Errata for Course Notes - J.S. Milne, Top

This file contains miscellaneous errata and additional remarks for my course
notes that I haven't yet incorporated into the versions on the web.

Most are taken from e-mail messages -- I thank everyone who has contributed!

Group Theory

Fields and Galois Theory

Algebraic Geometry

Algebraic Number Theory

Modular Functions and Modular Forms

Elliptic Curves

Abelian Varieties

Lectures on Etale Cohomology

Class Field Theory

Complex Multiplication

Algebraic Groups

Basic Theory of Affine Group Schemes

Lie Algebras, Algebraic Groups, and Lie Groups

Reductive Groups

Algebraic groups, Lie groups, and their arithmetic subgroups##
Group Theory v3.13 (GT)

p.40, Exercise 2-5. Q_n/Z(Q_n)
=‹a,b | a^2^(n-1), (a^2^(n-2))b^(-2), abab^(-1) and a^2^(n-2)›

=‹a,b | a^2^(n-2), b^2, abab›

=D_(2^(n-2)) (not D_(2^(n-1)))

or, order on left side is (2^n)/2=2^(n-1) while that on right side is 2^n so they don't fit. (Yunghyun Ahn)##
Fields and Galois Theory v4.50 (FT)

No known errors.
##
Algebraic Geometry v6.01 (AG)

No known errors.
##
Algebraic Number Theory v3.06 (ANT)

No known errors.
## Modular Forms and Modular Functions v1.30 (MF)

**p6** you said '... trivally on on H (Enis Kaya).

Tony Feng points out:**p.15, lemma 1.3** From Tony Feng:

In the proof of Lemma 1.3, I am not sure why V_1 \cap U_1 would be compact. I agree that it is relatively compact/closed, which is enough to get the desired U_2.## Abelian Varieties v2.00 (AV)

This draft is still very rough. Some proofs have been fixed in
the corrected version of my 1986 article on Abelian Varieties (xnotes).
##
Lectures on Etale Cohomology v2.21 (LEC)

From Section 9 on, I often abbreviate H^i(X_{et},F) to H^i(X,F) (Hongbo Yin).
## Class Field Theory v4.02 (CFT)

p.2, l7, Replace X^2+m with X^2-m (Jack Petok).
## Complex multiplication v0.00 (CM)

No known errors.
## Algebraic Groups v1.00 (iAG)

19.56 is misstated: It should say that every *split* reductive group...
T is a *split* torus.
## Basic Theory of Affine Group Schemes v1.00 (AGS)

p.30. The definition of a sub-coalgebra doesn't make sense unless, e.g., k is a field,
because in general D\otimes D is not a submodule of C\otimes C. (Roland Loetscher).
## Lie Algebras, Algebraic Groups, and Lie Groups, v2.00 (LAG)

p.15. In the final paragraph before "The isomorphism theorems", it states
that if a is a characteristic ideal of g, then every ideal in a is also an ideal in g.
It should read: If a is an ideal in g, then every *characteristic* ideal in a is also an ideal in g
(which is characteristic if a is characteristic).
## Reductive Groups, v1.00 (RG)

p.14, following the statement of 1.29: The centre (and radical)
of GL_n consists of the scalar matrices, not the diagonal matrices.
(Justin Campbell)
## Algebraic groups, Lie groups, and their arithmetic groups v3.00 (ALA)

p. 340 footnote 4. From Brian Conrad: you point out that a root datum is
really an ordered 6-tuple, not an ordered 4-tuple. In case it may be
of interest, the last bit of information in the 6-tuple (the bijective
map a |---> a^{\vee} from roots to coroots) is uniquely determined by
everything else, due to the requirements in the axioms of a root
datum -- see Lemma 3.2.4 in the book "Pseudo-reductive groups".

p. 372 footnote 16. Chapter II?? should be Chapter III.

p. 172, l.1: "and so is split be a finite" should be "and so is split by a finite" (Timo Keller).

Most are taken from e-mail messages -- I thank everyone who has contributed!

Group Theory

Fields and Galois Theory

Algebraic Geometry

Algebraic Number Theory

Modular Functions and Modular Forms

Elliptic Curves

Abelian Varieties

Lectures on Etale Cohomology

Class Field Theory

Complex Multiplication

Algebraic Groups

Basic Theory of Affine Group Schemes

Lie Algebras, Algebraic Groups, and Lie Groups

Reductive Groups

Algebraic groups, Lie groups, and their arithmetic subgroups

=‹a,b | a^2^(n-2), b^2, abab›

=D_(2^(n-2)) (not D_(2^(n-1)))

or, order on left side is (2^n)/2=2^(n-1) while that on right side is 2^n so they don't fit. (Yunghyun Ahn)

**p10**
The equality "Lambda(tau)=Lambda(tau')" should be interpreted
as saying that the two lattices are equal up to multiplication by a nonzero
complex number --- see
here.

**p.14, proof of 1.2, third paragraph**

As Keenan Kidwell and Tony Feng have pointed out, I muddle the
proof. I should say: G is a union of the interiors of the sets gV, g in G; fix a countable base;
the sets from the countable base contained in the interior of some gV form a countable cover
of G; now we only need to take enough g's to get each of the sets in the countable
cover at least once.

Keenan Kidwell points out, the sets gV are not open, hence not unions of open subsets. Fortunately this is easily fixed. As he writes:

In the proof of Proposition 1.2 in your exposition of modular forms and modular functions (beginning near the bottom of page 14), is it really possible to assume that the set $V$ is both open and compact? Initially I assumed that, because you referred to V as a neighbourhood of e, you meant that it was a set which contained e in its interior, but later you say that each set gV is a union of open sets in the countable base, which implies that gV is in fact open. For a general LCH group, we have a base of opens around each point consisting of sets with compact closure, but not necessarily open compact sets, right? But I think this is fixed fairly easily, since g is in the interior of gV, we have an open cover G=\bigcup (gV)^{int}, and by second-countability, this has a countable subcover, and so in particular we can write G=\bigcup g_n V for some sequence g_n, and then the proof goes through unchanged. So I think the only issue is when you say that each gV is a union of open sets.

Tony Feng points out:

In the third paragraph of the proof of Proposition 1.2, I was thinking that not every open set in the countable base need be contained in a single gV, but of course we can throw away such open sets from the base.

In the proof of Lemma 1.3, I am not sure why V_1 \cap U_1 would be compact. I agree that it is relatively compact/closed, which is enough to get the desired U_2.

**From Tony Feng:**
I think I've identified a small typo in your modular forms notes that was tripping me up for a while.

On **page 65,** in the right hand side for the inner product

p2. The claim in the footnote that every abelian surface is a Jacobian variety is not quite true. See the preprints of E. Kani, "The moduli spaces of Jacobians isomorphic to a product of two elliptic curves" and "The existence of Jacobians isomorphic to a product of two elliptic curves". (Kuang-yen Shih)

p26. now shows that dim(H^1(A,IF_l)) >= 2g: ')' missing (Timo Keller).

p37, line 6. The regular map alpha should go from T to the *dual* of A (Bart Litjens).

**From Bhupendra Nath Tiwari:** For AV, CFT, and CM

**From Tobias Barthel** See pdf file (2 pages)

**From Shaul Zemel** In the proof of Theorem 10.15, p49, concerning the map from Hom(A,B) \tensor Z_l
to the module Hom(T_l A,T_l B) (over Z_l), you start by proving that
if e_1,...,e_n are linearly independent over Z in Hom(A,B) then their
images are linearly independent over Z_l in Hom(T_l A,T_l B). But this
immediately proves that n cannot exceed the rank of the latter over
Z_l, i.e., 4dimAdimB (as can be even more clearly seen in Hom(V_l
A,V_l B) over Q_l, after tensoring the latter with Q_l). Hence you
immediately obtain the finiteness of the rank of Hom(A,B), and the
desired bound, without the need to involve decomposition into simple
Abelian varieties, different topologies, and polynomials. This is in
fact similar (as you have indicated there for something else) to the
fact that showing that if a (clearly torsion-free) subgroup of a real
vector space of dimension n is discrete then it's free of rank not
exceeding n.

**Timo Keller** points out that, in the proof of Theorem 10.15, p49, M should be defined to be a
Z-submodule of End^0(A), not End(T_l(A)) (the degree map P is defined on End^0(A), not on End(T_l(A)).
Also, when I write "Now choose the e_i to be
a Q-basis for End^0(A)." I seem to be assuming that End^0(A) is
finite dimensional over Q, which is what I'm trying to prove. The proof should be replaced by
this.

**From Everett Howe** In Prop. 13.2(b), I found a small typo, probably carried over from copying the result from [1986b] and not changing all of the notation: the "f" in the exponent should be an alpha.

**Tim Dokchitser** points out that I prove Zarhin's trick **(13.12)** only
over an algebraically closed field , and then immediately apply it in **(13.13)** over a finite field.
This is doubly confusing because (13.10) is certainly false over nonalgebraically closed field (over such a field
an abelian variety need not be isogenous to a principally polarized abelian variety).

However, I believe everything is O.K. Specifically, the proof of Zarhin's trick requires only (13.8), and,
because this holds over an algebraically closed field, it holds over every perfect field (see my 1986
Storr's article *Abelian Varieties* 16.11 and 16.14).

**From Sunil Chetty.**
Near the start of I 14 (Rosati involution): in (\alpha\beta)^\dagger = \beta\alpha there
should be a dagger on each of \beta and \alpha.

p.154, IV. Theorem 7.3. The property in (a) of the statement only
holds for the points P outside a closed subset of codimension 1. As **Martin Orr** writes:

[The theorem] asserts that the Siegel moduli variety M_{g,d} over the complex numbers satisfies: for every point P in M_{g,d}, there is an open neighbourhood U of P and a family A of polarized abelian varieties over U such that the fibre A_Q represents j^{-1}(Q) for all Q in M_{g,d} (I assume this should say "all Q in U" at the end).

I don't see how such a neighbourhood can exist around the elliptic point 0 in the j-line, because of the standard monodromy argument that there is no family of elliptic curves on all of M_{1,1}. Specifically, if we had such a U then the period mapping (which is just the inclusion U -> M_{1,1}) would lift to a map from U' to the upper half plane H, for some open neighbourhood U' of 0 (wlog lifting 0 to i). Then the image of this lifting contains an open neighbourhood V of i in H, and the map H -> M_{1,1} is not injective on V.

**From Roy Smith (on proofs of Torelli's theorem III 13)**

You ask on your website for advice on conceptual proofs of Torelli. ... here goes.

There are many, and the one you give there is the least conceptual one, due I believe to Martens.

Of course you also wanted short, ....well maybe these are not all so short.

The one due to Weil is based on the fact that certain self
intersections of a jacobian theta divisor are reducible, and is
sketched in mumford's lectures on curves given at michigan. Indeed
about 4 proofs are sketched there.

The most geometric one, due to Andreotti - Mayer and Green is to
intersect at the origin of the jacobian, those quadric hypersurfaces
occurring as tangent cones to the theta divisor at double points, thus
recovering the canonical model of the curve as their base locus, with
some few exceptions.

To show this works, one can appeal to the deformation theoretic
results of Kempf. i.e. since the italians proved that a canonical
curve is cut out by quadrics most of the time, one needs to know that
the ideal of all quadrics containing the canonical curve is generated
by the ones coming as tangent cones to theta. the ones which do arise
that way cut out the directions in moduli of abelian varieties where
theta remains singular in codimension three.

But these equisingular deformations of theta embed into the
deformations of the resolution of theta by the symmetric product of
the curve, which kempf showed are equal to the deformations of the
curve itself. hence every equisingular deformation of theta(C) comes
from a deformation of C, and these are cut out by the equations in
moduli of abelian varieties defined by quadratic hypersurfaces
containing C. hence the tangent cones to theta determine C.

This version of Green's result is in a paper of smith and varley,
in compositio 1990.

Perhaps the shortest geometric proof is due to andreotti, who
computed the branch locus of the canonical map on the theta divisor,
and showed quite directly it equals the dual variety of the canonical
curve. this is explained in andreotti's paper from about 1958, and
quite nicely too, with some small errata, in arbarello, cornalba,
griffiths, and harris' book on geometry of curves.

There are other short proofs that torelli holds for general
curves, simply from the fact that the quadrics containing the
canonical curve occur as the kernel of the dual of the derivative of
the torelli map from moduli of curves to moduli of abelian varieties.
this is described in the article on prym torelli by smith and varley
in contemporary mat. vol. 312, in honor of c.h. clemens, 2002, AMS.
there is also a special argument there for genus 4, essentially using
zariski's main theorem on the map from moduli of curves to moduli of
jacobians.

There are also inductive arguments, based on the fact that the
boundary of moduli of curves of genus g contains singular curves of
genus g-1, and allowing one to use lower genus torelli results to
deduce degree torelli for later genera.

Then of course there is matsusaka's proof, derived from torelli's
original proof that given an isomorphism of polarized jacobians, the
theta divisor defines the graph of an isomorphism between their
curves.

For shortest most conceptual, I recommend the proof in Arbarello,
Cornalba, Griffiths Harris, i.e. Andreotti's, for conceptualness and
completeness in a reasonably short argument..

p.56. "Grothendieck has banished [\'espace \'etal\'e] from mathematics." Grothendieck uses them in his 1957 Tohoku paper (p. 154-155) but banishes them in 3.1.6, p.25, of Chapter 0 in EGA I.

(Nikita Kozin)

p.78: "Proof: Let M be..." - change M (mathfrak?) to another font

p.80: "Remark 11.7: ... associated with the preseaf ... H^r" - must be "H^s"?

p.81: Diagram - must be "Sh(Y_et)" in left low corner?

p.154. Theorem 3.6 (Existence theorem). The theorem should say that there exists an L/K unramified at the primes not dividing m etc., otherwise the statement doesn't really make sense. (Li Jianing)

**From Kiran S. Kedlaya**

On page 115 of version 4.02, I found footnote 4 a bit confusing: it
seems to be arguing that given a family of sets with empty intersection,
the images under a map also have empty intersection. Of course this
works in this case because of the topology in the picture, but I found
it clearer to reinterpret this argument without taking quotients: since
I is open and the intersection of the compact sets N \cap U_K is
contained in I, the sets

(N \cap U_K) - I

themselves form a family of compact sets with empty intersection. Hence
some finite subfamily has empty intersection, and so on.

In II, Proposition 1.34, p.69, there is no explicit mention of the restrictive
hypothesis on the vanishing of the H^i(H,M), which might be confusing
for the reader who has never seen a spectral sequence before. It of
course gets used to deduce that

0 -> M^H -> M_*^H -> M_{\dagger}^H -> H^1(H,M) = 0

is exact, in order to finish the dimension shift.

p.261. For an improved exposition of the proof of Theorem 5.1, see RG I, 1.29.

p.259. There is a problem is with my definition of "almost-simple". Certainly "almost-simple" should imply "semisimple", so we should define an almost-simple algebraic group to be a noncommutative smooth connected algebraic group that is semisimple (so its radical over the algebraic closure is trivial) and has no nontrivial proper smooth connected normal algebraic subgroup. The quotient of such a group by its centre should be called simple. (The centre of a reductive group is the kernel of the adjoint representation on the Lie algebra. It suffices to check this over an algebraically closed field, see RG 2.12).

If we don't require that G be semisimple, then we get to the land of pseudo-reductive groups (Conrad-Gabber-Prasad), which is very complicated. Tits defines a pseudo-simple algebraic group to be a noncommutative smooth connected affine algebraic group with no nontrivial proper smooth connected normal subgroup. (Sebastian Petersen)

**From Bhupendra Nath Tiwari:** list of misprints for AGS, LAG, RG
odt; doc

**From David Calderbank**

Page 52, Weyl's Theorem 5.20. Part (a) states "If ad is semisimple, then g is
semisimple". However, semisimplicity of the adjoint representation is
characteristic of reductive Lie algebras (Proposition 6.2), not just semisimple
ones. The proof uses more than semisimplicity of the adjoint representation.

Page 56, Proposition 5.29. In the non-algebraically closed case, I don't think it is necessarily true that a Lie algebra with all elements semisimple is commutative. Indeed, so(3) over R would be a counterexample.

Page 57, Proposition 6.4. I don't see how the proof establishes the implication "g has a faithful semisimple representation => g is reductive".

**From Timo Keller**

p. 21: ... with [A,B] = AB - BA,and <- space missing

p. 25, l.-3: ) missing at the end of the equation

I, Theorem 16.21. The centre of a reductive group is of multiplicative type, but need not be reduced (e.g., SL_p). The correct statement is that the radical of G (which is a torus) is the reduced connected group attached to the centre. (There may be other errors of this type, where I incorrectly translated from the characteristic zero case to the general case.) There will be a new version of the notes probably in March 2012.

p. 372 footnote 16. Chapter II?? should be Chapter III.

p. 172, l.1: "and so is split be a finite" should be "and so is split by a finite" (Timo Keller).