Njr0 ,Hf , j = 1, 2, . . ., and

their contribution to (Hf /Fq , t) is given by exp ( j=1 r0 tjr0 /jr0 ) =

(1 tr0 )1 = j=0 tjr0 1 + tZ[[t]]. It follows that the zeta function of
Hf must be in 1 + tZ[[t]], as this function is the product of series of this
type.
Example 3.17 In the analysis of the Weil function of a variety, the
lemma above gives a special importance to the hypersurface case. We
discuss here the zeta function of hypersurfaces when n = 1, hypersur-
faces dened by polynomials of one variable. We will do so assuming
that q = p, so V is dened by a polynomial f (x) in Fp [x]. We shall
assume also that f (x) does not have multiples roots in Fp .
Let us decompose the dening polynomial f (x) into a product of ir-
reducible factors f = f1 fk , and let di denote the degree of fi , so

i di = d, where d is the degree of f . We then have that Nj,Hfi = di if
di | j and it is zero otherwise. For each root of fi induces a homomor-
phism Fp [x]/(fi ) Fp whose image is Fpdi , and therefore, the set of all
roots of fi must be Fpdi . Now if fi has a root in Fpk , then Fpdi Fpk
and Fpk is a eld extension of Fpdi . Then di | k and all the roots of fi are
in Fpk . Conversely, if di | k then Fpdi Fpk , and the larger eld must
contain all the roots of fi , which nishes the proof of the assertion.
Therefore, we have that (Hfi /Fp , t) = 1/(1 tdi ), and

k
k
(Vf /Fp , t) = (Hfi /Fp , t) = (1 tdi )1 .
i=1 i=1
Thus, (Vf /Fp , t) is a rational function with coecients in Z. Further,

(Vf /Fp , t) is a function of the form 1/p(t), where p(t) can be written
d
as i=1 (1 i t) Z[t], | i |= 1, i = 1, . . . , d. Finally, (Vf /Fp , 1/t) =
(1)k td (Vf //Fp , t).
Encoded in the zeta function (Vf //Fp , t) we have all the information
concerning the eld extensions Fq where all the roots of the polynomial
f lie.
The zeta function (3.13) of a smooth projective variety V dened

over a nite eld Fq was the subject of several famous conjectures by
A. Weil in 1949 [48]. The conjectures state that the zeta function of V
has properties similar to those of the zeta function of the elementary
Example 3.17 analyzed above. They are:
1. (V /Fq , t) is a rational function with coecients in Z.

2. The topology of the variety and its zeta function are intimately re-
lated in that
p1 (t)p3 (t) p2n1 (t)
(V /Fq , t) = ,
p0 (t)p2 (t) p2n (t)
where n is the dimension of V , pk (t) Z[t], p0 (t) = (1 t), pk (t) =
bk j
j=1 (1 k,j t), for some k,j C such that | k,j |= q , 1 k
2
2n 1, and p2n (t) = (1 q n t). When V is the reduction mod p

of a smooth variety over a eld embedded in C, the degrees bk of
the polynomials pk can be interpreted as the Betti numbers of the
complex points of V .
3. If E is the Euler characteristic of V , then
nE
(V /Fq , 1/q n t) = q 2 tE (V /Fq , t) ,
relation that implies that the sets {2ni,l }l and {q n /i,l }l coincide.
The rst major step in the settling of these conjectures was taken by
Dwork in 1960 [23, 24]. He proved the rationality of the zeta function by
applying techniques from p-adic analysis to the study of the problem. He
produced suitable lifts of characters on Fq to p , space where the said

techniques could be then used. We now discuss this issue further because
of its relation to lifts of Frobenius mappings, which do play a central role
in the theory of arithmetic dierential equations. Dworks use of lifts
of Frobenius is inspired in a philosophy that is perpendicular to ours.
Indeed, he views the p-adic numbers as arguments of functions, and not
as functions themselves. We shall not use Dworks idea otherwise, and
our discussion of his theorem will be very sketchy. We refer the reader
to Chapter V in [29] or II.6 and II.7 in [25] for an exposition of this in
its entirety. We shall follow here the rst of these references somewhat.
By Lemma 3.16, it suces to carry out the proof for a nonsingular
projective hypersurface V dened over Fq . In this case, if f (X0 , . . . , Xn )
is the dening polynomial of V , the nonsingular condition on V means
that f and (X0 f, . . . , Xn f ) do not have common zeroes in Fq , but in
eect, Dworks result is general, and it does not require this condition.
The dimension of the hypersurface V is n 1. Using the cell decom-
position PnF = (PnF \ Pn1F ) (Pn1
F \ Pn2
F ) (P1F \ P0F ) P0F , we can
express V as the disjoint union of ane hypersurfaces of dierent dimen-
sions, and so the rationality of (V /Fq , t) will follow if we can prove the
corresponding statement in the case where V is an ane hypersurface
dened by f Fq [x], which we assume hereupon. The argument is by
induction on n, the statement being clear when n = 1.
The bulk of the work lies in the proof that the zeta function of an
ane hypersurface V is a meromorphic function. In order to do that, let
us consider the modied zeta function

N
(V /Fq , t) = exp
j j
t , (3.14)
j=1
j
where Nj is now the number of Fqj -points (x1 , . . . , xn ) of V with no zero

coordinate. Then (V /Fq , t) = (V /Fq , t) exp ( j (Nj Nj )tj /j). If Hi
is the zero locus of gi = f (x1 , . . . , xi1 , 0, xi+1 , . . . , xn ), the exponential
factor is the zeta function of ni=1 Hi . Each Hi is either n-ane space or
an ane variety of dimension less or equal than n 2, and their zeta
functions are known either by using Example 3.14 or by the induction
hypothesis, and regardless of the case, they have the desired properties.
By an application of the inclusion-exclusion principle, cf. with the proof
of Lemma 3.16, we see that the said exponential is an element of 1 +
tZ[[t]], and by Lemma 3.16, we conclude that (V /Fq , t) 1 + tZ[[t]]
also.

Since the modied zeta function (V /Fq , t) in (3.14) is an element
of 1 + tZ[[t]], we may try to lift it to a p-adic meromorphic function
on p . Since the multiplicative group F q of nonzero elements of Fq is
cyclic of order q 1, the integer Nj can also be described as the number
j
of Fq -points (x1 , . . . , xn ) of V such that xqi = 1, i = 1, . . . , n. This

is the key reason for the introduction of (V /Fq , t), for now we can
use this condition in order to count Nj through a function that have a
suitable meromorphic lift to p . This will show that (V /Fq , t) itself is
meromorphic. Then it can be proven that (V /Fq , t) is actually rational.
Given a nite eld extension E of Fq of degree j, we have the trace
TrE/Fq and norm NE/Fq operations. These are dened by TrE/Fq (a) =
2 j1 j1
a + a q + a q + + aq and NE/Fq (a) = a aq aq , the sum and
product of the conjugates of a, respectively. Modulo a rational number,
these two operations can be cast as the trace and determinant of the
mapping E b ab E (cf. with the use of the same mapping in
3.2 when extending the p-adic norm to Qp ). They dene surjective
homomorphism TrE/Fq : E Fq and NE/Fq : E F q . We use the
rst of these to count Nj .
Let p be a nontrivial p-root of unity in p . Then
Fq p
(3.15)
a TrFq /Fp (a)
is an p -character on the additive group Fq , and we have that

TrFq /Fp (ba) q if b = 0,
=
0 if b F
q .
aFq
By this identity, we see that each Fqj -point (x1 , . . . , xn ) in V , all of

whose components are nonzero, contributes q j to x0 F j x0 f (x1 ,...,xn ) ,
q
and so taking away the contributions from x0 = 0, we obtain that

Tr (x0 f (x1 ,...,xn )) = q j Nj (q j 1)n . (3.16)
x0 ,...,xn Fj
q
At this point, we must analyze the left side of this expression in detail.
Given a Fq , its Teichm uller representative t = ta Z ur
p p is
dened to be a root of x x such that t a mod p. By Proposition
q
3.11, the set {ta }aFq is contained in the splitting eld of xq x over Qp .
(The reader may look back at our discussion in 2.5, where we proved
explicitly that for any element Fp a = 0, there was a (p 1)-root of
unity a such that a a mod p; cf. with the discussion of Teichm uller
representatives in Example 4.8.) We now have the following key result

due to Dwork (see Chapter V, 2 of [29] for its proof):
Lemma 3.18 Suppose that q = pj for some positive integer j. There

exists a p-adic power series (x) = an xn Qp ()[[x]] with ordp an
n/(p 1) such that the character function (3.15) on Fq can be obtained
j1
by evaluating (x)(xp ) (xp ) at the Teichmuller lift t of a.
This result is rather subtle. For instance, since the Teichm uller lift
a of a Fp is such that a a mod p, and since the power in Zp of
the p-root of unity depends only on its congruence class mod p, we
have Tr a = Tr a . However, as explained in Example 3.10 (cf. remark
ur p , while, for
3.12), a (pl 1)-root of unity lies in the boundary of Z p
instance, the ArtinHasse exponential function has radius of convergence
strictly less than 1, and could not include such a point in its domain of
denition. A rened argument must be used in order to yield a function
that lifts a Tr a to a function on a disc in p whose radius is greater
or equal than 1. Once that is accomplished, the general techniques of
p-adic analysis can then be applied to study the lifted function in p .
The function (x) used by Dwork is a suitable modication of the
ArtinHasse function Ep (x). It can be described in simple terms best
by using the binomial functions of Theorem 3.1. Let = 1. As
we saw in Example 3.10, has p-adic order 1/(p 1), and if we view
g(x) = g(x, ) = (1 + )x as a power series in x, the series will not
converge at the xs of interest. So g(x) is modied to
2
xp x 2 xp xp
g(x, y) = (1 + y)x (1 + y p ) p (1 + y p ) p2
and is dened to be

(x) = g(x, ) = an xn .
n=0
We can then verify the growth in the order of the an s that is stated in
the Lemma. This estimate for the p-adic order ensures that this series
1
converges on the disc {x p : ||x||p < p p1 }.
The root of unity corresponds to a unique solution of the equation
x p1
= p in p . Then we can dene

(x)p
(x)p
j
(xxp ) x+ p
pj
E (x) = e =e = Ep (x) e .
j=2
This function is equal to (x). Indeed, E (1) is the unique p-root of unity
congruent to 1 + modulo 2 , and for any c Zp such that cp = c,

j j1 pl
we have that E (c) = E (1)c , so if xp = x, then E (1) l=0 x =
j1
(x)(xp ) (xp ). The evaluation of (x) for any x such that
||x||p 1 is achieved by substitution into the dening power series, and
(xxp )
in general it
cannot be obtained directly by substitution into e =
p
x+ (x)
e p
.
Let us return to (3.16), and complete the promised sketch of the proof.
The coecients of x0 f (x1 , . . . , xn ) are all elements of Fq , so they can be
lifted to their Teichm uller representatives in p to obtain a polynomial
N
function F (X) = F (X0 , . . . , Xn ) = l=1 al X Il p [X0 , . . . , Xn ]. Here
Il = (il0 , il1 , . . . , iln ) is a tuple of integers, and X Il has the usual mean-
ing. Then using Lemma 3.18 in (3.16), we obtain that

n
qj Nj = (q j 1)n +
j1
(al xIl )(apl xpIl ) (alpj1 xp Il
).
x 0 , . . . , x n p l=1
q j 1
x = 1, i = 0, . . . n
i
j1
Let G(X0 , . . . , Xn ) = (al xIl )(apl xpIl ) (alpj1 xp Il ), and let Tq

be the operator acting on series and dened by Tq ( aI X I ) = aqI X I ,
where qI is the tuple obtained by multiplying each entry of I by q. Since
the al s are in the unit disc in p , the coecients of G(x) are such that
the traces of any power of the linear operator = Tq G are well dened
(see [29, Chapter V 3]), and we obtain that
q j Nj = (q j 1)n + (q j 1)n+1 Tr(j ) ,
so
n

n+1

n n+1
Nj = (1)l q j(nl1) + (1)l q j(nl) Tr(j ) .
i l
l=0 l=0
j j
Let (t) = exp( j=1 Tr( )t /j), the exponential computed in p .
This is a meromorphic function of t [29, Chapter V, 3], and by the
identity above, we obtain that

l+1
n n+1 l+1
n+1

n (1)
l (1)
l
nl1 nl
(Hf /Fq , t) = (1q t) (q t) ,
l=1 l=0
is also a meromorphic function.

Once it is known that (Hf /Fq , t) is meromorphic, it can be proved
that this meromorphic function is in fact a rational function [29, Chapter
V, 5]. Thus, the zeta function of Hf itself is rational.
4
Analytic functions on Zp
Let us pause to introduce some terminology before proceeding any fur-

ther with our work. We will need to consider some power series in several
variables below, and so we give the next denition with a higher degree
of generality from the one we have used until now.
Given a multi-index = (0 , . . . , k ) of nonnegative rational integers,
we shall say that 0, and use the conventional expression x to
0 xk . By the weight || of the multi-index
k
denote the monomial x 0
, we mean the rational integer || = 0 + + k .

Denition 4.1 We say that F (x) = 0 a x Qp [[x0 , . . . , xk ]] is
a restricted power series if lim|| a = 0.
Let

F (x) = aj xj (4.1)
j=0
be an element of Qp [[x]]. For this series to converge on Zp , it is both

necessary and sucient that
lim aj = 0 in Qp . (4.2)
j
In this case, the series F (x) denes an element of C(Zp , Qp ). Further,

as we indicated earlier, the series of n term-by-term dierentiations

F (n) (x) = j(j 1) (j n + 1)aj xjn ,
j=n
is also a convergent series on Zp , and it denes a continuous function.

Thus, power series of the type (4.1) whose coecients satisfy (4.2) are
48
Analytic functions on Zp 49
closed under dierentiation, and that makes the theory of analytic func-
tions over Qp simpler than the corresponding theory over C.
Denition 4.2 A function f : Zp Qp is said to be analytic if there

exists an integer N 1 and power series F0 , F1 , . . . , FpN 1 Qp [[x]]
with radius of convergence 1/pN , such that
f (x) = Fk (x k)
for each k {0, 1, . . . , pN 1}, and each x k + pN Zp .
Remark 4.3 An analytic function f on Zp denes an element of

C(Zp , Qp ). By compactness of Zp , all elements of C(Zp , Qp ) are bounded.
If f is locally given by a series of the type (4.1) with coecients satisfy-
ing (4.2), then the coecients of the series must be bounded. Thus, for
a suitable integer n, the function pn f ranges in Zp , that is to say, it is
an element of C(Zp , Zp ), and can be represented locally by series as in
Denition 4.2 whose coecients are in Zp .
It is natural to ask for a characterization of the functions in C(Zp , Zp )

that are analytic. We put this task in some perspective by reconsidering
the Mahlers expansion of a function f (x) in this space, whose coecients
an form a restricted sequence in Zp . By Mahlers expansion, it follows
immediately that f (x) is the uniform limit of polynomials, generalizing
in this sense Weierstrass theorem. The relationship (3.4) between the
an s and the values of f over the rational integers has another important
consequence: the sequence of partial sums of the Mahlers expansion of
f (x) yields an optimal polynomial of degree n approximating f uni-
formly. These approximating optimal polynomials are not unique, but
it is important to have one produced in a canonical manner, and in this
case, the polynomial is produced out of the knowledge of the values of
f on the set {0, . . . , n}.
However, it is dicult to nd conditions on the growth of the ||an ||p s
that would ensure that f (x) exists and is a continuous function also.
Mahler shows [36, ch. 7, Theorem 4] that if limn n ||an ||p = 0, then
f (x) C(Zp , Zp ) has a continuous derivative. In general, if a function
has a Mahler series with coecients an and its derivative exists and
is continuous in Zp , then the Mahlers coecients of the derivative are

given by an = k=1 (1)
k1
ak+n /n, which therefore, must be a null
sequence. But the converse does not hold.
Further, as in the real case, there are continuous functions on Zp that
50 Analytic functions on Zp
have continuous derivatives of all order, but are not analytic as dened
previously [36]. Indeed, the function

x
f (x) = an an = p[ n ] ,
n
n=0
is an element of C(Zp , Qp ) and can be continuously dierentiated in-

nitely many times. This follows [36, ch. 7, Theorem 5] by the fact that
for all > 0 we have that limn n ||an ||p = 0. However, for f (x) to
be analytic, it must be the case that ||an ||p < pn/p [36, p. 85], and the

coecients p[ n ] do not satisfy that condition.
Characterizing p-adic analytic functions becomes a rather dicult
task.
Denition 4.4 A function f : Zp Zp is said to be analytic of level m,
if for any a Zp there exists a restricted power series Fa Zp [[x]] such
that
f (a + pm u) = Fa (u)
for all u Zp . We say that the collection of series Fa represents f .
The denition above is not conventional. It was used in [14] for reasons
that will become clear later on.
Remark 4.5 We have the following:
1. In Denition 4.4, it is enough to take the as in a complete residue
system mod pm in Zp .
2. If we consider functions in C(Zp , Zp ), the notions of analyticity in
Denitions 4.2 and 4.4 are equivalent. Indeed, if f C(Zp , Zp ) is
locally representable by a restricted power series in Zp [[x]], then it
is analytic in the sense of Denition 4.2. Conversely, let as assume
that f C(Zp , Zp ) is locally represented by a restricted power series
F Qp [[x]] of the form (4.1), whose coecients satisfy (4.2), on a disc
of radius 1/pN , N 1. Thus we have that f (x) = F (x k) for some
k {0, . . . , pN 1} for all x k+pN Zp . Without loss of generality, let
us assume that k = 0. Computing the value of f at x = 0, we obtain
that f (0) = a0 , and so a0 Zp , it being in the range of f . By the
non-Archimedean property of the norm, the function f (x)a0 ranges
in Zp also, and we have that f (x) a0 = xg(x) for g C(Zp , Zp ),

and this g is represented by the series j=1 aj xj1 on the said disc.
Proceeding by induction, we then show that all the aj s are in Zp .
With this in mind, we can say that a function f : Zp Zp is
Analytic functions on Zp 51
analytic in the sense of [36] (cf. with [42], p. LG 2.4) if, and only if,
it is analytic of level m for some m.
Example 4.6 Given a subspace X Qp , a function f : X Qp is

said to be locally constant on X if for every X, there exists n such
that f is constant on the disc DX (, pn ) = ( + pn Zp ) X. Any locally
constant function f : Zp Zp is analytic. A typical example of that is
the characteristic function D(0,1/p) of the disc in Zp with center at zero
and radius 1/p, an analytic function of level 1.
Mahlers expansion of this characteristic function (see Theorem 3.2)
is given by

x
D(0,1/p) (x) = an ,
n
n=0
where
[n]

p

n jp n
an = (1) (1) ,
jp
j=0
and so, for instance,

x
D(0,1/2) (x) = 1 + n n1
n=1 (1) 2 in Z2 ,
n

n 1+ n x
D(0,1/3) (x) = 1 + n=1 (1)
n
2 cos 6 3 2 in Z3 .
n
In general, the p-adic coecient an can be expressed in terms of a prim-
itive p-root of unity p by the formula

p
an = p1 (pk 1)n .
k=1
Notice that this expression shows the rational integer an as a sum of

elements in Qp (p ). In Example 3.10, we proved that pk 1 has p-adic
norm 1/pp1 . Therefore,
||an ||p pn(p1)+1 ,
and we conclude that the Mahler power series of D(0,1/p) is a restricted
power series in Zp . Further, we see that {||an ||p } converges to zero faster
than pn/p , and this growth can be used to conclude that D(0,1/p) (x) is
an analytic function [36, p. 85]. But observe that drawing that conclusion
on the basis of properties of the an s is not an elementary result.
The example above illustrates the diculties we face

when dealing
x
with issues of analyticity. The polynomial functions are very
n
natural candidates to use when we wish to approximate uniformly a
continuous function f in an optimal manner. But if f is analytic, that
fact is not easily captured in properties of the coecients of the approx-
imation.
Example 4.7 The projection

Zp
n
Zp
cn
onto the n-th coecient cn in the p-adic expansion (2.4) of the p-adic
integer is a locally constant function on Zp . Notice that n is a linear
combination of the locally constant characteristic functions of the discs
a + pn+1 Zp , 0 a pn+1 1.
Example 4.8 For any Qp , there exists a unique representation

= j ()pj ,
jn
where all the coecients j () are (p 1)-roots of unity. This is the

Teichm uller representation of , and in some sense, it is a more natural
choice than that given by the representation (2.4). See its essential use
in the proof of Theorem 9.2 below.
In order to dene the functions j () in the Teichm uller repre-
sentation of , it suces to do so for a p-adic integer. So we proceed
to dene the functions j : Zp Zp for j = 0, 1, . . ..

Given Zp , we have = j=0 cj pj where the cj s are all rational
integers in the range 0 cj < p. Let us consider the root c0 of the
polynomial xp x associated to the rst coecient c0 , that is to say,
n
c0 = lim cp0 . By our discussion in 2.5, we have either c0 = 0 when
c0 = 0 or c0 a (p 1)-root of unity in Zp when c0 = 0. We dene the
Teichm uller character function by

Zp Zp
c0 .
It has the properties that
() = ()()
4.1 Strassmanns theorem 53
and that
||( + ) (() + ())||p < 1 ,
respectively.
The zeroth Teichm
uller coecient function is dened by
0 () = () .
We have the estimate
|| 0 ()||p max{|| c0 ||p , ||c0 0 ()||p } < 1 ,
and so 0 () mod p. The remaining Teichm uller coecient functions

can then be dened by the recursion formula

n1
j=0 j ()p
j
n () = , n = 1, 2, . . . .
pn
n
We then have that j=0 j ()pj mod pn+1 .
As is the case of the projection functions n in Example 4.7, all the
Teichmuller coecient functions n () are locally constant.
4.1 Strassmanns theorem

Let us consider a nontrivial restricted power series

F (x) = aj xj Qp [[x]] , lim ||aj ||p = 0 , (4.3)
j
j=0
as a function with domain of denition given by Zp . We have that

||aj ||p M for some constant M . In fact, the optimal constant M is
equal to maxj {||aj ||p }, which is achieved as the p-adic norm of some of
the coecients. Let N be an integer such that
||aN ||p = max{||aj ||p } , ||aj ||p < ||aN ||p for all j > N . (4.4)
We shall need the following theorem of Strassmann [47] later on. We

present its proof mostly for completeness, but also as a way to illustrate
the contrast between p-adic and the usual real analysis.
Theorem 4.9 Consider a nontrivial restricted power series of the form

(4.3), and view it as a function F : Zp Qp . Let N be an integer such
that (4.4) holds. Then F has at most N roots in Zp .
Proof. The result uses the completeness of Zp , and the non-Archimedean

property of the p-adic norm p . It follows via an induction argument
on N .
Indeed, if N = 0 and F (x0 ) = 0 then

a0 = aj xj0 ,
j=1
and by the non-Archimedean property of p , we obtain the contradic-

tory inequality

||a0 ||p = j
aj x0 max aj xj0 max ||aj ||p < ||a0 ||p .
j=1 j1 p j1
p
Suppose now that N > 0, and let x0 be a root of F (x). Then

j1
F (x) = F (x) F (x0 ) = aj (xj xj0 ) = (x x0 ) aj xk xj1k
0 .
j1 j1 k=0
We can interchange the order of summation to obtain that

j1

G(x) := j1k
a j xk x0 = ak+1+l xl0 xk = g k xk ,
j1 k=0 k=0 l0 k0
where we let the last equality dene the coecient gk . Thus, we see that
G(x) is also a restricted power series whose coecients are all p-adically
bounded by ||aN ||p , we have that ||gN 1 ||p = ||aN ||p and ||gj ||p < ||gN1 ||p
for all j N . By induction, the function G(x) has at most N 1 roots
in Zp , and so
F (x) = (x x0 )G(x)
can have at most N roots altogether.
Our use of this result will be in the form of an easy consequence: if
two power series of the indicated form agree over innitely many p-adic
integers, then they must be identical.
5
Arithmetic dierential operators on Zp
In going further, we now make a fundamental use of Theorem 2.12 to

introduce the notion of arithmetic dierential operators over the ring Zp .
This is the essential idea in our work here. In the following chapter, we
shall present the theory of these operators in general, discussing the jet
spaces whose global sections give rise to them. We shall even expand our
explanations to outline the general theory in the case of multiple primes,
something that we start also in a simplied manner here. But when we
revisit the study of these operators in the remaining chapters, there will
be little that we will do beyond our discussion of them over the ring Zp
until the very end. In the very last chapter, we compare the behaviour
of these operators over the p-adic integers with their behaviour over its
unramied completion.
The idea leading to arithmetic operators embodies a radically dif-
ferent philosophy from that used up until now. This philosophy arises
naturally when looking at the p-adic numbers in the setting of the an-
alytic functions of the previous section. For we treat p-adic numbers
now as functions, albeit functions on a space of dimension zero. These
functions all admit the representation (2.4), and so they ought to be con-
sidered analytic since these representations are convergent power series
in p. Once we agree with the idea of the p-adic numbers as analytic
functions, it is then only natural to dene their derivatives, the arith-
metic derivatives of our work. In this sense, the theory departs from the
idea that the rings we use in most of our analysis the ring Zp are
to be considered as ring of functions. This point of view leads to a very
strong tie between algebraic geometry and algebraic number theory.
The idea of treating numbers as functions is classical, and goes back
to at least R. Dedekin, W. Weber, and D. Hilbert. But the arithmetic
dierentiation that we discuss here is due to A. Buium in his work
55
56 Arithmetic dierential operators on Zp
exploiting Fermat quotients [8] (see also [10, 6]). This theory has been ex-
tended recently, and we now have a notion of arithmetic partial
dierential equations available [15, 16, 17], the implications of which
are currently under investigation.
Let cp (x, y) be the element of the polynomial ring Z[x, y] given by
xp + y p (x + y)p
cp (x, y) = . (5.1)
p
Let A be a ring and B be an algebra over A. If x A, we denote by x
its image in B also. We say that a map p : A B is a p-derivation if
p (1) = 0, and
p (x + y) = p (x) + p (y) + cp (x, y) ,

(5.2)
p (xy) = xp p (y) + y p p (x) + pp (x)p (y) ,
for all x, y A, respectively.
The rst of the identities above indicates that the p-derivation p is
additive modulo the zeroth order term cp (x, y) Z[x, y]. The second
of the identities is a sort of nonlinear Leibnizs rule mod p, where the
function factors in the two summands in the right side are composed
with the p-th power Frobenius map.
Any p-derivation p denes a ring homomorphism p : A B by the
expression
p (x) = xp + pp (x) . (5.3)
For by (5.2), we have that
p (x + y) = (x + y)p + pp (x) + pp (y) + pcp (x, y)

= p (x) + p (y) ,
while
p (xy) = xp y p + pxp p (y) + py p p (x) + p2 p (x)p (y)
= (xp + pp (x))(y p + pp (y)) = p (x)p (y) .
The homomorphism p commutes with the p-derivation p .
Conversely, given a homomorphism p : A B, the expression
p (x) xp
p (x) = (5.4)
p
denes a p -derivation, and we have that

p (x) xp p (p (x)) (p (x))p
p p (x) = p = = p p (x) .
p p
Arithmetic dierential operators on Zp 57
We say that the p-derivation p and the homomorphism p are associated

to each other.
Denition 5.1 A p -ring A is a ring A equipped with a p-derivation
p : A A. A morphism of p -rings is a ring homomorphism that
commutes with the p -derivations in each of the rings. A p -ring B is
said to be a p -ring over the p -ring A if it comes equipped with a p -ring
homomorphism A B. A p -ring A is a p -subring of the p -ring B if
A is a subring of B such that p A A.
We now study this general situation over the specic case of the ring
Zp . In this ring, the unique homomorphism p that lifts the p-th power
Frobenius isomorphism is the identity. Thus, a p-derivation on Zp is
associated to p (x) = x.
In fact, this assertion can be obtained independently. By Theorem 2.12,
x x mod p, and so we can consider the Fermat quotient
p
x xp
p (x) = . (5.5)
p
to dene a p-derivation p on Zp .
Theorem 5.2 The ring Zp carries a unique p-derivation p that is

dened by the expression (5.5).
Proof. By the observation made above, the set of p-derivations on Zp
are in 1-to-1 correspondence with ring automorphisms of Zp . Since the
identity is the only such homomorphism, the assertion follows.
Since p is xed,when referring to the p-derivation (5.5) we shall often
write x instead of p (x), except when confusion could arise.
We denote by i the i-th iterate of . We use the convention that
0
a = a.
Denition 5.3 A function f : Zp Qp is called an arithmetic dier-
ential operator of order m, or a p -function of order m, if there exists a
restricted power series F Qp [[x0 , x1 , . . . , xm ]] such that
f (a) = F (a, a, . . . , m a) (5.6)
for all a Zp .
We shall refer to an arithmetic dierential operator of order m simply

as a -dierential operator of order m also, or as -function of order m.
Lemma 5.4 If f, g : Zp Qp are arithmetic dierential operators

of orders m and n, respectively, then f g is an arithmetic dierential
operator of order m + n. In particular, the set of arithmetic dierential
operators f : Zp Zp is closed under compositions.
Proof. This is elementary.
Example 5.5 Using a slightly more general context, the theory of

arithmetic dierential operators in [6, 8] gives rise to several interesting
number theoretic locally constant functions that have nice representa-
tions as arithmetic dierential operators of low order. These have been
the main source of motivation for our work, and so we describe one
of them briey here, somewhat out of context. As we have indicated al-
ready, the general theory will be outlined in the following chapter. Lastly,
at the very end of our work, we present other examples that sparked our
interest.
Again, since functions of the type (5.6) are continuous and Zp is com-
pact, via a scaling we can reduce our consideration to the case where F
is a restricted power series with coecients in Zp . We can extend De-
nition 5.3 by considering functions f : ZN p Zp that can be represented
as in (5.6) with an F Zp [[x0 , . . . , xm ]], but where now each xj is an
N -tuple of variables. We call these arithmetic dierential operators of
order m also. If X is an ane scheme embedded into the ane N -space
over Zp , we let X(Zp ) ZN p be the natural inclusion at the level of Zp -
points. We then call a function X(Zp ) Zp an arithmetic dierential
of order m if it can be extended to an arithmetic dierential operator
ZNp Zp of order m. This suces for now to introduce the Legendre
symbol function as discussed in [10].
Let X be the multiplicative group scheme over Zp embedded into the
ane plane Spec Zp [v, w] via the map u (u, u1 ). Then we have that
X(Zp ) = Z p , and we can talk about arithmetic dierential operators
Zp Z p of order m. The Legendre symbol turns out to be a rather
remarkable example of one such operator, as we see here.
If p is an odd prime, the Legendre symbol
f : Z
p Zp
For the convenience of the reader not familiar with this or several other algebraic
concepts that we use, we recall them in 6.1 below.
Arithmetic dierential operators on Zp 59
is the function a (a/p), dened by

a 1 if a is a quadratic residue mod p,
=
p 1 if a is a quadratic nonresidue mod p.
for all a such that (a, p) = 1. This function is the -dierential operator
of order 1 on Z p given by

a p1 (2n 2)!pn
n1 n pn
= a 2 1+ (1) (a) a
p n=1
22n1 (n 1)!n!

1 (5.7)
p1 a 2
= a 2 1+p p .
a
(See Theorem 3.1.) Indeed, the three expressions involved in this equality
p1
are all congruent to a 2 mod p, and by (5.5), the square of the right
side is just

p1 a ap + pa
a 1 + p p = ap1 = 1.
a ap
We see then that if we dene the series

p1 (2n 2)!pn
F (x, y, , ) = x 2 1+ (1)n1 n y pn
n=1
22n1 (n 1)!n!
as a restricted power series of four variables, independent of , we have

that

1
1 1 p1 a 2
F (a, a , a, a ) = a 2 1+p p .
a
Notice that the Legendre symbol is a locally constant function of
level 1, that is to say, constant on discs of radius 1/p in its domain
of denition. It is therefore analytic. We shall see below that this latter
fact holds in general for any arithmetic dierential operator on Zp .
The theory of arithmetic dierential equations we just introduced is

intrinsically nonlinear. For let us recall that in classical analysis, a linear
dierential operator of order zero on a manifold M is just given by a
function on M , which is thought of as a multiplication operator on the
space of functions itself. Then a linear dierential operator P of order n
is dened to be a linear operator on the space of functions such that the
commutator [P, f ] = P f f P is a dierential operator of order n 1 for
any zeroth order operator f . The application of this line of reasoning to
arithmetic dierential operators is fruitless, and shows immediately their

intrinsic nonlinear nature. The proper analogue of linearity is otherwise.
Indeed, if f : Zp Zp is a Zp -valued function on Zp that we think of as
the multiplication operator given by a f (a)a, using the multiplicative
property of (5.5) encoded in the second identity in (5.2), we have that
[, f ](a) = (f (a)a) f (a)a = ap (f (a)) .
Thus, the commutator [, f ] is the product of (f ) and the p-th power
mapping. The alter egos of the n-th order linear dierential operators
in classical analysis would be compositions of iterates of the operator
and the p-th power Frobenius map up to order n, which makes of them
highly nonlinear expressions in the argument anyway: no commutator
[F, f ] of a nontrivial -operator F and a nontrivial function f can be
linear in the argument.
In appropriate cases, there is a convenient notion of linearity in the
context of arithmetic dierential operators, a notion that was already
present in [6, 8]. It refers to a group homomorphism property in cases
where the domain of the arithmetic dierential operator makes it possi-
ble to talk about such. We outline this notion here. Details will be given
in 6.3.
We place our discussion in the context introduced in Example 5.5
above. We recall that if X is an ane scheme embedded into the ane
N -space over Zp , a function X(Zp ) Zp is an arithmetic dierential
if it can be extended to an arithmetic dierential operator ZN p Zp .
Suppose that the X in this general set-up were to be a commutative
group scheme G over Zp . Then we could dene a linear arithmetic
dierential operator on G to be a dierential operator
G(Zp ) Zp
on G that is also a group homomorphism. Here, the range Zp in the right
side is viewed with its additive group structure. By making this set-
theoretical denition one that is scheme-theoretical varying the ring
Zp , we would obtain the notion of linear arithmetic dierential operator
of [6, 8] that was alluded to above. With this denition in place, we
recover the familiar property that linear operators have, namely that
the dierence of solutions to homogeneous equations is again a solution.
Example 5.5 illustrates on Zp the type of questions we would like
to ask of -functions over Zp with as much generality as possible. The
Legendre symbol admits a representation as an arithmetic dierential
operator of order 1. We shall prove that an analytic function on Zp of
level m can be represented as an arithmetic dierential operator whose

order is m.
Let us reiterate for the moment that when thinking of p-adic numbers
as functions, these functions are in fact analytic. This comes about via
the natural analogy that the representation (2.4) generates. It is an irony
that this analogy is at best somewhat incomplete: though we know of
analytic functions [14], we do not yet have the proper notion of what a
continuous arithmetic function ought to be.
Remark 5.6 Buiums general theory of arithmetic dierential opera-

ur , the completion of
tors [6, 8] is developed over the ring R := Rp := Z p
the maximal unramied extension of Zp , in the role of our Zp here. Let
k = R/pR be the residue eld, and let (R) be the multiplicative group
of roots of unity in R. We recall that the reduction mod p mapping
(R) k
denes an isomorphism whose inverse is the Teichm uller lift. Any ele-

ment of the ring R can be represented uniquely as a series i=0 i pi ,
where i (R) {0}. Using this representation of the elements of R,
we shall see that there exists a unique ring isomorphism
:RR
that lifts the p-th power Frobenius isomorphism on the residue eld
k, and for (R), we have that () = p . This is the associated
homomorphism (5.3) of the p-derivation on the ring R giving rise to the
general theory (see Theorem 6.9 below).
5.1 Multiple primes I

The theory of arithmetic dierential operators introduced above has
been extended to the case where there is more than one prime [17].
Elaborating on this idea requires the use of more complicated rings than
those that play a role in our work on the single prime case. We outline
some of the issues related to this next.
Let us begin by revisiting the presentation above and reformulating
it slightly, reversing somewhat the order followed when introducing the
concept of arithmetic derivative p in (5.5). We now begin our work at the
level of the ring Z instead of Zp , and replace the use of Theorem 2.12 on
Zp directly by that of Fermats little theorem at the level of Z. We then
use the fact that the identity is the unique automorphism in the latter
ring, and so given a prime p, Fermats little theorem ensures the existence
of a unique p derivation on Z associated to this homomorphism p = .
Now in 2.2 the ring Zp is presented as the p-adic completion of Z at the
ideal (p) Z (cf. denition in 6.1). The homomorphism p uniquely
extends by continuity to Zp , and the condition p (a) ap mod p in Zp
will hold for the extension by continuity also. We thus obtain a unique p-
derivation p on Zp associated to p . Notice that under this presentation,
the analogy between (C[x], d/dx) and (Z, ) discussed in the introduction
becomes clear: the fundamental theorems of algebra and arithmetic make
the linear polynomials in C[x] correspond to the primes in Z, and when
p > 2, the set {1, 0, 1} Z is the analogue of the eld of constant
polynomials in C, for these integers are the only solutions of the equation
p n = 0 in Z. When passing to Zp by continuity, the set of constants is
enlarged to encompass the roots of unity of order prime to p.
A theory involving at least a second prime forces us to consider as the
simplest ring to use one that carries suciently many automorphisms.
The ring Z is unsatisfactory in this respect, and the procedure above
requires modications, which we present now. In this way, we point to-
wards the introduction of arithmetic operators in the case of multiple
primes, and the general diculties we face when doing so. The basic ring
to use is essentially obtained from Z by adjoining at least one root of
unity, and then taking its completion relative to the prime ideal gener-
ated by the set of primes we are using. The details illustrate the diculty
that the reader will no doubt see clearly: how do we make two or more
primes interact with each other in a reasonable way.
Given two distinct primes p and q in Z, we consider the polynomial
cp,q in the ring Z[x0 , x1 , x2 ] dened by
cq (xp0 , px1 ) cp (xq0 , qx2 ) p q p q p q
cp,q (x0 , x1 , x2 ) = x + x , (5.8)
p q q 2 p 1
where for each prime p, cp Z[x, y] and p are the polynomial and
p-derivation in (5.1) and (5.5), respectively. Notice that this polynomial
cp,q lies in the ideal (x0 , x1 , x2 )min{p,q} Z[x0 , x1 , x2 ].
We now have the following [17].
Denition 5.7 Let P = {p1 , . . . , pd } be a nite set of primes in

Z. A P -ring is a ring A endowed with pk -derivations pk : A A,
k = 1, . . . , d, such that
pk pl a pl pk a = cpk ,pl (a, pk a, pl a) (5.9)

for all a A, k, l = 1, . . . , d. A homomorphism of P -rings A and

B is a homomorphism of rings : A B that commutes with the
pk -derivations in A and B, respectively.
Let P = {p1 , . . . , pd } be a nite set of primes in Z, and A be a
P -ring with pk -derivations pk , as in the denition. For each index
k in the range 1, . . . , d, we let pk denote the homomorphism (5.3)
associated with the pk -derivation pk . Then condition (5.9) implies the
commutativity relations pk pl = pl pk . Indeed, for all a A, we
obtain that
pk pl (a) = pk (apl + pl pl a)
= apl pk + pk pk apl + pl (pl a)pk + +pk pl pk pl a
= pl pk (a) + pk pk apl pl pl apk + pl (pl a)pk
pk (pk a)pl + pk pl (pk pl a pl pk a)
= pl pk (a) ,
where the last equality follows by substituting cpk ,pl (a, pk a, pl a) for
pk pl a pl pk a, and simple cancellations. In turn, the commutation
of the associated homomorphisms implies the commutation relations
pk pl a = pl pk a.
Conversely, suppose the homomorphisms k , l satisfy the commuta-
tion relations above, and that the pk s are nonzero divisors in A. Then
the conditions (5.9) hold, and we have that
pk pl a = pl pk a
for all a A.
A nontrivial example of a P -ring ring requires it to carry suciently
many automorphisms that can be the homomorphisms associated with
the various arithmetic derivations being considered. Out of necessity, the
basic example of ring to use must be larger than Z and with suciently
many automorphisms.
Example 5.8 Let S Z be a multiplicative system of integers coprime
to p1 , . . . , pd , and let ZS = S 1 Z be the corresponding ring of fractions.
For a given integer m that iscoprime
to p1 , . . . , pd , we consider the m-th
root of unity m = exp 2 m1 , and the polynomial ring A = ZS [m ]
in m . Notice that m is a root of the separable polynomial xm 1, which
has m distinct roots in the algebraic closure of Q, all of them roots of
unity. We have that A is contained in the eld Q(m ), and this extension
of Q is Galois and commutative.
If is a primitive m-root of unity and is an automorphism of Q(m ),
then () is another m-root of unity, which therefore must be of the form

n() for some integer n = n(). The map
G(Q(m )/Q) n() mod m
denes an isomorphism between the Galois group G(Q(m )/Q) and
(Z/mZ) , respectively.
The primes p1 , . . . , pd under consideration dene classes in (Z/mZ) .
We let p1 , . . . , pd be the corresponding elements of the Galois group
G(Q(m )/Q). Then pk (a) apk mod pk for k = 1, . . . , d and a A, and
A becomes a P -ring with respect to the pk -derivations pk associated
to the homomorphisms pk , k = 1, . . . , d.

All the pk -adic completions (see 6.1 for denition) A(pk ) at the ideal

(pk ) Z are P -rings in a natural way. For pl extends to A(pk ) by conti-

nuity, the condition pk a apk mod pk in A(pk ) holds by continuity, and

the condition pl a apl mod pl in A(pk ) holds because pl is invertible

in A(pk ) .
Example 5.9 Let us take a closer look at the ring ZS of the ring
ZS [m ] in the example above. We can take S = dj=1 (Z \ (pj )) and the
ring of fractions ZS will then be
ZP = {m/n : m, n Z such that (n, pj ) = 1 for j = 1, . . . , d} .
By Fermats little theorem, if a ZP , then we have that a apj mod pj ,
j = 1, . . . , d. So we obtain a P -ring structure on ZP . This ring is in fact
the intersection dj=1 Z(pj ) of the localizations (see 6.1 for denition;
cf. with Example 6.1) Z(pj ) of Z at the ideals generated by the various
primes under consideration.
If we were to develop a multiple primes theory parallel to the one we
study here, ZP would have to take on the role of the ring Z. Let us
consider the ideal iP = (p1 , . . . , pd ) generated by the primes in P. This

ideal induces a topology on ZP , and the completion ZiPP of ZP at iP is the

ring that then takes on the role of Zp (again, see 6.1 for the denitions
of the unexplained terms used here). This general approach is a natural
idea to adopt [17].
6
A general view of arithmetic
dierential operators
We present now a fuller view of arithmetic dierential operators by out-

lining their general theory. We shall not use this theory with this degree
of generality anywhere else in this monograph, and the additional de-
tails that we provide later on are only there to make of our treatment a
self contained one when emphasizing the dierent avor these operators
have when the ring Zp is enlarged.
Arithmetic dierential operators over Zp have a special behaviour in
comparison to their behaviour when the ground ring is larger, as in the
general theory. In the general theory [6, 8], the role of the the ring Zp is
played by by the completion Z ur of the maximum unramied extension
p
of Zp , and the p-derivation is now given by
ur
Z

ur
Z
p p
(a) ap (6.1)
a ,
p
ur . The dierences that result
where is the unique lift of Frobenius to Z p
between the ensuing two theories over these rings are analogous to the
dierences between number theoretic statements about nite elds, and
algebraic geometric statements over their algebraic closures. The theory
as we discuss it pertains a single prime p, as in [6, 5]. The extended
theory for several primes may be found in [17]. We shall provide some
details of it also, in 6.4, outlining further what was started in 5.1.
The reader interested in a quick review of the entire theory may nd it
convenient to consult the survey article [13].
As a guiding principle, let us observe that arithmetic dierential
operators were originally dened paralleling the theory of dierential
operators on bundles over a manifold. In order to clarify this analogy,
we recall the latter briey. The clasic denition of dierential operators
65
66 A general view of arithmetic dierential operators
is based on the theory of jet bundles [40], all of which have arithmetic
analogues to be discussed below. Our exposition is aimed at making the
comparison easy.
Given a manifold M and a vector bundle E M of rank l over it, the
k-th jet bundle J k (E) M is dened as follows. Let s (M, E) be a
section of the bundle, and p M . Consider a basis s1 , . . . , sl of local sec-
tions about p dened on a neighborhood U of a local chart (U, ), (q) =
(x1 (q), . . . , xn (q)). We represent s nearby p as s = f1 s1 + + fl sl for
some smooth functions f1 , . . . , fl that are supported in U , and dene s to
be in Zpk (M, E) if, and only if, (x11 xnn fj )(p) = 0 for all multi-indices
= (1 , . . . , n ) such that || k, for each j. The ber Jpk (E, M ) of the
k-th jet bundle over p is by denition the quotient (M, E)/Zpk (M, E). If
we let jk (s)(p) denote the equivalence class of s in Jpk (E, M ), J k (E, M )
is given the topology that makes jk (s) (M, J k (E, M )) a smooth
section.
If E is locally trivialized by U Rl , then J k (E, M ) is locally trivi-
alized by U kj=0 Sj (U, Rl ), where Sj (U, Rl ) is the space of symmetric
j-linear maps from U into Rl . Let {e1 , . . . , en } be the standard basis for
Rn , and set e i to be the i -tuple (ei , . . . , ei ). Then a local section T
i
of S (U, R ) can be identied with {T (e

j l
1 , . . . , en )}||=j ||=j R ,
1 n l
and jk (s) (M, J (E, M )) is then given by the (k + 1)-tuple of map-

k
pings jk (s) = (s, Ds, D2 s, . . . , Dk s), where by denition we have that

(D|| s)(e 1 , . . . , en ) = (x1 xn s), and thus, jk (s) corresponds to
1 n 1 n
{x1 xn s}||k in ||k R .

1 n l
An operator F : (M, E1 ) (M, E2 ) between sections of two bun-

dles E1 and E2 over M is said to be a dierential operator of order k if
it factors through the jet bundle J k (E1 , M ) over M , that is to say, if for
any s (M, E1 ), we have that
F (s)(x) = F (jk (s))(x) = F (x, s(x), (Ds)(x), (D 2 s)(x), . . . , (D k s)(x)) .
We shall see later in this chapter that arithmetic dierential operators
can be introduced in a manner that parallels this.
In the arithmetic theory of dierential operators, all the concepts
above have analogues. At the very elementary level, the role of the
manifold M can be taken to be now that of the scheme Spec A, where
A = Z[q], the analogue of the total space E of a bundle E M that
of a scheme X of nite type over A, and the analogue of the space of
sections (M, E) that of the set X(A) of A-points of the scheme X.
Suitable arithmetic jet spaces can be dened, and with them in place,
the arithmetic dierential operators can be presented as mappings that
factor through them, in exactly the same way as the one described above
for the standard dierential operators on the manifold M .
6.1 Basic algebraic concepts

Since we have arrived at the study of p-adic analysis and arithmetic
dierential operators from the point of view of the classical analyst, we
attempt to make the presentation self-contained for all readers by briey
recalling the standard concepts in algebraic geometry and commutative
algebra that we shall use. The reader with a strong background in algebra
may want to bypass this section altogether. Our basic references for the
concepts recalled here are [1] and [26], respectively.
Let A be a commutative ring with unity. Given a prime ideal p in A,
we let S be the multiplicatively closed set A \ p. Then the localization Ap
of A at p is the ring of fractions S 1 A, where a/b and c/d are considered
equivalent if v(ad bc) = 0 for some v S. Notice that the elements of
S become invertible in Ap , ring that only encodes information nearby p.
Let f A and S = {f n }n0 . In this case, it customary to write Af

for S 1 A.
The mapping a a/1 denes a ring homomorphism A S 1 A.
Notice that here we make use of the unit of the ring A in an essential
manner. Given an ideal i in A we denote by ip the ideal generated by its
image in Ap . This establishes a 1-to-1 correspondence between the set
of prime ideals in A that are contained in p and the prime ideals of Ap .
A local ring is a ring with a unique maximal ideal. The ring Ap is
local, with pp being its unique maximal ideal.
Example 6.1 If A = Z and p = (p) is the ideal generated by a prime

p, then Ap = {m/n : m, n Z , (n, p) = 1}. If f A, Af = {m/f n :
m, n Z , n 0}.
Example 6.2 Let A be the polynomial ring A = k[t] where k is a

eld and t is an n-tuple of indeterminates. If p is a prime ideal of A,
the localization Ap is the ring of all rational functions f (t)/g(t) where
g p. Suppose that k is algebraically closed, and let V = V (p) = {x
k n : f (x) = 0 for all f p}, the variety dened by p. Then Ap can be
identied with the ring of rational functions f (x)/g(x) in k n such that
g(x) = 0 at at least one point x in V .
The prime spectrum Spec A of a ring A and its Zariski topology are
among the most important concepts in algebraic geometry. As a set,
Spec A is the set of all prime ideals of A. Given an ideal a in A, we dene
V (a) to be the set of all prime ideals containing a. With more generality,
given E A, let V (E) denote the set of all primes ideals containing E.
We have that V (E) = V (aE ) where aE is the ideal generated by E. We
obtain a topology on Spec A by declaring its closed sets to be all sets
of the form V (E) = V (aE ) for E A. This is the Zariski topology on
Spec A.
Example 6.3 1. For the ring A = Z, we have that
Spec A = {(0)} {(p)| p a prime number} .
2. Let k be algebraically closed, and consider the polynomial ring in two
variables x = (x1 , x2 ), A = k[x] = k[x1 , x2 ]. Then
Spec A = {(0)}{(p)| p(x) irreducible}{(x1 a, x2 b)|(a, b) k 2 } .
In the case of a multiplicative closed set S, the homomorphism

: A S 1 A
induces an associated mapping a : Spec(S 1 A) Spec A that is
injective. The image US = a(Spec(S 1 A)) Spec A, which consist
of the set of prime ideals of A that are disjoint from S, is provided with
the subspace topology, and the inverse map : US Spec(S 1 A) is
continuous. Thus, Spec(S 1 A) is homeomorphic to US Spec A. When
S = {f n : n 0} for f in A, then US is the open set US = Spec (A\V (f )).
These sets are called principal open sets, and are denoted it by D(f ).
They form a basis for the open sets of the Zariski topology.
The sheaf OSpec A on Spec A is dened as follows. Given any open set

U Spec A, O(U ) consists of all functions s : U pU Ap such that
for each p U we have that s(p) Ap , and there exists a neighborhood
V U of p, and elements a, b of A such that, for each q V , b q, and
s(q) = a/b in Aq . The sections so dened form a sheaf OSpec A with the
induced sum and product operations.
The spectrum of A is the topological space Spec A together with the
sheaf OSpec A on Spec A.
A ringed space is a pair (X, OX ) consisting of a topological space X
and a sheaf of rings OX on X. If F is a sheaf of rings on a topological
space X, and p X, the stalk Fp at p is dened to be the germ of
sections at p, sections dened in a neighborhood of p where any two

such are considered equivalent if their restrictions agree on a smaller
neighborhood that contains p. The stalk Fp is a ring. A morphism of
ringed spaces : (X, OX ) (Y, OY ) is given by a continuous mapping

X Y and a collection of homomorphism U : OY (U ) OX (1 (U )),
U open in Y such that U 1 (U ),1 (V ) = U,V V for any open
sets U , V , U V , U,V the restriction homomorphism of the sheaf. An
isomorphism between ringed spaces is a morphism of ringed spaces that
has an inverse.
A locally ringed space (X, OX ) is a topological space X with a sheaf
OX on X such that the stalk OX,p is a local ring for each point p X. A
morphism : (X, OX ) (Y, OY ) of locally ringed spaces is a morphism
of ringed spaces that induces a local homomorphism between the stalks
of OY and the stalks of OX , a homomorhism being local if for every
x X the maximal ideal of the local ring or stalk at (x) Y is
mapped to the maximal ideal of the local ring at x X.
An ane scheme is a locally ringed space (X, OX ) that is isomorphic
as a locally ringed space to the spectrum of some ring. A scheme is a
locally ringed space (X, OX ) with the property that every point p X
has a neighborhood U so that the ringed space (U, OU ) is isomorphic to
(Spec A, OSpec A ) for some ring A.
A morphism of schemes is a morphisms of locally ringed spaces, and
an isomorphism is a morphism with a two-sided inverse.
Given a ring R, the ane n-space An over R is dened to be the locally
ringed space (Spec R[x1 , . . . , xn ], OSpec R[x1 ,...,xn ] ). This notion coincides
with the earlier one introduced in 3.3.
Example 6.4 Let A be a ring. The subspace of Spec A consisting of all

maximal ideals of A with the induced topology is the maximal spectrum
of A, and is denoted by Max A. Let OMax A be the sheaf of rings in this
case. Since the inverse image of a maximal ideal need not be maximal,
Max A does not have the same nice functorial properties of Spec A.
This algebraic set-up ties well with classical analysis. For we may
realize a paracompact Hausdor space X in terms of its ring of contin-
uous functions C(X). Indeed, for each x X, we may dene
mx = {f C(X) : f (x) = 0} ,
= Max C(X), we obtain a mapping
a maximal ideal. If X
X
X
x mx
By Urysohns lemma, this map is injective, and by a partition of unity

argument it can be proven to be surjective if we assume X to be com-
pact. In this case, since the open sets Uf = {x X : f (x) = 0} and
f = {m X
U : f m} form bases for the topologies of X and X,
respectively, it is easy to see that this map is a homeomorphism.
If K is an algebraically closed eld, the set of maximal ideals in A =

K[x1 , . . . , xn ] corresponds to the set of n-tuples (k 1 , . . . , kn ) Kn , as
a maximal ideal m is of the form m = (x1 k 1 , . . . , xn k n ) for some
k 1 , . . . , k n K. With the induced topology, Max A is identied with Kn .
An algebraic variety X over K is a locally ringed space that is locally
isomorphic to (Max A, OMax A ), where A is a nitely generated algebra
over K. An algebraic variety is an ane variety if it is isomorphic to
(Max A, OMax A ).
Schemes are the rightful generalizations of algebraic varieties. They
parallel the idea of a manifold in classical analysis or geometry, in
particular when we think of reconstructing a paracompact space by
its ring of continuous functions, as in Example 6.4. Schemes allow for
viewing algebraic varieties invariantly, capturing their essential algebraic
properties, and as such, lead to other ideas and methods in their study,
extending the range of applicability of the concept.
A Noetherian ring is a ring where every ideal is nitely generated. A
scheme (X, OX ) is locally Noetherian if it can be covered by open ane
subsets Spec Ai where each Ai is a Noetherian ring. The scheme (X, OX )
is said to be Noetherian if it can be covered by a nite number of open
ane subsets Spec Ai where each Ai is a Noetherian ring.
Let S be a xed scheme. A scheme X over S is a scheme (X, OX )
together with a morphism X S. A morphism f : X Y of schemes
X, Y over S is a morphism compatible with the morphisms to S from
X and Y , respectively. A scheme X over a ring A is a scheme (X, OX )
together with a morphism (X, OX ) (Spec A, OSpec A ).
Given a scheme X over A, the set X(A) of A-points of X are the
sections of X Spec A, that is to say, HomA (Spec A, X).
A scheme X over a ring A is of nite type if X admits a nite covering
of the form X = Ui , where Ui = Spec Ai and Ai is a ring that is a
nitely generated algebra over A.
Remark 6.5 Let S = {schemes} be the category of schemes, and let

F = {functors {rings} {sets}}. Schemes can be reconstructed up to
isomorphism from the functor
: S F,
obtained by dening (X) to be the functor that associates to a ring B

f
the set of B-points X(B) of X, and by sending a morphism X Y
f
to the mapping X(B) Y (B) that takes u Hom(Spec B, X) into
f u Hom(Spec B, Y ). For instance, if we consider the ane space
X = SpecA[x, y] = A2 over A, then its set of B points is A2 (B) = B B,
while for the ane scheme X = Spec(A[x, y]/f (x, y)) the set of B points
is X(B) = {(b1 , b2 ) B B : f (b1 , b2 ) = 0}.
A scheme X can be reconstructed up to isomorphism by its image
under .
If the ground ring is algebraically closed, we can consider the category
V of varieties and morphisms. We can then dene a functor
:VS
that identies a variety with the corresponding scheme, and a morphism

between varieties with the corresponding morphism of schemes.
We shall need the concept of completion of a ring A at an ideal,

which geometrically concentrates attention on a formal neighborhood of
a point, or on a Zariski closed subscheme of its spectrum Spec A.
For if i is an ideal of the ring A, then it induces a topology on A where
a basis of open neighborhoods of 0 is given by the nested sequence of
powers
A i i2 in .
If An = A/in and n : An An1 is the natural mapping, we obtain a

sequence
n n1 2
An An1 A2 A1 .

The completion Ai is the inverse limit of this sequence as n goes to

innity: Ai = lim
(An , n ). It is a complete topological ring.
Example 6.6 1. The p-adic integers Zp , as constructed in (2.7), are
the completion of Z at the ideal (p) generated by p. The completion
of Zur ur
p at (p) is the basic ring Zp of Buiums theory.
2. The completion of a polynomial ring A[x] at the ideal generated by
x is the ring A[[x]] of formal power series.

If A is a Noetherian ring, its completion Ai at i is again Noetherian. We

obtain a canonical continuous homomorphism i : A Ai whose kernel

consists of the zero divisors a such that a 1 i. If the homomorphism
i is an isomorphism, then the ring A is said to be complete with respect

to i. The ring Ai is complete with respect to i(i)Ai Ai .

Let A be a Noetherian ring that is complete with respect to i. On
X = V (i) Spec A we dene the sheaf of topological rings OX by
(D(f ), OX ) = limAf /in Af for D(f ) = D(f ) X. We call the pair

(X, OX ) the formal spectrum of A, and denote it by Spf(A), referring to
i as the dening ideal of Spf(A). A formal scheme is a topological local
ringed space that is locally isomorphic to a formal spectrum.
In a ring A, a nite strictly increasing chain of prime ideals p0 p1
pn is said to have length n. The dimension of A is the supremum
of the lengths of all chains of prime ideals. For instance, dim Z = 1. In a
Noetherian local ring A of dimension n, let m be its maximal ideal. Then
m/m2 has the structure of a vector space over the eld k = A/m. The
Noetherian ring is said to be regular if dimk m/m2 = n. A scheme is said
to be regular if all of its local rings are regular local rings. A scheme X
of nite type over a eld k is said to be a smooth scheme if the scheme
obtained from X by pullback from the eld k to its algebraic closure k
is a regular scheme. These two notions are closely related to each other,
and coincide if the eld k is perfect. In particular, a smooth scheme of
nite type over an algebraically closed eld k is a nonsingular algebraic
variety.
Example 6.7 We consider the classical notion of an ane variety X

over C from the algebraic point of view summarized above.
Let X be the maximal spectrum of some nitely generated C-algebra
A. We then have A = Cn /p for some ideal p. The maximal spectrum
Max C[x1 , . . . , xn ] of the polynomial algebra C[x1 , . . . , xn ] identies with
Cn , and by Hilberts basis theorem, the ring C[x1 , . . . , xn ] is Noethe-
rian. Thus, the ideal p viewed in the polynomial algebra is generated by
nitely many polynomials p1 , . . . , pk , and the ane variety X becomes
identied with the maximal spectrum of C[x1 , . . . , xn ]/(p1 , . . . , pk ), the
ring of functions on X.
On the other hand, if we start with the ring of functions OX , we can
take a nite number of generators x1 , . . . , xn , and using the Hilberts
basis theorem, nd a nite number of relations p1 , . . . , pk among them.
Then we can set X to be the ane variety in Cn cut out by the poly-
nomials p1 , . . . , pk , X = {x Cn : p1 (x) = = pk (x) = 0}.
The embedding of the ane variety X into Cn corresponds to a choice

of generators x1 , . . . , xn of OX . Each generator yields a map X C,
and all of them together the map X Cn . This may be stated in an
invariant manner. Indeed, we consider the embedding into the dual of
the vector space x1 , . . . , xn C spanned by the generators, where x X
gets identied with the evaluation at x mapping evx : x1 , . . . , xn C
that carries f C[x1 , . . . , xn ] into evx f = f (x). If the variety X is
irreducible, the ideal of functions (p1 , . . . , pk ) vanishing on X Cn is
prime. This is reected in the fact that OX contains no zero divisors. By
denition, a variety is irreducible if it cannot be written as the union of
two nonempty closed subsets.
The coordinate free approach follows the ideas in Example 6.4. The
points x of an ane variety X are in one-to-one correspondence with
the maximal ideals ix OX of functions vanishing at x. So if A is a
ring, we associate with it the ane variety X = Spec A whose points
are the maximal ideals in A. We get the identications OX = A and
X = Spec A, respectively. If the ring A has no zero divisors, the variety
X is irreducible.
The identication outlined can be extended to arbitrary ane schemes,
allowing nilpotents in their ring of functions. These would correspond
to multiplicities, or innitesimal directions, in the scheme.
6.2 General p -functions: arithmetic jet spaces

We use the remaining parts of this chapter to provide the reader the an-
nounced fuller view of arithmetic dierential operators, and their general
theory.
For simplicity, we set R = Z ur , and let k = R/pR be the residue eld.
p
This ring carries a p -derivation in a natural way, as we will see below,
and it is the ground ring of the general theory of arithmetic dierential
operators. It is a much richer ring than Zp in that both, R and k, have
very special properties, with the rst being complete and the latter being
algebraically closed. The theory of arithmetic dierential operators is
built for any p -ring A over R, that is to say, a p -ring A together with
a p -ring homomorphism R A. In what follows, A will stand for any
ring of this type unless otherwise indicated.
On a smooth scheme X over R, the ring of arithmetic dierential
operators of order r identies with the ring of global functions of certain
formal scheme J r (X), called the arithmetic r-th p-jet space of X [8]. We
begin with the discussion and denition of J r (X).
Let s1 (x0 , y0 , x1 , y1 ) and s2 (x0 , y0 , x1 , y1 ) be two elements of the poly-
nomial ring Z[x0 , y0 , x1 , y1 ]. Paraphrasing the denition of p -operator
given earlier, we say that a mapping
:AB
is a -operator over an A-algebra B associated to the pair of polynomials
{s1 , s2 } if
(x + y) = s1 (x, y, x, y) , and
(xy) = s2 (x, y, x, y) .
The pair of polynomials {s1 , s2 } Z[x0 , y0 , x1 , y1 ] is said to be
-generic if for a ring A of characteristic zero there exists an operator
: A A such that, if g A[x0 , x1 ] has the property that g(a, a) = 0
for all a A, then g = 0.
Two -operators 1 : A B and 2 : A B are said to be equivalent
over a subring A0 of A if there exists a constant a0 A 0 and f A0 [x]
such that 1 a = a0 2 a + f (a) for all a A.
We have the following [5] result, characterizing equivalent -operators
for generic polynomials:
Theorem 6.8 Let {s1 , s2 } Z[x0 , y0 , x1 , y1 ] be a generic pair over
a ring A of characteristic zero, and let : A B be a -operator
associated to this pair. Assume that the localization Z(p) is contained in
A for some xed prime p. Then is equivalent over Z(p) to the -operator
associated with one of the following four pairs:
1. s1 (x0 , y0 , x1 , y1 ) = x1 + y1 , s2 (x0 , y0 , x1 , y1 ) = x0 y1 + y0 x1 ,
2. s1 (x0 , y0 , x1 , y1 ) = x1 + y1 , s2 (x0 , y0 , x1 , y1 ) = x0 y1 + y0 x1 + x1 y1 ,
3. s1 (x0 , y0 , x1 , y1 ) = x1 + y1 , s2 (x0 , y0 , x1 , y1 ) = x0 y1 + y0 x1 + pn x1 y1 ,
where n is some integer in N, and
n n n
4. s1 (x0 , y0 , x1 , y1 ) = x1 +y1 +(xp0 +y0p (x0 +y0 )p )/p, s2 (x0 , y0 , x1 , y1 )
n n
= xp0 y1 + y0p x1 + px1 y1 , where n is some integer in N.
Notice that the -operator associated with the rst three generic pairs
in the theorem above restricts to the zero map on Z, while the -operator
associated to the last pair yields
n
m mp
(m) = ,
p
which coincides with the restriction of (5.5) to Z in the case when n = 1.

We now dene the basic p -ring structure on R, as follows.
Let (R) be the multiplicative group of roots of unity in R. The reduc-
tion mod p mapping (R) k denes an isomorphism whose inverse is
the Teichmuller lift. By our discussion in 3.2, any x R can be written
uniquely as a series

x= i pi ,
i=0
where i (R) {0}. We thus obtain a mapping
p : R R (6.2)
dened by

p n pn = np pn .
n=0 n=0
Clearly p () = for (R).

p
Theorem 6.9 The mapping (6.2) above is a ring homomorphism that

lifts the p-th power Frobenius homomorphism on k.
Proof. Let N be a positive integer coprime to p, and let N be a

primitive N -root of unity. The Galois group G(Q(N )/Q) is naturally
isomorphic to (Z/N Z) , with the Galois element s corresponding to the
n
class of the integer n such that sN = N (see discussion in Example
5.8).
We consider the class of p in (Z/N Z) . It corresponds to an auto-
p
morphism s = sp of Q(N ) such that sN = N . The automorphism s
induces an automorphism of Z[N ], and therefore, an automorphism of
the completion AN of this ring at any prime ideal PN containing p. In
this manner we obtain an induced automorphism s = sp of the P -adic
completion A of the direct limit of all of these AN s, where P A is the
direct limit of a sequence of ideals PN . Since A is isomorphic to R, and
p
clearly the automorphism s sends p to p and any N into N , we conclude
that s must be equal to p , and so p is a ring homomorphism.
Since p -operators on a p -ring A are in 1-to-1 correspondence with
homomorphisms of the ring A (see (5.4)), we may associate to the
homomorphism p the p-derivation on R given by
p (x) xp
p x = . (6.3)
p
The pair (R, p ) constitutes the basic p -ring of the general theory of
arithmetic dierential operators.
Let us consider an N -tuple of indeterminates y = (y1 , . . . , yN ) over R,
and let y (i) be a family of N -tuple of indeterminates over R parametrized
by Z0 , with y (0) = y. We set
R{y} := R[y (i) |i0 ]
for the polynomial ring in the indeterminates y (i) . This ring has a natural
p -structure on it.
Proposition 6.10 There exists a unique p-derivation
p : R{y} R{y}
(i)
that extends the p-derivation (6.3) on R, and satises the relations p yk =
(i+1)
yk .
Proof. This follows by a rather explicit construction. For let
p : R{y} R{y}
be the unique ring homomorphism extending the homomorphism (6.2)

of Theorem 6.9, and satisfying the relation
p (y (i) ) = (y (i) )p + py (i+1) .
Then, we may dene a p-derivation p : R{y} R{y} by the expression

p (P ) P p
p P := ,
p
where
p (P (x(0) , x(1) , x(2) , . . .) = P p ((x(0) )p + px(1) , (x(1) )p + px(2) , . . .) .
Here, P p stands for the polynomial obtained from P by twisting its co-
ecients by p , and (x(i) )p is the tuple of p-th powers of the components
of x(i) .
This denition implies that p y (i) = y (i+1) , as desired.
We now take advantage of the fact that R is complete, and that the
residue eld k = R/pR is algebraically closed, to develop arithmetic
analogues of the jet spaces that were outlined above. In the sequel, for
a given smooth or formal scheme X over R, we shall denote its ring of
sections by O(X).
Denition 6.11 Let S = {S n }n0 be a sequence of rings. Suppose

we have ring homomorphisms : S n S n+1 , and p-derivations p :
S n S n+1 such that p = p . We then say that (S , , p ),
or simply S , is a p -prolongation sequence. A morphism u : S
S of p -prolongation sequences is a sequence un : S n Sn of ring
homomorphisms such that p un = un+1 p and un = un+1 ,
respectively.
If A is any p -ring over R, we obtain a natural p -prolongation

sequence A by setting An = A for all n, and taking the ring homomor-
phisms to be all equal to the identity. A p -prolongation sequence over
A is a p -prolongation sequence S equipped with a morphism A S
of prolongation sequences.
For the ring R{y} discussed above, we consider the subrings
n
SR := R[y, p y, . . . , pn y] .
n+1 n
We view SR as an SR -algebra via the inclusion homomorphism, and we

observe that p SR SR . Therefore, the sequence of rings SR
n n+1
= {SRn
}
denes a p -prolongation sequence. Let (p) be the ideal generated by p.
We obtain the p-adic completion prolongation sequence

R[y, p y, . . . , pn y](p) ,
and the prolongation sequence
R[[y, p y, . . . , pn y]]
of formal power series rings, with their corresponding p -structures.

We now consider a scheme X of nite type over R, and dene its
p -jet spaces. We begin with the case where X is an ane scheme of
nite type, that is to say, the spectrum of a nitely generated R-algebra
X = Spec (R[y]/I), where y is a tuple of indeterminates, and I is some
ideal. Then we set

Jpn (X) := Spf R[y, p y, . . . , pn y](p) /(I, p I, . . . , pn I) . (6.4)

Notice that Jp0 (X) = X (p) , and that the sequence of rings of sections
given by {O(Jpn (X))}n0 has a natural structure of p -prolongation se-
quence O(Jp (X)) induced by the p-derivation p in Proposition 6.10.
The p -prolongation sequence so constructed has the following univer-
sality property.
Proposition 6.12 Let X be an ane scheme over R. Then for any

p -prolongation sequence S over R and any homomorphism u : O(X)
S 0 , there exists a unique morphism of p -prolongation sequences
u = (un ) : O(Jp (X)) S ,
such that u0 = u.
Proof. It suces to consider the case where X = Spec (R[y]/I) with
I = {0}. We dene

un : R[y (0) , . . . , y(r) ](p) S r
by
un (y (i) ) = pi (u(y (0) )) .
This denes a morphism of prolongation sequences. This morphism has
the desired properties.
The construction X Jpr (X) above for ane schemes over R is com-
patible with localization in the following sense:
Corollary If X = Spec B and U = Spec Bf , f B, then

O(Jpr (U )) = (O(Jpr (X))f )(p) .
Equivalently,

Jpr (U ) = Jpr (X) X (p)
U
(p)
.

Proof. The rings (O(Jpr (X))f )(p) has the structure of a prolongation
sequence by dening

np
g f p g g p p (f n )
p =
fn f n p (f n )

np

2
f p g g p p (f n ) p f p f
= 1 p + p2
f 2np fp fp

3
p f
p3 + .
fp
This prolongation sequence satises the same universality property as
that of the rings O(Jpr (U )).
Denition 6.13 Let X be a scheme of nite type over R. For any
ane Zariski open covering of X

m
X= Ui
i=1
we dene the p -jet space of order r of X by gluing Jpr (Ui ) along Jpr (Ui
Uj ). We denote the resulting formal schemes by Jpr (X), and refer to it
as the r-th p-jet space of X.
The resulting formal schemes have a universality property whose for-

mulation is parallel to the one in Proposition 6.12. Using this property
we see that, up to isomorphism, the formal schemes Jpr (X) depend only
on X in a functorial manner: for any morphism X Y of ane schemes
of nite type, there exist induced morphisms of schemes
Jpr (Y ) Jpr (X) .
Notice that if X itself is ane, we may use the tautological covering of

X consisting of the covering of X given by just one open set, namely X
itself. The space Jpr (X) with respect to this tautological covering coin-
cides with the space Jpr (X) in (6.4). This justies the abuse of notation
we have incurred into when using the same terminology to denote the
jet space of X, whether X is ane or not.
Denition 6.14 Let X be a scheme of nite type over R. The p -jet

spaces of X are given by the projective system of p-adic formal schemes

Jpr (X) Jpr1 (X) Jp1 (X) Jp0 (X) = X (p) . (6.5)
Denition 6.15 Let X and Y be smooth schemes over the xed

p -ring A. By a p -morphism of order r we mean a rule f : X Y
that attaches to any p -prolongation sequence S of p-adically complete
rings, a map of sets X(S 0 ) Y (S r ) that is functorial in S in the
obvious sense.
For any p -prolongation sequence S , the shifted sequence S [i] de-

ned by S[i]n := S n+i is a new p -prolongation sequence. Thus, any
morphism f : X Y of order r induces maps of sets X(S i ) Y (S r+i )
that are functorial in S . We can compose p -morphisms f : X Y ,
g : Y Z of orders r and s, respectively, and get p -morphisms
g f : X Z of order r + s. There is a natural map from the set
of p -morphisms X Y of order r into the set of p -morphisms X Y
of order r + 1, induced by the maps Y (S r ) Y (S r+1 ) arising from the
S r -algebra structure of S r+1 .
Remark 6.16 By the universality property of jet spaces, we have that
Jpn (X Y ) Jpn (X) Jpn (Y )
where the product on the left hand side is taken in the category of
schemes of nite type over R, and the product on the right hand side is
taken in the category of formal schemes over R.
Remark 6.17 By the universality property of jet spaces, the set of

order r p -morphisms X Y between two schemes of nite type over R
naturally identies with the set of morphisms over R of formal schemes

Jpr (X) Jp0 (Y ) = Y (p) .
Proposition 6.18 Let X be a smooth ane scheme over A, and let

u : A[y] O(X) be an etale morphism, where y is a N -tuple of indeter-
minates. Let y (i) be N -tuples of indeterminates parametrized by i Z0 .
Then the natural morphism

O(X (p) )[y (i) |1in ](p) O(Jpn (X))
that sends y (i) into pi (u(y)) is an isomorphism. In particular, we have
an isomorphism of formal schemes over A

Jpn (X) X (p) (AnN )(p) .
Corollary If Y X is an etale morphism of smooth schemes over

A, then

Jpn (Y ) Jpn (X) X (p)
Y
(p)
.
Remark 6.19 The functors X Jpr (X) from the category C of A-

schemes of nite type to the category C of p-adic formal schemes natu-

rally extends to a functor from B to C, where B is the category whose
objects are the same as those in C, hence Ob C = Ob B, and whose mor-
phisms are dened by

HomB (X, Y ) := HomC (X (p) , Y (p) )
for all X, Y Ob B.
We now spell out the relationship between p-jet spaces and arithmetic
dierential operators in this degree of generality. Let X be a scheme of
nite type over R. A global function f O(Jpr (X)) is an arithmetic
dierential equation of order r. It induces a map of sets f : X(R) R.
We want to cast this denition in a manner reminiscent of that given

in Denition 5.3, and reminiscent also of the denition of a classical
dierential operator recalled at the beginning of this chapter.
For any nonnegative integer r, we obtain a natural map r between
the set of R-points X(R) and Jpr (X)(R),
r : X(R) J r (X)(R) ,
and an R-algebra map
O(Jpr (X)) Or (X)
f f .
In order to see this, and since the assertion is local, it will suce to
consider the ane case where X = Spec (R[y]/I) for some ideal I. We
then recall that an R-point x in X(R) is a morphism of R-schemes
Spec R X, which induces a homomorphism O(X) R that we denote
by x also, abusing the notation some. By the universality property in
Proposition 6.12, we obtain a map r (x) : O(Jpr (X)) R, which yields
an R-point of Jpr (X)(R). Therefore, given f O(Jpr (X)), we dene
f(x) = r (x)(f ) .
If we are given an embedding i : Spec (R[y]/I) Spec R[y] = AN into
ane space, and take f to be the class of an element

F R[y (0) , y (1) , . . . , y (r) ](p) ,
we would obtain that
r (x) = (i(x), p (i(x)), . . . , pr (i(x))) ,
f(x) = F (i(x), p (i(x)), . . . , pr (i(x))) ,
which shows that f is a p -function of order at most r. This of course is

the content of Denition 5.3, when the role of R is played by the ring Zp .
By gluing over the nonempty intersections of the elements of a covering
of a scheme X of nite type over R, we obtain an intrinsic global notion
of an arithmetic dierential operator.
Let us point out that for an arithmetic dierential operator of order
r to be dened for a given scheme X of nite type over R, the formal
scheme Jpr (X) must admit globally dened functions.
6.3 The analogue of a p -linear operators for

group schemes
A group variety is a variety X together with a morphism X X X
such that the points of X with this operation form a group, and such
that the inverse x x1 is also a morphism of X. An algebraic group

is a synonym for a group variety. If X S is a scheme over S, a group

law is a morphism X S S X. Let i : X X and : S X be
morphisms of schemes such that = 1lS , where 1lS stands for the
identity morphism of S. Suppose that , i and satisfy the relations
( , 1lX ) = (1lX , ) = 1lX , (i, 1lX ) = (1lX , i) = , and
(1lX , ) = (, 1lX ), where (1lX , ) and (, 1lX ) are the two natural
morphisms X S X S X X that and 1lX dene. Then we say that
the scheme X over S together with the morphisms , i and is a group
scheme over S.
In the general set-up of the previous section, we take X to be a com-
mutative group scheme G over A. An arithmetic dierential operator
gives rise to a mapping G(A) A. If this mapping is also a group ho-
momorphism when A is viewed with its additive group structure, we say
that the operator is a p -character. These are the analogues of the linear
dierential operators in the arithmetic case, as proposed in [6, 8].
The ane line A1 under addition is an algebraic group, denoted by
Ga . Thus, we have that Ga = Spec R[y] is the additive group scheme
over R. On the other hand, A1 \ {0} = Spec R[y, y 1 ] is a group scheme
under multiplication, denoted by Gm . Let us recall that given a scheme
X over the ring A, if B is an A-algebra, we let X(B) denote the set
of all morphisms of A-schemes Spec B X, and any such morphism
is called a B-point of X. If X = A1 = Ga = Spec A[y], then X(B) is
simply the set B itself because a morphism Spec B Spec A[y] is the
same as a morphism A[y] B, and the latter is uniquely determined
by the image of y in B. If on the other hand, X = Spec A[y, y 1 ] =
Gm = Spec A[x, y]/(xy 1), then X(B) = B because we have the
identication HomA (A[y, y 1 ], B) = B via the map f f (y). Finally,
if X = Spec A[x, y]/(f (x, y)), then X(B) = {(a, b) B 2 : f (a, b) = 0}.
These three examples cover all the algebraic groups of dimension one, as
the latter of them encompasses an ane chart of an elliptic curve (E, A)
over A. All the geometric bers of group schemes of relative dimension
one are of this form.
Denition 6.20 Let G and H be smooth group schemes over a

6.3 The analogue of a p -linear operators for group schemes 83
p ring A. We say that G H is a p -homomorphism of order r

if it is a p -morphism of order r such that, for any prolongation se-
quence S , the maps G(S 0 ) H(S r ) are group homomorphisms. A
p -character of order r of G is a p -homomorphism G Ga of order r,
where Ga = Spec A[y] is the additive group scheme over A. The group
of p -characters of order r of G will be denoted by Xrp (G).
Example 6.21 Let us consider the p -jet space of the group schemes
A1 and A1 \ {0}, respectively. In the rst case, we have that A1 = Ga =
Spec A[y] is the additive group scheme over A, and

Jpn (Ga ) = Spf A[y, p y, . . . , pn y](p) .
In the second case, A1 \ {0} = Gm = Spec A[y, y 1 ], and

Jpn (Gm ) = Spf A[y, y 1 , p y, . . . , pn y](p) .
If on the other hand the scheme X is an elliptic curve over A given

on an ane chart by Spec A[x, y]/f (x, y), if we let x(0) denote the tuple
(x, y) then

Jpn (E, A) = Spf A[x(0) , x(1) , . . . , x(n) ](p) /(f, p f, . . . , pn f ) ,
where p f (x0 ) = (f ((x(0) )p + px(1) ) f p (x0 ))/p. The entire jet space
is obtained by using a covering, and gluing the resulting ane pieces
together.
These three cases describe the arithmetic jet spaces of all algebraic
groups of dimension one.
Notice that when X is a group scheme of nite type over R, then (6.5)
is a projective system of groups in the category of p-adic formal schemes
over R.
If G is a group scheme of nite type over R, then the group Xrp (G)
of order r p -characters G Ga identies with the group of homomor-

(p)
phisms Jpr (G) Ga , and thus, it identies with an A-submodule of
O(Jpr (G)). Let
O
p (G) := lim Op (G)
r

be the p -ring of all p -morphisms G A1 , and let
X
p (G) := lim Xp (G)
r

be the group of p -characters G Ga . Then O p (G) has a natural

structure of R[p ]-module, and X p (G) is an R[p ]-submodule. Here,
(R[p ], +, ) denotes the ring generated by R and the symbol p sub-
jected to the relation
p a = ap p ,
for all a R, where ap := p (a).

If X is a scheme of nite type over R, and f O(Jpr (X)) is an arith-
metic dierential equation of order r, let f : X(R) R be the corre-
sponding map of sets. If X = G is a group scheme, f is a p -character
of order r if the said map of sets is a homomorphism. The pre-image
f 1 (0) of 0 R under f is, by denition, the set of solutions of the
arithmetic dierential equation f = 0.
If G is a group scheme of dimension one, the structure of the
R[p ]-module O p (G) of p -characters has been described by Buium [8]
in complete detail. His results are as follows.
Theorem 6.22 [8] On the additive group G = Ga over R, any p

character f is of the form f = ai ip with ai R. That is to say,

Op (G) is a free R[p ]-module of rank 1 with the identity as generator.
Proof. Let K be the fraction eld R[1/p]. Given f Xnp (Ga ), since

Jpn (Ga ) = Spf A[y, p y, . . . , pn y](p) , we may identify f with an element
f R[[y (0) , y (1) , . . . , y (n) ]]. We have the mapping
: K[[y0 , y1 , . . . , yn ]] K[[y (0) , y(1) , . . . , y (n) ]]
dened by (y0 ) = y (0) , (y1 ) = p (y (0) ) = (y (0) )p + py (1) , . . . , (yn ) =

n
np (y (0) ) = (y (0) )p + + pn y (n) . This is a K-algebra isomorphism,
and 1 f is additive in y0 , . . . , yn . But the only additive elements in
K[[ip |0in ]] are those in the K-linear span of {ip |0in }. Thus, we

can complete the proof by showing that if i ai ip pR[ip |0in ], we
then have ai pR, and this is fairly clear.
Theorem 6.23 [8] On the multiplicative group G = Gm over R, there

exists a nonzero p -character f : Gm (R) = R R of order 1, unique
up to multiplication by a constant, and given by
n1

n
1 n1 p p x
fp (x) = (1) ,
n xp
n1
that generates O
p (G) as a free R[p ]-module.
6.3 The analogue of a p -linear operators for group schemes 85
In light of the more complicated nature of the multiplicative group, the

argument to describe its characters is more elaborate than the previous
one for Ga .
Proof. Let us rst observe that

1 1 p (x) 1 p x
fp (x) = log = log 1 + p p .
p xp p x
Since we already know that the n-th jet space of Gm is given by Jpn (Gm ) =

Spf A[y, y 1 , p y, . . . , pn y](p) , we see easily that fp1 is an element of the
group X1p (Gm ), a character of order 1 of Gm .
Given any f Xnp (Gm ), we can compose it with the map

(p)
(p)
Ga Gm
x exp (px)
to obtain an additive character f, which is then an element of the free

R[p ]-module in Theorem 6.22. Now observe that if y = 1 + T and
log y = l(T ) = T T 2 /2 + T 3 /3 + , then
1
fp1 = (p p)l(T ) R[[T ]][T (1) ](p) ,
p
and so

1
(ip fp1 ) exp (pT ) = ip (p p) l(exp (pT )) = i+1
p (T ) pp (T ) ,
i
p
which shows that the family {ip fp1 }i0 is a linearly independent collec-
tion of characters. Thus, f is in the span of {ip fp1 }n1
i=0 , and therefore,
f must be in the R[p ]-span of fp1 . This nishes the proof.
Now if G = E is an elliptic curve over R, then O p (E) is also a free
R[p ]-module of rank 1. But the basis is a character of order 1 or 2
depending on whether E has a canonical lift or not. An elliptic curve E
over R is called a canonical lift if there exists a morphism of schemes
E E over Zp whose reduction mod p is the absolute Frobenius. We
simply formulate the theorem for completeness. The reader is referred
to [8] for its proof, whose discussion exceeds the goals we have for this
monograph.
Theorem 6.24 (Buium [8]) If E is any elliptic curve over R, then

there exists a a nonzero character f 2 : E(R) R of order two whose
group of solutions contains pn E(R) as a subgroup of nite index, and
f 2 generates Op (E) as an R[p ]-module. If E has ordinary reduction
and is a canonical lift of its reduction, there exists a nonzero p -character

f 1 : E(R) R of order 1, unique up to multiplication by a constant,
and in this case, f 2 can be written as a linear combination of f 1 with
coecients taken out of R[p ], and the character f 1 generates the free
R[p ]-module Op (E).
The character fp1 in Theorem 6.23 is the arithmetic analogue of the

Kolchin logarithmic derivative [30], while the character f 2 in Theorem
6.24 is the arithmetic analogue of the Manin map [37].
6.4 Multiple primes II

In extending the theory above to the partial dierential setting, there
are two ways of proceeding, one already explained in 5.1. These ways
are obtained when we consider geometric directions of dierentiation
in addition to the arithmetic one, or when we consider the various
arithmetic directions associated to a set of multiple primes. These two
cases can be briey illustrated using rst the simple minded extension
d
of the analogy between (C[x], dx ) and (Z, p ) used earlier.
Indeed, we now consider the polynomial ring C[x1 , x2 ]. Given primes
p1 , p2 , we may propose as the analogue of the algebra (C[x1 , x2 ], x1 , x2 ),

xi := x i
, i = 1, 2, the triple (Z, p1 , p2 ), where the two arithmetic
n npi
derivatives p1 , p2 : Z Z are dened by pi n := , i = 1, 2.
pi
This captures the essence of the case of several arithmetic directions.
On the other hand, we could simply extend the theory to the partial
dierential case by taking as the analogue of (C[x1 , x2 ], x1 , x2 ) the
triple (Z[q], p , q ), where Z[q] is a polynomial ring in the indeterminate

n ( an q pn ) ( an q n )p
q, and p and q are dened by p ( an q ) :=
p
n d n

n
and q ( an q ) := q an q = nan q , respectively. In this
dq
context, we think of p and q as arithmetic and geometric directions, re-
spectively. The latter case is beyond the scope of this monograph, and
we refer the interested reader directly to the source [15, 16], where the
needed jet spaces foundational to the theory are introduced, and com-
plete statements on the characters on one dimensional group schemes
are made, paralleling the ones above for a single prime. On the other
hand, the rst of the cases above is also outside the scope of this mono-
graph, but we started in 5.1 to explain the diculties in going beyond
a single prime, and we shall elaborate on it further in this section, in a

rather sketchy form. The interested reader may nd the complete details
in [17].
Given a family P = {p1 , . . . , pd } of primes, let A0 be a P -ring, for
instance, the ring of Example 5.8. We denote by K0 its fraction eld.
We let A be the ring A = A0 [[q]] of power series in q, and dene a family
of homomorphism pk : A A, 1 k d, by

pk ( cn q n ) = pk (cn )q npk .
Then A is a P -ring also with respect to the pk -derivations pk associated

to the homomorphism pk s.
Given an n-tuple of variables x, we consider n-tuples of variables xi
indexed by vectors i = (i1 , . . . , id ) in Zd0 such that x = x(0,...,0) . We set
i
P = pi11 . . . pidd , and iP = ip11 . . . ipdd .
Let K0 {x} be the ring of polynomials
K0 {x} := K0 [xi : i 0]
with K0 -coecients in the variables xi with i Z0 . The

homomorphisms pk : A0 A0 are extended to ring endomorphisms
pk : K0 {x} K0 {x} by pk (xi ) = xi+ek , so that we have that
xi = iP x for all i. Clearly pk pl (a) = pl pk (a) for all a K0 {x}, all
k, l. If we consider the pk -derivations pk : K0 {x} K0 {x} associated
to the pk s, then K0 {x} is a P -ring that is generated as a K0 -algebra
by the elements P i
x, i 0:
K0 {x} = K0 [P
i
x : i 0] .
We dene the ring of P -polynomials A0 {x} to be the A0 -subalgebra

of K0 {x} generated by all the elements P
i
x:
A0 {x} := A0 [P
i
x : i 0] .
The ring A0 {x} is strictly larger than the ring A0 [xi : i 0]. And the
family {P
i
x : i 0} is algebraically independent over A0 , so A0 {x} is a
i
ring of polynomials in the variables P x.
Lemma 6.25 We have that pk A0 {x} A0 {x} for k = 1, . . . , d, and

so A0 {x} is in a natural manner a P -ring.
Proof. By the denition of a p-derivation (see (5.2)), the sets Sk :=

{a K0 {x} : pk a A0 {x}} are A0 -subalgebras of K0 {x}. Therefore,
it suces to show that P i
x Sk for all i and k. This can be done
using the the commutation relations (5.9) via an induction argument on

(i, k) Zd0 Z0 with respect to the lexicographic order.
Proceeding as before the statement of the lemma, we see that the
system
i
A0 [P x : i r]
has a natural structure of P -prolongation system.
Example 6.26 Let T be a tuple of indeterminates, and let
Ar = A0 [[P
i
T : i r]] .
Then, the structure of P -prolongation sequence above induces a struc-

ture of P -prolongation sequence on the sequence of rings (Ar ). For the
pk -derivation pk sends the ideal
i
Ir := (P T : i r) Ar
into the ideal Ir+ek Ar+ek .
We are now ready to dene the P -jet spaces. As in the case of a single
prime [8] discussed above, we now have the following existence result for
a universal prolongation sequence.
Proposition 6.27 Let A0 be a nitely generated A0 -algebra. Then

there exists a P -prolongation sequence A over A0 , with Ar nitely gen-
erated over A0 , satisfying the following property: for any P -prolongation
system B over A0 and any A0 -algebra homomorphism u : A0 B 0 ,
there exists a unique morphism of P -prolongation systems u : A B
such that u0 = u.
Proof. We express the nitely generated algebra A0 as

A0 [x]
A0 =
(f )
for a tuple of indeterminates x, and a tuple of polynomials f . Then we
set
A0 [Pi
x : i r]
Ar = i f : i r)
.
(P
We check easily that A = (Ar ) has the universality property in the

statement.
Denition 6.28 Let X be an ane scheme of nite type over A0 . Let

A0 = O(X) and let Ar be as in Proposition 6.27. Then the scheme
JPr (X) := Spec Ar
is called the P -jet space of order r of X.
By the universality property in Proposition 6.27, up to isomorphism

the scheme JPr (X) depends on X alone, and is functorial in X: for any
morphism X Y of ane schemes of nite type, there are induced
morphisms of schemes
JPr (Y ) JPr (X) .
Notice that when P consists of a single prime p, the p-adic completions

of the schemes Jpr (X) are those introduced and studied in [8, 11, 6],
and discussed earlier. For arbitrary P, the schemes JPr (X) above were
independently introduced by Borger in [4], where they are denoted by
Wr (X).
Lemma 6.29 Let X be an ane scheme of nite type over A0 and let
Y X be a principal open set of X, O(Y ) = O(X)f . Then O(JPr (Y ))

O(JPr (X))fr where fr = ir iP (f ). In particular, the induced mor-
phism JPr (Y ) JPr (X) is an open immersion whose image is principal,
and if we view this morphism as an inclusion, and Z X is another
principal open set, then we have that
JPr (Y Z) = JPr (Y ) JPr (Z) .
Proof. We can check that O(JPr (X))fr has the universality property
of O(JPr (Y )). The pk -derivations on O(JPr (X))fr are dened via the
formula
a bpk a apk b
pk pk
pk = .
b bp pk (b)
Denition 6.30 Let X be a scheme of nite type over A0 . An ane

open covering

m
X= Xi (6.6)
i=1
is called principal if
Xi Xj is principal in both, Xi and Xj , (6.7)

for all i, j = 1, . . . , m. If c = {Xi }m

i=1 is a principal covering of X, we
dene the P -jet space of order r of X with respect to this covering by
gluing JPr (Xi ) along JPr (Xi Xj ), and it is denoted by Jc,Pr
(X).
If X is ane, we may use the tautological covering c of X, that is to

say, the covering of X with a single open set, the set X itself. Then the
r
space Jc,P (X) coincides with the space JPr (X) of Denition 6.28. Notice
that any quasi-projective scheme X admits a principal covering.
r
In general, the schemes Jc,P (X) in the denition above depend on the
covering c in a nontrivial manner. As shown and indicated already in
[17], even though they have this deciency, they suce for the extension
of the theory we seek. Because it is the case that the rings of global
functions O(Jc,P
r
(X)p
j ) on the pj -adic completion Jc,P
r
(X)p
j of Jc,P
r
(X)
do not depend on the covering c, and are functorial in X. In fact, we
have the following [17, Proposition 2.22].
Proposition 6.31 Let us assume that X is a smooth scheme of nite

type, quasi-projective and with connected geometric bers over A0 =
Z(P) := dl=1 Z(pl ) . Let c and c be two principal coverings of X, and
r
Jc,P (X) and Jcr ,P (X) be the corresponding jet spaces. Then, there is a
natural isomorphism
O(Jc,P
r
(X)pk ) O(Jcr ,P (X)pk ) .
We therefore drop the covering from the notation, and denote the
isomorphism class of rings O(Jc,P r
(X)pk ) simply by O(JPr (X)pk ).
When P consists of a single prime p, we have interesting formal func-
tions f O(Jpr (X)p
) for a handful of interesting situations. In the case of
several primes, we would like to glue together family of these elements
fpj O(JPr (X)p
j ), j = 1, . . . , d. This gluing cannot be done directly
since, for instance, in the case where X is ane, each fpj is a function
on the tubular neighborhood Spf O(JPr (X))p
j of Spec O(JPr (X)) Fpj
in Spec O(JPr (X)), and these tubular neighborhoods are disjoint. We
solved this in [17] by dening the notion of the analytic continuation
of the various elements on the family.
Denition 6.32 Let A be a ring, I be an ideal in A, and P =

{p1 , . . . , pd } be a nite set of primes in Z that are noninvertible in A/I.
We say that a family

d
f = (fk ) Apk (6.8)
k=1

can be analytically continued along I if there exists f0 AI such that

the images of f0 and fk in the ring A(pk ,I) coincide for each k = 1, . . . , d.
In that case, we say that f0 represents f . If X = Spec A, we denote
by OI,P (X) the ring of families (6.8) that can be analytically continued
along I.
From this point on in the section, we take the P -ring A0 to be just

Z(P) . Given a Z(P) -point P : Spec Z(P) X, by the universality prop-
erty we obtain a unique lift to a point P r : Spec Z(P) JPr (X) that is
compatible with the action of P .
Denition 6.33 Let X be an ane scheme of nite type over Z(P) ,

P be a Z(P) -point P : Spec Z(P) X, and P r : Spec Z(P) JPr (X)
be its unique lift compatible with the action of P . We denote by P r
O(JPr (X)) also the ideal of the image of P r . By a P -function on X of
order r that is analytically continued along P we mean a family

d
f = (fk ) O(JPr (X))pk (6.9)
k=1
that can be analytically continued along P r . We denote by OrP,P (X) the

ring of all P -functions on X of order r that are analytically continued
along P .
d
Notice then that f = (fk ) k=1 O(JPr (X))pk can be analytically
continued along P r if, and only if, there exists an element
r
f0 O(JPr (X))P
that represents f , that is to say, such that the images of f0 and fk in

r )
O(JPr (X))(pk ,P
coincide for each k = 1, . . . , d. Thus, the the ring of all P -functions on

X that are analytically continued along P is
OrP,P (X) := OP r ,P (JPr (X)) .
We now have the following denition, which unravels the analytic

continuation concept in a pragmatic way under some general hypotheses
on X and P .
Denition 6.34 Let X be a smooth ane scheme over Z(P) . A Z(P) -

point P : Spec Z(P) X of X is called uniform if there exists an etale
map of Z(P) -algebras Z(P) [T ] O(X), where T is a tuple of indeter-

minates, such that the ideal of the image of P in O(X) is generated by
T . We refer to T as uniform coordinates. Then T is a regular sequence
in O(X), and the graded ring associated to the ideal (T ) in O(X) is
isomorphic to Z(P) [T ]. We have that

O(X)T Z(P) [[T ]] ,

and similarly that

O(X)(pk ,T ) Zpk [[T ]] .
For a general scheme X, we say that a Z(P) -point P of X is uniform if
there exists an ane open set X X that contains P such that P is
uniform in X .
We now have the following.
Denition 6.35 Let X be a smooth quasi-projective scheme over Z(P)
with geometrically connected bers, and let P be a uniform point in some
ane open set X . Let us denote by OrP,P (X) the pre-image of OrP,P (X )
via the restriction map

d
d
d
O(JPr (X)pk ) O(JPr (X )pk ) = O(JPr (X ))pk .
k=1 k=1 k=1
Elements of the ring OrP,P (X)

are referred to as P -functions of order
r on X that are analytically continued along P . We dene the ring of
P -functions on X that are analytically continued along P by
O
P,P (X) := lim OP,P (X) ,
r

with its natural P -ring structure.

Example 6.36 Let X = An = Spec Z(P) [T ] be the n-dimensional
ane space over Z(P) , where T is an n-tuple of indeterminates. We let
i
P be the zero section, and let t be the tuple of indeterminates (P T )ir .
If ordpk denotes the pk -adic order, then

OrP,P (An ) { aj tj Z(P) [[t]] : lim ordpk aj for each k} .
|j|
Since in this case Spec Z(P) X, the lift to JPr (X) of other Z(P) -points,
and the description of the P functions that can be analytically continued
along them, is carried in a similar manner via a translation of the zero
section.
We end our discussion of the theory for several primes by stating the
main results about group characters and the analogues of those results
of Buium given in 6.3 for the case of a single prime. Again, this is just a
sketch, and the interested reader is referred to [17] for complete results.
Let G be a smooth group scheme over Z(P) with multiplication :
GG G, where, = Z(P) . If G is ane, by the universality property
we have that JPr (G) is a group scheme over Z(P) . In the nonane case,
the covering dependent denitions may create problems, but we attach
to this general situation a formal group law as follows.
Let Z be the identity Spec Z(P) G, and assume Z is uniform, with
uniform coordinates T (see Denition 6.34). This condition on Z is triv-
ially satised in the cases where G = Ga , G = Gm , and G = E, an
elliptic curve. For arbitrary G, the condition that Z be uniform is not
restrictive whenever we choose the primes in P to be suciently large.
Under the conditions above, Z Z G G is a uniform point with
uniform coordinates T1 , T2 induced by T . We have an induced homo-
morphism Z(P) [[T ]] Z(P) [[T1 , T2 ]] that sends the ideal (T ) into the
ideal (T1 , T2 ). By the universality property applied to the restriction
Z(P) [T ] Z(P) [[T1 , T2 ]], we obtain morphisms
Z(P) [P
i
T : i r] Z(P) [[P
i i
T1 , P T2 : i r]]
that send the ideal generated by the variables into the corresponding
ideal generated by the variables. Thus, we have an induced morphism
Z(P) [[P
i
T : i r]] Z(P) [[P
i i
T1 , P T2 : i r]] .
We call Gr the image of the variables {P

i
T : i r} under this last
homomorphism. Then the tuple G is a formal group law over Z(P) .
r
Denition 6.37 Let G be a quasi-projective smooth group scheme

over Z(P) with geometrically connected bers. Let us assume that the
identity is a uniform point. Then there are homomorphisms
, pr1 , pr2 : OrZ,P (G) OrZZ,P (G G)
induced by the product and the two projections. We say that a

P -function f OrZ,P (G) of order r on G that is analytically
continued along Z is a P -character of order r on G if
f = pr1 f + pr2 f .
We denote by XrP (G) the group of P -characters of order r on G. We

dene the group of P -characters on G to be

X
P (G) := lim XP (G) .
r

The condition that f be a P -character of order r on G is that there

exists
f0 Z(P) [[P
i
T : i r]]
that represents f such that
f0 (Gr (T1 , T2 )) = f0 (T1 ) + f0 (T2 ) . (6.10)
Here, Gr is the corresponding formal group law, and f0 (T ) stands for
i
f0 (. . . , P T, . . .).
The group XrP (G) of P -characters of order r on G is a subgroup of the
additive group of the ring OrZ,P (G). The group X P (G) of P -characters

on G is a subgroup of OZ,G (G).
For any P -character XrP (G) and any P -ring A, there is an
induced group homomorphism
A : G(AP ) AP , (6.11)
where AP is viewed as a group with respect to addition. Thus, we may
speak of the group of solutions Ker A of the character . The mapping
A A is functorial in A.
Denition 6.38 Let P be a Z(P) -point of G. We say that a
P -character XrP (G) can be analytically continued along P if
OrP,P (G).
We now exhibit the P -characters for the groups Ga and Gm , respec-
tively, paralleling the discussion in 6.3 for a single prime:
The additive group. Let us consider the additive group scheme over
Z(P) ,
Ga := Spec Z(P) [x] .
The zero section is uniform, with uniform coordinate T = x. We have
that
O(JPr (Ga )) = Z(P) [P
i
x : i r] .
Let us consider the polynomial ring

Z(P) [P ] := Z(P) [pl : l I] = Z(P) n ,
nN
where I := {1, . . . , d}, N is the monoid of the natural numbers generated

by P, and the pl s are commuting variables. If i = (i1 , . . . , id ) Zd0 ,
we set Pi = pi11 . . . pidd . If n = Pi we set n = iP = ip11 . . . ipdd . The
ring Z(P) [P ] is then viewed as the ring of symbols of the arithmetic
partial dierential equations that we dene. For r Zd0 , we consider
the symbol

:= cn n Z(P) [P ] .
n|Pr
We may consider the element = x O(J r (Ga )), and using the
diagonal embedding, identify it with an element

d
O(JPr (Ga ))pk .
k=1
This clearly denes a P -character on Ga . In what follows, and

shall be identied with each other.
The following result for the additive group is simple. Later on, a less
elementary analogue for the multiplicative group Gm will follow. We will
not discuss the multiple prime situation for elliptic curves here.
Theorem 6.39 Let be a P -character on Ga of order r. Then
1. can be uniquely written as = x with

= ( cn n )x , cn Z(P) .
n
2. can be analytically continued along any Z(P) -point P of Ga .
Proof. Let XrP (Ga ), and suppose is represented by a series

0 Z(P) [[P
i
x : i r]]. We view this representative 0 as an element
of
Q[[P
i
x : i r]] = Q[[iP x : i r]] .
Notice that we have that

0 (. . . , iP x1 + iP x2 , . . .) = 0 (. . . , iP (x1 + x2 ), . . .)
= 0 (. . . , iP x1 , . . .) + 0 (. . . , iP x2 , . . .) .

Then 0 = cn n x, where cn Q.
We have that
n x xn mod (P
i
x : i r)
in the ring Q[P

i
x : i r]. Therefore, we obtain that

(0 )|Pi x=0; 0
=ir = c n xn .
It follows that cn Z(P) , which completes the proof of the rst assertion.
The second assertion is clear. We merely refer to Example 6.36 for
(some) details.
Corollary The group of P -characters X
P (Ga ) is a free Z(P) [P ]-
module of rank one with basis x.
The multiplicative group Let us now consider the multiplicative
group scheme over Z(P) ,
Gm := Spec Z(P) [x, x1 ] .
The zero section is uniform, with uniform coordinate T = x 1.
The formal group law G0 corresponding to T is
G0 (T1 , T2 ) = T1 + T2 + T1 T2 ,
and the logarithm of this formal group law is given by the series

Tn
lGm (T ) = (1)n1 Q[[T ]] .
n=1
n
By Lemma 6.29, we have that

1
O(JPr (Gm )) = Z(P) i
P x, : ir .
iP (x)
For each k, we consider the series

n
pn1 pk x
pk = p1k := (1)n1 k
O(JPek (Gm ))pk ,
n=1
n xpk
and the element

pl
(n)

pk := 1 = n Z(pk ) [pl : l Ik ] .
pl n
lIk nNk
In this expression, we have set Ik := {1, . . . , d}\{k}, Nk is the monoid

of the natural numbers generated by pk := P\{pk }, is the Mobius
function and the pl s are a set of d-variables that commute among them-
selves. Let us then consider the family

d
(pk pk ) O(JPe (Gm ))pk ,
k=1
where e = e1 + + ed = (1, . . . , 1). As we see right away, this family is

a P -character that we shall denote by
e
m XeP (Gm ) . (6.12)
Theorem 6.40 The family (pk pk ) is a P -character of order e on

Gm .
Proof. We begin by observing that
pk (1 + T ) = pk T + cpk (1, T ) (T, pk T ) Zpk [[T, pk T ]] , (6.13)
where cp (x, y) is the polynomial (5.1) associated with p. Then the image
of pk pk in Qpk [[P
i
T : i e]] is equal the following series:

n
n 1 (1) n1
(1 + T )
(pk ) n1 (1+T ) p
pk n=1 n
pk
(1+T )pk = p k
1
pk k n=1 n (1 + T )pk

1 pk (1 + T )
= pk lGm 1
pk (1 + T )pk
1
= p (pk pk )lGm (T )
pk k
d
pl
= (1 ) lGm (T ) .
pl
l=1
Notice that by (6.13), this series is convergent in the topology given by

the maximal ideal of the ring. We shall denote this series by 0e .
We have that 0e has coecients in Q Zpk = Z(pk ) , and is the same
for all k = 1, . . . , d. Hence, 0e has coecients in Z(P) , and it represents
the family (pk pk ). We also have that 0e satises the condition (6.10)
because

d pl
0e (Ge (T1 , T2 )) = l=1 (1 pl ) lGm (T )(G0 (T1 , T2 )

d pl
= l=1 (1 pl
) (lGm (T1 ) + lGm (T2 ))
= 0e (T1 ) + 0e (T2 ) .
This completes the proof.
Remark 6.41 Let us consider two primes, so d = 2. Then we have

(1,1)
that the character m above has the factorization

p1 p2 Tn
1 1 (1)n1 Q[[T, p1 T, p2 T, p1 p2 T ]] .
p1 p2 n
This is very much analogous to the factorization of the 2-dimensional
positive Laplacian when it acts on the logarithm of a function. Indeed,

we have
1
log u = z z log u ,
4
which can be written as

zu z u
z = z ,
u u
Here, z = x + iy is the complex coordinate on C = R2 , and =
(x2 + y2 ).
The analogy above is of limited scope. For it is not clear how to tie
up dierent primes through a conjugation operation. However, there is
(1,1)
another way of looking at the factorization above for m , and in this
e
other way, we can consider quite naturally the character m for
an ar-
d pj
bitrary number of primes. For the product j=1 1 pj can be
thought of as the factorization of a d-th order operator whose symbol
decomposes into the product of linear factors with distinct roots along
the various arithmetic directions given by the primes. In the classical
case, these are of course the strictly hyperbolic dierential operators.
In this sense, the arithmetic Laplacians of [17] exhibit a hyperbolic be-
haviour. The pursue of this idea might be worthwhile to undertake in
the future.
We now show that m e

XeP (Gm ) generates the space of all
P -characters of Gm , and determine all of the P -characters that can
be analytically continued along any given Z(P) -point P of Gm .
Theorem 6.42 Let be a P -character of order r on Gm . Then
1. can be uniquely written as

=( e
cn n )m , cn Z(P) .
n
2. can be analytically continued along a Z(P) -point P of Gm if, and

only if, either P is a torsion point or n cn = 0.
Proof. Let 0 be the series representing . Then we have an equality
of the form

0 = ( dn n )l(T )
n|Pr
in Q[[P
i
T : i r]] for dn Q.
Indeed let e(T ) Q[[T ]] be the compositional inverse of l(T ). Then

the series
i i
(. . . , P (T ), . . .) := 0 (. . . , P (e(T )), . . .)
satises the identity
i i i
(. . . , P (T1 + T2 ), . . .) = (. . . , P T1 , . . .) + (. . . , P T2 , . . .) .
By the argument used in the proof of Theorem 6.39, we conclude that

= dn n (T ), with dn Q, which completes the proof of the asser-
tion.
By this preliminary result, if we set Pi
T = 0 for i = 0, we obtain that

( dn n ) l(T ) Z(P) [[T ]] ,
where n T := T n .

Now we can see that if a polynomial = ln n Q[p1 , . . . , ps ]
satises that
l(T ) Zpk [[T ]] Q
for some k {1, . . . , s}, then is divisible in the ring Q[p1 , . . . , ps ] by
pk pk .
Indeed, let us divide by pk pk in Q[p1 , . . . , ps ], so that

=( an n )(pk pk ) + bn n ,
for an , bn Q, and bn = 0 if pk |n. We prove that the remainder term
above is identically zero by showing that bn = 0 for all n.

Since (pk pk ) l(T ) Zpk [[T ]], it follows that ( bn n ) l(T )

Zpk [[T ]] Q. We may assume ( bn n ) l(T ) Zpk [[T ]]. We have that
T nm
( bn n ) l(T ) = (1)m1 bn . (6.14)
n m
m

Let us x integers n , 1. By looking at the coecient of T n pk in
(6.14), we obtain that
n nbn
(1) n Zpk

n pk
n|n
because the equality nm = n pk with n 0 mod pk implies that m = pk

and that Z, n = n , so n | n , and m = nn pk . Since n is odd for
bn = 0, it follows that

nbn pk Zpk ,
n|n
and since this is true for all , we obtain that

nbn = 0 .
n|n
As this is true for all n , we may now use the Mobius inversion formula
to conclude that bn = 0 for all n. And this completes the proof of the
assertion.
Applying this result, it follows that
d

p
dn n = ( cn n ) 1 k
pk
k=1
for some cn Q. We may proceed by induction on n to show that

cn Z(P) for all n. This completes the proof of the rst part of the
Theorem.
For the proof of the second part, let : Gm Gm be the translation
dened by the inverse of P , and let be the automorphism dened by
on the various rings of functions. Since OrZ,P (Gm ) (recall that Z
is the identity Spec Z(P) Gm ), we have that OrP,P (Gm ). But
if = (k ) then k = k k (Pk ) for all k. Now, if P is torsion or

if n cn = 0, it is then clear that k (Pk ) = 0. So k = k , hence
OrP,P (Gm ).
Conversely, let P be non-torsion and suppose that a Z (P) is a given

number. Let n cn = 0, and assume that OrP,P (Gm ). We derive
a contradiction. For let p = p1 and b := ap1 1 + pZ(p) , so b = 1.
By Mahlers p-adic analogue of the Hermite-Lindemann theorem [35,
3] , we have that log b Q, where here log : 1 + pZp pZp is the
p-adic logarithm. Since OrP,P (Gm ) and OrP,P (Gm ), it follows
that OrP,P (Gm ). But = (k (a)) so, in particular,
1 (b) Q. But
d

1
1 (b) = ( cn ) 1 log b .
n
pl
l=1

Since n cn = 0, it follows that log b Q, the desired contradiction.
Consider the augmentation ideal

Z(P) [P ] :=
+
cn n Z(P) [P ] : cn = 0 .
n n
Roughly speaking this says that the p-adic exponential function is transcendental
at nonzero algebraic arguments.
Corollary The group of P -characters X P (Gm ) is a free Z(P) [P ]-

e
module of rank one with basis m . The group of P -characters in X
P (Gm )
that can be analytically continued along a given non-torsion point P of
Gm is isomorphic with the augmentation ideal Z(P) [P ]+ as a Z(P) [P ]-
module. This group is the same for all non-torsion P s.
e
We now compute the group of solutions of the P -character m in
(6.12).
Theorem 6.43 Let A be the P -ring Z(P) [m ] in Example 5.8, and let
e
m,A : Gm (AP ) = A
P AP be the homomorphism (6.11) induced by
e
m . Then
e
Ker m,A = (A
P )tors .
Here, tors denotes the torsion group of an Abelian group .

e
Proof. The nontrivial inclusion to prove is that Ker m,A (A
P )tors .
Let us take Q = (Qk ) Ker m,A
e
so that
pk (pk (Qk )) = 0 (6.15)

for all k. Here Qk Apk = APk1 APk2 , where pk = Pk1 Pk2
is the prime decomposition of pk A. In order to show that Q is torsion,
we may replace Q by any of its powers. So we may assume that Qk
1 + pk Apk for all k. Then (6.15) produces that
d

(pl pl ) lGm (Qk 1) = 0 .
l=1
Now the map

Apk Apk
(pl pl )
is injective for all k, l, assertion that we prove below. Using this result,
we conclude that lGm (Qk 1) = 0, which implies that Qk = 1 by the
injectivity of lGm : pk Apk pk Apk . This completes the proof of the
Theorem.
For the proof of the assertion, let us assume that (pl pl ) = 0.
We also have that M pl = , M := [Q(m ) : Q]. Since the polynomials
pl pl , pl 1 Q[pl ] are coprime, it follows that = 0, as desired.
M
Elliptic curves. We present the statement for elliptic curves for com-
pleteness.
Theorem 6.44 [17] Let E be an elliptic curve over A with ordinary

reduction at all the primes in P. Then there exists a character E 2e
of
order 2e = (2, . . . , 2), unique up to a unit in A, and such that every
other P -character on E is obtained in an appropriate sense from it.
2e
Remark 6.45 There is a factorization for E analogous to the fac-
e
torization for m discussed in Remark 6.41, but not as explicit. For we
now have that

d
pj pj 2
1 aj + pj lE (T ) ,
j=1
pj pj
where lE (T ) is the logarithm of the formal group law attached to E,

and aj Z is the trace of the pj -power Frobenius on the reduction mod
2e
pj of E. We now have that the character E admits a factorization
with factors that are operators of order 2 along each of the arithmetic
directions given by the d primes in P. This property makes of E 2e
an
operator that exhibits a hyperbolic nature, as in the case of the character
e
m of Remark 6.41 that arose when treating the multiplicative group.
2e
In this sense, the order of E should be taken to be the scalar 2d, and
not the multi-index 2e.
7
Analyticity of arithmetic dierential
operators
We now retake the study of arithmetic dierential operators over the

ring Zp by proving that any arithmetic dierential operator
F (x, x, . . . , r x) : Zp Zp
is an analytic function in the sense of [36, 42]. The following Lemma lays
the ground work to accomplish this task [14].
As a polynomial function in x, the arithmetic derivative (x) is a
p-adically continuous function, and the same is true of general arith-
metic dierential operators, or -functions, on Zp . What we show here
is that these functions hae another property, in addition to their mere
continuity.
Lemma 7.1 If x = a + pn u is an element of the disc a + pn Zp , let us

write k x as a polynomial function in u of degree pk ,
k

p
k
x= cka,j uj ,
j=0
with coecients cka,j Qp . Then the following p-adic estimates hold:

1. cka,0 p 1.
1
2. cka,1 p = nk .
p
1
3. cka,j p (nk+1)j1 , 2 j pk .
p
Proof. The rst of the estimates follows by the identity cka,0 = k (a),
which is a p-adic integer since ranges in Zp . We prove the remaining
two estimates by induction on k.
103
104 Analyticity of arithmetic dierential operators
The estimates are clear for k = 0. We assume now that they hold for
k 1. That is to say, we have
k1
p
k1
x= ck1 j
a,j u ,
j=0
where, for j 1, the coecients ck1

a,j satisfy the estimates
k1 1
c
a,j p ,
p(nk+2)j1
and the equality holds in the case when the index j is 1. By (5.5), we
have that
pk1 k1 j pk1 k1 j p
k j=0 ca,j u ( j=0 ca,j u )
x= .
p
We use the multinomial expansion in order to express the p-th power
above as a polynomial in u. We obtain

k1 0
k1
p
ck1
a,j p (ca,0 ) (ck1
a,pk1
)pk1 j
k
x= u
j
u,
p 0 , . . . , pk1 p
j=0
where for convenience we have set j = 00 + 11 + + pk1 pk1 , and

the last sum is over all nonnegative multi-indices = (0 , . . . , pk1 ) of
weight p. Notice that exponent j of u in that sum ranges in between 0,
corresponding to the multi-index = (p, 0, . . . , 0), and pk , corresponding
to the multi-index = (0, . . . , 0, p). This shows that k x is a polynomial
of degree pk in u, as desired. There remains only the estimation of the
p-adic norm of its coecients.
If we write
k

p
k
x= cka,j uj ,
j=0
then we have the relations

ck1
a,j

p
k1 0
(ca,0 ) (ck1
a,pk1
)pk1
cka,j =
p 0 , . . . , pk1 p
j =j
for 0 j pk1 , or

p

(ck1
a,0 )
0
(ck1
a,pk1
)pk1
cka,j =
0 , . . . , pk1 p
j =j
Analyticity of arithmetic dierential operators 105
for pk1 < j pk . Since each multinomial coecient is a rational inte-

ger, by the induction hypothesis applied to each of the coecients ck1 a,l ,
we have that

k1 0

p (ca,0 ) (ck1
a,pk1
)pk1 p
(nk+1)j .
0 , . . . , pk1 p p
p
If j = 00 + 11 + + p k1
pk1 = j, the right hand side of this
expression is simply p(nk+1)j+1 , and we may now estimate the p-adic
norm of cka,j for any j 1 using the expression for the coecient cka,j
given above. Indeed, by the non-Archimedean property of p , and the
estimate just derived, it follows that
k 1
ca,j
p p(nk+1)j1
for any j 1.
The estimate above is sharp for j = 1. Indeed, we have that
ck1
a,1
cka,1 = (1 p(ck1
a,0 )
p1
), (7.1)
p
and so
k p 1
ca,1 = p ck1 = = nk ,
a,1 p
p pnk+1 p
as desired. This completes the proof.
Remark 7.2 The coecient cka,1 is given by

k1
cka,1 = pnk (1 p( l (a))p1 ) ,
l=0
as follows by the recursion formula (7.1). Since the l (a)s are all
p-adic integers, by the non-Archimedean property of p applied to
this expression, it follows easily that the p-adic norm of this coecient
is 1/pnk .
Theorem 7.3 Any -function is an analytic function.
Proof. Let f (x) = F (x, x, . . . , r x) by an operator of order r given
by the restricted power series F Qp [[t0 , . . . , tr ]]. Thus,

f (x) = a x0 (x)1 ( r x)r ,
=(0 ,...,r )
where a 0 p-adically as || .
106 Analyticity of arithmetic dierential operators
1 n
The family of discs {a+pn Zp }pa=0 forms a covering of Zp . On each one
of these discs, we consider the power series fa (u) = f (a + pn u) Qp [[u]].
We use Lemma 7.1 to show that this power series converges on Zp for a
suitable choice of n.
Indeed, we have that
0 1 r

p

p

p
n
f (a+p u) = a ( c0a,j0 uj0 )0 ( c1a,j1 uj1 )1 ( cra,jr ujr )r .
=(0 ,...,r ) j0 =0 j1 =0 jr =0
p k
By the multinomial theorem, we expand each factor ( jk =0 cka,jk ujk )k
k pk k j
in this expression into a polynomial j=0 ca,j u in u of degree k pk .
Since each multinomial coecient is an integer, and these have p-adic
norms bounded above by 1, by the multiplicative and non-Archimedean
property of p , and referring to the estimates of Lemma 7.1, we con-
clude that
k 1
c .
a,j p (nk)j
p
It follows that
k
r k

r kp k=0k p
( cka,jk ujk ) = c j
a,j u ,
k=0 jk =0 j=0
where we may estimate the coecient c

a,j p-adically by
1
c .
a,j p
p(nr)j
Therefore, if we choose and x n to be any integer greater than the order
r of the operator f (x) = F (x, x, . . . , r x), the coecients c a,j will all
have uniformly bounded p-adic norm, and we have that
r k

k p
k=0
f (a + pn u) = a c
a,j u
j
=(0 ,...,r ) j=0
is a power series in u whose coecients go to zero p-adically as ||

because so do the a s. This nishes the proof.
8
Characteristic functions of discs in Zp : p-adic
coordinates
It is now natural to attempt to nd the analytic functions on Zp that

are -functions. The set of characteristic functions of discs in Zp is a
dense subset of C(Zp , Qp ) [36], so we investigate this problem for these
n
characteristic functions. Thus, given the covering {j + pn Zp }pj=01 of Zp
by discs of radius 1/pn , we ask if the characteristic function of any of
these discs can be realized as an arithmetic dierential operator. We
shall initially concentrate most of our attention in the case n = 1. The
explicit results we obtain in this case will guide the rest of our work.
Throughout this Chapter we use the coordinates of elements of Zp
that result when choosing the complete residue system {0, 1, . . . , p 1}
in realizing their expansions (2.4). These are referred to as the standard
p-adic coordinates.
8.1 Characteristic functions of discs of radii 1/p

We consider rst the case where p = 2. We ask if an operator of order
1 can be the characteristic function of a disc in the cover {j + 2Zp }1j=0
of Z2 . Of course, it suces to treat the question for the disc centered at
the origin, so we ask if we can nd coecients am,n so that the order 1
arithmetic operator

f (x) = am,n xm (x)n ,
m,n0
can be the characteristic function of the disc 2Z2 .

We use the changes x = 2u and x = 1 + 2u on the discs 2Z2 and
1 + 2Z2 , respectively. Then, for this to be the case, we must have the
107
108 Characteristic functions of discs in Zp : p-adic coordinates
relations

f0 (u) = a (2u)m (u 2u2 )n = 1 ,
m,n0 m,n m n 2 n (8.1)
f1 (u) = m,n0 am,n (1 + 2u) (1) (u + 2u ) = 0 ,
as series in u. And this system admits 2-adic integer solutions for the
amn s that go to zero as m + n .
Indeed, a simple commutation argument yields the explicit expressions

f0 (u) = a0,0 + k=1 c0,k uk , (8.2)
where

k kn n
c0,k = n=[ k + 2kn m=0 am,n (1)
kmn
2] kmn

[ k1
2 ] kn
kn kmn n
+ n=0 2 am,n (1) ,
m=k2n
kmn
while
k
f1 (u) = m=0 am,0 + k=1 c1,k u , (8.3)
where

k m+n
n kn
c1,k = n=0 am,n (1) 2 +
m=k
kn

k1 k n kn m+n
n=[ km + am,n (1) 2 ,
m=0 2 ] kn
respectively. Here, [ ] and [ ]+ are the greatest integer less or equal than
and the smallest integer greater or equal than functions, respectively.
Strassmans theorem [47] (see Theorem 4.9 in 4.1), (8.1) implies that
all but one of the coecients above vanish. We have:
Theorem 8.1 The characteristic function of 2Z2 is an arithmetic dif-

ferential operator of order 1. The system of equations (8.1) can be solved
for the coecients am,n , with am,n = 0 for all m 2, and
1
||am,n ||2 .
2n
Proof. We proceed by induction on n. Since am,n = 0 for m 2,
the two equations in (8.1) suce to solve for a0,n and a1,n , respectively.
Indeed, by (8.2), the coecient of u0 in f0 (u) is a0,0 , so a0,0 = 1. By

(8.3), the said coecient in f1 (u) is m=0 am,0 = a0,0 + a1,0 , and so
a1,0 = 1. Similarly, the coecient of u in f0 (u) is a0,1 + 2a1,0 , so a0,1 =

2, and the said coecient in the series f1 (u) is given by 2 m=1 mam,0

m=1 am,1 a0,1 = 2a1,0 a1,1 a0,1 = 0, and that yields a1,1 = 4.
Assume that all the am,n s have been chosen for all pairs m, n with
n < k, and satisfying the desired estimates for some k 1. Since the co-
ecient c0,k of f0 (u) must be zero, using its explicit form given by (8.2),
we solve the resulting equation for a0,k , and obtain a linear combination
of the previously found coecients. In this combination, am,n appears
multiplied by 2kn times some rational integer. Thus, a0,k is uniquely
determined, and it satises the desired estimate.
Having found a0,k , the coecient c1,k of uk in the series f1 (u) given
explicitly in (8.3), which must also be zero. allows us to solve for a1,k as
a linear combination of the previously found coecients, including a0,k .
In this combination, am,n appears again multiplied by 2kn times some
rational integer. Thus, a1,k is uniquely determined also, and it satises
the desired estimate. This nishes the proof.
We now generalize the result above to discs of radius 1/p for an arbi-
trary prime p. Our argument extends the one just given for the prime
2, realizing the said characteristic function as an operator of the form
p1
m=0 am,n x (w1 (x)+ +wp1+dp
m p1+dp
n0 (x))n , where the wj s
are suitably chosen coecients, and where dp is the degree of p, as de-
ned below. A dierent generalization will be given in Chapter 9, where
we will obtain a stronger result.
We begin our work here by associating to any prime p a canonical
set of p-adic numbers. In order to do so, we dene rst the associated
matrix A of p.
For p = 2, let A to be the number 1, a 1 1 matrix. For p 3 and
any integer l 0, we let vl be the vector

l (2)
l (3)

vl = .. , (8.4)
.
l (p 1)
and dene A to be the (p 2) (p 2) matrix whose columns are the
vectors v1 , . . . , vp2 :

(2) 2 (2) ... p2 (2)
(3) 2 (3) ... p2 (3)

A := .. .. .. .. . (8.5)
. . . .
(p 1) 2 (p 1) ... p2 (p 1)
We call A the matrix associated to p.
Denition 8.2 The Wronskian of the prime p is the determinant det A

of its associated matrix A. The degree dp of p is the p-adic order of its
Wronskian. We say that p is a singular prime if its degree dp is positive.
Otherwise, p is said to be nonsingular.
Example 8.3 Generally speaking, it is a rather dicult task to com-

pute the Wronskian of a prime p. For p = 2 it is 1. For p = 3, it is 2,
and for p = 5, it is the rational integer
223 319 77 11 134 17 29 61 89 15951542086918898827556696509 .
Thus, the rst 3 primes are all nonsingular.

In fact, among the rst 100 primes, only p = 29 and p = 311 are
singular. This assertion can be proven without having to nd explicitly
their Wronskian, merely proving that this Wronskian is divisible by p.
Even then, the quick growth of the integers involved in the calculations
makes the task computationally complex when the prime in question is
moderately large.
Let p be a prime of order dp . For any l in the range from 1 to p 2,

we dene Al to be the matrix obtained from the matrix A in (8.5)
associated to p by replacing its l-th column by the vector vp1+dp in
(8.4). The following conclusion is recorded for use later on. It follows by
Cramers rule.
Lemma 8.4 The vector

1
( det A1 , . . . , det Ap2 )
det A
is the unique solution of the equation Ax = vp1+dp .
Denition 8.5 The ordered (p 1 + dp )-tuple

1
( det A1 , . . . , det Ap2 , 0, . . . , 0, det A)
det A
is the set of Wronskian quotients associated to p.
Example 8.6 If p is nonsingular, the Wronskian quotients associated

to p are all p-adic integers. And once again, these numbers are dicult
to compute also. For p = 3, they are the integers (1, 1). For p = 5, we
use the value of the Wronskian given in Example 8.3, and obtain that
the third Wronskian quotient associated to the prime 5 is given by
det A3 3 13 31 37 251 9324919439243 q
= ,
det A 15951542086918898827556696509
where q is a prime whose decimal expansion contains 222 digits .
Remark 8.7 The naive thought that the set of of Wronskian quotients
associated to a prime p should all be p-adic integers fails for p = 29, the
rst of the singular primes.
Lemma 8.8 Let dp be the degree of the prime p, and let the (p1+dp )-
tuple (w1 , . . . , wp1+dp ) be the set of Wronskian quotients associated to
it. Then the -function
h(x) = w1 (x) + + wp1+dp p1+dp (x)
vanishes at x = a, 0 a p 1, and if
pp1+d

p
h(a + pu) = wa,j uj , 0 a p 1,

n=1

j
then ||wa,1 ||p = pp2+dp , and (wa,j /wa,1 ) 1/pj1 for any j.
p
Proof. For a vector w = (w1 , . . . , wp1+dp ) of constants yet to be

determined, we consider the function
h(x) = w1 (x) + + wp1+dp p1+dp (x) .
Since regardless of what the positive integer k may be both 0 and 1 are in
the kernel of k (x), the set of conditions h(a+pu)|u=0 = 0, 0 a p1
yields the system of equations

(2) 2 (2) ... p1+dp (2) w1
(3)
2 (3) ... p1+dp (3)

w2

.. .. .. . . = 0.
. . . .. ..
(p 1) 2 (p 1) ... p1+dp (p 1) wp1+dp
Notice that the matrix of this system is simply the augmented matrix
(A | vp1 . . . vp1+dp ). Since its row echelon form is given by the matrix
q = 4694858077226952009296804634445529738803634927448956662114977029779
96111969322231331735328135927753617578318483953518947685436992331399697
08914287050991506040052273191095308486354235128103259585157551339337818
2840897402417.
(1l | A1 vp1 . . . A1 vp1+dp ), we know that the solution space of the

system above has (dp + 1)-degrees of freedom. By Lemma 8.4, the set
of Wronskian quotients (w1 , . . . , wp1+dp ) associated to p is the unique
solution of this system whose last (dp + 1)-components are (0, . . . , 0, 1).
We choose and x this particular solution hereafter.
We pick a rational integer a such that 0 a p 1. Following the
notation in Lemma 7.1, we let cka,j denote the coecient of uj in the poly-
k1
nomial k (a + pu). By Remark 7.2, cka,1 = p1k l=0 (1 p( l (a))p1 ),
and if wa,j denotes the coecient of uj in h(a + pu), we have that

p1+dp
wk
k1
wa,1 = (1 p( l (a))p1 ) .
pk1
k=1 l=0
Since wp1+dp = 1, by the non-Archimedean property of p , we see

that ||wa,1 ||p pp2+dp . In fact, we have equality. For otherwise, the
p2+dp
expression above for wa,1 multiplied
by p would lead to the con-
tradictory statement that wp1+dp p < 1. This uses the fact that if
dp > 0, the k-th Wronskian quotient associated to p is zero for ks such
that p 1 k p 2 + dp , and that all of these
numbers
times pdp are

p1+d
p-adic integers. Thus, ||wa,1 ||p = pp2+dp = ca,1 p .
p
In general, we may write the wa,j in terms of the cka,j s for arbitrary
values of j as
p1+dp
wa,j = w1 c1a,j + + wp1+dp ca,j .
If we now use Lemma 7.1 with n = 1 and the non-Archimedean property

of p , we obtain that

w
a,j 1 ck
j = p(p2+dp )j wa,j max
wa,1 p p(p2+dp )j 1kp1+dp a,j p
p
1
.
pj1
This completes the proof.
Remark 8.9 Given the Wronskian quotients w1 , . . . , wp1+dp asso-

ciated to p, and the coecients wa,j of uj in the polynomial function
h(a + pu) above, the scaling indicated in this result seems to be optimal
if we are to make the most general statement. For instance, we could
ask instead if the quotients wa,j /wa,1 are p-adic integers. This is true
for the primes p = 2 and p = 3. It is false already for p = 5.
Remark 8.10 Notice that if p is a nonsingular prime, the arithmetic

operator h(x) in Lemma 8.8 is given by a polynomial in Zp [x0 , . . . , xp1 ]
(see Denition 5.3), and has order p 1. Else, it is given by a polynomial
in Qp [x0 , . . . , xp1+dp ]. In both cases, the polynomials are independent
of x0 .
Theorem 8.11 Let dp be the degree of p. Then, given a disc in Zp of

radius 1/p, there exists a formal power series of the form

p1
am,n xm (w1 (x) + + wp1+dp p1+dp (x))n ,
n0 m=0
such that am,n 0 p-adically, which converges pointwise to its charac-

teristic function. Here, (w1 , . . . , wp1+dp ) are the Wronskian quotients
associated to p.
In particular, when all of the Wronskian quotients of p are p-adic

integers (if at all), we may conclude that the characteristic function of a
disc of radius 1/p is an arithmetic dierential operator of order p1+dp
of the form indicated in the Theorem. Thus, for nonsingular primes p
the said characteristic functions are arithmetic dierential operators of
order p 1. For other primes, the assertion is only valid at the level of
formal power series that converge pointwise.
Proof. It suces to show that the characteristic function of pZp is a

formal operator of order p1+dp of the type indicated in the statement.
The argument that we provide works for any prime p, and generalizes the
one we have seen already in Theorem 8.1 for the case of the nonsingular
prime p = 2.
Let w1 , . . . , wp1+dp be the set of Wronskian quotients associated to
the prime p, and consider the (p 1 + dp )-th arithmetic operator
h(x) = w1 (x) + + wp1+dp p1+dp (x)
of Lemma 8.8. Then we know that for any a such that 0 a < p, we
have that
pp1+dp

h(a + pu) = wa,j uj ,
j=1

j
where ||wa,1 ||p = pp2+dp and wa,j /wa,1 1/pj1 . We proceed by
p
induction to determine the coecients am,n in the operator

p1

f (x) = am,n xm (h(x))n
n0 m=0
so that, if fa (u) = f (a + pu), we have that

f0 (u) = 1,

f1 (u) = 0,
.. (8.6)

.

fp1 (u) = 0 ,
and
1
||am,n ||p , n = 0, 1, 2, . . . . (8.7)
pn
For the starting point of the induction, we choose the coecients am,0 ,
0 m < p, such that

p1
F0 (x) = am,0 xm
m=0
solves (8.6) to order zero. This implies that a0,0 is 1, and that the equa-
tion

1 1 ... 1 a1,0 1
2 22 ... 2p1 a2,0 1

. .. .. .. .. = .
.. . . . . .
.
p 1 (p 1) . . . (p 1)
2 p1
ap1,0 1
holds. This Vandermonde system can be readily solved for the am,0 s,
with the solution being a vector of p-adic integers, and the estimates
(8.7) holding for n = 0.
For the k-th step of the induction, we assume that we have found

k1 p1
Fk1 (x) = am,n xm (h(x))n
n=0 m=0
such that Fak1 (u) = Fk1 (a + pu) solves system (8.6) to order k 1,
with the required estimates for the coecients up to that point. Let us
consider
p1
k1
p1
m n
Fk (x) = am,n x (h(x)) + am,k xm (h(x))k .
n=0 m=0 m=0
We determine the coecients am,k so that Fak (u) = Fk (a + pu) solves

(8.6) to order k.
The coecient of uk in Fak (u) is determined by the am,n s with n k.
In fact, by the multinomial expansion, this coecient is given by
pp1+dp

p1
p1
k1 l
k m
wa,1 am,k a + am,n cn,,m,r,a,p wa,l ,
m=0 n=0 m=0 r, l=1
where

n m
cn,,m,r,a,p = amr pr
1 , . . . , pp1+dp r
and where the last sum in the second term above is over all nonnegative
multi-indices of weight n and indices r in the range 0 r m, such
that 1 + 22 + + pp1+dp pp1+dp + r = k. Hence, the Fak (u) solves
(8.6) to order k if we have
pp1+dp l

p1
p1
k1 l=1 wa,l
am,k a =
m
am,n cn,,m,r,a,p k
(8.8)
m=0 n=0 m=0 r,
wa,1
for all as in the range 0 a p 1.

Notice that
pp1+dp l pp1+dp
l
l=1 wa,l 1 wa,l
k
= r l
.
wa,1 wa,1 wa,1
l=2
By Lemma 8.8, it follows that the p-adic norm of this number is bounded
1
by pkn+(p1+d p )r
. Now by the induction hypothesis on the p-adic norm
of the am,n s, we conclude that the right side of (8.8) has p-adic norm no
larger than 1/pk .
The equation for a0,k is uncoupled from the remaining am,k s. Indeed,
when a = 0 the only nonzero term in the left side of (8.8) is a0,k . Thus,
we have a0,k = g0k , where g0k is a p-adic number with p-adic norm no
larger than 1/pk . For the other coecients, (8.8) yields the system
k
1 1 ... 1 a1,k g1
2 2 2
. . . 2p1 a2,k g2k

. .. .. .. .. = .
.. . . . . .
.
p1 (p 1)2 ... (p 1)p1 ap1,k k
gp1
where the right side is a vector of p-adic numbers whose components all
have p-adic norm no larger than 1/pk . This system can be solved with
the solution having p-adic norm equal to the p-adic norm of the right
side. This completes the last step in the induction, and the proof.
8.2 Characteristic functions of discs of radii 1/pn

The argument given above can be generalized in a rather naive manner
for the study of the same problem for discs of radii 1/pn as well. How-
ever, we should expect now a singular behaviour for n > 1. Indeed, if we
think of p-adic numbers as analytic functions, as we have been doing so
far, larger powers of p should exhibit a behaviour analogous to branching
points. Our naive generalization shows that, and we obtain formal power
series in Qp [[x0 , . . . , xn ]] which converge pointwise to the characteristic
function of a disc of radius 1/pn , but the series are not necessarily re-
stricted. As in the next Chapter we shall improve substantially upon our
current result to this point, we merely outline the generalized argument,
leaving several of the details to the interested reader.
We associate with pn the (pn 2) (pn 2) matrix given by

p 2 (2)
n
(2) 2 (2) ...
(3) 2 (3) ... p 2 (3)
n

Apn := .. .. .. . . (8.9)
. . . ..
n
2
(pn 1) 2 (pn 1) . . . p (pn 1)
Denition 8.12 We say that det Apn is the Wronskian of pn , and call
its p-adic order the degree of pn , which we denote by dpn .
The integer dpn > 0 measures the singularity of pn . By the compu-

tational complexity of the problem, it is a dicult task to nd ps such
that dpn > 0 for some n > 1.
We can dene the ordered (pn 1 + dpn )-tuple
1
( det A1 , . . . , det Apn 2 , 0, . . . , 0, det Apn )
det Apn
as the Wronskian quotients associated to pn . Here, Aj is the matrix
obtained from Apn by replacing its j-th column by the vector whose k
component is p 1+dpn (2 + (k 1)), 1 k pn 2. We then have:
n
Lemma 8.13 Let dpn be the degree of pn , and let (w1 , . . . , wpn 1+dpn )
be the set of Wronskian quotients associated to it. Then the -function
n
1+dpn
hn (x) = w1 (x) + + wpn 1+dpn p (x)
8.2 Characteristic functions of discs of radii 1/pn 117
vanishes at x = a, 0 a pn 1, and if
pn 1+dpn
p

h(a + pu) = wa,j uj , 0 a pn 1 ,
n=1

n
2+dpn j
then ||wa,1 ||p = pp , and (wa,j /wa,1 ) 1/pj1 for any j.
p
We now consider n-tuples = (0 , . . . , n1 ) in Zn such that 0

0 , . . . , n1 p 1. This set contains pn elements. We provide it with
the lexicographical order.
Lemma 8.14 There are polynomials p0 (x), . . . , pn1 (x) in Z[x] such
that the pn pn matrix V = (vij ) whose entries are given by
vij = (p0 (i))0 (p1 ((i)))1 (pn1 ( n1 (i)))n1 , 0 i pn 1 ,
where (0 , . . . , n1 ) is the j-th element of the set { = (0 , . . . , n1 )
Zn : 0 0 , . . . , n1 p 1} in the lexicographical order, is invertible
over Zp , that is to say, its determinant is an invertible element of Zp .
We skip the details of the proof. A more elaborate argument will be
given in Lemma 9.4 below, where a closely related result is proven using
the constant coordinates of the arithmetic dierential theory. In that
argument, we can see that the statement above can be proven by taking
pj (x) = x for all j.
Based on this result, the generalized version of Theorem 8.11 for discs
of arbitrary radii reads as follows.
Theorem 8.15 Let p be a prime and n N. Assume that dpn is
the degree of pn , and consider the disc pn Zp in Zp of radius 1/pn . Let
hn (x) and p0 (x), . . . , pn1 (x) be the arithmetic dierential operator and
polynomials of Lemmas 8.13 and 8.14 above, respectively. Then, there
exists a restricted power series of the form

F (x0 , . . . , xn ) = a,n (p0 (x0 ))0 (pn1 (xn1 ))n1 (xn )m
m0 0j <p
in Zp [[x0 , . . . , xn ]] such that F (x, (x), . . . , n1 (x), hn (x)) converges to

the characteristic function of the said disc pointwise.
The argument parallels the one for n = 1 that proves Theorem 8.11.
In the earlier case, the family of polynomials given by Lemma 8.14 con-
sists of the single polynomial p0 (x) = x. In the general argument, the
pn pn -matrix V given in Lemma 8.14 now plays the role of that of the
Vandermonde matrix in the proof of Theorem 8.11.
Remark 8.16 The the representation of the characteristic function of

any disc in Zp as an arithmetic dierential operator is not unique.
9
Characteristic functions of discs in Zp:
harmonic arithmetic coordinates
Theorems 8.11 and 8.15 in the previous Chapter do not provide any
indication of the optimal order required to realize characteristic func-
tions of discs as arithmetic dierential operators of the said order. This
indication lacks even for nonsingular primes, where the order given by
Theorem 8.11 is the lowest. In this Chapter we prove a result that is
somewhat of a surprise, addressing this issue. It ties up the order of the
operator that realizes the characteristic function of the disc with its level
of analyticity (see Denition 4.4), and in that sense, it strengthens what
Theorem 7.3 says about them.
No characteristic function of a disc of radius 1/p can be an operator
of order zero, so both Theorems 8.1 & 8.11 yield the optimal result for
the prime 2. However, what about the remaining primes?
Remark 9.1 No nontrivial arithmetic dierential operator
f (x) = F (x, x, . . . , r x)
with F (t0 , . . . , tr ) Zp [t0 , . . . , tr ] can equal a nontrivial locally constant

function, no matter the order. For such a function would be a polynomial
in x, of a certain degree, that cannot be zero. Since the degree k in u of
fa,n (u) = f (a + pn u) must be the same as the degree of f (x) in x, and
since for suciently large n, this function would have to be constant, by
taking k analytic derivatives of fa,n , and setting u = 0, we would see
that k = 0. So F (t0 , . . . , tr ) must be the constant polynomial
In the work carried out in the previous Chapter, we used the coordi-
nate representation of the p-adic number that arise by considering the
complete residue system {0, 1, . . . , p 1} as the coecients in the ex-
pansion (2.4). In the context of arithmetic dierential operators, we will
119
120 Characteristic functions of discs in Zp : harmonic coordinates
now show that it is far more natural to consider the coordinates that
arise from the arithmetic constants, the solutions of (a) = 0. These
are the (p 1)-roots of unity and zero, and they lead to the Teichm
uller
representation of Example 4.8. In a sense, these are analogous to the
harmonic coordinates on a Riemannian manifold, although the analogy
is quite incomplete in that we do not have a notion of Riemannian met-
ric on Zp . However, the judicious use of these coordinates lead to the
proof of the following result [14]:
Theorem 9.2 Any analytic function f : Zp Zp of level m is an

arithmetic dierential operator of order m.
As it will turn out, the characteristic functions of discs in Zp of radii

1/pm , which are analytic functions of level m, can be realized as arith-
metic operators of the form F (x, x, . . . , m x), where F (x0 , . . . , xm ) is a
restricted power series in Zp [[x0 , . . . , xm ]]. This is a vast improvement on
the representation that, for instance, Theorem 8.11 yields for arbitrary
primes, even the nonsingular ones (cf. with the remark that follows the
statement of the theorem, prior to its proof).
We have already mentioned that representation of a characteristic
function of a disc in Zp as an arithmetic dierential operator is not
unique. Of all of these representations, those given by restricted power
series with coecients in Zp itself are to be preferred. They are the most
natural examples from the point of view of number theoretic considera-
tions.
9.1 A matrix associated to pm

Let us consider the set of all p-adic integer roots of the function x m x:
Cm := {a Zp : m a = 0} Zp .
Since the m-th iterate of the operator is given by a polynomial in

its argument of degree pm with Qp -coecients, Cm can have at most
pm elements. In fact, its cardinality is pm as the number of solutions of
m a = 0 is exactly pm . For we have.
Lemma 9.3 The composition
Cm Zp Zp /pm Zp
is bijective.
9.1 A matrix associated to pm 121
Proof. We prove the assertion by induction on m. For m = 0, we

have that C0 = {0}, and the result is clear. We assume now that the
statement is true for m 1 > 0, and prove it for m.
Given a Cm1 , we consider the polynomial tp t + pa Zp [t]. By
Hensels lemma, it has p distinct roots that we denote by a1 , . . . , ap Zp .
Notice that we have aj = a for all js, and since m1 a = 0, it follows
that m aj = 0.
Observe that if a, a Cm1 and we have that
aj aj mod pm (9.1)
for some j, j , then a = a and j = j . Indeed, if (9.1) holds then a a

mod pm1 , and by the induction hypothesis, a = a , and hence j = j
as well.
We thus have that each of the pm1 element of Cm1 yields p distinct
elements of Cm , so Cm contains a set of pm elements, and by the ob-
servation above, this set injects into Zp /pm Zp . As Cm has at most pm
elements, this forces the map Cm Zp /pm Zp to be bijective.
We use Lemma 9.3 to write
Cm = {a0 , a1 , . . . , apm 1 } , (9.2)
where a mod pm for all s in the set
I = {0, . . . , pm 1} . (9.3)
On the other hand, we x an ordering of the set
I = { = (0 , . . . , m1 ) Zm : 0 0 , . . . , m1 p 1} , (9.4)
and consider the pm pm -matrix
W = (w )I,I , (9.5)
whose entries are given by
w := (a )0 (a )1 ( m1 a )m1 Zp ,
where we have used the convention that a0 = 1 for all a Zp . Up to a

permutation of its columns, the matrix W is intrinsically associated to
the number pm .
Lemma 9.4 The determinant of the matrix W in (9.5) is invertible

in Zp .
Proof. We use the reduction mod p mapping

Zp Fp := Zp /pZp
a a.
Now let us notice that if a a mod pm then a a mod pm1 . For
we have that a = a + pm u with u Zp . So
a (a )p a + pj u (a + pj u)p
a = =
p p
a ap
= + pm1 u = a + pm1 u ,
p
for some u Zp , and so a a mod pm1 . We may iterate this
argument to conclude that i a i a mod pmi for any i m, and so
the function
Zp Fm p
a (a, a, . . . , m1 a)
induces a bijection between Zp /pm Zp and Fm p . Now for any element

= (0 , . . . , m1 ) in Fm
p and any I , we set

v = 00 11 m1
m1
.
Notice that for any a Cm we have that i a i mod p if i m1.
Therefore, the desired result here will follow by Lemma 9.3 if we merely
show that det (v ) = 0 Fp .
Let us assume that det (v ) = 0. This means that there exist con-
stants 0 ...m1 Fp for (0 , . . . , m1 ) I , not all zero, such that

p1

p1

... 0 ...m1 00 11 . . . m1
m1
=0
0 =0 m1 =0
for all Fm
p . We may then proceed by induction on m to prove that
these relations imply that that all the s vanish, which is a contradiction.
This nishes the proof.
9.2 Analytic functions and arithmetic dierential

operators
We now carry out the proof of Theorem 9.2 by proving the following
stronger result. In what follows, I and I are the sets of indices (9.3) and
(9.4), respectively.
9.2 Analytic functions and arithmetic dierential operators 123
Theorem 9.5 Let f : Zp Zp be an analytic function of level m. Then

there exists a unique restricted power series F Zp [[x0 , x1 , . . . , xm ]] with
the following properties:
m1 n
1. F (x0 , x1 , . . . , xm ) = n0 I a,n x0 0 x1 1 xm1 xm .
2. f (a) = F (a, a, . . . , a), a Zp .
m
Proof. We start by proving the existence of the power series F . Notice

that if
g : Zp Zp
is any arithmetic dierential operator of order m, and a Zp , then
h(x) := g(x + a)
is also an arithmetic dierential operator of order m; cf. Lemma 5.4.
Thus, without losing generality, we may assume that the function f in
the statement of the Theorem is zero on all discs of radius 1/pm except
for pm Zp . Without losing generality also, we may additionally assume
that there exists an l 0 such that f (pm u) = ul for all u Zp .
We recall the set Cm = {a0 , . . . , apm 1 } in (9.2). The family of discs
{a+pm Zp }aCm forms a covering of Zp . Let us notice that Cm a0 = 0,
and this is the center of the one disc where f is nonzero.
By Lemma 7.1, for a Cm and 0 k m, we have that k (a+pm u) =
pm k j k k
j=0 ca,j u , with ca,0 = a, and
k k 1 k 1
ca,0 1 , ca,1 = , ca,j , 2 j pk .
p p pmk p p(mk+1)j1
(9.6)
We may view (a+p u) as an element in the ring of polynomials Zp [u].
k m
Since m a = 0, we have that cm a,0 = 0.

We proceed to determine by induction the coecients a,n in the series
F appearing in the statement so that
"
1
||a,n ||p min 1, nl , n 0 , I , (9.7)
p
and so that, if Fa (u) = F (a + pm u) for a Cm , then we have that

Fa (u) = ul if a = 0 ,
(9.8)
Fa (u) = 0 if a = 0.
Here we view (9.8) as equalities of functions of u Zp . However, since
each Fa (u) is dened by a restricted power series in Zp [[u]], it is enough
to check (9.8) as equalities in the ring of formal power series Zp [[u]].
We consider the polynomials Fak (u) Zp [u] dened by

0 0 m1 m1 m n

k
p p
p
Fak (u) := a,n c0a,j uj . . . cm1
a,j u
j cm
a,j u
j
n=0 I j=0 j=0 j=1
so that Fak (u) converge u-adically to Fa in Zp [[u]]. We nd the a,n s

inductively so they satisfy the estimate (9.7), and such that the following
congruences hold in the ring Zp [u]:
k
Fa (u) ul mod uk+1 if a = 0 ,
(9.9)
Fak (u) 0 mod uk+1 if a = 0.
In what follows we denote by ij the Kronecker symbol.
For the starting point of the induction, we choose the coecients a,0 ,
I , such that (9.9) and (9.7) hold. This can be achieved by solving
the system of equations

w a,0 = l0 0 , I ,
I
where W = (w ) is the matrix (9.5). By Lemma 9.4, this system can

be readily solved for the a,0 s, with the solution being a vector of p-adic
integers.
For the k-th step of the induction, let us notice that for a = a , the
coecient of uk in Fak (u) is given by

k1
(cm
a,1 )
k
w a,k + a,n b,n,k , (9.10)
I n=0 I
where b,n,k is the coecient of uk in

0 0 m1 m1 m n
p p
p
c0a,j uj . . . cm1
a,j u
j cm
a,j u
j
.
j=0 j=0 j=1
Thus, b,n,k is a Z-linear combination of products of the form

m1
0

n
ca,j0i . . .
0
ca,jm1,i
m1 m
ca,jmi ,
i=1 i=1 i=1
with

m1 r
n
jri + jmi = k .
r=0 i=1 i=1
9.2 Analytic functions and arithmetic dierential operators 125
We may assume that there are integers sr such that jri 1 for i sr
and jri = 0 for i > sr . So we have

sr
m1 sr
n
sr jri , jri + jmi = k . (9.11)
i=1 r=0 i=1 i=1
By (9.6) and the induction hypothesis,

"
1
||a,n b,n,k ||p min 1, . (9.12)
pnl+
where

m1 sr
n
= [(m r + 1)( jri ) sr ] + jmi n .
r=0 i=1 i=1
Now, by (9.11) we have that

m1 sr n
r=0 [2( i=1 jri ) sr ] + i=1 jmi n
m1 sr n
r=0 i=1 jri + i=1 jmi n
= k n.
Hence "
1
||a,n b,n,k ||p min 1, . (9.13)
pkl
Now, by the induction hypothesis also, we can ensure that Fak (u) satises
(9.9) if we have that

k1
k
w a,k = (cm
a ,1 ) kl 0 a,n b,n,k , I.
I n=0 I
(9.14)
By (9.6) and (9.13), the p-adic norm of the right hand side of (9.14) is
bounded above by min{1, 1/pkl }. Again, by Lemma 9.5, we can solve
the system (9.14) for the a,k s, with the solution satisfying the estimates
(9.7). This completes the induction, and hence the existence part of the
Theorem.
In order to prove the uniqueness, we need to show that if a restricted
power series F satises conditions (1) and (2) in the Theorem for f = 0,
then a,n = 0 for all I , n 0. This can derived by an induction on
n, in view of the equalities

k1
k
w a,k = (cm a ,1 ) a,n b,n,k , I.
I n=0 I

If n = 0, this expression says that I w a,0 = 0, I, and since
the matrix (w ) is invertible, we have that a,0 = 0 for all I . For

larger values of n, the expression then says that I w a,k = 0 also
for I, since according to it, these values are determined by the ak s
with 0 k < k, that are zero. So a,k = 0 for all I . This completes
the proof.
10
Some dierences between arithmetic
ur
dierential operators over Zp and Z p
In this Chapter we explore some dierences between arithmetic dier-

ential operators over the ring Z ur and arithmetic dierential operators
p
over the coarser ring Zp . In particular, we discuss the naive analogue of
Theorem 9.2 over Z ur , which we show to be false.
p
Indeed, in the context of the general theory of [6, 8] discussed in 6,
where the role of the ring Zp is played by Z ur and the operator is
p
associated to the lift of Frobenius (6.2) of Theorem 6.9, we now have
the following:
Theorem 10.1 Let

0 xr
a x 0 r
F (x) =
ur [[x]]. Assume that the mapping

be a restricted power series in Z p
x F (x, x, . . . , r x)
ur . Then F (x) itself is a constant
is constant on a disk of some radius in Z p
in Zp .
ur
The proof of this result requires to use the fact that Fp = Z ur /pZ
ur is
p p
algebraically closed. This explains, in part, our earlier assertion in Chap-
ter 6 that the dierences between the theories of arithmetic dierential
operators over the rings Zp and Z ur are analogous to the dierences
p
between number theoretic statements about nite elds and algebraic
geometric statements over their algebraic closures.
We begin the proof of Theorem 10.1 by recalling a key result in [11].
For convenience, we denote by R the ground ring, as before. If X is a
smooth scheme over R of nite type, and {Ui } is a covering of X by
ane open sets, we recall that the p-adic completions of the schemes
127
128 ur
Some dierences between -operators over Zp and Z p
Spec O(J n (Ui )) glue together to produce the formal scheme Jpn (X),
the p-jet space of X of order n. If we glue the p-adic completions of
Spec O(J (Ui )) instead, we obtain the innite p-jet space of X. Since
the sheaf of rings O(Jpn (X)) are topologically generated by O(Jpn1 (X))
and O(Jpn1 (X)), we obtain a projective system of formal schemes (6.5),

where Jp0 (X) = X (p) is the p-adic completion of X, and Jp (X) is the
p-adic completion of the inverse limit of the Jpn (X)s. By taking the re-
duction mod p, we obtain a projective system of k = R/pR-schemes

(p)
J0r (X) J0r1 (X) J01 (X) J00 (X) = X0 .
By the universality property of the p-jets, there exists a natural lifting

mapping
: X(R) Jp (X)(R) .
In the case where X is the ane line over R, this mapping is just

R RN
x x = (x, x, 2 x, . . .) .
By composition with the reduction mod p mapping, we obtain the map-

ping
0
R kN
x 0 x = (x, x, 2 x, . . .) ,
where we denote by x the reduction x mod p of an element x in R.

We shall need the following only in a particular case, but state it in
general. A Witt vector over a commutative ring A is a sequence x =
(x0 , x1 , x2 , . . .) of elements of A. The Witt polynomials are dened by
n ni
Wn (x) = i=0 pi xpi , n = 0, 1, . . .. Then there exists a unique ring
structure on the set of Witt vectors over A such that:
1. The addition and product operations are given by universal polyno-

mials with integral coecients, and
2. Every Witt polynomial is a homomorphism from the ring of Witt
vectors over A into A.
For instance, we have that

xp +y p (x +y )p
(x0 , x1 , . . .) + (y0 , y1 , . . .) = (x0 + y0 , x1 + y1 + 0 0 p 0 0 , . . .) ,
(x0 , x1 , . . .) (y0 , y1 , . . .) = (x0 y0 , xp0 y1 + y0p x1 + px1 y1 , . . .) .
ur
Some dierences between -operators over Zp and Z 129
p
Remark 10.2 A p-derivation p over a ring A is a theory of length two

of the ring of Witt vectors of A, that is to say, a theory that involves
the ring W2 (A) of the rst two components of a Witt vector of A in
a manner compatible with the ring structure they inherit from that of
the Witt vectors. Or said dierently, if A is a ring and B is an algebra
over A, a p-derivation p : A B associated to the homomorphism
p : A B is such that the map A a (a, p (a)) W2 (B) is a ring
homomorphism, where in the right side, the rst component a stands
for its image in the B algebra over A.
Example 10.3 1. The Witt ring of any commutative ring A in which
p is invertible is isomorphic to AN . For the Witt polynomials produce
a homomorphism from the ring of Witt vectors to AN , and if p is
invertible, the said homomorphism is an isomorphism.
2. Consider the eld Fp = Zp /pZp . Then its Witt ring is Zp . On the other
hand, we have that the algebraic closure Fp is equal to Fp = Zur ur
p /pZp ,
and every x Fp has a unique Teichm uller representative x Zp (see
ur
Example 4.8), which is a root of unity and projects to x in Zur ur

p /pZp .
It follows that the Witt ring of Fp is Z ur .
p
Assume now that R is a commutative local ring with maximal ideal

pR such that the residue eld k = R/pR is of characteristic p. Of course,
this will be the case if R = Z ur . Then the ring W (k) of Witt vectors
p
N
of k is k . If the ring R is Hausdor and complete for the topology of
ideals dened by pR p2 R , and if k = R/pR is a perfect ring of
characteristic p, there exists one and only one multiplicative system of
representatives : k R [44]. These are multiplicative mappings such
that = 1l, and such that

kN R

a = (0 , 1 , . . .) (a) = j=0 (ajpj )pj
is a ring isomorphism.
Lemma 10.4 (Lemma 2.6, [11]) The composition mapping

kN R 0 kN

is a bijection. In fact, there exist integer coecients universal polynomi-

als Pn in n-variables, n 2, such that
0 (0 , . . . , n , . . .)=(0 , 1 , 2 +P2 (0 , 1 ), . . . , n +Pn (0 , . . . , n1 ), . . .) .
We now have the following.
130 Some dierences between -operators over Zp and Z ur
p

Lemma 10.5 Let F (x) = a x0 x
0
r
r
R[[x]] be a restricted
power series such that F (x, . . . , x) = 0 for all x R. Let F denote the
r
reduction mod p of F . Then F = 0 in the polynomial ring k[w0 , . . . , wr ].

Proof. Since F (x, . . . , r x) = 0 for all x R, we have that
F (x, x, . . . , r x) = 0
for all x R. By Lemma 10.4 above, any vector w = (w0 , w1 , . . . , wr )
k r+1 is the projection onto the rst (r + 1)-components of a vector
0 ((a)) = (a, a, . . . , r a, . . .). Thus, F (x) = F (w0 , w1 , . . . , wr ) = 0 for
any w k r+1 , as desired.
Notice that both Lemmas 10.4 and 10.5 hold in general, without re-
quiring the quotient eld k = R/pR to be algebraically closed. This
hypothesis is needed in order to complete the proof of Theorem 10.1, as
we now see.
Proof of Theorem 10.1. By subtracting the constant, we can assume
that F (x, x, . . . , r x) = 0, and so in order to prove the result, it will
suce to prove that p|F , for by iteration we will then able to conclude
that F must be identically zero. By Lemma 10.5, we have that
F (w0 , . . . , wr ) = 0
for any (w0 , . . . , wr ) k r+1 . Thus, F is in the zero ideal of the poly-
nomial ring k[w]. Since the residue eld k = Z ur /pZ
ur is algebraically
p p
closed, we may apply Hilberts Nullstellensatz to conclude that F = 0.
This nishes the proof.
The most general assertion that can be made along these lines is the
following:
Theorem 10.6 Let X is a smooth scheme over R = Z ur . If f is an
p
element in the ring O(Jp (X)) of global functions on the p-jet space of
r
order r such that the induced map X(R) R is 0, then f itself is 0.
The proof of this generalized version can be given in a manner parallel

to the one above. We merely need to replace the use of Lemma 10.4 as
stated here by its most general version in [11].
Remark 10.7 We may elaborate further on our results over the ring
Zp , and compare them to Mahlers theorem 3.2 about the structure of
continuous Zp -valued functions on Zp .
Let Z0 be the set of all nonnegative integral vectors = (0 , 1 , 2 , . . .)
ur
p
with nite support, that is to say, vectors = (0 , 1 , 2 , . . .) in Z

such that j 0 for all j, and j = 0 for j suciently large. If Z 0 ,

the weight || = i0 i is well-dened. Given a sequence of variables
x0 , x1 , x2 , . . . and an Z 0 1 2
0 , we set x for x0 x1 x2 . . .. In this case,
we say that a power series

F (x0 , x1 , x2 , . . .) = a x , a Zp ,
Z
0
is restricted if lim|| a = 0.
Mahlers theorem 3.2 is equivalent to the following, as proven, for
instance, in [18]:
Theorem 10.8 Let f : Zp Zp be a continuous function. Then

there exists a restricted power series F (x0 , x1 , x2 , . . .) in the variables
x0 , x1 , x2 , . . ., with Zp -coecients, such that
f (a) = F (a, a, 2 a, . . .)
for all a in Zp .
Our Theorems 7.3 and 9.2 imply that the series F in Theorem 10.8
can be chosen to depend on nitely many variables if, and only if, f is
analytic.
If in Denition 4.4 we replace Zp by Z ur

p , we would then obtain the
notion of an analytic function of level m over Z ur
p .
Theorem 10.9 Let

0 xr , r 1 ,
a x 0 r
F (x) =
ur [[x]] such that the mapping

be a restricted power series in Z p
x F (x, x, . . . , r x)
is analytic. Then F (x) must be a constant function. On the other hand, if

ur , it cannot be an arithmetic
F (x) is an analytic function of level m on Z p
dierential operator unless m = 0.
ur , an arithmetic dierential operator that is not of order
Thus, over Z p
zero is never analytic, and an analytic function that is not of level zero
is never an arithmetic dierential operator.
Proof. The rst statement amounts to the uniqueness of the repre-
sentation of an analytic function on Z ur , uniqueness that was proven
p
132 ur
combining Lemma 10.5 and the Nullstellensatz in the proof of Theorem

10.1, cf. with Remark 8.16. The last assertion then follows.
10.10 In Example 5.5 we exhibited the Legendre symbol

Remark
a ap over Z p as a dierential operator of order 1. The right hand
ur provided that a is now dened by
side of (5.7) makes sense for a Z p
(6.1), where is the lift of Frobenius. The extended function is no longer
locally constant, and, of course, it cannot be equal to the characteristic
function of any disc.
Example 10.11 Example 5.5 has an analogue for ane elliptic curves.
We discuss it briey.
Let X be the locus of v 2 = u3 + au + b in the ane plane Spec Zp [u, v]
over Zp , where 4v 3 + 27w2 is invertible, and view X as embedded in
3-space via the map (v, w) (v, w, (4v 3 + 27w2 )1 ). Let N (p, a, b) be
the number of Fp -points of X, that is to say, the number of points of
the reduction mod p of the said curve. These can be expressed in terms
of the traces of Frobenius ap (a, b), which are given as the coecient of
p1
xp1 in (x3 + ax + b) 2 .
The
number
of solutions v Fp to the equation v2 = w is equal to
1+ w
p
. Therefore, counting the point at , we must have that

3
u + au + b

u3 + au + b
N (p, a, b) = 1+ 1+ = p+1+ .
p p
uFp uFp
We could then try to use (5.7) to obtain an explicit expression for

N (p, a, b) as an arithmetic dierential operator in a and b, but that
will not work because we would obtain a series whose terms involve the
denominator (u3 + au + b)np , and for each a and b, there could be a
u Fp that annihilates
it.
p1
The identity ap a 2 mod p is analogous to N (p, a, b) ap (a, b)
mod p. The function ap (a, b) can can be represented [8, 10] as a quotient
of arithmetic dierential operators of order 2 dened on
X(Zp ) = {(a, b) Zp Zp : 4a3 + 27b2 Z

p },
of certain restricted power series in a, b, a, b, 2 a, 2 b and (4a3 +27b2 )1 ,

though these series are not as explicit as that given by (5.7) for the Leg-
endre symbol. We sketch the argument.
Let us recall that the R[p ]-module of characters of an elliptic curve
is generated by an operator f of order two or order one, depending upon
ur
p
the type of curve. Let = du/v be the standard invariant dierential.

Then we have a local expression
f (P ) = f (p1 log P ) , P X(pR)
about the origin for a unique operator f = 2 2p + 1 p + 0 R[p ],

the Picard-Fuchs operator associated to the character generator f . It
can be proved [8, 10] that, depending upon the type of elliptic curve, we
have that either f = 2p ap p + p or f = p , thus identifying
fully the coecients of f as an element of R[p ]. On the other hand
[10, Theorem 6.6], the Picard-Fuchs operator is proven to be of the form
f = 2p (f 2 /f 1 )p + h for certain arithmetic operators f 1 , f 2 given
by restricted power series in a, b, a, b, 2 a, 2 b and (4a3 + 27b2 )1 . The
desired assertion about N (p, a, b) follows by a comparison of coecients.
Theorems 9.2 and 10.1 show that in going from Zp to Z ur

p we loose
the property of expressing the characteristic function of a disc as an
arithmetic dierential operator. Since the maximal unramied extension
p of Qp is obtained by adjoining to Qp all the roots of unity of order
Qur
relatively prime to p, and the set of unramied integers Zur
p is just Zp =
ur
{x Qp : ||x||p 1}, it is natural to ask at what point we loose this

ur
property. We nish our work by proving that the said property is strictly
a p-adic one.
Let = n be any root of unity of order n with n relatively prime to p,
and let Qp () be the unramied extension of Qp obtained by adjoining
. This is a vector space over Qp whose dimension is the degree of n ,
which if n is a primitive root coincides with n. In what follows we do
not loose generality if we make this assumption. We denote by Zp () the
ring of integers of the extension, and let
p,n : Zp (n ) Zp (n )
be the homomorphism associated to the p operator on Zp ().
Lemma 10.12 The homomorphism p,n is not an analytic function.
Proof. Suppose that p,n (x) is equal to a restricted power series on

some disc of radius 1/pN . Without loss of generality, we may assume
that the disc is centered at the origin in Zp (n ). Since
p,n (x + y) = p,n (x) + p,n (y)
we may take (standard) derivatives with respect to y and evaluate the

134 ur
result at y = 0 to conclude that

d d
p,n (x) = p,n (0) ,
dx dx
and so the derivative is constant. Hence the power series giving p,n (x)
must contain linear or constant terms only. But since p,n is a homo-
morphism that extends the lift of Frobenius, this implies that the power
series and so p,n itself must be the identity. This is a contradiction.
For if the identity were a lift of Frobenius on Zp (n ), then it will induce
the identity as a Frobenius mapping on the quotient Zp (n )/pZp (n ). But
the latter is a nite eld, and the Frobenius mapping on a nite eld is
the identity if, and only if, the eld is Fp .
The result above leads us to think that Theorem 9.2 fails to hold as
soon as an unramied root of unity is adjoined to Zp . If we are to prove
that result in this context, we can no longer appeal to the use of Hilberts
Nullstellensatz as in the proof of Therem 10.1.
Analytic functions in Qp (n ) have unique power series expansion rep-
resentations [43]. We may use this fact in proving the following result,
the details of which we leave to the interested reader.
Theorem 10.13 On Zp (n ), no arithmetic dierential operator that
is not of order zero is analytic, and no analytic function that is not of
level zero is an arithmetic dierential operator.
References
[1] M.F. Atiyah & I.G. McDonald, Introduction to commutative algebra.

Addison-Wesley Publishing Co., Reading, Mass.-London-Don Mills, Ont.
1969, ix+128 pp.
[2] G. Bachman, Introduction to p-adic numbers and valuation theory, Aca-
demic Press, New York-London 1964, ix+173 pp.
[3] D. Bertrand, Sous groupes ` a un paramètre p-adiques de varietes de
groupes, Invent. Math. 40, 1977, pp. 171193.
[4] J. Borger, The basic geometry of Witt vectors, arXiv:math/0801.1691.
[5] A. Buium, Arithmetic analogues of derivations, J. Algebra, 198, 1997, pp.
290299.
[6] A. Buium, Arithmetic Dierential Equations, Mathematical Surveys and
Monographs, 118. American Mathematical Society, Providence, RI, 2005,
xxxii+310 pp.
[7] A. Buium, Dierential Algebra and Diophantine Geometry, Actualites
Mathematiques. Hermann, Paris, 1994, x+182 pp.
[8] A. Buium, Dierential characters of Abelian varieties over p-adic elds,
Invent. Math. 122, 1995, pp. 309340.
[9] A. Buium, Dierential modular forms, J. Reine Angew. Math. 520, 2000,
pp. 95167.
[10] A. Buium, Geometry of Fermat adeles, Trans. Amer. Math. Soc. 357,
2005, pp. 901964.
[11] A. Buium, Geometry of p-jets, Duke Math. J. 82, 1996, pp. 349367.
[12] A. Buium, Intersections in jet spaces and a conjecture of S. Lang, Ann.
of Math., 136, 1992, pp. 557567.
[13] A. Buium & S.R. Simanca, Arithmetic dierential equations in several
variables, Ann. Inst. Fourier. Grenoble 59, 2009, pp. 26852708.
[14] A. Buium, C.C. Ralph & S.R. Simanca, Arithmetic dierential operators
on Zp , J. Number Theory 131, 2011, pp. 96105.
[15] A. Buium & S.R. Simanca, Arithmetic Partial Dierential Equations, I,
Adv. Math. 225, 2010, pp. 689793.
[16] A. Buium & S.R. Simanca, Arithmetic partial dierential equations, II:
modular curves, Adv. Math. 225, 2010, pp. 13081340.
135
136 References
[17] A. Buium & S.R. Simanca, Arithmetic Laplacians, Adv. Math. 220, 2009,
pp. 246277.
[18] P.-J. Cahen & J.-L. Chabert, Integer Valued Polynomials, Mathemati-
cal Surveys and Monographs, 48. American Mathematical Society, Prov-
idence, RI, 1997, xx+322 pp.
[19] P. Cassidy, Dierential algebraic groups, Amer. J. Math. 94, 1972, pp.
891954.
[20] C. Chevalley, Class Field Theory, Nagoya University, Nagoya, 1954,
ii+104 pp.
[21] C. Chevalley, La theorie du symbole de restes normiques, J. reine angew.
Math., 169, 1933, pp. 140157.
[22] P. Deligne, La conjecture de Weil. I, Publ. Math. IHES, 43, 1974, pp.
273307.
[23] B. Dwork, On the rationality of the zeta function of an algebraic variety,
Amer. J. Math. 82, 1960, pp. 631648.
[24] B. Dwork, On the zeta function of a hypersurface, Publ. Math. I.H.E.S.
12, 1962, pp. 568.
[25] B. Dwork, G. Gerotto & F.J. Sullivan, An Introduction to G-functions,
Annals of Mathematical Studies 133, Princeton University Press, Prince-
ton, New Jersey, 1994.
[26] R. Hartshorne, Algebraic Geometry, Graduate Texts in Mathematics, 52.
Springer-Verlag, New York-Heidelberg, 2000, xvi+496 pp.

[27] H. Hasse, Uber die Darstellbarkeit von Zahlen durch quadratische Formen
im K orper der rationalen Zahlen, J. reine angew. Math., 152, 1923, pp.
129148.
[28] K. Hensel, Theorie der Algebraischen Zhalen, Leipzig 1908.
[29] N. Koblitz, p-adic numbers, p-adic analysis, and zeta-functions. Second
edition. Graduate Texts in Mathematics, 58. Springer-Verlag, New York,
1984, pp. xii+150.
[30] E.R. Kolchin, Dierential Algebra and Algebraic Groups, Pure and
Applied Mathematics, 54. Academic Press, New York-London, 1973,
xviii+446 pp.
[31] I. Kra & S.R. Simanca, On Circulant Matrices, preprint 2011.
[32] T. Kubota & H.W. Leopoldt, Eine p-adische theorie der zeta-werte I, J.
Reine und eigew. Math. 214/215, 1964, pp. 328339.
[33] B. Mazur & A. Wiles, Class elds of abelian extensions of Q, Invent.
Math. 76, 1984, pp. 179-330,
[34] S. Lang, Cyclotomic Fields I and II. Combined second edition. With an
appendix by Karl Rubin. Graduate Texts in Mathematics, 121. Springer-
Verlag, New York, 1990, xviii+433 pp.
[35] K. Mahler, Ein Beweis der Transzendenz der p-adische Exponenzial-
funkzion, J. reine angew. Math. 169, 1932, pp. 6166.
[36] K. Mahler, Introduction to p-adic numbers and their functions,
Cambridge Tracts in Mathematics. No. 64. Cambridge University Press,
London-New York, 1973, ix+91 pp.
References 137
[37] Yu.I. Manin, Algebraic curves over elds with dierentiation, Izv. Akad.
Nauk SSSR, Ser. Mat. 22, 1958, pp. 737756 = AMS translations Series
2, 37, 1964, pp. 5978.
[38] Yu.I. Manin, Periods of cusp forms and p-adic Hecke series, Math.
Sbornik., 92, 1973, pp. 378401.

[39] A. Ostrowski, Uber einige L
osungen der Funktionalgleichung (x)(y) =
(xy), Acta Math. 41, 1918, pp. 271284.
[40] R. Palais, Foundations of Global Non-Linear Analysis, W. A. Benjamin,
Inc., New York-Amsterdam 1968, vii+131 pp.
[41] J.F. Ritt, Dierential Algebra, American Mathematical Society Collo-
quium Publications, XXXIII, American Mathematical Society, New York,
N. Y., 1950, viii+184 pp.
[42] J-P. Serre, A Course in Arithmetic, Graduate Texts in Mathematics, 7.
Springer-Verlag, New York-Heidelberg, 1973, viii+115 pp.
[43] J-P. Serre, Lie algebras and Lie groups, Benjamin, New York, 1965.
[44] J-P. Serre, Local elds, Graduate Texts in Mathematics, 67. Springer-
Verlag, New York-Heidelberg, 1979, viii+241 pp.
[45] S.R. Simanca, A mixed boundary value problem for the Laplacian. Illinois
J. Math. 32, 1988, pp. 98114.
[46] S.R. Simanca, Pseudo-dierential operators. 236, Pitman Research Notes
in Mathematics, Longman Scientic & Technical, Harlow, 1990, 123 pp.

[47] R. Strassman, Uber den Wertevorrak von Potenzreihen im Gebiet der
p-adischen Zahlen, J. Reine Angew. Math. 159, 1928, pp. 1328.
[48] A. Weil, Number of solutions of equations in nite elds, Bull. Amer.
Math. Soc. 55, 1949, pp. 497508.
Index
p analyticity of a -function, 105

metric completion of Qp , 37 arithmetic dierential operator, 57
p -character, 82 arithmetic jet space, 74, 79
p -derivation on Zp , 57 ArtinHasse exponential, 31
Qp , 8 Bernoulli numbers, 2
algebraic properties, 17 binomial series, 24
analytic properties, 17
Cauchy sequences, 9
construction, 9 characteristic function of disc
locally compact, 17 expansion as a -function, 108
norm, 10 Mahlers expansion, 51
topology of, 15 compactness of Zp , 16
unramied extension Qur p , 36 completion, 9
Zp , 13 completion of a ring at an ideal, 71
compactness, 15 congruence, 14
denition, 14 continuous functions, 25
unramied integers Zur p , 36 convergence of series, 23
p-adic coordinates
distance, 8 harmonic p-adic coordinates, 120
integers, 14 standard p-adic coordinates, 107
norm, 8
order, 8 discs, 15
range of norm, 10 Dworks lemma, 32
representation, 12 exponential function, 30
p-adic norm, 10 Fermat quotient, 1, 5
extension to Qp , 35 eld
p-derivation, 56 p-adic numbers, 1, 8
additivity, 56 algebraically closed, 33, 70
associated homomorphism, 57 of fractions, 15
Leibnizs rule, 56 residue eld, 61
linear, 60
Hansels lemma, 20
adele ring, 3
hypersurface, 41
ane n-space, 69 id`
ele group, 3
ane space, 38 index of ramication, 35
ane variety, 38, 70 isosceles triangle property, 9
algebraic closure
of Qp , 34 jet bundles, 66
algebraic variety, 70 classical jet spaces, 66
analytic function, 49 Legendre symbol, 58
of level m, 50 Lift of Frobenius, 75
138
Index 139
local ring, 67
localization at prime ideal, 67
locally constant function, 51
locally ringed space, 69
Mobius function, 30
Mahlers expansion, 26
Neumann operator, 40
Newtons iteration, 18
non-Archimedean norm, 9
prime
degree of a prime, 110
singular, 110
Wronksian of pn , 116
Wronskian of p, 110
Wronskian quotients, 110
projective space, 38
projective variety, 38
rational normal curve, 38
restricted power series, 48
restricted sequence, 26
Riemann zeta function, 2, 41
ringed space, 68
roots of unity
in Qp , 20
scheme, 69
ane scheme, 69
locally Noetherian, 70
regular, 72
spectrum of a ring, 68
principal open sets, 68
Zariski topology, 68
Teichm uller representation, 52
Theorem
HasseMinkowskis, 2
Mahlers, 26
Strassmanns, 53
Wilsons, 33
Weil zeta function, 40
Weil conjectures, 43
Witt ring, 128

(London Mathematical Society Lecture Note Series) Claire C. Ralph, Santiago R. Simanca-Arithmetic Differential Operators Over The P-Adic Integers. 396-Cambridge University Press (2012)

Загружено:

Сведения о документе

Оригинальное название

Авторское право

Доступные форматы

Поделиться этим документом

Поделиться или встроить документ

Параметры публикации

Этот документ был вам полезен?

Это неприемлемый материал?

Авторское право:

Доступные форматы

(London Mathematical Society Lecture Note Series) Claire C. Ralph, Santiago R. Simanca-Arithmetic Differential Operators Over The P-Adic Integers. 396-Cambridge University Press (2012)

Загружено:

Авторское право:

Доступные форматы

L O ND O N MATH EMAT I C A L SO C I E T Y L E C T U R E N O T E SER I ES

Managing Editor: Professor M. Reid, Mathematics Institute,

Arithmetic Differential Operators

C. Ralph and S. Simanca, 2012

This publication is in copyright. Subject to statutory exception

First published 2012

Printed in the United Kingdom at the University Press, Cambridge

ISBN 978-1-107-67414-1 Paperback

Cambridge University Press has no responsibility for the persistence or

9 Characteristic functions of discs in Zp : harmonic co-

Our purpose in this monograph is to provide a concise and complete

numerator of B2 , B3 , . . . , Bp3 and that Fermat last theorem holds for

Its ring structure is obtained by dening addition and multiplication

functions of elliptic curves. The p-adic integers Zp were known to appear

to dene the notion of arithmetic derivative of x in the direction of p.

the notion of a p -character of a group scheme, the arithmetic version

rst appearance in Chapter 5, where we tie them to their associated ho-

The eld Qp of p-adic numbers is the completion of the eld Q of rational

We shall denote by dp the metric on Q that this norm induces.

ad and the biggest power dividing bc, we have that

This is a Cauchy sequence also, and we have

{qn }{rn } = {0, . . . , 0, 1, 1, . . .} = {1, 1, . . .} {1, . . . , 1, 0, 0, . . .} .

If I were an ideal with M I properly, then I would contain a non-null

||||p = lim ||qn ||p ,

extending in this manner the p-adic norm on Q to a p-adic norm on Qp .

We have the mapping

2.1 A pragmatic realization of Qp

and v is the inverse of bn mod p. Therefore,

Proposition 2.1 Any p-adic number can be written in the form

Example 2.2 We nd the expansion of 1/5 in Q3 . This rational has

Using the fact that

2.2 The p-adic integers Zp and their eld of fractions

the set of p-adic numbers whose canonical expansion (2.4) begins at

where Zp . The said sequence {an } Z satises the congruences

for each n 1. Any p-adic integer satises an innite set of congruences

Example 2.3 Let us consider the sequence of rational integers given

We have that an+1 an = 3 10n , and so an+1 an mod 5n . Thus, the

The discussion above can be reversed [42]. For if An = Z/pn Z is the

2.3 The topology of Qp

It then follows that if two discs have a nontrivial intersection, one of

proving the desired result.

Remark 2.6 Compactness of Zp can be proven in yet another way,

Lemma 2.7 The set of p-adic numbers Qp is totally disconnected.

2.4 Analytic and algebraic properties of Qp

||f ()||p < 1, by the congruences (2.6) for , we can nd a1 Z such

converges p-adically, and the limit is a p-adic integer root of f (x).

Theorem 2.8 Let a be any rational integer not divisible by p2 . Con-

a). If p | a, the equation has no solution in Qp .

We now reconsider the distinct primes p and p of the beginning, with

Remark 2.10 The Newtons iteration algorithm on Zp discussed above

a unique a Znp such that F (a) = 0 and a a1 mod p.

2.5 (p 1)-roots of unity in Qp

has exactly p1 roots in Qp , the maximum number of possible solutions,

converges. For using the non-Archimedean property of p , it suces

We have the eld identications Zp /pZp = Z/pZ

study arithmetic dierential operators can be done with more generality

||sn sm ||p = ||an + + am+1 ||p max ||aj ||p 0 ,

showing that {sn } is a Cauchy sequence, and so its limit exists in Qp .

(F (x + y) F (x))/y. For, by the binomial expansion, we have that

and if ||y||p < ||x||p , we have

Hence, the series above converges uniformly in y so long as y is p-adically

In a sense to be made precise below, the polynomial functions x

Theorem 3.1 Let y Qp be a p-adic number such that ordp y 0.