contents u bu · 2020. 4. 30. · notes to u-invariants in characteristic 2 zhengyao wu (吴正尧)...
TRANSCRIPT
NOTES TO u-INVARIANTS IN CHARACTERISTIC 2
ZHENGYAO WU (吴正尧)
Abstract. Notes for discussion with Yong Hu (胡勇) and Peng Sun (孙鹏) on
26 May 2018 at SUSTC. Main references: [Bae82b] and [MMW91].
Contents
1. The square number vs u and u 1
2. Henselian discrete valued fields 6
3. Fields with prescribed u and u 14
4. What is known 22
References 25
Throughout this note, let K be a field. For a quadratic form over K, let D(q)
be the set of elements of K∗ represented by q. We avoid 〈 〉 because we reserve
it for diagonal quadratic forms over fields of characteristic not 2. Also, [ ] is
easier to type than
\langle \rangle
Let W (K) be the Witt ring of nonsingular symmetric bilinear forms over K.
Let I = I(K) be the ideal of W (K) of even dimensional forms. Let Wq(K)
be the Witt group of nonsingular quadratic forms over K. We have Wq(K)
is a W (K)-module. The ideal InWq(K) is generated by n-fold Pfister forms
〈〈a1, . . . , an−1〉〉 ⊗ [1, b] for a1, . . . , an−1 ∈ F ∗ and b ∈ F .
1. The square number vs u and u
1.1 Definition
Let q : V → K be a quadratic from with associated symmetric bilinear form bq.
We call q nonsingular if Rad(bq) = 0; regular if Rad(q) = 0.
1.2 Lemma
Suppose charK = 2.1
2 ZHENGYAO WU (吴正尧)
(1) Every nonsingular form over K is regular.
(2) Every anisotropic quadratic form over K is regular (but not necessarily
nonsingular). An anisotrpic Pfister quadratic form over K is nonsingular.
(3) The form [c], c ∈ K∗, i.e. q(x) = cx2 is anisotropic and singular.
(4) The hyperbolic plane [0, 0], i.e. H(x, y) = xy is isotropic and nonsingular.
(5) The form [a, b], a, b ∈ K, i.e. q(x, y) = ax2 + xy + by2 is nonsingular.
(6) Every regular quadratic form has shape
[a1, b1] ⊥ · · · ⊥ [an, bn]︸ ︷︷ ︸nonsingular part
⊥ [c1] ⊥ · · · ⊥ [cm]︸ ︷︷ ︸singular part
, ai, bi ∈ K, cj ∈ K∗
In particular, every nonsingular quadratic form over K has even dimension and
u(K) is even.
1.3 Definition
u(K) = sup{dimK V | ∃ quadratic form q : V → K anisotropic and nonsingular.}u(K) = sup{dimK V | ∃ quadratic form q : V → K anisotropic. }
1.4 Lemma
(1) If charK 6= 2, then u(K) = u(K).
(2) If charK = 2, then u(K) ≤ u(K).
(3) There exists a field K such that charK = 2 and u(K) < u(K).
Proof. (3) Let K be a quadratically closed field of characteristic 2. Then
u(K) = 1. Also every 1-dimensional quadratic form is of the form [c], c ∈ K,
which is singular by Lemma 1.2(3), so u(K) = 0. �
1.5 Example
This is [Bae82b, Rem. 1.2 b)]. Suppose charK = 2. If K is perfect, then either
u(K) = 0 or u(K) = u(K) = 2.
Proof. Suppose u(K) > 0. By Lemma 1.2(5), 2 ≤ u(K). By Lemma 1.4(2),
u(K) ≤ u(K). Since x 7→ x2 is an automorphism of K, [a, b] is universal for all
a, b ∈ K. If q is anisotropic of dimension > 1, then q ' [a, b] for some a, b ∈ Kby Lemma 1.2(6) then u(K) ≤ 2. �
Let K2 be the subfield of square elements of K. We call [K : K2] the square
number of K.
NOTES TO u-INVARIANTS IN CHARACTERISTIC 2 3
1.6 Lemma
This is [Bae82b, Rem. 1.2 b)c)d)]. Suppose charK = 2.
(0) If finite, [K : K2] is a power of 2.
(1) u(K) ≥ [K : K2].
(2) If u(K) = 0, then u(K) = [K : K2].
(3) There exists K such that u(K) > 0 and u(K) > [K : K2].
Proof. (0) Since K/K2 is purely inseparable, we may decompose K/K2 into a
tower of simple extensions of degree 2. For details see [Bou81, Ch. V, § 5, no.
2, Prop. 4].
(1) The form [c1] ⊥ · · · ⊥ [cm], cj ∈ K is anisotropic iff c1, . . . , cm are linearly
independent over K2. By Lemma 1.2(6), u(K) ≥ [K : K2].
(2) When u(K) = 0, by Lemma 1.2(2)(5)(6), every anisotropic quadratic form
has shape [c1] ⊥ · · · ⊥ [cm], cj ∈ K. Then u(K) = [K : K2].
(3) Since [1, 1] is anisotropic over F2, u(F2) = 2 > 1 = [F2 : F22]. �
1.7 Lemma
This is [MMW91, Prop. 1, Cor. 1]. Suppose charK = 2. Suppose q = [a1, b1] ⊥· · · ⊥ [an, bn] ⊥ [c1] ⊥ · · · ⊥ [cm] is anisotropic over K.
(1) n+m ≤ [K : K2].
(2) u(K) ≤ 2[K : K2] (and hence u(K) ≤ 2[K : K2]).
(3) If n+m = [K : K2] and n ≥ 1, then u(K) = u(K) = 2[K : K2].
Proof. (1) For all di ∈ D([ai, bi]), [d1] ⊥ · · · ⊥ [dn] ⊥ [c1] ⊥ · · · ⊥ [cm] is
anisotropic as a subform of q. Then d1, . . . , dn, c1, . . . , cm is linearly independent
over K2.
(2) dim(q) = 2n+m ≤ 2(n+m) ≤ 2[K : K2] by (1).
(3) Our goal is to find an anisotropic nonsingular form of dimension 2n+ 2m.
Then 2[K : K2] ≤ u(K) and the rest follows from (2).
Next, we construct such a form. We have a K2-basis d1, d2, . . . , dn, c1, . . . , cm
of K. Let t : K → K2 be the K2-linear map defined by t(d1) = 1, t(d2) =
· · · = t(dn) = t(c1) = · · · = t(cm) = 0. Then the composition t ◦ [a1, b1] is a
2[K : K2]-dimensional quadratic form over K2. Since [a1, b1] is nonsingular,
t ◦ [a1, b1] is nonsingular.
Finally, we show that t◦ [a1, b1] is anisotropic. If v 6= 0 such that t([a1, b1](v)) =
0, then [a1, b1](v) ∈ SpanK2{d2, . . . , dn, c1, . . . , cm}, a contradiction to the fact
that q is anisotropic. �
4 ZHENGYAO WU (吴正尧)
1.8 Corollary
This is [MMW91, Cor. 2]. Suppose charK = 2. Then u(K) 6= 2n − 1 for all
n ≥ 2.
Proof. If not, then u(K) = 2n − 1. By Lemma 1.6(1) and Lemma 1.7(2), we
have 2n−1 − 0.5 ≤ [K : K2] ≤ 2n − 1. By Lemma 1.6(0), [K : K2] = 2n−1.
Suppose q = [a1, b1] ⊥ · · · ⊥ [ar, br] ⊥ [c1] ⊥ · · · ⊥ [cs] is anisotropic over
K such that 2r + s = 2n − 1. Then r + s ≤ 2n−1. Hence s is odd and
s+2n−1 = 2(r+s) ≤ 2n. Then s = 1, r = 2n−1−1 and r+s = 2n−1 = [K : K2].
It follows from Lemma 1.7(3) that u(K) = 2[K : K2] = 2n, a contradiction. �
1.9 Lemma
This is [Bae82b, p. 107, First paragraph]. Let k be a field of characteristic 2.
If K/k is finite, then
[K : K2] = [k : k2].
Proof. Let (wi)1≤i≤n be a k-basis of K. It follows from the identity (a1w1 +
· · ·+ anwn)2 = a21w
21 + · · ·+ a2
nw2n, (ai ∈ k) that (w2
i )1≤i≤n is a k2-basis of K2.
Then [K : k] = [K2 : k2]. Thus
[K : K2] =[K : k2]
[K2 : k2]=
[K : k][k : k2]
[K2 : k2]= [k : k2].
�
1.10 Corollary
This is [MMW91, Prop. 4(iii)]. Let k be a field. Suppose char k = 2. Let K/k
be an extension such that [K : k] <∞. Then1
2u(k) ≤ u(K) ≤ 2u(k).
Proof.
u(k) ≤ 2[k : k2], by Lemma 1.7(2).
= 2[K : K2], by Lemma 1.9.
≤ 2u(K), by Lemma 1.6(1).
u(K) ≤ 2[K : K2], by Lemma 1.7(2).
= 2[k : k2], by Lemma 1.9.
≤ 2u(k), by Lemma 1.6(1).
�
NOTES TO u-INVARIANTS IN CHARACTERISTIC 2 5
1.11 Lemma
This is [Bae82b, Lem. 1.3 A.1]. Let k be a field of characteristic 2. Suppose
K/k is separable and algebraic. If (wi)1≤i≤n are linearly independent elements
of K over k, then (w2i )1≤i≤n are linearly independent over k.
Proof. Suppose E = k(w1, . . . , wn). Let (wi)1≤i≤m (m ≥ n) be a k-basis of E
containing (wi)1≤i≤n. Let F = Spank(w2i )1≤i≤m. Suppose wiwj =
m∑l=1
cijlwl for
cijl ∈ k. Suppose a =m∑i=1
aiw2i and b =
m∑i=1
biw2i .
• a± b =m∑i=1
(ai ± bi)w2i ∈ F ;
• ab =m∑l=1
(m∑
i,j=1
aibjc2ijl
)wl ∈ F ;
• For a ∈ F ∗, 1a
= a( 1a)2 where ( 1
a)2 ∈ E2 ⊂ F and hence 1
a∈ F .
Thus F is a subfield of E.
Finally, since E/k is separable, E/F is separable. Then E = F since E/E2 is
purely inseparable. Then (w2i )1≤i≤m are linearly independent over k. Therefore
its subset (w2i )1≤i≤n are linearly independent over k. �
1.12 Lemma
This is [Bae82b, Lem. 1.3 A.2]. Let k be a field of characteristic 2. Suppose
K/k is separable and algebraic. If (ei)1≤i≤n are linearly independent elements
of k over k2, then they are also linearly independent over K2.
Proof. Suppose a2i ∈ K2 (ai ∈ K) such that a2
1e1 + · · · + a2nen = 0. We want
to show that a2i = 0 (i.e. ai = 0) for all 1 ≤ i ≤ n. Suppose E = k(a1, . . . , an).
Let (wi)1≤i≤m be a k-basis of E. Then ai =m∑j=1
bijwj for bij ∈ k. Then
0 =n∑i=1
a2i ei =
n∑i=1
(m∑j=1
bijwj
)2
ei =m∑j=1
(n∑i=1
b2ijei
)w2j
By Lemma 1.11, we haven∑i=1
b2ijei = 0 for all 1 ≤ j ≤ m and hence bij = 0 since
(ei)1≤i≤n are linearly independent over k2. Therefore ai = 0 for all 1 ≤ i ≤n. �
1.13 Lemma
This is [Bae82b, Lem. 1.3]. Let k be a field of characteristic 2. If K/k is
6 ZHENGYAO WU (吴正尧)
separable and algebraic, then
[K : K2] = [k : k2].
Proof. Let (wi)1≤i≤n be elements of K linearly independent over K2. Then
(wi)1≤i≤n are linearly independent over E2 where E = k(w1, . . . , wn) is subfield
of K such that E/k is finite. Then [K : K2] ≤ [E : E2]. By Lemma 1.9,
[E : E2] = [k : k2]. Hence [K : K2] ≤ [k : k2].
Conversely, let (ei)1≤i≤n be elements of k linearly independent over k2. By
Lemma 1.12, (ei)1≤i≤n are linearly independent over K2. Hence [k : k2] ≤ [K :
K2]. �
1.14 Example
This is [Bae82b, p.108, two paragraphs before § 2]. Let k = F2(X1, X2, . . . , Xn)
(resp. k = F2(X1, X2, . . .)). Then [k : k2] = 2n (resp. ∞). Let K be the
quadratic separable closure of k, i.e. elements in K − k are in an algebraic
closure of k whose minimal polynomial is a quadratic separable. It follows
from Lemma 1.13 that [K : K2] = [k : k2] = 2n (resp.∞). By Lemma 1.4(3),
u(K) = 0. By Lemma 1.6(2), u(K) = 2n (resp.∞).
2. Henselian discrete valued fields
Let (K, v) be a Henselian discrete valued field with valuation ring A, uni-
formizer π and residue field K = A/πA.
2.1 Lemma
If f(T ) = a2T2 + a1T + a0 is irreducible in K[T ], then 2v(a1) ≥ v(a0) + v(a2).
Proof. Suppose ai = uiπni for i ∈ {0, 1, 2}. If 2n1 < n0 + n2, then
g(T ) = πn2−2n1f(πn1−n2T ) = u2T2 + u1T + u0π
n0+n2−2n1 .
Hence g(T ) = T (u2T + u1). By Hensel’s lemma, g(T ) is reducible and thus
f(T ) is reducible, a contradiction. �
2.2 Lemma
Schwarz inequality. This is [MMW91, p. 342 (4)]. Let q : V → K be an
anisotropic quadratic form with associated symmetric bilinear form bq. Then
2v(bq(x, y)) ≥ v(q(x)) + v(q(y)) for all x, y ∈ V . This is true regardless of
charK and charK,
NOTES TO u-INVARIANTS IN CHARACTERISTIC 2 7
Proof. Suppose x, y are linearly independent. Since q is anisotropic, f(T ) =
q(xT+y) = q(x)T 2+bq(x, y)T+q(y) is irreducible. By Lemma 2.1, 2v(bq(x, y)) ≥v(q(x)) + v(q(y)).
Suppose x, y are linearly dependent. Then v(bq(x, y)) = v(0) = +∞. �
2.3 Lemma
This is [Bae82b, After Lem. 2.3]. Let q : V → K be an anisotropic quadratic
form. Let
Mi = {x ∈ V | v(q(x)) ≥ i}, i ≥ 0.
Then Mi+1 ⊂Mi and
(1) Mi is an A-module.
(2) Mi+2 = πMi for all i ≥ 0.
(3) If x ∈Mi and y ∈Mi+1, then v(bq(x, y)) ≥ i+ 1.
Proof. For all a ∈ K and x ∈Mi, v(q(x)) ≥ i. We have
v(q(ax)) = v(a2q(x)) = 2v(a) + v(q(x)).
(1) For all a ∈ A v(a) ≥ 0. Then v(q(ax)) ≥ i and hence ax ∈Mi.
(2) For all x ∈ Mi, v(q(πx)) = 2 + v(q(x)) ≥ i + 2 and hence πx ∈ Mi+2.
Conversely, for all y ∈ Mi+2, then v(π−1y) = −2 + v(q(y)) ≥ −2 + (i + 2) = i
and hence y ∈Mi.
(3) By (2) symmetric, it suffices to assume i = 0. By Lemma 2.2, v(bq(x, y)) ≥v(q(x)) + v(q(y))
2≥ 0 + 1
2=
1
2and thus v(bq(x, y)) ≥ 1. �
2.4 Lemma
This is [Bae82b, After Lem. 2.3]. Let q : V → K be an anisotropic quadratic
form. Let V0 = M0/M1 and V1 = M1/M2. They are vector spaces over K. Let
q0(x) = q(x) for all x ∈ M0. Let q1(x) =
(q(x)
π
)for all x ∈ M1. We call q0
and q1 residue forms of q. Then (Vi, qi) are anisotropic over K for i ∈ {0, 1}.
Proof. Suppose q1(x) = 0 in K for some x ∈M0. Then v(q(x)) ≥ 1, i.e. x ∈M1.
Hence x = 0 in V0 and thus q0 is anisotropic.
Suppose q1(x) = 0 in K for some x ∈ M1. Then v
(q(x)
π
)≥ 1, v(q(x)) ≥ 2,
i.e. x ∈M2. Hence x = 0 in V1 and thus q1 is anisotropic. �
8 ZHENGYAO WU (吴正尧)
2.5 Lemma
This is [Bae82b, p. 109-111(2.6)-(2.9)]. Let ai, bi ∈ A for all 1 ≤ i ≤ n and
uj ∈ A∗ for all 1 ≤ j ≤ m. Suppose
q = [a1, b1] ⊥ · · · ⊥ [an, bn] ⊥ [u1] ⊥ · · · ⊥ [um]
is anisotropic over K. Let q0 = [a1, b1] ⊥ · · · ⊥ [an, bn]. Then
q = q0 ⊥q0
π⊥ [u1,
u1
π] ⊥ · · · ⊥ [um,
umπ
]
is anisotropic and nonsingular over K.
Proof. By Lemma 1.4(5), q is nonsingular. If q is isotropic, then q0(s)+q0(t)
π+
m∑j=1
ujx2j +xjyj +
ujπy2j = 0. Suppose s = x0s
′ and t = y0t′ such that x0, y0 ∈ K
and s′ 6≡ 0 modπ, t′ 6≡ 0 modπ. Then ?
πx20q0(s′) + y2
0q0(t′) +m∑j=1
πx2juj + πxjyj + y2
juj = 0
Here xj , yj (0 ≤ j ≤ m) are not all 0 in K. Since q is anisotropic, we must have
min(v(πx20), v(y2
0), v(πx2j), v(y2
j ), v(πxjyj)(1 ≤ j ≤ m)) > 0. Three cases:
(1) min = πx2k for some 0 ≤ k ≤ m;
(2) min = y2k for some 0 ≤ k ≤ m;
(3) min = πxkyk for some 1 ≤ k ≤ m.
(1) Dividing ? by min = πx2k, we have(
x0
xk
)2
q0(s′)+1
π
(y0
xk
)2
q0(t′)+m∑j=1
(xjxk
)2
uj+
(xjxk
)(yjxk
)+
1
π
(yjxk
)2
uj = 0
Here v
(xjxk
)=v(πx2
j)− v(πx2k)
2≥ 0; v
(1
π
(yjxk
)2)
= v(y2j ) − v(πx2
k) ≥ 0
and thus ≥ 1 since v(y2j ) is even and v(πx2
k) is odd.
Modulo π, we have (x0
xk
)2
q0(s′) +m∑j=1
(xjxk
)2
uj = 0.
Since
(x0
x0
)s′ = s′ 6= 0 if k = 0; or
(xkxk
)= 1 if 1 ≤ k ≤ m, we have((
x0
xk
)s′,
(x1
xk
), . . . ,
(xmxk
))6= 0,
NOTES TO u-INVARIANTS IN CHARACTERISTIC 2 9
a contradiction to the fact that q is anisotropic.
(2) Dividing ? by min = y2k, we have
π
(x0
yk
)2
q0(s′)+
(y0
yk
)2
q0(t′)+m∑j=1
π
(xjyk
)2
uj+π
(xjyk
)(yjyk
)+
(yjyk
)2
uj = 0
Here v
(yjyk
)=
v(y2j )− v(y2
k)
2≥ 0; v
(π
(xjyk
)2)
= v(πx2j) − v(y2
k) ≥ 0
and thus ≥ 1 since v(πx2j) is odd and v(y2
k) is even; v
(π
(xjyk
)(yjyk
))=
1 + v(πx2j) + v(y2
j )
2− v(y2
k) ≥1
2and thus ≥ 1.
Modulo π, we have (y0
yk
)2
q0(t′) +m∑j=1
(yjyk
)2
uj = 0.
Since
(y0
y0
)t′ = t′ 6= 0 if k = 0; or
(ykyk
)= 1 if 1 ≤ k ≤ m, we have((
y0
yk
)t′,
(y1
yk
), . . . ,
(ymyk
))6= 0,
a contradiction to the fact that q is anisotropic.
(3) min = πxkyk. Then
−1 ≥ −v(π) + v(πxkyk)− v(y2k) = v
(xkyk
)= v(πx2
k)− v(πxkyk) ≥ 0
a contradiction. �
2.6 Definition
Let B be a valuation ring with field of fractions F = Frac(B). Let M be
the maximal ideal of B with residue field F = B/M . Let v : F ∗ → Γ be the
valuation to an ordered group. Let V be a finite dimensional vector space
over F . Let P be a full B-lattice in V , i.e. a B-module such that P ⊂ V and
FP = V . The residue defect of P is
rd(P ) = dimF V − dimF (P/MP ).
2.7 Lemma
This is [MMW91, Prop. 5(i)(ii)].
(1) rd(P ) ≥ 0.
(2) rd(P ) = 0 iff P is a finitely generated B-module.
10 ZHENGYAO WU (吴正尧)
Proof. (1) Suppose p1, . . . , pk ∈ P such that p1, . . . , pk ∈ P/MP are linearly
independent over F . We show that p1, . . . , pk are linearly independent over F
and thus dimF V ≥ dimF (P/MP ).
If not, then there exists ci ∈ F , 1 ≤ i ≤ k not all zero andk∑i=1
cipi = 0. Suppose
v(cj) = min1≤i≤k
(v(ci)). Then cj 6= 0 and∑i 6=j
cicjpi+pj = 0, where
cicj∈ B. Modulo
M , we have∑i6=j
cicjpi + pj = 0, a contradiction to the fact that p1, . . . , pk are
linearly independent over F .
(2) If P is a finitely generated B-module, then P is a free over B of finite rank
by [Bou64, Ch. VI, § 4, no. 1, Lem. 1]. Suppose P ' Bn. Then FP ' Fn
and P/MP ' P ⊗B B/M ' Bn ⊗B F ' Fn. Hence rd(P ) = dimF V −
dimF (P/MP ) = n− n = 0.
Conversely, suppose (p1, . . . , pk) is a F -basis of P/MP . By (1), (p1, . . . , pk)
are linearly independent over B (then over F ). Since rd(P ) = 0, we have
dimF V = k and hence (p1, . . . , pk) is a F -basis of V . It suffices to show that
(p1, . . . , pk) span P .
If not, then there existsk∑i=1
cipi ∈ P − (Bp1 ⊕ · · · ⊕ Bpk) such that v(cj) =
min1≤i≤k
(v(ci)) < 0. Then1
cj∈ M ,
∑i 6=j
cicjpi + pj =
1
cj
k∑i=1
cipi ∈ MP . Modulo M ,
we have∑i 6=j
cicjpi+pj = 0, a contradiction to the fact that p1, . . . , pk are linearly
independent over F . �
2.8 Lemma
This is [MMW91, Prop. 5(iii)].
Let W ⊂ V be an F -subspace. Let Q = P∩W . Then rd(P ) = rd(Q)+rd(P/Q).
Proof. Since P/Q = P/P ∩W ' (P + W )/W ⊂ V/W , we may assume that
P/Q ⊂ V/W . Since FQ = F (P ∩W ) = FP ∩W = V ∩W = W and F is
flat over B, we have F (P/Q) ' FP/FQ = V/W . Then P/Q is torsion-free
over B. By [Bou64, Ch. VI, § 4, no. 1, Lem. 1] again, P/Q is flat over B. By
[Bou07, Ch. X, § 4, no. 4, Cor. of Prop. 5], TorB1 (B/M,P/Q) = 0. From the
exact sequence 0→ Q→ P → P/Q→ 0, we have another exact sequence
0→ Q/MQ→ P/MP → P/Q
M(P/Q)→ 0.
NOTES TO u-INVARIANTS IN CHARACTERISTIC 2 11
Therefore
rd(Q) + rd(P/Q) = dimF W − dimF (Q/MQ) + dimF (V/W )− dimF
P/Q
M(P/Q)
= dimF V − dimF (P/MP ) = rd(P ).
�
2.9 Definition
Let (K, v) be a Henselian discrete valued field with valuation ring A, uni-
formizer π and residue field K = A/πA. Let q : V → K be an anisotropic
quadratic form. Let q0 and q1 be residue forms of q as in Lemma 2.4. The
residue defect of q is
rd(q) = dimK q − dimK q0 − dimK q1.
2.10 Lemma
This is [MMW91, p. 342 (5)]. rd(q) = rd(M0).
Proof. First KM0 = V . Then dimK V = dimK KM0.
Also, by Lemma 2.3,
dimK q0 + dimK q1 = dimK(M0/M1) + dimK(M1/M2)
= dimK(M0/M1) + dimK(M1/πM0) = dimK(M0/πM0).
Therefore
rd(q) = dimK q−dimK q0−dimK q1 = dimK KM0−dimK(M0/πM0) = rd(M0).
�
2.11 Lemma
This is [MMW91, Th. 1(a)]. Let (K, v) be a Henselian discrete valued field with
valuation ring A, uniformizer π and residue field K = A/πA. Let q : V → K
be an anisotropic quadratic form. If charK = 2 and q is totally singular, then
rd(q) ≤ [K : K2]− 2[K : K2].
12 ZHENGYAO WU (吴正尧)
Proof. Since q(cy) = c2q(y) for all c ∈ K and q is anisotropic, we have
dimK V = dimK2 D(q). Similarly, dimK(M0/πM0) = dimK
2(q(M0)/π2q(M0)).
rd(q) = rd(M0), by Lemma 2.10.
= dimK V − dimK(M0/πM0)
= dimK2 D(q)− dimK
2(q(M0)/π2q(M0)) Since q is a totally singular.
= rd(q(M0)) = rd(A)− rd(A/q(M0)), by Lemma 2.8.
≤ rd(A) = [K : K2]− dimK
2(A/π2A)
= [K : K2]− 2[K : K2], since dimK(A/π2A) = 2.
�
2.12 Theorem
This is [MMW91, Th. 1(b)]. Let (K, v) be a Henselian discrete valued field with
valuation ring A, uniformizer π and residue field K = A/πA. Let q : V → K
be an anisotropic quadratic form.
(1) If charK = 2, then
rd(q) = rd(Rad(bq)) ≤ [K : K2]− 2[K : K2].
(2) If charK 6= 2, then rd(q) = 0.
Proof. Let y1, . . . , yk be a K-basis of V . Since KM0 = V , we may assume that
yi ∈ M0. It follows from Lemma 2.2 that bq(z, yi) ∈ A for all 1 ≤ i ≤ k and
z ∈ M0. Let E = Ay1 ⊕ · · · ⊕ Ayk. Then γ : M0 → E∗ = HomA(E,A) such
that γ(z) = bq(z, •) is A-linear. Then ker(γ) = Rad(bq) ∩M0. Since E∗ is
a free A-module of finite rank, its sub-A-module Im(γ) is also free and thus
rd(Im(γ)) = 0. Then M0 ' ker(γ)⊕ Im(γ) ' Rad(bq) and
rd(q) = rd(Rad(bq) ∩M0) + rd(Im(γ)) = rd(Rad(bq)).
(1) Suppose charK = 2. Since q|Rad(bq) is totally singular, by Lemma 2.11,
rd(q) = rd(Rad(bq)) ≤ [K : K2]− 2[K : K2].
(2) Suppose charK 6= 2. Since q is anisotropic, it is nonsingular and hence
Rad(bq) = 0. Therefore rd(q) = rd(Rad(bq)) = 0. �
2.13 Theorem
This is [MMW91, Th. 2]. Let (K, v) be a Henselian discrete valued field with
NOTES TO u-INVARIANTS IN CHARACTERISTIC 2 13
valuation ring A, uniformizer π and residue field K = A/πA. Suppose charK =
charK = 2.
(1) u(K) = 2u(K) (analogous to Springer’s theorem for charK 6= 2).
(2) max{2u(K), [K : K2]} ≤ u(K) ≤ 2u(K) + [K : K2]− 2[K : K2].
Proof. First, it follows from Lemma 2.5 that there exists an anisotropic nonsin-
gular quadratic form q over K such that q0 ' q1 ' q and dimK q = 2 dimK q.
So 2u(K) ≤ u(K).
Next, let q be an anisotropic form over K. By Lemma 2.4, q0 and q1 are
anisotropic over K. Then
dimK q = dimK q0 + dimK q1 + rd(q) ≤ 2u(K) + rd(q).
(1) Suppose q is nonsingular and anisotropic. Then rd(q) = 0 and thus u(K) ≤2u(K).
(2) By Theorem 2.12, u(K) ≤ 2u(K) + rd(q) ≤ 2u(K) + [K : K2]− 2[K : K2].
By Lemma 1.4(2), we have u(K) ≤ u(K) and hence 2u(K) ≤ u(K).
By Lemma 1.7(1), we have [K : K2] ≤ u(K). �
2.14 Example
Let k0 be an algebraically closed field of char k0 = 2. Let k1 = k0(t1, . . . , tn−1)
be the rational function field over k0 with n−1 variables. Let L = k1(s1, . . . , sm−n+1)
where si ∈ k1((X)) and (s1, . . . , sm−n+1) are algebraically independent. Let v
be the restriction of the X-adic valuation on k1((X)) to L. Let K be the
Henselization of L relative to v. Then K = L = k1. By Theorem 2.13(1),
u(K) = 2u(k1) = 2n.
2m = [L : L2], since tr.deg.(L/k0) = m.
= [K : K2], by Lemma 1.13.
≤ u(K), by Lemma 1.6(1).
≤ 2m by Tsen-Lang theorem and k0 is C0.
2.15 Theorem
This is [Bae82b, Th. 1.1] and [MMW91, Cor. 3]. Let (K, v) be a complete
discrete valued field with residue field K. Suppose charK = 2. Then u(K) =
u(K) = 2u(K).
14 ZHENGYAO WU (吴正尧)
Proof. By Theorem 2.13(1), 2u(K) = u(K). By Lemma 1.4(2), u(K) ≤ u(K).
Since K is complete, we have [K : K2] = 2[K : K2]. By Theorem 2.13(2),
u(K) ≤ 2u(K). �
2.16 Corollary
This is [Bae82b, Cor. 2.10]. Let k be a field of characteristic 2. Let K =
k((T1))((T2)) . . . ((Tn)). Then u(K) = u(K) = 2nu(k).
Proof. For n = 1, use k((T1)) = k and Theorem 2.15. For n ≥ 2, use induction.
�
3. Fields with prescribed u and u
3.1 Review
Cassels-Pfister theorem ([EKM08, Th. 17.3] without assumption on charK).
Let q : V → K be a quadratic form. If a polynomial f ∈ K[T ] is represented
by qK(T ), then f is also represented by qK[T ].
3.2 Review
A field k is Cn if every homogeneous form of degree d in > dn variables has a
nontrivial zero. In paticular, if a field k is Cn, then every quadratic form in
> 2n variables is isotropic and hence u(k) ≤ 2n.
Chevalley-Warning theorem: If k is finite, then it is C1.
Tsen-Lang theorem: If k is Cn, then k(T ) is Cn+1.
3.3 Lemma
This is [Bae82b, Eg. 2.11]. Let k be a field. Suppose char k = 2. Let K =
k(T1, . . . , Tn) be the function field in n variables.
(1) The n-fold Pfister form qn = 〈〈T1, . . . , Tn−1〉〉 ⊗ [1, Tn] is anisotropic and
nonsingular.
(2) Suppose u(k) = 2. If a ∈ k − ℘(k), then the (n + 1)-fold Pfister form
qn = 〈〈T1, . . . , Tn〉〉 ⊗ [1, a] is anisotropic and nonsingular.
(3) Suppose u(k) = 0 and u(k) = 2. If a ∈ k− k2, then the (n+ 1)-fold Pfister
form qn = 〈〈a, T1, . . . , Tn−1〉〉 ⊗ [1, Tn] is anisotropic and nonsingular.
Proof. The form qn is nonsingular because it is orthogonal sums of forms the
shape [a, b] since it has a tensor factor [1, Tn] (resp. [1, a], [1, Tn]).
NOTES TO u-INVARIANTS IN CHARACTERISTIC 2 15
(1) We use induction to show that qn is anisotropic for all n ≥ 1. First, [1, T1]
is anisotropic since T1 6∈ ℘(K) where ℘(x) = x2 + x for all x ∈ K.
Suppose q′ = 〈〈T2, . . . , Tn−1〉〉[1, Tn] is anisotropic over K. If qn is isotropic,
then it is hyperbolic as it is Pfister. Since qn = q′ ⊥ T1q′, we have q′ ' T1q
′.
Then q′ represents T1 over k(T2, . . . , Tn)(T1). By the Cassels-Pfister theorem,
Then q′ represents T1 over k(T2, . . . , Tn)[T1]. Suppose q′(f1, . . . , f2n−1) = T1
where fi ∈ k(T2, . . . , Tn)[T1] for all 1 ≤ i ≤ 2n − 1. If T1|fi for all i, then
T 21 |q′(f1, . . . , f2n−1) = T1, a contradiction. Take T1 = 0, (fi(0))1≤i≤2n−1 are
not all 0. Then q′ is isotropic over k(T2, . . . , Tn). Hence q′ is isotropic over K,
a contradiction.
(2) Similar proof for q0 = [1, a] and q′ = 〈〈X2, . . . , Tn〉〉 ⊗ [1, a].
(2) Similar proof for q1 = 〈〈a〉〉⊗[1, Tn] and q′ = 〈〈a,X2, . . . , Tn−1〉〉⊗[1, Tn]. �
3.4 Example
This is [Bae82b, Eg. 2.11]. Let k be a field. Suppose char k = 2. Let K =
k(T1, . . . , Tn) be the function field in n variables.
(1) If k is algebraically closed, then u(K) = u(K) = 2n.
(2) If k is C1 such that u(k) = 2 or u(k) = 0 and u(k) = 2, then u(K) =
u(K) = 2n+1.
Proof. (1) By Lemma 3.3(1), the n-fold Pfister form 〈〈X1, . . . , Xn−1〉〉⊗ [1, Xn]
is anisotropic and nonsingular over K, then 2n ≤ u(K). By Lemma 1.4(2),
u(K) ≤ u(K). Since k is a C0-field, by the Tsen-Lang theorem, K is a Cn-
field. Then u(K) ≤ 2n.
(2) If u(k) = 2, then there exists some anisotropic nonsingular form [c, d], we
take a = dc∈ k − ℘(k) (resp. u(k) = 0 and u(k) = 2, by Lemma 1.4(3), k is
not quadratically closed, we take a ∈ k − k2). By Lemma 3.3(2), the (n + 1)-
fold Pfister form 〈〈X1, . . . , Xn〉〉 ⊗ [1, a] (resp. 〈〈a, T1, . . . , Tn−1〉〉 ⊗ [1, Tn] ) is
anisotropic and nonsingular over K, then 2n+1 ≤ u(K). By Lemma 1.4(2),
u(K) ≤ u(K). Since k is a C1-field, by the Tsen-Lang theorem, K is a Cn+1-
field. Then u(K) ≤ 2n+1. �
3.5 Definition
(1) Suppose charK 6= 2. Let ν(K) = min{n | InWq(K) is torsion-free.}.(2) Suppose charK = 2. Let
ν(K) = min{n | In+1Wq(K) = 0}
16 ZHENGYAO WU (吴正尧)
= sup{n | ∃n-fold anisotropic Pfister form over K.}.
3.6 Theorem
(1) If L/K is finite separable, then ν(K) ≤ ν(L) ≤ ν(K) + 1.
(2) There exists L/K finite separable such that ν(L) = ν(K) + 1.
(3) If L/K is finite purely inseparable, then ν(K) = ν(L).
We omit its proof since we care more about u and u, see [AB89].
3.7 Example
This is [MMW91, Prop. 2]. Let k be a field. Suppose char k = 2. Let K/k be
a finitely generated field extension such that tr.deg.(K/k) = n.
(1) If k is algebraically closed, then u(K) = u(K) = [K : K2] = 2n.
(2) If k is finite, then u(K) = u(K) = 2[K : K2] = 2n+1.
Proof. We first show that [K : K2] = 2n. Let (X1, . . . , Xn) be a transcendence
basis of K/k. Let L = k(X1, . . . , Xn). It follows from [Bou81, Ch. V, § 14, no.
7, Prop. 17] that K/L is finite. By Lemma 1.9, [K : K2] = [L : L2] = 2n.
(1) If k is algebraically closed, then by Lemma 3.3(1), 〈〈X1, . . . , Xn−1, Xn]] is
anisotropic over L. Then
2n ≤ 2ν(L), since ν(L) ≥ n.≤ 2ν(K), by Theorem 3.6 ν(L) ≤ ν(K).
≤ u(K), by Definition 3.5.
≤ u(K), by Lemma 1.6(1).
≤ 2n, by Tsen-Lang theorem and k is C0.
(2) If k is finite, then u(k) = 2. By Lemma 3.3(2), 〈〈X1, . . . , Xn−1, Xn, a]] is
anisotropic over L for a ∈ k − ℘(k). Then
2n+1 ≤ 2ν(L), since ν(L) ≥ n+ 1.
≤ 2ν(K), by Theorem 3.6 ν(L) ≤ ν(K).
≤ u(K), by Definition 3.5.
≤ u(K), by Lemma 1.6(1).
≤ 2n+1, by Tsen-Lang, Chevalley-Warning theorems.
�
3.8 Definition
A field extension K/k is separably generated if tr.deg.(K/k) = n and there
NOTES TO u-INVARIANTS IN CHARACTERISTIC 2 17
exists a transcendence basis (X1, . . . , Xn) of K/k such that K is separable and
algebraic over k(X1, . . . , Xn).
3.9 Example
This is [MMW91, Prop. 3]. Let k be an algebraically closed field. Suppose
char k = 2. LetK/k be a separably generated extension with tr.deg.(K/k) = n.
Let X be an indeterminate. Then u(K(X)) = u(K(X)) = 2[K : K2] = 2n+1.
Proof. Let (X1, . . . , Xn) be a transcendence basis ofK/k. Let L = k(X1, . . . , Xn).
By Lemma 1.13, [K : K2] = [L : L2] = 2n. Let a1, . . . , an be a 2-basis of
K/k. Then 〈〈a1, . . . , an〉〉 is anisotropic over K. Since X is an indeterminate,
〈〈a1, . . . , an, X]] is anisotropic over K(X). Then
2n+1 ≤ 2ν(K(X)), since ν(K(X)) ≥ n.≤ u(K(X)), by Definition 3.5.
≤ u(K(X)), by Lemma 1.6(1).
≤ 2n+1, by Tsen-Lang theorem and k is C0.
�
3.10 Open Problem
By Corollary 2.16, for all field K such that charK = 2 we have
u(K((X))) = u(K((X))) = 2u(K).
Does u(K(X)) = u(K(X)) for all K?
3.11 Lemma
This is [MMW91, p. 340, line 4]. Suppose charK = 2. Then for every simple
algebraic extension L/K, u(K(X)) ≥ u(L).
Proof. Suppose π ∈ K[X] is irreducible such that L ' K[X]/(π). Let E be
the (π)-adic completion of K(X). Then u(K(X)) ≥ u(E). By Theorem 2.15,
u(E) = u(E) = 2u(L). �
We have2u(K) ≤ u(K(X)), by Lemma 3.11.
≤ u(K(X)), by Lemma 1.4(2).
≤ 2[K(X) : K(X)2], by Lemma 1.7(2).
= 4[K : K2]
≤ 4u(K), by Lemma 1.6(1).
18 ZHENGYAO WU (吴正尧)
3.12 Lemma
This is [Mor84, Lem. 1]. Suppose charK = 2. Let L/K be a cyclic extension
and Gal(L/K) has generator s. Take β ∈ L − ℘(L). Let M = L(α) where
℘(α) = α2 + α = β. If there exists γ ∈ L such that βs + β = γ2 + γ and
TrL/K(γ) 6= 0, then M/K is cyclic. (The converse is also true [Alb37, p.197,
Th. 3], but is irrelavant to this note.)
Proof. Extend s to t : M →M such that t|L = s and t(α) = α+ γ. The map t
is well-defined since
℘(t(α)) = (α+ γ)2 + (α+ γ) = (α2 + α) + (γ2 + γ) = β + (βs + β) = βs.
Suppose [L : K] = 2e. Then the order of t in Gal(M/K) is a multiple of 2e.
t[L:K](α) = α+ (γ + s(γ) + · · ·+ s2e−1(γ)) = α+ TrL/K(γ) 6= α.
Then the order of t is > 2e and a factor of 2e+1 = |Gal(M/K)|, so it can only
be 2e+1. Thus Gal(M/K) is generated by t and hence cyclic. �
3.13 Lemma
This is [Mor84, Lem. 2]. Suppose charK = 2 and [K : ℘(K)] = 2. Take
β ∈ K−℘(K). LetM = K(α) where ℘(α) = α2+α = β. Then [M : ℘(M)] = 2.
Proof. First, ℘(M) = K ⊕ ℘(K)α. For all x+ yα ∈M , x, y ∈ K,
℘(x+ yα) = x2 + y2α2 + x+ yα = (x2 + x+ y2β) + (y2 + y)α ∈ K ⊕ ℘(K)α.
Conversely, ℘(K) ( ℘(K) + F2β ⊂ ℘(K)(β) = K. We have ℘(K) + F2β = K
since [K : ℘(K) + F2β] < [K : ℘(K)] = 2.
Suppose z + ℘(y)α ∈ K ⊕ ℘(K)α where y, z ∈ K. Suppose z = ℘(w) + nβ for
w ∈ K, n ∈ {0, 1}. Suppose y2β = ℘(v) +mβ for v ∈ K, m ∈ {0, 1}.If m = n, then
℘(v + w + yα) = ℘(v) + ℘(w) + ℘(y)α+ y2β = z + ℘(y)α.
If m 6= n, then
℘(v + w + (y +m+ n)α) = ℘(v) + ℘(w) + ℘(y +m+ n)α+ (y2 +m2 + n2)β
= ℘(v) + ℘(w) + ℘(y)α+ (y2 +m+ n)β = z + ℘(y)α.
Hence ℘(M) = K ⊕ ℘(K)α.
NOTES TO u-INVARIANTS IN CHARACTERISTIC 2 19
Next, the map
f : E = K(α) → K/℘(K)
x+ yα 7→ y + ℘(K), (x, y ∈ K).
is a sujective additive group homomorphism with ker(f) = K⊕℘(K)α = ℘(M).
Therefore M/℘(M) ' K/℘(K) and [M : ℘(M)] = [K : ℘(K)] = 2. �
3.14 Definition
Suppose charK = 2. Let Ks be the separable closure of K. The 2-separable
closure K2,s of K is the direct limit of finite separable extensions of K in Ks
with 2-power degree.
3.15 Theorem
This is [Mor84, Th. 1]. Suppose charK = 2 and α ∈ K − ℘(K). The following
are equivalent:
(1) If L is an extension of K such that α 6∈ ℘(L), then L = K.
(2) [K : ℘(K)] = 2.
(3) Every finite separable extension of K with a 2-power degree is cyclic.
(4) Gal(K2,s/K) ' Z2.
Proof. (1) implies (2). Since α 6∈ ℘(K), we have [K : ℘(K)] ≥ 2.
For all β ∈ K − ℘(K), let L = K(b) where ℘(b) = β. If α+ ℘(K) 6= β + ℘(K),
then α 6∈ ℘(L). By (1), L = K and hence β ∈ ℘(K), a contradiction. Therefore
α+ ℘(K) = β + ℘(K) and [K : ℘(K)] = 2.
(2) implies (3). Let L/K be a finite separable subextension of K2,s/K with a 2-
power degree. By the primitive element theorem, L = K(λ). Then the splitting
field of λ, i.e. the Galois closure L is finite over L. Suppose [L : K] = 2m. We
have a tower of fields
K = L0 ⊂ L1 ⊂ · · · ⊂ Lm = L
such that [Li : Li−1] = 2 for all 1 ≤ i ≤ m.
Next, we use induction to show that Li/K is cyclic and [Li : ℘(Li)] = 2. For
i = 0, L0 = K and [K : ℘(K)] is given by (2). Suppose it is true for i−1. Take
β ∈ Li−1 − ℘(Li−1) such that Li = Li−1(δ) and ℘(δ) = β. Here (β, δ) can be
replaced by (β+℘(x), δ+x) for all x ∈ Li−1. It follows from Lemma 3.13 that
[Li : ℘(Li)] = 2.
20 ZHENGYAO WU (吴正尧)
We have Li−1 = ℘(Li−1) + F2β by [Li−1 : ℘(Li−1)] = 2. Suppose Gal(Li−1/K)
is generated by s. Then βs = ℘(γ) + nβ for some γ ∈ Li−1 and n ∈ {0, 1}.Since βs
2i−1
= β, we have n = 1 and hence βs + β = γ2 + γ. Since Li−1/K is
separable, TrLi−1/K 6= 0 (see [Bou81, Ch. V, § 8, no. 2, Cor. to Prop. 1]). Since
(β, δ) can be replaced by (β+℘(x), δ+x) for all x ∈ Li−1, we may assume that
TrLi−1/K(γ) 6= 0. By Lemma 3.12, we have Li/K is cyclic for all 1 ≤ i ≤ m.
In particular, L = Lm is cyclic over K and thus abelian. Then the subgroup
Gal(L/L) of Gal(L/K) is normal, or equivalently, L/K is Galois. Therefore
L = L and L/K is cyclic.
(3) implies (4). K2,s = lim−→
L where L runs through finite separable extensions
of K with 2-power degrees. By (3), Gal(L/K) ' Z/2nZ for [L : K] = 2n.
Gal(K2,s/K) = lim←−
Gal(L/K) ' lim←−
Z/2nZ ' Z2.
(4) implies (2) and (1). First, we show (2). If [K : ℘(K)] > 2, then take
α1, α2 ∈ K−℘(K) such that α1−α2 6∈ ℘(K). Suppose ℘(δi) = αi for i ∈ {1, 2}.Then Gal(K(δ1, δ2)/K) ' Z/2Z× Z/2Z, a contradiction to (4).
Next, we show (1). Suppose L/K and α ∈ K − ℘(L). If K ( L, then there
exists δ ∈ L −K such that ℘(δ) = α. Let L1 = K(δ). Then [L1 : K] = 2 and
by the proof of Lemma 3.13, ℘(L1) = K ⊕ ℘(K)δ. Then K ⊂ ℘(L1) ⊂ ℘(L), a
contradiction. Therefore K = L. �
3.16 Lemma
This is [Sal77, Th. 3]. Let K be a field of charK = 2. If [K : ℘(K)] <∞, then
Br2(K) = 0.
Proof. It suffices to show that every quarternin algebra (a, b] splits over K.
Here (a, b] is generated by 1, i, j, ij + 1 with relations i2 = a, j2 + j = b,
ij + ji = i. We will use the fact that (a, b] is split over K iff its norm form
〈〈a, b]] ' 〈〈a〉〉 ⊗ 〈〈b]] is isotropic over K
If K is perfect, then a ∈ K2 and hence (a, b] is split.
Now suppose K is not perfect and a 6∈ K2. Since finite fields are perfect, K
is infinite and thus K2 is infinite. Then K(i)2/K2 is an infinite dimensional
F2-vector space. Let (ak)k∈I be representatives of a F2-basis of K(i)2/K2.
Let c1, . . . , cn be representatives of a F2-basis of K/℘(K). We have cl+℘(K) =
c2l + ℘(K) for all 1 ≤ l ≤ n. For ak ∈ K, suppose akc
2l =
n∑j=1
amklc2m + zk for
NOTES TO u-INVARIANTS IN CHARACTERISTIC 2 21
aikl ∈ F2 and zk ∈ ℘(K). Since Mn(F2) is finite and (amkl)ml ∈ M2(F2) for all
k ∈ I. By the pigeonhole principle, there are two representatives as, at with
the same matrix.
Suppose a∗ = 1 + as + at. Then a∗c2m + c2
m = zs + zt ∈ ℘(K) for all 1 ≤ m ≤ nand a∗ ∈ K(i)2 −K2 since as +K2 6= at +K2. Then (a, b] ' (a∗, b∗] for some
b∗ ∈ K. Suppose (a∗)2 = a∗ and b∗ =n∑
m=1
b∗mc2m+z for b∗m ∈ F2 and z ∈ ℘(K).
Finally, since (b∗m)2c2ma∗ + b∗mc2
m = b∗m(c2ma∗ + c2
m) ∈ ℘(K), we have
(a∗, b∗] ' (a∗, b∗+n∑
m=1
(b∗m)2c2ma∗] ' (a∗, b∗+
n∑m=1
(b∗m)2c2m] = (a∗, z] 'M2(K).
Using [EKM08, Fact 98.14(1)(4)]1. �
3.17 Theorem
This is [Mor84, Th. 3]. For all integer n > 0, there exists a field K, charK = 2
such that u(K) = 2 and u(K) = 2n (resp.∞).
Proof. Let k0 be an algebraically closed field in characteristic 2. Let k =
k0(X1, X2, . . . , Xn) (resp. k = k0(X1, X2, . . .)). Take α ∈ k − ℘(k). Suppose
K/k is a finite separable extension with a 2-power degree such that α 6∈ ℘(K)
and is maximal with repect to this property.
(1) First, we show that u(K) = 2. Since α ∈ k − ℘(K), the form [1, α] is
anisotropic. Then u(K) ≥ 2.
Also, by Theorem 3.15, [K : ℘(K)] = 2. Then Br2(K) = 0 since Lemma 3.16.
Then the norm form of (a, ab], i.e. the 2-fold Pfister form 〈〈a, ab]] ' [a, b] ⊥[1, ab] is isotropic and hence hyperbolic. Let [a, b] ⊥ [c, d] be any nonsingular
form of dimension 4. Then [a, b] ⊥ [c, d] ⊥ [1, ab] ⊥ [1, cd] ' H4. Since
[1, ab] ⊥ [1, cd] ' [1, ab+ cd] ⊥ H (see [EKM08, e.g. 7.23]). Cancel2 H = [0, 0],
we have
[a, b] ⊥ [c, d] ⊥ [1, ab+ cd] ' H3
Since [a, b] ⊥ [c, d] has dimension 4 > 3, it is isotropic. Therefore u(K) ≤ 2.
(2) If k = k0(X1, X2, . . . , Xn) we show that u(K) = 2n. By Tsen-Lang’s
theorem, k is a Cn field. Since K/k is finite, K is also a Cn-field. Then
u(K) ≤ 2n.
1The quaternion algebra
[a, b
F
]in [EKM08] corresponds the notation (a, ab].
2Althrough Witt cancellation does not hold in general, we can cancel nonsingular forms
[EKM08, Th. 8.4].
22 ZHENGYAO WU (吴正尧)
Also, [k : k2] = 2n and ℘(k) ( k. By Lemma 1.6(1), u(K) ≥ [K : K2]. By
Lemma 1.13, [K : K2] = [k : k2]. Therefore u(K) ≥ 2n.
(3) When k = k0(X1, X2, . . .), similarly u(K) ≥ [K : K2] = [k : k2] =∞. �
4. What is known
See tables after this page.
NOTES TO u-INVARIANTS IN CHARACTERISTIC 2 23
Assumptions for field K,
charK = 2
u(K) u(K) References
u(K) = 0 0 [K : K2] [Bae82b, Rem. 1.2 c)],
Lemma 1.6
Quadratically closed (e.g.
alg. closed)
0 1 [Bae82b, Rem. 1.2 a)],
Lemma 1.4(3)
Quadratic separable clo-
sure of k, [k : k2] = 2n (e.g.
k = F2(X1, X2, · · · , Xn))
0 2n [Bae82b, Rem. 1.2 e) ], Ex-
ample 1.14
Quadratic separable clo-
sure of F2(X1, X2, · · · )0 ∞ [Bae82b, p. 108], Exam-
ple 1.14
Prefect and u(K) > 0 (e.g.
finite)
2 2 [Bae82b, Rem. 1.2 b)], Ex-
ample 1.5
Complete discrete valued
field
2u(K) 2u(K) [Bae82b, Th. 1.1], Theo-
rem 2.15
Henselian discrete valued
field
2u(K) below3 [MMW91, Th. 2], Theo-
rem 2.13(2)
k((T1))((T2)) . . . ((Tn)) 2nu(k) 2nu(k) [Bae82b, Cor. 2.10], Corol-
lary 2.16
k(T1, . . . , Tn), k alge-
braically closed
2n 2n [Bae82b, Eg. 2.11], Exam-
ple 3.4(1)
k(T1, . . . , Tn), k is C1,
u(k) = 2 or u(k) = 0 and
u(k) = 2
2n+1 2n+1 [Bae82b, Eg. 2.11], Exam-
ple 3.4(2)
K/k finitely generated
with tr.deg.(K/k) = n, k
alg. closed
2n 2n [MMW91, Prop. 2], Exam-
ple 3.7(1)
K/k finitely generated
with tr.deg.(K/k) = n, k
finite
2n+1 2n+1 [MMW91, Prop. 2], Exam-
ple 3.7(2)
K/k separably generated
with tr.deg.(K/k) = n, k
alg. closed
2n 2n [MMW91, Prop. 3], Exam-
ple 3.9
3max{2u(K), [K : K2]} ≤ u(K) ≤ 2u(K) + [K : K2]− 2[K : K2]
24 ZHENGYAO WU (吴正尧)
Assumptions for field K,
charK = 2
u(K) u(K) References
TBA 6 ∞ [MTW91, Th. 4.1]
TBA 6 6 [MTW91, 5.1]
2m ≥ 2n ≥ 4, m ≥ n − 1
TBA
2n 2m [MTW91, 5.2]
n ≥ 2 TBA 2n ∞ [MTW91, 5.3]
k = k0(T1, . . . , Tn), k0 alg.
closed, K/k maximal rela-
tive to α ∈ k − ℘(K)
2 2n [Mor84, Th. 3], Theo-
rem 3.17
n ≤ m, k0 alg. closed,
k1 = k0(t1, . . . , tn−1),
L = k1(s1, . . . , sm−n−1) ⊂k1((X)), K a Henseliza-
tion of L rel. v such that
K = L = k1,
2n 2m [MMW91, p. 344, Eg.],
Example 2.14
K = k((X)), u(k) =
u(k) = 6, [k : k2] = 4
12 12 [MMW91, p. 344, Rem.
(i)]
[K : K2] = 2n TBA 12 [2n, 2n + 4] [MMW91, p. 344, Rem.
(i)]
n ≥ 2 TBA 2n · 3 [MMW91, p. 344, Rem.
(i)] claimed in [Lag15, p.
808]
n ≥ 2 6= 2n − 1 [MMW91, Cor. 2], Corol-
lary 1.8
TBA 6= 5 [MMW91, Cor. 2] need
and could not obtain
[Bae82a]
NOTES TO u-INVARIANTS IN CHARACTERISTIC 2 25
References
[AB89] R. Aravire and R. Baeza, The behavior of the ν-invariant of a field of characteris-
tic 2 under finite extensions, Rocky Mountain J. Math. 19 (1989), no. 3, 589–600,
Quadratic forms and real algebraic geometry (Corvallis, OR, 1986). MR 1043232→16
[Alb37] A. Adrian Albert, Mordern higher algebra, The University of Chicago Press, Chicago
Illinois, 1937. →18
[Bae82a] R. Baeza, Algebras de cuaterniones, formas cuadraticas y un resultado de c. arf.,
Notas. Soc. Mat. Chile (1982), no. 2, 1–19. →24
[Bae82b] , Comparing u-invariants of fields of characteristic 2, Bol. Soc. Brasil. Mat.
13 (1982), no. 1, 105–114. MR 692281 →1, →2, →3, →4, →5, →6, →7, →8, →13,
→14, →15, →23
[Bou64] N. Bourbaki, Elements de mathematique. Fasc. XXX. Algebre commutative.
Chapitre 5: Entiers. Chapitre 6: Valuations, Actualites Scientifiques et Industrielles,
No. 1308, Hermann, Paris, 1964. MR 0194450 →10
[Bou81] , Elements de mathematique, Masson, Paris, 1981, Algebre. Chapitres 4 a 7.
MR 643362 →3, →16, →20
[Bou07] , Elements de mathematique. Algebre. Chapitre 10. Algebre homologique,
Springer-Verlag, Berlin, 2007, Reprint of the 1980 original [Masson, Paris;
MR0610795]. MR 2327161 →10
[EKM08] Richard Elman, Nikita Karpenko, and Alexander Merkurjev, The algebraic and geo-
metric theory of quadratic forms, American Mathematical Society Colloquium Pub-
lications, vol. 56, American Mathematical Society, Providence, RI, 2008. MR 2427530
→14, →21
[Lag15] Ahmed Laghribi, Quelques invariants de corps de caracteristique 2 lies au u-
invariant, Bull. Sci. Math. 139 (2015), no. 7, 806–828. MR 3407516 →24
[MMW91] P. Mammone, R. Moresi, and A. R. Wadsworth, u-invariants of fields of character-
istic 2, Math. Z. 208 (1991), no. 3, 335–347. MR 1134579 →1, →3, →4, →6, →9,
→10, →11, →12, →13, →16, →17, →23, →24
[Mor84] Remo Moresi, On a class of fields admitting only cyclic extensions of prime power
degree, Bol. Soc. Brasil. Mat. 15 (1984), no. 1-2, 101–107. MR 794732 →18, →19,
→21, →24
[MTW91] P. Mammone, J.-P. Tignol, and A. Wadsworth, Fields of characteristic 2 with pre-
scribed u-invariants, Math. Ann. 290 (1991), no. 1, 109–128. MR 1107665 →24
[Sal77] David J. Saltman, Splittings of cyclic p-algebras, Proc. Amer. Math. Soc. 62 (1977),
no. 2, 223–228. MR 0435044 →20
Department of Mathematics, Shantou University, 243 Daxue Road, Shantou, Guangdong,
China 515063
E-mail address: [email protected]