M. A. Tsfasman, S. G. Vlăduţ Auth. Algebraic-Geometric Codes PDF
M. A. Tsfasman, S. G. Vlăduţ Auth. Algebraic-Geometric Codes PDF
M. A. Tsfasman, S. G. Vlăduţ Auth. Algebraic-Geometric Codes PDF
Managing Editor:
M . HAZEWINKEL
Centre for Mathematics and Computer Science, Amsterdam, The Netherlands
Editorial Board:
Volume 58
Algebraic-Geometric
Codes
by
M. A. Tsfasman
Institute of Information Transmission,
Academy of Sciences,
Moscow, U.S.S.R.
and
S. G. Vlädut
Centrai Economical Mathematical Institute,
Academy of Sciences,
Moscow, U.S.S.R.
'Et moi, ..., si j'avait su comment en revenir, One service mathematics has rendered the
je n'y serais point aIle.' human race. It has put common sense back
Jules Verne where it belongs, on the topmost shelf next
to the dusty canister labelled 'discarded non-
The series is divergent; therefore we may be sense'.
able to do something with it. Eric T. Bell
O. Heaviside
Mathematics is a tool for thought. A highly necessary tool in a world where both feedback and non-
linearities abound. Similarly, all kinds of parts of mathematics serve as tools for other parts and for
other sciences.
Applying a simple rewriting rule to the quote on the right above one finds such statements as:
'One service topology has rendered mathematical physics .. .'; 'One service logic has rendered com-
puter science ...'; 'One service category theory has rendered mathematics .. .'. All arguably true. And
all statements obtainable this way form part of the raison d' etre of this series.
This series, Mathematics and Its Applications, started in 1977. Now that over one hundred
volumes have appeared it seems opportune to reexamine its scope. At the time I wrote
"Growing specialization and diversification have brought a host of monographs and
textbooks on increasingly specialized topics. However, the 'tree' of knowledge of
mathematics and related fields does not grow only by putting forth new branches. It
also happens, quite often in fact, that branches which were thought to be completely
disparate are suddenly seen to be related. Further, the kind and level of sophistication
of mathematics applied in various sciences has changed drastically in recent years:
measure theory is used (non-trivially) in regional and theoretical economics; algebraic
geometry interacts with physics; the Minkowsky lemma, coding theory and the structure
of water meet one another in packing and covering theory; quantum fields, crystal
defects and mathematical programming profit from homotopy theory; Lie algebras are
relevant to filtering; and prediction and electrical engineering can use Stein spaces. And
in addition to this there are such new emerging subdisciplines as 'experimental
mathematics', 'CFD', 'completely integrable systems', 'chaos, synergetics and large-scale
order', which are almost impossible to fit into the existing classification schemes. They
draw upon widely different sections of mathematics."
By and large, all this still applies today. It is still true that at first sight mathematics seems rather
fragmented and that to find, see, and exploit the deeper underlying interrelations more effort is
needed and so are books that can help mathematicians and scientists do so. Accordingly MIA will
continue to try to make such books available.
If anything, the description I gave in 1977 is now an understatement. To the examples of
interaction areas one should add string theory where Riemann surfaces, algebraic geometry, modu-
lar functions, knots, quantum field theory, Kac-Moody algebras, monstrous moonshine (and more)
all corne together. And to the examples of things which can be usefully applied let me add the topic
'finite geometry'; a combination of words which sounds like it might not even exist, let alone be
applicable. And yet it is being applied: to statistics via designs, to radar/sonar detection arrays (via
finite projective planes), and to bus connections of VLSI chips (via difference sets). There seems to
be no part of (so-called pure) mathematics that is not in immediate danger of being applied. And,
accordingly, the applied mathematician needs to be aware of much more. Besides analysis and
numerics, the traditional workhorses, he may need all kinds of combinatorics, algebra, probability,
and so on.
In addition, the applied scientist needs to cope increasingly with the nonlinear world and the
v
vi SERIES EDITOR'S PREFACE
extra mathematical sophistication that this requires. For that is where the rewards are. Linear
models are honest and a bit sad and depressing: proportional efforts and results. It is in the non-
linear world that infinitesimal inputs may result in macroscopic outputs (or vice versa). To appreci-
ate what I am hinting at: if electronics were linear we would have no fun with transistors and com-
puters; we would have no TV; in fact you would not be reading these lines.
There is also no safety in ignoring such outlandish things as nonstandard analysis, superspace
and anticommuting integration, p-adic and ultrametric space. All three have applications in both
electrical engineering and physics. Once, complex numbers were equally outlandish, but they fre-
quently proved the shortest path between 'real' results. Similarly, the first two topics named have
already provided a number of 'wormhole' paths. There is no telling where all this is leading -
fortunately.
Thus the original scope of the series, which for various (sound) reasons now comprises five sub-
series: white (Japan), yellow (China), red (USSR), blue (Eastern Europe), and green (everything
else), still applies. It has been enlarged a bit to include books treating of the tools from one subdis-
cipline which are used in others. Thus the series still aims at books dealing with:
- a central concept which plays an important role in several different mathematical and/or
scientific specialization areas;
- new applications of the results and ideas from one area of scientific endeavour into another;
- influences which the results, problems and concepts of one field of enquiry have, and have had.,
on the development of another.
In designing an (error correcting) code one naturally, given certain desiderata such as the number of
errors that can be automatically corrected, desires to code as efficiently as possible. For a long time
there was a conjectural bound, the Gilbert-Varshanov bound. Then came Goppa with his algebraic-
geometry based codes (now usually called Goppa codes) and using these ideas the bound was
broken by Th. Zink and the authors of the present volume (1981).
There are now two books on these codes in this series and they complement and supplement
each other. The first, by Goppa himself, stresses the intuitive basic ideas. But these codes are based
on algebraic geometry, mostly the problem of the number of rational points of curves defined over
finite fields. And that is a technically complicated subject. The present book is a complete full treat-
ment of the subject by two authors who have been in the forefront of the research on these codes
for the full 10 years of their history.
The shortest path between two truths in the Never lend books, for no one ever returns
real domain passes through the complex them; the only books I bave in my library
domain. are books that other folk have lent me.
J. Hadamard Anatole France
La physique ne nous donne pas seulement The function of an expert is not to be more
l' occasion de resoudre des problemes ... elle right than other people, but to be wrong for
nous fait pressentir la solution. more sophisticated reasons.
H. Poincare David Butler
* *
IIxll = I {i I xi "$ O} I
then 0 < Kq(d) < 00 and the question is that of its value.
One is able to give lower and upper bounds for
Kq(d)/(d - 1) which do not depend on d . Roughly speaking,
for any d we have
~ ~ Kq(d)/(d - 1) ~ (q - l)/q
then
k-1
q (q - 1)
0q(k) = qk
- 1
xxi
xxii ADVICES
algebraic-geometric are
codesdenoted (X, 'P ,D) L '
(X,'P,D)Q' etc., while in many other papers they are denoted
C(G,D) and C* (G,D) , respect1vely,
. see section 3.1.1;
In means loge
exp(x) = eX
the symbol
• denotes end of proof (or its absence);
COD E S
This part is an introduction to the theory of
error-correcting codes. Presenting principal notions and
examples we always keep in mind the "algebraic-geometric
philosophy", which will come out explicitly in Part 3, but
here we do not touch any results using algebraic-geometric
codes.
Chapter 1.1 contains principal definitions and
properties of codes, and also some motivations. Chapter 1.2
is devoted to some examples, or, to be precise, to diverse
interesting classes of codes and various constructions of
new codes starting with those already known. In Chapter 1.3
we discuss different asymptotic problems (i.e. the rigorous
way to put questions about codes of large length), the
algorithmic point of view included. We end this part with
some notes on history and bibliography.
4 CODES Part 1
5
6 CODES Part 1
Remark 1. 1. 1.
In the coding theory there is a good
tradition attaching symbols C, n, k , d , q, R, c5
to the defined notions. We try as far as possible to hold to
this tradition.
1.1.2. [n,k,d]q-systems
m (k) ~ q + 1 ;
q
if q is even then
mq (3) ~ q + 2
mq(k) = q + 1
n n-r r
L A·x .y
r=O r
'\'
L. x n-lIvll .y IIvll
vee
n n r
'\' n r
L. A ·x - or W'e ( y ) L A .y
r=O r r=O r
n-d i n-d-i
L Ad+i·Y·x
i=O
Chapter 1.1 CODES AND THEIR PARAMETERS 15
a
xn + L A .·X
i
n-1.
i=O
a
L B 1... (x - l)i
i=O
where
B.1.
n-i
L A. E (-1) n+ i + j . ( j.). B .
j=n-a 1. j=n-i n-1. )
ci = { X E ~n I x·y
q
o for any y E C }
n
(here x-y L x.-y_ is the inner product).
i=1 1. 1.
-k x - y)
W (x:y) q 'We(x + (q - l)'Y
Col
k + d ~ n + 1 - g
ol
(n - k) + d ~ n + 1 - g
a
L B"(X-1)i
~
i=O
Chapter 1.1 CODES AND THEIR PARAMETERS 17
(?). (qa-i+l
... _ 1) 2: Bi 2: max { (
0 ,n )
i· (q a-i-g+l - 1) }
and for i ~ a - 2g + 1
w -k
(x: y) q ·wC(x + (q - l)·y x - y)
C.l
.1
a =n - k + g - 1 and we
get
a.l
xn + L B '.. (x - 1) i
i=O ~
q - k . ( (x + q - 1) n + aLB .. q.
i (x - 1) n-~ .J
i=O ~
.1
a
L
i=O
18 CODES Part 1
The lower bound and the equality are proved. The upper bound
will be proved below, after we interpret the values Bi in
terms of [n,k,d] -systems (see Theorem 1.1.26 and Exercise
q
1.1.27) .
•
Remark 1.1.17. It is easy to see that B = B'
k n-k
Exercise 1.1.18.
Show that if k+d~n+1 then
k + d =n + 1
and (n - k) + d.L = n + 1 t.e. that in
this case the code is of genus 0 (an MDS-code) .
Ao = 1, Al = 2 , A2 = 1 , A
3
= A
4
= o. It is a code of
genus at most 2 a = 3 the calculation of B.
~
gives
B3 = Al 2, B2 = 3Al + A2 = 7 , Bl = 3Al + 2A2 + A3 = 8,
Bo = Al + A2 + A3 + A4 = 3. Let us point out that in this
[ ~ i ~ ~ ~ ~l
000 110
000 011
E x(a'b) __ { qO
for b * °
aelF
q for b °
Proof: Let b '" 0, choose ao such that X (a 'b) '" 1.
o
Then
X1 (U'V)
then
is an additive character of
/\
1
L f(u) Ter nee
L f(u)
necl.
/\
Proof: L f(u) = L n L Xu(v) -f(v)
ueC velF q uee
(1.1.1)
L x 1 (O) lei •
uee
x (u) - L X (u) = L X (u + u)
v 0 uee v uee v 0
for v ~ el.
for v e el.
•
22 CODES Part 1
f .• IFq
n ----7 A , f (u) = x n-liull 'y lIuli
= L X (u v
1 1 1
)·x
v 1 elf q v 2 elFq
q-1 q-1
n 1-V. v.~ n
IT LX 1 (u.v.)·x
~ ~
~
.y IT(x + y' L
*X 1 (u.v.))
~ ~
i=l v.elF
~ q
n
IT (x - Y + y' L X (u.· a) )
i=l aelF q 1 ~
n u,?-l 1-u,?-1
IT( (x - y) ~ . (x + (q - 1) .y) ~ )
i=l
(x + (q - 1) 'y)n-lIuli. (x _ y) lIuli
Let
Theorem 1.1.26. C be a linear [n,k,~d]q-code,
n-d
xn + LB . . (x - 1) i
i=O ~
where for dL _ 1 ~ i ~ 0
B. = (~) . (qk-i
~
- 1)
and for n - d ~ i ~ dL
B.
~
~ max {0, (n)i . (qk-i - 1 )}.
•
Exercise 1.1.27. Check the following interpretation of
B.
~
in terms of [n,k,d]q-systems. Let 1> = {P l , · · · ,Pn } be
an [n,k,d] -system, P. e V . By H. denote the hyperplane
* q ~ ~
in V corresponding to Pi For 'R s;; 1> let
£ ('R) dim ( n Hi) . Then
P .e'R
~
B.
~
'Rc1>, I 'R I =i
2 2
Wc(x:y) = P(x + (q - l)·y : y·(x - y»
•
Remark 1.1.30. Of course, not every polynomial of this
form is in fact an enumerator of some linear code. For
Chapter 1.1 CODES AND THEIR PARAMETERS 25
1.1.4. Bounds
a) [n + 1 , k , dJ q
b) [n - 1 k - 1 dJ q
c) [n - 1 k d - IJ q
d) [n k - 1 d Jq
e) [n k d - IJ q
k-l
n ~ L
i=O
= max I~ n HI n - d .
H
q+l
(q + 1)-(n - d) ~ L IHi n ~I IV n 'PI + q-IH' n 'PI
i=1
= n + q- (n - d - d') ,
Chapter 1.1 CODES AND THEIR PARAMETERS 29
d s
n·qk . (q - 1)
k
(q - l)·q
1
d s E d(x,y)
x,yeC
n
E d(x,y) E E (1 - IS )
x,yeC i=l x,yeC xi'Yi
n
L E (1 - IS )·X .·X . s n·max Q(Z)
i=l a,belF a,b a,~ b,~ Z
q
fd;11
n - k L
i=O
(1!)
~
. (q -
n
Proof: Consider spheres in Wq of radius t I --,d-211
centered at the code vectors (by definition a sphere of
radius t centered at a is
t
IB t (a) I L (~). (q - 1) i
i=O •
Exercise 1.1.43. Prove the lemma.
n - k ~ (n)
logq v + v·logq (q - 1) - logq d + logqA
k
o
Lemma 1.1.45. Let A(n,d) = q , k = max{k} , the
o
maximum being taken over all [n,k,d] -codes C vith given
q
nand d ; let A(n,v,d) be the maximum possible number of
vectors of veight v in IF n such that the distance betveen
q
any tvo of them is at least d . Then for any
A(n,d,v) A (n ,d)
(Z).(q - l)v qn
•
Exercise 1.1.46. Prove the lemma. (Hint: use the fact
that for any code C there exists a sphere of radius v
such that the volume of its intersection with the code C
is at least the average volume of intersection of C with
such spheres).
32 CODES Part 1
d
A(n,d,w)
-
2
·W
+ q 2 ·W
(q _ l).n
1
whenever the denominator is positive .
•
Exercise 1.1.48. Prove the lemma. (Hint: Estimate the
sum of all pairwise distances in a spherical code, as it is
done in the proof of Theorem 1.1.39).
-k
W (x: y) q 'Wc(x + (q - l)y x - y)
Cl.
n
A'. q -k L AJ.·P i (j)
~
j=O
L P. (x) ·zi
~
i=O
Chapter 1.1 CODES AND THEIR PARAMETERS 33
n
E A). . Pi (j) ?! 0
j=O
n
We want to give an upper bound for qk = 1 + E Ai' i.e. to
i=d
solve the following linear programming problem (note that
n
E A .• p (j) ?! 0 for A.?! 0 ):
j=O J 0 )
n
H 1 + E x. -----+) max
i=d ~
n
(j) . (q - E
i=d
PJ.(j) ·Xi ?! 0 for j 1, ... , n
n
1 + E a.·P.(j) :S 0
i=l ~ ~
•
Let us remark that this result can be strengthened if
we apply the linear programming method to the
34 CODES Part 1
n-k
q <
d-2
L (~J'.). (q - 1) j
j=O
n-k
d-l
q < L (~). (q -
i=O
37
38 CODES Part 1
n
Ev'P : L(a) ~ IFq
Ev'P : f ~ ( f (P 1 ) , • • • , f (Pn »
n
P . e"P
(x - Pi)
-1
,
~
ge (x) for e ~ n - a - 2.
0 ~ Consider a function
F (x)= f (x) . go (x) , f (x) being a polynomial. Recall the
definition of the residue of F at Pi
where
Yi =n
. . (P.~ - P J.)
-1
,
J*~
k-1 .
xn + L (~).{qk-2 - l)'{x _ l)i
i=O 2
•
Exercise 1.2.5. Establish this formula for Reed-
Solomon codes by a direct computation. Check that for i * 0
s j (v) L V.·p~
1. 1.
o s j s 2t - 1 .
Pie'P
t
l
g(x) L y·x =c·TT(x-p.)
l=O l ieI 1.
t
L y{sJ.+l = 0, 0 s j s t - 1 ,
l=O
t t .+l
L y{S j+l L L y{e .. p~
1. 1.
l=O l=O P .e'P
1.
L e . . prg(P.) 0
1. 1. 1.
ieI
t l
g' (x) L y'·x
l=O l
Chapter 1.2 EXAMPLES AND CONSTRUCTIONS 43
then, setting
t-1
Fj(X) .n.
~eI,~*J k=O
L b
k
k
·X
e .. F . (P . ) . g' (P.)
~ J ~ ~
t-1
= L L e .. bk·~·g' (P.)
ieI k=O ~ ~ ~
t-1 t
'r'
L
'r'
L
b k·Ye·
' 'r'
L e.· pkH.
.
k=O e=o ieI ~ ~
is an [n, m + I, n - q
m-l ]q-COde with n ~ q
~
also [(a·P.)
~
= 0 for all a ~ 0, 1 Therefore the number
m
of zeroes of [ in is at most q q -- 11 . The maximum
q
m+l _ 1
cardinality of is (take for example all
q - 1
non-zero elements of IF m+ 1 such that their first non-zero
coordinate is 1 ). We
q
obtain an [n, m + I, n -
qm _
q - 1
1]
q
Chapter 1.2 EXAMPLES AND CONSTRUCTIONS 45
qm+l _ 1
code for n :s ........----,:--- in particular a code C with
q - 1
parameters
m+l
[q q _- 1
1 m]
,m + 1, q
q
qm+l _ 1
[n, n - m - 1, 3 1q n :s
q - 1
m+1 1
(and for n -- q q _ - 1 we know the spectrum of such codes).
These codes are good if we are interested in codes with
n = qm+1 - 1
d=3. For they lie on the Hamming bound
q - 1
(cf. Theorem 1.1.41).
46 CODES Part 1
Ev ep : Lm (r) -----+ ~~
m-a-1
d = (q - b) .q
r (m)i
[2m'i~O ,2
m-r] 2- code .
Chapter 1.2 EXAMPLES AND CONSTRUCTIONS 47
then
Let us
consider the case (n, q) = 1 , then the
polynomial n
(x - 1) has n distinct roots in ifq'. let
W be the splitting field of this polynomial, Le. the
qm
smallest extension of IFq where (x n - 1) decomposes
into a product of linear factors.
The group *
Wqm is cyclic, its subgroup Iln of n-th
roots of 1 is also cyclic, fix its generator a, then
i 2i 1 (n-1)i 1
1
1 a a a
H=
i
r
2i
r
(n-1)i r
1 a a a
Chapter 1.2 EXAMPLES AND CONSTRUCTIONS 49
Summing up we get:
i a(n-1)i 1
1
1 a
H =
ir a(n-1)i r
1 a
BCH-code
A C is a cyclic code such that
I ;2 b + 1 ,
{b , , b+ l!. - 2} ...
for some b ~ 0 and
l!. ~ 2 We call. l!. the designed distance of C. A BCH-code
in the narrow sense is a BCH-code with b = 1, a primitive
BCH-code is a BCH-code of length n = qm - 1 (in this case
ex is a primitive element of F ).
qm
use d ~ d' ).
- r~ ~ 11
S;;
n - k :s m· (d - 1 - r ~ 11)
d
m
Theorem 1.2.15. For any m~ 2 and n = q - 1
there exists a primitive q-ary BCH-code in the narrow sense
whose parameters are
n·q
d = 2, 3, ... , 1 + m. (q _ 1)
vl+ 1 = - E
i=l ~
v. . Prove that for i == -1 (mod 4) the new code
C' is self-dual.
Chapter 1.2 EXAMPLES AND CONSTRUCTIONS 53
let
n vi
L
i=l Z - Pi
The code
C = (v e ~~ I Rv (z) = 0 (mod G (z) ) }
n-l
y = (l,a,a , •.• ,a
2
),
i. e.
f (2~).
i=1 ~
(q - 1) i !O n - k + 1 •
n-1 .
(Hint: The line (c ,c iC ,c ·a i ••• iC ,a ·c) 1ncludes
1 1 2 2 2 n n i-I
at least n + 1 - k different q-ary vectors (ci,a ·C i )
of length 2m i estimate the total weight).
or
244823040 .
= X 24 -W24 (l/x)
= (x - 1)' (x
54.
+ x - x
3
+ x
2
- 1)' (x
532
- x + x - 1) .
or [23,12,7]2' or [11,6,5]3 •
•
58 CODES Part 1
1 if i j
f i (gj"H) ={ 0 if i ;t j
[n - d, k - 1, l!: 1~ r ] q -code.
62 CODES Part 1
[n + 1, k, d + 1]2-code •
d' ~ d , k' ~ n - m· (n - k) •
{ O} = C~ c C~ c c Ci
i m.
such that Ki = L m. , qi = r
~
Consider C'
t
as a direct
j=1 J
t i
sum Ci = e
V. such that C'. = e V. Let w.
~
be the
j=1 J ~ j=l J
concatenation of V.
~
and Ci (note that dim V.
~
= m. ).
~
Set C = e W ••
i=1 J
restriction C' = C n IF nr s; IF nr •
It is more difficult with the concatenation. We need
the following notion: erasure is an error whose position we
Chapter 1.2 EXAMPLES AND CONSTRUCTIONS 65
up to t ' :s rd'-
--2-
1l , where C is an [n,k,dJ q -code and
r
T :s D'; 1l errors, D' :s D. Then the concatenated
[n'N,k'm,d'DJr-code C' has a decoding algorithm correcting
ASYMPTOTIC PROBLEMS
67
68 CODES Part 1
1
\
\
\
\
A \\
\
\ D
\
c
B
1
Fig.1.1
k
.5 :s q . (q - 1)
(~ - 1)·q
°>
(XI ,
~ ~ ~q .4
•
Exercise 1.3.7. Prove the theorem. (Hint: Cf. the
proof of Theorem 1.3.1; the Plotkin bound yields
CX q ( (q - l)/q) = 0, the rest follows from the spoiling
lemma) .
•
Exercise 1.3.9. Prove that
!'log ( ~
n q i=O
(~).(q
~
- l)i) =H
q n
(£) + 0(1)
q - 1
1 - Hq ( - q - -
q - 1
q .
I 1 -
q. 0 '
q::-r )
q - 1·0
q
2 2
min (1 + h(u) - h(u + 20·u + 20»
O<u~1-20
where
h (x)
•
Functions R4 and R4 (2l are sometimes called the
bounds of four. Let us point out that for q =2 and for
o 0.273
~ the bound R4 (2l equals R4 (since the minimum
is achieved for u = 1 - 20 ).
min
Chapter 1.3 ASYMPTOTIC PROBLEMS 75
0 S L S 1 , 0 s ~/L s 1/2 0 s v s 1,
0 s 1} s (q - 2) -v/(q - 1)
0 s E- 1} s min {v - 1), 1 - v}
- 1})-h(~ - 1)
~) s v - q - 1
(3 (1
- 1} 1 - 1} q 2-1}
IS ~ 2~ + (2(3 + (v - (3) -K
q-1 (_1}_))
v - (3 - (1 - L)
where
x- (1 - x) - y- (1 - y)
h(x,y) ,Os x, y s 1
1 + 2/y- (1 - y) ,
q - 2 q - 3 2
q _ 1 - q _ 1- x - q _ 1- / (q - 2) -x- (1 - x)'
RGV (
q - 1 _
q
x) 2(q-1)·lnq·x
q2 2 + 0 (x2)
0 g: - 1
0 t
q + q - 1
and
t
t·g
R(o 0 ) 1 + - 10gq(qt + q - 1)
qt + q - 1
•
Chapter 1.3 ASYMPTOTIC PROBLEMS 77
Theorem 1.3.19.
k
CXq(oS) ~
n cx
q
k(~'oS)
If C is linear, the same is valid also for cx qlin
0
•
Corollary 1.3.22.
",sd
0q = I'
~m
d
sup Ii
operations in IFq
We define families of codes having a polynomial
decoding procedure in a similar way. Decoding of a code C
Chapter 1.3 ASYMPTOTIC PROBLEMS 81
Uq * = {( ~ ,R) I *
0 ~ R ~ cx q (~) }
Moreover, *
CXq(O) *
= 1, CXq(~) =0 for (q - 1) /q ~ ~ ~ 1 ,
and on the segment [0, (q - l)/q] the functions are
decreasing (here the upper * means ei ther or
pol,lin ) .
•
Exercise 1.3.26. Prove the theorem, supposing that we
already know that cxpol,lin(~) > 0 for 0 ~ ~ < (q - l)/q
q
(this is really so, see Theorem 1. 3.31 and Exercise 1. 3.32
below) .
Of course,
CX PqOl,lin(~)
o ~ CXpOl(~)
q U ~ CX q (~)
0 ,
Remark (the
1. 3.33 Blokh-Zyablov bound) . using
generalized concatenation one can do better than in
Theorem 1.3.31. The bound obtained is
dx
o
where
q - 1
-"--q-=-- - 8 ----7 ° ).
•
Corollary 1.3.36. a~OI(8) ~ m~x { ~.a:~I(a·8) },
where max is taken over all q-ary codes ( [n,k,d]q C
being the parameters of C ). The same is valid for
pol,lin
aq (max being taken over all linear codes).
•
Exercise 1.3.37. Prove the theorem and the corollary.
max
q'sq
I · . f (nlog- k)n
~m ~n
q
the limit being taken over all [n,k,dJ -codes with fixed q
lin q
and d. The definition of K-q (d) (for linear codes) is
similar.
Chapter 1.3 ASYMPTOTIC PROBLEMS 87
Theorem 1.3.40.
such a sphere is _~ t
v t = i~O (n)i . (q - 1) i ~ (n)t . (q - 1) t . The
"q(d) ~ f 11
d -2
•
88 CODES Part 1
d - 1
IC q (d) = ICqlin (d) 2
[n = 2 m , k
d - --
= n - m'---
2
2
, d]2-codes
m d - 2
[n = 2 - 1 , k = n - m'---
2-- , d - 1]2-codes.
lim sup d
n
Theorem 1.3.43.
5qlin (k) =
Chapter 1.3 ASYMPTOTIC PROBLEMS 89
•
There are other possible asymptotic questions. Since
n ~ t o , and we have already considered the situations
when either k or d is constant, the only cases left are
those when n, k, d ~ to and either kin ~ 0, or
din ~ 0 (if neither, we come to the main asymptotic
problem). Here are some examples.
dl{>(n) = sup(d)
< min {I - a , q ~ I}
(Hint: The lower estimate is given by the Hamming bound,
the upper by the Gilbert-Varshamov bound and by BCH-codes,
cf. Theorems 1.1.41, 1.1.51, and 1.2.15).
a-I
d
I{> (n) - ~·n
q
ki !: i
k. + d. !: n + 1 - g
~ ~
(n - k i ) + d:
~
!: n + 1 - g
92
NOTES 93
CUR V E S
This part contains the algebraic geometry we need.
Almost all constructions of our book use only algebraic
curves, that is the reason why in this part we concentrate
mostly on the theory of algebraic curves. Multi-dimensional
algebraic-geometric obj ects appear as instruments for the
study of curves. We try to use the possible minimum of
technical devices, and to work mostly with objects lucid for
one's geometric intuition. For example schemes appear rather
late when we are just absolutely unable to avoid them.
Notwithstanding our strife to restrict ourselves to
elementary means, the scope of information we really need
for algebraic-geometric codes is such that this part may
happen to be rather difficult for a non-algebraic-geometric
reader. However we hope that (using may be some extra books
and papers quoted in our bibliographic notes) the reader
with any background will be able to read this part, or at
least sections he needs most.
Chapter 2.1 presents principle definitions and tools
concerning curves (Jacobians included); for curves over the
field C of compex numbers we also discuss the connections
wi th Riemann surfaces. Chapter 2.2 is consentrated around
differential forms and the Riemann-Roch theorem; we also
discuss the Hurwitz formula and some properties of the
cartier operator. In Chapter 2.3 the curves are mostly over
a finite field, we are interested in their points and
98 CURVES Part 2
ALGEBRAIC CURVES
101
102 CURVES Part 2
By IAn
we denote here and below the n-dimensional
affine space over k ; its points P E IAn are n-tuples
P = (x1' ••• 'xn ) , xi E k. The projective space of
dimension n is denoted by IP n its points Q E IP n are
equivalence classes of n-tuples Q = (yo : Y1 : ••• : Yn ) , where
not all of Y.;... vanish, and n-tuple (y : •• • :y)
1 n
is
equivalent to (A·y :A·y : ••• :A·y)
o 1 n
for A E k , A*-O.
Note that there is a natural embedding IAn ~ IP n , given
by (x , ••• ,X) ~ (l:x : ..• :x) Hear and below we use
1 n 1 n
this inclusion of IAn into IP n unless the contrary is
mentioned explicitly. When it is necessary to stress the
dependence of IAn and IP n on k , we write IAn (k) and
IP n (k) •
Sometimes it is convenient to use a coordinateless
definition of IP n which follows. Let V be a vector space
over k of dimension (n + 1) • By IP(V) we denote the set
of lines in V, i.e. the set of equivalence classes {v},
v E V - CO} , v being equivalent to A·V for
A E k - {O}. Fixing a basis in V one obtains
identifications V ~ IA n + 1 , IP(V) ~ IP n .
n
X (P (x, ••• ,x)
1 n
E IA I
O} •
Chapter 2.1 ALGEBRAIC CURVES 103
deg Fi
E
j=O
for i
with
deg F.
~
E F .. (T , •.• , T ) = F. (T , ... , T ) •
j=O ~J 1 n ~ 1 n
2.1.10.
Exercise Show that k[X] is an integral ring
(i.e. that it has no non-trivial divisors of zero).
Exercise Show
2. 1. 15. that a rational map
f : X - - 7 IAn n
defines a rational map f : X -----7 IP . If f
is regular then f is also regular. The converse of the
last statement is false (give a counter-example).
Exercise 2.1.17.
Let If>: keY) '----+k(X) be an
arbi trary embedding over k Show that there exists a
rational dominant map f : X -----7 Y such that If> = f *
•
Corollary 2.1.23. k[X] k for a projective X
•
Corollary 2.1.24. An affine variety of positive
dimension is never isomorphic to a projective variety .
•
In contrast with Corollary 2.1.23 an affine variety is
completely determined by its algebra of regular functions.
Exercise 2.1.27.
Check that Op has a unique maximal
ideal mp formed by f e Op with f (P) = 0 (recall that a
commutative ring with unique maximal ideal is called local) .
2
The factor-group mplmp is an 0plmp-module, Le. it
2 *
is a k-vector space. The space sp = (mplmp) dual to
2
mplmp is called the tangent space to X at P (and mp/mp2
itself is called the co-tangent space to X at P ).
To construct we set
N = (m + l)·(n + 1) - 1 = m"n + m + n . Let Tij be
homogeneous coordinates on rPN for i=O,1, .. ",n and
j O,l, •.. ,m For P = (x :
o :x ) e X
m and
Q rp(P, Q) = (xi"Yj)
from a family
Morphism p: g~X to a family
pi : g' ~ is a regular map
X f : g ~ g' such that
p = pi of and for any P e X the induced map of fibres
f P : lJp ~ lJp is k-linear. Isomorphism of families is
defined in a natural way.
The simplest example of family of vector spaces is the
product X x Am with its natural projection
Px : X x Am ~ X. This family is called trivial. If U
is open in X and p: g ~ X is a family then the
restriction pi : p-l (U) ~ U is also a family of
p-l (U)
vector spaces over U, which is called the restriction of
g to U, and is denoted glu.
Further on we consider only line families (Le. with
dim lJp = 1 for any P eX). A line family p::e ~ X
is called a line bundle on X iff for any P e X there
exists an open set U, P e U eX, such that glu is
isomorphic to the trivial family. We denote line bundles by
:e, M, N etc., the trivial line bundle is denoted by 0. A
regular map s : X ~:e such that pos = id x is called a
section of the bundle p::e ~ X . In particular any line
bundle :e has the zero-section So such that
so(P) = 0 e Yp. The set of sections s : X ~ :e is a
vector space over k with operations
n
Sometimes we call the union X
X. = U
of several
i=1 ~
curves a (reducible) curve; in this case curves are
called irreducible components.
116 CURVES Part 2
aF(p)
ax
aFi
[ax. (P)
1•
) ~=1, ••• ,N
j=1, ... ,n.
c) P = (0,0,0) on X : {X 2
1
2
x2 + x3
2
x1
2
x3 + x2} .
Power series expansions. Non-singular points have the
following important property: a function which is regular at
a non-singular point p has a unique power series expansion
provided that a local parameter tp is fixed.
k k+1
ordp(f) max { k I f E mp f E mp }
2.1.3. Divisors
zeroes and poles and thus ordp(f) '" 0 only for a finite
number of points p eX. Note also that
where
(f)"" = L (-ordp(f»·P
ordp(f) <0
(2.1.2)
=I:a. , p.p
~
i = O,l, ••• ,n ,
and let
-a Q .
where fi = tQ f i By the definition of D any f'.
~
is
regular at Q and there exists io with f'. (Q) '# 0 •
~o
Hence I{>' is regular at Q. It is sufficient to note that
I{> and I{>' coincide as rational maps .
•
Corollary 2.1.58. Let be a rational
map , X and Y being smooth complete curves. Then I{> is
regular .
•
Chapter 2.1 ALGEBRAIC CURVES 125
for P ~ Supp D
(2.1.3)
for P e Supp D
D
s
= pE (ord(s) - ap)'p
DI =Ea .p
U PeU P
D=Ln.·F.
~ ~
2.1.4. 3acobians
and
-1
G ~ G, !peg) g
are regular.
132 CURVES Part 2
ip : X ~ J x , ip P ~P-P
° ° °
is regular for any Po E.X ;
•
The abelian variety JX is called the Jacobian of X
The dimension of the Jacobian is called the genus of X
and is denoted g(X). One can show that this definition
coincides with the definition of genus in terms of
134 CURVES Part 2
N+l N 2
{z = (zo, ••• ,ZN) E A (C) - {O} I L Iz.1 I}
i=O ~
RIEMANN-RoCH THEOREM
141
142 CURVES Part 2
N aFi
L aT .. dt J. o for i 1, .•. , m (2.2.1)
j=l J
dt . f .·dt
J J
~)
0
Uo = 1
{x e IP I u '" O} ,
P·dy
w G' (2.2.2)
x
(m - l)-(m - 2)
g(X) 2
*
If> (w)
148 CURVES Part 2
Check that
Exercise 2.2.9. I(> * (w) is well defined,
Le. it does not depend on the choice of representation
W = L f _-dg _ • Show that if w is regular at Q E Y then
~
the inverse image
~
I(>
* (w) is also regular at any PEl(>
* (Q) •
Automorphisms. An isomorphism g: X ~ X of a
curve X onto itself is called its automorphism. The set of
automorphisms of X is denoted Aut(X), or Autk(X) i f it
is necessary to indicate its dependence on k ; it is clear
that Aut (X) is a group.
Show that
Exercise 2.2.12. Aut «(pI) = PGL(2,k) ,
PGL (2, k) being the factor-group of GL ( 2 , k) over its
center (recall that the center of GL (2, k) consist of
matrices of the form (~~) for a E k* ).
Chapter 2.2 RIEMANN-ROCH THEOREM 149
i(D) = dimkI(D) .
Chapter 2.2 RIEMANN-ROCH THEOREM 151
deg(D') - g + 1 = (deg(D) - g + 1) + 1 •
f L a. ·t i
~
f=-M
w
o
being regular at P . Let us choose another local
parameter u at P and let Resp(w) be the residue of w
at P defined by this choice. By the properties b) and c)
we have
n-1
since char k = 0 the rational function gn = t / (n - 1)
is well defined for n 2:: 2 and we obtain
Resp(dt/t ) = ReSp(dgn ) = 0
n by the property d) • Hence
Resp(w) = Resp(w)
•
Here is the most important result on residues.
o (2.2.5)
•
154 CURVES Part 2
a) deg KX = 2g - 2 ;
deg J!. - 9 + 1
•
Corollary 2.2.25. a) deg X
X
= 2g - 2 ;
t(K - P - QJ ~ t(K) - 1 = g - 1 •
t (P) = 1 + 1 - g + t (K - P) 2
e(K - P - Q) = e(K - P) = g - 1
t(P + Q) 2 - g + 1 + t(K - P - Q) 2
P, pi E X and by Riemann-Roch
t(K - P) 2g - 2 - 1 + 1 - g + t(P) = g - 1
f ® f' ~ f·f'
f' f"
f X~X' ~Y
Kp {a e K I a P e K}
deg Bf =L (e p - 1)
PeX
* e p -1 ep
f (dtQ) = h·ep·t Q ·dtp + tp ·dh
g(X) 1 - n + !. L (e - 1) (2.2.9)
2 PeX P
•
i. e.
By Riemann-Roch
b) 1 is a gap at P ;
a) P is a Weierstrass point;
b) Divisor g.p is special;
g
w(P) = L (a. - i)
i=l ~
point
A P with w(P) g' (g - 1)/2 is called a
hyperelliptic point.
(2.2.10)
Set
C(w) = y p-1 'dt
RA TIONAl POINTS
169
170 CURVES Part 2
s = deg P
s s
V= nV., k·V= U V .•
i=l ~ i=l ~
such that V= K n Op .
172 CURVES Part 2
on X with
deg D •
Conversely, let
Chapter 2.3 RATIONAL POINTS 173
and let ap = a crp for any cr Gal (k/k) Then there exists
a unique k-rational divisor D on X such that D is
obtained from D by the above construction.
Hence we can look at a k-rational divisor on X in two
ways: as at a divisor D on X or as at a divisor D on
X which is invariant under the Galois action. Further on we
identify the divisors D and D and use these two points
of view simultaneously.
s s
g. = L cr.(w.·f) = L cr.(w.)·cr.(f)
J i=l ~ J i=l ~ J ~
co
Z(t) Z(X~t) exp ( E N rot r / r )
r=1
g 2g 2g-1
where Px(t) = q ot + Plot + 00. + P2g_1ot + 1 E ?let]
is a polynomial with integral coefficients. Moreover
1 dS
(s - 1)! - ( l o g Z(t» It=o
dt S
s
Ns qs + 1 - L w.s (2.3.1)
~
i=l
In particular,
Let
Theorem 2.3.15. 11 = I1(X) be the number of line
bundles of degree zero on X (i.e. the cardinality of the
set of ~q-rational points on the Jacobian J x of X). Then
11 = PX (l) = ~(W. - 1) .
j=l J
In particular,
•
The number Nq(g). For any X over ~q the inequality
(2.3.2) implies
Theorem 2.3.16.
b) N 4 (2) = 10 , N 9 (2) = 20 ;
q + 1 + 2 f2Vq1
Chapter 2.3 RATIONAL POINTS 179
if q is special then
Nq (2) r+ 2 rvCfl
q + 2 rvCfl - 1
for
for
2v'q' - r 2 vCfl > v'5' 2 - 1
q = 2 3 4 5 7 8 9 11 13 16 17 19 25
Nq (3) = 7 10 14 16 20 24 28 28 32 38 40 44 56
•
Proofs of these results are rather delicate; they have
two parts: the first gives estimates for Nq(g) which are
based on so called "explicit formulae" for the number of
rational points of a curve over a finite field (cf. Remark
2.3.22) and uses some technique from algebraic number
theory; the second part is a construction of curves with
required number of points and uses various methods of
algebraic geometry.
Z(t) (2.3.5)
co
Z(t) 1 + L R. ·t i
i=l ~
2.3.3. Asymptotics
A(q) ~ f2v'ql
A(q) ~ v'q - 1
hence
m we have
m m
0 s
r
L a. 12 L a.r-s
~
r=O ~
r,s=O
(2.3.7)
m
m + 1 + L (m + 1 - r) . (a ir + a i-r ) .
r=l
m 2g
o s 2g·(m + 1) + 2· L (m + 1 - r)· L a.r s
~
r=l i=l
(2.3.8)
m
s 2g· (m + 1) + 2· L (m + 1 _ r). (qr/2 + q-r/2 - N .q-r/2)
r=l 1 •
Hence
N m m
1 L m + 1 - r -r /2 s 1 + 1 L m + 1 ~ r. (qr/2 + q-r/2) .
g m + 1 .q g r=l m +
r=l
N
1 1
s 1 + c
g
,;q - 1
fN/dl
"/.., c md . t m Then we have an "explicit formula"
m=1
which follows
d'B 'I{J (q
-1/2 )
d d
V-l·e
where cxk = e k and CX t ' ••• ,cx 2gare ordered in such a
way that cx k ' cx k +g = 1 • If c m 2: 0 for any m = 1, ... , N
and f (cp) 2: 0 for any cp then the "explicit formula"
implies that
or for
2
f(cp) (a o + at·cos cp + ... + an·cos(n·cp» '(1 + cos cp)
A(q) = .;q - 1
•
184 CURVES Part 2
b)
A( ) ~ ..fl. (r - 1) - 2l
q l - 1
IXn(~ q) I IXo(~q) I
g(Xn ) - 1 g(X o ) - 1
log H
2ologq(~ - 1) ~ ~ g ~ 2 ologq (~ + 1)
N 1 (X)
lim g(X) A > 0
g~oo
Then we have
Proposition 2.3.26.
lim inf
g(X)~oo
we have
2g
H n(wo - 1)
i=1 ~
Hence
2g
In H goln q + L In(l
i=l
- a ° oq-1/2 )
~
2g
-
00
goln q m -m/2/m
L L aooq
~
i=l m=l
-m/2 2g m
- m=l
00
_q_ _ o
goln q L m L ao
i=l ~
186 CURVES Part 2
2g
- L a.m (2.3.9)
~
i=l
CD -m/2 2g m
L ~. L a i
m=b+l m i=l
Note that
b b 1 b qm
L L L m
m=1 m=l m m=l
and
L q -m1m
q
In q=-r
m=l
CD -m -(b+l) CD _j
L ~ < q L q q - (b+l) < .!._q_
q - l·q g q - 1
m=b+l m j=O
b
L m· q
Nm -m
;:: N .
b
L ~
-m
~-
-m
L~
CD -m)
m m
m=l 1 m=l m=b+l m
Chapter 2.3 RATIONAL POINTS 187
b
~
L.. q -m/2 . (Nm·q -m/2 _ qm/2 _ q-m/2)/m >
m=1
N 1 ·q q
> N ·In -q-- g(q _ 1) - b - In q _ 1
1 q - 1
2g m co -m/2
L (Xi :::s 2g· L -q-- <
i=1 m=b+l m
Hence
In M ?! g'ln q + N
1
'In --q-
q - 1
- o(g) .
logqM
lim g 1 + A'lOgq(q ~ 1)
g-----7CO
lim
g-----7CO
188 CURVES Part 2
lim inf
log M
q ~ 1 + L A olog q
r (m )
g~m g m=l m q qm - 1
r
Moreover, if L moArn!(qm/2 - 1) 1 then
m=l
N 1 (X)
g(X)
..;q - 1
we have
Nr (X) - Ni (X)
lim g(X)
o
g~m
lim o
g~m
So(c+1-S)
Chapter 2.3 RATIONAL POINTS 189
n-1 n oa n + 1 - (n + l)oa n + 1
L (j + 1) oa j (2.3.10)
j=O (a - 1) 2
n-1 n oa n + 1 - (n + 1) oa n + 1
L (n-j)oa- j (2.3011)
j=O a n-1 (a - 1) 2 0
n-1 n+1
L (n - j) oa j
a - (n + 1) °a + n (2.3012)
j=O (a - 1) 2
1 _ a n +1
1 - a
m
L (m + 1 - r) 0 (N oq -r/2 - q r/2 - q -r/2) :s go (m + 1)
r=l r
m m
L (m + 1 - r) oq-r/2 0roB - L (m + 1 - r)0(qr/2 + q-r/2) +
r=l r r=l
(2.3.13)
m
+ N 0 L (m + 1 - r) q -r /2 :s go (m
0 + 1)
1 r=l
B ~ . [(m+2)/2
q S/2 _.;;oq_ _ _ __
s s· (m + 1 - s) (v'q"' _ 1)2
+
(2.3.14)
+ g + m' (v'q"' + 1) .
v'q"' - 1 (v'q"' _ 1)2 1
Setting m c s we end the proof.
•
~
ELLIPTIC CURVES
191
192 CURVES Part 2
bijective.
223
Then the functions 1, x, y, x'y, x , y , x lie in
L(6'Po ) and thus are linearly dependent. Since only yOl
x 3 + a .x 2 + a .x + a (2.4.1)
246
x· (x - 1)· (x - A) (2.4.2)
2
X ·X x . (x - x ). (x - A·X )
o 2 1 1 0 1 0'
where
2
x = x Ix
1 0
, y = x Ix
2 0
• Note that Po = (0:0:1) • Let
Yo = a· (a - 1)· (a - A) • The 1 ine x1 = a· x
0
intersects E
at points Po' Q = (l:a:yo) ,and Q' = (l:a:-yo) • The
points Q and Q' are opposite to each other (i. e.
Q = -Q' ). Now it is easy to describe the addition law in
geometric terms and to write it out. If P1 = (l:X l :Yl) and
P 2 = (1:X 2 :Y2) then Q = P1 + P2 can be obtained as
follows: let Q' (1: X3: y) be the third point of
intersection of the line joining P
1
and P
2
with E then
P 1 +P 2 =Q is the reflection of Q' (see
Fig. 2.1).
Chapter 2.4 ELLIPTIC CURVES 195
Fig. 2.1
r
P3 = Pl + P2 • (Answer:
-Y
x3 - x1 - x2 -
[~: 2
x2
+ A + 1
(2.4.3)
Y3
X 2 "Y l
X
1
-
-
Xl
X
2
"Y 2
_ X
3
" [YX l
1
-~:l ) .
j j (E) (2.4.4)
i\' Ell. = {i\, 1/i\, 1 - i\, 1/(1 - i\), i\/(i\ - 1), (i\ - 1)/i\}
which can be verified by brute force. If we write an
equation of E in the form (2.4.2) then the projection
f(x,y) = x defines a map f E ~ 1P1 of degree 2 with 4
ramification points 0, 1, i\ , and 00 • Let
y2 = x. (x - 1)· (x - i\')
E and E'
b) Let
be elliptic curves, let i\ and
i\' be their Legendre moduli and let j(i\) = j(A') .
Considering i\' as a variable and A as a parameter, we
obtain an equation of degree 6 in i\', vanishing on A.
Therefore it has no other roots and hence E and E' are
isomorphic.
2 3
Y = X + a·x + b (2.4.5)
Chapter 2.4 ELLIPTIC CURVES 199
(2.4.6)
x 3 + a 4 ·x + a 6 (2.4.7)
Then j(E) = a 12
1
/A , where
2.4.3. Isogenies
Sketch of proof:
From the above argument it follows
that the proposition holds for a = 1 . For any a ~ 1 it
can be easily deduced by induction over a .
•
Supersingular curves. For a fixed p there exists
only a finite number of non-isomorphic supersingular curves
in characteristic p. Their moduli can be found out:
o (2.4.8)
•
Therefore all supersingular values of the modulus i\
and of the i-invariant lie in a finite field. Moreover, one
has
a) En~ (E) = 10
b) Endo(E) is a complex quadratic field;
a) End(E) = ~
b) End(E) =+ c·Ok
~ where c E ~ , p~c, Ok being
the maximal order in the complex quadratic field
k = En~ (E) (in this case c is called the conductor of
End (E) ) ;
c) End (E) is a maximal order in the quaternion
algebra End (E) °
•
204 CURVES Part 2
b) For p ~ 2, 3
'2
1 - m·t + q·t
ZE(t) = (1 - t ) ' ( l - qt)
(1) (q ,m) 1 ;
2
(2) q is a square, m = ±2·.;q , and G '" (Z/A) ,
where A = .;q + 1
MF
I = (m B- 1 -A)
1 we have Tr M'F-
"" m , det H'F ""- q ,
a·A =A , d·B =B (mod N) •
7'(z) + (2.4.10)
212 CURVES Part 2
1
'P' (z) =- 2' L
(z - w) 3
weA
Theorem 2.4.34.
weA-{O} weA-{O} •
2.4.35.
Exercise Prove Theorem 2.4.34. (Hint:
Consider expansions of 'P (z) and 'P' (z) into Laurent
series at the origin).
rp : !C ~ 11'2 (!C)
(l:7'(z):7"(Z)) ,if Z E A
z f------7 {
(0:0:1) , if Z E A
•
It is said that elliptic functions uniformize elliptic
curves.
SINGULAR CURVES
215
216 CURVES Part 2
2.5.1. Normalization
Set
(Hint: Op 2 k + c p ).
Chapter 2.5 SINGULAR CURVES 219
•
Exercise 2.5.14. a) Let be defined by
Y
223
X=+ x for char (k) '" 2 and let P = ( 0 , 0) • Show
that XV = fAl and that V can be given by
vet) (t 2 - I, t- (t 2 - 1» t being a coordinate on
fAl. Show that
S(P) {Q, Q' } where Q = {t = I} ,
Q' = {t' = -I} e fAl • Show that Op consists of quotients
P(t)/Q(t) where Q(t) is divisible neither by (t + 1)
2
nor by (t - 1) and that C p = (t - l)-Op,Op=k+c p '
2
be defined by Y = x3
p = (0,0) Show that then vet)
S(P) =Q= {t = O} ,
D = L L n .Q
Q
P QeS(P)
(m - 1) (m - 2) _
g 2 L ~p (2.5.1)
p
g(X/)
Proposition 2.5.20.
X . In fact one can notice that this definition does not use
Chapter 2.5 SINGULAR CURVES 223
G = A'H + B'F
•
In other words, the condition (G) ;e (H) + D is
sufficient for the divisibility of G by 11 in the
homogeneous coordinate ring RX = k[XO ,X1 ,X2 ]/(F) of X' ,
where G and 11 are the images of G and H in RX
224 CURVES Part 2
deg D = (r - l)·(r - 2) - 2g
•
We also need expressions for the divisors (dx) and
(G~) of the differential form dx and of the function
G~, respectively. Let Bx (respectively By ) be the
ramification divisor of the map X ---+ 1P1 defined on X'
by (x,y) ~ x (respectively by (x,y) ~ y ).
Proposition 2.5.25.
(dx) = Bx - 2'(X)m
2.5.4. Desinqularization
(2.5.2)
-1
u IL : (x:x:x)
012
t------+ (x :x :x ;(3:a)
012
0-- 1 (X - (P })
o
n U2 is also non-singular. Therefore the map
o-I X Xl ~ X is the normalization map.
1
(2.5.4)
where the sum is taken over the set of all infinitely close
points of X.
•
In fact the techique of monoidal trans.formations makes
it possible to desingularize a plane curve efficiently. A
desingularization algorithm will be given below in Part 4
230 CURVES Part 2
if X1.. is smooth at Qs
otherwise
a
Ps(xs,xs·t s ) = xss'P~(Xs,ts) and xskP~(xs,ts)
(2.5.5)
(2.5.6)
(2.5.7)
232 CURVES Part 2
aPt aPt
~t °
if -a-(O,O) -a-(O,O)
xt Yt
At (2.5.8)
, otherwise
{:
aPt aPt
At
if ax (0,0)
t
aY t (0,0)
° (2.5" 12)
, otherwise
233
234 CURVES Part 2
rp * Spec A ~ Spec B
its closure •
For example, if A is a domain then (0) is a prime
ideal whose closure coincides with Spec A .
2
ep = Homk(p) (PIP ,k(P»
Chapter 2.6 REDUCTIONS AND SCHEMES 239
dim X = dim A
any i .
2.6.12.
Exercise Similarly to Exercise 2.6.10 define
the sheaf of rational functions on a variety X. Show that
this sheaf is the constant sheaf with the stalk A = k(X) .
Note that all the groups O(U) are rings and all the
maps ~UV are ring homomorphisms. A presheaf (respectively,
a sheaf) possessing this property is called a presheaf
(respectively, a sheaf) of rings.
244 CURVES Part 2
(Spec B , 0) - - + (Spec A , 0) ;
2.6.4. Schemes
M
k . The first property means that t(X) = U Spec A. where
i=l ~
A. is
a finitely generated k-algebra. To explain the
~
2.6.30.
Exercise Show that the above definitions do
not depend on the choice of U. (Hint: In fact they depend
only on 0x,p).
2.6.31.
Exercise Check that dim X is well-defined,
i.e. that it does not depend on the choice of U.
b) x = U. t/I.~ (X.)
~
;
~
c) t/I.(u .. )
~~)
= t/I.(X.)
~ ~
n t/I).(X).)
let
aXjS
F(XjS) ) G(XjS)
IF(f l IG(f)
ayjS
F(Y jS) ) G(Y jS)
Chapter 2.6 REDUCTIONS AND SCHEMES 255
256
NOTES 257
A G - COD E S
This part is the central one. At last we come to
algebraic-geometric constructions of codes. Besides the
striking link between two disciplines so far from each
other, algebraic-geometric codes (which we abbreviate as
AG-codes) have very good parameters, if of course we choose
the varieties they come from in a proper way. Their
advantages become obvious when we consider asymptotic
problems. The best studied AG-codes are those from algebraic
curves. The main algebraic-geometric hero is a modular
curve, in this part we only list its properties, leaving the
detailed analysis for Part 4. Almost all coding corollaries
are discussed in this part.
In Chapter 3.1 we give the constructions and study some
properties of AG-codes. Chapter 3.2 presents some examples.
In particular we discuss codes from curves of small genera;
the case studied best of all is that of elliptic curves.
Chapter 3.3 is totally devoted to the problem of decoding
AG-codes. In Chapter 3.4 we collect the fruit, applying
AG-codes to asymptotic problems. Algebraic-geometric codes
ameliorate almost every lower bound for the main
asymptotics: for linear and for non-linear codes (for q
large enough), for polynomially constructable codes (for any
q), for polynomially decodable codes, for self-dual codes,
for constant-weight codes, etc. As usual we end the part
with historical and bibliographic notes.
264 AG-CODES Part 3
3.1.1. AG-constructions
3.1.2. Some additional remarks
3 . 1. 3·. Dual i ty and spectra
265
266 AG-CODES Part 3
3.1.1. AG-constructions
C = (X,'P,D) L •
k = t (D) ,
Proof: Let D = D1 - D2 D1 ~ 0 , D2 ~ O. A
non-zero function f e L(D) has at most a 1 = deg D1
poles and at least a 2 = deg D2 zeroes in Supp D, since
D + (f)o - (f)oo ~ 0 • Hence the number of its zeroes out of
Supp D is at most - a = a . We have supposed that
2
a < n , therefore we EvT(f) * 0 for any function
f * 0 , i.e. EvT is any embedding. Moreover EvT(f) has
at least (n - a) non-zero coordinates, thus
d ~ n - a .
k = t(D) ~ a - g + 1
The map
C = (X,1',D)O •
270 AG-CODES Part 3
(In some papers on the subj ect the code (X," ,D) n is
denoted by C* (G,D), where G is our D and D is our P).
Let us estimate the parameters.
k;!:n-a+g-1
d ;!: a - 2g + 2
n(p - D) ~ L(K + P - D)
~ (2g - 2 + n - a) - g + 1 =n - a + g - 1 ,
since
deg (K + P - D) 2g - 2 + n - a •
deg(w)
...
= deg(w)
0
- 2g + 2 ~ a - 2g + 2 + deg D2 ,
n - g + 1
this problem.
S
choose a local parameter t. . If D" = L b. 'Q. then for
~
i=l ~ ~
b.
t. ~·f
any f e L (D) the function
~
is regular at Q.
~ .
Consider the map
where {P l , • • • ,Pr } = 'P - Supp D" • The code C' = Ev' (L(D»,
which we denote by (X,'P,D)', is a lengthening of (X,'P' ,D)L
(where 'P' = 'P - Supp D" ) by s positions corresponding to
the points of Supp D" = Supp D n 'P • The parameters of C'
also satisfy the estimates of Theorem 3.1.1.
In fact the described construction of lengthening can
be changed for the following much more conceptual one, which
is valid for any variety X.
n
HO(f) ~ e '!P.
i=l ~
_ _ _ _~) F n
Germ'P q
~ n!_
s :e p .
i=1 1.
k~a-g+1
k + d =n - g + 1
k' ~ a - g + 1 = k ,
d' ~ n - a
d=n-a
i. e.
k + d n - g + 1
•
Remark 3.1. 12. strictly speaking the code C = (X, <J>,!£) H
depends on the choice of trivializations, and also (just as
the codes (X,<J>,D)L and (X,<J>,D)n) on the ordering of
<J> . However any other choice of trivialization corresponds
to the multiplication of i-th basis vector of IF n by a
non-zero constant
*
y. elF,
q
and any other choice of
1. q
ordering of <J> to some permutation of the coordinates, i.e.
both choices lead to an equivalent code (see section
Chapter 3.1 CONSTRUCTIONS AND PROPERTIES 275
Exercise 3 .1.16.
Prove that the projective systems
(X, "P, f) H and (X, "P, f) p are isomorphic, L e. that the
corresponding codes are equivalent.
n . qk-1. (q - 1)
n-a=d ~d~
c qk _ 1
Thus
n ~ a·
else .
Chapter 3.1 CONSTRUCTIONS AND PROPERTIES 281
As usual,
a = deg D = L a i deg Pi' 0 n = 11>1 0
m
k ;: n - L (a; - (3 0 + W 0 ) 0 deg Po + g - 1 ,
i=l .... ~ ~ ~
m
d ;: a + Leo deg P 0 0 - 2g + 2
i=l ~ ~ •
k ;: n + g - 1 - a - (3 - w
m
Exercise 3.1.27. Investigate the case D = a L ~i(Q),
o
i=l
where Q is a point of X of degree m, and ~ is a
282 AG-CODES Part 3
m .
where R = LUI(Q) ).
i=1
2 n (P - a1 2 ••• 2 n (P - D + R ) 2 n (P - D)
.R - R2 )1 m
b.
I .
where Ri = L u~ (P.) for a point Pi of degree bi ' and
j=1 I I
Exercise 3.1.29.
Let X be a hyperelliptic curve such
that no hyperelliptic point of it lie in X(F q) , and let
2g+2
D = L a i ·Qi be an Fq-divisor. write out the estimates
i=1
for k and d given by Proposition 3.1.24.
Therefore a (f (P ) , ••• ,f (P »
code vector is mapped
to
*
(g (f)
(P 1
1 *
(Pn » = ( f(g(P 1
),··. ,g (f)
n
,f(g(Pn »,··· ») ,
which is also a code vector. Thus G is naturally mapped to
Aut(C) n Sn and, according to the properties of group codes
(cf. section 1.2.2), we have C ~ ~q[G/H1]@ ••• @~q[G/Hm]
where Hi is the stabilizer of a point Qi e ~ • We obtain
the parameters
(b - g) 'm + (m - 1)
[n , ~ n - a·(q - l)/q - 1 , ~ a - 2g + 2] q
where n = I~ I , a = deg D
•
Remark 3.1.42. Using a subtler technique of linear
algebra it is possible to show that the statement of Theorem
3.1.40 holds in the following more general situation. For
an arbitrary divisor DIet Dl be such that D ~ q'D 1 '
then
L ResQw o.
QeSupp(w) co
since all the poles of the form f·w lie in l' (possible
poles of f are killed by zeros of w ). Thus any code
vector of CL is orthogonal to any code vector of
(v,v') = ri t.(s.)·t'.(s~)
~ ~ ~ ~
=
(i.e. if
and v are
u y-orthogonal). A code is called
quasi-self-dual if there exists such y and self-dual if
y = (1, ••• ,1).
an n
[ n, 2"' ~_n
~ 2 - g + 1] q - co d e C = (X, ~ ,D) L is quasi-self-
dual. Moreover, there exists a unique (up to a
multiplicative constant) form W
o
E n(p - 2D) such that
the code C is quasi-self-dual with respect to
y = (Yl' ••• ' Y n ) , where y.
~
= Res p i (w)
0
* O. In particular,
n n
L f(P.)·g(P.)·y. L Resp w o ,
i=l ~ ~ ~ i=1 i
N
P l + ••• + P r L e i . Pi (mod 2) ,
i=r+1
where e.
~
= 0 or 1 . Reorder P r+1' ... 'PN so that e. = 1
~
e. = 0
~
for i > s Then P +
1
+ P S :; 0 (mod 2) ,
S O!: r O!: N - 2g - 1 , and we can take "P = {Pl,···,Ps } Of
course s is always even since a divisor of an odd degree
is never zero modulo 2 •
•
Corollary 3.1.49. In the notation of Theorem 3.1.48
there exists a quasi-self-dual (self-dual if q is even)
n n
En, 2' O!: 2 - g + l]q-Code with some even n O!: N - 2g - 1
•
Remark 3.1.50. The sufficient condition of Theorem
3.1.46 is not in general necessary. Consider an elliptic
curve y2 + Y = x 3 • This curve has 3 points over IF 2 '
i. e. IX(IF 2 ) I = 2 + 1 - (w + w) = 3, hence W = ±v'2"i ,
and therefore IX(1F 4 ) I = 4 + 1 - (w + w )
2 2 9 Theorem
2.4.31 shows that X(IF) '" (71./3)2. Let P be the
4 o
infinite point, P (0,0) P 2 = (e,l), P = (e,e) ,
P = (e, ( ) ,
2 where
1
is a generator of
*
IF 4
3
The code
4
CL = (X, {P , ••• , P }, 2P ) L is a [4,2,3] 4 -code, and the dual
140
Chapter 3.1 CONSTRUCTIONS AND PROPERTIES 293
n n-r r
n-v(v) v(v)
L x 'y L Ar'x .y
vee r=d
n-t
Bt = L (n~j)'Aj ,
j=n-a
where we put A
1 Ad - 1 = 0 .
Chapter 3.1 CONSTRUCTIONS AND PROPERTIES 295
a t
L Bt' (x - 1)
t=O
Bt = (~), (qa-t-g+1 - 1) ,
and for a 2: t 2: a - 2g + 2
(~),(qr(a-t)/21+1_1) 2:Bt2:max{o,(~),(qa-t-g+l_l)}.
k+d2:n-g+1,
(n - k) + d~ 2: n - g + 1
a - t - g + 1 ::s (a - t) /2 + 1 .
EXAMPLES
297
298 AG-CODES Part 3
n :s q + 1
d = l , ... ,n
k=n+1-d
k = 1, ... , n - 1 ,
+ 1 , else .
300 AG-CODES Part 3
k=1, ••• , n - 2
Nq (2} =
r+ 20
10
4Vq"" + 1 for
for
for
q *" 4, 9
q= 9
q= 4
q + 2· f2v'q'l + 1 if q is non-special,
k+d~n-2
n :S Nq(3), k = 1 , ... , n - 3
Chapter 3.2 EXAMPLES 301
q 2 3 4 5 7 8 9 11 13 16 17 19 25
7 10 14 16 20 24 28 28 32 38 40 44 56
3.2.5.
Example Consider a curve X of genus 3 over
~a with 24 points (cf. Example 3.2.58) and a bundle of
degree 11 . We obtain a [24,9,13]a-code c. Concatenating
it with the parity-check [4,3,2]2-code we get a
302 AG-CODES Part 3
x n + k-1
L (n) o(qk-i
0
k
-1)o(X-1)~+Bko(X-l),
0
i=O ~
k + d ~ n
Chapter 3.2 EXAMPLES 303
~ ~ ~
L (qe(Q) - 1)
Qc'P, I Q I =k
Bk = (q - l)·M
a
Proof: Let h = L h_ be a representation of h as a
i=l ~
sum of different elements, then for any h o E H there is a
a
representation h + a-h = L (h + h) and all the elements
o i=l 0
also different. Therefore the numbers of
representations of hand (h + a-ho) are equal .
•
Corollary 3.2. 8. If a and the order N of Hare
co-prime, then for any h E H
M(H,h,a)
k-l
L (~)- (qk-i - l)-(X - l)i +
i=O
Let
e (h)
fo(h) n p p
piN
cS
Let t IN and t r-Tp p . Define an integer
p
f. mi n ( 0 , ex .)
jEG P J
=n p p
piN
M(H,a) ~. [ L
ml (a ,N)
( NI m) .tJ.(m).(-1) aim _
aim
-
r
2 H
ml (a/2,N/2)
L (N/2m)
a/2m ·tJ.(m) 1 •
1 N ) a-aim
M(H,a) ( 1m .tJ.(m).(-1)
If L aim
ml (a,N)
•
Exercise 3.2.13. Prove the theorem.
308 AG-CODES Part 3
1 [
~"L (NI m) "fJ.(m) " (-1) a-aim -
Bk ,m~n
" (N,a)
N ml(a,N) aim
- 4"
ml (a/2,NI2)
L ( NI2m) "fJ.(m)
a/2m
1 '
N ) a-aim
Bk ,m~n
" (N,a) rT -
~" 1 [ L ( 1m "fJ.(m)
aim
"(-I) -
N m I (a ,N)
- 2"
ml (a/2,NI2)
L ( NI2m)
a/2m "fJ. (m) 1 ,
Bk ,m~n
" (N,a)
q - 1
--N-"
~
L.. (Nlm)
aim
"fJ. (m)
ml (a,N)
•
Exercise 3.2.15. Prove the theorem.
Chapter 3.2 EXAMPLES 309
xn +
k-1
L
i=O
(l!)~ ° (qk-~
°
- 1) ° (x _ 1) i +
k+d~n-1
Bk - 1 - (k~l)o(q - 1)
B' q
n-k+1
B'n-k
B'
n-k-1
For
elliptic codes constructed from proper subsets
~ c X(~ q} the answer is not 50 nice. However it is still
possible to prove the negative answer to the above question.
First of all note that for q :s 11 there is no
non-trivial MDS-code longer than q + 1 (cf. section 1.2.1,
the remark after Problem 1.1.12).
For the rest values of q we have
representation.
Proof: Let
{h E ~ -h E ~, h - -h} •
I~ *I ~ n - 4 - (N - n) = = 2n - 4 - N .
R ~* - {±h")
~
"-1 , ... ,s+t
~-
IRI ~ 2n - 4 - N - 2s - 2t = 2n - 4 - s - a .
Lemma 3.2.24. If
n < IN 2 2
N - - --3 - 3 (N
A (N) 1N -
3
2 2
- 3(N - 1)
r 1
i1! N -
3
2 r
Since q + r2VqJ + 1 ~ N > n ~ q + 2 we have
N - n :s r2VqJ - 1
2~ - 1 ~ r2~ - 11 ~ N - n ~ A(N) ~
N - 2 ~ q + 1
3 3
q the inequalities
n 2: q + 2 , N::s q + r 2Vq 1 + 1 , N - n 2: 1N ; 2 r
are contradictory, i.e. the condition of the lemma for
(N - n) is always satisfied (for q 2: 13).
2· r2Vq'l + 8 2: 2' (N - n) + 10 ~ n ~ q + 2
max IX(IF ) I
X q
2
over ~q' where q r , r pm being an arbitrary power
of a prime.
where a,b e ~
q satisfy (The
existence of such b follows from the fact that ~* is
2 q
a cyclic group of order r - 1 . The existence of such a
is also easy to prove).
since
r2 + 1 + 2' gr' r
(u - a) P - r'Q
a,~
(r + l)'P o ,~ - (r + l)·Q if
(v - ~) {
L P - (r + l)·Q if
ar+1=~r+~ a,~
(w) = (r 3 + r2 - r - 2)·Q - D
Chapter 3.2 EXAMPLES 319
uu u + '1 uv v + '1r·u + 0
'P = Supp P = {P
a, {3 } being the set of affine
IFq-points of X~.
3 2
In particular if q is even and m (r +r -r-2)/2 ,
the code Cm is self-dual.
To calculate k m = dim Cm we need the following
i·r + j·(r + 1) s m}
i ~ 0 , 0 ~ j ~ r - 1 and let
m t'r + s , 0 ~ s ~ r - 1 .
Set
2
v(m) if m~ r - r - 2
km m + 1 - r2 -2 r if r
2
- r - 2 < m < r
3
3
r3 - v(r 3 +r 2 -r-2-m) if m ~ r
i'r + j·(r + 1) ~ m , 0 ~ i , 0 ~ j ~ r - 1
3
Proposition 3.2.46. Let m= i -r + j- (r + 1) :$ r - 1
for sOme i ~ 0 and 0 :$ j :$ r - 1 I f either j = 0 or
3
i :$ r - r
3
2
(in particular, if m :$ r 3 2
r ) - then
d = r - m
m
u G ~
IFq v Gr ----7 IF q
r
qm _ 1
IH(lF q ) I
q - 1 '
djn
m _ 1
q ---=-
~ r· ....
q - 1
m-1
max{ IS(IF q> I} r.qm-1 + q - 1
q - 1
S
m+1
[ qq -
- 1
1 , (m;r) , q
m- (r - 1)·qm-l ]
q
Chapter 3.2 EXAMPLES 325
[ n, m + 2, n - r·qm-1
qm _ 1
q - 1 ]
q
.
x.t - Z2 = 0
{
3
Y -y·t 2 - x2· z 0
Chapter 3.2 EXAMPLES 329
DECODING
331
332 AG-CODES Part 3
Consider an
AG-code C = (X,:P,D)n over If
deg D = a , I:PI = n , 2g - 2 < a s n + g - 1 , its
designed parameters are
k
c
= n - a + g - 1
dc =a - 2g + 2
s (v, f) = L e.' f (P .) •
ieI ~ ~
Let t = III .
Consider an auxiliary divisor D' of degree b such
that Supp D' n :P = fZJ By .
gl,···,gl we denote a basis of
L (D') , and by h 1 , • • • , hr a basis of L(D - D') . Of course
gi·hj e L(D) Let .
s ..
~J
Chapter 3.3 DECODING 333
o j 1, ..• , r (3.3.1)
Proposition 3.3.1. If
leD') >t
a - b > t + 2g - 2
then for any solution y= (Y1' ••• 'Yl) of this system, the
function
l
g = E y . . g.
Y i=1 ~ ~
since
deg (D - D' - L P.) ~ 2g - 1 •
ieI ~
t(D-D') a - b - g + 1 .
L f.(P.)·Z. (3.3.2)
ieI J ~ ~
Y
Proposition 3.3.2. If
a - b > 2g - 2
L f. (P.) ·x. 0,
ieI y J ~ ~
for which g
y (P.)
~
= 0 •
a - b > t + 2g - 2
t :S t
o
max {min (leD') , a - deg D' - 2g + 2}} - 1
D'
errors, the maximum being taken over all divisors D' such
that Supp D' nT =
•
(lJ
Proposition 3.3.6.
Hence
to ~ a - b - 2g + 1 a - to - 3g + 1 ,
i. e.
2to ~ a - 3g + 1
•
Exercise 3.3.7. Check that if the bases of L (D) ,
L (D' ) , and L (D - D') are already given then the basic
algorithm uses at most con 3 operations in the field IF q ,
where the constant c does not depend on q. (Hint: In
2
fact it needs at most c«d + g2)on) operations, which is
better than c n o 3 when q grows. To prove this just note
that a system of m linear equations in n variables needs
2
O(m on) operations in IFq ).
t(K - E - (a l - 2b l + 1) 'H) 0,
and therefore
t(K - (a
1
- 2b 1 + 1) 'H) ::s deg E
t(K - (a l - 2b l + 1) 'H)
t( (a
1
- 2b
1
+ 1) 'H) - a ·h
1
+ 2b 1 ·h - h + g - 1 •
340 AG-CODES Part 3
~ f_(_a_l_-_2_b_l_+----:~=-)-·-h-+-h-+-11 _ S (H)
__ fal'h - 2g + 11
2 + g - bl·h + h - S(H)
We get
~ t + 2g - 1 - a ·h + b ·h
1 1
deg E + 1 .
which is contradictory •
•
This proposition and Exercise 3.3.8 show that when t
satisfies the condition of Proposition 3.3.9 we have
v L(b ·H -
1
t !!O r dc ~ 1 1- S(H)
S(H) !: fh ; 11
and that if either h or r ~ 11
g is even then
S(H) S r h1 -
g ; 1 ,
S (H) S r +~ -
g 11 .
(Hint: To prove the second inequality use the fact that
S(H) is the maximum of the difference of values at integers
of two functions. One of them is almost linear, and the
other function l(i'H) has a polygon lower bound consisting
of the horizontal line l(i'H) ~ 1 and the line given by
the Riemann-Roch theorem. Maximum is attained at their
meeting point).
can be decoded up to r dc ~ 1 1 In
particular decoding exists for any curve X of genus g s 2
(for an appropriate choice of the divisor), since any curve
of genus two is hyperelliptic.
S(H) r (h -8 1)2 + 1
-2- 1 •
tends to ~ when m ~ ~ ).
S (P) =m'(m 8 - 4) + 1 .
1
(Hint: Consider the map Tl: X ~ IP, Tl (x:Y: z) = (Y:x - z)
correct
II de errors.
AG-code up to ~ dc ~ 1 1.
Remark 3.3.23. Problem 3.3.22 being open, there are
however some results which can be summed up as follows.
346 AG-CODES Part 3
-
Exercise 3.3.25. write out a decoding algorithm up to
~dC - ~ - 11
for q-ary restriction of c/'-ary AG-codes (m
being fixed), which needs at most c N3 operations in IFq o
2) k(A) ~ t ;
3) d (Bl.) > t ;
ASYMPTOTIC RESULTS
349
350 AG-CODES Part 3
gi
lim inf < 1
Ni
gi - 1
Le. Ri =l- N. - o ..~
Passing to the limit over i, we
~
see that the points (oi,R i ) are dense on the line segment
R 1 - '1 - 0,
•
Consider the quantity
N
A(q) = lim sup g
1
A(q)
Corollary 3.4.2.
•
Recall (cf. Theorem 2.3.22) that
1 - (v'q" - 1) -1 - 0
H
q
(8) - 8 = (vq - 1)-1
R = 1 - (logq(2 q - 1) - 1) - 8
C
a
(1') { D e Div+(X) deg D a , Supp D n X(lFq) ~ l' }
C
a,r { D e Div+(X) I deg D = a , ISupp D n X(lFq) I = r }
By J
a,d
we denote the set of line bundles !£ e Pic(X)
such deg !£ = a
that and the minimum distance of
C = (X,X(lFq)'!£)H equals exactly d •
ex Ja d , '----+ Ca , n-d
In particular
Proof: Let !£ e J
a,
d. Then there exists a section
s e HO(!£) such that the divisor of its zeros satisfies
n - d
a. - L
lim inf
1. 1.
> a
gi
t.
i\ lim
1.
g. > a
gi~<XI 1.
Then
where
a"H (L/a.)
1. 21. 1.
for t ./a.1. > (q-l) /q
1.
,
f {
a. It.
1. 1. a.1. 'log 2 q - t.1. -log2 (q-l)
356 AG-CODES Part 3
(3.4.1)
(3.4.2)
So (c + 1 - s)
Chapter 3.4 ASYMPTOTIC RESULTS 357
(we have
s·t
s
Thus
d :5
a ,t
(3.4.3)
a-e
+ max { L
s=l
lOg2(t s + b s -
bs
1)}
the maximum being taken over the same sequences
(b 1 ,b2 , ••• ) , and pen) being the number of partitions, the
well known number theory function satisfying the classical
inequality
:s log2 ( t s + b s ) :s
bs
+ 0 (g) •
a - l} + 0 (g) ,
t5
attained for where is given by the
equation
a-t 5·t 5
L a - t .
5=1 11 5 - 1
a-e a-e
f
a, e
5=1
L t 5 .log 2 (
11
5
11
5
- J + L x 5 'log2 (JJ.5)
5=1
a-t
5=1
L t 5 .log 2 (
11
5
11
5
- J + (a - t)· log 11
2
(a - t) . t a-t
a-e
11 - 1
a-e
i. e. :S 11 - 1 , hence
c 9 g.q S/2
1
L
In 2 s=2
c
L + o(g) :s
s=2 s· (c + 1 - s)·q S
9g c
L q-S/2 + O(g) = O(g)
c·ln 2 s=2
4 a-t a -- t + o (g)
:s L (q//J.)S + 0 (g) :s const·-c o (g)
c·ln 2 s=c+1
c s·t S c 9g·qS/2· S
L S
:s L :s
s=2 /J. - 1 s=2 s· (c + 1 - s)
9g c
:s -c - ·L s (vq//J.)s + o (g) = o(g)
s=2
Chapter 3.4 ASYMPTOTIC RESULTS 361
<Xl
Therefore
a-l s·t s
~ - 1
(a - l) - L a - l - 0 (g)
s=2
L e.
~ a: l + 0 (g)
a'H (lla)
2
+ o(g).
l 1
> 1 -
a q
f
a ,l (a - l) '10g 2q + l'10g2 q q 1 + o(g)
t
In this case (since a - t ~ q-=-r + o(g) ) we get
1
tla :5 1 -
q
1
Note also that for tla 1 -
q the formula for a)
and for b) coincide.
IX. (~ )I
such that lim ~ q = ~ - 1 and let be the
gi
lower asymptotic bound for linear codes of the form
c (X.,X.(~
~ ~ q )'£)H . Then
a) If q < 49 , then
H
q
(8) + -q-.
q - 1
(1 - 0) 1 + a
and 0'
2
and 8' be the roots of
3
1 + a
Then
where
for 8 e [0 o']U[8' q - 1]
'1 4' q
The
behaviour of the expurgation bound Ry for
2m
q =p ~ 49 is shown on Fig.3.1 on the next page. Cf. also
Diagrams A.2.2 and Tables A.2.4 of Appendix.
364 AG-CODES Part 3
Fig.3.1
-1
k ~ a - g + 1 ~ ~ -R-g
kin ~ R
5 = din ~ 1 - R - ~
-1
where de = n- (5 - e), 5 = RV (R). Indeed from this
inequality and Lemma 3.4.8 one can see that the number of
line bundles of degree a such that the minimum distance is
at most de is less than the total number of f'q-bUndles of
degree a:
E 13 I:s (3.4.5)
d:Sd a,d
e
366 AG-CODES Part 3
(3.4.6)
max
d~d
{(~ - l)·H (1 -
2
~)
n
+ fig}
a,n-d
.
e
Let
for
Chapter 3.4 ASYMPTOTIC RESULTS 367
and that
-1 -1
fey) + (1' 'R + l' . (l-y) 'log2(q-l) ,
for
1 - Y 1
R + X :s 1 - q
log
1 - y
--- - log2
:..:R~+-:--=:1'_---....:l=--+~yL.
2 Y 1 - Y
for
1 ~ 1 - Y ~ 1 _ 1
R + l' q
and
log 1 - Y + 1 og2 (q - 1)
2 Y
for
y:s 1 - (R + 'Y)
• 3 q- 1 < -=--~l=--:--_
l+R+1'
max f(d/n)
d:sd
c
368 AG-CODES Part 3
-1
f ( eS) :5 1 + '1 . (1 - log 2 (q - 1» (3.4.7)
For 1 ~ 1 - eS ~ 1 _ 1 we have
R + '1 q
- '1 - 1 + log2(q - 1» =
1 + '1 (3.4.8)
1 - (R + D) :5 IS :5 1 - q - 1. (R + D) . (3.4.9)
q
1 - eS 1
For :5 1 - we get
R + '1 q
-1
- '1 - 1 + 1 og2 (q - 1» = '1 . (log q) . (H
2 q ( eS) + R - 1) .
Chapter 3.4 ASYMPTOTIC RESULTS 369
R = 1 - Hq (0) (3.4.10)
o ~ 1 - q - l·(R
q
+ ~) (3.4.11)
1 - q - 1. (R + ~)
q
•
The bound Ry is the best estimate known for a~in
(but not for a q , cf. section 3.4.4).
1 - (VcI' - 1) -1 - 8
•
To prove this theorem it suffices to give a polynomial
algorithm of construction for AG-codes on some family of
Chapter 3.4 ASYMPTOTIC RESULTS 371
Vq"' - 1
•
Exercise 3.4.17. Prove the theorem.
Unfortunately we do
not know the precise value of
RKTV(lin) (0) • The reason is, of course, that we do not
know the parameters of all q-ary codes. However, each code
372 AG-CODES Part 3
(3.4.12 )
1 - R - - 2a lO g q a
o
RGV(~q-1 _ x) _2(q -
q2
1) .In q·x
2 for x ------+ 0
q _
RBz (--q- -
1)
x - 6 (q
q3
2 ·X
3
- 1) ·In q
2m· (q - 1)
R.rcr(c5) = mmax { 1 - -- -
q. (ct'/2- 1)
Sketch of proof:
Let ct' be an even power of a prime.
Consider a family of curves X of growing genus g over
F ct' such that the ratio of the number of Fqm-Points to the
genus is asymptotically maximal, i.e.
lim sup ~
g
= ct'/2 - 1
R - 1 - m' (q - 1)·(3 ,
15 - q' (3 - 2'a' ,
•
Chapter 3.4 ASYMPTOTIC RESULTS 377
RO
LT
(0) = max {(lOg
M
p2m).(pm - 2)·2m
(pm _ 1).n
_ 2~ .0J}
here the maximum is taken over all [n,k,d] 2m-codes, and a
•
Exercise 3.4.29. Prove the theorem. (Hint: Apply
Exercise 1.3.38 on the alphabet extension to the bound RKTV
for pol)
<Xp2m .
_pol(~)
.... q 0
> R GV (~)
0 •
<X1(0)~1- 1 -c5
q Vii" 1
378 AG-CODES Part 3
50 = (q - 1) 1 (q' + q - 1)
(
1 q - 1
1 - R _ 1 + q' + q
The curve RGy (5) and the line RLZ (5) intersect in
two points iff R LZ (5 0 ) > R Gy (5 0 ) , which is clearly
equivalent to
i.e.
(3.4.13)
m m m
l!: q > 22m'2 1(2 -1) _ 22m + 1 = 22m. (22m/(2 -1) - 1) + 1
Chapter 3.4 ASYMPTOTIC RESULTS 379
where
R" (0)
LT
= max (R(q') (0) -log (q')}
V q
2
Exercise 3.4.32. Check that for q = 256 there exist
.; such that (Hint: Put .;
o
= 0.0146
o
Chapter 3.4 ASYMPTOTIC RESULTS 381
Hq(8) + (1 - 8)'logq(q - 1) = 1 + ~q
for q 256 2 ) .
Prove that
~PqOl
.... dec,lin(~)
u ~ RZ ( 0 )
2
1 - - 0
v'q' - 1
•
Exercise 3.4.35. Prove the theorem. (Hint: It follows
from corollary 3.3.4 and Proposition 3.3.6).
•
Chapter 3.4 ASYMPTOTIC RESULTS 383
3m· (q - 1)
q. (qm/2 _ 1
•
Exercise 3.4.39. Prove Theorems 3.4.37 and 3.4.38.
(Hint: Similar to the proofs of Theorems 3.4.16 and 3.4.23,
taking into account Proposition 3.3.6).
osd
q
= lim sup 0
o~d ~ 0Sch
•
Exercise 3.4.43. Prove the theorem. (Hint: Use
Corollary 3.1.49 for a family of curves with
N/g - ,;q' - 1 ).
Prove that
-(w - -2-
<3 w'v'W"' ) ·log w
1 - v'W"' 2
•
Exercise3.4.48. Prove the theorem. (Hint: Apply
Exercise 3.4.46 to AG-codes). Show that for small enough w
the bound of Theorem 3.4.47 is better (for 0 in some
interval) than that of Exercise 3.4.45.
386
NOTES 387
395
396 MODULAR CODES Part 4
H ={ z eel Im z > 0 } .
a·z + b
'¥ z 1------+ c.z + d '
c = 0 (mod N)} .
Sometimes to define r(N) and ro(N) one writes them in
the following form:
r(1)/r(N) ~ PSL(2,7lIN) ,
~l(~/N) being the projective line over the ring ~/N, i.e.
the set of one-dimensional free ~/N-submodules of (~/N)2.
{ -N3 ·n (1 -
-2
P ) for N ~ 3
a) [r(l) :r(N)] = : piN
for N 2
A congruence subgroup of r
r (1) is a subgroup
containing r(N) for some N. Further on by r we denote
an arbitrary congruence subgroup of r (1) (in fact we are
mostly interested in the cases r = r(N) and r = r a (N) ).
Suppose that r is equipped with a discrete topology
while H has a usual complex topology (hence H is
homeomorphic to an open disc).
Show that
Exercise 4.1.5. the action of r N on H
is free, Le. has no fixed points. Moreover, for any v e H
there exists an open neighbourhood Uv homeomorphic to its
image V vcrN \H •
Since rN
= r/rN is a finite group, we can define the
Riemann surface r\H as a factor of rN\H over r N.
Exercise 4.1.7. Show that the Riemann surface structure
on r\N does not depend on the choice of N.
if c·r + d =0 .
Let r\H *
e
H--~ r(l)\H ~ C
X(N) = Xr(N)
YeN) = Yr(N) ,
'1:Zf--+ (a.z+b)
c·z+d
L
m
= ___1__ .(0-m 10)
v'i1I'
e SL(2,~)
Thus,
0'
Aut(Xo(N» ;;2 (Z/2) 0,
4.1.16.
Exercise Let a e H * , and let -a be the
image of a in r'\H * . Show that the ramification index of
the projection ITr , r' at a equals [ra : r a' ] .
(N - 6) -IlN
g(N) 1 + 12 -N
11. v v v
1+_-2-2-~
12 4 3 2
where
for 41N
v, - { ~ [1 n 11+ for HN
where
for p 2
(-~ ) for
for
p
p
= l(mod
= 3(mod
4)
4)
Chapter 4.1 CODES ON CLASSICAL MODULAR CURVES 405
v, ={ p[1 + [-~ II
for
for
91N
9fN
where
for P 3
for P !Ii l(mod 3)
for p !Ii 2 (mod 3)
vm = L II' «d,N/d» ,
diN
d>O
where
e1 = i,
= 11, e4
e" e = I- I- •
2 Q2
6·(1-+1) 4·(H1)
L P. + 2· L R +
i=l 2 m=l m
2
X (p ·n) ----+ X (p·n)
o 0
a) p ~ 5 andpin , or
p 3 and 31n n ~ 3 n ~ 9 , or
p 2 and 21n n ~ 2 n ~ 2·l , n ~ 4·l ,
l being a prime;
b) n is divisible by three different primes;
c) n is divisible by m
2
, where p~m and p 2 ·n ~ 36;
Exercise 4.1.27.
Prove the above proposition, for each
case pointing out a non-gap at 00 not exceeding go (p2. n) .
(Warning: This exercise is intended only for an assiduous
reader) .
m
f = L am'q
m=-N
IX>
fez)
so that
•
Thus, the modular equation ~NCx,j) =0 defines an
affine model of the curve XoCN) . Moreover, since ~N has
integral coefficients, the curve XO CN) is defined over
(IJ • Further on we shall be interested mostly in the curve
Xo(N) over (IJ , however to study its properties one has to
use the Riemann surface ro(N)\H* as well.
Y (N) ~ Y(l)
o
(s , t) 1----+ e(s,t) ,
Let e (s, t)
__ e2rri' (a·b'-a'·b)/N for s = (a,b) and
t = (a' ,b') •
Chapter 4.1 CODES ON CLASSICAL MODULAR CURVES 415
maps e 2rri / N to 1 .
m x idE
E xs E xs E )E (associativity) ;
rm
E
idE x m
1
E Xs E
m E
i ) E Xs E
E Xs E (commutativity),
1 E
1 E
m
(the existence of a
neutral element);
E E
'" x idE
E Xs E )E Xs E (the existence
1
of an inverse).
r
11 m
p e
E ~s ~ E
Chapter 4.1 CODES ON CLASSICAL MODULAR CURVES 417
(Y(N)/7l[1/N) Y(N) •
F/ r 8f
~HomSCh/A (. ,SF)
commutes, the functor morphism 8f being induced by f.
Y(N)/Z[l/N]
•
Chapter 4.1 CODES ON CLASSICAL MODULAR CURVES 421
To
clarify the situation described by Proposition
4.1.45 and Corollary 4.1.46, let us describe the case of
Y (N) and Y p (N) considered over Q. To put it more
exactly, let
and
g'
N
= g(X') = g(N)
N
and let be the number of
Theorem 4.1.52.
lim (4.1.1)
n-----?IXI
(4.1.1')
n - 1
a) the number of such points equals ~N'~
Thus
~N' (N - 6)
g'N = g(N) 1 + 12.N
'
RN = (
P1- ) ogN + 2/3
cogN + a (gN
2/3)
for some c > 0 0 From the proof of Theorem 2.3.22 one can
conclude that
N
1
~ (p - l)og + o (g/log g) ,
4.1.3. Codes
d~n-a,
430 MODULAR CODES Part 4
3
> N . (p - 1) . IT 2
n' - 24 I I (1 - e-) ,
o elN
N· (p - 1)
12
n = 1 N· (p - 1)
12
Exercise 4.1.59.
a) Show that the number pen) of the
positive integers of the form
m =1 N·(p - 1)
12
·n(l
e IN
+ e- r , m ~ n
1) ,
Chapter 4.1 CODES ON CLASSICAL MODULAR CURVES 431
p - 1 n
is asymptotically at least ~-ln n' i. e. pen) 2: f(n),
where lim 12-f(n) -In n = 1 (Hint: Consider the case of a
(p - 1)-n
n~co
N 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22
n 3 3 6 4 6 6 9 6 12 7 12 12 12 9 18 10 18 16 18
N 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40
n 12 24 15 21 18 24 15 36 16 24 24 27 24 36 19 30 28 36
go (2m) 2m- 3 + 1 - 0
m
where
0
{3.2 m/ 2-2 for an even m
m 2 (m-l)/2
for an odd m
m-3
no 2! 2 . (p - 1) ,
k 2! k(m,a) ,
k(m,a) n
<-.L
iff m-2-t
2 m-3 - 2 : S a < 2 m- 3 - 2m- 3 - t ,
v 00 = 1500 I
is a PGL(2,7lIN)-module of dimension
Hence considered
C' as a PGL(2,7lIN)-module has a
filtration of the form: ::> C , where
ao
Cm for b ~ m ~ ao is the image of L(moFm) in C', and
all the factors Cm+1/Cm for m = ao, ••• ,b - 1 are
PGL(2,7lIN)-modules of dimension ~NIN. In fact, applying a
more subtle algebraic-geometric and group theory technique,
one can determine the isomorphism classes of
PGL(2,7lIN)-modules of the form Cm+1/Cm. However this
procedure is far beyond the methods of this book.
435
436 MODULAR CODES Part 4
Field QJ Field k
Field Ie Field k 00
Elliptic curve E Elliptic module rp
Lattice A c Ie Lattice r c k 00
positive integer N Ideal I c A
(structure level) (structure level)
n i
L a.·~
i=O ~
D K{1:} ~ K
n i
D La.·1: f----+ ao
i=O 1.
1/>: A~K{1:}
~ (T) =i (T) + c 1 ·F + c 2
.~ '
q2 1
and d
2
= A - ·c2
for
for o
d
then M (A/(a» y , hence
d d
IMI IA/ (a) I y lal y
corresponding to a lattice r.
zeL
z· n (1 - Z'o-l) •
oer-{O}
cp : A - . . EndlF' (L)
q
2· I al i
cp (a) = i (a) + L cp. (a)'F ,
i=l ~
2· la I )
ip2.lal (a) E HO ( S,L®(l- Q )
never vanishes on S.
v (I) = {P e Spec A I P 2 I} .
By
F(I) : (Sch) 0 --~) (Sets)
a'z + b
Z 1-----+ c.z + d
where
where m = deg PI ' and the product is taken over the set of
irreducible monic polynomials r dividing PI •
•
The reducibility of H(I)®AkctJ is closely connected
with the fact that the above field k(I) is the natural
field of definition for this curve. This is parallel to the
fact that classical modular curves X(N) are defined over
«l (i: n ) •
l
M(I) ) M(I)
rr(I)
M1
1
(I) det g ) M1
rr (I)
(I)
{(~~) e GL(2,A) I c e I} ,
{(~ ~) e GL(2,A) I c e I , a - 1 e I} ,
H(I) { (~ ~) e GL (2 , A) I b e I , c e I , a - d e I} .
e(I) X(I)/H(I) •
In our assumptions
on the ideal I these curves
possess a number of nice properties:
Proof: a)
By Proposition 4.2.21 the fibres of the
morphism XCI) ~ Ht (I)®AlFq are absolutely irreducible.
using Proposition 4.2.2 and surjectivity of the map
det : G ----+ (A/I) * where G is any of the groups
1 0 (1), 1 1 (1), or H(I) , we get the assertion.
b) This assertion follows from the fact that in our
assumptions about I the determinant defines an isomorphism
- - q* ~
H(I)/lF - (A/I)/lF q * *
•
Exercise 4.2.24. Give a detailed proof of Proposition
4.2.23.
* * m _ 1
I (A/I) /lFql = qq _ 1 - l(mod(q + 1»
•
Since is identified with the factor of
over the action of (A/I) */lFq'
* we obta1n
.
452 MODULAR CODES Part 4
consists of m
b) The set Cusps 1 (I) 2· (q - l)/(q - 1)
points .
•
Exercise 4.2.28. Prove Proposition 4.2.27, using
Proposition 4.2.19. (Hint: b) follows from a) and Corollary
4.2.26; to prove a) consider the set ro(I)\~l(k) ).
m
P (T) ~ a o~ a E IF 1 .
I k~O k 'k q
2
P k + 1 (x,y) = xoPk(X,y)q + yoPk(X,y)q for k 2:: 0 0
Let
m
F(x,y) L akoPk(x,y)
k=O
FI = 0
{ o
aFI/ax
aFI/ay o
2m-2 2 2m-2
= Z(q -l)/(q -1). (Z + Z )q + Z ·H(Z :Z :Z )
1 12 0012
q2 1 q-1
F (c -·a c ·b) = 0,
I '
if and only if
a' 1\
q 2 - 1 • a , b' = i\ q-l . b , c' i\
-1
.c
(Hint:
c) We have
2 1
F (c q - ·a o •
I
(X (I)n{x = O})U{P ,P )
1 1 2
m-l
reT) L r oTk E A
k=O k
where
m-l
A L rkoPk(x,y) .
k=O
e (a,b,l) ~ (a,b,c) ,
where
m-l
c L rkoPk(a,b)
k=O
ao~ + boF
•
IF 2-rational pointso Similarly to the curves Xo (N) /p
q
and Xp (N) /p, the curves and C(I) have many
IF -rational points.
2
q
460 MODULAR CODES Part 4
N(I) _
lim g(I) - q - 1 .
m(I)~
lim N(I)/g(I) ~ q - 1 ,
e(I) ~ e(I)/PGL(2,A/I) = p1
it would yield
No (I)
lim
m(I)~m
go (I) =q - 1 '.
F(X) o
Chapter 4.2 CODES ON DRINFELD CURVES 463
(qm + l)/(q + 1)
•
The action of H = (A/I) */IF q* makes it possible to
describe the decomposition of F(x) into factors
irreducible over IF Let Hi be the subgroup in H
q2
generated by the image of the element T and let hl = IHil
be its order and its index in H, h = h ·h
1 2
4.2.3. Codes
n = qm.(q2m - l)/(q + 1) ,
n = (qm + l)/(q + 1) ,
k ~ a - q. (q2t - 1) / (q2 - 1) + 1 ,
d ~ (qm + l)/(q + 1) -a ,
where t = ["m/21 .
466 MODULAR CODES Part 4
2£ 2
go(I) :s q' (q - l)/(q - 1) ,
n q2 - q + 1
k2:a-q+1,
d 2: q2 - q + 1 - a .
k ~ a - q. (q 2 + 1) + 1 ,
432
d~q -q +q -q+1-a.
F 00 P
PeCusps(I)
Noting that
r(I)\PGL(2,A)/B = PGL(2,A/I)/B ,
POL YNOMIALITY
471
472 MODULAR CODES Part 4
1 :s g :s (p - 1) / 2 , 0:$ h :s t - 2 ,
and
n = (p - 1) (t + 1)
k ~ (2g - 1)· (l - 1) - 2h ,
Proposition 4.3.3.
I l1otoQl + 11 0Q
2
+ toQ
3
+ Q
4
J Q + toQ + 110Q + 11 0t oQ
1 2 3 4
•
Exercise 4.3.4. Check the formula for the action of
TV110t on the 4-tuple (Ql,Q2,Q3,Q4). (Hint: compute the
action of TV11 t 0 on r 0 ( 11 0 t) \!Pl (aJ) ).
We also need
an explicit description of the space
L(goF - hoQ 2 - hoQ4)used in the definition of the code.
Our immediate purpose is to show that this space contains a
basis of some special form consisting of rational functions
Chapter 4.3 POLYNOMIALITY 475
dim V ~ 1 (4.3.3)
2
for some
f e V1 n L ( 2· (6t + 5) . (I + J) - D) •
Hence
f e V
g n L(2g·(6t + 5)·(I + J) - g·D) ,
i=s
a. e K
~
, then by definition
n
Rn(f) L a. ·t
~
i
eK[t,t
-1
] , Qn (f) an e K .
i=s
478 MODULAR CODES Part 4
-1
K( (t» ~ K[t, t ], Qn K«t»~K.
A
m, t (H) = 0 for any m ~ g
is equivalent to
ordQ (i1) 2: g + 1 .
t
/1 (Ill + 1) ·g·s
v = (i + 11) ·g·s ,
11:.' K - g·n 2 - g. (i - 1)
A' A - g·n 2 - g. (i - 1)
Let
V' U Vp ,
PE(Sing X)n 1A2
g·n /1
( n ( n t Ker Am t»n( n «Ker Am,l )n(Ker A m,4 »n
teV' m=O I m=A-g·n l
v
n( m=O
n {(Ker A
m,2 )n(Ker A m,2 ) )nWg ,
v /1
u'g ug n m=K'
n
Ker Am 2)
,
n
m=A'
n Ker Am ,4) ,
482 MODULAR CODES Part 4
VO
g
n L(2g· (6£ + 5)' (I + J) - g'D - g'F) •
•
K-gn 2 A-gn 1
U
g,h (n
m=K-gn 2 -h
Ker A
m,2 ) n m=A-gn
(n -h
Ker A
m,4
)
1
°
Vg n L(2g· (6£ + 5)' (I + J) - g'D - h'Q2 - h'Q4)
•
To construct the code we need the following notation.
We write ss (X) s; X (IF ) for the set of points P E X (IF )
~ ~
such that a p = x (P) and (3p = j (P) are supersingular
values of j-invariant in characteristic p. If P E SS (X)
is a singular point, then the tree rP and functionals
Am, t are defined, where t E Vp' is the free node of rp
Chapter 4.3 POLYNOMIALITY 483
for
E v' (p - 1) 0 (l + 1) ,
Pess(X) P
1 : U h ® IF - - + IF n
g, p2 p2
Cg, h·
12· t+ll
m
F (
N x,]
.) -
- x 12· (t+1) + J.12· (t+1) + ~
t.. mn .x .yn
C
m,n=O
IX!
(1 + 504· E ~ (n)·qn)J
n=l J
j (q)
Chapter 4.3 POLYNOMIALITY 485
2 2
A (x, t) (t 121 · g ·X - 1 - c ·t 121 · g )
1
D(x,t)
(4.3.6)
j(a(z))
Let
12' t+11
R (F(x,j)) = x 12 ·(l!.+1) + L Qa(t).x a ,
o a=O
-1
where Qa(t) e IFp[t,t ] •
11·t
since F(x,j) lies in IFp[X] «q)) and q = t
-1
the Laurent polynomials Qa(t) I ie in fact in IFp[q,q ] ,
and we can therefore write
12 (HI)
-m
0
L rom oq
m=O
we have:
12 0 e+ll
c a , 12 oHll-i r a ,12 Hll-i L c am °v m, 12 oHll-i
m=12° (Hl)-i
0
12·Hl
L Cab'RoUb) for 1::s a ::s 12'£ + 11
b=l
thus for any 1 ::s i ::s 1~'£ + 11 and 1 ::s a ::s 12·R. + 11 we
have:
12 '£+11
L C
am ·v.
m~
m=i
12·Hl
L C
om
·V •
m~
m=i
FN = 0
{ 8FN/8X o
8FN/8j o
•
Corrolary 4.3.18. Let a 0 be a root of the equation
R(X) = 0 .Then the second coordinates ~p of the singular
.
points P E S1.ng -
X n
IA 2 with a p a0 are defined from the
system
o
•
To describe the conductor D of X, we must find an
expression for numbers nt' where
D = ~ n .Q
L. t t
and
Vt = max {m I b m * O} .
Then
{
j(Qt>
nt
vt - 2 for j(Qt> 1728
v t - e. + f. for Qt = Qi e {Ql,···,Q4}'
~ ~
where
10, e 4 = 11·l - 1,
(F')
x
= Be + D - 2' (6£ + 5)' I - 12· (£ + 1)' J •
Then
•
Exercise 4.3.22. Prove Proposition 4.3.21.
12· £+11
F N ( x, J. ) -_ 12 . (l+ 1 ) + .12' (l+ 1 ) + ~ a .b
x J L. cab' x . J
a,b=O
G(t) t·Rj.1_1(j(t»,
00
where J'(t) =
m~oCm·t m ) is the Fourier expansion
t- 1 • (1 +
coefficients {Cab} .
__l __ · W
W'm
Pm(x,y) m
496 MODULAR CODES Part 4
q2 -ary code on X' defined by the divisor D' and the set
of supersingular ~ -points on X' • The code C' has
q2 m,b
the following parameters:
n = n(C) (qm + l)/(q + 1)
k = k(C) ~ b - g' + 1 ,
d = d(C) ~ n - b ,
V V Vp , etc.
1 2
2
L (n *D') S;; V
m
P(x,y) = Z~s'H(ZO:Z1:Z2) ,
Therefore we have:
By Proposition 2.5.26,
we have
H = Z ·A
o
+ F'B
•
To construct L(n *D') , we have to consider functionals
Ar,t defined in 4.3.1, t being a free node of one of the
trees r 1 or r 2 Note that since all the points from
cusps' are defined over ~q' the functionals A
r,t are
defined on the space Wm Note also that the functionals
A
r,t
are trivial for r < -q4m ·n t , where
n t = ordQt (Zo) , since any P e Wm can be written in the
form P = G(Z
o
,Z ,Z )/Zs
1 2 0
for s ~ q4m
500 MODULAR CODES Part 4
-b
n
teV'
n
r=-n'q
4m Ker(Ar,t) C Wm
Proof: Since m
*
our only purpose is to
L(n D') s; V
sort out from Wm the elements p e Wm such that
ordQ (p) ~ -b for any t e V' • These are singled out by
t
the conditions for t e V' and
4m
r = -n·q , ... ,-b
•
Realization of C' . The algebraic-geometric modular
code C' = C'm,b
q2
is given by its generator matrix
over IF
M = (m (3)'
a, a running over the set of indices of
IF -rational supersingular points (Pal e Xo(I) (lFq2) and
2
q
over that of the basis of L (D')
ma, (3 = f(3 (P a ) e IF
q2
m
E : Wb ® IF 2 ~ (IF 2) ,
q q
E : p(x,y) ~ h- 1 . L p(a,O)
aeR(j
Chapter 4.3 POLYNOMIALITY 501
/l
Step 2. Construction of r.
Input: PI(T)
Output: The graph where and
r2 are desingularization trees of and
respectively.
4.3.3. Complexity
where P(Z) = Z
s + a _ 'z s-l + ... + a ' The roots of the
s 1 o
s-l
polynomial P(Z) are {{3,{3P, ••• ,{3P }
s-1 s-1 i op
Lao ozi (mod P) ~ L aI?oz (mod P)
i=O ~ i=O ~
F(O,O) = °, Fy(O,O) ;t °
Suppose that x and y satisfy F(X,y) = 0. Since
Fy (0, 0) ;t ° ,y allows a power expansion in x. Formulae
given in section 2.1.2 for power expansion of functions in
local parameters make it possible to construct segments of
corresponding power series bounded by a polynomial in d, s,
and b with polynomial complexity.
By (2.5.3), we have
o s l = g(X) = P a (X) - L _ 8p
PeSing X
r . (r - 1) /2
Q Q
PeSing X Q~P
L _ (v P - V p') = L
PeSing X P
therefore
E V' ~ E vp ~
PeSing it P PeSing it
Proposition 4.3.32.
511
512 MODULAR CODES Part 4
SPHERE PACKINGS
This part is mostly devoted to another kind of problems
which resemble greatly the questions about codes - namely,
to quite a classical problem of dense packing of equal
non-overlapping spheres in [RN. It comes out that both
direct application of algebraic-geometric codes and the use
of intuition developped while studying them are quite
useful. Moreover, in this part - especially in the last
chapter - one can see even better then before the marvelous
integrity of mathematics, two more parts of which - number
theory and that of packings being added to coding theory and
algebraic geometry.
In Chapter 5.1 we give basic definitions, put the
packing problem (both for lattice and non-lattice packings),
look at examples and discuss asymptotical results. Chapter
5.2 is devoted to some constructions of sphere packings
using codes; algebraic-geometric codes happen to be rather
useful, especially (as one would expect) for asymptotic
problems. Chapter 5.3 is a very short analogue of Part 2, we
give a brief textbook on number theory, mostly without
proofs, concentrating on the needs of what follows. Chapter
5.4 is the core of this part: we consider analogues of
algebraic-geometric codes from many different points of
view. We come to explicit algebraic-geometric and
number-theoretic constructions of asymptotically good
lattice packings. Algebraic-geometric approach should be
also quite useful to construct examples of dense packings
"on a finite level", but we do not say much about that here.
518 SPHERE PACKINGS Part 5
5.1.1. Parameters
5.1.2. Examples
5.1.3. Asymptotic problems
5.1.4. Codes and packings
5.2.1. Constructions
5.2.2. Results
5.1.1. Parameters
d = d(P) inf lu - vi ,
v,ueL, v*u
519
520 SPHERE PACKINGS Part 5
where
5.1.2.
Exercise Check that the volume of the
fundamental domain
N
F { L x.·e. ~ ~
10::5 x.
~
i=l
Chapter 5.1 DEFINITIONS AND EXAMPLES 521
A(L)
we write this as
1
--log V
N 2N 2
1
--log - -
2 N
(2rre)
Thus for N ~ 00 we get
1
- --log A(L)
N 2
522 SPHERE PACKINGS Part 5
i\(P) -(log2b.(P»/N
i\(P) ~ 0
~(1) = 1 , i\(l) = 0 •
~(2) rr/2v"3".
Clearly,
~l(N) ~ ~(N)
7(L) 4·(I5(L»2/N.
.\ (N) - 1 og2 I 2N
rroe0'1(N)
5.1.2. Examples
1 .
N+1
AN = { La.·e. o }
i=l ~ ~
a1
and that
~ , det AN = vN + 1
ai e ~ , L ai =0 (mod 2) }
2 •
Chapter 5.1 DEFINITIONS AND EXAMPLES 527
E
a
= {~
i=l
a.·e.
~ ~
2·a.eZ,a i - a . e Z ,
~ J
~
i=l
a.e2'Z}
~
8
E7 {x L a ~.·e.~ I x e Ea a7 -aa}
i=l
8
E6 = {x = L a ~.·e.~ I x e E7 , a6 = a7} .
i=l
Show that d(EN ) = V2' and det(EN ) = 9 - N .
Note that At = Z, D3 = A3 E4 = A4 Es = Ds The
lattice families A , D , and E are root lattices which
arise in many questions: in the theory of Lie groups and
algebras, in the singularity theory, in the theory of
rational surfaces, etc.
det rN 1
The lattices A
1
= 71 , A
2
E6 ,
A
3
= D3 D
4 '
D
5
E7 E8 = r 8 have the density coinciding with that from
Table A.4.1; they are the densest lattices in their
dimensions. A proof of this fact can be obtained by the
reduction theory of quadratic forms; we do not give it here.
Note that for all the described families
1
v.~,c -_·u. 1 ~ i ~ 24 , e e C 24
VB"" ~ , e
where u.~,e
has +3 in i-th position and ±1 in all other
positions, and the upper sign is chosen for a position where
the codeword e has 1
1
r.r.
~,J,o
. ., -_·x
VB"" i,j,a
1 ~ i , j ~ 24 , i ~ j , a e (Z/2)2 ,
where . . a
x ~,J, has ±4 on positions i and j with
arbitrary sign distribution which is defined by a and
zero entries in all other positions (there are 1104 vectors
of this type) . These vector types are denoted (3 1123) ,
530 SPHERE PACKINGS Part 5
and hence
Proposition 5.1.16.
Theorem 5.1.17.
xexp :s 1 .
note that A (PN > - - + 1 for N --+ 00. Check that our
algorithm is exponential in N .
•
The situation with X~xp is more complicated. One can
prove
Theorem 5.1.19.
n - 1
Ip= { -~ , ••• , 0, ••• , E....=....!21} c Z .
For a positive
integer and for X E IF~ set
IIxll 0' = Elx ~.10', where la I for a E IFp denotes the
absolute value of x as of an integer. Let c be a linear
subspace and let k = dim C , IV = min IICIIO' where minimum is
taken over all c E C - {O}; C is called an [n,k,IVJ p -
"O'-code". The usual linear codes correspond to the
degenerate case 0' = 0 . A proof of Theorem 5.1.19 can be
Exercise 5.1.20.
Establish for "O'-codes" an analogue
of the Gilbert-Varshamov bound.
Chapter 5.1 DEFINITIONS AND EXAMPLES 535
det L = pn-k
Theorem 5.1.23.
~~xp :s 1
•
Remark 5.1.24. Moreover the same result is also valid
for packing any bounded convex body whose automorphisms
include the reflections through the coordinate hyperplanes.
5.2.1. Constructions
539
540 SPHERE PACKINGS Part 5
the composition map (fN '------+ 71.N '------+ rRN also by C. The
p
lattice C-71. = {a-c I a e 71., c e C} is of rank k. Set
L = C-71. + p-71. N ,
Le.
L = (x e 71.N I x(mod p) e C} •
det L = P N-k •
5.2.2.
Exercise Let C be the [8,4,4]2-code obtained
from the Hamming code by adding the parity check. Show that
L "" Es
d) b = -molog t
p
is an integer.
r r
-pol -exp
· IT
xpol xexp ,X
Fq -----+ F~
where
Fp ~ {O, 1, ..• , P - 1} c I ,
i
c) V(L i ) == n·v(L) - b' L k.
j=l J
for a j e l. , c. e C i
J
and y e Li - 1 Since p'T . sends
L.~- 1 into itself we can assume that o :s a. :s p - 1 for
any j , Le. that x = y + Ti~o(C) with
J
c e C.
~
If .
*
z e IF'q then ~o(z) eK!;L and since KI (T-1L n K) O! IF'q
we get ~o (z)
_ T-1L
Hence. x = Ti (T-iy + ~o(c» where
T-1y e T- 1 (L ) = (T-1L) n let c = (z 1 , ••. ,zm) ;t 0 then
o
~o(Xj) has at least d i entries which do not lie in T-1L •
Therefore
p(i) , d(p(i) )2
V
U Px Pi v
xeXv
where
a
Exercise 5.2.10. Prove that v (Q) i!: n·v
o - b· L ki
i=l
1 [ 1100]
1-1 0 0
'2" 0 0 1 1
o 0 1-1
1 1 1 1 1 1 1 1
1-1 1-1 1-1 1-1
1 1-1-1 1 1-1-1
1 1-1-1 1 1-1-1 1
4" 1 1 1 1-1-1-1-1
1-1 1-1-1 1-1 1
1 1-1-1-1-1 1 1
1-1-1 1-1 1 1-1
3 -1 -1 -1 -1 -1
-1 -3 1 1 1 -1
-1 1 I
1 -1 1 1 -3
1
1 1 1
4" -1 1 1 -1 -11
. 1
-1 -1 -3 1 1 -1
5.2.2. Results
co
A({Am}) + L (1 - R i )
•
S
i=1
co
~ + ~'iE1 (1 - Ri )
•
A({LN }) A({Am})
Chapter 5.2 ASYMPTOTICALLY DENSE PACKINGS 547
Theorem 5.2.19.
a) ~pol ~ ~exp + 4/3
l l
b)
-pol "exp + 1
AT ~ "T
00
ex>
and since !. ~ (1 - R ) 1 , we get b) .
2 i~l i •
Asymptotic parameters. Let us look at the asymptotics.
Theorem 5.2.22.
•
Further on we use AG-codes.
Theorem 5.2.23.
:;:exp ::s 1 •
/\T 30•
Corollary 5.2.25.
-pol
AT 2.30 .
•
:5
Theorem 5.2.27.
•
Asymptotical results known to us now are displayed on
Diagram A.4.3.
CHAPTER 5.3
NUMBER FIELDS
551
552 SPHERE PACKINGS Part 5
xm + a m-l . x m- 1 + . . . + a 1 . x + a 0
m-1
a 'a o
m 0
•
Proposition 5.3.4. The ring Ok is finitely generated
as all-module.
for i j
c5 ••
l.J for i * j
Ok = :lw
1
+ ... + :lWn
N(x)
k - - - 7 IR
s x ([; t ,
a t----+ (~1 (a) , ••• ,~ s (a) ;~ s+l (a) , ••• ,~ s+t (a) ) •
Proposition 5.3.10.
d
11 , ••• ,xS1 ,Y 11 ,z 11 '···'Y t 1 'Zt 1
X
det . . . . . . . . . . . . . . . .
[
x1n,···,xsn'Y1n,z1n'···'Ytn,Ztn
].
556 SPHERE PACKINGS Part 5
It is clear that d
d* = det [~ 1 ~ , : . : ' ~s ~ , ~ 1 ~ ~ . ~ ~ Z 1 ~ , ~ 1 ~ ~ ~. ~ 1 ~ , : . : 1
X1n '···'XSn 'Y1n + v=!'Zln'Y 1n - v=I·z ln ,···
Exercise 5.3.11. *
Check that a e Ok iff Nk/~(a) = ±1.
log O~ ~ ~s+t
2
a ~ ( 1 og I CT (a) I , ••. ,log I CT (a) I ; 1 og I CT
1 S S
+ 1 ( a) I , ••• )
2 2
I CT 1 (a) I .•••. I CT s (a) I . I CT s+ 1 (a) I .•••. I CT s+t (a) I = IN (a) I 1.
IIx-yll = IIxll-llyll ;
558 SPHERE PACKINGS Part 5
ord (x)
p
where N(p) = 10k /pl. For such an absolute value (and for
anyone equivalent to it) a stronger condition holds:
Cl·b = {x·y I x e Cl , Y e b} •
5.3.2. Extensions
where a ~ e °
0 for i = 0 I • •• ,m •
Let us define the (relative) trace and norm as
•
Unramified extensions. An algebraic field extension is
called unramified iff DK/k = (1) • Theorem 5.3.8 (together
with Exercise 5.3.14) implies that fD has no unramified
extensions.
The field
k = D(v'2',v3'S'7'23'29 ')
and 1058.57 .
•
On the other hand using "explicit formulae" (cf. Remark
2.3.23) one can obtain a lower bound for IDk I 1 / n :
Un. 1 2
lim inf ID I ~ ~ (4ne 7 + )(1"(4ne 7 ) "['
ki
° °
f2l = k
Rings of the form Os or OF can be
characterized as those having one-dimensional irreducible
regular spectra of finite type over Z .
The number field case is mostly more difficult than the
Chapter 5.3 NUMBER FIELDS 565
n
n
A fractional ideal tl = p p where p runs over
p
-1
prime ideals of Ok' corresponds to a divisor on X , tl
566 SPHERE PACKINGS Part 5
There are some other analogies which are less clear but
also useful, and we use them in the next chapter.
CHAPTER 5.4
ANALOGUES OF AG-CODES
567
568 SPHERE PACKINGS Part 5
det L = 2- t _/T'n""""'r
·vlDkl •
Proposition 5.4.1.
and if t o then
deL) ,fir.
Chapter 5.4 ANALOGUES OF AG-CODES 569
Proof: Let
We have
I<T(£) I = I j=l]
Ex~ + ~ (y~]
j=l
+ z]~)
I E x~
j=l]
+ ~ (y~]
j=l
+ z]~) ~·I
v'2"
E x~
<:
j=l]
+ 2· ~ (y]~
j=l
+ z~)] <:
=/ ~2 + . IN (£) I liN
s
t
K/ILl '2 + t
/~ x~
N l/2N
.;N". ( nx~ ) v'N". IN (£) I liN v'N"
•
2: 2:
j=l ] j=l ] K/Q
2n _,.......,.
For Ko = ID (cos 11 ,v-46) we get A - 2.2218 (by
Theorem 5.3.20 Ko has an infinite class field tower).
On the other hand, Theorem 5.3.21 shows that for any
family of fields K we cannot get asymptotically less than
1.193. . . (and 1. 694 ..• assuming the generalized Riemann
hypothesis).
k(v) = IR for real places and k (v) = II; for complex ones.
Let 0. be a fractional ideal such that SfnSupp(o.) = 121
where Supp (0.) = {1' E Spec Ok I l' 2 o.} • For any V E S
there is a natural map
-1
CT V : 0. L(o.) ~ k(v) .
~ E> k(v) •
YES
r ~ N(Y)C(V) ~ q
n = lSI = ISfl + s + t .
Let
the cube U' into q equal parts, and identify the set of
qN small cubes with ~N. Define
q
mapping each point to the small cube it lies in; let 'Pv
be the component of 'PCXJ corresponding to v E SCXJ For
v E Sf let 'Pv be the map
Let
be defined as
a) d ~ n + 1 - a
Chapter 5.4 ANALOGUES OF AG-CODES 573
On one hand
a/N
s _r__
q
•
Proposition 5.4.5 is also valid for t * 0 , but the
proof is slightly more difficult.
574 SPHERE PACKINGS Part 5
logq~
lim inf --~---~-
K n
'1 :s const·~
ql/4
const· log q
q
1/4
•
It is not difficult to see that for large q on some
segment of ~-axis these codes are better than the
Gilbert-Varshamov bound.
Im IPs C H = {x e IR n I LXi = O}
h(K) = heal
s+t
T H ----7 H {x E ~n I LX' O} = H e~n-s-t
o i=l ~ 1
where
{x E ~s+t I LX. = O} ,
~
~s + t ·R
and
where m = [K:aJ] .
To estimate d(LS) it is sufficient to estimate the
product Tlmax {l,lI f ll v } for f E K *-{±1} It is known .
VES
that for any totally real field K of degree m
*
O~ (f E K* I Supp(f) ~ ~}
Chapter 5.4 ANALOGUES OF AG-CODES 579
n
{x e 71. I L x~ 0 O} •
We set
b) rk Lcp = n - 1 and
IX(lFq) I - q - l]g
det Lcp :s ..;noIJx(1Fq) I :s ..;no [1 + q + --"-g-----
580 SPHERE PACKINGS Part 5
Proof: a) Let f E *
0'P , f e IF q*
Then
n- 1 = v'il
b) We know that det A and
Then
and
Therefore
(5.4.2)
Chapter 5.4 ANALOGUES OF AG-CODES 581
Indeed, IJX(IF ) I
q
= 0(1
i=l
- w.)
~
, W.
~
being the
2g
n(l
i=l
- Wi) n(q
i=l
+ 1 - w.~ - w.)
~
~
~
riE, (q +
1 - w.
~
- Wilr
g
g
- L (w. + w.)
i=l ~ ~
•
Asymptotic behaviour.
As in Part 3 we consider
families of curves of growing genus with
and set
IX(lFq)I
- -----7 A > 0
g
N
yields an asymptotically good family of lattices {L N c IR }
with
Remark 5.4.15.
For an algebraically closed basic field
k of characteristic zero one can prove that deg f ~ g/2
for any "general" curve X and for any f E k (X) • If it
were the case for modular curves over ~q (which we do not
know) we would get the estimate
a = deg D L ai·r i .
We write f == 1 (mod D) iff ordp . (f - 1) 2: a. for any
~
.
~
p.
~
E Supp D Suppose that 'P n Supp D = IZJ Let
(f E * I f
0'P - l(mod D)}
584 SPHERE PACKINGS Part 5
a) d(L~,D) ~ ~ ,
b) rk L~,D = n - 1 and
A*
(xeO p . I x=1(moda i 'Pi)}'
~
We have *
O~,D and
[r-T0A*p . :r-T°p.,a.
A* ]
~ ~ ~
TI (
r.~ r .. (a.-1»)
~ ~
I I (q - 1)·q
Chapter 5.4 ANALOGUES OF AG-CODES 585
A ({L N }) oS - log ~ 1
-2 + -·log(ln
2
q) +
+
vg ·log q - log(q - 1)
.;q - 1 •
For q 2209 47 2 we get A < 1.3888 . . . .
0*
S,o.
* I f a l(mod o.)}
{f e Os
b) rk Ls ,0. = n - 1 and
det L
S,o.
Proof: a) Let
1 _2 . LX.
2: -·Llx.1
v'il' ~ v'il' x .>0 ~
~
We have
L xi L In IIfliv
x.>O yeS
~
IIfll >1
v
L L In IIf - 111v +
yeS yeS
00 00
No
II. ({L N }) ~ - log /2rr ' + --.!....logID - - ' (log rr - 1) +
e n K no
o 0
ISo I
•
Chapter 5.4 ANALOGUES OF AG-CODES 589
2rr _~
For the totally complex field ~(cos11,v-46) and
50 = 5 00 we get A ~ 11.1512... • For the totally real field
~(1I'2"',v'305o7o23o29') and 5 = 5 we get A < 8.80 . This
0 0 0 -
are not best choices but what we get is always much worse
than for FMLD-case.
*
O~,D
d log 1
Res~
------------------~) ~~
590 SPHERE PACKINGS Part 5
where
d log *
O~,D ~ n(E Pi - D)
f 1------+ df/ f ,
Le. e e' n n
IF p
Then
a) d ~ d' ~ a - 2g + 2 ;
b) k ~ n - 1 - m·~·a
p
pk = lei = [L
~,D
:B]
-1
B] . [A n - 1 :L~ ,D]
n n-1
n- 1: An- 1 n (pZ) ] = p
We have [A The multiplication
by p maps isomorphically An - 1 onto A n - 1 n (pZ) nand
Chapter 5.4 ANALOGUES OF AG-CODES 591
m·P-l·a
~ qdeg D - deg D' p
[L'P , D' : L'P , D' ] p
Summing up we get
~A ,
2·m· (p - 1)
~ A>
p
-1
8 < p - 2·A
m· (p 1)
592 SPHERE PACKINGS Part 5
4.
Each construction considered above was encoded by
three or four letters. Formally speaking there are 16
possibilities. What can we say about those we have not
mentioned?
594
NOTES 595
597
A.l. Codes of finite length
A.I.I. Bounds
A.I.2. Parameters of certain codes
A.I.3. Parameters of certain constructions
A.I.4. Binary codes from AG-codes
A.l.l. Bounds
2) (~) . (q -
A
q·w 2
d - 2'w + (q _ 1).n > 0
one has
5) let
n
1 + E ai'Pi(j) ~ 0
i=l
n
qk ~ 1 + E ai' (~) '(q - l)i
i=l
(the linear programming bound);
B) Conversely, if
a) [n,n,1]q-codes;
b) the even weight code: [n,n - 1,2]q
c) the repetion code: [n,1,n]q •
2) Hamming codes:
- 1 qm _ 1
- 1 ' q - 1
3) Reed-Solomon codes:
[q + 1, k, q + 2 - k]q
[qm _
q _ 1 ' m , q
1 JlI-1]
q
r ii/ql
E E (-1)
JO ° (m) (m-1+; _qJO) (q - b) oqm-a-1] ,
i=O j=O m-1
0 0 ... ,
J q
r = ao(q - 1) + b , 1 ~ b ~ q - 1
602 APPENDIX
In particular, for q 2:
r
L
i=O
6) BCH-codes (primitive):
[n , k , n + 1 - g - k] q
8) Golay codes:
9) Goppa codes:
and for q 2:
m· r~l ' dJ q
f (2~).(q -
i=l ~
l)i s n - r + 1 .
t
~ mJ.,D.]
[N, j=l 1. q
k
i=l
L m.·k.
1. 1.
i = 1, ••• ,t
·+1
[N, ~L m .,Di + 1] ;2
J=l J q
d ~ mi n { d .. D . }
lsist 1. 1.
;2[N, ~ m.,D.]
j=l J 1. q
A) d 25
n 91 92 93 94 95 96
k 24 24 25 26 27 27
B) d 27
n 159
k 69
C) d = 29
169 170
71. 044 72.044
APPENDIX 607
A) "True" bounds:
is the "true" bound for q-ary codes (Theorem
1.3.1).
is the "true" bound for linear q-ary codes
(Exercise 1.3.2).
is the "true" bound for polynomial families
of q-ary codes (Theorem 1.3.25).
a.~ol,lin is the "true" bound for polynomial families
of linear q-ary codes (Theorem 1.3.25).
a.~Ol dec is the "true" bound for families of q-ary
codes with polynomial complexity of
construction and decoding (Exercise 3.4.33).
a.~Ol dec,lin is the "true" bound for families of
lineary q-ary codes with polynomial
complexity of construction and decoding
(Exercise 3.4.33).
608 APPENDIX
B) Upper bounds:
Rp(d) - 1 - q q- 1 .~
where
APPENDIX 609
C) Lower bounds:
RGV(o)
RGv(o) - 0' J dx
_~ -1
R (0) = 1 - (vq - 1) - 0
TVZ
H (0)
q + (1 - o)'log (q - 1)
q = 1 + (Vq'" - 1)-1
2m' (q - 1)
q' (q
m/2 - 1)
RKTV , KT(o)
where
R;V(8) max {1
m
- 3m- (q - 1)
q- (q
m/2 - 1)
m- (q - 1)
q
-8}
where maximum is taken over all positive integers
m such that q m is a square;
where
2m
B) Comparison of bounds for 2 < q = P < 49
RSV
~-
Rsv (lin)
---
Comments: 1. If q = 4 (and also if q = 3 ) then we
also have RS ~ RH •
2. If 2 < q = pm < 49 and m is odd, this diagram
should be slightly changed. One should then set
R TVZ (0) = 1 - r q - 0 where rq is not quite known. One
should also consider RLT which plays an important role,
e.g. for q = 47 it intersects RGv .
3. One can also consider the case of 2 < q < 49 ,
q * pm. Then there is no proper notion of linear bounds, and
non-linear constructions leading to RLT play a crucial role.
APPENDIX 615
! ~11<T
~TV(lin),KT
RS~~_
--
RSV(lin) ~11<TV(lin)
- ....... 1)3z ~ RTVZ
Rsv
RSV(lin)
} 288 3 °log x
- -_ox
343 2
RKTV }
1152 3
- 343 oX olog2 x
RKTV(lin)
RKT - 8 °log 28 < 0
2
-_ox 2
RGV - 8 0log 2 0 In 2
1 2
R4 (2) - 2"°8 olog2 0 - 2 oX olog x
2
2 2
R4 In 2°° - 20x °logx
2
RBE
1
°
- 2"0 olog2 0
__
l_oX
In 2
1
RH - 2"°8 olog2 0 > 0
Rp 2°0 2 x 0
Rs > 0
APPENDIX 617
Bound 1 - R(5)
for 5 ~ 0
3
In q-5 (log 5) 2 q _x 3
2 q 6(q - 1)2-ln q
-2- q - q 1 -5-logq 5
(-.I([ - 1) - q4 3
- 2- -x -log X
(-.1([3 _ 1) 3 q
-2-5-logq 5 for q = p2m
- 2-
( q + 1) 2 -q 6 -x 3 -log X
RKTV(lin (q3_ 1}3 q
for q = p 2m+1
__
2 __ 5
R4
In q - 2-x 2 -IOlj x
1
- --a -logq 5 q
RBE 2 2-ln q -x
1
RH - --
2
a -logq 5 > 0
q q
Rp q - 1- a q - 1- x
Rs 5 > 0
618 APPENDIX
q 2
q = 4
q 16
q 64
q 256
H (w)
2
- W· H
2
(.....£.)
2w
- (1 - w)· H (
2 2
13 2w )
and
-1
W K K K REZ(W'oO) RG(W,oO) fl(w,o ) %
1 2 0 0
1
vq - 3
7l.
N 1 2 3 4 5 6 7 8
--
fJ. e (N) 1 0.907 0.740 0.617 0.465 0.373 0.295 0.254
8 e (N) 0.5 0.229 0.177 0.125 0.088 0.072 0.063 0.063
i\.e (N) 0 0.070 0.144 0.174 0.221 0.237 0.251 0.247
ve (N) -1 -1. 792 -2.5 -3 -3.5 3.792 -4 -4
'le(N) 1 1.155 1.260 1.414 1.516 1.665 1.811 2
8 e (N)-2 4 12 32 64 128 192 256 256
APPENDIX 627
LN 5 (L N ) '¥(L N )
7l.N 2- N 1
2
-N/2 2
AN
v'N + r !IN + i
2
-N/2 2
DN ' N ~ 3
2
!J4
2
-N/2 2
EN' 4 :s N :s 8
v'9 - N !J9 - N
2- N/2
rN ' N ~ 8, 2
N - o (mod 4)
A 1 4
24
628 APPENDIX
r r t
-pol :s 2.30 -exp :s 1. 30 ----+ :s 1.30
l' ·I'
-pol :s 2.30 -exp :s 1
l'
:s 1
Weierstrass points at
[At] A. O. L. Atkin, cusps of
Ann.Math., 85:1 (1967), 42-45.
631
632 BIBLIOGRAPHY
647
648 AUTHOR INDEX
651
652 INDEX
A. General notation.
B. Coding theory.
c. Algebraic geometry.
D. AG-codes.
E. Modular curves.
F. Sphere packings.
G. Number theory.
A. General notation
7l ring of integers;
G/H factor-group;
[G:H] index of a subgroup;
<g> cyclic group generated by g
X addive character.
B. Coding theory
y·C
Section 1.2.1. Pi (x) ; Ev~ ; Res? ; Res~ Sj(V) .
Section 1.2.2. C
23
, C
24
, C
11
, C12 .
Section 1.2.3. C eC
1 2
; c l ; C ®C
1 2
; c®l ; (C
1
I C2 ) .
ylin ; u lin lin
section 1.3.1. Yq ; Uq ; a q ; q q
aq
C. Algebraic geometry
Section 2.2.5. C ,• Cq
Nq(g) ; Bs •
section 2. 4 • 1. A •
section 2.4.2. j (E) •
Section 2.5.1. AL XV
Section 2.5.2. Cp np ~p D .
Section 2.5.3. Pa (X) •
Section 2.5.4. (j ; C' rQ ; rp .
section 2.6.2. Hax(A) Spec A ; k(p)
666 LIST OF SYMBOLS
D. AG-codes
E. Modular curves
Chapter 4. 2 . C k ; k
co
F. Packings
G. Number theory
.
section 5.3.1.
Section 5.3.2.
k ; Ok ;
° ; 5, t; Dk Rk Cl k hk
DK/k ; :l)K/k K
1
Section 5.3.3. Os OF
Section 5.4.1. sf S
00 Evs .
Section 5.4.3. o(q,r) .