Welcome to Scribd, the world's digital library. Read, publish, and share books and documents. See more
Download
Standard view
Full view
of .
Look up keyword
Like this
1Activity
0 of .
Results for:
No results containing your search query
P. 1
mazur

mazur

Ratings: (0)|Views: 20|Likes:
Published by rauindia

More info:

Published by: rauindia on Nov 11, 2011
Copyright:Attribution Non-commercial

Availability:

Read on Scribd mobile: iPhone, iPad and Android.
download as PDF, TXT or read online from Scribd
See more
See less

11/11/2011

pdf

text

original

 
BULLETIN (New Series) OF THEAMERICAN MATHEMATICAL SOCIETYVolume 45, Number 2, April 2008, Pages 185–228S 0273-0979(08)01207-XArticle electronically published on February 6, 2008
FINDING MEANING IN ERROR TERMS
BARRY MAZUR
In memory of Serge Lang 
Introduction
Four decades ago, Mikio Sato and John Tate predicted the shape of probabilitydistributions to which certain “error terms” in number theory conform. Theirprediction—known as the Sato-Tate Conjecture—has been verified for an importantclass of cases thanks to the recent work of Laurent Clozel, Michael Harris, andRichard Taylor [3], and of Michael Harris, Nicholas Shepherd-Barron, and RichardTaylor [18], combined with Richard Taylor’s most recent [54], which establishes this advance in our understanding.Part of the beauty of this breakthrough is how it pulls together progress madeover the past quarter century and work from significantly different viewpoints—from the theory of automorphic representations, from algebraic geometry, and fromGalois deformation theory—a demonstration, yet again, of the intense unity of mathematical thought.My aim is to discuss, in concrete terms, two “sample problems”—one still open,and one settled by the recent work—that give rise to error terms, about which theSato-Tate Conjecture makes precise predictions.I thank Andrew Granville, Michael Harris, Nick Katz, Mark Kisin, PhillipeMichel, Michael Rubinstein, Peter Sarnak and Richard Taylor for much enlighten-ing discussion and for their exceedingly helpful comments. I’m grateful to WilliamStein for conversations and advice about the substance and the format of this arti-cle and for the computations and graphs that appear in it, and also to ChristopherSwierczewski for computations, among which are those that produced the q-q plots.All plots were created using the free software SAGE (
sagemath.org
). For a file con-taining all figures in this article, with codes, see
http://wstein.org/mazur/sato.tate.figures
. I also want to thank Susan Holmes, who explained to me the natureand utility of q-q plots.
Received by the editors September 9, 2007.2000
Mathematics Subject Classification.
Primary 11-02, 11F03, 11F80, 11G05, 11G40.Part I of this article was presented in the
Current Events Bulletin 
section of the winter meetingof the AMS on January 7, 2007, in New Orleans. The title of the talk was “The structure of errorterms in number theory and an introduction to the Sato-Tate Conjecture”. Part I and some of Part II were published in the
Current Events Bulletin 
of the AMS that was distributed at themeeting.
c
2008 American Mathematical SocietyReverts to public domain 28 years from publication
185
 
186 BARRY MAZUR
Contents
Part I. The general question of error terms. Our first “sample problem”.1871. Error terms and the Sato-Tate Conjecture1871.1. Why are there still unsolved problems in number theory?1871.2. Much of the depth of the problem is hidden in the structure of theerror term1881.3. Strict square-root accuracy1891.4. Some sample arithmetic problems1891.5. The “next question”1921.6. The distribution of scaled error terms1921.7. Rates of convergence (first version)1941.8. Error term roulette1961.9. Eisenstein series as
good approximation 
and error term as
cusp form 
196Part II. An elliptic curve. Our new “sample problem”.1982. The number of points of an elliptic curve when reduced mod
p
, forvarying
p
1982.1. The elliptic curve that we will be working with1982.2. Rates of convergence (second version)2002.3. Overcounts versus undercounts2012.4. Error terms modulo
m
2042.5. Correlations205Part III. About the proof of Sato-Tate for the elliptic curve
.2053. Reducing the problem to a question about analytic continuation of 
L
-functions2053.1. The Sato-Tate distribution2053.2. Bases for the ring of polynomials2073.3.
L
-functions2083.4. Sato-Tate and the Generalized Riemann Hypothesis2093.5. Meromorphic extension of 
L
-functions2104. Replacing the problem of analytic continuation of 
L
-functions byquestions about automorphic forms2104.1. The reciprocity “divide”2104.2. Automorphic representations, automorphic forms2114.3. Galois representations associated to the symmetric
m
-th powers of our data2154.4. Digression on compatible families and Galois characters2164.5. Langlands reciprocity2184.6. Potential automorphy2184.7. Galois deformation theorems and the pivotal role played by residualrepresentations2194.8. Hopping from one prime to another2214.9. A rich source of potentially automorphic Galois representations2224.10. Concluding the theorem2234.11. Interpreting Sato-Tate as a statement about
equidistribution 
2234.12. Expository accounts of this recent work225About the author225References226
 
FINDING MEANING IN ERROR TERMS 187
Part I. The general question of error terms. Our first“sample problem”.
1.
Error terms and the Sato-Tate Conjecture
1.1.
Why are there still unsolved problems in number theory?
Eratos-thenes, to take an example—and other ancient Greek mathematicians—might haveimagined that all they needed were a few powerful insights and then everythingabout numbers would be as plain, say, as facts about triangles in the setting of Euclid’s
Elements of Geometr
. If Eratosthenes had felt this, and if he now—transported by some time machine—dropped in to visit us, I’m sure he would bequite surprised to see what has developed.To be sure, geometry has evolved splendidly but has expanded to higher realmsand more profound structures. Nevertheless, there is hardly a question that Euclidcould pose with his vocabulary about triangles that we can’t answer today. And,in stark contrast, many of the basic naive queries that Euclid or his contemporariesmight have had about primes, perfect numbers, and the like would still be open.Sometimes, but not that often, in number theory we get a complete answer to aquestion we have posed, an answer that finishes the problem off. Often somethingelse happens: we manage to find a fine, simple,
good approximation 
to the dataor phenomena that interests us—perhaps after some major effort—and then wediscover that yet deeper questions lie hidden in the error term, i.e., in the measureof how badly our approximation misses its mark.A telling example of this, and of how in the error term lies richness, is themanner in which we study
π
(
) := the number of prime numbers less than
. Thefunction
π
(
) is shown in Figure1.1in various ranges as step functions giving the“staircaseof numbers of primes.As is well known, Carl Friedrich Gauss, two centuries ago, computed tables of 
π
(
) by hand, for
up to the millions, and offered us a probabilistic “first” guessfor a nice smooth approximating curve for this data, a certain beautiful curve that,experimentally, seems to be an exceptionally good fit for the staircase of primes.The data, as we clearly see, certainly cries out to us to guess a
good approxima-tion 
. If you make believe that the chances that a number
is a prime is inverselyproportional to the number of digits of 
, you might well hit upon Gauss’s guess,which produces indeed a very good fit. In a letter written in 1849 Gauss claimedthat as early as 1792 or 1793 he had already observed that the density of prime
Figure 1.1.
The step function
π
(
) counts the number of primesup to
.

You're Reading a Free Preview

Download
scribd
/*********** DO NOT ALTER ANYTHING BELOW THIS LINE ! ************/ var s_code=s.t();if(s_code)document.write(s_code)//-->