2014.12.

第 11 期
 目录

研究探讨

Study Note of ”Compressible Fluid Flow and Systems of Conservation Laws

in Several Space Variables” . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Yu Wenhua 1
一个优美恒等式的证明 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 郑可迪 23

专题介绍

解析流形上的李群作用 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 郑志伟 25
Introduction to Heegner Points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Qiu Congling 31
一类特殊的非线性算子的一致有界定理 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 孙奥 49

经验交流

墨尔本交流感受 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 林艺儿 55
 Study Note of ”Compressible Fluid Flow and
Systems of Conservation Laws in Several Space
Variables”
Yu Wenhua∗

1 Chapter 1
The first part of my studying note was written about Chapter 2.1 of A.Majda’s
book ”Compressible Fluid Flow and Systems of Conservation Laws in Sev-
eral Space Variables”, focusing on the local existence
∑N of smooth solutions for
the general system of conservation laws, ∂u
∂t + ∂
j=1 ∂xj Fj (u) = S(u, x, t). I
set S = 0 here for simplicity in exposition, since all the proofs given below
remain valid in the general case.

1.1 Main Theorem

Theorem 1.1. Assume u0 ∈ H s , s > N/2 + 1 and u0 (x) ∈ G, G1 ⊂⊂ G,
Then there is a time interval [0, T ] with T > 0, so that

∂u ∑ ∂
N
+ Fj (u) = 0, (1.1)
∂t ∂xj
i=1
u(x, 0) = u0 (x) (1.2)

have a unique classical solution u(x, t) ∈ C 1 (RN × [0, T ]). Furthermore,u ∈

C([0, T ], H s ) ∩ C 1 ([0, T ], H s−1 ) and T depends on ∥ u0 ∥s and G1 ,
i.e. T (∥ u0 ∥s , G1 )

We now separate the proof of Theorem 1.1 into two parts, Theorem 1.2
and Theorem 1.3.

Theorem 1.2. Under the hypotheses of Theorem 1.1, there is a unique

classical solution u ∈ C 1 ([0, T ] × RN )to the equations in (1.1) with u ∈
L∞ ([0.T ], H s ) ∩ Cw ([0, T ], H s ) ∩ Lip([0, T ], H s−1 )
∗
Department of Mathematical Sciences, Tsinghua University

1
Theorem 1.3. Any classical solution of (1.1) with u(x, t) ∈ G2 ⊂⊂ G satis-
fying the regularity stated in the conclusion of Theorem 1.2 on some interval
[0, T ] satisfies the additional regularity u ∈ C([0, T ], H s ) ∩ C 1 ([0, T ], H s−1 )
We first begin the proof of Theorem 1.2

1.2 Proof of Theorem 1.2

First, we smooth the initial data to avoid technical difficulties regarding the
smoothness of the coefficients in the associated linearized problems for this
iteration scheme. ∫
We choose j(x) ∈ C0∞ (RN ), suppj ⊆ {x| | x |≤ 1},j ≥ 0, j = 1, set
jϵ = ϵ−N j(x/ϵ) we define Jϵ u ∈ C ∞ (RN ) ∩ H s (RN ) by
∫
Jϵ u(x) = jϵ (x − y)u(y)dy (1.3)
RN

we have properties:
For u ∈ H s ∥ Jϵ u − u ∥s → 0 as ϵ → 0. (1.4)

For u ∈ H 1 and ϵ ≤ ϵ0 , ∥ Jϵ u − u ∥0 ≤ Cϵ ∥ u ∥ (1.5)

set ϵk = 2−k ϵ0 , k = 0, 1, 2, 3, . . . and set uk0 (x) = Jϵ u0 , k = 0, 1, 2, . . . ,
where ϵ0 > 0 will be chosen later.
Differentiate the nonlinear terms in (1.2)and get

∂u ∑
N
∂u
+ Ãj (u) =0 (1.6)
∂t ∂xj
i=1

apply the symmetrizing matrix A0 (u), which satisfies

CI ≤ A0 (u) ≤ C −1 I (1.7)
∗
A0 (u) = A0 (u) (1.8)
with a constant C uniform for u ∈ G1 with G1 ⊂⊂ G

A0 Ãj = Aj with Aj = A∗j , j = 1, 2, . . . , N. (1.9)

Now it is sufficient to prove Theorem 1.2 for the quasi-linear symmetric
system

∂u ∑
N
∂u
A0 (u) + Aj (u) =0 (1.10)
∂t ∂xj
i=1
u(x, 0) = u0 (x) (1.11)
2
 we set

u0 (x, t) = u00 (x) (1.12)

then for k = 0, 1, 2, 3, . . . , we define uk+1 (x, t) inductively as the solution of

∂uk+1 ∑
N
k ∂uk+1
A0 (u ) + Aj (uk ) =0 (1.13)
∂t ∂xj
i=1
x,0
uk+1 (x, 0) = u0 (x) (1.14)

However, we need to prove this iterates in (1.13) is well defined. That is,
∃G2 with G2 ⊂⊂ G s.t.uk ∈ G2 for k = 1, 2, . . . . From (1.4),we know that
when ϵ → 0, then∥ u0 − u00 ∥s → 0. We suppose G2 to be an open set
satisfyingG1 ⊂ G2 ,G2 ⊂⊂ G, since u0 ∈ G1 , so there exists R > 0 s.t.
∥ u − u00 ∥L∞ < Cs R implies that u ∈ G2 , where Cs is the constant in the
Sobolev embedding estimate ∥ v ∥L∞ ≤∥ v ∥s . So we get:

∥ u − u00 ∥s ≤ R implies that u(x) ∈ G2 (1.15)

Also when ϵ0 is small enough, we have:

∥ u0 − uk0 ∥s ≤ CR/4, C is the constant in (1.6) (1.16)

Then at interval [0, Tk ], where Tk satisfies ∥ uk − u00 ∥s,Tk ≤ R, from the

discussion above we know the eqnarray of uk+1 is well defined on [0, Tk ].
Actually, because u0 is smooth, and by induction hypotheses we suppose uk
is smooth and is already known, then (1.13) is a linear equation of uk+1 with
smooth coefficients and initial data, so we know uk+1 ∈ C ∞ ([0, Tk ] × RN ).
It will be known from the lemma below that Tk has a positive low bound, i.e
∃T∗ > 0 s.t Tk ≥ 0, k = 1, 2, . . . , Then if the lemma hold, we will get a series
of smooth solutions {uk }∞ k=0 on [0, T∗ ) × R , which later will be proven to
N

be convergent to u, the solution of Equation(1.10).

Lemma 1.1. (Boundness in the High Norm) There are constants L > 0,
T∗ > 0, so that the solutions uk+1 (x, t) defined in (1.10) for k = 0, 1, 2, . . . .
satisfy:

∥ uk+1 − u00 ∥s,T∗ ≤ R (1.17)

∂uk+1
∥ ∥s−1,T∗ ≤ R (1.18)
∂t
3
 Proof of Lemma 1.1: we will prove this lemma inductively we set v k+1 =
uk+1− u00 and compute v k+1 satisfies

∂v k+1 ∑
N
∂v k+1
A0 (uk ) + Aj (uk ) = Hk (1.19)
∂t ∂xj
i=1
v k+1 (x, 0) = k+1
u0 (x) − u00 (x) (1.20)

Where
∑
N
∂U00
H =−k
Aj (uk ) (1.21)
∂xj
j=1

by induction hypothesis, uk (x, t) ∈ G2 ,(x, t) ∈ RN × [0, T∗ ] where T∗ will

be chosen later so as to be not related with k. For similarity we ignore the
superscripts and consider u ∈ C ∞ ,v ∈ C ∞ ,u(x, t) ∈ G2 ,(x, t) ∈ RN × [0, T∗ ]
satisfying

∂v ∑
N
∂v
A0 (u) + Aj (u) =H (1.22)
∂t ∂xj
i=1
v(x, 0) = v0 (x) (1.23)

Denote vα = Dα vfor | α |< s. Divide the equation by A0 (u) on both

hands and differentiate with respect to x αtimes.

∂v ∑ N
∂v
α
D ( )+ Dα (A0−1 (u)Aj (u) ) = Dα (A−1
0 (u)H) (1.24)
∂t ∂xj
i=1

∂vα ∑ −1 ∂vα ∑ α −1
N N
∂v
+ A0 (u)Aj (u) + D (A0 (u)Aj (u) )−
∂t ∂xj ∂xj
i=1 i=1
∑N
∂vα
− A−1
0 (u)Aj (u) = Dα (A−1
0 (u)H) (1.25)
∂xj
i=1

∂vα ∑
N
∂vα
A0 (u) + Aj (u) = A0 (u)Dα (A−1
0 (u)H) + F α (1.26)
∂t ∂xj
i=1

∑
N
∂vα ∂v
Fα = A0 (u)[A−1
0 (u)Aj (u) − Dα (A−1
0 (u)Aj (u) )] (1.27)
∂xj ∂xj
i=1
4
 We assume that we can find a constant C, depending only on G2 , ∥ u ∥s ,
R, s, s.t
∑ ∑
( ∥ Fα ∥20 ) + ( ∥ A0 Dα (A−1
0 H) ∥0 ) ≤ C(G2 , ∥ u ∥s , R, s)(∥ v ∥s +1),
2 2

1≤|α|≤s |α|≤s
(1.28)
∫ 1
Where ∥ v ∥0 = ( RN | v |2 ) 2 . Now we have control the nonlinear terms on
the right hand side of (1.22), so now we
∫ can use Energy method to estimate
(1.22). Denote Eα (t) = (A0 vα , vα ) = RN v T A0 v

∂
Eα (t)
∂t
= (A0t vα , vα ) + 2(A0 vαt , vα )
∑
N
= (A0t vα , vα ) + 2(A0 (u)Dα (A−1
0 (u)H) + Fα − Aj vαxj , vα )
j=1

∑
N ∑
N
= (A0t vα , vα ) + 2(A0 (u)Dα (A−1
0 (u)H) + Fα , vα ) − (Aj vα , vα )xj + (Ajxj vα , vα )
j=1 j=1

= ((divA)vα , vα ) + 2(A0 (u)Dα (A−1

0 (u)H) + Fα , vα )
≤∥ divA ∥L∞ ∥ vα ∥20 +((∥ Fα ∥20 ) + (∥ A0 Dα (A−1
0 H) ∥0 )) ∥ vα ∥0
2 2
(1.29)

Sum over | α |≤ s and consider the bound of the nonlinear part, we get

∂
C ∥ v ∥2s ≤ (∥ divA ∥L∞ +C + 1) ∥ v ∥2s +C (1.30)
∂t
By Gronwall Inequality,we get

∥ v(T ) ∥2s ≤ exp(C −1 (∥ divA ∥L∞ +C + 1)T )(∥ uk+1

0 − u00 ∥s +C −1 CT )
(1.31)

So

∥ v ∥s,T = max ∥ v(t) ∥2s

0≤t≤T
−1
≤ exp(C (∥ divA ∥L∞ +C + 1)T )(∥ uk+1
0 − u00 ∥s +C −1 CT )
(1.32)

Choose ϵ0 small enough, then it follows that ∥ uk+1

0 − u00 ∥s ≤ R/2. So
∥u k+1 − u0 ∥s,T ≤ exp(C (∥ divA ∥L∞ +C + 1)T )(R/2 + C −1 CT )
0 −1

Let T∗ satisfies exp(C −1 (∥ divA ∥L∞ +C + 1)T )(R/2 + C −1 CT ) < R ,

then the first part of lemma 1 proved.

5
 For the second part, on [0, T∗ ]we have
∂uk+1
∥ ∥s−1,T∗
∂t
∑N
∂uk+1
≤C ∥ Aj (uk ) ∥s−1,T∗
∂xj
j=1

≤ C ∥ uk+1 ∥s,T∗ max ∥ Aj ∥L∞ ,G2

1≤j≤N

≤ C(∥ u00 ∥s +R) max ∥ Aj ∥L∞ ,G2 (1.33)

1≤j≤N

So L is independent of T∗ .
Lemma 1.1 proved
We also need another lemma
∑
Lemma 1.2. ∃T∗∗ ≤ T∗ , α < 1, and {β}∞ k=1 with | β |< ∞,s.t. uk+1 satisfy
∥ uk+1 − uk ∥0,T∗∗ ≤∥ uk − uk−1 ∥0,T∗∗ +βk .
Proof of Lemma 1.2:
First we get

∂(uk+1 − uk ) ∑
N
∂(uk+1 − uk )
A0 (uk ) + Aj (uk ) = Fk (1.34)
∂t ∂xj
i=1

0 (x) − u0 (x)
(x, 0) = uk+1
k+1 0
v (1.35)
where

∑
N
∂uk
Fk = − (Aj (uk ) − Aj (uk−1 )) (1.36)
∂xj
j=1

For the same reason when we use energy estimates we can get

∥ uk+1 − uk ∥o,T ≤ CeCT (∥ uk+1

0 − uk0 ∥0 +T ∥ Fk ∥0,T ) (1.37)
From lemma 1.1 and Taylor expansion we know

∥ Fk ∥0
∑
N
∂ ∂uk
≤ max ∥ Aj ∥L∞ ,G2 ∥ uk − uk−1 ∥0
1≤j≤N ∂t ∂xj
i=1
∂
≤ max ∥ Aj ∥L∞ ,G2 ∥ uk ∥1 ∥ uk − uk−1 ∥0
1≤j≤N ∂t
≤ C ∥ uk − uk−1 ∥0 (1.38)
∂
C = max ∥ Aj ∥L∞ ,G2 (R+ ∥ u00 ∥s,T∗ ) (1.39)
1≤j≤N ∂t
6
 ∀T < T∗ , we have

∥ Fk ∥0,T ≤ C ∥ uk − uk−1 ∥0,T (1.40)

Then we choose ϵ0 small enough so that we can let ∥ uk+1 −uk ∥0 ≤ C2−k .
So by (1.37) and (1.38) we can choose proper T = T∗∗ and βk = C2−k , we
get lemma 1.2 proved.
Now we get back to the proof of Theorem 1.2:
∑∞First from the lemma 1.2, by inductive process it easily follows that
k=0 ∥ uk+1 − uk ∥
0,T∗∗ < ∞
When we consider uk , its initial data uk0 is in H s (RN ) ∩ C ∞ (RN ). So
by Energy Estimate, we can easily know that ∀t < T∗∗ , uk (•, t) ∈ H s (RN ).
k
Also from lemma 1.1 we know that ∥ ∂u ∂t ∥≤ L. So ∀t1 < t2 < T∗∗ ,

∥ uk (x, t1 ) − uk (x, t2 ) ∥2H s−1 (RN )

∑∫
= | Dα uk (x, t1 ) − Dα uk (x, t2 ) |2 dx
N
|α|≤s R
∑∫ ∂Dα uk (x, t∗ ) 2
= | | dx(t1 − t2 )2
N ∂t
|α|≤s R

∂uk 2
=∥ ∥ s−1 (t1 − t2 )2
∂t H
≤ L2 (t1 − t2 )2
⇒∥ uk (x, t1 ) − uk (x, t2 ) ∥H s−1 (RN ) ≤ L | t1 − t2 |
⇒ uk ∈ Lip([0, T∗∗ ], H s−1 (RN )) (1.41)

Especially, we get uk∑ ∈ C([0, T∗∗ ], L2 (RN )).

By lemma 1.2 we know ∞ k=1 ∥ u
k+1 − uk ∥
0,T∗∗ < ∞,
So ∃M , s.t ∀k, ∥ u ∥0,T∗∗ < M . i.e ∀0 < t < T∗∗ , ∥ uk ∥0 < M .
k

Also, ∀0 < t1 < t2 < T∗∗ , ∀k, ∥ uk (x, t1 ) − uk (x, t2 ) ∥0 ≤ L | t1 − t2 |.

So by Arezola-Ascoli Theorem,{uk } has a subsequence convergence to
some u ∈ C([0, T∗∗ ], L2 (RN )]). And since uk is a cauchy sequence in the
space, so limk→∞ uk = u in C([0, T∗∗ ], L2 (RN )]), i.e limk→∞ ∥ uk − u ∥0,T∗∗ =
0.
By lemma 1.1 we can assume that

∂uk
∥ uk ∥s,T∗∗ + ∥ ∥s−1,T∗∗ ≤ C (1.42)
∂t

uk (x, t) ∈ G2 ⊂⊂ G, (x, t) ∈ RN × [0, T∗∗ ] (1.43)

7
 and by Sobolev space interpolation inequalities
1−s′ /s ′
∥ v ∥s′ ≤ Cs ∥ v ∥0 ∥ v ∥ss /s (1.44)
Then ∥ uk − ul ∥s′ ,T∗∗ ≤ C(∥ uk − ul ∥0,T∗∗ ) 1−s′ /s .
So
lim ∥ uk − ul ∥s′ ,T∗∗ = 0, ∀s′ < s (1.45)
t→∞

So if s′ > N/2 + 1(since s > N/2 + 1,so such s′ exists)

Then ∥ v ∥C 1 (RN ) ≤ C ∥ v ∥H s′ (RN ) .
′
Then ∥ v ∥C 1 (RN ),T∗∗ ≤ C ∥ v ∥H s′ (RN ),T∗∗ . Because uk → u in C([0, T∗∗ ], H s (RN )),
then from the norm control showed above we know uk → u in C([0, T∗∗ ], C 1 (RN )).
And then, by considering

∂uk+1 ∑ N
∂uk+1
= −A−1 k
0 (u ) Aj (uk ) (1.46)
∂t ∂t
j=1
k k
∂t ∈ C([0, T∗∗ , C(R )]) and ∂t → ∂t inC([0, T∗∗ ], C(R ))
We get ∂u N ∂u ∂u N

so u ∈ C ([0, T∗∗ ] × R ) is the classical solution of (1.10).

1 N

What left to be proved is u ∈ Cw ([0, T∗∗ ], H s ) ∩ Lip([0, T∗∗ ], H s−1 )

The proof of the first part:

′
Since uk → u in C([0, T ], H s ) fors′ < s, [ϕ, uk (t)] → [ϕ, u(t)] for any
′ ′
ϕ ∈ H −s uniformly on[0, T∗ ]. And since ∥ uk ∥s,T∗∗ ≤ R+ ∥ u00 ∥s and H −s is
dense in H −s , ∀ϕ ∈ H −s ∃ϕ s.t | ϕ − ϕ |H −s < 3(R+∥u
ϵ
0∥ )
0 s
⇒| (ϕ, u − uk ) |=| (ϕ, u − uk ) + (ϕ − ϕ, u − uk ) |≤| (ϕ, u − uk ) | + |
(ϕ − ϕ)u | + | ((ϕ − ϕ, uk )) < ϵ when k is big enough
⇒ [ϕ, uk (t)] → [ϕ, u(t)] uniformly on[0, T∗∗ ] for anyϕ ∈ H −s since (ϕ, uK )
is continuous, so u ∈ Cw ([0, T∗∗ ], H s )

The proof of the second part: From Equation(1.41) we know ∥ uk (x, t1 ) −

uk (x, t2 ) ∥H s−1 (RN ) ≤ L | t1 − t2 |, and since s − 1 < s, by Equation (45) we
know uk → u in C([0, T∗∗ ], H s−1 ). So ∥ u(x, t1 ) − u(x, t2 ) ∥H s−1 (RN ) ≤ L |
t1 − t2 |. Then u ∈ Lip([0, T∗∗ ], H s−1 )
So Theorem 1.2 get proved. ∑ ∫
Now we define the norm ∥ v ∥2s,A0 (t) by ∥ v ∥2s,A0 (t) = |α|≤s RN (Dα v, A0 (t)Dα v)dx
for 0 ≤ t ≤ T∗∗ , A0 (t) is the short-hand notation for A0 (u(x, t)). As a con-
sequence of (1.42),(1.43)and(1.45), we can conclude that

A0 (u(x, t)) ∈ C([0, T∗ ], C(RN )) (1.47)

8
 CI ≤ A0 (U (x, t)) ≤ C −1 I, (x, t) ∈ RN × [0, T∗∗ ] (1.48)

From (1.47),(1.48),we can get two facts:

C ∥ v ∥2s ≤∥ v ∥2s,A0 (t) ≤ C −1 ∥ v ∥2s (1.49)

lim ∥ v ∥2s,A0 (t) = lim ∥ v ∥2s,A0 (t) (1.50)

t↓0 t↓0

for any v(t) ∈ Cw ([0, T∗∗ ], H s )

Now we start the proof of Theorem 1.3

1.3 Proof of Theorem 1.3

First, it is sufficient to prove that u ∈ C([0, T ], H s ) Since we soon concludes
from (2.18) that u ∈ C 1 ([0, T ], H s−1 ).
Furthermore, we only need to prove the strong right continuity of u at
t=0, since by translation we can use the same method to prove the strong
continuity property of each point. Similarly we know when we take the
inverse direction of time we can prove the left continuity property.
Because CI ≤ A0 (u) ≤ C −1 I, so ∥ • ∥s,A0 (t) and ∥ • ∥s are equivalent.
Moreover, considering the existing conclusion of uniform convexity in
Functional Analysis, if wn → w weakly in Hilbert space, then wn → w
strongly is equivalent to ∥ w ∥≥ limn→∞ ∥ wn ∥
So by (1.49)(1.50), we will prove the right continuity at 0 in H s if we
prove ∥ u0 ∥2s,A0 (0) ≥ limt↓0 ∥ u(t) ∥2s,A0 (0) ≥ limt↓0 ∥ u(t) ∥2s,A0 (t)
In this way we unify the norm by using (1.49)(1.50).
We need the following proposition:
Proposition 1.1. (Moser-type Calculus Inequalities)
(A) For f, g ∈ H s ∩ L∞ and | α |≤ s

∥ Dα (f g) ∥0 ≤ Cs (| f |L∞ ∥ Ds g ∥0 + | g |L∞ ∥ Ds f ∥0 ) (1.51)

(B)For f ∈ H s , Df ∈ L∞ , g ∈ H s−1 ∩ L∞ and | α |≤ s

∥ Dα (f g) − f Dα g ∥0 ≤ Cs (| Df |L∞ ∥ Ds−1 g ∥0 + | g |L∞ ∥ Ds f ∥0 ) (1.52)

(C) Assume g(u) is a smooth vector-valued function on G, u(x) is a con-

tinuous function with u(x) ∈ G1 , G1 ⊂⊂ G, and u(x) ∈ H s ∩ L∞ , then for
s ≤ 1,
9
 ∂g
∥ Ds g(u) ∥0 ≤| | | u |L ∞ ∥ D u ∥0
s−1 s
(1.53)
∂u s−1,G1
Also we need the following lemma:
Lemma 1.3. If u is the local solution of Theorem 2 on some interval[0, T∗∗ ],
then is f (s) ∈ L1 [0, T∗∗ ] s.t

∥ u(t) ∥2s,A0 (t)

∑∫
= (Dα u, A(u(t))Dα u)dx
N
|α|≤s R
∑∫ ∫ t
≤ (Dα u0 , A(u0 )Dα u0 )dx + | f (s) | ds
N
|α|≤s R 0
∫ t
=∥ u0 ∥2s,A0 (0) + | f (s) | ds (1.54)
0

Proof: because uk are in C ∞ ∩ H s ,(by Energy estimates), and by energy

estimates ,for 0 ≤ t ≤ T∗∗ ,
∫
∂ ∑
(Dα uk+1 , A0 (uk )Dα uk+1 )dx
∂t N
|α|≤s R
∫ ∫
= (divA(uk )uk+1 , uk+1 ) + 2 (Fsk+1 , uk+1 ) (1.55)
RN RN

where

uk+1
α = Dα uk+1 , (1.56)
Fsk+1
∫
∂uk+1 k ∂u
k+1
= A0 (uk )[A−1 k k
0 (u )Aj (u )
α
− Dα (A−1
0 Aj (u ) )]
1≤|α|≤s,1≤j≤N ∂xj ∂xj
(1.57)

From the high norm estimate of uk in Lemma 1.1,and by Proposition

1.1, we know that the right hand side of (1.56) can be controlled by an
integrable functionf (x). In fact f (s) ∈ L∞ ([0, T∗∗ ]). Integrating the both
sides of (1.56) with respect to t, then we get
∀0 ≤ t ≤ T∗∗ ,
∑∫
(Dα uk+1 (t), A0 (uk (t))Dα uk+1 (t))
N
|α|≤s R
∑∫ ∫ ∫ t
≤ R N
(Dα uk+1 k α k+1
0 , A0 (u0 )D u0 ) + | f (s) | ds(1.58)
|α|≤s RN 0
10
From (1.3) we know
∑∫ ∫
0 , A0 (u0 )D u0 ) =∥ u0 ∥s,A0 (0)
(Dα uk+1 α k+1
lim RN k 2
(1.59)
k→∞ RN
|α|≤s

Also when k → ∞ from (1.45) and Sobolev embedding estimate we know

max | A0 (uk (t)) − A0 (u(t)) |L∞ → 0 (1.60)

0≤t≤T∗

Moreover from the weak convergence when k → ∞, for fixed t > 0 it

follows that

lim (uk+1 (t), uk+1 (t))s,A0 (t) ≤ (u(t), u(t))s,A0 (t)

k→∞
∑∫
→∥ u(t) ∥2s,A(t) ≤ lim (Dα uk+1 (t), A0 (uk (t))Dα uk+1 (t))
k→∞ N
|α|≤s R
(1.61)

Then (1.54) follows from (1.59) and the equation above, this complete the
prove of lemma 1.3.
Now Theorem 1.3 is an immediate result of lemma 1.3. So Theorem 1.3
proved.
Now we have proved Theorem 1.2 and Theorem 1.3, so Theorem 1.1
proved.

2 Chapter 2
This part is my studying note on Chapter 2.2 , focusing on a continuation
principle for smooth solutions of the hyperbolic system:

∑
N
A0 (x)ut + Aj (u)uxj = 0, u(x, 0) = u0 (x) (2.1)
j=1

where

CI ≤ A0 (u(x, t)) ≤ C −1 I (2.2)

2.1 Main Theorem

Theorem 2.1. Assume that u0 ∈ H s for some s > N2 + 1. Let T > 0 be
some given time. Assume that there are fixed constants M1 , M2 and a fixed
open set G1 with G1 ⊂⊂ G (all independent of T∗ ) so that for any interval

11
of classical existence [0, T∗ ], T∗ ≤ T for u(t) from Theorem 1.1 in Chapter
1, the following a priori estimates are satisfied:
⇀ |L∞ ≤ M1 , 0 ≤ t ≤ T∗
| div A (2.3)
| Du |L∞ ≤ M2 , 0 ≤ t ≤ T∗ (2.4)
u(x, t) ∈ G1 ⊂⊂ G, (x, t) ∈ R × [0, T∗ ]
N
(2.5)

Then the classical solution u(t) exists on the interval [0, T ], with u(t) in
C([0, T ], H s ) ∩ C 1 ([0, T ], H s−1 ). Furthermore, u(t) satisfies the a priori es-
timate

∥ u ∥s,T∗ ≤ Cexp((M1 + M2 )CT∗ ) ∥ u0 ∥s (2.6)

for T∗ with 0 ≤ T∗ ≤ T and the two constants C in (2.6) depend only on s

and G1 , i.e., C(s, G1 ).

2.2 Proof of Theorem 2.1

: In order to prove this theorem, we first prove the a priori estimates (2.6).
We need the following proposition:

Proposition 2.1. (Moser-type Calculus Inequalities)

(A) For f, g ∈ H s ∩ L∞ and | α |≤ s

∥ Dα (f g) ∥0 ≤ Cs (| f |L∞ ∥ Ds g ∥0 + | g |L∞ ∥ Ds f ∥0 ) (2.7)

(B)For f ∈ H s , Df ∈ L∞ , g ∈ H s−1 ∩ L∞ and | α |≤ s

∥ Dα (f g) − f Dα g ∥0 ≤ Cs (| Df |L∞ ∥ Ds−1 g ∥0 + | g |L∞ ∥ Ds f ∥0 ) (2.8)

(C) Assume g(u) is a smooth vector-valued function on G, u(x) is a con-

tinuous function with u(x) ∈ G1 , G1 ⊂⊂ G, and u(x) ∈ H s ∩ L∞ , then for
s ≤ 1,

∂g
∥ Ds g(u) ∥0 ≤| | | u |L ∞ ∥ D u ∥0
s−1 s
(2.9)
∂u s−1,G1

2.2.1 Energy Estimate

We first take α-times derivative of both side of Equation(2.1), then we get
the following equations:

∂uα ∑ ∂uα
N
A0 (u) + = Fα (2.10)
∂t ∂xj
j=1
12
Where
∑
N
∂uα ∂u
Fα = A0 (u)[A−1
0 (u)Aj (u) − Dα (A−1
0 (u)Aj (u) )] (2.11)
∂xj ∂xj
i=1

∑
∥ Fα ∥0
1≤|α|≤s
∑ ∂uα ∂u
= ∥ A0 (u)[A−1
0 (u)Aj (u) − Dα (A−1
0 (u)Aj (u) )] ∥0
∂xj ∂xj
1≤|α|≤s,1≤j≤N
∑ ∂uα ∂u
≤ C −1 ∥ A0−1 (u)Aj (u) − Dα (A−1
0 (u)Aj (u) ) ∥0
∂xj ∂xj
1≤|α|≤s,1≤j≤N
∑ ∂u ∂u
≤ C −1 Cs (| D(A−1
0 Aj ) |L∞ ∥ D
s−1
∥0 + | |L∞ ∥ Ds (A−1 j ) ∥0 )
0 A(2.12)
∂xj ∂xj
1≤|α|≤s,1≤j≤N

The first ”≤” holds because of (2.9), and the second ”≤” holds because of
Proposition 2.1(B).
By Proposition 2.1(C), we have

∥ Ds (A−1
0 Aj (u)) ∥0
∂(A−1
0 Aj (u))
≤ Cs | |s−1,G1 | u |s−1
L∞ ∥ D u ∥ 0
s
∂u
≤ C̃(s, G1 ) ∥ Ds u ∥0
(2.13)

We assume a priori estimates | ∂u

∂xj |L∞ ≤ M2 , then we have So
∑
∥ ∂Fα ∥0 ≤ C̃(s, G1 )M2 ∥ Ds u ∥0 (2.14)
1≤|α|≤s
∑
Denote E(t) = o≤|α|≤s (A0 vα , vα ) So from the equation we know that

∂E ∑ ⇀ α , uα ) + 2(Fα , uα )
= (div Au (2.15)
∂t
o≤|α|≤s

13
 ∂
E(t)
∂t ∑
= ⇀ α , uα ) + 2(Fα , uα )
(div Au
o≤|α|≤s
∑
≤ M1 (uα , uα ) + 2(Fα , uα )
o≤|α|≤s
∑
≤ C −1 M1 (A0 uα , uα ) + 2 ∥ Fα ∥0 ∥ uα ∥0
o≤|α|≤s

≤ C −1 M1 E + 2C̃M2 ∥ uα ∥20
≤ C −1 (M1 + 2C̃)E (2.16)

So by Gronwall Inequality,
−1 (M +2C̃M )t
E(t) ≤ eC 1 2
∥ u0 ∥0 (2.17)

So
∑ −1 (M +2C̃M )t
∥ uα (t) ∥2 ≤ C −1 E(t) ≤ C −1 eC 1 2
∥ u0 ∥0 (2.18)
o≤|α|≤s

So priori estimate (2.6) gets proved.

Then if the H s classical solution u only exists on [0, T∗ ) for some T∗ ≤ T ,
then we apply the a priori estimate (2.6), and use the local existence theorem
in Chapter 1, beginning at T − ϵ, where ϵ > 0 is appropriately small, then
we can continue this H s classical solution beyond T∗ , contradiction. So
now we get u(t) ∈ C([0, T ], H s ), which ∑ follows Du(t) ∈ C([0, T ], H s−1 ).
−1 N
Then use the Equations(1): ∂t = −A0
∂u ∂u
j=1 Aj ∂xj , we can easily know
that ut ∈ C([0, T ], H s−1 ) too. So u(t) ∈ C([0, T ], H s ) ∩ C 1 ([0, T ], H s−1 )
So Theorem 2.1 get proved
Theorem 2.1 has two corollaries. Here is the first one:

Corollary 2.1. Assume u0 ∈ H s for some s > N2 +1. Assume that u(x, t) is
a classical solution of (2.1) on some interval [0, T ] with u ∈ C 1 ([0, T ] × RN ),
then on the same time interval [0, T ] necessarily u ∈ C([0, T ], H s ). In partic-
ular, if u0 ∈ ∩s H s , on any interval [0, T ], where u belongs to C([0, T ], H s0 )
for some s0 , s0 > N2 + 1, automatically u is a function in C ∞ ([0, T ] × RN ).

2.3 Proof of Corollary 2.1

If u ∈ C 1 ([0, T ] × RN ), then the a priori estimates (2.3),(2.4),(2.5) satisfies
automatically. Also since u0 ∈ H s , and a priori estimate (2.6), we know
u ∈ C([0, T ], H s ). Especially when the latter situation also holds, then
on interval [0, T ] we have u0 ∈ ∩s Hs hold. Then by the a priori estimate
(2.6) we have u ∈ ∩s C([0, T ], H s ). So u ∈ C([0, T ], C ∞ (RN )) by the Sobolev
14
 ∑N
Imbedding theorem. Also the equation ∂u
∂t = −A−1
0
∂u
j=1 Aj ∂xj implies that
ut ∈ C ∞,so we get u ∈ C ∞ ([0, T ] × RN ).
Here is the second one:

Corollary 2.2. Assume u0 ∈ H s for some s > N2 + 1. Then [0, T ) with

T < ∞ is a maximal interval of H s existence if and only if either

| ut |L∞ + | Du |L∞ → ∞ as t ↑ T,
or (2.19)
as t ↑ T, u(x, t) escape every compact subset K ⊂⊂ G. (2.20)

2.4 Proof of Corollary 2.2

If we choose some T∗ < T , then neither of situation(2.19),(2.20) holds for
T∗ . Then it is obvious that a priori estimates (2.4),(2.5) hold for T∗ , we only
need to check the a priori estimate (2.3). We assume that there exists a
∂Aj (u)
constant M s.t ∀0 ≤ j ≤ N , | ∂u |L∞ < M , and a constant p s.t. | ut |L∞
+ | Du |L∞ < P , ∀t ∈ [0, T∗ ]. Then we have the following calculation:

∂A0 (u) ∂u
(A0 )t = (2.21)
∂u ∂t
∂Aj (u) ∂u
(Aj )xj = (2.22)
∂u ∂xj

Then
⇀ |L∞ ≤ M P, 0 ≤ t ≤ T∗
| div A (2.23)

So now a priori estimate (2.3) still holds. Then by theorem 2.1 we know
the H s classical solution exists on [0, T∗ ]. So the H s classical solution exists
on the interval [0, T ). Also, we suppose [0, T0 ) is the maximum interval of
H s local existence, then either of situation (2.19),(2.20) implies T0 ≤ T . So
[0, T ) is the maximum interval of H s local existence.

3 Chapter 3
This part focus on Chapter 2.3 of A.Majda’s book ”Compressible Fluid Flow
and Systems of Conservation Laws in Several Space Variables”. In Chapter
1 and Chapter 2 we have given a complete treatment of the H s classical ex-
istence theory under the assumption that u0 (x) ∈ H s (RN ). However, such
conditions on the initial data are not always natural since for example the ini-
tial density might be a smooth plane wave ρ(x•w) with limx̃→+∞ ρ0 (x̃) = ρ+
and limx̃→−∞ ρ0 (x̃) = ρ− with ρ+ , ρ− > 0 and ρ− ̸= ρ+ . The proofs in the
previous two notes were based upon the global energy principle for symmet-
ric hyperbolic system. However, it is well known that solutions of hyperbolic
15
equations have finite propagation speed and obey a local energy principle .
To take advantage of the above local energy principle and also to allow for
initial data like the density in the previous paragraph, we now focus on the
uniformly local Sobolev space Hul s introduced by Kato.

In this note we still focus on the hyperbolic system

∑
N
A0 (u)ut + Aj (u)uxj = 0,
j=1

where u(x, t) ∈ G1 , G1 ⊂⊂ G, u(x, 0) = u0 (x) (3.1)

CI ≤ A0 (u) ≤ C −1 I for u ∈ G1 (3.2)

First we want to introduce the uniformly local Sobolev spaces Hul s . These

spaces defined in the following fashion: let θ ∈ C0∞ (RN ) be a function so

that θ ≥ 0 and θ(x) = 1, if | x |≤ 12 ; θ(x) = 0, if | x |> 1 and define
θd,y (x) = θ( x−y
d ).

Definition 3.1. The function u belongs to the uniformly local Sobolev spaces
u ∈ Hul
s , provided that there is some d > 0 so that

sup ∥ θd,y u ∥s =∥ u ∥s,d < ∞ (3.3)

y∈RN

Proposition 3.1. The norms ∥ . ∥s,d are all equivalent norms on Hul
s as d

varies; in particular,

∥ u ∥s,d1 ≤ C ∥ u ∥s,d2 (3.4)

Proof For d1 ≤ d2 , it is obvious that ∥ u ∥s,d1 ≤∥ u ∥s,d2 . What left to be

proved is that there exists a constant C which only related with s, d1 and d2 ,
s.t. ∥ u ∥s,d2 ≤ C(s, d1 , d2 ) ∥ u ∥s,d1 . To prove this, we only need to proved
the situation when s = 0, i.e ∥ u ∥s,d2 ≤ C(d1 , d2 ) ∥ u ∥s,d1 . When s > 0 the
proof is similar.

We define Cov(d, D)=minimum number of d2 -balls it takes to cover D-

balls. We can see that ∀d, D > 0, Cov(d, D) < ∞, and is only related with
d, D. Then if ∥ u ∥0,d2 = M ≥ 0, then supy∈RN ∥ θd2 ,y u ∥0 = M , so ∀ϵ > 0,
there exists y0 ∈ RN s.t

∥ θd2 ,y0 u ∥0 > M − ϵ (3.5)

16
 d1
Take t = Cov(d1 , d2 ) points z1 , . . . , zt , s.t. t 2 -balls B(z1 , d21 ), . . . , B(zt , d21 )
cover B(y0 , d2 ). Then we have

∑
t ∑
t
t ∥ u ∥0,d1 = ∥ u ∥0,d1 ≥ ∥ θd1 ,zi u ∥0
i=1 i=1
∑
t
≥ ∥ 1B(z d1 u ∥0 ≥∥ 1∩t d1 u ∥0
t, 2 ) i=1 B(zt , 2 )
i=1
≥∥ 1B(y0 ,d2 ) u ∥0 ≥∥ u ∥0,d2 −ϵ (3.6)

So let ϵ → 0, we get

∥ u ∥0,d2 ≤ Cov(d1 , d2 ) ∥ u ∥0,d1 . (3.7)

Proved.

Next, we formulate the local energy principle for the linear symmetric
hyperbolic equation

3.1 Energy Estimate

∑
N
A0 (u)vt + Aj (u)vxj = F, v(x, 0) = v0 (x) (3.8)
j=1

where u(x, t) ∈ G1 , G1 ⊂⊂ G. We recall that

CI ≤ A0 (u) ≤ C −1 I (3.9)

and also, there is a number D > 0 so that

∑N
max |( Aj (u)wj v, v) |≤ D | v |2 (3.10)
|w|=1,u∈G1
j=1

Then we define the number R by R = D C.

Now we try to make out an energy estimates by using Green’s Formula.
First we denote

Ω(y, T ) = {(t, x) ∈ RN +1 || x − y |≤ d + T − t, 0 < t < T } (3.11)

From the equation we get

(A0 vt , v) + (Aj vxj , v) = (F, v) (3.12)

17
So integrate over Ω(y, T ), and use Green’s Identity, we get:
∫
2 (F, v)dxdt
Ω(y,T )
∫
=2 (A0 vt , v) + (Aj vxj , v)dxdt
Ω(y,T )
∫
= (A0 v, v)t − (A0t v, v) + (Aj v, v)xj − (Ajxj v, v)dxdt
Ω(y,T )
∫ ∫
=− ⇀
(div Av, v)dxdt + (A0 v, v)t + (Aj v, v)xj dxdt
Ω(y,T ) Ω(y,T )
∫ ∫ ∫
=− ⇀
(div Av, v)dxdt + (A0 (u)v, v)(T )dx − (A0 (u)v0 , v0 )dx
Ω(y,T ) |x−y|≤d |x−y|≤d+RT
∫ T ∫
1 x1 xn
+ ((A0 v, v), (A1 v, v), . . . , (AN v, v)) • √ (R, ,..., )
0 |x−y|=d+R(T −s) 1+R 2 |x| |x|
(3.13)
1
Here √1+R 2
x1
(R, |x| , . . . , x|x|n ) is the unit normal vector on S = {(x, t) || x−y |=
d + R(T − t), 0 < t < T } pointing outwards.
Then we use the bounds (3.9) and (3.10),we have
x1 xn
((A0 v, v), (A1 v, v), . . . , (AN v, v)) • (R, ,..., )
|x| |x|
≥ RC | v |2 −D | v |2 = 0 (3.14)

So
∫
(A0 (u)v, v)(T )dx
|x−y|≤d
∫ ∫
≤ (A0 (u)v0 , v0 )dx + 2 (F, v)dxdt
|x−y|≤d+RT Ω(y,T )
∫
+ ⇀ v)dxdt
(div Av,
Ω(y,T )
∫
≤ (A0 (u)v0 , v0 )dx
|x−y|≤d+RT
∫ T∫
+ ⇀ v)dxdt
2(F, v) + (div Av, (3.15)
0 |x−y|≤d+R(T −s)

Now we have this energy estimates,we get back to the original equation:

∑
N
A0 (u)ut + Aj (u)uxj = 0, u(x, 0) = u0 (x) (3.16)
j=1

We need the following proposition:

18
 s )
Proposition 3.2. (Calculus Inequalities f orHul
∞
(A) For f, g ∈ Hul ∩ L and | α |≤ s
s

∥ Dα (f g) ∥0,d ≤ Cs,d (| f |L∞ ∥ g ∥s,2d + | g |L∞ ∥ f ∥s,2d ) (3.17)

s ∩ C 1 , g ∈ H s−1 ∩ L∞ and | α |≤ s
(B)For f ∈ Hul ul

∥ Dα (f g) − f Dα g ∥0,d ≤ Cs,d (| f |C 1 ∥ g ∥s−1,2d + | g |L∞ ∥ f ∥s,2d ) (3.18)

(C) Assume g(u) is smooth on G, u(x) continuous with u(x) ∈ G1 , G1 ⊂⊂

G, and u(x) ∈ Hul
s , then

∥ g(u) ∥s,d ≤ C(G1 , s, d)(1+ ∥ u ∥s,2d ) (3.19)

We first take α-times derivative of both side of Equation(3.1), then we

get the following equations:

∂uα ∑ ∂uα
N
A0 (u) + = Fα (3.20)
∂t ∂xj
j=1

Where
∑
N
∂uα ∂u
Fα = A0 (u)[A−1
0 (u)Aj (u) − Dα (A−1
0 (u)Aj (u) )] (3.21)
∂xj ∂xj
i=1

∑
∥ Fα ∥0,d
1≤|α|≤s
∑ ∂uα ∂u
= ∥ A0 (u)[A−1
0 (u)Aj (u) − Dα (A−1
0 (u)Aj (u) )] ∥0,d
∂xj ∂xj
1≤|α|≤s,1≤j≤N
∑ ∂uα ∂u
≤ C −1 ∥ A−1
0 (u)Aj (u) − Dα (A−1
0 (u)Aj (u) ) ∥0,d
∂xj ∂xj
1≤|α|≤s,1≤j≤N
∑ ∂u ∂u
≤ C −1 Cs,d (| A−1
0 Aj |C 1 ∥ ∥s−1,2d + | |L∞ ∥ A−1
0 Aj ∥s,2d )
∂xj ∂xj
1≤|α|≤s,1≤j≤N
(3.22)

The first ”≤” holds because of (3.9), and the second ”≤” holds because of
Proposition 3.2(B).
By Proposition 3.2(C), we have

∥ A−1
0 Aj (u) ∥s,2d ≤ C(G1 , s, d)(1+ ∥ u ∥s,4d ) (3.23)
19
We assume a priori estimates | ∂u
∂xj |L∞ ≤ M2 , then we have
∑
∥ Fα ∥0,d ≤ C̃(G1 , s, d)M2 (1+ ∥ u ∥s,4d ) (3.24)
1≤|α|≤s

Because ∥ • ∥s,4d is equivalent to ∥ • ∥s,d by Proposition 1, so actually we

get the following:
∑
∥ Fα ∥0,d ≤ C̃(G1 , s, d)M2 (1+ ∥ u ∥s,d ) (3.25)
1≤|α|≤s

Now we back to the energy estimate (2.13), we assume a priori estimate

⇀ |≤ M1 , then from above we have we have:
| div A
∫
(A0 (u)uα , v)(T )dx
|x−y|≤d
∫ ∫ T ∫
≤ (A0 (u)u0α , u0α )dx + ⇀ α , uα )
2(Fα , uα ) + (div Au
|x−y|≤d+RT 0 |x−y|≤d+R(T −s)
∫ ∫ T ∫
≤ C −1 | u0,α |2 dx + | Fα |2 +(1 + M1 ) | uα |2
|x−y|≤d+RT 0 |x−y|≤d+R(T −s)
(3.26)

So
∑∫
C (uα , uα )(T )dx
|α|≤s |x−y|≤d
∑∫
≤ (A0 (u)uα , v)(T )dx
|α|≤s |x−y|≤d
∑∫ ∫ T ∑∫
−1
≤ C | u0,α | dx + 2
| Fα |2 +(1 + M1 ) | uα |2
|α|≤s |x−y|≤d+RT 0 |α|≤s |x−y|≤d+R(T −s)
∑∫ ∫ T ∑∫
−1
≤ C ∥ u0,α ∥ dx + 2
| Fα |2 +(1 + M1 ) | uα |2
|α|≤s |x−y|≤d+RT 0 |α|≤s |x−y|≤d+R(T −s)
∫ T ∑ ∫ T
−1
≤C ∥ u0 ∥s,d+RT + ∥ Fα ∥0,d+R(T −s) +(1 + M1 ) ∥ u ∥s,d+R(T −s) (3.27)
0 |α|≤s 0

By Proposition 2.1, we know

∑
∥ Fα ∥0,d+R(T −s)
|α|≤s
∑
≤ C(d, d + RT ) ∥ Fα ∥0,d
|α|≤s

≤ C(G1 , s, d, RT )M2 (1+ ∥ u ∥s,d ) (3.28)

20
 ∑ ∫
Take supreme of y ∈ RN in |α|≤s |x−y|≤d (uα , uα )(T )dx, we get

C ∥ u(T ) ∥s,d
∫ T
−1
≤C ∥ u0 ∥s,d+RT +C(G1 , s, d, RT )M2 (1+ ∥ u ∥s,d )
0
∫ T
+(1 + M1 ) ∥ u(t) ∥s,d+R(T −s) (3.29)
0

i.e.
∫ T
∥ u(T ) ∥s,d ≤ C̃(C, d, RT, G1 , s)(1+ ∥ u0 ∥s,d+RT +(M1 + M2 ) ∥ u(t) ∥s,d )
0
(3.30)

Define E(t) =∥ u(t) ∥s,d , then

∫ T
E(T ) ≤ C̃(1 + E(0) + M E(t)dt) (3.31)
0

where M = M1 + M2 . So by Gronwall Inequality, we have:

E(T ) ≤ C̃(E(0) + 1)eC̃M T (3.32)

i.e.

∥ u(t) ∥s,d ≤ C̃(∥ u0 ∥s,d +1)eC̃M T (3.33)

Thus we get the core energy estimate. Then there is nothing different
left to prove the extension of Theorem 3.1 and Theorem 3.2. We state these
results:

3.2 Main Theorem

Theorem 3.1. (Annex) Assume u0 ∈ Hul s , s > N + 1 and u (x) ∈ G ,
2 0 1
G1 ⊂⊂ G. Then there is a time interval [0, T ] with T > 0 so that the
equation in (1) have a unique classical solution u ∈ C 1 ([0, T ] × RN ) with
u(x, t) ∈ G2 , G2 ⊂⊂ G for (x, t) ∈ RN × [0, T ]. Furthermore, u(t) ∈
s ) ∩ C 1 ([0, T ], H s−1 ) and u(t) ∈ L∞ ([0, T ], H s ); also T depends
C([0, T ], Hloc loc ul
only on s and ∥ u0 ∥s,d .
Theorem 3.2. (Annex) If u0 ∈ Hul s , s > N + 1, and for any time inter-
2
val [0, T∗ ] of local Hul existence with T∗ ≤ T , the following three a priori
s

estimates are satisfied:

⇀ |L∞ ≤ M1 , 0 ≤ t ≤ T∗
| div A (3.34)
| Du |L∞ ≤ M2 , 0 ≤ t ≤ T∗ (3.35)
u(x, t) ∈ G1 ⊂⊂ G, (x, t) ∈ R × [0, T∗ ]
N
(3.36)
21
 Then the classical Hul s solution of (1) exists on the interval [0, T ] with
s ) ∩ L∞ ([0, T ], H s ). Also, a similar a
u(t) ∈ C([0, T ], Hloc ) ∩ C 1 ([0, T ], Hloc
s
ul
priori estimate like the one in the notes before is also valid in Hul s norms:

∥ u(t) ∥s,d,T∗ ≤ C̃(∥ u0 ∥s,d +1)eC̃M T (3.37)

Acknowledgement
I would like to express my gratitude towards Professor Ning Jiang of Math-
ematical Sciences Center for his guidance in this note.

22
 一个优美恒等式的证明

郑可迪*

来看一个优美的恒等式：
∫ 1 ∑ 1 ∞
1
dx = (1)
0 xx nn
n=1

据说这个等式是约翰-伯努利发现的，下面用微积分的方法给出一个证明。
实际上，左边的 1
xx = e−xlnx 。由泰勒展开：
∞
∑ xk
e−x = (−1)k
k!
k=0

马上可以得到：
∫ ∫ ∫ ∞
1∑
1
1 1
−xlnx (xlnx)k
dx = e dx = (−1)k dx (2)
0 xx 0 0 k=0 k!

设f (x) = xlnx，对它求导，得到：f ′ (x) = lnx + 1。易知x = e−1 时，f (x)

取到最小值− 1e ，而我们知道

lim f (x) = f (1) = 0

x→0+

k
在区间 (0,1] 上，|f (x)| 的最大值为e−1 。设uk (x) = (−1)k (xlnx)
k! 。在区间
(0,1] 上，
1
|uk (x)| ⩽
k!ek
∑
∞
我们找到了一个强级数。由魏尔斯特拉斯比较判别法知，函数级数 uk (x)
k=0
在 (0,1] 上一致收敛。于是 (2) 式右端的积分可以变为级数求和：
∫ ∞
1∑ ∑ ∞ ∫
(xlnx)k 1
(xlnx)k
(−1)k dx = (−1)k dx (3)
0 k=0 k! 0 k!
k=0
*
20130
23
由分部积分进行递推（其实这里有点像 Wallis 公式的推导）：
∫ 1 ∫ 1 1 ∫ 1
1 1  k
(xlnx)k dx = lnk xdxk+1 = (xk+1 lnk x) − xk lnk−1 xdx
0 k + 1 0 k + 1 0 k + 1 0
∫ 1 ∫ 1
k k! k!
=− xk lnk−1 xdx = ... = (−1)k k
xk dx = (−1)k
k+1 0 (k + 1) 0 (k + 1)k+1

把这个结果代入到 (3) 中，马上得到：

∞ ∫
∑ ∑∞ ∞ ∑ 1
1
(xlnx)k 1
(−1)k dx = k+1
= (4)
0 k! (k + 1) nn
k=0 k=0 n=1

结合 (2)(3)(4) 马上可以得到 (1) 式成立。

24
 解析流形上的李群作用

郑志伟*

摘要

该理论的局部部分由 Sophus Lie 本人发展，标志着李群李代数整套理论的开始。

该理论的整体部分是很新的结果，本文会做简单的介绍。
关键词：李群和李代数，无穷小作用，局部，整体。

1 引言
我们关心的是李群 G 在解析流形 M （以下只考虑解析的流形）上的作用的无穷小刻
画。记 g 为 G 的李代数，Ja (M ) 为流形 M 上的解析向量场构成的李代数。如果 G 在 M
上作用，则 t 7−→ exp(−tX) 7−→ exp(−tX) · x 给出了 M 上的一族曲线，由此给出 M 上
的一个向量场。记这样的映射为 τ : g −→ Ja (M )。李的第一个基本结果是说这样得到的 τ
是一个李代数同态。
再考虑 M 上所有的微分同胚构成的群 Dif f (M )，则李群作用给出 G 到 Dif f (M )
的一个群同态，记为 α。若把 Dif f (M ) 近似看做一个李群，则其李代数为 Ja (M )，可将
τ 看作 α 的微分。所以自然地，称 τ 被 G 在 M 上的作用决定，称每一个 g 到 Ja (M ) 的
李代数同态是 G 在 M 上的无穷小作用。李的第二个基本结果是说对每个 G 在 M 上的
无穷小作用，都可以局部地（某种意义下唯一的）构造 G 在 M 上的作用，恰好决定该无
穷小作用。
李本人并没有考虑整体的情况。事实上，并不是对每个无穷小作用，都可以找到一个整
体作用与之对应。这里就可以举一个简单的观点：设 τ 是 G 在 M 上的无穷小作用，并设其
由一个 G 在 M 上的整体作用决定。取 M ′ 为 M 的开子流形，令 τ ′ (X) = τ (X)|M ′ (X ∈ g)，
则 τ ′ 无法由一个整体作用决定。
最后，文章会简单介绍该理论整体意义下的结果。
*
基数 01

25
 2 基本定义
下面一直令 n = dim(G), m = dim(M ).

定义 2.1. D 为 G × M 中 1 × M 的开领域，φ 为从 D 到 M 的解析映射，满足

1. φ(1, x) = x

2. E = {(g, h, x) ∈ G×G×M |(h, x), (gh, x), (g, φ(h, x)) ∈ D 且 φ(gh, x) = φ(g, φ(h, x))}
是 1 × M 的开领域。

则称 φ 为 G 在 M 上的一个局部作用。
如果上述定义中 D = G × M 并且 E = G × G × M ，则称 φ 为 G 在 M 上的一个整
体作用。
将 φ(g, x) 简记为 g · x。

定义 2.2. 称每一个 g 到 Ja (M ) 的李代数同态是 G 在 M 上的无穷小作用。

定义 2.3. G 上所有左不变微分算子构成的代数称为 G 的包络代数，记为 O。∀g ∈ G，可

将 O 与 Tg∞ (G) 等同。

记 φx : g 7−→ g −1 ·x 在 G 的单位元附近定义且解析，∀x。记 τφ∞ (a)x = (dφx )∞

1 (a), ∀a ∈
O，这是 M 上 x 附近的一个微分算子。如果 a = X1 X2 . . . Xr ，则 τφ∞ (a)x 由下式决定：
∂ ∂
τφ∞ (a)x (f ) = ( ... f (exp(−tr Xr ) . . . exp(−t1 X1 ) · x))0 (1)
∂t1 ∂tr
引理 2.1. ∀a ∈ O，τφ∞ (a)(x 7−→ τφ∞ (a)x ) 是 M 上的一个解析的微分算子。并且 τφ∞ (a 7−→
τφ∞ (a)) 是从 O 到 M 上微分算子代数的同态。

引理的前半部分是显然的。我们只需要验证 τφ∞ 是一个代数同态, 为此利用 (1) 进行

计算即可。具体证明省略。

推论 2.1. τφ = τφ∞ |g 是从 g 到 Ja (M ) 的李代数同态。

于是，τφ 是 G 在 M 上的一个无穷小作用，它被 τ 决定。

3 作用的图像与唯一性
令 τ (X 7−→ τ (X)) 是 G 在 M 上 的 无 穷 小 作 用。 相 应 的 有 τ̄ (X) : (g, x) 7−→
(Xg , τ (X)x )，这是流形 G × M 上的解析向量场。易见，τ̄ 是从 g 到 Ja (G × M ) 的
嵌入李代数同态。令
Lτ(g,x) := {τ̄ (X)(g,x) |X ∈ g}

26
于是有
Lτ : (g, x) 7−→ Lτ(g,x)

是 G × M 上一个相容的切向量系统, 维数等于 dim(G) = n。用 Frobenius 定理，对

(g, x) ∈ G × M ，可记 S(g,x) 为 Lτ 的经过 (g, x) 的极大积分子流形。

引理 3.1. 1. ∀h ∈ G，定义
λh : (g, x) 7−→ (hg, x)

为 G × M 上的一个解析同胚。则

λh (S(g,x) ) = S(hg,x)

2. 如果 φ 是 G 在 M 上的整体作用使得 τ = τφ ，则 ∀x ∈ M ，映射：

αφ,x : g 7−→ (g, g −1 · x)

是从 G 到 S(1,x) 的解析同胚。

3. 假设 φ 是 G 在 M 上的局部作用使得 τ = τφ 。给定 x0 ∈ M ，则存在 G 中 1 的连

通开领域 V (V = V −1 )，M 中 x0 的开领域 U，满足：∀x ∈ U, αφ,x 将 V 解析同胚
到 S(1,x) 的一个经过 (1, x) 的开子流形上。

因为 Lτ 在 λh 诱导的切映射下不动，∀h ∈ G，所以 λh (S(g,x) ) 是经过 (hg, x) 的积分

子流形，反向分析马上证得引理第一部分。
设 φ 是 G 在 M 上的整体作用使得 τ = τφ ，则 ∀X ∈ g, x ∈ M 有：

(dαφ,x )g (Xg ) = (Xg , τ (X)g−1 ·x ) = τ̄ (X)αφ,x (g)

记 Ax = αφ,x (G) 并给其从 αφ,x 得到的解析结构。由之前分析知道 Ax 是 S(1,x) 的开子流

形。作为 G × M 中的图像（graph），Ax 是闭的，从而等于 S(1,x) 。我们证明了引理的第
二部分。
引理的第三部分是第二部分的局部形式，证明类似。

推论 3.1. 令 τ 为 G 在 M 上的一个无穷小作用。如果 φ1 和 φ2 均是 G 在 M 上的整体

作用，并且 τφ1 = τφ2 = τ ，则 φ1 = φ2 。如果 φ1 和 φ2 均是 G 在 M 上的局部作用，并
且 τφ1 = τφ2 = τ ，则在 1 × M 的某个领域中 φ1 = φ2 。

先 考 虑 整 体 的 情 况。 记 pG 为 G × M 到 G 的 投 影，pM 为 G × M 到 M 的 投
影，pG,x = pG |S(1,x) 。 由 上 面 引 理 知 道 pG,x 是 从 S(1,x) 的 解 析 同 胚， 且 φ1 (g, x) =
pM ◦ p−1 −1
G,x (g ) = φ2 (g, x)，所以 φ1 = φ2 。

27
 如果 φ1 和 φ2 仅仅是局部的作用。固定 x0 ∈ M ，如之前引理取 U, V 。固定 x ∈ U ，
记 Ax,i = αφi ,x (V )(i = 1, 2)。则 Ax,1 和 Ax,2 均是 S1,x 的经过 (1, x) 的开子流形，从而
Ax,1 ∩ Ax,2 同时是 Ax,1 和 Ax,2 的开子流形。考虑集合

Wx = {g|g ∈ V, φ1 (g, x) = φ2 (g, x)}

它对应与 Ax,1 ∩ Ax,2 从而是 V 中开的，显然它也是闭的，所以 Wx = V 。于是在 V × U

上 φ1 = φ2 ，由于 x0 的任意性，必有在 1 × M 的某个开领域上 φ1 = φ2 。
上述推论给出了唯一性。

4 李的基本定理
首先我们给出一个引理：

引理 4.1. 令 τ 是 G 在 M 上的一个无穷小作用，Lτ 是如之前一样定义的切向量系统。给

定 x0 ∈ M ，则存在 G 中 1 的连通开领域 V (V = V −1 )，M 中 x0 的开领域 U，以及从
V × U 到 M 的解析映射 ψ，使得：

1. ∀x ∈ M, ψ(1, x) = x。

2. (g, x) 7−→ (g, ψ(g −1 , x)) 是从 V × U 到 (1, x0 ) 的一个开领域的解析同胚。

3. ∀x ∈ U ，映射
g 7−→ (g, ψ(g −1 , x))

是 V 到 S(1,x) 的一个包含 (1, x) 的开子流形的解析同胚。

引理的证明略琐碎，以后有空我再码上。现在我们可以陈述和证明李的第一个基本定
理：

定理 4.1. 令 τ 是 G 在 M 上的一个无穷小作用，给定 x0 ∈ M ，则存在 M 中 x0 的开领

域 U，以及 G 在 U 上的局部作用 φ，使得 ∀X ∈ g, τφ (X) = τ (X)|U 。

固定 x0 ∈ M ，取上述引理中的 V, U, ψ。令 D = {(g, x) ∈ V × U |ψ(g, x) ∈ U }。易知

D 是 1 × U 在 G × U 中的一个开领域。令 φ = ψ|D，我们证明 φ 是 G 在 U 上的局部作
用，使得 ∀X ∈ g, τφ (X) = τ (X)|U 。
固定 x ∈ U 。由定义，φx : g 7−→ (g, g −1 · x) 是 V 到 S(1,x) 的开子流形的解析同
胚。于是，任意 X ∈ g，(dφx )1 (X1 ) = τ (X)x 。我们剩下只需要验证的是 φ 在 U 上作
用的局部结合律。为此，取 G 中单位元的连通开领域 V1 = V1−1 , V2 = V2−1 ，M 中 x 的

28
连通开领域 U1 , U2 满足：V22 ⊂ V1 , V12 ⊂ V, ψ(V2 × U2 ) ⊂ U1 , ψ(V1 × U1 ) ⊂ U 易见，对
(g, h, y) ∈ V2 × V2 × U2 , h · y, gh · y, g · (h · y) 有定义。欲证明：gh · y = g · (h · y), ∀(g, h, y) ∈
V2 × V2 × U2 , i.e. 这与 ψ(h−1 g −1 , y) = ψ(h−1 , ψ(g −1 , y)), ∀(g, h, y) ∈ V2 × V2 × U2 一样。
固定 g ∈ V2 , y ∈ U2 ，令 z = ψ(g −1 , y)，定义

α(h) = (gh, ψ(h−1 g −1 , y)), β(h) = (gh, ψ(h−1 , z)), h ∈ V2

由之前引理，α( 相应的 , β) 是从 V2 到 S(1,y) ( 相应的，S(g,x) ) 的开子流形 A（相应

的，B）的解析同胚。A 和 B 均为 Lτ 的积分子流形，(g, z) ∈ A ∩ B，所以 A ∩ B 是
A 和 B 的非空开子流形。类似上一节推论中的论述，可以证明在 V2 上 α = β，从而
ψ(h−1 g −1 , y) = ψ(h−1 , z)，由 x ∈ U 的任意性，我们完成了证明。

5 向整体推广的第一步
上一节中得到的李的基本定理，从 G 和 M 的变量来看都是局部的。很自然的，我们
问能否把定理推广到整体的情况。下面，我们首先考虑对 M 这边的推广。我们需要一个
引理：

引理 5.1. 令 {Ui : i ∈ I} 为 M 的一个局部有限开覆盖。任意的 (i, j) ∈ I × I 和任意的

x ∈ Ui ∩ Uj ，令 Uijx 为 x 的包含于 Ui ∩ Uj 的一个开领域。则任意的 x ∈ M ，可以选择
x 的开领域 Ux 使得：

1. 若 x ∈ Ui ∩ Uj ，则 Ux ⊂ Uijx

2. 若 Ux ∩ Uj ̸= ∅，则存在 i ∈ I 使得 Ux ∪ Uy ⊂ Ui

引理的证明完全是点集拓扑的，关键在于对每个 x 都设法取其足够小的领域。
利用这个引理我们有了第一步的整体推广。

定理 5.1. 令 τ 是 G 在 M 上的一个无穷小作用，则存在 G 在 M 上的局部作用 φ，使

得 τφ = τ 。

根据上节定理，我们可以选取 M 的一个局部有限开覆盖 {Ui |i ∈ I}，以及 φi 为 G 在

Ui 上的局部作用，满足：τφi (X) = τ (X)|Ui , (X ∈ g, i ∈ I)。对每个 (i, j) ∈ I × I，每个
x ∈ Ui ∩ Uj ，可以选取 x 的开领域 Uijk ⊂ Ui ∩ Uj ，使得 φ1 , φ2 在 1 × Uijk 的某个领域上
定义且相等。由上面的引理构作 Ux 。对每个 x ∈ M ，令 I(x) = {i ∈ I|x ∈ Ui }，令 Wx
是使得所有 φi (i ∈ I(x)) 可定义且取相同的值的 (g, y) ∈ G × Ux 的集合。I(x) 有限，故
Wx 是 1 × Ux 的开领域。令 φx = φi |Wx , i ∈ I(x)。设有 x, y ∈ M, s.t., Wx ∩ Wy ̸= ∅，于

29
是 Ux ∩ Uy ̸= ∅，所以存在 i ∈ I(x)，使得 Ux ∪ Uy ⊂ Ui ，这意味着 i ∈ I(x) ∩ I(y)，所以
φx |Wx ∩ Wy = φy |Wx ∩ Wy = φi |Wx ∩ Wy ，从而可以把全部 φx 拼成 φ，它就是我们要的
局部作用！

6 G 上变量的整体推广
这一节我们不加证明地给出关于 G 上变量的整体推广的结论。这一部分较之前更加
有趣，但是更难。首先，我们有定理 5.1 的推论如下：

定理 6.1. 设 M 紧，G 单连通。则任意 τ 是 G 在 M 上的一个无穷小作用，都存在 G 在

M 上的整体作用 φ，使得 τφ = τ 。

如果 M 不是紧的，则不一定可以找到一个整体作用决定我们给的无穷小作用，即使
G 是单连通的。但是我们现在有很好的结果：

定理 6.2. 令 τ 是 G 在 M 上的一个无穷小作用。如果：

1. G 单连通

2. ∀X ∈ g, τ (X) 是 M 上的整体向量场

则存在唯一的 G 在 M 上的整体作用 φ，使得 τφ = τ .

事实上，这个定理还可以更强。可以把第二个条件减弱为：对 g 的一个代数生成集中
的任意向量场 X，τ (X) 是整体的。

参考文献
[1] V.S.Varadarajan: Lie Groups, Lie Algebras, and Their Representations Page 121-132

30
 Introduction to Heegner Points
Qiu Congling∗

The theory of Heegner points started with Heegners paper in, and then Birch first un-
dertook a systematic study in the 70’s. Then, Gross, Zagier, Kolyvagin etc. It played an
important role in the number theory of the next three decades, for example, the proof of
Gauss class number 1 theorem. Now, it has been studied deeply, but is still not fully un-
derstood by us. Someone, like Zhang Shouwu, hopes to use this theory to prove the BSD
conjecture.
The original application the theory of Heegner points is to find nontrivial rational points
on elliptic curves. But in the later 80’s, the Gross-Zagier formula linked it to L-functions
and Kolyvagin’s Euler system showed the functorial properties of Heegner points, just like
Weber’s theory of CM.
In this paper, we will establish some basic properties of Heegner points, the existence and
some actions on them. Also, we introduce the work of Gross-Zagier and Kolyvagin. But we
can not fully described the theory of L-functions and elliptic curves. So, some important
results will be omitted, for example, Kolyvagin’s theorem also shows the system of Heegner
points controls the size of Selmer group.

1 Preliminary knowledge
Some necessary preliminary knowledge in number theory is given below. Also some about
elliptic curves. Let K be an imaginary quadratic field, OK the integer group. An order O in
K is a subring of OK , at the same time a Z-module of rank [K : Q]. The conductor of O is

c = {x ∈ OK |OK ⊂ O},

which is an integral OK -ideal. Notice that c ⊂ O and is the largest O-ideal which is also an
OK -ideal. A fractional ideal a of O is a finitely generated O-submodule in K. a is called
invertible if there is some fractional ideal b such that

ab = O.

These invertible ideals form an abelian group and

a−1 = {x ∈ K : xa ⊂ O},
∗
20110

31
which is the biggest fractional ideal such that

aa−1 ⊂ O.

The importance of conductor is that it gives a criterion whether an integral ideal a is invert-
ible, saying
(a, c) = O ⇐⇒ a is invertibe.
√
When K is an quadratic field Q( d), where d is a square free integer, define the discrim-
inant {
d d ≡ 1( mod 4)
dK =
4d otherwise
then √
dK + dK
OK = Z[wK ], where wK = .
2
An order of K must be of form

O = Z[cwK ], where c = [OK : O]

and c = cOK is the conductor of O. The discriminant of O is defined as D = c2 dK . Notice

that O is uniquely determined by c or D. In fact, the definition of discriminant comes from
the discriminant of a Z-basis [?], in this way, we can define discriminant generally.
We also recall the ramification theory in quadratic fields, a prime number p is said

split if p = p1 p2 , then O/pi ∼

= Z/pZ, i = 1, 2;
ramified if p = p2 , then O/p1 ∼
1 = Z/pZ;
inert if p is still prime, then O/p ∼
= Fp2 .
The importance of discriminant is that the ramified prime numbers are precisely those
dividing the discriminant dK (this is valid for all number field). So when p ̸= 2, p||D
(exactly divides) is equivalent to that p is ramified in K. 2 ramifies if and only if d ̸≡ 1(
mod 4).
Class field theory and CM theory. The ring class field of an order O is defined by class
field theory, and could be generated by value of j(a) over K, i.e.

H(O) = K(j(a))

here a is any ideal, regarded as a lattice, j is Klein j-invariant. In fact, two ideal in a same class
have same j-value, and j-value between different classes are conjugate algebraic integers. In
fact, they are roots of the so called Hilbert polynomial,[?] . There is an isomorphism (known
as Artin map) between the class group Pic(O) of O and the galois group GH/K . Denote σ(d)
the Artin symbol (the image of artin of Artin map) d, then the main theorem of CM say

d(j(a)) = j(ad−1 ).

32
 Elliptic curves [?] Ch3, Ch8. The invariant differential of an elliptic curve defined over k
with Weierstarss equation

E : y 2 + a1 xy + a3 y = x3 + a2 x2 + a4 x + a6 , ai ∈ k

is defined as
dx
ω=
2y + a1 x + a3
which is invariant under translation, i.e. if P ∈ E, TP (Q) = P + Q is a translation, then

TP∗ (ω) = ω.

Check directly that it has no poles and zeros, i.e. div(ω = 0). If

ϕ : E1 −→ E2

is an isogeny, and ω1 , ω2 are the corresponding invariant differential, then

ϕ∗ ω2 = cω1 , c ∈ k̄(x).

But since

div(c) = div(ϕ∗ ω2 ) − div(ω2 ) = ϕ∗ div(ω2 ) − div(ω2 ) = 0 − 0 = 0,

c ∈ k̄. When E is defined over C and given by a Weierstarss uniformization f : C/L → E,

then
f ∗ ω = cdz, c ∈ C
.
The canonical height on E is a real quadratic form ⟨·, ·⟩ on E, ⟨P, P ⟩ = 0 if and only if P
is torsion. And it can be extended to a Hermitian form on E(k) ⊗ C.
The L-series of an elliptic curves defined over Q, it’s imaginary quadratic twist, and it’s
L-series over imaginary quadratic field can’t be stated in a few lines, so we omit them.

2 CM elliptic curves
The classical correspondence

{Rank − 2 lattice in C}/homothety ←→ {elliptic curves over C},

which realizes complex tori as elliptic curves by the Weierstarss uniformizations. A isogeny
between two complex tori is a holomorphic homomorphism keep group structure, must be
multiplication by a complex number. Up to homothety, a degree-N cyclic isogeny (i.e. the
kernel is Z/N Z) could be given by

C/Z[τ ] → C/Z[1/N, τ ], z 7→ z. (1)

33
It can also be defined in a totally algebraic way. Use Tate-module and some algebraic
classification theroem, we can proof that End(E) is Z or an order in some imaginary quadratic
field K when E is defined over some field F with characteristic 0[?]. In the latter case, we
say E has CM (complex multiplication). When F = C, the proof is much easier[?]. We
sketch the proof. Let L = Z[τ ] be a lattice, ∀α ∈ End(EL ) = {x ∈ K : xL ⊂ L}, we have

α = a + bτ, ατ = c + dτ

for some integer a, b, c, d, which gives

α2 − (a + d)α + ad − bc = 0.

So End(EL ) is Z-integral. If Z ⫋ End(EL ), suppose the above α ̸∈ Z, then b ̸= 0. We have

bτ 2 − (a − d)τ − c = 0.

So, O = End(EL ) is an order in K. By definition, L is a fractional ideal of O = End(E).

Moreover L is invertible. Indeed, It ca be shown by a more careful discussion. As in [?],
suppose τ satisfies

Ax2 + B 2 + C = 0, A, B, C ∈ Z, gcd(A, B, C) = 1

then √
D+ D 1
Oτ = OD = Z[ ] and LL = O
2 A
i.e. L−1 = AL. One has to realize that it doesn’t mean that any fractional ideal is invertible,
it depends on the order (one fractional ideal of O is fractional ideal of any order contained
in O. Further more, [?] shows disc(O) = D = B 2 − 4AC. We define disc(L) to be the
discriminant of the orders of L.

3 Moduli curve X0(N )

Y = Y0 (N ) is the open modular curve of ordered pairs (E, E ′ ) with an degree-N cyclic
isogeny
ϕ : E → E ′ , ker ϕ ∼
= Z/N Z.
As in (1), it may be given by

(L, L′ )τ , L = Z[τ ], L′ = Z[1/N, τ ] = Z[N τ ]

Let
Γ0 (N ) = {∈ SL2 (Z) : N |c}.
We have an isomorphism

Y0 (N ) ∼
= Γ0 (N ) \ H, (L, L′ )τ 7→ τ.

34
X0 (N ) is the natural compactification of Y0 (N ), classification the so-called generalized elliptic
curves. let H∗ = H ∪ Q,
X0 (N ) ∼= Γ0 (N ) \ H∗ .
The function field of X0 (N )(C) is C(j(τ ), j(N τ )), and in fact this two functions satisfy a
polynomial defined over Q, called the N -th modular equation. Thus, X0 (N ) is defined over
Q, have a planar model realized by

Γ0 (N ) \ H∗ −→ X0 (N ), τ 7−→ (j(τ ), j(N τ )).

4 Heegner points: existence

A Heegner on Y0 (N ) is a pair (E, E ′ ) with the same endomorphism ring O. Choose
L = a, L′ = b as fractional ideals of O with a ⊂ b and b/a ∼ = Z/N Z. Then n = ab−1 is
∼
an invertible integral ideal, with O/n = Z/N Z. If we have such an n, choose an invertible
fractional ideal a, (in fact an ideal class [a]), let b = an−1 . Then Ea , Eb are two elliptic curves
with obvious isogeny. Denote this Heegner point as (O, n, [a]).
Use the results and symbols in 1, we can easily conclude that the above happens, i.e.
τ = N τ (after a homothety) and D′ = D if and only if
′

A = N A′ , C ′ = N C, B = B, gcd(A′ , B ′ , C ′ ) = gcd(A, B, C) = 1.

Thus the existence of Heegner points is equivalent to that D = B 2 − 4N C (replace N AC by

N C) has an integer solution with (N, B, C) = 1.
I also give another description of the existence of Heegner points. It can be derived from
the previous one by analyze the equation, but we can see it from another viewpoint.
Let IK (c) be the monoid of integral OK -ideals prime to c, IO (c) the monoid of integral
O-ideals prime to c, i.e. invertible integral O-ideals. Then there is a multiplicative bijection
(isomorphism) between this two monoids given by

a 7→ a ∩ O, a ∈ IK (c); a 7→ aOK , a ∈ IO (c) (2)

From now on, for the discussion about Heegner point, we always suppose that N prime to c.
So, we always have N O is an invertible ideal, so is every factor of N . Now, for a ∈ IK (c),
consider a homomorphism
O/a ∩ O → OK /a,
which is injective. Since c is prime to a, multiplication by c induces an isomorphsim of OK /a.
But cOK ⊂ O, the homomorphism is surjective. So, we conclude that the existence of an
integral O-ideal n with
O/n ∼= Z/N Z
is equivalent to the existence of an integral OK -ideal m with O/m ∼ = Z/N Z. Further more,
it’s equivalent to that every prime number p dividing N is split or ramified in K ( if p
is inert, their will appear Fp2 in the quotient ), p with p2 dividing N is split in K ( if it’s
ramified, p||D, but it’s shown in 3 that we must have D ≡ B 2 ( mod 4N ), so p|B, and P 2 |D,

35
a contradiction ). In the split case, p has only one prime p of O dividing n over it, otherwise,
there will be a subgroup (Z/pZ)2 in O/n. Since we assumed (c, N ) = 1, p| gcd(D, N ) means
p|dK is ramified, so p||N . Moreover, if p ̸= 2, p||dK , if p = 2, 4|dK . Usually, we accept the
Heegner hypothesis assuming gcd(dK , N ) = 1, or equivalently, all p|N are split. Then, there
exist Heegner points for each c coprime to N .
We can count the number of Heegner points.Suppose the prime factorization ofN is
∏
r ∏
s
b
N= pai i qj j , pi |dK , qj ∤ dK
i=1 j=1

(As we have shown, in fact ai = 1, and if assuming the Heegner hypothesis, r = 0) and h(O)
be the class number. Then

#{heegner points of level N with order O} = 2s h(O)

5 Galois action on Heegner points

Consider the action of Aut(C) on X(C), which is given by

E : y 2 = x3 + ax + b 7−→ E σ : y 2 = x3 + σax + σb, σ ∈ Aut(C).

Notice Aut(C) is a semi product of it’s normal subgroup AutK (C) and the order-2 group
generated by the complex conjugation. For the complex conjugation, we have

(O, n, [a]) = (O, n, [a]). (3)

For σ ∈ AutK (C), since the j-invariants j(a) of all ideal classes of an order are conjugate
algebraic integers, So the order of (O, n, [a])σ is O too . In a special case, this point could
be given by a explicit formula by the CM theory. Denote σ(d) the Artin symbol of d, then
d(j(a)) = j(ad−1 ). Since ad−1 (bd−1 )−1 = ab−1 , we have

(O, n, [a])σ = (O, n, [aσ ]) = (O, n, [ab−1 ]). (4)

In particular, we know that (O, n, [a]) are over K(j(O)), the ring class field of O).

6 Hecke operators and Atkin-Lehner involutions

For any congruence subgroup Γ, there ia a natural correspondence between wight-k in-
variant functions respect to Γ and degree-k homogeneous functions defined on Y (Γ). When
Γ = Γ0 (N ), it’s given by
f (τ ) = F (C/Z[τ ], C/Z[1/N, τ ])
where F satisfies for all m ∈ C∗ ,

F (C/(mZ[τ ]), C/(mZ[1/N, τ ])) = m−k F (C/Z[τ ], C/Z[1/N, τ ]).

36
Extend F additively to a function defined on Div(X0 (N )), then an operator T from DivX0 (N ))
to DivX0 (N )) acts on F by T F (D) = F (T D), so on f . Thus actions on modular space induce
actions on modular forms. But historically, always the latter is discovered earlier.
The actions we discuss here are the Hecke operators (or Hecke correspondences) and the
Atkin-Lehner involutions. We give their their actions on X0 (N ) at first, since it make more
sense. Then we compute their actions on modular forms. And we shall prove that the Hecke
operators and the Atkin-Lehner involutions on X0 (N ) map Heegner points to Heegner points
of K.
Let x = (E, E ′ , ϕ) be a point in X0 (N ), where ϕ is a given isogeny. Every m-cyclic
subgroup C of E, with C ∩ ker ϕ = ∅ gives another point (E/C, E ′ /ϕ(C), ϕ/C) ∈ X0 (N ),
where ϕ/C means the induced isogeny. Let E(m) be the set of cyclic subgroups of order m,
∑
Tm (x) = (E/C, E ′ /ϕ(C), ϕ/C).
C∈E(m)
C∪ker ϕ=∅

We mainly focus on m with gcd(m, N ) = 1, then the classical product formula follows quickly.
The point is tha when gcd(m, n) = 1, taking m-cyclic and a n-cyclic gives a nm-cyclic, but
taking apply Tp and Tpk may gives a (Z/pZ)2 ⊕ (Z/pk−1 Z), the quotient by it is the same
with the quotient by a Z/pk−1 Z.
As a correspondence X0 (N ) ⇝ X0 (N ), Hecke operator Tm , is defined by the following
diagram
α β
X0 (N ) ← X0 (mN ) → X0 (N )
Since (m, N ) = 1 (E, G) ∈ X0 (mN ) must be of form G = C ⊕ D, where C, D are cyclic of
order N, m. Then α forgets D, β quotients by D:
α : (E, G) 7→ (E, C)
β : (E, G) 7→ (E/D, (C ⊕ D)/D)
In particular, Tp is the composition
α∗
∑ β∗
∑
(E, C) 7→ (E, (C ⊕ D) 7→ (E/D, (C ⊕ D)/D
D∈E(p) D∈E(p)

By a bookkeeping computation,
∑
p
Tp (Z[τ ], Z[1/N, τ ]) = (Z[γj τ ], Z[1/N, γj τ ])
j=0
[ ] ] [
1 j p 0
where γj = for j ̸= p, which is given by < j+τ p
> and γp = which is given by
0 p 0 1
< p1 >.
To get it’s action on Sk (Γ0 (N )), only need to notice that
[ ]−1 [ ]
1 0 1 0
Γ0 (pN ) = Γ0 (N ) ∩ Γ0 (N )
0 p 0 p
and The following technical lemma [?]

37
Lemma 6.1. Let α ∈ GL2 (R)+ , and for a congruence subgroup Γ,
Γ = ⨿(Γ ∩ α−1 Γα)βi
then
ΓαΓ = ⨿(Γαi ), αi = αβi .
Thus we get the usual double coset operators, which forms an abstract Hecke ring. And
the actions on Sk (Γ0 (N )) is given by
∑
Tk (p)f = pk/2−1 f |[αi ]k .
i

If (Z[τ ], Z[1/N, τ ]) is a heegner point of K, proceed as in section 2 and 4, we can show

that disc(Z[γj τ ]) and disc(Z[1/N, γj τ ]) have the same discriminant dividing Dp2 (recall as
we defined in section 2, it’s equal to the discriminant of minimal polynomial of γj τ and
N γj τ ). So, they are of the same order, and (Z[γj τ ], Z[1/N, γj τ ]) is a Heegner point. But we
usually can’t say much about this order in general without explicit computation. Thus, we
have a formula ∑
Tp (O, n, a) = (End(b), nb , b). (5)
[a:b]=p

[?] claim that nb = nEnd(b) ∩ End(b).

Moreover, disc(Z[γj τ ])|p2 D (D = disc(Z[τ ])). So, the conductor of (M, M ′ ) is prime to
N , provided so is (L, L′ ).

The Atkin-Lehner involutions are from taking dual isogeny. We define wN (E, E ′ , ϕ) =
b where ϕb is the dual isogeny of ϕ. Generally, let Q|N, gcd(Q, N/Q) = 1, G = ker ϕ,
(E ′ , E, ϕ),
G = ker ϕ′ , G[Q], G′ [Q] the (unique) order-Q subgroup part of G, G′ . wQ is the composition
′

E/G[Q] → E/G ∼
= E ′ → E ′ /G′ [Q]
It’s visible that
Observation. wQ is a intermediate process of wQ′ if Q|Q′ , when Q = N , we just take dual
isogeny. It’s clear wQ1 wQ2 = wQ1 Q2 for gcd(Q1 , Q2 ) = 1. Since E/E[Q] ∼
= E, we also have
wQ = 1. Thus we find that all the wQ ’s forms a group isomorphic to (Z/2Z)r , denoted as
2

W , where r is the number of factors of N .

For the computation of the matrices of the actions on Sk (Γ0 (N )), as usual, take E =
C/Z[τ ], G = ⟨1/N ⟩, so G[Q] = ⟨1/Q⟩. It’s direct
[ ] ⟨ ⟩
1 τ 1
wp (E, G) = (C/Z ,τ , + ).
Q Q N
Only need to find a Z-basis (ω1 , ω2 ) of Z[ Q1 , τ ], such that
τ 1
+ = ω1 /N,
Q N

38
then wQ (E, G) = (Z[ω2 /ω1 ], 1/N ). Any choice of integers x, y such that xQ − y(N/Q) = 1
produce the desired ω2 = xτ + y/Q. Thus
[ ]
Qx y
ω2 /ω1 = τ,
N Q

denote this matrix as WQ , In [?], it’s given by

[ ]
Qx y
WQ =
N z Qw

. Also, a direct computation shows WQ is in the normalizer of Γ0 (N ) and the action doesn’t
depend on the choice of a, b. We have now obtain the Atkin-Lehner involution on Sk (Γ0 (N )).
Compute directly, we can verify the observation in the end of the last paragraph in true when
regarded as actions on Sk (Γ0 (N )).
Proceed as previous, by comparing discriminant, we can show that wQ (O, n, a) is of the
same order with (O, n, a). [?] gave a precise formula. Let wp = wpordp (N ) . We showed in
section 4, p has only one prime p of O dividing n over it so

wp ((O, n, [a])) = (O, [np−k pk , ap−k ]). (6)

∏
This coincides with our previous observation. Moreover wN = p|N wp given by

wN ((O, n, [a])) = (O, n, [an−1 ]) (7)

is the dual isogeny, and if a = Z[ −N

1
τ
] (up to homothety), then
[ ]
−1 1
an = Z .
−N τ
Combining the actions of W and G = GH/K , we see that W × G acts transitively on the
set of Heegner points, and Wp ∈ W with p|dK acts trivially. Thus, W/⟨Wp : p|dK ⟩ × G acts
transitively and freely on the set of Heegner points.

7 Hecke algebra of Γ0 and Atkin-Lehner theory in gen-

eral
Define three commutative algebras

T0 = Z[Tp , p is prime and gcd(p, N ) = 1 ],

T = Z[Tp , p is prime and gcd(p, N ) = 1; Tq , q is prime and q|N ],

(recall the definition, as in section 6, we can similarly show Tq is defined as sum of γj , j ̸= p,
satisfying Tq Tn = Tqn .)

Tw = Z[Tp , p is prime and gcd(p, N ) = 1; wq , q is prime and q|N ].

39
Recall the action of wN is
−1 −1
2
f( )
wN (f ) =
Nτ Nτ2
Notice Tq and wq are not commutative. For the recursive relation of T, we have a formula
by formal L-series
∑
∞ ∏ ∏
Tn ns = (1 − Tp p−s + p1−2s )−1 (1 − Tp p−s )−1 .
n=1 p∤N p|N

Then Under the Petersson inner product for Sk (N ), both of Tp and wq are hermitian, but
Tq not. Thus there exists a basis fi consisting of normalized eigenforms of all elements in T0 ,
and all eigenvalues are real. When f is normalized, i.e. a0 (f ) = 1, we have Tp (f ) = ap (f )f .
There exists a so-called old subspace, i.e. space formed by f (eτ ) with f (τ ) ∈ Sk (M ) and
eM |N . Its orthogonal complement is called new space, which is stable under T and wq .
The eigenspaces are either old or new, moreover, each new space is spanned by a single
normalized form, called a new form. The action of Tq , wq on new form is as follows:

wq (f ) = ±f ; Uq (f ) = aq (f )f, if q||N, Uq (f ) = −q k−1 wf , if q 2 |N, aq = 0,

For the recursive relation, we can state as, for a new form

anp = an ap − p2k−1 an/p ; anq = an ap .

Denote ie (f (τ )) = f (eτ ) , S2 (Γ0 (M ))+ as new space, we have

⊕ ⊕
S2 (Γ0 (N )) = ie (S2 (Γ0 (M ))+ ),
M |N eM |N
⊕
S2 (Γ0 (M ))+ = Cf.
new form f

Attention: ie may not keep the property of being eigenform or new form, but it’s true that if
g is an eigenform of level N there must be some new form f of level M |N with an (g) = an (f )
for almost all n. ([?] Prop 5.8.4)
In the case k = 2, the previous discussion can be interpreted as follows. S2 (Γ0 (N ) could
be identified with U = H 0 (X(C), Ω1 ) (the holomorphic 1-forms on C) by

f 7→ 2πif (z)dz = ωf

which is a subspace of the first (deRham) cohomology group H 1 (X(C)). Generally, for any
compact Riemann surface S, denote the space of degree-1 smooth differential forms on S by
E 1 (S). E 1 (S) has a decomposition, say any w ∈ E 1 (S) can be write as f dz + gdz. Define
the ⋆-operator as
⋆(f dz + gdz) = i(f dz − gdz)
and a hermitian product ∫
⟨w1 , w2 ⟩ := w1 ∧ ⋆w2 .
S

40
Under this hermitian product, there is an orthogonal decomposition

E 1 (S) = dE 0 (S) ⊕ H 0 (S(C), Ω1 ) ⊕ H 0 (S(C), Ω1 ).

Now the Petersson inner product is the restriction of this hermitian product (up to a positive
number,).
Each eigenspace V can also considered as a real character (not Dirichlet) of T by setting
V (T ) equal to the corresponding eigenvalues. It’s the same (and more usually) we define for
every normalized eigenform a character

λf : T → R, T f = λV (T )f,

The image is Z[an (f ) : n ∈ Z>0 ], and the kernel is If = ker(λf ) = {T ∈ T : T f = 0} then,

we have an isomorphism
T/If ∼
= Z[an (f ) ].
The algebra T ⊂ EndC (S2 (Γ0 (N ))) is obviously free, and seems of infinitely rank, but in
fact, let
1
g = genus(X) = rankZ H1 (X, Z) = dimC S2 (Γ0 (N )) = dim U,
2
we have

Theorem 7.1. T is of a finitely generated free Z-module of rank g.

Proof: [sketch] Instead of U, we consider it’s dual Hom(U, C), and the pullback of T. Then
H1 (X, Z) is a sublattice in Hom(V, C) given by
∑ ∑ ∫
ai γi 7→ ai
γi

The quotient by H1 (X, Z) defines the Jacobian of X, isomorphic to Pic0 (X) (Abel-Jacobi).
Verifying the actions of Tp are the same under the isomorphism implies H1 (X, Z) is stable
under T ([?], sec 6.5). But it seems rather circuitous. In fact, we can express a loop by a pair
{τ, γτ } (a integral modular symbol) independent of the choice of a ∈ H∗ , where γ ∈ Γ0 (N ).
Then he action is given by ∑
Tp {τ, γτ } = {τ, γi τ }
here γi is defined by the double coset operatorTp . Verify directly H1 (X, Z) is stable under
T.[?]
Since the action of complex conjugation on X0 (N )(C) induces an action on H1 (X, Z)
which commutes with T. Thus, T in fact act on two submodules of rank g, and so is a
commutative subring of Mg (Z). Then its rank is at most g.
For the other side, consider the pairing

T ⊗ C × (S2 (Γ0 (N )), ⟨T, f ⟩ = a1 (T f ) (8)

it’s non-degenerate on the right. So rankZ T = dimC T ⊗ C ≥ g. 2 Thus we know that the

41
minimal polynomial of Tp is monic integral. So, the eigenvalues are algebraic integer, and so
is the coefficients of an eigenform f . Moreover,the number field of f ,

Kf = Q ⊗ Z[an (f )] = Q[an (f )]

is a finite real extension of Q.

In fact, we know more from the proof. Since the pairing in the proof is non-degenerate on
both sides,

Theorem 7.2. The space of modular forms with integer coefficients in S2 (Γ0 (N )) is a rank-g
lattice. Thus, S2 (Γ0 (N )) has a basis of forms with integer coefficients

Proof: It’s just the dual lattice of T in T ⊗ R ⊂ T ⊗ C 2

Surely, these integral forms are usually not newforms, even not eigenforms. But they can
be derived from newforms of each level M |N ( [?] Cor6.5.6).

8 Eichler-Shimura Construction and rational points on

Elliptic curves
A modular parametrization of level N of E is a nonconstant map Φ : X = X0 (N ) → E.
Then the pull-back of the invariant differential on E is then of the form ωf , where f ∈
S2 (Γ0 (N )). When E is defined over Q, we hope Φ is defined over Q. Suppose so it is, we can
use this map to find rational points on E.
Let K be a imaginary quadratic field as before, τ ∈ K, so Eτ is CM, but we do not require
P = (Eτ , EN τ ) to be heegner. let H be the ring class field of the order Oτ ∩ ON τ . Usually,
this two orders are not contained in each other, so, the intersection is smaller than both, and
the ring class field is bigger. By CM theory, we know that j(τ ), j(N τ ) ∈ H. So, we have a
CM point P in X(H). And we have Φ(τ ) ∈ E(H).
To find rational point over Q, the standard method is to apply trace map
∑
TrH/Q P = P σ.
σ∈G′

Here, the sum is being computed with the group law of Ef , and obvious, it’s over Q. Since
Φf is defined over Q, it communicate with G′ = GH/Q (we have defined G = GH/K .) Thus,
suppose τ is given by ([a], [b]), then
∑ ∑
TrH/Q (P ) = Φf (([a], [b]))σ = Φf (([σa], [σb]))
σ∈G′ σ∈G′

The action of σ on a, b is just restriction, and could be given by Artin map and the main
theorem of CM. Notice that since G′ is big, so TrH/Q P is very possible to be 0.
Now, we state the Eichler-Shimura construction, which provides such a modular parametriza-
tion.

42
 For a new form f , let N denote it’s level, as usual, denote J0 (N ) the jacobian of Γ0 (N ),
defined as
J0 (N ) = Hom(U, C)/H1 (X, Z)
For σ : Kf → C an embedding, define
∑
fσ = σ(an (f ))q n

it can be show that this action keeps the space of new forms S2 (Γ0 (N ))+ . Define the equiv-
alent class of f
[f ] = {f σ : σKf → C},
and the space spanned by this class

Vf = C([ωf ]) ⊂ U.

Then the action of H1 (X, Z) can be restricted to Vf , denote it as Λf , which is a sublattice of

Hom(Vf , C). The restriction map is a natural homomorphism from J0 (N ) to Hom(Vf , C)/Λf .
Use the pairing (3) and other arguments, we can proof
∼
Af := J0 (N )/If J0 (N ) −→ Hom(Vf , C)/Λf .

And Hom(Vf , C)/Λf is a dimension [Kf : Q] complex torus. Since Tp f − ap (f ) = 0, the

Hecke operator Tp acts on Af by multiplying ap (f ) (i.e. addition for ap (f ) times).
Moreover, if f has integral coefficients, i.e. [Kf : Q] = 1, we get an elliptic curve (not
denote it Af , but) Ef ∼
= Hom(Vf , C)/Λf . As we mentioned above, that we could choose a
generator of Λf to be {τ0 , γ(τ0 )}. It’s image (denote as Λf again) in Hom(Vf , C) ∼ = C is a
sublattice {∫ }
γ(τ0 )
Λf = ωf : γ ∈ Γ0 (N ) .
τ0

Now, we have a composition

Φx0 πf
Φf : X0 (N ) / J0 (N ) / Ef (9)

where Φx0 is the canonical inclusion with some base point x0

∫ x
Φx0 : x 7→
x0

and πf is the previous quotient map. One important fact is that, Φf gives a pullback of the
invariant differential ω on Ef , which is a nonzero multiple of ωf = 2πif (z)dz, i.e.

Φf (ω) = ω ◦ Φf = c2πif (z)dz, c ̸= 0. (10)

Besides the fact X0 (N ) could be defined over Q, it’s well known that Weil and Chow
proved that J0 (N ) is also defined over Q, and we can choose x0 such that Φx0 is also defined

43
over Q. Also, as we will expect, πf and Af are both defined over Q, c.f. [?]. So, the previous
conposition is defined over Q, known as the modular parametrisation. Also, c in (10) is also
in Q∗ , it’s called the Manin constant. ∫x
In particular, take x0 = i∞, then Φi∞ (x) = ∞ ∈ J(X). Then the composition

Φf : Γ0 (N ) \ H∗ / X0 (N ) / J0 (N ) / Ef

could be given explicitly using(9)(10)

Theorem 8.1. Let Ψf : C/Λf −→ Ef be the Weierstrass uniformisation, then
∫ τ ∑∞
an n
Φf (τ ) = Ψf (zτ ), zτ = c 2πif (z)dz = c q . (11))
i∞ n=1
n

The proof is just by definition and change of variables.

From the Eichler-Shimura construction, we can get other modular parametrization. The
pullback of f (τ ) ∈ S0 (Γ0 (N )) under the canonical covering map

π : X0 (M N ) → X0 (N ), Γ0 (N )τ 7→ Γ0 (M N )τ

is f (M τ ). And the canonical covering map is also defined over Q. Thus Composing with
π produces another modular parametrization for E. Also, by composing an isogeny over Q
also works. But in general a modular parametrization can be any morphism from modular
curves of any level to E, even not factoring through X0 (N ). Of course, something from
Eichler-Shimura construction construction is easy to handle.
Some theorem about the Eichler-Shimura construction follows.
Theorem 8.2. L(E, s) coincides with L(Ef , s).
L(E, s) It can be proved by the Eichler-Shimura relation about the reductions of E.
Theorem 8.3. The conductor of Ef is N .
From this, we know only N is a conductor of some elliptic curves, there might be new
forms of level N with integer coefficients. Thus, usually, there no new forms with integer
coefficients.
Theorem 8.4. The modular parametrization of E at the least level must comes from the
Eichler-Shimura construction (up to an isogeny over Q.)
The Eichler-Shimura construction is excellent, but it seems what we want is the other
direction: to find modular parametrization of a particular Elliptic curve. This embarrassing
condition is solved by the modularity theorem of Wiles.
Theorem 8.5. Let E be an elliptic curve over Q of conductor N . Then there exists a
newform f ∈ S2 (Γ0 (N )) such that L(E, s) = L(f, s).
Then, by the isogeny theorem of Faltings, and some other facts about Tate-module, we
know E is isogeny to Ef over Q.

44
9 Heegner points and rational points on elliptic curve
Now, we turn to Heegner points. Suppose Oτ = ON τ = OD , the ring class field H is
smaller than the general cases. Use the symbol previous, denote the Heegner point of τ as
(O, n, [a]). Let Φf : X0 (N ) → E be a modular parametrization given by Eichler-Shimura
construction, define
P = Φ(O, n, [a]) ∈ E(H)
Then use formula (3)
∑ ∑
TrH/K (P ) = Φf (O, n, [a])σ = Φf ((O, n, [a])σ (b)
σ∈G b∈Pic(O)
∑ ∑
= Φf (O, n, [ab−1 ]) = Φf (O, n, [b]) ∈ Ef (K)
b∈Pic(O) b∈Pic(O)

Similarly, use formula (4) we have

∑
TrH/K (P ) = Φf (O, n, [b])
b∈Pic(O)

Recall our discussion on Atkin-Lehner involution, we see that wN (f ) = εf, ε = ±1. If

ε = 1, by the definition of Φf , we have Φf ◦ wN = Φf . Thus, by (7),

Φf (O, n, [b)] = Φf ◦ wN (O, n, [nb]) = Φf (O, n, [nb])

and ∑ ∑
TrH/K (P ) = Φf (O, n, [nb]) = Φf (O, n, [b]) = TrH/K (P ) (12)
b∈Pic(O) b∈Pic(O)

So TrH/K (P ) ∈ E(Q). Similarly, if ε = −1,

TrH/K (P ) = −TrH/K (P ) (12′ ).

Besides, the ε appears in the functional equation of the L- series of f (or of Ef over Q),
let
Λ(f, s) := (2)s Γ(s)N s/2 L(f, s)
then
Λ(f, s) = −Λ(wN (f ), 2 − s) = −εΛ(f, 2 − s). (13)
Note that L(E, s) vanishes to even (resp. odd) order at s = 1 when ε = −1 (resp. ε = 1).
It’s also natural to consider E(H) as a G-module, and introduce the characters of G.
Generally, Let χ be a 1-dimensional (primitive) complex character of G, define

R = Z[χ(σ) : σ ∈ G].

45
For a Z[G] module M , we have a (Z[G], R)-bimodule M ⊗ R. Check directly M χ = {m ∈
M ⊗ R : mσ = χ(σ)m, ∀σ ∈ G} is a sub bimodule of M ⊗ R. For a sub Z[G]-module N , we
have an exact sequence of bimodules
0 / Nχ / Mχ / (M/N )χ .

For example, when χ is trivial, the case degenerates to be taking G−inviant elements.
Now, take M = E(H), G = GH/K . Like trace map, define
∑
Pχ = χ(σ)P σ .
σ∈G

It’s direct to very that

Pχ ∈ E(H)χ ⊂ V = E(H) ⊗ C
it’s called a Heegner vector. By (3)(4),
∑
Pχ = χ(b)Φf ((O, n, [ab])).
b∈O

If χ = 1, it’s just the trace map, providing a point over K.

We have just considered Heegner points for a single order, i.e. in one ring class field, and
didn’t see much magic. But as the previous examples shows, the trace map can relate the
rational points in different fields. It’s more meaningful when we consider Heegner points
P1 , P2 , P3 , ... on Ef defined over a tower of ring class fields H1 ⊂ H2 ⊂ H3 ... of K. They are
required to satisfy two properties, known as Euler system conditions. One relating the trace
maps and the Hecke operators, called norm compatibility. Rough speaking, these trace maps
(on Jacobian or divisor groups) should almost coincide with the Hecke operators. Also, by
modular parametrization Φf , the Hecke operators Tp induce multiplications by ap (f ), they
also should coincide with the trace maps on Ef . The other is the congruence relation, comes
from class field theory and Eichler-Shimura congruence relation. It’s very big and significant
a theme, contributing directly to many progresses in number theory since 1980’s. It’s direct
start is the conjecture of Gross [?] 11.2.
Here, we describe the Euler system of Heegner points introduced by Kolyvagin. To make
life easy, assuming the Heegner hypothesis gcd(dK , N ) = 1, or equivalently, all p|N are split
in K. Thus there exists an integral ideal of OK with OK /n ∼ = Z/N Z. Now, for an order Oc
with conductor c coprime to N , define nc = n ∩ O, so that O/nc ∼ = Z/N Z. Also define Hc be
its ring class field. Moreover, we are satisfied with the most easy case. For a Heegner point
(O, n, [a]), we always choose a = O, and denote it as
Qc = (O, nc , O).
Also define
Pc = Φf Qc ∈ E(Hc )
We further define S = {c : c square free, coprime to N, dK }. Thus for prime p|c, d =
c/p ∈ S. Define a trace map ∑
Tre (Qc ) = Pcσ
σ∈GHc /Hd

46
transferring Heegner points defined over Hc to a divisor defined over Hd . Easy to see it
coincides with the trace map on E(Hc ),
Φf (Tre (Qc )) = Tre (Pc )
here we take similar symbol for the trace map of E.
Then we obviously have norm compatibility,
Proposition 9.1. Let Te as usual be a Hecke operator, then
Tre (Qc ) = Tp (Qd ).
Consequently, for E,
Tre (Pc ) = ap (f )Pd
It’s almost direct by checking definition. Also we have congruence relation
Proposition 9.2. Each prime factor λc of e in Hc divides a unique prime λd in Hd , and we
have the congruence Pc = Frob(λd )Pd ( mod λc ). Here Frob is the Frobenius map.
Use this Euler system , Kolyvagin could construct certain Galois cohomology classes, which
was used to give partial proof of Gross’s conjecture.
Theorem 9.3. Let PK = TrH1 /K P1 ∈ E(K). Suppose PK is not torsion, then E(K) is of
rank 1..
The attractiveness of this theorem is that, the existence of a point of infinite order (big)
surely implies the the rankE(K) > 1, but it unexpectedly bounds the rank to be 1 (small).
This is the magic of Heegner points.
In this case, by(12)(12’), PK ∈ E(Q) if and only if ε = 1, and by (13) and only if L(E, 1)
vanish to an odd order. And PK ∈ E(Q) implies the rankE(Q) = 1.
The other side of Gross’s conjecture was solved by the Gross-Zagier-Zhang formula, which
relates the Euler system to the special value of L(E/K, s). Recall our definition of Heegner
vectors.
Theorem 9.4. Let PK be defined as in theorem 9.3, Pn,χ be the Heegner vector of Pn , we
have
1, ⟨PK , PK ⟩ = L′ (E/K, 1);
2, ⟨Pn,χ , Pn,χ ⟩ = L′ (E/K, χ, 1).
Thus, PK is not torsion if and only if L′ (E/K, s) dose not vanish; Pn,χ non-zero if and
only if L′ (E/K, χ, s) dose not vanish .
This two theorem can be combined to prove the following theorem concerning the BSD
conjecture
Theorem 9.5. If E is an elliptic curve over Q and ords=1 L(E, s) ≤ 1, then
rank(E(Q)) = ords=1 L(E, s).
Part of this theorem is easier. When ords=1 L(E, s) = 1, ε = 1 and non-torsion Heegner
must belongs to E(Q), so, rankE(Q) = 1. Some results about L-series of elliptic curves
guarantee that we can choose a K satisfying heegner Hypothesis and L(E/K, s) vanish to
order 1. Then Theroem 9.4 implies PK is not torsion, so rankE(Q) = 1.

47
References
[1] L.Atkin; J.Lehner. Hecke operators on Γ0 (m), Math. Ann., 185, 1970, 134160.

[2] D.A. Cox. Primes of the form x2 + ny 2 : Fermat, class field theory, and complex multi-
plication. Wiley-Interscience, 1997.

[3] H. Darmon. Rational points on modular elliptic curves. CBMS Regional Conference Series
in Mathematics, 101.

[4] F. Diamond; J. Shurman. A first course in modular forms. GTM 228. Springer-Verlag,
New York, 2005.

[5] B.H. Gross. Heegner points on X0(N). Modular forms (Durham, 1983), Horwood, Chich-
ester, pages 87105, 1984.

[6] B.H. Gross. Kolyvagin work on modular elliptic curves. L-functions and arithmetics.
L-functions and arithmetic (Durham, 1989), 235256, 1991.

[7] B.H. Gross and D. Zagier. Heegner points and derivatives of l-series. Invent. math,
84(2):225320, 1986.

[8] A.W. Knapp, Elliptic curves. Mathematical Notes, 40. Princeton University Press,
Princeton, NJ, 1992.

[9] S. Lang. Elliptic functions. GTM 112. Springer-Verlag, New York, 1987.

[10] J.S.Milne. Modular Functions and Modular Forms. www.jmilne.org/math/CourseNotes/mf.html

[11] J. Neukirch. Algebraic number theory. Springer-Verlag, Berlin, 1999.

[12] J.H. Silverman. The Arithmetic of Elliptic Curves, GTM 106, Springer-Verlag, New
York, 1986.

48
 一类特殊的非线性算子的一致有界定理

孙奥*

一致有界定理，有时也称为共鸣定理，是泛函分析中一个最基本的定理。定理叙述如
下

定理 1. 假设 E,F 是两个 Banach 空间，设 {Ti }i∈I 是一族 E → F 的连续线性算子。

如果有
sup ∥Ti x∥ < ∞ ∀x ∈ E
i∈I
那么就有结论
sup ∥Ti ∥ < ∞
i∈I

从定理的叙述我们可以明白为什么这个定理被称为一致有界定理。
一致有界定理是对线性算子成立的。那么是否能够推广到一般的（非线性）算子呢？
从定理的叙述我们知道，我们应该恰当地定义非线性算子的“有界”。同样，由于非线性
算子太过丰富，即使是加上连续性的条件，仍然不一定能够有类似于一致有界定理的条
件。所以我们还要恰当地给予非线性算子一些限制。本文即试图从这几个方面出发考察非
线性算子的情况。

1 主结果与证明

以下，我们总假设 E, F 是 Banach 空间，并且为了便于说明，均设为实线性空间，∥∥˙

是其上的范数。另外以下提到的所有算子除非特殊声明都是 E → F 的连续映射。我们用
B(x, r), x ∈ E, r > 0 表示集合 {y : ∥y − x∥ ⩽ r}

定义 1. 我们称一个算子 T 是齐性的，如果它满足对任意 x ∈ E，均有

T (x) = ∥x∥T ( ∥x∥
x
)
*
数 11

49
 容易验证，齐性算子（不一定线性）构成了一个线性空间。而且我们知道，所有的线
性算子都是齐性算子。

定义 2. 设 T 是一个齐性算子，我们称 sup ∥T (x)∥ 为其范数。

|x|=1
并且在不混淆的情况下，也记作 ∥T ∥

事实上，可以验证这个范数就是齐性算子空间的范数。但是我们这里用不到这个结
论，所以我们以后提 T 的范数仅仅是用于指其某种意义上的“界”。

定义 3. 我们称一个齐性算子 T 是有界的，当它的范数是有限的。并且此时容易验证，有
∥T (x)∥ ⩽ ∥T ∥∥x∥ ∀x ∈ E

尽管齐性算子与线性算子有很多相似之处，但是我们仍然不能得出类似于线性算子的
齐性算子的一致有界定理。下面是一个例子

例 1. 考虑复数域 C 作为 2 维的实线性空间。利用 Urysohn 引理可以在单位圆周 S 1 上

构造连续实值函数 gn (n ⩾ 2)，使得其值域为 [0, n]，且满足
2πi
gn (e n )=n
∪ 3
gn (e2πiθ ) = 0, ∀θ ∈ [0, 2n
1
] [ 2n , 1]
定义一族非线性算子 {Tn }∞
n=1 是 C → R 的算子，满足

Tn (x) = ∥x∥gn ( ∥x∥

x
)
于是这一族算子是齐性的，并且每一个都是有界的。由于对于任意的 x ∈ C，仅有有
x
限个 gn 使得 gn ( ∥x∥ ̸= 0)，故
sup ∥Ti x∥ < ∞ ∀x ∈ E
i∈Z
但是由于 ∥Tn ∥ = n，我们知道 lim → ∞∥Tn ∥ = ∞。从而这一族算子不满足一致有界
n
定理。

下面我们引入一个特殊的条件。

定义 4. 我们称一个线性算子 T 是 k-控制的，如果 ∀x, y ∈ E，有

∥T (x + y)∥ ⩽ k max{∥T (x)∥, ∥T (y)∥}
由于 ∥T (2x)∥ = 2∥T (x)∥，我们知道这里的 k ⩾ 2。
特别的，对于线性算子 L，由范数不等式 ∥L(x + y)∥ ⩽ ∥L(x)∥ + ∥L(y)∥, 我们知道
线性算子是 2-控制的。

可以验证，上一个例子中的算子不是 k-控制的（∀k > 0）。

50
例 2. 由范数不等式，我们知道 E 上的范数就是一个 2-控制的齐性非线性算子。因此我
们知道，以上的定义都不是平凡的（即满足这些性质的算子最终只能是线性算子）。

下面是我们的主定理

定理 2. 如果一族齐性 k-控制算子（k ⩾ 2）{Ti }i∈I 满足

sup ∥Ti x∥ < ∞ ∀x ∈ E
i∈I
那么就有结论
sup ∥Ti ∥ < ∞
i∈I
P roof
先证明一个引理：设 T 是一个齐性 k-控制算子，则 ∀x ∈ E, r > 0 我们有
sup ∥T x′ ∥ ⩾ kr ∥T ∥
x′ ∈B(x,r)
引理的证明：∀x ∈ E, ∀y 使得 ∥y∥ ⩽ 1，由 k-控制的条件，我们有
∥T (ry)∥ ⩽ k max{∥T (ry + x)∥, ∥T (x)∥ ⩽ k sup ∥T x′ ∥}
x′ ∈B(x,r)
上式左端对所有的满足要求的 y 取上确界，即有
r∥T ∥ ⩽ k sup ∥T x′ ∥
x′ ∈B(x,r)
整理即得引理成立。
回到原命题，现设 α > 0 是使得 α > 2k 2 + 1 的正实数。我们用反证法，假设结论不
对，那么 sup ∥Ti ∥ = ∞
i∈I
我们可以从中选出一个子列 {Tn }∞
n=1 ，使得 ∥Tn ∥ ⩾ (α + 1)
n

下面我们选取 E 中的一个子列 {xn }∞

n=1 如下：首先取 x0 = 0。对 n ⩾ 1，假设我们

已经取好了 x0 , x1 , · · · , xn−1 ，利用引理，我们可以选择 xn 使得

∥xn − xn−1 ∥ ⩽ α−n ，且 ∥Tn xn ∥ ⩾ 1 −n
2k α ∥Tn ∥
注意到，由于 k ⩾ 2 我们有 α > 9，从而 {xn }∞
n=1 是 Cauchy 列。假设其收敛到点

x ∈ E。那么我们有
∥x − xn ∥ = lim ∥xm − xn ∥ ⩽ lim α−n−1 + α−n−2 + · · · + α−m ⩽ α−n α−1
1
m→∞ m→∞
现在考虑 ∥Tn xn ∥ ⩽ k max{∥Tn x∥, ∥Tn (xn − x)∥}，而我们又有
k∥Tn (xn − x)∥ ⩽ k∥Tn ∥∥xn − x∥ ⩽ kα−n α−1
1
∥Tn ∥
1 −n
又 ∥Tn xn ∥ ⩾ 2k α ∥Tn ∥，于是由 α > 2k 2 + 1，我们有
∥Tn xn ∥ > k∥Tn (xn − x)∥
从而只可能 ∥Tn xn ∥ ⩽ k∥Tn x∥，由此我们得到
∥Tn x∥ ⩾ k1 ∥Tn xn ∥ ⩾ 1
2k2
α−n ∥Tn ∥ ⩾ 1 α+1 n
2k2
( α )

51
 则有 lim ∥Tn x∥ = ∞，与假设矛盾。故反设不成立。从而定理得证。
n→∞

推论 1. 由于有界线性算子均是齐性 2-控制算子，所以特别地，我们得到线性算子的一致
有界定理。

2 注记

2.1

这个证明主要思想来自参考文献 [1]。齐性、k-控制等名词都是我自己命名的，不知道
学术界是否已有固定的名称。

2.2

现在常见的一致有界定理的证明有两种。参考文献 [2] 中列出来的是一个用 Baire 定

理的证明。而历史上真正最早出现的是 Hahn 的被称为”gliding hump” 的方法，那是与参
考文献 [1] 类似的方法。
事实上，在连续算子的情况下，本文的结论利用参考文献 [1] 的证明是很容易得到的。
本文真正有趣的结论在于，主定理的证明完全没有利用连续性。于是完全照抄之前的证
明，我们可以得到这样的定理

定理 3. 如果一族齐性有界 k-控制算子（k ⩾ 2）{Ti }i∈I 满足

sup ∥Ti x∥ < ∞ ∀x ∈ E
i∈I
那么就有结论
sup ∥Ti ∥ < ∞
i∈I
这个结论对不连续的齐性 k-控制算子 也对。

注意，不连续的齐性 k-控制算子是存在的。

例 3. 设 k > 2，T 是 C → R 的算子，满足

T (z) = kz，当 ℜ(z)ℑ(z) > 0
T (z) = z，当 ℜ(z)ℑ(z) ⩽ 0
则 T 是一个不连续的齐性 k-控制算子。而且是有界的。

对于不连续的情况，参考文献 [2] 中的证明由于要用到连续算子的闭集原像仍然是闭

集这样一个条件，所以失效了。而本文中的主定理的证明完全没有用到连续性，故依旧成
立。

52
 当然，由于对于线性算子，连续与有界是等价条件，从而没有这种区别。但是非线性
算子，这两者不再等价。这也体现了非线性算子的特殊之处。

2.3

最后稍微说一下怎么发现这个条件的。参考文献 [2] 中的证明可以得到一个廉价的推

广：
只要连续算子 T 满足齐性以及次线性：
∥T (x + y)∥ ⩽ ∥T x∥ + ∥T y∥, ∀x, y ∈ E
那么就有一致收敛定理。
我当时觉得次线性这个推广太廉价了，就想找一个更加强的推广。可以证明 k-控制
这个条件比次线性要强。次线性可以推出 2-控制，但是反过来不一定。

例 4. 设 k > 2，T 是 C → R 的算子，满足

T (z) = kz，当 ℜ(z)ℑ(z) > 0
T (z) = z，当 ℜ(z)ℑ(z) ⩽ 0
这里 ℜ 表示一个复数的实部，ℑ 表示一个复数的虚部。则 T 是齐性的 k 控制算子。
√
但是由 |T (1 + i)| = k 2 > |T (1)| + |T (i)|，可知 T 不是次线性算子。

在我寻找各种例子的过程中，发现许多例子都是不连续的算子。这使得我在阅读了参
考文献 [1] 后，发现推广到不连续算子的这个结果更加强一些。
当然，除开这一点来说，参考文献 [1] 中的证明也是很漂亮的不利用纲性的证明。

关于齐性这个条件，事实上很多文献中指的是下面的定义

定义 5. 称 T 是实线性空间 E 中齐性算子，若对任意的 α ∈ R, x ∈ E，有

T (αx) = αT (x)

注意到这个条件可以推出本文开头的条件，但反过来不对（考虑取范数这个齐性算
子）。但如果没有 k-控制这样的条件，这一类算子依旧不能推出类似于一致有界定理。只
需将例 1 中的例子稍作修改，使得其中的每个算子的取值关于原点中心反对称即可。

53
参考文献

[1] A really simple elementary proof of the uniform boundedness theorem, Alan D. Sokal,
Amer. Math. Monthly 118, 450-452 (2011);

[2] Functonal Analysis, Sobolev Spaces and Partial Differential Equations, Haim Brezis,
Springer;

54
 墨尔本交流感受

林艺儿*

2014 年 7 月 5 日，我们数学系六人来到澳大利亚墨尔本参加为期两周的墨尔本夏令
营。本次夏令营主要是为了向我们展示数学、物理、化学中的前沿进展以及不同学科之间
的交叉联系。经过长达 11 小时的飞行，抵达了这座世界最宜居住的城市。墨尔本位于南
半球，刚下飞机，一阵寒意袭来，赶紧换上冬装。我们的住处是墨尔本大学的学生宿舍，
单人间虽然不大，但干净整洁各种设施俱全。
本来双休日食堂是不设午餐的，但考虑到我们旅途劳顿，特意为我们加开了午餐，使
我们倍感温暖。之后两天参观了校园和实验室，墨大校园虽然不大，但干净而典雅，由于
正值寒假，校园里的学生并不多。
利用晚上的休息时间，我们坐电车去了市中心。墨尔本的交通十分便利，利用电车和
火车可以到达城市和郊外的每一个角落。墨尔本是一个融合了各国文化的城市。走在市中
心街头，可以看到欧洲、亚洲各种不同的面孔，各国风味的餐馆，在这一个城市里就可以
品尝到世界各地的美食。汉语是这里的第二语言，城市里华人很多，走在街上也随处可见
各种汉语的标语和招牌，听见人们用中文交谈，倍感亲切。
白天的时间我们主要是在实验室跟着老师做实验，或者上课。数学课上两位教授分别
介绍 Markov 链在 PageRank 问题上的应用，以及证明了平面对称群共有 17 种，这两种
问题都结合了理论和生活，十分有趣。
双休日是自由时间，因此我们利用这两天游玩了城里城外值得去的地方。第一去了战
争纪念馆，皇家植物园和水族馆。战争纪念馆位于皇家植物园旁边，用于悼念为国捐躯的
澳洲人。植物园很大，里面各种植物枝繁叶茂，感觉仿佛进入了原始森林，美景令人心旷
神怡。值得一提的墨尔本的天气，虽然是蓝天白云，但不时下起淅淅沥沥的小雨。相比之
下，水族馆就显得较为一般，除了一些在中国看不到的珍奇动物，比不上国内的水族馆。
晚登上墨尔本最高的建筑 skydeck88，在这座 88 层建筑俯瞰被灯光点缀的整座城市，
美不胜收。虽然墨尔本的商店晚上 5 点（周四周五会晚一些）关门，但居然都不关灯。原
*
数 23 班

55
来是政府为了防盗，规定“人走灯亮”。
第二天我们从 Belgrave 乘坐去郊外的 puffing billy 火车，虽然墨尔本的火车很多，铁
轨四通八达，但像 puffing billy 这辆以蒸汽作为动力的是独一无二的。火车穿过原始森林，
到达郊外。与传统火车不同的是，乘坐时可以坐在“窗沿”上将双脚伸出去，火车运行时，
一排排树木飞快地向后移去，清风拂面，可以看到火车头的滚滚蒸汽。火车经过之地，路
人都会向火车里的游客招手，十分友好。只不过坐的时间长了难免会感觉寒冷（7 月是墨
尔本最冷的时候），郊外 Landlake 十分幽静，向森林深处走去，呼吸新鲜的空气，在北京
待久了我都快遗忘了清新空气的味道，因此不管吸多少口都感觉不够。待了两个小时，返
程的火车来了，这才恋恋不舍地离开。
每天的晚上都是自由时间，还去了很多有趣的地方，例如世界第三大摩天轮 Melbourne
Star，Victory Market。此外我们还用了一整天的时间去了动物园，与袋鼠和树懒等澳洲
国宝亲密接触。两周的时间虽然不长，但也使我们很好的感受了墨尔本这座城市的科学、
文化和风土人情，我对这座城市有很好的印象，如果有机会，我还会再来的。

56
 征稿启事
《荷思》是清华大学数学系学生自主创办的数学学术刊物，面向各个院系中对数学感兴趣
的本科生及研究生。
本刊欢迎全校师生任何与数学有关的投稿，无论是长篇的论述，还是精彩的小品，抑或
学习/教学的心得、习题的妙解。在原作者的允许下，推荐他人的作品也同样欢迎。
为了编辑方便，建议投稿者能够提供电子版，并且采用 LATEX (推荐) 或 Word 排版。来
稿请注明作者，联系方式。

投稿请寄：[email protected]
欢迎访问荷思网站：http://math.tsinghua.me/hesi/

《荷思》编辑部
2014-12

主办：清华大学 数学科学系 《荷思》编辑部

主编：杨宇轩
顾问：张端阳
编委：（按姓名拼音为序）
高安凝哲 蒋 旃 林艺儿 孟 成 邱聪灵 曲日浩
孙 奥 王 健 王力汉 王 怡 叶峪廷 殷思瑶
郑志伟
封面：陈凌骅
排版：杨宇轩

联系本刊：
[email protected]
北京市 海淀区 清华大学 28 号楼 226

57